Large-scale evacuation : the analysis, modeling, and management of emergency relocation from hazardous areas 9781482259858, 1482259850, 9781315119045, 9781482259865, 9781351645324, 9781351635837

Table of contents :
Content: Natural and technological hazards requiring evacuation management --
Protective actions and protective action decision makin --
Who leaves and who does not --
When do evacuees leave --
Managing evacuation logistics --
Evacuation behavioral forecasts --
Strategies for managing evacuation demand and capacity --
Evacuation traffic modeling and simulation --
Evacuation termination and reentry --
Case studies.

Citation preview

Large-Scale Evacuation

Large-Scale Evacuation introduces the reader to the steps involved in evacuation modeling for towns and cities, from understanding the hazards that can require largescale evacuations, through understanding how local officials decide to issue evacuation advisories and households decide whether to comply, to transportation simulation and traffic management strategies. The author team has been recognized internationally for their research and consulting experience in the field of evacuations. Collectively, they have 125 years of experience in evacuation, including more than 140 projects for federal and state agencies. The text explains how to model evacuations that use the road transportation network by combining perspectives from social scientists and transportation engineers, fields that have commonly approached evacuation modeling from distinctly different perspectives. In doing so, it offers a step-by-step guide through the key questions needed to model an evacuation and its impacts to the evacuation route system as well as evacuation management strategies for influencing demand and expanding capacity. The authors also demonstrate how to simulate the resulting traffic and evacuation management strategies that can be used to facilitate evacuee movement and reduce unnecessary demand. Case studies, which identify key points to analyze in an evacuation plan, discuss evacuation termination and re-entry, and highlight challenges that someone developing an evacuation plan or model should expect, are also included. This textbook will be of interest to researchers, practitioners, and advanced students. Michael K. Lindell is Emeritus Professor at the Hazard Reduction & Recovery Center at Texas A&M University and is Affiliate Professor at the Institute for Hazard Mitigation Planning and Research at University of Washington. Pamela Murray-Tuite is Associate Professor in the Department of Civil Engineering at Clemson University. Brian Wolshon is the Edward A. and Karen Wax Schmitt Distinguished Professor and Director of the Gulf Coast Research Center for Evacuation and Transportation Resiliency at Louisiana State University. Earl J. Baker is Emeritus Professor in the Department of Geography at Florida State University.

This page intentionally left blank

Large-Scale Evacuation The Analysis, Modeling, and Management of Emergency Relocation from Hazardous Areas Michael K. Lindell, Pamela Murray-Tuite, Brian Wolshon, and Earl J. Baker

First published 2019 by Routledge 52 Vanderbilt Avenue, New York, NY 10017 and by Routledge 2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN Routledge is an imprint of the Taylor & Francis Group, an informa business © 2019 Taylor & Francis The right of Michael K. Lindell, Pamela Murray-Tuite, Brian Wolshon, and Earl J. Baker to be identified as authors of this work has been asserted by them in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging-in-Publication Data Names: Lindell, Michael K., author. | Murray-Tuite, Pamela M. (Pamela Marie), 1975author. | Lindell, Michael K., author. | Wolshon, P. Brian (Paul Brian), author. | Baker, Earl J., author. Title: Large-scale evacuation : the analysis, modeling, and management of emergency relocation from hazardous areas / Michael K. Lindell, Pamela Murray-Tuite, Brian Wolshonc, and Earl J. Baker. Description: New York, NY : Routledge, 2019. | Includes bibliographical references and index. Identifiers: LCCN 2018030325 (print) | LCCN 2018044305 (ebook) | ISBN 9781315119045 (Master) | ISBN 9781482259865 (WebPDF) | ISBN 9781351645324 (ePub) | ISBN 9781351635837 (Mobipocket/Kindle) | ISBN 9781482259858 (hbk) | ISBN 9781482259865 (ebk) Subjects: LCSH: Evacuation of civilians--Planning. | Emergency transportation--Planning. | Emergency management--Planning. Classification: LCC HV554 (ebook) | LCC HV554 .L56 2019 (print) | DDC 363.34/72--dc23 LC record available at https://lccn.loc.gov/2018030325 ISBN: 978-1-4822-5985-8 (hbk) ISBN: 978-1-315-11904-5 (ebk) Typeset in Sabon by Integra Software Services Pvt. Ltd.

Contents

List of Figures List of Tables Acknowledgements List of Acronyms

Chapter 1 Chapter 2 Chapter 3 Chapter 4 Chapter 5 Chapter 6 Chapter 7 Chapter 8 Chapter 9 Chapter 10 Chapter 11 Glossary Index

Introduction and Overview Natural and Technological Hazards Requiring Evacuation Management Protective Actions and Protective Action Decision Making Who Leaves and Who Does Not When Do Evacuees Leave? Managing Evacuation Logistics Evacuation Behavioral Forecasts Strategies for Managing Evacuation Demand and Capacity Evacuation Traffic Modeling and Simulation Evacuation Termination and Reentry Case Studies

x xii xiii xiv

1 15 46 67 100 121 142 177 219 255 281 316 320

This page intentionally left blank

Detailed Contents

List of Figures List of Tables Acknowledgements List of Acronyms

Chapter 1 Introduction and Overview 1.1 1.2 1.3 1.4

Evacuation Fundamentals Evacuation Modeling Need for Multiple Disciplines Intended Audience and Scope

x xii xiii xiv

1 2 4 9 11

Chapter 2 Natural and Technological Hazards Requiring Evacuation Management

15

2.1 2.2 2.3 2.4 2.5

15 21 24 25 34

Floods Tsunamis Wildfires Hurricanes Hazardous Materials Releases

Chapter 3 Protective Actions and Protective Action Decision Making

46

3.1 3.2 3.3

Overview of Sheltering In-place Overview of Evacuation PAR Decision Making

46 52 53

Chapter 4 Who Leaves and Who Does Not

67

4.1 4.2

68 76

Evacuation Behavior Predictors of Evacuation

viii Detailed Contents

Chapter 5 When Do Evacuees Leave?

100

5.1 5.2 5.3 5.4

Authorities’ Decision Times Warning Dissemination Times Evacuation Preparation Times Evacuation Departure Times

100 102 106 113

Chapter 6 Managing Evacuation Logistics

121

6.1 6.2 6.3 6.4 6.5 6.6

122 124 126 129 130 133

Driver Evacuation Behavior Evacuation Accommodations Evacuation Destinations Evacuation Travel Times Travel Mode Evacuation Routes

Chapter 7 Evacuation Behavioral Forecasts

142

7.1 7.2 7.3

142 143 146

Hazard Maps Defining Geographical Zones Producing Evacuation Demand Estimates

Chapter 8 Strategies for Managing Evacuation Demand and Capacity

177

8.1 8.2 8.3 8.4

178 187 206 212

Demand Management Strategies Supply Management Strategies Traffic Management in Emergency Operations Centers Future Systems

Chapter 9 Evacuation Traffic Modeling and Simulation

219

9.1 9.2 9.3 9.4 9.5 9.6 9.7

220 222 224 240 241 242 247

Background Levels of Analysis Models of Key Evacuation Variables and Assumptions Model Refinements and Validation Megaregion Simulation Scenario Development for Evacuation Simulation Need for Multiple Scenarios

Detailed Contents

ix

Chapter 10 Evacuation Termination and Reentry

255

10.1 10.2 10.3 10.4 10.5 10.6

257 259 260 262 265 275

Permanent Migration Reentry Plan Compliance Evacuee Concerns Evacuee Information Sources An Overview of Disaster Recovery Planning Reentry Planning Principles

Chapter 11 Case Studies 11.1 11.2 11.3 11.4

Glossary Index

Case Study 1—Hurricane Katrina Case Study 2—2007 Southern California Wildfires Case Study 3—Nuclear Power Plant Traffic Management Study Case Study 4—Hazmat Transportation Evacuation Planning

281 282 287 293 306 316 320

Figures

1.1 1.2 1.3 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8

General Evacuation Modeling and Planning Framework Generalized Timeline for Short Notice Events Generalized Timeline for No-Notice Events Stage Rating Curve Map of the Distribution of Precipitation From a Storm FEMA Flood Insurance Rate Map Tsunami Emergency Notification/Warning System State of Texas Hurricane Risk Areas Hurricane Tracking Map Effects of Wind Speed on Plume Dispersion Wind Rose from 3am to 6pm on the First Day of the TMI Accident 2.9 Vulnerable Zones Around a Fixed-site Facility and Transportation Route 2.10 EPZ Around a Nuclear Power Plant with 2, 5, and 10 Mile Zones and Eight Directional Sectors and Keyhole Evacuation Zone Shaded 3.1 Precautionary Shelter In-Place Zones on Either Side of a Keyhole Evacuation Zone 3.2 Effectiveness of Protective Actions in a Nuclear Power Plant Emergency 3.3 Hurricane Protective Action Zones 3.4 Evacuation Decision Tree 3.5 Evacuation Decision Arcs 4.1 Protective Action Decision Model 4.2 Percent of Respondents in New York City Following Hurricane Irene Saying It Would Be Unsafe to Stay at Home If Struck By a Category 1, 2, or 3 Hurricane, by Evacuation Zone (NS = Non-surge) 5.1 Authorities’ Decision Time Curve for Rapid Onset Disasters 5.2 Warning Dissemination Curves For Rapid Onset Incidents 5.3 Warning Dissemination Curves For Normal Hurricane Landfall 5.4 Evacuation Preparation Time Curves For Normal Hurricane Landfall

5 7 8 16 18 20 23 31 33 36 38 40

43 56 57 58 61 64 70

78 101 103 104 106

Figures

5.5 5.6 5.7 7.1 7.2 7.3 8.1 8.2

8.3

8.4 8.5 8.6

8.7 9.1 9.2 9.3 10.1 11.1 11.2 11.3

Evacuation Departure Times for Rapid-Onset Incidents Evacuation Departure Times for Hurricane Ike Evacuation Departure Times for Hurricane Floyd Example of Cumulative Probability for Travel Mode Framework for Generating Activity Chains Framework for Detailed Activity Plans for Advance Notice Events Freeway Contraflow Lane Use Configurations for Evacuations Schematic (Top) and Field Photo (Bottom) of Median Cross Over Contraflow Loading Configuration Interstate 10 at Loyal Avenue, Kenner, Louisiana Ineffective Loading of Freeway Contraflow at Interstate 10 at Loyola Avenue, Hurricane Katrina Evacuation of New Orleans Southeast Louisiana Regional Evacuation Route Management Plan Texas DOT Evaculane Road Sign, Houston Texas Hurricane Evacuation Route Directional Shoulder Pavement Markings (Normal Lanes at Left and Contraflow Lanes at Right), US-290 – Texas Sample Movements After Turn Restrictions Evacuation Modeling Spectrum Overview of Interactions for Scenario Construction Tornado Diagram for the San Patricio County Evacuation Analysis Primary Sources of Emergency Information during the Hurricane Ike Evacuation Evacuspot Sculpture Near the French Quarter Time of Day Scenario Tree Schematic Map of Linden New Jersey

xi 115 116 117 159 165 169 192

194

195 197 199

200 202 222 244 249 263 287 296 306

Tables

2.1 2.2 2.3 3.1 3.2 3.3 3.4 3.5 3.6 4.1 4.2 4.3

5.1 7.1 9.1 10.1 10.2 10.3 11. 1 11.2a 11.2b 11. 3 11. 4

Saffir-Simpson Hurricane Scale Track Forecast Accuracy (2012–2016) Atmospheric Stability Classes Characteristics of Warning Mechanisms EPA Protective Action Guides (PAGs) Facilities with Highly Vulnerable Populations Characteristics of Special Facility Users Outcome Matrix for the Decision to Evacuate Example ETE Table Evacuation Rates in Hurricane Floyd by Risk Area and Perceived Vulnerability in a 125 mph Storm Evacuation Rates in Hurricane Floyd by Risk Area and Type of Official Evacuation Notice Heard Percent of Respondents Saying They Relied a Great Deal on Information Sources During Hurricanes Isaac and Sandy Influential Factors in Family Reunification for No-Notice Events Sample Evacuation Participation Rates for Hurricanes Timing Scenarios Percent of Evacuees Traveling to Different Destinations in Hurricane Rita Disaster Recovery Functions Infrastructure Facilities Approximate Evacuation and Shadow Boundary Evacuation Extents Nuclear Power Plant Evacuation Scenarios for Unit 1 Nuclear Power Plant Evacuation Scenarios for Unit 2 Evacuation Statistics for Site 1 Evacuation Statistics for Site 2

27 34 36 54 56 59 60 62 63 80 84

88 110 153 245 264 266 271 295 297 298 300 302

Acknowledgements

This book is based on work supported by the National Science Foundation under Grants CMMI-0654023, SES-0826873, ENG-0219155, SBE0527699, SBE-0838654, IIS-1212790/IIS-1540469 and CMMI-1760766. Other work was supported by the US Army Corps of Engineers through an extensive series of contracts to E.J. Baker and to M.K. Lindell. Additionally, some work was supported by contracts with the Texas Division of Emergency Management and the Virginia Transportation Research Council. A considerable amount of work performed by P.B. Wolshon was supported by agencies such as the United States Department of Transportation through their University Transportation Centers Program, the Louisiana Department of Transportation and Development, The Transportation Research Board and National Cooperative Highway Research Program, and the United States Nuclear Regulatory Commission through the Sandia National Laboratories. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the project sponsors.

List of Acronyms

ACH ATIS CCTV CMS CV DMS DOT DOT-ERG DPW EAS EHS EMA EOC EOP EPCRA EPZ ERPA ERS ETE FEMA FIRM GIS HAR HOV IC IDLH ITS LOC MEOW MOM MPC NHC NOAA NPP NRC PADM PAG

Air changes per hour Advanced Traveler Information System Closed Circuit Television Changeable Message Sign Connected Vehicle Dynamic Message Sign Department of Transportation DOT-Emergency Response Guidebook Department of Public Works Emergency Alert System Extremely Hazardous Substance Emergency Management Agency Emergency Operations Center Emergency Operations Plan Emergency Planning and Community Right to Know Act Emergency Planning Zone Emergency Response Planning Area Evacuation Route System Evacuation Time Estimate Federal Emergency Management Agency Flood Insurance Rate Map Geographic Information System Highway Advisory Radio High Occupancy Vehicle Incident Commander Immediately Dangerous to Life or Health Intelligent Transportation System Level of concern Maximum Envelope Of Water Maximum Of MEOWs Manual Police Control National Hurricane Center National Oceanographic and Atmospheric Administration Nuclear Power Plant Nuclear Regulatory Commission Protective Action Decision Model Protective Action Guide

Acronyms

PAR SLOSH TAZ TLV TWC USACE V2I V2V VMS VZ WEA

xv

Protective Action Recommendation Sea, Lake, and Overland Surges from Hurricanes (hurricane surge model) Traffic Analysis Zone Threshold Limit Value Tsunami Warning Center US Army Corps of Engineers Vehicle to Infrastructure communication Vehicle to Vehicle communication Variable Message Sign Vulnerable Zone Wireless Emergency Alert

This page intentionally left blank

Chapter 1

Introduction and Overview

Evacuation is a protective action that involves people relocating from a threatened area to a safer area. As Perry (1978) noted, evacuations can differ with respect to a number of different dimensions. These include their timing in relation to disaster impact (pre-impact or post-impact) and their duration—ranging from perhaps a few hours to permanent relocation. In addition, they differ in their degree of pre-impact planning (from completely improvised to substantially planned), the number of people involved (ranging from one person to millions of people), and the distance to safety (ranging from a few feet to many miles). The simplest evacuations—such as well practiced building fire evacuations—involve only a few people, require walking only a short distance, are well planned and exercised, take place pre-impact, and last only a short time. At the other extreme are mass evacuations that involve millions of people evacuating tens or hundreds of miles in vehicles, require a significant amount of improvisation despite a substantial amount of planning, take place before or after disaster impact has disrupted communication and transportation systems, and displace people for weeks, months, years, or even permanently. It is these mass evacuations that are the focus of this book, particularly the need to develop evacuation plans that are based on empirical data about how households respond to environmental threats coupled with engineering models of traffic flows. For many decades, practitioners and researchers have sought new techniques and systems to move people faster and more safely during evacuations. Some of these methods and strategies have focused on evacuees directly—using improved methods of communication to help them make faster and better informed decisions. Others have focused on transportation systems to better utilize personnel, modal, technological, and infrastructure resources to move people. Over time, this evolution has brought about major changes in the way evacuations are planned and implemented. It has also resulted in the emergence of specialized areas of emergency management study in the physical and social sciences, engineering, planning, and public administration. This book summarizes the current state of knowledge in many of these fields, with a particular focus on the practical application of this knowledge. It also highlights

2 Chapter 1 · Introduction and Overview many of the latest emerging topics that have been identified for needed study in the aftermath of recent high profile evacuations. Many people are surprised to learn that mass evacuations are quite common. A study of emergencies over a 10-year period showed that evacuations involving 1,000 or more persons occur, on average, about every two weeks somewhere in the United States (Dotson and Jones 2005). However, the large scale attention-grabbing evacuations that capture news headlines are considerably less frequent. In fact, of the events studied, only about 25% of them involved more than 5,000 people and only about 5% of them included 100,000 or more people. Because of their infrequent occurrence, large-scale evacuations can be extremely challenging to implement, so that is why they are the main focus of this book. Decades of operational experience have shown that when a mass evacuation of an urban area is needed, the methods used to move people become quite complex and can require travel over long distances and over extended periods of time. Not only do such conditions increase the risk of harm in an evacuation zone, they also affect much larger areas. In extreme cases, evacuations can have regional impacts. Past hurricane evacuations in Miami and New Orleans, for example, have impacted travel conditions statewide throughout Florida and Louisiana (Wolshon 2007) and even affected bordering states. Despite the multitude of conditions that can influence any specific evacuation, the history of prior evacuations indicates that there is actually a small set of key variables and fundamental relationships that govern all evacuation processes. These variables can be expressed in spatial and temporal terms and quantified. This book examines these concepts, describes a theoretical foundation of evacuation processes, and shows how emergency management and transportation professionals can apply evolving scientific and engineering knowledge to improve the practice of large scale mass evacuations.

1.1 Evacuation Fundamentals The goal of an evacuation is to avoid injuries, loss of life and, to a lesser extent, property damage and economic loss. Thus, a primary objective is to move all evacuees outside of a threat area as safely and as quickly as possible. The time it takes to clear the last person from a danger zone after the recognition of a threat is commonly referred to as clearance time, which is also referred to as an evacuation time estimate (ETE). Clearance times for mass evacuations vary widely based on the ■ characteristics of a hazard, ■ size and response of the evacuating population,

Chapter 1 · Introduction and Overview

3

■ road network through which evacuees must move, ■ adverse travel conditions such as heat, darkness, and precipitation. The characteristics of these four variables effectively dictate the clearance times of all evacuations. And, although evacuations vary widely in terms of the specific attributes and scope of these four variables, they can be scaled up or down to describe, quantify, and assess all evacuations within a spatiotemporal framework. Ultimately, these four variables are used to define the demand and supply conditions of all evacuation processes. Evacuation demand is, fundamentally, the number of people—and more specifically, the number of vehicles—that seek to use an evacuation route system (ERS)—the portion of the road network that authorities encourage people to use for their trips to safety. Evacuation demand is more precisely described as the number of vehicles per hour that attempt to depart from each origin via each path to each destination. Conversely, evacuation supply is the ability of the ERS to serve the demand placed upon it. Supply, in an evacuation context, may be described in a number of ways but, fundamentally, it is the ERS’s outflow capacity in terms of the number of vehicles per hour that can exit the risk area. More specifically, supply is a function of link capacity and network geometry. Link capacity can be defined simply in terms of the number of vehicles per hour that can move through a given section of the ERS. Consequently, local authorities typically designate the highways with the greatest capacities as the ERS. However, network geometry is also an important determinant of evacuation supply because total ERS capacity is equal to the sum of the individual link capacities only if the links are parallel to each other. For example, if an ERS consisted of two parallel evacuation links, each with a capacity of 800 vph, it would have a capacity of 1,600 vehicles per hour (vph). However, total ERS capacity will be the smaller of the individual link capacities if the links are serial. For example, if an ERS consisted of two serial evacuation links, one with a capacity of 800 vph and the other with a capacity of 400 vph, it would only have a capacity of 400 vph. Transportation networks are typically more complex than this example as a given route consists of a series of links and nodes (e.g., intersections). Multiple routes need to have no common links in order for capacity to be additive across the routes. Another important consideration in evacuation analysis is that neither demand nor supply variables remain static throughout an evacuation. Both are influenced by spatial and temporal conditions that vary during an emergency. In most emergencies, evacuation traffic demand rises over time until it reaches a peak. For example, information about changing threat conditions and phased evacuation notices

4 Chapter 1 · Introduction and Overview produce different evacuee departure times from different origins travelling to different destinations via different paths. Evacuation supply can decrease due to bottlenecks at merging highways, lanes blocked by vehicle breakdowns, and hazards such as flooding. The dynamic nature of evacuation demand and supply adds an additional layer of complexity to evacuation planning and management. In summary, clearance time is estimated as a function of evacuation demand and supply. When supply exceeds demand, vehicles can evacuate at the rate defined by the level of evacuation demand. However, when demand exceeds supply, the situation becomes more complex because queues will form that can decrease link capacities below their nominal values and, thus, increase clearance time—sometimes dramatically. Thus, the challenge for emergency managers and transportation officials is to employ demand management techniques such as phased evacuations (Zhang, Spansel, and Wolshon 2014b) and supply management techniques such as contraflow (Wolshon 2001) to balance demand and supply and, thus, reduce clearance time. These techniques are described in detail later in this book.

1.2 Evacuation Modeling Among the most significant advances in evacuation analysis and planning over the past four decades has been the development of quantitative models of evacuation processes (see Murray-Tuite and Wolshon 2013b; Lindell 2013). One contribution has been the development of mathematical models of evacuee demand and another contribution has been the development of simulation and optimization models for computing clearance times. Mathematical models of evacuation demand have taken two forms, aggregate and microscopic. The aggregate models have been used to characterize evacuation model variables such as average evacuation rates (Baker 1991), average percentage of evacuees seeking accommodations in public shelters (Mileti, Sorensen, and O’Brien 1992), and the distributions of warning reception times (Lindell and Perry 1987). However, microscopic models are increasingly being used to predict these evacuation model variables. There has been an extensive line of research on the prediction of households’ evacuation decisions with models ranging from the simple cross-tabulation of evacuation rates by hurricane category and risk area (Lindell and Prater 2007) to multi-stage, multi-equation models involving social/ environmental cues; warning source, channel, and message; previous experience, social and environmental context, psychological variables, and demographic variables (see Huang et al. 2016a for an example and Huang et al. 2016b for a review). There has also been research on models to predict other evacuation model variables such as departure

Chapter 1 · Introduction and Overview

5

time (Hasan, Mesa-Arango, and Ukkusuri 2013) and evacuation destination (Mesa-Arango, Hasan, Ukkusuri, and Murray-Tuite 2013). There has also been substantial development of simulation and optimization models that can integrate data from evacuee demand models with increasingly detailed ERS models to generate ETEs. As noted by Davidson and Nozick (2017), optimization models define a problem in terms of decision variables (controllable variables whose optimal values are to be determined), an objective function (the overall measure of performance to be minimized or maximized), and constraints (restrictions on the permissible values of the decision variables). By contrast, simulation models define a problem in terms of causal relationships among variables. Moreover, evacuation models are typically stochastic (having some element of randomness to their inputs) and dynamic (modeling the system’s evolution over time). Evacuation modeling serves numerous purposes, the most important of which is to estimate the number of people reaching safety by a given time and, conversely, to determine the time by which authorities need to issue evacuation notices in order for everyone to reach safety prior to a hazard’s arrival. In addition, these models can identify traffic congestion locations, estimate the demand for space in public shelters, test scenarios that have not occurred previously, evaluate strategies that could facilitate evacuee movement, and assess the sensitivity of ETEs to plausible variations in the input parameters. Evacuation models should take into consideration the interactions of the hazard, population, evacuation management agencies (emergency management, transportation, police, and transit agencies), and the transportation infrastructure, as displayed in Figure 1.1. This figure

Figure 1.1 General Evacuation Modeling and Planning Framework Evacuation management agencies Demand management

Supply management

ETEs/ Clearance times

Local population

Hazard event

Transportation Infrastructure

Evacuation capacity

Evacuation demand

Actual or simulated evacuation

6 Chapter 1 · Introduction and Overview not only represents the process as it unfolds in an actual evacuation, but also as it is simulated. For example, the dashed lines connecting the hazard event to evacuation management agencies and local households represent information that these community stakeholders obtain about the unfolding event. The solid lines represent impacts that the hazard can have on the ability of these community stakeholders to respond. The dashed lines between the two stakeholder groups reflect the exchange of information between them, with evacuation management agencies seeking to influence households indirectly through the news media, but also directly through agency Internet sites and social media accounts. Households provide feedback by accessing agency rumor control centers and posting on social media. Both stakeholder groups obtain information about the transportation infrastructure—the evacuation management agencies through infrastructure monitoring devices such as CCTV and the households through the news media. In turn, the local population can degrade the transportation infrastructure through traffic incidents such as lane-blocking collisions, whereas evacuation management agencies can enhance the transportation infrastructure through supply management actions such as contraflow. The solid line from the hazard event to infrastructure represents adverse impacts that reduce ERS capacity whereas the solid line from evacuation management agencies to infrastructure represents interventions that maintain or increase ERS capacity. The solid line from households to simulated or actual evacuation represents the demand model and the solid line from the transportation infrastructure to the simulated or actual evacuation represents the supply model. Finally, the dashed line from the simulated or actual evacuation to the evacuation management agencies represents the feedback to them about clearance times and other measures of effectiveness (for actual evacuations) and ETEs (for simulations). A long history of research in the social sciences has explored the relationships among the hazard, population (and their preferences and constraints), and warning messages. This social science research has informed further research into the development of spatiotemporal travel demand models that can be used with traffic simulation tools to produce ETEs that predict network clearance time. Assessing demand requires addressing the following questions: ■ How many vehicles are entering the ERS? ■ When are they entering the ERS? ■ Where are they entering the ERS (i.e., what are their origins)? ■ What are their destinations? ■ What routes are they taking from their origins to their destinations? ■ Where do they expect to stay when they get to their destinations?

Chapter 1 · Introduction and Overview

7

The question of how evacuees get to their destinations has two elements. First, the mode of transportation (e.g., personal vehicle, transit) for each household is needed because the vehicle is the fundamental unit of analysis for mass evacuation traffic models. At the very least, an aggregate percentage is needed for each mode of transportation for each origin/destination/departure time triplet. Many households using personal vehicles take more than one, influencing the overall number of evacuating vehicles. Second, the paths that the evacuees use to reach the destination from the origins are needed. Some evacuation models determine these paths by making assumptions about how drivers choose among different evacuation routes, whereas others rely on data about drivers’ expected routes obtained through surveys. Some hazards also damage or otherwise make sections of the ERS unavailable; these impacts should also be incorporated into the simulation. Furthermore, the population itself can affect the transportation system through traffic incidents such as crashes and disabled vehicles. Taking into account all of these effects, emergency management agencies may consider the ETEs to be too high, leading them to try to modify the demand (e.g., by instituting staged evacuations) or supply (e. g., adopting contraflow). When considering hazards, one needs to identify the type, severity, location, and impact timing. The detection and monitoring systems for some hazards, such as hurricanes, provide days of forewarning, and can be considered short notice events that allow preimpact evacuation; these hazards allow enough time to track the conditions, make decisions, and prepare for evacuation, as illustrated in the generalized timeline shown in Figure 1.2. However, each household operates on a different timeline and some evacuate prior to an official evacuation warning. Moreover, depending on how long households take to prepare, they might not evacuate until after impact. Other hazards, such as explosions, are not detected ahead of time; these no-notice events trigger postimpact evacuations, as illustrated in Figure 1.3. (Note that some individuals or households may evacuate

Figure 1.2 Generalized Timeline for Short Notice Events

8 Chapter 1 · Introduction and Overview

Figure 1.3 Generalized Timeline for No-Notice Events

before the authorities detect the hazard.) In these events, the amount of time to perform all of the activities is generally much smaller than for short notice events. The specific hazard type (e.g., hurricane, flood, wildfire explosion, fire, hazardous material release) suggests the appropriate protective action (evacuate or shelter in-place), whether decontamination will be needed, whether law enforcement will investigate the event as a crime scene, whether infrastructure could be damaged and where, whether injuries or casualties are present, and whether transit services can be offered, among other factors (Murray-Tuite and Wolshon 2013a). Different types of hazards present different environmental cues (if any), which increase the public’s and authorities’ awareness of the danger and increase the likelihood of evacuation (Perry 1983). Within a given type of hazard for which evacuation is the appropriate protective action, events that affect larger areas and have more severe impacts lead to larger evacuations. A greater evacuation zone and number of evacuees generally requires a greater network clearance time. In turn, to be successful, earlier evacuation notices are needed. These issues become more complex when other considerations are incorporated, such as background traffic already in or driving through an evacuation zone when an evacuation notice is issued. This additional traffic, which is not part of the recommended evacuation, can significantly increase clearance time. Another consideration is when persons not advised to evacuate do so anyway due to their perceived risk of remaining. These individuals, commonly referred to as shadow evacuees, can also increase clearance time significantly. In a well managed incident, shadow evacuees are anticipated to be 20% of the residential population within five miles of the 10 mile emergency planning zone (EPZ) of a nuclear power plant (Jones, Walton, and Wolshon 2011). However, a poorly managed incident could produce a much greater shadow evacuation, as was the case in the 1979 Three Mile Island accident. There, the Governor’s evacuation recommendation for

Chapter 1 · Introduction and Overview

9

pregnant women and preschool children within five miles of the plant (no more than 10,000 people if the target population segment left with their entire families) produced an evacuation of approximately 150,000 people (Houts et al. 1984; Lindell and Perry 1983). To help address situations in which demand is expected to exceed ERS capacity, evacuation management agencies can attempt to influence demand and modify supply. These agencies also need to consider the influence of the hazard on the infrastructure so that the evacuation management strategies do not expose the evacuees to other risks. The effect of the evacuation management strategies and the “do nothing” case (where the evacuation management agencies make no attempt to manage demand or supply) are tested through traffic simulation to produce ETEs. Comparing the evacuation time in the “do nothing” case to the hazard’s anticipated arrival time indicates how long before impact authorities must issue evacuation notices. However, if they find that the estimated amount of lead time is unacceptable—over 36 hours for many hurricaneprone cities—they need to identify evacuation management strategies and determine how much these strategies can reduce the ETEs.

1.3 Need for Multiple Disciplines Prior to the 1970s, many professionals in the emergency management and transportation fields assumed that evacuations “just happened” and that little could be done to facilitate vehicle movement during large scale, regional evacuations. This opinion, expressed by some state-level DOT officials, was based on the belief that ERS capacity was fixed and the massive demand generated by a large-scale evacuation would quickly overwhelm it. After all, as it was believed at the time, if it was not even possible to move routine rush hour traffic congestion-free, how would it ever be possible to move an entire city or region on the same network without enormous traffic problems? Another issue affecting evacuation planning was that few transportation agencies viewed evacuations as one of their responsibilities at that time. As professionals who were engaged in the safe and efficient movement of traffic during routine periods, transportation officials tended to see evacuations as an emergency management problem that they were willing to support, but not as a need that they could play a leading role in addressing. Due in large part to a series of hard-learned lessons from Hurricane Floyd (1999), the terror attacks of September 11th 2001, Hurricanes Rita and Katrina (2005), and 2007 Southern California wildfires, it became apparent that there were many simple and effective traffic management strategies that could significantly improve evacuations. Another realization was that transportation professionals could bring enormous expertise to help assess, plan, and coordinate evacuation

10

Chapter 1 · Introduction and Overview

operations. However, the most significant advancement that occurred at this time was the linkage of transportation and emergency management officials to discuss needs and policies, share resources, and develop joint operational strategies for protecting the public. Today in the United States, strategies such as the National Response Framework (www. fema.gov/media-library/assets/documents/117791), National Incident Management System (www.fema.gov/national-incident-managementsystem), and Incident Command System (www.fema.gov/incident-com mand-system-resources) have brought a standardized approach to the command, control, and coordination of emergency response. These systems provide a common organizational structure that allows responders from agencies across government jurisdictions to collaborate more effectively. In addition to these operational efforts, the National Science Foundation (www.nsf.gov), Department of Transportation (www.ops. fhwa.dot.gov/eto_tim_pse/) and the National Cooperative Highway and Transit Research Programs (www.trb.org/securityemergencies/securityan demergencies1.aspx) have supported scores of research studies, reports, and guidance documents that have developed evidence-based solutions to emergency and security related issues within transportation. These resources are significantly enhancing the development of carefully crafted, robust evacuation plans through the collaboration of emergency management and transportation agencies in ways that were virtually unimaginable just a few decades ago (Matherly et al., 2013, 2014). Nonetheless, some of the key findings from this research have not received the recognition they deserve, in part because they have not crossed disciplinary boundaries. This is partly because social scientists and transportation engineers have each tended to present their research findings at their own conferences and in their own journals. In addition, they tend to speak to other researchers rather than to practitioners and, when they do, they speak to different practitioner audiences—social scientists to emergency managers and transportation engineers to transportation officials. The research described in this book is important to practitioners because it explains how households react to different types of environmental threats and the protective actions they are advised to take in response to these threats. These considerations involve understanding people’s perception of a hazard (an interaction of the social environment with the natural environment), different sources of warnings and other information (interaction within the social environment), and alternative protective actions (an interaction of the social environment with the built environment). Practitioners also need to know how people’s responses to an environmental threat manifest themselves in evacuation demand (social environment) and how this demand interacts with the transportation system (built environment) so that accurate ETEs can be produced. The research described in this book is important to other researchers because interdisciplinary approaches are essential to advances in evacuation modeling. Social scientists and transportation engineers have

Chapter 1 · Introduction and Overview

11

been collaborating increasingly over the past few decades in interdisciplinary evacuation research efforts. Prior to these efforts, transportation engineers’ efforts typically, albeit with a few exceptions, focused on the traffic flow models or simulation efforts whereas extensive empirical social science research focused on models for warning response (Lindell and Prater 2007). These two broad fields, each of which has several subdisciplines, took quite different, and largely nonoverlapping, approaches to evacuation studies. Social scientists approached evacuation modeling as building theories or “empirical tests that explain how people make sense of and act in situations where they are told to evacuate or may want to evacuate from a hazard on their own” (Trainor et al. 2013, p. 152). Transportation based evacuation models, on the other hand, were typically developed to address a specific planning problem and efforts were “focused on collecting information and designing processes that will allow for a solution to that problem” (Trainor et al. 2013, p. 153). This book continues the efforts to combine social science and transportation engineering perspectives to provide both practitioners and researchers with an understanding of how social and transportation systems interact to produce ETEs in the face of environmental hazards.

1.4 Intended Audience and Scope The topics addressed in this book have been conceived to support the work of two primary audiences within the fields of emergency management and transportation, either of which may be practitioners or researchers. The research community primarily comprises university and government researchers but also students at the graduate and undergraduate level who seek a better understanding of the key models and strategies involved in evacuation management. The practitioner community seeks to apply research findings, as well as lessons learned in other locations and from other hazards, to protect public health and safety. This book walks the reader step by step through the key questions needed to model an evacuation and to manage evacuations through effective demand and supply strategies. The book begins with the basic questions “to what hazards does evacuation apply?” and “how do agencies make the decision to advise evacuating?” and then addresses people’s responses to evacuation advisories, including factors associated with their decisions to evacuate, activities undertaken prior to the evacuation trip (e.g., gathering family members, buying fuel), the timing of their departures, their transportation modes, and their choices of routes, destinations, and accommodations. These activities determine evacuation demand. The book then discusses methods of simulating the resulting traffic and the evacuation management strategies that can facilitate evacuee movement, especially reducing unnecessary demand.

12

Chapter 1 · Introduction and Overview

The book continues with a largely neglected issue, organized reentry into the community after authorities determine that it is safe to do so. Finally, the book concludes with four case studies that illustrate key concepts needed to develop and analyze evacuation plans. This book supports a planning process involving stakeholder engagement by providing assistance in evacuation modeling, which is a key component to the development of plans. The scope of this book is the vehicular evacuation of towns, cities, and metropolitan areas, rather than the pedestrian evacuation of buildings, transportation vehicles such as airplanes and ships, or mountainous areas (e.g., flash floods). Although most of the principles presented here apply to all hazards, the book does address some hazard-specific issues. Moreover, this book is also broader in scope than procedural guidance for preparing evacuation plans. Instead, it also presents the fundamental principles upon which evacuation plans should be based. That is, it describes what scientific research has revealed about people’s behavior during different phases of the evacuation process. It also describes the management and operation of transportation infrastructure and evacuation assets, with illustrations of evacuation planning, evaluation, and results. In summary, this book focuses on self-evacuation by personal vehicle. It introduces the reader to the steps involved in evacuation modeling by providing an understanding why hazards trigger evacuations, how authorities decide to issue evacuation advisories, how households respond to those warnings, and how traffic management strategies can make evacuations faster and safer. At each step, the emphasis is on the use of mathematical models for use in simulations that can be employed to assess the effectiveness of alternative evacuation management plans. The goal is to help evacuation planners learn from simulated, as well as actual, experience. Using computer simulations to identify flaws in evacuation plans before they are implemented on the road will ultimately save lives.

References Baker, E.J. 1991. Hurricane evacuation behavior. International Journal of Mass Emergencies and Disasters 9 (2), 287–310. Davidson, R.A., Nozick, L.K. 2017. Computer simulation and optimization. In: Rodríguez, H., Donner, W., Trainor, J. (Eds) Handbook of Disaster Research, Springer, New York, pp. 331–356. Dotson, L.J., Jones, J. 2004. Identification and Analysis of Factors Affecting Emergency Evacuations: Main Report. NUREG/CR-6864 vol. 1, SAND 20045901. Washington DC: U.S. Nuclear Regulatory Commission. Hasan, S., Mesa-Arango, R., Ukkusuri, S. 2013. A random parameter hazard based model to understand the temporal dynamics of household evacuation timing behavior. Transportation Research Part C 27, 108–116.

Chapter 1 · Introduction and Overview

13

Houts, P.S., Lindell, M.K., Hu, T.W., Cleary, P.D., Tokuhata, G., Flynn, C.B. 1984. The protective action decision model applied to evacuation during the Three Mile Island crisis. International Journal of Mass Emergencies and Disasters, 2 (1), 27–39. Huang, S.K., Lindell, M.K., Prater, C.S. 2016a. Toward a multi-stage model of hurricane evacuation decision: An empirical study of Hurricanes Katrina and Rita. Natural Hazards Review, 18 (3), 1–15. Huang, S-K., Lindell, M.K., Prater, C.S. 2016b. Who leaves and who stays? A review and statistical meta-analysis of hurricane evacuation studies. Environment and Behavior 48 (8), 991–1029. Jones, J.A., Walton, F., Smith, J.D., Wolshon, B, 2008. Assessment of Emergency Response Planning and Implementation in the Aftermath of Major Natural Disasters and Technological Accidents. NUREG/CR-6981, SAND2008-1776P, U.S. Nuclear Regulatory Commission, Washington, DC. Jones, J.A., Walton, F., Wolshon, B. 2011. Criteria for Development of Evacuation Time Estimate Studies. SAND2010-0016P, NUREG/CR-7002. US Nuclear Regulatory Commission, Washington DC. Lindell, M.K. (2013). Evacuation planning, analysis, and management. In A.B. Bariru and L. Racz (Eds). Handbook of Emergency Response: A Human Factors and Systems Engineering Approach (pp. 121–149). Boca Raton FL: CRC Press. Lindell, M.K., Perry, R.W. 1983. Nuclear power plant emergency warning: How would the public respond? Nuclear News, 26, 49–53. Lindell, M.K, Perry, R.W. 1987. Warning mechanisms in emergency response systems. International Journal of Mass Emergencies and Disasters 5 (2), 137–153. Lindell, M.K., Prater, C.S. 2007. Critical behavioral assumptions in evacuation time estimate analysis for private vehicles: examples from hurricane research and planning. Journal of Urban Planning and Development 133 (1), 18–29. Matherly, D., Langdon, N., Kuriger, A., Sahu, I., Wolshon, B., Renne, J., Thomas, R., Murray-Tuite, P., Dixit, V. 2014. A Guide to Regional Transportation Planning for Disasters, Emergencies, and Significant Events, National Cooperative Highway Research Program, Report 777. National Research Council Transportation Research Board, Washington DC. Matherly, D., Mobley, J., Wolshon, B., Renne, J., Thomas, R., Nichols, E. 2013. A Transportation Guide for All-Hazards Emergency Evacuation, Strategic Highway Research Program, Report 740. National Research Council Transportation Research Board, Washington DC. Mesa-Arango, R., Hasan, S., Ukkusuri, S., Murray-Tuite, P. 2013. A householdlevel model for hurricane evacuation destination type choice using Hurricane Ivan data. Natural Hazards Review 14 (1), 11–20. Mileti, D.S., Sorensen, J.H, O’Brien, P.W. 1992. Toward an explanation of mass care shelter use in evacuations. International Journal of Mass Emergencies and Disasters 10 (1), 25–42. Murray-Tuite, P.M., Wolshon, B. 2013a. Assumptions and processes for the development of no-notice evacuation scenarios for transportation simulation. International Journal of Mass Emergencies and Disasters 31 (1), 78–97.

14

Chapter 1 · Introduction and Overview Murray-Tuite, P.M., Wolshon, B. 2013b. Evacuation transportation modeling: an overview of research, development, and practice. Transportation Research – Part C 27, 25–45. Perry, R.W. 1978. A classification scheme for evacuation. Disasters, 2 (2–3), 169–170. Perry, R.W. 1983. Population evacuation in volcanic eruptions, floods and nuclear power plant accidents: some elementary comparisons. Journal of Community Psychology 11, 36–47. Trainor, J., Murray-Tuite, P., Edara, P. Fallah-Fini, S., Triantis, K. 2013. Interdisciplinary evacuation modeling. Natural Hazards Review 14 (3), 151–162. Wolshon, B. 2001. ‘One-way-out’: contraflow freeway operation for hurricane evacuation. Natural Hazards Review 2 (3), 105–112. Wolshon, B. 2007. Emergency transportation preparedness, management, and response in urban planning and development. Journal of Urban Planning and Development 133 (1), 1–2. Zhang, Z., Spansel, K., Wolshon, B. 2014b. Effect of phased evacuations in megaregion highway networks. Transportation Research Record 2459, 101– 109.

Chapter 2

Natural and Technological Hazards Requiring Evacuation Management

This chapter addresses a set of hazards that require evacuation management because they strike too fast for spontaneous relocation (which is possible for lava flows from effusive volcanic eruptions) but not so fast that evacuation is unsafe (as is the case for tornadoes). Section 2.1 addresses floods, Section 2.2 addresses tsunamis, Section 2.3 addresses wildfires, Section 2.4 addresses hurricanes, and Section 2.5 addresses hazardous materials releases.

2.1 Floods Flooding is a widespread problem in the United States that accounts for three quarters of all Presidential Disaster Declarations. There are seven different types of flooding that are widely recognized. Riverine (main stem) flooding occurs when surface runoff gradually rises to flood stage and overflows its banks. Flash flooding is defined by runoff reaching its peak in less than six hours. This usually occurs in hilly areas with steep slopes and sparse vegetation, but also occurs in urbanized areas with rapid runoff from impermeable surfaces such as streets, parking lots, and building roofs. Alluvial fan flooding occurs in deposits of soil and rock found at the foot of steep valley walls in arid Western regions. Ice/debris dam failures result when an accumulation of downstream material raises the water surface above the stream bank. Surface ponding/local drainage occurs when water accumulates in areas so flat that runoff cannot carry away the precipitation fast enough. Fluctuating lake levels can occur over short term, seasonal, or multiyear periods, especially in lakes that have limited outlets or are entirely landlocked. Control structure (dam or levee) failure has many characteristics in common with flash flooding. Floods are measured either by discharge or stage. Discharge, which is defined as the volume of water per unit of time, is the unit used by hydrologists. Stage, which is the height of water above a defined level, is the unit needed by emergency managers because flood stage determines the level of casualties and damage. Discharge is converted to stage by

Chapter 2 · Hazards Requiring Evacuation Management

Figure 2.1 Stage Rating Curve 70

Plain Valley

60 50 Stage (feet)

16

40 30 20 10 0 1

3

5

7

9

11

13

15

17

19

21

23

25

27

29

Discharge (thousand cubic feet/second)

From Lindell et al. 2006

means of a rating curve (see Figure 2.1). The horizontal axis shows discharge in cubic feet per second and the vertical axis shows stage in feet above flood stage. Note that high rates of discharge produce much higher stages in a valley than on a plain because the valley walls confine the water. Flooding is affected by a number of factors. The first of these, precipitation, must be considered at a given point and also across the entire watershed (basin). The total precipitation at a point is equal to its intensity of precipitation (frequently measured in inches per hour) times its duration. Total precipitation over a basin is equal to precipitation summed over all points in the surface area of the basin. The precipitation’s contribution to flooding is a function of temperature because rain (a liquid) is immediately available whereas snow (a solid) must first be melted by warm air or rain. Moreover, as indicated by Figure 2.2, the precipitation from a single storm might be deposited over two or more basins and the amount of rainfall in one basin might be quite different from that in the other basin. Consequently, there might be severe flooding in a town on one river (City A) and none at all in a town on another river (City B) even if the two towns received the same amount of rainfall from a storm. Flooding is also affected by surface runoff, which is determined by terrain and soil cover. One important aspect of terrain is its slope, with runoff increasing as slope increases. In addition to slope steepness, slope length and orientation to prevailing wind (and, thus, the accumulation of rainfall and snowfall) and sun (and, thus, the accumulation of snow) are also important determinants of flooding.

Chapter 2 · Hazards Requiring Evacuation Management

17

Slope geometry is also an important consideration. Divergent slopes (e.g., hills and ridges) provide rapid runoff dispersion. By contrast, convergent slopes (e.g., valleys) provide runoff storage in puddles, potholes, and ponds. Mixed slopes have combinations of these, so slope mean (the average slope angle) and variance (the variability of slope angles) determine the amount of storage. A slope with a zero mean and high variance (a plain with many potholes) will provide a larger amount of storage than a slope with a zero mean and low variance (a featureless plain). Similarly, a slope with a positive mean and high variance (a slope with many potholes) will provide a larger amount of storage than a slope with a positive mean and low variance. Soil cover also affects flooding because dense low plant growth slows runoff and promotes infiltration. In areas with limited vegetation, surface permeability is a major determinant of flooding. Surface permeability increases with the proportion of organic matter content because this material absorbs water like a sponge. Permeability also is affected by surface texture (particle size and shape). Clay, stone, and concrete are very impermeable because particles are small and smooth, whereas gravel and sand are very permeable—especially when the particles are large and have irregular shapes that prevent them from compacting. Finally, surface permeability is affected by soil saturation because even permeable surfaces resist infiltration when soil pores (the spaces between soil particles that ordinarily are filled with air) become filled with water. Groundwater flows via local transport to streams at the foot of hill slopes and via remote transport through aquifers. Rapid in- and outflow through valley fill increases peak flows whereas very slow in- and outflow through upland areas maintains flows between rains. Evapotranspiration takes place via two mechanisms. First, there is direct evaporation to the atmosphere from surface storage in rivers and lakes. Second, there is uptake from soil and subsequent transpiration by plants. Transpiration draws moisture from the soil into plants’ roots, up through the stem, and out through the leaves’ pores (similar to people sweating). The latter mechanism is generally much higher in summer than in winter due to increased heat and plant growth, but transpiration is negligible during periods of high precipitation. Stream channel flow is affected by channel wetting which infiltrates the stream banks (horizontally) until they are saturated as the water rises. In addition, there is seepage because porous channel bottoms allow water to infiltrate (vertically) into groundwater. Channel geometry also influences flow because a greater channel cross-section distributes the water over a greater area, as does the length of a reach (distinct section of river) because longer reaches provide greater water storage. High levels of discharge to downstream reaches can also affect flooding on upstream reaches because flooded downstream reaches slow flood transit by decreasing the river’s elevation drop. Flooding increases when upstream areas experience deforestation and overgrazing, which increase surface runoff to a moderate degree on

18

Chapter 2 · Hazards Requiring Evacuation Management

Figure 2.2 Map of the Distribution of Precipitation From a Storm

Basin A Basin B

Storm City A City B

From Lindell et al. 2006

shallow slopes and to a major degree on steep slopes as the soil erodes. The sediment is washed downstream where it can silt the channel and raise the elevation of the river bottom. These problems of agricultural development are aggravated by flood plain urbanization. Cities throughout the world have been located in flood plains because water was the most efficient means of transportation until the mid-1800s. Consequently, many cities were located at the head of navigation or at transshipment points between rivers. In addition, cities have been located in flood plains because level alluvial soil is very easy to excavate for building foundations. Finally, urban development takes place in flood plains because of the aesthetic attraction of water. People enjoy seeing lakes and rivers, and pay a premium for real estate that is located there. One consequence of urban development for flooding is that cities involve the replacement of vegetation with hardscape—impermeable surfaces such as building roofs, streets, and parking lots. This hardscape decreases soil infiltration, thus increasing the speed at which flood crests rise and fall. Another factor increasing flooding is intrusion into the flood plain by developers who fill intermittently flooded areas with soil to raise the elevation of the land. This decreases the channel cross-section, forcing the river to rise in other areas to compensate for the lost space.

2.1.1 Flood Hazard Analysis Flood risk areas in the US are generally defined by the 100-year flood— an event that scientists estimate to have a 1% chance of occurrence in

Chapter 2 · Hazards Requiring Evacuation Management

19

any given year and, thus, a 100-year recurrence interval. It is important to understand that these extreme events are essentially independent, so it is possible for a community to experience two 100-year floods in the same century. Indeed, it is possible to have them in the same year even though that would be a very improbable event. This statistical principle is misunderstood by many people who believe there can be only one 100-year flood per century (Bell and Tobin 2007). The belief that a 100-year flood occurring this year cannot be repeated for another 100 years (or at least nearly 100 years) is a very dangerous fallacy. Moreover, a 100-year flood is an arbitrary standard of safety that reflects a compromise between the goals of providing long term safety and developing economically valuable land. A 50-, 200-, or even a 500year standard could be used instead. Indeed, the Dutch use a 4,000-year standard for coastal protection (Terpstra and Lindell 2013). Community adoption of a 50-year flood standard would provide more area for residential, commercial, and industrial development. However, the resulting encroachment into the flood plain would lead to more frequent damaging floods than would a 100-year flood standard. Alternatively, a community might use different standards for different types of structures. For example, it might restrict the 100-year floodplain to low intensity uses (e.g., parks), allow residential housing to be constructed within the 500-year floodplain, and restrict nursing homes, hospitals, and schools to areas outside the 500-year floodplain. The Federal Emergency Management Agency has a flood hazard mapping program that identifies flood hazards and assesses flood risks. Flood Insurance Rate Maps (FIRMs) are based on data for river flow, storm tides, hydrologic/hydraulic analyses, rainfall, and topographic surveys (FEMA 2016). These maps can be used to identify flood risk areas that may need to be evacuated. Figure 2.3 shows the FIRM for Mt. Vernon Washington. Zones labeled with the letter A have the highest flood hazard and are located within the 100-year flood plain, zones labeled with the letter B have moderate flood hazard and are located outside the 100-year flood plain but inside the 500-year flood plain, and zones labeled with the letter C have minimal flood hazard and are located outside the 500-year flood plain.

2.1.2 Flood Monitoring and Forecasting Flood monitoring takes place at both regional and local levels using radar for assessing rainfall amounts at variable points in a watershed, rain gauges for detecting rainfall amounts at predetermined points in a watershed, and stream gauges for detecting water depth at predetermined points along a river. Detection also can be achieved by manual devices such as spotters for assessing rainfall amounts, water depth, or levee integrity at specific locations (planned before a flood or

20

Chapter 2 · Hazards Requiring Evacuation Management

improvised during one). Once data on the quantity and distribution of precipitation have been collected, they are used to estimate discharge volumes over time from the runoff characteristics of a given watershed (e.g., soil permeability and surface steepness) at a given time (e.g., current soil saturation). Once discharge volume is estimated, it can be used together with downstream topography (e.g., mountain valley vs. plain) to predict downstream flood heights. A flood’s speed of onset, and thus the amount of forewarning, varies among flood types; flash floods and control structure failures have much more rapid onset than main stem floods. Consequently, areas exposed to flash floods and control structure failures are especially likely to have local stream gauges to detect hazard onset. Unlike some other environmental hazards, it is possible to provide quite accurate forecasts of a flood’s location (the section of river that will flood), timing of impact, the magnitude of the flooding (the height of the flood crest above flood stage), and the flood zones in which flooding can be expected. Timely and specific warnings of floods are often provided by commercial news media as well as NOAA Weather Radio. The NWS issues a flood advisory to make people aware of weather conditions that could cause some flooding that is not expected to be severe enough to threaten lives and property, whereas a flood watch is issued to prompt people to prepare for the possibility of flooding. The NWS issues a flood warning

Figure 2.3 FEMA Flood Insurance Rate Map

Chapter 2 · Hazards Requiring Evacuation Management

21

or a flash flood warning to indicate that people should take immediate action for flooding that is imminent or in progress. The most appropriate protective action for persons is to evacuate in a direction perpendicular to the river channel. Because flash floods in mountain canyons can travel faster than a motor vehicle, it is safest to climb the canyon wall rather than try to drive out. It also is important to avoid crossing running water. Just two feet of fast moving water can float a car and push it downstream with 1,000 pounds of force.

2.2 Tsunamis Tsunamis are sometimes mistakenly called “tidal” waves but they are, in fact, sea waves that are usually generated by earthquakes. In addition, tsunamis can be caused by volcanic eruptions or landslides that usually, but not always, occur undersea. Tsunamis are rare events because 15,000 earthquakes over the course of a century have generated only 124 tsunamis, a rate of less than 1% of all earthquakes and only 0.7 tsunamis per year. This low rate of tsunami generation is attributable to earthquake intensity; two thirds of all Pacific tsunamis are generated by shallow earthquakes exceeding 7.5 in magnitude. It is important to note that the initial phase of a tsunami might be a drop in the water level, rather than a rise. An initial trough is created if the seafloor suddenly drops, whereas an initial wave is created if the seafloor suddenly rises. In either case, the initial phase will be followed by the alternate phase (i.e., a trough is followed by a wave or vice-versa). Tsunamis can travel across thousands of miles of open ocean (e.g., from the Aleutians to Hawaii or from Chile to Japan) at speeds up to 400 mph in the open ocean, but they slow to 25 mph as they begin to break in shallow water and run up onto the land. Tsunamis are largely invisible in the open ocean because they are only 1–2 feet high. However, they have wave lengths up to 60 miles and periods as great as one hour. This contrasts significantly with ordinary ocean waves having wave heights up to 30 feet, wave lengths of about 500 feet, and periods of about 10 seconds. Tsunamis can have devastating effects in some of the places where they make landfall because the waves encounter bottom friction when the water depth is less than 1/20 of their wavelength. At this point, the bottom of the wave front slows and is overtaken by the rest of the wave, which must rise over it. For example, when a wave reaches a depth of 330 feet, its speed is reduced from 400 mph to 60 mph. Later, reaching a depth of 154 feet reduces its speed to 44 mph. This causes the next 650 feet of the wave to overtake the wave front in a single second. As the wave continues shoreward, each succeeding segment of the wave must rise above the previous segment because it cannot go down (water is not compressible) or back (the rest

22

Chapter 2 · Hazards Requiring Evacuation Management

of the wave is pressing it forward). Because the wavelength is so long and wave speed is so fast, a large volume of water can pile up to a very great height—especially where the continental shelf is very narrow.

2.2.1 Tsunami Hazard Analysis Tsunamis threaten shorelines worldwide and have no known temporal (i.e., diurnal or seasonal) variation. The physical magnitude of a tsunami is extremely impressive. Wave crests can arrive at 10–45 minute intervals for up to six hours and the highest wave, as much as 100 ft at the shoreline, can be anywhere in the wave train. The area flooded by a tsunami is known as the inundation zone, which is equivalent to a 100year floodplain. Because of the complexities in accounting for wave behavior and the characteristics of the offshore bathymetry and onshore topography, tsunami inundation zones must be calculated by analysts using sophisticated computer programs. Tsunamis cause deaths from drowning and traumatic injuries from wave impact, as well as property damage.

2.2.2 Tsunami Monitoring and Forecasting Tsunami warning centers (TWCs) base their detection of remote tsunamis on seismic monitoring to detect major earthquakes, followed by pressure sensors and tidal gauges located throughout the Pacific basin to verify tsunami generation. Once tsunami generation has been confirmed, alerts can be transmitted throughout the Pacific basin (Darienzo et al. 2005). If a detected event exceeds a predetermined threshold, TWCs notify emergency managers who, in turn, disseminate warnings through National Oceanographic and Atmospheric Administration (NOAA) Weather Radio, sirens, radio and television, route alert, door to door, and telephone to households, businesses, and special facilities at risk (see Figure 2.4). In many cases, these warnings can forecast tsunami arrival times and magnitudes many hours in advance. In such cases, there will be enough time for vehicular evacuation out of the inundation zone. Evacuation to a safe distance out of the inundation zone is obviously difficult on low-lying coasts because it would take many minutes to reach safety, even when traveling in a vehicle. Moreover, vehicular evacuation can even be problematic where there are nearby hills if the primary evacuation route runs parallel to the coastline. Even more troubling are instances, such as the 2004 Indian Ocean tsunami, in which warning systems were unavailable or failed. In such cases, coastal residents’ only cue to a remotely initiated tsunami was wave arrival at the coast or, more confusingly, the arrival of a trough

Chapter 2 · Hazards Requiring Evacuation Management

23

(making it appear that the tide went out unexpectedly). In such cases, the only available protective action is to seek shelter in a high-rise building or on a nearby hill. If a tsunami is initiated locally (i.e., within a hundred miles), coastal residents must infer the potential for a tsunami from severe earthquake shaking. In such cases, people should begin protective action as soon as the shaking stops. As is the case with warning system failure, the most effective protective action is to seek shelter in a high-rise building or evacuate on foot to a nearby hill. However, shelter in-place is most likely to be successful in steel reinforced concrete structures with deep pilings that can withstand wave battering and foundation scour. A TWC will issue an information statement to notify people that an earthquake has occurred but there is no threat of a destructive tsunami and a tsunami watch to indicate that a distant earthquake has occurred and that a tsunami is possible. The TWC will issue a tsunami advisory if there is a tsunami threat only to people in or very near the water and a tsunami warning if there is a tsunami threat to people farther from the coastline.

Figure 2.4 Tsunami Emergency Notification/Warning System Tsunami warning center

NWS offices

NOAA Weather Radio State EOC Sirens

Local Dispatch/ EOC

Television and Radio

Households

Route alert

Businesses

Door-to-door

Special facilities

Telephone

Adapted From Lindell and Prater 2010

24

Chapter 2 · Hazards Requiring Evacuation Management

2.3 Wildfires All fires require the three elements of the fire triangle: fuel, which is any substance that will burn; oxygen that will combine with the fuel; and enough heat to ignite fuel (and sustain combustion if an external source is absent). The resulting combustion yields heat (sustaining the reaction) and combustion products such as toxic gases and unburned particles of fuel that are visible as smoke. Wildfires are distinguished mostly by their fuel. Wildland fires burn areas with nothing but natural vegetation for fuel, whereas interface fires burn into areas containing a mixture of natural vegetation and built structures. Firestorms are distinguished from other wildfires because they burn so intensely that they create their own local weather and are virtually impossible to extinguish. Wildfires can occur almost anywhere in the United States but are most common in the arid West where there are extensive stands of conifer trees and brush that serve as ready fuels. Once a fire starts, the three principal variables determining its severity are fuel, weather, and topography. Fuels differ in a number of characteristics that collectively define fuel type. These include the fuel’s ignition temperature (low is more dangerous), amount of moisture (dry is more dangerous), and the amount of energy (resinous wood is more dangerous). A given geographical area can be defined by its fuel loading, which is the quantity of vegetation in tons per acre, and fuel continuity, which refers to the proximity of individual elements of fuel. Horizontal proximity can be defined, for example, in terms of the distance between trees. Vertical proximity can be defined in terms of the distance between different levels of vegetation (e.g., grasses, brush, and tree branches). Weather affects fire behavior by wind speed and direction as well as temperature and humidity. Wind speed and direction have the most obvious effects on fire behavior, with strong wind pushing the fire front forward and carrying burning embers far in advance of the main front. High temperature and low humidity promote fires by decreasing fuel moisture, but these can vary during the day (cooling and humidifying at sunset) as well as over longer periods of time. Topography affects fire behavior by directing prevailing wind currents and the hot air produced by the fire. Canyons can accelerate the wind by funneling it through narrow openings. Steep slopes (greater than 10°) take advantage of a fuel’s location in the fire’s heated updraft, which allows the advancing fire front to dry nearby fuels through radiant heating and also provide a ready path for igniting these fuels. A fire’s forward movement speed doubles on a 10° slope and quadruples on a 20° slope.

2.3.1 Hazard Analysis Wildland fires are a major problem in the US because an average of about 73,000 such fires per year burn over 3 million acres. Approximately 13%

Chapter 2 · Hazards Requiring Evacuation Management

25

of these wildfires are caused by lightning, but people cause 24% of them accidentally and 26% of them deliberately. The greatest loss of life from a US wildfire occurred in the 1871 Peshtigo, Wisconsin wildfire that killed 2,200 people (Gess and Lutz 2002). More recently, the 1991 Oakland Hills California wildfire killed 25 people, injured 150, and damaged or destroyed over 3,000 homes. Major contributors to the severity of this wildland urban interface fire were the housing construction materials (predominantly wood siding and wood shingle roofs), vegetation planted immediately adjacent to the houses, and narrow winding roads that impeded access by fire-fighting equipment.

2.3.2 Wildfire Monitoring and Forecasting The US Forest Service maintains a Fire Danger Rating System that monitors changing weather and fuel conditions (e.g., fuel moisture content) throughout the summer fire season. Some of the fuel data are derived from satellite observations and the weather data come from hundreds of weather stations. A wildfire’s speed of onset, and thus the amount of forewarning, varies with wind speed and direction. Thus, fire spread takes an elliptical shape with different rates of progression for the head fire, flank fire, and back fire (Van Wagner 1969). A number of different types of fire spread models have been proposed over the years and recent ones have begun to incorporate atmospheric prediction models (see Li, Cova and Dennison 2015). Appropriate protective actions include evacuation out of the risk area, evacuating to a safe location (e.g., an open space such as a park or baseball field having well-watered grass that will not burn), and sheltering in-place within a fire-resistant structure (e.g., a concrete building with no nearby vegetation). The NWS issues a fire weather watch to indicate that people should be prepared for hazardous weather conditions within the next 12–48 hours whereas it issues a red flag warning to indicate that hazardous weather conditions are expected within the next 24 hours or are in progress. An extreme fire behavior advisory indicates that one or more fires are out of control and behaving unpredictably because of factors such as a high rate of spread.

2.4 Hurricanes A hurricane is an intense tropical cyclone, which is a low pressure system that develops a rotating surface wind circulation over tropical waters. The earliest stages of hurricane development are marked by thunderstorms that intensify through earlier stages—Tropical Depression (less than 39 mph) and Tropical Storm (39–74 mph)—that result in a

26

Chapter 2 · Hazards Requiring Evacuation Management

sustained surface wind speed exceeding 74 mph. At this point, the storm becomes a hurricane that can intensify to any one of the five SaffirSimpson categories. Tropical cyclones draw their energy from warm seawater, so they form only when there is an increase in sea surface temperature that exceeds 80°F. This warm water evaporates at a high rate and rises into the upper atmosphere, which lowers the atmospheric pressure at the sea surface and causes wind to flow in to replace the rising air. When the warm moist air reaches a high altitude, it cools and releases its latent heat of evaporation, causing thunderstorms. As the earth rotates, it imparts a corresponding rotation to the tropical system’s wind—counterclockwise in the Northern Hemisphere. An easterly steering wind (which is named for its direction of origin, so an easterly wind blows from east to west) pushes these storms westward across the Atlantic. Storm intensity weakens as it reaches the North Atlantic (because it derives less energy from the cooler water at high latitudes) or makes landfall (which cuts the storm off from its source of energy and adds the friction of interaction with the rough land surface). The nature of atmospheric processes is such that few of the minor storms escalate to a major hurricane. In the average year there are 100 tropical disturbances, 10 tropical storms, six hurricanes, and only two of these hurricanes strike the US coast. Hurricanes in Categories 3–5 account for 20% of landfalls, but over 80% of damage. Category 5 hurricanes are rare in the Atlantic (three during the 20th Century), but are more common in the Pacific. Most hurricanes striking the US coast originate in tropical water off the West African coast, but others are born in the Caribbean and some even begin in the Gulf of Mexico. The hurricane season usually begins in June, reaches its peak in September, and then decreases through the end of November, but extended into December during the 2005 hurricane season. The main components of a hurricane are the eye, the eye wall, and the rain bands that spiral inward toward the eye wall. The hurricane eye is a relatively calm, clear area usually 20–40 miles across. It is surrounded by a dense wall of thunderstorms—the eye wall—that has the strongest winds within the storm. The storm’s outer rain bands range in width from a few miles to tens of miles and are 50–300 miles long. These rain bands have high wind speeds and can extend out hundreds of miles from the hurricane eye. The entire hurricane, which can be as much as 600 miles in diameter, rotates counterclockwise in the Northern Hemisphere. This produces a storm surge that is located in the right front quadrant relative to the storm track. The destructive force of hurricanes comes from high wind, tornadoes, coastal flooding from storm surge, and inland flooding from heavy rain. High wind is probably the most obvious hurricane threat, and wind speed defines a hurricane’s classification category into one of the five categories of the Saffir-Simpson scale (see Table 2.1). The strength of the wind can be seen in the third column, which shows that the pressure

Chapter 2 · Hazards Requiring Evacuation Management

27

Table 2.1 Saffir-Simpson Hurricane Scale Saffir/Simpson Category

Wind speed (mph)

Velocity Pressure (psf)

Expected Damage

One

74–95

19.0

• Vegetation: some damage to foliage. • Street signs: minimal damage. • Mobile homes: some damage to unanchored structures. • Other buildings: little or no damage.

Two

96–110

30.6

• Vegetation: much damage to foliage; some trees blown down. • Street signs: extensive damage to poorly constructed signs. • Mobile homes: major damage to unanchored structures. • Other buildings: some damage to roof materials, doors, and windows.

Three

111–130

41.0

• Vegetation: major damage to foliage; large trees blown down. • Street signs: almost all poorly constructed signs blown away. • Mobile homes: destroyed. • Other buildings: some structural damage to small buildings.

Four

131–155

57.2

• Vegetation: major damage to foliage; large trees blown down. • Street signs: all down. • Mobile homes: destroyed. • Other buildings: extensive damage to roof materials, doors, and windows; many residential roof failures.

Five

>155

81.3

• Vegetation: major damage to foliage; large trees blown down. • Street signs: all down. • Mobile homes: destroyed. • Other buildings: some complete building failures.

From Lindell et al. 2006

of the wind on vegetation and structures is proportional to the square of the wind speed. That is, as the wind speed doubles from 80 mph in a Category 1 hurricane to 160 mph in a Category 5 hurricane, the velocity pressure quadruples from less than 20 pounds per square foot (psf) to over 80 psf. Damage from high wind (and the debris that is entrained in the wind field) is a function of a structure’s exposure. Wind exposure is

28

Chapter 2 · Hazards Requiring Evacuation Management

highest in areas directly downwind from open water or fields. Upwind hills, woodlands, and tall buildings decrease exposure to the direct force of the wind but increase exposure to flying debris such as tree branches and building materials that have been torn from their sources. In general, wind effects depend upon an object’s size, shape, material strength, and anchoring but even Tropical Storm force (39 mph) wind can overturn high profile vehicles, such as recreational vehicles and buses. Thus, emergency managers prefer to complete evacuations before the arrival of Tropical Storm wind so they can avoid having such vehicles block evacuation routes. The wind speed experienced at a given location depends upon the intensity of the hurricane and that location’s distance and direction from the hurricane eye. For a large storm, hurricane-force wind extends out 150 miles from the eye and Tropical Storm force wind extends out 300 miles. However, hurricanes do not have symmetric wind patterns. Wind speed tends to be higher in a hurricane’s right forward quadrant (the location to the right of the track and forward of the hurricane eye as one looks from the sea toward the land) because the forward movement speed of the hurricane is added to the speed of the wind rotating around the eye. Conversely, wind speed tends to be lower in the left forward quadrant because the forward movement speed of the hurricane is subtracted from the speed of the wind rotating around the eye. Wind gusts can exceed the maximum sustained wind speed by 25% or more. Moreover, wind speed is lower the farther away along the coastline a location is from the point at which the eye makes landfall. This is a very important point to bear in mind because few people who have “survived” a Category 4 hurricane have actually borne the impact of wind exceeding 130 mph. This is a reason why many people overestimate their homes’ ability to withstand the impact of a major storm. Finally, wind speed decreases as a hurricane moves inland because the storm is cut off from its source of energy (warm water) and because of surface friction from rough terrain. There is an almost immediate 10% decrease in wind speed at landfall and a 50% decrease within the first 10 hours. Hurricanes can produce tornadoes, usually in the rain bands that spiral out from the eye. These tornadoes generally occur some distance from the center of the storm. Tornadoes are much more compact and, therefore affect a much smaller area, than the hurricane that spawned them. However, tornado wind speed can be even greater—up to 400 mph or more. There were 23 tornadoes associated with Hurricane Alicia that struck Galveston in 1983. However, most of the tornadoes were weak, with wind speeds between 40–72 mph. The strongest tornado, which struck near Tyler, had a wind speed of 113–157 mph. A storm surge is a large dome of water, often 50–100 miles wide, that in the Northern Hemisphere sweeps across the coastline to the right (looking from the sea toward the land) of the hurricane’s point of landfall. Storm surge is most commonly associated with hurricanes, but

Chapter 2 · Hazards Requiring Evacuation Management

29

also can be caused by extratropical cyclones (nor’easters). The height of a storm surge increases as atmospheric pressure decreases and a storm’s maximum wind speed increases. Thus, the height of the storm surge generally increases with hurricane category but it is possible to have a very large storm surge in a weak hurricane. For example, Hurricane Ike was only a Category 2 storm in terms of wind speed but it had a storm surge that was more typical of a Category 4 hurricane. Moreover, storm surge is only one factor affecting the depth of the water over the land. A normal high tide must be added to that of the storm surge (or a low tide subtracted from it), although this usually only accounts for two feet or less along the Texas coast. Storm surge is especially significant where coastal topography and bathymetry (submarine topography) have shallow slopes and the coast has a narrowing shoreline that funnels the rising water. These factors are magnified when the storm remains stationary through several tide cycles and the affected coast is defined by low-lying barrier islands whose beaches and dunes have been eroded either by human development or by recent storms. Storm surge—together with astronomical high tide, rainfall, river flow, and storm surf—floods and batters structures and scours areas beneath foundations as much as 4-6 feet below the normal grade level. More significant is the effect of wind-generated waves, which can be more than 50% higher than the still-water surge depth (the surge depth as measured by the “average” between the wave peak and trough). Breaking waves crash into buildings with tremendous force—smashing windows and doors, collapsing walls, and sweeping even the strongest swimmers to their deaths. These waves reach their greatest height on the open coast but are generally much smaller in protected bays and bayous. At one time, storm surge was the primary source of casualties in all countries, but inland flooding is now the primary cause of hurricane deaths in the US. However, surge is still the primary source of casualties in developing countries such as Bangladesh. In these countries, population pressure pushes the poor to farm highly vulnerable areas and poverty limits the development of dikes and seawalls, warning systems, evacuation transportation systems, and vertical shelters (wind resistant structures that are elevated above flood level). Hurricanes can also generate widespread torrential rainfall that results in deadly and destructive floods. These floods can threaten areas well inland from the effects of hurricane wind and storm surge. Rainfall rates during hurricanes can range up to four inches/hour for short periods of time; Hurricane Harvey produced 60 inches of rain over four days (Blake and Zelinsky 2018). Because slow moving hurricanes take many more hours to pass through an area, they generally deposit greater amounts of rainfall—some hurricane impact areas have experienced up to 30 inches of rainfall over a period of several days. Such downpours cause severe local ponding (water that fell and did not move) and inland flooding (water that fell elsewhere and flowed in).

30

Chapter 2 · Hazards Requiring Evacuation Management

2.4.1 Hurricane Hazard Analysis The National Hurricane Center uses the Sea, Lake, and Overland Surges from Hurricanes (SLOSH) model to generate estimates of surge heights, although other approaches have been developed by academic researchers. SLOSH is a physics-based model developed that can be applied to historical, hypothetical, or predicted hurricanes (National Hurricane Center 2014). To generate surge hazard maps for evacuation planning, analysts typically use hundreds to thousands of hypothetical hurricanes of varying directions, speeds, intensities (Saffir-Simpson categories), and other parameters. These hypothetical hurricanes are run through the SLOSH model to generate Maximum Envelopes of Water (MEOWs), which indicate the maximum water surface elevation for each of the SLOSH model grid cells for all of the hurricane tracks for a given direction, forward speed, and intensity (USACE 2002). Then, MOMs (maximum of maximums), representing the “maximum water surface elevation for each grid cell regardless of approach direction, forward speed, or track” are created for each hurricane category (USACE 2002, p. 18). The MOMs may be adjusted for tidal anomalies and high tides and then used to produce hazard maps. Hurricane hazard maps (see for example Figure 2.5) are used to identify evacuation zones based on factors such as the risk of surge flooding, whether a non-flood area would be isolated by flooding in surrounding areas, jurisdictional boundaries (e.g., state and county boundaries), and the ERS serving that area. Although the Saffir-Simpson scale consists of five categories, evacuation zones may be combined into a smaller number, such as three. On the Mississippi coast, Zone A includes areas subject to flooding in Category 1 and 2 hurricanes, Zone B includes areas flooded in a Category 3 hurricane, and Zone C includes areas flooded in Category 4 and 5 hurricanes (USACE 2002). For higher category storms, residents in lower zones are also evacuated, as are mobile home residents, the latter is due to their vulnerability to wind. The SLOSH model generally is accurate within ± 20%. For example, if the model predicts a 10 foot storm surge, the observed peak will be between 8 and 12 feet. The surge model does not include breaking waves or inland flooding caused by rainfall. Analysts use the data on surge depth and wind speed to define the boundaries of Risk Areas. In some Texas coastal counties, for example, each Risk Area comprises the portion of a county that is expected to be affected by the corresponding hurricane category (Figure 2.5). That is, Risk Area 1 is the area expected to be affected by hurricane Category 1 (74–95 miles per hour winds), Risk Area 2 is the area expected to be affected by hurricane Category 2 (96–110 miles per hour winds), and so on. Risk Area boundaries generally run parallel to the coast, but can be very irregular where there are rivers or bays. Risk Area boundaries are far apart when the terrain is flat and close together when the terrain is steeply sloped.

Figure 2.5 State of Texas Hurricane Risk Areas

Chapter 2 · Hazards Requiring Evacuation Management

31

32

Chapter 2 · Hazards Requiring Evacuation Management

2.4.2 Hurricane Monitoring and Forecasting The National Hurricane Center monitors the progress of hurricanes and provides critical information about approaching storms. Many emergency managers use the HURREVAC software tool to display potential wind effects along the coast as well as inland. By integrating live forecasts and data, communities can use HURREVAC to determine when damaging winds will reach them (USACE 2002). HURREVAC updates its displays after each NHC forecast/advisory, which provides current data on the hurricane eye’s latitude/longitude coordinates and the storm’s intensity. The forecast/advisory also contains forecasts of the eye location, hurricane intensity, and hurricane size for 12, 24, 36, 48, and 72 hours from the current time. The NHC issues a new forecast/advisory every six hours. The first one is normally issued when meteorological data indicate a cyclone has formed and subsequent advisories are issued at 4am, 10am, 4pm, and 10pm Central Daylight Time. A hurricane’s location is defined by the position of its eye in degrees (and tenths of a degree) of latitude or longitude. One degree of latitude equals 60 nautical miles (each nautical mile equals 1.15 statute miles), but degrees of longitude vary in length because longitudinal meridians are most widely spaced at the equator and converge at the poles. Sometimes the rain bands, and even the eye itself, are obscured by higher-level clouds, making it difficult to locate the storm’s position precisely by satellites. In such cases, the eye can be located only within about 18–30 miles, whereas a well-defined eye can be located within 6–12 miles. As a hurricane’s wind moves around the eye in a circular motion, the eye itself moves over the surface of the earth. The location of the hurricane eye over time, the hurricane track, is important to local authorities because it determines whether or not the hurricane will landfall in their jurisdiction. Most Atlantic hurricanes begin their lives by tracking due west across the Atlantic Ocean but gradually curve toward the north. However, hurricanes have followed many other tracks. ■ Some of them continue on a straight track until they make landfall. ■ Others have made sharp changes in direction, ■ Occasionally, hurricanes even make circular loops. Figure 2.6 shows a hurricane tracking map in which the positions of the hurricane eye, as reported by the NHC in forecast advisories, are represented by a succession of dots moving from right to left. The dot on the farthest right side of the track is the location at which the storm became classified as a hurricane. The dot on the farthest left side of the track is the hurricane’s current location. It is even more difficult to forecast a storm’s future location than to identify its current location because there is uncertainty about how the

Chapter 2 · Hazards Requiring Evacuation Management

33

Figure 2.6 Hurricane Tracking Map

Radius of Tropical Storm wind

Hurricane track

Current hurricane eye location

hurricane track will be affected by other weather systems. The NHC provides estimates of its track forecast accuracy in such a way that the forecast track has a two-thirds chance of falling within it. In general, the longer the time interval until landfall, the greater is the uncertainty about the hurricane track (see Table 2.2). In turn, this uncertainty about the hurricane’s track makes it difficult for local authorities to decide whether or not to evacuate their jurisdictions. The probable track forecast error at 36 hours (63 nautical miles on either side of the forecast track) is especially important because this is the ETE for many urbanized areas. It is important to recognize that this is not the maximum error. Consequently, the error could be larger than 63 nautical miles, even though these larger errors are unlikely. A hurricane’s forward movement speed is defined by the rate at which the hurricane eye moves along its track. This forward movement speed is important to local authorities because it determines how soon the hurricane will make landfall in their jurisdiction. The faster a hurricane’s forward movement speed is, the sooner it will arrive and, therefore, the sooner an evacuation must be initiated. A storm’s forward movement speed averages about 10 mph, but can range from 0–30 mph and can vary over time. However, hurricanes can speed up, maintain a constant speed, slow down, or stall (stop all forward motion). In general, the longer the time interval until landfall, the greater is the uncertainty about the hurricane’s forward movement

34

Chapter 2 · Hazards Requiring Evacuation Management

Table 2.2 Track Forecast Accuracy (2012–2016) Forecast period (hours)

Probable Track Forecast Error (nautical miles)

Average Wind Speed Error (knots)

72

107

14

48

78

13

36

63

10

24

45

8

speed. Consequently, uncertainty about hurricane forward movement speed makes it difficult to decide how soon to begin evacuations. A hurricane’s intensity (defined by its Saffir-Simpson category) can increase, remain the same, or decrease over time. Hurricane intensity is important to local authorities because it determines how far inland (i.e., how many risk areas) authorities must evacuate their residents. In general, the longer the time interval until landfall, the greater is the uncertainty about hurricane intensity. Consequently, uncertainty about hurricane intensity makes it difficult to decide how far inland to evacuate. As is the case with forecasts of hurricane location, forecasts of wind speed decrease in accuracy as the length of the forecast period increases. The average forecast error at 36 hours—a time when some jurisdictions must make a final decision whether or not to evacuate—is approximately 10 knots. A difference of this size can be large enough to change a hurricane by one category on the Saffir-Simpson scale. This is the average error, not the maximum error. Consequently, the error could be larger than 15 mph, even though very large errors are unlikely. To provide local authorities and coastal residents with meteorologists’ assessment of a hurricane threat, the NHC issues forecasts for where a hurricane will be, how strong it will be, and how large it will be, up to five days in advance of the time of a given forecast advisory, along with a number of probability based products to help account for uncertainty in the forecast. However, it does not issue a hurricane watch until Tropical Storm force wind conditions are forecast to arrive at the coast within 48 hours and does not issue a hurricane warning until Tropical Storm force wind conditions are forecast to arrive at the coast within 36 hours.

2.5 Hazardous Materials Releases Toxic industrial chemical releases are of special concern to emergency managers because the airborne dispersion of these chemicals can produce lethal inhalation exposures at distances as great as 10 miles and

Chapter 2 · Hazards Requiring Evacuation Management

35

sometimes even more. The spread of a toxic chemical release can be defined by a dispersion model that includes the hazardous material’s (hazmat’s) chemical and physical characteristics, its release characteristics, the topographic conditions in the release area, and the meteorological conditions at the time of the release. The chemical and physical characteristics of the hazmat include its quantity (measured by the total weight of the hazmat released), volatility (higher volatility means more chemical becomes airborne per unit of time), buoyancy (whether it tends to flow into low spots because it is heavier than air), and toxicity (the biological effect due to cumulative dose or peak concentration). It also includes the chemical’s physical state—whether it is a solid, liquid (a substance above its boiling point is a vapor), or a gas at ambient temperature and pressure. In general, vapors and gases are major hazards because they are readily inhaled and this is the most rapid path into the body. Release characteristics are defined by the chemical’s temperature and pressure in relation to ambient conditions, its release rate (in pounds per minute), and the size (surface area) of the spilled pool if the substance is a liquid. Temperature and pressure are important because the rate at which the chemical disperses in the atmosphere increases when these parameters exceed ambient conditions. The release rate is important because it determines the concentration of the chemical in the atmosphere. Specifically, a higher release rate puts a larger volume of chemical into a given volume of air, thus increasing its concentration (where the latter is defined as the volume of chemical divided by the volume of air in which it is located). Topographical conditions relevant to liquid spills include the slope of the ground and the presence of depressions. As is the case with flooding, steep slopes allow a liquid to rapidly move away from the location of the spill. Both flat slopes and depressions decrease the size of a liquid pool which, in turn, affects the size of the pool’s surface area and reduces the rate at which vapor is generated from it. Thus, dikes are erected around chemical tanks to confine spills in case the tanks leak and hazmat responders build temporary dikes around spills for the same reason. Topographical characteristics also affect the dispersion of a chemical release in the atmosphere. Hills and valleys are land features that channel the wind direction and can increase wind speed at constriction points—for example, where a valley narrows and causes wind speed to increase due to a “funnel” effect. Forests and buildings are rough surfaces that increase turbulence in the wind field, causing greater vertical mixing that dilutes the chemical’s concentration in the atmosphere. By contrast, large water bodies have very smooth surfaces that do not constrain wind direction and, because they provide no wind turbulence, allow a chemical release to maintain a high concentration at ground level where it is most dangerous to people nearby. The immediate meteorological conditions of concern during a hazmat release are wind speed, wind direction, and atmospheric stability class.

36

Chapter 2 · Hazards Requiring Evacuation Management

Figure 2.7 Effects of Wind Speed on Plume Dispersion Vulnerable zone

Vulnerable zone Anytown

Anytown

B: Strong Wind

A: No Wind

From Lindell et al. 2006

The effect of wind speed on atmospheric dispersion can be seen in Figure 2.7, which shows a release dispersing uniformly in all directions when there is no wind (Panel A). Thus, the plume isopleth (contour of constant chemical concentration) corresponding to the Level of Concern— the Vulnerable Zone or VZ—for this chemical is a circle. The nearby town lies outside the VZ so its inhabitants would not need to take protective action. However, Panel B describes the situation in which there is a strong wind, so the plume isopleth corresponding to the Level of Concern for this chemical takes the shape of an ellipse. In this case, the nearby town lies inside the VZ so people there would need to take protective action. As Table 2.3 indicates, atmospheric stability can vary from Class A through Class F. Class A, the most unstable condition, occurs during strong sunlight (e.g., midday) and light wind. This dilutes the released

Table 2.3 Atmospheric Stability Classes Strength of sunlight Surface Wind Speed (mph)

Strong

Moderate

Nighttime conditions Slight

Overcast ≥ 50%

Overcast < 50%

< 4.5

A

A–B

B

–

–

4.5–6.7

A–B

B

C

E

F

6.7–11.2

B

B–C

C

D

E

11.2–13.4

C

C–D

D

D

D

>13.4

C

D

D

D

D

A: Extremely Unstable Conditions B: Moderately Unstable Conditions C: Slightly Unstable Conditions D: Neutral Conditions (heavy overcast day or night) E: Slightly Stable Conditions F: Moderately Stable Conditions

Chapter 2 · Hazards Requiring Evacuation Management

37

chemical by mixing it into a larger volume of air. Class F identifies the most stable atmospheric conditions, which take place during clear nighttime hours when there is a light wind. These conditions have very little vertical mixing, so the released chemical remains highly concentrated at ground level. It is important to recognize that meteorological characteristics can sometimes remain stable for days at a time, but at other times can change from one hour to the next. Figure 2.8, which Lindell et al. (2006) adapted from McKenna (2000), displays the wind direction at each hour during the day of the 1979 accident at the Three Mile Island (TMI) nuclear power plant in terms of the orientation of an arrow. Wind speed is indicated by the length of the arrow. The figure shows wind speed and direction changed repeatedly during the course of the accident, so any recommendation to evacuate the area downwind from the plant would have referred to different geographic areas at different times during the day. This would have made evacuation notices extremely problematic because the time required to evacuate these areas would have taken many hours. Consequently, the evacuation of one area would have still been in progress when the order to initiate an evacuation in a very different direction was initiated. The ultimate concern in emergency management is the protection of the population at risk. The risk to this target population varies inversely with distance from the source of the release. Specifically, the concentration (C) of a hazardous material decreases with distance (d) according to the inverse square law (i.e., C = 1/d2). However, distance is not the only factor that should be of concern. In addition, the density of the population should be considered because a greater number of persons per unit area increases risk area population. Moreover, there might be differences in susceptibility within the risk area population because individuals differ in their dose-response relationships as a function of age (the youngest and oldest tend to be the most susceptible) and physical condition (those with compromised immune systems are the most susceptible). Toxic chemicals differ in their exposure pathways—inhalation, ingestion, and absorption. Inhalation is the means by which entry into the lungs is achieved. This is generally a major concern because toxic materials can pass rapidly through lungs to bloodstream and on to specific organs within minutes of the time that exposure begins. Ingestion is of less immediate concern because entry through the mouth into the digestive system (stomach and intestines) is a slower route into the bloodstream and on to specific organs. Depending on the chemical’s concentration and toxicity, ingestion exposures might be able to be tolerated for days or months. Authorities might choose to prevent ingestion exposures by withholding contaminated food from the market or recommending that those in the risk area drink boiled or bottled water. Absorption involves entry directly through the pores of

38

Chapter 2 · Hazards Requiring Evacuation Management

Figure 2.8 Wind Rose from 3am to 6pm on the First Day of the TMI Accident

N

W

E 5 mph 10 mph

S From Lindell et al. 2006

the skin (or through the eyes), so it is more likely to be a concern for first responders than for local residents. Nonetheless, some chemicals can affect local populations in this way, as was the case with the release of methyl isocyanate during the accident in Bhopal, India, in 1984. The harmful effects of toxic chemicals are caused by alteration of cellular functions (cell damage or death), which can be either acute or chronic in nature. Acute effects occur during the time period from 0–48 hours. Irritants cause chemical burns (dehydration and exothermic reactions with cell tissue). Asphyxiants are of two types; simple asphyxiants such as carbon dioxide (CO2) displace oxygen (O2) within a confined space or are heavier than oxygen so they displace it in lowlying areas such as ditches. By contrast, chemical asphyxiants prevent the body from using the oxygen even if it is available in the atmosphere. For example, carbon monoxide (CO) combines with the hemoglobin in red blood cells more readily than does O2 so the CO prevents the body from obtaining the available O2 in the air. Anesthetics/narcotics depress the central nervous system and, in extreme cases, suppress autonomic responses such as breathing and heart function. Chronic, or long term, effects can be general cell toxins, known as cytotoxins, or have organ specific toxic effects. In the latter case, the word toxin is preceded by a prefix referring to the specific system affected. Consequently, toxins affecting the circulatory system are called hemotoxins, those affecting the liver are hepatotoxins, those affecting the kidneys are nephrotoxins, and those affecting the nervous system are referred to as neurotoxins. Other chronic effects of toxic

Chapter 2 · Hazards Requiring Evacuation Management

39

chemicals are to cause cancers, so these chemicals are referred to as carcinogens. Mutagens cause mutations in those directly exposed, whereas teratogens cause mutations to the genetic material of those directly exposed and, thus, mutations in their offspring. The severity of any toxic effect is generally due to a chemical’s rate and extent of absorption into the bloodstream, its rate and extent of transformation into breakdown products, and its rate and extent of excretion of the chemical and its breakdown products from the body (i.e., the substances into which the chemical decomposes). Research on toxic chemicals has led to the development of dose limits. Some important concepts in defining dose limits are the LD-50, which is the dose (usually of a liquid or solid) that is lethal to half of those exposed, and the LC-50, which is the concentration (usually of a gas) that is lethal to half of those exposed. Based upon these dose levels, authoritative sources have devised dose limits that are administrative quantities that should not be exceeded. LOCs are values provided by the EPA indicating the Level of Concern or “concentration of an EHS [Extremely Hazardous Substance] above which there may be serious irreversible health effects or death as a result of a single exposure for a relatively short period of time” (USEPA 1987, pp. 2–4). IDLHs are values provided by NIOSH/OSHA indicating the concentration of a gas that is Immediately Dangerous to Life or Health for those exposed more than 30 minutes. TLVs are Threshold Limit Values, which are the amounts that the American Conference of Government Industrial Hygienists has determined that a healthy person can be exposed to 8–10 hours/day, 5 days/week throughout their working life without adverse effects.

2.5.1 Hazmat Hazard Analysis—Fixed Site Facilities These principles can be used to define the VZs around hazmat facilities. The process begins by identifying dangerous chemicals (i.e., those that are threats because of their flammability, reactivity, or toxicity), their locations, and the quantities stored at those locations. Once the chemical inventory has been developed, VZs can be computed using data on the chemical’s toxicity, its quantity available for release, the type of spill (liquid or gaseous), the postulated release duration (e.g., 10 minutes), assumed meteorological conditions (wind speed and atmospheric stability), and terrain (urban or rural). Available methods include manual computations (USEPA 1987), ALOHA (FEMA, no date, see information about CAMEO at www.epa.gov/cameo), or RMP*Comp (www. epa.gov/rmp/rmpcomp). Once the radii of the VZs for the different chemicals have been computed, these can be overlaid onto a map with the release point in the center of the circle and the radius drawn around it (see Figure 2.9).

40

Chapter 2 · Hazards Requiring Evacuation Management

Figure 2.9 Vulnerable Zones Around a Fixed-site Facility and Transportation Route

Fixed site facility

Facility Vulnerable Zone

Hazmat transportation route

Hazmat Inc.

Transportation route Vulnerable Zone

From Lindell 2006

2.5.2 Hazmat Hazard Analysis—Transportation Emergency managers also should identify the highway, rail, water, and air routes through which hazardous materials are transported. Once these routes have been identified, the number of tank trucks, railroad tank cars, and barges carrying each type of hazardous material can be counted in a commodity flow study. Information about hazardous materials transportation can be found on the US DOT Web site (hazmat.dot.gov) and specific guidance for commodity flow studies is found at hazmat.dot.gov/hmep/guide_flow_surveys.pdf. Once the chemicals being transported have been identified, analysts can use the same procedures that were used for fixed site facilities. As Figure 2.9 indicates, this will lead to the construction of rectangular VZs surrounding the transportation routes. The facility and transportation route VZs can then be examined to identify areas of residential, commercial, and industrial land use (see Lindell 1995, for an analysis of hazardous waste transportation to an incinerator). If VZs for the transportation routes have not been prepared before an incident occurs, local emergency responders need to search the incident scene for shipping papers, a numbered placard, or a placard with an orange numbered panel to obtain either the hazmat’s name or its ID number so they can identify the correct section of the Emergency

Chapter 2 · Hazards Requiring Evacuation Management

41

Response Guidebook (www.phmsa.dot.gov/hazmat/erg/erg2016-eng lish) to read. This document identifies these chemicals’ potential hazards (fires, explosions, and toxic exposures), recommended response actions by firefighters and hazmat technicians, appropriate protective actions for emergency responders and the public, and recommended first aid measures for those who have received excessive exposures. For chemicals posing the greatest inhalation hazard, the Emergency Response Guidebook directs emergency responders to a Table of Initial Isolation and Protective Action Distances that should be used to determine where protective action (either evacuation or shelter in-place) should be implemented. Emergency managers should understand that the Emergency Response Guidebook classifies releases only as small or large, so use of the procedures in the Technical Guidance for Hazards Analysis or more advanced methods identified in the Handbook of Chemical Hazard Analysis Procedures is preferred to the table in the Emergency Response Guidebook. In particular, the protective action distances in the Emergency Response Guidebook are not appropriate for use in computing VZs for fixed site facilities manufacturing or storing Extremely Hazardous Substances defined by the Environmental Protection Agency under the Emergency Planning and Community Right to Know Act (EPCRA, also known as Title III of the Superfund Amendments and Reauthorization Act—SARA Title III).

2.5.3 Hazmat Hazard Analysis—Nuclear Power Plants A special case of hazmat exposures arises in connection with nuclear power plants. To understand the radiological hazards of these power plants, it is necessary to understand the atomic fission reaction. The atoms of chemical elements consist of positively charged protons and neutrally charged neutrons in the atom’s nucleus, together with negatively charged electrons orbiting around the nucleus. Some unstable chemical elements undergo a process of spontaneous decay in which a single atom divides into two less massive atoms (known as fission products) while emitting energy in the form of heat and ionizing radiation. The ionizing radiation can take the form of alpha, beta, or gamma radiation. Alpha radiation can travel only a very short distance and is easily blocked by a sheet of paper but is dangerous when inhaled (e.g., Pu—plutonium). Beta radiation can travel a moderate distance but be blocked by a sheet of aluminum foil. Gamma radiation can travel a long distance and can be blocked only by very dense substances such as stone, concrete, or lead. Radioactive materials are used for a variety of purposes. Small quantities of some materials are used as sources of radiation for medical and industrial diagnostic purposes (e.g., imaging fractured bones and faulty welds). Large quantities of other radiological materials are used

42

Chapter 2 · Hazards Requiring Evacuation Management

as sources of heat to produce the steam needed to drive electric generators at power plants. In these nuclear power plants, enriched uranium fuel fissions when struck by a free neutron. The thermal energy released is used to heat water and, thus, produce steam. The free neutrons are used to continue a sustained chain reaction and the fission products are waste products that must eventually be disposed in a permanent repository. The nuclear fuel temperature is controlled by cooling water and the reaction rate is controlled by neutron absorbing rods. The nuclear fuel is located in a reactor vessel that is connected to the plant’s reactor coolant system (RCS), which picks up heat from the fuel and uses it to produce steam to drive a turbine that produces electricity. The reactor vessel is located in a containment building (the turbine is in an adjacent building), which is constructed with thick walls of steelreinforced concrete. However, it has many penetrations for water pipes, steam pipes, and instrumentation and control cables. These penetrations are sealed during normal operations, but the seals could be damaged during an accident that allows radioactive material to escape from the containment building into the environment. During a severe accident involving irreversible loss of coolant, the fuel will first melt through the steel cladding, then melt through the RCS, and finally escape the containment building as early as 45–90 minutes after core uncovery. If the core melts, the danger to offsite locations depends upon containment integrity. Early health effects are likely if there is early total containment failure and are possible if there is early major containment leakage. Otherwise, early health effects are unlikely. The problem is that containment failure might not be predictable (McKenna 2000). Exposure pathways for radiological materials are similar to those of toxic chemicals. Breathing air that is contaminated with radioactive materials can cause inhalation exposure and eating food (e.g., unwashed local produce) or drinking liquids (e.g., water or milk) that is contaminated can cause ingestion exposure. Contamination also can enter the body through an open wound such as a compound fracture, laceration, or abrasion, but radiological materials do not cause absorption exposures because they do not pass through the skin. However, because radiological materials release energy, they can produce exposures via direct radiation (also known as “shine”) from a plume that is passing overhead. If the plume has a significant component of particulates, these might be deposited on the ground, vegetation, vehicles, or buildings and the direct radiation from the deposited particulates would produce a continuing exposure. In some cases, the small particles of deposited material could become resuspended and inhaled or ingested. In this connection, it is important to recognize the distinction between irradiation and contamination. Irradiation involves the transmission of energy to a target that absorbs it, whereas contamination occurs when radioactive particles are deposited in a location within the body where they provide continuing irradiation.

Chapter 2 · Hazards Requiring Evacuation Management

43

The US Nuclear Regulatory Commission (NRC) conducted extensive analyses to define the emergency planning zones (EPZs) that should be designated around these facilities. The NRC’s analyses led to the establishment of a 10-mile radius plume inhalation EPZ in which state and local authorities should be prepared for people to initiate protective actions to avoid inhalation exposure and direct radiation from a radioactive plume (USNRC 1978). Protective actions within the 10-mile EPZ are expected to include sheltering, evacuation, and the use of potassium iodide. In addition, there is a 50-mile radius ingestion pathway EPZ in which authorities should prepare to monitor water, milk, and food (especially leafy green vegetables) for contamination. Within the 50-mile radius, actions include bans on contaminated food and water (USNRC 1980). A hazard map for the 10-mile plume exposure typically places the nuclear power plant at the center and draws the 2-, 5-, and 10-mile concentric circles around the plant (Figure 2.10). Lines then extend out from the plant to segment the circles into wedge shaped areas that are used to identify which municipalities are affected when an actual plume travels (USNRC 2014b). Consistent with Figure 2.7b, radiological emergency response planners anticipate implementing a “keyhole” evacuation involving everyone within a 2-mile radius in all directions around the plant with additional evacuations in the sectors that are 5, or possibly 10, miles downwind (see the shaded area in Figure 2.10). Of

Figure 2.10 EPZ Around a Nuclear Power Plant with 2, 5, and 10 Mile Zones and Eight Directional Sectors and Keyhole Evacuation Zone Shaded Wind Direction A Wind Direction B

44

Chapter 2 · Hazards Requiring Evacuation Management

course, as the discussion of Figure 2.7 indicated, radiological exposures in excess of EPA standards would be confined to the shaded area of Figure 2.10 only if the wind direction is stable for the entire duration of the evacuation. However, the EPA standards will be exceeded in a different sector if the wind changes from direction A to direction B during the release. As indicated in Figure 2.8, the shift in wind direction can be much more dramatic than what is indicated in Figure 2.10. US commercial nuclear power plants use the NRC’s Emergency Classification System when advising government agencies and the public about hazardous conditions. A Notification of an Unusual Event is initiated when conditions indicate the potential for degradation of plant safety but there are no radiological releases requiring offsite response. An Alert is declared when conditions involve a substantial degradation of plant safety, either potential or actual, but any exposures are expected to be well below EPA protective action guidelines. A Site Area Emergency indicates that major failures of plant safety functions are in progress or have occurred but are not likely to exceed EPA protective action guidelines beyond the site boundary. Finally, a General Emergency involves substantial core degradation with potential loss of containment, either actual or imminent, that is likely to produce exposures greater than EPA protective action guidelines outside the immediate plant area.

References Bell, H.M., Tobin, G.A. 2007. Efficient and effective? The 100-year flood in the communication and perception of flood risk. Environmental Hazards 7 (4), 302–311. Blake, E.S., Zelinsky, D.A. 2018. National Hurricane Center Tropical Cyclone Report: Hurricane Harvey AL092017. National Hurricane Center, Coral Gables FL. Darienzo, M., Aya, A., Crawford, G.L., Gibbs, D., Whitmore, P.M. Wilde, T., Yanagi, B.S. 2005. Local tsunami warning in the Pacific coastal United States. Natural Hazards, 35 (1), 111–119. FEMA—Federal Emergency Management Agency. 2016. National Flood Insurance Program: Flood Hazard Mapping. Federal Emergency Management Agency, Washington DC, accessed 24 June, 2017 at www.fema.gov/ national-flood-insurance-program-flood-hazard-mapping. FEMA—Federal Emergency Management Agency. no date. An Emergency Alert System Best Practices Guide, Version 1.0. Federal Emergency Management Agency, Washington DC. FEMA/DOT/EPA—Federal Emergency Management Agency, U.S. Department of Transportation, and U.S. Environmental Protection Agency. no date, a. Handbook of Chemical Hazard Analysis Procedures. Federal Emergency Management Agency, Washington DC.

Chapter 2 · Hazards Requiring Evacuation Management

45

Gess, D., Lutz, W. 2002. Firestorm at Peshtigo: A Town, its People, and the Deadliest Fire in American History. Macmillan, New York. Li, D., Cova, T.J., Dennison, P.E. 2015. A household-level approach to staging wildfire evacuation warnings using trigger modeling. Computers, Environment and Urban Systems, 54, 56–67. Lindell, M.K., Prater, C.S. 2010. Tsunami preparedness on the Oregon and Washington coast: Recommendations for research. Natural Hazards Review, 11 (2), 69–81. Lindell, M.K., Prater, C.S., Perry, R.W. 2006. Fundamentals of Emergency Management. Emmitsburg MD: Federal Emergency Management Agency Emergency Management Institute. www.training.fema.gov/EMIWeb/edu/ fem.asp or hrrc.arch.tamu.edu/publications/books/. McKenna, T.J. 2000. Protective action recommendations based upon plant conditions, Journal of Hazardous Materials 75 (2–3), 145–164. National Hurricane Center. 2014. Sea, Lake, and Overland Surges from Hurricanes (SLOSH). National Oceanic and Atmospheric Administration, National Hurricane Center, Coral Gables FL, accessed 27 May 2017 at www.nhc.noaa. gov/surge/slosh.php. Terpstra, T., Lindell, M.K. 2013. Citizen’s perceptions of flood hazard adjustments: an application of the protective action decision model. Environment and Behavior 45 (8), 993–1018. USACE—US Army Corps of Engineers. 2002. Mississippi Hurricane Evacuation Study: Technical Data Report. US Army Corps of Engineers, accessed 14 October 2016 at coast.noaa.gov/hes/hes.html. USEPA—US Environmental Protection Agency 1987. Technical Guidance For Hazards Analysis: Emergency Planning For Extremely Hazardous Substances. US Environmental Protection Agency, Washington, DC. USNRC—US Nuclear Regulatory Commission/Federal Emergency Management Agency. 1980. Criteria for Preparation and Evaluation of Radiological Emergency Response Plans and Preparedness in Support of Nuclear Power Plants. NUREG-0654, FEMA-REP-1, Rev.1. US Nuclear Regulatory Commission, Washington DC. Van Wagner, C. 1969. A simple fire-growth model. The Forestry Chronicle 45 (2), 103–104.

Chapter 3

Protective Actions and Protective Action Decision Making

This chapter provides an overview of two protective actions that are common to most environmental hazards—sheltering in-place and evacuation. These protective actions are based upon three fundamental protection factors—time (i.e., exposure duration), distance, and shielding. Evacuation reduces the amount of time exposed and increases distance from the hazard source, whereas sheltering in-place can provide shielding from the hazard. In many situations, such as hurricanes, there is adequate time to clear the risk area before hazard impact so evacuation is preferred because it provides the greatest level of protection. In such cases, shelter inplace might be recommended for areas on the fringe of the evacuation zone because it provides greater protection than being outside. In other cases, such as tornadoes, there is not likely to be enough time to evacuate safely, so shelter in-place is routinely recommended. Nonetheless, there are many situations in which it is not immediately obvious which protective action should be recommended. Section 3.1 addresses shelter in-place, Section 3.2 addresses evacuation, Section 3.3 discusses protective action recommendation (PAR) selection, and Section 3.4 discussed PAR timing.

3.1 Overview of Sheltering In-place Sheltering in-place—going indoors, closing doors and windows, and turning off heating, ventilation, and air conditioning systems—is an appropriate course of action when little time is available to react to an event and it would be more dangerous outside than staying indoors. In wildfires, a body of water may be considered a place to shelter (Cova et al. 2009). “Officials should recommend shelter in-place only when there is reasonable assurance that moving people beyond their residence, workplace, or school will endanger their health and safety more so than allowing them to remain in place” (Sorensen, Shumpert, and Vogt 2004, p. 1). In some cases, the current location is safer than evacuation routes (Cova et al. 2009), such as when roads may be blocked by fire or heavy concentrations of chemicals. Shelter in-place is recommended when the

Chapter 3 · Protective Actions and Decision Making

47

shelter provides adequate protection and evacuation cannot be performed with the available time (Sorensen, Shumpert, and Vogt 2004). Chemical accidents represent one type of event when shelter in-place may be more appropriate (USDOT 2006). Shelter in-place can be effective for reducing exposure to peak concentrations for a limited time but may not reduce the cumulative dose over a longer period (Lindell and Perry 1992). Longer releases allow more of a chemical to seep indoors, which means a longer exposure for people within the shelter unless the occupants leave their shelters as soon as the chemical plume has passed. Shelter in-place is generally not recommended for chemicals that are flammable or explosive when exposed to air, but it might still be preferable when evacuation cannot be performed quickly. As indicated in Chapter 2, the decision maker can determine whether shelter in-place is the appropriate protective action by considering the characteristics of the released chemical, meteorological conditions, and characteristics of structures (e.g., how airtight they are), among other factors (Lindell and Perry 1992; Sorensen, Shumpert, and Vogt 2004). Just as with evacuation, shelter in-place requires some time, although the amount of time will almost always be smaller. Authorities need time to detect the hazard, select PARs, and disseminate warnings, and households need time to implement those PARs. In the case of shelter in-place, people who are outdoors must go inside, and close windows and doors, and take further actions depending on the nature of the hazard. For tornadoes, the length of forewarning is so short that they should immediately go to the basement or, if none is available, to an interior room on the ground floor. Other hazards have more forewarning so additional home preparations are possible, such as clearing potential fuel when threatened by a wildfire, filling and placing sandbags in anticipation of a flood, and boarding windows before hurricane wind becomes too strong. For chemical releases, people in the risk area should shut off heating, ventilation, and air conditioning systems to reduce air infiltration and, if possible, tape doors and windows (Sorensen, Shumpert, and Vogt 2004). Sealing a room takes approximately 17 minutes (Rogers et al. 1990). In chemical emergencies, it is also important for people to tune to the emergency alert system (EAS) so they can receive further instructions such as when to open the doors and windows after the plume has passed (Sorensen, Shumpert, and Vogt 2004). If the message is that a combined strategy of shelter in-place and evacuation is recommended, communication becomes even more essential. Shelter in-place might be recommended for people in more airtight homes whereas people in older, “leaky” structures may need to evacuate or seek shelter elsewhere. Low compliance with shelter in-place recommendations should be expected when both shelter in-place and evacuation are advised (Sorensen, Shumpert, and Vogt 2004). Compliance can be improved by explaining why shelter in-place provides effective protection and why evacuation might result in “sheltering” in a car when overtaken by a chemical plume.

48

Chapter 3 · Protective Actions and Decision Making

PARs may also change over the course of an event as a hazard and the state of the ERS evolve. Wildfires, for example, may allow people to evacuate early or shelter in their homes or a local refuge. Some residents may stay, rather than evacuate, in an attempt to defend their homes. In this case, it could be difficult to detect differences in the goals of protecting life and protecting property (Cova et al. 2009). As the fire front progresses toward inhabited areas, roads can become too hazardous to travel, removing the option of late evacuation and leaving shelter in-place as the only option. For situations in which shelter in-place is clearly the better course of action, communication is critical. Authorities should promptly and clearly explain the dangers of being outside in the areas of high exposure, especially if environmental cues (sights, sounds, or smells) are lacking, as is the case with radiological releases and some chemical releases. In such cases, the only way the risk area population can become aware of potential danger is by social warnings from authorities, news media, or peers. Trust in the information and its source is critical to achieving the sheltering behavior. Warnings should indicate when to begin and when to end shelter in-place (Sorensen, Shumpert, and Vogt 2004). Poor communication could result in people not perceiving sheltering to be effective which, in turn, could lead them to evacuate into danger. There are three major mechanisms by which sheltering in-place provides protection. First, it can provide protection from wind and water forces. Second, it can provide protection from inhalation of hazardous materials. Third, it can provide protection from direct radiation. Regarding the first mechanism, protection from wind and water forces, structures can be vulnerable to environmental hazards because of inadequate designs, inadequate construction materials, or both. Older homes have been constructed under earlier building codes and most neighborhoods are relatively homogeneous with respect to age, so identifying older neighborhoods in hazard-prone areas will help set priorities of emergency management interventions. For example, many areas of the country have homes, built during the early part of the 20th century, that are usually less weathertight, so they are much more prone to infiltration of hazardous materials. There are two major issues in assessing structural vulnerability. First is the question of whether the structure has the strength or resilience to withstand environmental forces such as wind or water. In this case, the concern is about the impact on the structure itself and, consequently, the loss of function and the time and cost of rebuilding. The second issue concerns the ability of the structure to protect its occupants. This is especially important in connection with hazardous materials because they can infiltrate into a structure and kill the occupants without damaging the building. In high wind (including tornadoes and hurricanes) and explosions (usually technological in origin, but also including some volcanic

Chapter 3 · Protective Actions and Decision Making

49

eruptions), a substantial increase in air pressure can cause structures to collapse. Such structural failures are caused by deficiencies in either design or materials, or both (American Institute of Architects 1995; Institute for Business and Home Safety 1997). Fortified homes (see www.ibhs.org) have connections and braces that reinforce roofs and gable-end walls against wind attack. In addition to positive pressure on upwind walls, high wind creates negative pressure, or suction, as it is forced to flow up and over the roof. This suction is greatest in flat roofs, intermediate in gable-end roofs (which slope in two directions), and least in hipped roofs (which slope in four directions). Suction tends to lift the roof from the walls unless resisted by adequate connections to the walls—which in turn, must have adequate connections to the foundation. Adequate designs and materials also provide protection to building openings such as windows and doors, thus preventing the wind from pressurizing the interior and adding to the stress on roof and walls. The need for window and sliding glass door shutters is widely recognized because the shutters resist the direct pressure of the wind and the impact of flying debris. However, door reinforcement is also important—especially for double-wide (two car) garage doors that are highly susceptible to failure because their great width allows the wind to deflect them inward and pull the rollers out of their tracks. For hurricanes, structures on the open coast must be of sufficiently sturdy construction that they can resist the direct impact of storm surf as well as the force of extremely high winds. Both riverine flooding and hurricane storm surge also require the structure to have foundations anchored well enough to resist scouring by water currents that can undermine building foundations and cause structural collapse. Additional protection can be provided by expedient floodproofing that uses waterproof construction materials, sealing of cracks, provision of valves on sewer lines, steel bulkheads for lowerlevel openings, and sump pumps to eject seepage (USOEP 1972). Tsunami impact poses an even greater threat than inland flooding or surge from hurricanes and coastal storms. These sea waves can threaten areas as much as 100 feet above sea level, so destruction is highly likely for most structures located very near the shoreline. However, properly designed steel-reinforced concrete structures located a short distance inland are likely to survive even the largest tsunamis.

3.1.1 Protection From Inhalation Exposure The principal hazard from toxic and radiological materials is inhalation of airborne materials (USEPA 1987). An enclosed space will provide a barrier if it can be closed tightly enough to keep out the hazardous material and has enough oxygen to sustain those within it until the danger has passed. Unfortunately, most structures are leaky, allowing contaminated air to infiltrate even when the doors and windows are

50

Chapter 3 · Protective Actions and Decision Making

closed. The rate of air exchange increases with the amount of leakage area, the wind speed, and the temperature differential between the indoor and outdoor air. The rate at which indoor and outdoor air are exchanged is commonly measured in air changes per hour (ACH). However, emergency managers will find it more useful to think of air exchange in terms of turnover time, which is the reciprocal of the air exchange rate, or tB = 1/ ACH, where ACH is the number of air changes per hour (Mannan and Kilpatrick 2000). As Wilson (1987) emphasizes, an infiltration rate of 1.0 ACH does not imply that all the clean air will be gone in one hour. Rather, the proportion of contaminated air gradually rises until, at the end of 1.0 tB hours, 63% of the original air has been replaced by contaminated air and 95% of the original air has been replaced by the end of 3.0 tB hours. Thus, for the case of 1.0 ACH, it will take over three hours (not just one hour) for the indoor air to become almost completely contaminated. This result is extremely important because it indicates that in-place sheltering is more effective than most people might infer from the apparent implication of the number of air changes per hour. The reason for the difference between the apparent result and the correct result can best be illustrated by examining the difference between the apparent, but incorrect, mechanism of air exchange and the actual mechanism. It would only take one hour to replace the clean air (the incorrect result) if the contaminated air somehow “pushed out” the clean air, but this is not what happens. Rather, the contaminated air that infiltrates into the structure mixes with the clean air rather than “pushing it out”. Clearly, exfiltration of a mixture of clean air and contaminated air will take longer to exhaust the clean air in a structure than will exfiltration of (“pushing out”) clean air alone. Consequently, sheltering in-place is at least three times as effective in reducing inhalation exposure as it first appears to be. Although this time lag effect is important, it is not the only mechanism by which sheltering in-place can reduce adverse health effects. It is also important to recognize the impact of a damping effect in reducing the fluctuations in plume concentrations. As discussed earlier, these fluctuations in wind speed and direction arise from irregularities in meteorological conditions and local terrain. One way of measuring peak concentration is by estimating the value that is exceeded approximately 1% of the time. Wilson (1987) reports that in the outdoor (contaminated) air, such 1% peak concentrations are 400% as large as the mean concentration. For indoor air, the equivalent peak concentration is only 50% larger than the mean. As he notes, even when the indoor concentration (100 parts per million [ppm], for example) has risen after six hours to match the outdoor concentration, the indoor peaks would be expected to be approximately 150 ppm when the outdoor peaks would be 400 ppm. This is, of course, of considerable significance when peak concentrations are the principal health threats.

Chapter 3 · Protective Actions and Decision Making

51

As Wilson (1987, 1989) and others have observed, a major problem in assessing the effectiveness of sheltering in-place is uncertainty about whether indoor air concentrations will remain sufficiently low for a sufficiently long period of time. This can be answered definitively only if there is information about the hazardous material being released (especially the identity of the material released, and the rate and duration of the release), the meteorological data needed for a computerized plume dispersion model (wind speed, wind direction, and atmospheric stability), and the air exchange rates for the structures in the hazard impact area. Data on the release and the meteorological conditions will not be available until an incident occurs, but data on the efficacy of sheltering in-place can be collected in advance. First, Rogers et al. (1990) report that energy conservation research has shown air exchange in most US dwellings ranges from 0.5 to 1.5 ACH. Second, Wilson (1989) reports that the most important factor affecting leakage area is the presence of a vapor barrier in the walls and ceiling of a structure, a feature that is most common in houses built in cold climates after 1960. Thus, emergency managers could estimate the effectiveness of sheltering in-place by obtaining access to local data on the age of the housing stock within different neighborhoods within their jurisdiction, focusing especially on neighborhoods with older, smaller homes occupied by low-income households (Chan et al. 2007a, 2007b). Finally, the fact that so much of the research and data on infiltration of hazardous materials has been developed from studies of energy conservation suggests that emergency planners consult with their local utilities to determine what information is available regarding the air exchange rates of different types of structures (e.g., residences, schools, and commercial buildings) in their communities. Special facilities, especially those such as hospitals that have low mobility residents, should be examined individually to assess their air exchange rates.

3.1.2 Protection From External Radiation Exposure Although both toxic and radiological materials present an inhalation hazard, a plume of radioactive material released from a nuclear power plant or during a transportation accident also can cause harm by means of external gamma radiation from the cloud and from ground contamination. Dense building materials such as concrete, brick, and stone provide shielding from external gamma radiation and, thus, can provide a basis for in-place sheltering during radiological emergencies. The effectiveness of structures made from different types of building materials has been examined in studies by Burson and Profio (1977), Anno and Dore (1978a; 1978b), Aldrich, Ericson and Johnson (1978), and Aldrich, et al. (1982). These studies calculated the dose reduction factors (the ratio of the dose received while sheltering to the unprotected dose) for three exposure routes: external gamma radiation from the cloud, external gamma

52

Chapter 3 · Protective Actions and Decision Making

radiation from ground contamination, and inhalation of radioactive materials infiltrating into the structure. Burson and Profio (1977) found that sheltering in a wood frame dwelling provides little more protection from cloud and ground exposure than does “sheltering” in a vehicle while evacuating. Sheltering on the ground floor of a masonry home with no basement or in the basement of a wood frame home gave considerably higher levels of protection: about 50% of the unprotected exposure to the cloud and less than 20% of the unprotected exposure to the ground. As one might expect, the basement of a masonry house was even more effective: 40% of the cloud exposure and 5% of the ground exposure. A large office building was the most effective shelter of all, reducing cloud exposure to about 20% and ground exposure to 1%. The importance of the construction materials is underscored by Burson and Profio’s (1977) work indicating it is the cloud exposure that produces most of the whole body radiation dose received by those sheltering in a home. Infiltration into the structure would account for only about 5% of the gamma radiation dose. Anno and Dore (1978a) calculated cloud dose reduction factors for single family dwellings and large structures (e.g., office buildings, multistory apartment complexes). They considered 0.125 to 3 ACH to define the range of infiltration rates for single family dwellings and other structures that could be used as temporary public shelters. For single family dwellings, whole body dose reduction factors for low air exchange rates (0.125 ACH) were calculated to be 0.40–0.33 compared to 0.43 for more representative air change rates (3 ACH). For large structures, whole body dose reduction factors for low air change rates were calculated to be 0.08 compared to 0.17–0.11 for the more representative air change rates. These investigators also estimated thyroid (inhalation) dose reduction factors to be about 0.05–0.01 for low air change rates and from 0.25–0.10 for more representative air change rates for either single family dwellings or large structures.

3.2 Overview of Evacuation Following Urbanik et al. (1980), the time required for an individual resident or transient household to evacuate after incident initiation can be defined as a function of four time components as shown in Equation (3.1). tT ¼ f ðtd ; tw ; tp ; te Þ where tT is a household’s total clearance time, td is the authorities’ decision time, tw is the household’s warning receipt time,

Chapter 3 · Protective Actions and Decision Making

53

tp is the household’s evacuation preparation time, and te is the household’s evacuation travel time. Authorities’ decision times (td) are critically important in rapid onset incidents for which there are no environmental cues that precede lifethreatening exposures. This can be the case in some nuclear power plant accidents—the situation addressed by Urbanik et al. (1980)—as well as in some flash floods and control structure failures, wildfires, remote tsunamis, and toxic chemical releases. However, authorities’ decisions can be important in slow-onset incidents such as main stem floods and hurricanes because official PARs provide those at risk with an indication that it is now time to take protective action. The time to warning receipt (tw) is highly variable because it depends on the nature of the local warning mechanisms. As Table 3.1 indicates, these warning mechanisms differ in many ways—with those having the lowest sender and receiver requirements (i.e., specialized equipment) having the slowest rate of dissemination. For example, door to door warning requires no specialized equipment for either senders or receivers, so this warning mechanism is quite common in rural jurisdictions that have lack the financial resources to pay for more rapid warning dissemination. Household evacuation preparation time (tp) is also highly variable because it depends on the amount of forewarning and people’s perceptions of the threat. For example, people are likely to gather the family and leave immediately when they expect the imminent arrival of a lifethreatening hazard such as a toxic chemical release. Finally, household evacuation travel time (te) also varies depending on the capacity of the ERS and the demand for space on that ERS. This, of course, is the topic of the remaining chapters.

3.3 PAR Decision Making In most cases, the chief administrative officer of the affected community selects the appropriate PARs in consultation with the local emergency manager at the community EOC. However, the Incident Commander at the scene of a hazmat transportation incident will often issue the PAR based on the guidance in the Emergency Response Guidebook. Moreover, nuclear power plants and toxic chemical facilities might have special arrangements with their local jurisdictions to transmit PARs directly to a Dispatch Center that can immediately activate sirens and warn people to take immediate protective action if a release is imminent or in progress. Hazards that affect multiple communities might require decisions at the county level and hazards that affect multiple counties (e.g., hurricanes) often involve decisions by the state governor. In addition, radiological emergencies that require specialized expertise that is unavailable at the

High

Moderate

Sirens

Tone alert radio

High

Moderate

Door-to-door

Social media

Low

High

Moderate

High

Moderate

High

Moderate

Penetration of normal activities

High

High

High

High

High

Source: Adapted from Lindell and Prater (2010)

Low Moderate3

Low Low3

Moderate

Moderate

Low2 Low

High

High

High

Dissemination rate

Low

Low

Low

Low1 High

Susceptibility to distortion

Message specificity

1 Low for “tone only” systems; Moderate for voice broadcast systems 2 Low for Reverse 911 systems; High for telephone notification trees 3 Low for direct links to emergency management agencies; Moderate to high for peer groups

High

High

Telephone

Route alert

Commercial TV/ Moderate radio

Precision of dissemination

Warning mechanism

Table 3.1 Characteristics of Warning Mechanisms

Moderate

Low

Low

Low

Low

Moderate

Low

Receiver requirements

Moderate

Low

Low

Moderate

Moderate

Moderate

High

Sender requirements

Moderate

High

Moderate

High

Low

Low

Low

Feedback

Chapter 3 · Protective Actions and Decision Making

55

community or county levels are likely to involve decisions by the state governor that are based upon recommendations by the state radiation protection agency.

3.3.1 PAR Selection The first issue in PAR decision making is the selection of which geographical areas, demographic segments, and special facilities should evacuate, which ones should shelter in-place, and which ones can continue normal activities. With respect to geographic areas, shelter in-place should be recommended for those areas in which the risk is insufficient to justify a high cost of evacuation and a slight risk of traffic accidents. However, shelter in-place can also be appropriate for safe structures within otherwise hazardous areas. For riverine flooding, tsunamis, and hurricane storm surge, concrete structures with well-anchored foundations are strong enough to resist battering waves and provide the height to escape the rising water that could threaten contents and occupants. In other cases, it is the strength of construction in resisting hurricane wind loads that protects the structure, contents, and occupants. For chemical and radiological threats, it is the tightness of construction in preventing the infiltration of outside (contaminated) air into the structure that is the important protective feature. Finally, in the case of exposure to a cloud of radioactive material, the construction material can provide shielding from penetrating radiation and from surface contamination. Conversely, some structures are much more vulnerable than others. For example, there are many situations in which mobile home occupants should evacuate from high wind zones even if single family and multi-family structures are adequately safe. A useful way to guide decisions about PARs begins with a modification of the keyhole evacuation pattern in Figure 2.10. As Figure 3.1 indicates, where there is uncertainty about the wind direction over the course of the release, the keyhole for the evacuation zone (depicted by the horizontal hatching) can be extended by a shelter in-place notice in the areas adjacent to that evacuation zone (depicted by vertical hatching). Assuming that the wind direction is forecast to be relatively stable, people in the unshaded areas can continue normal activities. There is the possibility that a radiological release could occur before an evacuation can be initiated (as suggested in Figure 1.3). In addition, there is the potential for a continuing threat from exposure to ground deposition during an evacuation after the plume passes. To determine when protective action should be initiated, the EPA has developed Early Phase Protective Action Guides (PAGs), which are specific criteria for initiating population protective action in radiological emergencies (Conklin and Edwards 2000). As Table 3.2 indicates, there are two principal radiological threats —a threat to the thyroid from radioiodine and a threat to the whole body

56

Chapter 3 · Protective Actions and Decision Making

Figure 3.1 Precautionary Shelter In-Place Zones on Either Side of a Keyhole Evacuation Zone Wind Direction A

Table 3.2 EPA Protective Action Guides (PAGs) Organ Whole body Thyroid

EPA PAGsa (rem/Sv)

Protective Actionb

1–5 (.01–.05) 25 (.25)

Evacuation Stable Iodine (KI)

a Dose inhalation from and external exposure from plume and ground deposition. b Actions should be taken to avert PAG dose. * Evacuation is considered to be the most effective protective action for nuclear power plant accidents at American sites.

from radioactive noble gases and particulates. If authorities expect the thyroid PAG to be exceeded, the population at risk should take potassium iodide to flood the thyroid with stable iodine and, thus, prevent the uptake of radioactive iodine. If the authorities expect the whole body PAG to be exceeded, the population at risk should evacuate if possible because the dose will decrease with distance. If people cannot evacuate in time to avoid exposure, shelter in-place might be an appropriate protective action. As Figure 3.2 indicates, the probability of exceeding 200 rem (2 Sievert, SV, a measure of radiological dose) decreases with distance even if people continue normal activity. Sheltering in-place in a home basement provides almost no dose reduction for homes within one mile but makes a noticeable difference at five miles.

Chapter 3 · Protective Actions and Decision Making

57

Probability of exceeding 200 rem (2 Sv)

Figure 3.2 Effectiveness of Protective Actions in a Nuclear Power Plant Emergency 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1 mi (1.6 km)

3 mi (5 km)

5 mi (8 km)

1 Normal activity

4 Evacuation 1 hour before plume arrival

2 Home basement shelter

5 Evacuation after plume arrival

3 Large building shelter Source: Adapted from McKenna (2000).

The significance of exceeding 200 rem (2 SV), the reference value on the y-axis of Figure 3.2, can be seen in the EPA’s Table of Protective Action Guides, which calls for initiating evacuation when the projected dose is expected to be approximately 1% of that reference value. Thus, the PAG for evacuation is an extremely conservative criterion for recommending an evacuation. Sheltering in a large building produces a large dose reduction at one mile and virtually eliminates exposure at three miles or more. Evacuation of locations at one mile from the plant one hour before plume arrival produces a dose reduction that is about the same as sheltering in the basement of a large building at three miles. The same evacuation timing for locations three miles from the plant virtually eliminates exposure. Finally, evacuation one hour after plume arrival reduces the dose somewhat below the level for continuing normal activity, but is not nearly as effective as large building shelter or evacuation before plume arrival. With respect to demographic segments, there are some hazards for which there are significant differences in susceptibility to harm. For example, fetuses and neonates are particularly susceptible to radiation hazard so pregnant women and preschool children might be advised to shelter in-place even if other population segments are not advised to take any protective actions. This basic idea of precautionary zones adjacent to the evacuation zone is also applicable to hurricane threats, as illustrated in Figure 3.3, which shows a hurricane approaching a coastal jurisdiction. In the right-hand side of the figure, there is a small circle indicating the hurricane eye

Chapter 3 · Protective Actions and Decision Making

Figure 3.3 Hurricane Protective Action Zones

Caution Area Risk Area Caution Area

Safe Area Caution Area Safe Area

Safe Area

Safe Area

Evacuation scope ≈ Hurricane intensity/size

Safe Area

58

Predicted storm track

Hurricane eye

Radius of Tropical Storm wind

surrounded by a larger circle showing the radius of Tropical Storm force wind. The hurricane itself is moving along a track toward a coastal location labeled the Risk Area, which includes the hurricane’s expected point of landfall, the distance on either side of the eye’s landfall where there will be damaging wind and surge, and the distance inland from the coast that the damaging wind and surge will extend. There is a Caution Area next to the Risk Area along the coast because of the uncertainty about the storm track and another Caution Area inland from the Risk Area because of the uncertainty about the storm intensity. These Caution Areas might be advised to shelter in-place because the cost and safety hazard of evacuation exceeds the risk of being struck by storm conditions. In addition, shelter in-place for the inland Caution Area will reduce the number of vehicles on evacuation routes leading inland from the Risk Area—thus reducing the likelihood that people will be caught on the road in the Risk Area when the storm makes landfall. Finally, there are Safe Areas farther along the coast and also farther inland that are expected to escape damaging wind and surge. Of course, these areas are not completely safe because a major change in track or intensity could cause them to be affected. In general, the size of the evacuation zone should extend farther inland when the hurricane category is higher (a Category 5 hurricane must travel farther inland before its wind speed drops to a non-threatening level) and should be wider when the hurricane has a larger radius of hurricane wind (because dangerous wind speed extends farther along the coast from the point at which the center makes landfall). Special facilities pose problems because of their users’ characteristics. For example, some facilities’ users have mobility limitations, whereas

Chapter 3 · Protective Actions and Decision Making

59

other facilities’ users are unfamiliar with the area, lack adequate protection factors for shelter in-place, or lack access to vehicles. Table 3.3 has a list of special facilities and Table 3.4 indicates the typical users’ characteristics in those facilities. Nursing homes and hospitals are particularly significant because they are likely to take a long time to evacuate and, moreover, they house fragile patients that will have higher levels of mortality and morbidity when moved (Dosa et al. 2012). Unfortunately, it remains unclear which aspects of the evacuees (i.e., specific medical conditions) and the evacuation process (e.g., evacuation distance, receiving facility capacity) have the greatest effect on adverse health outcomes. One complication for decisions about whether to issue a PAR for a given area is that there is uncertainty before impact about whether the hazard will strike. In addition, emergency managers should not only consider the probability of a strike or miss but also the cost of a strike

Table 3.3 Facilities with Highly Vulnerable Populations Facility Classification Health Related Penal Assembly & Athletic

Examples

Hospitals

Nursing homes

Mental institutions

Halfway houses (drug, alcohol, mental retardation)

Jails

Detention camps

Prisons Auditoriums

Reformatories Gymnasiums

Theaters

Athletic stadiums or fields

Exhibition halls Amusement & Beaches Recreation Camp/conference centers

Parks/lakes/rivers Golf courses

Amusement parks/fairgrounds/race Ski resorts courses Campgrounds/Recreational Vehicle parks Community recreation centers Religious

Churches/synagogues/mosques

Evangelical group centers

High Density Residential

Hotels/motels

Dormitories (college, military)

Apartment/condominium complexes

Convents/monasteries

Mobile home parks Transportation Rivers/lakes

Ferry/railroad/bus terminals

Dam locks/toll booths Commercial

Shopping centers Central business districts

Commercial/industrial parks

Educational

Day care centers

Vocational/business/ specialty schools

Preschools/kindergartens

Colleges/universities

Elementary/secondary schools From Lindell and Perry 1992

60

Chapter 3 · Protective Actions and Decision Making

Table 3.4 Characteristics of Special Facility Users Characteristics Of Users

Special Considerations

Mobility of users

Ambulatory Require close supervision Non-ambulatory Require life support Facility residents Residents of the impact area, but not of the facility (e.g., prison guards) Transients Days of week/hours of day Special events

Permanent residence of users

Periods of use User density

Concentrated Dispersed

Sheltering in-place

Highly effective Moderately effective Minimally or not effective

Transportation support

Would use own vehicles Require buses or other high occupancy vehicles Require ambulances

From Lindell and Perry 1992

or miss. The nature of an emergency manager’s PAR decision can be described by the decision tree in Figure 3.4, which is adapted from the hurricane evacuation decision tree described by Lindell and Prater (2007). The choice of protective action is indicated by the square at the left, and the three alternatives—evacuate, shelter in-place, and no action—branch to the right. Each branch intersects a circle representing an uncontrollable event—the intensity of the hazard strike. Categorizing the hazard as striking with minimal intensity (or not at all), moderate intensity (i.e., will not exceed the protection factor of shelter in-place), or major intensity (i.e., will exceed the protection factor of shelter in-place), produces three branches emanating from each of the circles. Each of these three mutually exclusive and exhaustive branches has an associated probability—pminimal, pmoderate, and pmajor. Collectively, all of the branches on the right-hand side of the diagram represent the nine outcomes that can occur when there are three decision alternatives and the uncontrollable environmental event has three states. If the probability of each branch and the utility of each outcome (i.e., the expected number of lives lost and the economic costs incurred) can be specified quantitatively, a decision maker can identify the protective action that has the greatest expected utility (Clemen and Reilly 2013). A major challenge for this approach is that it requires a tradeoff between fatalities and economic costs, which decision makers resist making explicitly even though they must ultimately make them implicitly (see Hammond and Bier 2015).

Chapter 3 · Protective Actions and Decision Making

61

Figure 3.4 Evacuation Decision Tree Major intensity Evacuate Moderate intensity Miss Shelter in-place

Outcome A: No lives lost; Large costs incurred Outcome B: No lives lost; Large costs incurred Outcome C: No lives lost; Large costs incurred Outcome D: Some lives lost; Small costs incurred

Major intensity Outcome E: No lives lost; Small costs incurred Moderate intensity Outcome F: No lives lost; Small costs incurred Miss

No action

Major intensity Moderate intensity Miss

Outcome G: Many lives lost; No costs incurred Outcome H: Some lives lost; No costs incurred Outcome I: No lives lost; No costs incurred

Adapted from Lindell and Prater, 2007b

Given that the hazard strikes, Outcome A indicates that evacuation is a correct decision if the hazard strikes with major intensity because lives are saved although evacuation costs are incurred. Outcome E is a correct decision if the hazard strikes with only moderate intensity because no lives are lost and only small costs are incurred due to disruption of normal activities. Finally, Outcome I indicates that no action is a correct decision if the hazard fails to materialize because no lives are lost and no costs are incurred. By contrast, Outcomes B, C, and F indicate decision errors (“false positives”) because evacuation costs are incurred even though no lives are lost. Conversely, Outcomes D, G, and H also are decision errors (“false negatives”) because lives are lost although no evacuation costs are incurred. The decision tree in Figure 3.5 has an equivalent representation as the outcome matrix in Table 3.5. Uncertainty can also complicate households’ decisions about how to respond to those warnings, as well as all of the other information that is available for a slow onset hazard (e.g., hurricane). Hence, households must face essentially the same decision problems as the evacuation managers, except that each household is only making the decision for itself rather than for the community as a whole.

3.3.2 PAR Timing The second issue in PAR decision making is the determination of when to issue PARs if a potential hazard has been detected before impact. The fundamental principle is that when evacuation is appropriate, it should be

62

Chapter 3 · Protective Actions and Decision Making

Table 3.5 Outcome Matrix for the Decision to Evacuate Major intensity

Moderate intensity

Miss

Evacuate

Correct decision

High cost false positive

High cost false positive

Shelter in-place

Low cost false negative Low cost false negative

Correct decision Low cost false negative

Low cost false positive Correct decision

No action

completed before the arrival of hazardous conditions, for example, before storm surge can flood evacuation routes or high wind can overturn motor homes and other high profile vehicles. However, emergency managers typically want to wait as long as possible to begin evacuations so they can avoid the expense and disruption of a false alarm. The evacuation decision deadline is the time beyond which it is no longer safe to delay an evacuation. For example, if a hurricane evacuation begins after the evacuation decision deadline, the evacuation will still be in progress and so some evacuees will still be in the Risk Area when Tropical Storm force wind arrives. As noted earlier, this is the wind speed at which an evacuation could be stalled if high profile vehicles such as buses and recreational vehicles are overturned by high wind striking them from the side. A criterion for deciding when to initiate an evacuation can be developed by recalling a fundamental idea of basic algebra, distance (d) is equal to rate (or speed, s) multiplied by time (t). In the case of an approaching hurricane, d (the hurricane’s distance from a coastal jurisdiction when an evacuation must be initiated) is equal to s (the forward movement speed of the hurricane along its track) multiplied by t (the ETE for an evacuation zone of the size defined by the hurricane’s Saffir-Simpson Category). The resulting value of d is called an evacuation decision arc that converts a measure of time (the ETE) into a measure of distance that can be seen on a hurricane tracking map (Cova et al. 2017). For example, the right side of Figure 3.5 shows an approaching hurricane whose eye is surrounded by a Radius of Tropical Storm Wind. As in Figure 3.2, this is the wind speed at which emergency managers want the evacuation zone cleared of all vehicles. The left side of the figure shows the hurricane Risk Area from Figure 3.2 with three arcs. A conventional evacuation decision arc is defined by the most likely values of the ETE for that Risk Area and the current value of the hurricane’s forward movement speed (the speed at which the hurricane center is moving over the water). However, both the Risk Area’s ETE and the hurricane’s forward movement speed are uncertain quantities, so it would be better to estimate the minimum probable radius, most probable radius, and maximum probable radius for the evacuation decision arc. The procedure for constructing these evacuation decision arcs can be illustrated using the hypothetical data in Table 3.6, which contains the

Chapter 3 · Protective Actions and Decision Making

63

Table 3.6 Example ETE Table Risk Areas to be Evacuated

Minimum Probable ETE

Most Probable ETE

Maximum Probable ETE

Risk Area 1 only

6

8

10

Risk Areas 1–2

8

10

12

Risk Areas 1–3

13

16

20

Risk Areas 1–4

18

22

27

Risk Areas 1–5

25

30

36

minimum probable, most probable, and maximum probable ETEs for the evacuation of each of the five risk areas corresponding to the areas that would be affected by each of the five Saffir-Simpson categories. As will be discussed more completely in Chapter 8, an ETE is a function of demand and capacity. If congestion is ignored (in fact, it is often a major determinant), the ETE for a given set of risk areas is determined by dividing the traffic demand (in vehicles) by the evacuation route capacity (in vehicles per hour) to produce the number of hours needed to compete the evacuation. This leads to two general principles. First, for a given population size, the larger the evacuation route system (ERS) capacity, the smaller the ETE. Second, for a given ERS capacity, the larger the population size, the larger the ETE. Although evacuation analyses typically produce a single ETE for a specific set of risk areas, there is uncertainty in the parameters that are used to calculate ETEs. For example, as will be discussed in Chapter 6, hurricane evacuation surveys have shown the average number of evacuating vehicles per household is 1.3 but this has varied from 1.1 to 2.2. As later chapters will indicate, there are also many other parameters in the estimation of ETEs that are also uncertain. However, if multiple simulations are run using different values of these uncertain parameters, and certain percentiles of the resulting ETE distributions are associated with each of the three points (e.g., minimum probable is the 15th percentile, most probable is the 50th percentile, and maximum probable is the 85th percentile), it will be possible to fill the cells in Table 3.6. Consider the example of a Category 4 hurricane with a minimum probable ETE of 18 hours, most probable ETE of 22 hours, and maximum probable ETE of 27 hours. In addition, assume that the hurricane’s current forward movement speed is 10 mph. These figures produce evacuation decision arcs of 180, 220, and 270 miles. That is, local officials who are extremely cautious about the time required to complete an evacuation should only wait until the Radius of Tropical Storm Wind is 270 miles from the coastline. Local officials who are moderately cautious can wait until the Radius of Tropical Storm Wind is 220 miles from the coastline and local officials who are quite optimistic can wait until the Radius of

Chapter 3 · Protective Actions and Decision Making

Tropical Storm Wind is 180 miles from the coastline. These distances can be displayed on a hurricane tracking map as evacuation decision arcs. As Figure 3.5 indicates, emergency managers can monitor the situation until the Radius of Tropical Storm Wind reaches the relevant evacuation decision arc. If they delay issuing an evacuation notice after that point, they might not be able to clear the evacuation zone before the arrival of Tropical Storm wind speed at the coast. To illustrate the use of these evacuation decision arcs, suppose that the hurricane center is 600 miles from the coast and the Radius of Tropical Storm Wind is 100 miles. Thus, the Radius of Tropical Storm Wind is 500 miles from the coast. If the local emergency managers have decided they will initiate an evacuation PAR at the maximum probable ETE (27 hours), that will be when the Radius of Tropical Storm Wind is 270 miles from the coast (again, assuming a forward movement speed of 10 mph). If the hurricane continues its forward movement at this speed, the emergency managers have 23 hours (i.e., (500 miles–270 miles/10 mph) to monitor the situation before they need to decide whether to issue an evacuation notice. Similarly, if they wait until the most probable ETE (22 hours), they can wait 28 hours before making a decision and if they wait until the minimum probable ETE (18 hours), they can wait 32 hours before making a decision. There would be the lowest likelihood of a false negative decision error (failing to evacuate for a hurricane that strikes) if they issued an evacuation notice at the maximum probable ETE (27 hours). Conversely, there would be the lowest likelihood of a false positive decision error (evacuating for a hurricane that missed) if they issued an evacuation PAR at the minimum probable ETE (18 hours). The advantage of waiting until the minimum probable ETE is that emergency managers can monitor the situation for another nine hours to determine if an evacuation is necessary; the disadvantage is that the evacuation must be executed flawlessly in order to clear the risk area if the hurricane does strike.

Figure 3.5 Evacuation Decision Arcs Evacuation decision arc radius = Storm forward movement speed x ETE

Risk Area

64

Minimum probable ETE Most probable ETE Maximum probable ETE

Radius of Tropical Storm Wind Predicted storm track Strike probability ≈ Hurricane track

Hurricane eye

Chapter 3 · Protective Actions and Decision Making

65

It is important to recognize that the timing of evacuation decision arcs provide only one aspect of deciding when to issue evacuation notices. The time of day when the messages are sent to the public is an important decision as they are more likely to receive the information and to take action during the day. As an extreme example, suppose that the maximum probable ETE occurs at 7:00 pm, the most probable ETE occurs at midnight, and the minimum probable ETE occurs at 4:00 am. In this case, most of the evacuation would take place in the dark—a situation that emergency managers would prefer to avoid if at all possible. These issues and others associated with ETE estimates and their utilization are discussed in the following chapters.

References Aldrich, D.C., Ericson, D.M., Johnson, J.D. 1978. Public Protection Strategies for Potential Nuclear Reactor Accidents: Sheltering Concepts With Existing Public and Private Structures. Sandia National Laboratories, Albuquerque, NM. Aldrich, D.C. et al. 1982. Technical Guidance for Siting Criteria Development. NUREG/CR-2239, SAND 81-1549. Sandia National Laboratories, Albuquerque, NM. American Institute of Architects. 1995. Buildings at Risk: Wind Design Basics for Practicing Architects. American Institute of Architects, Washington DC. Anno, G.H., Dore, M.A. 1978a. Protective Action Evaluation Part I – The Effectiveness of Sheltering as a Protective Action Against Nuclear Accidents Involving Gaseous Releases. EPA-520-1-78-001A. Environmental Protection Agency, Washington, DC. Anno, G.H., Dore, M.A. 1978b. Protective Action Evaluation Part II – The Effectiveness of Sheltering as a Protective Action Against Nuclear Accidents Involving Gaseous Releases. EPA-520-1-78-001B. Environmental Protection Agency, Washington, DC. Burson, Z.G., Profio, A.E. 1977. Structure shielding in reactor accidents. Health Physics Journal 33 (4), 287–299. Chan, W.R., Nazaroff, W.W., Price, P.N., Gadgil, A.J. 2007a. Effectiveness of urban shelter-in-place—I: idealized conditions. Atmospheric Environment 41 (23), 4962–4976. Chan, W.R., Nazaroff, W.W., Price, P.N., Gadgil, A.J. 2007b. Effectiveness of urban shelter-in-place—II: Residential districts. Atmospheric Environment 41 (33), 7082–7095. Clemen, R.T., Reilly, T. 2013. Making Hard Decisions with DecisionTools. SouthWestern Cengage Learning, Mason OH. Conklin, C., Edwards, J. 2000. Selection of protective action guides for nuclear incidents. Journal of Hazardous Materials 75 (2-3), 131–144. Cova, T.J., Dennison, P.E., Li, D., Drews, F.A., Siebeneck, L.K., Lindell, M.K. 2017. Warning triggers in environmental hazards: who should be warned to do what and when? Risk Analysis 37 (4), 601–611.

66

Chapter 3 · Protective Actions and Decision Making Cova, T.J., Drews, F.A., Siebeneck, L.K., Musters, A. 2009. Protective actions in wildfires: evacuate or shelter-in-place? Natural Hazards Review 10 (4), 151–162. Dosa, D., Hyer, K., Thomas, K., Swaminathan, S., Feng, Z., Brown, L., Mor, V. 2012. To evacuate or shelter in place: Implications of universal hurricane evacuation policies on nursing home residents. Journal of the American Medical Directors Association, 13 (2), 190-1e1. Hammond, G.D., Bier, V.M. 2015. Alternative evacuation strategies for nuclear power accidents. Reliability Engineering & System Safety, 135, 9–14. Institute for Business and Home Safety. 1997. Is Your Home Protected From Hurricane Disaster? Institute for Business and Home Safety, Boston MA. Lindell, M.K., Perry, R.W. 1992. Behavioral Foundations of Community Emergency Planning. Hemisphere Press, Washington DC. Lindell, M.K., Prater, C.S. 2007a. Critical behavioral assumptions in evacuation time estimate analysis for private vehicles: examples from hurricane research and planning. Journal of Urban Planning and Development 133 (1), 18–29. Lindell, M.K., Prater, C.S. 2007b. A hurricane evacuation management decision support system (EMDSS). Natural Hazards 40 (3), 627–634. Mannan, M.S., Kilpatrick, D.L. 2000. The pros and cons of shelter-in-place. Process Safety Progress 19 (4), 210–218. Rogers, G.O., Watson, A.P., Sorensen, J.H, Sharp, R.D., Carnes, S.A. 1990. Evaluating Protective Actions For Chemical Agent Emergencies, ORNL-6615, Oak Ridge National Laboratory, Oak Ridge, TN. Sorensen, J.H., Shumpert, B.L., Vogt, B.M. 2004. Planning for protective action decision making: evacuate or shelter-in-place. Journal of Hazardous Materials 109 (1), 1–11. USDOT—US Department of Transportation. 2006. Catastrophic Hurricane Evacuation Plan Evaluation: A Report to Congress. US Department of Transportation, Washington DC, accessed 6 June 2017. at www.fhwa.dot.gov/reports/ hurricanevacuation/rtc_chep_eval.pdf. USEPA—US Environmental Protection Agency 1987. Technical Guidance For Hazards Analysis: Emergency Planning For Extremely Hazardous Substances. US Environmental Protection Agency, Washington, DC. USOEP—US Office of Emergency Preparedness 1972. Report to Congress: Disaster Preparedness. US Government Printing Office, Washington, DC. Urbanik, T., Desrosiers, A., Lindell, M.K., Schuller, C.R. 1980. An Analysis of Techniques for Estimating Evacuation Times for Emergency Planning Zones, NUREG/CR-1745. US Nuclear Regulatory Commission, Washington, DC. Wilson, D.J. 1987. Stay indoors or evacuate to avoid exposure to toxic gas. Emergency Preparedness Digest (Canada), 14, 19–24. Wilson, D.J. 1989. Variation of indoor shelter effectiveness caused by air leakage variability of houses in Canada and the USA. In T. Glickman and A. Ujihara (Eds.) Proceedings of the conference on in-place protection during chemical emergencies. Resources for the Future, Washington DC.

Chapter 4

Who Leaves and Who Does Not

Evacuation participation rates, defined as the percentage of households or individuals in a given population who respond to a threat by leaving their homes to go someplace safer, vary greatly from one place to another in the same event and from one event to another in the same place. Understanding why those variations occur is important to minimizing undesired evacuation behavior—incomplete evacuation from areas or structures (e.g., mobile homes) that have been advised to evacuate (noncompliance) and unnecessary evacuation from areas that authorities have not advised to evacuate (evacuation shadow). In addition, this information is also needed as input assumptions for evacuation modeling and traffic management. First, evacuation analysts should assume that households will evacuate as a unit whenever this is possible. Indeed this assumption is correct because disaster researchers have long known that this is generally the case (Moore et al. 1963). Second, analysts often assume that everyone will evacuate from an area told by authorities to do so. That is, they assume a 100% participation rate from the official evacuation zone but this is rarely, if ever, the case (Lindell and Prater 2007). Although overestimating the participation rate might seem like a reasonable way to add conservatism to an evacuation analysis, excessive overestimates of participation rates can produce unrealistically high ETEs. Third, evacuation analysts often assume there is no evacuation outside the official evacuation zone, but this is also known to be false. As noted earlier, evacuation shadow (evacuation from areas not advised to do so) was estimated to be more than 10 times the number of people who were advised to evacuate during the 1979 Three Mile Island nuclear power plant accident (Lindell and Perry 1983). Moreover, a Texas evacuation planning study estimated 686,000 evacuees from Harris and Galveston counties in a Category 5 hurricane (Lindell et al. 2002b) but 2,500,000 evacuated during Hurricane Rita (Berg and Wilson 2013)—most from beyond the risk area for a Category 5 hurricane. Fortunately, few evacuations are characterized by such high levels of evacuation shadow because underestimates of evacuation shadow can produce unrealistically low ETEs.

68

Chapter 4 · Who Leaves and Who Does Not

Fourth, analysts sometimes assume that people will receive a warning and prepare to leave within a very short period of time. Although this is true for some incidents such as a hazardous materials release that is imminent or in progress, it is not necessarily true for other events such as flash floods (Lindell and Perry 1992). Especially in events with substantial forewarning such as hurricanes, hazard monitoring and evacuation preparations can take place over an extended period of time. To provide emergency managers and evacuation analysts with an accurate understanding of the percentage of the risk area population that will leave and the time that it takes for them to begin their evacuation, the next section presents a model of the way in which people typically make and implement protective action decisions. Section 4.1 turns to a discussion of typical levels of compliance with protective action recommendations and of shadow evacuation and Section 4.2 provides a discussion of the factors affecting household evacuation decisions.

4.1 Evacuation Behavior Who evacuates and when they evacuate can be readily explained by the Protective Action Decision Model (PADM—Lindell 2018; Lindell and Perry 1992, 2012, 2004), which begins with environmental cues, social cues, and warnings (Figure 4.1). Environmental cues are sights, smells, or sounds generated by an approaching threat and social cues are generated by observing the behavior of others. Warnings are messages transmitted from a source (an authority, news media, or peer) to a receiver via a channel (print, electronic, face to face), which result in effects that depend on receiver characteristics (Lasswell 1948; Johnson, Maio, and Smith-McLallen 2005). The effects are changes in receivers’ beliefs and behaviors; receivers’ characteristics include their psychological—cognitive (e.g., primary and secondary languages), psychomotor (e.g., vision and hearing), and physical (e.g., strength)—abilities and disabilities (Stough and Mayhorn 2013). Other receiver characteristics are economic (e.g., money and vehicles) and social/political (informal community networks and formal community organizations) resources. Environmental cues, social cues, and warnings trigger predecision processes—exposure, attention, and comprehension—that affect perceptions of the environmental threat, alternative protective actions, and relevant stakeholders—especially information sources. These core perceptions of threat, protective actions, and stakeholders are the bases for protective action decision making which, together with situational facilitators and impediments, produce a behavioral response. There are three basic categories of response—information search, protective response (problem-focused coping), and emotion-focused coping (actions that reduce negative emotions but provide no protection).

Chapter 4 · Who Leaves and Who Does Not

69

Information search provides a feedback loop as information is sought from additional environmental/social cues and warnings. The sequence in Figure 4.1 describes the way people typically make decisions about protective actions, but not everyone follows every step, let alone follows them in this sequence. Much of the evidence for understanding evacuation behavior has been obtained from post-impact surveys of people a few weeks to months after a disaster has occurred. These are usually called actual response surveys (or revealed preference surveys) because respondents are asked what they actually did in that situation. These questionnaires ask people whether and when they evacuated, as well as a number of possible predictor variables such as the ones identified by the PADM (see Huang, Lindell, and Prater 2016b) for a recent review of hurricane evacuation studies and Thompson, Garfin, and Silver 2017 for a review of evacuation studies for a broader range of hazards). In addition, there is also an increasing amount of evidence that comes from intended response studies (stated-preference studies) in which respondents are asked if they would evacuate in specific hypothetical threat scenarios. Some of these intended response studies are surveys of people in risk areas that are virtually identical to the post-impact actual response surveys; the principal difference is that people are asked how they expect to respond to a specific threat. Other intended response surveys are experiments in which respondents are given a set of scenarios that differ systematically from each other (a within-subjects design), or in which respondents are assigned to different groups, each of which is given a different scenario (a between-subjects design), or a combination of the two (a mixed design). Intended response studies have the advantage of being able to study people in locations where no actual threat has yet occurred or to study hypothetical situations that are rare in the real world. They can also systematically examine the relative impact of specific variables such as hurricane proximity, severity, and strike probability (Baker 1995). They can also be used to study the effects of different types of graphic, numeric, and verbal information on people’s information searches, hurricane strike probability judgments, and expected response actions (e.g., Wu, Lindell, and Prater 2015a, 2015b). Concerns have been expressed about the validity of both actual response and intended response studies. A concern about actual response surveys is that they are vulnerable to retrospective errors in which people might fail to remember weeks or months after an event what they were thinking and when they took certain actions during an emergency. A concern about intended-response surveys is that people might not respond to real threats the same way they do to hypothetical threats. Fortunately, it appears that intended-response surveys and actual-response surveys tend to identify the same variables as predictors of evacuation behavior, but possibly with differing effect sizes (see the discussion in Huang, Lindell, and Prater 2016b). That is, an intended-response survey might find that a predictor

From Lindell 2017

Social/ environmental context

Warning messages

Information channels

Information sources

Social cues

Environmental cues

Stakeholder perceptions

Protective action perceptions

Threat perceptions

Protective action decision making

Information search strategy

Personal characteristics • Physical/psychological, material, social/ political, and economic resources • Past experience • Demographic attributes

Pre-decision processes • Exposure • Attention • Comprehension

Figure 4.1 Protective Action Decision Model

Situational impediments

Behavioral response • Information search • Protective response • Emotion-focused coping

Situational facilitators

70 Chapter 4 · Who Leaves and Who Does Not

Chapter 4 · Who Leaves and Who Does Not

71

has a 25% stronger effect on evacuation expectations than on actual evacuation. Less established is how well intended-response surveys predict actual response in later incidents. There are instances of both good and poor matches, with Kang, Lindell, and Prater (2007) finding that the correspondence between expected and actual behavior was best for activities that were most similar to those taken before routine vacations (e.g., packing bags) but was much poorer for novel activities such as protecting the house from storm damage). However, too few comparisons have been conducted to document the conditions under which the matches will be best. It does appear that the more that a hypothetical threat scenario description corresponds to the information that is available in an actual incident (i.e., the variables identified in the PADM), the better the intended response will match the actual response. But at least one study concluded that although people’s intentions to evacuate in high threat situations tended to match actual responses fairly well, people were significantly more likely to say they would evacuate in low threat situations than is typically observed in actual evacuations (Baker 1995). One frequent problem is that intended-response scenarios rarely provide the variety of information that is present in actual threat situations. For example, the very nature of these experiments makes the risk of staying more salient than the cost and inconvenience of evacuating. On the other hand, some self reports assessed months or even years after the evacuation occurred might be limited by people’s ability to recall what they heard, what they did, and why they did it. However, Lindell et al. (2016) reviewed some studies that indicate respondents might actually be able to provide reliable reports even if a survey is conducted months later. More generally, Schwarz (2007) concluded that people are able to provide accurate answers when asked questions about behaviors that are rare and important—two characteristics of disaster responses. Of course, as Kang Lindell, and Prater (2007) showed, there are significant differences among different disaster responses in their rarity and importance. For example, people are more likely to be able to recall accurately if they evacuated than what they packed before leaving. In addition, there are concerns about the accuracy of people’s reports of the time at which they took different actions. For example, they are probably more accurate in their reports of the date and hour at which they evacuated than in their reports of the number of minutes it took them to pack their bags for the evacuation or to get gas, food, and cash for the trip before leaving. To address the limitations of most intended-response and actualresponse surveys, a few studies—called real-time surveys—have interviewed people in areas under threat at the time of the interview (Meyer et al. 2014). Real-time surveys have the advantage of asking respondents about current physical and social contexts, their sources of information, their awareness of official evacuation notices, their perceptions of the threat, the actions they have taken to that point, and what

72

Chapter 4 · Who Leaves and Who Does Not

they intend to do in response to different scenarios. Repeated contacts with the same respondents or repeated sampling of the same population can then provide concurrent data (not vulnerable to retrospective errors) about the ways in which people’s responses evolve over time in an actual incident. Finally, researchers can also collect evacuation data by monitoring and recording devices such as traffic counters (Li et al. 2013). Although such devices are limited in their ability to provide insights into the reasons why people are evacuating at that time, the data can be correlated with variables such as time of day, location of the counter, and the issuance of evacuation notices by public officials. Wilmot (2004) has suggested that other mechanisms such as GPS devices and cell phone towers have similar potential. Most of the research typically cited for understanding evacuation behavior comes from peer reviewed literature in academic journals, and this chapter relies heavily on those sources. However, there are many methodologically sound evacuation studies conducted to support evacuation planning activities whose data do not find their way into journals. They lack the imprimatur of peer review, but they are often conducted by the same individuals who contribute to journals and they sometimes employ data that cannot be found in journal articles. Accordingly, the findings from the highest quality technical reports are also incorporated into this summary. Although Chapter 2 discussed numerous hazards for which evacuation is an appropriate protective action, the evidence concerning evacuation behavior is far from uniform among hazards. Hurricanes have received the most attention by a substantial margin. Huang, Lindell, and Prater (2016b) identified 38 actual-response and 11 intendedresponse studies dealing with hurricane evacuation behavior in peerreviewed journals. Thompson, Garfin, and Silver (2017) reviewed 83 peer-reviewed articles dealing with evacuation behavior from natural hazards in general, and 59 of them dealt with hurricanes or cyclones. This chapter draws heavily on hurricane studies but also includes examples from other hazards when it is available. At some level, findings tend to generalize across hazards. That is, variables found to predict evacuation for one hazard tend to predict evacuation for other hazards as well. However, the relative magnitude of influence of the variables might vary among hazards, and overall evacuation rates vary among hazards.

4.1.1 Compliance with Official Evacuation Notices Evacuation takes time to implement, so authorities must begin to evacuate the risk area well before the expected arrival of hazardous conditions. As noted in Section 3.3.2, the uncertainties in predicting the area that will actually be affected by the hazard will usually require the

Chapter 4 · Who Leaves and Who Does Not

73

evacuation of a larger area to ensure that it includes the eventual impact area. For some hazards, most notably hydrological hazards such as hurricanes, river flooding, and tsunamis, the potential risk areas can be identified and publicized well in advance of threats—although the specific portions of the areas needing to evacuate will be event specific. Similarly, risk areas for fixed site facilities such as chemical plants (USEPA 1987) and nuclear power plants (USNRC/FEMA 1980) can also be identified. However, risk areas for tornadoes and wildfires depend more on local conditions at the time of the event, so risk areas for these events are difficult to assess in advance—although rough estimates of risk areas for hazardous materials transportation incidents can often be made based on the types of materials being transported and the distance from transportation routes (recall Section 2.5.2). Whether the risk area that needs to be evacuated has been identified before the incident occurs or during the incident, scientific evidence indicates that people in areas that authorities have designated as being at greatest risk tend to evacuate at higher rates than people in areas of less risk. The most extensive evidence regarding variation in evacuation rates by risk area comes from hurricane evacuations. Well in advance of a threatening hurricane, the SLOSH model analyses will indicate areas that are expected to experience dangerous conditions. As indicated in Figure 2.5, state and local officials categorize the results into a manageable number of planning scenarios and adapt the areas identified in the computer analyses to conform to geographical boundaries such as water bodies and streets, as well as administrative boundaries such as county lines and ZIP Code zones. Those areas are then publicized as evacuation zones. In some locations, the evacuation zones are complicated by geographical features such as rivers, bays, and estuaries behind barrier islands, as well as by coastal protection structures such as the levees around New Orleans, Louisiana and the seawall in Galveston, Texas. But overall, the storm surge risk tends to be greatest in areas closest to the shoreline and diminishes inland at a rate depending on the rise in elevation. The wind risk is generally highest near the coast and also decreases with distance, although tornadoes can wreak havoc many miles inland. In general, residents in site-built homes are told to evacuate only because of risk from storm surge and inland flooding, but people living in mobile homes are told to evacuate even if their area is threatened exclusively by high wind. In the Tampa Bay region of Florida, evacuation zones are labeled A through E, with A being the most vulnerable area closest to the coastline. In 2004, when the region evacuated for Category 4 Hurricane Charley, the evacuation rates were 46% in Zone A, 30% in Zone B, 23% in Zone C, 21% in Zone D, 13% in Zone E, and 15% inland of surge-prone areas (Tampa Bay Regional Planning Council 2010). New York City has six zones, labeled 1-6, with Zone 1 being the most vulnerable. In Superstorm Sandy in 2012 the evacuation rates were

74

Chapter 4 · Who Leaves and Who Does Not

39% in Zone 1, 21% in Zone 2, 12% in Zone 3, 8% in Zone 4, 9% in Zone 5, 6% in Zone 6, and 5% inland of Zone 6 (Baker 2014). In many studies, sample sizes are not sufficient to provide statistically reliable estimates for every evacuation zone, so zones are often aggregated for analysis, but the pattern is the same in virtually every storm— people are more likely to leave from the more vulnerable zones. However, as in the two examples cited here, not everyone complies with official evacuation notices. More than half failed to do so from Zone 1 in New York City in Sandy and from Zone A in Tampa Bay in Charley. Nevertheless, location—as indicated by evacuation zone, surge inundation zone, or surrogates for those measures—is typically one of the strongest predictors of whether people evacuate in hurricanes (Huang, Lindell, and Prater 2016b). The same general tendency appears to be true for other hazards as well, although the number of studies is notably smaller. In the 1979 Three Mile Island nuclear power plant accident, the governor’s recommendation for an evacuation of pregnant women and preschool children within 5 miles of the plant produced household evacuation rates of 55% within 5 miles, 45% from 5–10 miles, 34% from 10–15 miles, 12% from 15–20 miles, and 5% from 20–40 miles (Flynn 1979). A number of intended response surveys around nuclear power plants have shown a similar distance decay pattern (e.g., Zeigler and Johnson 1984). In the Graniteville, South Carolina train derailment and chlorine leak, 98% of households within a mile of the incident evacuated, compared to 59% 1–2 miles from the incident scene (Mitchell, Cutter, and Edmonds 2007). Risk area is usually correlated with other factors that might help explain evacuation behavior. Higher risk areas are more likely to receive evacuation notices from public officials, for example. Also, residents of areas closest to a hazard source are more likely to at least be aware of the potential for hazard occurrence even before an incident occurs (Lindell and Hwang 2008). But for hurricanes, when researchers statistically control for whether interviewees believed they were told by officials to evacuate and for their perceived vulnerability in the event of a hurricane strike, risk area is still one of the strongest predictors of evacuation (Baker 2000a).

4.1.2 Shadow Evacuation Related to location within a risk area is a concept known as shadow evacuation. This refers to people evacuating from areas not included in official evacuation notices, typically farther from the hazard impact area. This issue first came to prominence following the 1979 Three Mile Island nuclear power plant evacuation when, as noted in the previous section, there were substantial evacuation rates beyond the

Chapter 4 · Who Leaves and Who Does Not

75

evacuation zone recommended by authorities (Flynn 1979; Zeigler, Brunn, and Johnson Jr. 1981; Cutter and Barnes 1982). Once researchers became aware of shadow evacuation, they soon found it in other disasters such as the eruption of Mt. St. Helens (Perry and Greene 1983). Shadow evacuation has become a contentious issue with regard to nuclear power plants because it can significantly increase the time required to complete an evacuation safely, possibly requiring too much time for all those at risk to avoid radiation exposure. Worries about shadow evacuation around the proposed Shoreham nuclear power plant on Long Island, New York contributed to public and local government opposition to the plant, resulting in operation of the facility being cancelled after construction was complete (Ross and Shaw 1993). Evacuation plans were nonexistent at the Three Mile Island nuclear power plant before the accident occurred in 1979 and had to be improvised during the incident (Chenault, Hilbert, and Reichlin 1980). However, procedures for messaging and official announcements for evacuation were substantially improved in the aftermath of the event (Lindell et al. 1985). Nevertheless, a number of intended-response surveys conducted subsequent to Three Mile Island yielded patterns similar to the shadow evacuation observed in that incident. When interviewees have been presented with hypothetical threat scenarios, including evacuation notices issued by public officials, many residents living beyond areas included in the evacuation notice say they would evacuate (Zeigler and Johnson 1984). A large recent survey (n = 821) conducted in the EPZs of 63 nuclear facilities nationwide showed mixed results with 75% of respondents saying they would probably shelter inplace rather than evacuate if instructed to do so, but in a separate question almost 60% said they would evacuate and not follow instructions to shelter in-place (Walton and Wolshon 2010). Of the total sample, 114 people said they had been asked to evacuate due to a natural disaster or industrial accident in the past, and 75% of those said they did evacuate. Of the 114, 23% said they had evacuated without being asked by officials to do so. Eighty-two people (of the 821) said they had been asked to shelter in-place during an emergency, and 79% of those said they complied. There is evidence of shadow evacuation in toxic chemical releases also. Following a 1979 train derailment in Mississauga Ontario, evacuation was recommended from an area that was 2.5–5 miles downwind from the incident site. Nonetheless, 60% of the residents in crosswind and upwind areas where evacuation was not recommended did so anyway (Burton et al. 1981). An accident at a pesticide and herbicide repackaging plant in West Helena, AR prompted an evacuation in 1997. People living within 2 miles downwind of the plant were told to evacuate, but those living 2–3 miles away were told to shelter inplace. Within the 2-mile zone, 90% left, but 68% evacuated in the 2–3 mile shelter in-place zone (Vogt and Sorensen 1999). In the Graniteville,

76

Chapter 4 · Who Leaves and Who Does Not

SC incident referenced earlier, mandatory evacuation orders were only issued for the area within one mile of the accident site. The 59% who evacuated in the 1-2 mile area were shadow evacuees according to the authors (Mitchell, Cutter, and Edmonds 2007). Some of the evacuees from the shadow zone said they weren’t sure if they were within the 1-mile radius or not. (In an affidavit taken as part of a lawsuit, the local sheriff stated that areas beyond the 1-mile zone were also evacuated, but allowed to return home earlier than others (Hunt 2005)). Shadow evacuation also occurs in hurricanes and has probably contributed to unexpectedly lengthy evacuation times for evacuees. In the Tampa Bay example cited earlier in Hurricane Charley, two of the four counties ordered evacuation in Zones A, B, and C as well as for mobile and manufactured homes. Another county ordered evacuation in Zones A and B, plus mobile and manufactured homes, and the fourth ordered evacuation only in Zone A, plus mobile and manufactured homes. Residents living in site-built homes who evacuated from Zones D and E and from areas inland of E were shadow evacuees, as well as some of whom left from Zones B and C. Of the households in Zone D, 16% left, along with 7% from Zone E, and 7% from non-surge areas (Tampa Bay Regional Planning Council 2010). As noted earlier, in Hurricane Rita, approximately 2,500,000 people evacuated for a Category 5 hurricane that Lindell, Prater, and Wu (2003) had earlier forecast would produce 685,000 evacuees. More than 30% evacuated from areas inland from the official evacuation zones in the Houston-Galveston vicinity, probably because local officials warned people to leave if they had ever been flooded in the past (Peacock et al. 2007). In fact, many parts of this area had been flooded during Tropical Storm Allison during 2001. In Hurricane Floyd in 1999, a massive evacuation from south Florida through the Outer Banks of North Carolina saw 32% of site-built home residents leave from non-surge areas of coastal counties and 21% leave from adjacent inland (non-coastal) counties (Baker 2009). Many evacuees in Rita and Floyd encountered major road congestion and longer evacuation times than anticipated. For example, Wu, Lindell, and Prater (2012a) reported that Rita evacuees took nearly seven hours longer than normal to reach their evacuation destinations.

4.2 Predictors of Evacuation 4.2.1 Perceived Risk It would make sense that people should be more likely to evacuate if they believe they are in danger from a hazard and evacuation will reduce that danger. Unsurprisingly, that is the case for every hazard for which the issue has been studied (e.g., Thompson, Garfin, and Silver

Chapter 4 · Who Leaves and Who Does Not

77

2017). There are at least two components to perceived risk—belief that one will experience the hazard event and belief that one will be adversely affected if it occurs. That is, the belief that storm conditions will reach your location and that those storm conditions will cause casualties (death or injury) or property damage from wind, surge, or inland flooding (see Huang, Lindell, and Prater 2016b). For example, Perry, Lindell, and Greene (1981) reported results of surveys in four communities that experienced riverine flooding in the western United States. If residents heard that water was rising or that flooding was creating possible danger, 36% evacuated. If they heard that flooding was occurring and water was approaching their location, 66% evacuated. Of those respondents who totally or somewhat disbelieved warnings that their homes would be flooded, 16% evacuated, compared to 59% who largely or totally believed the flood warnings. Of those who believed their personal risk from flooding was slight, 24% evacuated versus 87% who said their personal risk was severe. Hurricane evacuation studies provide especially rich insights into the intricacies of risk perception and its relationship to evacuation behavior. Many post-hurricane surveys simply ask respondents to state the reasons they did or did not evacuate. Routinely, most people cite reasons related to perceived safety rather than constraints such as lack of time or resources or obstacles such as disabilities or pets. That is, people who evacuate tend to say they did not feel safe; people who do not evacuate tend to say they did feel safe. For example, in Connecticut in Superstorm Sandy, the great majority of those who evacuated cited concern about storm surge and waves, wind, or river flooding. The great majority of those who did not evacuate said their home would not flood or was well built, the storm was not strong enough to pose a danger to their location, or the storm would miss (USACE New England District 2016). The role of evacuation notices from public officials will be discussed below, but their presence or absence was also cited frequently as a reason for evacuating or not. Many residents take the threat more seriously if public safety officials have judged the threat to be great enough to issue a formal evacuation notice. Interviews conducted with coastal residents in “real-time” as hurricanes Earl, Irene, Isaac, and Sandy approached provide some interesting clues to understanding perceived risk. Respondents believed that they were more likely to experience hurricane-force winds from the storms than the actual probability calculated by the National Hurricane Center. However, they believed the likelihood of experiencing damaging or dangerous wind was much lower. They perceived the likelihood of experiencing damaging or dangerous flooding from storm surge and waves to be even lower. Most respondents said the storm would pose no danger to their safety even if the storm struck their location. This was true even for people living within a block of water bodies (Meyer et al. 2014).

Chapter 4 · Who Leaves and Who Does Not

For hurricanes, there is consistent evidence that too many people in dangerous locations underestimate their vulnerability whereas too many people in relatively safe locations overestimate their vulnerability. Figure 4.2 shows data from a survey of New York City residents conducted following Hurricane Irene (Baker 2013). They were asked if it would be safe to stay in their homes, considering both wind and water, if storms of three different categories struck their location directly (wind velocities were given as well as storm category). At the time, the zones were approximately correlated with Saffir-Simpson wind scale categories. Zone A would need to evacuate in most Category 1 hurricanes, Zone B in most Category 2 hurricanes, and Zone C in most Category 3 or 4 hurricanes. (The number of zones has since been increased to six to account for variations within storm categories resulting from different storm tracks.) Respondents felt less safe in stronger storms, which is appropriate. But the lines in Figure 4.2 should be higher than they are on the left side of the graph and not as high as they are on the right side. For example, only 65% of the people living in the most vulnerable zone (A) said they would be unsafe in a Category 3 hurricane with winds of 125 mph. This is the zone that would have been told to evacuate for most

Figure 4.2 Percent of Respondents in New York City Following Hurricane Irene Saying It Would Be Unsafe to Stay at Home If Struck By a Category 1, 2, or 3 Hurricane, by Evacuation Zone (NS = Non-surge) 100 Percent Saying they Would be Unsafe

78

90

Cat 1

80

Cat 2

70

Cat 3

60 50 40 30 20 10 0 Zone A

Zone B

Zone C

Evacuation Zone

From Baker 2013

NS

Chapter 4 · Who Leaves and Who Does Not

79

Category 1 storms with winds under 96 mph. At the same time, in the non-surge area inland of all evacuation zones, almost 20% said their homes would be unsafe in a Category 1 storm, 30% in a Category 2, and 50% in a Category 3. These beliefs contribute to too few leaving from the more dangerous areas and too many leaving from safe areas. These results are not unique to New York City residents, who have had limited exposure to tropical cyclones. Lindell et al. (2001) found similar results in their survey of residents in Texas coastal counties (see Lindell and Prater (2007), for a table reporting results similar to those in Figure 4.2). People tend to be more concerned about danger from hurricane wind than from storm surge or inland flooding, although hurricane flood hazards account for far more deaths and are the basis for delineating hurricane evacuation zones. In real-time interviews, even people living less than a block from water bodies said they were more concerned about wind, and that concern probably contributes to people’s beliefs about the vulnerability of those who live inland from the surge zones and hence will be shadow evacuees (Meyer et al. 2014). This result also generalizes across storms and geographical areas; surveys after Katrina and Rita found that people’s expectations of injury or death from these storms was more strongly correlated with their perceptions of wind risk than with perceptions of surge or inland flood risk (Huang, Lindell, and Prater 2016a). One reason for the failure of people to appreciate their danger from storm surge is that they do not know the elevation of their home. A study in Florida found that only 14% of residents in coastal counties were able to specify the elevation of their building site correctly even when they were given five-foot ranges from which to choose; a majority refused to even guess their home’s elevation (Baker and Zhao 2012). This study has not been replicated elsewhere, but there is no reason to believe that other locations would fare any better. For example, most coastal respondents in New England said they did not know their elevation (USACE New England District 2016). If people are made aware of the National Hurricane Center’s storm surge height forecast when it is included in warnings, few would be able to make good use of that forecast unless they have accurate information about their home’s elevation. A new product issued by the National Hurricane Center might improve this situation. The agency produces a map showing how deep the flooding could be in different parts of a community. A survey in Florida in 2016 indicated that people who were told they would experience 3-6 feet of flooding were more likely to say they would evacuate than people who were told they would experience winds of 125 mph (Baker, Downs, and St. Germaine 2016). Of course, providing people with risk area maps presumes that they are able to identify their evacuation zone when given a map. Unfortunately, there is a consistent finding that, despite the widespread

80

Chapter 4 · Who Leaves and Who Does Not

distribution of printed brochures containing evacuation zone maps, websites with evacuation zone maps, and websites where residents can enter their address and have their evacuation zone indicated, many people are unable to identify their evacuation zones. When public officials issue an evacuation notice for Zones A and B, few people living in those zones know whether it is relevant to them. In two different studies, Texas researchers mailed evacuation zone maps to coastal residents and asked them to identify their evacuation zone, while referring to the map. In the first study, one third of the respondents from the lower Texas coast from Brownsville to Corpus Christi were unable to correctly identify their evacuation zones (Zhang, Prater, and Lindell 2004). In the second study, which sampled the entire Texas coast, two thirds of the respondents were unable to correctly identify their evacuation zone (Arlikatti et al. 2006). Moreover, respondents’ accuracy in identifying their risk area was uncorrelated with expectations of evacuating from a hurricane. There was no apparent relationship of respondents’ risk area accuracy with definition of risk areas by political boundaries (city or county boundaries) or geographical features (rivers, bays, topographical elevation). One explanation for the low level of risk area accuracy is that variation in elevation causes risk areas to appear as irregular polygons on the maps, which are very narrow when elevation changes rapidly over a short distance (Arlikatti et al. 2006). Whatever the contributing factors, perceived vulnerability is a strong, consistent predictor of evacuation. Table 4.1 shows an example from Hurricane Floyd in 1999, in communities from south Florida through North Carolina (Baker 2000a). Following the hurricane, residents were asked if it would be safe to stay if their homes if a hurricane with a peak wind speed of 125 mph (a Category 3 storm) were to strike their location directly. In each of the four risk area groups, those who said their homes would be unsafe were significantly more likely to have evacuated in Floyd.

4.2.2 Forecasts and Warnings As noted previously, information sources (authorities, news media, and peers) can transmit forecasts and warnings through a variety of different

Table 4.1 Evacuation Rates in Hurricane Floyd by Risk Area and Perceived Vulnerability in a 125 mph Storm Category 1 Surge Zone

Other Surge Zones

Non-surge Coastal

Non-coastal

Safe

42

35

19

17

Unsafe

79

62

55

38

From Baker 2000a

Chapter 4 · Who Leaves and Who Does Not

81

channels (print, electronic, face to face). Researchers have found that warning messages are most effective if they indicate the message source (and, thus, the credibility of the rest of the message), as well as include information about identity of the threat, affected (and safe) areas, protective action recommendations, implementation deadline, and sources to contact for additional information and official assistance (Lindell and Perry 2004, Chapter 5; Mileti and Sorensen 1987). Warning messages providing information that is specific (contains details about message elements), consistent (among sources at a given time and across time for a given source), certain, clear, and accurate (Bean et al. 2015; Mileti and Peek 2000) tend to produce more accurate situational risk perceptions about likely casualties, damage, and disruption to the community in general and to one’s family in particular (Huang et al. 2012). As noted in Chapter 2, it is important to recognize that the term warning has a very specific meaning to public agencies charged with informing the public about threatening hazards. Many agencies issue a Warning when a hazardous event is in progress or imminent. This is more serious than other conditions such as a Watch, an Advisory, or an Alert. These notices might be classified as warnings with a capital “W” to distinguish their status as official judgments by physical scientists that usually go beyond making forecasts or predictions about an event. Intended-response experiments, which permit the separation of Warnings from other factors, have found that people are more likely to say they would evacuate given a Hurricane Warning than given a Hurricane Watch and more likely given a Hurricane Watch than if neither a Watch nor a Warning has been issued (e.g., Baker 1995; Petrolia and Bhattacharjee 2010). Moreover, post-hurricane surveys find that some, but not many, respondents cite federally issued official Warnings (as opposed to locally issued official evacuation notices) as their reason for evacuating and others cite the absence of such Warnings as their reason for not evacuating (USACE New England District 2016). Nonetheless, communicating the distinction between Warnings and Watches is sometimes problematic for the federal hazard agencies. For example, when Superstorm Sandy threatened the US Atlantic coast in 2012, the storm ceased to satisfy the National Hurricane Center’s technical definition of a hurricane after a certain point, so the agency was unable to issue a Hurricane Warning. Instead, local National Weather Service agencies issued various notices such as Hurricane High Wind Warnings and Coastal Flood Warnings. The distinction was missed by the public, most of whom thought the National Hurricane Center was issuing Hurricane Watches and Warnings as usual (Meyer et al. 2014). People’s inability to distinguish between Watches and Warnings is not limited to Superstorm Sandy. Florida surveys have found that few people knew the correct lead times (hours before possible onset of Tropical Storm force wind conditions) for Hurricane Watches and Warnings, either pre-2010 or post-2010, when the lead times were

82

Chapter 4 · Who Leaves and Who Does Not

extended to 48 hours for a Hurricane Watch and 36 hours for a Hurricane Warning (Baker 2010a). This finding might explain why graphs of evacuees’ departure times show little or no change upon issuance of Watches or Warnings alone (USACE Philadelphia District 1996). The effect of Warnings on evacuation decisions is difficult to isolate because official Warnings from federal government hazard detection agencies (and preceding notices such as Watches) trigger a number of other activities. Local agencies are more likely to initiate public safety activities such as issuing evacuation notices, the media normally give the event more attention, and environmental cues might be more pronounced. Clearly, most people take storm severity into account, at least as measured by wind speed, and the storm’s proximity to their location is also a factor in evacuation decisions. These variables are correlated with actual response, intended response, and reasons given for evacuating or not in actual threats, but other storm characteristics are less clear in their effects. Seeing a forecast graphic depicting a narrow line indicating the hurricane’s forecast track has been shown not to deter evacuation from areas farther from the line, when shown along with an “error cone” graphic encompassing a larger area where the storm’s center might go (Meyer, Broad, and Petrovic 2013). Indeed, Wu et al. (2014) found that there were no significant differences among experiment participants who were shown a forecast track only, an uncertainty cone only, or a forecast track plus uncertainty cone. In all three conditions, participants provided non-zero strike probability judgments for locations far away from the forecast track—including locations in the opposite direction of the forecast track. Moreover, Cox, House and Lindell (2013) found no differences between an uncertainty cone display and a track ensemble display that presents a set of hypothetical tracks. However, Ruginski et al. (2016) found small differences among five display conditions (track only, cone only, track/cone, fuzzy cone, and track ensemble), but all five elicited a pattern of declining damage judgments with distance from the track centerline. Most recently, a study of dynamic decision making examined information search patterns of participants tracking approaching hurricanes (Wu, Lindell, and Prater 2015a), as well as their strike probability judgments for six cities around the Gulf of Mexico and their protective action recommendations for the jurisdiction to which they were assigned (Wu, Lindell, and Prater 2015b). Participants were allowed to select the numeric (hurricane parameter table), graphic (hurricane tracking map), and verbal (NHC watches and warnings) display elements they wished to view during each of six successive forecast advisories about four different hypothetical hurricanes. The information search data revealed a mixed pattern of preference for forecast tracks and uncertainty cones; participants viewed forecast tracks more frequently than uncertainty cones but spent more time viewing uncertainty cones than forecast tracks. Strike probability judgments for six cities around the Gulf of Mexico revealed a pattern

Chapter 4 · Who Leaves and Who Does Not

83

similar to that in Wu et al. (2014) and Ruginski et al. (2016)—a steady decline with a city’s distance from the forecast track. Most people do not know how accurate the forecasts for storm location or intensity are, but beliefs about those forecast errors have been uncorrelated with evacuation (Baker 2004). In real-time interviews, 55% of the respondents were able to correctly identify Hurricane Earl’s current wind category, but only 30% were able to correctly state its forecast wind category when it was expected to reach its closest point of approach to their location (Baker, Broad, and Meyer 2011). Most people said they did not know how high the storm surge was forecast to be for their location, and only 16% in Massachusetts and 21% in North Carolina could specify it correctly. As noted earlier, people tend to overstate their probability of experiencing hurricane force winds, compared to values provided by the National Hurricane Center along with forecasts.

4.2.3 Official Evacuation Notices Authorities’ communications that contain evacuation as a PAR have been given a number of different labels—such as “evacuation advisories”, “evacuation warnings”, “voluntary evacuations”, “evacuation recommendations”, “evacuation orders”, and “mandatory evacuations”—that vary in their implied degree of compulsion. Thus, to minimize problems of interpretation, this section refers to all of these communications by the relatively neutral term “notices”. Beyond simply indicating to the public that a wildfire, flood, or volcano poses a threat, public officials can convey their assessment of the threat level by explicitly recommending—and sometimes even ordering—the evacuation of people in a certain geographical area (e.g., people within 1 mile of a chemical release) or who have certain personal characteristics (e.g., pregnant women and preschool children). Political jurisdictions differ in their authorities regarding the issuance of evacuation notices (Wolshon et al. 2005). Public officials can universally recommend that people evacuate from areas at risk, but the authority to forcibly compel evacuation varies. In some states, only the Governor can make evacuation mandatory, whereas in others both the Governor and local elected officials can do so. In general, the research literature yields several consistent findings regarding evacuation notices. ■ Not everyone in designated evacuation zones is aware that they have been told to evacuate and others outside those evacuation zones believe they have been told to evacuate (Baker 2000). ■ In large scale evacuations such as hurricanes, authorities generally lack enough personnel to deliver door to door notices—let alone enforce

84

Chapter 4 · Who Leaves and Who Does Not

mandatory orders by arresting, transporting, and arraigning people who do not comply (Perry and Lindell 1980). ■ People are more likely to leave if they believe they have been told by officials to evacuate; indeed, this is the strongest predictor of evacuation decisions (Huang, Lindell, and Prater 2016b). ■ People are more likely to leave if they believe the evacuation is mandatory rather than recommended (Baker 1991). ■ For most events, the majority of people do not begin to evacuate until officials issue evacuation notices, but a minority—sometimes a substantial minority—evacuates early (Baker 2000b). Officials are more likely to issue evacuation notices for people living in more hazardous locations, so it is important to adjust for risk area when assessing the impact of official evacuation notices. Table 4.2 shows evacuation rates in Hurricane Floyd in communities ranging from south Florida through North Carolina in 1999 (Baker 2000a). It indicates the percentage evacuating in each of four zones, depending on whether the respondents said they heard from officials that it was mandatory that they evacuate, that it was recommended that they evacuate, or they did not hear any official evacuation notices that applied to themselves. Hearing evacuation notices had a substantial effect on evacuation rate in all four zones, and believing that evacuation was mandatory had a greater effect than believing that it was recommended. Actual evacuation notices varied among survey communities both in terms of whether they were mandatory as well as the zones to which they applied, but the data in Table 4.2 is based on what respondents said they heard. Some people in areas being told to leave said they did not hear the notice, either directly from authorities or indirectly through peers. But, some people in areas not told to evacuate thought they had been told to leave, particularly in the nonsurge areas of coastal counties and in the non-coastal counties. (Although mobile homes were told to evacuate in non-surge areas, and possibly in non-coastal counties, they constituted a small percentage of total households surveyed and do not account for the patterns inland of surge zones shown in Table 4.2.)

Table 4.2 Evacuation Rates in Hurricane Floyd by Risk Area and Type of Official Evacuation Notice Heard Category 1 Surge Zone

Other Surge Zones

Non-surge Coastal

Non-coastal

Heard Must

87

84

75

71

Heard Should

60

58

58

52

Heard Neither

48

33

22

17

From Baker 2000a

Chapter 4 · Who Leaves and Who Does Not

85

In a statistical analysis of the Floyd data that adjusted for five types of variables (risk area, perceived vulnerability, housing type, political jurisdiction, and demographics), people who said they heard evacuation notices were 19 percentage points more likely to evacuate than people who said they heard no evacuation notices. People who said they heard mandatory evacuation orders were 44 percentage points more likely to leave (Baker 2000a). Similarly, Huang et al. (2012) found that official warnings had a significant effect on evacuation from Hurricane Ike when controlling for risk area, expected personal impacts, social cues, “unnecessary” evacuation experience, and expected evacuation impediments. Moreover, Huang, Lindell, and Prater (2016a) found that official warnings had a significant effect on evacuation from Hurricanes Katrina and Rita when controlling for risk area, expected rapid onset, expected wind impacts, and expected evacuation impediments. Misconceptions about whether evacuation notices are mandatory might result from hearing them second hand but sometimes agencies use confusing wording. Some states and local governments use “voluntary” rather than “recommended” but might combine it with the word “order” which would seem to imply something less than a suggestion. During Hurricane Isaac, one county in Florida issued an “order of evacuation;” it was not intended to be mandatory, but residents could be excused for not understanding that. Most jurisdictions say they would not actually enforce a mandatory evacuation order (Wolshon 2009) but whether residents know that or not, the mandatory description probably adds to the gravity of the message. If fact, even if a jurisdiction lacks the authority to issue a mandatory evacuation order, they can make it sound as if people do not have a choice by using expressions like “the area is being evacuated” or “an evacuation is in effect for. . ..” Using a term like “voluntary” invites noncompliance. Some people hear an evacuation notice but do not understand that the notice applies to them. As noted earlier, many if not most people in hurricane areas do not know their evacuation zone, so the effectiveness of evacuation notices can be limited when they are issued for specific zones. This is also true for so called “no-notice” evacuations in those instances when evacuation notices are issued, but without the notifications or possibly even environmental cues that precede other evacuations. In the Graniteville, SC train derailment evacuation some residents said they did not know if they were in the 1-mile mandatory evacuation zone (Mitchell, Cutter, and Edmonds 2007). This lack of risk area awareness accentuates the importance of not just issuing an evacuation notice but disseminating it effectively (for a discussion of alternative warning technologies, see Lindell and Perry 1992; Lindell and Prater 2010). Door to door notification by a uniformed public safety official such as a police officer or firefighter is probably the most effective method, but many jurisdictions say they

86

Chapter 4 · Who Leaves and Who Does Not

lack the resources to disseminate warnings that way. Evacuation notices are typically transmitted through local news media, but TV stations do not necessarily provide maps or other tools to help people evaluate their own situations with respect to the notice and, of course, radio stations cannot provide maps. Police and fire vehicles sometimes drive through neighborhoods with loudspeakers, but not everyone hears them. Some communities have automated telephone notification (“reverse 911”) systems that can target specific geographical areas and have been shown to be effective (Neves et al. 2014), but they require landline telephones or voluntary registration by cell phone users to receive the calls. A 2016 survey in Florida indicated that only 37% of responding households said they had signed up with their local emergency management agency to receive emergency notifications (Baker, Downs, and St. Germaine 2016). Subsets of cell towers can be used to deliver messages, but that method still lacks the spatial precision needed for many hazards.

4.2.4 Environmental and Social Cues Environmental cues refer to conditions that people can observe in their physical environment that can indicate the presence of a threat. Examples include rising water, an exceptionally heavy rainfall that might result in flash flooding, high wind speed, an unusual chemical smell, ground shaking, and smoke. People who are aware of a hazard in their location can associate environmental cues with that hazard and infer that a specific event might be about to occur. Conversely, however, people sometimes infer that they are not at risk if they do not observe environmental cues. For example, Gruntfest, Downing, and White (1978) reported that people at the mouth of Colorado’s Big Thompson Canyon refused to evacuate when given an accurate flash flood warning because the morning sky was clear and sunny (the storm cell had produced intense rainfall and flooding high in the mountains overnight). However, they later evacuated when given an inaccurate warning of a dam break upstream in Estes Park because a dam break would not be expected to provide environmental cues before the flood water arrived. Social cues are indications that people receive from other people, such as observations of businesses closing or neighbors preparing to evacuate. Such observations motivate evacuation because they indicate that peers consider the threat serious enough to take action. The Huang, Lindell, and Prater (2016b) review of hurricane evacuation studies found that observation of peers evacuating was more strongly correlated with evacuation than was the observation of businesses closing or environmental cues. Indeed, peer evacuation was second only to receipt of an official warning as a predictor of evacuation.

Chapter 4 · Who Leaves and Who Does Not

87

4.2.5 Information Sources Literature reviews have produced somewhat mixed results when assessing the effect of information sources on evacuation. As Dow and Cutter (2000, 2002) have noted, coastal residents have many options for information during a hurricane threat beyond those issued by official government agencies. The Thompson, Garfin, and Silver (2017) discussion of warning sources concluded that community officials such as law enforcement officers have the greatest credibility and produce more evacuation compliance, but noted that peers and the news media also served as important warning sources. However, the Huang et al. (2016) review concluded that the very few hurricane studies of reliance on authorities, news media, and peers suggest these sources by themselves have a mixed or weak effect on evacuation. For example, Gladwin, Gladwin, and Peacock (2001) found that few people evacuate solely because authorities advise them to do so. Instead, authorities are influential because they are believed to provide credible information and it is that information that people use to make their evacuation decisions. One likely explanation for the difference between the conclusions of Thompson et al. (2017) and Huang et al. (2016) is that, although Thompson et al. (2017) examined a wider range of hazards, they conducted a conventional narrative review that simply counted the number of studies that reported a positive effect. By contrast, Huang et al. (2016) conducted a more powerful statistical meta-analysis (SMA) of the hurricane evacuation literature that reported weighted average effect sizes. The disadvantage of narrative reviews is that they often produce equivocal results when some studies show significant effects and others show nonsignificant effects. In such cases, the authors often report that there is “some evidence” of an effect. By contrast an SMA weights the effect size of each study in accordance with its sample size, thus producing a weighted average effect size. In addition, SMAs produce 95% confidence intervals for those effect sizes, allowing authors to determine if there is a high probability of a non-zero effect size in the population. Another part of the difficulty might be that the literature does not always distinguish between the medium delivering an evacuation notice with the delivery of more general information about the hazard, including evacuation notices but also forecasts, alerts, etc. In Hurricane Floyd, evacuees were asked to indicate which information source had the greatest influence on their decision to evacuate: public officials, news media, or peers. They were told to only consider media information other than evacuation notices from public officials. Fifty one percent said public officials had the greatest influence, followed by the news media (32%), and then peers (16%). In open-ended questions asking coastal residents why they did or did not evacuate in hurricanes, most people cite safety-related concerns or satisfaction rather than information from any particular source.

88

Chapter 4 · Who Leaves and Who Does Not

At least for hurricanes, most people rely primarily on television for information, although they have ample opportunity to use multiple sources of information given the amount of forewarning before landfall. In 2012, Hurricane Isaac threatened the Gulf Coast from northwest Florida through southeast Louisiana and Hurricane Sandy threatened the Middle Atlantic and Northeast states. Coastal residents in those regions were called by telephone as the storms were threatening and asked to indicate how much they were relying on various information sources about the storms. Table 4.3 indicates that the great majority relied a great deal on television, followed by the Internet, peers, radio, and social media (Baker et al. 2013a, 2013b). These results are consistent with findings from the Lindell et al. (2005) study of Hurricane Lili in Texas.

4.2.6 Experience It would be unwise to suggest that previous experience with a hazard or evacuation for the hazard has no effect on future evacuation behavior, but the role of experience is not as straightforward as it might seem. Recent literature reviews conducted for hurricane studies (Huang, Lindell, and Prater 2016b) and for natural hazards in general (Thompson, Garfin, and Silver 2017) found no consistent relationship between experience and evacuation. Some studies found positive relationships, some found negative relationships, but most found no relationship. Huang et al. (2016) found that, among 21 studies of household evacuation, 24% reported significant positive correlations, 10% reported significant negative correlations, and 66% reported nonsignificant correlations. Overall, the correlations for the actual evacuation studies ranged from r = −.12 to .29 with a nonsignificant weighted average correlation r = .01. Part of the problem is that experience has been measured in many different ways, ranging from “have you ever experienced a hurricane” to reports of specific impacts (deaths, injuries, property damage, utility loss, job and school disruption) to self and personal acquaintances. Demuth, Morss, Lazo and Trumbo (2016) shed some light on this issue by finding that different measures of hurricane experience (previous evacuation, property damage,

Table 4.3 Percent of Respondents Saying They Relied a Great Deal on Information Sources During Hurricanes Isaac and Sandy Television

Internet

Peers

Radio

Social Media

Isaac

89

33

32

18

10

Sandy

91

29

18

12

8

From Baker et al. 2013a, 2013b

Chapter 4 · Who Leaves and Who Does Not

89

financial loss, household casualties, personal distress, and overall impact severity from all hurricane experiences) varied in their effects on four mediating variables—cognitive risk perception, affective risk perception, self efficacy, and response efficacy—that, in turn, have direct effects on evacuation decisions. Significantly, some aspects of experience had effects that worked in opposite directions. For example, personal distress tended to increase evacuation intentions by increasing affective risk perception but decrease evacuation intentions by decreasing self efficacy. At the community level, there is ample evidence that it is not necessary for a community to have experience in order to have successful evacuations. In 1979, Hurricane Frederic produced a higher evacuation participation rate in Pensacola, Florida than Pass Christian, Mississippi and Panama City, Florida, both of which had more recent major damage from hurricanes (Baker 1991). In Hurricane Floyd, the Savannah, Georgia region was the area with the greatest evacuation participation rate of 11 clusters of coastal counties from south Florida through North Carolina, even though it was statistically one of the most “overdue” locations in the study area for a hurricane strike (USACE Savannah District 2000). Length of residence is not the same as experience, but it does provide an opportunity to gain more experience with a hazard and provides more opportunity to learn about it in other ways. In Floyd, after adjusting statistically for other factors, people who had lived in their home fewer than 10 years were about 10 percentage points more likely to evacuate, compared to people who had lived in their homes longer (Baker 2000a). Thompson et al. (2017) looking at a broad range of natural hazards, found length of residence mostly to have a negative relationship with evacuation. However, Huang et al. (2016) reported that, among the 12 hurricane evacuation studies examining this variable, the weighted average correlation of length of residence with evacuation was a nonsignificant r = -.02. A special case relating to experience is the experience of a previous “unnecessary” evacuation for a disaster that failed to occur or that struck elsewhere—the “cry wolf” effect. The concern is that people who are told to evacuate during one event and do so, will be less likely to do so in future events if the hazard does not subsequently affect their location. That is, if the evacuation was a “false alarm.” It is probably true that if people evacuate frequently enough without being affected by a hazard, the evacuation rates will decline. However, there is little evidence of that point being reached in hurricane evacuations. When non-evacuees are asked why they did not leave, few say it was because of having evacuated unnecessarily in the past. When evacuees, even those experiencing long transit times, were asked if they would do anything differently, few said they would not have left. In 1985 in Panama City Beach, Florida residents were told to evacuate three times

90

Chapter 4 · Who Leaves and Who Does Not

in one hurricane season. The evacuation rate was essentially the same in all three evacuations, despite the fact that the threatening storms missed in all three cases (Baker 1991). People who evacuate in one storm are mostly likely to do the same thing in future threats. This evacuation pattern was documented in Hurricanes Bertha and Fran in 1996, where 39% evacuated for both hurricanes, 37% remained home for both, and those who evacuated only for the first hurricane were offset by those who evacuated only for the second one (Dow and Cutter 1998). There were similar results for Hurricanes Ivan in 2004 and Katrina in 2005 (Murray-Tuite et al. 2012). There was a concern in Florida in 2004 that residents in some communities were experiencing “evacuation fatigue” due to multiple evacuations, but evacuation response rates for people who said they were told to evacuate in second or even third storms remained as high as in previous storms in evacuation zones (Baker 2006). Huang et al. (2016) reported that the correlations of a previous “unnecessary” evacuation ranged from r = −.16 to .15 in seven evacuation studies, with a nonsignificant weighted average correlation r = .01.

4.2.7 Housing Type As indicated previously, hurricane evacuation zones are based on vulnerability to storm surge and waves and, when a hurricane threat is deemed great enough, residents in all structures located in those zones are included in evacuation notices from public officials. However, mobile homes and manufactured housing are usually deemed to be more vulnerable than site-built housing to wind, and consequently residents in those structures are advised or ordered to evacuate even if they reside inland of surge-defined evacuation zones. These evacuation notices normally apply at least to coastal counties (parishes in Louisiana) but, depending on the strength of a storm and its forward speed, might extend farther inland. (Changes to manufactured housing construction standards in effect since 1994 might eventually lead to modifications in evacuation policies for mobile homes and manufactured housing, but that modification has yet to occur.) In short, residents of mobile homes are usually at greater risk than those in site-built homes. Mobile home residents generally recognize their greater risk and are therefore more likely than site-built home residents to evacuate in hurricanes, making this one of the four best predictors of evacuation (Huang, Lindell, and Prater 2016b). Interviews in Florida in 2004 and 2005, averaged over multiple storms, geographical areas of the state, and risk areas revealed that mobile home residents were more than twice as likely to leave, compared to residents of site-built homes (Baker 2010b). In Lee County Florida in Wilma in 2005, 70% of mobile home residents in mainland locations evacuated, whereas only 48% of

Chapter 4 · Who Leaves and Who Does Not

91

residents living in site-built homes on islands did so (Baker 2010b). After accounting statistically for risk area, hearing evacuation notices from public officials, perceived vulnerability, and demographic characteristics, mobile home residents are still more likely than site-built residents to leave. In Opal the difference was 29%, in Floyd it was 23%, and in Andrew it was 48% (Baker 2010b).

4.2.8 Pets Many pet owners are reluctant to evacuate if they cannot take their pets with them. If their only available evacuation accommodations will not allow pets, that provides a deterrent to evacuating (Heath, Beck, et al. 2001, Heath, Kass, et al. 2001). Statistically, however, pets have a relatively small or nonsignificant effect on overall evacuation participation rate (Hunt, Bogue, and Rohrbaugh 2012). Edmonds and Cutter (2008) reviewed surveys conducted following evacuations to summarize the percentage of non-evacuees citing pets as their reason for not leaving, and calculated a mean of 2.6%. They included 29 surveys, all but three of which followed hurricanes, and 11 were for different locations in Hurricane Floyd. Responses included areas not told by officials to evacuate, as well as areas that were. Other studies have compared evacuation rates between people who did and did not own pets, with Thompson et al. (2017) finding pet owners to be consistently less likely to evacuate, although they cited just four actual response studies in support of this conclusion. In Hurricane Wilma in Lee County Florida in 2005, pet owners were less likely to evacuate only if they believed their home would be safe in a storm like Wilma (Baker 2007). Pet owners who said their home would be unsafe were as likely as non-pet owners to leave. In Hurricane Floyd, with a sample of nearly 7,000, after adjusting statistically for risk area, hearing evacuation notices, perceived vulnerability, demographics, and other predictors, pet owners were about 5 percentage points less likely to have evacuated (Baker 2000a). In New York City and surrounding counties in Hurricane Irene, there was no difference in evacuation based on pet ownership, after adjusting statistically for risk area and political jurisdiction (Baker 2014). The issue of pets has led to agencies considering shelter provision that allows them. To assess the level of provisions for pets in evacuation planning, an NCHRP survey (Wolshon 2009) investigated how animals were being considered in plans and how sheltering plans included amenities for pets, livestock, and service companions. The survey showed that the majority of emergency preparedness agencies gave consideration to pets and service companions at some level and that more than 60% included them in their sheltering plans. It is important to note that respondents to the survey included accommodations for service animals. Although comparatively fewer in number, some agencies around the

92

Chapter 4 · Who Leaves and Who Does Not

country are also now beginning to make sheltering provisions for livestock, such as horses, particularly in areas such as California where this issue is important to the local populace. Edmonds and Cutter (2008) developed a pet estimation model that can be used to assist local officials in identifying the geographic distribution of families with pets that are likely to require shelter. They suggested that the use of widely available population datasets and the research methodology would provide the tools necessary to estimate the sheltering needs of pet-owning residents within their jurisdictions.

4.2.9 Demographic Characteristics The role of demographics in predicting evacuation behavior is somewhat contentious. Obstacles associated with a number of demographic variables can certainly make evacuation difficult or impossible without assistance for some people (Bian and Wilmot 2017). Social vulnerability indexes, some of which are aimed specifically at evacuation, rely heavily on demographic and economic variables (Chakroborty, Tobin, and Montz 2005; Schmidtlein et al. 2008). However, demographic variables do not predict evacuation nearly as well as other factors such as risk area, housing, hearing evacuation notices, and perceived vulnerability. The Thompson et al. (2017) review of evacuation from a broader range of natural hazards concluded that there was fairly consistent evidence that women are more likely to evacuate than men, Whites are more likely to evacuate than other racial groups, and the elderly are less likely to evacuate than other age groups. They found that having children in the home and having a disabled person in the home consistently resulted in less evacuation. They concluded that education, income, and home ownership were inconsistent predictors. However, as noted earlier, their review was a conventional narrative review that simply counted the number of studies that reported a positive effect. The Huang et al. (2016b) SMA reported average effect sizes that supported Baker’s (1991) conclusion that demographic variables are weak and inconsistent predictors of evacuation. Several authors have suggested that demographics probably play a role in the evacuation decision-making process, even if they are not correlated strongly or consistently with evacuation behavior (Riccheti-Masterson and Horney 2013; Huang et al. (2016a). Women are known to be more risk averse than men, for example, but other factors might mitigate their greater tendency to evacuate. Evacuations are typically household behaviors, so demographic measures that apply just to the survey respondent might not predict the group behavior—although Maghelal, Peacock, and Lee (2016) noted that some households evacuate in separate groups. It is likely that some of the inconsistencies observed in the literature result from similar demographics being associated with differing cultural norms from place to place. Baker and Peacock (2009) reported that non-

Chapter 4 · Who Leaves and Who Does Not

93

evacuees in households with incomes lower than $15,000 annually were far more likely to say they did not evacuate from Hurricane Floyd simply because they felt safe staying home (62%) as opposed to having no place to go (8%) or no transportation (4%). But in New Orleans when Katrina struck, a large percentage of the poor population was carless, and that contributed to the number of people who did not evacuate before the storm (Brodie et al. 2006). A surprising number of studies have done little to account for the role of risk when evaluating the role of demographics. As noted earlier, risk area is a strong predictor of who evacuates and who does not, but depending on the hazard and location, demographics might be correlated with risk area. This is clearly the case in most locations for hurricanes. Vulnerability to surge and waves generally decreases as a function of distance from the shoreline, as does the value of property, so income and possibly other demographics tend to decrease along with risk area. But some studies have not measured vulnerability or risk area at all, so it is not possible to determine whether survey respondents should have evacuated. A number of other studies measured risk area roughly by simply noting whether a respondent resided in or out of an evacuation zone, without capturing variations in vulnerability within the evacuation zone. Mobile home residents tend to have lower incomes, and because mobile homes are more vulnerable to wind than site-built housing, evacuation tends to be greater from mobile homes. Most hurricane studies have done a better job of measuring housing than risk area.

References Arlikatti, S., Lindell, M.K., Prater, C.S., Zhang, Y. 2006. Risk area accuracy and hurricane evacuation expectations of coastal residents. Environment and Behavior 38 (2), 226–247. Baker, E.J. 1991. Hurricane evacuation behavior. International Journal of Mass Emergencies and Disasters 9 (2), 287–310. Baker, E.J. 1995. Public response to hurricane probability forecasts. Professional Geographer 47 (2), 137–147. Baker, E.J. 2000a. Evacuation behavior in Hurricane Floyd. In: National Hurricane Conference, New Orleans, LA. Baker, E.J. 2000b. Hurricane evacuation in the United States. In Pielke, R., Jr and Peilke, R., Sr. Storms. Routledge, New York, pp. 306–319. Baker, E.J. 2004. Public beliefs about the accuracy of hurricane forecasts. In: Association of American Geographers Annual Conference, Philadelphia, PA. Baker, E.J. 2006. Evacuation fatigue in Florida in 2004. In: Association of American Geographers Annual Conference, Chicago, IL. Baker, E.J. 2007. Public Response to Hurricane Wilma and Hurricane Preparedness in Lee County, Florida. Hazards Management Group, Tallahassee, FL.

94

Chapter 4 · Who Leaves and Who Does Not Baker, E.J. 2009. Shadow evacuation in Hurricanes. In: Association of American Geographers Annual Conference, Las Vegas, NV. Baker, E.J. 2010a. Hurricane Preparedness in Florida Households in 2010. Hazards Management Group, Tallahassee, FL. Baker, E.J. 2010b. Mobile home evacuation in hurricanes. In: Association of American Geographers Annual Conference, Washington, DC. Baker, E.J. 2013. Hurricane Irene Post-storm Behavioral Survey. Hazards Management Group, Tallahassee, FL. Baker, E.J. 2014. New York City Hurricane Evacuation Study Behavioral Analysis. Hazards Management Group, Tallahassee, FL. Baker, E.J., Broad, K., Czajkowski, J., Meyer, R. 2013a. In: Hurricane Isaac realtime interviews. National Hurricane Conference, New Orleans, LA. Baker, E.J., Broad, K., Czajkowski, J., Meyer, R. 2013b. In: Hurricane Sandy realtime interviews. National Hurricane Conference, New Orleans, LA. Baker, E.J., Broad, K., Meyer, R. 2011. Effect of Perceived Vulnerability on Response to Hurricane Threats: Real-Time Survey Interim Project Report. Florida State University, Florida Catastropic Storm Risk Management Center, Tallahassee, FL. Baker, E.J., Downs, P., St. Germaine, J. 2016. 2016 Florida Household Disaster Preparedness Survey: Executive Summary. Prepared for the Florida Division of Emergency Management. Hazards Management Group, Tallahassee, Florida. Baker, E.J., Peacock, W.G. 2009. Evacuation effectiveness: meeting the unique challenges of low-income households. In: Safe Homes for All, 2009 Hurricane Risk Mitigation Leadership Forum, Washington, DC. Baker, E.J., Zhao, T. 2012. Perceived vs. actual homesite elevations in Florida coastal counties – Implications for evacuation. In: Association of American Geographers Annual Conference, New York. Bean, H., Sutton, J., Liu, B.F., Madden, S., Wood, M.M., Mileti, D.S. 2015. The study of mobile public warning messages: a research review and agenda. Review of Communication 15 (1), 60–80. Berg, K., Wilson, B. 2013. A multilevel analysis of private-vehicle evacuation clearance times along the US Gulf Coast. Environmental Hazards 12 (3–4), 291–310. Bian, R., Wilmot, C.G. 2017. Measuring the vulnerability of disadvantaged populations during hurricane evacuation, Natural Hazards 85 (2), 691–707. Brodie, M., Weltzien, E., Altman, D., Blendon, R.J., Benson, J.M. 2006. Experiences of Hurricane Katrina evacuees in Houston shelters: implications for future planning. American Journal of Public Health 96 (8), 1402–1408. Burton, I., Kliman, M., Powell, D., Schmidt, L., Timmerman, P., Victor, P., Whyte, A., Wojick, J. 1981. The Mississauga Evacuation. University of Toronto, Toronto. Chakroborty, J., Tobin, G.A., Montz, B.E. 2005. Population evacuation: assessing spatial variability in geophysical risk and social vulnerability to natural hazards. Natural Hazards Review 6 (1), 23–33. Chenault, W.W., Hilbert, G.D., Reichlin, S.D. 1980. Evacuation Planning in the TMI Accident. Human Sciences Research, McLean, VA.

Chapter 4 · Who Leaves and Who Does Not

95

Cox, J., House, D., Lindell, M.K. 2013. Visualizing uncertainty in predicted hurricane tracks. International Journal for Uncertainty Quantification 3, 143–156. Cutter, S., Barnes, K. 1982. Evacuation behavior and Three Mile Island, Disasters 6 (2), 116–124. Demuth, J.L., Morss, R.E., Lazo, J.K., Trumbo, C. 2016. The effects of past hurricane experiences on evacuation intentions through risk perception and efficacy beliefs: A mediation analysis. Weather, Climate, and Society 8 (4), 327–344. Dow, K., Cutter, S.L. 1998. Crying wolf: Repeat responses to hurricane evacuation orders. Coastal Management 26 (4), 237–252. Dow, K., Cutter, S.L. 2000. Public orders and personal opinions: household strategies for hurricane risk assessment. Environmental Hazards 2 (4), 143–155. Dow, K., Cutter, S.L. 2002. Emerging Hurricane evacuation issues: Hurricane Floyd and South Carolina. Natural Hazards Review 3 (1), 12–18. Edmonds, A.S., Cutter, S.L. 2008. Planning for pet evacuations during disasters. Journal of Homeland Security and Emergency Management 5 (1). Flynn, C.B. 1979. Three-Mile Island Telephone Survey: Preliminary Report on Procedures and Findings. Mountain West Research, Tempe, AZ. Gladwin, C.H., Gladwin, H., Peacock, W.G. 2001. Modeling hurricane evacuation decisions with ethnographic methods. International Journal of Mass Emergencies and Disasters 19 (2), 117–143. Gruntfest, E., Downing, T., White, G.F. 1978. Big Thompson flood exposes need for better flood reaction system. Civil Engineering,78, 72–73. Heath, S.E., Beck, A.M., Kass, P.H., Glickman, L.T. 2001. Risk factors for pet evacuation failure after a slow-onset disaster. Journal of the American Veterinary Medicine Association 218 (12), 1905–1910. Heath, S.E, Kass, P.H., Beck, A.M., Glickman, L.T. 2001. Human and pet-related risk factors for household evacuation failure during a natural disaster. American Journal of Epidemiology 153 (7), 659–665. Huang, S.K., Lindell, M.K., Prater, C.S. 2016a. Toward a multi-stage model of hurricane evacuation decision: An empirical study of Hurricanes Katrina and Rita. Natural Hazards Review 18 (3), 1–15. Huang, S-K., Lindell, M.K., Prater, C.S. 2016b. Who leaves and who stays? A review and statistical meta-analysis of hurricane evacuation studies. Environment and Behavior 48 (8), 991–1029. Huang, S-K., Lindell, M.K., Prater, C.S., Wu, H-C., Siebeneck, L.K. 2012. Household evacuation decision making in response to Hurricane Ike. Natural Hazards Review 13 (4), 283–296. Hunt, M. 2005. Affidavit of Sheriff Mike Hunt, US District Court, South Carolina Aiken Division, Graniteville Train Derailment. Hunt, M.G., Bogue, K., Rohrbaugh, N. 2012. Pet ownership and evacuation prior to Hurricane Irene. Animals 2 (4), 529–539. Johnson, B.T., Maio, G.R., Smith-McLallen, A. 2005. Communication and attitude change: causes, processes, and effects. In Albarracin, D., Johnson, B.T., Zanna, M.P. (Eds.), The Handbook of Attitudes. Erlbaum, Mahwah NJ, pp. 617–669.

96

Chapter 4 · Who Leaves and Who Does Not Kang, J.E., Lindell, M.K., Prater, C.S. 2007. Hurricane evacuation expectations and actual behavior in Hurricane Lili. Journal of Applied Social Psychology 37 (4), 87–903. Lasswell, H. 1948. The structure and function of communication in society. In: Bryson, L. (Ed.), Communication of Ideas. Harper, New York, pp. 43–71. Li, J., Ozbay, K., Bartin, B., Iyer, S., Carnegie, J.A. 2013. Empirical evacuation response curve during hurricane Irene in Cape May County, New Jersey. Transportation Research Record 2376, 1–10. Lindell, M.K. 2018. Communicating imminent risk. In: Rodríguez, H., Donner, W., Trainor, J. (Eds.), Handbook of Disaster Research, Springer, New York, pp. 449–477. Lindell, M.K., Bolton, P.A., Perry, R.W., Stoetzel, G.A., Martin, J.B. & Flynn, C.B. 1985. Planning Concepts and Decision Criteria for Sheltering and Evacuation in a Nuclear Power Plant Emergency, AIF/NESP-031. Atomic Industrial Forum, Behtesda MD. Lindell, M.K., Hwang, S.N. 2008. Households’ perceived personal risk and responses in a multihazard environment. Risk Analysis 28 (2), 539–556. Lindell, M.K., Lu, J-C., Prater, C.S. 2005. Household decision making and evacuation in response to Hurricane Lili. Natural Hazards Review 6 (4), 171–179. Lindell, M.K., Perry, R.W. 1983. Nuclear power plant emergency warning: How would the public respond? Nuclear News, 26, 49–53. Lindell, M.K., Perry, R.W. 1992. Behavioral Foundations of Community Emergency Planning. Hemisphere Press, Washington DC. Lindell, M.K, Perry, R.W. 2004. Communicating Environmental Risk in Multiethnic Communities. Sage Publications, Thousand Oaks CA. Lindell, M.K, Perry, R.W. 2012. The Protective Action Decision Model: Theoretical modifications and additional evidence. Risk Analysis 32 (4), 616–632. Lindell, M.K., Prater, C.S. 2007. Critical behavioral assumptions in evacuation time estimate analysis for private vehicles: examples from hurricane research and planning. Journal of Urban Planning and Development 133 (1), 18–29. Lindell, M.K., Prater, C.S. 2010. Tsunami preparedness on the Oregon and Washington coast: Recommendations for research. Natural Hazards Review, 11 (2), 69–81. Lindell, M.K., Prater, C.S., Perry, R.W, Wu, J-Y. 2002a. EMBLEM: An EmpiricallyBased Large Scale Evacuation Time Estimate Model. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S., Perry, R.W, Wu, J-Y. 2002b. Hurricane Evacuation Time Estimates for the Texas Gulf Coast. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S., Sanderson, W.G., Lee, H-M., Zhang, Y., Mohite, A., Hwang, S-N. 2001. Texas Gulf Coast Residents’ Expectations and Intentions Regarding Hurricane Evacuation. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S., Wu, H-C., Huang, S-K., Johnston, D.M., Becker, J.S., Shiroshita, H. 2016. Immediate behavioral responses to earthquakes in Christchurch New Zealand and Hitachi Japan. Disasters 40 (1), 85–111.

Chapter 4 · Who Leaves and Who Does Not

97

Maghelal, P., Peacock, W.G., Li, X. 2016. Evacuating together or separately: factors influencing split evacuations prior to Hurricane Rita. Natural Hazards Review DOI: 10.1061/(ASCE)NH.1527-6996.0000226. Meyer, R., Baker, E.J., Broad, K., Czajkowski, J., Orlov, B. 2014. The dynamics of risk perception: Real-time evidence from the 2012 Atlantic hurricane season. Bulletin of the American Meteorological Society 95 (9), 1389– 1404. Meyer, R., Broad, K., Petrovic., N 2013. Dynamic simulation as an approach to understanding hurricane risk response: Insights from the Stormview lab. Risk Analysis 33 (8), 1532–1552. Mileti, D.S., Peek, L. 2000. The social psychology of public response to warnings of a nuclear power plant accident. Journal of Hazardous Materials 75 (2–3), 181–194. Mileti, D.S., Sorensen, J.H. 1987. Why people take precautions against natural disasters. In Weinstein, N. (Ed.), Taking Care: Why People Take Precautions. Cambridge University Press, New York, pp. 296–320. Mitchell, J.T., Cutter, S.L., Edmonds, A.S. 2007. Improving shadow evacuation management: case study of the Graniteville, SC, chlorine spill. Journal of Emergency Management 5 (1), 28–34. Moore, H.E., Bates, F.L., Layman, M.V., Parenton, V.J. 1963. Before the Wind: A Study of the Response to Hurricane Carla. National Academy of Sciences – National Research Council, Washington, DC. Murray-Tuite, P.M., Yin, W., Ukkusuri, S., Gladwin, H. 2012. Changes in evacuation secisions between Hurricanes Ivan and Katrina. Transportation Research Record 2312, 98–107. Neves, T.T., Mann, S.C., Myers, L.B., Cosby, A.G. 2014. Assessing reverse 911: a case study of the 2007 San Diego wildfires. Journal of Emergency Management 12 (4), 315–325. Peacock, W.G., Maghelal, P., Lindell, M.K., Prater, C.S. 2007. Draft: Hurricane Rita Behavioral Survey Final Report. Texas A&M University Hazards Reduction and Recovery Center, College Station TX. Perry, R.W., Greene, M.R. 1983. Citizen Response to Volcanic Eruptions. New York: Irvington. Perry, R.W., Lindell, M.K. 1980. Predisaster planning to promote compliance with evacuation warnings. In Baker, E.J. (Ed.) Hurricanes and Coastal Storms: Awareness, Education and Mitigation. Florida State University, Tallahassee, FL, pp. 44-49. Perry, R.W., Lindell, M.K., Greene, M.R. 1981. Evacuation Planning in Emergency Management. Heath-Lexington Books, Lexington MA. Petrolia, D.R., Bhattacharjee, S. 2010. Why don’t coastal residents choose to evacuate for hurricanes? Coastal Management 38 (2), 97–112. Riccheti-Masterson, K., Horney, J. 2013. Social factors as modifiers of hurricane Irene evacuation behavior in Beaufort Count, SC. PLOS – Current DOI: 10.1371/currents.dis.620b6c2ec4408c217788bb1c091ef919. Ross, J., Shaw, B.M. 1993. Organizational escalation and exit: Lessons from the Shoreham nuclear power plant. Academy of Management Journal 36 (4), 701–732.

98

Chapter 4 · Who Leaves and Who Does Not Ruginski, I.T., Boone, A.P., Padilla, L.M., Liu, L., Heydari, N., Kramer, H.S., Hegarty, M., Thompson, W.B., House, D.H., Creem-Regehr, S.H. 2016. Nonexpert interpretations of hurricane forecast uncertainty visualizations. Spatial Cognition & Computation 16 (2), 154–172. Schmidtlein, M.C., Deutsch, R.C., Piegorsch, W.W., Cutter, S.L. 2008. A sensitivity analysis of the social vulnerability index. Risk Analysis, 28 (4), 1099–1114. Schwarz, N. 2007. Retrospective and concurrent self-reports: The rationale for real-time data capture. In: Stone, A., Shiffman, S.S., Atienza, A., Nebeling, L. (Eds.), The Science of Real-Time Data Capture: Self-Reports in Health Research. Oxford University Press, New York, pp. 11–26. Stough, L., Mayhorn, C.B. 2013. Population segments with disabilities. International Journal of Mass Emergencies and Disasters 31 (3), 384–402. Tampa Bay Regional Planning Council. 2010. Regional Behavioral Survey Report, Volume 3-8. Tampa Bay Regional Planning Council, Tampa FL, accessed 6 December, 2016 at www.tbrpc.org/tampabaydisaster/sres2010/docs/ Vol_3_Behavioral_Survey.pdf. Thompson, R.R., Garfin, D.R., Silver, R.C. 2017. Evacuation from natural disasters: a systematic review of the literature. Risk Analysis 37 (4), 812–839. USACE—US Army Corps of Engineers Philadelphia District. 1996. Hurricane Opal assessment: Review of the use and value of hurricane evacuation study products in the Hurricane Opal Evacuation, Alabama and Florida, October 3–4, 1995. Accessed 8 December 2016 at coast.noaa.gov/hes/docs/post Storm/H_OPAL_ASSESSMENT_REVIEW_USE_VALUE_HES_PRODUCTS_OPA L_EVACUATION_AL_FL.pdf. USACE—US Army Corps of Engineers Savannah District. 2000. Hurricane Floyd Assessment: Review of Hurricane Evacuation Studies Utilization And Information Dissemination. US Army Corps of Engineers, accessed 14 October 2016 at coast.noaa.gov/hes/postStorm.html?redirect=301ocm. USACE—US Army Corps of Engineers New England District. 2016. New England Hurricane Evacuation Study Technical Data Report. US Army Corps of Engineers, accessed 14 October 2016 at www.nae.usace.army.mil/portals/74/ docs/Topics/HurricaneStudies/2016%20State%20Updates/Massachu setts/New%20England%20Hurricane%20Evacuation%20Study.pdf. USEPA—US Environmental Protection Agency 1987. Technical Guidance For Hazards Analysis: Emergency Planning For Extremely Hazardous Substances. US Environmental Protection Agency, Washington, DC. USNRC/FEMA—US Nuclear Regulatory Commission/Federal Emergency Management Agency. 1980. Criteria for Preparation and Evaluation of Radiological Emergency Response Plans and Preparedness in Support of Nuclear Power Plants. NUREG-0654, FEMA-REP-1, Rev.1. US Nuclear Regulatory Commission, Washington DC. Vogt, B.M., Sorensen, J.H. 1999. Description of Survey Data Regarding the Chemical Repackaging Plant Accident West Helena, Arkansas. Oak Ridge National Laboratory, Oak Ridge, TN. Walton, F., Wolshon, B. 2010. Understanding public response to nuclear power plant protective actions. Risk, Hazards & Crisis in Public Policy 1 (3), 35–61.

Chapter 4 · Who Leaves and Who Does Not

99

Wilmot, C.G. 2004. Data collection related to emergency events. In: 7th International Conference on Travel Survey Methods, Costa Rica. Wolshon, B. 2009. Transportation’s Role in Emergency Evacuation and Reentry. National Cooperative Highway Research Program, Synthesis of Highway Practice 392. Washington DC. Wolshon, B., Urbina Hamilton, E., Levitan, M., Wilmot, C. 2005. Review of policies and practices for hurricane evacuation. II: traffic operations, management, and control. Natural Hazards Review 6 (3), 143–161. Wu, H-C., Lindell, M.K., Prater, C.S. 2012. Logistics of hurricane evacuation in Hurricanes Katrina and Rita. Transportation Research Part F 15 (5), 445–461. Wu, H-C., Lindell, M.K., Prater, C.S. 2015a. Process tracing analysis of hurricane information displays. Risk Analysis 35 (12), 2202–2220. Wu, H.C., Lindell, M.K., Prater, C.S. 2015b. Strike probability judgments and protective action recommendations in a dynamic hurricane tracking task. Natural Hazards 79 (1), 355–380. Wu, H.C., Lindell, M.K., Prater, C.S., Samuelson, C.D. 2014. Effects of track and threat information on judgments of hurricane strike probability. Risk Analysis 34 (6), 1025–1039. Zeigler, D.J., Johnson, J.H. 1984. Evacuation behavior in response to nuclear power plant accidents. The Professional Geographer 36 (2), 207–215. Zeigler, D.J., Brunn, S.D., Johnson, J.H., Jr. 1981. Evacuation from a nuclear technological disaster. Geographical Review 71 (1), 1–16. Zhang, Y., Prater, C.S., Lindell, M.K. 2004. Risk area accuracy and evacuation from Hurricane Bret. Natural Hazards Review 5 (3), 115–120.

Chapter 5

When Do Evacuees Leave?

Not all households evacuate at the same time and the circumstances that account for departure timing vary across incidents. As indicated in Chapter 1, authorities sometimes detect a hazard and issue an official warning that recommends people leave before impact (pre-impact, postwarning evacuation). In other cases, people are aware of an approaching hazard and decide to leave before authorities issue an evacuation notice (pre-impact, pre-warning evacuation). Finally, there are cases in which authorities are unable to detect hazard onset before impact. In most of these cases (e.g., flash floods), people evacuate in response to environmental cues and informal warnings from peers (post-impact, pre-warning) but there are some situations in which environmental cues are absent (e.g., some toxic chemical and radiological hazards) in which people evacuate after authorities detect an impact and issue an evacuation warning (post-impact, post-warning). In the most common case—pre-impact, post-warning evacuation—the percentage of the risk area population that evacuates at each successive time interval usually takes the form of an S-shaped curve in which, according to Urbanik et al. (1980), the time required for an individual resident or transient household to evacuate after incident initiation can be defined as a function of four time components in Equation 3.1—the authorities’ decision time (Section 5.1), the household’s warning receipt time (Section 5.2), the household’s evacuation preparation time (Section 5.3), and the household’s evacuation travel time (Chapter 9).

5.1 Authorities’ Decision Times Authorities’ decision times can be quite small in some instances, for example, when local Incident Commanders are well trained in the use of decision aids such as the Emergency Response Guidebook (Pipeline and Hazardous Materials Safety Administration 2016). These decision aids fulfill the three Cova et al. (2017) objectives defined in Chapter 3— identify the target groups that should take protective action (everyone

Chapter 5 · When Do Evacuees Leave?

101

within a specified distance), the most appropriate PAR for each target group (evacuate or shelter in-place), and the time that these PARs should be initiated (in some cases, immediately). However, it is unreasonable to assume that there will always be minimal delays in authorities’ evacuation decisions. To estimate the distribution of authorities’ decision times, Rogers (1994) collected data from public officials about chemical emergencies that occurred in 1990. His data, which are displayed in Figure 5.1, show that authorities were able to issue a PAR within 30 minutes in over half of the incidents. Although PAR issuance in some incidents took four hours or more, many of these were minor incidents in which it was unclear if any protective action would be needed. Authorities’ decision times are usually less of an issue in hurricane evacuations because these storms are detected many days before an evacuation decision must be made. For example, as Hurricane Lili approached the Lake Sabine area of Texas, local authorities advised their residents eight hours before the National Hurricane Center issued a hurricane warning that they would issue an evacuation notice early the next morning (Lindell, Lu, and Prater 2005). Nonetheless, there are important counter-examples such as the New Orleans mayor’s tardy issuance of an evacuation notice as Hurricane Katrina approached. In that event, the NHC had issued a hurricane warning at 11:00 p.m. CDT on Saturday, August 27 but the mayor failed to issue an evacuation notice until 10:00 a.m. CDT on Sunday, August 28. This was only

Figure 5.1 Authorities’ Decision Time Curve for Rapid Onset Disasters

Percentage of Warnings Issued

100 80 60 40 20 0

0

30

60

90

120 Time (min)

150

180

210

240

102 Chapter 5 · When Do Evacuees Leave? 20 hours before Katrina made landfall close to Buras, Louisiana, at 6:10 a.m. on Monday, August 29. The delays in Hurricane Katrina are not unique; Cova et al. (2017) report that local authorities tend to struggle with selecting PARs and, especially, determining what situational cues should trigger the change from “wait and see” to “take immediate action”. Lindell and Prater (2007) and Dye, Eggers, and Shapira (2014) called attention to local authorities’ concerns about false positive (evacuating for an event that strikes elsewhere) and false negative (failing to evacuate for an event that does strike) decision errors. In addition to their concerns about the costs of evacuations for local government, local officials are also concerned—needlessly as noted in Section 4.2.6—that false positives will reduce local residents’ willingness to evacuate in future incidents. Conversely, officials who have clearly defined authority, welldefined procedures, and an accurate understanding of research findings on people’s warning responses will be able to make prompt decisions (Sorensen and Mileti 2016).

5.2 Warning Dissemination Times Once authorities decide to issue an evacuation notice, it takes time for the message to be disseminated. This time component can be represented by a cumulative distribution function of the form in Equation (5.1).
pt ¼ 1  exp at b þ c where pt exp

c

is the proportion of the households that have been warned at time t, denotes the base of the natural logarithm (e), which is raised to the t multiplied by the coefficient a and raised to the b power, and is a constant that is positive when people become aware of a hazardous situation before the time that authorities issue an evacuation warning (which is defined as t = 0).

The most dangerous situations—and thus one about which emergency managers should be greatly concerned—are rapid onset events such as flash floods, local tsunamis, and hazardous materials releases. In addition, hurricanes can occasionally generate conditions that provide little forewarning—a late changing track or late intensification (Lindell, Prater, and Peacock 2007). A late changing track is a situation in which a hurricane tracks parallel to the coast but makes a sharp turn toward land before striking the coast (e.g., Hurricane Bret in 1999 and

Chapter 5 · When Do Evacuees Leave?

103

Hurricane Charley in 2004). A late intensification shortly before landfall is a situation in which a hurricane’s maximum wind speed, as indicated by its Saffir-Simpson category, increases by one or more categories shortly before the arrival of Tropical Storm wind—the most common deadline for clearing coastal evacuation zones. Figure 5.2 presents rapid onset warning dissemination distributions for six communities. Toutle WA, 15 miles northwest of Mt. St. Helens, and Woodland WA, 30 miles southwest of the mountain, were threatened by the mountain’s May 18, 1980 eruption (Lindell and Perry 1987, Perry and Greene 1983). In addition, there are data from four flood stricken communities (Perry, Lindell, and Greene 1981). Sumner WA and Fillmore CA were struck by flash floods, whereas Snoqualmie WA and Valley NE had more forewarning of their flooding. As Figure 5.2 indicates, the rate of warning diffusion was much more rapid in Toutle than in Woodland during the first 30 minutes but both communities achieved nearly complete warning dissemination within four hours. Warning dissemination in Sumner and Fillmore was somewhat slower than in Toutle initially but all three of these communities, together with Woodland, had achieved approximately the same level of warning dissemination at two hours (80–90%). Warning dissemination was significantly slower in Snoqualmie and especially in Valley. At one hour, Snoqualmie had only achieved just over half of Toutle’s level of warning dissemination and Valley had only achieved just over one third of Toutle’s level. Figure 5.3 presents warning dissemination data from Hurricanes Katrina and Rita that were computed from a questionnaire item in

Figure 5.2 Warning Dissemination Curves For Rapid Onset Incidents

104 Chapter 5 · When Do Evacuees Leave?

Figure 5.3 Warning Dissemination Curves For Normal Hurricane Landfall

which respondents were asked to report how many times per day they obtained information from the local news media and peers (0, 1–2, 3–4, 5–6, or 7 or more). These contact frequencies were then converted to contact intervals. As the figure indicates, the data imply that 5–30% of the respondents would have received a warning almost immediately—a figure that is lower than Dow and Cutter’s (2000) report that 58% of their respondents left a news channel on continuously once a hurricane that might threaten South Carolina was within 2–3 days of the coast. The news media curves for Katrina and Rita suggest that 85–90% of the risk area population would learn about an evacuation notice from the news media within 12 hours and there would be only a slight increase during the interval from 12–24 hours. This, of course, underestimates the total percentage of the risk area population who would become aware of an evacuation notice within that time period because it ignores the likelihood that some people would be contacted by peers or would see neighbors evacuating before they checked the news media. In addition, Figure 5.3 presents warning diffusion time distributions on the island of Mauritius during the 2004 Indian Ocean tsunami (Perry 2007) and in Boston during the 2010 water contamination incident (Lindell, Huang, and Prater 2017). Perry (2007) reported that it took around five hours after the earthquake for governmental authorities to learn about the tsunami and another hour before the first warnings were issued. Figure 5.3 displays an adjusted warning diffusion time distribution that has removed the time lag between the time that the earthquake occurred and the time that Mauritius authorities issued their first warnings. This adjusted distribution almost certainly

Chapter 5 · When Do Evacuees Leave?

105

underestimates the warning diffusion time distribution that would occur during a hurricane threat because the percentage of the Mauritius population that received a warning at the origin (t0) is only 5% whereas, as noted earlier, Dow and Cutter (2000) found that this was 58% in Hurricane Floyd. Nonetheless, the percentage of the respondents who were alerted about the tsunami rose to 80% within 10 hours (8:00 pm), flattened overnight, and increased by 10 percentage points over the next day. It is important to note that the analyses in Perry’s (2007) study made no distinction between respondents who were in coastal areas and those who were in inland areas, so it is impossible to tell how much faster residents of coastal areas received warnings than did residents of inland areas. Lindell, Huang, and Prater (2017) examined local residents’ warning sources, warning channels, and warning receipt times during the May 1–4 2010 Boston water contamination incident. Both the Massachusetts governor and the Boston mayor declared a state of emergency and initiated warnings through a wide range of channels. Figure 5.3 shows that in Boston, as in Mauritius, there was a very small percentage of respondents who received an immediate warning but the percentage rose steadily through the afternoon and evening hours. Here too, the rate of warning diffusion flattened at about 80% and remained at that level overnight. The somewhat slower rate of increase in the Mauritius and Boston warning diffusion data than in the Hurricane Katrina and Hurricane Rita warning diffusion data can be explained by differences in people’s attentiveness to unexpected events (the Indian Ocean tsunami and the Boston water contamination incident) in comparison to events with ample forewarning (most hurricanes). Moreover, Lindell et al. (2017) found that, under normal circumstances, people’s patterns of news media monitoring vary systematically over the course of the day. Respondents in the Boston water contamination reported that their TV, radio, newspaper, and total media access over the course of the typical day had a small peak in TV access from 6–8 am, with a much larger peak from 5–7 pm and a slightly smaller one at 10-11pm. Radio access had a more pronounced peak from 7–9 am, another at noon, and a third peak from 4–7 pm. Finally, newspaper access had a single major peak from 7–10 am with another minor peak from 5–6 pm. This timeof-day variation in news media access is an important consideration because the distribution of warning times in the Boston water contamination incident followed a logistic (S-shaped) distribution, with the largest increase in warnings reception taking place during prime TV news time (4–6 pm). That is, the Boston warning diffusion distribution, like the one in Mauritius, was much slower because most warning recipients were only monitoring the news media at their normal intervals for daily activities rather than the elevated intervals found in the aftermath of being alerted to an approaching hurricane.

106 Chapter 5 · When Do Evacuees Leave?

5.3 Evacuation Preparation Times Some aspects of evacuation behavior, particularly the evacuation of the nuclear family as a unit is axiomatic in the social sciences (Zelinsky and Kosinski 1991). When families are separated at the time a warning is received, those at home either wait for the others to return or else contact them and arrange to meet at a safe location (Perry et al. 1981, pp. 43–45). Nonetheless, some families that are together at the time of warning receipt will leave in separate groups at different times (Maghelal, Peacock, and Li 2016) or some family members may stay behind as required by their employers. However, other preparation activities vary by type of event. Lindell and Perry (1992) reported preparation time data from Lindell and his colleagues’ (1985) report that compared evacuation analysts’ assumed preparation time distributions for a nuclear power plant evacuation to empirical data from Perry and his colleagues’ (1981) four flood evacuation communities and Perry and Greene’s (1983) two volcano evacuation communities. The assumed distributions reached 100% between 15 and 105 minutes, whereas Figure 5.4 shows that none of the empirical distributions even reached 100% within 120 minutes. Indeed, the two flash flood stricken communities had the highest levels of evacuation at that time, but even they only reached just over 80%. Later, Lindell et al. (2001) collected data from Texas coastal residents on the amount of time they estimated it would take them to prepare for an approaching hurricane by completing six tasks—1)

Figure 5.4 Evacuation Preparation Time Curves For Normal Hurricane Landfall

Chapter 5 · When Do Evacuees Leave?

107

prepare to leave from work; 2) travel from work to home; 3) gather all of the persons who would evacuate with them; 4) pack the items they would need while gone; 5) protect their property from storm damage (e.g., board up windows); and 6) shut off utilities, secure the home, and leave. Respondents indicated if it would take them 15 minutes or less, 16–30 minutes, 31–45 minutes, 46–60 minutes, or 61 minutes or more. The validity of the expected preparation time distributions was supported by later research on some of the same respondents’ actual preparation times during Hurricane Lili (Kang, Lindell, and Prater 2007). Two later surveys collected data on the actual amount of time that people took to perform these same six tasks when preparing to evacuate from Hurricanes Katrina and Rita (Huang, Lindell, and Prater 2016, Lindell and Prater 2008, Wu, Lindell, and Prater 2012b). In addition, the Katrina and Rita questionnaires asked respondents to report the time that they decided to evacuate and the time that they actually evacuated, which yields an estimate of their evacuation delay time—an upper bound on the time required for evacuation preparation. Finally, Baker (2005) reported similar data on evacuation delay time from Hurricane Charley evacuees These surveys provided response alternatives in different formats, so the data in Figure 5.4 round up the total preparation time to the nearest integer. However, total preparation times greater than 6 hours were coded into three 6-hour categories—7–12, 13–18, and 19–24 hours. In order to indicate the time at which the five curves that had not reached asymptote at 6 hours did finally reach this point, the scale for the x-axis has been compressed. The resulting curves in Figure 5.4 reveal a number of noteworthy results. First, unsurprisingly, those at home expected to (TX Home Estimated), and actually did (Katrina Home Actual and Rita Home Actual), prepare more rapidly than those at work (TX Work Estimated, Katrina Work Actual, and Rita Work Actual). Indeed, those at work take approximately 90 minutes longer to reach 50% prepared and there is a difference of over two hours to reach 100%. Second, there are relatively few differences between the results for the estimated and actual results. The only notable difference is the one-hour difference to reach 100% prepared for those at home. Third, the departure delay curves for Charley, Katrina, and Rita are also quite similar to each other, but are notably different from the expected and actual preparation time curves. Specifically, the departure delay curves indicate that larger percentages of households leave almost immediately after they decide to do so and that the departure delay curves are much flatter than the expected and actual task preparation time curves. There are a number of plausible explanations for the differences produced by the two types of methods of estimating household mobilization times (expected/actual task completion vs. self-reported/computed evacuation delay). One possibility is that the task completion times only provided a partial measure of logistical preparation to evacuate because

108 Chapter 5 · When Do Evacuees Leave? some households might have performed additional tasks that were not in the list. This would produce an underestimate of their mobilization time. This explanation is consistent with the Lindell et al. (2001) data on coastal residents expectations of the time it would take them to prepare to leave from work and to travel from work to home (see also Kang et al. 2007). One of the reasons for this trip would be to reunify the household before leaving, as reported in previous evacuation studies. For example, Drabek and Boggs (1968) found that, in an unexpected flood, families that were separated but needed to evacuate immediately selected a destination they believed the missing family members would think of so that the family could reunite as soon as possible. Perry et al. (1981) reported somewhat similar results. Although only about 4% of the households in their four flooded communities had members missing, 60% of those households waited for the missing persons to return before evacuating, whereas 35% contacted the missing members and met them at a safe location. The evacuation of the family as a unit suggests that during short- and no-notice events, the focus of preparation activity will be on family reunification. Many households will have people who wish to remain at home to gather belongings, supplies, pets, and other family members— as well as obtain additional information or warn others—before evacuating (Lindell et al. 2015). In some cases, people try to return home even if it is not safe to do so. For example, Suppasri et al. (2013) reported that there were many cases in which people left a safe location to return home—thus becoming casualties in the 2011 Japan tsunami. More recent research has sought to go beyond merely describing family reunification to attempting to model the process based on interviews and surveys. One such study (Liu, Murray-Tuite, and Schweitzer 2012) used structured interviews with 233 parents and caregivers (out of 315 total interviewees) in the Chicago, IL region to identify anticipated behaviors for a minor and a major no-notice event. For context, the researchers defined two evacuation radii (5 and 25 miles) relative to the parent’s workplace. For the minor incident, nearly three quarters of parents who had a child at school or day care within the evacuation radius indicated they would pick up their children; this was much more than the normal child chauffeuring rate (the percentage who normally picked up their children from school or day care). Nearly one quarter would pick up the children even if they were outside the evacuation zone. In the major incident scenario, a smaller portion of parents indicated they would chauffeur their children than in the minor incident scenario, however, more would do so than on a normal day. This smaller percentage in the major incident could have been due to another caregiver located closer to the school assuming this role. Transportation options and travel distance may have led respondents to believe that they could not reach the school in a timely fashion and that someone else could better accomplish the child gathering or that the school

Chapter 5 · When Do Evacuees Leave?

109

would best be able to evacuate the child in the larger incident scenario. Gender differences, prominent in normal child-related travel, were less apparent in the two emergency scenarios. In households where one adult was a transit commuter and another was a car commuter, the person with the car was more likely to pick up the child for evacuation. The parent that was at a location (e.g., home or work) that was closer to the children’s location was more likely to pick them up in an evacuation (Liu, Murray-Tuite, and Schweitzer 2012). Jones, Walton, and Sullivan (2008) conducted a study of public knowledge of, confidence in, and expected responses to nuclear power plant PARs. The research, which included extensive telephone surveys and focus group sessions of residents living within the 10-mile nuclear power plant EPZs and focus group interviews with emergency response managers and personnel included questions about the sheltering and evacuation of school children during emergencies. The survey results were consistent with the findings of evacuation preparation trip studies (Murray-Tuite and Mahmassani 2003, 2004, Liu, Murray-Tuite, and Schweitzer 2012). Specifically, many parents were aware that schools within the EPZs have evacuation plans to transport school children to safe shelter areas and indicated they received annual emergency planning information from their children’s schools. Nonetheless, they planned go to the schools and pick up their children even if they were told the children were already being evacuated and parents should not come to the school. A general finding of the research was that parents did not think the schools would safely evacuate their children, often because of a perceived lack of buses. The report concluded that, in recognition of these opinions, evacuation analysts should anticipate traffic congestion around schools during nuclear power plant evacuations. Young children are not the only ones who may be picked up in evacuation preparation trips; spouses, adult children, parents, friends, and neighbors may also be picked up (see Table 5.1 for associated factors). Reasons for family reunification may be concern for their wellbeing or advance knowledge that they need assistance in an evacuation (Liu, Murray-Tuite, and Schweitzer 2014). A key conclusion of this study was that, when deciding whether or not to pick up a spouse, people first consider their own mobility for this trip, followed by their spouse’s willingness or need for transportation assistance. The researchers were unable to develop an acceptable statistical model for picking up adult-age children, probably because there were only 16 respondents in this category. Pets are also likely to be picked up from home because they are an important part of family life (Heath, Kass, et al. 2001). Based on their interviews, (Liu, Murray-Tuite, and Schweitzer 2014, p. 591) concluded “those who did keep pets were clear that they were part of their reasoning for gathering at home before long term regional evacuation. Since so many families planned on congregating at home prior to

Income

Education level

Correlated with ethnicity and education level. High income households are more likely (expect) to pick up children in normal and emergency situations.

Those with a car available are more likely to pick up a spouse. This is the dominating factor.

Car availability significantly affected child pick-up behavior/expectations when parents are far away from children.

Car availability

Highly correlated with car availability

Spouses’ education is more important than respondents’ education levels in predicting pick up expectations.

Nonsignificant. Men and women are almost equally likely to pick up spouses.

Women are more likely to pick them up in an emergency, but gender is less significant than on regular days.

Gender

Driver

Spousesb

Type of Person Gathered

Juvenile Childrena

Influential Factor

Table 5.1 Influential Factors in Family Reunification for No-Notice Events

Higher household income increases the odds of picking up parents in the evacuation zone.

Parentsb

Drivers are much more likely to pick up non-family members.

Non-Family Memberb

Unemployed parents are more likely responsible for chauffeuring children than employed parents.

Employment status

a (Liu, Murray-Tuite, and Schweitzer 2012) b (Liu, Murray-Tuite, and Schweitzer 2014a)

The farther parents are away from children, the less likely they will pick them up; parents closer to the children are more likely to pick them up in an emergency.

These individuals are less likely to pick up non-family members

Marital status – married

Distance

These individuals are less likely to pick up non-family members

Parents of a juvenile child

112 Chapter 5 · When Do Evacuees Leave? leaving the region entirely, gathering pets at the same time as they collected valuables or at-home family members made sense”. If the home were not in imminent danger, it would be a logical meeting place. Based on the interviews of 210 respondents expecting to pick up family members, 34% expected to do so at home (Liu, Murray-Tuite, and Schweitzer 2014). Liu et al. (2014a) found that men were more likely than women to report reuniting at home, more adults in the household increased the odds of reuniting at home, more children in the household decreased the odds of reuniting at home for parents (interaction term of parents and number children in the household). In a univariate case, children increase the odds of the household reuniting at home (Liu, Murray-Tuite, and Schweitzer 2014). Another behavioral intention survey found that preparing to leave the home was expected to take about an hour and about 68% of households with regular commuters would wait for them to return home before evacuating (Ma et al. 2009). Further support for the explanation that the difference between evacuation preparation times estimated by the expected/actual task completion time method and the departure time method is due to an incomplete task list can be found in two recent studies that have produced thorough examinations of trips taken outside the home in preparation for evacuation. Specifically, a study of Key West households’ responses to Hurricane Wilma found that 45% of them took one or more trips within the city as the hurricane approached and only 49% of these respondents evacuated (Noltenius 2008, Noltenius and Ralston 2010). Going to work accounted for more trips than any other activity (26%), followed by picking up persons (13%), getting medicines (9%), getting money (7%), and getting property protection materials (4%). Significantly, 41% of the trips involved a unique activity or combination of activities, none of which were included in the expected/actual task completion list for Hurricanes Lili, Katrina, or Rita. The authors reported average evacuation delay times of 44, 44, and about 60 minutes for the first, second, and third trips, respectively. Another study suggesting the difference between evacuation preparation times estimated by the expected/actual task completion time method and the departure time method is due to an incomplete task list can be seen in Yin and his colleagues’ (Yin 2013, Yin et al. 2014) survey of Miami Beach residents’ expected responses to a Category 4 hurricane. Their results indicated that only 49% (227/462) of the respondents expected to limit themselves to the types of in-home tasks addressed in the Lili, Katrina, and Rita surveys. The remaining 51% would make at least one tour—a travel circuit that begins and ends at the same location—in preparation for evacuation. The overwhelming majority of them (91%) would make one tour, 8% would make two tours, and the rest would make three tours. Of those making a single tour, 64% would make a single stop and 30% would make two stops.

Chapter 5 · When Do Evacuees Leave?

113

The most common purchases were gas (51% of all stops), food (43%), medications (19%), and cash (17%). Because this survey collected data on expected, rather than actual, responses, there are no data on delay times. However, it is clear from the Noltenius and Yin studies that a significant percentage of households responding to hurricane warnings will take a significant amount of time traveling in their cars to different locations throughout their communities to prepare for a later evacuation. One variation on the incomplete task list explanation is that some households might have completed the six listed tasks but waited until they reached what they believed was the last possible instant before they needed to leave, an additional “task” that can be called psychological preparation (as opposed to logistical preparation) to evacuate. A variation on this explanation is that the task completion time method fails to account for people’s known reluctance to evacuate at night. Thus, people who decided at 4 pm to evacuate and took four hours in logistical preparation, but waited until 8am the next day to leave would have a mobilization time of four hours when estimated by the task completion method but 16 hours when estimated by the departure delay method. Another possible problem with the task completion method is that it might overestimate households’ mobilization time because some households had different people perform the tasks concurrently. If this were the case, adding completion times for concurrent tasks would produce an overestimate of the household’s mobilization time. A variation on this explanation is that some households might have performed one or more of the tasks after they became aware of the threat but before they received an evacuation warning. For example, a household might have gathered all of the persons who would evacuate, packed the items they would need while gone, and protected their property from storm damage (e.g., boarded up windows) in anticipation of the possibility that they might need to leave. In the absence of data that can definitively distinguish among these explanations, evacuation analysts should assume that the curves for expected and actual preparation time at work produce the most conservative distributions up to four hours. Beyond that, evacuation delay time curves provide the most conservative distributions.

5.4 Evacuation Departure Times In theory, the cumulative distributions of warning times and preparation times can be combined to produce a normalized distribution of departure times. The distribution is normalized because it describes the proportion, not the absolute number, of vehicles beginning an evacuation at time t. Assuming statistical independence between warning time and preparation times, a synthetic departure time distribution can be

114 Chapter 5 · When Do Evacuees Leave? calculated by multiplying the probability of warning receipt within a given time interval by the probability of preparing within each successive time interval (Urbanik 2000, Lindell et al. 2001). In practice, however, it can be problematic to construct a distribution (over households) of departure time from the available data on warning receipt and evacuation preparation. First, some households who are monitoring the progress of a storm will begin preparing to evacuate before they receive an official evacuation warning. Indeed, the available evidence indicates that at least a few households will begin their evacuation before they receive an official evacuation warning. Second, constructing a synthetic departure distribution from warning diffusion and preparation time distributions requires an assumption about the correlation between warning receipt and evacuation preparation. Although a correlation of zero is computationally convenient, it is not necessarily plausible because people who receive later warnings are likely to accelerate their evacuation preparations. This negative correlation between warning receipt and preparation time is likely to be especially high when warnings are not initiated until shortly before disaster impact. These issues make it important to conduct research on evacuation time components in a wide range of situations to assess the validity of synthetic departure distributions. Nonetheless, it is possible to begin to address the issue of uncertainty about departure distributions by examining the variation in departure curves that have been reported for a variety of different hazards. Most evacuation studies have reported data from hurricanes, which are usually detected long before landfall so most evacuees are divided between preimpact, pre-warning evacuation and pre-impact, post-warning evacuation. That is, some households leave before public officials issue evacuation notices (pre-impact, pre-warning evacuation), whereas others leave very soon following issuance of evacuation notices and some wait until shortly before they expect the threatening event to arrive (pre-impact, post-warning evacuation). In some post-impact pre-warning evacuations, such as the 1978 Sumner flood documented by Perry et al. (1981), households have responded to environmental cues, such as observation of rising water, and informal warnings received from peers before authorities disseminate warnings throughout the impact area. There is little data on post-impact post-warning evacuations, such as the one that took place after Cyclone Tracy, because many of these are intended to provide temporary postdisaster housing rather than protect the population from an environmental hazard (Haas, Cochrane, and Eddy 1977). Many surveys documenting response following evacuations have asked evacuees to indicate the time (and date in the case of multi-day evacuations) when they departed their homes. The responses have been graphed to depict cumulative evacuation curves. The curves show how the percentage of evacuating households (on the y-axis) grows cumulatively over time (on the x-axis), typically with a few people leaving early and then increasing to the point at which all of the evacuees (but not all

Chapter 5 · When Do Evacuees Leave?

115

of those in the risk area) have eventually departed. The data rely on the recall of evacuees about fairly specific information whose accuracy might suffer with passage of time. Nevertheless, studies have been rather consistent with respect to a number of patterns. Data from Rogers and Sorensen’s (1989) study of hazardous materials incidents in Confluence and Pittsburg Pennsylvania provide good examples of variation in departure time distributions across incidents. Figure 5.5 indicates that the Confluence incident generated a rapid warning dissemination and almost everyone evacuated soon after receiving a warning. By contrast, the Pittsburgh incident generated a somewhat slower warning dissemination and only about half of those who received a warning evacuated and those who did evacuate took much longer to do so. The longer delays in Pittsburgh were due to disbelieving the initial warning, passively waiting to see how the situation developed, and actively seeking additional information. Although rare, there are examples of hurricane evacuation departure curves that also have relatively short time intervals, although in this context, “short” is 12 hours. Two cases are Pinellas County Florida’s response to Hurricane Elena during 1985 (Baker 1986) and northwest Florida’s response to Hurricane Opal during 1995 (USACE Philadelphia District 1996). Another case is northwest Florida’s response to Hurricane Eloise during 1975, which is the only documented example of hurricane evacuation departures occurring over a period of just six hours. However, the limited forewarning might be the reason why as little as 45% of the risk area population evacuated from some locations (Baker et al. 1976). Most hurricanes have many days of forewarning and some evacuation notices have been issued many days in advance, so evacuations

Figure 5.5 Evacuation Departure Times for Rapid-Onset Incidents

116 Chapter 5 · When Do Evacuees Leave? often take place over multiple days (Huang, Lindell, and Prater 2016, Huang et al. 2012, Lindell, Lu, and Prater 2005). For example, Figure 5.6 shows a typical pattern of evacuation departure times divided into six hour intervals (early morning, 12 am–6 am; late morning, 6 am–12 pm; afternoon, 12 pm–6 pm; and evening, 6 pm–12 am). In anticipation of Hurricane Ike, about 15% of households departed September 8–10 before the NHC Hurricane Watch, another 20% left in the next 18 hours before the NHC Hurricane Warning, and the remainder left thereafter. Over that time, there were consistent spikes in evacuation departures during the late morning and afternoon followed by a substantial decline in the evening. This is because people prefer not to evacuate at night but will do so if necessary. Hurricane examples of nighttime evacuations are Eloise (Baker et al. 1976), Elena (Baker 1986), and Opal (USACE Philadelphia District 1996) in Florida. The Hurricane Ike data are typical of the departure time data for other hurricanes as well. For example, Figure 5.7 shows curves for four coastal regions (southeast Florida, northeast Florida, northeast Georgia, and southeast North Carolina) that were threatened by Hurricane Floyd (USACE Savannah District 2000). Over a four-day period from September 12–16, 1999, this storm travelled roughly parallel to the Atlantic coast from south Florida until it made landfall in North Carolina. In each of the four regions, the evacuations were spread out over multiple days and each showed the same general pattern of minimal evacuations in the overnight hours (shaded in gray) followed by a dramatic increase during the day (8 am–4 pm), and a significant

Figure 5.6 Evacuation Departure Times for Hurricane Ike 25.00 Hurricane Watch 4:00 PM

Percent

20.00

15.00

10.00

5.00

0.00

Hurricane Warning 10:00 AM Thu Sep 11

Hurricane Eye Landfall 2:30 AM Sat Sep 13

Chapter 5 · When Do Evacuees Leave?

117

Figure 5.7 Evacuation Departure Times for Hurricane Floyd

Cumulative Percent of Evacuees

100

80

60

40 FL-SE FL-NE GA-NE NC-SE

20

0 12 4 8 12 4 8 12 4 8 12 4 8 12 4 8 12 4 8 12 4 8 12 4 8 12 4

| 12 Sept |

13 Sept

|

14 Sept

|

15 Sept

|

16 Sept

decline in the evening (4 pm–8 pm). Nonetheless, there were some noticeable differences, with the two Florida regions having a maximum of 40% of their evacuations take place on a single day, whereas the Georgia and North Carolina regions had a maximum of 60% of their evacuations take place on a single day. Several analyses of evacuation departure time distributions have found statistical relationships that might be used to predict households’ departure times (Koshute 2013, Fu et al. 2007, Li et al. 2013). However, the results of studies to date are rather inconclusive because the few studies addressing this issue have not identified a consistent set of predictors. Indeed, departure times in Hurricanes Katrina and Rita were uncorrelated with any of the 14 demographic variables assessed in those questionnaires (Lindell and Prater 2008).

References Baker, E.J. 1986. Hurricane Elena: Preparedness and response in Florida. Florida Policy Review 1 (2), 17–23. Baker, E.J. 2005. Hurricane Charley Evacuation Behavioral Analysis. Hazards Management Group, Tallahassee, FL. Baker, E.J., Brigham, J.C., Paredes, J.A., Smith, D.D. 1976. Social impact of Hurricane Eloise on Panama City Florida. In Immediate Response Project. Gainesville, FL: State University System of Florida Sea Grant Program.

118 Chapter 5 · When Do Evacuees Leave? Cova, T.J., Dennison, P.E., Li, D., Drews, F.A., Siebeneck, L.K., Lindell, M.K. 2017. Warning triggers in environmental hazards: who should be warned to do what and when? Risk Analysis 37 (4), 601–611. Dow, K., Cutter, S.L. 2000. Public orders and personal opinions: household strategies for hurricane risk assessment. Environmental Hazards 2 (4), 143–155. Drabek, T.E., Boggs, K.S. 1968. Families in disaster: reactions and relatives. Journal of Marriage and the Family 30 (3), 443–451. Dye, K.C., Eggers, J.P., Shapira, Z. 2014. Trade-offs in a tempest: stakeholder influence on hurricane evacuation decisions. Organization Science 25 (4), 1009–1025. Fu, H., Wilmot, C.G., Zhang, H., Baker, E.J. 2007. Modeling the hurricane evacuation response curve. Transportation Research Record 2022, 94–102. Haas, J.E., Cochrane, H.C., Eddy, D.G. 1977. Darwin, Australia, Christmas 1974: Consequences of a cyclone on a small city. Ekistics 44 (1), 45–51. Heath, S.E., Beck, A.M., Kass, P.H., Glickman, L.T. 2001. Risk factors for pet evacuation failure after a slow-onset disaster. Journal of the American Veterinary Medicine Association 218 (12), 1905–1910. Huang, S.K., Lindell, M.K., Prater, C.S. 2016. Toward a multi-stage model of hurricane evacuation decision: An empirical study of Hurricanes Katrina and Rita. Natural Hazards Review 18 (3), 1–15. Huang, S-K., Lindell, M.K., Prater, C.S., Wu, H-C., Siebeneck, L.K. 2012. Household evacuation decision making in response to Hurricane Ike. Natural Hazards Review 13 (4), 283–296. Jones, J.A., Walton, F., Sullivan, R.L. 2008. Review of NUREG-0654, Supplement 3, Criteria for Protective Action Recommendations for Severe Accidents: Focus Groups and Telephone Survey SAND2008-4195P, NUREG/CR-6953, Vol. 2. US Nuclear Regulatory Commission, Washington DC. Kang, J.E., Lindell, M.K., Prater, C.S. 2007. Hurricane evacuation expectations and actual behavior in Hurricane Lili. Journal of Applied Social Psychology 37 (4), 881–897. Koshute, P. 2013. Evaluation of existing models for prediction of hurricane evacuation response curves. Natural Hazards Review 14 (3), 175–181. Li, J., Ozbay, K., Bartin, B., Iyer, S., Carnegie, J.A. 2013. Empirical evacuation response curve during hurricane Irene in Cape May County, New Jersey. Transportation Research Record 2376, 1–10. Lindell, M.K., Bolton, P.A., Perry, R.W., Stoetzel, G.A., Martin, J.B. & Flynn, C.B. 1985. Planning Concepts and Decision Criteria for Sheltering and Evacuation in a Nuclear Power Plant Emergency, AIF/NESP-031. Atomic Industrial Forum, Behtesda MD. Lindell, M.K., Huang, S-K., Prater, C.S. 2017. Predicting residents’ responses to the May 1–4, 2010, Boston water contamination incident. International Journal of Mass Emergencies and Disasters 35 (1), 84–113. Lindell, M.K., Lu, J-C., Prater, C.S. 2005. Household decision making and evacuation in response to Hurricane Lili. Natural Hazards Review 6 (4), 171–179.

Chapter 5 · When Do Evacuees Leave?

119

Lindell, M.K, Perry, R.W. 1987. Warning mechanisms in emergency response systems. International Journal of Mass Emergencies and Disasters 5 (2), 137–153. Lindell, M.K., Perry, R.W. 1992. Behavioral Foundations of Community Emergency Planning. Hemisphere Press, Washington DC. Lindell, M.K., Prater, C.S. 2007. A hurricane evacuation management decision support system (EMDSS). Natural Hazards 40 (3), 627–634. Lindell, M.K., Prater, C.S. 2008. Behavioral Analysis: Texas Hurricane Evacuation Study. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S., Gregg, C.E., Apatu, E., Huang, S-K., Wu, H-C. 2015. Households’ immediate responses to the 2009 Samoa earthquake and tsunami. International Journal of Disaster Risk Reduction 12, 328–340. Lindell, M.K., Prater, C.S., Peacock, W.G. 2007. Organizational communication and decision making for hurricane emergencies. Natural Hazards Review 8 (3), 50–60. Lindell, M.K., Prater, C.S., Sanderson, W.G., Lee, H-M., Zhang, Y., Mohite, A., Hwang, S-N. 2001. Texas Gulf Coast Residents’ Expectations and Intentions Regarding Hurricane Evacuation. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Liu, S., Murray-Tuite, P., Schweitzer, L. 2012. Analysis of child pick-up during daily routines and for daytime no-notice evacuations. Transportation Research – A 46 (1), 48–67. Liu, S., Murray-Tuite, P., Schweitzer, L. 2014. Uniting multi-adult households during emergency evacuation planning. Disasters 38 (3), 587–609. Ma, Y., Krometis, J., Sen, A. 2009. Radiological emergency evacuation trip generation model developed from telephone survey. In: 88th Annual Meeting of the Transportation Research Board. Washington, DC. Maghelal, P., Peacock, W.G., Li, X. 2016. Evacuating together or separately: factors influencing split evacuations prior to Hurricane Rita. Natural Hazards Review DOI: 10.1061/(ASCE)NH.1527-6996.0000226. Murray-Tuite, P.M., Mahmassani, H.S. 2003. Model of household trip chain sequencing in an emergency evacuation. Transportation Research Record 1831, 21–29. Murray-Tuite, P.M., Mahmassani, H.S. 2004. Transportation network evacuation planning with household activity interactions. Transportation Research Record 1894, 150–159. Noltenius, M.S. 2008. Capturing Pre-Evacuation Trips and Associative Delays: A Case Study of the Evacuation of Key West, Florida for Hurricane Wilma. Dissertation. Department of Geography, The University of Tennessee, Knoxville TN. Noltenius, M.S., Ralston, B. 2010. Pre-evacuation trip behavior. In: Showalter, P., Lu, Y. (Eds.), Geotechnologies and the Environment: Geospatial Techniques in Urban Hazard and Disaster Analysis. Springer Science, Netherlands, pp. 395–413. Perry, S.D. 2007. Tsunami warning dissemination in Mauritius. Journal of Applied Communication Research 35 (4), 399–417.

120 Chapter 5 · When Do Evacuees Leave? Perry, R.W., Greene, M.R. 1983. Citizen Response to Volcanic Eruptions. New York: Irvington. Perry, R.W., Lindell, M.K., Greene, M.R. 1981. Evacuation Planning in Emergency Management. Heath-Lexington Books, Lexington MA. Pipeline and Hazardous Materials Safety Administration. 2016. Emergency Response Guidebook. Pipeline and Hazardous Materials Safety Administration, Washington DC. www.phmsa.dot.gov/hazmat/outreach-training/erg. Rogers, G.O. 1994. The timing of emergency decisions: modeling decisions by community officials during chemical accidents. Journal of Hazardous Materials 37 (2), 353–373. Rogers, G.O., Sorensen, J.H. 1989. Warning and response in two hazardous materials transportation accidents in the US. Journal of Hazardous Materials 22 (1), 57–74. Sorensen, J.H., Mileti, D.S. 2016. First Alert and/or Warning Issuance Delay Time Estimation for Dam Breaches, Controlled Dam Releases, and Levee Breaches and Overtopping. US Army Corps of Engineers Institute for Water Resources Risk Management Center, Davis CA. Suppasri, A., Shuto, N., Imamura, F., Koshimura, S., Mas, E., Yalciner, A.C. 2013. Lessons learned from the 2011 Great East Japan tsunami: performance of tsunami countermeasures, coastal buildings, and tsunami evacuation in Japan. Pure and Applied Geophysics 170 (6–8), 993–1018. USACE—US Army Corps of Engineers Philadelphia District. 1996. Hurricane Opal assessment: Review of the use and value of hurricane evacuation study products in the Hurricane Opal Evacuation, Alabama and Florida, October 3–4, 1995. Accessed 8 December 2016 at coast.noaa.gov/hes/docs/postStorm/ H_OPAL_ASSESSMENT_REVIEW_USE_VALUE_HES_PRODUCTS_OPAL_EVA CUATION_AL_FL.pdf. USACE—US Army Corps of Engineers Savannah District. 2000. Hurricane Floyd Assessment: Review of Hurricane Evacuation Studies Utilization And Information Dissemination. US Army Corps of Engineers, accessed 14 October 2016 at coast.noaa.gov/hes/postStorm.html?redirect=301ocm. Urbanik, T. 2000. Evacuation time estimates for nuclear power plants. Journal of Hazardous Materials 75 (2), 165–180. Urbanik, T., Desrosiers, A., Lindell, M.K., Schuller, C.R. 1980. An Analysis of Techniques for Estimating Evacuation Times for Emergency Planning Zones, NUREG/CR-1745. US Nuclear Regulatory Commission, Washington, DC. Wu, H-C., Lindell, M.K., Prater, C.S. 2012. Logistics of hurricane evacuation in Hurricanes Katrina and Rita. Transportation Research Part F 15 (5), 445–461. Yin, W. 2013. An Agent-based Travel Demand Model System for Hurricane Evacuation Simulation. Unpublished Ph.D. Dissertation, Virginia Tech Civil and Environmental Engineering, Blacksburg VA. Yin, W., Murray-Tuite, P.M., Ukkusuri, S.V.,Gladwin, H. 2014. An agent-based modeling system for travel demand simulation for hurricane evacuation. Transportation Research – Part C 42, 44–59. Zelinsky, W., Kosinski, L.A. 1991. The Emergency Evacuation of Cities: A CrossNational Historical and Geographical Study. Rowman and Littlefield Publishers, Savage, MD.

Chapter 6

Managing Evacuation Logistics

As noted in Chapters 4 and 5, most empirical research on household evacuations has addressed only the first phase of the process—the activities that take place during the time between an evacuation decision and departure from the home. This chapter addresses the next phase of evacuation—the activities that take place between the time that people leave their homes and the time they arrive at their evacuation destinations. The activities and associated resources associated with this phase can be called evacuation logistics. This definition of evacuation logistics is consistent with the definitions of logistics provided in a variety of textbooks on the topic. For example, “[l]ogistics describes the entire process of materials and products moving into, through, and out of a firm” (Johnson and Wood 1996, p. 4), “[b]usiness logistics is the study and management of goods and service flows and the associated information that sets these into motion” (Ballou 1987, p.6) and “[l]ogistics is concerned with physical and information flows from raw material through to the final distribution of the finished product” (Rushton, Oxley, and Croucher 2000, p. 4). Households’ evacuation logistics include selecting a transportation mode (personal vehicle, peer’s vehicle, public transportation) and the number of vehicles to take if they use personal vehicles, the destinations to which they expect to travel, the routes they expect to take, and the accommodations in which they expect to stay when they arrive at their destinations (commercial facilities such as hotels and motels, peers’ homes, or public shelters). As discussed in more detail below, households’ decisions about evacuation logistics are determined by their previous experience as well as information they receive either before departure or en route to their destinations. In addition, these decisions are affected by people’s hazard proximity (in the case of hurricanes, how far they are inland or along the coast from the point of landfall) and household characteristics (e.g., how many people there are in the household, whether they have a functioning vehicle, and whether they have enough money to pay for the evacuation expenses). Authorities’ evacuation logistics involve evacuation transportation support and evacuation traffic management (Lindell and Perry 1992). Evacuation transportation support involves providing vehicles that can

122 Chapter 6 · Managing Evacuation Logistics transport those who lack personal vehicles or who have medical conditions that make travel by personal vehicles too risky. Evacuation traffic management involves coping with the fact that the roads in an ERS lack the capacity needed to meet the demand of all the vehicles that are trying to leave the risk area. As discussed in Chapter 8, authorities have two ways to manage the excess demand—either reduce peak demand or increase ERS capacity. Demand management involves either shifting part of the demand from its peak to times when the ERS capacity is not exceeded or balancing demand over all of the available paths through the ERS. Capacity management involves strategies that restore capacity (e.g., clearing incidents promptly) or increase capacity beyond its normal maximum (e.g., contraflow that diverts outbound traffic onto lanes that normally carry inbound traffic). Both demand management and capacity management are more effective when local authorities use an Emergency Operations Center (EOC) as part of an Intelligent Transportation System (ITS). The EOC helps representatives of local, state, and federal agencies coordinate the movement of threatened populations out of risk counties, through transit counties, into host counties. Risk counties are the origins of the evacuation traffic, whereas host counties are the destinations of this traffic. In between them lie transit counties through which evacuating vehicles pass. In general, transit counties contribute few, if any, additional evacuees and—similarly—provide accommodations for few, if any, evacuees.

6.1 Driver Evacuation Behavior Drivers can be understandably fearful when they are trying to escape from danger and this can affect the way they operate their vehicles. However, some researchers and public officials have mistakenly assumed that drivers will panic in an emergency evacuation—leading to dozens or hundreds of deaths in traffic accidents. Belief in this disaster myth has been perpetuated in recent books and articles, such as the claim that 15% of those affected are out of emotional control, 75% are apathetic and lack initiative, and only 10% are calm (Leach 1994, Vorst 2010). These authors also claim that 20% of those at risk refuse to evacuate in response to an imminent threat and a significant proportion of survivors refuse to evacuate after impact. Other authors have made somewhat less extreme assertions about driver behavior. For example, Alsnih and Stopher (2004) concluded that those in the risk area might not comply with evacuation notices and expressed concern about identifying the destinations to which evacuees should be directed. Moreover, Abdelgawad and Abdulhai (2009) contended that driver stress and aggression would certainly increase in mass evacuations, as would driver confusion and traffic incidents (e.g., breakdowns and

Chapter 6 · Managing Evacuation Logistics

123

collisions, Chen, Chen, and Miller-Hooks 2007). Tu et al. (2010) concluded that drivers who feel a sense of urgency violate traffic regulations, ignore control devices, and change their vehicle operation behavior (acceleration, braking, and headway) in order to pressure lead vehicles to accelerate or change lanes. In particular, decreased headways produce higher acceleration and deceleration rates, increased emergency braking, and higher variance in speed. They reported that reductions in headway (time between vehicles) and minimum gap (distance between slow or standing vehicles) significantly reduced ETEs. Although drivers might experience some degree of stress that reduces their driving effectiveness, claims of drivers’ severe emotional impairment are wholly unsupported. Disaster researchers have found that people respond adaptively, even in rapid onset disasters. Decades ago, Fritz and Marks (1954) reported that, in the immediate aftermath of a tornado strike, 19% of the victims were calm and self-controlled and 51% were mildly or highly agitated but behaved in a controlled manner. Another 13% were shocked and confused and only 2% were mildly or highly agitated and behaved in an uncontrolled manner. In general, research has found that people’s behavior can be described as boundedly rational; that is, most people respond appropriately, given the limited information they have about the situation (e.g., Mileti, Drabek and Haas 1975, Drabek 1986, Lindell and Perry 1992, Tierney, Lindell, and Perry 2001, CDRSS 2006, Lindell, Prater, and Perry 2006). Indeed, some transportation researchers such as Bish and Sherali (2013) have acknowledged that social science references such as Midlarsky (1968), Barton (1969), and Lindell and Perry (1992) indicate that, not only are people capable of rational self protection, many engage in prosocial behavior. More generally, the PADM—which was introduced in Chapter 4 to explain evacuation decisions—can also provide a useful perspective on households’ evacuation logistics. Specifically, drivers who are trying to decide which route to take can observe environmental cues (traffic queues) or receive warning messages that might vary in their content from different sources through a variety of channels. As is the case with many other types of warnings, some sources are channel bound; that is, a given channel only carries information from a single source. Using en route information as an example, a dynamic message sign (DMS) only displays information provided by a transportation department. However, other channels might provide information from multiple sources, as when information received by radio might have originated from radio station personnel (e.g., relaying observations from its own helicopter crew), the transportation department (e.g., relaying information from freeway video cameras to the radio station), or peers (e.g., calling in traffic observations to station personnel). Drivers vary in their access to different communication channels; some have working radios or mobile phones and others do not. Indeed, channel access can vary over time, as drivers can check TV before they leave their homes but cannot access this

124 Chapter 6 · Managing Evacuation Logistics channel en route unless the TV stations also post evacuation information on their websites which are, in turn, accessed through passengers’ cell phones. In addition, some types of traffic information can be channel bound, as radio can only transmit verbal descriptions and not visual images. In the context of evacuation traffic, threat perceptions can be characterized by expectations of traffic queues that delay progress toward the evacuation destination. However, the inconvenience of delay might not be the only aspect of threat that is relevant; long queues can be debilitating or fatal during the periods of high heat typically found in hurricane evacuations during the summer months. In this context, the alternative protective actions (in this case, evacuation strategies) include seeking shelter locally or diverting to a less congested evacuation route. Evacuees are likely to choose among the alternative evacuation strategies by comparing them on safety and travel time to the destination. When none of the alternatives is obviously superior to the others, or when drivers do not know any alternatives to the current strategy, information seeking strategies for evacuation traffic disruptions are likely to involve seeking confirmation of initial threat cues by obtaining additional information about current and future traffic conditions and about alternative evacuation plans (alternate departure times, routes, or destinations).

6.2 Evacuation Accommodations Most evacuees stay in one of three types of accommodations—peers’ (friends’ or relatives’) homes, commercial facilities (hotels or motels), or public shelters (which are often, but not always, operated by the American Red Cross in the US). Very few households stay in other accommodations such as vacation homes or recreational vehicles. Research to date has consistently shown that the overwhelming majority of evacuees stay with friends or relatives. These percentages were 62% in Hurricane Bret (Prater, Wenger, and Grady 2000), 70% in Bonnie (Whitehead 2003), 60% in Floyd (Cheng and Wilmot 2011), 54% in Lili (Lindell, Kang, and Prater 2011, Lindell, Lu, and Prater 2005), 61% in Katrina/Rita (Wu, Lindell, and Prater 2012b), 63% in Ike (Wu et al. 2013), and 62% in Ivan (Yin et al. 2014b). The median estimate for accommodations in peers’ homes is 62% of the evacuating households, with a range from 54% to 70%. Many fewer evacuees stay in commercial facilities—27% in Hurricane Bret, 16% in Bonnie, 32% in Floyd, 29% in Hurricane Lili, 18% in Katrina/Rita, 30% in Ike, and 22% in Ivan. The median estimate for accommodations in commercial facilities is 27%, with a range from 16% to 32%. Finally, a small percentage stay in other facilities (e.g., second

Chapter 6 · Managing Evacuation Logistics

125

homes, recreational vehicles)—9% in Bret and 5% in Floyd, but the data on this type of accommodation are limited because few evacuation surveys have asked about it. Mileti, Sorensen and O’Brien’s (1992) review of 23 evacuations concluded that fewer than 15% of evacuees go to public shelters, which is consistent with Baker’s (2000b) report that fewer than 15% of evacuees go to public shelters and the percentage tends to be closer to 5%. The latter estimate is supported by data from other hurricanes— 3% in Hurricane Bret, 6% in Bonnie, 4% in Floyd, 3% in Lili, 3% in Katrina/Rita, 2% in Ike, and 2% in Ivan. The median estimate for these latter studies is 3%, with a range from 2% to 6%. The consistency of these results suggests that the pattern of public shelter usage is significantly lower for hurricane evacuations than Mileti and his colleagues (1992) estimate for evacuations in general—probably because hurricanes provide so much forewarning that people have an adequate amount of time to make arrangements for accommodations with peers or commercial facilities. The small percentage of evacuees seeking accommodations in public shelters during hurricanes might seem to suggest that local officials in host counties need not be overly concerned about demand for public shelters except in the very largest hurricane evacuations. However, in an evacuation of 2 million people, even a public shelter percentage as low as 3% still yields a need for 60,000 beds. Moreover, a rapid onset disaster, such as a major hurricane making a late turn such as Hurricane Bret (1999) or Charlie (2004), could compress the departure time distribution and make it difficult for evacuees to reach their preferred accommodations by nightfall. Consequently, a hurricane evacuation this large might increase the percentage of people seeking public shelters because the 16–38% of the evacuees who would ordinarily seek accommodations in commercial facilities will yield 320,000–760,000 persons—requiring 128,000–304,000 beds. Such demand is likely to exceed local capacity, thus increasing the need for space in public shelters, at least in the short term until households can move farther inland to search for hotel or motel rooms. To date, there has been little research on the predictors of evacuees’ choice of accommodations, but two reports have documented the characteristics of those who use public shelters. Mileti et al. (1992) concluded that public shelters are primarily used by ethnic minorities, the poor, and later departures. Similarly, Baker (2000b) reported that hurricane evacuation shelters are used by those with low incomes, ethnic minorities, late evacuees, and those who evacuate from nonsurge areas. More recent studies have also sought to characterize those who use other accommodations. For example, Lindell, Kang, and Prater (2011) found that the choice of accommodations during Lili was unrelated to evacuation departure time, and Wu et al. (2012b) reported a similar result for Katrina and Rita. However, Mesa-Arango et al.

126 Chapter 6 · Managing Evacuation Logistics (2013) found that those traveling longer distances for Ivan were more likely to stay in commercial facilities and less likely to stay in public shelters. Wu et al. (2012b) noted a similar finding for Katrina and noted that evacuees from Louisiana’s St. Charles and Jefferson parishes were more likely than Rita evacuees from Texas counties to stay in hotels/ motels. One possible explanation for this finding is that Louisiana evacuees’ peers lived too far away for convenient travel but it is also possible that their peers lived very near to them, and, thus, also had to evacuate. Wu et al. (2013) reported that Ike evacuees with larger households were more likely to stay in commercial facilities (r = .13) or public shelters (r = .12) whereas those with higher incomes were more likely to stay with peers (r = .18) and much less likely to stay in public shelters (r = -.24). In addition, Yin et al. (2014b) extended these findings by reporting that staying with peers was negatively correlated with pet ownership (r = -.11) whereas staying in commercial facilities was correlated with distance travelled (r = .24), evacuation departure delay (r = -.10), mandatory evacuation (r = .09), household size (r = .08), high school graduate (r = -.08), and pet ownership (r = .09). Staying in a public shelter was correlated with distance travelled (r = -.17) and evacuation departure delay (r = .08).

6.3 Evacuation Destinations Evacuees’ ultimate destinations (whether in-state or out-of-state) are discussed below; their local intermediate destinations—such as gas stations, grocery stores, and banks that are visited in preparation for evacuation out of town—are discussed in Chapter 7. Research on hurricane evacuations has found substantial variation in evacuees’ destinations and distances travelled, presumably due to the availability of accommodations. Prater et al. (2000) found that all of their respondents left their home counties but only about 25% of them went farther than Brownsville (260 km/160 miles) or San Antonio (about 250 km/150 miles). Dash and Morrow (2001) reported that 19% of the households in the Florida Keys who evacuated from Hurricane Georges only evacuated to another island despite county officials’ order to evacuate to the mainland. Dow and Cutter (2002) reported that 9% of those fleeing from Hurricane Floyd evacuated incounty (at most, 80 km/50 miles), 32% evacuated in-state (at most, 400 km/250 miles), and 56% evacuated out-of-state. Whitehead (2003) reported that Hurricane Bonnie evacuees travelled an average of 286 km/178 miles and Lindell et al. (2011) reported average travel distances of 311 km/193miles in Lili but these distances ranged from 108 km/67 miles in Chambers County to 212 km/132 miles in Orange County. The

Chapter 6 · Managing Evacuation Logistics

127

average evacuation distance in Katrina was estimated to be 429 km/266 miles (Wu, Lindell, and Prater 2012b) and in Rita was estimated to be 321 km/199 mi by Wu et al. (2012b) and 319km/198 miles by Siebeneck and Cova (2008). The average evacuation distance in Ike was estimated to be 253 km/157 miles (Wu et al. 2013). There are mixed findings for the correlation of distance travelled and evacuation departure delay. Lindell et al. (2011) and Wu et al. (2012) found nonsignificant relationships. However, Wu et al. (2013) reported that distance travelled was negatively related to evacuation departure delay (r = -.19) and Yin et al. (2014b) reported an even larger correlation (r = -.40), suggesting that those who had closer destinations were able to wait longer before leaving. As the data reported above indicate, there is variation in evacuation distances across hurricanes as well as variation among evacuees for a given hurricane. The differences in evacuation distances are probably due to differences in the size and intensity of the storms, as well as the size of the evacuation zone population—all of which affect the demand for safe places to stay. In addition, the differences in evacuation distances are probably also due to differences in the proximity of safe places to stay. Thus, the greater the demand and the smaller the supply, the farther some evacuees will need to travel. In this regard, the evacuation destination data for Katrina and Rita are quite informative because they reveal three different patterns, one in Katrina and two distinctly different ones in Rita for the Sabine Study Area (SSA) and Houston-Galveston Study Area (GSA). In Katrina, only about 1% of Hurricane Katrina evacuees stayed in their own parishes, a result that contrasts sharply with the 19% in Hurricane Georges in South Florida (Dash and Morrow 2001) and the 9% in Hurricane Floyd (Dow and Cutter 2002). Moreover, only 10% of the Katrina evacuees stayed within coastal Louisiana, whereas 38% went inland to Lafayette, Baton Rouge, Central Louisiana or North Louisiana and almost all of the remainder went to Houston (13%) or other out of state locations (39%). This 52% level of out-of-state evacuees is quite similar to the 56% observed in Hurricane Floyd (Dow and Cutter 2002). In Rita, the low level of residents evacuating to other locations within their own counties (less than 1% in SSA and GSA) is similar to that in Hurricane Katrina, but there was an even lower level of Texas evacuees going to other coastal counties (1% in SSA and 5% in GSA) than in Katrina (10%). Instead, the primary evacuation destination of Rita evacuees was directly north—45% of SSA going to East Texas and 41% of GSA residents going to Central Texas. Only 13% of SSA evacuees went to Central Texas whereas 40% of GSA evacuees went to East Texas. Equal percentages of SSA and GSA evacuees went to North Texas or Dallas/Ft. Worth (15%) but 22% of SSA evacuees went out-of-state, whereas only 6% of GSA evacuees did so.

128 Chapter 6 · Managing Evacuation Logistics The differences in evacuation destinations for Georges, Floyd, Katrina, and Rita can be accounted for in part by the size of the states involved and evacuees’ proximity to state borders. The Florida Keys are over 500 miles from the Georgia state line, so out-of-state travel during Hurricane Georges would have been unlikely. South Carolina is a small state (78,000 km2) compared to Louisiana (113,000 km2), and Texas (678,000 km2). South Carolina only extends about 322 km (200 miles) inland and the inland area of this state has only two cities over 40,000 people, the largest of which is only 116,000. Thus, South Carolina appears to have little capacity to house a substantial number of coastal evacuees in commercial facilities or public shelters. By contrast, Louisiana has seven inland cities with populations over 40,000 and two of these have approximately 1 quarter million people. However, most of Central and North Louisiana is northwest of the Katrina impact area so the highways leading directly north from that area go to Mississippi (e.g., I-55). This explains why there was also a relatively high proportion of out of state evacuees in Louisiana. Finally, Texas extends inland approximately 350 miles from the Gulf coast to the Oklahoma border and has 46 cities over 40,000 population that are inland from the coast—seven of which have populations exceeding 1 quarter million. There is little difference in SSA and GSA residents’ access to Dallas/Ft. Worth and North Texas, so it is unsurprising that 17% of SSA and GSA residents evacuated to that region of the state. However, SSA residents—who are immediately adjacent to the Texas/Louisiana border—would find it easier to go to Louisiana (12% of SSA evacuees) than to Central Texas (7% of SSA evacuees) because of the heavy traffic from GSA that had an earlier start because Rita was initially expected to make landfall there. By contrast, GSA evacuees could travel north on I-45 into Central Texas (28%) and East Texas (17%). There would be little reason for GSA evacuees to go to other states (4%) unless they had friends or relatives in Oklahoma or Arkansas. In summary, regional geography explains why the percentage of Louisianans who went out of state was substantial (33%) but smaller than the percentage of South Carolinians who did so (56%). Geography also explains why the percentage of SSA evacuees from Rita going out of state (12%) was even smaller than the percentage of Louisiana evacuees from Katrina and the percentage of SSA evacuees from Rita going out of state (4%) was smaller still. Although hurricane evacuation studies provide some insights into evacuees’ destination choice processes, they do not directly address host counties’ needs for estimates of the numbers of persons that will seek accommodations in commercial facilities and public shelters. That is, Section 6.2 provided data for estimating total demand for different types of accommodations, but the studies in this section provide only limited insights into the distribution of this demand across evacuation destinations. To address this problem, Cheng, Wilmot, and Baker

Chapter 6 · Managing Evacuation Logistics

129

(2008) investigated factors associated with the selection of the location (e.g., town or zone) of the accommodations. Based on survey data for Hurricane Floyd, they developed separate multinomial logit models for the destination selected for each of the two most common accommodation types—peers’ homes and commercial establishments. For both types of accommodations, they found that increased distance from the evacuees’ homes and higher risk were negatively associated with the likelihood of selecting that location for the accommodations. For peers’ homes, a greater population and the location being in a metropolitan area increased the likelihood of selection. For commercial facilities, the number of properties at the destination and proximity to the interstate increased the likelihood of selecting a particular destination. Both models included sociodemographic characterization of the destination. Cheng, Wilmot, and Baker (2011) later used gravity models to account for Hurricane Floyd evacuees’ choices of destinations. They began with a static model to compute a preliminary estimate of the origin-destination (O-D) matrix, after which they estimated time-dependent travel demand using a sequential logit model and finally computed time-dependent O-D trip tables based on travel cost, distance from the hurricane’s forecast point of landfall, and the available accommodations at each destination. They found that the estimated model had a satisfactory fit to the data, with two peaks in the trip length distributions at about 200 and 300 miles inland from the coast. They attributed the first peak to the perceived minimum safe distance from storm conditions and the second to Atlanta’s large population (and, thus, high probability of being the home of a relative or friend) and large number of commercial facilities. They cautioned that, although the estimated model fit the Hurricane Floyd data well, it is unclear how well the results would generalize to other hurricane evacuations. Accordingly, they recommended that additional studies be conducted on other evacuations to determine whether those data would generate similar parameter estimates that could be used in predictive models.

6.4 Evacuation Travel Times Travel distances are important, but they do not tell the entire story about travelers’ inconvenience during evacuation because major portions of the trip are conducted at “stop and go” speeds that are well below normal—sometimes spending much longer times in “stop” than in “go”. Thus, travel durations are also an important aspect of evacuation. For example, average evacuation travel times were 508 minutes for Katrina, between 653 minutes (Wu, Lindell, and Prater 2012b) and 690 minutes (Siebeneck and Cova 2008) for Rita, and 295 minutes for Ike (Wu et al. 2013). An alternative estimate of traveler inconvenience

130 Chapter 6 · Managing Evacuation Logistics can be seen in the additional travel times for these evacuations; that is the amount of time in excess of the normal travel times to the travelers’ destinations. Survey respondents reported that the additional travel times for these three hurricanes were 180, 417, and 80 minutes, respectively. These correspond to travel times that were 55, 177, and 37% of the normal travel times, respectively. It is important to note that these are average additional travel times for these evacuations; drivers who left early could expect their trips to take no more than the usual duration whereas those who left late would face even longer delays. Indeed, some Rita evacuees took 15 hours more than normal to reach their destinations.

6.5 Travel Mode There are three principal travel modes used in large scale evacuations— personal vehicles, carpooling (getting rides with friends, relatives, or neighbors), and official transportation (typically buses from schools or mass transit agencies). In addition, there have been cases in which authorities planned or actually used aircraft, trains, postal vehicles (e.g., Chenault, Hilbert and Reichlin 1980) or even fire trucks for very localized flooding (Perry, Lindell, and Greene 1981).

6.5.1 Personal Vehicles Personal vehicles are the overwhelming preference for evacuees in the US; 74% in four flood evacuations (Perry, Lindell, and Greene 1981), 88% in Hurricane Bret (Prater, Wenger, and Grady 2000), 90% in Lili (Lindell, Kang, and Prater 2011), 89% in Katrina (Wu et al. 2012), 91% in Rita (Siebeneck and Cova 2008), and 87% in Ike (Wu et al. 2013). Moreover, many of those households take multiple vehicles (25% according to Dow and Cutter 2002). The average number of evacuating vehicles per household was 1.3 in the Mt. St. Helens eruption (Lindell and Perry 1992), 1.34 in Hurricane Bret (Prater, Wenger, and Grady 2000), 1.7 in Georges (Dash and Morrow 2001), 1.26 in Floyd (Dow and Cutter 2002), 1.6 in Lili with a range from 1.10–2.15 across five counties/parishes (Lindell, Kang, and Prater 2011), 1.42 in Katrina/Rita (Wu, Lindell, and Prater 2012b), 1.5 in Rita (Siebeneck and Cova 2008), and 1.25 in Ike (Wu et al. 2013). The median value for these studies is 1.38 with a range from 1.25–1.70. In addition, Lindell et al. (2011) reported that there was a range across counties of 1.10-2.15 in Lili and Siebeneck and Cova (2008) reported a range of 1.15–1.85 in Rita. One finding about vehicle use that has significant implications for the estimation of departure time distributions is the report that many

Chapter 6 · Managing Evacuation Logistics

131

households evacuating from Hurricane Rita not only took multiple vehicles, but 9% of them evacuated in groups that departed at different times—8% in two groups and 1% in three groups (Maghelal, Peacock, and Li 2016). Moreover, household splitting rates were particularly high in the area at highest risk, Galveston (17%). Overall, the average lag after the first departure was nearly 14 hours for the second group and 21 hours for the third group. The variables predicting multiple group evacuation differed from those predicting evacuation itself— greater perceived risk of failing to reach a safe destination (Odds ratio, OR = .48) and, unsurprisingly, possession of two or more vehicles (OR = 2.78). As Yin et al. (2014c) noted, the extremely wide range in evacuating vehicles per household from 1.10–2.15 in Lili is problematic for evacuation planners because this would yield a range from 1.10–2.15 million vehicles in an evacuation of 1 million households. Such a large range would yield extremely large uncertainties in ETEs. Thus, some researchers have sought to reduce this uncertainty by identifying predictors of vehicle usage, one of which is the number of registered vehicles. The percentage of registered vehicles taken in evacuation can be interpreted as a prediction of the number of evacuating vehicles per household (evph) from the number of registered vehicles (rv) using the equation evph = b rv, where b is the percentage estimated from historical data. Baker (2000b) reported a range of 65–75% and, consistent with that estimate, Wu et al. (2013) found that 62% of the registered vehicles were taken in the Ike evacuation. However, Lindell and Perry (1992) reported that evacuees only took an average of 52% of all registered vehicles in the Mt. St. Helens eruption, which might reflect a greater tendency to leave some vehicles behind in a very rapid onset disaster. There has also been research on other predictors of the number of evacuating vehicles per household. Dow and Cutter (2002) found that larger households and those with higher incomes took more vehicles. Lindell et al. (2011) replicated this finding but also found that proximity to the coast also predicted the number of vehicles taken. Wu et al. (2012b) reported that Katrina/Rita evacuees took more vehicles if they were married (r = .19), had larger households (r = .24), and had higher incomes (r = .19). Similarly, Wu et al. (2013) found that Ike evacuees took more vehicles if they were married (r = .30), had larger households (r = .28), and had higher incomes (r = .31). Yin et al. (2014a) examined these variables together with other characteristics of the evacuation and the households. They found that, in Hurricane Ivan, evph = 1.38, which is close to the median of the studies reported above and the data in their Table 2 indicate that, consistent with Baker (2000b), 65% of all registered vehicles were taken. Their correlation analysis indicated that travel distance (r = -.08), mandatory evacuation order (r = .11), nonmandatory evacuation order (r = -.06), hurricane experience (r = .09), household size (r = .10), number

132 Chapter 6 · Managing Evacuation Logistics of household members over 80 (r = -.07), number of household members 18-80 (r = .15), post graduate education (r = -.07), pet (r = .16), evacuated to friend/relative house (r = -.07), and distance to the coast (r = -.06) were significantly correlated with evph. Regression analyses indicated that only distance to destination, departure delay, post graduate education, and mobile home residence were significant predictors in one model and those variables plus number of household members over 80 years old, hurricane experience, and pet ownership were significant predictors in another model. The models differed in their treatment of the limiting factor of the number of vehicles available to the household. As noted by Lindell and Prater (2007), it is also important to identify the number of evacuating trailers because these also occupy space on the evacuation routes. If there are 1.4 vehicles and .35 trailers per household, this will increase the effective vehicle demand by 25% whereas adding .12 trailers would only increase traffic demand by a third as much. Thus, local officials need to have accurate estimates of the number of evacuating trailers per household (etph) that are likely to be taken. Baker (2000b) reported that 5% of hurricane evacuees take trailers or recreational vehicles (i.e., less than .05 etph), but there are conflicting data from Lili (.35 etph), Katrina/Rita (.12 etph), and Ike (.12 etph). No published research to date has reported any variables that predict trailer usage in evacuations.

6.5.2 Carpooling Only a small minority of households get rides with others. Baker (2000b) reported that only about 5% of households evacuating from hurricanes require assistance—most of whom obtain rides with peers. This is consistent with other data—13% in four flood evacuations (Perry, Lindell, and Greene 1981), 7% in Hurricane Bret (Prater, Wenger, and Grady 2000), 9% in Lili (Lindell, Kang, and Prater 2011), 7% in Katrina/Rita (Wu, Lindell, and Prater 2012b), and 10% in Ike (Wu et al. 2013). Wu et al. (2012b) found that a lack of vehicle access and, consequently, carpooling was more common among older unmarried residents with low education and income. This finding extends the previously recognized problem of evacuating ethnic minorities from coastal cities (Litman 2006, Wolshon 2002) by indicating that the problem of limited mobility also exists for other categories of evacuees living in suburban and rural areas. Wu et al. (2012b) also found that reliance on carpools seems to have had no adverse consequences for the carless respondents because they left no later than those who were able to take their own vehicles. Indeed, they travelled shorter distances (r = -.22), were more likely to stay in the homes of peers (r = .24), and less likely to stay in public shelters (r = -.20).

Chapter 6 · Managing Evacuation Logistics

133

6.5.3 Public Transit Few households use public transportation—an average of 13% in four flood evacuations (Perry, Lindell, and Greene 1981), 1% in Hurricane Bret (Prater, Wenger, and Grady 2000), 1% in Lili, less than 1% in Katrina/Rita, and 1% in Ike. The noticeable difference between the four flood evacuations and the five hurricane evacuations is probably due to the substantially longer distances needed to reach the evacuation destinations in the hurricane evacuations than in the flood evacuations.

6.6 Evacuation Routes The criteria authorities use to select official evacuation routes are not well documented. They appear to be based on the road’s capacity, with larger capacity roads that have a radial direction from hazardous facilities or are perpendicular from the hazard (e.g., away from the coast for hurricanes) being more likely to be designated as evacuation routes. However, the factors influencing drivers’ route choices have received greater attention. Route choice decisions occur during two time periods, pre-trip and en route, which differ in the information sources, channels, and messages that are available. Many people’s pre-trip evacuation route choices are based upon their routine experiences in taking the “normal” routes from their homes to their intended evacuation destinations. For example, many Hurricane Bret evacuees from Corpus Christi planned to seek accommodations in the San Antonio area and beyond and therefore expected to take I-37, which is typically the fastest and most convenient way to get there. Such choices implicitly assume that travel conditions will be the same during evacuation as they are during normal trips and fail to consider the additional demand that will cause trips on that route to take significantly longer than usual—the additional travel times described in Section 6.4. However, some of those who have had previous hurricane evacuation experience on the “normal” route expect severe congestion in a mass evacuation so they choose alternate routes that they expect to be less crowded. For example, Lindell et al. (2001) reported that the percentage of households expecting to take unofficial evacuation routes varied from 9% to 37% across the five Texas Hurricane Study Areas. Still other evacuees learn from authorities, news media, or peers about congestion on the “normal” route and, consequently choose other routes. Of those who set out initially on the “normal” route, some adhere to that route regardless of the level of congestion experienced whereas others divert to alternate routes because they expect that the alternates will be better. As was the case with the choice of the initial

134 Chapter 6 · Managing Evacuation Logistics route, the choice of an alternate route might be based on previous experience or information communicated from authorities, news media, or peers. Although the route choice decision process is rather simple to describe, assessing the aggregate effect of thousands of drivers’ route choices is a major challenge for evacuation managers because there are limited data available about this process. Specifically, there are only a few evacuation studies that have collected data on people’s route choices and, although there is a more extensive literature on travel route diversion in response to daily traffic congestion, there is still much to learn about this process. The most mathematically tractable model of traffic routing is the System Optimal (SO) model, which minimizes travel times for all vehicles in the network. This approach assumes that drivers would be willing to personally incur some additional travel time for the benefit of others in the system. Although analysts do not believe this assumption to be empirically valid, it could be used to establish a minimum evacuation time. A conceptually simpler model of route choice is the User Equilibrium (UE) model, which assumes that each driver selects the route that minimizes his/her individual travel time. In an everyday traffic context, drivers could obtain information about the travel times throughout the network by means of repeated experience with similar conditions. If one were only considering a choice of routes from a single intersection (including freeway on-ramps and off-ramps), a driver reaching the intersection could look in each direction and choose the one that appears to have the least congestion. This model is attractive to evacuation analysts because it simplifies the process of modeling traffic flows. Data from routine traffic disruptions, such as traffic accidents, provides some support for the UE model but indicate that it provides only an incomplete account of driver behavior. Specifically, some studies have found that, consistent with the UE model, diversion rates increase with visible congestion on the current route (Khattak, Koppelman, and Schofer 1992, Khattak, Schofer, and Koppelman 1993, Khattak and Khattak 1998) and decrease with visible congestion on alternate routes (Bonsall and Palmer 1999). However, Knoop et al. (2010) found that 50% or more of the affected drivers switched from their original route to an alternate when there was a major traffic incident but did not switch for a minor one. The authors concluded that drivers would respond in a similar manner to queues during mass evacuations. That is, “travelers will understand that the situation is considerably different and they will reroute to a route that is shorter, or more reliable as long as they succeed of getting out of the affected area in time” (p. 127). Of course, this should make it clear that route switching will occur only if travelers believe that there is an alternate route that will be less congested. It is unclear what percentage of drivers caught in such

Chapter 6 · Managing Evacuation Logistics

135

queues actually do believe there are less congested alternatives but photos of the long lines of motionless traffic in many hurricane evacuations suggest that this percentage is not large. Other research has found that the UE model is oversimplified because travelers’ likelihood of diverting from a congested route to an alternate is significantly related to their familiarity with alternate routes (Bonsall and Palmer 1999, Khattak, Polydoropoulou and Ben-Akiva 1996), especially if they are aware of multiple alternatives (Khattak, Koppelman, and Schofer 1992). In addition, diversion rates increase when drivers have longer distances to drive before reaching their destinations (Khattak, Koppelman, and Schofer 1992). Conversely, diversion rates decrease when drivers consider alternate routes to be less safe (Khattak, Koppelman, and Schofer 1992), take more time because of congestion (Khattak and Khattak 1998), or have multiple traffic stops (Khattak, Koppelman, and Schofer 1992). The importance of past experience and, thus, route familiarity can also be seen in the finding that many routine commuters were reluctant to change to an alternate route even when information was provided that showed the alternate was better (Khattak et al. 2008). To some extent, evacuation transportation managers are trying to overcome these problems by using Intelligent Transportation System (ITS) information systems such as variable messages signs (VMS) and highway advisory radio (HAR) to provide timely route guidance. However, the degree to which these systems can produce driver behavior that more closely meets the assumptions of the UE model remains to be determined because disaster evacuation studies provide both theory and data that identify conflicts with the UE model. As noted earlier, the PADM indicates that people supplement observable social and environmental cues (in this case, observation of congestion on their evacuation route) with prior beliefs (in this case, from their travel experience on the ERS during normal conditions or previous evacuations) and messages received from different sources through a variety of channels. Thus, the UE model will produce accurate predictions only to the degree that ITSs provide timely information that drivers receive, heed, comprehend, and use. Disaster evacuation studies seem to suggest that ITSs implemented to date have not reached a level of effectiveness that is fully consistent with the UE model. Indeed, one important finding is that most evacuees prefer to take interstate highways in places where they are available (Chiu and Mirchandani 2008, Dow and Cutter 2002, Lindell and Prater 2008, Prater, Wenger, and Grady 2000). For example, Prater et al. (2000) reported that 60% of Hurricane Bret evacuees reported using I-37 and Dow and Cutter (2002) reported extreme congestion on I-26 and I-95 during the Hurricane Floyd evacuation because most drivers preferred to use those routes even though other routes were available and underutilized. Lindell et al. (2001) found that Texas coastal residents’ expected evacuation routes depended upon the availability of highways that

136 Chapter 6 · Managing Evacuation Logistics would take them to their expected evacuation destinations. Forty two percent of the residents of the Houston Galveston Study Area (GSA) expected to take I-45, which was consistent with their expected destinations such as Inland GSA, Dallas, and Central Texas. However, many respondents expected to take minor routes or multiple routes (33.4%) or unofficial routes (9.4%). Only 6.8% had not planned which route to take. Unfortunately, interstate highways leading to safety are not always available. Lindell et al. (2001) found that, in the Lake Sabine Study Area (SSA), which lacks an interstate highway leading away from the coast, many residents expected to use unofficial routes (34.4%) or had not planned which route to take (6.5%). Only 18.3% planned to take I-10, presumably because it runs parallel to the coast, and the rest were scattered over seven different routes. Evacuees report a variety of reasons for their route choices. Prater et al. (2000) reported that 69% of Hurricane Bret evacuees chose their evacuation route because it was “most logical” (i.e., based on their previous experience and the available information about the current situation) whereas 3% relied on official recommendations, 4% relied on hurricane evacuation maps, none relied on news media recommendations, and 24% cited other reasons. Similarly, Zhang, Prater, and Lindell (2004) reported that only 26% of Hurricane Bret evacuees reported receiving evacuation route information from local emergency management personnel, which was only slightly higher than the percentage reporting having received information identifying the risk areas in which they were located (19%). Dow and Cutter (2002) reported that the majority of Floyd evacuees carried road maps but just half of them used these maps to determine their evacuation routes. Instead, as noted above, they relied heavily on the interstate system with which they were most familiar. Lindell et al. (2011) found that evacuees from Hurricane Lili relied much more on personal familiarity with their evacuation routes (Mean rating = 3.93 on a 1–5 rating scale) and on expectations about time, safety or convenience (M = 3.83) than on recommendations of peers (M = 2.79), the news media (M = 2.66), local authorities (M = 2.48); or written materials received in advance (M = 1.61). In general, most sources of evacuation route information were positively correlated with each other; those who consulted any source tended to consult multiple sources. However, those who chose an evacuation route based on previous experience were less likely to rely on other sources of route information. Wu et al. (2012) found that Hurricane Katrina and Rita evacuees were more likely to choose their evacuation routes on the basis of past experience with the route (M = 3.39 on a 1-5 scale) than on the basis of traffic conditions encountered en route (M = 3.20), recommendations by the news media (M = 2.28) or local authorities (M = 2.20), or written materials received in advance (M = 1.86). This study also

Chapter 6 · Managing Evacuation Logistics

137

found that people neglected to use preimpact written materials such as evacuation maps (only 36% did so), which replicates findings from Dow and Cutter (2002) and Zhang et al. (2004). The importance of past experience with evacuation routes indicates that evacuees learn from experience and, to some extent, suggests that repeated hurricane evacuations within a given area can begin to produce the development of the equilibrium traffic conditions associated with routine traffic patterns (Transportation Research Board 2010). However, this is true only for the few communities that have experienced multiple hurricane evacuations within the lifetimes of the majority of their residents. Moreover, the significance of conditions encountered en route is broadly consistent with the assumption that evacuees respond adaptively (Sheffi, Mahmassani, and Powell 1982). However, the fact that both past experience and en route conditions were used extensively poses a problem for evacuation modeling because the nonsignificant correlation between them (r = .06, Wu et al. 2012) suggests that some evacuees rely primarily on past experience whereas others rely primarily on traffic conditions encountered en route. Moreover, those who rely on these two sources have different demographic profiles (older Whites with smaller families and higher education levels vs. females with children and lower incomes, respectively) and the statistically significant demographic variables have relatively poor predictive validity (all correlations less than .10 in absolute value). One unexpected finding from the Katrina/Rita data is that evacuees who relied on personal experience had lower evacuation travel times and additional travel times. By contrast, those who relied on other sources of information—especially officials’ recommendations and traffic conditions encountered en route—tended to have greater evacuation travel times and additional travel times. These results are a bit puzzling because these three evacuation route information sources were unrelated to coastal proximity and departure timing. Reliance on past experience was associated with a greater tendency to travel shorter evacuation distances and to stay with peers rather than in commercial facilities, which suggests that they have routinized the entire evacuation process to a greater extent than other evacuees. Finally, drivers’ higher dependence on previous experience and familiarity with the evacuation route than on recommendations by officials is likely to be problematic for analysts who try to forecast (Lindell and Prater 2007), and transportation officials who attempt to regulate (Wolshon et al. 2005), evacuation traffic flows. Many emergency managers try to provide hurricane hazard awareness information, especially evacuation route information, before an evacuation is imminent. However, the available data seem to indicate that preseason hurricane information has limited utilization and that the most important method of controlling evacuation traffic is for local officials and the media to provide evacuation route information during a hurricane’s

138 Chapter 6 · Managing Evacuation Logistics approach (Zhang, Prater, and Lindell 2004). The methods that can be used to manage demand for specific evacuation routes—traffic monitoring and traffic information provision—as well as evacuees’ responses to information, are discussed in Chapter 8.

References Abdelgawad, H., Abdulhai, B. 2009. Optimal spatiotemporal evacuation demand management: Methodology and case study in Toronto. In: 88th Annual Meeting of the Transportation Research Board, Washington, DC. Alsnih, R., Stopher, P. 2004. Review of procedures associated with devising emergency evacuation plans. Transportation Research Record 1865, 89–97. Baker, E.J. 2000b. Hurricane evacuation in the United States. In Pielke, R., Jr and Peilke, R., Sr. Storms. Routledge, New York, pp. 306–319. Ballou, R.H. 1987. Basic Business Logistics: Transportation, Materials Management, Physical Distribution, 2nd ed. Prentice-Hall, Englewood Cliffs, NJ. Barton, A. 1969. Communities in Disaster. Doubleday, New York. Bish, D.R., Sherali, H.D. 2013. Aggregate-level demand management in evacuation planning. European Journal of Operational Research 224 (1), 79–92. Bonsall, P.W., Palmer, I.A. 1999. Route choice in response to variable message signs—factors affecting compliance. In Emmerink, R., Nijkamp, P. (Eds.) Behavioural and Network Impacts of Driver Information Systems. Aldershot, Ashgate, England, pp. 181–214. CDRSS—Committee on Disaster Research in the Social Sciences. 2006. Facing Hazards and Disasters: Understanding Human Dimensions. National Academy of Sciences/National Research Council, Washington DC. Chen, M., Chen, L. Miller-Hooks, E. 2007. Traffic Signal timing for urban evacuation. Journal of Urban Planning and Development 133 (1), 30–42. Chenault, W.W., Hilbert, G.D., Reichlin, S.D. 1980. Evacuation Planning in the TMI Accident. Human Sciences Research, McLean, VA. Cheng, G., Wilmot, C.G. 2011. Dynamic gravity model for hurricane evacuation planning. In: 90th Annual Meeting of the Transportation Research Board, Washington, DC. Cheng, G, Wilmot, C.G., Baker, E.J. 2008. Destination Choice Model for Hurricane Evacuation, 87th Annual Meeting of the Transportation Research Board, Washington, DC. Cheng, G, Wilmot, C.G., Baker, E.J. 2011. Dynamic gravity model for hurricane evacuation planning. Transportation Research Record 2234, 125–134. Chiu, Y-C., and Mirchandani, P.B. 2008. Online behavior-robust feedback information routing strategy for mass evacuation. IEEE Transactions on Intelligent Transportation Systems 9 (2), 264–274. Dash, N., Morrow, B.H. 2001. Return delays and evacuation order compliance: the case of Hurricane Georges and the Florida Keys. Environmental Hazards 2 (3), 119–128.

Chapter 6 · Managing Evacuation Logistics

139

Dow, K., Cutter, S.L. 2002. Emerging Hurricane evacuation issues: Hurricane Floyd and South Carolina. Natural Hazards Review 3 (1), 12–18. Drabek, T.E. 1986. Human System Responses to Disaster: An Inventory of Sociological Findings. Springer-Verlag, New York. Fritz, C.E., Marks, E. 1954. The NORC studies of human behavior in disaster. Journal of Social Issues 10, 26–41. Johnson, J.C., Wood, D.F. 1996. Contemporary Logistics. 6th ed. Prentice-Hall, Upper Saddle River NJ. Khattak, A., Koppelman, F., Schofer, J. 1992. Stated preferences for investigating commuters’ diversion propensity. In: 71st Annual Transportation Research Board Meeting, Washington, DC. Khattak, A., Polydoropoulou, A., Ben-Akiva, M. 1996. Modeling revealed and stated pretrip travel response to advanced traveler information systems. Transportation Research Record 1537, 46–54. Khattak, A.J., Pan, X., Williams, B.M., Rouphail, N.M., Fan, Y. 2008. Traveler information delivery mechanisms: impacts on consumer behavior. Transportation Research Record 2069, 77–84. Khattak, A., Schofer, J., Koppelman, F. 1993. Commuters’ en route diversion and return decisions: analysis and implications for advanced traveler information systems. Transportation Research – A 27 (2), 101–111. Khattak, J.A., Khattak, A.J. 1998. A comparative analysis of spatial knowledge and en route diversion behavior in Chicago and San Francisco: implications for advanced traveler information systems. Transportation Research Record 1621, 27–35. Knoop, V.L., Hoogendoorn, S.P., van Zuylen, H. 2010. Rerouting behaviour of travellers under exceptional traffic conditions—an empirical analysis of route choice. Procedia Engineering 3, 113–128. Leach, J. 1994. Survival Psychology. Palgrave, London. Lindell, M.K., Kang, J.E., Prater, C.S. 2011. The logistics of household hurricane evacuation. Natural Hazards 58 (3), 1093–1109. Lindell, M.K., Lu, J-C., Prater, C.S. 2005. Household decision making and evacuation in response to Hurricane Lili. Natural Hazards Review 6 (4), 171–179. Lindell, M.K., Perry, R.W. 1992. Behavioral Foundations of Community Emergency Planning. Hemisphere Press, Washington DC. Lindell, M.K., Prater, C.S. 2007. Critical behavioral assumptions in evacuation time estimate analysis for private vehicles: examples from hurricane research and planning. Journal of Urban Planning and Development 133 (1), 18–29. Lindell, M.K., Prater, C.S. 2008. Behavioral Analysis: Texas Hurricane Evacuation Study. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S., Perry, R.W. 2006. Fundamentals of Emergency Management. Emmitsburg MD: Federal Emergency Management Agency Emergency Management Institute. www.training.fema.gov/EMIWeb/edu/ fem.asp or hrrc.arch.tamu.edu/publications/books/. Lindell, M.K., Prater, C.S., Sanderson, W.G., Lee, H-M., Zhang, Y., Mohite, A., Hwang, S-N. 2001. Texas Gulf Coast Residents’ Expectations and Intentions

140 Chapter 6 · Managing Evacuation Logistics Regarding Hurricane Evacuation. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Litman, T. 2006. Lessons from Katrina and Rita: what major disasters can teach transportation planners. Journal of Transportation Engineering 132 (1), 11–18. Maghelal, P., Peacock, W.G., Li, X. 2016. Evacuating together or separately: factors influencing split evacuations prior to Hurricane Rita. Natural Hazards Review DOI: 10.1061/(ASCE)NH.1527-6996.0000226. Midlarsky, E. 1968. Aiding responses: An analysis and review. Merrill-Palmer Quarterly 14, 229–260. Mileti, D.S., Drabek, T., Haas, J.E. 1975. Human Systems in Extreme Environments. University of Colorado Institute of Behavioral Science, Boulder CO. Mileti, D.S., Sorensen, J.H, O’Brien, P.W. 1992. Toward an explanation of mass care shelter use in evacuations. International Journal of Mass Emergencies and Disasters 10 (1), 25–42. Perry, R.W., Lindell, M.K., Greene, M.R. 1981. Evacuation Planning in Emergency Management. Heath-Lexington Books, Lexington MA. Prater, C.S., Wenger, D., Grady, K. 2000. Hurricane Bret Post Storm Assessment: A Review of the Utilization of Hurricane Evacuation Studies and Information Dissemination. Texas A&M University Hazard Reduction and Recovery Center, College Station TX. Rushton, A., Oxley, J., Croucher, P. 2000. The Handbook of Logistics and Distribution Management. Kogan Page, London. Sheffi, Y., Mahmassani, H.S., Powell, W.B. 1982. A transportation network evacuation model. Transportation Research – Part A 16A (2), 209–218. Siebeneck, L.K., Cova, T.J. 2008. An assessment of the return entry process for Hurricane Rita 2005. International Journal of Mass Emergencies and Disasters 26 (2), 91–111. Tierney, K.J., Lindell, M.K., Perry, R.W. 2001. Facing the Unexpected: Disaster Preparedness and Response in the United States. Joseph Henry Press, Washinton DC. Transportation Research Board. 2010. Highway Capacity Manual. Washington, D.C.: Author. Tu, H, Tamminga, G., Drolenga, H., de Wit, J., van der Berg, W. 2010. Evacuation plan of the city of Almere: simulating the impact of driving behavior on evacuation clearance time. Procedia Engineering 3, 67–75. Vorst, H.C. 2010. Evacuation models and disaster psychology. Procedia Engineering 3, 15–21. Whitehead, J.C. 2003. One million dollars per mile? The opportunity costs of hurricane evacuation. Ocean and Coastal Management 46 (11–12), 1069– 1083. Wolshon, B. 2002. Planning for the evacuation of New Orleans. Institute of Transportation Engineers. ITE Journal 72 (2), 44. Wolshon, B., Urbina Hamilton, E., Levitan, M., Wilmot, C. 2005. Review of policies and practices for hurricane evacuation. II: traffic operations, management, and control. Natural Hazards Review 6 (3), 143–161. Wu, H-C., Lindell, M.K., Prater, C.S. 2012. Logistics of hurricane evacuation in Hurricanes Katrina and Rita. Transportation Research Part F 15 (5), 445–461.

Chapter 6 · Managing Evacuation Logistics

141

Wu, H.C., Lindell, M.K., Prater, C.S. 2015b. Strike probability judgments and protective action recommendations in a dynamic hurricane tracking task. Natural Hazards 79 (1), 355–380. Wu, H-C., Lindell, M.K., Prater, C.S., Huang, S-K. 2013. Logistics of hurricane evacuation in Hurricane Ike. In: Cheung, J., Song, H. (Eds.), Logistics: Perspectives, Approaches and Challenges. Nova Science Publishers, Hauppauge, NY, pp. 127–140. Yin, W., Murray-Tuite, P.M., Gladwin, H. 2014. A statistical analysis of the number of household vehicles used for Hurricane Ivan evacuation. Journal of Transportation Engineering 140 (12), 04014060. Yin, W., Murray-Tuite, P.M., Ukkusuri, S.V.,Gladwin, H. 2014b. An agent-based modeling system for travel demand simulation for hurricane evacuation. Transportation Research – Part C 42, 44–59. Yin, W., Murray-Tuite, P.M., Ukkusuri, S.V.,Gladwin, H. 2014c. Pre-evacuation activities for hurricane evacuation: an analysis of behavioral intentions from Miami Beach, Florida. 93rd Annual Meeting of the Transportation Research Board, Washington, DC. Zhang, Y., Prater, C.S., Lindell, M.K. 2004. Risk area accuracy and evacuation from Hurricane Bret. Natural Hazards Review 5 (3), 115–120.

Chapter 7

Evacuation Behavioral Forecasts

The previous chapters discussed various aspects of evacuation related behavior, including the identification of sociodemographic, economic, and hazard-related characteristics associated with making a particular choice of whether and when to evacuate. The knowledge gained from the statistical analyses needs to be converted into estimates of demand that are suitable for use in transportation analyses. Thus, this chapter discusses techniques that range in complexity from simple contingency tables (e.g., participation rates as a function of storm category and risk area) to the use of population synthesizers with statistical models in agent-based models. This chapter addresses the issues that need to be considered when conducting the population analyses that serve as the foundation of the evacuation transportation analyses and evacuation management plans in Chapters 8 and 9. Section 7.1 reviews some relevant aspects of scientifically defined risk area maps whereas Section 7.2 discusses alternative types of zones for communicating PARs. Section 7.3 explains how to use census data, synthesized populations, and evacuation surveys to supplement the data from research described in Chapters 4–6 in producing site-specific forecasts of evacuation compliance, shadow evacuation, departure times, travel modes, destinations, and accommodations. At this point in both practice and research, travel times are estimated using traffic simulation tools (discussed in Chapter 9) that embed assumptions about how travelers select routes.

7.1 Hazard Maps As discussed in Chapter 2, hazard maps are typically developed based on scientific and engineering understanding of the physical (and, in the case of hazardous materials, chemical) characteristics of a hazard. Chemical facilities and nuclear power plants are fixed point sources whose primary hazards are airborne, so the risk area for a given release is approximately an ellipse (assuming constant wind speed

Chapter 7 · Evacuation Behavioral Forecasts

143

and direction) and risk area boundaries for all possible releases are circles. Hazmat transportation incidents are also point sources when they occur but, as noted in Chapter 2, the boundaries of their risk areas are lines that parallel the transportation routes. Volcanoes are also point sources for ash releases but their other major hazards—lava flows and lahars act as fluids. Consequently, their risk areas are similar to those of inland floods, whose risk area boundaries tend to parallel the river but are altered by local terrain; steeper terrain on one side of the river makes the risk area boundaries closer to the normal riverbank on that side than on the other side. Tsunamis and hurricane storm surges are effectively linear sources by the time they strike the coast, so their risk area boundaries are roughly parallel to the coast but, like other fluid flows, are constrained by terrain. Wildfires also begin as point sources but tend to expand at a rate and in a direction that is primarily determined by the wind. Because wildfires can begin almost anywhere and their movement is constrained only by fuel availability, it is infeasible to define risk area boundaries for this hazard. Similarly, risk maps are typically infeasible for terrorist attacks because the hazards—chemical, biological radiological, nuclear, and explosive (CBRNE) agents—are not confined to transportation routes as is the case for hazmat transportation. Moreover, unlike fixed site and transportation hazmat releases, probable or even maximum release magnitudes cannot be identified easily in advance.

7.2 Defining Geographical Zones 7.2.1 Zones for Communication, Emergency Planning, and Protective Action Risk area maps can be very useful for guiding hazard mitigation actions such as land use regulations and building codes. This is because users, such as land use planners and property developers, can meet one-on-one to consult large-scale maps, determine if a given property is in a hazard zone and, thus, whether its use is regulated. However, the situation is quite different for communicating the need for protective actions in emergencies. In such cases, emergency managers often attempt to provide PARs to a very large audience through the news media. At best, communication via TV is limited to smallerscale maps that make it difficult for viewers to determine if they are within risk area boundaries for hazards such as hurricanes that impact large areas. As noted in Chapter 4, studies of Texas coastal residents found that many people could not identify their hurricane risk area

144 Chapter 7 · Evacuation Behavioral Forecasts even when provided with a map (Arlikatti et al. 2006, Zhang et al. 2004). Of course, the problem is even worse when trying to communicate over radio, which is largely limited to communicating risk area boundaries in terms of distance (and, in some cases, direction) from the hazard source. This information would be satisfactory if people were able to accurately judge distances but that is not always the case. For example, Mondschein, Blumenberg, and Taylor (2010) found that people were, on average, able to report reasonably accurate distances to different landmarks; there was substantial individual variation in these estimates—especially among people who did not routinely drive automobiles. Consequently, Trainor et al. (2013) made a distinction between communication zones, which are simply and clearly defined for communicating specific hazard information or protective actions to the public, and action zones, which better correspond to social variations in the communities. The action zones, used for higher resolution modeling of the population’s actions that accounts for those social variations, should be smaller than the communication zones and have boundaries such that, when grouped, align with the boundaries of the larger communication zones. This typology can be integrated with recommendations by Lindell (2013) and Murray-Tuite and Wolshon (2013b) by recognizing that most emergency communications about events that are likely (e.g., within the next 5 days), imminent (e.g., within the next 24 hours), or in progress are communicated via news media channels such as TV, radio, and newspapers. In many cases, these channels have a broadcast area that is larger than the risk area for any plausible event—especially with the availability of cable transmission and web sites – and encompass numerous communication zones. Thus, for example, people in this broadcast area who are many miles inland from scientifically assessed hurricane surge zones will be able to monitor the onset of a hurricane. To avoid unnecessary protective actions such as evacuation shadow, it is important to designate protective action zones, that identify the geographic regions in which it is possible that protective action might need to be taken, depending on circumstances of a specific event. These protective action zones should be aligned with communication zones. Among (but not within) the communication zones, the recommended protective actions may vary. Moreover, emergency managers should designate emergency planning zones that, as in the case of nuclear power plant EPZs, comprise all of the protective action zones for a particular hazard. That is, there should be multiple protective action zones that do not overlap with each other and collectively cover an entire emergency planning zone. Indeed, some nuclear power plants addressed this problem after the Three Mile Island accident by designating areas within the NRC’s 10-mile EPZ that are defined by geographical or political boundaries as Protective Action Areas or Emergency Response Planning Areas (Dotson and Jones 2005). Emergency managers can make protective

Chapter 7 · Evacuation Behavioral Forecasts

145

action zones easier to understand if they align these zones’ boundaries with political boundaries and clearly distinguishable features such as roads, rivers, and other significant landmarks (Arlikatti et al. 2006, Zhang et al. 2004). In addition, the protective action zones should be color coded for easy interpretation and protective action zone maps should be at a sufficiently large scale that viewers can identify the landmarks that define zone boundaries. In general, no zones should be completely surrounded by another zone (Wilmot and Meduri 2005), although emergency managers might plan to evacuate a trailer court during a hurricane in which the surrounding protective action zone is advised to shelter in-place. A good test of the structure of the protective action zones is whether they can easily be described in words, such as ZIP codes (Wilmot and Meduri 2005) or “town name west of the river”. There are cases in which regional planning commissions, such as the Houston-Galveston Area Council, replaced hazard maps based on storm surge risk areas with evacuation zone maps that follow ZIP code boundaries. This is an effective solution for urban areas because ZIP codes cover such small areas that they provide a good approximation to the scientifically estimated risk areas. However, ZIP codes cover very large areas in rural areas because the population density is so small. Consequently, a single ZIP code can have significant variations in topography and, thus, exposure to storm surge. Consequently, the boundary of the protective action zone would need to be drawn much farther inland than the boundary of the scientifically estimated risk area. Some communities have attempted to minimize the problems with protective action zone maps by providing neighborhood markers. For example, some jurisdictions have placed poles with colors at different heights to indicate the water height in that location that would result from storms of different categories. At least one community has placed colored bands around stop sign posts corresponding to the evacuation zone in which the sign is located.

7.2.2 Traffic Analysis Zones Emergency planning zones contain smaller traffic analysis zones—TAZs (USACE 2002), which are familiar to transportation engineers and modelers who often use them for estimating demand from or to a zone. TAZs are aggregations of census block groups, potentially with a range of four to about 100 contiguous Census blocks (Nyerges 1995). A TAZ should be relatively homogeneous in land use (Wilmot and Meduri 2005) and residential or employment population (Nyerges 1995). Land uses that generate almost no evacuation traffic, such as nature reserves, should be separated from other uses (Wilmot and Meduri 2005). Although TAZs are relatively small, Trainor et al. (2013) note that future efforts might need to consider even smaller areas if those are

146 Chapter 7 · Evacuation Behavioral Forecasts better able to capture variations in housing types, occupants’ sociodemographic characteristics, and hazard risk. Of course, the use of smaller, more socially relevant areas would depend on the degree to which greater resolution yields more accurate estimates of evacuation demand and the degree to which relevant data are available for these areas.

7.3 Producing Evacuation Demand Estimates For traffic simulation and ETE computation, evacuation demand estimates require, at a minimum, the number of vehicles from each origin, their destinations, and time they enter the road network. More details can be included if desired and if the data and models are available to the analyst. From a behavioral perspective, it is not yet known which elements of evacuation travel are decided jointly and which are considered sequentially. That is, it is unknown what proportion of evacuees begin by choosing accommodations such as peer’s home (which, in turn, implies an evacuation destination and route) or begin by choosing an evacuation route (e.g., the shortest route out of the risk area) and choose the first available accommodations found after exiting the risk area. Thus, the order of the sections below does not necessarily correspond to the order in which evacuees make their decisions. People travel from one place to another throughout the course of the day, so some of the vehicles on the road when an evacuation begins will not driven by evacuees. As mentioned in Section 5.3, in an evacuation with ample forewarning, evacuees may spend several days making trips around town purchasing supplies for evacuating (e.g., gas) but nonevacuees may also be taking trips to purchase supplies for staying in their homes until the storm passes (e.g., food, water, batteries). Even for short- or no-notice evacuations, people engage in urgent preparation activities such as gathering family members. These trips can be categorized as background traffic and preparation trips. Although some earlier evacuation research treated preparation trips as part of background traffic, these two types of travel are distinctly different. “Background traffic refers to vehicles in the network that are not part of the active evacuation” (Murray-Tuite and Wolshon 2013b), whereas preparation trips involve travel undertaken to prepare the home and household for evacuation or sheltering in-place. Background traffic (McGhee and Grimes 2006) and preparation trips are among the least well documented demand considerations in the evacuation literature, yet can have critical impacts. Depending on the type of evacuation and the characteristics of the evacuation-inducing event, background traffic may exceed evacuation traffic. The type of hazard

Chapter 7 · Evacuation Behavioral Forecasts

147

also dictates the amount of time available for preparation trips as well as which types of trips are taken. For example, in preparation for hurricanes, common preparation activities outside the home include purchasing food, gasoline, and medicine and withdrawing cash—all of which might be distributed over several days (Yin et al. 2014c). However, in no-notice events, preparation activities might be limited to reuniting the family (Liu, Murray-Tuite, and Schweitzer 2012, 2014a) and chained with the evacuation trip. The following sections discuss the importance of background traffic and preparation activities for different types of hazards, assumptions typically made, and data sources used to characterize this demand, the types of activities accounting for the majority of out-of-home preparation activities, and the potential consequences of not accounting for background traffic and preparation travel. Such consequences include overly optimistic evacuation times and failure to capture the actual traffic patterns (Liu, Murray-Tuite, and Schweitzer 2014b, MurrayTuite and Mahmassani 2004, Noltenius and Ralston 2010).

7.3.1 Background Traffic It is important to distinguish background traffic, which consists of normal travel activities that are not captured by evacuation behavior models, from shadow evacuation which, as indicated in Section 4.1.2, consists of those who evacuate even though not advised to do so. Like shadow evacuation, background traffic interferes with evacuee movement by using roadways that the evacuees may need, thus reducing road capacity available to the evacuees, and contributing to greater congestion and longer evacuation times. Depending on the type of event, background traffic may be passing through the evacuation zone at the time an evacuation notice is issued or when evacuees begin to leave the evacuation zone. This immediately affects the evacuation process and is more likely to have a significant impact for no-notice and smaller scale events. Background traffic may also be found on the roads immediately outside of an evacuation zone, causing congestion that spills back into the evacuation zone and makes it more difficult for evacuees to escape hazard impacts. Background traffic is also present at reception centers in the evacuees’ destination cities. Depending on the time of day and level of traffic, this background traffic may make it difficult for evacuees to exit their routes. Spill back effects then cause upstream congestion that affects movement along the evacuation route. Since background traffic arises from normal travel behavior, a starting place for modeling is normal traffic for the study area. In turn, an understanding of normal traffic may be developed from two sources: traffic data and local transportation planning models.

148 Chapter 7 · Evacuation Behavioral Forecasts

7.3.1.1 Traffic Data Departments of Transportation (DOTs) collect many types of traffic data such as counts of the number of vehicles at specific locations over time. These counts are archived for a number of years, allowing the agencies to develop a historical record for given times of day, days of the week, month, etc. The counts indicate the amount of background traffic that should be expected at a given location at a specific time on a typical day. Travel times and speeds are also commercially available on road segments, based on probe vehicles, a subset of the total traffic flow. DOTs often purchase these travel times and post them on dynamic message signs (DMS) to provide travelers with information that they can easily comprehend (see also Section 8.3). Travel speed has a relationship with congestion, based on fundamental traffic flow diagrams. This congestion could allow inference of the volume of traffic. Emerging technologies, such as vehicle-to-vehicle (V2V) and vehicleto-infrastructure (V2I)—collectively labeled V2X—are expected to provide additional sources of traffic information in the near future. Once DOTs have access to at least the V2I data, they may have data similar to both the commercially available probe data and location-specific counts.

7.3.1.2 Transportation Planning Models The traffic data described in the previous section, as well as other data DOTs collect periodically, are used to calibrate transportation planning models. These models can serve as the basis for evacuation simulation models and a strong indication of normal traffic that can be used as initial estimates of background traffic. As information about an evacuation notice and the event itself becomes available, some adjustment to background traffic levels should be expected. Discretionary trips (e.g., recreational or social trips) may be canceled and trip timings may be adjusted. In events with substantial forewarning, mandatory trips (e.g., school and work trips) may be canceled either because the schools or work locations closed or because the driver decided to cancel the trip. For no-notice events or as hazards with substantial forewarning get closer, background traffic may spontaneously re-route to avoid at-risk areas. When running evacuation simulations, a variety of background traffic scenarios should be considered. Two such scenarios are no background traffic and the normal amount of background traffic. Although no background traffic is an unrealistic scenario, it represents the best-case scenario and sets a lower bound on the ETE. Conversely, the normal amount of background traffic sets an upper bound on the ETE. Further analyses should include additional scenarios that reduce the background traffic as the evacuation progresses or information about the hazard spreads.

Chapter 7 · Evacuation Behavioral Forecasts

149

A final consideration may be evacuees traveling into the study area from other areas. Ideally, the study area would incorporate as much of the evacuating area as possible or at least be coordinated with the hazard-specific study efforts of neighboring jurisdictions so that a good understanding of the number of vehicles expected to travel into or through the study area can be obtained.

7.3.2 Number of Evacuating Vehicles Many of the assumptions made to develop ETEs in early studies were not well grounded in evacuation behavior research. For example, social scientists have cautioned that evacuation analyses assuming that 100% of people who are warned will evacuate with no shadow evacuation are likely to overestimate the evacuees from the evacuation zones and underestimate the geographic dispersion of evacuees (Lindell and Perry 1992; Lindell and Prater 2007). Evacuation analyses have also skipped the consideration of households as the decision making unit and based the number of evacuating vehicles on the number of registered vehicles—with assumptions ranging from 70% (USACE 2002) to 100% (Radwan, Hobeika, and Sivasailam 1985). Determining the number of evacuating vehicles requires consideration of several types of evacuees, including residents, transients or visitors, and special facility users or residents. Residents and visitors are discussed in subsections below. Special facilities, such as those listed in Table 3.3, should be analyzed separately because the mobility of their users and users’ transportation needs, time of day during which they operate, the density of the users, and the availability of sheltering in-place differ from occupants of residences and office buildings (Urbanik 2000). Many of these facilities have special evacuation plans that should be taken into consideration by adding special traffic with specific destinations (e.g., other schools, hospitals, jails) to the simulation. Buses used for transportation are larger than cars and the additional space they use on the road may be accounted for by using passenger car equivalents or specific features of simulation tools that allow multimodal traffic. For residents and tourists, determining the number of evacuating vehicles requires several decisions. First, will the household or traveling group evacuate or stay? Second, how many of the household’s personal vehicles will be taken? Third, if no personal vehicles are taken, how will the evacuees be transported—transit or in the vehicles of family member, friends, or neighbors or by other means, depending on the hazard? Fourth, if personal vehicles are taken, will they be towing trailers? Each of these questions will be addressed in turn in the following sections. In addition, each of the questions can be answered by using one of two approaches. The first approach is based on aggregate information,

150 Chapter 7 · Evacuation Behavioral Forecasts typically used in practice. That is, the question can be answered by using the most typical value such as a mean or median, or by using a distribution of values. The second approach, currently more often found in research than practice, is a microscopic approach based on a statistical model that predicts a household’s most likely value from available variables such as geographic location (e.g., proximity to the hazard source) and household characteristics (e.g., household size, household income). The selection of an approach may be based on the needs of the scenario, the available time and data, and the level of detail desired. Moreover, these two approaches may be blended as needed.

7.3.2.1 Aggregate Approach Baker (2000) reports that most of the US hurricane evacuation behavioral analyses use a combination of data types to drive the estimations. The first involves interviews with people in many locations inquiring about what they did in multiple hurricane threats to identify patterns under various conditions and to develop an empirical general response model. The second and third types involve data from intended response and single event actual response surveys. Many single event actual response surveys can be combined to generate the general response model. Combining the intended and actual response approaches helps overcome the limitations of each individual approach previously discussed in Chapter 4. In the aggregate approach, the first step is to determine the number of people/households and visitor groups that evacuate. The second step is to convert the population into vehicles, recognizing that some people are transit dependent. The final step to generating the number of evacuating vehicles is to make adjustments for trailers being towed. 7.3.2.1.1 Evacuation Participation The aggregate approach is based on estimates (based on ranges or identified patterns) of participation rates. The basic idea is to estimate the percentage of people in a zone that evacuate by multiplying the population of residents (or visitors) in that zone by the participation rate, as shown in the following equation. ERz ¼ PRz  rRz where ERz PRz rRz

is the number of evacuating residents from zone z, is the number of residents in zone z, and is the participation rate of residents in zone z.

ð7:1Þ

Chapter 7 · Evacuation Behavioral Forecasts

151

The residential population can be estimated from the most recent Census, which is a reason why it is convenient for the emergency planning zone and the protective action zones to align with Census boundaries. However, if there are compelling reasons for the protective action zone boundaries to deviate from the Census unit boundaries, Geographical Information Systems can be used to overlay the evacuation zone shapefile onto the Census unit shapefile to estimate the evacuation zone population. In addition, the population estimate may need to be adjusted based on the number of births, deaths, immigrants and emigrants if it has been a while since the most recent Census (Lindell and Prater 2007). For certain hazards, such as hurricanes, the mobile home population may be considered as a subpopulation of residents. As noted in Section 4.2.7, their vulnerability to wind damage is higher than residents in site-built homes and they typically have higher participation rates for each zone. The population of mobile home residents should be subtracted from the overall population so that they are not double counted. For transients, local convention or visitor bureau data may provide information on the number of hotel/motel rooms (Lindell and Prater 2007), but their locations may be difficult to determine with respect to protective action zones. Reasonable assumptions may be needed for the distribution of these rooms, such as the majority of them being located near the shoreline for beach areas or near downtown or tourist attractions for other areas. Assumptions about the occupancy of the rooms should be based on the evacuation scenario (see Chapter 9), with consideration of the season and day of the week. Once the total transient population is estimated, an equation analogous to (7.1) can be applied. The participation rate for tourists may be different from that of the residents. The rate could be higher as there is no evidence that tourists would be more reluctant to evacuate, but the particular hazard and transportation access (e.g., flights to/from an island vs. driving on the mainland) would have to be considered as context (Baker 2000). For advance notice events, residents are likely to start the evacuation trip or trip chain from home and transients from their hotels/motels, so Census data on nighttime population may be adequate. However, for shorter or no-notice events that occur during the day, the population is more dispersed. Some data on the daytime population distribution may be available; otherwise, transportation planning models may help develop estimates of the population in TAZs based on trips to and from the TAZ for a particular time period. When estimating the evacuation participation rate from behavioral expectations (intended response) surveys, analysts might consider the pattern noted in Section 4.1 that, although people’s hurricane evacuation expectations are accurate predictors of their actual evacuation for severe storms, they might overestimate the actual evacuation rates for weak storms. It is plausible to assume that this general pattern would also be found in response to other hazards.

152 Chapter 7 · Evacuation Behavioral Forecasts When estimating the evacuation participation rate from past evacuation experience, it is important to recognize that most of the studies reporting these data have been hurricane evacuations, which indicate that participation rates vary substantially from one storm to another and from one risk area to another within each hurricane. For example, Table 7.1 presents a selection of previously estimated participation rates, which decrease with storm category and distance from the coastline (Lindell and Prater 2007). As the table indicates, those in the emergency planning zone should be divided into at least three categories—residents (of multi-family dwellings and site-built single family dwellings), mobile home residents, and transients. Moreover, each of these categories should be subdivided into those who are transit dependent and those who can self-evacuate, although the percentage of transit dependent households is probably the same among residents and mobile home residents, and is likely to be negligible among transients in most emergency planning zones. Estimation of the transit dependent population is particularly important when determining what mobility assistance will be needed and estimating the number of personal vehicles that will be used. Because their stay in the area is temporary and a return trip is already planned, visitors are highly likely to evacuate when advised to do so. This is especially likely if the hotel/motel where they are staying has business interruption insurance and can afford to refund any prepaid lodging expenses. Analysts are advised to assume at least a 90% participation rate for this category of hurricane evacuees (USACE 2002). Evacuation participation rates for special needs population segments vary, depending upon whether the individuals are institutionalized or dispersed (see Table 3.3, for a list of institutions). Hurricane evacuation decisions for institutionalized special population segments—such as people in nursing homes, hospitals, and jails—are typically made by the facility administrators in accordance with the provisions of the evacuation annex of their local jurisdiction’s Emergency Operations Plan (EOP). By contrast, evacuation decisions for dispersed special population segments—especially people who require close supervision, are non-ambulatory, or require life support—are often made by those individuals in conjunction with their caregivers. Emergency managers are encouraged to contact organizations, such as those that provide home nursing services, that have already created and regularly update contact lists of clients within this population. The special needs population should not be double counted if a separate category is established for them. Another consideration for residents and transients is whether they are outside of the evacuation zones and, thus, are likely to be part of the evacuation shadow. Shadow evacuation rates for residents can vary widely. For hurricanes, Baker (1991) reported that this comprises

Chapter 7 · Evacuation Behavioral Forecasts

153

Table 7.1 Sample Evacuation Participation Rates for Hurricanes Risk area

Hurricane Category 1

2

3

4

5

Residents 1

46 – 70%

2

36 – 70%

3

31 – 40%

49%

4

28a – 40%b

46%a

5

26 – 40%

Non-surge zones

a

b

a

b

a

b

a

10%

b

a

85b – 88%a

a

78 – 85%

a

64%

a

54%

b

91%a

70 – 73%

a

83%

87%a

70%a, b

80%a

84%a

a

79%

82%a

90–100% when advised

90–100% when advised

a

68 – 70% a

44%

b

100%a

a

b

a

98%a

b

88%

b

20%

Mobile Home Residents 1

90%b

95%b

2

90%b

95%b

3

70%

b

90%b

70%

b

90%b

5

70%

b

90%b

Non-surge zones

50%b

70%b

Any zone

90–100% when advised

4

Transients 90–100% when advised

90–100% when advised

a Lindell and Prater 2007 b USACE 2002

20–50% of residents in low risk areas and Table 7.1 provides quantitative ranges of shadow evacuation rates by risk area and hurricane category. As noted in Section 4.1.2, these shadow evacuation rates can be strongly influenced by past experience and authorities’ warning messages. Recall that there were approximately ten times as many evacuees from the Three Mile Island nuclear power plant EPZ as would have been expected from the Pennsylvania Governor’s limited evacuation advisory (Lindell and Perry 1983) and three times as many evacuees from the Houston/Galveston Study Area during Hurricane Rita as Lindell, Prater and Wu (2002) forecast based on data such as Table 7.1. In the Hurricane Rita example, the overwhelming majority of the shadow evacuees departed from areas that were outside the officially designated emergency planning zone (the five hurricane risk areas) but inside the broadcast area. This example underscores the importance of considering this group of evacuees in evacuation analyses. The Rita shadow evacuees, numerous though they were, brought traffic to a standstill far outside the emergency planning zone, so no one in the queues were at risk from hurricane wind or storm surge. However, queues caused by evacuation shadow might

154 Chapter 7 · Evacuation Behavioral Forecasts propagate back into the emergency planning zone, causing casualties in other situations. Few studies have specifically addressed hurricane shadow evacuation for transients (Lindell and Prater 2007). Given that many visitors are on vacations that would be adversely affected by the event, it is reasonable to assume that 100% of transients evacuate regardless of their hurricane risk area (an assumption made in Lindell et al. 2002). For other hazards, proximity to the hazard source and the type and duration of disaster impacts should be considered when making assumptions for transients outside the evacuation zones. Guidance for nuclear power plants does not specifically state 100% participation by everyone in the protective action zone (Jones, Walton, and Wolshon 2011), but it seems to be implied that there will be nearly 100% compliance if there is a recommendation for evacuation in the keyhole pattern described in Chapter 2 (e.g., a 2mile radius and 5 miles downwind). In addition, there is likely to be a substantial level of shadow evacuation in crosswind and upwind directions within the 10-mile EPZ (Lindell and Perry 1992) and, as in the 1979 Three Mile Island nuclear power plant accident, a significant level of shadow evacuation outside the 10 mile EPZ. The assumptions underlying the shadow evacuation rates should be clearly stated and justified. For a nuclear power plant evacuation, Jones, Walton, and Wolshon (2011) recommend that analysts assume a shadow evacuation rate similar to that in the Three Mile Island incident—20% for areas outside the 10-mile EPZ but within 15 miles of the plants. 7.3.2.1.2 Conversion to Vehicles The next step, converting the population into vehicles, is done in two steps. The first step converts the population into households and the second step moves the households into available vehicles. To convert the population into households, the conventional approach divides the population by the average number of people per household (Lindell and Prater 2007) as in equation (7-2). The number of persons per residential household in cities and counties is available from Census sites such as http://factfinder2.census.gov/faces/nav/jsf/ pages/index.xhtml. However, evacuation analyses will yield more precise estimates if this value is determined by using data from smaller census units such as census tracts or block groups. EHHRz ¼

ERz PHHRz

where ERz

is the number of evacuating residents from zone z,

ð7:2Þ

Chapter 7 · Evacuation Behavioral Forecasts

EHHRz PHHRz

155

is the number of evacuating residential households from zone z, and is the number of persons per households for zone z.

For transients, average group size may be used in place of persons per household if the previous step calculates the number of transients. If the number of rooms is assumed to account for group size, this may be sufficient to estimate the number of transient groups. 7.3.2.1.3 Personal Vehicles The next step converts groups (households or tourist groups) into vehicles. For residential households, assumptions or models estimating the number of evacuating household vehicles are needed. Generally, “most” of the vehicles from each household will be used (Jones, Walton, and Wolshon 2011), but assumptions or scenarios are needed to produce quantitative estimates. Baker (2000) reported that his extensive hurricane evacuation surveys have found that 65–75% of the registered vehicles are taken. Moreover, as discussed in Section 6.5, previous studies indicate that there is a median of 1.38 vehicles per household, with a range of 1.10–2.15 across hurricanes and a range of 1.15–1.83 across counties within a given hurricane. In part, this substantial variation across counties arises from a statistical artifact—small samples are likely to generate more extreme low and high values. The number of evacuating vehicles per residential household is multiplied by the number of evacuating residential households to determine the number of evacuating personal vehicles from residential households as shown in equation (7.3). EVRz ¼ EHHRz  VHHRz

ð7:3Þ

where EVRz EHHRz VHHRz

is the number of evacuating residential vehicles from zone z, is the number of evacuating residential households from zone z, and is the number of vehicles used per household for zone z.

Transients are assumed to be less likely to have access to multiple vehicles than permanent residents although few studies have investigated this issue (Lindell and Prater 2007). However, prior assumptions have included one vehicle per transient household (Lindell et al. 2002), in which case the number of evacuating transient vehicles is simply the number of transient households. Travel by tour bus should also be considered in some areas (Hobeika, Kim, and Beckwith 1994). These vehicles take significantly more space on evacuation routes than do

156 Chapter 7 · Evacuation Behavioral Forecasts automobiles, but compensate for this by carrying many more passengers. This yields a smaller number of total evacuating vehicles than assuming one vehicle per hotel room. 7.3.2.1.4 Transit Dependents Not all resident households own vehicles and not all transients rent or bring their own vehicles. These evacuees are not expected to contribute to the total number of vehicles but can delay departure times when they generate trips for drivers that pick them up before departing for an evacuation destination. Census or transportation survey information can be used to identify the number of households with no vehicles available. Some of the transit dependent resident evacuees will use transit (if provided) and others will seek rides from family, friends, or neighbors. Jones et al. (2011) assumed that up to 50% of these residents will find a ride with someone they know, but more recent data suggest this is an underestimate. Specifically, data from Hurricanes Lili (Lindell et al 2011), Katrina/Rita (Wu et al. 2012), and Ike (Wu et al. 2012) yielded estimates ranging 69–90%. Those who cannot carpool will need transportation support, which often takes place in two stages. The first stage transports evacuees on buses from neighborhood collection points, such as elementary schools, to a central staging area. From the central staging area, evacuees are transported to reception centers and public shelters in locations outside the emergency planning zone. The special needs population may need more specialized transportation support than buses. Some evacuees will have medical conditions that require ambulances, so transportation support must be tailored to meet the needs of this population. It is generally recommended that a special needs population roster be developed and updated frequently. However, such rosters should be the responsibility of those organizations, such as social service and home health organizations, that generate and maintain the rosters as part of their routine business operations (Daines 1991). 7.3.2.1.5 Towed Vehicles When the hazard detection and forecast system provides advance warning and the hazard has the potential to damage boats or other potentially mobile assets, evacuees are likely to tow them. Lists of marinas and boat storage facilities should be maintained as local traffic can be expected at these locations prior to and during the evacuation. For hurricane evacuation, studies have assumed 5% of households will pull a trailer or drive a motor home (USACE 2002), whereas higher rates have been reported in Hurricane Lili (35%, Lindell et al. 2005), Katrina/Rita (12%, Wu et al. 2012), and Ike (12%, Wu et al. 2013). One of the reasons for accounting for trailers is that they are not separate evacuating vehicles but affect the

Chapter 7 · Evacuation Behavioral Forecasts

157

amount of space the households’ vehicles use on the roadway. To handle this situation, personal vehicle equivalents should be calculated.

7.3.2.2 Microscopic Approach A more disaggregate approach (more typical in research than practice for the US Hurricane Evacuation Studies) can be built upon statistical models if the analysts have survey data from a recent incident in their jurisdiction (e.g., a recent post-storm evacuation survey) or a behavioral expectations survey in which risk area residents have been asked to indicate how they would respond to different hypothetical, but realistic, scenarios. Although these are seemingly very different procedures, a recent statistical meta-analysis shows that they produce similar results, at least for stronger storms (Huang et al. 2016). Models may reflect single decisions or combinations of them, such as the decision to evacuate and the departure time for those who do evacuate (e.g., Fu and Wilmot 2004). 7.3.2.2.1 Evacuation Participation Evacuation decisions for people inside the recommended evacuation zone (compliant evacuees) or outside the recommended evacuation zone (shadow evacuees) are typically modeled using a discrete choice approach, such as the logit model, although shadow evacuees have also been based on a percentage. This discrete choice model provides the likelihood of an outcome (evacuate or stay) given a set of explanatory variables, such as those discussed in Chapter 4. The logit model takes the form shown in equation (7.4), as presented in Ben-Akiva and Lerman (1985). 0

Pn ði Þ ¼

e μβ xin 0 0 e μβ xin þ e μβ xjn

ð7:4Þ

where Pn ði Þ is the probability household (or person) n selects alternative i, μ is a scale parameter, generally assumed to be 1, β is a vector of parameter values (estimated with statistical software), xin is a set of variables, the values of some of which may be specific to alternative i, and xjn is a set of variables, the values of some of which may be specific to alternative j. Once a model is obtained that contains statistically significant and theoretically meaningful variables and parameter values, the analyst

158 Chapter 7 · Evacuation Behavioral Forecasts should apply it to the household population to determine the likelihood of each household evacuating and to obtain an estimate of the number of evacuating households. If enough data are available, separate models may be estimated for different areas, such as TAZs or action zones (Trainor et al., 2013). To apply the logit model, data representing the households must be obtained, and the data should include values for each variable in the model. Hazard-based (e.g., hurricane category) and geographic (e.g., coastal proximity) variables may also be used. Attributes specific to households needed to generate the probability that the specific household evacuates are usually unavailable from general Census data, which provides aggregate data for a Census unit. Population synthesis techniques can be used to generate a set of individual synthetic households that, in aggregate, have the same degree of variation and correlation in their sociodemographic attributes (e.g., education, income, household size) as the Census unit in which they are located. The synthetic households are created by applying statistical analyses such as correlation structure and marginal sums to different data sources (Müller and Axhausen 2010), often including the Public Use Microdata Sample (PUMS) data from the Census in the US, which includes deidentified records for a sample of individuals and housing units. Once they have been created, the sociodemographic characteristics (and other characteristics based on statistically significant and theoretically meaningful variables from the choice model) of synthetic households are then used to generate essential evacuation transportation data such as evacuation decisions, departure times, and choices of transportation mode, evacuation destination, and type of accommodations. Once the population of synthetic households (for this step and the following steps) has been created with all of necessary attributes, a statistical model such as the logit model can be applied to each synthetic household. In the case of an evacuation decision, the dependent variable is a dichotomy so a binary choice (e.g., binary logistic) model would be used (if a partial household evacuation is considered, three options—all evacuate/some household members evacuate and some stay/all stay— would require a model such as the multinomial logit model). To determine whether a given synthetic household should be considered an evacuating household, a random number between zero and one is generated and compared to the probability of a synthetic household with the specified characteristics evacuating. If the random number is less than the probability of selecting Option 1 (e.g., evacuate), Option 1 is assigned to that household. Otherwise, the other option (e.g., stay) should be considered as the outcome. Following this procedure for the entire population of synthetic households yields the number and identity of the households that should be modeled as evacuating. A similar approach could be applied to the transient and special needs populations if sufficient data are available. To date, there do not appear to have been any detailed, microscopic studies conducted on

Chapter 7 · Evacuation Behavioral Forecasts

159

these populations. In the absence of data to support this level of modeling, the aggregate approach should be applied for these groups. 7.3.2.2.2 Travel Modes Mode of transportation should be considered for all types of trips. For normal conditions, discrete choice models (e.g., multinomial logit models and nested logit models) can be used to address mode choice. This technique can also be used for evacuations when the data is available to develop them (e.g., Sadri et al. 2014). The household characteristics needed for this model should be part of the population synthesis. As noted in Section 6.5, the model should consist of multiple travel modes—such as personal vehicle, carpooling with someone else, and public transit. When there are more than two options, the probabilities should be organized into a cumulative distribution, such as that shown in Figure 7.1. As noted in the previous section, a random number should then be generated and compared to the cumulative distribution to determine which outcome to use. For example, based on Figure 7.1, a random number between 0.0 and 0.75 should be associated with personal vehicle, between 0.75 and 0.90 should be associated with obtaining a ride from others, and greater than 0.90 should be assigned to transit.

Figure 7.1 Example of Cumulative Probability for Travel Mode

160 Chapter 7 · Evacuation Behavioral Forecasts For households using personal vehicles, the next question is how many vehicles that household will take. Simple models can use the median estimate provided in Section 6.5.1 (Md = 1.38) or, for a sensitivity analysis, the range of estimates in that section (e.g., 1.25–1.70). More advanced models for determining the number of vehicles that would be used by households can be developed from the analysis of survey data by Poisson distributions (Cova and Johnson 2002) or Poisson regression models (e.g., Yin, Murray-Tuite, and Gladwin 2014). The latter models use household characteristics such as number of household members between the ages of 18 and 80 to predict the number of vehicles taken. The general procedure for applying these models is similar to that previously described for predicting other aspects of evacuation behavior such as evacuation decisions. To date, there do not appear to have been any examples of choice models used to predict whether a household tows another vehicle. In the absence of such models, the aggregate approach should be applied.

7.3.3 Accommodations and Destinations As noted in Section 6.2, the most common categories of accommodations are (in order of popularity), the homes of friends or relatives, commercial facilities (hotels/motels), and public shelters. Just as for modeling evacuation travel mode choice, the model for accommodation type choice can be developed in two different ways. The first, aggregate approach, is based on the historical record and professional judgment. The second, microscopic modeling, is based on survey data and statistical models, such as the multinomial logit and nested logit models (e.g., Mesa-Arango et al. 2013).

7.3.3.1 Aggregate Approach As indicated in Section 6.2, the aggregate approach indicates that the best estimate of the percentage of hurricane evacuees seeking accommodations in the homes of peers is 62%, with a range from 54–70%. Many fewer households stay in commercial facilities, with a median estimate of 27% and a range from 16-32%. Finally, approximately 3% stay in public shelters, with a range of 2–6%. For planning purposes, these median estimates need to be adjusted because they do not sum to 100%. The adjusted percentages for hurricane evacuations are 67, 29, and 4% for accommodations with peers’ homes, commercial facilities, and public shelters, respectively. These percentages from hurricane evacuations do not necessarily apply to other types of hazards, especially those with less forewarning and geographic areas in which the evacuating households’ peer

Chapter 7 · Evacuation Behavioral Forecasts

161

networks are also in the evacuation zone (limiting the opportunities for accommodations with peers outside the evacuation zone) and where there are few commercial facilities near the evacuation zone (limiting the number of available rooms) or the evacuating households’ have lower incomes (limiting their ability to pay for rooms). If one assumes that the percentage of evacuating households seeking accommodations in public shelters will, indeed, be the Mileti et al. (1992) estimate of 15% and that accommodations in peers’ homes and commercial facilities will be in the same ratio as in hurricane evacuations, then one would expect that the percentages for peers’ homes and commercial facilities will be 59% and 26%, respectively. As noted in Section 6.2, the estimate of 15% accommodations in public shelters is more appropriate for no- or short-notice evacuations and, thus, is about triple what would be expected for most hurricanes and other evacuations with ample forewarning. Analysts should specify the percentages for each alternative as an assumption for the entire evacuation analysis or for specific scenarios. The number of evacuating vehicles can then be multiplied by the appropriate percentages to determine the number of vehicles traveling to each type of accommodation, although analysts should make an adjustment for transit vehicles traveling to public shelters. The geographic location of the accommodations must be specified for the traffic simulation. The locations of public shelters can be specified individually, whereas the locations of commercial facilities and peers’ homes can be specified at a geographic scale that depends on the number of households in the evacuation zone and the proximity of available commercial facilities and peers’ homes. For evacuations that displace a small number of households, evacuees are likely to find accommodations in the same county or neighboring counties. However, evacuees are likely to need to travel farther for large-scale evacuations such as those initiated in response to major hurricanes. Hurricane evacuation planning studies have reported assumptions of the percent of evacuees leaving their home counties range 40-65%, depending on surge zone and hurricane category (USACE 2002). However, this is significantly lower than the 81% of surveyed evacuees who left their counties for Hurricane Ivan in 2004 and over 89% who did so for Hurricane Katrina in 2005 (Murray-Tuite et al. 2012). The Ivan and Katrina data are consistent with those from the Lindell et al. (2001) evacuation expectations survey of Texas coastal counties, which found that the percentages of households expecting to leave their multi-county study areas ranged 80–99%. Other surveys also indicate that hurricane evacuees travel long distances to find accommodations. Dow and Cutter (2002) reported that the percentages of Myrtle Beach evacuees travelling in-county (an average of 24 km/15 mi) were 50% for Diana, 42% for Bertha, 31% for Fran, and 19% for Floyd. However, the percentages travelling

162 Chapter 7 · Evacuation Behavioral Forecasts out-of-county but in-state (an average of 242 km/150 miles) were 50% for Diana, 43% for Bertha, 41% for Fran, and 57% for Floyd. Finally the percentages travelling out-of-state (an average of 402 km/250 miles) were 15% for Bertha, 28% for Fran, and 38% for Floyd. A more recent series of studies found that the average distance households travelled to their evacuation destinations was 311 km (193 miles) in Hurricane Lili (Lindell et al. 2011), 428 km (266 miles) in Katrina and 199 in Hurricane Rita (Wu et al 2012) and 253 km (157 miles) in Hurricane Ike (Wu et al 2013). These values provide a substantial range, but analysts can narrow the range by selecting values that are consistent with the hazard scenario (longer distances for stronger storms) and local conditions (longer distances for coastal communities that have less densely populated inland areas nearby). Analysts can obtain even more precise estimates of evacuation destinations from the percentages of towns or cities reported in behavioral expectations survey data for a specific study area. For example, Lindell et al. (2001) and Lindell et al. (2013) conducted surveys in which respondents reported the type of accommodations they expected to use during a hurricane evacuation and the city (or geographical region if they did not plan to stay in a city) in which their expected accommodations were located. A second approach to assigning destinations is based on traditional transportation planning models such as the gravity model. In the evacuation context, this model is based on the total number of survey respondents evacuating from origin (protective action) zones, the total respondents arriving at destination zones, and the travel times between each pair of origin and destination zones. Different gravity models should be developed based on the type of accommodations (Cheng et al. 2011, Mesa-Arango, Hasan, Ukkusuri and Murray-Tuite 2013, Ng, Diaz, and Behr 2016, Wilmot, Modali, and Chen 2006). Hazards other than hurricanes may not require distinguishing incounty and out-of-county destinations if their evacuation zones are contained entirely within the originating county. As is the case for hurricanes, public shelters for smaller scale evacuations should be opened in geographical locations that are convenient for the evacuees expected to use them. For extremely small-scale evacuations, such as a downtown block, it is reasonable to assume that those who live outside the evacuation zone will return to their homes where they expect to reunite with their families. This information may be obtainable from transportation planning models. If the hazard does not spread in a particular direction (e.g., from wind effects as discussed in Section 2.5), evacuees living inside the evacuation zones may be assumed to disperse to zones in proportion to the population of those zones. These proportions may be assumed to correspond with the locations of peers’ homes (Murray-Tuite and Wolshon 2013a).

Chapter 7 · Evacuation Behavioral Forecasts

163

7.3.3.2 Microscopic Approach The microscopic approach, which requires collecting survey data and conducting statistical analyses (e.g., discrete choice models), is more time consuming but has the advantage of identifying variables associated with the use of specific types of accommodations. Some of these variables are hazard type and severity, size of the evacuation zone, time of day the evacuation occurs, age, ethnicity, income (Mileti, Sorensen, and O’Brien 1992), as well as education and pet ownership (Whitehead et al. 2000). These variables, and their statistical significance, provide additional information beyond what is needed to estimate the percentage of the evacuees seeking accommodations in public shelters. For example, a significant regression coefficient for ethnicity suggests a need for translators; a significant coefficient for age may give an idea of types of medications that are needed; and a significant coefficient for pets indicates a need to accommodate these animals. As noted in Section 7.3.2.2.1, the models are applied to the synthetic households generated through the population synthesis and the outcome for a particular household is based on comparing a random number to the cumulative distribution. To be more specific about the locations of the accommodations, several approaches use either post-storm or evacuation expectations survey data from the study area. The gravity model is one option. Another approach, discrete choice modeling, allows the incorporation of characteristics of destination zones (and, potentially, individual and household characteristics). The choices are the destination zones, which could be numerous— one reason why this approach has not been implemented frequently. Separate models should be developed based on the accommodation type. As discussed in Section 6.3, one of the few studies using this approach found that the significant predictor variables were the distance from the origin to the destination, the population at the destination (for the friends’/ relatives’ homes model), the number of hotels/motels (for that accommodation type model), an indicator of whether the destination would also be expected to be hazardous (high wind in the case of hurricanes), the percentage of the destination population that was White, an indicator of whether the destination was in a metropolitan statistical area, and an indicator of whether the destination zone contained an interstate highway (Cheng et al. 2008). The implementation of this approach is virtually identical to the one discussed for discrete choice models of mode choice.

7.3.4 Departure Times 7.3.4.1 Preparation Activities As discussed in Section 5.3, the time taken to conduct preparation activities affects evacuees’ departure times and, in turn, the time at which they exit the evacuation zone. Examples of preparation activities

164 Chapter 7 · Evacuation Behavioral Forecasts include buying supplies to prepare the home for hazard impact (e.g., plywood for windows, sandbags around doors and basement windows), refilling medicines, withdrawing cash from banks, buying food and water, and fueling vehicles (Yin et al. 2014b). Some activities, such as packing, are conducted within the home whereas others are conducted outside the home and thus involve travel. Regardless of where the activity takes place, the time required to conduct it delays the ultimate evacuation trip and spreads evacuees’ departure times since not everyone engages in the same activities, begins each activity at the same time, or completes that activity in the same amount of time (Kang et al. 2007). In multiday evacuations (e.g., for hurricanes), some trips are undertaken by people who intend to stay within their homes, but are nonetheless preparation trips because their purchases of supplies in anticipation of the storm are not part of normal background traffic. Other preparation trips are taken by evacuees who are making purchases needed before departing for their destinations. Evacuation analysts should expect the patterns or volumes of the preparation trips to be different from normal. At the aggregate level, for example, these trips to grocery stores, gas stations, and banks may be compressed into a few days rather than spread out over an entire week as is normal. The types of preparation trips undertaken and their timing depend on the hazard. For example, evacuation preparation activities may be limited to family reunification for no-notice events because the short time until impact forces people to prioritize their preparations. Gathering family members may only be a single stop (or, at most, a few stops) along the evacuation trip. Other hazards, such as hurricanes, allow a greater amount of time in which to conduct these activities. In these cases of substantial forewarning, preparation activities could be conducted over several days before the ultimate evacuation trip or chained together with the evacuation trip. Although parts of the preparation trips may be conducted on freeways, contributing to evacuation congestion (Noltenius and Ralston 2010), locations for preparation activities are more often accessed by local streets or arterials. This requires consideration of local travel behavior, not just long distance evacuation trips. Evacuation traffic management strategies are generally designed to facilitate evacuees’ movement along freeways and highways, but sometimes these evacuation strategies interfere with or cause additional travel on local roads. For instance, ramp closures may smooth freeway traffic flow at the expense of arterial flow (Ghanipoor Machiani et al. 2013). Increased congestion and travel times on arterials could delay the completion of the preparation activities and, thus, increase ETEs. To avoid such problems, the implementation of traffic management strategies should account for these local traffic evacuation preparation activities. This can mean giving local residents hours of advance warning about the time when the evacuation management strategy will be implemented, as well

Chapter 7 · Evacuation Behavioral Forecasts

165

as carefully considering very specific aspects of strategy implementation (e.g., which ramps to close or where to start a lane reversal) and the time when the strategy will be implemented. Another issue that should be anticipated is multiple directions of travel. Many people incorrectly assume that evacuation trips are only in the outbound direction, away from the threat. However, preparation trips may begin outside the risk area (e.g., at work), be conducted in multiple directions, and end at a home that is inside the risk area. In particular, the locations of family members may create a situation in which drivers must travel toward the hazard source rather than away from it (Zimmerman, Brodesky, and Karp 2007). This is especially likely to occur early in the evacuation process for incidents with ample forewarning and could occur at any time in short- or no-notice evacuations. Erroneously assuming that all trips will be outbound will cause the analyst to miss the interactions of outbound, inbound, and cross-traffic. 7.3.4.1.1 Incorporating Family Reunification into Evacuation Models Incorporating family reunification into evacuation models requires considering different time-of-day, day-of-the-week, and season scenarios. At night and during much of the weekend, many families are already united at home but midday evacuations during the work week while school is in session usually requires consideration of the family reunification issues described above. To incorporate family reunification into an evacuation model, trip or activity chains need to be created for the evacuating population using agent- or activity-based transportation models, as indicated in Figure 7.2. The activities in these chains can include trips out of the home to pick up household members (e.g., young children or elderly parents), as well as home-based activities such as gathering pets, waiting for other family members, and packing bags before leaving.

Figure 7.2 Framework for Generating Activity Chains Population synthesis

Household & individual characteristics

Activity selection / participation (statistical model) Activity chains Activity duration (statistical model or distribution)

Activity sequencing (statistical model or distribution)

166 Chapter 7 · Evacuation Behavioral Forecasts Evacuation analysts need to identify household members’ initial locations at the time the scenario begins. There is a significant amount of personal variability in these initial locations that cannot be easily captured because these vary from day to day. Nonetheless, aggregate distributions of these initial locations can be derived from household travel surveys, travel and activity diaries, and conventional transportation planning models. Household composition assumptions can be based on surveys, Census data, or a combination of these two sources. A population synthesis approach of the type described in Section 7.3.2.2.1 can also be used to approximate the details of households in the study area. The characteristics of the synthetic household and individuals can be used with statistical models of activity participation. As with the evacuate/stay decision, models (e.g., discrete choice models), estimate the probability of a household/individual with particular characteristics (and possibly scenario and hazard characteristics) participating in a given activity. Once activity participation is determined, the household members’ locations provide origins and destinations for trips (traveling from one location to another) or they indicate stops on an activity chain (a series of activities or trips made by an individual) When engaging in an activity (e.g., picking up children), an activity duration is needed (e.g., for finding the children at school and securing them in the vehicle (Murray-Tuite and Mahmassani 2004)). This duration may be estimated with a statistical duration model or drawn from a statistical distribution based on data from behavioral expectations or actual evacuation surveys. When modeling a household’s evacuation preparations, the activities may need to be sequenced and assigned to the individuals who will conduct them. In an emergency, parents primarily respond to childrelated travel according to their everyday roles. However, if cell phone voice and text communications are not working, both parents may independently attempt to pick up a child (Liu, Murray-Tuite, and Schweitzer 2012). In assigning individuals to family reunification activities, analysts should consider people’s everyday roles and the factors identified in Table 5.1, such as car availability. For sequencing the pickups of multiple children, behavioral expectations data may be collected. Alternatively, assumptions may be made, such as pickups being conducted on a “nearest-first” rule as in Liu, Murray-Tuite, and Schweitzer (2014b) or a “farthest-from-the-home” rule. When travel modes are not explicitly considered in the assignment and sequencing of activities, they need to be identified for the activity chains, ensuring that changes in transport mode are logical from one activity to the next. For example, individuals cannot use a personal vehicle for a pickup unless they had a vehicle for the previous travel segment or could access one at their previous location (Murray-Tuite, Schweitzer, and Morrison 2012). These detailed activity travel plans can then be used in conjunction with a departure time model for the first part of the chain as inputs to

Chapter 7 · Evacuation Behavioral Forecasts

167

traffic simulation models that examine the aggregate traffic conditions. Liu, Murray-Tuite, and Schweitzer (2014b) showed that for large-scale evacuations, family reunification leads to lower rates of network clearance and potentially varied impacts on speeds. This is because reunification trips that are inbound toward the hazard source rather than outbound away from it could reduce some directional traffic impacts, such as congestion and lower speeds, by distributing outbound departures over a longer duration. 7.3.4.1.2 Hazards With Substantial Forewarning For hazards with substantial forewarning, such as hurricanes, preparation activities are conducted over multiple days and may be more focused on trip and home preparations rather than reuniting family members. There is less need to analyze family reunification trips because family members spontaneously reunify over the course of the forewarning period. Although researchers have long recognized households’ involvement in evacuation preparation, it is only recently that researchers have begun to develop detailed inventories of pre-evacuation activities in the home (e.g., Kang et al. 2007) and in local trips (Noltenius 2008, Yin et al. 2014b). 7.3.4.1.3 Prior Findings on Pre-Evacuation Activities In an online Hurricane Wilma post-evacuation survey of Key West residents, Noltenius (2008) obtained 81 responses reporting preparation activities out of 287 survey respondents. About half of the respondents participating in these activities evacuated, whereas the remainder sheltered in Key West. In this relatively small sample, the average respondent reporting trips lacked childcare responsibilities so there was little information about this activity. Moreover, she was unable to identify any statistically significant predictors of preparation trips. However, qualitative findings suggested that when multiple trips were made, most respondents made the trips on the same day and about half used trip chaining whereas the others made discrete trips. Yin et al. (2014b) later conducted a behavioral expectations survey that produced a larger (707 respondent) dataset, of whom 606 indicated they would evacuate. Activity analysis of 462 evacuating households revealed that half reported only in-home activities whereas the other half also reported out-of-home activities requiring use of the transportation network. The most commonly reported out-of-home activities were buying food, gas, and medicine and withdrawing cash. As in Noltenius (2008), the stop frequency decreased with an increase in the number of tours and activities could be combined (chained) within a tour (a set of trips that begins and ends at the same location). With the larger dataset, Yin et al. (2014b) were able to develop a binary logit model that identified variables associated with respondents’

168 Chapter 7 · Evacuation Behavioral Forecasts expectations of participating in out-of-home activities. Their model indicated that larger households, college graduates, and those driving their own vehicles increased the likelihood of engaging in these activities. However, a larger number of older people (over 64) in the household decreased the likelihood of participating in these activities. The Yin et al. (2014b) survey lacked data on activity duration but Noltenius shared the activity duration data she had collected. Based on the Noltenius data, Yin et al. (2014b) concluded that shopping trip data followed a truncated normal distribution with a mean (M) = 25 minutes, a standard deviation (SD) = 15 minutes, a minimum (Min) =10 minutes, and a maximum (Max) = 60 minutes. They also assumed a truncated normal distribution for family/friend pickups with M = 57 minutes, SD = 76 minutes, Min = 45 minutes, and Max = 120 minutes. The activity sequences reported in Yin et al. (2014b) had many permutations. Purchasing food and fuel were the most commonly reported first activities, which were often followed by buying medicine and withdrawing cash. Consistent with Wu et al. (2012b), only a small number of households indicated they would pick up friends or relatives who rely on them for evacuation. In both Noltenius (2008) and Yin et al. (2014b), driving a personal vehicle was by far the dominant transportation mode for preparation trips. The households in Yin et al.’s (2014b) study expected to start evacuation preparations early, typically within the first 2–3 days of a five-day preparation period. However, households that expected to evacuate late tended to chain their preparedness trips with their ultimate evacuation. Most households anticipated performing their out-ofhome preparation activities in a temporally concentrated manner—in a series of stops on one tour or on the same day for multiple tours. The majority of tours started in the morning. 7.3.4.1.4 Incorporating Pre-Evacuation Activity Behavior into Evacuation Modeling The general framework in Figure 7.2 can be extended to the pre-impact evacuation context by considering the organization of activities into tours (beginning and ending at the same location, such as the home) and chains that occur over multiple days. Yin et al. (2014c) provided a detailed framework for incorporating evacuation preparation activities into an evacuation simulation, the activity portion of which is summarized in Figure 7.3. The 10 steps are to (1) determine whether the household participates in any out-of-home activities, (2) determine the number of separate tours/trips, (3) select specific activities, (4) determine activity combinations, (5) select activity travel modes, (6) identify specific activity locations, (7) determine whether the activities are chained with the evacuation trip, (8) determine the day(s) on which the tours/trips are conducted, (9) determine the time of day the tour/trip begins, and (10) estimate the

Chapter 7 · Evacuation Behavioral Forecasts

169

Figure 7.3 Framework for Detailed Activity Plans for Advance Notice Events

1 Out-of-the-home activity participation (statistical model) 4 Household and individual characteristics, Evacuation decisions

2 Number of tours (statistical distribution)

3 Activity combinations (statistical distribution)

5

Activity selection (statistical distribution)

6 Activity travel mode (statistical distribution)

Activity location (statistical distribution)

7

8

Activity-evacuation chain (statistical distribution)

Tour travel day (statistical model and distribution)

10

9 Tour time of day (statistical distribution)

Activity duration (statistical distribution)

Activity plans

duration of the activities. Analysts can replace any of the steps involving distributions with more advanced statistical models. They can also resequence the steps as needs dictate or additional evidence of decision making processes and sequences is obtained. The output of the process is an activity plan for each household. These activity plans, whether obtained directly from behavioral expectations surveys, actual evacuation surveys, or through simulation processes, can be analyzed on their own or with additional simulation tools (e.g., traffic simulation). As with travel modes, destinations, and accommodations, analysts can use aggregate or microscopic approaches to assign evacuating vehicles a departure time. With either approach, the generally accepted shape of the cumulative departure time or traffic loading curve is an “S”. As discussed in Section 5.4, the curvature of the “S” depends on the rates of warning dissemination and household evacuation preparation. When both of these phenomena take place rapidly, the acceleration of the cumulative departure time curve is quite steep. However, when either warning dissemination or household evacuation preparation (or both) is slow,

170 Chapter 7 · Evacuation Behavioral Forecasts the curve can be essentially linear—as indicated in the Pittsburgh departure time data in Figure 5.5. As indicated in Figure 5.6, multi-day evacuations, such as those for hurricanes, produce “S” shape distributions that are repeated over 2–3 days, with the largest percentage of departures typically occurring in the morning. It might seem surprising that these departure times are “S” shaped even though almost all households have received a warning and have prepared to leave. The explanation is that few people are willing to leave before dawn or after dusk. Evacuees prefer morning departures since they allow most, if not all, of the travel to take place during daylight. The cumulative time scale differs for different hazards (Murray-Tuite and Wolshon 2013a). The “S” shape indicates that an evacuation usually begins slowly but rapidly gains momentum. If people become aware of the hazard before an evacuation notice is given, some people will evacuate “early”, as seen in Figure 5.6. This is likely for hurricanes and other events with substantial advance warning (Zimmerman, Brodesky, and Karp 2007). As discussed in Section 5.4, it is not uncommon for early departures to comprise up to 10% of the evacuees (Wolshon, Jones, and Walton 2010) but 15% or more is rare (Baker 2000).

7.3.4.2 Aggregate Approach A logistic function, which is most commonly used to create an “S” shaped curve, is mathematically expressed as a function of the two parameters in equation (7.5)—the half loading time H and the response rate a (Yazici and Ozbay 2008). P ðt Þ ¼

1 1 þ e aðtH Þ

ð7:5Þ

where t P(t) a H

represents the time, is the cumulative percentage of total trips generated by time t, is the parameter representing the response and affecting the slope of the curve, and is the half loading time (the midpoint of the loading curve) (Yazici and Ozbay 2008).

A short half loading time and rapid response rate generate concentrated departure times (Murray-Tuite and Wolshon 2013a). A number of evacuation analysts have assumed arbitrary parameters to generate what they considered to be plausible looking curves. For example, Radwan, Hobeika, and Sivasailam (1985) used two sets of assumed parameters (i.e., based on professional judgment rather than empirical

Chapter 7 · Evacuation Behavioral Forecasts

171

data)—a = 0.1 and H = 0.45 and a = 0.1 and H = 0.35. However, Lindell and Perry’s (1992) comparison of the warning and preparation times assumed in nuclear power plant evacuation analyses substantially underestimated the response curves obtained in actual evacuations. Another formula used to generate the curve is the cumulative Rayleigh distribution suggested by Tweedie et al. (1986). This function can be expressed as in equation (7.6). Lindell and Prater (2007) indicate that the advantage of this distribution is that the curve becomes asymmetric as the statistical mode (most frequent observation) approaches zero. That is, the lower leg to the “S” disappears and the curve begins rising almost immediately—as was the case for evacuations following an earthquake that coastal residents expect to cause a local tsunami (Lindell et al. 2015). t 2

P ðt Þ ¼ 1  e 0:5ðβÞ

ð7:6Þ

where t P(t) β

represents the time, is the cumulative percentage of total trips generated by time t, and is the mode of the distribution (scale parameter).

To apply either curve, the analyst defines the parameters and decides on an increment of time. Plugging in the first time t1 (e.g., 3 hours after an evacuation notice) into Equation 7.5 or 7.6 predicts the cumulative percent of evacuating households at that time. Then, the analyst picks the second time and substitutes this value for t2 into the appropriate equation. Subtracting the percent obtained from the calculation with t1 from that obtained in t2 indicates the percent leaving in the time interval between t1 and t2. The percentages for each time interval can be multiplied by the total number of evacuating vehicles to obtain the number of vehicles departing in a given time interval.

7.3.4.3 Microscopic Approach An alternate approach that uses evacuation expectations or actual evacuation data to directly tie departure time with hazard characteristics, evacuation notices, time of day, and (potentially) household characteristics is a sequential logit model (Fu et al. 2007). Their model, developed for hurricanes, included variables indicating a time of day (categorized as morning, mid-day, and late afternoon-evening), type of evacuation notice (voluntary, mandatory, none), hurricane wind speed, and expected time to landfall. Still other approaches based on survey data include statistical approaches for the departure time, such as the hazard-based duration

172 Chapter 7 · Evacuation Behavioral Forecasts model (Hasan, Mesa-Arango, and Ukkusuri 2013) or the Cox model (Yin et al. 2014b). These types of models allow the incorporation of accommodation types and travel modes, as well as household characteristics. Depending on the level of detail in the behavioral data, analysts can model household departure time or its two components— warning time and mobilization time—in time periods using discrete choice techniques, such as ordered probit models (Sadri, Ukkusuri, and Murray-Tuite 2013). Analysts can apply the regression coefficients from these types of statistical analyses to each synthetic household to determine its likelihood of evacuating by a given time or during a time increment. As noted in Section 7.3.2.2.1, the analyst can generate a random number that is then compared to the cumulative distribution to determine the outcome for a particular household. The outputs of either the aggregate or microscopic approaches to forecasting evacuee behavior should be combined with activity chain modeling to obtain evacuees’ overall travel plans. This chapter’s departure time analysis, in particular, defines the timeframe in which the preevacuation activities can be undertaken. Evacuees’ travel plans can then be used in conjunction with traffic simulation to generate ETEs. Some traffic simulation models can handle assignments of individual vehicles whereas other models use matrices containing the total number of trips from an origin zone to a destination zone for a given time period. Individual vehicles’ origins and destinations can be geographically aggregated into defined zones as necessary.

References Arlikatti, S., Lindell, M.K., Prater, C.S., Zhang, Y. 2006. Risk area accuracy and hurricane evacuation expectations of coastal residents. Environment and Behavior 38 (2), 226–247. Baker, E.J. 1991. Hurricane evacuation behavior. International Journal of Mass Emergencies and Disasters 9 (2), 287–310. Baker, E.J. 2000. Hurricane evacuation in the United States. In Pielke, R., Jr and Peilke, R., Sr. Storms. Routledge, New York, pp. 306–319. Ben-Akiva, M., Lerman, S.R. 1985. Discrete Choice Analysis: Theory and Application to Travel Demand. The MIT Press, Cambridge MA. Cheng, G, Wilmot, C.G., Baker, E.J. 2008. Destination Choice Model for Hurricane Evacuation, 87th Annual Meeting of the Transportation Research Board, Washington, DC. Cheng, G, Wilmot, C.G., Baker, E.J. 2011. Dynamic gravity model for hurricane evacuation planning. Transportation Research Record 2234, 125–134. Cova, T.J., Johnson, J.P. 2002. Microsimulation of neighborhood evacuations in the urban-wildland interface. Environment and Planning, A 34 (12), 2211– 2229.

Chapter 7 · Evacuation Behavioral Forecasts

173

Daines, G.E. 1991. Planning, training and exercising. In: Drabek, T.E. and Hoetmer, G.J. Emergency Management: Principles and Practice for Local Government. International City Management Association, Washington DC, pp. 161–200. Dotson, L.J., Jones, J.A. 2005. Development of Evacuation Time Estimate Studies for Nuclear Power Plants, NUREG/CR-6863, SAND2004-5900. US Nuclear Regulatory Commission, Washington DC. Dow, K., Cutter, S.L. 2002. Emerging Hurricane evacuation issues: Hurricane Floyd and South Carolina. Natural Hazards Review 3 (1), 12–18. Fu, H., Wilmot, C.G. 2004. A sequential logit dynamic travel demand model for hurricane evacuation. In: 83rd Annual Meeting of the Transportation Research Board, Washington, DC. Fu, H., Wilmot, C.G., Zhang, H., Baker, E.J. 2007. Modeling the hurricane evacuation response curve. Transportation Research Record 2022, 94–102. Ghanipoor Machiani, S., Murray-Tuite, P., Jahangiri, A., Liu, S., Park, B., Chiu, Y. C., Wolshon, B. 2013. No-notice evacuation management: ramp closures under varying budgets and demand scenarios. Transportation Research Record 2376, 27–37. Hasan, S., Mesa-Arango, R., Ukkusuri, S. 2013. A random parameter hazard based model to understand the temporal dynamics of household evacuation timing behavior. Transportation Research Part C 27, 108–116. Hobeika, A.G., Kim, C., Beckwith, R. 1994. A decision support system for developing evacuation plans around nuclear power stations. Interfaces 24 (5), 22–35. Huang, S-K., Lindell, M.K., Prater, C.S. 2016b. Who leaves and who stays? A review and statistical meta-analysis of hurricane evacuation studies. Environment and Behavior 48 (8), 991–1029. Jones, J.A., Walton, F., Wolshon, B. 2011. Criteria for the Development of Evacuation Time Estimate Studies. SAND2010-0016P, NUREG/CR-7002, US Nuclear Regulatory Commission, Washington DC. Kang, J.E., Lindell, M.K., Prater, C.S. 2007. Hurricane evacuation expectations and actual behavior in Hurricane Lili. Journal of Applied Social Psychology 37 (4), 881–897. Lindell, M.K. (2013). Evacuation planning, analysis, and management. In A.B. Bariru and L. Racz (Eds.). Handbook of Emergency Response: A Human Factors and Systems Engineering Approach. Boca Raton FL: CRC Press, 121–149. Lindell, M.K., Ge, Y., Huang, S-K., Prater, C.S., Wu, H-C., Wei, H-L. 2013. Behavioral Study: Valley Hurricane Evacuation Study for Willacy, Cameron, and Hidalgo Counties, Texas. College Station TX: Texas A&M University Hazard Reduction & Recovery Center. Available at www.hrrc.arch.tamu.edu/publica tions/research%20reports/. Lindell, M.K., Kang, J.E., Prater, C.S. 2011. The logistics of household hurricane evacuation. Natural Hazards 58 (3), 1093–1109. Lindell, M.K., Lu, J-C., Prater, C.S. 2005. Household decision making and evacuation in response to Hurricane Lili. Natural Hazards Review 6 (4), 171–179.

174 Chapter 7 · Evacuation Behavioral Forecasts Lindell, M.K., Perry, R.W. 1983. Nuclear power plant emergency warning: How would the public respond? Nuclear News, 26, 49–53. Lindell, M.K., Perry, R.W. 1992. Behavioral Foundations of Community Emergency Planning. Hemisphere Press, Washington DC. Lindell, M.K., Prater, C.S. 2007. Critical behavioral assumptions in evacuation time estimate analysis for private vehicles: examples from hurricane research and planning. Journal of Urban Planning and Development 133 (1), 18–29. Lindell, M.K., Prater, C.S., Gregg, C.E., Apatu, E., Huang, S-K., Wu, H-C. 2015. Households’ immediate responses to the 2009 Samoa earthquake and tsunami. International Journal of Disaster Risk Reduction 12, 328–340. Lindell, M.K., Prater, C.S., Perry, R.W, Wu, J-Y. 2002. EMBLEM: An EmpiricallyBased Large Scale Evacuation Time Estimate Model. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S., Sanderson, W.G., Lee, H-M., Zhang, Y., Mohite, A., Hwang, S-N. 2001. Texas Gulf Coast Residents’ Expectations and Intentions Regarding Hurricane Evacuation. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S., Wu, J.Y. 2002. Hurricane Evacuation Time Estimates for the Texas Gulf Coast. College Station TX: Texas A&M University Hazard Reduction & Recovery Center. Liu, S., Murray-Tuite, P., Schweitzer, L. 2012. Analysis of child pick-up during daily routines and for daytime no-notice evacuations. Transportation Research – A 46 (1), 48–67. Liu, S., Murray-Tuite, P., Schweitzer, L. 2014a. Uniting multi-adult households during emergency evacuation planning. Disasters 38 (3), 587–609. Liu, S., Murray-Tuite, P., Schweitzer, L. 2014b. Incorporating household gathering and mode decisions in large-scale no-notice evacuation modeling. ComputerAided Civil and Infrastructure Engineering 29 (2), 107–122. McGhee, C.C., Grimes, M.C. 2006. An Operational Analysis of the Hampton Roads Hurricane Evacuation Traffic Control Plan. Richmond VA: Virginia Department of Transportation. Mesa-Arango, R., Hasan, S., Ukkusuri, S., Murray-Tuite, P. 2013. A householdlevel model for hurricane evacuation destination type choice using Hurricane Ivan data. Natural Hazards Review 14 (1), 11–20. Mileti, D.S., Sorensen, J.H, O’Brien, P.W. 1992. Toward an explanation of mass care shelter use in evacuations. International Journal of Mass Emergencies and Disasters 10 (1), 25–42. Mondschein, A., Blumenberg, E., Taylor, B. 2010. Accessibility and cognition: the effect of transport mode on spatial knowledge. Urban Studies 47 (4), 845–866. Müller, K., Axhausen, K.W. 2010. Population Synthesis for Microsimulation: State of the Art. ETH Zürich, Institut für Verkehrsplanung, Transporttechnik, Strassen-und Eisenbahnbau (IVT). Murray-Tuite, P.M., Mahmassani, H.S. 2004. Transportation network evacuation planning with household activity interactions. Transportation Research Record 1894, 150–159.

Chapter 7 · Evacuation Behavioral Forecasts

175

Murray-Tuite, P.M., Schweitzer, L., Morrison, R. 2012. Household no-notice evacuation logistics: how well do households optimize? Journal of Transportation Safety and Security 4 (4), 336–361. Murray-Tuite, P.M., Wolshon, B. 2013a. Assumptions and processes for the development of no-notice evacuation scenarios for transportation simulation. International Journal of Mass Emergencies and Disasters 31 (1), 78–97. Murray-Tuite, P.M., Wolshon, B. 2013b. Evacuation transportation modeling: an overview of research, development, and practice. Transportation Research – Part C 27, 25–45. Murray-Tuite, P.M., Yin, W., Ukkusuri, S., Gladwin, H. 2012. Changes in evacuation secisions between Hurricanes Ivan and Katrina. Transportation Research Record 2312, 98–107. Ng, M., Diaz, R., and Behr, J. 2016. Inter-and intra-regional evacuation behavior during Hurricane Irene. Travel Behaviour and Society, 3, 21–28. Noltenius, M.S. 2008. Capturing Pre-Evacuation Trips and Associative Delays: A Case Study of the Evacuation of Key West, Florida for Hurricane Wilma. Dissertation. Department of Geography, The University of Tennessee, Knoxville TN. Noltenius, M.S., Ralston, B. 2010. Pre-evacuation trip behavior. In: Showalter, P., Lu, Y. (Eds.), Geotechnologies and the Environment: Geospatial Techniques in Urban Hazard and Disaster Analysis. Springer Science, Netherlands, pp. 395–413. Nyerges, T.L. 1995. Geographical information system support for urban/regional transportation analysis, In: Hanson, S. (Ed.), The Geography of Urban Transportation. The Guilford Press, New York, pp. 240–265. Radwan, A, Hobeika, A., Sivasailam, D. 1985. Computer simulation model for rural network evacuation under natural disasters. ITE Journal 55 (9), 25–30. Sadri, A.M., Ukkusuri, S.V., Murray-Tuite, P. 2013. A random parameter ordered probit model to understand the elapsed time between evacuation decision and actual evacuation. Transportation Research Part C 32, 21–30. Sadri, A.M., Ukkusuri, S.V., Murray-Tuite, P., Gladwin, H. 2014. Analysis of hurricane evacuee mode choice behavior. Transportation Research Part C: Emerging Technologies 48, 37–46. Trainor, J., Murray-Tuite, P., Edara, P. Fallah-Fini, S., Triantis, K. 2013. Interdisciplinary evacuation modeling. Natural Hazards Review 14 (3), 151–162. Tweedie, S.W., Rowland, J.R., Walsh, S.J., Rhoten, R.P., Hagle, P.I. 1986. A methodology for estimating emergency evacuation times. The Social Science Journal 23 (2), 189–204. USACE—US Army Corps of Engineers. 2002. Mississippi Hurricane Evacuation Study: Technical Data Report. US Army Corps of Engineers, accessed 14 October 2016 at www.coast.noaa.gov/hes/hes.html. Urbanik, T. 2000. Evacuation time estimates for nuclear power plants. Journal of Hazardous Materials 75 (2), 165–180. Whitehead, J.C., Edwards, B., Van Willigen, M., Maiolo, J.R., Wilson, K., Smith, K. T. 2000. Heading for higher ground: factors affecting real and hypothetical hurricane evacuation behavior. Environmental Hazards 2 (4), 133–142.

176 Chapter 7 · Evacuation Behavioral Forecasts Wilmot, C.G., Meduri, N. 2005. A methodology to establish hurricane evacuation zones. In: 84th Annual Meeting of the Transportation Research Board, Washington, DC. Wilmot, C.G., Modali, N., Chen, B. 2006. Modeling Hurricane Evacuation Traffic: Testing the Gravity and Intervening Opportunity Models as Models of Destination Choice in Hurricane Evacuation. No. FHWA/LA. 06/407. Louisiana State University. Department of Civil and Environmental Engieering, Baton Rouge LA. Wolshon, B., Jones, J., Walton, F. 2010. The evacuation tail and its effect on evacuation decision making. Journal of Emergency Management 8 (1), 37–46. Wu, H-C., Lindell, M.K., Prater, C.S. 2012. Logistics of hurricane evacuation in Hurricanes Katrina and Rita. Transportation Research Part F 15 (5), 445–461. Wu, H-C., Lindell, M.K., Prater, C.S., Huang, S-K. 2013. Logistics of hurricane evacuation in Hurricane Ike. In: Cheung, J., Song, H. (Eds.), Logistics: Perspectives, Approaches and Challenges. Nova Science Publishers, Hauppauge, NY, pp. 127–140. Yazici, A.M., Ozbay, K. 2008. Evacuation modelling in the United States: does the demand model choice matter? Transport Reviews 28 (6), 757–779. Yin, W., Murray-Tuite, P.M., Gladwin, H. 2014. A statistical analysis of the number of household vehicles used for Hurricane Ivan evacuation. Journal of Transportation Engineering 140 (12), 04014060. Yin, W., Murray-Tuite, P.M., Ukkusuri, S.V.,Gladwin, H. 2014b. An agent-based modeling system for travel demand simulation for hurricane evacuation. Transportation Research – Part C 42, 44–59. Yin, W., Murray-Tuite, P.M., Ukkusuri, S.V.,Gladwin, H. 2014c. Pre-evacuation activities for hurricane evacuation: an analysis of behavioral intentions from Miami Beach, Florida. 93rd Annual Meeting of the Transportation Research Board, Washington, DC. Zhang, Y., Prater, C.S., Lindell, M.K. 2004. Risk area accuracy and evacuation from Hurricane Bret. Natural Hazards Review 5 (3), 115–120. Zimmerman, C., Brodesky, R., Karp, J. 2007. Using Highways for No-Notice Evacuations: Routes to Effective Evacuation Planning Primer Serie, FHWAHOP-08-003. Federal Highway Administrartion, Washington DC.

Chapter 8

Strategies for Managing Evacuation Demand and Capacity

Evacuation management strategies can be used on both sides of the evacuation supply and demand equation, to influence local authorities’ supply of space in the ERS and/or evacuees’ demand for that space. As discussed in Chapter 4, evacuation messages and their delivery channels are critical influences on the generation of demand during emergencies. Thus, one strategy for reducing unnecessary demand involves prior education and clear emergency communication to reduce shadow evacuation— either by encouraging sheltering in-place or the continuance of normal activities. Furthermore, clearly identifying the evacuation zone encourages background traffic to avoid the risk area, thereby facilitating the evacuation of those at risk. Another method of demand management involves demand phasing. This method has received attention in the optimization and simulation fields and could reduce evacuee travel time. In practice, demand phasing could be implemented by giving a series of evacuation notices. However, officials tend to be hesitant to implement this approach and there is a question of whether people would comply with requests to delay their departures. The potential problem is that the well-documented phenomenon of geographical evacuation shadow will have a temporal analogue in which people begin their evacuations before it is “their turn”. These issues and methods are discussed in more detail in Section 8.1. From the supply side, evacuation management strategies can be applied to freeways, limited access highways, and arterial roadways. These strategies are intended to increase the roadway capacity in the outbound direction, smooth traffic flow, or otherwise reduce congestion—and thus evacuation time. Capacity enhancement may also take the form of additional buses that provide more evacuation opportunities to limitedmobility evacuees. This chapter examines emerging knowledge and practices for capacity management through the closure of roads and ramps, the incorporation of shoulders and other lanes for use as evacuation routes, the retiming of traffic signals, and the use of contraflow lanes or facilities. The description of the popular contraflow strategy covers its recent development and evolution; operational characteristics; design, operation and management; and a variety of other related issues that affect its safety. This discussion also addresses drivers’ understanding of contraflow, its accessibility and enforcement,

178 Chapter 8 · Strategies for Evacuation Management and its initiation and termination. For arterials, strategies typically focus on modifying operations at intersections, such as turning restrictions, modified signal timing, and alternative intersection control (e.g., police directed). These strategies are described in Section 8.2. Traffic management systems that can monitor the conditions are discussed in Section 8.3, and Section 8.4 outlines potential future systems.

8.1 Demand Management Strategies 8.1.1 Pre-impact Hazard Education and Communication 8.1.1.1 Hazard Education Program Overview There are six basic functions that should be addressed in a community risk communication program (Lindell and Perry 2004). These are strategic analysis, operational analysis, resource mobilization, program development, program implementation for the continuing hazard phase, and program implementation for the escalating crisis and emergency response phases. The first function, strategic analysis, lays the foundation for later risk communication activities. Thus, the strategic analysis involves conducting analyses to identify hazards and locations at risk, as well as examining the community to identify important characteristics such as its ethnic composition and communication channels. In addition, hazard managers should work with community agencies to identify the community’s prevailing perceptions of environmental hazards and hazard adjustments and set appropriate goals for the risk communication program. Operational analysis seeks to identify appropriate hazard mitigation and emergency preparedness actions that can reduce the need for evacuations and to implement them more effectively when the need arises. In addition, operational analysis should engage households and businesses in working with the community agencies to encourage these hazard adjustments. Hazard managers should also work with community agencies to identify available risk communication sources and channels in the community, as well as to identify the differences among audience segments in terms of their access to and preferences for different types of media (e.g., radio) and specific channels (e.g., specific radio stations) within each medium. During resource mobilization, hazard managers should obtain the support of senior management by ‘‘selling’’ the importance of assessing community hazard vulnerability and identifying hazard mitigation, emergency preparedness, emergency response, and disaster recovery as effective solutions. They should also enlist the participation of government agencies, businesses, and non-governmental organizations in a collaborative strategy to coordinate all of the community’s risk communication programs. Hazard managers should also work with these organizations

Chapter 8 · Strategies for Evacuation Management

179

to develop collaborative relationships with the news media and with neighborhood associations and service organizations. During program development for all phases, hazard managers should acquire the facilities and equipment they need to warn vulnerable population segments more rapidly and completely. In addition, they should staff, train, and exercise a crisis communications team and establish procedures for maintaining an effective communication flow in an escalating crisis and during emergency response. Hazard managers should also work with community organizations to develop a comprehensive risk communication program that presents information about hazards and hazard adjustments in a form that attracts attention and is easily understood and retained. They should also encourage these organizations to use informal communication networks in the community. Hazard managers should also establish procedures for obtaining feedback from the news media and the public. Program implementation for the continuing hazard phase involves building source credibility by increasing perceptions of expertise and trustworthiness and using a variety of channels to disseminate hazard information. Hazard managers should also describe community-level hazard adjustments being planned or implemented—such as land use regulations or building codes. They should also work with local organizations to describe feasible household hazard adjustments risk area residents can take to protect themselves. Finally, hazard managers should evaluate risk communication program effectiveness by measuring the degree to which the program has achieved its objectives. Program implementation for the escalating crisis and emergency response phases involves working with community officials to classify the situation in terms of its severity and activate the crisis communication team promptly so that its members can make all appropriate contacts. Hazard managers should determine the appropriate time to release sensitive information by developing procedures that define when information is to be released and select the communication channels appropriate to the situation. Hazard managers also need to provide timely and accurate news releases for the public that are supplemented by fact sheets containing basic background information that is appropriate to any incident. Hazard managers should maintain source credibility with the news media and the public by being honest about what is and is not known and state that they do not know the answer to a question when this is the case. However, they should make a commitment to answer any unresolved questions at a later time. Last, they should evaluate performance through post-incident critiques involving all members of the crisis communication team.

8.1.1.2 Hazard Awareness Program Content and Channels One of the most important elements of a hazard education program is to identify the hazards to which the community is exposed and, especially,

180 Chapter 8 · Strategies for Evacuation Management map the geographical areas that are at different levels of risk. Of course, as indicated in Chapter 2, some hazards—such as hurricane surge, tsunamis, and inland floods—are more amenable to mapping than others—such as wildfires and tornadoes. Moreover, the scientifically estimated risk areas should be converted to protective action areas that are defined by easily recognizable boundaries such as streets, geographic features (e.g., rivers), political jurisdiction boundaries, or postal codes. This information should be supplemented with relevant information about each hazard such its environmental cues, speed of onset, scope and duration of impact, and appropriate protective actions. The purpose of this additional information is to inform people about nonobvious aspects of these hazards. For example, it is important to note that tsunami arrival might be indicated by a trough, a sudden recession of the shoreline, rather than a wave, and that additional waves can continue to arrive for hours after the first one. It would also be important to inform people that an adult can be knocked down by as little as 6 in of fast flowing water and that a wildfire’s leading edge can spread faster than people can run (approximately 13 mph/20 kph). The hazard awareness program should also provide information about hazard mitigation and emergency preparedness actions. In particular, land use practices can be used to limit development density in hazard prone areas and building construction practices can be used to increase structural resilience to hazard forces (Lindell et al. 2006). For example, developers can reduce flood risk, and thus the need to evacuate, by constructing buildings on the upland portions of land parcels and leaving the lower portions as open space. In addition, homeowners can reduce their wildfire vulnerability by creating defensible spaces around their houses and outbuildings. The hazard awareness program should also provide people with information about community warning systems such as those listed in Table 3.1. In addition, they should include information about how to register with any local emergency alert and notification system that can send warnings to cell phones via voice, text, or email. The hazard awareness materials should also encourage people to develop a household protective action plan. For hazardous materials releases, this would involve deciding whether the home is sufficiently weathertight that shelter in-place would be feasible. Similarly, households in wildfire areas should decide in advance whether to evacuate early or acquire the resources to stay and defend because late evacuations have a much higher casualty rate. Regardless of the hazard, households should decide in advance how to reunite if members are separated during the day, when the adults are working at different locations and the children are in different schools. Hazard awareness program content should be disseminated through as many channels as possible. Emergency managers often give talks and hand out hazard awareness materials at meetings of business organizations (e.g., Kiwanis, Rotary), neighborhood associations, and community events such as farmers’ markets and county fairs. They also provide these materials in

Chapter 8 · Strategies for Evacuation Management

181

libraries, in utility bills, and special inserts of local newspapers. Some particularly inventive emergency managers have collaborated with local grocery chains to print hurricane awareness information on paper shopping bags. Public outreach and education can help prepare the general public for an emergency and the information that may be communicated during the emergency. This education campaign should include evacuation routes and procedures, emergency provisions, and testing of the communication systems (Pretorius et al. 2006).

8.1.2 Hazard Warnings and Public Information The principal way that authorities can motivate compliance with PARs is to ensure that people have the information they need to accurately assess the threat and the resources they need to implement the recommended actions. Some of the needed information can be disseminated in hazard awareness programs before an incident occurs whereas other information is disseminated in warning messages. As Lindell’s (2017) recent summary of the warning literature concluded, officials should disseminate warning messages that identify an authoritative warning source; describe the hazard (and its consequences, if these are not already known to those at risk), the areas that will be affected, and the time that hazardous conditions will occur; and provide a PAR (Mileti and Sorensen 1990). Messages should also address whether there are high risk structures, such as mobile homes, that require protective action even though structures around them are safe or high risk population segments, such as pregnant women and preschool children, that should take protective action even though other people in their vicinity are safe. Moreover, if hazard impact is uncertain, warning messages should indicate how probable the event is (Drabek 1999). In addition, the warning message should identify sources of additional information and provide information about sources of assistance for those who need help in evacuating (Lindell and Perry 2004). If the recommendation is to shelter in-place for a hazmat release, the warning should explain that this means to go indoors; close all doors and windows; turn off heating, ventilation, and air conditioning (HVAC) systems; and—if time and materials are available—tape the edges of doors and windows to seal them more completely (Sorensen et al. 2004). In addition, people should be instructed to go outside and open their buildings after they receive an all clear signal so any hazmat that has infiltrated the structure can dissipate (Lindell and Perry 1992). In the case of evacuation, the information should include a safe route of travel and the location of a safe destination that has adequate accommodations (Perry et al. 1981). In addition, the Federal Highway Administration (2007) expands this list to include information about the type of evacuation notice (e.g., voluntary, recommended, or

182 Chapter 8 · Strategies for Evacuation Management mandatory), as well as the time the evacuation should begin and how long it is expected to last. Mileti and Peek (2000) also recommend that warning messages should be clear (stated in words that people can understand), specific, accurate, stated authoritatively, and consistent (both within a message and between messages). Finally, the warnings should be long enough to provide all of the relevant information but short enough to be readily understood, and they should be repeated so people can pick up any information they missed in previous transmissions. Any lack of consistency in the evacuation notices can cause confusion, delay decision making, and lead to unnecessary evacuation or at-risk people remaining behind. The traffic generated by shadow evacuees can impede evacuation for evacuees from higher risk areas, illustrating the importance of telling those who do not need to evacuate to stay (Vasconez and Kehrli 2010). The protective action information should be widely distributed, through numerous outlets and in multiple languages. The messages also need to be repeated and consistent. Emergency management agencies and related agencies should use designated spokespersons to provide timely and accurate messages. Agencies should have a media outreach plan that includes a schedule for media interviews (Federal Highway Administration 2007). During evacuations, regular media briefings should identify what information about the warnings and evacuation logistics are being repeated, as well as what information is new for that briefing. The locations of shelters and their current status (full or not) should be part of this information. Hotlines for this information should be established. As shelters fill, new ones should be opened and this updated information needs to be communicated to evacuees. This information may be communicated through the media, websites, and hotlines, as well as through transportation related communication systems, such as variable message signs (VMS or DMS), 511 messages, field staff, and welcome centers and rest areas. Transportation information should include the evacuation routes and real-time traffic and road conditions (Federal Highway Administration 2007). This information should be continually updated and communicated to the public through the methods previously mentioned as well as in-vehicle navigation systems, radio traffic reporting companies, trucking and automobile associations (Federal Highway Administration 2007), and traffic apps, among other en route communication systems. However, there are cases in which much or even most of this information is unnecessary. Sometimes, mere mention of the hazard is sufficient if people already know the remaining information. For example, many coastal residents of American Samoa could infer from hearing that a tsunami was coming what would be the consequence of remaining in their homes, what areas were likely to be affected, that the hazard could arrive soon, and that they should immediately evacuate to high ground (Lindell et al. 2015). Indeed, even detection of environmental cues can be

Chapter 8 · Strategies for Evacuation Management

183

sufficient because some of the Samoans knew that an earthquake could cause a tsunami and evacuated because the ground shaking signaled the tsunami threat.

8.1.2.1 Motivating Timely Departures There are two fundamental reasons why evacuees delay their departure after receiving a warning—logistical preparation and psychological preparation. As noted in Section 7.3.4, logistical preparation comprises a number of in-home tasks such as packing the items needed while gone, protecting property from storm damage (e.g., boarding windows), shutting off utilities, and securing the home. Logistical preparation also includes out-of-home tasks such as returning from work, picking up persons who will evacuate with the household, purchasing property protection materials, gas, medicines, and food, and withdrawing cash. Finally, logistical preparation is also defined by developing a strategy for responding to the threat by identifying a safe destination, suitable accommodations, mode of transportation (usually personal vehicles), and route of travel. Consequently, people who have already completed some of the logistical preparation tasks will be able to leave earlier than those who have not. By contrast, psychological preparation is defined by collecting information about the hazard and developing a consensus within the household that the threat has reached an unacceptable level that requires evacuation. Consistent with Czajkowski’s (2011) model, most people monitor the uncertain escalation of a threat and wait as long as they think it is possible before evacuating. To some extent, people’s departure times are determined by their own assessments of the amount of time it will take to evacuate before hurricane landfall. Thus, psychological preparation is significantly influenced by information, such as hurricane forecasts. If a forecast changes to indicate that landfall will occur sooner than previously anticipated, people will begin to evacuate earlier than otherwise planned. However, people’s departure times are also determined by authorities’ assessments of the amount of time it will take to evacuate before hurricane landfall. These authorities’ assessments influence public response primarily through evacuation notices and instructions provided by local officials. Thus, officials can affect the distribution of departures by the timing of evacuation notices and how they word the notices and related announcements. Evacuation notices for no-notice events usually include messages that communicate the urgent need for immediate departure, and the resulting response curves reflect that information. The research literature clearly indicates that the rate of evacuation departures is more strongly related to the perceived imminence and urgency of the threat than of the demographic characteristics of the risk area population. The prevalence of multi-day evacuations stems from good forecasts and a precautionary approach by public safety officials, particularly in major

184 Chapter 8 · Strategies for Evacuation Management hurricanes. However, evacuation notices cannot always be issued early enough to provide multiple days in which to evacuate. Even hurricanes sometimes make major changes in their tracks that lead to last minute evacuation warnings, as was the case in Hurricane Bret (1999) and Charley (2004). At the other extreme, flash floods and hazardous material incidents will rarely, if ever, provide the amount of forewarning that produces protracted evacuations. As the difference between Figures 5.5 and 5.7 indicate, evacuations of areas near the incident scene will need to be rushed to completion following issuance of evacuation notices, and the duration of evacuations will be much shorter than during hurricanes. From a practical operational perspective, demand management can be accomplished in a number of different ways, including reducing background traffic, reducing shadow evacuation, recommending sheltering in-place, and using “phased” (sometimes called staged or sequenced) evacuation. Reducing background traffic requires careful and effective communication so that people who do not need to be on the road in a hazardous area can cancel or reroute their trips. Such action falls primarily within the purview of emergency management agencies, which define this strategy as access control (Lindell & Perry 1992). Although access control is routinely practiced by police and transportation personnel for planned (e.g., athletic events) and unplanned (e.g., fires and multi-vehicle automobile accidents) events, it has rarely been addressed in the transportation literature. Reducing shadow evacuation can be accomplished through prior education and clear communication during an incident. However, since a household’s decision to evacuate is primarily based on its members’ risk perceptions, convincing people to remain rather than evacuate can be difficult. Sheltering in-place is, indeed, safer than evacuation for some disaster types (e.g., tornados, some chemicals), especially when there is a significant risk of being overtaken by the hazard before exiting the impact area. In such cases, people risk “sheltering” in their cars rather than in a building that is structurally stronger (in the case of tornadoes) or has a lower air exchange rate (in the case of toxic chemical releases). However, the protective advantage of sheltering in-place needs to be clearly communicated to the public—preferably prior to, but as a last resort, during an incident.

8.1.2.2 Phased Evacuation The basic idea of phased evacuation is to reduce the peak demand by spreading the evacuation demand over time, which is accomplished by evacuating different geographical areas or social groups at different times. This spreading of demand has the potential to reduce congestion and individually experienced travel times. To implement the strategy in practice, evacuees need to be able to identify themselves as members of specific groups. Examples of easily self-identifiable groups are residents and tourists. For example, the first group advised to evacuate might be those living

Chapter 8 · Strategies for Evacuation Management

185

on barrier islands who are advised to leave well in advance of an approaching hurricane. The next to be advised to leave might be mobile home residents and tourists, followed by inland residents who are advised to start their evacuation even later. Another common approach is to group evacuees by their evacuation zone. Generally, geographic zones or specific locations (e.g., mobile home parks) at higher risk and closer to the hazard are evacuated first. In order to be successful, authorities need to clearly communicate the boundaries of the different evacuation zones. As noted in Section 7.1, risk maps that define boundaries in terms of hazard probabilities or magnitudes are difficult for most people to understand (Arlikatti et al. 2006, Zhang, Prater, and Lindell 2004). Conversely, locations with pre-existing labels (e.g., postal codes and political jurisdictions) and landmarks such as rivers and major highways provide easily identifiable boundaries that are easy to understand, but rarely coincide with risk area boundaries. Thus, local authorities must often resolve the disparity by choosing identifiable boundaries that include areas they consider to be at minimal or no risk. Nonetheless, even with high quality communication of the evacuation zones, it is rarely possible to prevent evacuees from zones/ groups with a later recommended evacuation start time from departing during earlier phases. From a research perspective, the basic problem for phased evacuation is to find an optimal phasing of evacuation assignments that minimize evacuation time, risk, or the time for those in the highest risk areas to reach safety. One of the first studies in this area (Chen and Zhan 2004) involved comparing phased evacuation with simultaneous evacuation in three different network structures (grid, ring, and “real”, all based on San Marcos, TX). The researchers divided the emergency planning zone into four evacuation zones in order to implement the overall evacuation in four phases. They concluded from their simulation results that neither the phased nor the simultaneous evacuation strategy performed better across all network structures and population densities (Chen and Zhan 2004). In a subsequent study, Mitchell and Radwan (2006) noted the importance of considering population density when dividing the emergency planning zone into evacuation zones and the importance of correctly timing the initiation of demand from each evacuation zone; some splits of the emergency planning zone into evacuation zones resulted in underutilization of ERS capacity and caused later congestion. Further research has increased the number of evacuation zones, and combined the departure timing of each zone with other aspects of the evacuation problem such as destinations and routes. Some analyses have used complex optimization techniques rather than simulations. For instance, Sbayti and Mahmassani (2006) developed a bilevel formulation that determines time-dependent route assignments at the upper level with the method of successive averages and solves the route travel times at the lower level with traffic assignment simulation software. The modeling process assigns each vehicle a departure time,

186 Chapter 8 · Strategies for Evacuation Management route, and destination. They used this output to generate a time-dependent phasing plan for each origin and compared phased evacuation to simultaneous evacuation. In their simulation experiments, network clearance time decreased by 20% and average trip time decreased by approximately 60%. Abdelgawad and Abdulhai (2009) used evolutionary algorithms to determine the optimal destinations and scheduling of evacuation demand, coupled with an optimization model with dynamic traffic assignment to perform the routing. Chiu et al. (2006) proposed a model that determined optimal destinations, routes, and phases. Their model, which was based on the cell transmission model, minimized travel time for all evacuees in the network. Also based on the cell transmission model, Dixit and Radwan (2009) examined the sequencing of evacuation notices for four cities in a hurricane evacuation, incorporating a sequential logit demand model. Using the phasing of evacuation notices for the four cities led to an 11% reduction in vehiclehours compared to the simultaneous approach (Dixit and Radwan 2009). Optimization and simulation experiments agreed that phased evacuation reduced evacuation times compared to simultaneous demand loading (at least in many scenarios). Additional research suggests that providing evacuees with optimized destinations and routes (e.g., Han et al. 2006) could further reduce congestion and evacuation time. Conclusions about the effectiveness of phased evacuation are limited by the lack of behavioral research directly addressing risk area residents’ expectations of complying with such strategies. However, available studies of evacuation expectations and actual evacuations provide indirect indications that compliance with a phased evacuation strategy will almost certainly be less than perfect and might be quite poor. There are at least three major phenomena that must be better understood in order to assess the effectiveness of this strategy. First, given that the total area under the demand curve is a constant for an event of a given magnitude (e.g., hurricane category), “flattening” it to fit within the ERS capacity requires authorities to begin an evacuation earlier than they would otherwise (e.g., at 40 hours before arrival of Tropical Storm wind rather than 35 hours). However, the earlier authorities initiate an evacuation, the greater is the probability of a false alarm because the hurricane is farther offshore and, thus, prospective evacuees have greater uncertainty about whether it will make landfall in their jurisdiction. Residents of risk areas closest to the coast who are concerned about an “unnecessary” evacuation will delay their departures awaiting further information, thus reducing the effectiveness of authorities’ intended shift in demand to earlier departures. Second, as noted in Chapter 4, observations of other households evacuating is a significant predictor of decisions to evacuate. This suggests that inland residents’ observations of evacuation traffic in

Chapter 8 · Strategies for Evacuation Management

187

their communities, even if they only see the traffic on TV, will prompt them to evacuate before the time they have been scheduled to leave. Third, as also noted in Chapter 4, as much as 15%—and occasionally more—of evacuation zone residents leave before they receive an official warning (Baker 2000) and data from Hurricane Ike indicate that departure time was uncorrelated with risk area (Huang et al. 2012). Thus, some degree of early evacuation is likely to occur in all risk areas at all phases of a phased evacuation. The relative effects of the first two phenomena (late evacuation from coastal risk areas and early evacuation from inland risk areas) on evacuation traffic demand will depend on the relative numbers of vehicles per hour entering the ERS from the coastal and inland risk areas. In turn, this will depend on the percentages of households evacuating late from coastal risk areas and evacuating early from inland risk areas, the relative number of households in those risk areas, and the relative number of vehicles per household in those risk areas. In ideal circumstances, households departing early from inland risk areas would exactly compensate for the delayed departures from the coastal risk areas but this seems improbable. The uncertainty associated with relative effects of the first two phenomena is compounded by uncertainty about the magnitude of the third phenomenon (randomly distributed early departures). In summary, evacuation analysts should expect an evacuation shadow in time (i.e., across phases) that is similar to the extensive literature on evacuation shadow in space (i.e., across risk areas). Thus, one can expect that a significant percentage of the population will evacuate before they receive an official warning for their evacuation zone. The absence of data about households’ compliance with a phased evacuation advisory, coupled with an absence of data about the sensitivity of a phased evacuation strategy to different levels of compliance indicates that much research remains to be done before it is possible to identify the conditions in which a phased evacuation strategy can be recommended.

8.2 Supply Management Strategies A vehicular evacuation is, in many respects, similar to the movement of traffic during any routine peak hour condition or during planned special events such as athletic contests, concerts, and other major public gatherings. This similarity arises from several shared general characteristics. Most notable of these are that traffic flow tends to be directionally imbalanced, oriented in a single primary direction (outward, away from an origin), and that traffic volume loads into the network within a concentrated temporal or “surge” period. When

188 Chapter 8 · Strategies for Evacuation Management combined, these conditions mean that the demand quickly overwhelms the available network capacity, resulting in wide ranging and long lasting congestion. The high traffic density also means that any disturbances to flow within the network, such as a traffic incident or lane closure, can cause rapid, widespread, and persistent cascading effects that produce queues, congestion, and delay. However, evacuations obviously differ from routine peak period and special event conditions in the potential for evacuation traffic to have a direct life safety consequence in which people who are unable to quickly leave a threat area face the possibility of significant injury or death. As such, the traffic management strategies used for evacuation traffic management take on a greater significance and may require the application of strategies that would not be used under conditions that were not immediately life threatening. Although, in the past, these evacuation management strategies have most visibly focused on limiting the effects of congestion by implementing techniques to utilize all available road network capacity and preventing breakdowns and disturbances, they also extend into numerous other areas. To accommodate evacuations’ sudden and wide-ranging surges in demand, transportation agencies have worked in conjunction with emergency management and law enforcement agencies to develop and implement a variety of supply management strategies. Common methods to enhance system capacity include modifying signal controls to provide longer green time durations for the directions that have greater evacuation demand, adding lanes, and using nearby parallel routes that would otherwise be underutilized. Techniques also include closing routes, restricting, or rerouting travelers. However, no strategy will be effective unless it is consistent with driver expectation and effectively communicated to them. For example, attempting to shift traffic demand to other routes will only be effective if drivers believe they can still reach their intended destinations without excessive additional travel time, distance, and delay. In events such as hurricanes that can require hundreds of miles of driving in unfamiliar areas, drivers are known to favor familiar routes within a network, and often resist alternate, less familiar routes (Dow and Cutter 2002, Wolshon and McArdle 2011). Thus, authorities’ communication with drivers during an evacuation needs to be timely, accurate, and useful. Information that is out of date, incorrect, or irrelevant will diminish driver compliance with instructions. The 7th Edition of the Institute of Transportation Engineers’ (ITE) Traffic Engineering Handbook (Pande and Wolshon 2016) discussed a number of concepts associated with evacuation, emergency, and event transportation management. Below is a partial list of typical operational strategies that could be readily adapted for use in evacuations. Perhaps most significant, is that, with effective adaptation, they could also be used to fit a variety of specific locations and facility types and even event or non-event traffic conditions.

Chapter 8 · Strategies for Evacuation Management

189

■ Freeway traffic control and lane utilization—closing ramps, adding ramp capacity, eliminating weaving areas, using alternate routes, implementing contraflow, and metering onramps. ■ Street traffic control and lane utilization—establishing lane control, alternative lane operations, and lane closures; eliminating on-street parking; and providing “trailblazer” signing. ■ Intersection traffic control—modifying signal timings, establishing turn restrictions, and providing advance signing. ■ Traffic incident management resources—providing motorist service patrols and temporary signing. ■ Traveler information and surveillance—providing highway advisory radio, changeable message signs (CMS or VMS or DMS), Closed Circuit TV, and temporary signing. ■ Travel demand management—providing mass transit vehicles, pretrip traveler information, and High Occupancy Vehicle lanes. ■ Emergency vehicle access—providing dedicated access into event venue and incident scenes. In addition to discussing these strategies specifically, this section also highlights the broader roles that transportation agencies play in the direction and control of transportation systems to facilitate evacuations by using traffic demand and supply management measures. Today, transportation agencies provide invaluable support to emergency management and law enforcement officials by providing communications and coordinating with neighboring states when evacuations cover multistate regions. This section summarizes many established practices and recent developments in traffic control, as well as the role transportation agencies play in supporting plan development and decision making in evacuations. There is a specific emphasis on the application of these topics to freeways and other uninterrupted flow facilities. Although the strategies discussed in this chapter focus on roadways and modes of transport that would use these facilities, they also facilitate the evacuation of carless evacuees on buses. Moreover, a key need for any of these strategies to work is cooperative planning among transportation, emergency management, and police agencies. The resulting evacuation traffic management plan should also be exercised to observe its effectiveness and make necessary modifications.

8.2.1 Evacuation Traffic Management Measures on Freeways During evacuations, the management of traffic on freeways often includes strategies that restrict the movement of background traffic to favor the movement of evacuation traffic. Experience in managing evacuations (as well as planned or unplanned special events) has shown that it is

190 Chapter 8 · Strategies for Evacuation Management important to keep traffic moving along freeway corridors since, in the absence of designated alternate evacuation routes, the majority of evacuees will prefer freeways out of a threatened area. As noted earlier, this is because evacuees tend to select routes that are the most familiar to them and carry the highest volumes of traffic at the highest free-flow speeds during routine, non-emergency periods. There is a range of options that can be considered when seeking to expedite and increase the efficiency of traffic flow on freeways. The obvious goal is to enable the movement of evacuees away from the evacuation zone. In terms of freeways specifically, where access and egress can be better controlled, some recognized effective techniques include: ■ Reducing background traffic congestion in the areas surrounding the evacuation zone by preventing and reducing inbound flow; ■ Closing some strategic freeway access points (i.e., on-ramps) outside the evacuation zone to prevent background traffic access to the freeway corridor, hence reducing the chances of congestion and promoting the efficient outbound flow of evacuees from the evacuation zone; and ■ Closing some strategic freeway exit points (i.e., off-ramps) outside the evacuation zone to prevent queue spillbacks onto the freeway mainline as a result of arterial congestion. If allowed, these spillbacks propagate onto the mainline and significantly reduce the operational capacity in the outbound direction from the evacuation zone. During an evacuation, HOV lane restrictions may be lifted. Even if these are not explicitly considered, it is reasonable to assume that drivers will ignore HOV restrictions. Other strategies, such as contraflow operations, ramp closure, and intersection management require the implementation of highway department resources. The sections that follow discuss the objectives, operational needs, and expected outcomes of several freeway management strategies that have been used or are planned for use in evacuations.

8.2.1.1 Contraflow Contraflow is a form of reversible traffic operation in which one or more travel lanes of a roadway are used for the movement of traffic in the opposing direction (Wolshon and Lambert 2004). Although its use as a traffic management strategy for peak hour and planned special event conditions dates as far back as the 1920s, a lack of identifiable need effectively negated its implementation for evacuations until it was used on an impromptu, unplanned basis in Georgia and South Carolina in advance of Hurricane Floyd in 1999. Based on that experience and its use during several hurricanes along the Gulf Coast during the subsequent decade, every coastal US state threatened by hurricanes now has plans to

Chapter 8 · Strategies for Evacuation Management

191

implement contraflow when threatened by major storms; it has been developed for use from New Jersey to Florida on the Atlantic seaboard and from Florida west through Texas along the Gulf of Mexico (Wolshon 2001). Contraflow is effective because it provides an immediately significant increase in the directional capacity of a roadway without the time or cost required to plan, design, and construct additional lanes. Contraflow segments are most common and logical on freeways because they are the highest capacity roadways and are designed to facilitate high speed operation. Contraflow is also more practical on freeways because these routes lack at-grade intersections that interrupt flow or permit unrestricted access into the reversed segment. It can also be implemented and controlled with fewer personnel resources than unrestricted highways. Nearly all contraflow strategies currently planned on US freeways have been designed for the reversal of all inbound lanes. These configurations, were represented schematically in prior studies (Wolshon 2001, Wolshon and Lambert 2004) and are shown here in Figure 8.1. Inset 1d of the figure is referred to as a “One Way Out” or “All Lanes Out” evacuation—the most common form of contraflow. This is in contrast to Inset 1a, which shows a typical bi-direction operation of two lanes of flow in each direction that would be used during all routine, non-emergency periods. Though not as common, some contraflow plans have also included options for the reversal of only one inbound lane (Inset 1b) with another option to use one or more outbound shoulders (Inset 1c). Inbound lanes in these plans are maintained for entry into the evacuation zone for service vehicles to provide assistance to evacuees in need along the contraflow segment. Obviously, these two conditions would only be appropriate on multilane roadways. Another critical need for effective contraflow operation is ingress/ egress management. Without an adequate plan to load and unload vehicles from a reversible flow segment, potentially worse congestion is created at the inflow and outflow points of the section. Observed experiences in New Orleans, Houston, and elsewhere show that freeway contraflow works best when it is filled at multiple points (median crossovers, reversed ramps, etc.) in its origin area and when normaland reverse-flowing traffic streams are not permitted to re-merge at the terminations. Rather, these traffic streams should be split onto either intersecting or parallel routes. The Fang and Edara (2014) simulation study also suggested that the use of intermediate bidirectional crossovers, spaced at regular intervals throughout the contraflow section, will also help to balance traffic volumes and reduce differential congestion and delay in the normal- and reverse-flowing lanes. As a result of contraflow’s recent demonstrated effectiveness during Hurricane Katrina, it is also now viewed as a viable strategy for other mass evacuations. It is not without limitations, particularly for

INBOUND

LANE

INBOUND

LANE

SHLDR.

MEDIAN

SHLDR. OUTBOUND OUTBOUND SHILDR.

INBOUND

LANE

INBOUND

LANE

SHLDR.

From Wolshon and Lambert 2004

SHLDR.

MEDIAN LANE

SHLDR. OUTBOUND LANE

OUTBOUND SHILDR.

1c. Normal w/Shoulder and One Contraflow Lane

SHLDR.

1a. Normal Operation

INBOUND LANE

INBOUND LANE

SHLDR.

MEDIAN

SHLDR.

LANE

INBOUND

LANE

INBOUND

SHLDR.

MEDIAN

SHLDR.

LANE

INBOUND

LANE

LANE

INBOUND

LANE

SHLDR.

SHLDR. OUTBOUND OUTBOUND SHILDR.

1d. Normal Plus Two Contraflow Lanes

SHLDR.

1b. Normal Plus One Contraflow Lane

Figure 8.1 Freeway Contraflow Lane Use Configurations for Evacuations

192 Chapter 8 · Strategies for Evacuation Management

Chapter 8 · Strategies for Evacuation Management

193

no-notice and short-notice hazard events. And, although the basic concept of contraflow is simple, it can be complex to implement and operate in practice. If not carefully designed and managed, contraflow segments also have the potential to confuse drivers. To ensure safe operation, improper access and egress movements must be prohibited at all times during its operation. Segments must also be fully cleared of opposing traffic prior to initiating contraflow operations. These are not necessarily easy to accomplish, particularly in locations where segments are in excess of 100 miles and where interchanges are frequent. For these reasons, some transportation officials regard them as risky, for use only during daylight hours and—even then—only in the direst situations. These are also the reasons why contraflow for evacuation has been planned nearly exclusively for freeways, where access and egress can be most tightly controlled. Nonetheless, as one Texas emergency manager observed, access control can be more theoretical than real when freeways and the service roads that parallel them are separated only by grassy strips and many evacuating vehicles have four wheel drive. Evacuation freeway contraflow, like reversible roadways in general, has a number of physical and operational attributes that are common among all applications. The principal physical attributes are the highway’s overall length and number of lanes, as well as the configuration and length of the inbound and outbound transition areas. The primary operational attribute is the way in which the segment will be used; that is, the temporal control of traffic movements. The primary temporal components of evacuation contraflow are the time required to transition traffic from one direction to another and the duration of its use. Like all reversible flow roadways, evacuation contraflow lanes need to achieve and maintain full utilization to be effective. Although this sounds like an obvious fact, it can be challenging to achieve in practice—even in evacuations. The most common method to transition into and out of contraflow segments is to provide median crossovers at the inflow and outflow ends. Since these termini regulate the ingress and egress of traffic entering and exiting the segment and they are locations of concentrated lane changing as drivers weave and merge into the desired lane of travel, they effectively dictate the capacity of the entire segment. Another common method to transition into and out of contraflow segments is to reverse the direction of off ramps to allow vehicles to enter in the reverse-flowing direction. A simple median crossover contraflow transition is illustrated in Figure 8.2, which shows the initiation point for the I-10 contraflow segment in New Orleans during the Hurricane Ivan evacuation. In this location, evacuating traffic vehicles in the left and center outbound lanes of I-10 were transitioned across the median and into the contraflow lanes using paved lanes over the sodded median.

194 Chapter 8 · Strategies for Evacuation Management

Figure 8.2 Schematic (Top) and Field Photo (Bottom) of Median Cross Over Contraflow Loading Configuration Interstate 10 at Loyal Avenue, Kenner, Louisiana

From Wolshon 2002

Field observation and simulation studies have shown that contraflow entry points with inadequate inflow transitions result in traffic congestion and delay prior to the contraflow segment, thus preventing that contraflow segment from carrying capacity-level demand. Such a condition occurred at the aforementioned contraflow loading point in New Orleans during the Hurricane Ivan evacuation. Police control of this location limited the number of vehicles that could enter the contraflow lanes which, in turn, limited the downstream flow beyond the entry point to significantly below its vehicle carrying capability as shown in the photographs in Figure 8.3. Traffic queues upstream of the crossover

Chapter 8 · Strategies for Evacuation Management

195

Figure 8.3 Ineffective Loading of Freeway Contraflow at Interstate 10 at Loyola Avenue, Hurricane Katrina Evacuation of New Orleans

From Wolshon et al. 2006

extended more than 14 miles. This plan was significantly improved prior to the Katrina evacuation one year later by permitting vehicles to enter the contraflow lanes at multiple points, spatially spreading the demand over a longer distance and reducing the length and duration of the congested conditions. Inadequate design of the downstream end of contraflow segments can also greatly limit their effectiveness. Operational experience and simulation modeling have shown that an inability to move traffic from contraflow lanes back into normally flowing lanes will result in congestion backing up from the termination transition point in the contraflow lanes. Under demand conditions associated with evacuations, queue formation can occur quite rapidly and extend upstream for many miles

196 Chapter 8 · Strategies for Evacuation Management within hours. To limit the potential for such outcomes, configurations that require merging of the normal and contraflow lanes are discouraged; particularly if they also incorporate lane drops. One popular split design used to terminate contraflow is to route the two traffic streams at the termination onto separate routes and the other is to reduce the level of outflow demand at the termination by including egress points along the intermediate segment. In general, split designs offer higher levels of operational efficiency than merge designs. The obvious benefit of a split is that it reduces the potential for bottleneck congestion resulting from merging four lanes into two. Its most significant drawback is that it requires one of the two lane groups to exit to a different route, thereby eliminating route options at the end of the segment. In some older designs, the contraflow traffic stream was planned to be routed onto an intersecting arterial roadway. Of course, this type of split design needs adequate capacity on the receiving roadway. Reviews of practice suggest that until now, contraflow has been used for hurricane and wildfire evacuations but no other type of natural or technological hazard. The first reason for this is that hurricanes and wildfires affect much greater geographic areas and tend to be slower moving than many other hazards. Because of their large scope, they tend to create a need to move larger numbers of people over greater distances than other types of hazards. The second, and perhaps even more critical, reason is that contraflow requires considerable personnel and materiel resources as well as time to mobilize and implement. Prior hurricane evacuation experiences show that the positioning of traffic control devices and enforcement personnel can require at least six hours. Another time consideration is the time to clear the segment of opposing traffic. Although there are several different ways that this can be accomplished, if inbound vehicles are permitted to complete their entire journey on the route being reversed, this process could take several hours on long contraflow segments. Based on this significant resource demand, decisions to implement contraflow often must be made days in advance.

8.2.1.2 Route Closures Road segment closure is another tool that has been used to manage evacuation traffic. This can be used on any roadway functional classification, signalized or freely flowing. Closures can be used as a protective action to limit traveler exposure to a hazard or to limit cut-through traffic into areas unequipped to accommodate the increased demand. On a regional basis, route closures have also been used to prevent access to a downstream section that lacks the capacity to accommodate inflow traffic from multiple incoming routes.

Chapter 8 · Strategies for Evacuation Management

197

In Louisiana, the closure of lengthy segments of interstate freeways is part of the State’s regional evacuation plan for the southeastern area of the state, including New Orleans (see Figure 8.4). These regional closures, which have been used to high degrees of effectiveness, are also integrated with the regional contraflow plan. This plan also requires forced routing of traffic onto alternative routes, coordination of parallel non-freeway routes, and the reconfiguration of busy urban freeway interchanges to more effectively load evacuees from the surface street network into the ERS. Despite limiting capacity in the immediate vicinity of the closed routes, the closures are used to prevent westbound traffic from one freeway from merging with traffic from another freeway and backing up both routes. It should also be noted, however, that road closures (along with techniques such as contraflow, and turn prohibitions) can also become controversial because they can have economic impacts on freight movement, or may increase the risk of crashes and secondary traffic incidents. However, if there is a clear threat to life and safety such as a hazardous materials release, these drawbacks are significantly outweighed by their benefits.

Figure 8.4 Southeast Louisiana Regional Evacuation Route Management Plan

From State of Louisiana 2016

198 Chapter 8 · Strategies for Evacuation Management

8.2.1.3 Ramp Closures Ramp closures can also be used to promote effective freeway traffic flow during evacuations by reducing the number of merging and related speed reduction points on freeways created by on-ramp inflow. Under normal (non-evacuation) conditions, on-ramp closures have been shown to eliminate bottlenecks on mainline freeway lanes. Off-ramp closures can have a similar effect by preventing congestion formation both along the freeway and, perhaps more critically, on parallel and perpendicular arterial streets. Under some plans, off-ramp closures are also viewed as a way to prevent “pass through” and “background” traffic from entering an evacuation zone. Although there is often variation in the specific definition of these traffic sources, generally speaking, pass through traffic includes vehicles that move through an evacuation area, but do not originate or terminate their travel within it, whereas background traffic includes vehicles that start and/or end within the evacuation area and are traveling for routine, non-emergency, or other non-evacuation purposes. Although ramp closures can increase travel speed, they do not increase overall throughput. Another potential consequence can be that evacuees spend more time on arterials entering the ERS (e.g., Ghanipoor Machiani et al. 2013). In turn, this could lead to increased risk to evacuees if they might be overtaken by a hazard such as a wildfire or hazardous materials plume. Consequently, ramp closures should be prioritized appropriately and their benefits quantified using simulation. Similar to the route closures discussed in the previous sections, ramp closures can also be used to limit access to hazardous areas. However, evacuation analysts should be sure to identify enough ingress routes that responding personnel and emergency vehicles can promptly access the evacuation zone.

8.2.1.4 Use of Freeway Shoulders Another method that more completely utilizes all potential capacity on evacuation routes is to use shoulders as additional travel lanes on freeways. Although this sounds simple, there are numerous details that should be assessed. From a technical standpoint, the shoulder’s design characteristics should be evaluated to make sure that a full width is continuously available along its entire length. If the shoulder is raised, evacuation analysts should verify that it can be safely and efficiently traversed by all vehicles anticipated to be using it. The structural strength of the pavement must also be adequate to support the traffic loads that will be placed upon it. The idea of using shoulders for an additional travel lane is not new. The Virginia DOT has used shoulder lanes for many years to gain additional capacity during rush hours in the suburban Washington, DC area (Wolshon and Lambert 2004). Shoulder lanes were also suggested, though never implemented for hurricane evacuations in Florida. After the 2005 Hurricane Rita evacuation, the Texas DOT conducted extensive

Chapter 8 · Strategies for Evacuation Management

199

studies of ways in which they could learn from the shortcomings of that event. Among the recommendations of this research was the use of shoulders as Evaculanes on the outside shoulder of both the normal and contraflow sides of freeways (Ballard and Borchardt 2006). A follow up study by Ullman et al. (2007) used focus group input and surveys to develop guidelines for special use hurricane evacuation traffic control devices, including shoulder lane signs and pavement markings. Figures 8.5 and 8.6 illustrate the Texas Evaculanes. Their most notable benefits are that, if preplanned and marked, they can be activated rapidly and, since drivers are typically familiar with driving on freeways, their use does not involve any additional explanation or instruction. However, transportation authorities should encourage proper procedures for using Evaculanes (such as moving stalled vehicles out of the lane and onto the unpaved embankment area), identifying their locations, and posting their schedules of availability. Depending on the specific design characteristics of the facility on which they would be used, it may be necessary for transportation crews to inspect the shoulder on the segments in which Evaculanes would be used. This would ensure the removal of vehicles or other obstructions that might block flow as well as the removal of any items or hazards that could cause safety problems. Since Evaculanes are a nonstandard application, they are incompatible with several fundamental principles of roadway design and control.

Figure 8.5 Texas DOT Evaculane Road Sign, Houston Texas

200 Chapter 8 · Strategies for Evacuation Management

Figure 8.6 Hurricane Evacuation Route Directional Shoulder Pavement Markings (Normal Lanes at Left and Contraflow Lanes at Right), US-290 – Texas

Note: These Photos Were Not Taken Under Evacuation Conditions

Chapter 8 · Strategies for Evacuation Management

201

As such, the benefits of their use must be weighed against their costs and potential risks. For example, Evaculanes eliminate the emergency stopping area for vehicles. Thus, drivers in need of assistance will be required to stop on the unpaved embankment area outside of the shoulder. Evaculanes also eliminate the paved area often used by service and emergency response vehicles as they respond to calls and circulate within congested evacuation traffic. Properly planned Evaculanes should also be accompanied by adequate downstream capacity so that outflowing vehicles are not forced to merge back into the normal travel lanes at the end of the segment.

8.2.2 Evacuation Traffic Management Measures on Arterial Routes In general, the objective of arterial roadway evacuation management strategies is to limit congestion that results from the interaction of evacuation and background traffic that would impede the movement of evacuees out of the evacuation zone. For rapid onset events, including some hazmat incidents and wildfires, the rapid buildup of background traffic congestion presents severe challenges to maintaining traffic flow during high travel demand periods. Analysts should assume that such conditions are inevitable and will limit the outbound flow of evacuees from the evacuation zone through the arterials around the evacuation zone. Among the broad goals of arterial management measures during an evacuation are to: ■ Keep the evacuation zone closed to incoming background traffic and facilitate outbound flow of evacuees to freeways and surrounding arterials. ■ Use arterial road closures and turn restrictions to support freeway operations and reduce background interference. ■ Set arterial signal timing controls to facilitate outbound movement of traffic. ■ Support the movement of evacuation traffic into reception centers. Although the first consideration for assessing potential timing changes along arterial routes should be to modify traffic signal timings to maximize movement in the outbound direction away from the source of danger, consideration must also be given to vehicles traveling in other directions as well. This is particularly true in more densely populated areas, such as central business districts of urban areas, because traffic will still be moving in all directions. This includes perpendicular traffic travelling across mainline evacuation routes and some emergency response vehicles travelling into the evacuation zone. Recent research has shown that the delay produced by even moderate levels of cross-street volume would offset the travel time gains made by

202 Chapter 8 · Strategies for Evacuation Management vehicles in the mainline direction (Chen, Chen, and Miller-Hooks 2007). Observational experience has also shown that setting signals to a yellow-red flash mode to give exclusive right-of-way to evacuating traffic or providing manual police control to give directional flow preference to evacuees can essentially eliminate mobility in local areas as cross-town routes become unusable.

8.2.2.1 Turn Restrictions Turn restrictions are also often referred to as “crossing elimination.” As shown in Figure 8.7, this strategy improves flow in the primary outbound direction by removing many, if not all, conflicting traffic stream maneuvers —most notably left turn and minor street crossings (Xie, Lin, and Waller 2010, Xie and Turnquist 2011, 2009, Xie, Waller, and Kockelman 2011, Cova and Johnson 2003, Jahangiri et al. 2014, Liu and Luo 2012, Luo and Liu 2012). Implementing such restrictions requires police control or barricades or, in many cases, both. Despite the logical benefits of turn and crossing prohibitions, this strategy has not actually been used during evacuations. Most experience

Figure 8.7 Sample Movements After Turn Restrictions

Chapter 8 · Strategies for Evacuation Management

203

using these techniques has been during planned special events such as major sporting and civic events. In simulation studies, crossing elimination strategies have provided up to a 40% reduction in travel time but benefits depend on the network configuration and hazard scenario (Cova and Johnson 2003).

8.2.2.2 Signal Timing Modification Recent experience has shown that traffic signals can have a significant adverse impact on evacuation traffic flow. Although typically regarded as a strategy most applicable for use in urban areas (with denser road grids, higher traffic volumes, and more signals) particularly for nonotice events (e.g., Parr and Kaisar 2011), traffic signal re-timing has also been used to move hurricane evacuation traffic through rural areas. Most notably, these included signals in small towns where only one or two signals regulate crossing and turning maneuvers through and into a primary highway that is being used as a pass-through for evacuees moving to more distant large cities. A review of practice (Wolshon 2009a) showed that there are no “standardized” or “recommended” rules of operation for traffic signal control during evacuations. However, the primary goal is always to facilitate the outbound movement of traffic from the hazard zone. Within this objective, however, turning and crossing movements from intersecting streets must also be accommodated and, in many urban locations, there might not be a clearly defined primary movement direction. As a result, signal timing modifications must be assessed, evaluated, and planned in advance whenever possible. A simulation study examined the effects of varied traffic signal timings for urban area evacuations, focusing on arterial street corridors with varied cycle lengths (Chen, Chen, and Miller-Hooks 2007). Tests with cycles of 180, 240, and 300 seconds, as well as all-yellow and allred flashing modes, suggested that the “best” plan depended on what needed to be achieved. Longer cycle lengths with longer green times for the outbound directions were more effective at facilitating the movement of evacuation traffic. However, if volumes are near those of routine peak periods, then typical non-emergency outbound peak timing plans could be most effective. From a practical perspective, this suggests that routine peak hour timing plans, developed from a long history of experience and observation to move people out of central business districts during peak commuter periods, are probably the most effective way to move traffic in multiple directions during an evacuation through a urban network. Another technique is to use adaptive traffic signal control (e.g., Liu et al. 2007). Under this form of operation, traffic signal timing can

204 Chapter 8 · Strategies for Evacuation Management adapt automatically, based on directional traffic demand arriving at, or anticipated to be arriving at, each intersection approach. Sophisticated software and hardware can permit cycle lengths and signal phase timings to vary based on real-time demand sensor observations. For example, the District of Columbia and surrounding communities have developed signal timing plans specifically for evacuation on selected major arterials. In some adaptive control systems, such as the Sydney Coordinated Adaptive Traffic System, protected turn movement phases can be added or eliminated automatically, and networks and corridors of signals coordinated as needed, to achieve a desired objective. Similar to all other traffic management strategies, however, the modification of traffic signal timings has shortcomings that have been observed in several hurricane evacuations in which signals along arterials in less densely populated areas were set to flashing yellow to maintain uninterrupted flow along the main route. The unintended consequence was that local travelers in these areas seeking to enter the mainline route from minor intersecting roads were unable to find adequate gaps to access or cross the intersection. To address this issue, the signals were reset to include minor street green time to provide intermittent access opportunities for minor street traffic. Unfortunately, this method also leads to congestion, long queues, and delays as well as the potential for incomplete clearance of the hazard zone. Based on these conditions, some areas plan to use flashing yellow in conjunction with police enforcement (discussed in the following section). Unfortunately, in areas where officer availability is limited and the needs for their involvement are great, this can also be problematic.

8.2.2.3 Police Manual Traffic Control Many communities that lack sophisticated signal control systems have nevertheless been able to achieve demand-adaptive control at intersections through the use of police operated manual traffic control (Parr, Wolshon, and Murray-Tuite 2016). The use of manual traffic control based on direct observation, which is a common practice in cities throughout the world, is most common when there is abnormally high, directionally unbalanced, or widely varying traffic demand and, in particular, before and after planned special events and during emergencies. Police officers are effective under these conditions because they can directly observe and adapt to the changing environment by directly allocating right-of-way. During evacuations, police control also has the added benefit of putting “boots on the ground” to respond to problems and project the presence of authority during the event. Tactically, manual traffic control is most common at high volume intersections, especially ones at which traffic from one or more routes merges or conflicts with traffic from other routes. Depending on the

Chapter 8 · Strategies for Evacuation Management

205

amount of traffic, number of lanes involved, and complexity of the location, several officers may be required at a single intersection. Strategically, manual traffic control is most effective for roadways that combine high demand and regional connectivity. Intersections are also more likely to be manually controlled if they require additional access management guidance, such as those along contraflow corridors and in close proximity to grade separated interchanges (Parr and Wolshon 2016). In addition to selecting the most important intersections and training officers on the strategic objectives for their right-of-way allocation, Parr and Wolshon (2016) proposed that officers be trained on specific procedures including: ■ The use of reflective vests at all times; ■ The use of lighting for directing traffic in adverse weather; ■ The need for additional lighting at night from the police vehicle or additional spotlights; ■ Where to stand within the intersection; ■ How to position their body to command vehicles; ■ Uniform hand signals to start and stop the flow of traffic; ■ Safety when directing conflicting turn movements; and ■ The use of traffic control tools such as flashlights, whistles, illuminated batons, and flares. A recent simulation study (Parr, Wolshon, and Murray-Tuite 2016) compared the effects of manual traffic control to flashing yellow signals and crossing elimination during evacuation to determine which unconventional intersection control strategy was best suited for a given urban evacuation scenario. The study results suggested that manual traffic control was best suited for intersections immediately upstream of a bottleneck or for closely spaced, uncoordinated signals. Flashing yellow signals appeared to work well for intersections with high, unbalanced demand and low volumes on the minor approach. Crossing elimination strategies worked best when demand from non-conflicting directions was high and all other approach volumes were relatively low. The study further suggested that, in practice, any of these control strategies could also be used in combination with other evacuation management strategies such as phasing demand and implementing contraflow to increase ERS capacity and decrease clearance time.

8.2.2.4 Transit In the wake of Hurricanes Katrina and Rita, municipalities across the US placed increased emphasis on the planning of mass transportation systems to serve mobility limited populations during emergencies.

206 Chapter 8 · Strategies for Evacuation Management Similarly, the modeling and analysis of transit-based evacuations became an area of research emphasis. Over the past half decade, transit operations have been analyzed using modeling to optimize scheduling and operations (Abdelgawad, Abdulhai, and Wahba 2010a, 2010b, Bish 2011, He et al. 2009). Additional modeling studies of transit bus evacuation operations were conducted by Naghawi and Wolshon (2010, 2012). In these models, the Citizen Assisted Evacuation Plan for metropolitan New Orleans was evaluated to assess bus travel time and operating speeds under alternative evacuation routing plans and under a range of response urgencies. Although the practical application and results of such actions would likely vary significantly based on the particular conditions and response to an event, the results of the simulation experiments suggest that although the buses required to move more evacuees increased the amount of traffic on evacuation routes, their impact was minimal when they were routed exclusively to alternative non-freeway evacuation routes. However, when buses were routed to more heavily utilized freeways, travel delays were observed and congestion-related queuing increased significantly.

8.3 Traffic Management in Emergency Operations Centers Access to timely and accurate traffic information during evacuations is critical for the management of evacuations. Evacuation traffic managers need information about traffic flow rates and speeds, along with lane closures, hazard conditions, traffic incidents, and the availability of alternative routes to effectively guide evacuees. During many recent evacuations, access to and exchange of accurate and timely traffic information has been difficult. Comments from emergency management officials showed that they often found themselves “working blind,” with little quantitative knowledge of which evacuation routes were flowing well and which were in gridlock. As a result, they were unable to direct traffic away from congested routes to ones that were carrying little traffic.

8.3.1 System Monitoring One of the ways that transportation agencies have responded to the need for up to date evacuee traveler information is through the application of intelligent transportation system (ITS) technologies. ITSs include a broad range of technologies and applications that use sensors, communications, computer controls, and the like to monitor and control vehicle movement. ITSs are a fundamental component of routine

Chapter 8 · Strategies for Evacuation Management

207

transportation management and have been widely adapted for, and integrated into, emergency transportation planning. A review of transportation activities for evacuation showed that all of the transportation agencies assessed in the study incorporated existing and developing ITS systems into their evacuation management plans (De Maio, Musolino, and Vitetta 2012). An effective ITS requires a monitoring component for assessing traffic flow, a control component for metering flows throughout the different links of the ERS, and an Advanced Traveller Information System (ATIS) that can use a variety of communication channels to disseminate travel messages to evacuees. The most common ITS application is the use of loop sensors to provide real-time monitoring of travel conditions, especially traffic volume and speed. These loop sensors are connected to count stations that can operate on solar power and transfer small volumes of data to the ITS control center. In turn, this information can be used to reroute traffic by comparing current volumes and speeds to historical data to determine if these parameters are significantly higher or lower than normal and compare traffic demand to route capacity. Such data can also yield insights into the existence and location of flow impediments. Another common type of surveillance method is closed circuit television (CCTV) cameras that are capable of remotely monitoring traffic speeds and flows. CCTV cameras have an advantage over loop detection in that they can also provide direct visual confirmation of traffic and weather conditions at remote locations. They can also be used for detecting traffic incidents and verifying their removal. One limitation of CCTV is that, unlike the count stations described earlier, CCTV typically requires direct power and hardwired communication connections. This is often prohibitively expensive in remote locations along rural hurricane evacuation routes. Monitoring the system also allows evacuation traffic managers to react to changing conditions by modifying the system. As discussed in the preceding section, many modern traffic signal networks employ remote traffic signal control. During an emergency, these systems would permit traffic signal timings to be changed to give preference for a particular approach movement or travel direction. Most also incorporate emergency preemption that gives immediate right-of-way priority to emergency response vehicles. Ramp metering can be used to achieve similar preference-related goals.

8.3.2 Communicating Travel Information and Guidance In addition to enhancing the ability of transportation and emergency management agencies to receive transportation-related data, ITSs can

208 Chapter 8 · Strategies for Evacuation Management also disseminate traveler information to evacuees and traffic control commands to field devices. Some ATIS channels, such as commercial television, are limited to use before evacuees depart whereas others, such as roadside signs, are limited to use en route. ATIS can transmit information to evacuees about when to leave, what routes to take, where commercial facilities and public shelters are available, and where to go for evacuation buses. Two of most commonly utilized (and planned) methods of en route highway communication, other than commercial radio and the Internet, are highway advisory radio (HAR) and dynamic message signs (DMSs; also called changeable or variable message signs CMSs or VMSs). These systems can convey traffic information, but have limited signal ranges. HAR typically has a range of about 3-5 miles, although trailer-mounted systems can be moved to locations where they will be needed. As an alternative to conventional HAR, the Delaware DOT acquired its own commercial radio stations for use as a statewide travel information station. During non-emergency periods, these stations are used to disseminate general travel information. Similarly, DMS is limited to very specific locations and typically a single direction of travel per sign. Consequently, they are used to convey limited information such as shelter locations, alternative evacuation routes, congestion alerts, traffic incident information, and locations of services such as gas stations, rest areas, and lodging. The implementation of an ITS system can be illustrated by Houston TranStar, which manages traffic in the Houston metropolitan area during emergencies as well as during normal operations (www.houston transtar.org/). TranStar, which is jointly operated by the City of Houston, Harris County, the Metropolitan Transit Authority of Harris County, and the Texas Department of Transportation, monitors the system’s freeways with loop detectors and more than 900 CCTV cameras. In response to traffic incidents, TranStar staff can use DMS and HAR to warn motorists of traffic conditions ahead, control traffic signals to limit the number of vehicles entering the system, and dispatch emergency vehicles to incident scenes. During a major emergency, TranStar staff are augmented by personnel from other local, state, and federal agencies, as well as non-governmental organizations who report to the TranStar EOC. An ITS can facilitate the management of evacuation traffic but it is important to recognize that it is limited by ERS geometry. Specifically, an ATIS can be successful in diverting traffic around congestion only if evacuees have multiple points of access to alternate routes, there is adequate capacity on the alternate routes, the transitions from the original routes to the alternate routes are located far upstream of the traffic incident, and there is adequate ATIS capability (Robinson and Khattak 2012).

Chapter 8 · Strategies for Evacuation Management

209

8.3.3 Compliance With Route Guidance According to the PADM, drivers’ ability to comply with authorities’ route guidance will depend in part on the information channels that authorities use. That is, drivers are able to comply with authorities’ guidance only if they are able to receive, heed, and understand the information being transmitted from the information sources over the information channels. Moreover, drivers’ willingness to comply depends upon their perceptions of the source’s expertise (knowledge about the current and future state of evacuation traffic on different routes) and trustworthiness (willingness to provide accurate and timely information about current and future states of the evacuation traffic on different routes). Unfortunately, there appears to be no research that has directly addressed perceptions of evacuation information sources such as authorities, news media, and peers. However, there has been a modest amount of research that has examined different information channels. As noted in Section 4.2.5, there are some cases in which there is no difference between source and channel because the source is channel bound. For example, TV news anchors only appear on one communication channel—television. However, TV news anchors are often only an intermediate source relaying a message originating from authorities, such as local emergency managers who also transmit messages via other channels such as radio stations. In still other cases, such as DMS, only the channel is apparent; the source (an anonymous individual in the transportation department) is unknown. Dia et al. (2001) reported that drivers most frequently relied on radio traffic reports (74%) and personal observation (64%) but also relied on DMS (23%) to assess different travel routes. Khattak et al. (2008) found that TV (68%), radio (48%), and the Internet (23%) were the most commonly used information channels but there was a higher probability of route change associated with radio (Odds Ratio = 6.91) than the Internet (OR = 5.35) or TV (OR = 3.00). One study concluded that radio reports have more influence on route diversion because drivers think radio reports provide more reliable assessments of the congestion severity, especially if the delay is projected to be 20 minutes or more (Khattak, Schofer, and Koppelman 1993). However, a later study found that radio, DMS, mobile phone, global positioning system, and Internet information produced equivalent rates of diversion to alternate routes when indicating congestion (Robinson and Khattak 2011). Perhaps the most significant conclusion about information sources/ channels is that route diversion is more likely when information is available from multiple sources/channels (Khattak et al. 2008). However, it is important to examine the effects of DMSs very carefully because they can be considered the “channel of last resort”—the channel that evacuation managers can use to reach drivers that do not

210 Chapter 8 · Strategies for Evacuation Management have functioning radios or mobile phones. Chatterjee and McDonald (2004) reported data that were consistent with the PADM’s pre-decision processes of exposure (being able to receive traffic information), attention (noticing the traffic information), and comprehension (understanding the meaning of traffic information). They found that DMS awareness rates were about 61%, with a range of 33–89%, which likely depends on the number and prominence of DMSs. An average of 92% (range 80–100%) of those who noticed the signs were able to read and understand them. Moreover, these researchers found that only 32% of the drivers who noticed DMSs thought the signs were relevant to their trip or made them better informed about alternate routes—what Mileti and Sorensen (1988) call personalization. Consequently, only 13% (range 3–100%) of the drivers diverted to an alternate route. A number of studies have found that drivers do not consider DMSs to be reliable—findings that are related to the PADM’s concern with drivers’ perceptions of the source’s (i.e., the DOT’s) expertise and trustworthiness. Specifically, Chatterjee and McDonald (2004) also reported that drivers frequently considered DMS messages to be imprecise, not credible, irrelevant, and not useful in identifying a better route. Similarly, Robinson and Khattak (2011) explained the findings of some previous studies that found nonsignificant effects of DMSs (Levinson and Huo 2003; Scheisel and Demetsky 2000) by reporting that 22% of their respondents had little confidence in DMS information. This conclusion was also reached by Bonsall and Palmer (1999), who also found that diversion rates decrease with low DMS credibility and is consistent with the finding that route diversion is affected by the reliability of realtime information (Mahmassani and Liu 1999). Travelers tend to be more prone to route diversion when delays exceed those shown on DMS (Wardman, Bonsall, and Shires 1997). This theme of information reliability is supported by other findings that route diversion is affected by the reliability of real-time traffic information received through different channels (Mahmassani and Liu 1999, Abdel-Aty and Abdalla 2004, Chen, Srinivasan, and Mahmassani 1999, Madanat, Yang, and Yen 1995). In addition, compliance with authorities’ route guidance depends in part on the message content that authorities transmit. Consistent with the PADM’s protective action decision making stage, Tsirimpa, Polydoropoulou, and Antoniou (2007) proposed that an ATIS can affect drivers’ behavior in two ways—expanding their awareness of travel alternatives (e.g., travel modes, departure times, and routes) and their perceptions of the characteristics of those travel alternatives. For example, Dia, Harney, and Boyle (2001) proposed that drivers need credible quantitative information about the current state of the system (traffic incident locations and expected delays) and guidance about alternate routes—especially travel time on those routes. Indeed, many studies have found that diversion rates increase as a function of real-time

Chapter 8 · Strategies for Evacuation Management

211

information about traffic conditions such as an accident on the current route (Bonsall and Palmer 1999) and its consequent delay duration (Bonsall and Palmer 1999, Jou et al. 2004, Khattak et al. 1991). More specifically, Robinson and Khattak (2011) found that diversion rates increased with predicted delay duration from 31% when no duration was indicated to 77% for a 30-minute delay, 82% for a 60-minute delay, and 86% for a 120-minute delay. Other studies have also shown that diversion rates increase when there is guidance about alternate routes (Bonsall and Palmer 1999, Adler 2001, Adler, Recker, and McNally 1993), especially the expected amount of time required to reach an alternate route (Bonsall and Palmer 1999), traffic conditions on that route (Khattak et al. 1991), and expected travel time on that route (Adler, Recker, and McNally 1993). Khattak et al. (1996) found that respondents were far more likely to change routes in response to a recommendation for an alternate route (44%) than for qualitative (13%), quantitative (13%), or predictive (14%) information about delay on their current route. Robinson and Khattak (2010) conducted a stated preference study of reactions to a series of ATIS information notices that examined multiple variables addressed in the PADM—the “threat” (traffic queues of varying durations) and the “protective action” (an alternate route). They found that a radio broadcast of an alternate route, an ATIS posting “alternate route guided by state police”, and an ATIS posting of an alternate route with “gas/food/lodging available” generated slightly higher percentages of diversions than an ATIS posting of an alternate route. However, there was a much stronger effect of posting the expected queue duration. The diversion rate to the alternate route was 33% for an undesignated duration, 75–84% for a 30-minute duration, 80–90% for a 60-minute duration, and 84–92% for a 120-minute duration. Robinson and Khattak (2011) found that information about service availability increased the diversion rate by 6% over “Accident ahead” and alternate route guidance, and information about state police guidance increased the diversion rate by another 3%. However, compliance with a diversion recommendation depends on drivers’ risk preferences regarding uncertainty about travel conditions on the recommended alternate route (Adler, Recker, and McNally 1993). Similarly, Dia et al. (2001) asked drivers how they would respond to five types of information 1) qualitative information indicating unexpected congestion on their usual route, 2) the qualitative delay information in Condition 1 plus forecasts of expected delays of 15 and 30 minutes, 3) the qualitative delay information in Condition 1 plus a recommendation to take the best alternate route, 4) the qualitative delay information in Condition 1 plus the delay duration on the current route, and 5) the qualitative delay information in Condition 1 plus delay duration on the current route and travel time on the best alternate route. The researchers found a diversion rate of 58% for the qualitative

212 Chapter 8 · Strategies for Evacuation Management information only (Condition 1), 71% for the addition of forecast delay information (Condition 2), 76% for the addition of recommended alternative route (Condition 3), 54% for the addition of quantitative delay on the usual route (Condition 4), and 75% for the addition of quantitative delay on the usual route and expected travel time on the alternate route (Condition 5). Thus, forecasts of delay times on the current route, recommendations to take alternate routes, and forecast travel times on the alternate routes stimulated higher diversion rates than qualitative or quantitative information about the current route.

8.4 Future Systems In the future, it is likely that even more sophisticated technologies and systems such as connected vehicles and autonomous, self-driving vehicles will also be used in evacuations. Connected vehicle (CV) or vehicleto-vehicle (V2V) technologies are systems that permit vehicles to exchange information with other vehicles, roadway control features, mobile phones, and other communication systems. Because of their potential to also bring significant benefits to evacuations and other emergencies, the USDOT has been leading research into the application of these technologies and their associated control features in evacuations. The centerpiece of their evacuation work is EVAC, an application within the agency’s Response, Emergency Staging and Communications, Uniform Management, and Evacuation (R.E.S.C.U.M.E.) bundle that takes advantage of CV technologies for more effectively guiding vehicles during evacuations by distributing traffic loads among available routes and providing the availability and locations of services such as shelter availability, fuel, and food to evacuees (Liu et al. in press). A simulation model of EVAC functionality was used to assess its potential benefits using the temporal and spatial traffic movements from the Hurricane Katrina evacuation of the Greater New Orleans metropolitan region. The analyses were performed under seven different scenarios to evaluate varying usage (market penetration) and driver compliance rates. Overall, the EVAC functionalities showed positive results in several key aspects of hurricane evacuation travel including congestion reduction, improved mobility, and shorter travel times to resources. Autonomous and self-driving vehicles, as the names imply, are technologies that allow some, or even all, driving tasks to be accomplished independent of driver involvement. Although fully automated control is not yet widely available, most if not all of the technologies and systems required for its use are already in existence. In fact, systems such as adaptive cruise control, automatic braking, and lane-keeping have been in use for many years. Thus, it is likely that the rest of the technologies required for full automation will be gradually adopted more widely.

Chapter 8 · Strategies for Evacuation Management

213

Evacuation-related ideas for automated driving systems, especially for buses and other transit vehicles, relate to the autonomous pick up and transport of carless, low mobility, and special needs evacuees. Other ideas involve the integration of optimized time-dependent routing strategies with self-driving vehicles to maximize network productivity to reduce congestion and travel time during evacuations (Zhang et al. 2015).

References Abdel-Aty, M., Abdalla, M.F. 2004. Modeling drivers’ diversion from normal routes under ATIS using generalized estimating equations and binomial probit link function. Transportation 31 (3), 327–348. Abdelgawad, H., Abdulhai, B. 2009. Optimal spatiotemporal evacuation demand management: Methodology and case study in Toronto. In: 88th Annual Meeting of the Transportation Research Board, Washington, DC. Abdelgawad, H., Abdulhai, B., Wahba, M. 2010a. Multiobjective optimization for multimodal evacuation. Transportation Research Record 2196, 21–33. Abdelgawad, H., Abdulhai, B., Wahba, M. 2010b. Optimizing mass transit utilization in emergency evacuation of congested urban areas. In: 89th Annual Meeting of The Transportation Research Board, Washington DC. Adler, J.L. 2001. Investigating the learning effects of route guidance and traffic advisories on route choice behavior. Transportation Research Part C: Emerging Technologies 9 (1), 1–14. Adler, J.L., Recker, W.W., McNally, M.G. 1993. A conflict model and interactive simulator (FASTCARS) for predicting en-route driver behaviour in response to real-time traffic condition information. Transportation 20 (2), 83–106. Arlikatti, S., Lindell, M.K., Prater, C.S., Zhang, Y. 2006. Risk area accuracy and hurricane evacuation expectations of coastal residents. Environment and Behavior 38 (2), 226–247. Baker, E.J. 2000. Hurricane evacuation in the United States. In Pielke, R., Jr and Peilke, R., Sr. Storms. Routledge, New York, pp. 306–319. Ballard, A.J., Borchardt, D.W. 2006. Recommended Practices for Hurricane Evacuation Traffic Operations. FHWA/TX-06/0-4962-P2. Texas Transportation Institute, College Station TX. Bish, D.R. 2011. Planning for a bus-based evacuation. OR Spectrum 33 (3), 629–654. Bonsall, P.W., Palmer, I.A. 1999. Route choice in response to variable message signs—factors affecting compliance. In Emmerink, R., Nijkamp, P. (Eds.) Behavioural and Network Impacts of Driver Information Systems. Aldershot, Ashgate, England, 181–214. Chatterjee, K., McDonald, M. 2004. Effectiveness of using variable message signs to disseminate dynamic traffic information: evidence from field trials in European cities. Transport Reviews 24 (5), 559–585. Chen, M., Chen, L. Miller-Hooks, E. 2007. Traffic Signal timing for urban evacuation. Journal of Urban Planning and Development 133 (1), 30–42.

214 Chapter 8 · Strategies for Evacuation Management Chen, P.S.T., Srinivasan, K.K., Mahmassani, H.S. 1999. Effect of information quality on compliance behavior of commuters under real-time traffic information. Transportation Research Record 1676, 53–60. Chen, X., Zhan, F.B. 2004. Agent-based modeling and simulation of urban evacuation: relative effectiveness of simultaneous and phased evacuation strategies. In: 83rd Annual Meeting of the Transportation Research Board, Washington, DC. Chiu, Y-C., Villalobos, J., Gautam, B., Zheng, H. 2006. Modeling and solving optimal evacuation destination-route-flow-staging problem for no-notice extreme events. In: 85th Annual Meeting of the Transportation Research Board, Washington, DC. Cova, T.J., Johnson, J.P. 2003. A network flow model for lane-based evacuation routing. Transportation Research Part A: Policy and Practice 37 (7), 579–604. Czajkowski, J. 2011. Is it time to go yet? Understanding household hurricane evacuation decisions from a dynamic perspective, Natural Hazards Review, 12 (2), 72–84. De Maio, M.L., Musolino, G., Vitetta, A. 2012. The role of ITS in evacuation route choice. WIT Transactions on Information and Communication Technologies 44, 503–515. Dia, H., Harney, D., Boyle, A. 2001. Dynamics of drivers’ route choice decisions under advanced traveller information systems. Road & Transport Research 10 (4), 3. Dixit, V.V., Radwan, E. 2009. Optimal scheduling of evacuation orders for cities. In: 88th Annual Meeting of the Transportation Research Board, Washington DC. Dow, K., Cutter, S.L. 2002. Emerging Hurricane evacuation issues: Hurricane Floyd and South Carolina. Natural Hazards Review 3 (1), 12–18. Drabek, T.E. 1999. Understanding disaster warning responses. The Social Science Journal 36 (3), 515–523. Fang, L., Edara, P. 2014. Mobility benefits of intermediate crossovers on contraflow facilities during hurricane evacuation. Transportation Research Record 2459, 37–46. FHwA—Federal Highway Administration. 2007. Using Highways During Evacuation Operations for Events with Advance Notice. United States Department of Transportation, Washington DC. Ghanipoor Machiani, S., Murray-Tuite, P., Jahangiri, A., Liu, S., Park, B., Chiu, Y. C., & Wolshon, B. 2013. No-notice evacuation management: ramp closures under varying budgets and demand scenarios. Transportation Research Record 2376, 27–37. Han, L.D., Chin, S-M., Hwang, H-L., Yuan, F. 2006. Proposed framework for simultaneous optimization of evacuation traffic destination and route assignment. In: 85th Annual Meeting of the Transportation Research Board, Washington, DC. He, S., Zhang, L., Song, R., Wen, Y., Wu, D. 2009. Optimal transit routing problem for emergency evacuations. In: The 88th Annual Meeting of the Transportation Research Board, Washington, DC, January 11 –15, 2009.

Chapter 8 · Strategies for Evacuation Management

215

Huang, S-K., Lindell, M.K., Prater, C.S., Wu, H-C., Siebeneck, L.K. 2012. Household evacuation decision making in response to Hurricane Ike. Natural Hazards Review 13 (4), 283–296. Jahangiri, A., Murray-Tuite, P., Machiani, S., Park, B., Wolshon, B. 2014. Modeling and assessing crossing elimination for no-notice evacuations. Transportation Research Record 2459, 91–100. Jou, R.C., Lam, S.H., Weng, M.C., Chen, C.C. 2004. Real time traffic information and ITS impacts on route switching behavior of expressway drivers. Journal of Advanced Transportation 38 (2), 187–223. Khattak, A.J., Pan, X., Williams, B.M., Rouphail, N.M., Fan, Y. 2008. Traveler information delivery mechanisms: impacts on consumer behavior. Transportation Research Record 2069, 77–84. Khattak, A., Schofer, J., Koppelman, F. 1993. Commuters’ en route diversion and return decisions: analysis and implications for advanced traveler information systems. Transportation Research – A 27 (2), 101–111. Khattak, A., Polydoropoulou, A., Ben-Akiva, M. 1996. Modeling revealed and stated pretrip travel response to advanced traveler information systems. Transportation Research Record 1537, 46–54. Khattak, A.J., Schofer, J., Koppelman, F., Huchingson, R.D. 1991. Factors influencing commuters’ en route diversion behavior in response to delay. Transportation Research Record 1318, 125–135. Levinson, D., Huo, H. 2003. Effectiveness of DMS using empirical loop detector data. In: 82nd Annual Meeting of the Transportation Research Board, Washington, DC. Lindell, M.K. 2017. Communicating imminent risk. In: Rodríguez, H., Donner, W., Trainor, J. (Eds.), Handbook of Disaster Research, Springer, New York, 449–477. Lindell, M.K., Perry, R.W. 1992. Behavioral Foundations of Community Emergency Planning. Hemisphere Press, Washington DC. Lindell, M.K, Perry, R.W. 2004. Communicating Environmental Risk in Multiethnic Communities. Sage Publications, Thousand Oaks CA. Lindell, M.K., Prater, C.S., Gregg, C.E., Apatu, E., Huang, S-K., Wu, H-C. 2015. Households’ immediate responses to the 2009 Samoa earthquake and tsunami. International Journal of Disaster Risk Reduction 12, 328–340. Lindell, M.K., Prater, C.S., Perry, R.W. 2006. Fundamentals of Emergency Management. Emmitsburg MD: Federal Emergency Management Agency Emergency Management Institute. www.training.fema.gov/EMIWeb/edu/ fem.asp or hrrc.arch.tamu.edu/publications/books/. Liu, H.X., Ban, J.X., Ma, W., Mirchandani, P.B. 2007. Model Reference Adaptive Control Framework for Real-Time Traffic Management Under Emergency Evacuation, Journal of Urban Planning and Development 133 (1), 43–50. Liu, W., Corhadi, G., Roden, D., Wolshon, B. in press. Risk reduction impact of connected vehicle technology on regional hurricane evacuations: a simulation study. International Journal of Disaster Risk Reduction. Liu, Y., Luo, Z. 2012. A bi-level model for planning signalized and uninterrupted flow intersections in an evacuation network. Computer-Aided Civil and Infrastructure Engineering 27 (10), 731–747.

216 Chapter 8 · Strategies for Evacuation Management Luo, Z., Liu, Y. 2012. Optimal Location Planning of Signalized and UninterruptedFlow Intersections in Urban Network During Emergency Evacuation. In: 91st Annual Meeting of the Transportation Research Board, Washington, DC. Madanat, S., Yang, D., Yen, Y. 1995. Analysis of stated route diversion intentions under advanced traveler information systems using latent variable modeling. Transportation Research Record 1485, 10–17. Mahmassani, H.S., Liu, Y.H. 1999. Dynamics of commuting decision behaviour under advanced traveller information systems. Transportation Research Part C: Emerging Technologies 7 (2), 91–107. Mileti, D.S., Peek, L. 2000. The social psychology of public response to warnings of a nuclear power plant accident. Journal of Hazardous Materials, 75 (2-3), 181–194. Mileti, D.S., Sorensen, J.H. 1988. Planning and implementing warning systems. In: Lystad, M. (Ed.), Mental Health Response to Mass Emergencies. Brunner/ Mazel Publishers, New York, pp. 321–345. Mileti, D.S., Sorensen, J.H. 1990. Communication of Emergency Public Warnings. Oak Ridge National Laboratory, Oak Ridge, TN. Mitchell, S., Radwan, E. 2006. Heuristic priority ranking of emergency evacuation staging to reduce clearance time. Transportation Research Record 1964, 219–228. Naghawi, H., Wolshon, B. 2010. Transit-based emergency evacuation simulation modeling. Journal of Transportation Safety & Security 2 (2), 184–201. Naghawi, H., Wolshon, B. 2012. Performance of traffic networks during multimodal evacuations: a simulation based assessment. Natural Hazards Review 13 (3), 196–204. Pande, A., Wolshon, B. 2016. Traffic Engineering Handbook, 7th ed. John Wiley and Sons, Hoboken, NJ. Parr, S.A., Kaisar, E. 2011. Critical intersection signal optimization during urban evacuation utilizing dynamic programming. Journal of Transportation Safety & Security 3 (1), 59–76. Parr, S., Wolshon, B. 2016. Methodology for simulating manual traffic control. Transportation Research Record 2562, 9–17. Parr, S.A., Wolshon, B., Murray-Tuite, P.M. 2016. Unconventional intersection control strategies for urban evacuation. Transportation Research Record 2599, 52–62. Perry, R.W., Lindell, M.K., Greene, M.R. 1981. Evacuation Planning in Emergency Management. Heath-Lexington Books, Lexington MA. Pretorius, P., Anderson, S., Akwabi, K., Crowther, B., Ye, Q., Houston, N., Vann Easton, A. 2006. Operational Concept: Assessment of the State of the Practice and State of the Art in Evacuation Transportation Management. FHWA-HOP-08-020. Federal Highway Administration, Washington DC. Robinson, R.M., Khattak, A. 2010. Route change decision making by hurricane evacuees facing congestion. Transportation Research Record 2196, 168–175. Robinson, R.M., and A. Khattak. 2011. Selection of source and use of traffic information in emergency situations. Transportation Research Record 2234 71–78.

Chapter 8 · Strategies for Evacuation Management

217

Robinson, R.M., Khattak, A. 2012. Evacuee route choice decisions in a dynamic hurricane evacuation context. Transportation Research Record 2312, 141–149. Sbayti, H., Mahmassani, H.S. 2006. Optimal scheduling of evacuation operations. In: 85th Annual Meeting of the Transportation Research Board, Washington, D.C. Scheisel, R., Demetsky, M.J. 2000. Evaluation of Traveler Diversion Due to EnRoute Information. Mid-Atlantic Universities Transportation Center Report UVA/29472/CE00/103. Virginia Department of Transportation and US Department of Transportation, Charlottesville VA. Sorensen, J.H., Shumpert, B.L., Vogt, B.M. 2004. Planning for protective action decision making: evacuate or shelter-in-place. Journal of Hazardous Materials 109 (1), 1–11. State of Louisiana, 2016. Louisiana Emergency Preparedness Guide. Baton Rouge, LA. Accessed: 21 May, 2018 at http://www.lsp.org/pdf/2016Emer gencyGuide_English.pdf. Texas Department of Public Safety and Texas Department of Transportation. 2013. 2013 Inland Evacuation Map. Texas Department of Public Safety, Austin TX, accessed 25 May 2018 at www.co.chambers.tx.us/users/emerg/ mgmnt/2013%20EvacuLANES-DPS.pdf. Tsirimpa, A., Polydoropoulou, A., Antoniou, C. 2007. Development of a mixed multi-nomial logit model to capture the impact of information systems on travelers’ switching behavior. Journal of Intelligent Transportation Systems 11 (2), 79–89. Ullman, B.R., Trout, N., Ballard, A.J. 2007. Guidelines for Hurricane Evacuation Signing and Markings, SWUTC/07/0-4962-1. Southwest Region University Transportation Center, Texas Transportation Institute, Texas A & M University System, College Station TX. Vasconez, K.C., Kehrli, M. 2010. Highway Evacuations in Selected Metropolitan Regions: Assessment of Impediments.: Federal Highway Administration, Washington, DC. Wardman, M.R., Bonsall, P.W., Shires, J. 1997. Stated preference analysis of drivers’ route choice reaction to variable message sign information. Transportation Research Part C 5, 389–405. Wolshon, B. 2001. ‘One-way-out’: contraflow freeway operation for hurricane evacuation. Natural Hazards Review 2 (3), 105–112. Wolshon, B. 2002. Planning for the evacuation of New Orleans. Institute of Transportation Engineers. ITE Journal 72(2), 44. Wolshon, B. 2009. Transportation’s Role in Emergency Evacuation and Reentry. National Cooperative Highway Research Program, Synthesis of Highway Practice 392. Washington DC. Wolshon, B., and Lambert, L. 2004. Convertible Roadways and Lanes. NCHRP Synthesis of Highway Practice 340. Washington DC. Wolshon, B., McArdle, B. 2011. Traffic impacts and dispersal patterns on secondary roadways during regional evacuations. Natural Hazards Review 12 (1), 19–27. Xie, C., Lin, D-Y., Waller, S.T. 2010. A dynamic evacuation network optimization problem with lane reversal and crossing elimination strategies.

218 Chapter 8 · Strategies for Evacuation Management Transportation Research Part E: Logistics and Transportation Review 46 (3), 295–316. Xie, C., Turnquist, M.A. 2011. Lane-based evacuation network optimization: An integrated Lagrangian relaxation and tabu search approach, Transportation Research Part C: Emerging Technologies 19 (1), 40–63. Xie, C., Waller, S.T., Kockelman, K.M. 2011. Intersection origin–destination flow optimization problem for evacuation network design. Transportation Research Record 2234, 105–115. Zhang, Y., Prater, C.S., Lindell, M.K. 2004. Risk area accuracy and evacuation from Hurricane Bret. Natural Hazards Review 5 (3), 115–120. Zhang, Z., Parr, S.A., Jiang, H., Wolshon, B. 2015. Optimization model for regional evacuation transportation system using macroscopic productivity function. Transportation Research Part B: Methodological 81, 616–630.

Chapter 9

Evacuation Traffic Modeling and Simulation

Computer-based traffic simulation has served for decades as one of the fundamental tools in the traffic engineering toolbox. It is routinely used to rapidly and affordably evaluate alternative designs, control strategies, and traffic patterns for projects throughout the world. As traffic simulation systems have grown in computational speed and analytical detail, it has also become easier to apply them for evaluating ever larger and more complex networks, carrying ever increasing volumes of traffic. Today, general-purpose traffic simulation systems such as Dynus-T, TRANSIMS, VISSIM, and Paramics, now permit the creation and second-by-second tracking of the movement of hundreds of thousands of individual vehicles over thousands of miles of roadways for simulation periods of several days. Such massive and intricate models also offer the capability to produce high fidelity data on a large scale as well as the ability to study and evaluate the system-wide impacts of near-infinite sets of conditions. Because of these capabilities, these tools also lend themselves, quite nicely, for the modeling and simulation of evacuation traffic processes. Although increases in the number of programs that are being used and the number of analysts using them has been a positive development for evacuation planning, the selection of any particular system for a specific location and hazard can still be challenging. All simulation systems come with varying levels of development effort, input detail, computational speed, output fidelity, and so on. The selection of any specific system can also vary by purpose. Although some analysts prefer to begin with general purpose traffic simulation models applicable to any traffic condition and then adapt them to reflect evacuation-specific conditions, others have tended toward special purpose simulation packages developed specifically for evacuation traffic flow modeling. Some of the more notable of these special purpose evacuation systems include: MASS eVACuation (MASSVAC), NETwork emergency eVACuation (NETVAC), the Oak Ridge Evacuation Modeling System (OREMS), DYNamic

220 Chapter 9 · Evacuation Modeling and Simulation network EVacuation (DYNEV), and the Evacuation Traffic Information System (ETIS).1 This chapter discusses the range of traffic simulation platforms as well as some of the emerging behavioral models that can be used to estimate and represent many of the key inputs used to build the evacuation transportation models, including input assumptions and system selections used to represent key aspects of evacuation processes. This chapter also briefly discusses the modeling and simulation of nonauto modes of travel and the simulation of manual traffic control at intersections and multi-city mega-regions. Finally, although they were covered in detail in prior chapters, several behaviorally-related topics that are relevant to simulation are also highlighted, as are the applications of emerging techniques for the verification, validation, and calibration of evacuation transportation models.

9.1 Background The use of traffic simulation to analyze evacuations is quite different from its more typical applications. In addition to the large demands in concentrated areas over short time durations, the representation of evacuation traffic also requires the modeling of key aspects of the “evacuation process.” As described in Chapters 4–7, these include such factors as the temporal and spatial generation of evacuation and background travel demand, as well as various household activities that occur before and during evacuations. Traffic analyses may also include the movements of school children, carless populations, and various other special needs populations that may require assistance. Evacuation traffic simulation has been used for four decades, over which time simulations have also become larger, more complex, and more detailed—reflecting advances in computational power and knowledge gained from both scientific research and professional experience. Urbanik (1979) was one of the first to conduct a hurricane evacuation transportation analysis, which was immediately followed by ETE analyses that the US Nuclear Regulatory Commission required of all nuclear power plants after the 1980 Three Mile Island nuclear power plant accident. These ETE analyses (see Lindell and Perry, 1992, Chapter 8 for a review of some of these studies’ behavioral assumptions) were motivated by new emergency preparedness regulations (US Nuclear Regulatory Commission 1980) and were evaluated by criteria formulated by Urbanik et al. (1980).

1 Additional detailed discussion of the capabilities and requirements of these models and others can also be found online in “Appendix F: Hurricane Evacuation Models and Tools” of the USDOT “Report to Congress on Catastrophic Hurricane Evacuation Plan Evaluation” (USDOT 2006).

Chapter 9 · Evacuation Modeling and Simulation

221

Among all of the ETE studies conducted at that time, Sheffi et al. (1980) had the most significant impact on the field of subsequent evacuation traffic simulation modeling. Although research and analysis continued throughout the 1980s and 1990s, it was not until Hurricane Floyd in 1999 and a series of major storms in the late 1990s and early 2000s that a significant increase in evacuation simulation studies became evident in the literature. Many of these studies concentrated on topics related to unconventional traffic management techniques and strategies, including, for example, the operational effect of designs to load and unload contraflow to increase outbound roadway capacity (Theodoulou 2003, Lim and Wolshon 2005). In the wake of major hurricanes such as Katrina and Rita in the mid2000s, the rate of published works on evacuation and the use of simulation modeling increased significantly once again. These storms, along with other catastrophic disasters such as the terrorist attacks of September 11th 2001, the 2004 Indian Ocean tsunami, the 2007 Southern California wildfires, and the 2011 Fukushima Daiichi tsunami nuclear power plant Natech disaster (natural disaster triggering a technological disaster), ushered in a new period of enormous research interest in evacuations that led to simulation studies of transit assisted evacuations (Abdelgawad and Abdulhai 2010, Chen and Chou 2009, Zhang and Chang 2014, Naghawi and Wolshon 2010, 2014) and the evaluation of the effectiveness of temporal and spatial phased evacuation planning in urban networks (Liu, Lai, and Chang 2006, Chiu 2004). Over the era of the mid to late 2000s, there were notable advances in the way that evacuation traffic was represented during simulations. These included an evolution from macroscopic traffic modeling (Hobeika and Jamei 1985, Kirschenbaum 1992, KLD Associates 1984) of larger regional networks, particularly to estimate ETEs and network delay characteristics, to the emergence of mesoscopic modeling platforms that were able to undertake regional scale simulation studies at near-microscopic fidelity. Chiu et al. (2008) and Dixit, Montz, and Wolshon (2011), in particular, were valuable for the evaluation of evacuation traffic management strategies, such as contraflow, at the regional level for the Louisiana region during Hurricane Katrina and the Houston-Galveston area during Hurricane Rita. More recently, research has continued to push the computational limits of current computing power to expand the size of hurricane evacuation models to the megaregion level. A notable illustration of this was the Zhang, Spansel, and Wolshon (2013) effort in the early 2000–2010s to model the Gulf Coast megaregion evacuation using mesoscopic measures of performance. One focus of this work was to examine the effect of proactive evacuation traffic management strategies such as contraflow at varying levels of multi-regional evacuation demand. In the future, it is expected that ever larger geographic areas will be modeled at ever greater levels of detail and will, no doubt, also encompass multiple modes of transportation.

222 Chapter 9 · Evacuation Modeling and Simulation

9.2 Levels of Analysis A fundamental initial decision in the simulation process is the determination of the level of fidelity at which the simulation performance indicators should be represented. In broad terms, traffic simulation occurs at one of three levels of abstraction—macro-, meso-, or microlevel. At each of these levels, various aspects of traffic movement, control and design features of the evacuation travel network, and behavioral characteristics are represented in different ways to reflect varying network sizes, available input and desired output levels of detail, and labor required for development and coding. Analysts generally believe that these levels of abstraction also often represent a tradeoff among data input requirements; level of coding effort (and associated labor cost); and the number, type, and level of detail of the output system performance measures that can be collected for analysis. However, even within these three broad areas, there is a range of models that, when viewed all together, create a spectrum of modeling tools. Figure 9.1 shows a recent representation of the spectrum available for conducting analysis on evacuation operations and planning at the macro-, meso-, and micro-levels of simulation. At one end of this spectrum are macro-level simulation systems, in which the representation of traffic flow is often compared to fluid flow through a pipe. At this level of abstraction, only roads down to the functional classification level of “collector-distributor” are typically included in the simulation and the characteristics and movements of individual vehicles and people are aggregated to represent “group averages.” Although they may not achieve complete fidelity of specific traffic and driver behavior, macro-level systems can provide accurate results for small ERSs that have small populations and sparse road

Figure 9.1 Evacuation Modeling Spectrum ETIS HEADSUP

TRANSIMS

CUBE Avenue

DYNASMART-P

OREMS

Integration 2.0

EMME/2

TransCAD

MACRO

From Hardy and Wunderlich 2007

MESO

VISSIM DYNAMIT

Corridor

Region

PCDYNEV

CORSIM

Paramics

MICRO

Chapter 9 · Evacuation Modeling and Simulation

223

networks. In addition, they can often provide approximate results quite rapidly and cheaply for large ERSs that have large populations and dense road networks. In the latter case, these tools are often favored by high level decision makers who only require a broad view of how certain evacuation management strategies are likely to impact evacuee movement rather than precise estimates of the delay at each intersection in the ERS. Some recent macro-level models have even been developed for use in real-time decision support (Post Buckley Schuh and Jernigan Inc. 2006). At the other end of the spectrum are micro-level simulation systems, which yield output at much higher levels of fidelity than the other simulation levels. In microsimulations, detailed performance measures can be produced for individual vehicles and specific locations, even specific intersection approach lanes. This level of analysis permits very focused analyses of the effect of specific designs such as added lanes and contraflow median crossovers, as well as control features such as traffic signals and police manual traffic control. However, this additional detail comes with a price. One major drawback to microscale modeling is the large coding effort required to represent the characteristics of the network in significant detail. This can mean the number, width, and grade of all travel lanes; freeway ramp configuration; traffic signal phasing, cycle length, and coordination; stop and yield signing; and even on-street parking areas and the location of entrance driveways to homes and businesses. In turn, this large coding effort leads to the other major drawback—the limited size of the geographic area and time duration that can reasonably be represented in a model. As a result, most microscale models are used to represent traffic conditions in much smaller areas over several hours, rather than over an entire day or two such as would be required for a hurricane evacuation simulation. These issues can also limit the applicability of microscale modeling for the simulation of other large-scale evacuation scenarios. In between the micro- and macroscale models are mesoscale models which, depending on the modeling system, can incorporate attributes of both micro and macro systems. For example, some mesoscale systems analyze networks by subdividing a longer segment of road into smaller subsegments in which the movement of vehicles can be aggregated to represent “average” flow rates and speeds. An example of such an approach is the “cell transmission” model in which the individual cells used to represent road subsegments can be as long as several miles (if the segment is relatively homogenous in its design and control conditions) or as short as a single vehicle. Models at the latter level effectively permit the creation of near micro-level network representations. When applied strategically, a mesoscale model allows analysts to analyze geographic areas that are considerably larger than those that would be attempted with microscale models, while still permitting the computation of more disaggregate output results than macroscale models. This

224 Chapter 9 · Evacuation Modeling and Simulation can be extremely valuable in evacuation simulation because capacity restrictors such as bottlenecks and traffic incidents can be modeled at high detail, whereas long stretches of rural freeway can be abstracted to a single segment. Because of the obvious advantages that this offers, mesoscale modeling systems are growing in popularity, both in general as well as for the simulation of evacuation scenarios, more specifically. Another approach that has been used in evacuation modeling has been a combination of separate models at different levels of abstraction. As it is well recognized that corridor (and area wide) traffic movements can be effectively governed by the capacity of a few localized areas (such as intersections and ramps) where congestion is likely to occur, there can be advantages to first micro-modeling these locations to assess their flow characteristics then to include those conditions in a separate network-level macro model. Such an approach was evident in the modeling work of the Florida DOT (FDOT) to simulate potential traffic conditions associated with an evacuation of the Florida Keys. Early FDOT analyses showed that the ability of traffic to flow through a few signalized intersections could significantly impact, if not completely determine, the overall clearance of evacuees over the 100-mile segment of US 1 from Key West to the Florida mainland. Another issue was the creation of merge congestion caused by the elimination of travel lanes through more populated islands where additional capacity was needed to serve local traffic. However, neither of these conditions could be captured in detail or would be observable using macro-level modeling. Based on these results, FDOT was able to apply microscale modeling to analyze the influence of localized, facility-specific design features such as intersections, and then use a macroscopic model to analyze traffic conditions at high levels over the full length of US 1. This procedure, conducting an initial microscale modeling coupled with later macroscale analyses, could also be used for integrating microscale analyses of freeway merge areas and contraflow loading and unloading points into regional hurricane evacuation analyses.

9.3 Models of Key Evacuation Variables and Assumptions Although terms such as “models” and “modeling” are often used virtually synonymously with “simulation” and are even commonly used together to describe a “simulation model” or a “traffic model,” a simulation is a representation of key aspects of an actual system of relevance to an analyst that is made up of a group of individual “models” that quantitatively represent particular processes and behaviors. Some of these individual models may be descriptive, meaning that they seek to represent observed phenomena (e.g., car following models), while others may be prescriptive,

Chapter 9 · Evacuation Modeling and Simulation

225

meaning that their outcomes suggest a course of action. An example of the latter is an optimization model, such as might be found in traffic assignment, which provides routes to each vehicle so as to minimize a function of travel time. As such, traffic simulation systems may have models to represent aspects of driver behavior (e.g., car following, lane changing, gap acceptance behaviors) and vehicle performance (e.g., acceleration, braking, climbing) that are influenced by various design and traffic control features that have been implemented at a particular time and location in the simulation process. Similarly, evacuation conditions also create interrelationships among human behavior, perceived threat conditions, and road network characteristics. These are broadly referred to as “evacuation process” variables and are, themselves, also quantified with models. The differences between these categories of variables, as well as the way they impact the simulation of nuclear power plants evacuations, were shown in an, as yet, unpublished study for the USNRC (Wolshon et al. 2018). In this research, either empirically based or theoretically assumed values were used to describe the physical features and processes of evacuations and those used to quantify and mathematically represent the physical features and processes of drivers, vehicles, and roadway characteristics. These were varied within the computational routines of the simulation code in order to determine the potential impacts of variation in these values on the range of ETEs. The variables in the study that were related to evacuation processes included, but were not limited to, the population size within the EPZ; the mobilization time evacuees were expected to require; the amount of background traffic within the simulation road network; and the number of large trucks, buses, and other heavy vehicles within the network. The computational simulation process input variables studied included the free-flow speed of routes in the model; the location and duration of flow disruptions; adverse weather conditions; time step size for system status updating; and the seed for the pseudorandom number generator. Each of the variables was assessed under a range of values that spanned from a realistic “minimum value” to a potential “maximum value.” The goal of the research was to allow the NRC to better understand the influence these variables had on clearance time calculations and deepen the understanding of the sensitivity of evacuation time estimate studies to certain key input parameters. In the sections that follow, a number of process variables important to the simulation of evacuation traffic are discussed in terms of recent simulation modeling applications and research and development efforts to represent their effects during a mass evacuation. These models are grouped into two primary categories, including those associated with the supply side of transportation systems and networks and those associated with the demand side of the systems.

226 Chapter 9 · Evacuation Modeling and Simulation

9.3.1 Demand Side Many traffic simulation tools treat demand as input. Some tools use aggregate demand represented as the number of vehicles that travel from a given origin zone to a given destination zone, leaving at specific time increments. Others allow the specification of individual vehicles and their intermediate stops. The effects of demand management strategies discussed in Section 8.1 are typically modeled outside of the traffic simulation tool. That is, the analysts specify how the demand management strategy affects the rate at which vehicles leave each origin for each destination and then input that demand to the traffic simulation tool. However, as discussed below, the choice of evacuation routes is generally simulated by a model of driver behavior, such as the system optimal (SO) model or the user equilibrium (UE) model, that makes the computations more tractable even though it does not correspond exactly to the pattern of evacuation route choice that has been documented in Section 6.5. As shown in Chapter 7, the number of behavioral components to consider in an evacuation analysis can be quite large, especially when specific models are not available for a particular combination of hazard, timing, and area. For example, there are extensive behavioral data for the response of risk area residents (area) to hurricanes (hazard) that provide ample forewarning (timing). However, the empirical base is quite sparse regarding the behavior of risk area residents (area) to near-field tsunamis (hazard) that provide only a few minutes of forewarning (timing). When possible, percentages (for the aggregate approach in Section 7.3.2.1) or models based on survey information (for the microscopic approach in Section 7.3.2.2) should be used. The analyst also needs to decide which behavior should be incorporated into the study and to have data, models, or (least preferable) assumptions with which to represent those behaviors. For example, the analyst needs to decide whether preparation activities will be modeled or simply the evacuation trip. The demand is likely to vary based on several factors, as discussed in earlier chapters. For planning purposes, since a single estimate of demand is unlikely to precisely represent conditions that would be experienced for a future event, different demand scenarios may be created and used in conjunction with traffic simulation to represent a range of conditions that may be experienced.

9.3.1.1 Evacuation Participation The evacuating population consists of both area residents and visitors from areas inside and outside evacuation zones. Visitors are highly likely to evacuate when advised to do so because their stay in the area is temporary and a return trip is already planned. Of all of the types of

Chapter 9 · Evacuation Modeling and Simulation

227

evacuees, the analyst is most likely to have models for the behavior of residents within the evacuation zones. These models should be used as the basis for developing scenarios by varying the input values, such as conditions that correspond to the hazard and timing parts of the overall scenario. When such models are not available, ranges of participation rates should be considered, as suggested in Table 7.1. Full compliance inside the evacuation zone is typically thought to be a conservative estimate, but this is an inappropriate assumption for two reasons. As discussed in Chapter 4, evacuation warnings are extremely unlikely to gain complete compliance and they almost always produce evacuation shadow. Although these two phenomena might be thought to offset each other, this is not the case because the lowest levels of evacuation shadow are likely to occur with the least threatening hazards—which will produce the lowest levels of evacuation warning compliance. Thus, assuming complete compliance is likely to significantly overestimate evacuation rates. Conversely, the highest levels of compliance are likely to produce the highest levels of evacuation shadow. In this case, assuming complete compliance with the evacuation warning is likely to significantly underestimate evacuation rates. The output of this component of scenario definition is the number of evacuating households and their origins.

9.3.1.2 Travel Modes Travel mode should be considered for the intermediate, background, and evacuation trips. The techniques discussed in Chapter 7 should be used for the evacuation trips, including the number of household vehicles used. As part of creating different demand scenarios, in the aggregate approach, the number of vehicles used per household should be varied using the ranges presented in Section 7.2.2.1.2. In the microscopic approach, data values to which the statistical models are applied can be varied according to the scenario parameters (e.g., storm category). Scenarios should also include transit considerations, but there is virtually no reliable survey data addressing these issues. For example, disaster researchers generally believe that many evacuees without access to reliable personal vehicles will find a ride with someone they know. Some evidence in support of this belief is that Jones et al. (2011) found that 50% of their survey respondents would be willing to give a ride to someone they saw at a bus stop waiting for a bus. However, it is unknown how likely it is that evacuating drivers would actually see people in this situation; unfortunately, the probability can reasonably assumed to be small. It is more likely that those without reliable transportation will contact, or will be contacted by, people that they know personally and offered a ride. However, there is a problem here, too. People who lack reliable vehicles are likely to have friends and

228 Chapter 9 · Evacuation Modeling and Simulation neighbors who themselves lack reliable vehicles. The most appropriate resolution to this problem is to make different assumptions in each TAZ. In TAZs where the vast majority of households have reliable transportation, it seems likely that 50% of those who lack reliable transportation will get rides in other private vehicles. By contrast, in TAZs where the vast majority of households lack reliable transportation, it seems likely that most of those who lack reliable transportation must rely on public buses. In addition to addressing the percentage of evacuees needing to be picked up by peers or public transit (see Section 6.5), evacuation analysts need to estimate the number of buses needed to transport evacuees to public shelters. Jones et al. (2011) recommend estimating bus capacity as 50% of the reported seating capacity due to the space required to accommodate evacuees’ luggage. Other considerations that also need to be addressed are the frequency of initial pickup trips (e.g., from neighborhood schools, the routes taken, and timing of evacuation transit services beginning and ending. The output of this part of the scenario definition should include the transportation mode split or modes assigned to each trip and the number of household vehicles used in the evacuation. It should also explain the rationale for any assumptions about the modal splits for residents lacking reliable transportation.

9.3.1.3 Accommodations and Destinations The question of where these vehicles are going has two components— the type of accommodations and the location of those accommodations. From an emergency management perspective, public shelters need to be provided, but their number and locations should be based on estimation of the public’s needs for such accommodations. As noted in Section 7.2, public shelter use varies according to the size of evacuating population and the timing of disaster impact. As Mileti et al. (1992) reported, a middle of the night evacuation encourages the use of public shelters, at least temporarily. Chapter 7 presents two approaches to determining accommodation distributions. In the aggregate approach, scenarios can be developed by varying the percentage for each accommodation type within the ranges shown in Section 6.2. In the microscopic approach, data values—particularly those related to the hazard—can be varied for different scenarios when the model is applied. For example, the strength of the storm, its time of arrival, and the type and timing of evacuation notices can be varied to obtain different likelihoods of accommodation selection (provided these variables are in the model). The traffic simulation requires the analyst to identify the geographic location of each type of accommodation. For analyses of small evacuation zones, such as for a small chemical facility with a 1-mile VZ radius

Chapter 9 · Evacuation Modeling and Simulation

229

in a suburban area, it is possible to specify the locations of individual public shelters and to identify the locations of individual hotels and motels. However, the locations of peers’ homes can only be identified in terms of geographic areas. For large evacuation zones, such as for a major hurricane striking a densely populated urban area, it is only possible to specify the locations of all three types of accommodations —public shelters, commercial facilities, and peers’ homes—at an aggregate level. For example, Lindell et al. (2001) coded Houston/Galveston Study Area (GSA) coastal residents’ expected evacuation destinations as Inland GSA, East Texas, North Texas, Dallas, Central Texas, Austin, San Antonio, South Texas, and Out of state US. When using the aggregate approach in Chapter 7, the percentages can be varied within ranges derived from post-impact surveys of actual evacuations or from local evacuation expectations surveys. Greater travel distances should be expected for events that have larger evacuation zones. For gravity models, destination percentages can be varied and travel times between origins and destinations can be varied. For discrete choice models, also, some of the data can be varied to correspond to other parts of the scenarios. The outcomes of this part of the scenario definition depend on whether the aggregate or microscopic approach is adopted. For the aggregate approach, the percentage for each accommodation type should be specified as well as the percentage for each geographic destination area. For the microscopic approach, accommodation type and geographic destination information should be produced for each household based on hazard or travel characteristics or both.

9.3.1.4 Departure Times As noted in Chapter 7, some evacuees and stayers will participate in preparation activities prior to evacuation. These intermediate trips should be obtained from behavior models. In the absence of such models, scenarios may be based on people evacuating to safety directly from their locations at the time they receive an evacuation notice, first returning home before evacuating, or directly leaving from home, depending on the hazard and size of the evacuation zone. When assuming that evacuees return home before departing the evacuation zone, traffic patterns (origins, destinations, and volumes) may be based on transportation planning surveys and models. This part of the scenario defines the pre-evacuation activities and origins of the ultimate evacuation trips. The number of evacuating vehicles trying to leave the evacuation zone at the same time strongly influences the amount of congestion. When using the aggregate approach discussed in Chapter 7, different values and combinations of the parameters from Equations 7-5 or 7-6

230 Chapter 9 · Evacuation Modeling and Simulation can form the basis of departure time scenarios. The USACE (1999) recommends three departure time distributions, classified as “slow,” “medium,” and “fast.” Past experience with evacuation modeling has shown that assuming increased response speeds, as would result from more rapid warnings and prior household evacuation preparation, has little effect on ETEs for large area evacuations because ERS clearance is limited by its capacity, particularly local bottlenecks (Tamminga et al. 2011). However, increased response speeds can have a substantial effect on ETEs for small area evacuations because vehicle flow never reaches the ERS’s capacity (see Lindell, Prater and Wu 2002). In the microscopic approach, the data can be varied to generate departure times aligned with the rest of the scenario characteristics. For example, in the Fu et al. (2007) model, type of evacuation notice and hurricane category might be varied for different scenarios. If departure time is estimated as part of a microscopic evacuate/stay model, then separate departure time scenarios are not needed. However, when departure time is analyzed separately, several links to other elements of the scenarios should be addressed. For example, departure times could be determined by the characteristics of the hazard (e.g., availability of environmental cues decrease average departure times) and the rate of warning dissemination (e.g., door to door warnings increase average departure times), and elements of the scenario (e.g., summer increases the number of people in remote areas of parks and, thus, average departure times). Other factors that could be included, depending on the sophistication of the modeling, are the people’s experience with the hazard and evacuation (which affect both informal warning dissemination and household preparation times) and the behavior of neighbors and others in the community (which affects observation of social cues). The outcomes of this part of the scenario definition should be individual departure time or aggregate percentages leaving by a given time.

9.3.1.5 Evacuation Routes Traffic simulation tools typically include at least one traffic assignment technique. It is important to understand and specify what the assignment technique is, since each one has its limitations. System optimum (SO)-based techniques assign traffic to minimize the total vehicle travel time, and could be used as a lower bound on network clearance time. However, some travelers are able to individually obtain a lower travel time if they do not follow the path the SO technique assigns them. The technique by which each individual driver obtains his/her own minimal travel time is the user equilibrium (or stochastic user equilibrium when using perceived travel time). However, the user equilibrium (UE) assignment approach requires knowledge of the travel times on all links in the network. This knowledge is typically assumed to be gained through

Chapter 9 · Evacuation Modeling and Simulation

231

experience, which is not a reasonable assumption during evacuations (Lindell and Prater 2007). It may be more realistic to suppose that evacuees will select good, but not necessarily optimal paths even though background traffic, at least initially, follows habitual paths (Zheng et al. 2010). If the software allows, background traffic and evacuating traffic may be assigned according to different techniques. Other considerations for routing behavior include whether transit routes will be simulated, the degree to which evacuees follow recommended evacuation routes (some take unofficial routes—Lindell et al. 2001), whether travelers have a bias toward using interstates (they do— Dow and Cutter 2002), and whether they can switch routes or determine their routes once the trip has begun (this varies, depending on local traffic management strategies). When allowing route switching, additional assumptions must be made, such as the frequency of information updates and the tolerance for a difference in travel times among routes (see Section 8.3.3). The analysts must specify the rules for determining the next part of the path for enroute route choice models (Pel, Bliemer, and Hoogendoorn 2012).

9.3.1.6 Background Traffic While an evacuation is occurring, particularly in no-notice evacuations, other background traffic is on the roads. To assess the level of background traffic, normal levels of traffic based on transportation planning models can be used. These traffic estimates vary by time of day and should correspond to the timing scenarios (e.g., morning peak, mid-day, evening peak, and night). It may be reasonable to assume that, as notice of an event spreads through official warnings and peer communications, the number of background trips passing near the evacuation zones will decrease. This reduction should be defined in scenarios. A 100% level of normal traffic is conservative, although increasingly implausible as time progresses, whereas no background traffic is overly optimistic. Trucks, which are likely to be part of the background traffic, differ from cars in their vehicle operating characteristics. The Highway Capacity Manual allows a relatively low proportion of trucks in the vehicle flow to be represented as personal car equivalents. The percent of traffic flow that consists of trucks can be estimated in local traffic and freight studies. This part of the scenario definition should provide a profile of the background traffic.

9.3.2 Supply Side Variables and Models All traffic simulation models start with a symbolic representation of the ERS. Many simulation tools allow a direct use of a Geographic Information System (GIS) layer depicting the roads in the ERS or the use of

232 Chapter 9 · Evacuation Modeling and Simulation such a layer as the base upon which the ERS can be drawn within the simulation tool. GIS layers are available from a variety of sources such as DOTs. The network should indicate how the roads are connected and identify their characteristics (e.g., number of lanes, shoulders, and speeds) at a level that is commensurate with the level of the model. At the very least, the simulation needs a measure of the capacity of each road segment. Intersection control and characteristics can also be included in some of the tools. In simulation studies, analysts should specify a base condition network that is close to “normal” conditions and has not been modified by any evacuation management strategies. Supply side strategies for evacuation management are generally associated with the network and its design and control features, which effectively dictate the capacity at the facility level (e.g., intersection, ramp, lane). The rest of this section describes eight evacuation traffic management strategies—contraflow, route closures, ramp closures, use of freeway shoulders, turn restrictions/crossing elimination, signal timing modifications, police manual traffic control, and transit. The next subsections describe the basic procedures for modeling each strategy within an evacuation traffic simulation.

9.3.2.1 Contraflow To implement contraflow in traffic simulation, the direction of the segments should be reversed or a new facility operating in the contraflow direction can be added to the model while removing links or reducing the capacity (or number of lanes) in the original network. Simulating access and egress to the contraflow segments typically requires the addition of links or cells at the appropriate locations. If an entire facility (rather than a subset of lanes on the facility) is being reversed, ramps in the normal direction should also to be removed or have their capacities reduced to zero unless the ramps are also reversible. The analyst should carefully consider the capacity of the lanes on the reversed facility. Although the Highway Capacity Manual provides the standard approaches to determining the capacity of different facilities, traffic counts recorded during the 2005 Hurricane Katrina evacuation in south Louisiana indicated that flow rates in the contraflow lanes were about 75% of the adjacent normally flowing lanes (Wolshon 2008). Although no firm explanation for these lower rates has been determined, this reduced flow is consistent with modeling predictions (Post Buckley Schuh and Jernigan Inc. 2000) and simulation studies (Theodoulou and Wolshon 2004, Lim and Wolshon 2005, Williams et al. 2007). The analyst may consider a scenario where the capacity of the contraflow lanes is 75% of that recommended by the Highway Capacity Manual. Later studies, based on additional evacuation traffic data, continued to show that evacuation traffic flow, even in conventionally-flowing

Chapter 9 · Evacuation Modeling and Simulation

233

lanes and particularly at near-capacity levels, differs from that in nonemergency periods (Wolshon and McArdle 2009). Because of the consistent and compelling evidence of the flow rate variation, FDOT recently recommended the use of Maximum Sustainable Evacuation Traffic Flow Rates (MSETFR) for modeling evacuation traffic in the Florida Keys. Although contraflow might never be used, establishing standard evacuation flow rates for conventionally flowing lanes is particularly important for this chain of islands because, as noted earlier, there is only a single route of egress for over 80,000 residents and visitors. Research designed to provide a quantitative basis and explanation of evacuation flow phenomena was conducted by Dixit and Wolshon (2014). Using field data collected during the evacuations from Hurricanes Ivan (2004), Katrina (2005) and Gustav (2008), along with observations from routine non-emergency conditions, the researchers found that a consistent and fundamental difference exists between traffic dynamics under evacuation conditions and those under routine non-emergency periods. Based on the analysis, two quantities were introduced including “maximum evacuation flow rates” (MEFR) and “maximum sustainable evacuation flow rates” (MSEFR). Based on observation of prior hurricanes, flow rates during evacuations were found to reach a maximum value of MEFR followed by a drop in flow rate to a MSEFR that was able to be sustained over several hours, or until demand dropped below that necessary to completely saturate the section. The researchers suggested that MEFR represents the true measure of “capacity”. These findings are important to a number of key policy shaping factors that are critical to evacuation planning. Most important among these is the strong suggestion of policy changes that would shift away from the use of traditional capacity estimation techniques and toward values based on direct observation of traffic under evacuation conditions.

9.3.2.2 Route Closures and Ramp Closures To model route and ramp closures, a few techniques are possible. One is to delete the appropriate links and nodes from the network. Another option is to place an “incident” on the link(s) with 100% capacity loss for the duration during which the route will be closed. If any traffic is assumed to follow habitual routes, the analyst may have to simulate information provision to the vehicles (e.g., place a CMS to divert the traffic around the closures) to prevent them from being trapped in the simulated network (although in reality, they would be able to divert on their own). These methods are particularly applicable for simulations that rely on computational routines for dynamic and/or user equilibrium-based traffic assignment that rely on routine, non-emergency assumptions of

234 Chapter 9 · Evacuation Modeling and Simulation route choice for an evacuation condition. In the aforementioned nuclear power plant emergency evacuation study (Wolshon et al. in preparation), the use of conventional traffic assignment routines resulted in vastly different assignments of traffic to alternative routes when certain ramp links were eliminated. Often, these new assignments were illogical based on expected real world conditions and “dummy” links and closures were required within the model to reflect more realistic evacuating vehicles movements.

9.3.2.3 Use of Freeway Shoulders To model the use of freeway shoulders in simulation, analysts can add a lane to the appropriate network links or otherwise increase the capacity of that section. The capacity of this link/section should be considered carefully since the Highway Capacity Manual indicates that capacity analyses should consider the presence and width of shoulders or clearance at the side of the road. The Highway Capacity Manual should be consulted to recalculate the capacity of the link, and this value (or a value based on evacuation observations—see Section 9.3.2.1) should replace the original capacity characteristic of the link.

9.3.2.4 Turn Restrictions/Crossing Elimination The way in which turn restrictions and crossing elimination are modeled in evacuation traffic simulation depends on the particular simulation program being used, but generally, movements or links (connectors) need to be removed. If there are no conflicting movements remaining at the intersection, signals can be removed; otherwise, signal timing should be adjusted to re-allocate the green time that was previously assigned to the movements that were removed. To identify the intersections at which crossing elimination should be implemented, techniques fall into two categories: professional judgment and optimization, with simulation being used to assess the impact of the crossing elimination plan. Local professional judgment and experience can indicate which intersections are likely to experience heavy delays and traffic flowing in a primary direction during the evacuation. Conflicts with the primary direction could be the ones eliminated. Optimization has been implemented in a number of different ways. For example, Cova and Johnson (2003) presented a network flow model for identifying optimal lane based evacuation routing plans within a complex urban road network. They used an integer extension of the minimum cost flow problem to generate routing plans that trade total vehicle travel distance for merging, while preventing traffic crossing conflicts at intersections. They next evaluated these plans using capacity analysis and microscopic traffic simulation techniques and

Chapter 9 · Evacuation Modeling and Simulation

235

showed up to a 40% reduction in travel time depending on the network configuration and hazard scenario. Several other researchers (Xie, Waller, and Kockelman 2011, Xie and Turnquist 2011) have expanded Cova and Johnson’s general idea. Traffic simulation has also been used in conjunction with heuristics, such as simulated annealing (Jahangiri et al. 2014) and evolutionary algorithms (Liu and Luo 2012), to identify locations for crossing elimination. The optimization/heuristic studies typically identified different locations or configurations and greater reductions in travel time or evacuation time than professional judgment-based plans tested using the same traffic simulation tool. Simulation can be used to evaluate both local and network wide impacts of the crossing elimination plan. Local impacts could be assessed in terms of queue lengths and delays. The network wide impacts can be assessed through a comparison of network clearance times.

9.3.2.5 Signal Timing Modification As indicated in Section 8.2, there are no “standardized” or “recommended” rules of operation for traffic signal control during evacuations (Wolshon 2009a). Signal timing parameters, particularly cycle length and green time allocation to different phases, can be modified in the simulation software according to the selected software’s user manual. Evaluation of the effects of the modified signal timing should include changes to clearance time and average delay (Chen et al. 2007), among other measures aligned with the purpose of the specific study.

9.3.2.6 Police Manual Traffic Control (MTC) There has historically been limited research on developing simulation models of police MTC at intersections. Parr and Wolshon (2016) analyzed facets of MTC to develop a modeling approach that could be integrated into simulation platforms. Their work examined the factors that influence police decision making, the way decisions were used to implement phase changes, and how effective MTC was compared to automatic signal control. Based on this understanding, the researchers created an algorithm that produced reasonably accurate predictions of police control behavior under a variety of conditions. The approach used to develop the algorithm relied on video observation of police MTC behavior at locations in Louisiana and Florida. The observations, which were made immediately after major sporting events, were used to systematically categorize the observed conditions and create a timeline of events (phase changes, phase length, lane groups, vehicle departures, etc.) that took place within each intersection. Then, data on individual vehicle departures, platoon gaps and intersection blockages, and gaps in the traffic platoon that occurred due

236 Chapter 9 · Evacuation Modeling and Simulation to poor coordination, lack of demand, or travel conditions between intersections were used to create a second by second record that linked departures for all intersection movements, lane groups, phase length and phase sequence, and intersection blockages and gaps. Finally, these data were analyzed using binary logit modeling to calculate the choice probabilities at single-second time steps that either changed the signal during a time interval or let the existing phase remain green. The research suggests that officers allocate right-of-way by first assigning an ordinal priority to the competing approach directions and then allocating green time based on easily observable gaps in the approaching traffic streams. Another notable finding of this work was the remarkable consistency of these behaviors over officers, events, traffic conditions, intersection configurations, and event locations— suggesting that this model is generalizable to evacuations. In follow up research, Parr, Wolshon, and Murray-Tuite (2016) applied the MTC model in a traffic simulation along with a dynamic programing algorithm to identify optimum control strategies for evacuation scenarios. The goal of the research was to assess whether MTC could be as effective as flashing yellow signals or crossing elimination for moving traffic in an urban evacuation. The research findings suggested that MTC was best suited for intersections immediately upstream of a bottleneck or for closely spaced, uncoordinated signals.

9.3.2.7 Transit As most traffic simulation platforms are focused on the movement of vehicles through controlled networks, they tend not to support the direct coding of transit vehicles and their associated pedestrians. For those that do, the analyst needs to designate stops/stations and routes and, if appropriate, the movement of individual persons to and from these locations. Examples of stops in the evacuation context include collection points and shelters. A schedule, frequency, or headway is also typically needed. Wolshon et al. (2009) conducted an evacuation simulation of New Orleans in the wake of Hurricane Katrina to evaluate transit evacuation alternatives for the city. To support this work the analysts adapted the TRANSIMS meso-scale traffic simulation system to create a transitbased evacuation model that included not only the movement of buses, but estimates of the number and location of transit-dependent populations as well as their movements from their homes to bus pickup evacuation locations. The multistep analytic process began by creating a coding procedure to represent the carless population and their activities within the evacuation plans. Then, based on discussions with local officials and other experts, the analysts built alternative evacuation routing and network loading scenarios, based on the 2007 New Orleans

Chapter 9 · Evacuation Modeling and Simulation

237

City Assisted Evacuation Plan (CAEP) and The Jefferson Parish Publicly Assisted Evacuation Plan, into the simulation and integrated them into the auto-based evacuation component of the model. Finally, the analysts ran a series of simulations to generate a distribution of results, mostly notably total evacuation time and average travel time, for analysis and comparison. The results of these analyses, discussed earlier were particularly useful in several respects. First, they were able to demonstrate the processes and conditions of carless evacuees who might be involved in the CAEP portion of the overall New Orleans evacuation. Second, it was able to show the movement conditions (speed, delay, travel time, etc.) of the CAEP buses specifically, separate from the self-evacuator traffic. And third, they made it possible to assess the conditions on different routes, at different times, and under different bus routing scenarios within the context of the observed Hurricane Katrina evacuation conditions.

9.3.2.8 Traffic Incidents As is the case during routine traffic operations, traffic incidents should be expected to occur in any evacuation. With the amount of traffic volume and the potential for more inexperienced drivers moving through unfamiliar areas (some of them with unreliable vehicles), the likelihood of traffic incidents might also be expected to be even higher than routine. Given the enormous volumes generated over compressed periods, the effects of on-road traffic incidents also have the potential to be much more wide ranging, long lasting, and safety threatening. For example, in the Florida Keys where only a single road of egress exists for the majority of an 80,000-person population, a lane-blocking incident could endanger the entire evacuation process. Despite conditions such as these, the effect of traffic incidents are not commonly factored into traffic simulation models or accounted for in evacuation transportation planning. Often this is because these events are deemed random and there are few reliable methods to predict their location and timing. Since many agencies also have plans to utilize motor assistance patrols and other wrecker services, it is possible that traffic incidents are not regarded as significant threat to the overall evacuation process. To begin addressing this issue, Robinson et al. (2009) quantified the effects of traffic incidents during large scale evacuations. They used the Virginia Hurricane Emergency Response Plan evacuation plan and simulation to evaluate traffic incidents in the Hampton Roads region. They conducted simulations of evacuation traffic over a 70-hour period, during which an average of almost 200 randomly selected accidents and 1,400 incidents (including disabled and abandoned vehicles) were generated. The traffic incidents significantly extended evacuation travel

238 Chapter 9 · Evacuation Modeling and Simulation time, although the time required to complete the evacuation of the region increased only marginally if evacuees could be rerouted to other roads within the network. More recently, Collins et al. (2015, 2017) conducted further analyses of data from the Hampton Roads area and concluded that traffic incidents only increased ETEs by about 10%. In Wolshon’s (2009) survey of state and local transportation, law enforcement, and emergency management and response agencies, 25 of 36 (about 70%) of the respondents indicated they had plans to provide en route fuel to out-of-fuel vehicles and 21 indicated that they planned to provide wrecker services to remove disabled vehicles from evacuation routes. From follow up interviews with some respondents, vehicles would only be moved from the travel lanes to locations where they would not create a blockage. Most often, this would be a shoulder rather than a local service facility. The survey questions also permitted a review of the types of agencies that would be coordinating these services. Emergency management agencies were found to most often take the lead in these activities, particularly at the state, rather than local, level. The relatively low numbers in the local agency category were not surprising since several of them indicated they did not have jurisdiction over any evacuation routes within their boundaries and their portion of the overall evacuation network was quite limited. Regional scale traffic modeling has also played a role in determining potential fuel demand during major regional evacuations. These data were then used to predict potential concentrations of fuel run-outs, and position mobile fuel supply trucks near these locations. Gottumukkala (2012) used surveys of evacuees to determine origins and destinations, numbers of vehicles, routes traveled, and amount of fuel at trip origin. This information was used to compute fuel demand, which was then compared to fueling stations’ locations and inventories along evacuation routes in the State of Louisiana. The analysis identified the segments of roadway with the highest fuel deficit. Similar efforts to maintain and predict fuel shortages as well as track supplies and evacuees have been developed through academic/commercial partnerships in Texas. The final concern about traffic incidents is evacuation safety, especially fatality rates. This is an important issue to address because some officials, such as the Pennsylvania governor during the Three Mile Island incident, have been concerned that panicked drivers would cause hundreds of crashes and dozens of deaths during a mass evacuation (Martin 1980). The first study to address this issue systematically was conducted by Hans and Sell (1974), who reviewed data from 54 evacuations involving more than one million persons. The analysts found that there were two automobile-related deaths in one incident, which led them to calculate that the evacuation fatality risk was 8.9 x 10-8 per person-mile, which was higher than the rate during normal conditions (2.4 x 10-8 per person-mile). However, both of the deaths

Chapter 9 · Evacuation Modeling and Simulation

239

were due to drowning, so their relevance to the estimation of traffic accident fatality rates was questionable. A later study of evacuation from the Mississauga train derailment (Burton et al. 1981) reported no evacuation fatalities associated with the incident and a report of evacuation during two Guadaloupe Island volcanic eruptions reported no traffic accidents at all (Bastien et al. 1985). Similarly, Mileti, Hartsough, and Madson’s (1982) study of traffic accidents associated with the Three Mile Island nuclear power plant accident reported that there were no traffic fatalities during the evacuation of an estimated 144,000 persons. Sorensen’s (1987) subsequent analysis of data from 300 chemical incidents involving the evacuation of an estimated 250,000 persons also found no fatalities due to traffic accidents. In a similar analysis, Witzig and Weerakkody (1987) collected data on 320 evacuations from a wider variety of hazards and concluded that fatality risks during evacuation were no greater than during normal driving conditions. Although this review of evacuation risks is limited to small number of studies, many of which are over 30 years old, there appears to have been little subsequent research on this topic. Neither Aumonier and Morrey’s (1990) review for British nuclear regulatory authorities nor Hammond and Bier’s (2015) study of alternative evacuation decision strategies identified any additional research on evacuation risks. However, there are compelling reasons to have confidence in the conclusion that evacuation traffic risks were no greater than during normal driving conditions. First, as Hans and Sell (1974) noted, evacuations generally have average speeds of approximately 35 mph. Indeed, some major evacuations, such as Hurricane Rita, have been noteworthy for freeways filled with stationary vehicles. These slow evacuation speeds are important because the overwhelming majority of motor vehicle fatalities occur at speeds greater than 40 mph. Second, many of the conditions that produce crashes during normal driving conditions are eliminated during evacuations. For example, Lindell et al. (1982) cited evidence from national highway safety data that most fatal accidents involve high speed head-on collisions. Since evacuations are well known for drivers traveling the same direction at low speeds, fatal accidents would be rather improbable. In addition, NHTSA (2008) reported that 36% of crashes involve vehicles that are turning or crossing at intersections and another 22% involve vehicles that run off the road. However, as noted in Chapter 8, turning or crossing maneuvers at intersections are closely managed on arterials and impossible to perform during contraflow operations. Similarly, running off the road is unlikely to occur when traffic is moving slowly. Third, not only are slow speed crashes unlikely, those that do occur are unlikely to produce fatalities or severe injuries. Specifically, NHTSA (2008) reported that less than 1% of all crashes resulted in fatalities and only 11% produced incapacitating injuries. The incidence of such

240 Chapter 9 · Evacuation Modeling and Simulation casualties in slow speed crashes is undoubtedly much smaller still. In summary, there is ample evidence to conclude that the rate of severe casualties during automobile travel in major evacuations is no more than in normal traffic conditions and is likely to be less.

9.4 Model Refinements and Validation The idea behind the need for a calibrated model is that, once it is fitted to field observation, any changes in the performance of the model that result from a modification to the model would be assumed to have a similar magnitude under actual conditions if the same change was made to the real system. Model calibration and validation is a key component of traffic simulation, but can be difficult to achieve for evacuation models because mass evacuations are relatively rare. Consequently, effective model calibration becomes a process of “matching” certain influential behaviors in the model to fit analogous activities under normal conditions. This could mean, for example, adjusting one or more key parameters to influence travel speeds or volumes in the model to more closely replicate observed values. In some prior modeling efforts, volume calibration has been accomplished by making certain routes more desirable than others by adjusting free flow speeds or route function preference. Historically, a problem of performing calibrations is that it has been difficult to acquire field observations on all routes in a simulation network, especially large scale, regional networks. By contrast to model calibration, validation can be accomplished somewhat more readily, although it can still require observations of field conditions and involve tedious confirmation efforts especially, once again, for large area networks coded at a microscopic level. Some typical validation checks for traffic models include confirmation of network connectivity to ensure that modeled vehicles can access evacuation routes the way they could during an actual evacuation, that modeled signal phasing and timing are consistent with actual signal operation, and that the modeled percentages of heavy vehicles and posted/desired travel speeds are similar to field conditions. Chiu et al. (2008) conducted a regional scale simulation study using DynusT to evaluate regional impacts of evacuation strategies for the Houston-Galveston Texas metropolitan region during Hurricane Rita. This model represented a significant advance over prior work in that it was among the first to evaluate the traffic impacts of an evacuation at a regional scale. By contrast, several earlier studies utilized survey data or normal day traffic to calibrate and validate their simulation models. Although this may not be a realistic representation of what actually occurs during a mass evacuation, previous modelers have had little choice due to lack of observed traffic data during evacuations.

Chapter 9 · Evacuation Modeling and Simulation

241

The Federal Highway Administration’s Volume III: Guidelines for Applying Traffic Microsimulation Modeling Software (Dowling, Skabardonis, and Alexiadis 2004), as well as an in depth textbook by Barceló (2010) identified various statistics for calibrating and validating simulation models. In a recent study, Dixit et al. (2011) concluded that the R2 value of the regression line y = b0 + b1x + e between the observed and simulated volumes is the most suitable measure for evaluating evacuation models. Once calibrated and validated, a simulation model should provide a realistic representation of the evacuation process. This model can then be used to quantitatively and qualitatively examine the spatiotemporal characteristics (speed, volume, density, queuing, and congestion formation/recovery) of the evacuating traffic and evaluate transit evacuation plans.

9.5 Megaregion Simulation Zhang, Spansel, and Wolshon (2013) developed a micro level traffic simulation for a megaregion that advanced the scale and scope of evacuation traffic simulation to the limits of the available technology. The simulation used hypothetical mass evacuations of the Gulf Coast from Houston to New Orleans that were first modeled using an experimental traffic demand generation process to create a spatiotemporal distribution of departure times, origins, and destinations based on past hurricane scenarios. The resulting evacuations featured traffic from multiple cities, over several days, with route assignments in multiple and overlapping directions. The modeling and research results suggested that it was possible to scale up and adapt existing models to reflect a simultaneous multi city evacuation covering a megaregion. The movements generated by the demand and operational models were both logical and meaningful and captured key elements of the traffic progression over a large area and extended time durations. The performance output also showed demonstrable benefits of proactive traffic management strategies and the effect of increased and decreased advance warning times (Zhang, Spansel, and Wolshon 2014a). This research project also revealed the limitations of existing modeling and computational processing capabilities, as some of the highest demand conditions would often produce run time errors. As part of a follow up study, Zhang, Spansel, and Wolshon (2014b) examined evacuation phasing in a megaregion. Because of their enormous populations, wide geographic expanses, and frequent locations along oceanic coasts or major inland waterways, megaregions were thought to be particularly vulnerable to evacuation problems. The results of the simulations suggested that the effectiveness of phased evacuations can vary widely based on the level of demand and congestion in the network.

242 Chapter 9 · Evacuation Modeling and Simulation As might be expected, the researchers found that evacuation phasing is only effective in highly utilized networks and, in general, no significant improvements would likely be gained by evacuation phasing if a network was only marginally congested or when no congestion occurred.

9.6 Scenario Development for Evacuation Simulation The scenarios that should be examined with transportation simulation tools depend on the modeling goals. Common goals include the calculation of minimum network clearance time, ETEs, identification of bottlenecks, replication of the most common behaviors (e.g., departures during daylight, favored use of a particular road segment), evaluation of evacuation management strategies (see Chapter 8), and sensitivity analyses. Due to the interactions of natural and physical systems with individual human decision makers, evacuation modeling involves many sources of uncertainty. To handle the scope of uncertainty, many assumptions need to be made in order to both define and limit the scenarios that are simulated and to fit within the time and budget available. These assumptions should be documented in the study report. “Simulation scenarios involve variations of the supply, demand (generated from behavior models), parameters affecting the rules, and/or the rules themselves” (Murray-Tuite and Wolshon 2013, p. 80). The desired performance measures can either be directly obtained from the simulation tool’s outputs or derived by post processing the outputs (provided the right tool was selected). The available time may also require tradeoffs among the number of scenarios examined, computational speed, and the level of detail involved in the behavioral and traffic analyses. Ideally, the scenarios account for the interactions of the hazard and its timing, emergency management, the transportation system, and citizen response. An overview of these interactions is presented in Figure 9.2. (Evacuation process elements of the scenarios were highlighted in Section 9.3.1.) Although a considerable amount of effort may be expended in modeling different aspects of behavior, as illustrated in previous chapters, explicit models are not likely to be obtainable for every possibility. However, the available models can serve as the basis for developing appropriate elements of scenarios.

9.6.1 Hazard Scenarios The hazard scenarios should specify the type of hazard (e.g., hurricane, flood, hazmat release) as well as the event characteristics such as magnitude, impact location, impact area size, and speed of onset (i.e.,

Chapter 9 · Evacuation Modeling and Simulation

243

progression over time). In addition, the scenario should specify the prevailing weather conditions (temperature, wind speed/direction, and precipitation) and the timing of the simulated incident (season, day of week, and time of day). As noted in Chapter 2, the hazard, together with the weather conditions, and time determine the appropriate PARs and their timing, the evacuation demand and evacuation supply strategies selected, and the effectiveness with which those strategies are implemented. They also affect the likely impacts on telecommunications and transportation infrastructure that would impede people’s ability to successfully implement those protective actions or increase the prevalence of traffic incidents. Moreover, they also indicate whether multiple jurisdictions will be affected and, thus, the need to coordinate protective action recommendations, management of evacuation transportation routes, and the number and location of public shelters (Murray-Tuite and Wolshon 2013). Finally, all of these factors affect the behavior of risk area residents and transients, background traffic, and shadow evacuees. The specific behaviors of interest are preparation activities, accommodations/destination choice, travel modes, departure time, and travel routes (see Section 9.3.1). The magnitude of an event helps guide logical assumptions and modeling of the size of the evacuation zone and infrastructure damage. As noted in Chapter 2, scientific and engineering models of specific hazards can be used to identify the size of the impact area. The severity of some events, such as hazmat releases, are indicated by the type of material, the amount released, release rate, and meteorological variables, all of which must be defined in the scenario. Based on the information and models available, analysts should define a set of reasonable scenarios for a given hazard. A single planning scenario is unlikely to accurately capture a future realized scenario, so multiple scenarios should be considered. The selected set of hazard scenarios is defined by the analyst’s judgment, ideally in consultation with subject matter experts who are knowledgeable about the hazard, so the scenarios analyzed reflect the range of plausible conditions.

9.6.2 Timing The timing of hazard impact affects how many people may need to be evacuated and where they are when the hazard strikes, evacuation notices are issued, or evacuation begins. Timing scenarios should consider season, day of the week, and time of day. Table 9.1 shows the permutations of one way of categorizing these considerations. The scenario designations in the table are arbitrary examples; analysts can use their own notation. The season of the year is important in many communities because it determines the numbers of residents and transients, especially tourists.

244 Chapter 9 · Evacuation Modeling and Simulation

Figure 9.2 Overview of Interactions for Scenario Construction

Hazard Size Intensity Location Progression

• • • •

• • • •

Timing Season Day of the week Time of day

• • •

Transportation System Hazard effect Weather Incidents Evacuation strategies

• • •

Emergency management Protective action recommendations and timing Transit provision Evacuation strategy implementation

Behavior •

Residents

•

Tourists

•

Shadow

•

Background traffic

• • • • •

Activities Destination Departure time Modes Routes

For example, summer in coastal areas can dramatically increase the number of tourists. Conservative assumptions typically involve full levels of residents and average tourist levels, unless a special event is taking place, when maximum tourist levels may be assumed (Jones, Walton, and Wolshon 2011). Conversely, winter in northern states can dramatically reduce vehicle speeds on roads or even prevent evacuation altogether. In this case, conservative assumptions may involve substantially lower assumptions about link capacities in terms of the number of vehicles per hour. Another key seasonal difference is the location of children. Rather than being located at schools and day care centers during the week, children may be dispersed throughout the community during the summer—attending camp, in day care, at home, or at other locations either with or without their parents or key caregivers. To limit the number of scenarios, the day of the week is often categorized as either weekday or weekend, although traffic studies indicate that Monday and Friday are different from Tuesday, Wednesday, and Thursday. The latter three days show the largest similarities for the week (Rakha and Aerde 1995). During school year weekdays, students may be assumed to be at schools or day care during the day. Adult permanent residents are dispersed throughout the area performing daily activities and at work locations (Jones, Walton, and Wolshon 2011).

Chapter 9 · Evacuation Modeling and Simulation

245

Table 9.1 Timing Scenarios Scenario designation Season

Day of the week Time of day

T1 T2 T3 T4 T5 T6 T7 T8

Weekday Weekday Weekend Weekend Weekday Weekday Weekend Weekend

Summer Summer Summer Summer Non-Summer Non-Summer Non-Summer Non-Summer

Daytime Evening Daytime Evening Daytime Evening Daytime Evening

During weekends, some residents will be at work, but staffing levels are generally lower than during workdays. Still other residents will be home or engaged in social, recreational, and shopping activities at the types of facilities listed in Table 3.3. Schools may be open for events such as sports and cultural performances, but these are also operating at lower capacities and associated with different transportation needs than on weekdays. Recreational facilities and commercial districts tend to have more users on weekends than weekdays. Day of the week scenarios involving evenings may be consolidated across all seven days of the week to limit the number of scenarios, although one may expect more dispersion for social activities on weekends than during weekdays. Time of day influences a variety of evacuation-related considerations, including people’s locations, the amount of background traffic, people’s access to warnings, and how quickly they can respond to protective action recommendations. Residents tend to be dispersed throughout the community during the daytime hours, but are generally at home during the evening and nighttime hours. These considerations can lead to substantially different daytime and nighttime populations for any given location as well as that population’s need to engage in evacuation preparations such as gathering family members. Evening and nighttime hours both typically have many residents at home, so most families will be united at the beginning of an incident. However, nighttime hours have a lower percentage of the population that is awake which, in turn, affects the warning dissemination and evacuation mobilization time distributions. Warning times, in particular, tend to be longer because a very small percentage of the population is using commercial TV, radio, or social media. Consequently, analysts should check with local emergency managers to determine if warnings would be delivered through channels that have a high penetration of normal activities. Similarly, it is for this reason that coastal authorities sometimes warn people before they go to bed to expect an evacuation notice early the next morning if an approaching hurricane continues on the same track and forward speed. As noted in Table 3.1, emergency managers might use a different

246 Chapter 9 · Evacuation Modeling and Simulation set of warning mechanisms during the night, which would require evacuation analysts to assume different warning distributions for daytime and nighttime events. Normal traffic volumes fluctuate throughout the day, so analysts should assume background traffic during an evacuation will be consistent with normal traffic volumes, origins, and destinations, at least initially (especially for no-notice events). Depending on the relative magnitude of background traffic to the evacuation traffic (which is hazard specific), time of day scenarios may need to be refined to consider morning peak, mid-day, evening peak, and night, to reflect a range of background traffic.

9.6.3 Transportation System During an evacuation, the transportation system, including the road network and the transit system, may be altered by the hazard, weather, and traffic incidents, as well as the evacuation management strategies described in Chapter 8. The hazard can have a direct impact on transportation infrastructure, as when floods or fires make roads impassible (Matherly, Murray-Tuite, and Wolshon 2016). Evacuation traffic managers will prohibit vehicles from using damaged infrastructure, with as much advance notice as possible, both spatially and temporally, so travelers can plan accordingly. Thus, for evacuation modeling, analysts should use hazard maps to identify infrastructure that will be unavailable. Other hazards, such as hazmat releases without a fire, would not necessarily cause infrastructure damage, but would require keeping people out of the area. Other indirect effects could include suspending transit service due to health and safety concerns. For example, rail service may be discontinued after a terrorist attack in the city whether the transit line was the target or not. To address these issues, scenarios should include full transit system availability, partial transit system availability, and full transit system unavailability, as well as variations in the time at which unavailability begins. Weather conditions also affect visibility and the road capacity. The Federal Highway Administration provides ranges of reductions in freeway average speed, free-flow speed, and capacity for light rain/snow, heavy rain, heavy snow, and general low visibility (Federal Highway Administration 2009). The Highway Capacity Manual also includes sections on weather effects. The ranges of traffic variables from these two documents can be incorporated into weather scenarios, particularly for adjusting capacity and speed parameters in the traffic simulation. When combining scenario elements, each combination should be internally consistent. Some combinations would be obviously absurd, such as snow effects combined with a summer hurricane. However, other

Chapter 9 · Evacuation Modeling and Simulation

247

combinations might be more subtly contradictory, as for example, if there were a release of ammonia gas during a rainstorm. Although the rain would reduce evacuation speeds, it would also significantly attenuate the severity of the release because ammonia is water soluble so much of the gas would collect on the ground near the release point rather than being transported downwind and inhaled by people within the protective action zone. Evacuation impediments include work zones and traffic incidents, such as disabled vehicles and vehicle collisions, that reduce road capacity. Traffic incident scenarios, which are characterized by location, severity, and duration, may be selected from historical records (Robinson et al. 2009, Murray-Tuite and Wolshon 2013), random traffic incident models (Murray-Tuite and Wolshon 2013), or a combination of these two sources. An aggregate approach to traffic incidents and construction effects assumes that “one segment of one of the top five highest volume roadways will be out of service and unavailable to evacuees” (p. 6) or closing an Interstate lane in the outbound direction (Jones, Walton, and Wolshon 2011). The effects of traffic incidents should be carefully considered when planning evacuation management strategies that are generally intended to reduce demand or increase capacity. If an evacuation management strategy allows evacuating vehicles to use highway shoulders, this will impair crews’ ability to move disabled vehicles out of the traffic flow. As noted in Chapter 8, provision of traffic information is also a strategy that can be used to manage evacuees’ responses to traffic incidents, road closures, service disruptions, and general congestion. If the analysts’ traffic simulation tool can model drivers’ responses to traffic information, the communication methods (e.g., fixed and portable variable message signs—DMS, 511, radio, Internet, V2X, etc.) and updating frequency should be specified, as well as the percentage of travelers assumed to be receiving this information and complying with its recommendations (Murray-Tuite and Wolshon 2013). Appropriate combinations of these factors depend on the hazard and evacuation zone. Combinations representing a best-case scenario should be considered for the lower bound on ETEs. Other combinations should be considered to generate more realistic scenarios, as well as a plausible worst case scenario that provides a reasonable upper bound.

9.7 Need for Multiple Scenarios Before an event occurs, it is impossible to predict the exact hazard conditions and traffic demand that will arise. Consequently, evacuation modeling studies should include multiple scenarios to obtain ETEs (or other performance measures) for the different situations that could

248 Chapter 9 · Evacuation Modeling and Simulation reasonably be expected. This is true whether an aggregate or microscopic demand modeling approach is selected. Scenario development should define the hazard characteristics, evacuation timing, infrastructure impacts, emergency management actions, and resident and visitor behavior. These elements influence each other and should be jointly considered to form logical scenarios. When using conservative values for each of the variables in a scenario, analysts should be careful about using “worst case” values, which could result in an unrealistic scenario (Jones, Walton, and Wolshon 2011). Although it may be theoretically interesting to understand the worst of the worst situations, practitioners and decision makers are more interested in obtaining reasonable values on which to base evacuation timing decisions. Reports of the evacuation analyses should provide detailed descriptions of the scenarios, the assumptions, and what they mean. One useful way of assessing the results of multiple scenarios is to conduct a sensitivity analysis of the results (e.g., ETEs) to systematic variation in the values of the scenario parameters. Lindell (2008) proposed that evacuation analysts compare their simulation results by defining the minimum, most, and maximum probable values for each evacuation variable. The ETEs associated with each of these values are used to produce a tornado diagram that displays the range in ETE values associated with the variation in the uncertain variables. For example, the tornado diagram for an evacuation analysis of San Patricio County in Figure 9.3 shows that the ETE is 14 hours when all five variables in the sensitivity analysis are set to their most likely values. When all other variables are left at their most likely values but ERS capacity is set to its minimum and maximum values, the ETEs are 12 and 18 hours, respectively. The bars for the other variables are determined in the same way, and the bars are ordered from top to bottom with respect to decreasing size. The tornado diagram indicates that, although there is uncertainty in all of the variables, it is the uncertainties in the ERS capacity and percentage of early evacuating residents that dominates the results. The results of this uncertainty analysis can be used in four ways. First, researchers can use the results of uncertainty analyses to guide the collection of additional survey data (e.g., compliance/spontaneous evacuation rates, early evacuation rates) that generalize across communities. Second, local authorities (emergency managers and transportation planners) can use uncertainty analyses to guide the collection of more precise local data (e.g., the proportion of transit dependent households, hotel/ motel capacities and occupancy rates). Third, emergency managers can use uncertainty analyses to guide changes in evacuee demand by adopting some of the evacuation management strategies discussed in Section 8.1. Fourth, evacuation traffic managers can use uncertainty analyses to guide changes in ERS capacity by adopting some of the evacuation management strategies discussed in Section 8.2.

Chapter 9 · Evacuation Modeling and Simulation

249

Figure 9.3 Tornado Diagram for the San Patricio County Evacuation Analysis

Lindell 2008

The reverse of a sensitivity analysis might be called a specificity analysis. A potential limitation of any evacuation simulation is that many traffic simulation tools require a random number seed to begin the sequence of random numbers used to generate specific values of random variables. If only one random number seed is used in all scenarios, the results might be specific to that random number seed. Thus, analysts should perform multiple replications of each scenario with a different random number seed in each replication before analyzing the distribution of the outputs. When that is done, a tornado diagram equivalent to the one in Figure 9.3 would report the average (across replications) of the minimum, most, and maximum probable values for each evacuation variable.

References Abdelgawad, H., Abdulhai, B. 2010. Managing large-scale multimodal emergency evacuations. Journal of Transportation Safety & Security 2 (2), 122–151. Aumonier, S., Morrey, M. 1990. Non-radiological risks of evacuation. Journal of Radiological Protection 10 (4), 287–290. Barceló J., 2010. Fundamentals of Traffic Simulation. Springer, New York. Bastien, M.C., Dumas, M., Laporte, J., Parmentier, N. 1985. Evacuation risks: a tentative approach for quantification. Risk Analysis 5 (1), 53–61. Burton, I., Kliman, M., Powell, D., Schmidt, L., Timmerman, P., Victor, P., Whyte, A., Wojick, J. 1981. The Mississauga Evacuation. University of Toronto, Toronto.

250 Chapter 9 · Evacuation Modeling and Simulation Chen, C-C., Chou, C-S. 2009. Modeling and performance assessment of a transit-based evacuation plan within a contraflow simulation environment. Transportation Research Record 2091, 40–50. Chen, M., Chen, L. Miller-Hooks, E. 2007. Traffic Signal timing for urban evacuation. Journal of Urban Planning and Development 133 (1), 30–42. Chiu, Y-C. 2004. Traffic scheduling simulation and assignment for area-wide evacuation. In: IEEE Intelligent Transportation Systems Conference, Washington, DC. Chiu, Y-C, Zheng, H., Villalobos, J.A., Peacock, W.G., Henk, R. 2008. Evaluating regional contra-flow and phased evacuation strategies for Texas using a largescale dynamic traffic simulation and assignment approach. Journal of Homeland Security and Emergency Management 5 (1). Collins, A.J., Foytik, P., Frydenlund, E., Robinson, R.M., Jordan, C.A. 2015. Generic incident model for investigating traffic incident impacts on evacuation times in large-scale emergencies. Transportation Research Record, 2549, 11–17. Collins, A.J., Robinson, R.M., Jordan, C.A., Khattak, A. 2017. Development of a traffic model involving multiple municipalities for inclusion in large microscopic evacuation simulations. International Journal of Disaster Risk Reduction, https://doi.org/10.1016/j.ijdrr.2017.12.010. Cova, T.J., Johnson, J.P. 2003. A network flow model for lane-based evacuation routing. Transportation Research Part A: Policy and Practice 37 (7), 579–604. Dixit, V., Montz, T., Wolshon, B. 2011. Validation techniques for region-level microscopic mass evacuation traffic simulations. Transportation Research Record 2229, 66–74. Dixit, V., Wolshon, B. 2014. Evacuation traffic dynamics. Transportation Research, Part C, 49, 114–125. Dow, K., Cutter, S.L. 2002. Emerging Hurricane evacuation issues: Hurricane Floyd and South Carolina. Natural Hazards Review 3 (1), 12–18. Dowling, R., Skabardonis, A., Alexiadis, V. 2004. Traffic Analysis Toolbox Volume III: Guidelines for Applying Traffic Microsimulation Modeling Software. United States Department of Transportation, Federal Highway Administration Report DTFH61-01-C-00181, Washington DC. FHwA—Federal Highway Administration. 2009. How Do Weather Events Impact Roads? US Department of Transportation, Washington DC, accessed 18 November 2017 at www.ops.fhwa.dot.gov/weather/q1_roadimpact.htm. Fu, H., Wilmot, C.G., Zhang, H., Baker, E.J. 2007. Modeling the hurricane evacuation response curve. Transportation Research Record 2022, 94–102. Gottumukkala, N.R. 2012. Fuel demand estimation for hurricane evacuation in Louisiana: evacuee behavior for Gustav, Ike, Katrina & Rita. In:National Evacuation Conference, New Orleans, LA. Hammond, G.D., Bier, V.M. 2015. Alternative evacuation strategies for nuclear power accidents. Reliability Engineering and System Safety 135, 9–14. Hans J.M., Sell, T.C. 1974. Evacuation Risks: An Evaluation, EPA-520/6-74-002. United States Environmental Protection Agency, Washington DC. Hardy, M., Wunderlich, K. 2007. Evacuation Management Operations (EMO) Modeling Assessment: TransportationModleing Inventory, Report No.

Chapter 9 · Evacuation Modeling and Simulation

251

DTFH61-05-D-00002. United States Department of Transportation, Research and Innovative Technology Administration, Washington DC. Hobeika, A.G., Jamei, B. 1985. MASSVAC, a model for calculating evacuation times under natural disasters. Simulation Series 15 (1), 23–28. Jahangiri, A., Murray-Tuite, P., Machiani, S., Park, B., Wolshon, B. 2014. Modeling and assessing crossing elimination for no-notice evacuations. Transportation Research Record 2459, 91–100. Jones, J.A., Walton, F., Wolshon, B. 2011. Criteria for the Development of Evacuation Time Estimate Studies. SAND2010-0016P, NUREG/CR-7002, US Nuclear Regulatory Commission, Washington DC. Kirschenbaum, A. 1992. Warning and evacuation during a mass disaster: a multivariate decision-making model. International journal of Mass Emergencies and Disasters 10 (1), 91–114. KLD Associates. 1984. Formulations of the DYNEV and I-DYNEV Traffic Simulation Models Used in ESF. Washington, DC: Federal Emergency Management Agency. Lim, E., Wolshon, B. 2005. Modeling and performance assessment of contraflow evacuation termination points. Transportation Research Record 1922, 118–127. Lindell, M.K. (2008). EMBLEM2: An empirically based large-scale evacuation time estimate model. Transportation Research A 42, 140–154. Lindell, M. K., Bolton, P. A., Perry, R. W., Stoetzel, G. A., Martin, J. B. & Flynn, C. B. (1985). Planning Concepts and Decision Criteria for Sheltering and Evacuation in a Nuclear Power Plant Emergency. Atomic Industrial Forum/National Environmental Studies Project. AIF/NESP-031. Lindell, M.K., Perry, R.W. 1992. Behavioral Foundations of Community Emergency Planning. Hemisphere Press, Washington DC. Lindell, M.K., Prater, C.S. 2007a. Critical behavioral assumptions in evacuation time estimate analysis for private vehicles: examples from hurricane research and planning. Journal of Urban Planning and Development 133 (1), 18–29. Lindell, M.K., Prater, C.S., Sanderson, W.G., Lee, H-M., Zhang, Y., Mohite, A., Hwang, S-N. 2001. Texas Gulf Coast Residents’ Expectations and Intentions Regarding Hurricane Evacuation. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S., Wu, J.Y. 2002. Hurricane Evacuation Time Estimates for the Texas Gulf Coast. College Station TX: Texas A&M University Hazard Reduction & Recovery Center. Liu, Y., Lai, X., Chang, G-L. 2006. Cell-based network optimization model for phased evacuation planning under emergencies. Transportation Research Record 1964, 127–135. Liu, Y., Luo, Z. 2012. A bi-level model for planning signalized and uninterrupted flow intersections in an evacuation network. Computer-Aided Civil and Infrastructure Engineering 27 (10), 731–747. Martin, D. 1980. Three Mile Island: Prologue or Epilogue. Ballinger Publishing, Cambridge, MA. Matherly, D., Murray-Tuite, P., Wolshon, B. 2016. Traffic management during planned and unplanned emergency events. In: Pande, A. Wolshon, B. (Eds.), Traffic Engineering Handbook. John Wiley & Sons, Hoboken, NJ, 599–636.

252 Chapter 9 · Evacuation Modeling and Simulation Mileti, D.S., Hartsough, D., Madson, P. 1982. The Three Mile Island Incident: A Study of Behavioral Indicators of Behavioral Stress. Shaw, Pittman, Potts and Trowbridge, Washington DC. Mileti, D.S., Sorensen, J.H, O’Brien, P.W. 1992. Toward an explanation of mass care shelter use in evacuations. International Journal of Mass Emergencies and Disasters 10 (1), 25–42. Murray-Tuite, P.M., Wolshon, B. 2013. Assumptions and processes for the development of no-notice evacuation scenarios for transportation simulation. International Journal of Mass Emergencies and Disasters 31 (1), 78–97. Naghawi, H., Wolshon, B. 2010. Transit-based emergency evacuation simulation modeling. Journal of Transportation Safety & Security 2 (2), 184–201. Naghawi, H., Wolshon, B. 2014. Operation of multimodal transport system during mass evacuations. Canadian Journal of Civil Engineering 42 (2), 81–88. NHTSA—National Highway Traffic Safety Administration. 2008. National Motor Vehicle Crash Causation Survey: Report to Congress, DOT HS 811 059. National Highway Traffic Safety Administration, Washington DC. Parr, S., Wolshon, B. 2016. Methodology for simulating manual traffic control. Transportation Research Record 2562, 9–17. Parr, S.A., Wolshon, B., Murray-Tuite, P.M. 2016. Unconventional intersection control strategies for urban evacuation. Transportation Research Record 2599, 52–62. Pel, A.J., Bliemer, M.C.J., Hoogendoorn, S.P. 2012. A review on travel behaviour modelling in dynamic traffic simulation models for evacuations. Transportation 39 (1), 97–123. Post Buckley Schuh and Jernigan Inc. 2000. Southeast United States Hurricane Evacuation Traffic Study. Post Buckley Schuh and Jernigan, Tallahassee, FL. Post Buckley Schuh and Jernigan Inc. 2006. Southeast United States Hurricane Evacuation Study: Evacuation Traffic Information System (ETIS) User Manual. Post Buckley Schuh and Jernigan, Tallahassee, FL Rakha, H., Van Aerde, M. 1995. Statistical analysis of day-to-day variations in real-time traffic flow data. Transportation Research Record 1510, 26–34. Robinson, R.M., Khattak, A.J., Sokolowski, J.A., Foytik, P., Wang, X. 2009. Role of Traffic Incidents in Hampton Roads Hurricane Evacuations. In: 88th Annual Meeting of the Transportation Research Board, Washington, DC. Sheffi, Y., Mahmassani, H.S., Powell, W.B. 1980. NETVAC: A Transportation Network Evacuation Model. Massachusetts Institute of Technology Center for Transportation Studies, Cambridge MA. Sorensen, J.H. 1987. Evacuations due to off-site releases from chemical accidents: experience from 1980 to 1984. Journal of Hazardous Materials 14 (2), 247–257. Tamminga, G., Tu, H., Daamen, W., Hoogendoorn, S. 2011. Influence of departure time spans and corresponding network performance on evacuation time. Transportation Research Record 2234, 89–96. Theodoulou, G. 2003. Contraflow Evacuation on the Westbound I-10 out of the City of New Orleans. Unpublished Masters Thesis, Louisiana State University, Baton Rouge LA.

Chapter 9 · Evacuation Modeling and Simulation

253

Theodoulou, G., Wolshon, B. 2004. Alternative methods to increase the effectiveness of freeway contraflow evacuation. Transportation Research Record 1865, 48–56. USACE—US Army Corps of Engineers. 1999. Northwest Florida Hurricane Evacuation Study Technical Data Report. US Army Corps of Engineers, accessed 14 October 2016 at coast.noaa.gov/hes/hes.html. USDOT—US Department of Transportation. 2006. Catastrophic Hurricane Evacuation Plan Evaluation: A Report to Congress, accessed 11 September 2016. http://www.fhwa.dot.gov/reports/hurricanevacuation/appendixf.htm. USNRC—US Nuclear Regulatory Commission/Federal Emergency Management Agency. 1980. Criteria for Preparation and Evaluation of Radiological Emergency Response Plans and Preparedness in Support of Nuclear Power Plants. NUREG-0654, FEMA-REP-1, Rev.1. US Nuclear Regulatory Commission, Washington DC. Urbanik, T. 1979. Hurricane evacuation demand and capacity estimation. In: Baker, E.J. (Ed.), Hurricanes and Coastal Storms: Awareness, Education and Mitigation, Report 33. Florida State University, Tallahassee, FL, 32–37. Urbanik, T., Desrosiers, A., Lindell, M.K., Schuller, C.R. 1980. An Analysis of Techniques for Estimating Evacuation Times for Emergency Planning Zones, NUREG/CR-1745. US Nuclear Regulatory Commission, Washington, DC. Williams, B.M., Tagliaferri, A.P., Meinhold, S.S., Hummer, J.E., Rouphail, N.M. 2007. Simulation and analysis of freeway lane reversal for coastal hurricane evacuation. Journal of Urban Planning and Development 133 (1), 61–72. Witzig, W.F., Weerakkody, S.D. 1987. Evacuation risks: quantification and application to evacuation scenarios of nuclear power plants. Nuclear Technology 78 (1), 24–33. Wolshon, B. 2008. Empirical characterization of mass evacuation traffic flow. Transportation Research Record 2041, 38–48. Wolshon, B. 2009a. Transportation’s Role in Emergency Evacuation and Reentry. National Cooperative Highway Research Program, Synthesis of Highway Practice 392. Washington DC. Wolshon, B., McArdle, B. 2009. Temporospatial analysis of Hurricane Katrina regional evacuation traffic patterns. Journal of Infrastructure Systems 15 (1), 12–20. Wolshon, B., Lefate, J., Naghawi, H., Montz, T., Dixit, V. 2009. Application of TRANSIMS for the Multimodal Microscale Simulation of the New Orleans Emergency Evacuation Plan - Draft Final Report. Federal Highway Administration, United States Department of Transportation, Washington, DC. Wolshon, B., Parr, S., Jones, J., Herrera, N., Tuncer, E. In preparation. Enhancing Guidance for Evacuation Time Estimate Studies, Sandia National Laboratory, Albuquerque, NM. Xie, C., Turnquist, M.A. 2011. Lane-based evacuation network optimization: An integrated Lagrangian relaxation and tabu search approach, Transportation Research Part C: Emerging Technologies 19 (1), 40–63. Xie, C., Waller, S.T., Kockelman, K.M. 2011. Intersection origin–destination flow optimization problem for evacuation network design. Transportation Research Record 2234, 105–115.

254 Chapter 9 · Evacuation Modeling and Simulation Zhang, X., Chang, G-L. 2014. A transit-based evacuation model for metropolitan areas. Journal of Public Transportation 17 (3), 129–148. Zhang, Z., Spansel, K., Wolshon, B. 2013. Megaregion network simulation for evacuation analysis.In: 92nd Annual Meeting of the Transportation Research Board, Washington, DC. Zhang, Z., Spansel, K., Wolshon, B. 2014a. Performance characteristics of megaregion traffic networks during mass evacuations. International Journal of Transportation Science and Technology 2 (3), 53–72. Zhang, Z., Spansel, K., Wolshon, B. 2014b. Effect of phased evacuations in megaregion highway networks. Transportation Research Record 2459, 101–109. Zheng, H., Chiu, Y-C., Mirchandani, P.B., Hickman, M. 2010. Modeling of evacuation and background traffic for optimal zone-based vehicle evacuation strategy. Transportation Research Record 2196, 65–74.

Chapter 10

Evacuation Termination and Reentry

Managing the reentry of evacuees into their communities at the end of an evacuation is a significant challenge for local officials because evacuees who return too early could frustrate efforts to maintain security in the evacuation zone. Moreover, those who enter the evacuation zone before damage assessment teams complete their work might become exposed to safety hazards from the collapse of unstable buildings and health hazards from moving back into homes that lack essential services such as water, sewer, electric power, and gas (McEntire and Cope 2004). Alternatively, returnees who are prevented from reentry might find themselves parked at traffic control points for hours or days waiting for police to let them into the evacuation zone (Dash and Morrow 2001). Controlling reentry is important because it facilitates more complete debris removal, infrastructure restoration, and property security. However, extended delays can have negative effects on the community. A study examining the evacuation and reentry process in Baños, Ecuador during the predicted volcanic eruption of Mount Tungurahua between October 1999 and January 2000 found that the economic well-being of local businesses was a significant factor in the timing of the reentry process (Lane, Tobin, and Whiteford 2003). Despite the dangers of a potential volcanic eruption, many evacuees attempted to reenter the city prior to issuance of an all-clear message. In another account of this event, Lane (2001) stated “some evacuees believed that their economic recovery was contingent upon a return to the area of high risk” and “since most of the residents of Baños were either directly or indirectly involved with the dominant tourist industry, employees were all dependent on tourists returning to the community” (p. 31). In discussing reentry processes following natural and technological disasters, Stallings (1991) suggested that researchers examine the reentry process by “treating the process of issuing the ‘all-clear’ as analogous to issuing a pre-disaster warning and the decision to return as analogous to the decision to evacuate” (p. 183). In many respects, the analogy is apt. As Wolshon (2009b) notes, local government agencies must prepare for reentry into evacuation zones by assessing hazards (e.g., damaged roads) and facilitating traffic movement (e.g., removing debris and restoring

256 Chapter 10 · Evacuation Termination and Reentry traffic control devices). They are involved in traffic management, including coordinating buses that return carless population segments to their homes. In addition, these agencies need to collaborate with law enforcement in controlling access to evacuated areas until it is safe for residents to return. Such access controls take the form of total exclusion, personnel restrictions, spatial restrictions, and temporal restrictions. Total exclusion quite obviously means that no one is allowed into the evacuation zone other than emergency personnel. In addition to police officers and fire fighters, this includes transportation crews clearing debris from roads and repair crews restoring other infrastructure such as electric and phone lines. Personnel restrictions permit access only to specific categories of people, most commonly those who can provide documentation of residence within the evacuation zone. Spatial restrictions permit access only to certain areas of the evacuation zone and temporal restrictions permit access only at certain times of day. Personnel, spatial, and temporal restrictions are usually imposed in conjunction with each other as, for example, when specific portions of the evacuation zone that have cleared roads are opened to local residents during daylight hours—a policy sometimes called “look and leave.” In many cases, reentry is staged or phased as successive portions of the evacuation zone are cleared for residents to return. Another—unfortunate—similarity to evacuation is that returnees have demonstrated limited compliance with official reentry plans following past disasters, with some seeking to reenter the evacuation zone before local authorities want them to return (Siebeneck and Cova 2008). This lack of compliance is substantially based on lack of information or outright misconceptions about conditions in the evacuation zone—ranging from the availability of electric power and other infrastructure to the prevalence of looting. Of course, there are some important differences between evacuation and reentry, so the latter is not the exact reverse of the former. For example, it is much more difficult for risk area authorities to communicate with residents who might be dispersed over many counties or, in some cases, many states. Moreover, unlike an evacuation in which traffic diverges from the risk area into a much larger safe area, reentry traffic converges from the safe area back into the evacuation zone. This converging traffic can produce significant congestion and substantial frustration for returning evacuees and local officials. Consequently, evacuation planners need to understand the concerns that motivate people’s reentry into the evacuation zone and their likely level of compliance with official reentry plans. Once the typical patterns of evacuee response are understood, it will be possible to develop more effective reentry plans. The following sections of this chapter summarize research and practice on the termination of evacuations when authorities judge that environmental conditions have returned to a level that is safe enough

Chapter 10 · Evacuation Termination and Reentry

257

to allow evacuees to return to their homes. Safety considerations include lingering conditions of the original hazard, such as burning embers from wildfires and residual substances from biological/chemical/radiological releases, as well as damage to buildings and infrastructure in the evacuation zone. Specifically, the next sections describe the extent to which evacuees have complied with past plans for reentry into evacuated areas, the concerns that motivate their behavior, and the sources from which they receive information about the reentry process. The last sections explain the broader context of disaster recovery within which a reentry plan is developed and describe some basic issues that a reentry plan should address.

10.1 Permanent Migration Some households that evacuate from a major disaster will not return, an issue that is relevant to reentry planning because this reduces the number of households that will seek to reenter the evacuation zone. However, only a few studies have examined the overall effects of disasters on permanent migration from evacuation zones. Smith and McCarty’s (1996) study of Hurricane Andrew found that 52% (187, 200) of the households in severely damaged South Dade County moved out of their homes whereas only 10% (166,100) of those in less damaged North Dade County did so. Roughly equal percentages left South Dade because of structural damage to their homes (43%) and infrastructure loss (41%), whereas most of those in North Dade left because of infrastructure loss (87%). The infrastructure loss was more readily restored, so 45% of North Dade residents were able to return to their homes within a week whereas only 6% of South Dade residents were able to return this quickly. The location of the impact zone in the highly urbanized Miami/Dade County metropolitan area meant there was a substantial amount of nearby vacant housing available for temporary occupancy. Consequently, most people were able to relocate elsewhere within the county (80% for North Dade and 74% for South Dade). Nonetheless, one third of those who were displaced by Hurricane Andrew failed to return to their homes during the next two years. In South Dade, the rate of permanent relocation was lowest for those who moved elsewhere in Dade County (28%) and highest for those who evacuated out of state (90%). Oliver-Smith (2005) reached similar conclusions, reporting that Hurricane Andrew displaced 353,000 people, of whom 11% left permanently; half of those resettled in communities within a half hour drive. Hori et al. (2009) examined outmigration from four severely damaged Louisiana parishes following Hurricanes Katrina and Rita. They reported that outmigration ranged from 19–76% and the overwhelming

258 Chapter 10 · Evacuation Termination and Reentry majority of this movement was storm-related (86–98%). Some outmigration was to nearby parishes, ranging from 33% for Orleans to 46% for St. Bernard. This close proximity made it feasible to implement temporary reentry (“look and leave”). There was mixed in- and out-migration in four buffer parishes (–.43 to +5.1% net change in population). The Myers et al. (2008) analysis of Gulf Coast outmigration after Hurricanes Katrina and Rita found that the rates of outmigration were systematically related to community characteristics. There were higher rates of outmigration in communities with greater percentages of disadvantaged populations (lower levels of income per capita, median home valuation, median rent, health insurance; high rates of unemployment, poverty, high school dropouts, and female-headed households), higher levels of housing damage, and denser development. At the household level, a fundamental cause of permanent migration is the loss of housing. Studies of post-storm housing recovery show that it takes approximately two years for moderately damaged housing to recover and extensively damaged housing takes more than four years (Peacock et al. 2014). However, there are documented cases in which housing reconstruction has been delayed indefinitely—leading to “ghost towns” (Comerio 1998). Research to date has identified a number of characteristics of household outmigration after disasters. One consistent predictor of post-disaster return is homeownership. Elliott and Pais (2006), Kim and Oh (2014), Landry et al. (2007), and Paxson and Rouse (2008) all found homeowners were more likely to return. Another predictor of post-disaster return is storm damage. Landry et al. (2007) and Elliott and Pais (2006) found that those with less damage were more likely to return and Kim and Oh (2014) found that households with lower financial losses were more likely to return. A third predictor of post-disaster return is risk perception. Baker et al. (2009), Kim and Oh (2014), Landry et al. (2007), and Paxson and Rouse (2008) found that households with lower expectations of another major hurricane were more likely to return. However, Shaw and Baker (2010) reported that the importance of risk perceptions decreased over time. A fourth predictor of post-disaster return is ethnicity. Fussell et al. (2010), Groen and Polivka (2010), Kim and Oh (2014), and Paxson and Rouse (2008) found that Whites were more likely to return. However, ethnicity might be a proxy for other variables because Fussell et al. (2010) found that race became nonsignificant when controlling for the level of home damage. Similarly, Li et al. (2010) speculated that the higher return rates for Vietnamese-Americans than African-Americans was due to significant community support organized by the local Vietnamese church during the early phases of the recovery process. There also are consistent effects of community bondedness on postdisaster return. Chamlee-Wright and Storr (2009), Fussell et al. (2010), Groen and Polivka (2010) and Paxson and Rouse (2008) reported that households with higher levels of community ties (e.g., integration into

Chapter 10 · Evacuation Termination and Reentry

259

peer networks and emotional attachments to the community) were more likely to return. However, Landry et al. (2007) reported that being born in the stricken community had a negative effect on return likelihood and community tenure (years lived in the community) had no effect on return. There are other variables for which the evidence is less consistent. Regarding age, Landry et al. (2007), Fussell et al. (2010), and Groen and Polivka (2010) reported older evacuees are more likely to return but Baker et al. (2009), Kim and Oh (2014), and Elliott and Pais (2006) reported nonsignificant effects for this variable. The results for income are also inconsistent. Landry et al. (2007) reported higher income households were more likely to return, Baker et al. (2009) and Kim and Oh (2014) reported no effect, and Elliott and Pais (2006) reported higher income households were less likely to return. Finally, there are some variables for which there is insufficient evidence. Landry et al. (2007) reported that evacuees were more likely to return if they were married, less highly educated, and employed before the evacuation. However, this is the only study that reported the effects of these variables.

10.2 Reentry Plan Compliance Several studies have examined issues arising during the reentry process. Dash and Morrow’s (2001) study of Hurricane Georges in the Florida Keys examined the effects of delays in issuing a reentry message. They reported that many residents of the evacuation zone wanted to return to their homes the morning after the hurricane made landfall so they could assess the extent of their property’s damage. There is only one highway from the mainland to the Keys and authorities restricted access to it because they considered it to be unsafe. Over a period of two days, the road was occasionally opened to residents of the Upper Keys, who were less affected by storm damage. Dash and Morrow found that the majority of individuals who experienced reentry delays indicated that they were still likely to evacuate from future hurricane events. However, the data suggested that delays in issuing a reentry message might have had a negative impact on compliance with future evacuation notices by those who learned of the negative reentry experiences only through secondary sources such as the news media. In a study examining the reentry process after Hurricane Rita, Siebeneck and Cova (2008) found low compliance with reentry orders, as only 46.4% of evacuees returned either on or after the reentry dates that State authorities scheduled for their communities. In addition, these researchers found that communicating the reentry plan to evacuees was problematic, as only about half of the evacuees received an all-clear

260 Chapter 10 · Evacuation Termination and Reentry message and even fewer (19.5%) were aware of the Texas Department of Transportation staged reentry plan, in part due to authorities’ inability to communicate reentry messages to evacuees dispersed throughout many cities. The Siebeneck et al. (2013) study of the Hurricane Ike evacuation replicated the results of Siebeneck and Cova (2008) in finding that only a minority (38%) of households complied with official reentry plans. The researchers also found that only 36% of the respondents said they had received an official message about their community’s reentry plan but message receipt was not the determining factor because there was no significant relationship between receipt of a reentry message and compliance with it. However, those who were concerned about reentry traffic were more likely to comply (r = .19) whereas those who perceived greater physical risk of returning were less likely to comply (r = –.20). Finally, there were few demographic variables that were significantly correlated with reentry plan compliance or reentry day, and even the correlations that were statistically significant were quite small.

10.3 Evacuee Concerns Siebeneck and Cova’s (2012) study of the 2008 Cedar River flood in Iowa found that evacuees’ perceptions of flood risk increased from M = 1.9 (on a scale of not at all = 1 to a very great extent = 5) two days before the flood to M = 3.5 on the day they evacuated and further to M = 4.3 while at the evacuation destination. After that, risk perceptions declined to M = 3.7 upon returning home, and M = 3.4 upon entering their homes. Unlike the low level of reentry plan compliance after Rita and Ike, just over 72% of the Cedar River flood respondents complied with the reentry plan schedule and those who did comply had a higher level of risk perception than those who did not comply. As was the case in evacuation departures (Maghelal, Peacock, and Li 2016), some households returned in multiple groups. After the Cedar River flood, nearly one third of the respondents returned with a different group than the one they evacuated with. Those who returned in a different evacuation group had higher risk perceptions, were more likely to have children in the household, and were more likely to have sustained home damage than those who returned in the same evacuation group. In seeking to explain why reentry plan compliance was so low after Hurricane Ike, Siebeneck et al. (2013) found that evacuees were most concerned about loss of utilities (M = 4.1 on a scale of not at all = 1 to a very great extent = 5), followed by looting (M = 3.0), return traffic (M = 2.9), physical risk from damaged structures (M = 2.6), and loss of income (M = 2.4). Regarding physical safety, households believed their homes were much more dangerous when the storm made landfall (M = 4.0)

Chapter 10 · Evacuation Termination and Reentry

261

than when they evacuated (M = 2.6). However, they thought that the danger had returned to normal when they decided to return home (M = 2.6) and actually returned home (M = 2.4). In general, they overestimated the problems they would encounter during the reentry process. The greatest difference between expected and actually experienced reentry issues was the fear of looting (mean difference = 1.81, which was 45% of the response scale), and the smallest difference was the loss of utilities (mean difference = 0.17, which was only 4% of the response scale). The finding about looting is notable because six decades of disaster research have documented that the extent of looting in disasters is grossly exaggerated (Tierney, Bevc, and Kuligowski 2006). Unfortunately, this myth continues to be perpetuated in the movies and news media. Nonetheless, the data from the Hurricane Ike reentry showed that concerns about looting were unrelated to reentry compliance, similar to Lindell and Prater’s (2008) finding that concern about looting did not appear to have inhibited people’s evacuation from Hurricane Katrina or Hurricane Rita. Siebeneck et al. (2013) also examined the extent of differences in the expected and actually experienced reentry problems across sociodemographic segments. The correlation results indicated that ethnic minorities and less educated individuals had greater expectations of looting in the evacuation zone than did other sociodemographic groups. In fact, these demographic segments did report actually experiencing looting at significantly higher levels than other demographic segments. In addition, when making the decision to return home, females, non-elderly persons, lower income households, and renters expected greater problems with lost income than other respondents. Indeed, upon returning home, evacuees who were younger, had lower incomes, were renters, had children under 18, and were ethnic minorities did experience greater levels of this problem. Moreover, females and ethnic minorities expected greater problems with lack of utilities when making the decision to return home. However, only females and less educated evacuees reported actually experiencing these problems. Finally, evacuees with children under 18, those with less education, and ethnic minorities expected greater problems with traffic during the reentry trip than other demographic segments. However, only lower income households and those with lower education levels reported actually experiencing traffic problems. The Siebeneck et al. (2013) results indicate that there was not a significant relationship between geographic factors and reentry concerns. However, the location of the return destination, in particular the magnitude of hurricane risk, was related to higher return concerns. For example, households from hurricane risk areas closer to the coast expressed heightened concerns with looting. This can be attributed to respondents’ beliefs that homes and businesses in these areas had experienced significant damage to their exterior walls and, thus provided a reduced level of protection for the building contents. Moreover, when examining the relationship between county of residence and return concerns, returnees

262 Chapter 10 · Evacuation Termination and Reentry to Galveston and Harris counties experienced significantly less concern about lost income as a result of Hurricane Ike than did residents located elsewhere. This is a surprising finding given the amount of damage to these counties and the consequent disruption of business. One possibility is that these counties have higher percentages of salaried workers who have flexibility in their work schedules. Alternatively, these counties had experienced several major storms in recent decades, so businesses in these areas might have established plans to allow workers to telecommute or work remotely during the recovery process. Finally, Siebeneck et al. (2013) examined whether greater concerns about being stuck in traffic, encountering physical risk associated with damaged structures, and lack of local utilities would yield higher reentry plan compliance and whether greater concern about looting and losing income while away from work would result in lower levels of reentry plan compliance. The results indicate that there was no statistically significant relationship between access to utilities and reentry compliance. Likewise, concern about looters was also unrelated to compliance with official reentry plans. In addition, there was a nonsignificant relationship between expected physical risk and reentry plan compliance. However, evacuees who expected greater problems with being stuck in traffic during reentry were more likely to heed the official reentry plans. This result might be related to returnees initially being turned away during the reentry into Galveston during the week following Ike, where local officials were so overwhelmed by the number of people attempting to return that they had to close Interstate 40 and revise the reentry plan. Because of this, returnees—especially those from Galveston County—might have wanted to avoid the stress and possibility of being turned away, and therefore decided to comply with the reentry plan by waiting for an all-clear message. Last, a higher level of concern related to the physical risk caused by damaged structures was positively related to early return. This is contrary to the proposition that greater expectations of physical risk would be positively related to reentry plan compliance. One possible explanation for the unexpected result is that evacuees felt a need to return to their homes early in order to inspect damage, gather information for insurance claims, and salvage any personal belonging spared by the storm—and that these benefits of an early reentry outweighed any concerns about physical risks.

10.4 Evacuee Information Sources Siebeneck and Cova (2008) reported that evacuees received all-clear messages mostly from local authorities and local news media, followed by peers, national news media, and the Internet. In addition, the extent of evacuees’ reliance on particular sources in making their decision to

Chapter 10 · Evacuation Termination and Reentry

263

return home was (on a 1–5 scale) peers (M = 3.5) > local news media (M = 3.0) > national news media (M = 2.7) > local authorities (M = 2.5) > Internet (M = 1.6). The results also indicated that younger people and individuals with children tended to rely on peers more than other sources. Female gender and lower education levels were positively related to compliance with reentry orders. In a related study of the Hurricane Ike reentry data, Lin et al. (2014) examined the extent to which people change their reliance on different information sources over the course of an evacuation. As Figure 10.1 indicates, more people relied on local news (60%) than any other source before and on the day of the evacuation but the percentage dropped to about 40% while at the evacuation destination and thereafter. The other sources were under 20% each before and on the day of the evacuation but the percentage monitoring national news tripled from 10 to 30% while at the evacuation destination and then returned to its previous level after that. Few people relied on peers until the day they decided to return, at which point this source tripled from about 10 to over 30%. People had nearly uniformly low reliance on the Internet (about 9%) and local authorities (about 12%) throughout the course of the incident. Part of the reason for low reliance on local officials is that evacuees had scattered themselves over many different counties and even states. For example, Table 10.1 shows that evacuees from Houston-Galveston Risk Areas and Lake Sabine Risk Areas found temporary shelter over thousands of square miles.

Figure 10.1 Primary Sources of Emergency Information during the Hurricane Ike Evacuation

From Lin et al. 2014

27

47

44

53

46

37

41

Harris

Hardin

Jasper

Jefferson

Newton

Orange

Total

East TX

* Includes Fort Worth Adapted From Wu et al. 2012

28

Galveston

Evacuation County

20

18

12

12

8

14

40

42

Central TX

9

10

14

4

9

12

10

8

North TX

3

2

2

1

4

3

6

4

West TX

2

1

2

2

0

1

2

6

Coastal TX

5

5

1

6

7

5

6

6

Dallas/TX*

Evacuation Destination

0.3

1

0

0

0

0

2

1

Own City

Table 10.1 Percent of Evacuees Traveling to Different Destinations in Hurricane Rita

18

26

22

21

25

17

8

5

Other State

1

1

0

1

3

1

0

1

Multiple Destinations

796

148

85

140

106

138

52

127

Total (N)

Chapter 10 · Evacuation Termination and Reentry

265

The correlations of demographic variables with information sources tended to be nonsignificant and those that were significant were small. However, the two notable patterns were that households with children were somewhat more likely to rely on local news in their home city (r = .15) and evacuation destination (r = .12), national news (r = .17), and the Internet (r = .16). Those with higher education and income tended to rely less on local news in their evacuation destinations (r = –.12 and –.23, respectively) and the national news (r = –.22 and –.25, respectively). Finally, reliance on local news in their home city, local news in their evacuation destination, and national news were significantly correlated with each other (average r = .43), but noticeably less related to reliance on the Internet (average r = .21), and much less related to reliance on peers (average r = .11). There was a significant relationship between reentry compliance and reliance on peers (r = .18) but not any of the other information sources (average r = .04). Siebeneck and Cova’s (2012) examination of evacuation reentry after the Cedar River flood found that the extent of evacuees’ reliance on information sources in making their decision to return home was (on a 1–5 scale) peers (Mean—M = 3.5) > local news media (M = 3.3) > local authorities (M = 2.7) > Internet (M = 1.8) > national news media (M = 1.6). The pattern of information receipt about the reentry plan was different from the sources of information more generally. More people learned of the reentry plan from the local media (53%) and local authorities (35%) than peers (13%), the Internet (9%), and national news media (5%); 27% did not hear about the reentry plan. The researchers found that evacuees who relied more on local authorities, local news media, and national news media were more likely than those who relied on peers and the Internet to receive information about the reentry plan. Moreover, evacuees who evacuated a longer distance from their homes were more likely to use the Internet as an information source. However, it is important to note that the median evacuation distance was only 13.84 km (8.6 miles) and the average return trip time was only 32 minutes—both of which were much smaller than in hurricane evacuations. However, only reliance on local authorities, local news media, and the Internet were associated with compliance with the reentry plan (returning after the scheduled return date). Finally, respondents who had greater reliance on local authorities (r = .13, p < .05), local news media in hometown (r = .20, p < .01), and local news media in evacuation destinations (r = .16, p < .01) were more likely to report they received reentry plans.

10.5 An Overview of Disaster Recovery Planning In addition to understanding how evacuees’ concerns and information sources influence their compliance with reentry plans, it is important for

266 Chapter 10 · Evacuation Termination and Reentry evacuation planners to understand how their plans fit into their communities’ broader disaster recovery plans. This understanding needs to be achieved before a disaster because local government needs to perform many tasks very quickly after a disaster strikes and these need to be coordinated. For a community’s reentry plan to be compatible with its broader disaster recovery plan, evacuation planners should be aware of the specific functions that must be performed in the disaster aftermath. Table 10.2 identifies four principal disaster recovery functions—disaster assessment, short-term recovery, long term reconstruction, and recovery management (Lindell, Prater, and Perry 2006). The recovery phase’s disaster assessment function should be integrated with the emergency response phase’s emergency assessment function in identifying the physical impacts of the disaster. The tasks performed within this function identify the need for other recovery activities. Short term recovery focuses on the immediate tasks of securing the impact area, housing victims, and establishing conditions under which households and businesses can begin the process of recovery. Many of these must be completed before reentry. Long term reconstruction actually implements the reconstruction of the disaster impact area and manages the disaster’s psychological, demographic, economic, and political impacts. These can be initiated after

Table 10.2 Disaster Recovery Functions Disaster Assessment ■ Damage assessment

■ “Lessons learned”

■ Victims’ needs assessments Short Term Recovery ■ Hazard source control and area protection ■ Emergency demolition ■ Impact area security

■ Repair permitting

■ Temporary shelter/housing

■ Donations management

■ Infrastructure restoration

■ Disaster assistance

■ Debris management Long Term Reconstruction ■ Land use practices

■ Infrastructure resilience

■ Building construction practices

■ Historic preservation

■ Public health/mental health recovery

■ Environmental recovery

■ Economic development

■ Disaster memorialization

Recovery Management ■ Agency notification and mobilization

■ Public information

■ Mobilization of facilities and equipment

■ Recovery legal authority and financing

■ Internal direction and control

■ Administrative and logistical support

■ External coordination

■ Documentation

Source: Lindell et al. 2006

Chapter 10 · Evacuation Termination and Reentry

267

reentry. Finally, recovery management monitors the performance of the disaster assessment, short term recovery, and long term reconstruction functions. It also ensures these functions are coordinated and provides the resources needed to accomplish them. The following section describes each of these functions in greater detail, emphasizing the tasks that are most directly relevant to post-disaster reentry into the evacuation zone.

10.5.1 Disaster Assessment Disaster assessment includes both physical and social impact assessment. Physical impact assessment, which is usually called damage assessment, addresses residential, commercial, and industrial buildings. In addition, there is a need to conduct damage assessment for infrastructure such as water, sewer, electric power, fuel, transportation, and telecommunications systems. Finally, damage assessment also must address critical facilities such as hospitals, police stations, and fire stations. In addition, there is a need for two tasks that are usually initiated after reentry. The first is a social impact assessment, usually called victims’ needs assessment to assure that the available recovery programs are meeting victims’ needs. Second, a “lessons learned” assessment examines the disaster’s physical and social impacts to identify ways in which the mitigation actions can be taken to reduce the community’s hazard vulnerability.

10.5.1.1 Damage Assessment There are three basic types of damage assessment, the first of which is rapid assessment (Schwab et al. 1998). The purpose of rapid assessment is to identify the areas affected by the disaster and the approximate magnitude of the disaster’s physical impacts. It is especially important to assess the need for lifesaving activities very quickly, so rapid assessment should be completed within one to three hours after disaster impact. In turn, this allows emergency managers to determine where there are collapsed buildings requiring search and rescue operations and whether there is a potential for secondary impacts such as hazardous materials releases after an earthquake. A rapid assessment is performed by available police, fire, and public works personnel—both on shift and recalled to duty—to conduct assessments in predetermined geographic sectors of the community. Supplementary data can be provided for a rapid assessment from the private sector organizations that own or operate lifelines and critical facilities. The second type of assessment is the preliminary damage assessment, which is designed to produce counts of destroyed, major damaged, minor damaged, and affected structures (Federal Emergency Management Agency 2016). This level of assessment should be completed within 3 to 4 days, depending on the size and accessibility of the

268 Chapter 10 · Evacuation Termination and Reentry impact area and the number and prior training of the damage assessment teams. The data from the preliminary damage assessment are used to support requests for state and federal disaster declarations. A preliminary damage assessment is often performed by having local government personnel conduct windshield surveys by driving along all of the streets in the impact area. Buildings can then be tagged red, yellow, or green depending on the level of damage and occupant safety, with red tagged buildings being unsuitable for occupancy. Once a preliminary damage assessment has been completed, jurisdiction personnel have the basic information they need to begin reentry planning. Finally, there are site assessments, which are addressed here because they are part of the disaster recovery process even though they do not need to be completed before reentry begins. Site assessments produce detailed estimates of the cost to repair or replace each affected structure. Site assessments might take weeks to complete, depending on the size and accessibility of the impact area as well as the number and training level of the assessment personnel.

10.5.1.2 Other Disaster Assessment Tasks The effects of disasters are not confined to physical damage. Affected communities also need to perform victims’ needs assessments to assess the needs of those individuals and groups who have lost property, been injured, or lost family members. In addition, communities often establish “lessons learned” committees to study the event, determine the ways, if any, that the jurisdiction should modify its shelter in-place, land use plan, building code, and other community operations in light of the disaster impact. Like site assessments, the victims’ needs assessment and “lessons learned” assessment need not be completed before beginning reentry into the evacuation zones.

10.5.2 Short Term Recovery 10.5.2.1 Impact Area Security and Reentry First, there is a need to maintain security in the impact area to ensure residents do not return before it is safe to do so and also to protect vulnerable property from the threat of looting. Addressing these issues requires jurisdictions to develop procedures for residents’ reentry. Unfortunately, Stallings (1991) concluded that there is little research on ending evacuations to guide the planning process and this deficit continues. However, there is anecdotal evidence of problems that have arisen after disasters. The available evidence indicates a need to provide for temporary reentry to remove

Chapter 10 · Evacuation Termination and Reentry

269

essential items (e.g., clothing and medications) and permanent reentry for continuous habitation. In both cases, hazardous conditions must have abated sufficiently to allow people to enter safely. In some cases, hazard abatement might include the demolition of severely damaged buildings and the removal of heavy debris. In addition, proper identification listing a local address is needed to ensure that only residents or authorized reconstruction personnel are allowed to enter. Finally, a jurisdiction must establish basic habitability criteria, such as the restoration of transportation and sewer systems. It is possible to allow people to return before electric power is available because some people have their own generators, but the housing habitability criteria should be established before a disaster strikes. If the disaster has had a regional impact, reentry should be coordinated with neighboring jurisdictions. As noted earlier, some jurisdictions have established access restrictions by zone and time. Local residents’ reentry is restricted to daylight access in areas where infrastructure services are unavailable (“look and leave”). However, as is the case with the original evacuation notice, compliance with the “leave” component of “look and leave” is potentially problematic. If returnees decide to stay, then authorities must decide whether it is worth allocating their overtaxed police force’s limited resources trying to compel people to leave. In any event, as areas are declared to be safe and services are available, they are cleared for permanent access.

10.5.2.2 Infrastructure and Critical Facility Restoration As noted earlier, there are often many households and businesses that cannot resume normal functioning simply because of the lack of potable water, sewer, electric power, fuel, telecommunications, or transportation—not because of damage to their homes or places of business. Consequently, there is a need to inspect and repair any damage to pipelines and power lines, as well as streets, bridges, street signs, and streetlights. In addition to returning these households and businesses to normal functioning, restoration of infrastructure to these areas also provides places where emergency workers and construction crews can live while they are rebuilding the structures that are too badly damaged to be occupied. Since it is not possible to restore all infrastructure facilities immediately, communities need to establish a coordinated set of priorities for infrastructure restoration before a disaster strikes. There will also be a need to quickly repair critical facilities such as hospitals, police stations, and fire stations. Moreover, a community’s public infrastructure is also served by other critical facilities such as water treatment plants, transit bus barns, public works equipment yards, and government offices. There is also privately operated

270 Chapter 10 · Evacuation Termination and Reentry infrastructure that includes electric power stations, television and radio facilities (both stations and broadcast towers), and telephone switching facilities. As is the case with infrastructure restoration, communities need to establish a coordinated set of priorities for critical facility restoration before a disaster strikes.

10.5.2.3 Temporary Shelter and Housing Since Quarantelli (1982), disaster victims’ accommodations have commonly been classified as progressing from emergency shelter (unplanned and spontaneously sought locations that are intended only to provide protection from the elements) through temporary shelter (which includes food preparation and sleeping facilities) and temporary housing (which allows victims to reestablish household routines in nonpreferred locations or structures) to permanent housing (which reestablishes household routines in preferred locations and structures). Evacuees typically find temporary shelter outside the evacuation zone in the homes of friends and relatives, commercial facilities such hotels and motels, or mass care facilities such as auditoriums and gymnasiums. Evacuees whose homes remain habitable will typically seek to return to them as soon as possible. In addition, many evacuees whose homes have been damaged or destroyed often seek temporary housing, and later permanent housing, as close as possible to their homes.

10.5.2.4 Temporary Business Operation Just as households need temporary housing, so too businesses need temporary operating locations when their normal locations have been severely damaged or destroyed. Many small businesses have customers who are loyal enough to travel an extra distance, but loyalty does have its limits. Consequently, government might need to permit the establishment of temporary business operations in parking lots or other open spaces that are close to the displaced businesses’ normal locations. Such temporary locations might be needed for as much as a year (and even longer in some cases).

10.5.2.5 Other Tasks Most of the natural disasters, and explosions among the technological disasters, can destroy many structures. In addition, chemical or radiological hazards can contaminate structures. In turn, this can produce significant problems of debris management. For those structures that are in imminent danger of collapse, procedures are needed to determine if they require emergency demolition. By contrast, structures that have minimal to moderate damage will require repair permitting to

Chapter 10 · Evacuation Termination and Reentry

271

determine which of them will be eligible for reoccupancy based upon the percent damage to the different elements of the building—foundation, wall, and roof systems, exterior walls, interior walls, floors and flooring materials, plumbing, electrical systems, and heating/ventilation/ air conditioning systems (see Table 10.3). Major disasters frequently produce an outpouring of material assistance that typically includes a large number of useful items that must be sorted and useless items that must be discarded, which creates a problem of donations management. Finally, the large number of other people attempting to contact many different government agencies and nongovernmental organizations for food, clothing, shelter, and financial support requires coordinated disaster assistance, often at a single site that is conveniently located to returnees’ homes. Debris management, emergency demolition, repair permitting, donations management, and disaster assistance should all be initiated as soon as possible after a disaster but do not need to be completed before reentry into the evacuation zone.

10.5.3 Long Term Reconstruction This function comprises nine tasks that are essential parts of the recovery process but none of them needs to be completed before reentry into the evacuation zone. Effective hazard source control and area

Table 10.3 Infrastructure Facilities Facility Type

Examples

Water

Pumping stations and pipelines

Sewer

Pumping stations and pipelines

Electric power

Generating stations and power lines

Liquid (e.g., oil) and gas (e.g. natural gas) fuels

Pumping stations and pipelines

Telecommunications

Broadcast studios, transmission towers Cellular telephone towers Telephone switching centers and telephone lines

Transportation

Bus/truck terminals and roads Rail yards and rail lines Sea and inland marine ports Airports

Public safety and health

Police and fire stations Ambulance garages Hospitals

Source: Lindell et al. 2006

272 Chapter 10 · Evacuation Termination and Reentry protection anticipates induced growth in the protected area if measures will be implemented to control a hazard source (e.g., an upstream dam) or protect specific areas (e.g., local levees). Local planners should reexamine their communities’ land use practices by revising existing land use plans and passing new ordinances that will limit the potential for increases in hazard exposure by controlling development in hazard prone areas. In addition, planners should reexamine their communities’ building construction practices, such as requirements for elevating structures located in floodplains because this will also limit increases in structural vulnerability and, if done extensively, even reduce it. There also is a need for public health/mental health recovery tasks by involving mental health professionals as victim advocates, especially for victims who are unaccustomed to working with white collar bureaucracies (Salzer and Bickman 1999). Other recommendations include designing community interventions to provide social support by establishing victim locator systems, facilitating self-help groups, and community organizing (Salzer and Bickman 1999). Planners should also provide guidance on the economic development of the disaster stricken areas and increasing the resilience of community infrastructure. In the latter case, for example, roads and bridges can be strengthened and aboveground lines can be undergrounded to reduce their vulnerability to wind and ice. In some cases, pipelines for water, sewer, and fuel and major transmission lines for electric power and telephone can be rerouted to reduce vulnerability. Historic preservation can be facilitated by decreasing the vulnerability of undamaged historic structures to determine how to protect them from future disasters (Spennemann and Look 1998) and environmental remediation might be necessary if hazardous materials spills have occurred (Lindell and Perry 1997, Showalter and Myers 1994). Finally, disaster memorialization can play an important part in the recovery of a community’s sense of identity and pride by providing a carefully designed, transparent, and participatory process that promotes community healing. All of these tasks are important components of disaster recovery but none of them need to be completed before reentry into the evacuation zone.

10.5.4 Recovery Management 10.5.4.1 Agency Notification and Mobilization Unlike the incident management function performed during emergency response (see Perry and Lindell (2007)), the recovery management function performed during the disaster recovery does not require special procedures for agency notification and mobilization because agencies are well aware of the disaster by the time recovery is initiated.

Chapter 10 · Evacuation Termination and Reentry

273

10.5.4.2 Mobilization of Recovery Facilities and Equipment Recovery management does require the mobilization of recovery facilities for donations management, debris management, and disaster assistance. As noted earlier, a community with a large population of displaced victims and a small housing vacancy rate might need to develop one or more mobile home parks to provide enough temporary housing. Rapid mobilization of such facilities requires preimpact screening to identify appropriate sites. Site selection criteria should, of course, include suitable zoning and access to utilities such as water/sewer, fuel and electricity. In addition, planners should focus on sites that have access to public transportation and close proximity to the types of jobs that will be held by a low income population.

10.5.4.3 Internal Direction and Control There is a need for internal direction and control among agencies within the jurisdiction because many aspects of the recovery process require multiagency coordination. Disaster recovery typically involves local government agencies in tasks that are more like their normal duties than is the case for the emergency response. Thus, the allocation of recovery functions to agencies will be relatively simple. In addition, disaster recovery does not require an equivalent to an Incident Commander who oversees the emergency response. Instead, different departments will usually be coordinated by a disaster recovery committee. Finally, there is less time pressure during the disaster recovery than during the emergency response, so this committee’s meetings can be scheduled for daily or, later, weekly frequency.

10.5.4.4 External Coordination There is a need for external coordination, especially in presidentially declared disasters, in order to access the resources of other jurisdictions and higher levels of government. As is the case for internal direction and control, there should be a clear understanding of which agencies will address each disaster recovery function. In addition, local agencies need to understand the restrictions associated with different state, federal, nongovernmental organization, and community based organization programs.

10.5.4.5 Public Information There is also a need for public information, especially to inform disaster victims about recovery policies and procedures. In addition, there is a need to inform other citizens about the progress of the recovery. Thus, the recovery plan should describe procedures for disseminating public

274 Chapter 10 · Evacuation Termination and Reentry information during disaster recovery. These should describe which agencies will be the source of each type of information, what will be the general content of their messages, and what communication channels they will use. General information about the recovery process and sources of additional information can be distributed through the traditional news media (TV, radio, and newspapers), as well as the Internet and social media. Brochures can be targeted at individuals and organizations located in vulnerable zones (before a disaster strikes) or impact areas (after a disaster strikes). Telephone hotlines and local government websites can be useful for answering questions about the recovery process, and a full time public information officer should be on staff at the disaster assistance center during short term recovery.

10.5.4.6 Recovery Legal Authority and Financing The recovery committee needs to obtain legal authority for a wide range of short term recovery actions including a development moratorium, temporary repair permits, demolition regulations, and zoning for temporary housing (Schwab et al. 1998). This committee also needs to explore the feasibility of, and legal authority for long term recovery tasks.

10.5.4.7 Administrative and Logistical Support During the recovery period, the pace of operations decreases so the management of specific emergency response and recovery functions does not need to be focused at incident scenes in the Incident Command System (ICS) or centralized in the EOC. Thus, the activities performed by the ICS’s Planning, Logistics, and Administration Sections within the IMS are gradually dispersed back to the jurisdiction’s normal departments. Nonetheless, special provisions are required to support the additional staff generated by obtaining mutual aid personnel from other jurisdictions and volunteer personnel such as architects and engineers used as building inspectors. Moreover, records accumulated by the ICS’s Finance Section must be available to provide a justification for expenditures on disaster recovery and hazard mitigation that are reimbursable by state and federal agencies.

10.5.4.8 Documentation As is the case in the emergency response, documentation is needed during disaster recovery to provide the basis for organizational learning. Maintaining an event log of who took what actions in response to what conditions will provide the recovery committee with the information it needs to produce the “lessons learned” document and, later, to revise the recovery plan. In addition, detailed documentation provides the

Chapter 10 · Evacuation Termination and Reentry

275

jurisdiction’s legal counsel with the information that might be needed to defend against any lawsuits.

10.6 Reentry Planning Principles The Siebeneck et al. (2013) finding that Hurricane Ike respondents were unable to accurately gauge the extent to which they would encounter problems during reentry could be problematic because returnees’ overestimation of the problems they can expect to encounter during the reentry phase could affect their level of preparedness upon returning home. Specifically, returnees that are expecting to encounter problems that do not materialize may be less prepared to handle the problems that actually do arise either while making the return trip or upon arriving at their homes. Conversely, returnees’ ability to accurately predict the extent to which they will encounter problems such as loss of utilities may enable them to cope more effectively by adjusting their reentry timing or by preparing in advance to deal with this issue upon returning home. Overall, the differences among demographic segments in these expected and experienced reentry problems are consistent with research on socially vulnerable population segments—people who have limited resources for responding to, and recovering from, disasters. The available evidence indicates that evacuees with limited resources experience heightened concern when making the reentry decision and experience greater problems upon returning home. For example, evacuees with hourly wage jobs might experience greater pressures to return early because of concerns about lost earnings while away from home (Lane 2001). This would be a moot point if the businesses that employ them remain closed but evacuees may need to reenter the evacuation zone just to find out if those businesses are closed. Thus, local officials should consider the needs of socially vulnerable population segments when devising their reentry plans. In addition, they should work with local businesses to facilitate communication between management and employees.

10.6.1 Returnees’ Information Sources Reentry plans should recognize that reliance on information sources shifts over time (Siebeneck and Cova 2008). Nonetheless, the primary source that households utilize for reentry information is local news media even when they decide to return home. Although people rarely report relying most on peers (e.g., friends, relatives, neighbors, or coworkers) as sources of general emergency information, they do rely on them extensively for

276 Chapter 10 · Evacuation Termination and Reentry reentry information. This result underscores the need for emergency managers to consider the issue of returnees’ reliance on different information sources during their community reentry process. For example, if returnees rely more on peers than local authorities when they decide to return, inaccurate information may be spread. Thus, emergency managers need to improve their communication of reentry information in order to enhance households’ reliance on them for reentry information. One important action for emergency managers is to advise residents before and during evacuation to seek reentry information on their home community’s website and social media such as Facebook.

10.6.2 Perception of Danger Households have greater perceptions of danger at their homes only during physical threat conditions (Siebeneck and Cova 2008). Conversely, their perceptions of danger are likely to be relatively similar at the time they evacuate, decide to return, and actually return home. In addition, only households that live in multi-family structures with one or two stories have greater risk perceptions. If authorities make evacuees aware of dangerous conditions at their homes in a storm’s aftermath, they might be less likely to return home early rather than complying with official orders.

10.6.3 Demographics and Reliance on a Particular Source Understanding different population segments’ preferences for information sources can assist emergency managers in targeting the sources that will distribute reentry information most effectively. Unfortunately, the results regarding the correlations of demographic characteristics with reliance on information sources were not all consistent with previous research. Indeed, the only consistent result is the significant negative correlation between education level and the extent of reliance on national news media (Siebeneck and Cova 2008, Siebeneck et al. 2013). Future research might find significant relationships between demographic characteristics and reliance on different information sources, but it is quite possible that this search will prove no more successful than the search for demographic correlates of evacuation decisions, which Baker (1991) and Huang et al. (2016) have found to be inconsistent across studies. As a practical matter, the apparent absence of well defined audience segments—specific population segments that only receive information through specific media channels— indicates that local authorities need not develop audience segmentation strategies that target individual demographic groups. Nonetheless, it is important to disseminate reentry and recovery information through a

Chapter 10 · Evacuation Termination and Reentry

277

wide variety of channels because individual households vary in their utilization of specific channels even if broad demographic groups do not.

10.6.4 Reentry Compliance and Reliance on a Particular Source The available research reveals that no particular reentry information source is related to reentry compliance. Thus, local authorities need to distribute reentry information to a variety of different sources in hopes of improving the rate of reentry compliance.

10.6.5 Reentry Plan and Reentry Process The limited data available indicates that people who rely on many information sources are more likely to receive their community’s reentry plans and tend to be more satisfied with the reentry process. People who receive official messages about their communities’ reentry plans are also more likely to have higher satisfaction with the reentry process and a better understanding of those reentry plans. Nonetheless, the extent to which people understand their communities’ reentry plans tends to be low, regardless of the number and type of their reentry information sources. Moreover, there have been no significant correlations between respondents’ understanding of reentry plans and their reentry plan compliance, or between their receipt of reentry plans and their reentry plan compliance. Thus, local authorities need to identify effective ways to communicate with evacuees that have relocated to distant communities. For example, hurricane preparedness brochures distributed in coastal jurisdictions might let risk area residents know that they should look for reentry information on specific web pages of their home jurisdictions’ web sites and social media accounts.

References Baker, E.J. 1991. Hurricane evacuation behavior. International Journal of Mass Emergencies and Disasters 9 (2), 287–310. Baker, E.J. 2009. Shadow evacuation in Hurricanes. In: Association of American Geographers Annual Conference, Las Vegas, NV. Chamlee-Wright, E., Storr, V.H. 2009. There’s no place like New Orleans: sense of place and community recovery in the Ninth Ward after Hurricane Katrina. Journal of Urban Affairs 31 (5), 615–634.

278 Chapter 10 · Evacuation Termination and Reentry Comerio, M.C. 1998. Disaster Hits Home: New Policy for Urban Housing Recovery. University of California Press, Berkeley CA. Dash, N., Morrow, B.H. 2001. Return delays and evacuation order compliance: the case of Hurricane Georges and the Florida Keys. Environmental Hazards 2 (3), 119–128. Elliott, J.R., Pais, J. 2006. Race, class, and Hurricane Katrina: social differences in human responses to disaster. Social Science Research 35 (2), 295–321. FEMA—Federal Emergency Management Agency. 2016. Damage Assessment Operations Manual. Federal Emergency Management Agency, Washington DC, accessed 12 March, 2017 at www.fema.gov/media-library/assets/docu ments/109040. Fussell, E., Sastry, N., VanLandingham, M. 2010. Race, socioeconomic status, and return migration to New Orleans after Hurricane Katrina. Population and Environment 31 (1–3), 20–42. Groen, J. A., Polivka, A.E. 2010. Going home after Hurricane Katrina: determinants of return migration and changes in the affected areas. Demography 47 (4), 821–844. Hori, M., Schafer, M.J., Bowman, D.J. 2009. Displacement dynamics in southern Louisiana after Hurricanes Katrina and Rita. Population Research and Policy Review 28 (1), 45–65. Huang, S-K., Lindell, M.K., Prater, C.S. 2016b. Who leaves and who stays? A review and statistical meta-analysis of hurricane evacuation studies. Environment and Behavior 48 (8), 991–1029. Kim, J., Oh, S.S. 2014. The virtuous circle in disaster recovery: Who returns and stays in town after disaster evacuation? Journal of Risk Research 17 (5), 665–682. Landry, C.E., Bin, O., Hindsley, P., Whitehead, J., Wilson, K. 2007. Going home: evacuation-migration decisions of Hurricane Katrina survivors. Southern Economic Journal 74 (2), 326–343. Lane, L.R. 2001. Hazard Vulnerability in Socioeconomic Context: An Example from Ecuador. M.A. Thesis, University of South Florida, Tampa FL. Lane, L.R., Tobin, G., Whiteford, L.M. 2003. Volcanic hazard or economic destitution: hard choices in Baños, Ecuador. Global Environmental Change Part B: Environmental Hazards 5 (1–2), 23–34. Li, W., Airriess, C.A., Chen, A., Leong, K.J., Keith, V. 2010. Katrina and migration: evacuation and return by African Americans and Vietnamese Americans in an Eastern New Orleans suburb. Professional Geographer 61 (1), 103–118. Lindell, M.K., R.W. Perry. 1997. Hazardous materials releases in the Northridge earthquake. Risk Analysis 17, 147–156. Lindell, M.K., Prater, C.S. 2008. Behavioral Analysis: Texas Hurricane Evacuation Study. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S., Perry, R.W. 2006. Fundamentals of Emergency Management. Emmitsburg MD: Federal Emergency Management Agency Emergency Management Institute. www.training.fema.gov/EMIWeb/edu/ fem.asp or hrrc.arch.tamu.edu/publications/books/.

Chapter 10 · Evacuation Termination and Reentry

279

Maghelal, P., Peacock, W.G., Li, X. 2016. Evacuating together or separately: factors influencing split evacuations prior to Hurricane Rita. Natural Hazards Review DOI: 10.1061/(ASCE)NH.1527-6996.0000226. McEntire, D.A., Cope, J. 2004. Damage Assessment After the Paso Robles (San Simeon, California) Earthquake: Lessons for Emergency Management. Quick Response Report 166. University of Colorado Natural Hazards Center, Boulder, CO. Myers, C.A., Slack, T., Singelmann, J. 2008. Social vulnerability and migration in the wake of disaster: the case of Hurricanes Katrina and Rita. Population and Environment 29 (6), 271–291. Oliver-Smith, A. 2005. Disasters and forced migration in the 21st century. In: Understanding Katrina: Perspectives from the Social Sciences. understandingkatrina. ssrc. org. Paxson, C., and C.E. Rouse. 2008. Returning to New Orleans after Hurricane Katrina. American Economic Review 98 (2), 38–42. Peacock, W.G., Van Zandt, S., Zhang, Y., Highfield, W.E. 2014. Inequities in longterm housing recovery after disasters. Journal of the American Planning Association 80 (4), 356–371. Perry, R.W., Lindell, M.K. 2007. Emergency Planning. John Wiley and Sons, Hoboken, NJ. Quarantelli, E.L. 1982. Sheltering and Housing After Major Community Disasters. Ohio State University Disaster Research Center, Columbus, OH. Salzer, M.S., Bickman, L. 1999. The short- and long-term psychological impact of disasters: Implications for mental health interventions and policy. In: Gist, R., Lubin, B. (Eds.), Response to Disaster: Psychosocial, Community, and Ecological Approaches. Brunner Mazel, Philadelphia, pp. 63–82. Schwab, J., Topping, K.C., Eadie, C.C., Deyle, R.E., Smith, R.A. 1998. Planning for Post-Disaster Recovery and Reconstruction. American Planning Association, Chicago, IL. Shaw, W.D., Baker, J. 2010. Models of location choice and willingness to pay to avoid hurricane risks for Hurricane Katrina evacuees. International Journal of Mass Emergencies and Disasters 28 (1), 87–114. Showalter, P.S., Myers, M.F. 1994. Natural disasters in the United States as release agents of oil, chemicals, or radiological materials between 1980– 1990: analysis and recommendations. Risk Analysis 14, 169–182. Siebeneck, L.K., Cova, T.J. 2008. An assessment of the return entry process for Hurricane Rita 2005. International Journal of Mass Emergencies and Disasters 26 (2), 91–111. Siebeneck, L.K., Cova, T.J. 2012. Spatial and temporal variation in evacuee risk perception throughout the evacuation and return-entry process. Risk Analysis 32 (9), 1468–1480. Siebeneck, L.K., Lindell, M.K., Prater, C.S., Wu, H-C., Huang, S-K. 2013. Evacuees’ reentry concerns and experiences in the aftermath of Hurricane Ike. Natural Hazards 65 (3), 2267–2286. Smith, S.K., McCarty, C. 1996. Demographic effects of natural disasters: a case study of Hurricane Andrew. Demography 33 (2), 265–275.

280 Chapter 10 · Evacuation Termination and Reentry Spennemann, D.H.R., Look, D.W. 1998. Disaster Management Programs For Historic Sites. Association for Preservation Technology, San Francisco CA. Stallings, R.A. 1991. Ending evacuations. International Journal of Mass Emergencies and Disasters 9 (2), 183–200. Tierney, K., Bevc, C., Kuligowski, E. 2006. Metaphors matter: disaster myths, media frames, and their consequences in Hurricane Katrina. American Academy of Political and Social Science 604 (1), 57–81. Wolshon, B. 2009b. The role of transportation in evacuation and reentry: a survey of practice. Journal of Transportation Safety & Security 1 (3), 224–240.

Chapter 11

Case Studies

The preceding chapters have explained the general principles for creating, analyzing, and implementing evacuation plans, but this is only the first step toward planning and preparing more effectively for future incidents. In addition, it is important to see how those principles can be applied. Thus, this chapter provides brief summaries of four past evacuation planning experiences and their outcomes. Although not all aspects of evacuation planning are included within each case study, these examples illustrate key elements of these efforts including: ■ the stages of the evacuation planning process, beginning with evacuation zone definition and ending with post-event evacuee reentry; ■ general characteristics and threat conditions of hazards for which evacuation plans are used; ■ relevant preparedness characteristics of the locality that influenced the evacuation process; ■ the state of local agencies’ emergency planning and training; and ■ key components of the emergency response processes. More specifically, the case studies focus on the key transportation and traffic management aspects of the evacuation process, highlighting key aspects of: ■ planning and preparedness; ■ evacuation direction and control; ■ inter-agency communication, risk area warning, and public information; ■ assisted evacuation; and ■ post-event reentry. To illustrate the breadth and depth of evacuation management, the first three case studies focus on larger scale events that also provide longer periods of forewarning. The fourth case study addresses the

282 Chapter 11 · Case Studies more common small scale, no-notice events, which tend to occur in unpredictable locations with little warning and, as such, require a significant amount of improvisation. Nonetheless, potential incident locations can be identified by means of techniques such as commodity flow studies (Bierling et al. 2011, USEPA 1993). The first two case studies examine actual evacuations, the first of which is about the largest and most commonly discussed (at least in the United States) environmental hazard—hurricanes. The second case study examines a wildland-urban interface fire. The third and fourth case studies are somewhat different in focusing on planning for future evacuations rather than assessing past evacuations. These four case studies describe prior work by the authors and their coauthors. These works also include excerpts from reports prepared by the Sandia National Laboratory for the United States Nuclear Regulatory Commission (Jones et al. 2008) and the National Cooperative Highway Research Program (NCHRP) study (Wolshon 2008) as well as a study conducted by the Parsons Inc. for a regional transportation agency (Parsons Inc. 2016). Readers interested in more detailed descriptions and findings of these works, as well as other related information, are encouraged to review those resources.

11.1 Case Study 1—Hurricane Katrina In many ways, Hurricane Katrina was a watershed event for emergency managers and transportation officials in the United States. When it occurred, it was by far the most costly natural disaster in the history of the country and remains among the costliest in terms of loss of life. From a transportation perspective, it also demonstrated many of the best and worst aspects of the state of mass evacuation planning and operation at the time. In the years since, it has, in many ways, come to be viewed as the starting point for several now-common evacuation engineering, planning, and management methods.

11.1.1 Hazard Conditions Hurricane Katrina made landfall on the Gulf Coast near Buras, Louisiana on August 29, 2005 as a Category 3 hurricane (Knabb, Rhome, and Brown 2005). At one point, the storm was approximately 400 miles across with sustained winds exceeding 170 miles per hour, which prompted the evacuation of approximately 2 million people along the Gulf Coast from Louisiana to Florida. The highway-based portion of the evacuation was generally considered a success as an estimated 1.3 million people moved away from coastal southeast Louisiana and the

Chapter 11 · Case Studies

283

New Orleans metropolitan region in the 48 hours that preceded Katrina’s arrival (Wolshon and McArdle 2009). Katrina was extremely unusual, in that the major hazard threat did not occur until after the storm when sections of the levee system failed and flooded most of the city. This produced a post-impact evacuation in which people who did not evacuate before the storm had to be evacuated after it passed.

11.1.2 Preparedness and Planning As Katrina revealed, evacuation planning and preparedness in south Louisiana was, at the same time, both extensive and greatly lacking. Like most areas at that time, officials had undertaken significant efforts to develop plans for auto-based self-evacuators, but had few if any significant plans for people with low or no access to reliable personal vehicles. One of the most significant factors that helped state and local transportation officials in the area was the fact that they had conducted an exercise of their EOP about a year earlier. Moreover, the shortcomings identified in the Hurricane Ivan evacuation in September 2004 gave officials the opportunity to integrate lessons learned from that event and revise the State’s contraflow plan, which was a key factor in the successful evacuation of New Orleans and areas to its south in Katrina. The 2005 New Orleans emergency management plan, which was finalized only weeks before Katrina, provided for the use of all available resources to evacuate threatened areas, but made only limited arrangements to evacuate persons unable to transport themselves. Although estimates suggested the need to evacuate upwards of 100,000 residents who lacked personal transportation, city planners anticipated that many of these people would receive rides with friends and family members and that hospital patients and routine users of senior centers would have transportation arranged by these facilities. As it turned out, the lack of reliable personal vehicles tended to be systematically concentrated within specific neighborhoods rather than randomly distributed throughout the city. Consequently, the friends and relatives of those who lacked reliable personal vehicles also tended to lack reliable personal vehicles. Thus, the inadequate public transportation support by buses tended to strand families in those neighborhoods. Outside of Louisiana, communities in coastal areas of Alabama and Mississippi also implemented their emergency plans during Katrina. Many of these had also been developed or updated based on lessons learned from recent hurricanes. In Alabama, officials had developed and exercised the State’s contraflow plan to reduce the time needed for implementation and used new proactive communications strategies to facilitate the response (Jones et al. 2008). Because of the close population and transportation linkages between Gulf States, local officials often coordinated within their states and

284 Chapter 11 · Case Studies across political boundaries into neighboring states. It is interesting to note that even though many parishes in Louisiana and counties in Mississippi and Alabama issued mandatory evacuation orders, New Orleans was initially under a “voluntary” evacuation notice. Although a mandatory evacuation was ordered later, this delay was thought by some to have contributed to the city’s fatalities (Jones et al. 2008).

11.1.3 Evacuation Direction and Control The evacuation plan used for Katrina included a phased evacuation and a contraflow plan revised to improve shortcomings identified after the previous year’s evacuation for Hurricane Ivan. Using this updated plan, contraflow controls for Katrina were prepared and implemented in less time than expected, thus facilitating the movement of more than 90% of the population of southeast Louisiana. Based on traffic count data collected from routes in the vicinity of New Orleans, traffic flow had dropped to a trickle about 8 hours prior to storm landfall, suggesting that everyone with the means and desire to evacuate had done so (Wolshon and McArdle 2009). Nevertheless, approximately 70,000 people remained in the city (Jones et al. 2008). Late in the evacuation process, when the New Orleans mayor issued a mandatory order about a day prior to storm landfall, the Regional Transit Authority transported people from pickup points around the city to the Superdome and took special needs and other persons requiring transportation assistance to Baton Rouge. Evacuations in Mississippi were generally phased, with lower-lying areas, mobile home communities, and residences along waterways encouraged to evacuate prior to those in less vulnerable areas.

11.1.4 Public Information and Warnings Because of the long-standing and well recognized threat of hurricanes, communities throughout the Gulf Coast use a variety of methods to inform residents about the hurricane threat and household preparedness at the start of each hurricane season. These include television and radio broadcasts, newspaper articles, and the Internet. At the local level, direct mailings using brochures and literature in utility bills and telephone directories have been common (Jones et al. 2008). As a large and slow moving storm that had crossed over south Florida the previous week, Hurricane Katrina received considerable media attention as it moved toward the Louisiana Gulf Coast. As is the case with most major hurricanes, local and national broadcast media began to increase their coverage of the storm many days before

Chapter 11 · Case Studies

285

landfall. As the storm drew closer, some communities implemented route alert warnings by driving streets in high risk areas using loud speakers to notify residents of mandatory evacuation orders. Others had local law enforcement personnel go door to door to hand out pamphlets instructing residents on the order in which they were to evacuate during phased evacuations. Residents in mobile homes, along inland waterways, and in low-lying areas were especially encouraged to evacuate early (Jones et al. 2008). DMSs were used to inform drivers on key highways about current storm and traffic conditions. Given that New Orleans, and the Gulf Coast more broadly, is a major tourist destination for visitors from around the world and that the area has many local recent immigrants, emergency information was also issued in Spanish and Vietnamese. In some areas of Florida, written emergency information was also distributed in Spanish, French, and German (Jones et al. 2008).

11.1.5 Assisted Evacuation An element of the Hurricane Katrina response that received considerable media coverage was the evacuation of low-mobility and special needs populations. Prior to Katrina and, arguably today still, the extent of the public responsibility to provide transportation resources to these groups remains ill-defined. Although there have been enormous changes in how provisions for evacuation assistance is viewed, especially in New Orleans, Houston, and other locations where major catastrophic disasters have occurred, it is unclear how it would be accomplished in other locations. Even in New Orleans, practical and cost considerations have made it impossible to maintain pre-arranged contracts for passenger rail-based evacuation services. It is likely that all cities will need to rely heavily on voluntary assistance. As it was long recognized that there were no formalized and contractual arrangements for the movement of many low-mobility and special needs populations despite the acknowledged need, residents with room in their vehicles were encouraged to extend help to those in need. To some extent, this call for informal assistance was successful because some people without their own vehicles did evacuate with someone else (Wu, Lindell, and Prater 2012). Some of these rides might have been offered spontaneously, as has been the case in evacuations from Lili (Lindell, Kang, and Prater 2011), Rita (Wu et al. 2012), and Ike (Wu, Lindell and Prater,2013). Other rides might have been coordinated by religious and other community groups such as “Operation Brother’s Keeper.” Although the precise extent to which these different sources were successful in providing transportation assistance is unknown, it was clearly insufficient to accommodate everyone who needed help (Wolshon 2008).

286 Chapter 11 · Case Studies Jones et al. (2008) reported that police and fire department personnel were sent around New Orleans urging people to go to checkpoints where approximately 20 buses circulating the city would pick them up to take them to the Superdome. Unfortunately, pickup locations were not announced in advance so residents, whether or not they had mobility limitations, found it difficult to find them. In Gulfport, Mississippi, school buses were used to evacuate special needs populations to area shelters and ambulances were called upon to transport those who were nonambulatory. However, there were no address lists to direct drivers to the homes of those with special needs. In most cases, these individuals were able to make arrangements with friends or and family (and in some cases call 911) for transportation to shelters. In an effort to provide support to resolve these issues, voluntary and civic groups have been engaged. One such group in New Orleans is Evacuteer, which is a nonprofit organization that annually recruits, trains, and manages 500 evacuation volunteers (Evacuteers) to assist in the city’s assisted evacuation plan. Formed after Katrina, their goal is to support the estimated 35,000 New Orleanians who lack the ability or a safe destination to evacuate. Evacuteers will work at the 17 neighborhood pickup points, the Union Passenger Terminal, and City Hall to assist with the 3-1-1 hotline. The pickup locations, known as Evacuspots are marked by 14-foot tall stainless steel statues, shown in Figure 11.1, designed by local artist and funded by FEMA, the Arts Council of New Orleans, and the New Orleans Office of Homeland Security and Emergency Preparedness (Evacuteer 2016). Surveys suggested that about 70% of elder care facilities did not evacuate from Hurricane Katrina. In most cases, nursing home managers prefer to shelter in-place during a hurricane, as it recognized that evacuating is stressful to patients and can lead to health complications. The ambulances and specialized buses needed to evacuate hospitals and nursing homes are also very costly and such costs are not reimbursed if a hurricane does not strike that jurisdiction. As a result, most decisions to evacuate care facilities are made late, even though such decisions are better made early (Jones et al. 2008). Similar reviews also stated that contracted buses were not always available and alternative vehicles sometimes lacked air conditioning or broke down along the way. Trips often took longer than expected; food and water sometimes needed to be rationed; and medicine, oxygen tanks and incontinence supplies were often left behind. As a result of poor planning, bad decisions, and unfortunate circumstances, over 200 nursing home patients died because of Hurricane Katrina (Jones et al. 2008). Mobility limited populations are not limited to the elderly, infirm, and indigent. A range of other persons, including those in local and state correctional facilities, also need to be evacuated. The Louisiana Department of Corrections stated that even though the evacuation presented a “logistical challenge,” it was nevertheless safe and efficient

Chapter 11 · Case Studies

287

Figure 11.1 Evacuspot Sculpture Near the French Quarter

From Cherney, 2013

as over 6,000 inmates were evacuated for Hurricane Katrina (Jones et al. 2008). Despite the many lessons learned in Katrina and the clear need to address the issue of low-mobility evacuations, it is apparent that addressing these issues remains a work in progress throughout the US. Although there are many plans of varying breadth and depth, it is difficult to conclude that any of them could be implemented successfully. Both before and after Katrina some public officials asserted that the decision to evacuate is ultimately a “personal choice.” And to the extent that some people have limited, or no, ability to move themselves, they will necessarily rely partially—and, in some cases, completely—on others. It is clear that in the US, the greater the dependence upon others to evacuate, the greater risk of becoming a victim. Although Katrina exposed this reality and efforts have been made to improve the situation, provisions for low mobility evacuation assistance remains inadequate in the US.

11.2 Case Study 2—2007 Southern California Wildfires In the autumn of 2007, the Southern California Wildfires burned areas in counties from Los Angeles and San Bernardino in the north to San

288 Chapter 11 · Case Studies Diego and Imperial on the Mexican border. These fires burned over a half million acres and more than 3,000 buildings, and caused 139 injuries and 12 deaths. The total number of people evacuated for these fires, nearly a million, was the largest in California history (Jones et al. 2008).

11.2.1 Preparedness and Planning Wildfires present distinct challenges to emergency officials making evacuation decisions. Although many coastal communities have detailed plans with specific timelines for initiating hurricane evacuations, wildfires can be even more challenging for pre-incident evacuation planning because it is not possible to predict their point of origin or direction and speed of travel (Wolshon 2008). Wildfires generally advance at a rate of 1–5 mph when the wind is calm, but have the potential to move much faster. Since wildfire spread is often based on the movement of windborne embers, the movement of a fire is largely a function of wind speed and direction. A hot dry Santa Ana wind of 40 mph or more can produce fire spreading speeds of 60–70 mph and perhaps more. This high rate of speed is achieved when embers from the flame tops are blown to new locations where new flames are ignited. Because a wildfire’s characteristics (e.g., point of origin, direction and speed of travel) cannot be predicted in advance, local authorities must improvise their evacuation plans using basic templates. In addition to defining the evacuation zones and routes, emergency managers must establish trigger points, which are critical locations that—when the fire front reaches them—signal that it is time to begin the evacuation (Cova et al. 2017). The decisions whether and when to evacuate during the 2007 California wildfires consistently came from fire departments, which provided Incident Commanders (ICs) at the local level (Wolshon 2008). Later discussions with fire officials revealed that the fire departments developed “evacuation boxes”, which were geographical areas that were defined by readily recognizable physical boundaries such as highways and waterways, and relayed that information to law enforcement and the city/county EOCs. Discussions with local officials revealed that, although fire department officials led the effort to decide where and when to evacuate, their primary job was to fight the fires rather than to evacuate people. Thus, once an IC decided to recommend evacuation, law enforcement and transportation officials were responsible for implementing the evacuation. The evacuation process was led by local law enforcement agencies but were supported by local DOTs and Departments of Public Works (DPWs), which provided traffic barricades and roadway DMSs to close roads based on the guidance of law enforcement officials (Jones et al. 2008).

Chapter 11 · Case Studies

289

Because of the large scope of the fire threat, state agencies supported the local agencies in evacuation planning and implementation. The California Department of Transportation (CalTrans) had representatives in the local EOCs in addition to establishing their District Command Centers, which included key management and staff. CalTrans assisted with the coordination of emergency response, evacuations, and route closures, in conjunction with the California Highway Patrol. CalTrans also mobilized maintenance and construction crews to assist in traffic control and field damage assessments (Wolshon 2008). As is typical of most large-scale evacuations, the exact number of residents who evacuated, when they left, and where they went was not known for certain. However, it is generally accepted that over 900,000 people evacuated as a result of the wildfires, with some sources saying the number that actually evacuated approached one million people. Regardless of the total number of people who evacuated, it is accepted that this was the largest evacuation in California history. News media reports and personal interviews indicated that both mandatory and voluntary evacuation orders were issued during the event. The type of evacuation notice and when it was issued depended on the fire’s direction and speed. Although first responders in San Diego County recognized that a mandatory evacuation order did not give them the authority to force citizens from their homes, they did believe that child endangerment laws gave them the authority to forcibly remove children from a house. They reported that if they threatened the parents this way, the families decided to evacuate “99.9 percent of the time” (Wolshon 2008). Because transportation officials could not predict evacuation zones far enough in advance, they found it difficult to implement traffic management techniques such as contraflow or priority signalization. Nonetheless, some of these traffic management actions were used (as described in the following section), although they were viewed somewhat negatively because they often required additional manpower that was in scarce supply. Contraflow operations were seriously considered for areas north of San Diego but, ultimately, were never used (Jones et al. 2008). One method of traffic control utilized by the Mayor of San Diego was to ask people in unaffected areas to stay off the roads to free capacity for emergency response vehicles and evacuation traffic. Although the effectiveness of this attempt to reduce background traffic was not measured, it demonstrates a proactive message and effective utilization of the media to convey to citizens how they can facilitate emergency actions.

11.2.2 Evacuation Direction and Control There were 15 major highways that were closed at various times during the wildfires, but these closures did not appear to have affected the

290 Chapter 11 · Case Studies evacuations. Most notably, all three of the most heavily traveled highways—I-5, I-8 and I-15—were closed at some time (Wolshon 2008). To compensate for these closures, local officials worked with their federal counterparts at the Camp Pendleton Marine Corps Base to permit public use of on-base roads so evacuation traffic could access northbound I-5 when I-15 was closed. One of the ways in which CalTrans assisted with road closures was through the release of the CALTRANS COMMUTER ALERT, which provided locations and details about road closures throughout the affected areas. These road closures were also illustrated through GISs by providing detailed maps that depicted the perimeters of the wildfires and the resulting road closures. Both San Diego County and CalTrans provided mapping services to assist responders and the citizenry during this period (Wolshon 2008).

11.2.3 Assisted Evacuation The 2007 California wildfires also required assisted evacuations for care facilities and limited mobility evacuees. Jones et al. (2008) found that 14 nursing homes with about 1,200 residents and 85 assisted-living facilities with about 2,200 elderly residents were evacuated in San Diego County. The fires also required the short-term closure of two hospitals and a psychiatric hospital. Six fatalities resulted from the evacuation of the nursing home and elder care facility residents. Transit buses were also employed to transport elderly carless evacuees. The transit-assisted evacuation included an additional 1,000 seniors in San Diego County. Due to San Diego’s location near the Mexican border, the county is home to a large migrant worker population. During the 2007 wildfires, there were several challenges to meeting the evacuation needs of this diverse group, including warning recipients’ limited: ■ English-speaking proficiency which may have produced misunderstanding of evacuation and shelter in-place orders; ■ trust of public officials because of possible illegal immigration status or prior negative encounters with law enforcement and immigration agencies; and ■ financial resources to cover non-working periods. Although no one was killed or injured as a result of failing to evacuate because of language barriers, city officials in San Diego cited a “chronic lack of translators, which hindered the ability to evacuate and/or provide other emergency services” in the city’s After Action Report (City of San Diego 2007).

Chapter 11 · Case Studies

291

11.2.4 Communication and Public Information Similar to Hurricane Katrina and other events worldwide, there were significant issues with interoperable communications equipment among the first responder agencies during the 2003 California wildfires. In the following four years, agencies within the region worked aggressively to address this shortfall and first responder communications were handled quite effectively both within and between responding agencies. Among the deficiencies that remained, however, were a shortage of 800 MHz radios among firefighting crews, a lack of tactical channels for unit to unit communication, and an overcrowding and delay of information exchange on the available channels (Jones et al. 2008). San Diego County emergency management agencies also benefited from web-based emergency management communication tools, which allowed agency representatives to process resource requests through a single system and create situation reports. These sitreps provided users with instantaneous and comprehensive situational awareness, as well as an ability to respond to resource requests throughout the course of the incident (Jones et al. 2008). Jones et al. (2008) reported that San Diego officials utilized a variety of communication assets to disseminate warnings and provide public information, including: ■ Door to door warnings by first responders; ■ Police and fire/rescue vehicle sirens; ■ Police and fire/rescue vehicle and helicopter flights; ■ Constant monitoring and information flow to news media outlets for dissemination to the public; ■ Emergency Alert System warnings disseminated via television; ■ AlertSanDiego mass notifications; ■ Community Access Phone System; and ■ 2-1-1 Information Line. The AlertSanDiego system was implemented after the 2003 wildfires to improve direct alerts to the public. This call system, which was populated with listed and unlisted phone numbers provided by the county’s 911 database, was able to operate at a rate of 12,000 calls per hour. The system, with its ability to permit fire departments to identify “evacuation boxes”, was able to restrict calls to persons located within a designated evacuation zone rather than notifying an entire zip code or area code. Although only designed to provide alerts through landlines, people can register mobile devices and sign up for text messaging (Jones et al. 2008).

292 Chapter 11 · Case Studies Another improvement over the 2003 fires was the ability for local officials to use the 2-1-1 call system to relay non-emergency information to the public. More than 120,000 calls were made to San Diego County phones to provide current information about evacuations, shelters, road closures, volunteer and recovery information and services (Jones et al. 2008).

11.2.5 Reentry San Diego County, like many others, lacked a formalized plan for reentry following wildfire containment. Officials contended that it is practically impossible to control reentry to evacuation zones in the San Diego area. Nonetheless, although there were no formal plans for reentry, authorities followed guidelines for allowing reentry into certain areas. As noted in Chapter 10, the primary concern for reentry is public safety. Utility companies focused on ensuring utilities were secured, but this did not mean that those utilities had been restored (Wolshon 2008). In addition, CalTrans damage assessment teams certified the safety of state and federal roadways. The CalTrans damage assessment teams addressed immediate safety needs for reopening route segments, with their immediate priorities being slope stabilization, erosion control, guardrail installation, signage replacement, culverts/drainage repair, and electric power restoration for call boxes, and lighting. CalTrans efforts were able to re-open all route segments within two weeks. Although a controlled reentry was not possible for all of San Diego County, there were examples of isolated neighborhoods in which reentry was controlled. In these instances, a reentry assistance center was set up at the entry point to that area. This center included many different services to help people in the affected area restart their lives. In addition to managing access, the reentry assistance center was also meant to provide security against looters, safety hazards within the area, and unscrupulous contractors (Wolshon 2008). Before allowing reentry into these areas, the Fire Department conducted assessments to check for natural gas, electrical, and other hazards. Once individuals obtained the necessary credentials, they were required to check in and were granted access only during daylight hours. This process was repeated daily until authorities allowed unrestricted reentry. San Diego County officials also increased the usefulness of the reentry assistance centers by co-locating grief counselors to assist those who were experiencing difficult emotional issues as a result of the wildfires (Wolshon 2008).

Chapter 11 · Case Studies

293

11.3 Case Study 3—Nuclear Power Plant Traffic Management Study Evacuation planning associated with nuclear power plant (NPP) emergencies is somewhat different from planning evacuations for other hazards. One of those most notable areas is its study. As the power industry in the US is regulated by the United States Nuclear Regulatory Commission (NRC), formalized evacuation time estimate (ETE) studies are required for all licensed reactor sites. Because these studies involve traffic- and behaviorally-based study based on well researched and documented criteria, they tend to be better assessed than evacuations for other hazards, certainly more so than natural hazards. Another notable aspect of NPP emergency planning is that evacuation plans are coordinated with relevant local stakeholders and extensive efforts are undertaken to communicate them to potential evacuees within the 10mile EPZ. Although some or most of these activities also occur for other hazards, they are not typically undertaken at the same level of detail and formality as for NPPs. Although this level of planning and preparedness is obviously a function of the threat posed by a radiological release, it was also motivated by the highly publicized 1979 Three Mile Island emergency. Perhaps most important however, the extensive level of preparedness for NPPs also reflects the emergency preparedness regulations of the NRC and Federal Emergency Management Agency (FEMA). Although these agencies do not plan or order evacuations themselves, they require plant operators to follow established onsite emergency planning procedures that are documented in their regulatory guidance (USNRC/FEMA 1980). FEMA supports off-site emergency response activities and protective action decision making for activities such as evacuations made by off site response organizations. NRC regulatory guidance also requires each nuclear power plant licensee to submit ETEs that describe evacuation analyses for a variety of conditions. These include different meteorological conditions (wind direction and speed, precipitation); background traffic; shadow evacuation response; and EPZ population occupancy and distribution scenarios for different times of day, days of the week, and seasonal variation. The NRC originally published guidance for ETE studies in 1980 (Urbanik et al. 1980) and recently updated it (Jones, Walton, and Wolshon 2011). This case study discusses a traffic management analysis that was undertaken based on the findings of an ETE study (Parsons Inc. 2016). The following sections describe the traffic management approaches that were proposed, the assumptions and analytic techniques that were used in a simulation model, and the traffic conditions that were expected. A unique aspect of this study was the scale and level of detail at which it was undertaken. Historically, evacuation planning, particularly for

294 Chapter 11 · Case Studies mass evacuations, has typically taken place at high levels of analysis in which large areas were analyzed using coarse modeling. Analysis at this level occurs because of the substantial time, effort, and expense required to characterize all of the nodes and links in a road network, as well as to run simulations for the many different scenarios that might occur. Thus, analysts typically are forced to assume that a large scale, macrolevel analysis is sufficient to inform broad decision making and develop appropriate evacuation traffic management plans. In contrast to these practices, the goal of the agencies who funded the study and the efforts of the consultants who conducted the analyses were to understand the detailed, location-specific impacts of specific control actions. Consequently, the analysis was conducted at the level of individual signal controlled intersections, turning movement restrictions at specific locations, and ramp closures, in addition to numerous other traffic management actions within a region-wide evacuation. Because the work described in this case also involved assessments of some proprietary and security-related information, it is not possible to discuss all details related to specific locations or agencies involved in the study. Moreover, the following discussion changes some details in order to maintain the confidentiality of sensitive information. However, none of the information that has been withheld or altered affects the key points of this case study—the primary concepts that guided this work and the lessons learned from it, particularly as they apply to evacuations at other NPPs and, more broadly, other large area hazards.

11.3.1 Background, Goal, and Intent of Analyses The conditions discussed in this case study are notable because of the NPP’s proximity to a major metropolitan center. Although most NPPs are located in areas with low population densities, there are some exceptions. This is expected to become an increasing challenge in the future as suburbs expand into previously rural areas. This case site is also interesting, though not exceptional, because the EPZ is traversed by at least one heavily traveled freeway. Such heavily used transportation routes create the potential for substantial background traffic that can increase volumes well above the levels generated by resident evacuee populations. Heavily traveled routes can also present impediments to travel because they may need to be crossed by evacuating traffic.

11.3.2 Demand Generation Similar to other emergencies, the first step of this traffic analysis was to forecast and generate the expected vehicle volumes during a specific

Chapter 11 · Case Studies

295

evacuation. In this case, the primary sources of traffic during the evacuation study were assumed to include: ■ Risk area evacuees, ■ Shadow evacuees, and ■ Background traffic. Next, it was necessary to determine how the traffic was spatially distributed and how it would temporally load into the network. For NPPs, traffic demand is typically generated for various sectors of the EPZ and in the shadow areas immediately outside of the EPZ. As noted in Chapter 2, the EPZ has an approximately 10-mile radius surrounding the plant that is subdivided into three zones (0–2 miles, 2–5 miles, and 5–10 miles from the plant) and 16 compass sectors, 22.5° each in width (N, NNE, NE, etc.). Planners used these zones/sectors (e.g., 0–2 miles, N 2–5 miles, N 5–10 miles) to define Emergency Response Planning Areas (ERPAs), which are clarified by using readily identifiable borders based on local geographic features and/or political boundaries to enhance people’s comprehension of emergency warnings and public information. The evacuation and shadow demand was based on population and vehicle ownership rates from within the EPZ at three distances from each of the two units at the site. For each evacuation zone, it was assumed that 100% of persons in each area and 30% in the shadow zone would evacuate, as shown in Table 11.1. The areas in this table are shown here in approximate terms. The apparent overlap of the areas shown in the table are due to various temporal and spatial considerations that took place during the traffic generation process for the analysis. Assumptions governing the entry of evacuating vehicles into the ERS followed commonly accepted temporal response patterns of other major evacuations as described in Chapter 5. In the nuclear power plant analysis, however, the response timeline was compressed to reflect the

Table 11. 1 Approximate Evacuation and Shadow Boundary Evacuation Extents Area of Evacuation

Radius of Shadow Boundary

2 miles

5 miles

5 miles

10 miles

10 miles

15 miles

Adapted from Parsons Inc. 2016

296 Chapter 11 · Case Studies assumption that all evacuees would enter the network within three hours of an order to do so. In NPP ETE analyses, it is generally recognized that each population group (the general public, transit dependent residents, special facility residents, and schools) will differ based on different considerations for trip generation times. Thus, surveys of residents are carried out to develop departure time distributions. Another significant consideration in the analysis was generation and movement of background traffic, which can vary significantly, based on time of day, day of week, and season. Because background traffic can often be a primary factor that produces congestion, and thus affects clearance time, the analysts believed that efforts should be made to limit background traffic entry into the EPZ. This could be accomplished by encouraging drivers to cancel nonessential trips in the area and closing inbound access into the EPZ. Based on these assumptions, 12 base cases of time of day and day of week demand generation scenarios with varying distances were developed as shown in the schematic summary of Figure 11.2. These scenarios reflect a variety of times at which traffic volume on the area road network could be at its highest and its predominant direction of travel oriented with or against the general direction of evacuation flow.

Figure 11. 2 Time of Day Scenario Tree 5mi

Night

10mi 20mi 5mi

Morning

10mi 20mi

Study 5mi

Mid-day

10mi 20mi 5mi

Afternoon

10mi 20mi

Adapted from Parsons Inc. 2016

Chapter 11 · Case Studies

297

The analysts then adjusted these combinations of conditions by replicating the analysis for each of the two units and then adding or deleting conditions such as planned improvements to the ERS. This yielded 36 scenarios covering a variety of area conditions (see Tables 11.2a and 11.2b). Daytime and weekday scenarios represented conditions during an average workday. The evening scenarios represented periods when residents are generally at home with lower background traffic on the roads. It is possible that a radiological release could take place during a special event—such as a concert or sporting event—that could attract a large number of transients to the threat area for a short period of time. Typical evacuation analysis guidance suggests that such events should be assessed within the expected estimated population, duration, and season of the event. However, none of these conditions was assumed for the purposes of this study.

Table 11.2a Nuclear Power Plant Evacuation Scenarios for Unit 1 Scenario

Time of Day

Area of Evacuation(*)

1

2 miles

2

5 miles

3

5 miles w/1 yr. planned improvement

Night Time

5 miles w/5 yr. planned improvement

Case Study Site 1

4 5

10 miles

6

2 miles

7

5 miles

8 Morning

5 miles Winter conditions

9

5 miles w/1 yr. planned improvement

10

5 miles w/5 yr. planned improvement

11

Daytime

Afternoon

2 miles

12

2 miles

13

5 miles

14

5 miles winter conditions

Evening 15

5 miles w/1 yr. planned improvement

16

5 miles w/5 yr. planned improvement

* Distances are approximated From Parsons Inc. 2016

298 Chapter 11 · Case Studies

Table 11.2b Nuclear Power Plant Evacuation Scenarios for Unit 2 Scenario Time of Day

Area of Evacuation(*)

17

2 miles

18

5 miles

19

5 miles w/1 yr. planned improvement

Night Time

5 miles w/5 yr. planned improvement

20 21

10 miles

22

2 miles

23

5 miles

24

10 miles Winter conditions

Case Study Site 2

Morning 25

5 miles w/1 yr. planned improvement

26

5 miles w/5 yr. planned improvement

27

2 miles

28

5 miles

29 30

10 miles Daytime

Afternoon

5 miles w/1 yr. planned improvement

31

5 miles w/5 yr. planned improvement

32

2 miles

33

5 miles

34

10 miles winter conditions

Evening 35

5 miles w/1 yr. planned improvement

36

5 miles w/5 yr. planned improvement

* Distances are approximated From Parsons Inc. 2016

Background and pass through traffic make relatively small contributions to the total demand estimate in most cases because they are relatively small compared to the resident and transient population. However, this EPZ is traversed by a busy freeway, so the level of background traffic can be substantially higher. In this case study, background traffic represented a significant percentage of the overall volume in the network and greatly influenced the onset and movement of congestion downstream of the two NPP units. NRC guidance also addresses shadow evacuations, assuming that 20% of the permanent resident population will shadow evacuate from

Chapter 11 · Case Studies

299

areas five miles beyond the evacuation zone being assessed. In this case, the analysts made a more conservative assumption that there would be a 30% shadow evacuation.

11.3.3 Evacuation Traffic Management Strategies The focus of the analyses in this case study was the evaluation of evacuation traffic management strategies. Recent professional experience and simulation research has shown that short term evacuation traffic management strategies, such as those described in Chapter 8, can significantly facilitate overall traffic movement during large scale evacuations. In this case study, the analysts tested separate evacuation traffic management techniques for freeways and arterials because of the differences between the two types of roadways. Traffic flow on freeways, in contrast to that of arterial streets, is uninterrupted by traffic signals and stop signs; thus, inflow to, and outflow from, the route can only occur at on- and off-ramps. Evacuation management strategies on freeways near and within the study region favored flow along freeways away from the EPZ while also preventing traffic from entering the freeways. This was accomplished by closing freeway access for background traffic into the EPZ and strategically closing other on-ramps to prohibit background traffic from combining with the outbound flow of evacuation traffic. Similarly, off-ramp closures outside of the EPZ were used to prevent queuing spillback onto the freeway mainline from downstream arterial congestion. Arterial management measures involved techniques such as turn restrictions, traffic signal timing modifications, and police MTC to facilitate outbound flow of evacuees onto area freeways and surrounding arterials by also restricting background traffic access into the EPZ.

11.3.4 Analyses and Interpretation of the Study Results Tables 11.3 and 11.4 list the primary input assumptions for each scenario in the leftmost columns (Time of Day through Background Demand Reduction) and the resulting Clearance Time for each scenario in the rightmost column. Overall, these results demonstrate the impact time of day (and its associated levels of background traffic) had on clearance time. Evacuations initiated during nighttime hours had lower clearance times than any other time of day. In many cases, the nighttime clearance times were about half as long as any other time period. Conversely, the mid-day scenarios (with the highest background traffic volumes) showed the longest clearance times. Another significant

Day Time

Morning

Night Time

Time of Day

114,000

5 miles

132,000

132,000

5 miles

5 miles winter conditions

5 miles w/5 yr. 147,000 planned improvement

5 miles w/1 yr. 133,000 planned improvement

81,000

95,000 7.30 pm

46,000 193,000 7.30 pm

43,000 176,000 7.30 pm

43,000 175,000 7.30 pm

43,000 175,000 7.30 pm

14,000

9.00 pm

9.00 pm

9.00 pm

9.00 pm

9.00 pm

11.00 pm

11.00 pm

270,000 180,000 450,000 9.30 pm

46,000 173,000 9.30 pm

11.00 pm

11.00 pm

11.00 pm

2 miles

10 miles

80,000 9.30 pm

43,000 157,000 9.30 pm

14,000

113

113

106

106

70

121

103

103

96

60

218

215

200

188

166

61

39

39

35

18

331

328

306

294

236

182

142

142

131

78

Required Road Closures Completion Time Time of Evafor Traffic Management cuation Order Arterial Freeway Total Plan

49,000 186,000 9.30 pm

5 miles w/5 yr. 137,000 planned improvement

5 miles w/1 yr. 127,000 planned improvement

66,000

2 miles

Evacuation Volume (veh) Area of Evacuation(*) EPZ Zone Shadow Total

Table 11. 3 Evacuation Statistics for Site 1

0%

0%

0%

0%

0%

0%

0%

0%

0%

0%

Background Traffic Reduction

15.00

19.300

20.00

15.00

9.30

12.00

5.30

6.00

6.30

4.30

Evacuation Time Estimate (hh :mm)

133,000

5 miles winter conditions

* Distances are approximated From Parsons Inc. 2016

5 miles w/5 yr. 159,000 planned improvement

5 miles w/1 yr. 147,000 planned improvement

132,000

5 miles

Evening

14,000

81,000

2 miles

95,000 4.30 pm

95,000 1.30 pm

49,000 208,000 4.30 pm

46,000 193,000 4.30 pm

43,000 176,000 4.30 pm

43,000 175,000 4.30 pm

14,000

81,000

Afternoon 2 miles

6.00 pm

6.00 pm

6.00 pm

6.00 pm

6.00 pm

2.30 pm

113

113

106

106

70

70

218

215

200

188

166

166

331

328

306

294

236

236

0%

0%

0%

0%

0%

0%

17.00

21.30

22.15

11.00

8.30

9.30

Day Morning Time

Night Time

Time of Day

10 miles

5 miles w/5 yr. planned improvement

87,000

80,000

36,000 257,000 7.30 pm

71,000

221,000

5 miles

5 miles w/1 yr. planned improvement

30,000 101,000 7.30 pm

31,000

2 miles

32,000 119,000 7.30 pm

36,000 116,000 7.30 pm

40,000 7.30 pm

34,000 108,000 9.30 pm

74,000

10 miles

9,000

32,000 101,000 9.30 pm

36,000 246,000 9.30 pm

94,000 9.30 pm

36,000 9.30 pm

Time of Evacuation Order

69,000

5 miles w/5 yr. planned improvement

210,000

64,000

5 miles w/1 yr. planned improvement

9,000

30,000

27,000

Shadow Total

5 miles

EPZ Zone

Evacuation Volume (veh)

2 miles

Area of Evacuation(*)

Table 11. 4 Evacuation Statistics for Site 2

9.00 pm

9.00 pm

9.00 pm

9.00 pm

9.00 pm

11.00 pm

11.00 pm

11.00 pm

11.00 pm

11.00 pm

Completion Time for Traffic Management Plan

75

75

70

71

71

67

67

63

67

67

Arterial

195

193

168

152

164

26

24

36

19

14

270

268

238

223

235

93

91

99

86

81

0%

0%

0%

0%

0%

0%

0%

0%

0%

0%

Background Traffic Freeway Total Reduction

Required Road Closures

7.00

7.30

11.00

7.25

6.00

5.00

5.15

9.45

5.15

4.00

Evacuation Time Estimate (hh : mm)

5 miles w/5 yr. planned improvement

From Parsons Inc. 2016

Evening

87,000

80,000

43,000 176,000 4.30 pm

221,000

10 miles

5 miles w/1 yr. planned improvement

43,000 175,000 4.30 pm

71,000

5 miles

49,000 208,000 4.30 pm

46,000 193,000 4.30 pm

95,000 4.30 pm

14,000

31,000

2 miles

32,000 112,000 1.30 pm

34,000 121,000 1.30 pm

80,000

87,000

5 miles w/5 yr. planned improvement

Afternoon 5 miles w/1 yr. planned improvement

36,000 257,000 1.30 pm

221,000

10 miles

40,000 1.30 pm

71,000

5 miles

9,000

30,000 101,000 1.30 pm

31,000

2 miles

6.00 pm

6.00 pm

6.00 pm

6.00 pm

6.00 pm

3.30 pm

3.30 pm

3.30 pm

3.30 pm

3.30 pm

75

75

70

71

71

75

75

70

71

71

195

193

168

152

164

195

193

168

152

164

270

268

238

223

235

270

268

238

223

235

0%

0%

0%

0%

0%

0%

0%

0%

0%

0%

8.45

8.45

9.45

11.15

9.15

7.00

12.30

14.15

12.00

11.30

304 Chapter 11 · Case Studies contributor was the use of inbound access closures, preventing vehicles from entering the EPZ. Another key determinant of clearance time in this case study was the geographic extent of the evacuation notice. As would be expected, the evacuation notices covering the largest areas generated the largest amount of traffic (both primary and shadow) and, correspondingly, resulted in the longest clearance times. For the largest evacuations, clearance time also reflected the effect of traffic management efforts to lower background traffic within the road network outside the EPZ. In these cases, demand management was assumed to lower background traffic by 20%. Nonetheless, despite these efforts, clearance times remained in excess of 15 hours and even reached about 20 hours in the case of a mid-day evacuation. This is a staggering result, considering that most travel distances would be less than 10 miles. In theory, the last evacuees to exit the EPZ would be able to move faster on foot than in a vehicle. However, the radiological exposure to the last evacuees would depend on their location in relation to the plume and the duration of their exposure to it. There would be little or no exposure to evacuees who were exiting the EPZ directly upwind from the plant (i.e., in exactly the opposite direction of the plume travel. This point underscores the importance of integrating the evacuation analysis with radiological release projections and meteorological forecasts. Another notable input assumption shown in the two tables is the number of ramp and road closures and movement restrictions that were assumed for the freeways and arterial roadways in each scenario. The intent of these closures was to reduce the congestive effects of background traffic within the evacuation network. Although this was quite simple to carry out in computer simulations, the resources required to actually implement these strategies in an actual evacuation would need to be closely examined. The hundreds of traffic enforcement personnel needed to implement the movement restrictions in many of these scenarios would simply not be available in an actual emergency.

11.3.5 Study Findings and Conclusions The location of this study within a highly populated metropolitan area represents an extreme example of traffic conditions within the spectrum of potential NPP emergencies. In fact, the traffic volumes generated by an emergency within this study area are about half of those observed during Hurricane Katrina across all of southeast Louisiana, despite the fact that the study area encompassed a road network about one tenth to one twentieth that size. Moreover, the presence of several major freeways that carry some of the highest daily traffic volumes in the world

Chapter 11 · Case Studies

305

means that an evacuation in this study area would likely experience background traffic 5–10 times greater than the Katrina evacuation. Finally, the geographic characteristics of the location, which would limit the available direction of travel to half of a normal 360-degree outward movement, also contributed enormously to the large ETEs. The simulation results showed that ETEs were expected to vary between about 4.5–20 hours for one set of conditions and about 4–12 hours for the other. The primary obstacle to rapid clearance of evacuees from the EPZ in either set of conditions was background traffic. In fact, to achieve even these clearance times, especially for an emergency that could occur during evening peak commute periods, extremely aggressive region-wide traffic management strategies would be needed. Study results suggested that authorities would need to close or force redirection at more than 230 individual locations for each unit. Given this large number of closure locations and other types of traffic management strategies (e.g., signal timing changes, travel demand management, manual intersection control) needed during an evacuation, there would be an enormous demand on the personnel resources of the agencies responsible for implementation. As is the case with any mass evacuation scenario, this finding emphasizes the need for considerable advance planning and coordination among all agencies involved in managing the evacuation. An additional finding of particular note in this study was the fact that most of the traffic management measures, despite having beneficial effects on the movement of evacuation traffic, also negatively affected the broader movement of traffic within the region. Of course, this was not unexpected or unique to this study area because, as outbound directional movements are favored, it is typical to see significant reductions to the movement of opposing and intersecting traffic. Facilitating the movement of these latter groups of vehicles on a regional basis also requires considerable advance planning and coordination of traffic management strategies that require considerable personnel and material resources in their implementation. More broadly, these results indicate that an evacuation of the study region at any time of the day or night would be accompanied by substantial delay and traffic congestion. However, this finding was neither unexpected nor unique to this location. Rather, it illustrates a fundamental limitation of all mass evacuations. Urban area road networks are planned and designed to process weekday peak period traffic. Even at these times, urban areas routinely experience congestion and delays. Obviously, any evacuation that would coincide with a commuter period would add to the difficulty of movement. And although they were not explicitly considered as part of this study, additional adverse conditions—such as traffic incidents, inclement weather, and road construction or other types of system failures—could also compound travel difficulties.

306 Chapter 11 · Case Studies

11.4 Case Study 4—Hazmat Transportation Evacuation Planning Planning for hazmat transportation evacuations is challenging because, as noted in Chapter 2, evacuation analysts do not know, before an incident occurs, what will be the specific location, what will be the type of hazmat released, and what will be the quantity released. A procedure for addressing these uncertainties can be illustrated by a case study of hazmat transportation to a proposed incinerator site in Linden, New Jersey (Lindell 1995). As indicated in Figure 11.3, US 1/9 was the major highway that would bring the hazmat shipments into Linden, but it was

Figure 11.3 Schematic Map of Linden New Jersey

Lindell 1995

Chapter 11 · Case Studies

307

unclear whether South Wood Avenue, Stiles Street, or Lower Road would be the most suitable route from US 1/9 to the incinerator site. The principal concern for evacuation planning was a neighborhood that had a limited number of egress routes because it was bounded on the south by an industrial area, the east by the New Jersey Turnpike, and the north by a cemetery and petroleum tank farms. Thus, the only routes out of the risk area would be two arterials (South Wood Avenue and Stiles Street) and three residential streets that provided access to US Route1/9 on the neighborhood’s west side.

11.4.1 Hazard Analysis The evacuation analysis began by obtaining an inventory of waste products that were expected to be shipped to the incinerator and identifying the chemicals that were listed within the Department of Transportation Emergency Response Guidebook (for the current edition, see PHMSA 2016) or listed in the EPA’s (1987) Technical Guidance for Hazards Analysis. One set of wastes was identified for further analysis because the DOT-ERG recommended protective action (either shelter inplace or evacuation) within 0.5 mile if a truck transporting these wastes were involved in a fire. Another set of wastes was included because they were included in the DOT-ERG’s Table of Protective Action Distances. The last set of wastes was identified for further analysis because they were listed in Exhibit C-1 of the EPA’s Technical Guidance. Data for these wastes were summarized in a table listing the chemical name, shipment quantity, expected release quantity, level of concern (if listed in the Technical Guidance as an Extremely Hazardous Substance—EHS), liquid factor ambient (all of the wastes would be shipped in liquid form), and release rate in pounds per minute (calculated from the release quantities and liquid factors). In addition, the table listed the Vulnerable Zone (VZ) distances for the EHSs derived from Exhibit 3-2 in the Technical Guidance on the assumption of urban terrain, an F atmospheric stability class, and 3.4 mph wind speed. Finally, the table listed the isolation/evacuation distances for those chemicals listed in the DOTERG. The analyses indicated that most (19 of 22) materials had VZ estimates of 0.5 mile or less. The wastes would be diluted rather than pure product, so this VZ estimate provides a conservative protective action distance to place on either side of the transportation route.

11.4.2 Protective Action Analysis Once the size of the VZ had been identified, the next step of the analysis was to determine if there was reasonable assurance that people within

308 Chapter 11 · Case Studies the VZ could be protected by evacuating (the first priority) or sheltering in-place (the second priority) if a release occurred. Any evacuation was expected to be of short duration because the source quantity was small so the plume would be of short duration and dissipate to safe concentrations within a few hours. Thus, people would need to take no preparatory actions other than gathering household members. Moreover, because the area was relatively small (roughly 1200 households) and the 0.5 miles distance to a safe location was short, it was unlikely that an evacuation would produce the types of significant queues that inflate evacuation travel time during much larger evacuations. Instead, the critical ETE components would be the times for incident detection and notification, protective action decision making by local authorities, and warning dissemination.

11.4.3 Detection and Notification It has long been documented that other motorists provide prompt detection of highway incidents and notification of local emergency dispatch centers (Lindell and Perry 1980). Based on data from the early 1990s, Evanco (1999) reported that the average notification time was 5.2 minutes in urban areas and 9.6 minutes in rural areas. Brodsky (1993) reported that approximately 95% of all urban accident notifications were received within ten minutes during daytime and evening hours (7am– 11pm) and the percentage only dropped to 90% during the overnight hours (11pm–7am). Notifications were somewhat slower in rural areas, with the corresponding percentages being 80% and 65%, respectively. These notification times have undoubtedly been decreasing over time because they depend in part on the proportion (p) of passing vehicles that are equipped with mobile phones. For example, Mussa’s (1997) mathematical model was able to produce notification time curves like those reported by the Evanco (1999) and Brodsky (1993) with p = .01. The proportion of vehicles with either hands-free devices or passengers who could make a call is clearly higher than that and has undoubtedly increased over time. Detection would be essentially instantaneous if a vehicle carrying hazmat were equipped with an automatic crash notification device (Clark and Cushing 2002). A crash sensor and geographic locator system could immediately notify the carrier’s dispatcher who, in turn, could notify the local jurisdiction’s dispatch center.

11.4.4 Protective Action Decision Making There is limited research on the time required to make a protective action decision, with the available data on hazardous materials incidents

Chapter 11 · Case Studies

309

indicating that this varies substantially from one incident to another. For example, Rogers (1994) reported that 50% of decisions are made within 30 minutes, 75% within 60 minutes, and 90% within 120 minutes. The longest delays are probably due to decision makers’ uncertainties about the situation. These include uncertainties about the characteristics of the release and meteorological conditions. As noted in Chapter 2, important characteristics of the release include the chemical’s toxicity, volatility, and water solubility; its likelihood of being released (if a release is not already in progress); the quantity of chemical available for release; and the current release rate (and possible maximum release rate). Important meteorological conditions include wind direction, wind speed, atmospheric stability, and precipitation. However, unnecessary delays can be avoided if the emergency response personnel have been trained in the use of the DOT-ERG and the community’s EOP clearly indicates when the authority for protective action decision making will be assumed by higher level authorities such as the local emergency manager or mayor.

11.4.5 Warning and Public Information Most local EOPs anticipate warning those in the risk area by means of route alerting with loudspeakers broadcasting warnings from emergency vehicles and by face to face contacts with emergency responders going door to door. Both of these warning mechanisms would be supplemented by emergency information disseminated through the Emergency Alert System (EAS), which is a network of broadcast, cable, satellite, and wireline service providers that have agreed to transmit official emergency messages to the public during disasters (FEMA, no date). These three warning mechanisms are the most commonly used methods for disseminating warnings in no-notice incidents because they require little investment in specialized equipment, provide adequate penetration of normal activities, are not susceptible to significant message distortion, and achieve adequate rates of dissemination over time (see Chapter 3, Table 3.1). They are particularly suitable for hazmat transportation incidents that have a rapid onset and the small (0.5 miles) scope of impact identified in the Linden analysis. Rogers and Sorensen (1989) reported data from five hazmat incidents in which the times to complete warning dissemination were less than 2.5 hours. In addition, many communities rely on other warning mechanisms such as the National Oceanographic and Atmospheric Administration’s Weather Radio, whose receivers remain in standby mode until receiving a tone alert that turns on the receiver for a warning message (see Lindell and Prater 2010, Table 3). In addition, many communities have adopted automated telephone notification systems that call the telephone numbers

310 Chapter 11 · Case Studies and transmit recorded warnings to recipients. Such calls can reliably be directed to all of the landlines located within a geographic risk area but a substantial proportion of the population has only cell phones. This population segment can receive warnings over the Wireless Emergency Alert (WEA) system that broadcasts to all phones within range of targeted cell towers. Alternatively, cell phone users can register with a Reverse 911 system that targets their phone numbers. These rapid notification systems are especially important for special facilities such as those noted in previous chapters that have large populations with limited mobility. Finally, research on a wide range of incidents has found that informal peer warning systems play a significant role in relaying warnings. That is, friends, relatives, neighbors, and coworkers who receive a warning typically relay it to others whom they think are at risk (Lindell 2017).

11.4.6 Protective Action Implementation Given the limited number of egress routes from the area, the Linden incinerator analysis then examined the feasibility of sheltering in-place for residential areas. The analysis also determined whether there was any combination of release sites and wind directions that would force residents of the risk area to evacuate through a release plume to reach safety. Next, the analysis determined whether the ERS had the capacity to evacuate all evacuating residents without the last one being overtaken by the plume. Finally, the analysis examined the feasibility of evacuating or sheltering in-place from special facilities within the VZs. These special facilities are distinctive because of complications caused by their users’ mobility (e.g., nonambulatory nursing home residents), permanent residence (e.g., transients who are unfamiliar with local roads), periods of use (e.g., special events that occur only a few times a year), user density (e.g., high capacity sports stadiums), availability of structures to shelter in-place (e.g., athletic fields), and available transportation support (e.g., bus riders).

11.4.6.1 Sheltering In-Place for Residential Structures As was the case for most communities at that time, the Linden EOP made no mention of shelter in-place. To validate the consideration of this protective action, local emergency managers should, at minimum, conduct a curbside survey within the VZs for hazmat transportation routes to determine the extent to which local structures are likely to have adequately low air infiltration rates. The Linden analysis noted that, if storm windows and doors are present, it is likely that lower cost measures such as weather stripping have also been adopted. Thus, all

Chapter 11 · Case Studies

311

else being equal, the presence of storm windows and doors suggests that residences have significant levels of resistance to air infiltration. However, if many structures are old, there is likely to be leakage elsewhere. Consequently, further analyses should be conducted to determine typical levels of air exchange rates per hour. Such analyses are especially important for special facilities housing relatively immobile populations such as hospitals, nursing homes, schools, and jails. Once data on air infiltration rates have been obtained, it is possible to examine alternate hazmat release scenarios to identify the situations in which shelter inplace provides adequate protection.

11.4.6.2 Evacuation of the General Public The feasibility of the general public evacuation was assessed with respect to two objectives. First, everyone in the VZ should have at least one evacuation route that is safe, regardless of the hazmat release location and wind direction. Second, the ERS must have the capacity to handle the volume of evacuating vehicles to ensure that the last vehicles in line are not overtaken by the plume. As Figure 11.3 indicates, the greatest threat to meeting the first condition (at least one evacuation route) would involve hypothetical Plume A, which would prevent residents of the Tremley neighborhood directly east from evacuating on either of the normal routes out of the area—Lower Road and South Wood Avenue. However, this neighborhood could be evacuated successfully using a controlled access gate onto the New Jersey Turnpike. The greatest threat to meeting the second condition would involve hypothetical Plume B, which would force evacuation of the neighborhood onto a single route, Lower Road. Parsons Brinckerhoff (1992) estimated the number of dwelling units in this neighborhood to be approximately 1,200 and data available at the time of the study indicated that there would likely be approximately 1.33 evacuating vehicles per household (Lindell and Perry 1992). Consequently, an evacuation of the neighborhood would involve approximately 1600 vehicles. The loading of the evacuation route was estimated by assuming that 30% of the evacuees depart in the first 15 min after warning receipt, another 50% depart in the next 15 minutes, and the last 20% depart in the following 15 minutes. This assumed loading function was deliberately selected to be more rapid than any actually reported in the evacuation literature (see Lindell and Perry 1992, Table 8.3 and Figure 8.8), and thus more conservative because it would be more likely to produce congestion. According to the assumed loading function, there would be 480 vehicles on the road in the first 15 minutes, 800 in the second 15 minutes, and 320 in the third 15 minutes. Under

312 Chapter 11 · Case Studies these conditions, some congestion could result if all of the evacuees are routed onto Lower Road and the traffic capacities were as low as the 1200 vehicles per hour per lane expected for forced flow on a two lane undivided rural road with one lane in each direction (FEMA 1984, pp. 2–22). The Linden analysis concluded that even in the unlikely event that such congestion occurred, it could be managed by routing some of the traffic east onto South Wood Avenue onto the New Jersey Turnpike or by implementing contraflow on Lower Road. Significantly, even if traffic congestion did form on this route, the tail of the queue would develop in an area that is at a right angle to, not under the plume. Consequently, there would be no chemical exposures.

11.4.6.3 Evacuation of Special Facilities Nursing homes and hospitals pose significant evacuation problems because their residents have reduced mobility and require transportation support—especially those on life support systems. Evacuation would not be required if the structure provides adequate protection against air infiltration and ventilation systems can be shut down promptly. Schools and day care facilities also have occupants with reduced mobility (partly because they require supervision) and limited access to transportation. Thus, assurance of their buildings’ effectiveness for shelter in-place is important if evacuation transportation assistance is not readily available. One significant complication for school evacuations is the likely convergence of parents. Even if they are told ahead of time that their children will be transported to a safe location, it is likely that some parents will clog roads near the school in an attempt to pick up their children as soon as possible. Thus, an effective traffic management plan is needed. Apartment complexes have relatively high population densities, and mobile users that often have their own transportation. Like the residential structures addressed in Section 11.4.6.1, apartment complexes can be variable in their suitability for shelter in-place because this protective action depends on the quality of original construction and continuing maintenance. Similarly, hotels and motels have relatively high population densities, and mobile users that often have their own transportation. However, they are occupied by transients who may be able to leave rapidly but require specific directions for exiting the risk area. Nonetheless, these facilities are very likely to provide adequate protection against air infiltration and ventilation systems can be shut down promptly. Commercial/industrial parks often have a high density that could clog evacuation routes, especially during shift changes. However, such

Chapter 11 · Case Studies

313

facilities can often be notified rapidly through public address systems and are occupied by mobile users that often have their own transportation. These facilities can be quite variable in their ability to support shelter in-place. Local business districts have relatively high population densities and mobile users that vary in access to their own transportation. The number of parking spaces on-street and in parking lots can provide an indication of the number of vehicles that will try to enter the evacuation route system. Like apartment complexes, stores in local business districts vary in their suitability for sheltering in-place depending on the quality of their original construction and continuing maintenance. Athletic fields and parks provide no opportunity for shelter in-place and, to the degree that they concentrate many people from outside the VZ, they can add significantly to the size of the evacuating population. However, users of such facilities are typically ambulatory and have their own transportation—although some users might have arrived on public transportation and, thus, need transportation assistance. These facilities are also most likely to be used during evenings and weekends when there is reduced traffic from commercial and industrial facilities. Community recreation centers are likely to have the same patterns of usage time and user mobility as athletic fields and parks but provide an opportunity for shelter in-place. Churches and other religion-based facilities also have users that are mobile, have their own transportation, and are in use during off-peak hours of commercial and industrial activity.

References Bierling, D.H., Rogers, G.O., Jasek, D.L., Protopapas, A.A., Warner, J.E., Olson, L.E. 2011. Guidebook for Conducting Local Hazardous Materials Commodity Flow Studies. Transportation Research Board, Washington, DC. Brodsky, H. 1993. The call for help after an injury road accident. Accident Analysis and Prevention 25 (2), 123–130. Cherney, M. 2013. Evacuteer.org. Tulane University’s Public Service Internship Blog, New Orleans. Accessed 27 February, 2016 at tulanepsip.wordpress. com/2013/07/22/evacuteer-org-by-meredith-cherney/. City of San Diego Response. 2007. After Action Report – October 2007 Wildfires. City of San Diego, San Diego CA. Clark, D.E., Cushing, B.M. 2002. Predicted effect of automatic crash notification on traffic mortality. Accident Analysis & Prevention 34 (4), 507–513. Cova, T.J., Dennison, P.E., Li, D., Drews, F.A., Siebeneck, L.K., Lindell, M.K. 2017. Warning triggers in environmental hazards: who should be warned to do what and when? Risk Analysis 37 (4), 601–611. Evanco, W.M. 1999. The potential impact of rural mayday systems on vehicular crash fatalities. Accident Analysis & Prevention 31(5),455–462.

314 Chapter 11 · Case Studies Evacuteer. 2016. The Power of Us. Accessed 17 January, 2018 at www.evacu teer.org/. Fang, L., Edara, P. 2014. Mobility benefits of intermediate crossovers on contraflow facilities during hurricane evacuation. Transportation Research Record 2459, 37–46. FEMA—Federal Emergency Management Agency. no date. An Emergency Alert System Best Practices Guide – Version 1.0. Federal Emergency Management Agency, Washington DC. FEMA—Federal Emergency Management Agency. 1984. Transportation Planning Guidelines for the Evacuation of Large Populations, CPG-2-15. Federal Emergency Management Agency, Washington DC. Jones, J.A., Walton, F., Smith, J.D., Wolshon, B. 2008. Assessment of Emergency Response Planning and Implementation for Large Scale Evacuations. SAND2007-1776P, NUREG/CR-6981. US Nuclear Regulatory Commission, Washington DC. Jones, J.A., Walton, F., Sullivan, R.L. 2008. Review of NUREG-0654, Supplement 3, Criteria for Protective Action Recommendations for Severe Accidents: Focus Groups and Telephone Survey SAND2008-4195P, NUREG/CR-6953, Vol. 2. US Nuclear Regulatory Commission, Washington DC. Jones, J., Walton, F., Wolshon, B. 2011. Criteria for Development of Evacuation Time Estimate Studies. SAND2010-0016P, NUREG/CR-7002. US Nuclear Regulatory Commission, Washington DC. Knabb, R.D., Rhome, J.R., Brown, D.P. 2005. Tropical Cyclone Report: Hurricane Katrina, 23-30 August 2005. Miami FL: National Hurricane Center. Lindell, M.K. 2017. Communicating imminent risk. In: Rodríguez, H., Donner, W., Trainor, J. (Eds.), Handbook of Disaster Research, Springer, New York, pp. 449–477. Lindell, M.K., Kang, J.E., Prater, C.S. 2011. The logistics of household hurricane evacuation. Natural Hazards 58 (3), 1093–1109. Lindell, M.K., Perry, R.W. 1980. Evaluation criteria for emergency response plans in radiological transportation. Journal of Hazardous Materials 3 (4), 335–348. Lindell, M.K., Perry, R.W. 1992. Behavioral Foundations of Community Emergency Planning. Hemisphere Press, Washington DC. Lindell, M.K., Prater, C.S. 2008. Behavioral Analysis: Texas Hurricane Evacuation Study. Texas A&M University Hazard Reduction & Recovery Center, College Station TX. Lindell, M.K., Prater, C.S. 2010. Tsunami preparedness on the Oregon and Washington coast: Recommendations for research. Natural Hazards Review 11 (2), 69–81. Mussa, R.N. 1997. Evaluation of driver-based freeway incident detection. ITE Journal, 67(3), 33. Parsons, Inc., 2016. Nuclear Emergency Evacuation Traffic Management Plan – Phase 2. Preliminary Draft Report. Parsons Inc., Pasadena CA. Parsons Brinckerhoff. 1992. GAF Hazardous Waste Facility: Linden New Jersey Assessment of Alternative Access Routes. Parsons Brinckerhoff, West Trenton NJ.

Chapter 11 · Case Studies

315

PHMSA—Pipeline and Hazardous Materials Safety Administration. 2016. Emergency Response Guidebook. Pipeline and Hazardous Materials Safety Administration, Washington DC. www.phmsa.dot.gov/hazmat/outreach-training/ erg. Rogers, G.O. 1994. The timing of emergency decisions: modeling decisions by community officials during chemical accidents. Journal of Hazardous Materials 37 (2), 353–373. Rogers, G.O., Sorensen, J.H. 1989. Warning and response in two hazardous materials transportation accidents in the US. Journal of Hazardous Materials 22 (1), 57–74. USEPA—US Environmental Protection Agency. 1993. Hazard Analysis on the Move. EPA 550-F-93-004. US Environmental Protection Agency, Washington, DC, accessed 31 January 2018 at www.epa.gov/epcra/hazards-analysismove-sara-title-iii-epcra-and-conducting-commodity-flow-study. Wolshon, B. 2009a. Transportation’s Role in Emergency Evacuation and Reentry. National Cooperative Highway Research Program, Synthesis of Highway Practice 392. Washington DC. Wolshon, B. 2009b. The role of transportation in evacuation and reentry: a survey of practice. Journal of Transportation Safety & Security 1 (3), 224–240. Wolshon, B., McArdle, B. 2009. Temporospatial analysis of Hurricane Katrina regional evacuation traffic patterns. Journal of Infrastructure Systems 15 (1), 12–20. Wu, H-C., Lindell, M.K., Prater, C.S. 2012. Logistics of hurricane evacuation in Hurricanes Katrina and Rita. Transportation Research Part F 15 (5), 445–461. Wu, H-C., Lindell, M.K., Prater, C.S., Huang, S-K. 2013. Logistics of hurricane evacuation in Hurricane Ike. In: Cheung, J., Song, H. (Eds.), Logistics: Perspectives, Approaches and Challenges. Nova Science Publishers, Hauppauge, NY, pp. 127–140.

Glossary

Access control: official measures taken to prevent unauthorized persons from entering an evacuation zone. Activity chain: a sequence of activities (e.g., working followed by shopping) in which an individual participates. Adaptive traffic signal control: a traffic management strategy in which traffic signal timing changes, or adapts, based on actual traffic demand. Arterial: roadways that are designed to carry moderate to heavy traffic volumes. Although they are not access controlled like freeways, their primary purpose is to serve a mobility, rather than access, function. However, they commonly do serve often abutting land and flow on these roads is interrupted by intersection control and other access points. Background traffic: vehicles engaged in normal travel activities that are not part of an active evacuation. Compliance: following a recommended course of action. Congestion: conditions that occur when traffic demand nears or exceeds the capacity of a road or an incident occurs to prohibit or limit the flow of people and vehicles. Congested conditions are typically characterized by slower speeds, longer trip times, and increased vehicular queueing. Contraflow: a temporary traffic operation strategy in which flow in an opposing direction of travel is reversed to serve flow in the other direction. It may include some or all lanes of a roadway and is controlled using a variety of restrictions depending on the speed and access conditions of the roadway. Crossing elimination: a traffic management strategy that involves the prohibition of some intersection movements that would be allowed under normal conditions. “Cry wolf”: a “false alarm” or warning of a disaster strike that does not materialize at the warned location. Cut-through traffic: vehicular traffic that passes through an area (typically residential) without stopping or without at least an origin or a destination within the area, typically for the purpose of shortening travel time and/or distance.

Glossary

317

Evacuation capacity: the ability of the evacuation route system to handle evacuation traffic, in vehicles per hour. Evacuation demand: the amount of traffic attempting to use the Evacuation Route System, in vehicles per hour. Demand may be further characterized by origin, destination, and departure time. Evacuation fatigue: the decrease in households’ willingness to comply with evacuation recommendations as a function of repeated evacuation recommendations. Evacuation participation rate: the percentage of the households in the officially designated evacuation zone that evacuate. Evacuation Route System (ERS): the portion of the available road network that officials have designated for use as evacuation routes. Evacuation shadow: household evacuation from areas that are not officially designated as part of an evacuation zone. Evacuation supply: see evacuation capacity. Evacuation zone: a geographical area in which people have been advised to evacuate. Evaculane: road shoulders or center turn lanes that can be operated in the outbound direction to increase capacity (Texas Department of Public Safety and Texas Department of Transportation 2013). Flood discharge: the volume of water per unit of time. Flood stage: the height of water above a defined level. Hazard zone/hazard area: a geographic area within which scientific assessments have indicated that the threat to human health and safety would exceed a designated level of concern. Hazmat: hazardous materials, which include chemical, biological, radiological, nuclear, explosive (CBRNE) agents. Ingress/egress management: see access control. Macro-level simulation: representation of traffic flow as fluid flow through a pipe, in which only roads down to the functional classification level of “collector-distributor” are typically included in the simulation and the characteristics and movements of individual vehicles and people are aggregated to represent “group averages.” Meso-level simulation: representation of traffic flow by incorporating attributes of both micro-level and macro-level simulations. Micro-level traffic simulation: simulation in which the movement of individual vehicles are computed and updated on a second-bysecond (or sub-second) basis. This relies on random number generation and probability distributions to generate vehicles, select routing decisions, and determine the behavior of individual drivers. This results in a stochastic set of results that requires a single model to be run several times with different random number seeds to obtain an “averaged” result of likely conditions. Model: a simplified representation of reality, usually mathematical or computational. Model calibration: the process of “matching” certain influential behaviors in the model to fit analogous activities under normal conditions.

318 Glossary Model validation: the process confirming that the elements of a model are consistent with the features of the real world system being modeled. Model verification: the process of determining that a model implementation and its associated data accurately represent the intended conceptual description and specifications. Optimization: a mathematical technique that prescribes a course of action based on an objective function and a set of constraints. Pass through traffic: pass through traffic includes vehicles that move through an area, but do not originate or terminate their travel within it. Phased evacuation: a strategy in which the evacuations of different zones are initiated at different times, usually beginning with the area at greatest risk. Preparation trip: travel performed to obtain materials to protect the home or for the evacuation. Queue: a line of customers (e.g., vehicles) waiting for service (e.g., to go through an intersection). Reception Center: a safe location at which evacuees register for services provided by government agencies and non-governmental organizations. Reentry: the process of entering an evacuation zone after determining that it is safe to do so. Note that some residents are likely to reenter the evacuation zone before local officials authorize reentry. Risk area: see hazard zone/hazard area. Road network: the entire set of roads within a designated area. Saffir-Simpson Category: a classification of hurricane magnitude that is determined by maximum wind speed. Scenario: a set of conditions that represents the context of the hazard, warning, and demand considerations. Sequenced evacuation: see phased evacuation. Shadow evacuation: see evacuation shadow. Shelter: a public facility at which people can sleep, eat, and use other services free of charge while they are evacuated. Shelter in-place: a protective action in which people go indoors and close doors and windows to protect themselves from a hazard. In general, they would stay in an interior room, but would go to a basement for tornadoes and radiological releases, or to a higher floor for floods or storm surges. Simulation: the representation of the primary operation and/or desired essence of a real-world process or system. While this representation of this system may be captured in a “model,” simulation is used to represent the operation of the system over time. Staged evacuation: see phased evacuation. Storm surge: a dome of water that is most commonly associated with hurricanes, but also can be caused by extratropical cyclones (nor’easters).

Glossary

319

Transient: persons in an evacuation zone who do not normally work or reside there. Transient traffic: evacuation traffic associated with population cohorts who visit, but do not reside, in an evacuation area. This can include tourists, shoppers, employees, etc. Trip: travel from one location (origin) to another (destination) Trip Chain: travel from an origin to a destination with a number of stops in between. Tour: a series of trips that begins and ends at the same location. Turn restriction: a movement through an intersection that is prohibited.

Index

Note: page references in italics indicate figures; bold indicates tables. Abdelgawad, H. 122, 186 Abdulhai, B. 122, 186 access control 184, 193, 256; see also ingress/egress management accommodations: estimating demand 160–3; modeling and simulation 228–9; temporary recovery 270; types 124–6 ACH see air changes per hour action zones 58, 144–5, 151, 154, 158, 162, 247 activity chains 165, 165–6, 172 actual response surveys 69, 150 adaptive traffic signal control 103–4 Advanced Traveler Information System (ATIS) 207, 208, 210, 211 agency notification, disaster recovery 272 aggregate model: accommodations/ destinations 160–2; departure times 170–1; vehicle numbers 150–7, 153 air changes per hour (ACH) 50, 51, 52 Aldrich, D.C. 51 Alert 44, 81; see also Emergency Classification System AlertSanDiego system 291 alluvial fan flooding 15 Alsnih, R. 122 Anno, G.H. 51, 52 Antoniou, C. 210 apartment complexes 52, 312 arterial roadways 177, 196, 201–6, 202, 304 athletic fields 313 ATIS see Advanced Traveler Information System

atmospheric stability 36, 39, 51, 307, 309 atomic fission reaction 41 Aumonier, S. 239 authorities’ decision times 100–2, 101 background demand reduction 299 background traffic 145–9, 231 Baker, E.J. 88, 92, 107, 125, 128, 129, 131, 132, 150, 152, 155, 258, 259, 276 Barceló, J. 241 Barton, A. 123 behavior 68–74, 70; compliance 72–4; drivers 122–4; panic 122, 238; rational 123; shadow evacuation 74–6 behavioral forecasting 142–72; accommodations/destinations 160–3; background traffic 147–9; departure times 153, 159, 163–72, 165, 169; evacuation zones 142–6; hazard maps 142–3; vehicle numbers 149–60, 153, 159 Ben-Akiva, M. 157 Bier, V.M. 239 Bish, D.R. 123 Blumenberg, E. 144 Boggs, K.S. 108 Bonsall, P.W. 210 boundedly rational behavior 123 Boyle, A. 210 Brodsky, H. 308 building construction practices 180, 272 Burson, Z.G. 51, 52 business operation, temporary 270

Index carpooling 132 Caution Area 58, 58 CCTV see Closed Circuit Television Chamlee-Wright, E. 258 Changeable Message Sign (CMS) 189, 208, 233 Chatterjee, K. 210 Chen, M. 235 Cheng, G. 128, 129 children 9, 57, 74, 83, 92, 108–12, 110–11, 165–6, 244 Chiu, Y-C. 186, 221, 240 churches 258, 313 clearance time 2–6, 5, 8, 52, 186, 205, 225, 230, 235, 242, 296, 299, 304–5; see also Evacuation Time Estimate (ETE) Closed Circuit Television (CCTV) 6, 189, 207, 208 CMS see Changeable Message Sign Collins, A.J. 238 commercial/industrial parks 312–13 commodity flow studies 40, 282 communication see public information communication zones 143–5 community centers 313 compliance: reentry plan 259–60, 277; route guidance 209–12; see also noncompliance, participation rates Connected Vehicle (CV) 212 contraflow 4, 190–6, 192, 194, 195, 232–3 control structure failures (dams or levees) 15, 20, 53 Cova, T.J. 100, 102, 127, 130, 159, 234, 260, 262, 265 Cox, J. 82 critical facility restoration 269–70 crossing elimination 202–3, 205, 234–5 “cry wolf” effect 89 Cutter, S. 87, 91, 92, 104, 105, 126, 131, 135, 136, 137, 161 cut-through traffic 196 CV see Connected Vehicle Cyclone Tracy (1974) 114 Czajkowski, J. 183 dam failure 15, 20, 53 damage assessment 267–8 Dash, N. 126, 159 Davidson, R.A. 5 day care facilities 108, 244, 312 debris management 270, 271, 273 debris/ice dam failures 15 demand see evacuation demand

321

demographics: as predictors of evacuation 92–3; reentry planning principles 276–7 demolition, emergency 270, 271 Demuth, J.L. 88 Department of Public Works (DPW) 288 Department of Transportation (DOT) 9, 10, 40, 41, 47, 53, 100, 148, 198–9, 208, 210, 212, 220, 224, 232, 233, 260, 288, 289, 307, 309 departure times 113–17, 115, 116, 117; calculating 153, 159, 163–72, 165, 169; modeling and simulation 229–30; motivating timely departures 183–4 destinations 126–9; estimating demand 160–3; modeling and simulation 228–9 Dia, H. 209, 210, 211 disaster assessment 266, 267–8 disaster assistance 271, 273, 274 disaster memorialization 272 disaster recovery management 267, 272–5; agency notification and mobilization 272; mobilization of recovery facilities and equipment 273; internal direction and control 273; external coordination 273; public information 273–4; recovery legal authority and financing 274; administrative and logistical support 274; documentation 274–5 disaster recovery planning 265–75, 266; assessment 267–8; long term 271, 271–2; short term 268–71 disaster re-entry: returnees’ information sources 275–7; perception of danger 276; re-entry plan compliance 277; re-entry plan and process 277 Dixit, V. 186, 221, 233, 241 DMS see Dynamic Message Sign documentation, recovery 274–5 donations management 271, 273 Dore, M.A. 51, 52 DOT see Department of Transportation DOT-Emergency Response Guidebook (DOT-ERG) 40–1, 53, 100, 307, 309 Dow, K. 87, 104, 105, 126, 131, 135, 136, 137, 161 Downing, T. 86 DPW see Department of Public Works Drabek, T.E. 108 driver behavior 122–4 Dye, K.C. 102

322 Index Dynamic Message Sign (DMS) 123, 148, 182, 189, 208, 209, 210, 247, 285, 288 EAS see Emergency Alert System Edara, P. 191 Edmonds, A.S. 91, 92 education programs 178–81 Eggers, J.P. 102 EHS see Extremely Hazardous Substance Elliott, J.R. 258, 259 EMA see Emergency Management Agency Emergency Alert System (EAS) 47, 291, 309 emergency assessment 266 Emergency Classification System 44, emergency demolition 270, 271 Emergency Management Agency (EMA) 19, 86, 267, 293 Emergency Operations Centers (EOC) 23, 53, 122, 206–12, 274, 288, 289 Emergency Operations Plan (EOP) 152, 283, 309, 310 Emergency Planning and Community Right to Know Act (EPCRA) 41 Emergency Planning Zone (EPZ) 8, 43, 75, 109, 143–5, 151, 152, 153–4, 156, 185, 225, 293, 294, 295, 296, 298, 299, 300, 304, 305 Emergency Response Guidebook see DOT-Emergency Response Guidebook Emergency Response Planning Area (ERPA) 144, 295 emergency shelter 270 environmental cues, as predictor of evacuation 86 environmental remediation 272 EOC see Emergency Operations Center EOP see Emergency Operations Plan EPCRA see Emergency Planning and Community Right to Know Act EPZ see Emergency Planning Zone equipment, mobilization of recovery 273 Ericson, D.M. 51 ERPA see Emergency Response Planning Area ERS see Evacuation Route Systemevacuation accommodations see accommodations ETE see Evacuation Time Estimate evacuating vehicles see travel modes

evacuation capacity see evacuation supply evacuation assistance, Hurricane Katrina (case study) 285–7, 287; California wildfires (case study) 306 evacuation decision arcs 62–5, 64 evacuation decision deadline 62 evacuation demand 3; estimation 147–72, 153, 159, 165, 169; generation (case study) 294–9, 295, 296, 297, 298; management 181–7; modeling and simulation 226–31 evacuation departure times see departure times evacuation destinations see destinations evacuation disciplines, need for multiple 9–11 evacuation fatigue 90 evacuation fundamentals 2–4 evacuation logistics see logistics evacuation management see management strategies evacuation modeling 4–9, 5, 7, 8 evacuation participation rates see participation rates evacuation preparation times see preparation times evacuation routes see routes/ roadways Evacuation Route System (ERS) 3, 63, 313, 317 evacuation shadow 8, 67, 74–6 evacuation supply 3; arterial routes 201–6, 202; freeways 189–201, 192, 194, 195, 197, 199, 200; variables and models 231–40 Evacuation Time Estimate (ETE) 2, 5, 5, 6, 7, 9, 10, 11, 33, 62–5, 67, 123, 131, 146, 148, 149, 164, 172, 220, 221, 225, 230, 238, 242, 247, 248, 293, 296, 305, 308; see also clearance time evacuation travel times see travel times evacuation zones 142–6 evacuee concerns, while evacuated 260–2 evacuee information sources, while evacuated 262–4, 263 Evaculanes 199, 199, 201 Evanco, W.M. 308 experience 12, 88, 88–90 external radiation exposure 51–2 extreme fire behaviour advisory 25

Index Extremely Hazardous Substances (EHS) 39, 41, 307 false alarm 62, 89, 186; see also false positive decisions false negative/positive decisions 64 family reunification 165–7 Fang, L. 191 FEMA see Federal Emergency Management Agency financing, disaster recovery 274 fire weather watch 25 firestorms 24 FIRM see Flood Insurance Rate Map fixed-site facilities, hazmat releases 39, 40 flash flood warnings 21, 86 flash flooding 15, 86 flood advisory 20 flood discharge 15–17, 16, 20 Flood Insurance Rate Map (FIRM) 19 flood stage 15–16, 20 flood warnings 20, 77 flood watch 20 floods 15–21, 16, 18; hazard analysis 18–19; monitoring and forecasting 19–21, 20; types 15 forecasting: floods 19–21, 20; hurricanes 32–4, 33, 34; tsunamis 22–3, 23; wildfires 25 forecasts, as predictor of evacuation 80, 80–3 fortified homes 49 freeway traffic management 189–201; contraflow 190–6, 192, 194, 195; modeling and simulation 234; ramp closures 198; route closures 196–7, 197; use of shoulders 198–201, 199, 200, 234 Fritz, C.E. 123 Fu, H. 230 fuels, wildfire 24 Fukushima Daiichi tsunami nuclear power plant Natech disaster (2011) 221 Fussell, E. 258, 259 future systems 212–13 Garfin, D.R. 72, 87 General Emergency 44; see also Emergency Classification System Geographic Information Systems (GIS) 231–2, 290 Gladwin, C.H. 87 Gladwin, H. 87 Gottumukkala, N.R. 238

323

Greene, M.R. 77, 106 Groen, J.A. 258, 259 Gruntfest, E. 86 Hammond, G.D. 239 Handbook of Chemical Hazard Analysis Procedures 41 Hans, J.M. 238, 239 HAR see Highway Advisory Radio hardscape 18 Harney, D. 210 Hartsough, D. 239 hazard analysis: floods 18–19; hazardous materials releases 39–44, 40, 43; hazmat transportation evacuation planning (case study) 307; hurricanes 30–1, 31; tsunamis 22; wildfires 24–5 hazard awareness program see hazard education program hazard education program (pre-impact) 178–81; content and channels 179–81; overview 178–9 hazard maps 142–3 hazard preparation activities 167 hazard scenarios 242–3; impact timing 243–6; transportation system 246–7 hazard source control and area protection 271–2 hazard warnings 181–7 hazard zone/area 143, 203, 204; see also Risk Area hazardous materials (hazmat) releases 34–44, 36, 36, 38; fixed-site facility analysis 39, 40; nuclear power plant analysis 41–4, 43; transportation analysis 40, 40–1 hazardous materials (hazmat) transportation planning (case study) 306, 306–13; detection and notification 308; hazard analysis 307; protective action analysis 307–8; protective action decision making 308–9; protective action implementation 310–13; warning and public information 309–10 hazards 15–44; floods 15–21, 16, 18, 20; hazmat releases 34–44, 36, 36, 38, 40, 43; hurricanes 25–34, 27, 31, 33, 34; tsunamis 21–3, 23; wildfires 24–5 hazmat see hazardous materials High Occupancy Vehicle (HOV) 60, 189, 190

324 Index Highway Advisory Radio (HAR) 135, 189, 208 Highway Capacity Manual 231–2, 234, 246 historic preservation 272 Hobeika, A.G. 170 Hori, M. 257 horizontal proximity 24; see also fuels, wildfire hospitals 19, 51, 59, 152, 267, 269, 283, 286, 290, 311, 312 host counties 122, 125, 128 hotels 121, 124–6, 151–2, 160, 163, 229, 248, 270, 312 House, D. 82 housing: permanent 270; temporary 270; types and vulnerability 90–1; see also accommodations; mobile homes HOV see High Occupancy Vehicles Huang, S-K. 72, 85, 86, 87, 88, 89, 90, 92, 105, 276 Hurricane Alicia (1983) 28 Hurricane Andrew (1992) 91, 257 Hurricane Bertha (1996) 90, 161, 162 Hurricane Bonnie (1998) 124, 125, 126 Hurricane Bret (1999) 102–3, 124, 125, 130, 132, 133, 135, 136, 184 Hurricane Charley (2004) 73, 76, 103, 107, 125, 184 Hurricane Diana (1984) 161, 162 Hurricane Earl (2010) 77, 83 Hurricane Elena (1985) 115, 116 Hurricane Eloise (1975) 115, 116 hurricane eye 26, 28, 32, 33, 57–8, 58, 64, 116 hurricane eye wall 26 Hurricane Floyd (1999) 9, 76, 80, 80, 84, 84–5, 87, 89, 91, 93, 105, 116, 117, 124, 125, 126, 127, 128, 129, 130, 135, 136, 161, 162, 190, 221 Hurricane Fran (1996) 90, 161, 162 Hurricane Frederic (1979) 89 Hurricane Georges (1998) 126, 127, 128, 130, 259 Hurricane Gustav (2008) 233 Hurricane Harvey (2018) 29 Hurricane Ike (2008) 29, 85, 116, 116, 124, 125, 126, 127, 129, 130, 131, 132, 133, 156, 162, 187, 260, 261, 262, 263, 263, 275, 285 Hurricane Irene (2001) 77, 78, 78, 91 Hurricane Isaac (2012) 77, 85, 88, 88

Hurricane Ivan (2004) 90, 124, 125, 126, 131, 161, 193, 194, 233, 283, 284 Hurricane Katrina (2005) 9, 79, 85, 90, 93, 101–5, 107, 112, 117, 124–33, 136–7, 156, 161–2, 191, 195, 195, 205, 212, 221, 232, 233, 236, 237, 257, 258, 261, 282; assisted evacuation 285–7, 287; evacuation direction and control 284; hazard conditions 282–3; preparedness and planning 283–4; public information and warnings 284–5 Hurricane Lili (2002) 88, 101, 107, 112, 124, 125, 126, 130, 131, 132, 133, 136, 156, 162, 285 Hurricane Opal (1995) 91, 115, 116 Hurricane Rita (2005) 9, 67, 76, 79, 85, 103–5, 107, 112, 117, 124–33, 136, 137, 153, 156, 162, 198, 205, 221, 239, 240, 257–61, 264, 285 Hurricane Sandy (also Superstorm Sandy: 2012) 73, 74, 77, 81, 88, 88 hurricane track 30, 32, 33, 33, 62, 64, 64, 82, 102 Hurricane Warning 34, 81–2, 101, 113, 116, 116 Hurricane Watch 34, 81–2, 116, 116 Hurricane Wilma (2005) 90, 91, 112, 167 hurricanes 25–34, 27; hazard analysis 30–1, 31; monitoring and forecasting 32–4, 33, 34 IC see Incident Commander ice/debris dam failures 15 Immediately Dangerous to Life or Health (IDLH) 39 impact area security and reentry 268–9 Incident Commander (IC) 53, 100, 273, 288 Indian Ocean tsunami (2004) 22, 104, 105, 221 information sources, as predictor of evacuation 87 information see public information infrastructure, short term recovery 269–70 ingestion pathway EPZ 43; see also Emergency Planning Zone ingress/egress management 191, 193, 198; see also access control inhalation exposure 49–51

Index Intelligent Transportation System (ITS) 122, 135, 206–8 intended response studies 69, 71, 74, 75, 81–2, 150–1 interface fires 24–5, 282 ITS see Intelligent Transportation System jails 149, 152, 311 Johnson, J.D. 51 Johnson, J.P. 234 Jones, J.A. 109, 154, 156, 227, 228, 286, 290, 291 Kang, J.E. 71, 125 Khattak, A.J. 209, 210, 211 Kim, J. 258, 259 Knoop, V.L. 134 lake levels, fluctuating 15 land use practices 180, 272 Landry, C.E. 258, 259 Lane, L.R. 255 Lazo, J.K. 88 Lee, H-M. 92 legal authority, disaster recovery 274 Lerman, S.R. 157 “lessons learned” assessment 267, 268 levee failure 15, 20, 53 levels of analysis 222–4 Level of Concern (LOC) 36, 39, 262, 307 Li, W. 258 Lindell, M.K. 37, 54, 60, 71, 72, 76, 77, 79, 82, 85, 86, 88, 102, 105, 106, 108, 123, 125, 126, 127, 130, 131, 132, 133, 135, 136, 144, 153, 161, 162, 171, 181, 229, 239, 248, 261 Liu, S. 112, 166, 167 LOC see level of concern local business districts 313 local drainage/surface ponding 15, 29 logistical preparation 107, 113, 183 logistics 121–38; accommodations 124–6; destinations 126–9; disaster recovery 274; driver behavior 122–4; routes 133–8; travel modes 130–3; travel times 129–30 long term reconstruction 266, 271, 271–2

325

macro-level simulation see level of analysis Madson, P. 239 Maghelal, P. 92 Mahmassani, H.S. 185 main stem (riverine) flooding 15, 49, 55, 77 management strategies 177–213; demand management 178–87; future systems 212–13; hazard warnings and public information 181–7; pre-impact hazard education 178–81; supply management 187–206; traffic management 206–12; traffic on arterial routes 201–6, 202; traffic on freeways 189–201, 192, 194, 195, 197, 199, 200 Manual Police Control (MPC) 202, 204, 235 Marks, E. 123 Maximum Envelope of Water (MEOW) 30 Maximum of MEOWs (MOM) 30 McCarty, C. 257 McDonald, M. 210 McKenna, T.J. 37, 57 mental health recovery tasks 272 MEOW see Maximum Envelope of Water Mesa-Arango, R. 125 micro-level simulation see level of analysis microscopic model 157–60; accommodations and destinations 163; departure times 171–2; number of evacuating vehicles 157–60, 159; participation 157–9; travel modes 159, 159–60 Midlarsky, E. 123 migration, permanent 257–9 Mileti, D.S. 125, 161, 182, 210, 228, 239 minimum/most/maximum probable radius 62 Mitchell, S. 185 mobile homes 30, 55, 73, 90–3, 132, 151–2, 153, 185, 273, 284–5 model calibration 240–1 model validation 240–1 MOM see Maximum of MEOWs Mondschein, A. 144 monitoring: floods 19–21, 20; hurricanes 32–4, 33, 34; tsunamis 22–3, 23; wildfires 25 Montz, T. 221 Morrey, M. 239

326 Index Morrow, B.H. 126, 259 Morss, R.E. 88 motels 121, 124–6, 151–2, 160, 163, 229, 248, 270, 312 motivating timely departures 183–4 MPC see Manual Police Control Murray-Tuite, P. 144, 166, 167, 236 Mussa, R.N. 308 Myers, C.A. 258 Naghawi, H. 206 National Hurricane Center (NHC) 30, 32, 33, 34, 77, 79, 81, 82, 83, 101, 116 National Oceanographic and Atmosphere Administration (NOAA) 20, 22, 23, 309 natural hazards see hazards network clearance time 6, 8, 167, 186, 230, 235, 242 NHC see National Hurricane Center NOAA see National Oceanographic and Atmosphere Administration Noltenius, M.S. 167, 168 noncompliance 67, 85; see also compliance, participation rates no-notice evacuations 7, 8, 85, 108, 110, 146–8, 151, 164–5, 183, 193, 231, 246, 282, 309 Notification of an Unusual Event 44 Nozick, L.K. 5 NPP see Nuclear Power Plants NRC see Nuclear Regulatory Commission Nuclear Power Plant Traffic Management Study 293; background/goal/intent 294; demand generation 294–9, 295, 296, 297, 298; findings and conclusions 304–5; results 299–304, 300–1, 302–3; strategies 299 Nuclear Power Plants (NPP) 8, 37, 51, 53, 56, 57, 67, 73, 74–5, 106, 109, 142, 144, 153, 154, 171, 220, 221, 225, 234, 239, 293–305; hazmat releases 41–4, 43 nursing homes 19, 59, 152, 286, 290, 310, 311, 312 Oakland Hills California wildfire (1991) 25 O’Brien, P.W. 125

official evacuation notices, as predictor of evacuation 83–6, 84; compliance with 72–4 Oh, S.S. 258, 259 Oliver-Smith, A. 257 optimization models 4, 5, 186, 225 PADM see Protective Action Decision Model PAG see Protective Action Guide Pais, J. 258, 259 Palmer, I.A. 210 panic 122, 238 PAR see protective action recommendation parallel links 3, 22 parks 313 Parr, S. 205, 235, 236 Parsons Brinckerhoff 311 participation rates 67–93; evacuation behavior 68–74, 70; modeling and simulation 226–7; predictors of evacuation 76–93, 78, 80, 84, 88; producing demand estimates 150–4, 153, 157–9; see also compliance, noncompliance pass through traffic 198, 298 Paxson, C. 258 Peacock, W.G. 87, 92 Peek, L. 182 perceived danger/risk: as predictor of evacuation 76–80, 78; in reentry planning 276 permanent housing 270 permanent migration 257–9 Perry, R.W. 1, 77, 104, 105, 106, 108, 114, 123, 131, 171 personal vehicles: estimating demand 155–6; as evacuation travel mode 130–2 personalization 210 personnel restrictions 256 pets 77, 91–2, 109, 112 phased evacuation 4, 184–7 plume inhalation EPZ see Emergency Planning Zone police manual traffic control (MTC) 204–5, 235–6 Polivka, A.E. 258, 259 Polydoropoulou, A. 210 postimpact evacuations 7 Prater, C.S. 54, 60, 71, 72, 76, 79, 85, 86, 102, 105, 125, 126, 132, 135, 136, 153, 171, 261 predictors 76–93; demographic characteristics 92–3; environmental

Index and social cues 86; experience 88, 88–90; forecasts and warnings 80, 80–3; housing type 90–1; information sources 87–8; Official Evacuation Notices 83–6, 84; perceived risk 76–80, 78; pets 91–2 pregnant women 9, 57, 74, 83, 181 preliminary damage assessment 267–8 preparation activities 163–70, 165; behavioral modeling 168–70, 169; family reunification 165–7; hazards with substantial forewarning 167; prior findings on 167–8 preparation times 106 110–11, 106–13 preparation trips 109, 146–7, 164–5, 167–8 preparedness 91, 168, 178, 180, 220, 283–4, 288–9 prescriptive models 224 Profio, A.E. 51, 52 protective action 46–65; decision making 53–65, 54; evacuation as 52–3; geographical zones for 143–5; hazmat transportation evacuation planning (case study) 307–11; overview 46, 52–3; sheltering in-place 46–52 Protective Action Areas 144, 180 Protective Action Decision Model (PADM) 68, 69, 70, 71, 123, 135, 209, 210, 211 Protective Action Guide (PAG) 55, 56, 57 protective action recommendation (PAR) 53–65, 54; selection 55–61, 56, 56, 57, 58, 59, 60, 61; timing 61–5, 62, 63, 64 protective action zones 58, 144–5, 151, 154, 162, 247 psychological preparation 113, 183 public health recovery tasks 272 public information 87–8; disaster recovery 273–4; hazmat transportation evacuation planning (case study) 309–10; Hurricane Katrina (case study) 284–5; management strategies 181–7; for returnees 275–6; sources for evacuees 262–5, 263, 264; Southern California Wildfires (case study) 291–2; statements 23; on travel 207–8 public transit: estimating demand 156; as evacuation travel mode

327

133; management strategies 205–6; modeling and simulation 236–7 Quarantelli, E.L. 270 Radwan, A. 170, 185, 186 rain bands 26, 28, 32 ramp closures 198, 233–4 rapid assessment 267 reach 17 receiver characteristics 68 reception centers 147, 156, 201 reconstruction, long term 271, 271–2 recovery see disaster recovery red flag warning 25 reentry: evacuee concerns 260–2; impact area security 268–9; information sources 262–5, 263, 264; plan compliance 259–60; planning principles 275–7; Southern California Wildfires 2007 (case study) 292 religion-based facilities 313 repair permitting 270–1, 274 right front quadrant 26 Risk Areas 30, 31, 58, 58, 62, 63 risk communication program see hazard education program risk counties 122 risk see perceived danger/risk riverine (main stem) flooding 15, 49, 55, 77 RMP*Comp 39 Robinson, R.M. 210, 211, 237 Rogers, G.O. 51, 101, 115, 309 Rouse, C.E. 258 routes/roadways 133–8; arterial 201–6; closures 196–7, 197, 233–4; guidance compliance 209–12; modeling and simulation 230–1, 233–4; see also freeways Ruginski, I.T. 82, 83 Safe Areas 58, 58 Saffir-Simpson categories 26, 27, 30, 34, 63, 78, 103 Sbayti, H. 185 scenarios 242–7; hazards 242–3; need for multiple 247–9, 249; timing 243–6, 244, 245; transportation system 246–7; need for multiple scenarios 247–9, 249 schools 19, 46, 74, 108–9, 311, 312 Schwarz, N. 71

328 Index Schweitzer, L. 166, 167 Sea, Lake, and Overland Surges from Hurricanes (hurricane surge model) (SLOSH) 30, 73 security, impact area 268–9 Sell, T.C. 238, 239 sensitivity analysis 160, 242, 248–9 September 11th 2001 terror attacks 9, 221 sequenced evacuation see phased evacuation serial links 3 shadow evacuation 8, 67, 74–6 Shapira, Z. 102 Shaw, W.D. 258 Sheffi, Y. 220 shelter, temporary 270; see also accommodations, housing sheltering in-place: external radiation exposure 51–2; hazmat transportation evacuation planning (case study) 310–11; inhalation exposure 49–51; overview 46 Sherali, H.D. 123 shielding 46, 51, 55 short notice events 7, 7–8, 161 short term recovery 266, 268–71 Siebeneck, L.K. 127, 130, 259, 260, 261, 262, 265, 275 signal timing modification 203–4, 235 Silver, R.C. 72, 87 simulation models see traffic modeling and simulation Site Area Emergency 44 site assessments 268 Sivasailam, D. 170 SLOSH see Sea, Lake, and Overland Surges from Hurricanes (hurricane surge model) Smith, S.K. 257 social cues, as predictor of evacuation 86 Sorensen, J.H. 115, 125, 210, 239, 309 Southern California Wildfires (2007) 9, 221, 287; assisted evacuation 290; communication and public information 291–2; evacuation direction and control 289–90; preparedness and planning 288–9; reentry 292 Spansel, K. 221, 241 spatial restrictions 256 special facilities 22, 59–60, 149, 310–13 specificity analysis 249 spill back effects 147

stage 15 staged evacuations see phased evacuations Stallings, R.A. 255, 268 stay and defend 180 Stopher, P. 122 storm surge 26, 28–30, 49, 55, 62, 73, 77, 79, 83, 90, 143, 145, 153 Storr, V.H. 258 strategies see management strategies Sullivan, R.L. 109 Superstorm Sandy (Hurricane Sandy: 2012) 73, 74, 77, 81, 88, 88 Supprasi, A. 108 surface ponding/local drainage 15, 29 surveys, actual response 69, 72, 150; intended response/behavioural expectation 69, 71–5, 81–2, 150–1, 157, 162, 166–77, 169 synthetic households 158, 163, 166, 172 system monitoring 206–7 System Optimal (SO) model 134, 226, 230 Taylor, B. 144 TAZ see Traffic Analysis Zone Technical Guidance for Hazards Analysis 41, 307 technological hazards see hazards temporal restrictions 256 temporary business operation 270 temporary shelter/housing 270 termination 255–77; disaster recovery planning 265–75, 266, 271; evacuee information sources 262–5, 263, 264; permanent migration 257–9; see also reentry Thompson, R.R. 72, 87, 89, 91, 92 Threshold Limit Value (TLV) 39 time (exposure duration) 46 time (travel duration) 129–30 timing 100–17; authorities’ decision times 100–2, 101; departure times 113–17, 115, 116, 117, 229–30; modeling and simulation 229–30; motivating timely departures 183–4; preparation times 106, 110–11, 106–13; protective action recommendations (PAR) 61–5, 62, 63, 64; scenario modeling and simulation 243–6, 244, 245; warning dissemination times 102–5, 103, 104 TLV see Threshold Limit Value topography 22, 24, 29 tornado diagram 248–9, 249

Index total exclusion 256 tour 112 tourists 149, 151, 155, 184–5, 243–4, 255 towed vehicles 156–7 toxic chemical releases see hazardous materials (hazmat) releases toxins 38–9 Traffic Analysis Zone (TAZ) 145–6, 151, 158, 228 Traffic Engineering Handbook 188 traffic incidents 6–7, 122, 207–8, 237–8, 246–7 traffic management: on arterial routes 201–6, 202; in emergency operations centers 206–12; on freeways 189–201, 192, 194, 195, 197, 199, 200 traffic modeling and simulation 219–49; background 220–1; incidents 237–40; key variables and assumptions 224–40; levels of analysis 222, 222–4; megaregions 241–2; need for multiple scenarios 247–9, 249; refinements and validation 240–1; scenario development for 242–7, 244, 245 Trainor, J. 144, 145 transient persons 52, 100, 149, 151–6, 153, 158, 243, 297, 298, 310, 312 transit counties 122 transit see public transit transportation analysis, hazardous materials (hazmat) releases 40, 40–1 transportation planning models 148–9 transportation system scenarios 246–7 travel modes 130–3; carpooling 132; modeling and simulation 227–8; personal vehicles 130–2; producing demand estimates 159, 159–60; public transit 133 travel times 129–30 trip chain 151, 167 tropical cyclones 25, 79 Tropical Storm Allison (2001) 76 Tropical Storm Wind 25, 28, 34, 58, 62–4, 64, 81, 103, 186 Trumbo, C. 88 Tsirimpa, A. 210 tsunami advisory 23 tsunami warning 23

329

Tsunami Warning Centers (TWC) 22–3, 23 tsunami watch 23 tsunamis 21–3; hazard analysis 22; monitoring and forecasting 22–3, 23 Tu, H. 123 turn restrictions 202, 202–3, 234–5 turnover time 50 TWC see Tsunami Warning Centers Tweedie, S.W. 171 Ullman, B.R. 199 Urbanik, T. 52, 53, 100, 220 US Army Corps of Engineers (USACE) 230 User Equilibrium (UE) model 134, 226, 230, 233 V2I see vehicle to infrastructure communication V2V see vehicle to vehicle communication variable message signs (VMS) 135, 182, 189, 208 vehicle demand estimates 154–5; calculation methods 149–60; see also personal vehicles vehicle to infrastructure communication (V2I) 148 vehicle-to-vehicle communication (V2V) 148, 212 vertical proximity 24; see also fuels, wildfire victims’ needs assessment 267 VMS see Variable Message Signs Vulnerable Zones (VZ) 36, 39–41, 228, 307–8, 310–11, 313 Walton, F. 109, 154 warnings 80, 80–3; dissemination times 102–5, 103, 104; hazmat transportation evacuation planning (case study) 309–10; Hurricane Katrina (case study) 284–5; strategic management of 181–7 WEA see Wireless Emergency Alert system weather 24 Weerakkody, S.D. 239 White, G.F. 86 Whitehead, J.C. 126 wildfires 24–5; see also Southern California Wildfires (2007)

330 Index wildland fires 24 Wilmot, C.G. 72, 128, 129 Wilson, D.J. 50, 51 windshield surveys 268 Wireless Emergency Alert (WEA) system 310 Witzig, W.F. 239 Wolshon, B. 144, 153, 205, 206, 221, 233, 235, 236, 238, 241, 255

Wu, H-C. 76, 82, 83, 125, 126, 127, 131, 132, 136, 153, 168 Yin, W. 126, 127, 131, 167, 168 Zhang, Y. 136, 137 Zhang, Z. 221, 241