Recent Trends in Transportation Infrastructure, Volume 2: Select Proceedings of TIPCE 2022 981992555X, 9789819925551

Table of contents :
Contents
About the Editors
Transportation Infrastructure: Planning and Design
Estimation of Transit OD Matrix Using Boarding and Alighting Data: A Case Study of Haridwar-Rishikesh Metro Corridor
1 Introduction
2 Methodology
3 Methodology Testing
4 Case Study
5 Conclusion
References
Measuring Perceived Satisfaction of Choice Riders Towards Factors Influencing Accessibility to Metro Stations: An Evidence from Delhi
1 Introduction
2 Methods
3 Survey Design, Data Collection and Database Preparation
4 Data Analysis, Results and Discussion
4.1 RIDIT Analysis
4.2 GRA
4.3 TOPSIS Analysis
4.4 Comparative Attribute Rankings of Three MADM Techniques
5 Conclusion
References
Impact Assessment of Ring Road in Haridwar and National Highway Connectivity on Haridwar-Rishikesh Metro Ridership
1 Introduction
2 Study Area and Data
2.1 Overview of the Study Area
2.2 Availability and Source of the Data
2.3 Modal Share
2.4 Utility Parameters
3 Methodology
3.1 Calibration for the Base Year (2017)
3.2 Calibration for the Year 2021
3.3 Modeling for the Year 2021 with Ring Road and NH Up-Gradation
3.4 Estimation of Metro Ridership for the Year 2021 with Ring Road and NH Up-Gradation
3.5 Findings from the Survey
4 Results
4.1 Impact of Ring Road on Change in Metro Share
4.2 Impact of NH Up-Gradation Between Haridwar-Rishikesh on Change in Metro Share
5 Conclusion
References
Passenger Delay and Journey Time Reliability Analysis of Ferry Transport Across Waterway: Case Study of National Waterway-1, India
1 Introduction
2 Literature Review
2.1 Literature Review Related to Passenger Delay
2.2 Literature Review Related to Travel Time Reliability
3 Methodology
3.1 Data Collection
3.2 Data Extraction
3.3 Descriptive Statistics of the Extracted Parameters
4 Analysis and Results
4.1 General Observations
4.2 Reliability and Relevant Indices from Observed Data
4.3 Improvement of Total Journey Time (by Varying Frequency)
4.4 Inferences from the Change in Frequency
5 Concluding Remarks
Appendix
References
Is CAWI-Based Transit Passengers’ Satisfaction Survey Feasible in India?—Insights During COVID-19
1 Introduction
2 Survey Methodology
3 Insights from CAWI Survey Methods
3.1 Survey Distribution and Response Losses
3.2 Why High Non-responses?
3.3 Completeness and Biasedness in Responses
3.4 Modelling the Non-responses
4 Conclusions
References
Joint Versus Standalone Estimation of First/Last Mile Mode Choice Utilising Revealed and Stated Preference Datasets
1 Introduction
2 Combined Estimation with RP and SP Data
3 Empirical Analysis
3.1 Data Collection and Survey Design
3.2 Modelling Results
4 Conclusions
References
Do Service Attributes Influence the Mode Choice of Access-Egress Trips in the Context of Delhi Metro?
1 Introduction
2 Study Area
3 Methodology
4 Survey Design and Data Collection
4.1 Attitudinal Variables
4.2 Interdependence of Attitudinal Variables
5 Model Development
6 Results
7 Discussion
8 Conclusion
References
Assessment of Gender and Income-Based Inequality in Travel Behavior of Vadodara City
1 Introduction Multinomial Logit
2 Data
3 Data Analysis
4 Model Development
5 Conclusion
References
Traffic Impact Study of an Integrated Township and Formulation of Improvement Measures—A Case Study of Technocity in Thiruvananthapuram
1 Introduction
2 Objectives
3 Literature Review
4 Study Area
5 Methodology
6 Data Collection and Analysis
7 Travel Demand Modeling
7.1 Estimation of Trips Generated
7.2 Modal Split
7.3 Traffic Assignment
8 The “With” and “Without” Scenario
9 Improvement Proposals
10 Conclusion
References
Rollover Stability Analysis of Trucks-Effect of Curve Geometry and Operating Speed
1 Introduction
2 Methodology
2.1 Development of Vehicle Model and Validation
2.2 Sensitivity Analysis
3 Results and Discussion
3.1 Effect of Geometry on Rollover Stability of Vehicles
3.2 Relationship Between Radius of Curve and Steering Wheel Angle
3.3 Lateral Acceleration Prediction Models
4 Conclusions
References
Analysing Willingness to Pay and Attitude Towards Safety for Indian Motorcyclists
1 Introduction
1.1 Motivation
2 Methodology
2.1 Questionnaire Design
2.2 Data Collection
3 Result
4 Conclusion
References
Use of Advanced Techniques for Functional Evaluation of Pavements: A Review and a Pilot Study
1 Introduction
2 Review of Studies on the Use of Advanced Techniques for Functional Evaluation of Pavements
2.1 Studies on Roughness Evaluation
2.2 Studies on Surface Distress Evaluation
2.3 Studies on Skid Resistance Evaluation
3 Results of Pilot Study
4 Concluding Remarks
References
A Machine Learning-Based Active Learning Framework to Capture Risk and Uncertainty in Transportation and Construction Scheduling
1 Introduction
2 Literature Review
3 Methodology
3.1 Fuzzy Mathematical Formulation of Risk and Uncertainty
3.2 Modeling Uncertainty with ModAL
3.3 An Example with modAL to Model Activity Duration with Uncertainty
4 Example and Discussion
4.1 Model Limitations
5 Conclusions and Future Works
References
Earthwork Logistics Optimization in Road Construction Project
1 Introduction
2 Literature Review
3 Research Methodology
3.1 Problem Definition
3.2 Model for Distance-Volume Minimization
4 Results and Discussions
5 Conclusions
References
Urban and Rural Transportation
Reduction of Vehicular Emission at Urban Road Junctions Through Traffic Interventions
1 Introduction
2 Literature Review
3 Study Area
4 Methodology
4.1 Modeling of Signalized Intersection
4.2 Emission Estimation
4.3 Alternative Scenarios
5 Results and Discussions
5.1 Effect of Traffic Characteristics on Emission
5.2 Effect of Intersection Geometry and Control on Emission
5.3 Summary of Emission Model for Signalized Intersection
6 Conclusions
References
Development of PCU Model for Unsignalised Intersection: A Case Study of Ranchi City
1 Introduction
2 Background Literature
3 Data Collection and Extraction
4 Methodology
5 Results and Discussions
6 Conclusion
References
Modeling Pedestrian Waiting Time Delay at Signalized Midblock Crosswalks Under Non Uniform Arrivals and Non Compliance Behavior
1 Introduction
2 Methodology and Data Collection
3 Pedestrian Field Waiting Delay
4 Pedestrian Waiting Delay Model
5 Conclusions
References
Assessment of Adherence Level to Helmet Usage on Varying Roads in Delhi
1 Preamble
2 Objectives and Scope of Study
3 Literature Review
4 Data Collection Process
4.1 Study Location
4.2 Study Details
5 Binary Logistic Regression (BLR)
5.1 Purpose of Implementing Binary Logistic Regression
6 Level of Helmet Usage Across Varying Functional Classification of Roads
7 Concluding Remarks
References
Assessing the Impact of Underground Utility Works on Road Traffic and Users: A Study from an Indian City
1 Background
2 Methodology
3 Site Description and Data Collection
3.1 Site Description
3.2 Data Collection
3.3 VISSIM Simulations
4 Description of the Model Used
4.1 Vehicle Technology and Emissions Rates
4.2 Driving Behaviour and Vehicle Activity
4.3 Vehicle Fleet Distribution
5 Results
5.1 Average Vehicular Speed
5.2 Exhaust Emissions
5.3 Fuel Consumption
6 Discussions and Conclusions
References
Characteristics of e-rickshaw Dominated Mixed-Mode Traffic in Suburban Arterial Corridors
1 Introduction
2 Objectives and Study Scopes
3 Study Design
3.1 Model and Method
3.2 Field Data
4 Application and Interpretation of Results
4.1 Distributional Characteristics of Flow Parameters
4.2 Macroscopic Relations of Traffic Parameters Using Complex Models
5 Conclusions
References
Qualitative and Quantitative Evaluation of Urban Car Parking System: A Case Study of Bhopal City
1 Introduction
1.1 Parking Statistics
1.2 Objectives
2 Literature Review
3 Study Methodology
4 Study Area and Data Collection
4.1 Study Area
4.2 Data Collection
5 Data Analysis
5.1 Quantitative Data Analysis
5.2 Qualitative Data Analysis
6 Result and Conclusion
References
Intercity Transportation
Estimation of Risk Exposure Index for Road Network in Landslide-Prone Areas
1 Introduction
2 Scope and Objective
3 Methodology
3.1 Data Collection
3.2 Risk Assessment
3.3 Risk Exposure Index Estimation
4 Results and Discussions
4.1 Risk Assessment
4.2 Risk Exposure Index
5 Conclusion
References
Mode Choice Behaviour of Textile Shippers in India
1 Introduction
2 Literature Review
3 Questionnaire Design and Data Collection
4 Data Summary
5 Qualitative Factors of Preference Ratings
6 Heterogeneity Among Shippers
7 Mode Choice Modelling
8 Conclusion
References
Optimal Coal Transport Mode Choice for Near Plant Coal Mine by Using Integrated Fuzzy Analytical Hierarchical Process and Fuzzy Goal Programming Model
1 Introduction
2 Questionnaire and Data Collection
3 AHP Methodology and FGP Model
3.1 Use of Fuzzy Comparison Technique for Judgement in Decision-Making Phase
3.2 Triangulation Fuzzy Number (TFN)
3.3 Fuzzy Prioritization Approach
3.4 Solving the Fuzzy Prioritization Problem and FGP Model
4 Model Development Using Fuzzy AHP Methodology and FGP Model
5 Results and Discussion
6 Conclusion
References
Economic Analysis of FasTag on Highway Toll Collection
1 Introduction
2 Literature Review
3 Methodology
3.1 Site Selection
3.2 Data Collection
3.3 Service Time Analysis
3.4 Probability of Payment of Toll
3.5 Revenue Forecast
4 Data Analysis
4.1 Service Time
4.2 Probability of Payment by a Vehicle at Toll Plaza
4.3 Revenue Forecast
5 Results
6 Discussion
7 Conclusion
References
Sustainable Transportation
Factors Affecting Public Transportation Usage Rate: Geographically Weighted Regression
1 Introduction
2 Past Research
3 Methodology
3.1 Ordinary Least Square Regression
3.2 Geographically Weighted Regression
3.3 Kernels
4 Data and Analysis
4.1 OLS Results
4.2 GWR Results
4.3 Interpretation of the GWR Model
4.4 Comparisons of OLS and GWR Model
5 Conclusion
References
Evaluation of Active Transport Systems: A Glance at Recent Studies
1 Introduction
2 Procedures of Literature Searches
3 Evaluation Approaches of Active Transport
3.1 Global Perspective
3.2 Perspective with Reference to Cities in India
4 Performance and Sustainability Assessment Framework
4.1 Sustainability Assessment Frameworks: A Review
4.2 Factors Associated with Active Transport
5 Discussion, and Proposal of a Conceptual Framework
6 Conclusion
References
Study of Factors Affecting Pedestrian Movement in Mass Leisure Gatherings
1 Introduction
2 Literature Review
3 Study Area and Data Collection
4 Methodology and Results
5 Conclusions and Future Scope
References
What’s Hindering EV Mass Adoption in Urban India: From Potential User’s Perspective
1 Introduction
1.1 Problem Statement
1.2 Research Aim
2 Literature Review and Identification of Barriers
3 Methodology
3.1 Logistic Regression Model
3.2 Performance Evaluation
4 Case Study
4.1 Study Area
4.2 Data Collection
5 Results and Discussion
5.1 Model
5.2 Performance Evaluation of the Model
6 Conclusion
References
Measuring Users’ Perceived Importance Towards Factors Affecting Their Willingness to Use Bicycle: Patna as a Case Study
1 Introduction
2 Study Area
3 Methodology
3.1 Identification of Factors from Literature
3.2 Design of Questionnaire and Data Collection
3.3 Method Used for Analysis (RIDIT)
4 Data Analysis and Results
4.1 Socio-economic Profile and Trip Characteristics
4.2 Analysis of Importance Rating Data
4.3 Results Based on RIDIT Analysis
5 Conclusion
References

Citation preview

Lecture Notes in Civil Engineering

Amit Agarwal S. Velmurugan Akhilesh Kumar Maurya   Editors

Recent Trends in Transportation Infrastructure, Volume 2 Select Proceedings of TIPCE 2022

Lecture Notes in Civil Engineering Volume 347

Series Editors Marco di Prisco, Politecnico di Milano, Milano, Italy Sheng-Hong Chen, School of Water Resources and Hydropower Engineering, Wuhan University, Wuhan, China Ioannis Vayas, Institute of Steel Structures, National Technical University of Athens, Athens, Greece Sanjay Kumar Shukla, School of Engineering, Edith Cowan University, Joondalup, WA, Australia Anuj Sharma, Iowa State University, Ames, IA, USA Nagesh Kumar, Department of Civil Engineering, Indian Institute of Science Bangalore, Bengaluru, Karnataka, India Chien Ming Wang, School of Civil Engineering, The University of Queensland, Brisbane, QLD, Australia

Lecture Notes in Civil Engineering (LNCE) publishes the latest developments in Civil Engineering—quickly, informally and in top quality. Though original research reported in proceedings and post-proceedings represents the core of LNCE, edited volumes of exceptionally high quality and interest may also be considered for publication. Volumes published in LNCE embrace all aspects and subfields of, as well as new challenges in, Civil Engineering. Topics in the series include: • • • • • • • • • • • • • • •

Construction and Structural Mechanics Building Materials Concrete, Steel and Timber Structures Geotechnical Engineering Earthquake Engineering Coastal Engineering Ocean and Offshore Engineering; Ships and Floating Structures Hydraulics, Hydrology and Water Resources Engineering Environmental Engineering and Sustainability Structural Health and Monitoring Surveying and Geographical Information Systems Indoor Environments Transportation and Traffic Risk Analysis Safety and Security

To submit a proposal or request further information, please contact the appropriate Springer Editor: – Pierpaolo Riva at [email protected] (Europe and Americas); – Swati Meherishi at [email protected] (Asia—except China, Australia, and New Zealand); – Wayne Hu at [email protected] (China). All books in the series now indexed by Scopus and EI Compendex database!

Amit Agarwal · S. Velmurugan · Akhilesh Kumar Maurya Editors

Recent Trends in Transportation Infrastructure, Volume 2 Select Proceedings of TIPCE 2022

Editors Amit Agarwal Department of Civil Engineering Indian Institute of Technology Roorkee Roorkee, Uttarakhand, India

S. Velmurugan CSIR-Central Road Research Institute Delhi, India

Akhilesh Kumar Maurya Department Civil Engineering Indian Institute of Technology Guwahati Guwahati, Assam, India

ISSN 2366-2557 ISSN 2366-2565 (electronic) Lecture Notes in Civil Engineering ISBN 978-981-99-2555-1 ISBN 978-981-99-2556-8 (eBook) https://doi.org/10.1007/978-981-99-2556-8 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Contents

Transportation Infrastructure: Planning and Design Estimation of Transit OD Matrix Using Boarding and Alighting Data: A Case Study of Haridwar-Rishikesh Metro Corridor . . . . . . . . . . . Samsoor Mohmmand, Rupam Fedujwar, and Amit Agarwal Measuring Perceived Satisfaction of Choice Riders Towards Factors Influencing Accessibility to Metro Stations: An Evidence from Delhi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Manaswinee Kar, Shubhajit Sadhukhan, and Manoranjan Parida Impact Assessment of Ring Road in Haridwar and National Highway Connectivity on Haridwar-Rishikesh Metro Ridership . . . . . . . Nidhi Kathait, Pushpa Choudhary, Amit Agarwal, and Rajat Rastogi Passenger Delay and Journey Time Reliability Analysis of Ferry Transport Across Waterway: Case Study of National Waterway-1, India . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Debabrota Das, Anuj Kishor Budhkar, and Ayush Patel

3

15

27

41

Is CAWI-Based Transit Passengers’ Satisfaction Survey Feasible in India?—Insights During COVID-19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Vishwajeet Kishore Verma and Rajat Rastogi

53

Joint Versus Standalone Estimation of First/Last Mile Mode Choice Utilising Revealed and Stated Preference Datasets . . . . . . . . . . . . . B. S. Manoj and Arkopal Kishore Goswami

67

Do Service Attributes Influence the Mode Choice of Access-Egress Trips in the Context of Delhi Metro? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rashmi Choudhary, Jogendra Kumar Nayak, and Manoranjan Parida

81

Assessment of Gender and Income-Based Inequality in Travel Behavior of Vadodara City . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pankaj Prajapati and Reshma Khan

95

v

vi

Contents

Traffic Impact Study of an Integrated Township and Formulation of Improvement Measures—A Case Study of Technocity in Thiruvananthapuram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 V. S. Sanjay Kumar, P. N. Salini, Ebin Sam, and S. Akshara Rollover Stability Analysis of Trucks-Effect of Curve Geometry and Operating Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Y. K. Remya, Jacob Anitha, and E. A. Subaida Analysing Willingness to Pay and Attitude Towards Safety for Indian Motorcyclists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Anand Kumar Saurav, Hillol Chakravarty, and Ranja Bandyopadhyaya Use of Advanced Techniques for Functional Evaluation of Pavements: A Review and a Pilot Study . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 N. H. Riyaz Khan and S. Vasantha Kumar A Machine Learning-Based Active Learning Framework to Capture Risk and Uncertainty in Transportation and Construction Scheduling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 Manoj K. Jha, Nicodeme Wanko, and Anil Kumar Bachu Earthwork Logistics Optimization in Road Construction Project . . . . . . . 179 Furqan A. Bhat, Debopam Roy, Prasanta K. Sahu, Sanjay C. Choudhari, and A. Bahurudeen Urban and Rural Transportation Reduction of Vehicular Emission at Urban Road Junctions Through Traffic Interventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 Sumaiya Rahman and Mithun Mohan Development of PCU Model for Unsignalised Intersection: A Case Study of Ranchi City . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 Aarohi Kumar Munshi and Ashish Kumar Patnaik Modeling Pedestrian Waiting Time Delay at Signalized Midblock Crosswalks Under Non Uniform Arrivals and Non Compliance Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 Sandeep Manthirikul and Udit Jain Assessment of Adherence Level to Helmet Usage on Varying Roads in Delhi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 Kanishk Singh, S. Velmurugan, and Nishit Patel Assessing the Impact of Underground Utility Works on Road Traffic and Users: A Study from an Indian City . . . . . . . . . . . . . . . . . . . . . . 243 Ashish Verma, P. Anbazhagan, Sai Kiran Mayakuntla, and Furqan A. Bhat

Contents

vii

Characteristics of e-rickshaw Dominated Mixed-Mode Traffic in Suburban Arterial Corridors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 Pankaj Kumar, Sabyasachi Mondal, Pritam Saha, and Sudip K. Roy Qualitative and Quantitative Evaluation of Urban Car Parking System: A Case Study of Bhopal City . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 Hrishabh Chouhan, Pritikana Das, and Dungar Singh Intercity Transportation Estimation of Risk Exposure Index for Road Network in Landslide-Prone Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291 P. N. Salini, P. Rahul, U. Salini, and Samson Mathew Mode Choice Behaviour of Textile Shippers in India . . . . . . . . . . . . . . . . . . 305 V. Ansu and M. V. L. R. Anjaneyulu Optimal Coal Transport Mode Choice for Near Plant Coal Mine by Using Integrated Fuzzy Analytical Hierarchical Process and Fuzzy Goal Programming Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317 Sudeep Kumar Mishra and Sunny Deol Guzzarlapudi Economic Analysis of FasTag on Highway Toll Collection . . . . . . . . . . . . . 333 Anand Mishra and Abhisek Mudgal Sustainable Transportation Factors Affecting Public Transportation Usage Rate: Geographically Weighted Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 Pankaj Prajapati and Divyesh Abhani Evaluation of Active Transport Systems: A Glance at Recent Studies . . . 359 Sakshi Sharma, Rajat Rastogi, and Debasis Basu Study of Factors Affecting Pedestrian Movement in Mass Leisure Gatherings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373 Durba Kundu, Anuj Budhkar, Angshuman Pandit, and Bimalendu Mandal What’s Hindering EV Mass Adoption in Urban India: From Potential User’s Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385 Shaurya Mall and Ramesh Anbanandam Measuring Users’ Perceived Importance Towards Factors Affecting Their Willingness to Use Bicycle: Patna as a Case Study . . . . . 397 Manan Monga, Shubhajit Sadhukhan, and Aditya Manish Pitale

About the Editors

Dr. Amit Agarwal is currently an Assistant Professor at the Department of Civil Engineering and Joint Faculty at Mehta Family School of Data Science and Artificial Intelligence, IIT Roorkee. Before joining IIT Roorkee, he has worked at Robert Bosch, Singapore as a senior research scientist. He got his Ph.D. from TU Berlin, Germany, in 2017 and M Tech from IIT Delhi, in 2012 and B. Tech from MNIT Jaipur in 2009. His research interests include Intelligent Transportation Systems, MultiAgent Transport Simulation framework (MATSim), Crowdsourced Data, Sustainable Transport, Air Pollution Exposure, Travel Demand Modeling. He is working on many research and consultancy projects funded by IAHE (MoRTH), SERB, iHub DivyaSampark (TIH) Roorkee, UKMRC, etc. His team was a winner in the Smart Move Innovative Urban Mobility Challenge, 2021. He received Outstanding Young Faculty Award, DAAD Scholarship for Master Project, Ph.D. and NBCC Award of Excellence. Dr. S. Velmurugan is currently a Chief Scientist at CSIR-Central Road Research Institute (CRRI), New Delhi. He obtained his Ph.D. from the Indian Institute of Technology (IIT) Bombay, in 1995. He is an expert in the area of traffic engineering, transportation engineering, road safety, transport economics, and transport environment. Prof. Akhilesh Kumar Maurya is an accomplished civil engineering Professor at Indian Institute of Technology (IIT) Guwahati, with Ph.D. from IIT Kanpur and MTech in CAD from IIT Roorkee. His research interests encompass traffic flow modeling, road safety audit, accident analysis, and traffic data collection, reflected in 100+ papers published in international journals and conferences. He has received the DAAD fellowship and is a certified “Road Safety Auditor” by the International Road Federation (India) and Australian Road Research Board. Maurya is the President of “Transportation Research Group of India (TRG)”, a member of WCTRS, and a life member of “Indian Roads Congress.”

ix

Transportation Infrastructure: Planning and Design

Estimation of Transit OD Matrix Using Boarding and Alighting Data: A Case Study of Haridwar-Rishikesh Metro Corridor Samsoor Mohmmand, Rupam Fedujwar, and Amit Agarwal

Abstract Transit OD matrix is fundamental to service planning and operation. Traditionally, transit OD matrices are generated through on-board personal interview surveys, albeit highly expensive, time-consuming, and labor-intensive. Therefore, utilizing the readily available data in the estimation process has been the subject of many research for the past few decades. This study focuses on the estimation of transit OD matrix using boarding and alighting passenger counts, which can further be extended to smart-card data. Four different methods from literature are explored, namely, IPF, Markov Model, Uncertainty Maximization, and Compressed Sensing. Furthermore, a new model for route-level transit OD estimation is introduced. Unlike compressed sensing, the proposed model employs l∞ norm regularizer instead of Euclidean norm and uses spot information about the critical cell in each direction. The proposed model is found to outperform the existing methods when tested for RMSE value and trip length distributions for true OD matrices. In the most favorable scenario, the RMSE value was found to be 18 compared to 145 and 95 for compressed sensing and IPF, respectively. Lastly, the trip matrix for the HaridwarRishikesh metro corridor was estimated using the above methods and the results were compared based on their similarities and differences. Keywords OD matrix · Boarding & alighting · IPF · Markov · Entropy · Compressed sensing

S. Mohmmand (B) · R. Fedujwar · A. Agarwal Department of Civil Engineering, Indian Institute of Technology Roorkee, Roorkee, Haridwar, India e-mail: [email protected] R. Fedujwar e-mail: [email protected] A. Agarwal e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_1

3

4

S. Mohmmand et al.

1 Introduction Background. Transit OD matrix can be crucial in route designing, route modification (extending, splitting, or merging), transit and crew scheduling, ridership forecasting, vehicle composition, and so on (Mishalani et al. 2016). They can be generated at route or network levels. At the route-level, no transfers are taken into account and trips are assumed to be independent of each other, whereas working at the network level necessitates dealing with linked trips, which makes the estimation process more complicated (Luo et al. 2017). The difficulties involved in the conventional ways of OD matrix generation have forced the researchers to seek feasible alternatives. One such alternative is utilizing automatic data collection systems such as Automatic Fare Collection (AFC), Automatic Passenger Count (APC), and Automatic Vehicle Location (AVL). With such systems in place, it is more convenient to collect large-scale information about passengers’ mobility over longer time periods (He and Trépanier 2015). Large sample sizes collected from automatic systems help in achieving more comprehensive trip matrices (Farzin 2008; Luo et al. 2017). The problem of OD estimation using boarding and alighting data is ill-posed in nature. That is, the number of unknowns (OD flows) to be determined is far higher than the number of equations (constraints) available. Thus, there is no unique solution to the problem. A number of approaches can be found in the literature that address such problems. These methods can be classified into balancing, Bayesian, and optimization categories (Kumar et al. 2019), which are briefed below. Balancing methods. In these methods, the rows and columns of the base matrix are balanced iteratively in such a way that their totals match the observed boarding and alighting values as closely as possible. Mosche et al. (1985) reviewed the alternative methods for estimating route-level trip tables. Their findings showed that in the presence of sample on-board survey and deterministic ride-check data, Iterative Proportional Fitting (IPF), Constrained Generalized Least Square (CGLS), and Constrained Maximum Likelihood Estimate (CMLE) methods produce similar results. Nevertheless, IPF is preferred over the other two for its simplicity. Simon and Furth (Simon and Furth 1985) showed Tsygalnitsky method to be accurate enough for simple and complex routes. However, it cannot be substituted for the general OD surveys, as these surveys can generate data which is nowhere else available (e.g., trip purpose, access/egress mode, socio-demographics, etc.). IPF with null seed (i.e., comprised of ones and zeros only) produces the same results as that of Tsygalnitsky approach (Furth and Navick 1992). They also found that IPF-null is plausible for short and less interfered routes. McCord et al. (2010) and Mishalani et al. (2011) indicated that survey-based IPF not only results in higher accuracy than IPF-null, but also better than the results of the on-board survey alone. They further stated that in absence of prior information, IPF-null can be as good as or better than on-board survey, thanks to the large sample size. Bayesian methods. Maher (1983) proposed a Bayesian inference method which uses link counts as input. The proposed model has superiority over Entropy Maximization (EM) and information minimization approaches for the high emphasis it

Estimation of Transit OD Matrix Using Boarding and Alighting Data …

5

lays on the prior information. Hazelton (2010) developed a computationally efficient Markov Chain Monte Carlo (MCMC) sampler, which eliminates the need for enumeration of the candidates. Candidates can rather be sampled directly from a set of feasible OD trip vectors. Li (2009) developed a Markov model and used on/ off counts for the estimation purpose. This approach eliminates the non-structural zero problem (i.e., the presence of zeros due to erroneous/ incomplete input data). Ji et al. (2015) used (MCMC) algorithm and addressed the enumeration problem by introducing a novel sampler. Whereas, the sampler used by Hazelton (2010) was based on Markov model (Li 2009) which misses on important regions of the target distribution under a realistic OD structure, making the convergence of simulation sequence difficult. Optimization methods. Willumsen (1978) and Van Zuylen (1978) employed entropy maximization and information minimization techniques respectively for OD estimation using link count data. Kikuchi and Nopadon (2009) applied uncertainty maximization approach to the Yokohama Seaside Line. The model is capable of incorporating qualitative as well as quantitative spot information, and it is highly sensitive to such information though. In another study conducted by Kikuchi and Nopadon (2010), the OD table for Hankai Line in Osaka was estimated using fragmented data. Kumar et al. (2019) employed the concept of compressed sensing to estimate the OD matrix for a bus route in Twin Cities, Minnesota. Their proposed model forces the solution to be sparse. The method was shown to produce small errors only. Research gap and objectives. Most of the methodologies developed for OD estimation using on/off counts up to date suffer from major drawbacks. The nonstructural zero problem, the need for a base matrix, and error magnification are the main shortcomings balancing methods suffer from. Similarly, most of the developed methods are not only computationally inefficient, but also produce significant errors in the estimation process. This research aims to review the existing methodologies and propose an approach for transit OD matrix estimation using boarding and alighting counts. The main objective is to develop a model that can estimate reliable trip matrices, with minimal requirement for prior information and higher computational efficiency. The proposed method will be validated using several test matrices and the results will be compared with those generated by some of the existing methods.

2 Methodology The methods considered for this study are briefly explained below. Iterative Proportional Fitting (IPF). IPF method estimates the OD flows along a transit line in such a way that the estimated OD flows are proportional to the corresponding entries in the base matrix. The constants of proportionality alter the elements of the base matrix, making them be consistent with the known boarding

6

S. Mohmmand et al.

and alighting values. As there is no prior information available, we use a null seed in this research. Refer to Moshe et al. (1985) for details about IPF. Markov model. Given the boarding and alighting counts for a route, initially the passenger alighting probability (q j ) for each transit stop is computed using Eq. (1). Ʌ

qˆ j =

αj + z j ∑ j−1 α j + β j + k=1 (yk − z k )

j = 2, . . . , n − 1

(1)

A conjugate prior is considered for q j : q j ∼ beta(α j , β j ). beta(α, β) is α beta distribution with mean equal to α+β and variance of (α+β)2α.β . The hyper(α+β+1) parameters α j and β j are calculated using prior knowledge. In the absence of such knowledge, their values are taken as one; yk and y j are the number of passengers boarding and z k and z j are the number of passengers alighting at stop k and j, respectively. Subsequently, the transition probabilities are found using Eq. (2). pi j is probability of a passenger traveling from stop i to j. pi j = q j

j−1 π

(1 − qk ) and pi(i+1) = qi+1 ( j = i + 2, . . . , n)

(2)

k=i+1

xˆi j = pi j · yi (i = 1, ..., n − 1; j = i + 1, ..., n)

(3)

Having prepared the transition matrix, the entries of this matrix are multiplied by the boarding count for the respective stops to get the corresponding OD flows (see Eq. (3)). Uncertainty maximization. Employed by Kikuchi and Nopadon (2009) for the first time, the method uses spot information to constrain the solution and reduce the degree of ill-posedness. A non-linear optimization problem was formulated solving which would ensure maximum entropy and conformity to the given information. Spot information can be of qualitative (e.g., the flow between stop i and j is almost the same as that between stop m and n) or quantitative (e.g., end-to-end OD flow is 5% of the total trips) forms. They can be incorporated into the structure of the problem through their membership functions. The problem is formulated as follows: Objective: maximi ze h S(H ) ≥ h

(4)

Subjected to n ∑ j=1

Bi =

n ∑

Ai j

(5)

i=1

S Q k (qk ) ≥ h

(6)

Estimation of Transit OD Matrix Using Boarding and Alighting Data …

7

( ) ∑ 1 where entropy is H = i pi j log2 pi j , Bi and A j are the total boarding and alighting at stop i and j, respectively, h is an artificial variable, S(H ) and S Q k (qk ) are membership function for entropy and spot information, respectively. Compressed sensing. This approach was proposed by Kumar et al. (2019). It is based on the concept of convex optimization. The basic assumption is that passengers boarding at a station are unlikely to be alighting at all the subsequent stations, rather they will be attracted by a small number of stations. Therefore, the solution sought needs to be sparse (i.e., contain predominantly zero elements). To achieve a sparse solution, the l0 norm of the solution vector needs to be minimized. However, minimizing the l0 norm of the vector is an NP hard problem. √∑Instead, l1 is used as a surrogate norm. Typically, l p norm of x is defined as l p = p i |xi | p where p ∈ R. In addition to Eq. (5), the following constraints are considered:

k ∑ i=1

xii = 0

(7)

xi j = 0 i f i > j

(8)

(bi − ai ) =

k n ∑ ∑

xi j

(9)

i=1 j=k+1

Equations (7) and (8) show the flow within the same stop and that to the prior stops in a single direction is equal to zero. Further, section loads can be constrained using Eq. (9). All these constraints can be expressed in a general form as Eq. (10): A(x) = B

(10)

The optimization problem can be written as Minimize ||Ax − B||2 + μ||x||1

(11)

Subject to: x >= 0. μ controls the degree of sparsity in the solution. Proposed method. Inspired from compressed sensing, the present study proposes a modified version of the compressed sensing approach. The following are the novelty of the proposed model: 1. Unlike compressed sensing, the current model uses l∞ norm: l∞ norm also called maximum norm is used to spot the maximum absolute value in the error vector |Ax − B|. 2. μ = 0.2 is adopted as a measure of sparsity: The concept of sparsity has been used in image processing and communication networks to compress high-resolution

8

S. Mohmmand et al.

images; thus, minimizing the storage requirement. In the context of transit OD estimation, sparsity is being controlled by the μ value. A higher μ value implies higher sparsity, which can be translated into more number of zero elements in the OD matrix (i.e., flow between a large number of OD pairs will be zero). Whereas, a lower value of μ reduces the number of zero entries in the OD matrix (i.e., there will be some amount of passenger flow between a large number of transit stops). The μ value may be adjusted according to the characteristics of transit system, ridership, and data available. Having tested different values in [0,1] range, we found that μ = 0.2 results in the lowest RMSE value for most of the test matrices. Thus, we adopt this value for our further case study. 3. Rigid or approximate spot information about the critical cell in either direction are provided: Small piece of information (also called spot information) about the flow pattern of passengers between some OD pairs can prove handy in reducing the degree of freedom of our solution. Such information can be obtained from an old OD matrix, experts’ opinion, or surveys conducted. The on/off counts used as input may come from the APC system. APC data includes important details such as date, time, bus ID, trip ID, direction, and the number of passengers boarding and alighting at each stop. Raw APC data needs cleaning before being supplied to the model. The file may be imported to a database and must be checked for consistency and completeness with the help of other datasets such as GTFS. APC data can be obtained at different aggregation levels. In the current study, we have used daily boarding/ alighting counts; however, in case of higher temporal variability in transit ridership, we can generate OD tables at a disaggregated level (i.e., trip-level in different time bins). As pointed out earlier, the μ value would need to be calibrated according to the data available. Figure 1 illustrates the procedure for the proposed method. Initially, boarding/ alighting data is used to identify the critical cell (i.e., the cell with one of the highest errors) in the matrix. It is found that the cell corresponding to the highest boarding and alighting values can be considered as a critical cell. The value of the critical cell can be supplied from sources such as experts’ opinion and on-board survey. But at a non-informative scenario, it can be supplied from IPF-null. The spot information is supplied to the model as a constraint and the model is run on Python API of CVXPY. The method has the ability to generate remarkably good results provided that accurate spot information is injected into the model.

Fig. 1 Proposed model

Estimation of Transit OD Matrix Using Boarding and Alighting Data …

Minimize ||Ax − B||∞ + (0.2)||x||1

9

(12)

Subject to: x >= 0. xcr = p Equation (12) shows the proposed model, where xcr is the flow for the critical cell and p is the value of xcr provided by one of the aforementioned sources.

3 Methodology Testing True OD matrices. In the absence of true OD matrix, we have performed our tests on true matrices collected from the literature (see Table 1). Validation RMSE. At first, root mean square error (RMSE) is estimated for each matrix, which are shown in Table 2. In general, the estimation error is higher for longer transit lines with heavy ridership (see Table 1). Matrix one and three have shown the highest RMSE values among all. As expected, IPF and Markov approaches have similar error values. The proposed method, on the other hand, shows significantly lower RMSE values for both matrices and almost similar RMSE values for the rest of the matrices. Trip length distribution. The trip length distribution is compared for all matrices as show in Fig. 2. These distributions are drawn based on the number of stops for each trip. Similar to RMSE validation, Markov and IPF methods produced similar trip length distributions. These two techniques generate OD matrix of intermediate accuracy; a few large deviations from the actual numbers are observed. From trip length distribution perspective, compressed sensing turned out to be the most inaccurate method. Even though uncertainty maximization approach requires spot information Table 1 Transit OD matrices S. No.

City

Ridership

Study

1

Yokohama Seaside Line, Kanazawa, Japan

22,759

Kikuchi and Kronprasert (2009)

2

Hankai Line, Osaka, Japan

3717

Kikuchi and Kronprasert (2010)

3

Northeastern region of the United States

17,242

Ulusoy et al. (2010)

4

East Fort-Kovalam, Kerala, India

374

Cyril et al. (2017)

5

Line 16, Los Angeles

266

Simon and Furth (1985)

6

Guangzhou, China

417

Gao et al. (2015)

10

S. Mohmmand et al.

Table 2 RMSE values for test matrices Matrix 1

IPF

Markov

Entropy

94.735

41.734

95.008

Compressed sensing 145.675

Proposed method 18.737

2

6.969

7.194

6.482

9.895

6.677

3

156.647

156.637

164.574

233.071

106.822

4

2.581

2.692

3.584

2.788

2.163

5

3.58

3.503

3.574

6.191

3.435

6

3.616

3.697

5.258

2.951

2.917

and involves formulation of membership functions, its results are not satisfactory in general. The proposed method, on the other hand, generated the most consistent results for all test matrices.

(a) Test matrix 1

(d) Test matrix 4

(b) Test matrix 2

(e) Test matrix 5

Fig. 2 Trip length distribution for the test matrices

(c) Test matrix 3

(f) Test matrix 6

Estimation of Transit OD Matrix Using Boarding and Alighting Data …

11

4 Case Study As a case study, the proposed Haridwar-Rishikesh metro corridor is considered (ICRA 2017). The corridor is 34 km long (20 stations), which connects Haridwar and Rishikesh cities in Uttarakhand, India. The boarding/alighting data used in this research is borrowed from the detailed project report (DPR) (ICRA 2017). The OD matrix for the given metro line is estimated using IPF, Markov, Compressed Sensing, and proposed approach. Comparing the results obtained, it can be clearly observed that Markov model can generate the same results as IPF method with less number of iterations and less effort involved. Nonetheless, the difference in OD flows is significant as we compare these two with compressed sensing and the proposed approach (see Fig. 3). OD flows originating at Arya Nagar Chowk and destined at Jwalapur, Har ki Podi, Shantikunj, Rajwala, Bibiwala and Rishikesh and those originating at Haridwar Railway Station and destined at the same destinations are showing higher deviations from the IPF outcome. From Rishikesh to Haridwar Railway Station, the estimated passenger trips using IPF, Compressed Sensing and the proposed method are 1560, 2130, and 1964, respectively, which depict considerable differences of 404 (20.5%) and −166 (−8.5%) with respect to the proposed method. Similarly, the differences for the trips between Haridwar Railway Station and Har Ki Podi are estimated as −438 (−37.5%) and −819 (−70%), respectively. In general, the estimations are more diverse for those stops which are major passenger attractions, whereas they are more or less the same for minor stops. Having discussed the reliability of the results generated by the proposed approach, this study can be pivotal for transportation planning and operation. However, the main limitation of the study is that the results of the case study are not validated from the field study. This is mainly because the subject corridor is at the planning phase and is not operational yet. Thus, validation of the method through an empirical study can be the subject of another study.

5 Conclusion In this research, four methods were discussed for transit OD matrix estimation. The performance of Markov and IPF is almost similar. Uncertainty maximization approach requires spot information and its outcome is highly sensitive to the quality of the spot information. The problem with this approach is that it is difficult to form unbiased membership functions for the given qualitative information. Compressed sensing technique forces the solution to be sparse, making it more suitable for triplevel OD estimation and may result in major errors if used for daily ridership data. To overcome the high errors, a modified compressed sensing model is proposed. The possibility of major errors is reduced by lowering the sparsity coefficient (μ). The results of the validation process depict that the proposed model outperforms

12

S. Mohmmand et al.

(a) IPF

(c) Compressed sensing

(b) Markov model

(d) Proposed method

Fig. 3 Estimated OD matrix for Haridwar-Rishikesh metro line

all four methods discussed in this paper. For instance, for test matrix 1 the RMSE values for compressed sensing were the highest at 145 followed by IPF and entropy maximization methods at 95 and 41, respectively. The RMSE value for the proposed model was only 18. Furthermore, it can be seen from the trip length distribution that the proposed method appears more reliable than the rest of the methods.

References Cyril A, George V, Mulangi RH (2017) Electronic ticket machine data analytics for public bus transport planning. In: 2017 International conference on energy, communication, data analytics and soft computing (ICECDS). IEEE, pp 3917–3922 Farzin JM (2008) Constructing an automated bus origin–destination matrix using fare card and global positioning system data in Sao Paulo, Brazil. Transp Res Rec 2072(1):30–37 Furth PG, Navick DS (1992) Bus route OD matrix generation: Relationship between biproportional and recursive methods. Transp Res Rec

Estimation of Transit OD Matrix Using Boarding and Alighting Data …

13

Gao LX, Li GY, Hu JH, Liang JX (2015) A calculation method of od matrix in multi-modal transit network based on traffic big data. In: 2015 International conference on transportation information and safety (ICTIS). IEEE, pp 295–298 Hazelton ML (2010) Statistical inference for transit system origin-destination matrices. Technometrics 52(2):221–230 He L, Trépanier M (2015) Estimating the destination of unlinked trips in transit smart card fare data. Transp Res Rec 2535(1):97–104 ICRA (2017) Traffic and transportation study for preparation of DPR for Dehradun, Rishikesh and Haridwar metro. Technical report, ICRA Management and Consulting Services Ltd. Ji Y, You Q, Jiang S, Zhang HM (2015) Statistical inference on transit routelevel origin–destination flows using automatic passenger counter data. J Adv Transp 49(6):724–737 Kikuchi S, Kronprasert N (2009) Constructing a transit origin–destination table using the uncertainty maximization concept. Transp Res Rec 2112(1):43–52 Kikuchi S, Kronprasert N (2010) Constructing transit origin–destination tables from fragmented data. Transp Res Rec 2196(1):34–44 Kumar P, Khani A, Davis GA (2019) Transit route origin–destination matrix estimation using compressed sensing. Transp Res Rec 2673(10):164–174 Li B (2009) Markov models for bayesian analysis about transit route origin–destination matrices. Transp Res Part B Methodol 43(3):301–310 Luo D, Cats O, van Lint H (2017) Constructing transit origin-destination matrices with spatial clustering. Transp Res Rec J Transp Res Board 2652(1):39–49. https://doi.org/10.3141/265 2-05 Maher MJ (1983) Inferences on trip matrices from observations on link volumes: a bayesian statistical approach. Transp Res Part B Methodol 17(6):435–447 McCord MR, Mishalani RG, Goel P, Strohl B (2010) Iterative proportional fitting procedure to determine bus route passenger origin–destination flows. Transportation research record, pp 59– 65 Mishalani RG, McCord MR, Reinhold T (2016) Use of mobile device wireless signals to determine transit route-level passenger origin–destination flows: methodology and empirical evaluation. Transp Res Rec 2544(1):123–130 Mishalani RG, Ji Y, McCord MR (2011) Effect of onboard survey sample size on estimation of transit bus route passenger origin–destination flow matrix using automatic passenger counter data. Transportation research record, pp 64–73 Moshe EBA, Peter PM, Hsu PS (1985) Alternative methods to estimate route-level trip tables and expand on-board surveys. Transportation research record Simon J, Furth PG (1985) Generating a bus route od matrix from onoff data. J Transp Eng 111(6):583–593. https://doi.org/10.1061/(ASCE)0733-947X(1985)111:6(583) Ulusoy YY, Chien SIJ, Wei CH (2010) Optimal all-stop, short-turn, and express transit services under heterogeneous demand. Transp Res Rec 2197(1):8–18 Willumsen LG (1978) Estimation of an OD matrix from traffic counts–a review. Technical report, Institute of Transport Studies, University of Leeds Van Zuylen J (1978) The information minimizing method: validity and applicability to transport planning. New developments in modelling travel demand and urban systems

Measuring Perceived Satisfaction of Choice Riders Towards Factors Influencing Accessibility to Metro Stations: An Evidence from Delhi Manaswinee Kar, Shubhajit Sadhukhan, and Manoranjan Parida

Abstract The present research aims at identifying the priority attributes influencing accessibility to the metro stations based on choice riders’ (private vehicle users who choose to ride metro) perception. Tablet-based face-to-face interviews were conducted at the parking lots of metro stations in Delhi, India, to collect 1050 choice rider responses in terms of satisfaction towards the selected accessibility influencing attributes on a six-point Likert-type ordinal rating scale. Three established multi-attribute decision-making (MADM) methods, namely, Relative to an Identified Distribution Integral Transformation (RIDIT), Grey Relational Analysis (GRA) and Technique for Order Preference by Similarity to the Ideal Solution (TOPSIS) were employed to analyse the collected responses and obtain priority rankings of attributes based on user satisfaction. The study findings indicated users’ priority towards improvement of ‘Presence of Variable Message Signs’, ‘Safety’, ‘Security’, ‘Accessibility in Extreme Weather’, ‘Universal Design Considerations’, ‘Parking Guidance Information’ and higher satisfaction with ‘Egress Time’ and ‘In-Vehicle Travel Time’. A comparison of attribute rankings obtained from the three methods was made using Spearman’s rank-order correlation analysis. The coefficients indicated a consistency in the attribute rankings and a positive relationship among the ranking methods at 99% confidence level. The study outcomes can assist the transport planners and policymakers in devising strategies for priority-based judicious allocation of resources towards the improvement of poorly performing attributes influencing accessibility to metro stations.

M. Kar (B) Department of Civil Engineering, Indian Institute of Technology Roorkee, Roorkee 247667, Uttarakhand, India e-mail: [email protected] S. Sadhukhan Department of Architecture and Planning, Indian Institute of Technology Roorkee, Roorkee 247667, Uttarakhand, India M. Parida CSIR-Central Road Research Institute (CRRI), New Delhi 110025, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_2

15

16

M. Kar et al.

Keywords Perceived satisfaction · Choice riders · Accessibility · Metro · RIDIT · GRA · TOPSIS

1 Introduction Rapid urbanisation and economic growth in developing countries like India have resulted in a huge dependence on private vehicles to meet the increasing travel needs of the commuters. However, an increased level of private motorisation worsens the traffic conditions on roads by resulting in congestion, enormous delays and consequent increase in travel time, reduced safety and pollution. Hence, several programmes and schemes have been implemented at the national level to reduce the usage of private vehicles by strengthening and encouraging the use of public transport system. Along with bus services which serve as the primary public transport mode in all Indian cities, metro rail facilities exist in the major metropolitan cities of India namely, Delhi, Kolkata, Mumbai, Bengaluru and many other cities (Goel and Tiwari 2014). However, the metro systems cannot provide door-to-door connectivity to commuters. Accessibility to metro stations plays a significant role in the successful utilisation of the facility. In such a context, Park-and-Ride facilities can provide suitable accessibility to metro stations by allowing the choice riders (private vehicle users who choose to ride metro) to park their vehicles at parking lots near the stations and take metro to their destinations (Bolger et al. 1992; Dickins 1991; Lam et al. 2001; Noel 1988). Such attractive facilities promise seamless connectivity and encourage the choice riders to shift to metro. Due to inadequate literature on the evaluation of metro accessibility involving user perspectives in an urban Indian context, it is imperative to study and understand the perception of choice riders towards various attributes influencing accessibility to the metro stations. The present study aims at bridging the gap by analysing and prioritising the attributes influencing accessibility to the metro stations based on choice riders’ satisfaction. This attribute prioritisation can assist in the optimum allocation of resources towards the improvement of poorly performing attributes. The present study is demonstrated in the context of Delhi, the capital city of India. Delhi, as a case study, is found suitable because it is one of the highly congested and polluted cities in India with a substantial share of registered private vehicles plying on the roads. On the other hand, Delhi owns the most successful and ambitious metro network in the country with 12 operational lines and 285 stations (DMRC 2020). 105 out of 285 metro stations have existing parking facilities which serve as potential Park-and-Ride facilities. Accessibility to the metro stations in terms of such Parkand-Ride facilities can attract the private vehicle users to avoid traffic congestion and delays by shifting to metro thereby, reducing the private vehicle usage and improving the quality of urban life in Delhi. The subsequent section describes the methods employed in the present work in brief. Section 3 broadly illustrates the design of the survey instrument, data collection and database preparation. The results of the data analysis are reported and discussed

Measuring Perceived Satisfaction of Choice Riders Towards Factors …

17

in Sect. 4. Lastly, Sect. 5 summarises the major findings of the present study and highlights the potential avenues of future research.

2 Methods The present study used three well-known multi-attribute decision-making (MADM) techniques, namely, Relative to an identified distribution integral transformation (RIDIT) analysis (Majumdar et al. 2020; Roy and Basu 2019; Sadhukhan et al. 2015, 2018; Wu 2007), Gray relation analysis (GRA) (Majumdar et al. 2020; Roy and Basu 2019; Sadhukhan et al. 2015; Wu 2007) and Technique for order preference by similarity to ideal solution (TOPSIS) (Majumdar et al. 2021; Roy and Basu 2019; Sadhukhan et al. 2015), to identify if there existed any variation in the derived rankings of accessibility influencing attributes specified by these three methods. The rationale behind choosing the three above-mentioned methods was distribution-free (RIDIT and GRA), suitable for analysing Likert-type ordinal data, could be used to evaluate incomplete, poor, uncertain and unreliable data (Kuo et al. 2008). Additionally, TOPSIS involved simpler scalar quantity estimations and showed good calculation efficiency. The theoretical background and the steps involved in RIDIT, GRA and TOPSIS methods are well-defined and can be suitably found in the literature (Roy and Basu 2019; Sadhukhan et al. 2015).

3 Survey Design, Data Collection and Database Preparation A comprehensive set of 17 essential service attributes influencing accessibility to metro stations for choice riders reported in an earlier work by the authors (Kar et al. 2022) is considered for the present study and summarised in Table 1. The attributes were finalised based on an extensive review of related literature (Abdul Sukor et al. 2017; Bergman et al. 2011; Goel and Tiwari 2016; Ministry of Housing and Urban Affairs Government of India 2021; Policy 2014; Saiyad et al. 2021; Wardman and Tyler 2000; Yang et al. 2015) and contextual relevance. A tablet-based survey questionnaire was designed using the selected attributes. A pilot survey (100 samples) was undertaken to validate the designed survey instrument and check its user acceptability. Based on the inferences drawn from the pilot study, the questionnaire incurred subtle improvisations in terms of sequencing of questions, deletion of ambiguous questions and simplification of technical terminologies with help text. Some other modifications were also incorporated, such as inclusion of feeder bus/shuttle bus as an option in the availability of other access modes, inclusion of a question in travel information to mention the name of the boarding metro station and reason in case,

18

M. Kar et al.

it was different from the metro station nearest to home. The final survey questionnaire comprised of three sections. The first section included questions related to the socio-economic information such as gender, age group, occupation, monthly family income and vehicle ownership. The second section included questions related to the travel information such as trip origin and destination, travel frequency, trip purpose, boarding and alighting metro stations, travel group size, travel time, travel cost, available access modes and the mode chosen most often by the respondents to access the metro station. The third section recorded responses of the choice riders in terms of satisfaction towards the selected accessibility influencing attributes on a six-point Likert-type ordinal rating scale ‘0–5’. While ‘1’, ‘2’, ‘3’, ‘4’, ‘5’ indicated ‘very low’, ‘low’, ‘average’, ‘high’ and ‘extreme’ satisfaction with the attribute respectively, ‘0’ indicated the respondents’ satisfaction in the case of non-existing facilities where measuring satisfaction was not possible. A simple random sampling technique was used for data collection. The choice riders were intercepted at the metro station parking lots. The choice riders were approached randomly and requested to take the survey. Detailed responses of the choice riders were recorded for those who agreed to participate in the survey. User responses from 105 metro stations of Delhi metro with existing parking facilities were collected on weekdays and weekends. Surveyors’ assistance not only helped the respondents in completing the survey but also ensured collection of a good quality database. An aggregate of 1050 choice rider responses was recorded. The database was extracted and checked for incompleteness and ambiguity. Tablet-based data collection ensured the collection of a significantly high percentage of complete samples. Then, 1022 complete and valid user samples were cleaned using inter-quartile range method. After removal of outliers (5.58%), 965 cleaned responses (adequate as per (Krejcie and Morgan 1970)) were obtained and considered for further analysis. The satisfaction ratings recorded by the survey respondents with respect to the finalised attributes is summarised in Fig. 1. The internal consistency of the satisfaction rating data was confirmed by a Cronbach’s Alpha Coefficient (α) value of 0.84 (Gliem and Gliem 2003). While the Kolmogorov–Smirnov and Shapiro–Wilk test statistics ranged between 0.206–0.420 and 0.658–0.881 respectively, the p-values were 0.000 at 0.05 significance level (Shapiro and Francia 1972). Hence, both the tests indicated non-normal data. The sample data comprises 75.6% male and 24.4% female choice riders. The survey participants predominantly belong to age group 18–40 years (~83%). The percentage share of respondents aged 40 years old and above is very low because the older people are found to be more vulnerable owing to the COVID-19 pandemic. The majority of respondents (60.93%) exclusively own two-wheelers, 4.87% have solely cars, and 34.2% own both types of vehicles. Two-wheelers are found to be used by a sizable portion of the respondents to access the metro stations. The survey respondents, however, are more inclined to buy and use cars in the future. The recorded database has identified ‘work’ to be the predominant purpose of travel (72.64%) during the prevalent pandemic situation.

Measuring Perceived Satisfaction of Choice Riders Towards Factors …

19

Table 1 List of service attributes influencing choice riders’ accessibility to metro (Source Kar et al. 2022) Sl. No

Attribute

Notation

Description

1

Travel route information

F1

It indicates adequate information regarding the travel route is available to the users to reach the Metro station from home

2

Safety

F2

It indicates the access mode provides a safe travel to users against difficult road traffic conditions enroute to the Metro station

3

Security

F3

It indicates the access mode ensures security to the users against pick-pocketing or any violent activities, etc. during travel from home to the Metro station

4

Riding comfort F4

It indicates the easiness with which the users ride the access mode to the Metro station

5

Convenience

F5

It indicates the suitability and usefulness of the access mode for the users in reaching the Metro station

6

In-vehicle travel time

F6

It indicates the time spent inside the access mode during travelling from home to reaching the parking facility at Metro station

7

Egress time

F7

It indicates the time spent in getting off the access mode and walking from the parking lot till reaching the Metro platform

8

Operational and maintenance cost

F8

It indicates the money spent in maintaining the access mode in good operational conditions

9

Luggage space F9

It indicates the availability of adequate space in the access mode to keep the luggage while travelling to the Metro station

10

Extreme weather conditions

F10

It indicates the easiness of travelling to the Metro station using the access mode during extreme weather conditions such as very hot and humid or rainy or cold weather

11

Universal design considerations

F11

It indicates availability of facilities to make the access mode inclusive for all user groups such as children, pregnant ladies, old people and differently-abled people

12

Designated parking facilities

F12

It indicates presence of designated spaces at Metro stations for systematic and organised parking of different types of vehicles along with separate spaces for users with disabilities

13

Illumination

F13

It indicates provision of adequate lighting inside the Metro parking lot areas to ensure visibility and safety to the users

14

Presence of variable message signs (VMS)

F14

It indicates the provision of real-time display of information regarding the total and remaining spaces available to the users for parking vehicles in the Metro parking lot

15

Parking guidance information

F15

It indicates the presence of directional information in the form of arrows for guiding the users in parking vehicles at the Metro parking lot

16

Parking cost

F16

It indicates the existing fee structure applicable for parking vehicles at the Metro parking lot (continued)

20

M. Kar et al.

Table 1 (continued) Sl. No

Attribute

Notation

Description

17

Parking security

F17

It indicates the security provisions in the form of security guard, CCTV cameras available for safeguard of the vehicles parked at the Metro parking lot

Fig. 1 Summary of satisfaction ratings of survey respondents

4 Data Analysis, Results and Discussion 4.1 RIDIT Analysis The present study attempts to make a suitable comparison of the results obtained by analysing the perception of choice riders using RIDIT analysis which is reported in an earlier work by the authors (Kar et al. 2022) with the results obtained from two other pronounced MADM methods, GRA and TOPSIS. RIDIT analysis was performed on the cleaned database (965 responses) and the results are summarised in Table 2. The statistical significance of the RIDIT analysis results is confirmed by the Kruskal–Wallis Value (W) value 31,519.22, which is greater than the critical Chi-squared value [χ2 (17 − 1) = 26.296] at 0.05 significance level.

Measuring Perceived Satisfaction of Choice Riders Towards Factors …

21

Table 2 Attribute ranking using RIDIT analysis Attributes

RIDIT score (ρi )

Lower bound

Upper bound

Travel route information

0.8360

0.8174

0.8545

Rank 9

Safety

0.6162

0.5976

0.6348

15

Security

0.5735

0.5549

0.5921

16

Riding comfort

0.7518

0.7333

0.7704

12

Convenience

0.8586

0.8400

0.8772

7

In-vehicle travel time

1.0261

1.0075

1.0447

3

Egress time

1.3265

1.3079

1.3450

1

Operational and maintenance cost

0.9473

0.9287

0.9659

6

Luggage space

0.8296

0.8111

0.8482

10

Extreme weather conditions

0.6829

0.6643

0.7014

13

Universal design considerations

0.6367

0.6181

0.6553

14

Designated parking facilities

0.9818

0.9632

1.0003

5

Illumination

0.9924

0.9739

1.0110

4

Presence of variable message signs (VMS)

0.1439

0.1253

0.1625

17

Parking guidance information

0.7712

0.7526

0.7898

11

Parking cost

1.0279

1.0093

1.0465

2

Parking security

0.8517

0.8332

0.8703

8

This implies statistically significant but different perceptions of choice riders towards the satisfaction associated with the selected attributes in the study. Further, it may be observed from Table 2 that the respondents perceived ‘Egress Time’ as the highest satisfactorily performing attribute (ρi = 1.3265) followed by ‘Parking Cost’ (ρi = 1.0279), ‘In-Vehicle Travel Time’ (ρi = 1.0261), ‘Illumination’ (ρi = 0.9924) and ‘Designated Parking Facilities’ (ρi = 0.9818). This is because Delhi Metro Rail Corporation (DMRC) has offered provisions of parking lots adjacent to the metro stations, facilitating the choice riders to park their vehicles and reach the metro platform in very less time. At the same time, the respondents seem to be quite satisfied with the existing parking fare structure as compared to other lot-related attributes due to the subsidised rates enforced by DMRC. Higher user satisfaction with in-vehicle travel time can be justified because the access modes are private vehicles which are driven by the users. On the other hand, the respondents attached low satisfaction with ‘Presence of VMS’ (ρi = 0.1439) and ‘Parking Guidance Information’ (ρi = 0.7712) due to the absence of such facilities in parking lots of majority of the stations. Also, ‘Security’ (ρi = 0.5735), ‘Safety’ (ρi = 0.6162), ‘Universal Design Considerations’ (ρi = 0.6367), ‘Extreme Weather Conditions’ (ρi = 0.6829), ‘Riding Comfort’ (ρi = 0.7518) were associated with poor satisfaction levels which indicated the need for immediate interventions to improve accessibility to the metro stations. The need for improving the above four attributes can be clearly

22

M. Kar et al.

understood with respect to a two-wheeler as it is an open vehicle (as opposed to a fourwheeler) and more prone to safety and security-related issues. Also, travelling in a two-wheeler during bad weather is uncomfortable, inconvenient and can be risky (for example, reduced visibility during heavy rains, fog). Moreover, such a vehicle does not include suitable facilities to provide universal accessibility to various user groups like differently-abled, elderly, pregnant ladies, etc. Apart from the above-mentioned attributes, the respondents attached moderate satisfaction levels with ‘Operational & Maintenance Cost’, ‘Convenience’, ‘Parking Security’ (towards the higher side) and ‘Travel Route Information’, ‘Luggage Space’ (towards the lower side). This indicates a need to instal informatory boards near potential trip origin locations to keep the choice riders informed about the travel route to the metro stations and also, improve the design aspects of access modes, especially two-wheeler in terms of luggage space requirements.

4.2 GRA The choice riders’ satisfaction responses were also analysed by Gray Relational Analysis, and the findings are reported in Table 3. From Table 3, it is evident that GRA identifies an additional attribute ‘Operational & Maintenance Cost’ to be associated with high degree of satisfaction apart from other satisfactorily performing attributes namely ‘Egress Time’, ‘In-Vehicle Travel Time’, ‘Parking Cost’ and ‘Illumination’ as identified in the RIDIT analysis. The respondents are moderately satisfied with ‘Designated Parking Facilities’ which indicates the need to provide parking lots near metro stations which do not have such facilities. Other attributes, namely, ‘Travel Route Information’, ‘Convenience’, ‘Parking Security’ and ‘Luggage Space’ are found to be moderately performing according to the users. Moreover, both the methods (RIDIT and GRA) indicate the same set of attributes specifically, ‘Presence of VMS’, ‘Security’, ‘Safety’, ‘Universal Design Considerations’, ‘Extreme Weather Conditions’, ‘Parking Guidance Information’ to be poorly performing based on user perception. This further strengthens the research finding that the above-mentioned attributes influencing the choice riders’ accessibility to metro stations are found to be priority areas of resource allocation and seek immediate improvement.

4.3 TOPSIS Analysis The present study considered the 17 selected accessibility influencing attributes as alternatives and the 6 different levels in the rating scale (each level weighted equally) as criteria to record satisfaction towards the attributes. The satisfaction levels 4, 5 were maximised and 0, 1, 2, 3 were minimised to obtain the positive ideal solution. Similarly, the levels 0, 1, 2, 3 were maximised and 4, 5 were minimised to obtain the

Measuring Perceived Satisfaction of Choice Riders Towards Factors …

23

Table 3 Attribute ranking using GRA Attributes

Smax

Smin

dmax

dmin

Average gray score

Travel route information

5

2

3

0

0.7581

Rank 6

Safety

5

1

4

0

0.6946

15

Security

5

1

4

0

0.6781

16

Riding comfort

5

1

4

0

0.7349

11

Convenience

5

0

5

0

0.7558

8

In-vehicle travel time

5

2

3

0

0.8019

2

Egress time

5

2

3

0

0.8749

1

Operational and maintenance cost

5

0

5

0

0.7766

5

Luggage space

5

0

5

0

0.7532

10

Extreme weather conditions

5

1

4

0

0.7113

13

Universal design considerations

5

1

4

0

0.6969

14

Designated parking facilities

5

0

5

0

0.7561

7

Illumination

5

2

3

0

0.7915

4

Presence of variable message signs (VMS)

5

0

5

0

0.4176

17

Parking guidance information

5

0

5

0

0.7250

12

Parking cost

5

2

3

0

0.7956

3

Parking security

5

1

4

0

0.7550

9

negative ideal solution. TOPSIS scores (C∗j ) were then computed for all the selected accessibility influencing attributes using the respective positive (S+ i ) and negative ) ideal solutions. The results are summarised in Table 4. (S− i From Table 4, it can be observed that the respondents arehighly satisfied with the  ∗ travel time-related attributes, ‘Egress Time’ Cj = 0.9201 and ‘In-Vehicle Travel   Time’ C∗j = 0.8737 similar to RIDIT and GRA attribute ranking methods. Also, the respondents attachhigher satisfaction levels with other attributes   like ‘Illumi∗ ∗ nation’ Cj = 0.8463 , ‘Travel Route Information’ Cj = 0.8164 and ‘Opera  tional & Maintenance Cost’ C∗j = 0.8156 . In terms of poor performance, TOPSIS analysis identifies the same attributes, i.e. ‘Presence of VMS’, ‘Universal Design Considerations’, ‘Security’, ‘Safety’, ‘Extreme Weather Conditions’, ‘Parking Guidance Information’ as obtained in RIDIT and GRA.

24

M. Kar et al.

Table 4 Attribute ranking using TOPSIS analysis Attributes

S+ i

S− i

C∗j

Travel route information

0.0471

0.2094

0.8164

4

Safety

0.0928

0.1862

0.6673

14

Security

0.1162

0.1757

0.6021

15

Riding comfort

0.0569

0.2062

0.7839

10

Convenience

0.0517

0.2035

0.7973

8

In-vehicle travel time

0.0310

0.2145

0.8737

2

Egress time

0.0194

0.2237

0.9201

1

Operational and maintenance cost

0.0460

0.2035

0.8156

5

Luggage space

0.0510

0.2029

0.7992

7

Extreme weather conditions

0.0840

0.1855

0.6883

13

Universal design considerations

0.1271

0.1721

0.5752

16

Designated parking facilities

0.0489

0.1884

0.7941

9

Illumination

0.0380

0.2093

0.8463

3

Presence of variable message signs (VMS)

0.2168

0.0375

0.1476

17

Parking guidance information

0.0794

0.1823

0.6965

12

Parking cost

0.0469

0.2054

0.8140

6

Parking security

0.0611

0.1922

0.7589

11

Rank

4.4 Comparative Attribute Rankings of Three MADM Techniques Due to a difference in the approach, the three MADM techniques employed in the present study yielded different scores with the same response database. Hence, Spearman’s rank-order correlation test was performed using IBM SPSS Statistics 26.0 software to check the monotonicity in the attribute rankings obtained from the three MADM techniques and the results are reported in Table 5. The results of the analysis indicated a statistical significance of the correlation coefficients and a positive relationship among the attribute rankings by the three techniques at 99% confidence level. Although the attribute rankings obtained from all the methods are acceptable, the present study prefers the priority rankings of accessibility influencing attributes obtained from GRA due to a higher correlation (indicative from the coefficients) with the rankings proposed by RIDIT and TOPSIS methods.

Measuring Perceived Satisfaction of Choice Riders Towards Factors … Table 5 Results of Spearman’s rank-order correlation analysis

25

RIDIT

GRA

TOPSIS

RIDIT

1.000

0.975**

0.890**

GRA

0.975**

1.000

0.953**

TOPSIS

0.890**

0.953**

1.000

**

Correlation is significant at 0.01 level (2-tailed)

5 Conclusion The present study examines the attributes influencing accessibility to metro stations and determines attribute priority based on choice riders’ satisfaction in an urban Indian context. The choice riders have indicated a need for immediate improvement of qualitative attributes influencing metro accessibility than the quantitative attributes (‘Egress Time’, ‘In-Vehicle Travel Time’, ‘Parking Cost’ and ‘Operational & Maintenance Cost’). The qualitative attributes include parking lot-related attributes like ‘Presence of VMS’, ‘Parking Guidance Information’ and access moderelated attributes like ‘Security’, ‘Safety’, ‘Extreme Weather Conditions’, ‘Universal Design Considerations’. The results of Spearman’s rank-order correlation analysis confirm consistency and statistical significance among the RIDIT, GRA, and TOPSIS-derived attribute rankings based on satisfaction of choice riders at 99% confidence level. However, the present research adopts the ranking suggested by GRA method due to a higher correlation with rankings suggested by RIDIT and TOPSIS as indicated by the correlation coefficients. Apart from highlighting the priority improvement areas, the present research work instigates to further consider importance ratings of choice riders with respect to the attributes and perform Importance-Satisfaction analysis to identify gaps. Estimating the perceived benefit to the choice riders based on their willingness-to-pay for improvements in the metro accessibility facilitating attributes and also the influences of these improvements on metro ridership can be some possible avenues for future research. While the prioritisation approach demonstrated in the present study can assist the transport planners and policymakers to formulate strategies for judicious allocation of resources towards the improvement of the attributes, it can be applied with suitable alterations in other contexts which involve multi-attribute decision-making.

References Abdul Sukor NS, Jarani N, Muhammad Fisal SF (2017) Analysis of passengers’ access and egress characteristics to the train station. Eng Herit J 1:01–04. https://doi.org/10.26480/gwk.02.2017. 01.04 Bergman Å, Gliebe J, Strathman J (2011) Modeling access mode choice for inter-suburban commuter rail. J Public Transp 14:23–42. https://doi.org/10.5038/2375-0901.14.4.2

26

M. Kar et al.

Bolger D, Colquhoun D, Morrall J (1992) Planning and design of park-and-ride facilities for the calgary light rail transit system. Transp Res Rec 2:141–148 Delhi Metro Rail Corporation (DMRC) (2020) DMRC annual report 2019–2020 Dickins IANSJ (1991) Park and ride facilities on light rail transit systems. Transportation (AMST) 18:23–36 Gliem JA, Gliem RR (2003) Calculating, interpreting, and reporting cronbach’s alpha reliability coefficient for Likert-type scales. In: Midwest research-to-practice conference in adult, continuing, and community education, pp 82–88 Goel R, Tiwari G (2014) Promoting low carbon transport in India: case study of metro rails in indian cities Goel R, Tiwari G (2016) Access–egress and other travel characteristics of metro users in Delhi and its satellite cities. IATSS Res 39:164–172. https://doi.org/10.1016/j.iatssr.2015.10.001 Kar M, Sadhukhan S, Parida M (2022) Measuring heterogeneity in perceived satisfaction of private vehicle users towards attributes affecting access to metro stations : a case study of Delhi. Case Stud Transp Policy. https://doi.org/10.1016/j.cstp.2022.07.009 Krejcie RV, Morgan DW (1970) Determining sample size for research activities. Educ Psychol Meas 30:607–610 Kuo Y, Yang T, Huang GW (2008) The use of grey relational analysis in solving multiple attribute decision-making problems. Comput Ind Eng 55:80–93. https://doi.org/10.1016/j.cie. 2007.12.002 Lam WHK, Holyoak NM, Lo HP (2001) How park-and-ride can be successful in eastern Asia. J Urban Plan Dev 127:63–78 Majumdar BB, Mitra S, Pareekh P (2020) On identification and prioritization of motivators and deterrents of bicycling. Transp Lett 12:591–603. https://doi.org/10.1080/19427867.2019.167 1042 Majumdar BB, Sahu PK, Patil M, Vendotti N (2021) Pedestrian satisfaction-based methodology for prioritization of critical sidewalk and crosswalk attributes influencing walkability. J Urban Plan Dev 147. https://doi.org/10.1061/(asce)up.1943-5444.0000718 Ministry of Housing and Urban Affairs Government of India (2021) Harmonised guidelines & standards for universal accessibility in India 2021 Noel EC (1988) Park-and-ride: alive, well, and expanding in the United States. J Urban Plan Dev 114:2–13 Ministry of Urban Development (MoUD) (2014) National urban transport policy. Gov. India Roy S, Basu D (2019) Ranking urban catchment areas according to service condition of walk environment. J Transp Eng Part A Syst 145:04019005. https://doi.org/10.1061/jtepbs.0000225 Sadhukhan S, Banerjee UK, Maitra B (2015) Commuters’ perception towards transfer facility attributes in and around metro stations: experience in Kolkata. J Urban Plan Dev 141:04014038. https://doi.org/10.1061/(asce)up.1943-5444.0000243 Sadhukhan S, Banerjee UK, Maitra B (2018) Preference heterogeneity towards the importance of transfer facility attributes at metro stations in Kolkata. Travel Behav Soc 12:72–83. https://doi. org/10.1016/j.tbs.2017.05.001 Saiyad G, Srivastava M, Rathwa D (2021) Assessment of transit accessibility through feeder modes and its influence on feeder mode choice behavior. Arab J Sci Eng. https://doi.org/10.1007/s13 369-021-06082-9 Shapiro SS, Francia RS (1972) An approximate analysis of variance test for normality. J Am Stat Assoc 67:215. https://doi.org/10.2307/2284728 Wardman M, Tyler J (2000) Rail network accessibility and the demand for inter-urban rail travel. Transp Rev 20:3–24. https://doi.org/10.1080/014416400295310 Wu C-H (2007) On the application of grey relational analysis and RIDIT analysis to Likert scale surveys. Int Math Forum 2:675–687. https://doi.org/10.12988/imf.2007.07059 Yang M, Zhao J, Wang W, Liu Z, Li Z (2015) Metro commuters’ satisfaction in multi-type access and egress transferring groups. Transp Res Part D Transp Environ 34:179–194. https://doi.org/ 10.1016/j.trd.2014.11.004

Impact Assessment of Ring Road in Haridwar and National Highway Connectivity on Haridwar-Rishikesh Metro Ridership Nidhi Kathait, Pushpa Choudhary, Amit Agarwal, and Rajat Rastogi

Abstract An exclusive mass rapid transit (MRT) system is planned to connect Dehradun, Rishikesh, and Haridwar cities. A detailed project report (DPR) was prepared to estimate the ridership for different corridors of the proposed metro routes for the base year (2017) and horizon years. In the last few years, a few developments have been proposed in the region; they are a ring road in Haridwar and an upgradation of National Highway connectivity between Haridwar and Rishikesh. The Haridwar-Rishikesh metro line coincides with the National Highway between the two cities. In this case study, an assessment of the impacts of these developments on the metro ridership for the horizon year 2021 is performed. A model is developed using the available Origin-Destination (OD) matrix data and scenarios are developed for both base and horizon years. In addition to model development, a questionnaire survey was carried out to assess the current situation (i.e., choices post-up-gradation of the National Highway between Haridwar and Rishikesh). The results show that the introduction of the ring road is likely to reduce the metro share in the range of 0.1–1.4%, which is only marginal, and NH up-gradation is expected to reduce the metro share in the range of 3.5–5.0%, which is minor yet expected based on the input OD matrix. Keywords Metro neo · Metro ridership · Scenario comparison · OD matrix

N. Kathait (B) · P. Choudhary · A. Agarwal · R. Rastogi Department of Civil Engineering, IIT Roorkee, Roorkee, Uttarakhand, India e-mail: [email protected] P. Choudhary e-mail: [email protected] A. Agarwal e-mail: [email protected] R. Rastogi e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_3

27

28

N. Kathait et al.

1 Introduction India is witnessing rapid urbanization due to the enhancement of the infrastructure development sector under various urban development schemes contributing to growth and development across the country. The state of Uttarakhand is in the stage of planning a Mass Rapid Transit (MRT) system to connect the major cities to promote sustainable transport connectivity and improve overall economic development. In a recent Traffic and Transportation Study for preparation of the Detailed Project Report (DPR) (ICRA 2017), a metro neo system is proposed, which is an articulated or biarticulated trolley bus system with overhead electric traction. The proposed metro neo system will connect Dehradun, Rishikesh, and Haridwar cities, which see huge tourist footfall several times every year. These cities are not only witnessing rapid urban development but are also the key economic centers for the state. Looking at the forecasted passenger demand, instead of an MRT system, a fully elevated bus rapid transit system is proposed, which is named metro-neo. The project is being undertaken by Uttarakhand Metro Rail Urban Infrastructure & Building Construction Corporation (UKMRC). In the DPR, the year 2017 is considered the base year, and ridership estimation is carried out for different corridors of the proposed metro routes for different horizon years (2021, 2031, 2041, and 2051). An increase in population, tourism, and per capita trip rate is considered for estimating the ridership and future growth in trips for all the horizon year. Any infrastructure project may affect the overall vehicular composition, trip patterns, travel behavior, etc., in the region. Two major developments are proposed for the Dehradun metropolitan region, which may affect the overall metro ridership estimated in the DPR. These two developments are (i) Ring Road in Haridwar and (ii) Upgradation of National Highway (NH) connectivity between Haridwar and Rishikesh. The idea behind the proposal for a ring road in Haridwar and the upgradation of NH between Haridwar and Rishikesh is to reduce the congestion levels on the streets by avoiding through traffic in the city. The proposed alignment for the metro is along the NH connecting Haridwar, Rishikesh, and Dehradun. Clearly, such developments in a region are likely to impact travel behavior and thus affect the overall modal share. This paper presents a case study that focuses on assessing the impact of these two developments on estimated overall ridership on the Haridwar-Rishikesh line for the horizon year 2021. For this, a model is developed using the Origin-Destination (OD) matrix, and two scenarios are calibrated to estimate the metro share for both base (2017) and horizon year (2021). This study uses a multi-agent transport simulation framework (MATSim) for scenario development, calibration, and Graphhopper Routing Engine to estimate the change in metro share due to recent developments in the region. The novelty of the case study lies in the following tasks: (a) use of a multinomial logit model to estimate the mode-specific constant (b) use of the Graphhoper routing engine to estimate the travel time for a new road (in addition to the

Impact Assessment of Ring Road in Haridwar and National Highway …

29

existing road network) (c) use of a framework which integrates the real-time congestion pattern with the Graphhopper routing engine to evaluate the real-time travel times for given OD pairs. The rest of the paper is structured as follows: The study area is demonstrated in Sect. 2, where an overview of the study area is presented, followed by details related to available data, OD matrix, and modal share. The methodology is described in Sect. 3, where calibration for the base year and horizon year is discussed. Model for the year 2021 with ring road and NH up-gradation is also described in Sect. 3. Results are discussed in Sect. 4, and the study is concluded in Sect. 5.

2 Study Area and Data 2.1 Overview of the Study Area The study area of the reference study (ICRA 2017) is Dehradun, Rishikesh, and Haridwar. The same has been taken in the present study; however, as per the requirement, the insights are focused on the Haridwar-Rishikesh metro line only. Both cities witness major tourist footfall a couple of times a year to take the dip in the holy river Ganga. A Coordinate Reference System (CRS) was missing from the zonal shapefile, which is aligned on the map using AutoCAD. For this, a few potential reference points from the zonal shapefile are matched with the map (see Fig. 1). These points are chosen based on the understanding of the zones and road network. After this, all points in the zonal shapefile are adjusted to match the reference points. This process provides a CRS, which is used in the present study.

2.2 Availability and Source of the Data The study utilizes the already available data from DPR (ICRA 2017). The data includes 1. 2. 3. 4.

Zone shapefile OD matrix for 2017, 2021 Traffic volume count Alignment of Haridwar Ring Road.

30

N. Kathait et al.

Fig. 1 Zones in the study area

2.3 Modal Share The modal share for the Dehradun, Rishikesh, and Haridwar regions is taken from DPR (ICRA 2017). The modal share is for motorized trips only. In the absence of any other additional information, it is assumed that the OD matrices comprise motorized trips only and, therefore, included in the scenario. Due to ambiguity in usages of auto and Intermediate Public Transport (IPT) modes, it is excluded, and the share of auto mode (i.e., 5%) is redistributed in the remaining modes. The modal share for the base year (2017) is shown in Table 1. Table 1 Derived modal share for the base year Travel modes

Motorized two-wheeler

Car

Shared IPT

Bus

Share (%)

38

21

22

19

Impact Assessment of Ring Road in Haridwar and National Highway …

31

2.4 Utility Parameters The utility parameters of various modes are available from DPR (ICRA 2017). These can be written as follows: U M T W = ASC M T W + 12.38 · dist M T W + 53.17 · time M T W

(1)

Ucar = ASCcar + 9.09 · distcar + 95.18 · timecar

(2)

U I P T = ASC I P T + 16.30 · dist I P T + 38.50 · time I P T

(3)

Ubus = ASCbus + 18.37 · distbus + 44.14 · timebus

(4)

Umetr o = ASCmetr o + 5.12 · distmetr o + 45.27 · timemetr o

(5)

where U is utility and different travel modes are subscribed for each parameter. ASC is an alternative (or mode) specific constant, dist, time are travel distance and travel time between origin and destination. However, alternative specific constant of any travel mode is not available; these are required to study the impact of any transport infrastructure-related changes/ developments on the metro ridership (i.e., modal share).

3 Methodology The work is split into the following steps: • • • •

Calibration for the base year (2017) Calibration for horizon year (2021) Scenario development for ring road in Haridwar and up-gradation of NH Comparison of metro shares for the two developments.

3.1 Calibration for the Base Year (2017) The modal share and utility parameters for the base year (2017) are given. The Alternative Specific Constants (ASCs) are not available, which are required to study the shift from one transport mode to another. To determine the ASCs, a Dynamic Traffic Assignment (DTA)-based model is set up. In this work, a Multi-Agent Transport Simulation (MATSim) framework is used (Horni et al. 2016). This tool is previously used for calibration purposes in the Indian context (Agarwal et al. 2018, 2020) and

32

N. Kathait et al.

embeds the multinomial logit model. Typically, it is an activity-based model, i.e., trips are surrounded by activities performed by a person. However, in the absence of the activity information for the persons, type of activities, location of activities, etc., the MATSim is used as a trip-based model by making the scoring (utility computation) for activities zero. This means that only the trip-based utility function will be used in the MATSim controller. The following steps are taken for the scenario calibration. 1. Road Network: A road network is required to model the trips. A road network is extracted from the OpenStreetMap1 (OSM). The road network is further cleaned based on the various manual checks. The same OSM network is also used to add the ring road in Haridwar and further evaluation (see Sect. 3.3). 2. Trips: Typically, a MATSim person plan consists of the location of activities, their end times, and travel mode between each pair of activities. Since MATSim is used as a trip-based model, a plan is created by drawing points (as per random uniform distribution) for origins and destinations. To accelerate the process, the number of points is taken as 10% of the total trips between each OD pair. For each trip, a mode is assigned in a way that the resulting modal share turns out to be the actual modal share (see Table 1). A departure time is required for each trip. For this, the traffic volume counts data is used to determine the mean and standard deviation of morning and evening peaks and the share of trips in the morning and evening. The analysis of the traffic counts data shows that 54% of the trips are made in the morning and the rest in the evening. The mean and standard deviations are (09:00, 2 h) and (18:00, 2 h) for morning and evening peaks, respectively. Thus, the trip start times are drawn as per these time distributions. 3. Configuration parameters: The network loading algorithm in MATSim is suitable for Indian traffic conditions (Agarwal et al. 2018). The embedded link dynamics resemble the simplified kinematic wave model. For this scenario, a simulation run requires about 100 iterations to converge. For the initial 80% of the iterations, innovation is switched on (i.e., mode choice, route choice, and time choice are permitted). This is required to assign the appropriate travel modes to the trips. For the rest of the iterations, a MATSim plan is selected randomly from the choice set based on a probability distribution that converges to a multinomial logit model. A different variation of alternative specific constants can be attempted in each run so that the modal share after convergence is close to the modal share in Table 1. Thus, after about 175 runs (each with 100 iterations), the resulting modal share is close to the original (actual) modal share, as shown in Table 2.

1

https://www.openstreetmap.org/.

Impact Assessment of Ring Road in Haridwar and National Highway …

33

Table 2 Original and calibrated modal share for the base year Travel modes

Motorized two-wheeler

Car

Shared IPT

Bus

Original modal share

38

21

22

19

Calibrated modal share

38.28

20.18

21.91

19.63

3.2 Calibration for the Year 2021 To determine the change in the metro ridership, the alternative specific constant for the metro is also required. However, similar to ASCs for travel modes in the base year, it is also unavailable. The ASCs for all modes except the metro are known from Sect. 3.1. However, the same methodology cannot be adapted to determine the ASC for metro because the modal share for the year 2021 is unavailable, i.e., the calibration cannot be performed. Therefore, a different approach is adopted for this purpose. The steps are explained next. 1. Total trips, as well as metro trips for the year 2021, are known. This provides the share of metro trips (i.e., pmetr o ). 2. Equation (5) can be re-written as Eq. (6), in which Vmetr o is already known. Further, the application of the multinomial logit model to determine metro share  can be given by Eq. (7). Here, eUr est is the sum of the exponential of utilities of all travel modes (except metro). Rearranging the equation results in Eq. (9), in which the right side can be determined if travel distance and travel times are known for a given OD pair Umetr o = ASCmetr o + Vmetr o eUmetr o  + eUr est    pmetr o ∗ eUr est Umetr o = ln (1 − pmetr o )    pmetr o ∗ eUr est − Vmetr o ASCmetr o = ln (1 − pmetr o ) pmetr o =

eUmetr o

(6)

(7)

(8)

(9)

where Vmetr o is the time and distance component of the utility for the metro (see Eq. (5)). pmetr o is the share of the metro, i.e., the probability of opting for metro in the given choice set. Ur est is the utility of the travel modes other than the metro. 3. To calculate the utility of all modes (see Eqs. (1–5)), travel distance and travel times are required. Therefore, a novel tool is used. The tool is developed by

34

N. Kathait et al.

Choudhary et al. (2022)2 and can be used in the web browser as well as an Application Programming Interface (API). In the present study, it is referred to as “GHC Router.” The process is briefly explained here. 4. The GHC Router uses Graphhopper Routing Engine,3 in which the shortest/ fastest path is suggested for a given OD pair. Therefore, for each OD pair, fifty random origins and destinations are identified, which represent fifty trips. Some of the zones are big and cover the forest, hills, rivers, etc. Therefore, to avoid the illogical points, forests, hills, etc., are excluded from these zones, and thus, a possible location of origin or destination is used. Hence, the configured Graphhopper API (Choudhary et al. 2022) estimates travel distances for each trip using different travel modes. 5. Since this stage is for the base year, the travel time for each trip is estimated using the average speeds given in the DPR (ICRA 2017). This way, utilities of all modes are known, and using Eq. (9), ASC for metro can be estimated for each trip in each OD pair. 6. The average ASC of fifty trips is stored for each OD pair. This ASC is further used in the next stage. The number fifty appears to be enough to have random points from all over a zone to capture enough spatial context. An increase in these random points will increase the computation time.

3.3 Modeling for the Year 2021 with Ring Road and NH Up-Gradation This is the stage in which the impact on the ridership will be estimated. For this, the utility equations of all travel modes can be used. However, the road network needs to be updated, i.e., it should have a ring road. Therefore, the following approach is adopted. 1. The Graphhopper-based routing tool (GHC Router) is used, and the road network from OpenStreetMap) is updated by adding the ring road. For this, the JOSM tool4 is used (see Fig. 2 for the alignment of the ring road in Haridwar). The updated network (existing + ring road) is used with the GHC Router (Choudhary et al. 2022). For instance, the right side of Fig. 3 confirms that in the real world, the ring road does not exist (i.e., the map layer does not show the ring road). However, the left side of Fig. 3 shows a route including the ring road for a given origin and destination, i.e., all the calculation is processed on the digital road network on the back-end. Thus, the impact of Ring Road in Haridwar can be included in the model. 2. In the next step, the impact of up-gradation of the NH between Haridwar and Rishikesh is required. For this, during the estimation of travel distances and 2

https://github.com/teg-iitr/congestion-emission-routing-system. https://github.com/graphhopper/graphhopper. 4 https://josm.openstreetmap.de/. 3

Impact Assessment of Ring Road in Haridwar and National Highway …

35

Fig. 2 Haridwar ring road in JOSM

Fig. 3 Map showing a possible route using ring road (left) and absence of ring road in real-world (right)

travel times, the impact of the real-time congestion should be included (i.e., the evaluation year is the same as the horizon year (2021)). Therefore, here Maps Traffic Flow Information API5 is integrated with the GHC Router. It is a realtime, crowdsourced API that can provide the real-time average speeds on most of the arterials and major roads for a given area (bounding box). These average 5

https://developer.here.com/documentation/traffic/dev_guide/topics/what-is.html.

36

N. Kathait et al.

speeds are integrated into the GHC router such that the routes suggested by the GHC router include the impact of real-time congestion patterns. Since the upgradation of the NH between Haridwar and Rishikesh was completed before this study, the impact of the up-gradation of the NH between Haridwar and Rishikesh can be included in the model with the help of real-time crowdsourced API. 3. Though the GHC router supports multi-modal routing (e.g., car, motorcycle, etc.), the metro support is not present. Therefore, the following approach is adopted. a. A QuadTree data structure of all metro stops is created. This will provide a metro stop close to any given coordinate. b. Whenever a request to find a metro route is placed, the nearest metro stations close to origin and destinations are identified. c. Since the metro alignment is along the NH, a car route is identified from the origin metro stop (access) to the destination metro stop (egress). d. Thus, the travel distance for the metro route is the sum of access distance, travel distance by metro, and egress distance. e. The proposed metro (“metro neo”) alignment is assumed as fully access controlled. Therefore, to estimate the travel time, the average speed is taken from the DPR (ICRA 2017). f. In the absence of details about the access/egress mode share, access/egress travel is assumed by walk, and travel time is estimated using an average walk speed of 5 km/h. 4. In this work, the study area is only Haridwar-Rishikesh; therefore, the metro line in Combination 3 (in Table 6–14 of DPR (ICRA 2017)) is used. The trips beyond the study area are excluded post-calibration step. 5. As shown in Fig. 4, many zones close to metro alignment were having zero metro trip production, which appears to be an error in the input data. The production value from the neighboring zone is taken to fix this issue. 6. After step 5, the utilities of all modes can be estimated. Thus, similar to the calibration for the year 2021 (Sect. 3.2), fifty random trips are drawn for each OD- pair, utilities of all travel modes are estimated, and new metro trips are estimated using the multinomial logit model (Eq. (7)).

3.4 Estimation of Metro Ridership for the Year 2021 with Ring Road and NH Up-Gradation In order to determine the ridership at each stop, the nearest metro stop is required for each zone so that the OD trips can be aggregated to metro stops on HaridwarRishikesh and Rishikesh-Haridwar lines. This process is to be performed for the OD Matrix for metro trips in the year 2021. The QuadTree data structure for metro stations is used again in the next step. In other words, the nearest metro stop is identified from each zone, and boarding/ alighting numbers are aggregated for both cases (see Sect. 4) on both Haridwar to

Impact Assessment of Ring Road in Haridwar and National Highway …

37

Fig. 4 Haridwar zones showing trip productions (red is close to zero and blue is highest)

Rishikesh and Rishikesh to Haridwar lines. This provides the percent change in the metro ridership from each stop.

3.5 Findings from the Survey A questionnaire survey was planned to assess the willingness to shift to the metro in the current scenario, which includes the impact of the development of the NH between Haridwar and Rishikesh. Additionally, the survey includes the question related to the metro usage after Ring Road in Haridwar. The data was collected using Pen-and-Paper Personal Interviews (PAPI) in various parts of Haridwar. The survey questionnaire has different sections to capture the effect of recent developments in the area. The development of the NH gets embedded in the form of new travel times and travel costs, which is already known to the respondents as they are using the system for their travel. In total, 1099 records were collected. A few survey records were excluded during the data cleaning process. Finally, 658 records were analyzed. The modal share from cleaned responses is shown in Fig. 5. This modal share is a result of the recent up-gradation of NH connectivity between Haridwar and Rishikesh. It was observed that out of total cleaned responses, only 0.9% of persons reported that they would start using the ring road. These users reported that most likely,

38

N. Kathait et al.

Fig. 5 Modal share from survey responses

they will use the metro, or they may use the metro. This number is well within the estimated range from the model (see Sect. 4.1).

4 Results As per the objective of this case study, the following two policy scenarios are considered: • Impact of Ring Road only on the metro ridership • Impact of NH up-gradation on the metro ridership. To estimate the changes in the metro shares for the two policy scenarios, the OD matrix is first flattened, where each row represents one OD relation. For each OD relation, the methodology described in Sects. 3.3 and 3.4 is used. The results are presented in the following two sections.

4.1 Impact of Ring Road on Change in Metro Share As an impact of the ring road in Haridwar, the joint metro share on HaridwarRishikesh and Rishikesh-Haridwar lines are likely to decrease in the range of 0.1– 1.4%, which is only marginal and as per the general intuition given the different alignments of the ring road and metro alignment.

Impact Assessment of Ring Road in Haridwar and National Highway …

39

4.2 Impact of NH Up-Gradation Between Haridwar-Rishikesh on Change in Metro Share As an impact of the up-gradation of NH connectivity between Haridwar and Rishikesh, the joint metro share on Haridwar-Rishikesh and Rishikesh-Haridwar lines is likely to decrease in the range of 3.5–5.0%, which is minor yet expected. However, in the future, with an increase in congestion, this effect is likely to diminish, i.e., the metro share may further increase.

5 Conclusion This work was carried out to assess the impact of Ring Road in Haridwar and the up-gradation of the National Highway between Haridwar and Rishikesh on the metro ridership, which was estimated in the DPR (ICRA 2017). Firstly, a model was developed using the OD matrix from the DPR (ICRA 2017), and two policy scenarios were developed. In addition to this, to assess the current situation (i.e., choices post-upgradation of the National Highway between Haridwar and Rishikesh), a questionnaire survey was planned. The questionnaire also included the question about the choice after Ring Road in Haridwar. The results from the model indicated that the introduction of the ring road is likely to reduce the metro share in the range of 0.1–1.4%, and NH up-gradation is likely to reduce the metro share in the range of 3.5–5.0%. The former is also confirmed by the survey results. The impact of both developments in the Haridwar-Rishikesh region is minor, and the decrease in the metro share may diminish as the congestion increases in the region. Overall, the change in the metro ridership may be attributed to the prediction variability inherited in the modeling and may be considered an insignificant variation in the metro ridership on the whole. This work uses open-source tools and thus is useful to study the impact of transportation infrastructure in other cities based on the available data. The present study has a few limitations. The first limitation is that auto and IPT modes are not included in the modal share due to the ambiguity in the usage provided, because based on the PAPI survey more than 30% of people use these modes. Therefore, it signifies that a large portion of the population uses IPT modes for daily travel which is significantly different from the reported share of auto mode, i.e., 5%, in the DPR. Therefore, it is redistributed in the remaining modes. In the future study, a rational modal share should be obtained and used in the model for better results. The second limitation is that the impact assessment is done only for the horizon year 2021. Third, similar to the calibration, the impact may be assessed for the whole Dehradun Metropolitan Region. Future work for the same or different areas may overcome these limitations using the proposed methodology.

40

N. Kathait et al.

References Agarwal A, Lämmel G, Nagel K (2018) Incorporating within link dynamics in an agent-based computationally faster and scalable queue model. Transp A Transp Sci Agarwal A, Ziemke D, Nagel K (2020) Calibration of choice model parameters in a transport scenario with heterogeneous traffic conditions and income dependency. Transp Lett Int J Transp Res Choudhary R, Ratra S, Agarwal A (2022) Multimodal routing framework for urban environments considering real-time air quality and congestion. Atmos Pollut Res Horni A, Nagel K, Axhausen KW (Eds) (2016) The multi-agent transport simulation: MATSim ICRA (2017) Traffic and transportation study for preparation of DPR for Dehradun, Rishikesh and Haridwar metro

Passenger Delay and Journey Time Reliability Analysis of Ferry Transport Across Waterway: Case Study of National Waterway-1, India Debabrota Das, Anuj Kishor Budhkar, and Ayush Patel

Abstract The ferry movements across the river Hooghly, constituting the National Waterway-1 (NW-1) of India, are very significant owing to infeasible road bridge crossings. Sometimes the passengers of the ferry stations across it have to experience delays that have not been studied from a transportation planning perspective. The objective of this paper is to study the ferry journey time, passenger delays, reliability, vessel capacity and their variation and to provide some suggestive measures to control or minimize delay and improve the efficiency of the ferries from an operator’s perspective. Videographic data of total of 138 ferries are collected across six locations in NW-1 and the passenger, as well as ferry arrival and departure, are noted. Based upon these travel time reliability in the form of buffer time index for the journey time as well as waiting time at the ferry station is calculated. Furthermore, these parameters are studied after a hypothetical increase in vessel frequency. It is observed that different reliability parameters, capacity, and delay decrease with the frequency which indicates that ferries with smaller passenger capacity and higher frequency are found to be more time and capacity-reliable, to cater for a surge in demand in peak hours. The study outputs can suggest some measures from the passengers’ and ferry operators’ perspectives. Keywords Water transport · Travel time reliability · Passenger delay · Waiting time · Ferry operation

D. Das (B) Department of Civil Engineering, Indian Institute of Technology Kanpur, Kanpur, India e-mail: [email protected] A. K. Budhkar Department of Civil Engineering, Indian Institute of Engineering Science and Technology, Shibpur, India A. Patel Department of Civil Engineering, Rajkiya Engineering College Azamgarh, Azamgarh, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_4

41

42

D. Das et al.

1 Introduction Waterways are the channels through which the ships and barges are allowed to navigate. Worldwide, major rivers like the Nile, the Rhine, and the Danube are important navigable inland waterways. In India, there are 111 National Waterways (NWs) under the Inland Waterways Authority of India (IWAI) out of which 13 NWs are operational for shipping and navigation and cargo/passenger vessels are moving on them (Ministry of Ports, Shipping and Waterways 2018). The National Waterway-1 or the NW-1 (Fig. 1) is a 1620 km long channel of The Ganga-Bhagirathi-Hooghly River system between Prayagraj and Haldia (Inland Waterways Authority of India 2021), passing through the state of Uttar Pradesh, Bihar, Jharkhand, and West Bengal making it the longest operable waterway in India (“National Waterway 1” 2021). As there are only a few major bridges across the NW-1, ferry movement is one of the important modes of transportation to cross the river. The passengers wait at ferry stations depending on the ferry frequency, causing a delay in their overall travel. Furthermore, they need to wait within the ferry too to start. A few studies have been conducted on the NW-1 but not according to the transportation perspective. Previous studies on inland waterways have shown the ship traffic characteristics (Xin et al. 2019), factors affecting ship accidents risk (Quy et al. 2006), ship traffic characteristics and environmental conditions on ship collision frequency (Weng et al. 2020), etc., but the delay of the passengers have not been studied from ferry/ship owners’ perspective. Therefore, there is a need to develop a detailed study of delays to the passengers in the waterway to provide some suggestive ways to minimize the delay and improve the efficiency and safety of the ferries. Hence in this paper, a systematic study of passenger waiting time and travel time across important ferry routes along the NW-1 is conducted. Furthermore, reliability and frequency are analyzed from passengers’ and operators’ perspectives.

Fig. 1 National Waterway-1 and Kolkata port (Source Google map)

Passenger Delay and Journey Time Reliability Analysis of Ferry …

43

2 Literature Review 2.1 Literature Review Related to Passenger Delay Delay at the stop is affected by stop characteristics, vehicle characteristics, traffic conditions, passenger volumes, etc. Stop characteristics include stop location, stop place, connectivity of the stop with other places and accessibility to the people. Vehicle characteristics include vehicle size, number of seats and number of passages or doors in the vehicle, allowance to the people to stand over the journey period after filling the seats, etc. Traffic conditions include the arrival rate of vehicles and people, service time, stop capacity, the volume of passengers and the size of belongings with the passengers (Huo et al. 2018), which will increase the loading and unloading time affecting the delay (Chen 1999). These results may be applicable for ferry transport also.

2.2 Literature Review Related to Travel Time Reliability Carrion and Levinson (2012) reviewed the value of travel time reliability of road transportation using centrality-dispersion (or mean–variance) and scheduling models and a meta-analysis was performed. The studies of Chen et al. (2003) show that people value both travel time and its predictability which is a good way to evaluate the benefits of a service. Pu (2011) discussed different travel time reliability measures such as 95th percentile travel time, standard deviation, per cent variation, buffer time index (BTI), planning time index, travel time index which asserts the Federal Highway Administration report (2006). Travel time reliability can be similarly calculated for ferries. From the study of Xin et al. (2019), it has been concluded that the ship traffic in the waterways follows a linear speed-density relation. In this study, the ship’s acceleration and deceleration data were taken into account by the extraction of ship traffic characteristics. The increase in the proportion of large-scale ships would result in a slow decline in theoretical channel capacity, though the human factors were not taken into account. But numerous researches have been conducted on the safety of the ship (Weng et al. 2018; Lin 1999; Weng and Li 2019; Altan and Otay 2018). Weng et al. (2020) carried out a study on the effects of ship traffic characteristics and environmental conditions on ship collision frequency using a dynamic ship domain model and found that the collision frequency is greater for the bigger flow, the bigger density, the narrower traffic lane width, and adverse weather conditions. Although there are many studies that have been conducted on the rivers, these are mainly located outside India. Since inland waterways in India are also very important and have mixed traffic conditions like roads, the applicability of the previous works should be studied. So, ferry traffic, delay as well as conflicting movement studies

44

D. Das et al.

should be conducted in inland waterways so that they can be useful and helpful for future research scopes.

3 Methodology 3.1 Data Collection As per the objective of this paper, passenger delay, capacity, ferry arrival, and departure need to be studied. For this purpose video data of ferry and passenger movement were collected at six different ferry stations along the NW-1, the details of which are illustrated in Table 1. It has been ensured that any external factors or climate (such as rain, storm, wind, artificial activities, pandemic, etc.) were not affecting the collection. The locations are depicted in Fig. 2a and a view from the imposed camera of one of the locations, Metiabruj-Najirgunj (location 5) is shown in Fig. 2b.

3.2 Data Extraction In this study, it is required to extract the data of passenger delay, the journey time of ferries across the river, vessel capacity and ferry arrival and departure frequencies. Table 1 Details of the locations of data collection Location

Name

Date

Time

High/ low tide

No. of ferries observed

Average flow (pass/h)

1

Naihati-Mechhuabajar, North 24 Pgrs, West Bengal (WB)

04/01/ 22

09:45–12:30

Low

36

440

30

473

17

280

8

226

33

355

16:00–17:45

High

05/01/ 22

09:15–12:15

Low

15:30–17:30

High

Shyamnagar-Telenipara, North 24 Pgrs, WB

07/01/ 22

08:00–12:30

Low

15:45–18:00

High

3

Serampore-Barrackpore, Hooghly, WB

27/08/ 21

09:15–11:15

Low

17:15–18:15

Low

4

Howrah-shipping Ghat, Howrah, WB

12/01/ 22

10:30–13:00

Low

5

Metiabruj-Najirgunj, Kolkata, WB

12/09/ 21

11:15–13:45

Low to High

9

1431

6

Budge Budge-Bauria, South 24 Pgrs, WB

10/09/ 21

11:45–14:00

High

5

70

2

Passenger Delay and Journey Time Reliability Analysis of Ferry …

a

45

b Location 1 Location 2 Location 3 Location 4 Location 5

NW-1

Najirgunj

Location 6

NW-1

Foot Bridge

Ferry

Jetty

Fig. 2 a Locations of data collection along the NW-1, b Details of location 5 of data collection Source Google maps

For this purpose, the minute-wise number of persons entering the jetty is counted from the collected videos of different locations. The number of belongings and the vehicles with the passengers such as large luggage, bicycles, motorcycles, or bikes are noted as these also possess some space and may affect the total capacity of the ferry. The arrival and departure times of each ferry in that station are also recorded. When a passenger enters the area of the ferry station, at first he/she may wait for the ferry. This waiting time is out-of-vehicle delay (OVD). After the ferry comes, he/ she boards it and again he/she has to wait for the ferry to start and leave the station, while other passengers are to access it. This waiting time is expressed as in-vehicle delay (IVD). The average OVD, IVD and the total delay, i.e., the sum for each minute are calculated and tabulated. The tidal conditions of the water of the NW-1 at that time (i.e., high tide or low tide) are also noted down. The total journey time of each of the ferries is also extracted from the collected videos. A sample data extraction table for the first 10-min data collected from location 5 is shown in Table A1 of the Appendix.

3.3 Descriptive Statistics of the Extracted Parameters The numbers of passengers entering the stations and their belongings/vehicles are found to be discrete values while the delays (OVD, IVD, and total delay), and the arrival and departure rates of ferries are observed as continuous. Weighted delay is the product of total delay (OVD + IVD) and the corresponding number of passengers entering the station in that minute. Average delay expressed in second and average flow of passengers in passengers/h is calculated as per Eqs. 1 and 2 respectively. The observed luggage and vehicle data would be used for future studies. ∑

Average delay, T D =

weighted delay total number o f passenger s

(1)

46

D. Das et al.

Average f low =

T otal N umber o f passenger s entering into jett y ) ( number o f minute obser ved 60

(2)

4 Analysis and Results 4.1 General Observations The ferries are found to maintain almost fixed frequency and fixed journey time for a particular location. However, the passenger arrival rate is not the same throughout the time of data collection. The capacities of the ferries are also not constant for all the locations. The waiting times of the ferries are affected by the movement of the ships/ barges along the waterway, thus the delays and overall journey time of the passengers are also increasing. The tidal water current is not much affecting the journey time. As soon as the ferry arrives it is observed that passengers hurry to board and alight. This is evident from observed flow levels to/from the ferry. When there is a ship crossing the ferry path, the arrival and departure times of the ferry are increased up to 15% and 32% of the average arrival and departure times respectively.

4.2 Reliability and Relevant Indices from Observed Data For reliability analysis, the following assumptions are taken into account: 1. Waiting times are assumed to be similar for jetties at both sides of the locations. 2. Passengers are arriving at a uniform rate within respective one-minute intervals. The total journey time of the passengers is the sum of the total delay and journey time. Average journey time (Eq. 3), 95th percentile journey time, BTI for journey time (Eq. 4), average delay (Eq. 1), 95th percentile delay, standard deviation (Eq. 5), and BTI for delay (Eq. 6) are shown in Table 3. ∑

J our ney times no. o f obser vations

(3)

95th per centile jour ney time − average jour ney time × 100 average jour ney time

(4)

verage jour ney time = Bu f f er time index f or jour ney time =

/ standar d deviation o f delay =

∑n m i=1

( )2 n i T Di − T D n−1

(5)

Passenger Delay and Journey Time Reliability Analysis of Ferry …

47

where n i = number of minutes; n = total number of persons; T D i = total delay at i th minute. Bu f f er time index f or delay =

95th per centile delay − T D TD

× 100

(6)

From Table 2, it is observed that standard deviations are within 94 and 44% less than average journey time and average delay respectively for all locations. Buffer time index for waiting delay is high because of interfering ship traffic along NW-1 for which ferries’ departures get affected. Therefore there is large unreliability in the waiting time and it needs to be addressed. The tidal effect on the journey time is not much significant for all locations. Table 2 Different reliability parameters of journey time and delay of all locations Location

Average frequency (s/veh)

Journey time of the ferry Average (s)

Standard deviation(s)

Waiting delay at the ferry station Buffer time index (%)

Average (s)

Standard deviation (s)

Buffer time index (%)

1

618

455

84

53.2

374

214

441

2

1115

404

78

28.7

280

267

254

3

1050

415

34

8.8

460

257

223

4

584

480

30

9.5

207

150

344

5

939

303

48

26.2

462

285

153

6

2443

465

66

13.7

1233

1050

474

Table 3 Different parameters of delay and capacity of fixed frequency in location 5 Particulars

Existing average arrival frequency Fixed frequency (min) (min:sec) 15:39

5

8

10

Maximum theoretical delay (s)

1115

270

450

570 870

462

148

234

290 425

270

450

570 870

Average delay (s) 95th percentile delay (s) Standard deviation of delays (s)

1168 285

15

83.6 135.5 170 256.1

Buffer time index of delay (%)

153

82.7 92.6

97

Average capacity (pass/ferry)

582

119

193

239 336

105

Standard deviation of capacity

138

1.3

3.6

4.6

15.4

48

D. Das et al.

4.3 Improvement of Total Journey Time (by Varying Frequency) Although journey time reliability cannot be improved due to route constraints, the waiting time can be more reliable by improving the frequency. An experiment was conducted considering the same passenger arrival rate as observed from the field, but ferries are departing at different frequencies than the existing average frequency. Maximum theoretical delay, average delay, 95th percentile delay, its standard deviation, BTI of delay, average capacity, and standard deviation are also calculated based on their sample size and the number of passengers in the corresponding minutes. The maximum delay is calculated assuming a passenger arrives just when the previous ferry departs therefore he/she has to wait for the duration equal to frequency. These calculations are conducted considering ferries departing at fixed frequency but different epochs, and the values are averaged out to consider variation in passenger arrival rate. The detailed calculation for location 5 is illustrated with the existing average frequency of ferry in the 2nd column of Table 3. The maximum theoretical delay (Dmax ) experienced by a passenger under each fixed frequency remains constant: Dmax = 60(n − 1) + 30; where n = fixed frequency of the ferry in minute. The plot of maximum theoretical delay, average delay, 95th percentile delay, the standard deviation of delays, BTI of delay, average capacity, the standard deviation of capacity for fixed as well as existing frequency for location 5 is provided in Fig. 3. From the plot, it is observed that maximum theoretical delay, average delay, 95th percentile delay and average capacity, standard deviations of delay and capacity decrease with frequency. For existing frequency, the capacity of the ferry is higher than varying frequency which indicates that smaller ferries with higher frequency are more reliable. Maximum theoretical delay (s) Average delay (s)

Particulars

1200 1000 800 600 400

95th percentile delay (s)

200 0

Frequency of ferry (min)

Standard deviation of delay (s) Buffer time index of delay (%)

Fig. 3 Plot of different particulars with varying and existing frequency for location 5

Passenger Delay and Journey Time Reliability Analysis of Ferry …

49

4.4 Inferences from the Change in Frequency Inference for the Travelers. The Standard deviation of waiting delay is reduced by 71%, 52%, and 40%, for 5 min, 8 min, and 10 min frequencies respectively that of the existing frequency. BTI is also reduced by 46%, 39%, and 37% respectively than that of existing. Inference for the Operators. If ferry frequency has to be improved, more vessels are required in the fleet with less vessel capacity. This will incur an additional cost on the operator, but the fuel cost per trip will reduce due to lesser vessel capacity. Considering an average journey time of five minutes, in one direction, to achieve frequencies of 5, 8, and 10 min, an additional 3, 1, and 1 vessel(s) are required respectively in addition to the existing two vessels existing in location 5 under consideration. But assuming proportionate variation in ferry operation cost and vessel capacity, the operation cost reduces by 65%, 43%, and 29% respectively.

5 Concluding Remarks In this paper, the trends of the ferries and the passengers have been studied and the key factors like average delay, maximum delay, journey time, and the effect of the big ships along the NW-1 on the ferries’ increased arrival and departure time are studied. Travel time reliability is calculated based on average travel time, 95th percentile travel time, standard deviation, and buffer time index. It is observed that the BTI for waiting delay is much greater than that of journey time. The large unreliability may be addressed by improving frequency. It is observed that frequency improved to 10, 8, and 5 min from the existing 15 min and 39 s and BTI reduces by 46%, 39%, and 37% respectively indicating significant improvement. The disadvantage will be the higher operating cost but it may be compensated by smaller vessels with less capacity for the operators. Therefore there is a huge scope in the improvement of the journey times, which can be conducted using following a schedule of ferry departure on time which will be broadcast to the regular users. The studies in this paper can be supplemented with detailed passenger management data at ferry boarding/alighting points, and the capacity of ferries. Furthermore, the effect of luggage and vehicles can be checked on boarding/alighting, and a comprehensive analysis of ferry movements can be studied.

Appendix See Table 4.

…

10

…

8

5

19

13

4

8

22

7

0

3

1

6

2

0

0

With luggage

9

20

12

6

4

5

13

3

6

7

12

1

19

0

2

Persons

Min

…

0

0

2

0

0

1

0

2

1

0

With cycle

Total

…

0

0

0

0

0

0

0

0

0

0

With motorcycle

178

…

18

27

28

12

23

8

19

16

8

19

No. of passengers

…

0

0

0

0

0

0

5

40

100

160

OVD (s)

Table 4 A sample data extraction table for the first 10 min of data collected from location 5

19

68

128

188

248

308

368

423

448

448

448

IVD (s)

…

68

128

188

248

308

368

428

488

548

608

Total Delay (s)

54,297

…

1224

3456

5264

2976

7084

2944

8053

7808

4384

11,552

Weighted delay (pass.s)

Ferry Departed at 10:38

Ferry Arrived at 03:10

Comment

50 D. Das et al.

Passenger Delay and Journey Time Reliability Analysis of Ferry …

51

References Altan YC, Otay EN (2018) Spatial mapping of encounter probability in congested waterways using AIS. Ocean Eng 164:263–271. https://doi.org/10.1016/j.oceaneng.2018.06.049 Carrion C, Levinson D (2012) Value of travel time reliability: a review of current evidence. Transp Res Part A Policy Pract 46(4):720–741 Chen L (1999) Research on level of service and service volume of bus reserved lanes. Doctoral dissertation, Master’s thesis, Institute of Civil Engineering, National Taiwan University Chen C, Skabardonis A, Varaiya P (2003) Travel-time reliability as a measure of service. Transp Res Rec 1855(1):74–79 Federal Highway Administration (2006) Travel time reliability: making it there on time, all the time. US Department of Transportation Huo Y, Li W, Zhao J, Zhu S (2018) Modelling bus delay at bus stop. Transport 33(1):12–21 Inland Waterways Authority of India. http://www.iwai.nic.in/waterways/national-waterways/nat ional-waterways-1?id=2523. Last Accessed 13 July 2021 Lin SC (1974) Physical risk analysis of ship grounding. In: MacDuff T (ed) The probability of vessel collision. Massachusetts Institute of Technology (1999), Massachusetts. Ocean Ind 9(1):144– 148 Ministry of Ports, Shipping and Waterways. Operational national waterways in the country (press release). https://pib.gov.in/Pressreleaseshare.aspx?PRID=1557459. Last Accessed 27 Dec 2018 National Waterway-1. Wikipedia. https://en.wikipedia.org/wiki/National_Waterway_1. Last Accessed 28 May 2021 Pu W (2011) Analytic relationships between travel time reliability measures. Transp Res Rec 2254(1):122–130 Quy NM, Vrijling JK, Gelder PHAJMV, Groenveld R (2006) On the assessment of ship grounding risk in restricted channels. In: The 8th international conference on marine sciences and technologies-black sea conference, Varna, Bulgaria Weng J, Liao S, Wu B, Yang D (2020) Exploring effects of ship traffic characteristics and environmental conditions on ship collision frequency. Marit Policy Manag 47(4):523–543 Weng J, Li G (2019) Exploring shipping accident contributory factors using association rules. J Transp Saf Secur 11(1):36–57. https://doi.org/10.1080/19439962.2017.1341440 Weng J, Yang D, Qian T, Huang Z (2018) Combining zero-inflated negative binomial regression with MLRT techniques: an approach to evaluating shipping accident casualties. Ocean Eng 166:135–144. https://doi.org/10.1016/j.oceaneng.2018.08.011 Xin X, Liu K, Yang X, Yuan Z, Zhang J (2019) A simulation model for ship navigation in the “Xiazhimen” waterway based on statistical analysis of AIS data. Ocean Eng 180:279–289

Is CAWI-Based Transit Passengers’ Satisfaction Survey Feasible in India?—Insights During COVID-19 Vishwajeet Kishore Verma and Rajat Rastogi

Abstract Present study examined the feasibility of conducting a CAWI-based survey for collecting transit commuters’ travel and satisfaction data when travel restrictions were imposed in a city. Four CAWI methods, namely, through email, text SMS, snowball chain-referral and organization source, were conducted during the COVID-19 lockdown in New Delhi, India. A web-based questionnaire was designed. Its link was shared through email and texting services with passenger and organizational databases. The following were observed: (a) high distribution and response losses with respect to email and SMS-based methods; (b) passenger’s bias in snowball sampling method due to chain-referral strategy; (c) nil response when individuals were contacted through email, and limited responses if contacted through organization; and d) incomplete responses in open ended and conditional questions. It was concluded that the methods have low feasibility, provide limited control on nonresponse quality, connect also with non-eligible respondents, and create low interest in the survey. These may reduce if the web-based survey is administered in physical mode. Keywords Public transport - Passenger Satisfaction Survey (PSS) · CAWI survey · COVID-19 · Non-response

1 Introduction Various modes of survey are being used by transit agencies to get responses from the passengers regarding the offered services and operations. These include personal interviews, telephonic surveys, self-administered surveys, internet surveys, etc. The V. K. Verma (B) · R. Rastogi Department of Civil Engineering, Transportation Engineering Group, Indian Institute of Technology Roorkee, Roorkee, India e-mail: [email protected] R. Rastogi e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_5

53

54

V. K. Verma and R. Rastogi

selection of survey mode depends on the study purpose, geographic and socioeconomic characteristics of the targeted passengers, and survey instrument type. Going by the study context, the passengers’ satisfaction with transit services is examined through Passengers’ Satisfaction Survey (PSS). PSS is done periodically by service providers to assess acceptability of service quality or to identify the improvements needed to the service quality delivered to the passengers (Verma and Rastogi, 2022). This helps the operator in getting insights on how to retain existing or add potential commuters to the transit. To get this idea, the survey can be conducted across geographical boundaries with or without a transit service. The methods of data collection can be - face-to-face interview, telephonic interview, questionnaire pick and drop, etc. Targeted respondents can be contacted onboard or at station (on platform or outside the station). However, these approaches of contacting respondents and conducting survey amid ongoing COVID-19 pandemic was not doable. COVID-19 pandemic had placed many of the Indian cities under lockdown, with restricted availability and usage of transport (with 30 or 50% occupancy) (UITP 2020; Subbarao and Kadali 2022). Such restrictions caused difficulty in contacting the passengers who would have been eligible for the survey. Also, the requirements of social distancing and personal safety against infection induced modal shift from transit to private modes (Das et al. 2021) thus reducing ridership and the possibility of getting adequate responses to work in conclusive manner. Therefore, understanding passengers’ perception towards transit service during operation of limited transit services was difficult and the demand of the time was to formulate a survey methodology which can fetch responses while maintaining the COVID protocols (Dong et al. 2021; Abdullah et al. 2020). A look at the survey methods indicated that contrary to the use of pen and paperbased personal interview, the computer-assisted survey provides seamless connect between interviewee and the targeted respondents (Web-Based Survey Techniques 2006; Greenlaw and Brown-Welty 2009). The respondents can use their leisure time to respond to the questionnaire based on their previous travel experience (Schleyer and Forrest 2000; Couper 2011). However, survey participation, response rate, and completion to the web-based survey are influenced by numerous factors like respondents’ socioeconomic characteristics and their access to internet and smart devices, questionnaire length and language, survey purpose, questionnaire delivery method (email, text message, cards having survey URL) and many more. Table 1 presents a glimpse of the response rate and completion rate from web-based survey-related studies conducted at different locations having different survey objectives. From the table, varying levels of response rates (9–44%) and survey completion rates (21–98%) are observed. On literacy, age-group and internet penetration aspects, it is reported that younger respondents, higher socioeconomic status, technical knowledge ability, etc. are indirect success measures for an online-based survey (Dillman et al. 2014). Nevertheless, considering these aspects, the authors explored the feasibility of Computer Assisted Web Interview (CAWI) survey in New Delhi, India. The city has the highest internet penetration rate (68%) compared to national average of 45% (Keelery 2021). The survey intended to examine the passengers’ satisfaction with

Is CAWI-Based Transit Passengers’ Satisfaction Survey Feasible …

55

Table 1 Response rate and survey completion rate of web-based survey Survey objective

Survey location

Response rate (in percent)

Completion rate (in percent)

References

Economic status of residents

American Evaluation Association Group, USA*

52

98

Greenlaw and Brown-Welty (2009)

To address nonprofit financial strategies and organizational performance

New Jersey, USA*

44

21

Lin and Ryzin (2012)

Bus passengers’ perception survey

San Francisco, USA

9

76

Agrawal et al. (2017)

Survey to victims of laws and crime

Finland

25

–

Laaksonen and Heiskanen (2014)

Perceptions surveys for various projects related to different transportation services, ride info apps and utilities

Madrid, Helsinki, London, and Vitoria-Gasteiz

14–21

64–80

Monzon et al. (2020)

*

Organization level survey link distribution

services made available in public transport like Delhi Metro and Delhi Transport Corporation (DTC) bus services.

2 Survey Methodology Before conducting CAWI, a web-based questionnaire was designed to collect satisfaction scores of public transport passengers. It consisted of four parts sequentially related to passengers’ travel information, passengers’ satisfaction with service quality in transit, change in satisfaction level during COVID-19, and socioeconomic information of respondents. While designing the questionnaire, due importance was given to question type and format (for response effort, closed form, minimal writing, field coded), choices (binary), order and grouping (question order, choice order, information order, grouping by possible intent) and scale effect (mixed use of nominal and ordinal scale). Likert scale was presented as a slider with a range of 0–10. Bilevel pilot testing of questionnaire was done to examine the issues with respondents’ understanding and responses made. Modified questionnaire was placed on a cloud and a shareable link was generated. The link was processable on computers, laptops and mobile phone, and allowed remote access and processing of the respondents’ responses along with required download on a remote computer. Alongside, a database was prepared, with assistance from certain agencies, of eligible population in New Delhi, India. This included name, email, contact

56

V. K. Verma and R. Rastogi

and gender. The survey was conducted during June–July 2021. During that time the city was under lockdown and limited travel options were available (Velmurugan et al. 2023). Four CAWI Survey methods, namely, email (E), text messages (TM), snowball (S), and organizational level (OL) were used. Except for snowball method, random sampling was used to contact the targeted respondents. For snowball method, Goodman’s snowball-based reference sampling technique was used (Goodman 1961). Questionnaire link was shared through email and WhatsApp messaging interface with the respondents. Cover letter describing survey and eligibility criterion was also sent through email and requests for responses were also posted through SMS. Survey requests were floated to respondents and organizations during the daytime from morning 10 am to evening 06 pm on weekdays as well as weekends.

3 Insights from CAWI Survey Methods The insights from the four CAWI methods on the responses received on the questionnaire are discussed with respect to the survey distribution and response losses, causes of nonresponses, issues in modelling nonresponses and biases if present.

3.1 Survey Distribution and Response Losses Outcomes of each of the four methods with respect to the questionnaire distribution and respondents’ participation proportion are given in Table 2. Out of the requests made, a substantial number of the email IDs and WhatsApp IDs were found either invalid or not active. This is defined as distribution loss. Out of successfully distributed requests, the respondents who clicked on the survey link and responded were considered as successful responses and defined as response success. The distribution success is the proportion to which the web link was successfully delivered, and response loss was the proportion of respondents who did not participate in the survey. It can be noted that E-method had 38% distribution loss, whereas it was 50% in TM-method. This indicates that people stick to their email IDs reasonably but switch their cell numbers more frequently, or it may be associated with entry of respondents’ erroneous or invalid contact details while preparation of their database, or the database is not up to date (Himelein et al. 2020). This caused higher distribution loss in TM-method. It was nil in rest of the methods. This might be due to the use of chain-referral strategy in S-method wherein the information on survey was passed on to friends and relatives by the receiver. In spite of the better distribution success in E-method, the response success was nil. This might be due to unfamiliarity of persons holding emails with the surveyor, lack of trust in absence of physical contact and valid personal ID of the surveyor, and the psychological fear of revealing travel information to the unknown person, or maybe request delivered

Is CAWI-Based Transit Passengers’ Satisfaction Survey Feasible …

57

Table 2 Insights from CAWI methods Method

Shared

Invalid IDs

Distributed (in percent)

Response (in percent)

E

3200

1213

1987 (62.1)

Nil (0)

3452

TM

6857

3405 (49.6)

50 (1.5)

S

42*

0

42 (100)

15 (35.7)

OL

14 Organizations

–

–

1–10 (7.14)

*

18 (Start of Chain) + 24 (1st contact of start of chain)

to the spam folder (Web-Based Survey Techniques 2006). Response success was negligible in TM-method too. It was reasonable in S-method (36%) which may be due to familiarity across the respondents connected through a chain. In case of OLmethod, 10 responses were received but all were from a single organization which is 7.14% response success rate. However, New Delhi failed to draw large response rates overall.

3.2 Why High Non-responses? The low or negligible responses in online surveys had been observed in other studies too. Kagerbauer et al. (2013) found that CAWI survey methods possess less response rate than Computer Assisted Telephonic Interview (CATI) and conventional paperpen-based surveys. In general, non-response in any form of surveys is attributed to social exchange theory and leverage salience theory (Pani and Sahu 2019; Hostert and Griffiths 2012). First one is related to respondents’ lack of trust in the survey or surveyor, and second is related to the population sub-groups which have different response tendency. In the present study, these causes got coupled with the mobility service disruption due to COVID-19, which enhanced the non-responses highly. In TM-method, 43 out of 50 respondents explicitly indicated the reason for nonresponse. These are shown in Fig. 1. Majorly, it was either not being targeted respondent as enumerated in the survey or not interested to respondent. The blind contact made in the CAWI format of survey increases the propensity of contacting the persons who might be choice riders or attitudinally personal vehicle users. They will have least of the information on transit usage and hence found themselves not eligible to respond to the survey. In generalized form, the targeted respondents for a survey will be missing if the survey is not administered at a location of interest, say in the present context at transit station or inside the transit. Further, during COVID-19 lot many restrictions were imposed on transit usage, there were fears of getting in contact with an infected person during travel and many of the employed were at home either due to loss of job or work from home. This is also clear from Fig. 2 which indicates that around 75% of respondents’ either decreased trip frequency or remained at home completely. These might have made the possible eligible respondents uninterested in the survey. S-method violated the social exchange theory due to survey referral

58

V. K. Verma and R. Rastogi

Fig. 1 Reasons of non-responses from TM-method

Fig. 2 Trip frequency impacted due to COVID-19

programme which involved friends and relatives and helped in getting better response success. In OL-method, the organization whose employees responded was a research centre which promotes such survey-based research. Hence, some better response rate was observed in this case.

3.3 Completeness and Biasedness in Responses The length of questionnaire affects its completeness (Burchell and Marsh 1992; Berdie 1973). However, the completeness of the survey is entirely dependent on respondents’ interest in the survey. In the present study, 66% of the received responses were found complete, and rest of the responses were partially complete. This syncs well with the survey completion rate presented in Table 1. Further, Table 3 presents questionnaire’s section-wise nonresponses across the survey methods. The nonresponses had been found relatively lower in section-1 which is related to the travel information. Non-response increased in section-2 which was related to capturing satisfaction scores with respect to the service attributes of transit. Rest of the

Is CAWI-Based Transit Passengers’ Satisfaction Survey Feasible …

59

two sections had the same non-response rate. It shall be noted here that respondents seemed to be unwilling in sharing their socioeconomic information sought in section 4. Negligible usage of transit or travel during lockdown period might have impacted the responses in the first three sections. Figures 3, 4 and 5 show non-response to the individual questions in section-1, section-2, and combined sections-3 and 4 respectively. In section-1, non-response was relatively higher for questions like pass validity, service type, trip purposeother and access mode-other. Surveyor’s intervention on these questions might have improved the response rate as two of the questions were open-ended. In section-2, questions related to safety and parking were mostly unanswered. Though parking may be irrelevant to some of the passengers (accessing transit by mode other than personal vehicle), but safety shall not be the case. This is something different than normal belief . In rest of the questions in section-2 and all the questions in section-3 and 4, the non-response has been found almost same across the questions. Section 3 was related to change in satisfaction score if the respondent travelled using transit before and during COVID-19. As there was big time lag when the respondent travelled last (memory lapse) and limited exposure to transit during COVID-19, the comparison might become difficult and translated in uniform non-response to the questions. Further, many of the respondents might not have continued the survey till last, hence there is non-response to personal questions presented towards the end of the survey. This is in line with the observations made by Richardson et al. (1995) and Dillman et al. (2014). Table 4 shows socioeconomic characteristics of respondents who participated to survey through TM, S and OL-method. From the table, it can be observed that for any method, around 70 percent of respondents were at least graduates, 70 percent were of age between 15 and 45 years who largely were either student or had middlelevel jobs. Thus, it can be inferred that though the survey has observed participation from higher socioeconomic respondents, still a large section of potential respondents does not respond to survey irrespective of the method tried. This has caused high non-responses in E and TM-methods too. Male participation was found high in TM and OL method compared to S-method which showed trend opposite to the usually reported participation rates by gender. It showed that female friends and relatives responded more, may be during their leisure time. Participation of respondents with school education only was lower and those with higher education was higher in OL-method which is self-explanatory for an organizational employee breakup. In Smethod, the proportion of graduates was quite high which is not as per normal norms. Better representation by occupation was observed in OL-method, but it was skewed Table 3 Overall non-responses in data Survey method

Section 1

Section 2

Section 3

Section 4

TM

27.8

39.5

34.0

34.0

S

23.6

49.2

46.6

46.6

7.4

17.1

10.0

10.0

OL

60

V. K. Verma and R. Rastogi

Fig. 3 Non-response in section-1 of survey questionnaire

Fig. 4 Non-response in section 2 of survey questionnaire

in TM-method (towards middle-level employees) and S-method (towards students). This indicated that control on category of survey respondents is difficult in TM and S-methods, and it becomes representative by itself in OL-method due to employee structure. The representative sample by vehicle ownership was available in TM and OL-method but was skewed towards ‘vehicle occupancy = Nil’ in S-method. It seems that respondents have looked at this attribute as self-ownership rather than household ownership. This is clear from the response in S-method wherein 50% were students and obviously 50% did not own a vehicle. Two issues cropped here, one the absence of surveyor caused non-responses which otherwise could have been rectified during survey itself through proper explanation, and second, the S-method is dependent

Is CAWI-Based Transit Passengers’ Satisfaction Survey Feasible …

61

Fig. 5 Non-response in sections 3 and 4 of survey questionnaire

upon the selection of respondent by the receiver or survey operator. This caused biasness in the responses. The same was reiterated as limitations or advantages of snowball sampling depending on the research objectives by researchers (Salganik and Heckathorn 2004; Heckathorn 2002). As a result, it can be said that the E and TM-methods are not effective in metropolitan cities for the reason being cited above.

3.4 Modelling the Non-responses There is research on development of non-response driven prediction models for a survey. These models help in studying and predicting the potential causes of non-responses. Statistical and Machine Learning (ML) modelling techniques are usually deployed by researchers to predict the survey non-response. Pani et al. (2019) majorly created prediction models for Establishment-Based Freight Surveys (EBFS) by using several ML (supervised and unsupervised) predictor models. However, unlike EBFS, development of non-response prediction models for transit commuters is difficult. In EBFS survey, the characteristics of EBFS, their location, freight commodity, their business models, etc. are available beforehand. These variables act as response variables, and respondents’ response to participate in the survey act as predictor variable (binary) in the ML prediction technique. However, in case of transit commuters’ survey, neither pre-information of respondent characteristics, previous similar surveys nor response patterns in the area were available. Due to this, the non-responses in transit service satisfaction survey are difficult to model. This, off-course, is possible in the organization-based surveys.

62 Table 4 Respondents’ socioeconomic characteristics as per TM, S and OL-method

V. K. Verma and R. Rastogi

Characteristics

TM-method S-method OL-method

Gender Male

63.6

37.5

66

Female

36.3

62.5

34

–

–

–

Don’t want to reveal Age group 15–25

42.4

50.0

33.3

26–45

48.5

50.0

33.3

46–60

9.1

–

33.4

–

–

–

Up to 12th

24.3

25.0

11.2

Graduate

39.4

62.5

44.4

PG or higher

36.3

12.5

44.4 11.2

Education Illiterate

Occupation 36.4

50.0

Unemployed

Student

3.0

–

Supporting staff

–

3.0

12.5

22.2

Middle level employee 39.4

25.0

33.3

15.2

12.5

33.3

3.0

–

2-Wheeler

18.2

37.5

11.2

Car

30.3

–

33.3

Both

24.3

12.5

22.2

None

27.2

50.0

33.3

Higher level employee Retired

–

Vehicle ownership

4 Conclusions The usability of the CAWI mode of surveys was examined to collect the responses from the users of any transit system. The survey was floated in New Delhi and the obvious choice as transit was Delhi Metro and DTC. As the time was impacted by COVID-19 spread there were city and area lockdowns. Transit was operating services with limited occupancy. Fears of spread of infection through contact were dominant,

Is CAWI-Based Transit Passengers’ Satisfaction Survey Feasible …

63

and most of the working population was working from home. This impacted negatively the retention of memory on previous travel experiences. Under these circumstances, on conducting CAWI-based transit service satisfaction survey, the following were observed: 1. Validation of the database containing respondents’ information is essential prior to the conduct of the survey. Same is recommended by Lin and van Ryzin (2012). This would reduce distribution loss. 2. Remote contact increases the propensity of selection of untargeted population which further increases the possibility of non-responses and distribution losses. 3. The engagement and response rate also gets affected negatively due to survey emails treated as e-marketing emails or getting them dumped in a spam folder. This is being warned by Das et al. (2021) and Heckathorn (2002). 4. Lack of trust, psychological fear of revealing self-information related to socioeconomic and day-to-day travel, issue of reliability of surveyor due to absence of physical contact and valid IDs, and possibility of hacking the system through unknown links also affected the response success and resulted in higher response loss. 5. Unlike emails, WhatsApp provides prompt and one-to-one interaction. Addressing request direct through the respondent’s name is found appealing and resulting in survey acceptance. But few complained issues regarding their privacy. 6. Females were more courteous and participated at higher rate than males in TM and S-methods. In the first method, it might be due to lack of awareness with technology related offenses by external and unknown contacts, and in second method it might be due to receiving requests from known contacts. 7. Familiarity with the sender or surveyor or starter of the chain emulated higher response rate in S-method. However, its effect got diminished as the chain progressed. Such issues are already reported by researchers Subbarao and Kadali (2022) and UITP (2020). 8. Higher non-response to personal information might be an outcome of loss of interest in the survey because of its length. This is supported by Richardson et al. (2006) and Dillman and Brown-Welty (2009). 9. The conduct of survey during pandemic indicated that responses to the memorybased replies, open-ended questions, contextual comparisons based on personal experiences got affected the most and thus resulted in item or data-based nonresponse. This was also true for the situational attributes (say safety and security at different levels) which lost their relevance and importance as respondents were at home for a long period and rated them immaterial. Intervention by a surveyor during physical contact might have reduced the item-based nonresponse. 10. The importance of conducting the contextual surveys at location of attraction through any feasible mode of survey got strengthened as this will not cause non-response due to contact with ineligible respondent. Eligible respondents

64

V. K. Verma and R. Rastogi

found at such locations will also be interested in providing responses and thus will reduce non-response due to lack of interest. 11. The representativeness of the survey sample is affected in CAWI surveys. It may get translated by itself in OL-methods, might be due to organizational structure and instructions passed on from primary organizational contact. Coverage bias is reported as a serious issue associated with the online survey methods wherein target population would not be covered in the sampling frame Web-Based Survey Techniques (2006). Analysis revealed that though CAWI mode seems superior to reach the targeted respondents, but it lacks respondents’ response and response quality. The study concludes that CAWI-based survey methods need to be examined under unrestricted conditions once the COVID-19 lockdown impacts get subsided.

References Abdullah M, Dias C, Muley D, Shahin M (2020) Exploring the impacts of COVID-19 on travel behavior and mode preferences. Transp Res Interdiscip Perspect 8. https://doi.org/10.1016/j. trip.2020.100255 Agrawal AW, Granger-Bevan S, Newmark GL, Nixon H (2017) Comparing data quality and cost from three modes of on-board transit surveys. Transp Policy 54:70–79. https://doi.org/10.1016/ j.tranpol.2016.06.010 Berdie DR (1973) Questionnaire length and response rate. J Appl Psychol 58:278–280. https://doi. org/10.1037/h0035427 Burchell B, Marsh C (1992) The effect of questionnaire length on survey response. Qual Quant 26:233–244. https://doi.org/10.1007/BF00172427 Couper MP (2011) The future of modes of data collection. Public Opin Q 75:889–908. https://doi. org/10.1093/poq/nfr046 Das S, Boruah A, Banerjee A et al (2021) Impact of COVID-19: a radical modal shift from public to private transport mode. Transp Policy 104743. https://doi.org/10.1016/j.tranpol.2021.05.005 Dillman DA, Smyth JD, Christian LM (2014) Internet, phone, mail, and mixed mode surveys, fourth Dong H, Ma S, Jia N, Tian J (2021) Understanding public transport satisfaction in post COVID-19 pandemic. Transp Policy 101:81–88. https://doi.org/10.1016/j.tranpol.2020.12.004 Goodman LA (1961) Snowball sampling Greenlaw C, Brown-Welty S (2009) A comparison of web-based and paper-based survey methods: testing assumptions of survey mode and response cost. Eval Rev 33:464–480. https://doi.org/ 10.1177/0193841X09340214 Heckathorn DD (2002) Respondent-driven sampling II: deriving valid population estimates from chain-referral samples of hidden populations. Soc Probl 49:11–34. https://doi.org/10.1525/sp. 2002.49.1.11 Himelein K, Eckman S, Lau C, Mckenzie D (2020) Mobile phone surveys for understanding COVID-19 impacts: part i sampling and mode. In: World bank blog. https://blogs.worldb ank.org/impactevaluations/mobile-phone-surveys-understanding-covid-19-impacts-part-i-sam pling-and-mode. Last Accessed 5 Jul 2022 Hostert P, Griffiths P (2012) Handbook of survey methodology for the social sciences. Springer, New York, NY

Is CAWI-Based Transit Passengers’ Satisfaction Survey Feasible …

65

Kagerbauer M, Manz W, Zumkeller D (2013) Analysis of PAPI, CATI, and CAWI methods for a multiday household travel survey. In: Transport survey methods: best practice for decision making, pp 289–304. Emerald Group Publishing Limited Keelery S (2021) Internet usage in India Laaksonen S, Heiskanen M (2014) Comparison of three modes for a crime victimization survey. J Surv Stat Methodol 2:459–483. https://doi.org/10.1093/jssam/smu018 Lin W, van Ryzin GG (2012) Web and mail surveys: an experimental comparison of methods for nonprofit research. Nonprofit Volunt Sect Q 41:1014–1028. https://doi.org/10.1177/089976401 1423840 Monzon A, Julio R, Garcia-Martinez A (2020) Hybrid methodology for improving response rates and data quality in mobility surveys. Travel Behav Soc 20:155–164. https://doi.org/10.1016/j. tbs.2020.03.012 Pani A, Sahu PK (2019) Modelling non-response in establishment-based freight surveys: a sampling tool for statewide freight data collection in middle-income countries. Transp Policy. https://doi. org/10.1016/j.tranpol.2019.10.011 Richardson A, Ampt E, Meyburg A (1995) Survey methods for transport planning Salganik MJ, Heckathorn DD (2004) Sampling and estimation in hidden populations using respondent-driven sampling. Sociol Methodol 34:193–240. https://doi.org/10.1111/j.00811750.2004.00152.x Schleyer TKL, Forrest JL (2000) Methods for the design and administration of web-based surveys. J Am Med Inform Assoc 7:416–425. https://doi.org/10.1136/jamia.2000.0070416 Stopher P, Stecher C (2006) Travel survey methods quality and future directions. Elsevier Ltd. Subbarao SSV, Kadali R (2022) Impact of COVID-19 pandemic lockdown on the public transportation system and strategic plans to improve PT ridership: a review. Innov Infrastruct Solut 7 UITP (2020) Covid-19 has hit Indian public transport from many directions. In: UITP. https://www. uitp.org/news/covid-19-has-hit-indian-public-transport-from-many-directions/. Last Accessed 20 Apr 2022 Velmurugan S, Padma S, Advani M, Sharma R, Singhal R, Patel C, ... & Bhuyan PK (2023) Transportation amid pandemics impact of COVID-19 on transportation in urban India.Elsevier 275–292 Vishwajeet, Verma R, Rastogi M, Parida A, Maji S, Velmurugan A, Das (2022) Proceedings of the fifth international conference of transportation research group of India5th CTRG Volume 3.An overview of approaches and methods for evaluating public transport performance 21– 45.Springer Nature, Singapore Web-Based Survey Techniques (2006) Transportation Research Board, Washington, D.C.

Joint Versus Standalone Estimation of First/Last Mile Mode Choice Utilising Revealed and Stated Preference Datasets B. S. Manoj

and Arkopal Kishore Goswami

Abstract In recent years, revealed preference (RP) and stated preference (SP) surveys have been extensively employed in variety of sectors where consumer choice is involved so as to estimate demand for a service/product. This paper focuses on investigating the differences in modelling outputs arising from the use of RP, SP and joint RP-SP datasets. A case study of first and last mile mode choice to the upcoming metro rail stations in Mumbai is considered. Data was collected from random individuals residing in the adjoining neighbourhoods of five upcoming metro stations. Multinomial logit (MNL) models are estimated considering only RP data, only SP data and joint RP-SP data. Results show that estimates of several parameters of the standalone SP and RP models either overestimate or underestimate the propensity of using an alternative. However, the joint model provides intuitive parameter estimates, and the overall goodness of fit of the joint model is significantly greater than stand-alone models. Joint models have a greater number of variables explaining the choice behaviour when compared to standalone models. Also, the joint RP–SP model enables in developing the scale factor, which accounts for the temporal differences between the RP and SP data, and eliminates the hypothetical biasness and behavioural differences that could have a major impact on the findings of the standalone RP and SP models. Keywords Stated preference data · Revealed preference data · Choice model · First and last mile · Multinomial logit model

B. S. Manoj (B) · A. K. Goswami Ranbir and Chitra Gupta School of Infrastructure Design and Management, Indian Institute of Technology Kharagpur, Kharagpur 721302, India e-mail: [email protected] A. K. Goswami e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_6

67

68

B. S. Manoj and A. K. Goswami

1 Introduction Discrete choice models have been employed widely to predict commuters travel mode using RP data (Morikawa 1994). Most of the models using RP data predict the choice probabilities of the existing alternatives. In a few instances, these models predict the likelihood to switch to new alternatives or to continue with the existing choice of alternatives. However, change in the choice of alternatives in connection with the introduction of new transportation service(s) is of major interest to many researchers and policymakers (Ben-Akiva and Morikawa 1990). Valuable insights regarding this change of mode choice or switching behaviour can be obtained from SP (panel) survey data. The respondents are presented with hypothetical scenario(s) about the future changes in the transportation system or about a new upcoming transportation service and are asked how they would change their choices. One such example is presented in this paper where a survey was conducted to estimate the individual’s choice of last mile mode by asking the respondents to state their travel mode choice when an upcoming metro station becomes operational near their neighbourhood. It is generally considered that the decision-making resulting from stated preferences may differ from the one that individuals actually carry out, as it is widely assumed that a survey of stated intents would give responses with considerable bias and huge random errors (Morikawa 1994). For estimation of mode choice when a new transport service is not yet introduced, SP data sets have a few advantages over the RP datasets, such as (a) RP data are consistent with current behaviour of the respondents; (b) SP data can consider hypothetical or non-existing services or choice or alternatives; (c) Trade-off between the alternatives of choices are clearly visible (Ben-Akiva et al. 1992). However, one of the most prominent issues of SP surveys is that, while they are capable of imitating the real world through careful design, this does not mean they truly represent it. A similar remark might be made about RP surveys as well, which are plagued with non-available alternatives and measurement inaccuracy in general. The issue of realism has been largely related with the use of SP data in prediction, where alternative specific constants and scale must be adjusted to true market shares or some assumption must be made to foresee new alternatives that are not yet available in the market (Cherchi and Hensher 2015). Hensher (2004) in his paper highlights the relevance of collecting SP data from the respondents which likely results in complex heterogeneity of choices. The typical way to reduce the cognitive burden from the respondents is to design the SP questionnaires as blocks, so as to reduce the number of scenarios. However some researchers (Cherchi and Hensher 2015) even question the block design method. It is important to ensure that the combination of attribute levels presented is realistic and avoid keeping dominating alternatives, to make it a more realistic of sets of choices. It is easy to pick any dominating alternative in a choice set (Bliemer et al. 2017). Although RP data is preferred over SP data because of the data validity issues, both data types have their unique advantages and disadvantages. To overcome the shortcomings of each type of data separately, several researchers have developed methods for combining RP and SP data. Researchers use RP and SP data to construct

Joint Versus Standalone Estimation of First/Last Mile Mode Choice …

69

a combined model by assuming that people make similar trade-offs between significant attributes like price and quality (Morikawa 1994) which imitates user behaviour. The fundamental concept behind this approach is to use RP data as the benchmark for comparison to collect information about market equilibrium, while using SP data to enhance attribute trade-offs that are unavailable in RP data. According to Ben-Akiva et al. (1992), the main benefit of combining RP and SP data is efficiency, i.e. by increasing the number of observations and using all available data. In addition, the respondents’ bias correction is reduced in SP data by using actual market insights from RP data. Furthermore, in RP-SP combined data helps in identification by estimating attribute trade-offs and the consequences of new modes that aren’t apparent from RP data (Bradley and Daly 1992). To eliminate the possibility of hypothetical bias, experts may need to create highly simple survey tasks that respondents must complete in a short amount of time, rather than putting them under a lot of stress and jeopardising the quality of their responses. Individual respondents have limited capacity to process the information and do not always put much effort while choosing an alternative in the survey. In terms of data combination methods, Dissanayake and Morikawa (2002) created an RP-SP coupled Nested Logit model to construct two-level choice of vehicle ownership and mode choice for developing nations. Morikawa et al. (2002) developed a joint RP-SP discrete choice model, integrated with linear structural equation, considering latent attributes and also using attitudinal data as indicators. Hass et al. (2018) developed mode choice estimation models using cross-sectional survey and panel data. Using integrated RP-SP data, Meister et al. (2022) built a MDCEV model to simulate mode choice behaviour during the COVID-19 pandemic in Switzerland. Many scholars have developed ideas that attempt to understand user behaviour by capturing user sentiments. Ajzen (1991) used the theory of planned behaviour to investigate user attitudes toward various modes. Likewise, several investigations have been developed using different data combinations. Research on mode choice is not only limited to RP-SP data, but with the evolution of big data, mode choice models for ride hailing services (Song and Zhong 2020), travel demand analysis of different modes (Vij and Shankari 2015) have been estimated using such data. However the joint RP-SP data is widely used for access mode choice (Morikawa 1994; Ben-Akiva and Morikawa 1990; Yang et al. 2013). The major drawback in these methods is the potential correlation between the RP and SP responses that are collected from the same individuals. This paper addresses the standalone models of RP and SP data and the joint RP-SP model and its implications on the outcomes.

2 Combined Estimation with RP and SP Data Commuters choose a mode of travel based on the principles of utility maximisation wherein respondents choose the mode with the highest utility. According to random utility theory, the utility of an alternative ‘i’ for a person ‘n’ is a utility function with

70

B. S. Manoj and A. K. Goswami

attributes and characteristics of person and will have a form: u n (i ) = u(xin , sn )

(1)

where x in is the utility attributes of the selection ‘i’ for a traveller ‘n’ and S n is the travellers’ characteristics that differs for each traveller ‘n’. The above utility function Eq. (1) is simplified by McFadden (1973) in a linear form, which is expressed as follows: u(xin , sn ) = V (xin , sn ) + εin

(2)

where V is the deterministic component of random utility function and εin is the unobserved attribute that generally represents the random error term. The RP data represents actual behaviour of the traveller but it cannot replicate the travellers’ preference in case of introduction of new mode, and may cause model errors due to correlation issue (Qiao et al. 2016). The multinomial logit model (MNL), which is based on the assumption of independently and identically distributed error terms εin , has been majorly used for the mode choice modelling, when choices of modes are greater than two. Hence the MNL model formulations used in this study are as follows. The general utility equation and probability of RP MNL model is written as shown below. RP RP RP u RP (xin , sn ) = V RP (xin , sn ) + εin RP eVn,i Pn (i, RP) =  JRP Vn, j j=1 e ⎛ ⎞ N RP   ⎝L = PiqRP ⎠

(3)

(4)

(5)

n=1 Aiε A(q)

Equation (5) represents the maximum likelihood function of the RP MNL model where the co-efficient of the RP model will be estimated. Alternately, SP data have no correlation issue and represents travellers’ choice under hypothetical scenarios (Wang et al. 2000; Ahern and Tapley 2008). The general utility equation and probability of SP MNL model is shown below. Equation (8) represents the maximum likelihood function of the SP MNL model where the coefficient of the SP model will be estimated.  SP  SP SP u SP xin , Sn = V SP xin , Sn + εin

(6)

SP

e Vn,i

Pn (i, SP) =  JSP

j=1

e Vn, j

(7)

Joint Versus Standalone Estimation of First/Last Mile Mode Choice …

⎛

⎞

N   SP

⎝L =

71

PiqSP ⎠

(8)

n=1 Aiε A(q)

The random components of the utility functions of both the RP and SP models are assumed to be independent with means as zero. The main difference between the RP and SP models is the data generating process of RP and SP model, which is represented by variance. The differences between the variances of error terms between RP and SP can be represented as the function of variance of error (σ ) terms between RP and SP data 2

2 σRP = μ2 σ SP

(9)

where µ is the scale co-factor. Subsequently, combined RP-SP model was proposed by Ben-Akiva and Morikawa (1990). After adopting the formulation for RP and SP data the utility equation can be written as follows:  RP  RP RP u RP xin , Sn = V RP xin , Sn + εin

(10)

 SP   SP SP μ ∗ u SP xin , Sn = μ ∗ V SP xin , Sn + εin

(11)

The probability of a respondent choosing an alternative i for RP and SP data would be e Vn,i RP Pn (i, RP) =  jRP Vn, j j=1 e

(12)

and eμVn,i

SP

Pn (i, SP) =  JSP

j=1

(13)

eμVn, j

The coefficients of the joint RP-SP (β, μ) model are estimated using the maximum likelihood method using software Apollo (Hess and Palma 2019) with the hypothesis that the two RP-SP probabilities are independent. ⎛ L(β, μ) = ⎝

RP N 



n=1 Ai ε A(q)

⎞⎛ PiqR P ⎠⎝

N   SP

n=1 Aiε A(q)

⎞ PiqS P ⎠

(14)

72

B. S. Manoj and A. K. Goswami

3 Empirical Analysis 3.1 Data Collection and Survey Design Data was collected in the city of Mumbai, which is a part of the Mumbai Metropolitan Region (MMR). Mumbai has a population of 12.5 million people, making it one of the most densely populated (76,790 per sq. mile) cities in the world. Mumbai’s public transportation system includes the suburban rail system/local train, bus services, taxis and autorickshaws. The suburban rail network delivers about 75 lakh passengers to and from the MMR region on a daily basis. Every day, the BEST (Brihanmumbai Electricity Supply and Transport Limited) buses transport about 37 lakh people (Best 2018). The Mumbai Metro is being constructed to improve urban mass transportation and supplement the suburban train system. Once completed, it will be 235 km long with 200 stations when completed. Currently, Line 1 is operating with a route length of 11.4 km and 12 stations. A total of 435 individual samples were collected from within 800–1000 m of future metro rail station sites at Seepz (16% of total sample), Mahalakshmi (24%), Grant Road (23%), Girgaon (17%) and Cuffe Parade (20%). These stations were selected based on their type of land use forms. The main intention to include these metro stations was to consider different land use types so as to additionally understand the influence of land use on last mile mode choice. Seepz station is classified as an Industrial/commercial land use zone, Mahalakshmi station is considered a commercial zone, Grant Road station as mixed land use, Girgaon station as Residential zone and Cuffe Parade station as a Residential zone as per MMRCL (2011). The purpose of considering 800–1000 m from prospective metro stations was to identify the access mode preference of metro users of specific stations. The study site’s geography is such that the Mumbai suburban rail line and the planned metro line alignments run parallel and in close proximity. Suburban railway stations and metro rail stations are around 2 km apart. In addition, the access distance covered by modes in Mumbai varies from 900 m for walk to 3500 m for personal vehicles (Rastogi and Rao 2003). As a result, for metro commuters, the study considers a catchment radius of 800– 1000 m as all access modes would be found within the range, whereas for anything greater than 1000 m the survey would likely miss the walking mode. Face-to-face household surveys were conducted to better understand the behaviour of public transport users’ access trips in the current context as well as future possible access modes when Mumbai Metro Line-3 becomes operational. The data was collected between January and February 2020 by a team of 8 well-trained enumerators using mobile tablets. The authors administered the survey sheets and the data collection. The survey was designed using Google Forms. The data was collected from those household members who responded ‘yes’ to the screening question, ‘would you travel by metro rail in the future?’. A total of 640 random households were approached, wherein 68% of respondents, one respondent per household, replied ‘yes’. In the survey, the respondents were asked about their present and future first mile and last-mile travel choices for their daily commute trips. The survey questionnaire

Joint Versus Standalone Estimation of First/Last Mile Mode Choice …

73

was divided into three major sections—(a) current first and last-mile travel characteristics, such as access mode, frequency of use, travel time of first mile/last mile trip, perception of quality of infrastructure in the first/last mile, perceived comfort of the trip, and mod involved in different trip purposes (work/education, shopping and entertainment), (b) detailed socio-demographic data of the respondents including age, gender, monthly household income, education qualification and household vehicle ownership and (c) future first and last-mile mode choice as per given choice sets. The Revealed Preference (RP) survey section aids in understanding individuals’ current access mode choice (no metro availability), whereas the Stated Preference (SP) section includes hypothetical scenarios of future metro station access mode choice (when metro becomes operational). The SP component of the survey contained 15 unique choice sets, each with three levels: poor, average and good, from which respondents had to choose an access mode. Using JMP software, D-optimal design, an efficient method was employed to develop scenarios of choice sets with attribute levels. The attributes covered in the choice sets were time duration to access metro station, ability to multitask during access trip, cost of access trip, infrastructure quality, comfort, safety and security, and weather condition experienced during the future access trip to metro station. The attributes are selected from various literature and levels of each attribute are defined as shown in Appendix 1 (https://tinyurl.com/ 2p8jxurd). The access travel time to metro stations was calculated using Google Maps for various modes of transportation and for peak and non-peak hours of the day, and then divided into three groups based on mean value. The average cost incurred when using the modes is computed, which includes fuel cost, maintenance cost, and parking cost in the case of personal cars, and drop-off, only fare cost in the case of bus and IPT. The ability to multitask when accessing metro stations was grouped into three groups (good being 3–4 tasks, average being 2 and poor being 0–1) based on commuters’ skills to read/write, use a phone, socialise and shop for groceries. Additionally, the land use data around the metro stations were also collected, which allowed in assessing its relationship with access mode choice. The land use type around each metro station is considered as per the detailed project report (DPR) prepared by Mumbai Metro Rail Corporation Limited (MMRCL 2011). To have a better understanding of the land uses around the metro station sites currently, the different land use percentages are estimated using the data collected from the Municipal Corporation of Greater Mumbai with the help of QGIS software and Google Earth.

3.2 Modelling Results Standalone MNL models, using RP and SP data separately, were developed to test their performance when compared to the joint RP-SP model, as shown in Table 1. The standalone models were discretely estimated, considering the RP and SP data as independent sets.

74

B. S. Manoj and A. K. Goswami

Table 1 MNL model results Model details Number of individuals 435 Number of observations 8676 Model parameters

RP

SP

Joint RP-SP

Estimate t-ratio* Estimate t-ratio* Estimate t-ratio* Alternative specific constant (ASCs) Walk (base)

0.00

NA

0.00

NA

0

NA

Bicycle

−0.45

−1.01

3.69

6.40

−5.18

−13.44

Personal vehicle (PV)

−3.03

−1.75

6.01

12.30

−1.65

−7.15

Bus

−2.11

−0.05

2.10

7.12

0.31

2.15

IPT

−1.14

−0.94

7.15

20.10

−2.78

−14.52

Drop-off

NA

NA

8.02

22.89

−1.85

−7.12

Socio-demographics Age: Base—Minors (Below 18 years) Young millennial dummy for bus

−0.58

−3.78

–

–

−0.42

−3.89

Old age dummy for bus

−0.62

−2.87

–

–

−0.45

−2.84

Monthly HH income: base—low income HH (0–20 k) Middle income dummy for PV 0.29

2.35

0.37

1.96

0.34

3.09

High income dummy for PV

–

0.70

2.52

0.58

3.62

–

Education qualification (base—below 10th) Graduate dummy for Bus

–

–

–

–

0.02

1.89#

Graduate dummy for IPT

–

–

–

–

−0.29

−2.14

Car ownership only: IPT

–

–

–

–

−0.36

−2.89

Car ownership only: drop off

NA

NA

–

–

0.30

2.59

Motorcycle ownership only: bus

–

–

–

–

−0.20

−2.01

Motorcycle ownership only: IPT

–

–

–

–

−0.38

−4.54

Bicycle ownership only: PV

–

–

–

–

−1.01

−3.69

Bicycle ownership only: bus

–

–

–

–

−0.50

−2.88

Bicycle ownership only: IPT

–

–

−0.89

−3.99

−0.95

−5.10

Multiple vehicle ownership: PV

–

–

0.94

9.87

−0.36

−2.89

Multiple vehicle ownership: bus

−0.78

−4.18

−0.38

−4.78

0.30

2.59

Multiple vehicle ownership: IPT

−0.89

−2.60

–

–

−0.20

−2.01

Vehicle Ownership

(continued)

Joint Versus Standalone Estimation of First/Last Mile Mode Choice …

75

Table 1 (continued) Model details Number of individuals 435 Number of observations 8676 Model parameters

RP

SP

Joint RP-SP

Estimate t-ratio* Estimate t-ratio* Estimate t-ratio* Travel characteristics and mode attributes Land use (base—commercial) Residential dummy for bicycle –

–

0.58

3.01

–

–

Residential dummy for PV

–

–

0.20

2.36

0.18

2.03

Mixed dummy for Bicycle

−1.88

−1.99

–

–

–

–

Industrial dummy for Bus

–

–

1.88

1.98

0.30

3.85

Bicycle

–

–

–

–

−0.87

−18.43

PV

–

–

−0.41

−4.73

−0.61

−22.15

Bus

–

–

−0.25

−2.96

−0.10

−3.86

Drop-off

NA

NA

–

–

−0.87

−18.43

−1.76# 0.90

10.52

0.85

9.89

NA

1.68

12.56

1.25

9.21

2.55

2.10

8.52

1.16

13.85

Travel time for access trip

Quality of infrastructure (base—poor) Average quality dummy for bus

−0.10

Good quality dummy for drop NA off Good quality dummy for bus

0.78

Safety and security (base—poor) Average safety and security dummy for PV

NA

NA

0.52

6.85

–

–

Average safety and security dummy for Bus

NA

NA

1.35

6.05

–

–

Average safety and security dummy for Drop-off

NA

NA

–

–

3.10

10.89

–

–

–

–

0.80

4.50

Frequency of trip Shopping trip for PV

Trip purpose (base—entertainment trip) Work trip dummy for PV

–

–

–

–

0.60

2.17

Work trip dummy for IPT

–

–

–

–

0.70

3.10 (continued)

76

B. S. Manoj and A. K. Goswami

Table 1 (continued) Model details Number of individuals 435 Number of observations 8676 Model parameters

RP

SP

Joint RP-SP

Estimate t-ratio* Estimate t-ratio* Estimate t-ratio* Weather (base—monsoon) Summer dummy for drop off

NA

NA

1.76

7.01

1.61

10.02

Winter dummy drop off

NA

NA

6.36

13.52

4.03

13.54

Cost_biycle

–

–

–

–

−0.18

−6.54

Cost_PV

0.23

2.35

−1.63

−13.25 −0.78

−16.54

Cost_bus

0.11

1.99

−1.01

−9.99

−0.25

−8.79

Cost_drop-off

NA

NA

–

–

−0.70

−13.52

Access cost

Scale Parameters for RP-SP joint model mu_RP

–

–

–

–

1

NA

mu_SP

–

–

–

–

0.84

17.34

Log-likelihood at constant only (0, whole model)

−2817.32

Log-likelihood at convergence −1003.23 (final, whole model)

−11,691.20

−15,545.31

−5331.89

−3947.07

‘*’ represents t-ratio of all the attributes are considered at 95% significance level ‘# ’ represents the t-ratio are significant at 90% C.I ‘–’ represents the attributes are insignificant at 95% C.I ‘NA’ represents the variable is not applicable to the model

Literature shows that individual RP and SP models either underestimate the coefficients of attributes, likely to be in the case of RP models, or overestimate the coefficients of attributes, likely to be in the case of SP models. Whereas, the joint RP-SP model is likely to provide optimised coefficient estimates for the attributes (Morikawa 1994). Research also shows that the overall goodness of fit of a combined RP-SP model is significantly better than separate RP and SP models. In addition, the coefficients of alternate specific constants (ASC) in some of the cases are likely to be significantly different when modelled separately (Ben-Akiva and Morikawa 1990). The difference in ASC’s estimate values in individual models may lead to misinterpretation of access modes that are likely to be chosen by the commuters/ respondents. Furthermore, in standalone RP and SP models, some variables could be insignificant, whereas the same attributes are likely to be significant in case of a joint RP-SP model (Lavasani et al. 2017). The reason for the insignificance of parameters in standalone models is due to the high correlation and low variance among the explanatory variables in the data. Similar observations of insignificant parameters

Joint Versus Standalone Estimation of First/Last Mile Mode Choice …

77

and incorrect signs are noted in many literatures (Bhat and Castelar 2002; Brownstone et al. 2000; Hensher et al. 1998). These differences in individual models have major impact on decision-making. The first section of Table 1 includes the alternative specific constants (ASCs) for available access modes (bicycle, personal vehicle, bus, IPT, drop-off), where walk mode is considered base. Comparing the estimates of ASCs of the standalone RP model indicate that all else being equal, utility of walk mode is the predominant first mile and last mile model compared to other modes. Alternately, the SP model shows the utility of drop-off, personal vehicle, bus, bicycle and IPT is greater than walk mode. However, the joint RP-SP model indicates that the utility of bus mode for the first and last-mile trip is higher than walk mode, whereas other modes like personal vehicle, drop-off, IPT and bicycle have lower utility than walk. Comparing the estimates of the income parameters of the standalone SP model and the joint RP-SP model, it can be observed that the standalone SP model overestimates the propensity of using personal vehicles (0.37) for the access trip as opposed to the joint model (0.34). However, the stand-alone RP model, underestimates the same parameter (0.29). Similar trends in the results can be seen for other parameters, such as multiple vehicle ownership, residential and industrial land uses and quality of infrastructure. The underestimated and overestimated attributes result in the decrease and increase in willingness to pay for a unit increase of a particular mode, respectively. Secondly, in several cases, standalone SP models could not capture the significance of attributes that the joint RP-SP model could capture. For example, the age parameter was insignificant in the standalone SP model. In contrast, the joint model predicts that as age increases, there is a higher likelihood of walking than availing of the bus for the access trip. Ignoring the specific attributes will result in a bias towards a particular section of society using a particular mode to access the metro station. As such, the joint model helps to better understand user behaviour. Thirdly, modelling results show that the joint model estimates fall between the standalone RP and SP models. For example, the estimates of the quality of infrastructure in RP standalone model show that when the bus-related infrastructure improves from poor to good, the commuters are more likely to access the metro stop by bus (0.78). However, this propensity is even higher in case of the standalone SP model (2.10). In case of the joint RP-SP model though, the parameter estimate lies between the estimates of the standalone models (1.16). The scale parameters for RP data are normalised to 1 and the SP data scale parameter is estimated in joint RP-SP MNL model. The scale parameter for SP is significant and the value is less than unity. This signifies that the SP data have more variability with comparison with the RP data, which seems to be ideally logical as the SP data have more variability in the hypothetical scenarios.

78

B. S. Manoj and A. K. Goswami

4 Conclusions In this research, user demand for first and last-mile modes to future metro stations is analysed using RP, SP, joint RP-SP MNL model. The main objective of this paper is to examine the effects of different modelling techniques to address the various issues associated with the use of different data sets like RP, SP and combined RP-SP. Different types of data were collected while identifying the attributes that significantly affect commuters’ behaviour while choosing a mode for their travel to/from origin/destination to a transit station. The scale factors for SP data in joint RP-SP MNL model are statistically significant and have value less than unity. The significant value shows that the scale factor is able to capture the unobserved variances between the datasets and has more variability than RP data. Joint RP-SP model enables to develop scale factor, which accounts for the temporal differences between the RP and SP data, and eliminates the hypothetical biasness and behavioural differences that could arise in SP data and have a major impact on the findings of the standalone RP and SP models. RP and SP individual MNL models underestimate and overestimate the co-efficient of attributes respectively. Whereas, the joint RP-SP model is likely to provide optimised coefficient estimates for the attributes. Present research also shows that the overall goodness of fit of a combined RP-SP model is significantly better than separate RP and SP models when we compared the log-likelihood values of the respective models. Standalone RP and SP models display some variables that are insignificant, whereas the same attributes are likely to be significant in case of a joint RP-SP model. Hence, the joint RP-SP models overcome the shortcomings of standalone models which better reflect the user behaviour and thus provide better guidance to policymakers for implementation.

References Ahern AA, Tapley N (2008) The use of stated preference techniques to model modal choices on interurban trips in Ireland. Transp Res Part A Policy Pract 42(1):15–27. https://doi.org/10.1016/ j.tra.2007.06.005 Ajzen I (1991) The theory of planned behavior. Organ Behav Hum Decis Process 50(2):179–211. https://doi.org/10.1016/0749-5978(91)90020-T Ben-Akiva M, Morikawa T (1990) Estimation of switching models from revealed preferences and stated intentions. Transp Res Part A Gen 24(6):485–495. https://doi.org/10.1016/0191-260 7(90)90037-7 Ben-Akiva M, Morikawa T, Shiroishi F (1992) Analysis of the reliability of preference ranking data. J Bus Res 24(2):149–164. https://doi.org/10.1016/0148-2963(92)90058-J Bhat CR, Castelar S (2002) A unified mixed logit framework for modeling revealed and stated preferences: formulation and application to congestion pricing analysis in the San Francisco Bay area. Transp Res Part B Methodol 36(7):593–616. https://doi.org/10.1016/S0191-2615(01)000 20-0 Bliemer MCJ, Rose JM, Chorus CG (2017) Detecting dominance in stated choice data and accounting for dominance-based scale differences in logit models. Transp Res Part B Methodol 102:83–104. https://doi.org/10.1016/j.trb.2017.05.005

Joint Versus Standalone Estimation of First/Last Mile Mode Choice …

79

Bradley M, Daly A (1992) Estimation of logit choice models using mixed stated-preference and revealed-preference information. In: 6th international conference on travel behaviour, Quebec, pp 209–232 Brownstone D, Bunch DS, Train K (2000) Joint mixed logit models of stated and revealed preferences for alternative-fuel vehicles. Transp Res Part B Methodol 34(5):315–338. https://doi.org/ 10.1016/S0191-2615(99)00031-4 Cherchi E, Hensher DA (2015) Workshop synthesis: stated preference surveys and experimental design, an audit of the journey so far and future research perspectives. Transp Res Procedia 11:154–164. https://doi.org/10.1016/j.trpro.2015.12.013 De Haas MC, Hoogendoorn RG, Scheepers CE, Hoogendoorn-Lanser S (2018) Travel mode choice modeling from cross-sectional survey and panel data: the inclusion of initial nonresponse. Transp Res Procedia 32:268–278. https://doi.org/10.1016/J.TRPRO.2018.10.049 Dissanayake D, Morikawa T (2002) A combined RP/SP nested logit model of vehicle ownership, mode choice and trip chaining in developing countries. In: Proceedings of the conference on traffic and transportation studies, ICTTS, pp 588–595. https://doi.org/10.1061/40630(255)83 Hensher DA (2004) Identifying the influence of stated choice design dimensionality on willingness to pay for travel time savings. J Transp Econ Policy 38(3):425–446 Hensher D, Louviere J, Swait J (1998) Combining sources of preference data. J Econ 89(1–2):197– 221. https://doi.org/10.1016/S0304-4076(98)00061-X Hess S, Palma D (2019) Apollo: A flexible, powerful and customisable freeware package for choice model estimation and application. J Choice Model 32:100170. https://doi.org/10.1016/j.jocm. 2019.100170 Lavasani M, Hossan MS, Asgari H, Jin X (2017) Examining methodological issues on combined RP and SP data. Transp Res Procedia 25:2330–2343. https://doi.org/10.1016/J.TRPRO.2017. 05.218 M. BEST (2018) Rates of regular tickets, daily bus pass and periodical bus-passes. https://www. bestundertaking.com/in/pdf/FareRevisionApril2018.pdf. Accessed 30 Aug 2021 McFadden D (1973) Conditional Logit analysis of quantitate choice behaviour Meister A, Mondal A, Asmussen KE, Bhat C, Axhausen KW (2022) Modeling urban mode choice behavior during the COVID-19 pandemic in Switzerland using mixed multiple discretecontinuous extreme value models. 36119812210895. https://doi.org/10.1177/036119812210 89545 MMRCL (2011) Detailed project report for through metro corridor: Colaba-Bandra-Seepz. https:/ /www.mmrcl.com/sites/default/files/dpr-metro-line-III.pdf. Accessed 30 Aug 2021 [Online] Morikawa T (1994) Correcting state dependence and serial correlation in the RP/SP combined estimation method. Transportation 21(2):153–165. https://doi.org/10.1007/BF01098790 Morikawa T, Ben-Akiva M, McFadden D (2002) Discrete choice models incorporating revealed preferences and psychometric data. Adv Econ 16:29–55. https://doi.org/10.1016/S0731-905 3(02)16003-8/FULL/XML Qiao Y, Huang Y, Yang F, Zhang M, Chen L (2016) Empirical study of travel mode forecasting improvement for the combined revealed preference/stated preference data-based discrete choice model. Adv Mech Eng 8(1):1–8. https://doi.org/10.1177/1687814015624836 Rastogi R, Rao KVK (2003) Travel characteristics of commuters accessing transit: case study. J Transp Eng 129(6):684–694. https://doi.org/10.1061/(ASCE)0733-947X(2003)129:6(684) Song M, Zhong H (2020) Efficient weighted univariate clustering maps outstanding dysregulated genomic zones in human cancers. Bioinformatics 1–10. https://doi.org/10.1093/bioinformatics/ btaa613 Vij A, Shankari K (2015) When is big data big enough? Implications of using GPS-based surveys for travel demand analysis. Transp Res Part C Emerg Technol 56:446–462. https://doi.org/10. 1016/J.TRC.2015.04.025

80

B. S. Manoj and A. K. Goswami

Wang D, Borgers A, Oppewal H, Timmermans H (2000) A stated choice approach to developing multi-faceted models of activity behavior. Transp Res Part A Policy Pract 34(8):625–643. https:/ /doi.org/10.1016/S0965-8564(99)00045-2 Yang L, Choudhury CF, Ben-Akiva M, Abreu e Silva J, Carvalho D (2013) Stated preference survey for new smart transport modes and services: design, pilot study and new revision. J Sustain Tour 21(4):596–613

Do Service Attributes Influence the Mode Choice of Access-Egress Trips in the Context of Delhi Metro? Rashmi Choudhary, Jogendra Kumar Nayak, and Manoranjan Parida

Abstract Modeling travel choices is a critical step in the planning process. The present study explores commuters’ mode choice preference for the access-egress trips concerning metro stations in Delhi, India. Data was collected from five hundred ten respondents using a tablet-based face-to-face interview at ten metro stations. The available modes included walk, intermediate public transit, bus, and private vehicles. Observable parameters such as travel and socio-demographic characteristics, and non-observable parameters (attitudinal parameters such as satisfaction with comfort, ease in transfer, safety and security, and cost) extracted using exploratory factor analysis, were considered in the multinomial logit model. Key findings of the study show that attitudinal parameters significantly influence mode choice, while sociodemographic characteristics apart from vehicle ownership parameter fail to do so. The results suggest that sustainable modes (walk and IPT) can be promoted by reducing access-egress distance and improving comfort for commuters. Keywords Access-egress · Multinomial logit model · Mode choice

1 Introduction Accelerated industrialization, urbanization, and population growth have caused migration toward cities, increasing the demand for smooth urban mobility. The capacity enhancement of road transport infrastructure in urban areas is widely limited R. Choudhary (B) Department of Civil Engineering, Indian Institute of Technology Roorkee, Roorkee, Uttarakhand 247667, India e-mail: [email protected] J. K. Nayak Department of Management Studies, Indian Institute of Technology Roorkee, Roorkee, Uttarakhand 247667, India M. Parida CSIR - Central Road Research Institute (CRRI), New Delhi 110025, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_7

81

82

R. Choudhary et al.

by physical constraints leading to increasing pressure on the existing transit system (Sadhukhan et al. 2015). Access and egress trips should be improved and assessed to provide smooth transit trips to commuters. Access-egress trips refer to sections of trips from home to transit station and from the transit station to the destination (Clifton and Muhs 2012). The commuters’ access-egress modes impact traffic on roads, congestion, and emissions. Mode choice modeling is one of the most complex stages of the four-stage modeling process, including trip generation, trip distribution, modal split, and trip assignment. Choice modeling is a decision-making technique for commuters to choose between alternative modes based on several observable and unobservable parameters. Observable parameters include travel distance, time, and cost, whereas the unobservable or latent parameters combine perception and attitudes toward modespecific characteristics and services provided in access-egress trips. Although socioeconomic attributes may indicate commuters’ attitudes, the latent attributes are essential for understanding a person’s choice. Including attitudinal parameters in choice, modeling ensures better models and accurate predictions about commuters’ choices compared to conventional choice models (Ben-Akiva et al. 2002). However, an aggregated impact of commuters’ socio-demographic, trip characteristics, and attitudinal parameters on the mode choice of access-egress needs to be explored in detail. Mode choice primarily contributes to analyzing an area’s future travel demand by considering all possible alternative modes. In previous studies, mixed logit, and multinomial logit (MNL) models were used to analyze the mode choice for access and egress considering demographic and travel characteristics (Halldórsdóttir 2017). MNL models were used by Bastarianto et al. (2019) for commuters’ tour-based choice behavior. Haggar et al. (2019) provided insights on how passengers’ habits can impact their intentions and mode choice in different scenarios. Ben-Akiva et al. (2002) has elaborated on incorporating both observable and latent factors to construct a mode choice model. Including attitudinal variables allows the researchers to have robust economic and behavioral approaches. Studies have focused on exploring different areas with traditional DCM methods (Murtagh et al. Sep 2012). Ben-Akiva et al. (2002) stated that apart from the level of service, the trip purpose and number of modes in the trip also affect the mode choice. Studies based on passengers’ travel behavior in access-egress were considered to select attributes related to access-egress facilities. Various parameters were found in the literature that impact passengers’ decision of the mode (Bastarianto et al. 2019). Studies related to service quality of access and egress services were also accounted for shortlisting the parameters, impacting the mode choice and the passengers’ satisfaction. Researchers have either constructed perception scales or adapted available scales to collect data (Bouscasse et al. 2018). Givoni and Rietveld (2007), de O˜na et al. (2016) have employed various socio-demographic and travel characteristics to assess access-egress services. Hine and Scott (2000) employed the psychological and temporal factors while studying the accessibility to transit facilities. Namgung and Akar (2014) estimated the effect of attitudinal parameters on the transit choice by determining the relationship between the passengers’ attitude and their transit use through logit modeling.

Do Service Attributes Influence the Mode Choice of Access-Egress …

83

The study suggested that the model’s explanatory power increases by including attitudinal parameters such as passengers’ satisfaction level with cost, comfort, or safety. The attitudinal parameters such as individuals’ comfort and perceptions also reflect the commuters’ travel choices (Bahamonde-Birke et al. 2017). Studies focus on the services offered by the access-egress to the metro rail without considering attitudinal variables (Goel and Tiwari 2016). The previous studies focused on the mode choice modeling for access-egress and last-mile public transit services. It can be derived that socio-demographic characteristics, travel attributes, and attitudes of individuals toward the transportation services were included individually, and the combined influence of all parameters was not explored (Namgung and Akar 2014). The gap from previous studies can be identified as lacking an integrated choice model for access-egress modes, including the attitudinal parameters. The current study aims to understand and discuss the impact of socio-demographics, travel characteristics, and attitudinal parameters on the choices made by commuters for access-egress trips. The study is focused on the stations of the Delhi metro rail network. Based on the results, the societal impacts, and benefits from the inclusion of commuters’ attitudes in choice models for policymakers and transport planners are suggested.

2 Study Area The study is conducted on the access-egress services of Delhi Metro Rail Systems as Delhi is India’s busiest metro rail system. Being a metro city with a population of 16.75 million, Delhi has the highest motor-vehicle ownership in India (GoI: Census of India 2011). The metro network of Delhi is expanding rapidly to cover the city and its national capital region. It currently has nine metro lines, represented by different colors in Fig. 1 in all the zones of Delhi. For the 234 stations of the metro system, the average daily ridership for five years from January 2014–December 2019 ranged between 400 and 71,000 passengers per day (DMRC: Delhi metro rail corporation annual report 2019). It shows that the ridership of different metro stations depends on the facilities and land use characteristics. Ten metro stations were earmarked for data collection based on the ridership data and landuse characteristics.

3 Methodology The study explores the impact of attitudinal variables, socio-demographic factors, and travel characteristics on the mode choice for access and egress trips. The trips considered are between home and metro station, and between metro station and destination. The alternatives for mode choice include walk, IPT, bus, and private vehicles, which are nominal type-dependent variables. Therefore, the MNL predicts the relationship between various independent variables and choice alternatives. Attitudinal parameters are taken from commuters’ satisfaction levels toward services in

84

R. Choudhary et al.

Fig. 1 Study area: Delhi metro network and selected stations

the access-egress trips. The latent exogenous factors (attitudinal) are obtained using the exploratory factor analysis (EFA) on the attitudinal variables’ satisfaction levels collected from users in the form of ratings on the Likert scale. A choice model using multinomial logistic regression is developed to find the impact of various parameters on users’ preferences of all the available modes. The relationship between mode choice for access-egress trips and the variables impacting mode choice is derived using a discrete choice model. MNL is a traditional logit model widely used in transportation because of its simplicity and reliability (Bastarianto et al. 2019). Exploratory factor analysis extracts latent factors from the users’ perception ratings for access-egress variables. Outcomes of multinomial logistic regression provide the utilities for each available access and egress mode. The sign and magnitude of coefficients are observed for the MNL model’s interpretation. These signs represent whether the coefficients increase or decrease passengers’ preference for that mode of travel. The assumptions for applicability of MNL models include the dependent variable to be nominal, which is the mode alternatives, and no multi-collinearity between the independent variables. The multi-collinearity is checked and verified by Karl Pearson’s correlation test and Variation Inflation Factor (VIF). The theoretical structure based on random utility establishes the concept of selecting an alternative with maximum utility out of available alternatives. The impact of variables on the choice of access-egress modes is derived from the magnitude and direction of coefficients obtained from the MNL model’s outcomes.

Do Service Attributes Influence the Mode Choice of Access-Egress …

85

4 Survey Design and Data Collection A questionnaire comprising three sections is used for the primary data collection, including travel characteristics of access and egress trips, commuters’ perception of the services provided, and the commuters’ demographic characteristics. Distance and time of the connecting trips from home to metro stations or metro stations to the destination, the trip frequency, and purpose are included in travel characteristics. Commuters’ perceptions are collected in the form of importance and satisfaction levels for comfort, accessibility, cost, safety, and security for access and egress trips. Age, gender, education, household size, income, and vehicle ownership data are collected through respondents’ demographics. A five-point Likert scale ranging from 1 (least satisfied/least important) to 5 (most satisfied/most important) is used to collect perceived importance and satisfaction levels. The data is collected near the entry/exit gates of ten selected stations between 5th December and 30th December 2019. A tablet-based face-to-face interview method was used to collect metro passengers’ responses (Mandhani et al. 2020). The sample size was estimated for the primary data collection using Cochran’s method based on the population and the required accuracy. At a 95% confidence level and 0.05 degree of variability, the minimum size obtained is 384. The authenticity of data collected for the choice modeling impacts the prediction accuracy of models. 510 responses were collected on-site to eliminate the risk of incomplete and inappropriate responses. Out of which, 492 were found as complete and appropriate for analysis, and each respondent provided the perception levels based on access and egress trips of the transit system. Table 1 shows the descriptive statistics and distribution of 492 respondents based on different aspects.

4.1 Attitudinal Variables The usage of any mode depends on the service quality provided by it (Mandhani et al. 2020); therefore, attitudinal variables explaining passengers’ satisfaction with the service provided in access-egress are considered for explaining commuters’ choice. Twenty-three items are selected for the questionnaire to account for the perception levels about access-egress service parameters, out of which nineteen items are found to be loaded on five different factors extracted through Exploratory Factor Analysis (EFA). The Principal Component Analysis (PCA) and orthogonal varimax are used for factor extraction, and orthogonal varimax is used for rotation. Keiser-MayerOlkin measure value is observed as 0.801, which is an indicator of adequate sampling (Hadia et al. 2016). Bartlett’s test of sphericity got a significant value (p < 0.001) with a chi-square of value 1973, confirming the correlations’ overall significance. The total variance is 53%, which is considered acceptable if greater than 50%. The five factors obtained from EFA are satisfaction with transfer between modes, satisfaction with safety and security, satisfaction with accessibility, satisfaction with comfort,

86

R. Choudhary et al.

Table 1 Descriptive statistics of respondents Variable

Category

Sample (%)

Variable

Category

Sample (%)

Gender

Female

41.5

Occupation

Government

12.0

Male

58.5

Housewife

7.9

< 18

4.5

Other

5.5

18–25

28.9

Private

51.6

Student

23.0

55

11.0

Income (in dollars)

>1370

11.8

Household Size 1

2.4

342–684

33.7

2

3.5

684–1026

27.4

3

15.1

1026–1370

13.5

4 or more

79

Business

9.5

Daily

57.8

Education

18.7

Once a month

18.1

Leisure

16.7

Once a week

10.9

Shopping

7.7

Twice a week

13.2

Work

47.4

Frequency of usage

Purpose

and satisfaction with cost. Table 2 elaborates factors, items, factor loadings, and Cronbach alpha value for each factor. The Cronbach alpha values for all five factors are higher than 0.60, representing adequacy and data reliability (Hair et al. 2009). The Cronbach alpha value represents the internal consistency of items under each factor.

4.2 Interdependence of Attitudinal Variables The correlation coefficient’s value varies from -1 (strong inverse correlation) to 1 (strong positive correlation), and 0 represents no relationship. The pair of variables with a close correlation value may cause multi-collinearity. The correlation values vary between 0.1 and 0.4, showing no strong correlation between any two parameters. Along with Karl Pearson’s correlation test, the variance inflation factor (VIF) is also checked, a multivariate measure of the variance explained on a variable by other variables in a model. The threshold value of VIF is taken as 5 (Gareth et al. 2013). For the attitudinal parameters, the VIF values varied from 1.158 to 1.336, indicating no strong correlation between variables in compliance with both collinearity tests.

Do Service Attributes Influence the Mode Choice of Access-Egress …

87

Table 2 Exploratory factor analysis Factors

Items

Factor loading

Cronbach alpha

Satisfaction with transfer between modes

Decrease in waiting time for metro

0.419

0.731

Connectivity between different modes

0.492

Availability of access-egress mode

0.517

Availability of customer care at/near metro stations

0.591

Availability of information boards at/near metro stations

0.625

Availability of facilities for People with Disability (PwD)

0.635

Availability of information boards for access-egress modes

0.687

Satisfaction with safety and security

Safety in severe traffic conditions

0.620

Security in waiting areas

0.701

Security in access-egress modes

0.791

Satisfaction with accessibility

Ease in finding the metro station

0.405

Decrease in egress distance

0.681

Decrease in access distance

0.823

Satisfaction with comfort

Comfort in access

0.467

Comfort in egress

0.538

Availability of lifts and escalators inside the metro station

0.614

Ease in buying tickets/cards for metro

0.664

0.624

0.60

0.660

Satisfaction with cost Cost of access

0.746

Cost of egress

0.797

0.641

5 Model Development The explanatory variables used for choice modeling consisted of socio-demographic characteristics, trip characteristics, and attitudinal parameters. Mode choice modeling was carried out for these explanatory variables using r-studio software through a multinomial logistic regression technique (mlogit-package). Among trip characteristics, the frequency of using the mode, the purpose of the trip, access/egress distance, and time were considered. In socio-economic traits, age, gender, occupation, household size, monthly income, and vehicle ownership were considered for the model. The latent factors obtained from EFA were taken as attitudinal variables, which included satisfaction with transfer between modes, satisfaction with safety

88

R. Choudhary et al.

and security, satisfaction with accessibility, satisfaction with comfort, and satisfaction with cost. A total of 15 variables were considered to influence the mode choice in access-egress trips. For the choice model, walk was considered as the reference mode. The other alternatives of modes were Intermediate Public Transit buses and private vehicles.

6 Results In Table 3, the positive sign of beta coefficients indicates that the contribution of a particular parameter is higher in the decision-making for a given mode concerning the reference mode. The |t| values were calculated to use the coefficients and standard error ratio. The |t| values of variables higher than 1.96 signify that the variables obtained are statistically significant at a 95% confidence level. The |t| greater than 1.64 were statistically significant for 90%, and |t| greater than 2.576 were significant for a 99% confidence level. In the MNL model, the goodness of fit is defined by ρ2 (Pseudo R2 ). The value of ρ2 from 0.2 to 0.4 shows an excellent fit (McFadden 1986). The value of ρ2 was found as 0.344, satisfying the goodness of fit condition. It can be observed that the variables of distance, time, and satisfaction with the cost positively impact the preference for bus over walk. In contrast, vehicle ownership negatively influences the selection of buses for leisure trips and satisfaction with comfort, safety, and security. It can be quoted from the results that distance, time, and attitudinal variables affect the choice of mode quite significantly compared to other variables. Among the attribute levels of frequency of usage of the metro services, the commuters traveling once a week prefer private vehicles followed by IPT over walk and buses. For the trip purpose, it was observed that shopping and leisure negatively impact the usage of buses, IPT, and private vehicles, which indicates that walking is the preferred mode for access/egress in shopping and leisure trips. With the increasing distance of access and egress of metro services, commuters use private vehicles followed by buses, then IPT and walk. Walking is not the most accessible alternative for long distances, and with increasing distance, people prefer other modes for more comfort. Age positively influences commuters’ preference for private vehicles over walking, indicating that commuters of higher age may prefer private vehicles more. Being female commuters negatively impacts buses and IPT over walk quite significantly. An increase in income positively impacts the choice of private vehicles, while increasing household size negatively impacts the same. Vehicle ownership has a negative impact on buses and a positive on their vehicles. From the model coefficients of attitudinal variables, satisfaction with transfer between modes positively impacts IPT, and satisfaction with cost has a significant positive impact on buses. Satisfaction with safety and comfort negatively influenced buses, and satisfaction with both cost and comfort negatively impacted IPT. The explanatory power of the model increases with the use of attitudinal variables. The predicted probabilities show that 51% of the passengers will prefer IPT, 31% will choose to walk, 14% will use private vehicles, and only 4% will select a bus for the access-egress trip. The limitations of the choice model

Do Service Attributes Influence the Mode Choice of Access-Egress …

89

because of using MNL includes the unavailability of effects of the interrelationship of parameters on the mode choice. For instance, if the effect of gender on mode choice is considered, the effect of a particular user’s household income will not be considered, and all the parameters are considered independent. Table 3 Results of choice model (S.E.: standard error, β: coefficients) Mode (reference—Walk)

Bus

Variables

β

IPT

β

S.E

0.455

3.559

3.363***

1.739

−8.279*

2.797

Travel once a week

0.848

1.193

0.977***

0.594

1.359***

0.801

Travel twice a week

0.799

1.008

0.119

0.490

0.743

0.731

Travel once a month

0.343

0.918

−0.244

0.450

0.619

0.702

Education trip

0.817

1.427

0.406

0.768

1.190

1.156

Leisure trip

−1.806***

0.930

−0.722

0.699

−1.334*** 0.783

Shopping trip

−0.807

1.499

−1.504**

0.744

−1.045

1.135

Work trip

−0.971

0.992

−0.702

0.528

−0.823

0.751

2.393*

0.434

1.397*

0.179

2.503*

0.299

0.769***

0.440

Distance

β

Private vehicle

Intercept

S.E

S.E

−0.334

0.281

−0.157

0.344

Transfer (S)

− 0.493

0.504

1.022*

0.242

0.173

0.368

Safety (S)

−0.5717*** 0.336

−0.032

0.165

0.146

0.262

Accessibility (S)

−0.452

0.356

−0.278

0.199

0.080

0.291

Comfort (S)

−0.905***

0.491

−0.562**

0.252

−0.156

0.413

0.508***

0.307

−0.665*

0.159

−0.35***

0.204

Time

Cost (S)

0.114

0.244

−0.133

0.116

0.499*

0.199

Gender

−1.393**

0.621

−1.146*

0.318

−0.087

0.503

Housewife

−1.833

1.530

−0.715

0.822

−0.970

1.169

Other job

−0.608

1.650

0.003

0.786

−0.666

1.306

Private job

−0.377

0.878

−0.327

0.481

−0.442

0.649

Student

−1.240

1.482

−1.192*** 0.735

−2.155**

1.123

Household Size

−0.351

0.300

−0.123

0.169

−0.520**

0.258

Income

−0.175

0.231

0.073

0.122

0.262*

0.105

Ownership

−0.692**

0.346

0.120

0.146

0.844*

0.225

Age

*p-value < 0.01, ** 0.01 < p-value < 0.05, *** 0.05 < p-value < 0.10; (S)—Satisfaction with the parameter

90

R. Choudhary et al.

7 Discussion For the current study, a combination of observable and non-observable variables was considered to develop a choice model for the connecting trips of the mass rapid transit system. From the choice model results, it can be inferred that if commuters own their vehicles and earn a higher monthly income, they’ll prefer personal vehicles for access trips to the metro station in case of leisure trips. With increasing trip distance, preference for private vehicles increases rapidly, followed by buses and then IPT compared to walk, resulting from the unavailability of facilities and comfort in the trip. Expanding the mass transit network can decrease access and egress distances, making more people walk for access-egress trips. Due to the availability of market areas and recreational spaces near the metro stations in Delhi, people going for shopping and leisure prefer to walk more than choosing other modes. Trip purpose, which defines the activity choice, also impacts the mode choice. It confirms the result of Krygsman et al. (2007), which states that activity choice is performed before mode choice by the user. Older people who can afford personal vehicles prefer this alternative because of their comfort, less walking distance, and fewer transfers in the access trips. The study also showed that female commuters prefer to walk more than buses and IPT due to the lack of women safety in public transport systems, as Nayak and Benazeer (2017). The difference in preference in both studies can be justified by the difference in the study area and safety level provided in public transport systems. If commuters’ satisfaction with transfer between modes is smooth or if there are few transfers, commuters use IPTs more for their trips. With the increase in satisfaction with comfort, safety, and security, people may not prefer buses and choose more comfortable and safe modes such as private vehicles and IPTs. With increasing satisfaction of access-egress trips’ cost, the preference for buses increases, but IPT and private vehicles’ preference decreases. The bus fare is low compared to IPT and the cost of a private vehicle for the same distance. Cost plays a significant role in selecting modes for access-egress trips, which is confirmed by Tam et al. (2005). The behavioral variables were also considered in previous studies using MNL for developing the mode choice models for the urban areas (Birr 2018; Elharoun et al. 2018). Commuters’ preference for walking increases significantly with increasing satisfaction in comfort, which contradicts the results of Birr (2018). Comfort refers to the availability of lifts and escalators inside the metro station and comfort in access and egress modes individually. The study considered only single modes in access and egress trips, but (Heinen 2018) has shown that multimodality in any trip can attract people toward public transit or active transportation modes, which can be considered as the future scope of the study. To the authors’ knowledge, apart from Goel and Tiwari Mar (2016), Rastogi and Krishna Rao 2003), no significant studies were found for mode choice behavior in the access and egress trips toward transit systems in India. Other studies related to access-egress characteristics were found in the Indian context, but none of these studies focused on choice behavior. The novelty in the current study is shown by incorporating attitudinal variables

Do Service Attributes Influence the Mode Choice of Access-Egress …

91

with socio-demographic and travel characteristics in the mode choice modeling for access-egress trips toward rapid transit systems in the Indian context.

8 Conclusion MRT systems’ growth has proven efficient in shifting people from private modes to the transit system. Still, for the connectivity up to the transit stations, using personal vehicles, walk, intermediate public transit, and various other modes also cause negative externalities like congestion and emission. The study was performed to explore the mode choice of access-egress trips at the Delhi metro rail system’s metro stations, with valid responses from 492 respondents. A tablet-based face-to-face questionnaire survey was used to collect the socio-demographic characteristics, trip characteristics, and commuters’ perception of access-egress service users regarding importance and satisfaction levels. The relationship between the choice of mode and the explanatory variables impacting the mode choice was derived using the MNL model. The model’s explanatory variables included the frequency of using the transit services, trip purpose, distance and time for the trip section, age, gender, occupation, income, household size, and vehicle ownership. Along with these observable variables, attitudinal parameters describing passengers’ satisfaction level with transfer between modes, accessibility, safety and security, comfort, and cost were considered to explain the passengers’ attitude and behavior. The choices available for the passengers in access-egress sections were walk, intermediate public transit, bus, and private vehicles. The walk was taken as the reference mode, from which the selection probability of other modes is compared. The β coefficients explain the nature and impact of variables on each mode’s selection probability. From the sign of β coefficients, it can be understood that if the β coefficient is positive, then for a unit increase in the explanatory variable, there will be β unit increase in the outcome variable. If the β coefficient is negative, then for a unit increase in the explanatory variable, there will be β unit decrease in the outcome variable. The model’s significance at 90, 95, and 99% confidence levels was checked using the t-values. The applications of mode choice models can determine which parameters of service quality are essential to be improved for commuters to shift from private vehicles to other modes. Making the fare structure flexible may allow more people to use the public transit mode to access and egress the metro stations. The importance of attitudinal variables in mode choice models is growing with time, and it may play a crucial role in changing social dynamics over time. For travel demand modeling, the mode choice results showing variables with a different degree of dependence on the probability of choosing a given mode of transport can be applied. Several policy interventions can be provided for each latent factor based on the observed variables, but improving each parameter is not feasible in the Indian context. The study provided insights into linkages of different aspects of travel behavior, travel mode preference, satisfaction with access-egress services, passengers’ characteristics, and the attitudinal variables. The transit authorities and the transport planners for the urban areas can use the recommendations and outcomes

92

R. Choudhary et al.

from the choice model developed for access and egress services to improve service quality based on passengers’ perceptions. Although MNL is a traditional model, choice modeling is new for access and egress trips. Further, the choice modeling for the access-egress trip can be attempted through hybrid choice models that overcome the limitations of MNL.

References Bahamonde-Birke FJ, Kunert U, Link H, Ortúzar JDD (2017) About attitudes and perceptions: finding the proper way to consider latent variables in discrete choice models. Transportation 44(3):475–493 Bastarianto F, Irawan M, Choudhury C, Palma D, Muthohar I (2019) A tour-based mode choice model for commuters in indonesia. Sustainability 11(3):788. https://doi.org/10.3390/su1103 0788 Ben-Akiva M, Walker J, Bernardino AT, Gopinath DA, Morikawa T, Polydoropoulou A (2002) Integration of choice and latent variable models. Perpetual Motion: Travel Behav Res Oppor Appl Chall 2002:431–470 Birr K (2018) Mode choice modelling for urban areas. Czasopismo Techniczne 6. https://doi.org/ 10.4467/2353737xct.18.087.8692 Bouscasse H, Joly I, Bonnel P (2018) How does environmental concern influence mode choice habits? a mediation analysis. Transp Res Part D: Transp Environ 59:205–222. https://doi.org/ 10.1016/j.trd.2018.01.007 Clifton K, Muhs CD (2012) Capturing and representing multimodal trips in travel surveys. Transp Res Record: J Transp Res Board 2285(1):74–83. https://doi.org/10.3141/2285-09 de Oña J, de Oña R, Eboli L, Forciniti C, Mazzulla G (2016) Transit passengers’ behavioural intentions: the influence of service quality and customer satisfaction. Transp A Transp Sci 12(5):385–412 https://doi.org/10.1080/23249935.2016.1146365 DMRC: Delhi metro rail corporation annual report 2018–2019. Delhi Metro Rail Corporation (2019) Elharoun M, Shahdah UE, El-Badawy SM (2018) Developing a mode choice model for mansoura city in egypt. Int J Traffic Transp Eng 8:528–542 Gareth J, Daniela W, Trevor H, Robert T (2013) An introduction to statistical learning: with applications in R. Spinger Givoni M, Rietveld P (2007) The access journey to the railway station and its role in passengers’ satisfaction with rail travel. Transp Policy 14(5):357–365. https://doi.org/10.1016/j.tra npol.2007.04.004 Goel R, Tiwari G (2016) Access–egress and other travel characteristics of metro users in delhi and its satellite cities. IATSS Res 39(2):164–172. https://doi.org/10.1016/j.iatssr.2015.10.001 GoI: Census of India (2011) provisional population totals. New Delhi: Office of the Registrar General and Census Commissioner, Government of India (2011) Hadia NU, Abdullah N, Sentosa I (2016) An easy approach to exploratory factor analysis: marketing perspective. J Educ Soc Res. https://doi.org/10.5901/jesr.2016.v6n1p215 Haggar P, Whitmarsh L, Skippon SM (2019) Habit discontinuity and student travel mode choice. Transport Res F: Traffic Psychol Behav 64:1–13. https://doi.org/10.1016/j.trf.2019.04.022 Hair JF, Black B, Babin B, Anderson R (2009) Multivariate data analysis, Global. Prentice Hall, NJ Halldórsdóttir K, Nielsen OA, Prato CG (2017) Home-end and activity-end preferences for access to and egress from train stations in the Copenhagen region. Int J Sustain Transp 11(10):776–786. https://doi.org/10.1080/15568318.2017.1317888 Heinen E (2018) Are multimodals more likely to change their travel behaviour? a cross-sectional analysis to explore the theoretical link between multimodality and the intention to change mode choice. Transp Res F: Traffic Psychol Behav 56:200–214

Do Service Attributes Influence the Mode Choice of Access-Egress …

93

Hine J, Scott J (2000) Seamless, accessible travel: users’ views of the public transport journey and interchange. Transp Policy 7(3):217–226. https://doi.org/10.1016/s0967-070x(00)00022-6 Krygsman S, Arentze T, Timmermans H (2007) Capturing tour mode and activity choice interdependencies: a co-evolutionary logit modelling approach. Transp Res Part A: Policy Pract 41(10):913–933. https://doi.org/10.1016/j.tra.2006.03.006 Mandhani J, Nayak JK, Parida M (2020) Interrelationships among service quality factors of metro rail transit system: an integrated bayesian networks and PLS-SEM approach. Transp Res Part A: Policy Pract 140:320–336. https://doi.org/10.1016/j.tra.2020.08.014 McFadden D (1986) The choice theory approach to market research. Mark Sci 5(4):275–297 Murtagh N, Gatersleben B, Uzzell D (2012) Multiple identities and travel mode choice for regular journeys. Transp Res F Traffic Psychol Behav 15(5):514–524. https://doi.org/10.1016/j.trf.2012. 05.002 Namgung M, Akar G (2014) Role of gender and attitudes on public transportation use. Transp Res Record: J Transp Res Board 2415(1):136–144. https://doi.org/10.3141/2415-15 Nayak JK, Benazeer D (2017) Identifying and addressing the issue of women’s fear of victimization in public transport: a case of delhi. J Eastern Asia Soc Transp Stud 12:2392–2407 Rastogi R, Krishna Rao K (2003) Travel characteristics of commuters accessing transit: case study. J Transp Eng 129(6):684–694 Sadhukhan S, Banerjee UK, Maitra B (2015) Commuters’ perception towards transfer facility attributes in and around metro stations: experience in Kolkata. J Urban Plan Dev 141(4):04014038. https://doi.org/10.1061/(asce)up.1943-5444.0000243 Tam ML, Tam ML, Lam WH (2005) Analysis of airport access mode choice: a case study in Hong Kong. J Eastern Asia Soc Transp Stud 6:708–723

Assessment of Gender and Income-Based Inequality in Travel Behavior of Vadodara City Pankaj Prajapati and Reshma Khan

Abstract In developing country like India there is a wide variation in travel pattern among socio-economic groups. This paper examines gender and income-based difference in travel behavior for trip length, trip purpose, and trip mode independently and also interacting with them. These differences in travel behavior can be due to affordability and accessibility-related issues for different income groups and accessibility, social roles, empowerment, and safety-related issues for gender groups. For the study purpose a midsized city Vadodara, India, is taken. Household survey was conducted to collect data for all administrative wards of the city. Descriptive analysis along with chi-square and ANOVA test is done to see the variability among socio-economic groups. Using NLOGIT software, a multinomial logit model (MNL) was developed to describe people’s travel patterns and mode choices in Vadodara. It is found that women are less mobile and travel for shorter distances than male in each income groups. As the income increases mobility of females increases and their trip length also increases. Low and low-med income groups are more likely to travel shorter distances and rely more on NMT. There is a decline in cycle trips with increase in age among both gender groups and is seen much higher than in female. As a result of these findings, the city should implement a gender and income-based culturally sensitive transportation policy. Looking ahead, it is clear that the public transportation system must be developed and the fleet expanded to keep up with the population’s ever-increasing size and travel needs. Keywords Travel behavior · Socio-economic groups · NLOGIT

P. Prajapati (B) · R. Khan The Maharaja Sayajirao University of Baroda, Vadodara, Gujarat, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_8

95

96

P. Prajapati and R. Khan

1 Introduction Multinomial Logit Urban transportation planning is concerned with the process of constructing new facilities as well as identifying how to best utilize existing ones. Travel behavior refers to people moving from one location to another with the purpose of choosing a specific mode of transportation. Sustainable mobility planning needs identifying the groups of society who have mobility constrains, measure their mobility and accessibility, and designs policies and strategies to address these issues. This necessitates measuring variation in travel choices with respect to different socio-economic groups. These methods also help in predisposing which groups of society are moving toward or away from the overall sustainable mobility goals. Gender and income are the two most important sociodemographic factors that influence travel mode choice. Increasing transport cost hinders mobility for lowincome population. Mode choice behavior differs from place to place due to variances in socio-economic behavior, traveler attitude, and infrastructure facilities. Predicting travel choice behavior is one of the most important aspects of transportation modeling. Cities in the United Kingdom are addressing travel demand by providing proper public transportation, whereas cities in the United States are addressing the issue by providing better transportation. India is also experiencing traffic congestion, which can be reduced by conducting thorough research and forecasting of travel demand. The study aims to create a mode choice model for the city of Vadodara. The city is experiencing traffic congestion and has to develop policies to address the issue. This research aims to create a multinomial logit (MNL) mode choice model, using NLOGIT software. Role of gender in travel behavior: There is difference between travel behavior of women than that of man in most of the developing and some developed countries. For men and women of similar age, education, employment status, and household structure, commuting differences are seen (Hanson and Johnston 1985). Women trips are more complex, and they have multiple stops other than work trips such as drop and pick up children (Scholl 2002). Women makes more non-work trips such as household requirements, shopping, and childcare (Best and Lanzendorf 2005). Female choice in using mode is also different than man. They uses less cars and more public transportation (PT) even though it is available and they have to travel for longer distances (Srinivasan and Rogers 2005; Anand and Tiwari 2006). Role of Income in Travel behavior: Difference in level of mode choice and access to resources often influences destination choice, distance traveled, and mode choice (Dobbs 2005). In Japan man falling in higher income group is making more travel-related physical activity in comparison to low-income group (Matsushita et al. 2015). In a case study conducted in Huzhou city of China (Cheng et al. 2013) shows that urban low-income people in China generally have lower mobility than the nonlow-income. Lower income group residents of CBD travel shorter distance, make more trip rates, prefer NMT and PT for making trips, spend less time and money for making trips, than those staying away from CBD of same income groups. People of

Assessment of Gender and Income-Based Inequality in Travel Behavior …

97

low and low-med SEWS (socio-economic wellbeing score) make short trips and are mainly dependent on NMT for making trips (Jain and Tiwari 2020). Poor of developing as well as developed countries are mostly forced walkers, more dependent on public transportation and cycles. The quality of public transportation, accessibility, and cultural variables all have a role in low-income households’ mode selection. Low-income women were also shown a negative attitude toward public transportation. (Sun et al. 1998) found that household size, income, and vehicle ownership were all positively related to the number of household trips and vehicle miles traveled in a study based on the 1994 Portland activity-based travel survey (VMT). According to a study conducted for Upper Austria (Simma and Axhausen 2003), car ownership results in fewer trips by foot and public transportation and more trips by car.

2 Data Study area: Vadodara, commonly known as Baroda, is Gujarat’s third most populous town, behind Ahmedabad and Surat, and India’s 16th most populous city. According to the 2011 census, the city has a population of 16.6 lakh people living in 12 wards spread across a VMSS area of 125 square kilometers. Males make up 52% of the population, while females make up 48%. Vadodara’s population is concentrated in the city’s northern and central areas. Data Collection: For the study purpose, a revealed preference household survey for 12 administrative wards of Vadodara city from 1068 households. The number of households chosen was proportionate to the population of each ward as per Census 2011. For each surveyed household, data on socio-economic characteristics, asset ownership, and individual information was collected. Several households without any trips and wrong data were eliminated. So, the sample included 1042 households with 5670 trips after excluding unnecessary data during data cleaning. Minimum sample size of 384 household is sufficient for categorical data (Kotrlik and Higgins 2001). Trip purpose was classified into four groups such as work, education, shopping, and others. Trip purposes such as health, entertainment, leisure, and drop were all considered in other purpose. The households were categorized into four income groups, i.e., low, low-med, med-high, and very high based on household monthly income data. In case where income data was missing which accounted for about 20% of household data, linear regression was used to calculate missing income, taking income as dependent variable and vehicle ownership, house structure and employed person in family as independent variable. A telephonic interview was conducted to collect the mileage data of private vehicles to calculate the travel cost.

98

P. Prajapati and R. Khan

3 Data Analysis The gender distribution observed in the sample data is 32% female and that of male trips is about 68%. About 11% households were with income less than 15,000 INR falling in the low-income group. Table 1 shows that average trip length of male is higher than that of female for each mode used. Average trip length for trips made by bus is maximum. Results show that male walk for more distance compared to female. Average trip length of motorized 2-wheeler, car, vanpool, and auto is lowest for low-income group and increases with increase in income group. In case of cycle, the average trip length is highest for low income and decreases with increase in income groups. Table 2 shows variation in average travel cost for each mode by gender and income groups. It is seen that male spends more than female for traveling. This is obvious because males’ average trip length is higher than that of females. Car and vanpool are seen to be the more expensive. Vanpool is generally used for educational purposes and its charges are seen to be high. Average cost of motorized 2-wheeler is lowest as it is mainly used for shorter trips and also gives high mileage. When compared among income groups it is seen that travel expenditure on each mode increases with increase in income. Only for bus the travel cost decreases as usage of bus decreases with increase in income. Table 1 Average trip length (km) among socio-economic groups Modes

Gender groups

Income groups

Male

low

Female

Low-med

Med-high

Very high

M2W

5.85

4.27

4.14

5.11

5.69

6.10

Cycle

2.23

2.16

2.41

2.21

2.17

2.03

Car

7.51

5.43

4.75

5.73

7.20

7.72

Vanpool

5.04

4.49

2.47

3.87

5.63

5.07

Walk

0.83

0.75

0.77

0.85

0.70

0.79

Bus

10.52

7.46

8.49

8.57

11.21

5.23

Auto

4.81

3.3

3.12

3.68

4.88

6.44

Table 2 Average travel cost among socio-economic groups Mode

Gender Male

Income groups Female

Low

Low-med

Med-high

Very high

8.92

6.56

6.23

7.82

8.62

9.23

Car

26.68

18.47

17.79

20.18

24.51

28.62

Auto

16.27

16.86

12.53

17.14

18.7

19.56

M2W

Bus

18.45

15.15

18.12

17.29

16.91

15.94

Vanpool

23.72

24.61

20.18

23.87

24.85

25

Assessment of Gender and Income-Based Inequality in Travel Behavior …

99

100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

other shopping education

male

chi2(2)=24.511, sig=0.004

very high

med-high

low-med

low

very high

low-med

med-high

work low

share of trips

Trip purpose by socio-economic groups: Males are more likely to travel for work trips (60%) as compared to females (24%). Females are traveling more for shopping and other trips, i.e., leisure, health, entertainment, worship, drop, etc. This shows that females have more complex and multiple roles in making trips. About 40% females stay at home as compared to just 13% males. It is seen that there is less variation in activities involvement when compared among various income groups. Mode choice by socio-economic groups: Motorized 2-wheeler has the highest trip share in both the gender groups. Nearly half of the trips are conducted using motorized 2-wheeler. Males are more dependent on personal motorized vehicle as compared to female. The share of walk and PT is high in female as compared to male. Low and low-med income people walk trips are more than any other income groups. These walks trip decreases with increase in income. This shows that people within low-income groups have less access to resources and are thus forced walkers. Also, they are more dependent on public vehicles as they have no other choice. Cycle as a mode is less preferred by high-income groups as compared to low-income groups. As income increases people switches from NMT and PT to personalized vehicles. Trip length by socio-economic groups: The average trip length of male and female is 5.62 and 3.7 km respectively, while it is 3.25, 4.28, 5.59, and 6.04 km for low, low-med, med-high, and very high-income groups. There is a significant difference between distance traveled among both, gender groups and income groups (Fig. 1). Female makes shorter trips than male. About 70% female make trips within 4 km as compared to 50% male. When compared within income groups trip length increases with the increase in income. So, people with low-income thereby satisfy their needs within short distances. As per the ANOVA test shown by Table 3, there is a significant difference in trip length by gender and income groups both. Post hoc analysis as shown by Table 4 shows that there is no significant difference in trip length among med-high and very high-income groups.

female

chi2(2)=44.986, sig=0.000

Fig. 1 Trip purpose variation among income groups within gender groups

100

P. Prajapati and R. Khan

Table 3 ANOVA test for variation in trip length df

F

Sig.

Gender

Between groups Within groups Total

1 5668 5669

193.041

0.000

Income groups

Between groups Within groups Total

3 5666 5669

73.965

0.000

Table 4 Post hoc multiple comparisons—trip length by income groups Low

Low-med

Low-med

Mean difference (row-column) Significance

1.037 0.000

Med-high

Mean difference (row-column) Significance

2.441 0.000

1.403 0.000

Very high

Mean difference (row-column) Significance

2.789 0.000

1.752 0.000

Med-high

0.348 0.224

Trip purpose variation among income groups within gender groups: Work trips of females are seen to be increased with the increase in income. This shows that females get more opportunity to do work with the increase in household income. Shopping trips are highest for low-income females and decrease with increase in income groups this is due to the share of work trips increases from low to very high-income groups significantly. Whereas in male income groups shopping trips among each income groups do not differ much. Similarly, shopping and other trips of females are more as compared to males in each income groups. Mode choice variation among Income groups within gender groups: As shown in Fig. 2, there is significant difference in mode share among income groups within gender groups. Motorized 2-wheelers are the highest preferred mode among male income groups. Use of motorized 2-wheeler increases with increase in income groups among both the gender group but increases more for females. Low-income females walk share is highest compared to other modes. Walking as a mode is seen decreasing in both the gender groups with increase in income. Bus and auto share also decreases as income increases. This is due to people switches to personal vehicles with the increase in income. Cycle as a mode is used more by males compared to females in each income groups. In male cycle as a mode is used more in low and low-med income groups and least in med-high and very high-income groups. In female cycle is more used by low-med and med-high income groups and least by low and very-high income groups. With the increase in income cycle usage as a mode decline and it declines much faster in males as compared to females. Similar is the case for walk. Women in our culture wear saree and so it is difficult and unsafe to drive wearing saree. Although some females use cycle, their proportion is low compared to males.

101

100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

auto bus walk vp

male

very high

med-high

low-med

low

very high

med-high

low-med

car low

share of trips

Assessment of Gender and Income-Based Inequality in Travel Behavior …

cycle 2w

female

chi2(2)=541.185, sig=0.000

chi2(2)=342.202, sig=0.000

Fig. 2 Mode choice variation among income groups within gender groups

Trip length variation among income groups within gender groups: There is a variation in trip length among all income groups within gender groups and also between gender group within income groups. Table 5 shows that low-income male travel shortest distance (mean = 3.47 km) compared to other male income groups. Similarly, low-income females make shorter trips (mean = 2.78 km) compared to females of other income groups. There is not much variation in trip length of medhigh and very high-income groups in both within gender groups. Females in each income group make shorter trips as compared to male. Table 1 indicated that males’ average trip length by each mode is higher than females. Also, males walk more than females. This shows that females have less mobility compared to males. People of low and low-med income groups regardless of gender groups makes shorter trips and tends to satisfy their needs within shorter distances. As per ANOVA shown in Table 6, there is significant variation between income groups within gender groups and also between gender groups within each income groups. Post hoc analysis shown in Table 7 shows that within male income groups there is no significant difference in trips length between med-high and very highincome groups. Also, there is no significant difference between female’s low and low-med, med-high and very high- income groups. Table 5 Summary statistics for trip length Income groups Low

Male

Female

Mean

Std. Dev

Mean

Std. Dev

3.47

3.4

2.78

4.62

Low-med

4.81

4.56

3.12

3.29

Med-high

6.39

5.34

4.18

3.76

Very high

6.68

5.5

4.61

3.54

102

P. Prajapati and R. Khan

Table 6 ANOVA test for association in trip length among gender and income groups ANOVA test for variation in trip length between Income group within gender groups

ANOVA test for variation in trip length between gender within Income groups

df

F

Sig.

Male

Between groups Within groups Total

3 3874 3877

58.933

0.000

Female

Between groups Within groups Total

3 1788 1791

18.546

0.000

Low

Between groups Within groups Total

1 600 601

4.269

0.039

Low-med

Between groups Within groups Total

1 1815 1816

63.248

0.000

Med-high

Between groups Within groups Total

1 2337 2338

102.88

0.000

Very high

Between groups Within groups Total

1 910 911

33.958

0.000

Table 7 Post hoc multiple comparisons: trip length variation among income groups within gender groups Income groups

Male Low

Female Low-med Med-high Low

Low-med Med-high

Low-med

Mean 1.342 difference 0.000 (row-column) Significance

0.340 0.684

Med-high

Mean 2.918 1.576 difference 0.000 0.000 (row-column) Significance

1.406 1.066 0.000 0.000

Very high

Mean 3.212 1.869 0.000 0.000 difference (row-column) Significance

0.293 0.592

1.828 1.488 0.000 0.000

0.421 0.359

Variation in trip share among age groups within socio-economic groups Significant effect of age on daily mobility behavior across gender and income groups has been observed in the study. It is seen that mobility increases with age among both the gender groups. Old people above 59 years are seen less mobile than any other age group. Mobility of people falling in age group 27–59 is seen highest among both

Assessment of Gender and Income-Based Inequality in Travel Behavior …

103

the genders and income groups. People falling in this age group are mostly working people. Variation in trip length among gender groups within each age group Table 8 shows variation in trip length among gender and income groups within age groups. ANOVA test in Table 9 shows that there is no significant difference in trip length between and within gender groups among 0–16 years age group and above 59 years age group. Table 10 shows there is significant association between and within income groups among age groups. Table 8 shows that male makes comparatively longer trips than females. Children falling in age group 0–16 make shortest trips compared to all others age groups and their trips are mostly educational trips. This shows that the educational trips Table 8 Summary statistic of trip length Age

Gender groups Male m

Income groups

Female SD

m

Low SD

m

SD

Low med

Med high

Very high

m

m

m

SD

SD

SD

16

2.77

2.23

3.00

2.38

1.44

1.20

2.43

1.91

3.60

2.60

3.35

2.17

17–26

6.97

6.03

5.27

4.62

4.22

4.77

5.73

5.44

7.56

6.00

6.42

5.37

27–59

5.78

4.81

3.33

3.59

3.58

3.90

4.40

4.14

5.54

4.77

6.28

5.18

>59

4.22

3.91

3.54

3.31

2.18

1.93

3.19

2.94

3.44

2.89

6.56

4.87

Table 9 ANOVA Test for association between age and gender groups and for association between age and income groups Age

Gender groups

ANOVA test for association ANOVA test for association between age and gender groups between age and income groups df

0–16

17–26

27–59

> 59

F

Sig. 2.149

0.143

df

F

Sig.

3

38.262

0.000

19.299

0.000

31.461

0.000

18.988

0.000

Between groups

1

Within groups

934

Total

935

Between groups

1

Within groups

1535

Total

1536

Between groups

1

Within groups

2891

Total

2892

Between groups

1

Within groups

302

300

Total

303

303

932 935 29.012

0.000

3 1533 1536

173.968

0.000

3 2889 2892

2.237

0.136

3

104

P. Prajapati and R. Khan

Table 10 Post hoc multiple comparison among income groups within age groups Low Age 0–16 years

Age 17–26 years

Age 27–59 years

Age above 59 years

Low-med

Low-med

Mean difference (row-column) Significance

0.991 0.000

Med-high

Mean difference (row-column) Significance

2.158 0.000

1.166 0.000

Very high

Mean difference (row-column) Significance

1.903 0.000

0.911 0.002

Low-med

Mean difference (row-column) Significance

1.511 0.018

Med-high

Mean difference (row-column) Significance

3.347 0.000

1.836 0.000

Very high

Mean difference (row-column) Significance

2.207 0.000

0.695 0.338

Low-med

Mean difference (row-column) Significance

0.306 0.041

Med-high

Mean difference (row-column) Significance

0.298 0.000

1.142 0.000

Very high

Mean difference (row-column) Significance

0.347 0.000

1.886 0.002

Low-med

Mean difference (row-column) Significance

1.012 0.531

Med-high

Mean difference (row-column) Significance

1.256 0.354

0.243 0.958

Very high

Mean difference (row-column) Significance

4.382 0.000

3.370 0.000

Med-high

0.255 0.727

1.140 0.017

0.743 0.020

3.126 0.000

are satisfied within short distances for both male and female. Trip length increases with increase in income among each age group. Lower-income people among each age group are less mobile and are thus facing mobility issues. Post hoc analysis shown by Table 10 shows that there is insignificant difference in trip length among med-high and very high-income groups for age group 0–16 years, low-med and very high-income groups for age group 17–26 years. It is seen that people falling in above

Assessment of Gender and Income-Based Inequality in Travel Behavior …

105

60 age group have significant difference in trip length among low and very high, low-med and very high, med-high and very high-income groups only.

4 Model Development Multinomial Logit Model (MNL) is a frequent modeling approach for mode choice because it is simpler, provides a more economic model, and increases the power to detect relationships with other variables. The mode choice modeling software used in this study is NLOGIT, which is an extension of the LIMDEP econometric and statistical software package. Table 11 presents a list of variables used in the model. The descriptive analysis of the variables included in the study is given in Table 11. The final result with different utility equations for each mode gives parameter estimation as shown in Table 12. Walk mode is considered as base. Model result The probability of choosing walking is more for female compared to male. Walking is preferred by low-income group people and people with trip length up to 500 m have the highest utility for this mode and as trip length increases its utility decreases. With respect to education trip, people have higher probability to choose walking for work, shopping, and other purposes. The M2W users from age group 27–59 have the highest utility. The propensity of choosing this mode is the highest for work, followed by other purposes with respect to education and shopping trips. For households having more employed persons and family members, the probability of choosing M2W reduces. The probability of choosing bicycle is more for short trips compared to long trips. The utility of bicycling is higher for low-income group people. With respect to education and shopping trips, car is preferred for work and other trips. High-income group people have higher utility of car compared to low-income group people. The van pooling is preferred for education trips over other trips. The probability of choosing bus for low-income group people is the highest and as they have higher income, its probability is decreases. People have higher utility of bus for purposes other than work, education, and shopping trips. The utility of bus is higher for low access distance. When employed persons are increasing in household, they have higher probability of choosing bus while number of school going and college going members are increasing, their probability is decreasing. When the distance is more than 400 m from household, the utility of bus mode decreases and it is lowest when it is more than 1000 m. As the cost of travel increases, its utility decreases and is same for waiting time. The probability of choosing bus for long distance trip is higher. Like bus, the auto-rickshaw also has higher utility for other purposes than work, education and shopping. The higher propensity has observed for lower travel cost.

106

P. Prajapati and R. Khan

Table 11 List of variables and its description Socio-economic parameters

Travel parameters

Variables

Description

Variables

Description

GENDER1

Male

IVTT

In vehicle travel time in minutes

GENDER2

Female

NMTL1

Trip length ≤0.5 km

AGE1

Age of trip maker ≤16 years

NMTL2

Trip length 0.5–1 km

AGE2

Age of trip maker 17–26 years

NMTL3

Trip length 1–2 km

AGE3

Age of trip maker 27–59

NMTL4

Trip length ≥2 km

AGE4

Age of trip maker >60

ACS_L1

Access length ≤250 m

PRPS1

Purpose-work

ACS_L2

Access length 250–500 m

PRPS2

Purpose-education

ACS_L3

Access length ≥500 m

PRPS3

Purpose-shopping

ACS_LTH

Access length in m

PRPS4

Purpose-other

COST

Cost of trip in rupees

INC1

Household income (≤15,000)

COST1

Cost ≤5 rupees

INC2

Household income (1500–35,000) COST2

Cost 5–10 rupees

INC3

Household income (35,000–75,000)

Cost 10–20 rupees

COST3

INC4

Household income (≥75,000)

COST4

Cost 20–50 rupees

M2W

No of motorized two-wheeler in household

COST5

Cost ≥50 rupees

CAR

No of cars in household

WT_TIME

Waiting time in minutes

CYCLE

No of cycle in household

BUSDIS1

Home to bus stop distance ≤400 m

FIN_INC

Monthly household income

BUSDIS2

Home to bus stop distance 400–1000 m

HH_MEM

No of members in a house

BUSDIS3

Home to bus stop distance ≥1000 m

EMP_PRS

No of employed person in a house

CLG_PRS

College going person in a house

SCH_CHLD

School going children in a house

Also, it is preferred for longer distance journey compared to shorter one. Collegegoing students have higher utility of auto-rickshaw. Model validation If models are to be accepted and utilized to support decision-making, model validation is important element of the development process. Validation ensures that the model satisfies its objectives in terms of the methodology used and the outcomes obtained. The goodness of fit of the model is reasonable, and the data used in this model are found to be significant. From Table 13 every coefficient has significance level less than 0.05 which means better coefficient. Model result shows for our Pseudo-R2

Assessment of Gender and Income-Based Inequality in Travel Behavior …

107

Table 12 Result of MNL model Modes

Variables

Coefficient

p-value

Walk (BASE)

Gender (male = 1, female = 0)

−1.4817

0.0000

Income >35,000 (BASE) Income1 (≤15,000)

0.8059

0.0000

Income2 (15,001–35,000)

0.6868

0.0000

Trip length (≥2 km, BASE) Trip length (≤0.5 km)

5.9893

0.0000

Trip length (0.5–1 km)

4.3935

0.0000

Trip length (1–2 km)

2.7413

0.0000

0.8348

0.0000

Purpose2 (education = BASE) Purpose1 (work)

M2W

Purpose3 (shopping)

0.6445

0.0011

Purpose4 (other)

1.9720

0.0000

Constant

4.7366

0.0000

Gender (male = 1, female = 0)

−0.9812

0.0000

−2.6135

0.0000

Age4 (>59 years, BASE) Age1(35,000, BASE) Income1 (≤15,000)

0.8745

0.0007

Income2 (15,001–35,000)

0.5513

0.0017

Cycle ownership

0.3846

0.0038

IVTT (min)

−0.0407

0.0002

1.7399

0.0000

Trip length (≥2 km, BASE) Trip length (≤0.5 km) Trip length (0.5–1 km)

1.0948

0.0001

Trip length (1–2 km)

0.4948

0.0371 (continued)

108

P. Prajapati and R. Khan

Table 12 (continued) Modes

Variables

Coefficient

p-value

Car

Constant

6.3228

0.0000

Purpose (work and other = BASE) Purpose2 (education)

−5.0151

0.0000

Purpose3 (shopping)

−1.9716

0.0000

IVTT (min)

0.0290

0.0000

−1.3676

0.0051

Income4 (≥75,000, BASE) Income1 (≤15,000)

Vanpool

Income2 (15,001–35,000)

−0.6230

0.0042

Income3 (35,001–75,000)

−0.5021

0.0008

HH employed person

−0.6469

0.0000

Constant

1.0545

0.0089

Gender (male = 1, female = 0)

−1.6126

0.0000

3.2558

0.0000

−0.4411

0.0349

Purpose (work, shopping and other = BASE) Purpose2 (Education, BASE) Income 70,000) Bus

Constant

0.9591

0.0212

Gender (male = 1, female = 0)

−1.7170

0.0000

Income4 (≥70,000, BASE) Income1 (≤15,000)

2.1605

0.0000

Income2 (15,001–35,000)

1.5854

0.0000

Income3 (35,001–70,000)

0.8283

0.0136

0.7914

0.0055

Purpose (work, education and shopping = BASE) Purpose4 (other) Access length (≥500 m) (BASE) Access length (≤250 m)

2.9252

0.0000

Access length (250–500 m)

1.3373

0.0000

HH employed person

0.2487

0.0136

College going person

0.2826

0.0089

School going children

−0.3775

0.0003

Cost

−0.3312

0.0000

Waiting time (min)

−0.3380

0.0000

IVTT (min)

0.4240

0.0000

Bus stop distance (≤400 m, BASE) Bus stop distance (400–1000 m)

−1.094

0.0000

Bus stop distance (≥1000 m)

−1.763

0.0102 (continued)

Assessment of Gender and Income-Based Inequality in Travel Behavior …

109

Table 12 (continued) Modes

Variables

Coefficient

p-value

Auto-rickshaw

Constant

−17.1415

0.0000

Gender (male = 1, female = 0)

−1.6492

0.0000

Age

0.0340

0.0000

Income

−0.1035 * 10–4

0.0006

Cost (≥50 rupees, BASE) Cost (≤5 rupees)

19.4055

0.0000

Cost (5–10 rupees)

19.8047

0.0000

Cost (10–20 rupees)

17.5722

0.0000

Cost (20–50 rupees)

14.6662

0.0000

Access length (m)

−0.0008

0.0063

IVTT (min)

0.2431

0.0000

Purpose4 (other)

0.7633

0.0014

College going person

0.2707

0.0042

Purpose (work, education and shopping = BASE)

Table 13 Model fitting information

Model

Model fitting criteria

Likelihood ratio tests

−2 Log likelihood

Chi-square

Sig.

Intercept only

−7386.435

–

–

Final

−4042.8

9060.197

0.000

Pseudo R2 = 0.453 Adjusted R2 = 0.446

value of 0.453 we got the linear regression model’s R2 value of 0.87. For a discrete choice model, a pseudo-R2 of 0.453 is considered a good fit. In reality, for the linear model equivalent, pseudo-R2 values between 0.4 and 0.5 can be interpreted as R2 values between 0.8 and 1.

5 Conclusion The mobility of women in the city is low and that of low-income women is the lowest. Mobility of women is seen increasing with the increase in income. It is seen that the city has short trip length and female in each income groups has shorter trip lengths than male. As the income increases the trip length increases but the increase is higher for males than that of females. This indicated that there is a need to meet the demand of both females and males belonging to med-high and very high-income groups within short distances. This necessitates community-based land use planning.

110

P. Prajapati and R. Khan

Walk trips of women are higher than that of males but the average trip length of male is higher than female which indicates that male walks for longer distances. Also, females in each income group make more walk trips compared to males. As income increases walk share decreases among both the gender groups. Thus, it is essential to improve pedestrian infrastructure which will also attract potential pedestrians. As per the findings females and people within low and low-med income groups are more likely to travel shorter distances and rely more on NMT. Low and low-med income groups are more dependent on public transportation as they have no access to personal vehicles thus makes them captive users. As income increases people shifts from NMT and public transport (PT) to private modes and this shift is rapid in female than that of male. To encourage people in the middlehigh and very-high income groups to use the bus, strategies need to be developed. Also, it is seen that females are more dependent on PT such as vanpool, bus, and auto-rickshaw compared to males. In Vadodara, women take more walks and utilize PT at the same rate as men, indicating that gender-specific issues such as safety and security associated with PT and walking do not exist. Main trip purpose for men is work, that for women is education followed by shopping. Work trips for females increase with increase in income. Thus, opportunity of women to work increases with rise in household income. Cycle as a mode is mostly preferred by low and low-med income households in comparison to med-high and very high-income groups. People belonging to age group below 17 years are highest cycle users and there too females are using more cycle as compared to male. Their main trip purpose is education. Usage of cycle reduces with increase in age among both the gender groups and this reduction is at much faster rate among females. Result shows that as the access length and waiting time increases the probability to choose bus as a mode reduces. This indicates that if the access length is more for bus station its usage decreases significantly and people shift to some other mode. Thus, in order to attract people to use bus, bus stops should be easily accessible at shorter distances and also the frequency of bus has to be increased. However, there is a need to improve transportation infrastructure, which, while offered for the study city, the report does not go into detail in testing the quality of the infrastructure. Looking ahead, it is clear that the public transportation system must be developed and the fleet expanded to keep up with the population’s ever-increasing size and travel needs. As incomes rise and attitudes change, females are increasingly able to travel by personal motorized two-wheeler. Thus, the motorized two-wheeler fleet in Vadodara is expected to grow significantly in the coming years. The road network must be planned accordingly. Future Scope We were unable to account for location-based factors affecting accessibility and mobility in the study. Due to resource constraints, spatial planning-related interventions may be required to give equal opportunity to those in the low and low-med income groups within short distances. What types of transportation options should be considered if women’s mobility needs to be increased in order for them to access their

Assessment of Gender and Income-Based Inequality in Travel Behavior …

111

life choices? Would economic growth be accelerated if women were provided affordable transportation services, and if so, what would these modes of transportation be? These are issues that need to be researched and thought about further.

References Anand A, Tiwari G (2006) A gendered perspective of the shelter-transport-livelihood link: the case of poor women in Delhi. Transp Rev 26(1):63–80. https://doi.org/10.1080/01441640500175615 Best H, Lanzendorf M (2005) Division of labour and gender differences in metropolitan car use. an empirical study in Cologne, Germany. J Transp Geogr 13(2):109–121. https://doi.org/10.1016/ j.jtrangeo.2004.04.007 Cheng L et al (2013) Travel behavior of the urban low-income in China: case study of Huzhou city. Procedia Soc Behav Sci 96(Cictp):231–242. https://doi.org/10.1016/j.sbspro.2013.08.030 Dobbs L (2005) Wedded to the car: women, employment and the importance of private transport. Transp Policy 12(3):266–278. https://doi.org/10.1016/j.tranpol.2005.02.004 Hanson S, Johnston I (1985) Gender differences in work-trip length: explanations and implications. Urban Geogr 6(3):193–219. https://doi.org/10.2747/0272-3638.6.3.193 Jain D, Tiwari G (2020) ‘Gender and income based variability in travel choices in Vishakhapatnam, India. Transp Res Procedia 48:2870–2890. https://doi.org/10.1016/j.trpro.2020.08.232 Kotrlik JWKJW, Higgins CCHCC (2001) Organizational research: determining appropriate sample size in survey research appropriate sample size in survey research. Inf Technol Learn Perform J 19(1):43 Matsushita M, Harada K, Arao T (2015) Socioeconomic position and work, travel, and recreationrelated physical activity in Japanese adults: a cross-sectional study health behavior, health promotion and society. BMC Public Health 15(1). https://doi.org/10.1186/s12889-015-2226-z Scholl L (2002) Transportation affordability for low-income populations: a review of the research literature, ongoing transportation assistance programs (October), pp 1–96 Simma A, Axhausen KW (2003) Interactions between travel behaviour, accessibility and personal characteristics: the case of Upper Austria. Eur J Transp Infrastruct Res 3(2):179–197 Srinivasan S, Rogers P (2005) Travel behavior of low-income residents: studying two contrasting locations in the city of Chennai, India. J Transp Geogr 13(3):265–274. https://doi.org/10.1016/ j.jtrangeo.2004.07.008 Sun X, Wilmot CG, Kasturi T (1998) Household travel, household characteristics, and land use: an empirical study from the 1994 Portland activity-based travel survey. Transp Res Rec 1617:10–17. https://doi.org/10.3141/1617-02

Traffic Impact Study of an Integrated Township and Formulation of Improvement Measures—A Case Study of Technocity in Thiruvananthapuram V. S. Sanjay Kumar , P. N. Salini , Ebin Sam , and S. Akshara

Abstract Rapid urbanization and unplanned development paves way to traffic congestion, road crashes, and fatalities. Technocity, a technology park is being developed as a part of Technopark’s fourth phase, about 5 km away from the Technopark campus in Thiruvananthapuram, the capital of the Indian state Kerala. The project is expected to provide six lakh job opportunities and thus, heavy traffic will be drawn into the campus, significantly affecting the traffic scenario on the encompassing roads. Traffic Impact Assessment is an essential planning tool used to analyze the impact of the additional traffic generated by the proposed campus on existing transport system in an area. This paper aims to delve into the inquisitive understanding regarding the impact of traffic generated by Technocity on roadway segments under the influence of the campus and also on its internal roads. To assess the present traffic scenario, traffic volume survey was conducted at various locations. In accordance with the current situation of traffic and based on the results obtained through efficient analysis techniques, appropriate enhancement measures have been identified for smoothening the flow of traffic. Keywords Technocity · Traffic impact assessment · Travel demand modeling · Trip generation rate

1 Introduction Kerala has developed its Information Technology (IT) initiatives in a “hub and spoke model.” The major IT parks in the state, namely, Technopark Trivandrum, Infopark Kochi, and Cyberpark Calicut, act as the IT hubs of the state. Technopark is one of the India’s largest IT parks and the world’s greenest Technopolis spreads across V. S. S. Kumar (B) · P. N. Salini · E. Sam · S. Akshara KSCSTE-NATPAC, Thiruvananthapuram, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_9

113

114

V. S. S. Kumar et al.

766.86 acres of land with 106 lakhs sq. ft. built-up space (completed) and about 460 businesses operational at present. Technocity, an integrated township, is being developed as a part of Technopark’s fourth phase stretching over 182 ha. and is planned for a total built-up area of 27 million square feet with multiple buildings for its tenant organizations. The Technocity is being built on either side of National Highway 66 (NH 66), the major transport corridor in the state, making it two campuses. The Technocity project is expected to provide about six lakh job opportunities in the future; consequently, there will be heavy inflow of traffic from/into these campuses from the NH. Hence, the impact of traffic generated by the campus needs to be assessed to formulate traffic management measures. Traffic Impact Assessment (TIA) is an essential tool to analyze the impact of the additional traffic generated by a proposed facility on existing transport system in an area. This paper aims to assess the impact of traffic generated by the proposed development in Technocity campus on the encompassing roads and also on its internal roads.

2 Objectives In this case study, we aim to gain an in-depth understanding of the impact of Technocity, an integrated township under construction, on the roads under the influence of the campus and also on its internal roads. The objectives of the study are as follows: • To estimate the traffic generated by the proposed Technocity campus. • To assess the impact of the traffic generated on access roads and major intersections with NH 66. • To suggest appropriate mitigation measures to ensure safe and efficient movement of traffic.

3 Literature Review Padma et al. (2019) highlighted the importance of TIA and brought out the need to make TIA an integral and mandatory requirement in the case of new as well as capacity augmentation by considering case studies in Delhi and Tamil Nadu. They computed the percentage impact of the proposed complex on the horizon year traffic. Nafis et al. (2018) did an inquisitive understanding of the impact of Khilgaon flyover on roadway segments along the corridor and also the adjacent flyover. Minhans et al. (2013) used the regimes of Trip Rate Analysis, Cross-Classification Analysis, and Regression Analysis to assess the future traffic expected to be caused by the proposed Tesco hypermarket (TH) in Skudai. Regidor et al. (2005) conducted traffic impact studies for high-rise mixed-use condominiums located in central business districts in Metro Manila and did a systematic analysis of “with” and “without” development scenarios. Limapornwanitch et al. (2005) explained the current TIA applications in

Traffic Impact Study of an Integrated Township and Formulation …

115

two developing cities, namely, Thailand and Philippines and compared their strengths and weaknesses.

4 Study Area Technocity is being built in Thiruvananthapuram, Kerala, at Pallippuram. The campus is demarcated into 25 plots with varying land uses. It is an integrated township for electronics, software, and other information technology (IT) applications. In addition to IT firms, it also provides residential, commercial, hospitality, and educational facilities. It is a self-sufficient satellite city that would not strain the resources or infrastructure of Thiruvananthapuram. Technocity is being developed after proper planning that includes business parks, parking lots, bus terminals, motorcyclist parks, residential areas, and areas for commercial activity. Major part of the land is allocated to IT space. The study area of the project concentrated mainly on that aspect. Figure 1 shows the study area along with proposed land uses.

Fig. 1 Study area (Source Master plan for Technocity)

116

V. S. S. Kumar et al.

5 Methodology The study paves the way to investigate and develop traffic management plans that help to minimize traffic congestion. To start with, a reconnaissance survey was conducted to have an appraisal of the study area and to identify the links carrying substantial traffic flow in the region. Based on the reconnaissance and considering the site conditions, plans for executing detailed field surveys were developed. Primary survey included traffic volume survey which is conducted to understand the efficiency of the roads and the amount of traffic playing through them. Road inventory survey was conducted to collect the geometric and other parameters from the field. The secondary data pertaining to the study area was also collected. The analysis of data is done on the basis of the field surveys and other sources of available information. Travel demand modeling was performed to predict the future travel patterns. In accordance with the current situation of traffic plying on the National Highway and based on the results obtained through efficient analysis techniques, the appropriate traffic improvement measures have been identified for the smooth flow of traffic. The methodology framework formulated for the study is shown in Fig. 2.

Reconnaissance Survey Secondary Data Data Collection Primary Data

Anticipated Total Traffic towards NH Traffic Assignment

Modal Split

Traffic Improvement Measures

Fig. 2 Methodology framework

Traffic Volume Survey Road Inventory Estimation of Traffic from Technocity Link Volume Count

Projection of Existing Traffic

Intersection Volume Count

Data Analysis

Traffic Impact Study of an Integrated Township and Formulation …

117

6 Data Collection and Analysis Data collection and analysis are the basic steps in any study. To assess the present traffic scenario, traffic volume survey was conducted at Pallippuram and Mangalapuram, two major intersections on NH 66 in the vicinity of the Technocity campus and at Karamood, an intersection on Mangalapuram-Pothencode road which is running along the boundary of the study region in the east–west direction. Mid-block traffic volume survey was conducted in Pallippuram-Mangalapuram link of NH 66. Road inventory was done to get the road data parameters such as carriageway width and shoulder width. The secondary data collected includes the proposed master plan for the work center, tentative land use (Source: Technopark) and proposed road development plan for roads in the influence area, including the NH 66 (Source: NHAI). The analysis is based on the field studies and observations, and also on various secondary data. It helps in understanding the pattern of vehicular traffic prevailing under the present situation and is an important input for the design of facility improvements. Traffic data obtained from the field surveys were computed in terms of peak hour traffic volume in PCU as per IRC SP 41:1994 for intersections and IRC 106:1990 for urban roads. Peak hour is normally considered as 8–10% of the Average Daily Traffic. The existing scenario of traffic in the National Highway is analyzed and is tabulated as shown in Table 1. From Table 1, the highest peak hour traffic volume was observed at Mangalapuram Junction with 4205 PCU per hour. The traffic characteristics of the roads are changing as years are passing mainly due to change in pricing of vehicles and change in economy. The vehicle ownership levels are increasing tremendously. The existing vehicular composition at the major junctions is depicted in pie chart as shown in Fig. 3. From Fig. 3, it can be inferred that car composition forms the highest in Mangalapuram and Pallippuram junction followed by Two-wheeler whereas at Karamood Junction, two-wheeler composition forms the highest percentage followed by car. Table 1 Details of existing traffic volume (in PCU)

Location

Peak hour traffic volume in PCU

Pallippuram junction

3368

Mangalapuram junction

4205

Karamood junction Pallippuram–Mangalapuram link

924 3776

118

V. S. S. Kumar et al.

Vehicular Composition at Pallippuram

Vehicular Composition at Mangalapuram

4%

1% 1% 1% 5%

2% 1%

3% 1% 3%

7%

5%

27%

1%

1%

11%

44%

23%

44%

1%

7%

7%

KSRTC Minibus/Tempo Auto rickshaw LCV 3A Goods auto

KSRTC Other buses Minibus/Tempo Car Auto rickshaw Two wheeler LCV

Other buses Car Two wheeler 2A MAT

Vehicular Composition at Karamood 1% 1% 1%

3%

Minibus/Tempo

15% 7%

KSRTC

32%

Car Auto rickshaw Two wheeler

34%

2A 6%

3A

Fig. 3 Existing vehicular composition

7 Travel Demand Modeling Travel Demand Modeling, an important aspect of Transportation planning, is carried out to establish the spatial distribution of travel explicitly by means of an appropriate system of zones. Total trips generated from the Technocity campus are estimated by using Trip Rate method, considering the built-up area, trip rate per peak hour, and Floor Area Ratio (FAR). The trip rates are adopted from the previous studies (NATPAC 2020), whereas permissible FAR as per relevant rules are followed. The

Traffic Impact Study of an Integrated Township and Formulation …

119

total trips generated are converted into vehicular trips using occupancy values, which in turn is distributed mode-wise based on existing traffic proportion in the Technopark Phase I. Finally, it is assigned to the intersections and identified links in the vicinity of the campus based on the stochastic equilibrium approach, i.e., based on the perception of users.

7.1 Estimation of Trips Generated The total number of trips originating and ending in each zone of the campus is estimated using Trip Rate, Built-up Area, and Floor Area Ratio. Estimation of Trip Rate. Trip rate is computed from the total trips generated from Technopark Phase 1 by its total built-up area (NATPAC 2020). As per the previous studies (NATPAC 2020), it was found that 5980 vehicular trips per hour are generated from Technopark Phase 1, which possesses a total built-up area of 6 million sq. ft. This corresponds to a vehicular trip rate of 0.98 per 1000 sq. ft. Estimation of Floor Area Ratio. The total built-up area proposed is 645,800 sq. ft. for a land area of 10.33 acres, corresponding to an FAR of 1.4. The permissible FAR as per Kerala Municipal Building Rules is 3. The adopted FAR is used for estimating the proposed built-up area in all the land plots under consideration, wherever the corresponding values are not known. Based on the trip rate and floor area ratio considered, total number of trips generated in the peak hour is computed and is found to be 16,802 vehicular trips.

7.2 Modal Split The traffic generated from the Technocity is segmented into varied category of vehicles based on existing traffic proportion in the Technopark Phase I as obtained from the previous studies (NATPAC 2020). The modal split used is given in Fig. 4.

7.3 Traffic Assignment Traffic assignment is done based on stochastic equilibrium approach, i.e., based on the perception of users. According to this approach, all reasonable paths between the origin and destination will have flow.

120

V. S. S. Kumar et al.

Fig. 4 Modal split 0.5 0.5

0.5

0.5

10 43

45 Car/Jeep Public bus Other buses Mini truck

Two-wheeler Passenger Auto rickshaw Mini bus

The peak hour traffic along the major intersections in the vicinity of the proposed campus is given in Table 2. The peak hour traffic along the major links in the vicinity of the study region is given in Table 3 Table 2 Anticipated total traffic during peak hour Junction

Projected traffic

Traffic from technocity

Total traffic (PCU/peak hour)

Pallippuram

4507

7369

11,876

Technocity-NH 66 intersection

5412

7974

13,386

Karamood

1374

1374

2611

Mangalapuram

5627

1924

7551

Table 3 Anticipated traffic along the road links Links

Peak hour traffic from technocity

Existing peak hour traffic

Total peak hour traffic (PCU/hr)

Daily traffic (PCU)

Kallara-Pallippuram road

4083

649

4768

27,355

Technocity Internal road

5057

0

5095

29,232

Karamood-CRPF road

999

338

1347

7728

Mangalapuram-Karamood road

999

1515

1474

14,533

Traffic Impact Study of an Integrated Township and Formulation …

121

Volume

15000 Palippuram Jn. (A)

10000 5000 0 Without

With

Towards technocity Main Gate(B) Karamood Jn.(C)

Scenario 1 (2026) Fig. 5 Traffic volume “with” and “without” campus development

8 The “With” and “Without” Scenario The traffic at the major intersections with and without the proposed campus development is estimated for the year 2026 and is shown in Fig. 5. It is seen that the proposed campus has a significant impact on the surrounding roads, which makes it necessary to advocate for the traffic-related problems to suggest suitable redressal measures to mitigate it.

9 Improvement Proposals According to IRC SP 92:2017, grade separated intersection is justified when the total traffic in all arms of an intersection exceeds 10,000 PCU per hour. The anticipated traffic at Pallippuram Junction and Technocity-NH Intersection are well beyond 10,000 PCU/h. So, these junctions warrant grade separated facility in the form of flyover/vehicle underpass/overpass. National Highway Authority of India (NHAI) has already proposed grade separators at Mangalapuram Junction and Technocity-NH intersection as part of the six-laning of NH66. As per IRC 64:1990, a four-lane bi-directional traffic of 24 m width as shown in Fig. 6 is recommended along Kallara-Pallippuram road, which can accommodate 35,000 PCU. Two-wheelers form 45% of the total traffic and hence an exclusive lane for two-wheelers is also proposed. Also, enhanced pedestrian facilities in the form of footpath of 1.8 m width on either side are proposed to ensure pleasant and comfortable walking for pedestrians. It is recommended to have a 4-lane bi-directional road with separate lane for twowheelers and footpath along Technocity Internal road, which may have to cater peak hour traffic of 29,200 PCU.

122

V. S. S. Kumar et al.

Fig. 6 Typical cross section of roadway for Kallara-Pallippuram road

Fig. 7 Typical cross section of roadway for Technocity Internal road

10 Conclusion The study aims to analyze the impact of traffic generated from Technocity campus on the roads under the influence of the campus and also on its internal roads to ensure the smooth movement of traffic on NH. Traffic Impact Assessment (TIA) is used to determine the possible implications of the proposed development on the transportation and traffic system and thereby minimize the negative impacts. The study was accomplished by conducting detailed traffic volume studies. Thus, the existing scenario of traffic in the National Highway is analyzed and traffic is projected based on growth rate method. Then, Travel demand modeling was carried out to understand where the trips come from and where they go, and what modes and which routes are used. In view of current traffic conditions on the National Highway and in light of the results obtained by efficient analysis techniques, the appropriate traffic improvement measures have been suggested for managing traffic flow along the various roads.

Traffic Impact Study of an Integrated Township and Formulation …

123

References Azra N, Hoque S (2014) Implementation of traffic impact assessment in developing countries: case study of Bangladesh. Int J Transp Highw Railw Eng 1(3). [ISSN : 2475-2746] Anwari N, Islam MR, Hoque MS (2018) Traffic impact assessment of Khilgaon flyover. J Sci Technol Limapornwanitch et al (2005) The implementation of traffic impact assessment in southeast Asian cities: case studies of Thailand and the Philippines. J East Asia Soc Transp Stud 6:4208–4223 Minhans et al (2013) Traffic impact assessment: a case of proposed hypermarket in Skudai Town of Malaysia 65(3):1–7. www.jurnalteknologi.utm.my | eISSN 2180–3722 | ISSN 0127–9696 May LW, Rahman RA, Hassin MF, Diah JM, Mashros N, Abdullah ME, Bin Mohd Masirin MI (2019) An overview of the practice of traffic impact assessment in Malaysia. Int J Eng Adv Technol (IJEAT) 8(5c). ISSN: 2249 – 8958 Padma S, Velmurugan S, Kalsi N, Ravindera K, Erramapallia M, Kannan S (2020) Traffic impact assessment for sustainable development in urban areas. Transp Res Proced 48:3173–3187; World conference on transport research—WCTR 2019, Mumbai Regidor JRF, Teodoro RVR (2005) Traffic impact assessment for sustainable traffic management and transportation planning. In: Urban areas proceedings of the eastern Asia society for transportation studies, vol 5, pp 2342–2351 Technical Study for Technopark Way-In Expansion Works, NATPAC (2020) Withanaarachchi J, Setunge S, Bajwa S (2012) Traffic impact assessment and land use development and decision making. In: International conference disaster management, pp 256–273

Rollover Stability Analysis of Trucks-Effect of Curve Geometry and Operating Speed Y. K. Remya, Jacob Anitha, and E. A. Subaida

Abstract Road crashes have become a major concern worldwide. Rural highways account for more than 66% of total road fatalities as the speed of vehicles on these highways are very high. Rollover crashes at curved roads in these areas are mostly serious and cause severe damage and injury than other kinds of vehicle crashes. The relatively low rollover stability of heavy commercial vehicles promotes rollover and contributes to the number of heavy vehicle crashes. Inconsistency in geometric design of rural curves can present an unpredictable road profile before the driver resulting in erroneous and dangerous maneuvers. A way to improve safety on curves is improving its geometric design. Research has been conducted to evaluate the dynamic behavior of vehicles on curves and their relation with highway geometry. But uniformity was not observed in the geometric parameters identified to affect rollover stability. So, this research evaluated the influence of geometry of horizontal curves on rollover stability of vehicles with focus on two-axle trucks loaded up to its Gross Vehicle Weight (GVW). The lateral behavior of a cornering truck under varying curve geometry was examined using vehicle dynamics simulation software IPG TruckMaker and lateral acceleration experienced by the truck while cornering was used as a measure of its stability against rollover. It was observed that the most significant geometric variable affecting rollover stability of trucks is radius of the horizontal curve. Length of the curve and superelevation also affect truck stability. Models which predict lateral acceleration experienced by a truck at point of curvature, midpoint of curve, and Y. K. Remya (B) · E. A. Subaida Department of Civil Engineering, Government Engineering College, APJ Abdul Kalam Technological University, Thrissur 680009, India e-mail: [email protected]; [email protected] E. A. Subaida e-mail: [email protected] Y. K. Remya Department of Civil Engineering, SCMS School of Engineering and Technology, APJ Abdul Kalam Technological University, Karukutty 683576, India J. Anitha Department of Civil Engineering, Government Polytechnic College, Chelakkara, Thrissur 680586, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_10

125

126

Y. K. Remya et al.

point of tangency were developed. Maximum lateral acceleration prediction models were also developed. The outputs of this research can be used as road safety audit tools to identify potential stability issues on curves and to design safe curves. Keywords Rollover stability · Trucks · Highway geometry · Road safety

1 Introduction Road crashes are a leading cause of death and serious injury across all countries in the world. To improve global road safety, United Nations General Assembly has proclaimed the Decade of Action for Road Safety 2021–2030, with the ambitious target of preventing at least 50% of road traffic deaths and injuries by 2030. As stated in Highway Safety Manual published by American Association of State Highway and Transportation (AASHTO 2010), roadway factors contribute to 34% of road crashes. An inconsistent road design can mislead a road user, thereby creating human errors and can trigger a crash (Ahmed 2013). Rural highways are considered as deadliest as fatalities and serious injuries are 2.5 times more than on urban highways. Curves on rural highways are dangerous as operating speed of vehicles will be high. Studies have tried to relate road geometric design elements such as number of lanes, sight distance, superelevation, median width and type, lane and shoulder width, curve radius, gradient, and horizontal and vertical alignments to accident rates (Zhang et al. 2009; Abhilash and Ravikumar 2015; Geedipally 2017; Islam et al. 2018; Mohammed 2019). There are few studies in the literature that identifies factors causing a rollover crash. Heterogeneous negative binomial crash prediction model can estimate the number of rollovers as a function of road geometry, the environment, and traffic conditions (Zou et al. 2012; Mohammadi and Samarayanake 2014; Hosseinpour 2016). Weather, road surface conditions, and speeding were found to have heterogeneous and significant impact on rollover crash risk (Alrejjal et al. 2022). Straightening sharp horizontal curves, widening shoulder width, better design of centreline medians, gentle down gradient can reduce rollover crash rates (Hosseinpour 2016; Yin 2020). The effects of the interaction of horizontal alignment, speed, and superelevation are very important on vehicles’ rollover risk (Yin 2020). There are three ways to identify the rollover of vehicles: the real static models, the real dynamics models, and the mathematical or simulation models. However, due to the high costs associated with the first two kinds of tests, mathematical models are the most used in research (Moreno 2018). Studies have indicated that heavy commercial vehicles have a high percentage of involvement in rollover crash compared with their low percentage in the vehicle population (NHTSA 2006). A few studies were carried out to analyze heavy vehicle rollover stability and their relationship with curve geometry. Load mass and centroid position have the greatest impact on rollover stability of trucks (Zhao et al. 2014). Tight curves along with steeper downgrades would amplify the impact of the truck configurations on its rollover risk (Alrejjal and Ksaibati 2022). Previous literatures

Rollover Stability Analysis of Trucks-Effect of Curve Geometry …

127

were based on steady-state cornering conditions and no uniformity was observed in the geometric variables identified to affect rollover stability. Models which predict lateral acceleration experienced by a truck as function of road geometry are not developed yet. So, the objective of this study was to identify various geometric parameters that significantly affect rollover stability of a truck loaded up to its Gross Vehicle Weight (GVW). The effect of steering wheel angle on lateral stability of truck was also studied. Regression analysis was conducted to examine and quantify the relationship between lateral acceleration, operating speed, and curve geometry, there by developing lateral acceleration prediction models for salient points along the curve such as point of curvature, midpoint of curve, point of tangency, and maximum lateral acceleration.

2 Methodology 2.1 Development of Vehicle Model and Validation A two-axle truck with rear twin tires loaded up to its GVW, which is here after called ‘fully loaded truck’ was considered for the study. The dynamic response of the cornering truck was evaluated using vehicle dynamics simulation software IPG Truckmaker version 7.1.5. Lateral acceleration experienced by the truck while cornering was used as a measure of its stability against rollover. The simulation software was calibrated and validated using real field data collected with a two-axle truck equipped with acceleration sensor. Android phones with acceleration sensor installed in them were used for this purpose. Android phones with active sensor were placed at different positions inside the vehicle. Four sensors were placed just above the wheel positions and the remaining two sensors were placed above the front and rear axle centers. Two-axle truck used for the study has characteristics as shown in Table 1. The truck with active sensors was operated through more than 20 horizontal curves of known geometry for a number of trials. Table 2 shows the summary statistics of geometry of these horizontal curves which was used for model calibration and validation. The data collected by the sensors at different positions during various trials were retrieved and sorted to obtain lateral acceleration values along each curve at each sensor position and trial. This lateral acceleration data along each curve was subjected to Analysis of Variance (ANOVA) test to check whether there is any significant difference between lateral acceleration values at six sensor positions. It was observed that there is significant difference between lateral acceleration values at different sensor positions. Also, lateral acceleration experienced above rear left wheel position was the highest at curves toward right, whereas lateral acceleration experienced above rear right wheel position was the highest at curves toward left. This indicated that rear outer wheel position was the critical position exhibiting maximum lateral acceleration. Critical lateral acceleration experienced by the truck

128

Y. K. Remya et al.

during field survey was used to calibrate and validate the simulation software for this study. Road models were developed in the simulation software for the horizontal curves used during field study. Each road model consisted of a straight section, transition curve if any, horizontal curve followed by transition curve at the end and straight section. A road model developed in the software is shown in Fig. 1. A Vehicle model of the truck used during field survey was also generated. The initial speed of the truck entering the road model was made the same as that of real field conditions when the simulation began and the speed profile along each curve was plotted. Lateral acceleration experienced by the truck along each road model was also plotted. Speed and lateral acceleration profile at salient points along each road model were also determined. Salient points considered were point of curvature, point at quarter length from point of curvature, midpoint of curve, point at three-quarter length from point of curvature, and point of tangency. The truck model was calibrated using speed and lateral acceleration values at these salient points. Table 3 gives the summary of the calibration parameters used for calibrating the TruckMaker software. The results obtained during simulation using the calibrated truck model varied by at most 15% from field results. Also, Analysis of Variance (ANOVA) test at 95% significance level showed that there is no significant difference between simulation and field results for the salient points considered. 60% of the field results were used for calibrating the truck model and remaining 40% was used for validating the model. Figure 2 shows Table 1 Characteristics of test vehicle Sl No

Vehicle characteristics

1

Vehicle model

TATA SE 1613

2

Kerb weight

4850 kg

3

Gross vehicle weight

16200 kg

4

Suspension

Semi elliptical leaf springs

5

Overall length

6970 mm

6

Overall width

2434 mm

7

Track width

1890 mm

8

Wheel base

4225 mm

Table 2 Summary statistics of geometry of study curves used for validation of software Geometrical parameter

Minimum

Maximum

Mean

Standard deviation 221.5

Radius of curve (m)

96.6

900

366.6

Superelevation

0.002

0.046

0.025

0.0165

Curve length (m)

19.5

110

70

Deflection angle (deg)

6

40

15.17

9.95

Width of road at curve (m)

5.6

8.95

6.22

0.89

19.5

Rollover Stability Analysis of Trucks-Effect of Curve Geometry …

129

Fig. 1 Road model in IPG truckmaker (transition curve length at beginning = 23 m, transition curve length at end = 20 m, length of horizontal curve = 58 m, deflection angle of curve = 12°, radius of curve = 156 m)

Table 3 Calibration parameters of vehicle model Sl No

Vehicle Model Parameters

1

Vehicle body

Rigid

2

Gross vehicle weight

16,200 kg

3

Overall length

6970 mm

4

Overall width

2434 mm

5

Track width

1890 mm

6

Wheel base

4225 mm

7

Centre of gravity of unladen truck from rear outermost point at ground

X = 4.115 m, Y = 0 m, Z = 0.873 m

8

Centre of gravity of payload

X = 2.21 m, Y = 0 m, Z = 1.082 m

9

Vehicle suspension spring stiffness

K front = 723,276.5 N/m K rear = 615,234.375 N/m

10

Tire size

275/80R22.5 (Twin tires on rear axle)

the operating speed and lateral acceleration profile along a single road model during field survey as well as simulation.

2.2 Sensitivity Analysis To have a clear understanding about the relationship between different geometrical parameters of horizontal curve and rollover stability of a fully loaded truck, a sensitivity analysis was conducted. 384 test alignments of varying curve geometry were modeled for this purpose. Table 4 shows the summary statistics of geometric variables used for the test alignments during sensitivity analysis. The range of geometric

130

Y. K. Remya et al.

Fig. 2 Comparison of speed and lateral acceleration profile along a single study curve-fully loaded truck

variables was selected keeping in mind the practical values usually adopted on highways. The input variables for the simulation were the geometric variables such as radius of horizontal curve, length of curve, deflection angle, superelevation, and width of pavement at curve. These variables were varied one by one by keeping the other variables unchanged. This was done to understand the influence of each geometric variable alone on rollover stability. The speed of the vehicle while entering the test alignment was set as 65 km/h throughout the study. This speed was selected because of the fact that most Indian states have enforced a speed limit of 65 km/h in National and State highways. The output variables collected after simulation were lateral acceleration, operating speed of the vehicle, and steering wheel angle along each test alignment. Steering wheel angle was measured between front of the truck and front wheel position at each point along the test alignment. The influence of each geometric variable was determined by leaving all other variables unchanged. The effect of geometry of curve on steering wheel angle at various locations along the curve was also studied.

3 Results and Discussion 3.1 Effect of Geometry on Rollover Stability of Vehicles Figure 3 shows the variation of maximum lateral acceleration experienced by the truck with radius of horizontal curve for different curve lengths with superelevation values of 0.01 and 0.03. Separate plots were made for superelevation values of 0.05 and 0.07 also. It was observed that as length of curve increased, maximum

Rollover Stability Analysis of Trucks-Effect of Curve Geometry …

131

Table 4 Summary statistics of geometric variables used during sensitivity analysis Geometric variable

Median

Mode

Length of curve 125 (m)

125

50

0.04

0.04

0.01

Radius of curve 310.42 (m)

300

50

Deflection angle

34.94

23.41

57.32

Width of pavement at curve

7.75

7.8

7

Superelevation

Mean

Standard deviation

Skewness

Minimum

Maximum

55.97

9.32E-18

50

200

0.02

−2.1E-15

0.01

0.07

157.31

0.17

50

600

36.54

2.82

4.78

229.30

−3.1E-14

5.5

10

0.03

lateral acceleration experienced by the truck also increased. Superelevation at curve also influenced maximum lateral acceleration values. Maximum lateral acceleration decreased with an increase in radius of curve. But the variation in maximum lateral acceleration with radius of curve is not uniform. Variation in lateral acceleration goes on increasing as radius of curve decreases. Also, it can be seen that lateral acceleration values are irregular for very low radius curves. So it was decided to carry out further analysis for low, medium, and high radius curves separately as shown in Table 5. The study curves were classified into three categories by comparing the actual radius with ruling minimum radius as follows. Ruling design radius and minimum design radius of each test alignment were calculated using point mass equation (Eq. 1) with ruling and minimum design speed of 100 and 80 km/h respectively as specified by IRC 73-1980 for plain terrain. Maximum friction supply of 0.15 and superelevation of 0.07 were used during calculation. e+ f =

v2 gR

(1)

Fig. 3 Variation of maximum lateral acceleration with radius for curve for a e = 0.01, b e = 0.03

132

Y. K. Remya et al.

Table 5 Details of classification of curves into three categories Category

Details

Category 1: low radius Curves with radius less than minimum design radius curves

Sample size 95

Category 2: medium radius curves

Curves with radius between minimum and ruling design radius

111

Category 3: high radius curves

Curves with radius greater than ruling design radius

142

To understand the variation in lateral acceleration along a curve, lateral acceleration at point of curvature (PC), midpoint of curve (MC), and point of tangency (PT) for each test alignment were determined. Tables 6, 7, and 8 show summary statistics of lateral acceleration experienced by truck at PC, MC, PT, and maximum lateral acceleration for three categories of curves. It was observed that lateral acceleration experienced by a fully loaded truck decreased with increase in radius of curve and decrease in length of curve for all categories of curves. Lateral acceleration at the midpoint of curve decreased with increase in super elevation for all categories of curves except for low radius curves at low curve length. That is, for curves having very low radius, superelevation, and length, lateral acceleration at midpoint of the curve was very high. Maximum lateral acceleration experienced by the truck at low radius curves decreased with increase in superelevation especially for low radius curves. At point of tangency, lateral acceleration experienced by the truck was found to remain unaffected by change in length of curve and superelevation except for very low curve lengths. At very low curve lengths, lateral acceleration at point of tangency was found to decrease with increase in superelevation. Lateral acceleration at point of curvature was least affected by changes in length and superelevation of curve. While investigating the influence of length of curve on lateral acceleration, it was observed that lateral acceleration at midpoint of curve, point of tangency, and maximum lateral acceleration decreased with increase in curve length. Lateral acceleration at point of curvature was not affected by change in curve length. As lateral acceleration experienced by the truck and its margin of safety against rollover are inversely related, reducing the lateral acceleration will increase rollover stability of a truck. So, it can be concluded that an increase in radius of curve superelevation and length of curve has significant influence on increasing the rollover stability of a truck. Figure 4 shows the box plot representation of distribution of maximum lateral acceleration for three categories of curves. It was observed that the maximum lateral acceleration experienced by the truck is 0.19, 0.12, and 0.1 g respectively for low, medium, and high radius curves. Minimum lateral acceleration of 0.03 g was experienced for high radius curves compared to 0.07 g for low radius curves.

Mean

0.031

0.137

0.121

0.042

0.031

−0.002

−0.031

0.137

MC

PT

Max

0.024

0.087

PC

L = 200 m

Max

0.046

−0.012

PT

0.024

0.090

0.087

0.049

PC

MC

L = 150 m

0.044

0.020

0.137

PT

Max

0.031

0.057

0.087

0.098

PC

0.024

0.035

MC

L = 100 m

0.127

Max

0.021

0.033

0.124

0.066

MC

PT

0.024

SD

0.088

e = 0.01

PC

L = 50 m

Location

0.052

0.090

−0.091

−0.249

0.065

0.090

−0.077

0.180

0.020

0.086

0.127

0.181

0.034

0.127

0.108

0.065

−0.142

0.180

−0.066

0.090

0.127

0.145

0.065

0.182

0.088

0.178

0.130

Max

−0.027

0.087

0.024

0.086

0.066

Min

0.135

−0.029

0.001

0.091

0.135

−0.009

0.043

0.091

0.134

0.021

0.084

0.091

0.133

0.063

0.127

0.097

e = 0.03

Mean

0.027

0.039

0.111

0.023

0.027

0.042

0.095

0.023

0.028

0.042

0.064

0.023

0.041

0.020

0.039

0.025

SD

0.103

−0.084

−0.217

0.066

0.103

−0.072

−0.147

0.066

0.103

−0.056

−0.048

0.066

0.081

0.022

0.080

0.068

Min

0.178

0.019

0.086

0.131

0.178

0.032

0.109

0.132

0.178

0.054

0.142

0.131

0.192

0.088

0.187

0.141

Max

0.134

−0.025

0.002

0.095

0.133

−0.006

0.038

0.095

0.134

0.021

0.064

0.095

0.137

0.061

0.129

0.105

e = 0.05

Mean

0.023

0.035

0.099

0.022

0.024

0.037

0.096

0.023

0.023

0.040

0.075

0.023

0.053

0.020

0.048

0.032

SD

Table 6 Summary statistics of lateral acceleration values (g) at various points for low radius curves

0.114

−0.075

−0.185

0.067

0.108

−0.066

−0.150

0.067

0.113

−0.049

−0.070

0.067

0.074

0.020

0.073

0.069

Min

0.174

0.019

0.085

0.135

0.174

0.031

0.111

0.135

0.176

0.055

0.129

0.136

0.225

0.085

0.208

0.160

Max

Mean

0.130

−0.022

0.003

0.100

0.130

0.020

0.021

0.101

0.128

0.020

0.046

0.101

0.141

0.057

0.129

0.113

e = 0.07

0.173

0.031

0.085

0.023

0.172

0.036

0.099

0.024

0.173

0.037

0.087

0.024

0.065

0.025

0.058

0.037

SD

0.125

−0.061

−0.139

0.070

0.129

−0.060

−0.158

0.071

0.124

−0.040

−0.064

0.071

0.073

0.004

0.066

0.072

Min

0.020

0.019

0.083

0.139

0.020

0.031

0.101

0.140

0.031

0.055

0.123

0.140

0.255

0.083

0.237

0.182

Max

Rollover Stability Analysis of Trucks-Effect of Curve Geometry … 133

Mean

0.009

0.092

0.009

0.092

Max

0.004

0.004

0.080

0.027

MC

0.007

PT

0.053

PC

L = 200 m

Max

0.001

0.035

PT

0.007

MC

0.007

0.053

0.090

PC

L = 150 m

0.003

0.046

0.091

PT

Max

0.011

0.011

0.053

0.090

MC

0.007

0.010

PC

L = 100 m

0.069

Max

0.006

0.010

0.069

0.049

MC

PT

0.007

SD

0.055

e = 0.01

PC

L = 50 m

Location

0.080

0.021

0.074

0.044

0.081

0.034

0.081

0.044

0.076

0.043

0.076

0.045

0.056

0.042

0.056

0.046

Min

0.106

0.031

0.085

0.065

0.106

0.036

0.101

0.065

0.108

0.050

0.107

0.065

0.085

0.058

0.085

0.066

Max

0.096

0.026

0.082

0.055

0.097

0.033

0.094

0.053

0.093

0.049

0.086

0.053

0.063

0.045

0.062

0.055

e = 0.03

Mean

0.006

0.003

0.003

0.007

0.008

0.001

0.007

0.008

0.010

0.003

0.011

0.007

0.011

0.006

0.011

0.008

SD

0.088

0.021

0.078

0.046

0.086

0.033

0.085

0.043

0.080

0.045

0.071

0.044

0.049

0.038

0.049

0.045

Min

Table 7 Summary statistics of simulation output for medium radius curves

0.103

0.030

0.086

0.065

0.108

0.034

0.103

0.064

0.107

0.052

0.103

0.064

0.079

0.054

0.078

0.066

Max

Mean

0.102

0.026

0.081

0.054

0.100

0.032

0.098

0.054

0.098

0.050

0.081

0.054

0.058

0.040

0.055

0.055

e = 0.05

0.008

0.003

0.003

0.008

0.006

0.001

0.006

0.008

0.009

0.002

0.012

0.007

0.008

0.006

0.010

0.008

SD

0.092

0.020

0.076

0.045

0.093

0.031

0.091

0.044

0.086

0.047

0.065

0.045

0.049

0.032

0.042

0.046

Min

0.113

0.029

0.085

0.066

0.109

0.033

0.107

0.065

0.111

0.053

0.098

0.065

0.072

0.050

0.071

0.068

Max

Mean

0.104

0.026

0.080

0.056

0.105

0.030

0.104

0.056

0.101

0.051

0.073

0.056

0.061

0.033

0.048

0.058

e = 0.07

0.012

0.003

0.003

0.008

0.012

0.001

0.005

0.008

0.008

0.002

0.012

0.008

0.004

0.007

0.010

0.008

SD

0.081

0.020

0.076

0.047

0.081

0.029

0.097

0.046

0.090

0.048

0.058

0.047

0.056

0.025

0.035

0.048

Min

0.119

0.028

0.083

0.068

0.120

0.031

0.111

0.068

0.113

0.054

0.092

0.068

0.067

0.043

0.064

0.070

Max

134 Y. K. Remya et al.

Mean

0.008

0.066

0.008

0.065

Max

0.007

0.001

0.062

0.029

MC

0.005

PT

0.035

PC

L = 200 m

Max

0.003

0.030

PT

0.005

0.008

0.035

0.066

PC

MC

L = 150 m

0.004

0.037

0.060

PT

Max

0.008

0.009

0.035

0.058

PC

0.005

0.008

MC

L = 100 m

0.041

Max

0.005

0.008

0.041

0.033

MC

PT

0.006

SD

0.035

e = 0.01

PC

L = 50 m

Location

0.055

0.027

0.053

0.029

0.055

0.027

0.055

0.028

0.049

0.032

0.047

0.029

0.032

0.026

0.032

0.028

Min

0.078

0.031

0.073

0.043

0.079

0.035

0.079

0.043

0.075

0.042

0.074

0.044

0.054

0.041

0.054

0.045

Max

0.074

0.029

0.066

0.036

0.072

0.029

0.071

0.034

0.065

0.039

0.054

0.034

0.037

0.028

0.034

0.034

e = 0.03

Mean

0.007

0.001

0.006

0.005

0.008

0.003

0.008

0.005

0.008

0.004

0.009

0.005

0.005

0.005

0.008

0.006

SD

0.065

0.028

0.058

0.030

0.062

0.025

0.061

0.028

0.055

0.034

0.042

0.028

0.031

0.021

0.024

0.027

Min

Table 8 Summary statistics of simulation output for high radius curves

0.085

0.030

0.076

0.044

0.085

0.033

0.084

0.043

0.079

0.044

0.069

0.043

0.048

0.037

0.047

0.044

Max

0.078

0.028

0.065

0.035

0.078

0.026

0.077

0.034

0.072

0.041

0.047

0.035

0.042

0.022

0.027

0.035

e = 0.05

Mean

0.008

0.001

0.006

0.005

0.007

0.003

0.007

0.005

0.008

0.004

0.009

0.005

0.004

0.006

0.008

0.006

SD

0.068

0.025

0.056

0.028

0.069

0.022

0.068

0.028

0.062

0.037

0.036

0.028

0.037

0.015

0.017

0.028

Min

0.090

0.029

0.075

0.044

0.092

0.031

0.089

0.043

0.085

0.046

0.064

0.044

0.048

0.031

0.040

0.045

Max

SD

0.086

0.032

0.065

0.036

0.087

0.024

0.086

0.036

0.081

0.042

0.040

0.036

0.049

0.014

0.020

0.036

0.007

0.015

0.006

0.005

0.007

0.003

0.007

0.006

0.007

0.004

0.009

0.006

0.004

0.006

0.008

0.006

e = 0.07

Mean

0.077

0.025

0.057

0.029

0.078

0.020

0.077

0.029

0.072

0.037

0.028

0.029

0.044

0.007

0.010

0.029

Min

0.098

0.072

0.075

0.044

0.099

0.029

0.096

0.045

0.093

0.047

0.056

0.046

0.055

0.024

0.033

0.046

Max

Rollover Stability Analysis of Trucks-Effect of Curve Geometry … 135

136

Y. K. Remya et al.

Fig. 4 Box plot representation of distribution of maximum lateral acceleration for three categories of curves

3.2 Relationship Between Radius of Curve and Steering Wheel Angle An attempt was made to determine the influence of road geometry on steering wheel angle at PC, MC, PT and at location of maximum lateral acceleration, with focus on radius of horizontal curve. Figure 5 shows the variation in steering wheel angle with radius of curve at different points along the curve. It was observed that steering wheel angle decreased as radius of curve increased at all points along the curve. Also, steering wheel angle is highest at the point of maximum lateral acceleration followed by midpoint of curve. Steering wheel angle at PT is lower than that at PC. Figure 6 shows box plot representation of distribution of steering wheel angle at the location of maximum lateral acceleration for three categories of curves. It was observed that steering wheel angle was very high for low radius curves followed by medium radius curves. Steering wheel angle was very low for high radius curves.

Fig. 5 Variation in steering wheel angle with radius of curve at PC, MC, PT, and point of maximum lateral acceleration

Rollover Stability Analysis of Trucks-Effect of Curve Geometry …

137

Fig. 6 Box plot representation of distribution of maximum steering wheel angle for three categories of curves

3.3 Lateral Acceleration Prediction Models In order to improve the road design, it is crucial to evaluate and define the relationship between road geometric design elements and rollover stability. So an attempt was made to develop regression models to predict lateral acceleration at PC, MC, PT, and maximum lateral acceleration for low radius, medium radius, and high radius curves separately. Various independent variables tried were radius of curve (R), Length of curve (L), superelevation (e), operating speed of truck (V), width of curve (W). 95% confidence interval was selected for the regression analysis. Multicollinearity among independent variables was checked by determining the coefficient of correlation between each of the two independent variable and hence eliminated. Models for lateral acceleration at PC The model form found to be appropriate for predicting lateral acceleration at PC was a y PC = a0 + a1 R + a2 v

(2)

where a y PC denoted lateral acceleration experienced by the truck at point of curvature in m/s2 , R denoted radius of curve in m, v denoted operating speed of truck in m/s and the model parameters a0 , a1, and a2 were estimated for each category of curve and for all types of curves together; they are given in Table 9. The model was developed using 60% of data and the remaining 40% data was used for validating it. It was observed that, for all categories of curves, lateral acceleration experienced by fully loaded truck at PC could be predicted using radius of curve and operating speed of truck as independent variables. Lateral acceleration increases as radius of curve decreases, and operating speed increases.

138

Y. K. Remya et al.

Table 9 Models for lateral acceleration at PC Curve type

Model

R2

DF CD

RMSE VD

CD

VD

Low radius curves

−0.08-R/1790+V/67

0.96

52

37

0.005

0.004

Medium radius curves

−0.65-R/5814+v/23

0.96

68

37

0.001

0.001

High radius curves

−0.24-R/15465+v/58

0.95

99

37

0.001

0.001

All categories of curves

−0.45-R/560+v/11

0.78

337

37

0.019

0.016

Note DF = Degrees of freedom, CD = Calibration data set, VD = Validation data set

Models for lateral acceleration at MC, PT, and at point of maximum lateral acceleration The model form found to be appropriate for predicting lateral acceleration at MC, PT, and at point of maximum lateral acceleration was a y = a0 + a1 L + a2 e + a3 R + a4 v

(3)

where ay denoted lateral acceleration in m/s2 experienced by the truck in m/s2 . R denoted radius of curve (m), v is operating speed of truck (m/s), L is curve length (m), e is superelevation and the model parameters a0 , a1 , a2 , a3, and a4 were estimated for each category of curve and for all types of curves together. Tables 10, 11, and 12 give models for lateral acceleration at MC, PT, and at point of maximum lateral acceleration for three categories of curves separately and for all categories of curves together. All models were developed using 60% of data and validated using remaining 40% data. It was observed that lateral acceleration at MC, PT, and at location of maximum lateral acceleration is dependent on radius of curve, length of curve, superelevation, and operating speed of truck. For all categories of curves, radius of curve, length of curve, and superelevation were negatively related to lateral acceleration. It was found Table 10 Models for lateral acceleration at MC Curve type

Model

R2

DF CD

RMSE VD

CD

VD

Low radius curves

−1.16-L/7576–0.43e-R/ 1504+0.08v

0.84

50

35

0.017

0.076

Medium radius curves

1.91+L/1912-e/78-R/ 4547–0.1v

0.92

71

35

0.005

0.034

High radius curves

2.76+L/1406–0.17e-R/ 10101–0.15v

0.96

97

35

0.004

0.028

All categories of curves

−4-L/724–0.66e-R/ 557+0.3v

0.7

335

35

0.334

0.197

Note DF = Degrees of freedom, CD = Calibration data set, VD = Validation data set

Rollover Stability Analysis of Trucks-Effect of Curve Geometry …

139

Table 11 Models for lateral acceleration at PT Curve type

Model

R2

DF CD

RMSE VD

CD

VD

Low radius curves

−0.16-L/1822+0.03e-R/ 15535+0.01v

0.8

50

35

0.013

0.021

Medium radius curves

2+L/2294–0.19e-R/ 36101–0.11v

0.6

66

35

0.006

0.012

High radius curves

3.34+L/952–0.41e-R/ 34602–0.19v

0.7

97

35

0.006

0.016

All categories of curves

−1.63-L/356+0.07e-R/ 2596+0.13v

0.7

340

40

0.17

0.19

Note DF = Degrees of freedom, CD = Calibration data set, VD = Validation data set

Table 12 Models for maximum lateral acceleration on horizontal curve Curve type

Model

R2

DF CD

RMSE VD

CD

VD

Low radius curves

−0.05+L/5102–0.21e-R/ 1340+0.02v

0.92

50

35

0.01

0.017

Medium radius curves

−0.91+L/4673+0.13e-R/ 5917+0.06v

0.8

66

35

0.009

0.015

High radius curves

−0.73+L/5848+0.36e-R/12578 + 0.04v

0.94

97

35

0.004

0.01

All categories of curves

0.4+L/711+1.32e-R/481+0.05v

0.7

335

35

0.21

0.16

Note DF = Degrees of freedom, CD = Calibration data set, VD = Validation data set

that lateral acceleration for low radius curves decreases with increase in superelevation especially at MC. Thus for sharp curves, adequate superelevation needs to be designed to ensure lateral stability against rollover.

4 Conclusions The following conclusions were drawn from the study: • An increase in radius of horizontal curve will increase the rollover stability of a truck. Wherever space restrictions limit the maximum radius provided at a curve, providing higher length of horizontal curve and superelevation can increase the rollover stability. It was found that lateral acceleration for low radius curves decreases with increase in superelevation especially at MC. Thus for sharp curves, adequate superelevation needs to be designed to ensure lateral stability against rollover.

140

Y. K. Remya et al.

• Increase in rollover stability of truck with increase in radius of curvature is not uniform. The Analysis of Variance test conducted showed a significant difference in the lateral acceleration values for three categories of curves. So, models which predict maximum lateral acceleration and lateral acceleration at PC, MC, and PT were developed for low, medium, and high radius curves separately on two-lane rural highways. • Lateral acceleration predicted using the developed models for a particular curve can be used to determine margin of safety against rollover of the truck by comparing with rollover threshold. This will help in finding rollover potential of that truck at a curve. For example, if rollover threshold of a truck is 0.7 g and maximum lateral acceleration experienced by that truck at a curve is 0.4 g, the truck can experience an additional lateral acceleration of 0.3 g only before rollover. These results can hence be used to assign speed limit or install cautionary sign boards on such hazardous curves. • Rollover stability of a truck at any point along a curve is related to steering wheel angle of the truck at that point. Curve geometry, especially radius of curve has a direct influence on steering wheel angle. More studies have to be carried out to evaluate rollover stability of vehicles based on steering wheel angle. • This study provides insight into the importance of highway geometry in deciding the rollover stability of vehicles, especially trucks on rural horizontal curves. The outputs of this study can be used by transportation engineers to design safe curves. Also, these results can be used as road safety audit tools to identify potentially dangerous curves in terms of rollover.

References American Association of State Highway and Transportation Officials (AASHTO) (2010) Highway safety manual, 1st edn. Washington, DC Abhilash N, Ravikumar T (2015) Highway accident modeling—influence of geometrics. Int J Res (IJR) 2(Issue 10). e-ISSN: 2348–6848, p-ISSN: 2348–795X Ahmed I et al (2013) Road infrastructure and road safety. Transp Commun Bull Asia Pacific 83 Alrejjal A et al (2022) Investigating factors influencing rollover crash risk on mountainous interstates. J Saf Res 80:391–398 Alrejjal A, Ksaibati K (2022) Impact of mountainous interstate alignments and truck configurations on rollover propensity. J Saf Res 80:160–174 Geedipally SR, Pratt MP, Lord D (2017) Effects of geometry and pavement friction on horizontal curve crash frequency. J Transp Saf Secur. https://doi.org/10.1080/19439962.2017.1365317 Hosseinpour M et al (2016) Evaluating the effects of road geometry, environment, and traffic volume on rollover crashes. Special issue on the impact of vehicle movement on exploitation parameters of roads and runways, Transport Taylor and Francis 31(2):221–232 Islam MH et al (2018) Relationship of accident rates and road geometric design. In: IOP Conference series, earth and environmental science, vol 357 Mohammadi MA, Samarayanake VA (2014) Crash frequency modeling using negative binomial models: an application of generalized estimating equation to longitudinal data. Anal Methods Accid Res 2:52–69

Rollover Stability Analysis of Trucks-Effect of Curve Geometry …

141

Mohammed HA (2019) The influence of road geometric design elements on highway safety. Int J Civ Eng Technol 4(4):146–162 Moreno G et al (2018) Stability models of heavy vehicle. Contemp Eng Sci 11 National Highway Safety Transportation Administration (NHTSA) (2006) http://www.nhtsa.dot. gov/cars/testing/ncap/Rollover/Index.htm.Safercar.gov-Rollover Yin Y (2020) The influence of road geometry on vehicle rollover and skidding. Int J Environ Res Public Health Zhang Y et al (2009) Analysis of the relation between highway horizontal curve and traffic safety. In: International conference on measuring technology and mechatronics automation Zhao Z et al (2014) Analysis of heavy vehicle rollover and stability. Comput Model New Technol 18(12D):157–161 Zou Y et al (2012) Application of finite mixture of negative binomial regression models with varying weight parameters for vehicle crash data analysis. Accid Anal Prev 50

Analysing Willingness to Pay and Attitude Towards Safety for Indian Motorcyclists Anand Kumar Saurav, Hillol Chakravarty, and Ranja Bandyopadhyaya

Abstract Vulnerable Road Users (VRUs), like Motorised two-wheelers (MTWs), are the most affected class of road users. Road safety project evaluation requires analysis of social cost of road crashes. Researchers attempted to understand Willingness to Pay (WTP) for different classes of road users. In this study, WTP for reduction in the chance of fatality is estimated and influence of individual’s perception of safety, their socio-demographic characteristics and whether they can make independent financial decisions on WTP is analysed. WTP was observed to vary with the amount of safety improvement the expense will offer. Payment card method was adopted for estimating WTP for various levels of risk reduction and to understand the variation of WTP with variation in the chance of fatality reduction for Indian MTWs. The WTP-CV Payment Card Questionnaire was designed. The questionnaire includes socio-demographic data, risk exposure, financial independence rated on a scale of 1 to 7 and lastly, the valuation questions for various levels of fatality risk reduction. A total of 541 data were collected through face-to-face interviews, online and telephonic interviews. It could be observed that the MTWs with higher degrees of education, higher monthly incomes and more crash history are prepared to pay more. The WTP and probability of risk reduction could be observed as not a linear function. Also, an individual’s attitude towards safety influences the WTP greatly. Keywords WTP-CV · Payment card · MTWs · Fatality reduction · Risk exposure · SPSS Statistics

A. K. Saurav · H. Chakravarty · R. Bandyopadhyaya (B) National Institute of Technology Patna, Patna, Bihar 800005, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_11

143

144

A. K. Saurav et al.

1 Introduction According to the World Health Organization (WHO), as more people select highspeed roads, almost half of all road collision victims are becoming Vulnerable Road Users (VRUs). In 2013, the World Health Organization (WHO 2015) defined VRUs as pedestrians, cyclists and motorised two-wheelers (MTWs). Here riders of a motorcycle can be referred to as both motorcyclists and Motorised two-wheelers (MTWs) riders. The World Health Organization (WHO 2015) said that 50% of road accident victims were Vulnerable Road Users (VRUs). Amongst VRUs, MTWs riders’ deaths account for half of all fatalities recorded for VRUs. In emerging countries like India, MTWs account for approximately 37.1% of total road accident fatalities among vehicle categories in 2019, as shown in Fig. 1 (MORTH 2020). The two-wheeler accounts for the most prevalent vehicle type in street mishaps and accounts for the majority of registered vehicles in the country. The majority of fatalities of MTWs riders are a result of head injuries, which are the most serious public health issue in developing countries like India. The major cause of head injury is the refusal to wear a helmet or the use of an inappropriate type of helmet (Cini et al. 2014; Erhardt et al. 2016; Yu et al. 2011). It is seen that thousands of people are killed or wounded on our roadways every day. Millions of individuals will spend weeks in hospitals each year as a result of serious wrecks, and many will never be able to live, work, or play as they once did. Thus, Road vehicular accidents have a significant socioeconomic impact on the victims, their families and the country (Haddak et al. 2016; Mon et al., 2018). It is also possible to conclude that India’s economy is suffering as a result of MTW accidents. Willingness to pay (WTP) is the maximum amount of money that a customer is willing to pay for a product or service (Haddak et al. 2016). The contingent valuation (CV) method is a hypothetical market situation aimed at obtaining a respondent’s preferences by conducting a questionnaire survey and asking respondents about their willingness to pay for a given reduction in their security risk level (Mofadal et al. 2015). In the payment card, respondents can pick what they are willing to pay from a list of values on the record, and there is also a space to write an alternative price if the options are not adequate, making the valuation process easier (Mofadal et al. 2015). Buses , 4.3% Trucks/Lorries , 9% Cars, Taxis, Vans & LMVs , 15.8% Auto-Rickshaws , 4.4%

Others, 9.5%

Pedestrian, 17.1% Bicycles, 2.8%

Two-wheelers, 37.1%

Fig. 1 Road traffic fatalities by road-user category in India (MORTH, 2019)

Analysing Willingness to Pay and Attitude Towards Safety for Indian …

145

Many researchers have attempted to quantify WTP by lowering road mortality for a variety of road users, including bicycles, pedestrians, bus passengers and cars (Bala- rishnan & Karuppanagounder 2020; Mofadal et al. 2015; Mon et al. 2018). In developing countries, several WTP-CV studies have been conducted to assess accident costs for various groups of road users, such as pedestrians and MTWs (Bhattacharya et al. 2007; Chaturabong et al. 2011; Jou and Chen 2015). Mon et al. (2018) estimated WTP at 50 per cent of fatality risk reduction, and Mofaal et al. (2015) estimated WTP at five different probabilities of pedestrian fatality risk reduction by creating five different scenarios for demonstrating valuation questions. Both respondents and researchers find the direct calculation of WTP utilising the contingent valuation approach with a payment card to be simple to grasp and analyse, as studied by Bhattacharya et al. (2007); Mofadal et al. (2015); and Mon et al. (2018). This paper aims to understand the variation of WTP with variation in the chance of fatality reduction for Indian MTWs using the direct contingent valuation approach with the payment card method. And to understand the relationship of WTP with sociodemographic information and crash experience. The study also aims to understand the influence of an individual’s financial decision-making independence with their WTP.

1.1 Motivation A preliminary survey was conducted in Patna city. The data was collected through police records by visiting the police stations of three different areas of Patna city. Both being a victim or accused of an accident on MTWs can result in loss of health and wealth; thus, both are counted as the overall number of MTWs accidents and compared to the total number of accidents with all types of vehicles in that year, as shown in Fig. The ratio of MTWs being the victim or accused to the total number of vehicle accidents in that year is plotted. It can be noticed from the graph that the trendline has a positive slope from 2015 to 2019, as shown in Fig. 2. This served as the motivation for studying the WTP for helmet usage for MTWs to subsequently reduce road accidents.

2 Methodology The WTP-CV Questionnaire was used in this study with the payment card technique. The data for various levels of fatality risk reduction were collected through both online and offline modes to get a better and quick survey. The respondent’s Motorized two-wheeler (MTWs) who have finished at least high school were chosen because they were thought to understand the circumstances better. The questionnaire survey

A. K. Saurav et al. Percentage of MTWs involved accident

146 50

40.73

40 30

37.92

43.46

28.42

23.6

20 10 0

2015

2016

2017 Year

2018

2019

Fig. 2 Graph shows the increase in the percentage of MTWs involved accident in the total number of road traffic accidents from the year 2015 to 2019

was conducted in person and online mode. The scenario of WTP with various levels of fatality risk reductions was discussed, along with several helmet safety accessories that may be added to provide varying levels of helmet protection. The collected data was examined using regression analysis with the help of software (SPSS Statistics). The WTP values are taken as the dependent variable. The regression analysis outcomes were examined to determine the relationship between WTP and socio-demographic information, risk exposure and percentage fatality risk reduction. A person’s financial independence was also compared with WTP. The flowchart of methodology is shown in Fig. 3.

Fig. 3 Flowchart of the methodology adopted

Analysing Willingness to Pay and Attitude Towards Safety for Indian …

147

2.1 Questionnaire Design Most of the respondents had their mother tongue as Hindi language, so the questionnaire was designed both in English and Hindi so that the respondents quickly understood the questions. The questionnaire is divided into three segments. The first segment contained socio-demographic information such as the number of dependents in the family, marital status, education, profession, personal and monthly household income, respondent age and other factors that were found to influence WTP decision-making (Bhattacharya et al. 2007; Haddak et al. 2016; Mofadal et al. 2015). The second segment includes questions about the number of crashes encountered by respondents in the past two years, the severity of the incidents and if respondents are financially free to make the decision, rated on a scale of 1 to 7. The third and last segment is essential, which includes valuation questions presented in five different percentages of fatality risk reduction. The WTP values of all scenarios with varying percentages of MTWs riders’ fatality risk reduction owing to head injury were compared using the valuation questions (Mofadal et al. 2015). The valuation questions, which are based on five scenarios, are consistent with past research (Bhattacharya et al. 2007; Mofadal et al. 2015); they have been shown in Table 2. Helmet Safety Scenarios Erhardt et al. (2016) stated that the type of helmet used was directly analogous to the probability of head injuries. The probability of surgery was two times higher in individuals who wore an open-face helmet than in those who wore a full-face helmet (Cini et al. 2014; Yu et al. 2011). Yu et al. (2011) stated that MTWs with poorly fastened helmets were at two times higher risk of head injury compared to those with adequately fastened helmets. We prepared a table of varying helmet types with various safety features and drawbacks and was printed in colour format for better visuality, as shown in Table 1. The table was presented before the MTWs user so that they could become acquainted with the scenario. The five distinct scenarios for the chance of fatality risk reduction for MTW riders were generated, using examples of different types of helmet with differing safety features, as shown in Table 2. Two alternative helmets were described to each rider of MTWs (i.e. Helmet A and Helmet B). The helmet they are now wearing is marked as Helmet A. Helmet B was stated to be safer than Helmet A and capable of lowering the probability of fatality risk at different percentages that they should consider purchasing. It was considered that both helmets are owned by the same company and have the same warranty period for manufacturing flaws. The cost of helmets varies according to quality standards and the safety features added to the helmets during their manufacturing. As per traffic police records from Patna, India, 15 fatalities per 100,000 people were estimated as the probability of a head injury in 2019. Each MTW rider was given a Payment Card with various alternatives and a blank space for the amount they were willing to pay, as shown in Fig. 4.

148

A. K. Saurav et al.

Table 1 Types of helmets with their safety features and drawbacks Types of helmet

Image

Description

Drawback

Half helmet

Cover the top of your head and Do not come provide minimal protection equipped with a visor or face shield

Open-Face(3/4) helmet

Equipped with partial or full visors to protect the eyes and face from sunlight and covers the head top, back and sides

Lack of a chin bar and does not protect you against weather conditions and road debris

Modular (flip-up) helmet

The chin bar and visor can flip up to open the front of the helmet, including a visor for eye protection and a secondary interior visor for added sun protection

Weighs slightly more than the full-face helmet due to the additional design hinge feature

Full face helmet

Designed with a chin bar and does not lift at high speed

Dual sport helmet

Feature a larger visor for eye protection than a full face and have improved soundproofing

Smart helmet

Uses artificial intelligence to function

Payment Card For More Safer Helmet: In Rs 5 75

10 100

15 125

20

200 600

300

350

400

500

1000

1500

2000

3000

Any additional amount (Not included above). Rs………………………………………………….

40

150

> 3000

Analysing Willingness to Pay and Attitude Towards Safety for Indian …

149

Table 2 Five situations involving motorised two-wheelers (MTWs) safety were evaluated using a set of questions: Scenario number

Helmet A

Helmet B

Scenario 1: Probability of fatality

15/100000 per year

12/100000 per year

Question: How much extra money are you willing to spend for the helmet to reduce your chances of fatality from a head injury in an MTWs crash by 20%? Scenario 2: Probability of fatality

15/100000 per year

9/100000 per year

Question: How much extra money are you willing to spend for the helmet to reduce your chances of fatality from a head injury in an MTWs crash by 40%? Scenario 3: Probability of fatality

15/100000 per year

7.5/100000 per year

Question: How much extra money are you willing to spend for the helmet to reduce your chances of fatality from a head injury in an MTWs crash by 50%? Scenario 4: Probability of fatality

15/100000 per year

6/100000 per year

Question: How much extra money are you willing to spend for the helmet to reduce your chances of fatality from a head injury in an MTWs crash by 60%? Scenario 5: Probability of fatality

15/100000 per year

3/100000 per year

Question: How much extra money are you willing to spend for the helmet to reduce your chances of fatality from a head injury in an MTWs crash by 80%?

Payment Card For More Safer Helmet: In Rs 5 10 15 75 100 125 300 350 400 1000 1500 2000 Any additional amount (Not included above).

20 40 150 200 500 600 3000 > 3000 Rs...........................................................

Fig. 4 Payment card with several WTP values and a space for any additional amount

2.2 Data Collection MTWs who have finished high school and are not below 18 years of age are chosen for the survey as they are assumed to be licenced two-wheeler riders and have a better understanding of the scenario. The survey was intended for daily commuters who use MTWs to get to work, college, hospitals and other destinations. In this study, 541 responses are collected for analysis after excluding irrelevant forms. Incomplete forms or mismatched options were all considered irrelevant. The offline survey was conducted through a face-to-face interview. The respondents were given a hard copy of the questionnaire, and questions were demonstrated

150

A. K. Saurav et al.

Fig. 5 Photographs were taken during the offline questionnaire survey

to them for better understanding. The survey took about a quarter of an hour on average per person. Figure 5 depicts the scenarios that occurred during the face-toface interview. The Google form was created to make collecting online survey data as simple as possible. The Google form was forwarded to the MTWs riders via email and social media platforms. Some of them are also contacted via phone directly. The details of the respondents according to their characteristics and frequency are arranged and are shown in Table 3.

3 Result The regression analysis was done through SPSS Statistics, taking log (WTP) as a dependent variable and gender, age, marital status, household income, education, personal income, household size, crash experience and percentage risk reduction as the independent variables. The result obtained is shown in Table 4. The Sig. values generated below 0.05 were considered significant and had a considerable impact on WTP values. It was also observed that the coefficient for gender, age and marital status is negative, while the coefficient of education, income, household size, crash experience and risk reduction is positive. The responder’s financial independencies coefficient is also obtained as negative, and its significance value is above 0.05. A graph was drawn between average WTP as the dependent variable and percentage fatality risk reduction as the independent variable and is shown in Fig. 6. The equation of the non-linear graph derived from the average WTP value and the percentage fatality risk reduction is shown in Eq. 1. y = 54.51x2 − 88.033x + 367.43

(1)

Analysing Willingness to Pay and Attitude Towards Safety for Indian …

151

Table 3 Details of categorised respondent’s socio-demographic information Respondent characteristics

Categories

Code

Frequency

Gender

Male

1

484

Female

0

57

Below 18 yrs

1

0

18yrs to 45 yrs

2

369

46yrs to 60yrs

3

127

Above 60yrs

4

45

Matric

1

43

Intermediate

2

63

Graduate

3

313

Post-graduate

4

118

Others

5

4

Married

1

307

Unmarried

0

234

Below rs 20,000

1

129

20,000 to 60,000

2

287

60,000 to 1 lakh

3

93

Above 1 lakh

4

27

Below 3

1

8

3 to 5

2

346

Above 5

3

187

Age

Education

Martial status Income

Household size

Table 4 Result of regression analysis using SPSS Model

1

(Constant)

Unstandardised coefficients

Standardised coefficients

B

Beta

Std. Error

t

Sig

2.155

0.218

9.906

0.000

Gender

−0.265

0.067

−0.157

−3.928

0.000

Age

−0.022

0.037

−0.027

−0.594

0.553

0.057

0.028

0.091

2.066

0.039

−0.083

0.051

−0.080

−1.622

0.105

Education Martial Status Income

0.074

0.036

0.120

2.081

0.038

Household Size

0.076

0.043

0.074

1.791

0.074

Crash Experience

0.027

0.012

0.101

2.245

0.025

−0.006

0.017

−0.016

−0.358

0.721

1.116

0.097

0.428

11.530

0.000

Financial Independent Risk Reduction

152

A. K. Saurav et al.

1500 1000 500 0

20

40

50

60

80

Fig. 6 Graph of average WTP v/s Percentage fatality risk reduction

It can be inferred that female riders of MTWs are willing to pay more than male riders of MTWs to reduce the risk of road fatality. The similar trend was seen in India, Sudan and Sweden (Bharti et al. 2022; Mofadal et al. 2015; Svensson and Vredin Johansson 2010). The negative coefficient for the age indicates that older MTWs riders are less willing to spend more money on their safety in traffic accidents with respect to younger MTWs riders. A similar result was seen in Thailand (Chaturabong et al. 2011). MTWs riders who are married are willing to pay more for their safety than unmarried individuals. A similar trend was seen in Sudan (Mofadal et al. 2015), but the study in Myanmar contradicts the result (Mon et al., 2018). Riders on MTWs with a higher degree of qualification are willing to pay more. A similar result was seen in studies in Taiwan and Sudan (Jou and Chen 2015; Mofadal et al. 2015). Riders on MTWs who earn more per month are willing to pay more, supporting the economic theory. In Thailand and China, studies found a similar result (Chaturabong et al. 2011; Yang et al. 2016). Riders on MTWs with a higher number of household members are more likely to spend more money on traffic accident prevention than riders on MTWs with smaller household members. A similar trend was seen in Myanmar (Mon et al. 2019). Riders on the MTWs who have had more past crashes are more inclined to spend money on traffic accident prevention as they may become more concerned about the potential implications. The negative result for the respondent’s financial decision suggests that MTW riders would choose to ride four-wheelers rather than invest more in helmet safety. Additionally, the less significant value indicates that it has no effect on WTP values. Further, it was seen that the WTP has a non-linear variation with percentage fatality risk reduction, as shown in Fig. 6. It contradicts the study done in Rhône Département, France, where it was seen that WTP was unaffected by the degree of risk reduction (Haddak et al. 2016), and the study in Sweden got a satisfactory result for the relation of WTP with percentage fatality risk reduction (Andersson 2007).

Analysing Willingness to Pay and Attitude Towards Safety for Indian …

153

4 Conclusion The objective of this study was to compare the WTP with socio-demographic information, previous crash experience and risk reduction. Data were collected both physical and online through the use of a questionnaire, which included 541 respondents. The WTP-CV payment card method was used. In this study, several helmet safety features such as visor, chin bar, weight, protection region, GPS and artificial intelligence were used as distinct scenarios in the questionnaire. These scenarios were shown to be quite efficient in familiarising responders with the circumstance of how spending more money on a helmet might lessen their chances of someone being killed. The contingent valuation approach using a payment card was found to be the most effective way of estimating WTP values at different levels of mortality risk reduction. In this study, education, wealth, crash experience and risk reduction were shown to be positive and significant factors of MTW riders’ WTP, although gender was found to be negative and significant. Individual financial independence has no effect on WTP values. The study’s findings will aid developing-country safety agencies in determining the cost of a helmet with additional safety features.

References Andersson H (2007) Willingness to pay for road safety and estimates of the risk of death: Evidence from a Swedish contingent valuation study. Accid Anal Prev 39(4):853–865. https://doi.org/10. 1016/j.aap.2006.12.008 Balakrishnan S, Karuppanagounder K (2020) Estimating the cost of two-wheeler road accident injuries in India using the willingness to pay method. Aust J Civ Eng 18(1):65–72. https://doi. org/10.1080/14488353.2020.1721951 Bharti S, Bandyopadhyaya R, Raju NK (2022) Estimation of willingness to pay and value of statistical life for road crash fatality reduction for motorcyclists: a case study of Patna, India. J Inst Eng (India): Ser A 103(4):1315–1323 Bhattacharya S, Alberini A, Cropper ML (2007) The value of mortality risk reductions in Delhi. India. J Risk Uncertain 34(1):21–47. https://doi.org/10.1007/s11166-006-9002-5 Chaturabong P, Kanitpong K, Jiwattanakulpaisarn P (2011) Analysis of costs of motorcycle accidents in thailand by willingness-to-pay method. Transp Res Rec 2239:56–63. https://doi.org/ 10.3141/2239-07 Cini MA, Prado BG, Hinnig PDF, Fukushima WY, Adami F (2014) Influence of type of helmet on facial trauma in motorcycle accidents. Br J Oral Maxillofac Surg 52(9):789–792. https://doi. org/10.1016/j.bjoms.2014.05.006 Erhardt T, Rice T, Troszak L, Zhu M (2016) Motorcycle helmet type and the risk of head injury and neck injury during motorcycle collisions in California. Accid Anal Prev 86:23–28. https://doi. org/10.1016/j.aap.2015.10.004 Haddak MM, Lefèvre M, Havet N (2016) Willingness-to-pay for road safety improvement. Transp Res Part a: Policy Pract 87:1–10. https://doi.org/10.1016/j.tra.2016.01.010

154

A. K. Saurav et al.

Jou RC, Chen TY (2015) The willingness to pay of parties to traffic accidents for loss of productivity and consolation compensation. Accid Anal Prev 85:1–12. https://doi.org/10.1016/j.aap.2015. 08.021 Mofadal AIA, Kanitpong K, Jiwattanakulpaisarn P (2015) Analysis of pedestrian accident costs in Sudan using the willingness-to-pay method. Accid Anal Prev 78:201–211. https://doi.org/10. 1016/j.aap.2015.02.022 Mon EE, Jomnonkwao S, Khampirat B, Satiennam T, Ratanavaraha V (2019) Estimating the willingness to pay and the value of fatality risk reduction for car drivers in Myanmar. Case Stud Transp Policy 7(2):301–309. https://doi.org/10.1016/j.cstp.2019.02.010 Mon EE, Jomnonkwao S, Khampirat B, Satiennam W, Ratanavaraha V (2018) Willingness to pay for mortality risk reduction for traffic accidents in Myanmar. Accid Anal Prev 118(April):18–28. https://doi.org/10.1016/j.aap.2018.05.018 Svensson M, Vredin Johansson M (2010) Willingness to pay for private and public road safety in stated preference studies: Why the difference? Accid Anal Prev 42(4):1205–1212. https://doi. org/10.1016/j.aap.2010.01.012 Transport research wing, ministry of road transport and highways, and government of India. (2020). Road Accidents in India 2019. https://morth.nic.in/sites/default/files/RA_Uploading.pdf Yang Z, Liu P, Xu X (2016) Estimation of social value of statistical life using willingness-to-pay method in Nanjing, China. Accid Anal Prev 95:308–316. https://doi.org/10.1016/j.aap.2016. 04.026 Yu WY, Chen CY, Chiu WT, Lin MR (2011) Effectiveness of different types of motorcycle helmets and effects of their improper use on head injuries. Int J Epidemiol 40(3):794–803. https://doi. org/10.1093/ije/dyr040 World Health Organization (WHO) (2015) Global status report on road safety. Injury Prevention. http://www.who.int/violence_injury_prevention/road_safety_status/2013/en/index.html

Use of Advanced Techniques for Functional Evaluation of Pavements: A Review and a Pilot Study N. H. Riyaz Khan

and S. Vasantha Kumar

Abstract In India, 37 kms of national highways are being newly constructed every day and in the last seven years itself, the length of national highways have almost doubled in the country. The road pavement condition plays a crucial role on safety, comfort, traffic, travel times, vehicle operating cost and emission levels. In the present study, an extensive literature survey was carried out on the functional evaluation of pavements which usually involves the three components, namely, the distresses such as cracks, potholes, surface roughness, and skid resistance. It was found that in developed countries like USA, Europe, and China, the visual condition survey is gradually being replaced by use of advanced technologies like Light detection and ranging (LIDAR), where the accurate three-dimensional (3D) model of the road surface was created and using which, mapping of common surface distresses was done, which will eliminate any subjectivity of the evaluator. Though smartphones were used for functional evaluation of pavements in India, however studies using LIDAR could not be found. This necessitates the need for such studies in India, where the manual evaluation is a labor-intensive and time-consuming task considering the length of roads in the country. In addition to the review of literature, a pilot study was also carried out in Vellore, India using a 3D laser scanner in order to explore its usefulness in pavement distress evaluation. A point cloud of 8.23 million data points with X, Y, and Z values was collected and using which the pothole characteristics were calculated. It was found that the LIDAR derived point cloud data has very high potential in mapping the pothole and other distresses as it can record the minute height differences of even 1 mm. Keywords Pavement functional evaluation · LIDAR scanner · Distress study

N. H. Riyaz Khan (B) · S. Vasantha Kumar School of Civil Engineering, Vellore Institute of Technology (VIT), Vellore 632014, Tamilnadu, India e-mail: [email protected] S. Vasantha Kumar e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_12

155

156

N. H. Riyaz Khan and S. Vasantha Kumar

1 Introduction Indian economy heavily relies on the road network, which serves as the arteries among various regions. The development of road infrastructure would create a multiplier effect on the nation’s socio-economic growth by improving accessibility between urban and rural areas. Efficient and safe operation of transportation facilities depends upon the effective road maintenance. Road maintenance is an important factor which needs more attention because the requirement for it grows as the road system matures. Aside from that, road maintenance is vital for road safety. Road management authorities are in a difficult position to make road repair and reconstruction decisions because of declining road maintenance budget and increased environmental challenges. To overcome this difficulty, the concept of Pavement Management System (PMS), a systematic technique for organizing, planning, and allocating funds as well as scheduling all pavement repair work to assist road agencies in making decisions was introduced (Khahro et al. 2021). In PMS, the performance of a road pavement is dictated by the functional and structural conditions. The functional conditions deal with the roughness, skid resistance, and distresses, whereas the structural conditions deal with the pavement soundness in serving the load and the traffic flow (Rusmanto and Handayani 1977). Until 2000, most of the developing countries used manual surveys for collecting data regarding pavement conditions. It was time-consuming and could not be able to quantify the real pavement serviceability (Justo-Silva et al. 2021). But, in the last decade, automated and semi-automated techniques for functional evaluations gained more importance due to ease of quantification techniques. At the network level, functional evaluation is very useful for maintenance prioritization, whereas in the project level it focuses on identifying the root causes for the distress formation. It mainly deals with the evaluation of the following three components, i.e., roughness, distress, and skid resistance. The quality index is generally calculated based on the examination of these three features, which will explain about the condition of pavements. In the past, visual condition surveys and destructive testing were mostly used for evaluating the condition of pavements. However, these conventional methods are laborious and time-consuming, and the results are not accurate for quantification of the pavement condition. Hence there is a need for exploring the newer technologies for evaluating the functional condition of pavements. The main aim of this article is to present a review of studies which used newer technologies such as light detection and ranging (LIDAR) for the functional evaluation of pavements. In addition to that, a pilot study was carried out to show how the LIDAR technology can be used to assess the potholes on roads.

Use of Advanced Techniques for Functional Evaluation of Pavements …

157

2 Review of Studies on the Use of Advanced Techniques for Functional Evaluation of Pavements In this section, first the studies on roughness were presented, which is then followed by pavement distresses and finally the studies on skid resistance were presented.

2.1 Studies on Roughness Evaluation The most often utilized condition parameter in objectively evaluating pavement conditions is road roughness. Roughness is the deviation of the road surface from its true planar surface, and it has strong correlation with the vehicle operating cost as well as comfortability (Arianto and Suprapto 2018). Generally, the road roughness measuring methods are divided into four classes such as Precision Profiler Method, Other Profiler Methods, Response Type Road Roughness Measuring System (RTRRMS), and Subjective Ratings (Sayers et al. 1986). Despite the extensive usage of these instruments until now, there are still some drawbacks such as they are very huge in size, less coverage of road networks, cumbersome, expensive and depending on another instrument for calibration (Cundill 1991; Yi and Ma 2009; Choubane et al. 2002; Perera et al. 2006). In recent years, the smartphone-based roughness studies are getting popular due to their cost effectiveness and easy usage. These studies have considered the effect of vehicle speed, smartphone mounting position, and acceleration collection sampling rate for estimating roughness accurately (Alatoom and Obaidat 2021; Wang et al. 2020; Janani et al. 2021; Zang et al. 2018). Android-based applications combined with an accelerometer sensor provided in the smartphone are used to collect data more efficiently in this method (Islam et al. 2014). Researchers have addressed the need to investigate new techniques in functional evaluation of pavements especially in roughness measurement using different scanning technologies on replacement to inertial profilers (Alhasan et al. 2017). In recent years, transportation organizations have started using advanced technologies such as LIDAR for pavement evaluation in order to achieve greater benefits with less and fewer resources. Laser scanners, which are based on this technology, produce dense point clouds that provide exceptionally accurate and high-resolution 3D data (Blasiis et al. 2021). Many LIDAR-based studies have been reported worldwide for roughness measurement (Alhasan et al. 2017; Blasiis et al. 2021; Alhasan and White 2015; Kumar and Angelats 2017). Unlike other systems, these LIDAR scanners are not limited to terrain areas and traffic speed (Díaz-Vilariño et al. 2016). Nevertheless, in India, the usage of LIDAR scanners for pavement roughness evaluation is very limited and rare. This necessitates the need for research on the use of advanced technologies like LIDAR for pavement roughness calculation in India. Also, the reported studies from India on roughness calculation have mostly carried out the study on national highways (NH), state highways (SH), major district roads (MDR), other

158

N. H. Riyaz Khan and S. Vasantha Kumar

district roads (ODR), and village roads (VR) (Sandra and Sarkar 2013). However, the studies from urban roads were not found especially from India.

2.2 Studies on Surface Distress Evaluation One of the most essential criteria to ensure a durable pavement is the correct quantification of pavement distress, which normally defined as any indication of poor or unfavorable pavement performance or signs of impending failure. Types of distresses include Cracking, Patching, Rutting, Raveling, and Potholes. The most preferred method to identify distress is to visually assess the pavements and have subjective human specialists evaluate them. This is a traditional method which costs a lot of money, takes a long time, and often yields unreliable and variable findings (Ragnoli et al. 2018). The modern systems are made up of one or more acquisition devices and post-processing apps that use computer vision and image processing algorithms to perform semiautomatic or automated data extraction methods. In recent years, pavement distress inspection by remote sensing procedure using Unmanned Aerial Vehicles (UAVs) and LIDAR technology is widely employed in many countries (Huang et al. 2004; Li et al. 2019; Farhadmanesh et al. 2021). Díaz-Vilariño et al. (2016) proposed a method for quantification and classification of the distresses by creating a digital elevation model from the point cloud obtained by LIDAR system through image segmentation. Feng et al. (2022) measured the crack area and volume using five crack detection algorithms applied on the point clouds acquired through Terrestrial Lidar System (TLS). Ravi et al. (2020) proposed a fully automated technique for pothole detection and quantification, in which they have used 3D point cloud from MLS and reported an accuracy of ±1–2 cm. Barbarella et al. (2022) created a Digital Elevation Model (DEM) from the point cloud to analyze the distresses by considering velocity and acquisition rate of the survey vehicle. It was found that most of the studies were reported from other developed nations only and no study was found in India on the use of LIDAR for pavement distress evaluation. Only a few studies from India have explored the use of smartphones for roughness computation (Lekshmipathy et al. 2021). For example, Lekshmipathy et al. (2021) explored the feasibility of employing smartphones with accelerometers sensors for detecting potholes. However, the use of advanced instruments like LIDAR could not be found. Also, most of the reported studies have considered only one particular type of distress such as cracking or rutting and did not consider all the distresses. It is important to mention here that distresses on Indian roads especially in an urban context sometimes may be unique and will have a combination of many distresses. Identifying such distresses would be more challenging and hence demands advanced technologies such as LIDAR.

Use of Advanced Techniques for Functional Evaluation of Pavements …

159

2.3 Studies on Skid Resistance Evaluation One of the most essential parameters in condition evaluation is skid resistance of a pavement. The skid resistance measured by various devices on the same surface may vary because it is greatly influenced by the type of measured wheel, its loading, tread pattern, and the tire pressure (Yu et al. 2020). Skid resistance tester is one of the most popular conventional equipment for measuring the skid resistance of pavement in India and abroad and many studies have reported on the same. Aldagari et al. (2022) developed a predictive model for skid resistance of asphalt pavements by examining the surface friction details of 35 road sections. They used a skid trailer for measuring the skid number, whereas the micro-texture and macro-texture were measured by dynamic friction tester and circular texture meter. Pomoni et al. (2020) used a calibrated Grip Tester system for measuring skid resistance; it consists of fixed slip of the test wheel coupled with the water film depth to measure unitless friction measurement called the Grip Number. Xiao et al. (2018) studied the effect of aggregate morphology on skid-resistance of single-grade micro-surfacing using Aggregate Image Measurement System (AIMS). Though skid resistance tester is most popular, in recent years, advanced technologies like LIDAR were reported for measuring skid resistance. For example, Meegoda and Gao (2015) established a correlation between skid resistance and mean profile depth acquired through skid resistance trailer and vehicle mounted laser, respectively. Chen et al. (2021) proposed a three-dimensional evaluation method for classifying the texture of asphalt pavements by considering macroscopic and micro-texture features using 3D laser scanner. Though there are studies reported from other countries, however studies from India are very limited and this necessitates the need for research on these promising and potential areas.

3 Results of Pilot Study A pilot study was carried out to explore how LIDAR can be used to study the characteristics of a pothole on a road. A 100 m section on Jimmy carter road located within our university campus was taken as the study stretch for the pilot work. The reason for selecting this stretch is, a pothole of approximately 30 cm in length and breadth was found on this road as seen in Fig. 1. The Light Detection and Ranging (LIDAR) popularly called laser scanning is actually a remote sensing method where the light is emitted in a continuous manner from a laser instrument. This laser pulse after reflecting from earth features such as roads, tress, and buildings returns to the LIDAR scanner where the three-dimensional coordinate, i.e., X, Y, and Z of the target point was calculated using the distance computed from travel time and speed. The LIDAR system can be of either aerial or ground-based systems and in case of ground-based systems, the scanner would be normally kept on a tripod and then scanner rotates to capture the 3D of the surrounding area.

160

N. H. Riyaz Khan and S. Vasantha Kumar

Fig. 1 Photo of pothole studied using LIDAR

In the present study, ground-based laser scanning was carried out using one of the popular LIDAR scanners called Leica BLK 360 scanner as shown in Fig. 2. The scanner has a capability of emitting 3,60,000 laser pulses per second and thus can produce a point cloud of millions of data points of X, Y, and Z values. The range of the scanner is 60 m, which means the objects that are within a distance of 60 m, can be covered in one scan. For distances of more than 60 m, the instrument needs to be shifted with sufficient overlapping between the subsequent scans. As it can be seen from Fig. 2, the scanner was kept at a location in such a way that the pothole was within the range of 60 m from scanner. The scan time is generally 3 min/scan, and the advantage of BLK 360 is that it provides photos also taken with high-definition camera along with point cloud data. In other LIDAR scanners which are available in the market, only point cloud can be obtained while transferring the data, whereas the BLK 360 model can provide photos also taken during the survey along 360 degrees, which would be helpful at the time of registration of scans either through automatic or manual mode. In the present study, one scan was taken on April 12, 2022 during the mid-day. The collected data were then transferred to the Cyclone Register 360 software for post-processing of the data. The obtained point cloud data collected using LIDAR is shown in Fig. 3. A total of 82,34,787 (8.23 million) data points with X, Y, and Z values were collected at the study location. Using the point cloud data, the X, Y, and

Use of Advanced Techniques for Functional Evaluation of Pavements …

161

Fig. 2 Photo showing the LIDAR instrument and the pothole (red circle)

Z values were noted down along the pothole in longitudinal and transverse directions in order to find the depth of the pothole. The results are shown in Fig. 4. It can be seen that the elevation of the road or the road level is −1.0635 m (average of the left and right extreme Z values in Fig. 4). The reason for the negative value of the “Z” is that the road is located below the scanner by a depth of 1.0635 m as the

162

N. H. Riyaz Khan and S. Vasantha Kumar

Fig. 3 Screenshot showing the point cloud data from LIDAR survey

Fig. 4 3D coordinates along transverse direction of the pothole

scanner location was taken as 0, 0, 0 for X, Y, Z respectively. As we can see from Fig. 4, the pothole depth is 0.9 cm near the left edge (difference of 1.074 and 1.065) and 1.1 cm near the right edge (difference of 1.073 and 1.062). The maximum depth of pothole was observed near the center with a depth of 2.2 cm (difference between 1.084 and 1.062). In order to check the LIDAR derived pothole depths, the actual pothole depth was also measured using ruler as shown in Fig. 1. The actual depth was found to be almost close to 2.2 cm and this indicates that the LIDAR derived point cloud data has very high potential in mapping the potholes and other surface distresses. The main advantage of LIDAR data is it can measure the minute height differences of even 1 mm as the Z values reported by the scanner are in three decimal

Use of Advanced Techniques for Functional Evaluation of Pavements …

163

places (1.001–1.000 = 0.1 cm or 1 mm). Another advantage with point cloud data is the calculation of accurate surface area of the pothole and it was found to be 0.054 sq m for the pothole shown in Fig. 1. Thus, the results of pilot study revealed that LIDAR technology could be one of the promising tools in pavement distress evaluation.

4 Concluding Remarks The safety, comfort, traffic, travel times, vehicle operating cost, and emission levels on roads generally depends on the pavement condition. The present study has focused on the functional evaluation of pavement condition which involves the distresses, roughness, and skid resistance mainly. Literature review on the topic revealed that the developed countries have already shifted to the use of advanced technologies such as LIDAR for functional evaluation of their pavements. But in India, still we are using the conventional procedures like visual judgment for distress identification and evaluation, bump integrator for roughness measurement, etc. Though the use of smartphones was recently evolving for functional evaluation in the country, however the use of advanced technologies like LIDAR is still in nascent stage. High initial cost of LIDAR instruments may be the reason for why such advanced instruments have not become popular in the country. Because they normally cost in the range of 20 lakhs to 1 crore which may not be affordable by all. Recently in VIT Vellore, a ground-based LIDAR instrument from Leica was purchased and it was explored through a pilot study within the university campus for pothole identification. The results were found to be very promising as one can get 3D coordinates in “mm” accuracy which can be best utilized for functional evaluation of pavements in the country.

References Alatoom YI, Obaidat TI (2021) Measurement of street pavement roughness in urban areas using smartphone. Int J Pavement Res Technol Aldagari S, Al-Assi M, Kassem E, Chowdhury A, Masad E (2022) Development of predictive models for skid resistance of asphalt pavements and seal coat. Int J Pavement Eng 23:695–707 Alhasan A, White DJ, De Brabanter K (2017) Spatial pavement roughness from stationary laser scanning. Int J Pavement Eng 18:83–96 Alhasan AA, White DJ (2015) Terrestrial laser scanning roughness assessments for infrastructure Arianto T, Suprapto M (2018) Pavement condition assessment using IRI from roadroid and surface distress index method on national road in sumenep regency. In: IOP conference series: materials science and engineering, vol 333, pp 1–8 Barbarella M, Di Benedetto A, Fiani M (2022) A method for obtaining a DEM with curved abscissa from MLS data for linear infrastructure survey design. Remote Sens 14 Chen B, Xiong C, Li W, He J, Zhang X (2021) Assessing surface texture features of asphalt pavement based on three-dimensional laser scanning technology. Buildings 11

164

N. H. Riyaz Khan and S. Vasantha Kumar

Choubane B, McNamara RL, Page GC (2002) Evaluation of high-speed profilers for measurement of asphalt pavement smoothness in Florida. Transp Res Rec 62–67 Cundill MA (1991) MERLIN a low cost machine for measuring road roughness in developing countries. Transp Res Rec 1291:106–112 De Blasiis MR, Di Benedetto A, Fiani M, Garozzo M (2021) Assessing of the road pavement roughness by means of lidar technology. Coatings 11:1–19 Díaz-Vilariño L, González-Jorge H, Bueno M, Arias P, Puente I (2016) Automatic classification of urban pavements using mobile LiDAR data and roughness descriptors. Constr Build Mater 102:208–215 Farhadmanesh M, Cross C, Mashhadi AH, Rashidi A, Wempen J (2021) Highway asset and pavement condition management using mobile photogrammetry. Transp Res Rec 2675:296–307 Feng Z, El Issaoui A, Lehtomäki M, Ingman M, Kaartinen H, Kukko A, Savela J, Hyyppä H, Hyyppä J (2022) Pavement distress detection using terrestrial laser scanning point clouds—accuracy evaluation and algorithm comparison. ISPRS Open J Photogramm Remote Sens 3:100010 Huang B, Fwa TF, Chan WT (2004) Pavement-distress data collection system based on mobile geographic information system. Transp Res Rec 54–62 Islam S, Buttlar WG, Aldunate RG, Vavrik WR (2014) Measurement of pavement roughness using android-based smartphone application. Transp Res Rec 2457:30–38 Janani L, Sunitha V, Mathew S (2021) Influence of surface distresses on smartphone-based pavement roughness evaluation. Int J Pavement Eng 22:1637–1650 Justo-Silva R, Ferreira A, Flintsch G (2021) Review on machine learning techniques for developing pavement performance prediction models. Sustain 13:1–27 Khahro SH, Memon ZA, Gungat L, Yazid MRM, Rahim A, Mubaraki M, Nur NI (2021) Low-cost pavement management system for developing countries. Sustainability 13 Kumar P, Angelats E (2017) An automated road roughness detection from mobile laser scanning data. Int Arch Photogramm Remote Sens Spat Inf Sci ISPRS Arch 42:91–96 Lekshmipathy J, Velayudhan S, Mathew S (2021) Effect of combining algorithms in smartphone based pothole detection. Int J Pavement Res Technol 14:63–72 Li Z, Cheng C, Kwan MP, Tong X, Tian S (2019) Identifying asphalt pavement distress using UAV LiDAR point cloud data and random forest classification. ISPRS Int J Geo-Inf 8 Meegoda JN, Gao S (2015) Evaluation of pavement skid resistance using high speed texture measurement. J Traffic Transp Eng 2:382–390 Perera RW, Kohn SD, Wiser LJ (2006) Factors contributing to differences between profiler and the international roughness index. Transp Res Rec 81–88 Pomoni M, Plati C, Loizos A, Yannis G (2020) Investigation of pavement skid resistance and macrotexture on a long-term basis. Int J Pavement Eng 0:1–10 Ragnoli A, De Blasiis MR, Di Benedetto A (2018) Pavement distress detection methods: a review. Infrastructures 3 Ravi R, Habib A, Bullock D (2020) Pothole mapping and patching quantity estimates using lidarbased mobile mapping systems. Transp Res Rec 2674:124–134 Rusmanto U, Syafi’I Handayani D (2018) Structural and functional prediction of pavement condition (A case study on south arterial road, Yogyakarta). AIP Conf Proc 1977 (2018) Sandra AK, Sarkar AK (2013) Development of a model for estimating International Roughness Index from pavement distresses. Int J Pavement Eng 14(8):715–724 Sayers M, Gillespie T, Paterson W (1986) Guidelines for conducting and calibrating road roughness measurements. World Bank technical paper number 46 Wang G, Burrow M, Ghataora G (2020) Study of the factors affecting road roughness measurement using smartphones. J Infrastruct Syst 26:04020020 Xiao Y, Wang F, Cui P, Lei L, Lin J, Yi M (2018) Evaluation of fine aggregate morphology by image method and its effect on skid-resistance of micro-surfacing. Materials (Basel) 11 Yi Z, Ma RG (2009) A study of pavement roughness measurement system based on laser ranger finder. In: Proceedings 2009 international conference on image analysis and signal process IASP, vol 1, pp 295–299

Use of Advanced Techniques for Functional Evaluation of Pavements …

165

Yu M, You Z, Wu G, Kong L, Liu C, Gao J (2020) Measurement and modeling of skid resistance of asphalt pavement: a review. Constr Build Mater 260:119878 Zang K, Shen J, Huang H, Wan M, Shi J (2018) Assessing and mapping of road surface roughness based on GPS and accelerometer sensors on bicycle-mounted smartphones. Sensors (Switzerland) 18:1–17

A Machine Learning-Based Active Learning Framework to Capture Risk and Uncertainty in Transportation and Construction Scheduling Manoj K. Jha , Nicodeme Wanko, and Anil Kumar Bachu

Abstract In the execution of transportation and construction projects, the regulatory approval requirements and changes in original scope of work arising due to weather, design changes, and other unforeseen issues often result in significant cost overruns, thereby delaying the project completion schedule. In this paper, we develop a machine learning-based active learning framework to capture the risk and uncertainty in transportation and construction scheduling. In a typical critical path method, the duration times of individual tasks are assumed to be given and deterministic. In order to account for risk and uncertainty, we assume that the estimated activity duration is available from a large random sample between specified bounds. We conduct a case study for a 20-activity transportation and construction project. The result shows that the true extent of delay and cost overrun due to risk and uncertainty can be captured in the planning stages, thereby allowing the agencies to be prepared for such delays and cost overruns after the execution of the project. Another benefit is that an alternate critical path can be reworked if the extent of risk and uncertainty is known in advance. The result can be further enhanced in future works by using more comprehensive activity and project-specific historical data. Keywords Active learning framework · Machine learning · Fuzzy logic · Critical path method · Uncertainty · Project management

M. K. Jha (B) The Brite Group, Leesburg, VA, USA e-mail: [email protected] N. Wanko Morgan State University, Baltimore, MD, USA A. K. Bachu Indian Institute of Technology, Patna, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_13

167

168

M. K. Jha et al.

1 Introduction In the execution of transportation and construction projects, the government imposed regulatory approval requirements (e.g., environmental permitting process) and changes in original scope of work arising due to weather, design changes, and other unforeseen issues (e.g., shortage in skilled labor) often result in significant cost overruns, thereby delaying the project completion schedule. This issue is commonplace in India, the US, and other countries across the globe. For example, according to Lepartner (2007), by 2030, the U.S. will spend $25 trillion in construction, and if construction uncertainties and delays are not addressed, it will result in a waste of $3 trillion. Likewise, in the Indian context, a study done by Patil et al. (2015) identified the causes of delay in Indian transportation infrastructure projects. The study found land acquisition, environmental impact of the project, financial closure, change orders by the client, poor site management, and supervision by contractor as the top five important causes of construction delays in transportation projects. In another recent study by Rivera et al. (2020), the authors studied the causes of delay in road construction projects across 25 developing countries, including India. The study observed that the lack of experience of the construction manager, inadequate planning/scheduling, and influence on people’s land alongside the road construction project (expropriation for the construction of the project) have more significant impacts than frequent changes in the design (which was listed as the most frequent cause of delay). In this paper, we develop a machine learning-based active learning framework called modular Active Learning (modAL) to capture the risk and uncertainty in transportation and construction project scheduling. Critical project schedules are usually created using the Critical Path Method (CPM) (e.g., Parsa 2021; Chaher & Benseghir, 2018). In a typical CPM, the duration times of individual tasks are assumed to be given and deterministic. In order to account for uncertainty, we assume that the estimated duration is available from a large random sample between specified bounds. This idea is similar to the research conducted by Fang and Zhu (2014) in developing an active learning framework with uncertain knowledge. Fang and Zhu (2014) improvised on the uncertain labelling knowledge by introducing the diversity density concept. In the construction project scheduling delay context, as an example, let’s assume that we are uncertain about the duration of a particular activity, say the regulatory approval process. All we know is it can be any number between 15 and 75 days. We can construct a sufficiently large sample or feasible scenarios between these bounds and guide the search for a sub-optimal task duration based on historical data. We can then use this sub-optimal task duration in the CPM to obtain the critical path and sequence of activities. We consider the impacts of uncertainty in project management by exploiting fuzzy mathematical programming in conjunction with the modAL (2021) framework. A 20-activity example is used to test the effectiveness of the proposed methodology in capturing uncertainty and further enhancing the CPM-based scheduling.

A Machine Learning-Based Active Learning Framework to Capture …

169

2 Literature Review The problem of risk-based transportation and construction scheduling is not new. In the construction context, it was initially studied by Carr and Tah (2001). The authors developed a construction project risk management system using a fuzzy approach to construction project risk assessment and analysis. A hierarchical risk breakdown structure was described to represent a formal model for qualitative risk assessment. The relationships between risk factors, risks, and their consequences were represented on case and effect diagrams. Risk descriptions and their consequences were defined using descriptive linguistic variables. Using fuzzy approximation and composition, the relationships between risk sources and the consequences on project performance measures were identified and quantified consistently. Some recent works in this area include (1) a graph-based approach for unpacking construction sequence analysis to evaluate schedules (Hong et al. 2022); (2) a construction project scheduling methodology considering COVID-19 pandemic measures (Aslan and Turkakin 2022); (3) a new surrogate measure for resilient approach to construction scheduling (Milat et al 2021); (4) multi-objective multimode resource constrained project scheduling with fuzzy activity durations in prefabricated building construction (Yuan et al. 2021); and (5) development of the construction scheduling based on fuzzy discrete event simulation for a TFT-LCD plant (Su and Chang 2019). While the list of recent literature presented above is by no means exhaustive, it nevertheless highlights three important points: (1) risk and uncertainty are widely recognized in construction project scheduling; (2) fuzzy logic is predominantly used for modeling risk and uncertainty; and (3) ML-based active learning framework has not been applied before in construction project scheduling. In previous works, the first author developed a Dynamic Bayesian Network for considering risk and uncertainty in extreme events (e.g., terrorist threat, hurricane evacuation, and planning) (Jha and Keele 2012; Jha 2009; 2006b; Jha and Marroquin 2006). Additional significant papers that used fuzzy logic in modeling risk and uncertainty in the scheduling context are (1) Castro-Lacouture et al. (2009); (2) Chan et al. (2009); and (3) Fayek and Oduba (2005). Castro-Lacouture et al. (2009) used fuzzy mathematical models and the critical path method for construction project scheduling with time, cost, and material restrictions. The authors evaluated the viability of fuzzy mathematical models for determining construction schedules and for evaluating the contingencies created by schedule compression and delays due to unforeseen material shortages. Chan et al. (2009) provided an overview of the application of fuzzy techniques in construction management research. Fayek and Oduba (2005) developed fuzzy expert systems for predicting industrial construction labor productivity. The factors that affected the productivity of each activity were identified, and fuzzy membership functions and expert rules were developed. The models were validated using data collected from an actual construction project. The resulting models were found to have high linguistic prediction accuracies.

170

M. K. Jha et al.

While the first author has used an ensembled-based ML model (e.g., a random forest classifier) in transportation problems (Jha and Ogallo 2021, 2022), the application of modAL for minimizing labeling error due to uncertainty has not been used in the transportation or construction context in previous works. This gives us an opportunity to apply modAL in conjunction with fuzzy logic to model risk and uncertainty in transportation and construction scheduling problems.

3 Methodology We develop a delay minimization formulation as follows: Min D =

n 

δi p(ai )

(1)

i=1

subject to li ≤ pi (ai ) ≤ u i for all pi

(2)

δi ≤ 1

(3)

D≤M

(4)

  p j a j − pi (ai ) ≥ 0

(5)

where D = total delay; δi = uncertainty factor in the completion of the ith activity; pi (ai )=projected completion of the ith activity; M = maximum allowable delay; and l i and ui are lower and upper bounds of the completion time of task i, respectively. Equation (5) implies that activity j cannot be completed before activity i.

3.1 Fuzzy Mathematical Formulation of Risk and Uncertainty Next, we use fuzzy logic to formulate the risk and uncertainty in project completion times. Let M(x) be the fuzzy set that represents the user’s perception of risk and uncertainty for project completion times measured on a scale of 1–10. It describes the user’s perception of risk as low, medium, or high. The corresponding fuzzy membership function is μ M (x) whose value ranges between 0 and 1, and it is represented as a triangular fuzzy number. In previous research, it is shown that triangular fuzzy

A Machine Learning-Based Active Learning Framework to Capture …

171

numbers best represent the uncertainty scenarios (see, for example, Kikuchi and Jha 2006). A typical triangular fuzzy number is most often presented in the form: A = (a1 , a2 , a3 )

(6)

where a1 is lower (left) boundary of the triangular fuzzy number, a2 is number corresponding to the highest level of presumption, and a3 is upper (right) boundary of the fuzzy number. The membership function μ A (x) of the fuzzy number A is ⎧ 0, ⎪ ⎪

⎪ ⎪ x−a1 ⎨ , a −a μ A (x) = a2 −x1

3 ⎪ , ⎪ ⎪ a −a ⎪ ⎩ 3 2 0,

⎫ ⎪ ⎪ ⎪ ⎪ a1 ≤ x ≤ a2 ⎬

x ≤ a1

a2 ≤ x ≤ a3 ⎪ ⎪ ⎪ ⎪ ⎭ x ≥ a3

(7)

By the similar notion, we assume that μ M (x) can be represented as a triangular fuzzy number with M = (m1 , m2 , m3 ). For example, if M = (2, 6, 10) then μ M (x) can be expressed as ⎧ ⎪ ⎪ ⎨

⎫ 0, x ≤ 2 ⎪ ⎪ ⎬ x − 0.5, 2 ≤ x ≤ 6 4 μ M (x) = ⎪ − x + 2.5, 6 ≤ x ≤ 10 ⎪ ⎪ ⎪ ⎩ 4 ⎭ 0, x ≥ 10

(8)

Using a triangular fuzzy number, we calculate the risk perception based on Fuzzy Logic. The above formulation is an extension of the work by Jha (2006a, 2007); and Jha and Davy (2019).

3.2 Modeling Uncertainty with ModAL In a typical CPM, the duration times of individual tasks are assumed to be given and deterministic. In CPM algorithms, an upper and lower bound as well as the most likely time of completion of a task are specified. But these values are deterministic and assumed to be specified based on the user’s own experience. This creates the reliability of the CPM solution doubtful since any inaccuracy in the task duration time will lead to an inaccurate critical path. Fang and Zhu (2014) faced a similar issue in the ML context in problems dealing with labeling information for training instances in a supervised learning task. They developed a formulation for an active learning paradigm with uncertain information. They offered an algorithm that used an error-reduction sampling estimation.

172

M. K. Jha et al.

In the present research, we account for uncertainty by assuming that the estimated duration is available from a large random sample between specified bounds. This idea is similar to the research conducted by Fang and Zhu (2014) in developing an active learning framework with uncertain knowledge. The mathematical formulation for the active learning framework with uncertain knowledge can be found in Fang and Zhu (2014), and therefore, it has been skipped here for brevity. The active learning framework to account for uncertainty is developed using a Python package called modAL. We have adapted modAL in the current research to account for uncertainty in estimating task duration times. modAL is an active learning framework for Python-3, designed with modularity, flexibility, and extensibility in mind. Built on top of the Python package called Scikit-learn, it allows us to rapidly create active learning workflows (modAL 2021). modAL was designed primarily to measure the uncertainty of predictions based on the large training datasets.

3.3 An Example with modAL to Model Activity Duration with Uncertainty Let’s assume that we are uncertain about the duration of a particular activity. But we do know that it can be any number between 3 and 10 days. We construct 100 numbers using the modAL package in Python which are randomly distributed in the neighborhood of 3 and 10. As can be seen in Figs. 1, 2, the initial regression is very poor. It is based on the ensemble method.

Fig. 1 Initial Regression (x-axis represents probability distribution between 0–1 and y-axis represents number of days of completion of a task in question)

A Machine Learning-Based Active Learning Framework to Capture …

173

Fig. 2 Final Regression (x-axis represents probability distribution between 0–1 and y-axis represents number of days of completion of a task in question)

After active learning and prediction after 20 queries, the regression fitting is much better as shown in the set of plots below. Figure 2 shows that active learning is very effective in performing the best regression over time and therefore an activity duration of 8 days is the best estimate that should be used for this activity in the CPM process. While it is a relatively simple example, it makes the point that the active learning algorithm improves the solution over successive iterations.

4 Example and Discussion We apply the proposed methodology in a 20-activity transportation and construction project. Using the active learning framework, we modify the projected completion time (PCT) of individual tasks by considering risk and uncertainty. We introduce a risk level, assign a risk numerical grade, and calculate risk perception using fuzzy logic. We also introduce a risk uncertainty factor after which we calculate the PCT and delay. Table 1 shows the data. The actual and projected completion times are shown in Fig. 3. It is observed that the projected completion time of an activity after considering risk and uncertainty is generally higher than without their consideration. This implies that real-world projects often incur delays in execution due to the inability of capturing risk and uncertainty because additional funds are generally needed whose actual value cannot be ascertained in advance, resulting in delays. Next, we assign unit cost of the delay in the completion of the activities and calculate the costs incurred due to the delay. Table 2 shows the result. The highest

29

11

10

6

18

5

6

7

8

27

11

5

14

35

10

17

16

12

24

5

5

10

18

19

20

28

27

28

12

18

15

16

17

19

11

12

31

15

11

13

26

17

8

12

9

10

25

29

24

23

16

16

3

30

4

2

17

12

18

1

Actual completion time (ACT) in days

Projected completion time (PCT) (obtained from CPM in days) in days

Activity

Table 1 Example of risk perception with fuzzy logic

L

M

H

H

M

H

M

M

L

L

L

H

H

M

L

L

L

H

M

M

Risk level

2

6

8

8

5

9

5

5

1

2

3

8

7

5

2

2

3

8

6

5

Risk numerical grade

0.25

1

0.5

0.5

0.75

0.25

0.75

0.75

0.25

0.25

0.25

0.5

0.75

0.75

0.25

0.25

0.25

0.5

1

0.75

Risk perception based on fuzzy logic

13

10

8

15

32

15

9

19

24

19

15

12

32

11

13

14

20

24

36

21

PCT after considering risk by fuzzy logic

0.57

0.51

0.87

0.8

0.38

0.57

0.23

0.31

0.17

0.57

0.88

0.6

0.7

0.29

0.04

0.66

0.37

0.06

0.32

0.09

Risk uncertainty factor (RUF)

16

8

9

18

25

19

6

14

22

24

23

13

31

8

10

18

22

17

24

13

6

3

4

8

7

7

1

3

3

9

11

5

13

2

0

7

6

1

6

1

PCT after Delay = considering risk/ PCT x UF uncertainty factor (in days)

174 M. K. Jha et al.

A Machine Learning-Based Active Learning Framework to Capture …

175

Fig. 3 Comparison of actual and projected completion times (x-axis represents number of tasks and y-axis represents number of days)

cost is incurred for Activity 15 because it is an expensive activity which suffers a delay of 7 days. Consider boring a tunnel in a highway construction project, which is an expensive activity. Any delay in the execution of this activity after it has already been scheduled (that is, labor, equipment, and material have already been allocated and are on standby) will significantly inflate the cost. On the other hand, activity 6 has no delay due to risk and uncertainty as a result of which it does not affect the cost.

4.1 Model Limitations While the approach presented here is interesting and offers useful insight to project delays due to risk and uncertainty, its application in real-world projects requires the following: (1) The lower and upper bounds of activity completion will be based on historical data and project specific. Therefore, a good modAL application will require a more realistic capture of these bounds; (2) The fuzzy factors to model risk and uncertainty will be activity and project specific. They need to be fine-tuned based on historical data; and (3) The unit delay cost will be activity and project specific. They need to be fine-tuned.

176 Table 2 Cost of delay due to risk and uncertainty

M. K. Jha et al.

Activity

Delay (in Days)

Unit cost of delay ($/day)

Total delay cost

1

1

453,061

453,061

2

6

514,296

3,085,776

3

1

791,435

791,435

4

6

123,429

740,574

5

7

625,829

4,380,803

6

0

79,878

0

7

2

868,505

1,737,010

8

13

113,525

1,475,825

9

5

615,596

3,077,980

10

11

524,010

5,764,110

11

9

340,324

3,062,916

12

3

774,261

2,322,783

13

3

15,225

45,675

14

1

545,656

545,656

15

7

979,385

6,855,695

16

7

156,348

1,094,436

17

8

28,552

228,416

18

4

238,871

955,484

19

3

423,641

1,270,923

20

6

723,212

4,339,272

5 Conclusions and Future Works We developed a fuzzy mathematical approach that uses an active learning framework called modAL to account for perceived delays in activity durations. We performed an example to assess the effectiveness of the developed methodology. The result showed that delay for individual activities for the example studied ranged from 0 to 13 days. This is significant because any revision to the activity durations will change the critical path since the sequence of the activities leading up to the critical path will change. This will require readjusting material and personnel resulting into significant cost overrun. While length of delay and resulting cost overrun will be project specific, what is important to note is that any changes in the individual activity durations may change the sequence of intermediate activities resulting into significant inflated cost. Future works may include using historical project specific data to fine-tune the upper and lower limits of the activity duration and a more robust formulation of risk and uncertainty using empirical data. Future work may also include an estimation of the inflated costs due to risk and uncertainty in order to realize the full potential of the developed methodology.

A Machine Learning-Based Active Learning Framework to Capture …

177

Acknowledgements This work is part of the second author’s doctoral work and not tied to any external grant or funding. All authors have contributed to the study and have approved the paper for publication.

References Aslan S, Turkakin O (2022) A construction project scheduling methodology considering COVID-19 pandemic measures. J Safety Res 80:54–66 Carr V, Tah JHM (2001) A fuzzy approach to construction project risk assessment and analysis: construction project risk management system. Adv Eng Softw 32(10):847–857 Castro-Lacouture D, Gürsel AS, Gonzalez-Joaqui J, Yates JK (2009) Construction project scheduling with time, cost, and material restrictions using fuzzy mathematical models and critical path method, ASCE. J Constr Eng Manag 135(10):1096–1104 Chaher A, Benseghir H, Ibrahim A, Kabir MN (2018) A Fuzzy-based risk model for construction project management. IEEE Conference on Systems Process and Control (ICSPC) 2018:129–134 Chan APC, Chan DWM, Yeung JF, (2009) Overview of the application of fuzzy techniques in construction management research, ASCE J Constr Eng Manag, 35(11) Fang M, Zhu X (2014) Active learning with uncertain labeling knowledge. Pattern Recogn Lett 43:98–108 Fayek AR, Oduba A (2005) Predicting industrial construction labor productivity using fuzzy expert systems. ASCE J Constr Eng Manag 131(8):938–951 Hong Y, Xie H, Hovhannisyan V, Brilakis I (2022) A graph-based approach for unpacking construction sequence analysis to evaluate schedules. Adv Eng Inform 52:101625. https://doi.org/10. 1016/j.aei.2022.101625 Jha MK, Ogallo H (2022) Using machine learning for the development of a maintenance management system: case study of Kenya. Int J Econ, Financ Manag Sci 10(4):166–172 Jha MK, Ogallo H, (2021) Studying the dynamic sight distance problem with a machine learning algorithm. In: Proceedings of the 2021 Annual TRB Meeting, Washington, D.C., Paper Number: TRBAM-21–03783 Jha MK, Davy B, (2019) Investigating the Benefit-Cost of MEMS application for structural health monitoring of transportation infrastructure. In: 2007 Cleantech Conference and Trade Show Cleantech 2007: The Cleantech Conference, Venture Forum and Trade Show (1st ed.). CRC Press. https://doi.org/10.1201/9780429187469, Laudon M, Romanowicz P (eds), 326–329 Jha MK, Keele R, (2012) Using dynamic bayesian networks for investigating the impacts of extreme events, In Bayesian Networks, W. Premchaiswadi (ed), 35–48, InTech Open (www.intechopen. com) Jha MK (2009) A dynamic bayesian network for predicting the likelihood of a terrorist attack at critical transportation infrastructure facilities. J Infrastruct Syst 15(1):31–39 Jha MK, (2007) A Fuzzy-Evolutionary algorithm for optimizing benefit/cost ratio of 3-dimensional highway alignments. In: Proceedings of the 2nd International Conference on Modeling, Simulation, and Applied Optimization, Abu Dhabi, UAE, March 2007 (available from www.resear chgate.net) Jha MK (2006a) Feasibility of computer visualization in highway development: a fuzzy logic-based approach. Comput-Aided Civ Infrastruct Eng 21(2):136–147 Jha MK (2006b) Applying bayesian networks to assess vulnerability of critical transportation infrastructure. Applications of Advanced Technology in Transportation. ASCE Press, Alexandria, VA, pp 1–6 Jha MK, Marroquin O, (2006) A dynamic bayesian network for hurricane planning and preparedness, In: Proceedings of the Seventh International Congress on Advances in Civil Engineering,

178

M. K. Jha et al.

October 11–13, 2006, Yildiz Technical University, Istanbul, Turkey (available from www.resear chgate.net) Kikuchi S, Jha MK (2006) A Method for reconciling values of parameters, In: Proceedings of the 2006 Annual INFORMS AI/DM Workshop, Pittsburgh, PA (available from www.researchg ate.net) Lepartner BB, (2007) Broken buildings, buster budgets: how to Fix America’s trillion-dollar construction industry, The University of Chicago Press, Chicago, ISBN-13:978-0-226-47267-6 Milat M, Knezic S, Sedlar J (2021) A new surrogate measure for resilient approach to construction scheduling. Procedia Comput Sci 181:468–476 modAL (2021), modAL, a Python Package. https://github.com/modAL-python/modAL#:~:text=% 20modAL-python%20/%20modAL%20%201%20Introduction.%20Built,4%20Installation.% 20%205%20Citing.%20%20More, accessed 1/27/2022 Parsa S (2021) Estimation of activities duration in control project using fuzzy logic. https://papers. ssrn.com/sol3/papers.cfm?abstract_id=3930312 Patil SK, Gupta AK, Desai DB, Sajane AS (2013) Causes of delay in Indian transportation infrastructure projects. Int J Res Eng Technol 2(11):71–80 Rivera L, Baguec H Jr, Yeom C (2020) 2020, A study on causes of delay in road construction projects across 25 developing countries. Infrastructures 5:84. https://doi.org/10.3390/infrastru ctures5100084 Su T-S, Chang L-M (2019) Development of the construction scheduling based on fuzzy discrete event simulation for a TFT-LCD plant. Procedia Manuf 38:1130–1137 Yuan Y, Ye S, Lin L, Gen M (2021) Multi-objective multi-mode resource-constrained project scheduling with fuzzy activity durations in prefabricated building construction. Comput Ind Eng 158:107316

Earthwork Logistics Optimization in Road Construction Project Furqan A. Bhat, Debopam Roy, Prasanta K. Sahu, Sanjay C. Choudhari, and A. Bahurudeen

Abstract Earthwork requires the movement of large quantities of soil from borrow areas to fill areas, typically for embankment and subgrade preparation. These operations contribute to a significant component of a project and deal with the balancing of cutting and filling volumes cost-effectively. Earthworks contribute about 25% of the total cost of a road project, out of which one-fourth is spent on the earthmoving operations alone. Careful planning of earthmoving operations can make considerable savings in the logistics cost due to many feasible options available to the project team. This paper aims to demonstrate the optimization technique’s ability to provide costefficiency in managing the logistics of raw material (earthwork) supply chain in the road project. This study focuses on planning the logistical operations of earthwork for a 14 km road construction project between two Indian towns. A large quantity of soil is needed to be procured to prepare these embankment and subgrade layers for which 14 borrow sites were identified. This paper presents the transportation problem as a linear programming formulation for the optimal planning and distribution of earthwork from borrow areas to different segments of the road (fill areas). The objective function is to minimize the total cost of transportation of soil from borrow sites for the construction of embankment and subgrade layers constrained by the total demand at construction sites and total supply at borrow sites. Keywords Earthwork logistics optimization · Linear programming · Road construction · Embankment · Subgrade F. A. Bhat (B) Department of Civil Engineering, Indian Institute of Science Bengaluru, Karnataka 560012, India e-mail: [email protected] D. Roy National Institute of Construction Management and Research, Pune, Maharashtra 411 045, India P. K. Sahu · A. Bahurudeen Department of Civil Engineering, Birla Institute of Technology and Science Pilani, Hyderabad, Telangana 500078, India S. C. Choudhari Operations Management and Quantitative Techniques, Indian Institute of Management Indore (IIM Indore), Indore, India © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_14

179

180

F. A. Bhat et al.

1 Introduction Infrastructure industry characteristics are very different from other industries in terms of planning, scheduling and sequencing, resource allocation, site production, and temporary organization (Koskela 2000). Planning, sequencing, and scheduling any construction project are difficult and crucial tasks as the choices made at the early planning stage of a project are bound to affect the project’s successful execution from the initial conceptual stage to the late completion stage (Waly and Thabet 2003). These activities for all the infrastructural projects require the necessary allocation of raw materials at the construction sites as per the need. Improper planning can lead to scarcity of the necessary construction resources at the specified site, which can adversely affect the project performance measures such as time, cost, and safety at the construction site (Mawdesley et al. 2004). Among all infrastructures, road projects are prioritized as these projects are essential for any country’s economic development and require a significant budget to plan and execute such projects. However, most of these projects are affected by the cost and time overrun, which are accepted as key performance indicators of any project. Several studies (Manavazhi and Adhikari 2002; Kaliba et al. 2009; Mahamid 2013) have systematically identified the factors affecting road project performance, such as organizational weakness, government regulations, suppliers’ default, inclement weather, inflation and suggested the ways to mitigate their effects. The typical cross-section of road construction includes various layers such as embankment, subgrade, wet mix macadam base, etc. A large quantity of materials is moved for these multiple layers, and relocation amounts to a substantial monetary cost. Earthwork is one of such critical activity in preparing these layers, especially for embankment and subgrade. Earthwork is carried out at the earlier stages of a project and impacts the planning and scheduling of the entire project. Many heavy linear projects such as roadway and railway construction require substantial earthmoving operations (Moselhi and Alshibani 2009). These operations contribute to a major component of a project, involve high cost, use heavy machinery, and deal with the balancing of cutting and filling volumes in a cost-effective manner. Earthworks contribute about 25% of the total cost of a road project, out of which 20–25% is spent on the earthmoving operations alone (Hare et al. 2011), impacting the rest of the project activities and the overall project performance (Mawdesley et al. 2004). The earthmoving operation costs are incurred due to the transportation of large quantities of soil for several kilometers to the point of placement. Any effort in monitoring the transportation of earthmoving is an essential factor in optimizing the road construction cost. It is observed that the road construction industry mostly relies on the judgment of construction managers and the convenience of contractors to decide on the logistic plan for earthmoving. Inadequate planning, coupled with arbitrary decision-making, results in considerable cost overruns in earthmoving. Optimization of logistics planning can improve cost benefit to the road project, mostly due to substantial savings in logistics cost (Sobotka et al. 2012a). Even a small improvement in logistics cost can provide cost efficiency leading to financial benefit for the

Earthwork Logistics Optimization in Road Construction Project

181

project. This research aims to demonstrate the optimization technique’s usefulness to provide the cost benefits in managing the raw material (earthwork) supply chain in the road project. This paper uses a linear programming formulation for application in the optimal planning and distribution of earthwork from borrow areas to different segments of the road (fill areas) to minimize logistics cost. As per our observation, in the proposed model, the logistic cost is directly related to the distance by which the soil is transported from sources to different road segments. Hence, this work optimizes the distance as a proxy to logistics cost. The model’s outcome provided an optimal distribution of the earthwork from the given supply locations to the road construction sections’ demand locations. This paper has four more sections in addition to this introduction. The next section gives a brief overview of the earlier studies related to optimizing the construction materials’ logistics process. The research methodology is provided subsequently in the third section, which describes the objective function and the constraints. Results are discussed in the fourth section, and the final section concludes the paper.

2 Literature Review A significant portion of earthmoving operations characterizes linear construction projects such as roads, railways, pipelines, and dams. These operations are carried out at the initial stages of the projects, involve huge costs, and optimize the haul distances between cutting and filling sections to maximize profit. As such, a considerable amount of effort is put into estimating and planning such works. Different techniques and models have been developed for optimizing earthmoving operations. These include (i) economical allocation techniques such as mass-haul diagram (Stark and Mayer 1983), (ii) linear programing method developed by Stark and Nicholls (Stark and Nicholls 1972) which was later enhanced by other researchers (Stark and Mayer 1983; Mayer and Stark 1981; Easa 1987; Jayawardane 1990), (iii) queuing models (Halpin and Woodhead 1976), (iv) expert systems and artificial intelligence (Christian and Xie 1996), (v) simulation and genetic algorithmbased models (Marzouk and Moselhi 2003, 2004; Hajjar and AbouRizk 1999), and (vi) models using commercial software. This section describes linear programming models developed and used in earlier studies in managing logistic operations. Linear programming models to minimize the cost of earthmoving operations were first used by Mayer and Stark and later advanced by many other researchers (Mayer and Stark 1981). These mathematical models are developed to optimize an objective function using certain decision variables and are bound by some physical constraints. The cost of transportation of materials is minimized while satisfying the minimum requirements at various demand locations using maximum available material from different source locations. Mayer and Stark analyzed the unit cost of earthmoving operations using the linear programming optimization method (Mayer and Stark 1981). A slight modification to this method was proposed by Easa (1987) to formulate a linear programming model to analyze the disparity in unit cost as a function

182

F. A. Bhat et al.

of the quantity of earth moved in a stepwise form. Linear programming models were further advanced into mixed-integer linear programming models (MILP). Easa (1987) formulated a mixed-integer linear programming optimization model to minimize the logistical cost of earthwork between borrowing and filling areas by additionally including the cost of new cutting and filling areas (Easa 1988). Similarly, Jayawardane (1990) developed a MILP model to select construction equipment to optimize the cost and finish the project within the constrained period. There has been excessive research on this subject in the last couple of decades (Sobotka et al. 2012a; Son et al. 2005; Wong et al. 2010; Lima et al. 2012). To optimize the cost of earthmoving operations, Son et al. formulated an LP model to find out the shortest distance possible between the borrow and the filling areas (Son et al. 2005). Similarly, Sobotka et al. proposed a mathematical linear programming model to optimize construction materials’ movement between suppliers and consumers based on time and cost of transportation (Sobotka et al. 2012a). On the other hand, researchers developed a mixed-integer linear programming model to plan construction facilities with minimum logistical costs (Wong et al. 2010; Lima et al. 2012). The logistical planning of an infrastructure project is characterized by the large-scale movement of construction materials and construction equipment between various production and consumption points. Planning of these movements can be optimized to make the overall process cost-effective. The optimization model characterized by only consumption and supply points is called the transportation models. In contrast, a model with a processing stage and demand and supply stage is called a transshipment model. Transportation and transshipment models can be mathematically developed and solved either by linear programming or mixed-integer linear programming optimization technique (Ma and Suo 2006; He et al. 2012; Steinrücke and Jahr 2012).

3 Research Methodology This section describes the methodology used for this study. We provide details about the road earthwork project. This is followed by a brief description of the model used for planning the logistics movement of the earthwork.

3.1 Problem Definition This study focuses on planning the logistical operations of earthwork for a road construction project between the two Indian towns. The project is a part of a highway project and incorporates a stretch of 44 km road. The soil being transported is needed to prepare embankment and subgrade before laying the final bituminous layers. A large quantity of soil needed to be procured to prepare these two layers, and hence 14 borrow sites were identified by the contractor and approved by the client. All 14 sites

Earthwork Logistics Optimization in Road Construction Project

183

provide soil with different properties and quality. This quality is measured in terms of an index called California Bearing Ratio (CBR), which indicates the soil’s strength for road construction. As per the IRC:37–2018 (IRC:37-2018 2018) guidelines, soils with CBR value minimum of 5% for subgrade and 3% for embankment is suggested to be used. The stretch of 44 km road was estimated to use soil from 14 borrow areas to prepare embankment and subgrade. For the earthwork operation planning, the road work was divided into 44 equal segments (1 segment = 1 km), and each road segment was required to be prepared for both embankment and subgrade layers (see Fig. 1). The total soil (in cubic meter) for each road section was estimated based on the road alignment plan and profile across all the sections. The cost of soil procurement at the construction site includes the cost of material and the cost of soil transportation from the borrow areas to the fill areas. The cost of transportation can be obtained by multiplying the cost per kilometer (as quoted by the transporter) and the distance between the respective borrow and fill sites. The cost of raw materials (soil) from each borrowed site was constant irrespective of their difference in properties and quality. Since the transportation cost (per kilometer) and the cost of soil (per cubic meter) is constant, the model was insensitive to these parameters. Based on these factors, it is logical to assume that minimizing the total distance (in km cubic meter) will reduce the total cost of preparing embankment and subgrade. Fig. 1 A Schematic network diagram for earthwork distribution from borrow (quarries) areas to road segments

184

F. A. Bhat et al.

3.2 Model for Distance-Volume Minimization The model proposes soil transportation from borrow areas to road segments to prepare embankment and subgrade to minimize the total distance (in km cubic meter), eventually reducing the cost. This model can be viewed as a logistics network optimization transportation problem and solved by linear programming (LP) optimization technique. LP technique optimizes (minimize/maximize) a linear objective function, subject to linear equality and linear inequality constraints. To formulate the problem, distances from borrow areas to all the road segments, the total demand for soil at all road segments, and available soil at each borrow area are considered. The following generalized formulation is proposed considering the various parameters and decision variables, as defined in Table 1.

Minimize Distance Volume

I  J 

Di j ∗ E i j +

i=1 j =1 J 

I  J 

D i j ∗ Si j

(1)

i=1 j =1

E i j + S i j ≤ S i for i = 1, . . . I where I = 14

(2)

j =1 I 

E i j ≥ M j for j = 1, . . . J where J = 44

(3)

i=1

Table 1 Description of the notations Notations

Description

Index i

Source (borrow area, i.e., quarries) index (from 1 to I);

j

Demand (road segment) point index (from 1 to J);

Decision variables E ij

Quantity of soil shipped from borrow area i to road segment j in cubic meter for embankment;

S ij

Quantity of soil shipped from Borrow area i to road segment j in cubic meter for subgrade;

Parameters I

Total number of quarries = 14;

J

Total number of road segments = 44;

Dij

Distance between borrow area i to road segment j (mid-point) in kilometers (km);

Si

Total available supply of soil at each borrow area i in cubic meter;

Mj

Total demand of soil at road segment j in cubic meter for embankment;

Nj

Total demand of soil at road segment j in cubic meter for subgrade;

Earthwork Logistics Optimization in Road Construction Project I 

S i j ≥ N j for j = 1, . . . J where J = 44

185

(4)

i=1 J 

Si j ≤ 0 for i = 5, 6, 7, 9, 10, 12

(5)

j =1

E i j ≥ 0, S i j ≥ 0

(6)

The objective function is presented by Eq. (1), followed by constraints given by Eqs. (2–6). The objective of the study is to minimize the cost of transportation of the earthwork. This is achieved by Eq. (1) by minimizing the distance over which the material is transported as distance-volume. Equation (2) enforces that the total raw material moved from the source cannot exceed the maximum available quantity. This means the total sum of soil shipped from a borrow area i for both embankment Eij and subgrade Sij at road segment j cannot exceed the available supply of soil Si . Equations (3) and (4) ensure that the total sum of material that reaches the demand point meets the required quantity at each respective location. This condition states that the total quantity of soil required for the preparation of embankment Eij and subgrade Sij at a road section j from all the combinations of borrow areas should be greater than or equal to the minimum requirement Mj and Nj respectively. Equation (5) meets the quality requirement at the demand location, especially for subgrade. This condition ensures that the borrow area’s soil with a CBR value of less than five is not used for subgrade. Similar constraints are not required for embankment due to the CBR value of more than 3% of all the quarries. Finally, Eq. (6) imposes that all the decision variable values are always either zero or more than zero. The model formulation included 1232 decision variables, 88 demand constraints, and 14 supply constraints. The data required for the model was sought from the project team, including the original plan of earthwork distribution for possible benchmarking. It comprises the distance between each quarry to each road segment in kilometers, the quantity of soil available at each quarry in cubic meters, CBR values of soil at each quarry, and the required amount of soil at each road segment for both embankment and subgrade in cubic meters. The paper minimizes the distancevolume for moving the soil from quarries to road segments. Furthermore, we also sought information about transportation costs to determine the logistics cost and the potential saving over the original plan. The Excel spreadsheet Solver does not support more than 200 decision variables. Hence, the current paper used Premium Solver to solve the proposed formulation in an Excel spreadsheet. The Solver output was then interpreted to move the soil from quarries to embankment and subgrade layers of the given 44 road segments. The optimal distribution plan was then compared with the project team’s original plan for determining the significance of realized saving in the logistics cost and discussed in the next section.

186

F. A. Bhat et al.

4 Results and Discussions The proposed model formulation was run in the premium Excel Solver. The optimal solution minimizes the total distance-volume of soil shipments from fourteen borrow areas to road segments for embankment and subgrade layers. The optimal logistics plan was then compared with the original plan prepared by the project team. The project team also had prepared the logistics plan for the movement of soil from sources to road segments in the Excel spreadsheet using common sense and experience (Sobotka et al. 2012b). The detailed comparison of both optimal and original plans, including the cost and saving, is shown in Table 2. The total distance-volume saving of the optimal plan over the original plan shows the difference of 2,17,039 cubic meter kilometers. Nonetheless, given the transportation cost, the optimal plan was expected to save Rs 1.8275 million over the original plan. This suggests approximately 7.28% saving over the original estimate of the project team. This is a very significant saving considering the competitive bid among the contractors who work with a 10 to 20% margin for road construction work. Any project manager’s critical objective is to complete the project cost-effectively within the planned timeframe. The borrow area (i.e., quarries) meets the requirement of several road segments for embankment and subgrade. The final percentage of soil to be consumed from various borrow areas for both the plans are shown in Table 3. The maximum available soil supply from each quarry, including the CBR values, was available to the project team. Table 3 shows both percentage and quantity in cubic meter from the available supply at each quarry. Further, it shows the percentage of quantity consumed for embankment and subgrade from the quarry’s procured supply. It also determines the number of truck trips that need to be planned from each quarry to the road segment. It is to be noted that the procured soil’s total quantity in both the plans is the same from both the quarries. However, the percentage of quantity procured from a few quarries varies. This has resulted in saving the cost in the optimal plan compared to the original plan due to reduced distance-volume. It is observed that many borrow areas are completely used; some are used partly while two are not used at all in both the plans. The quantities that are procured partly vary in proportions between both the plans. For example, the quantity planned to be procured from quarry Q5 was far more in the original plan than suggested by the optimal plan. However, the quantities to be procured in quarries Q3, Q4, Q6, and Table 2 Cost comparison of optimal and project team plan Original plan a. Total distance quantity for meeting the 29,81,311 logistics plan (m3 km) b. Total logistics cost (Rs)#

Optimal plan 27,64,272

Saving

Percentage saving

2,17,039 7.28

2,51,02,639 2,32,75,173 18,27,466 7.28

of transportation based on details given by transporter = Rs 8.42 per unit of m3 km (So, b = a * 8.42)

# Cost

Q3

Q4

Q5

√

√

√

√

7634

Procured (% of supply quantity)

Procured (m3 )

Q3

30.98

61.85

0

763

Subgrade (%)

No. of trips from quarriesb

Q2

100

Q1

9.54

Details

Procured (% of supply quantity)

Optimal plan of procurement

12,000

38.15

Embankment 100 (%)

31.48

Q3

9295

33.19

66.81

120,000 92,946

Q2

100

Q1

9.54

Details

Q4

25.29

Q4

6571

65.86

34.14

65,706

21.90

41.95

Q5

4532

NA

100

45,323

75.54

Q5

80,000 120,000 300,000 300,000 60,000

Q2

Original plan of procurement by the project team

CBR > = 5a

Supply quantity (m3 )

Q1

Quarry no (source for earthwork) Q7

Q8

Q9

Q10

Q11

Q12

Q13

Q14

Total

Q6

18.43

Q6

3373

NA

100

33,731

16.87

Q7

Q8 100

Q9 0

100

Q7

3000

NA

100

100

Q8

5000

100

0

0

Q9

0

NA

0

30,000 50,000 0

100

√

Q10

74.34

Q10

8383

NA

100

83,831

69.86

0

Q11

0

0

0

0

0

Q11

√

Q12

Q13 100

100

Q14

√

Total 44.00

100

Q12

2500

NA

100

100

Q13

2400

95.05

4.95

100

Q14

10,000

57.77

42.23

(continued)

44

Total

67,817

41.00

59.00

25,000 24,000 100,000 678,171

100

√

200,000 30,000 50,000 55,000 120,000 75,000 25,000 24,000 100,000 1,539,000

Q6

Table 3 Comparison of optimal and project team logistics procurement original plan

Earthwork Logistics Optimization in Road Construction Project 187

Q3

763

No. of trips from quarriesb

9444

33.65

7587

80.81

19.19

75,866

Q4

Q1

0

Q2

0

Q3

1490

Q4

10,160

Q6

3685

NA

100

36,854

Q6

−20,151 3124

Q5

2517

NA

100

25,172

Q5

Q8

Q9

0

Q7

3000

NA

100

0

Q8

5000

100

0

0

Q9

0

NA

0

30,000 50,000 0

Q7

Diff in Truck 0 0 149 1016 −2015 312 0 0 0 trips √ Tick ( ) indicates that CBR is meeting the requirement of subgrade, NA (quarry not compatible) a CBR value more than 5, b consider the capacity of a truck is 10m3

% Diff in (m3 )

Difference in Optimal and Original procurement plan

12,000

50.62

0

Subgrade (%)

66.35

120,000 94,436

49.38

7634

Q2

Embankment 100 (%)

Procured (m3 )

Q1

Quarry no (source for earthwork)

Table 3 (continued)

538

5377

Q10

8921

NA

100

89,209

Q10

0

0

Q11

0

0

0

0

Q11

Q13

Q14

Total

0

0

Q12

2500

NA

100

0

0

Q13

2400

91.52

8.48

0

0

Q14

10,000

53.13

46.87

0

0

Total

67,817

41.00

59.00

25,000 24,000 100,000 678,171

Q12

188 F. A. Bhat et al.

Earthwork Logistics Optimization in Road Construction Project

189

Q10 are more in an optimal plan than the original plan. There are many feasible solutions for shipping soil from borrow areas to road segments due to the many options available to the project team. However, the solution obtained by the means of common sense could be far from optimal due to the complexity involved in such cases, as can be observed by the results of this case study. Such a situation requires a lot of computational efforts and should not be handled without using optimization method. It is possible to make considerable savings in logistics cost by optimization rather than allocating resources based on experience and thumb rules. Such small savings can provide a significant cost benefit to a project because of a cost-efficient plan. The project logistics team can interpret and use the distribution plan for planning, scheduling, and monitoring the entire project’s earthwork in an optimal way. It indicates the quantity of soil to be procured from each borrow area and moving it along the optimal links to various road segments both for embankment and subgrade. The total quality to be procured can determine the number of truck trips that need to be made from quarries to road segments. The spreadsheet solution also provides the details regarding the quality of soil to be transported on each link as described by decision variables such as Eij and Sij . The entire shipment plan can be linked with the project planning network and monitored accordingly for distribution. The team can make an appropriate contract with the logistic provider and schedule the transportation carrier as per the optimal plan and project schedule.

5 Conclusions The paper addresses the usefulness of an optimization model for earthwork cost minimization in a road project. The outcome of the model presents the optimal plan of shipment from several feasible options. The solution provides logistics distribution of soil from different borrow areas to various road segments of embankment and subgrade. It is expected to result in better efficiency leading to significant cost benefits compared to the original plan. The project team can use an optimized plan in synchronization with the project planning network to plan and schedule the project’s entire earthwork. Since road construction engineers and managers often use spreadsheets, this article demonstrates the ability of Excel Solver to provide solutions for optimizing resources leading to financial benefits. Thus, the proposed approach will be helpful to project leads while planning the project logistics. This research effort will be useful for the project and construction managers while illustrating the development and ability of a Solver-based optimization model to plan and schedule raw material movement in the road project supply chain. Further, the schedule distribution can be created by linking it to the project schedule and determining the required resources such as truck trips. It can also be of use to understand the impact of such an application in real-life implementation regarding time performance and compare it with the project team’s actual logistics plan based on some judgment.

190

F. A. Bhat et al.

References Christian J, Xie TX (1996) Improving earthmoving estimating by more realistic knowledge. Can J Civ Eng 23:250–259 Easa SM (1987) Earthwork allocations with nonconstant unit costs. J Constr Eng Manag 113:34–50 Easa SM (1988) Earthwork allocations with linear unit costs. J Constr Eng Manag 114:641–655 Hajjar D, AbouRizk S (1999) Simphony: an environment for building special purpose construction simulation tools. Metall Mater Trans A 30:2221 Halpin DW, Woodhead RW (1976) Design of construction and process operations. Wiley, New York Hare WL, Koch VR, Lucet Y (2011) Models and algorithms to improve earthwork operations in road design using mixed integer linear programming. Eur J Oper Res 215:470–480. https://doi. org/10.1016/j.ejor.2011.06.011 He T, Ho W, Man CLK, Xu X (2012) A fuzzy AHP based integer linear programming model for the multi-criteria transshipment problem. Int J Logist Manag 23:159–179. https://doi.org/10.1108/ 09574091211226975 IRC:37–2018 (2018) Guidelines for the design of flexible pavements. Indian Road Congress, New Delhi Jayawardane BAKW (1990). 116:18 Kaliba C, Muya M, Mumba K (2009) Cost escalation and schedule delays in road construction projects in Zambia. Int J Proj Manag 27:522–531. https://doi.org/10.1016/j.ijproman.2008. 07.003 Koskela L (2000) An exploration towards a production theory and its application to construction. VTT Publ de Lima RX, Júnior EFN, de Prata BA, Weissmann J (2012) Distribution of materials in road earthmoving and paving: mathematical programming approach. J Constr Eng Manag 139:1046– 1054. https://doi.org/10.1061/(ASCE)co.1943-7862.0000666 Ma H, Suo C (2006) A model for designing multiple products logistics networks. Int J Phys Distrib Logist Manag 36:127–135. https://doi.org/10.1108/09600030610656440 Mahamid I (2013) Effects of project’s physical characteristics on cost deviation in road construction. J King Saud Univ Eng Sci 25:81–88. https://doi.org/10.1016/j.jksues.2012.04.001 Manavazhi MR, Adhikari DK (2002) Material and equipment procurement delays in highway projects in Nepal. Int J Proj Manag 20:627–632 Marzouk M, Moselhi O (2003) Object-oriented simulation model for earthmoving operations. J Constr Eng Manag 129:173–181. https://doi.org/10.1061/(asce)0733-9364(2003)129:2(173) Marzouk M, Moselhi O (2004) Multiobjective optimization of earthmoving operations. J Constr Eng Manag 130:105–113. https://doi.org/10.1061/(ASCE)0733-9364(2004)130:1(105) Mawdesley MJ, Askew WH, Al-Jibouri SH (2004) Site layout for earthworks in road projects. Eng Constr Archit Manag 11:83–89. https://doi.org/10.1108/09699980410527885 Mayer RH, Stark RM (1981) Earthmoving logistics. J Constr Div 107:297–312 Moselhi O, Alshibani A (2009) Optimization of earthmoving operations in heavy civil engineering projects. J Constr Eng Manag 135:948–954. https://doi.org/10.1061/(asce)0733-936 4(2009)135:10(948) Sobotka A, Jaskowski P, Czarnigowska A (2012a) Optimization of aggregate supplies for road projects. Procedia Soc Behav Sci 48:838–846. https://doi.org/10.1016/j.sbspro.2012.06.1061 Sobotka A, Jaskowski J, Czarnigowska A (2012b) Optimizing of aggregate supplies for road projects. Procedia Soc Behav Sci 48:838–846 Son J, Mattila KG, Myers DS (2005) Determination of haul distance and direction in mass excavation. J Constr Eng Manag 131:302–309. https://doi.org/10.1061/(asce)0733-9364(2005)131: 3(302) Stark RM, Mayer RH (1983) Quantitative construction management: uses of linear optimization. Wiley Stark RM, Nicholls RL (1972) Mathematical foundations for design: civil engineering systems. Courier Corporation

Earthwork Logistics Optimization in Road Construction Project

191

Steinrücke M, Jahr M (2012) Tactical planning in supply chain networks with customer oriented single sourcing. Int J Logist Manag 23:259–279. https://doi.org/10.1108/09574091211265387 Waly AF, Thabet WY (2003) A virtual construction environment for preconstruction planning. Autom Constr 12:139–154. https://doi.org/10.1016/S0926-5805(02)00047-X Wong CK, Fung IWH, Tam CM (2010) Comparison of using mixed-integer programming and genetic algorithms for construction site facility layout planning. J Constr Eng Manag 136:1116– 1128. https://doi.org/10.1061/(ASCE)co.1943-7862.0000214

Urban and Rural Transportation

Reduction of Vehicular Emission at Urban Road Junctions Through Traffic Interventions Sumaiya Rahman and Mithun Mohan

Abstract The road transportation sector is one of the dominant sources of vehicular emission, which is causing very high levels of air pollution. Any technique to effectively control pollution is only possible by precisely estimating vehicle exhaust emissions. The objective of the research is to estimate vehicular emissions near the signalized intersection under the effect of traffic, control, vehicle, and road characteristics. This will enable to establish the link between emissions and the most likely influencing and measurable characteristics of Indian traffic conditions. The simulation results generated in VISSIM are imported into EnViVer to calculate the total emissions and emissions of individual vehicle classes. By simulating various combinations of vehicular, traffic, geometric, and control conditions at the intersection, the researcher will be able to arrive at the optimal combination that will result in minimal vehicular emission. The mathematical models that could be developed out of the research will be helpful to field practitioners in selecting the best strategy to tackle air pollution resulting from vehicular traffic. CO2 emission reduces significantly by 45.79% by decreasing 2W (Two-wheeler) and cars in the traffic stream by 75% and substituting these with busses. Further, a reduction of pollution levels by 91.1% occurs when all conventional cars are replaced by electric vehicles. Hence, encouraging the use of public transportation and the adoption of electric-powered vehicles could be the right step to tackle the ever-increasing pollution levels on Indian roads. Keywords Vehicle emissions · Simulation · Signalized intersections · Indian traffic

S. Rahman · M. Mohan (B) Department of Civil Engineering, National Institute of Technology Karnataka, Surathkal, Karnataka 575025, India e-mail: [email protected] S. Rahman e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_15

195

196

S. Rahman and M. Mohan

1 Introduction The rapid growth of urbanization in India over the past decade led to a tremendous increase in the number of vehicles, increasing the growth of the road transportation sector, especially in Delhi, the metropolitan city of India. The number of vehicles per thousand population increased considerably from 317 in 2005-06 to 598 in 201718. Poor air quality caused by the increase in vehicles is one of the main causes of environmental stress in various developing countries. Transport sector is the major contributor to air pollution through vehicle exhaust emissions in India. In India, it is the third most CO2 emitting sector. Road transport within the transport sector contributed more than 90% of total CO2 emissions (International Energy Agency (IEA), 2020; Ministry of Environment Forest and Climate Change 2018). The poor urban air quality near traffic intersections is generally due to the interruption in traffic flow, frequently occurring delays and start–stops. Emission rates depend on the characteristics of traffic, vehicles, control, and road geometry. Traffic characteristics such as traffic flow rate, vehicular compositions, and road characteristics such as free left turn movement could potentially impact the traffic operation and resulting pollution at intersection. These traffic and road characteristics combined with vehicle and control characteristics result in higher exhaust concentration at road junctions. Investigating the effect of each of these parameters and their combination in the field presents the investigator with the challenge that the occurrences of such conditions cannot be controlled. On the contrary, microscopic traffic simulation models coupled with an emission estimation module could be used to precisely estimate vehicular exhaust emissions and suggest viable rectification. The Indian heterogeneous traffic consists of a wide mix of vehicles having diverse static and dynamic characteristics. The mix consists of the various composition of both motorized and non-motorized vehicles. The absence of lane markings, and lane discipline is another feature of this traffic condition. Nowadays, transportation engineers mostly use VISSIM, as a microscopic traffic simulation tool, to simulate the actual traffic conditions and to allow speed-based emission estimation. A calibrated model can be obtained by changing default parameters to represent field conditions accurately and can be built for precise emission estimation. VISSIM supports an addon module called EnViVer, a microscopic exhaust emission modeling tool, which estimates emissions based on the speed-time profile of individual vehicles. It is used for estimating CO2 , NOx and PM10 emissions by considering the speed-time trajectories of vehicles. The report presents the preparation, simulation, and emission estimation of a VISSIM model. This model estimates the emissions at a four-legged signalized intersection in Delhi, India. It deals with heterogenous traffic flow and is controlled by four-phase signal system having signal cycle length of 120 s. The base data gathered is considered as input parameters in VISSIM, and the simulation results are imported into EnViVer to estimate emissions.

Reduction of Vehicular Emission at Urban Road Junctions Through …

197

2 Literature Review Chauhan (2019) estimated emission for signalized intersection using VISSIM and EnViVer, which investigated the effect of vehicle composition and green interval of signals on vehicle exhaust emission at Vadodara city. Satish Chandra (Bedada et al. 2020) calculated the amount of traffic a flyover can handle based on fuel consumed, emission generated and acceptable delays. The impact of real-world driving conditions on vehicle exhaust emissions was explored by Nesamani and Subramanian (2006) using the IVE (International Vehicle Emission) model. An emission model was developed by Partin et al. (2012) using real traffic data collected for FLAME and VISSIM traffic flow models. Sekhar et al. (2013) estimated delay and fuel loss during idling vehicles at signalized intersections in Ahmedabad city. Singh et al. (2020) developed an inventory for the fleet to compare past emission, evaluate control policies, estimate state-wise vehicle emission inventories, and identify significant emitters in the fleet. However, despite the extensive efforts in the literature to estimate vehicular emissions, most of the studies mainly focused on vehicular emission estimation without considering the different scenarios that might have resulted in emissions. The combination of heterogeneous traffic modeled in microscopic traffic simulation software and instantaneous emission models needs to be explored in the immediate future for enabling road users to provide a possible measure for emission reduction by considering the Indian traffic conditions.

3 Study Area India’s capital, Delhi has been on the front pages of news frequently for being one of the most polluted capitals in the world. It is located at Latitude 28°35' N and Longitude 77°12' E, with an area of 1483 km2 and a resident of 16 million people in the northern part of India. Although the authorities have tried multiple ways to curb the pollution levels in the city, their efforts have not delivered significant outputs. The winter in Delhi is marked by frequent cold, dry air and ground-based inversion with low wind conditions, increases pollutant concentrations (Central Pollution Control Board (CPCB) 2000). The city has an adequate road network and more registered vehicles than combined vehicles of Chennai, Mumbai and Kolkata (Central Pollution Control Board (CPCB) 2000), which leads to traffic congestion, thereby increasing air pollution. Therefore, this study focuses on traffic emissions at a typical signalized intersection located in Delhi.

198

S. Rahman and M. Mohan

4 Methodology The methodology of designing signalized intersections starts with developing a base intersection in VISSIM. The traffic inputs and their behaviors are modified to simulate the traffic operations in Delhi. The major pollutants emitted from transport sector are CO2 , NOx , and PM10 . Once the VISSIM models are prepared, different cases will be simulated to investigate the effect of traffic, geometric and control conditions on the resulting air pollution levels. The impact of each of these and their combinations will be studied through simulation runs. The output of simulations is then imported into EnViVer to calculate the total emissions and emissions of individual vehicle classes. The results will then be subjected to optimization to determine the combination that would yield minimal vehicular emission at signalized intersections. Mathematical models are obtained to tackle air pollution resulting from vehicular traffic. The sequence of events undertaken in this study are explained in the subsequent sections.

4.1 Modeling of Signalized Intersection VISSIM, a microscopic time-step tool, was used in the study to create a base model that accurately represents the typical signalized of Delhi. The creation of this base model begins with the addition of traffic and geometric data of the intersections, such as traffic volume, routing decisions, traffic composition, vehicle types, vehicle classes, and desired speed distribution. These are added as fixed inputs their values are given in Table 1. For signalizing the intersection in VISSIM, traffic signal controls for traffic movements are set up according to the signal time listed with green signal time of 26 s, a red time of 2 s and amber time of 2 s. VISSIM uses the built-in car-following models for simulating the driving behavior parameters of an individual vehicle class. By modifying the default values of the parameters from previous studies (Mohan and Chandra 2017), traffic movement found on Indian roads is simulated. Table 1 Fixed input data S. no

Vehicle composition

Input traffic volume

Traffic route allocation

Class

Approach

Movement

Speed

Relative flow

Volume (Veh/ hr)

Relative flow

1

2W

40

0.46

WB

2000

LT

0.15

2

3W

25

0.15

NB

2000

TM

0.60

3

CAR

30

0.33

EB

2000

RT

0.25

SB

2000

4

BUS

20

0.035

5

LCV

25

0.015

6

HCV

20

0.01

Reduction of Vehicular Emission at Urban Road Junctions Through …

199

4.2 Emission Estimation VISSIM’s add-on module EnViVer is used for estimating real-world emissions using speed-specific vehicle trajectory generated by the simulation model. EnViVer is capable of estimating CO2 , NOx , and PM10 emissions. The vehicle record files (. fzp format) of VISSIM are imported into EnViVer to estimate emissions. The vehicle record file contains the information about coordinates, vehicle types, simulation second, speed, lane/link/gradient, etc. In EnViVer, customization is done as per the local traffic and road characteristics for the correct estimation of emissions. EnViVer emission model provides vehicle class customization for light-duty vehicles, heavyduty vehicles, and busses. In this research, two-wheelers, three-wheelers, cars, and LCVs are considered as light-duty vehicles, while all trucks are considered heavyduty vehicles. The data relating to vehicle fleet properties such as fuel distribution, age distribution, introduction year of emission standards and average CO2 emission are needed as inputs in EnViVer to compute the emission. Since the study focuses on traffic simulation and emission modeling of typical signalized intersections in Delhi, the registered vehicle population for Delhi from 2005 to 2021 is collected from the ‘VAHAN’ server, maintained by the Ministry of Road Transportation and Highways, Govt. of India. The registered vehicles are available under motorized two-wheelers (2W), three-wheelers (3W), cars, busses, light-duty, and heavy-duty vehicles. The age distribution requires computation as given in Table 2, while rest of the properties are directly taken from previous studies (Dhyani 2017; Emission factor development for Indian vehicles ARAI, Air Quality Monitoring Project-Indian Clean Air Programme 2007). Information on the age distribution of vehicles for the present year is computed using the registered vehicle population and a survival function. Equation (1) is used to estimate the survival function of vehicles. Survival rate Su (i, a) is a function of vehicle type ‘i’ of age ‘a’, while shape factor ‘α ‘relates to the onset of significant retirement and ‘L50 ’ is the age by which 50% of vehicles have retired taken from the previous studies (Pandey 2014). The vehicle population of a particular category (type) each year ‘a’ is multiplied with the estimated Su (i, a) to obtain the new vehicle population on-road for the given reference year. The details of the age distribution of different types of vehicles are presented in Table 2. Fuel distribution of vehicles for the fleet is another important parameter for emission calculations. According to the fuel type, vehicles in Delhi are powered by petrol, diesel, CNG, LPG and electricity, and their distribution in the traffic mix has been taken from the previous studies performed in Delhi (Dhyani 2017). Euro introduction year and average CO2 emission are gathered from previous research (Emission Table 2 Vehicle age distribution Vehicle class

2W

3W

CAR

BUS

LCV

HCV 14.9

Newer than 1 year (%)

9.5

9.2

6.7

7.7

10.9

Average vehicle age (years)

5.4

5.5

7.4

7.5

5.5

5.0

10.6

10.9

19.0

13.0

10.0

13.0

Average exit age (years)

200

S. Rahman and M. Mohan

factor development for Indian vehicles ARAI, Air Quality Monitoring Project-Indian Clean Air Programme 2007). Su(i, a) =

1

1+e

)] [ ( α 1− La

(1)

50

4.3 Alternative Scenarios Traffic emission can be reduced by improving traffic, vehicular, control, and road characteristics at intersection. Different alternative scenarios are tested by modifying these characteristics of base model. Traffic characteristics are modified by changing the vehicle composition and traffic volume. One major factor that was investigated was the effect that a greater share of public transportation has on vehicular emissions. This was tested by decreasing the volume of motorcycles and cars and increasing the volume of busses proportionally to check their effect on emissions. In modified compositions, the volumes of motorcycles and cars were decreased at a rate of 0.25%, 0.5%, and 0.75% separately and together, respectively, and subsequently, the volume of busses increased. To study the impact of traffic volume on emissions, the input traffic volume from the approaches were changed from 1500 to 500 veh/h, with the saturation flow being around 2000 veh/h. The impact of vehicular characteristics on emission is considered by changing a certain percentage of cars that runs on conventional fuels to electric-powered. The percentage of conventional fuelled cars were replaced by 25, 50, 75 and 100% of electric-powered cars at different volume levels to estimate their effect on air quality at intersection. Travel pattern of the vehicle is yet another factor that could possibly have an influence on the emission. The geometric characteristics of the intersection in VISSIM was modified to allow for free left turn with different traffic volume governing the effect of travel pattern on the emission. The effect of control characteristic on emissions is studied by changing the green signal timings to 20, 32 and 38 s. The findings of the testing are explained in the next section.

5 Results and Discussions 5.1 Effect of Traffic Characteristics on Emission The quantity of pollutants generated from the vehicle exhaust for the base model of signalized intersection having a green signal time of 26 s was found to bear an exponential relationship with the traffic volume at the intersection. This is shown in Fig. 1. When this intersection is dealing with a traffic volume of 6535 veh/h, the

Reduction of Vehicular Emission at Urban Road Junctions Through …

201

levels of different pollutants are as follows: CO2 = 1273 kg/h, NOx = 4.156 kg/h and PM10 = 0.4268 kg/h. Among these, CO2 is the most critical pollutant from vehicle exhaust emission. It was found that providing efficient public transport facilities for the safe and economical movement of people could offer a significant impact in reducing air pollution. To test this statement, the proportions of 2W and car were decreased individually and together through 75% and subsequently the proportion of busses was increased to replace the reduced private vehicles. The change in pollution levels when 2W and cars are replaced by busses is shown in Fig. 2. It could be seen that the level of emission is the least when both 2W and cars are replaced by busses. This is due to the increase in public vehicles and significant percent decrease in diesel cars. It could be seen from the graph that replacing 2W and cars together with busses are effective only when the proportion replaced exceeds 40%.

Fig. 1 Effect of traffic volume on emission

Fig. 2 Effect of vehicle composition on emission

202

S. Rahman and M. Mohan

Fig. 3 Effect of different share of electric vehicles on emission

With the aim of reducing vehicle’s exhaust and dependency on conventional fuel type, the government is promoting the adoption of electric vehicles to replace the present vehicles on the road. Electric vehicles produce zero carbon emissions and zero nitrous oxides compared to conventional fuel vehicles. Figure 3 shows the effect of introducing 25, 50, 75 and 100% of electric vehicles on emission by replacing the conventional cars in the traffic stream at the intersection.

5.2 Effect of Intersection Geometry and Control on Emission To investigate the influence of geometry on emission, free left turns were provided by modifying the base model in VISSIM. Although a separate left turn lane was provided to begin much ahead of the signal, it did not have any significant impact on emission. The amount of emission produced when the intersection had free left turns for different traffic volume levels is presented in Fig. 4, and the comparison of emissions between the intersections with and without left turns is depicted in Fig. 5. It could be observed from Fig. 5 that there are not many differences between the emissions produced the base intersection and the modified intersection having a separate left turn lane. This might be because of the queued vehicles deny the opportunity to the left turning vehicles to access the reserved left turn lanes. Thus, in the present case, the provision of a separate left turn lane will not be an uneconomical solution. The study also explored the effect of control characteristics of the signalized intersection on vehicular emissions. For this, the green time for all the approaches was varied at constant input traffic volume. The emission estimated with the variation

Reduction of Vehicular Emission at Urban Road Junctions Through …

203

Fig. 4 Effect of free left turn on emission

Fig. 5 Comparison of emission with and without free left turn

in green time is plotted in Fig. 6. It could be seen that there is a slight reduction in the emission levels with an increase in the green time. However, the effect drops for higher values of green times. At the same time, modifying green time for a slight reduction in emission may not be a good option as the green time is set as a part of the signal design for the intersection.

5.3 Summary of Emission Model for Signalized Intersection This section summarizes the mathematical models that were developed by the use of best-fit lines based on non-linear regression analysis. The models presented in Table 3 can be used to estimate the total CO2 emitted (in kg/h) using Traffic Volume (in veh/ h) and Green Time (in sec) for the different traffic, geometry and control conditions discussed earlier. The mathematical model for traffic and road characteristics follows

204

S. Rahman and M. Mohan

Fig. 6 Effect of green time on emission

Table 3 Mathematical model Characteristics

Mathematical models

Remarks

Traffic

Total CO2 = 104.52e0.0004*Traffic Volume

Emission based on change in traffic volumes

Road

Total CO2 = 99.515e0.0004*Traffic Volume

Emission on free left turn intersection with respect to traffic volume

Control

Total CO2 = 0.1597(Green Time)2 − 13.947(Green Time) + 1518.2

Emission with respect to change in green time

an exponential relationship, while the control characteristic follows a polynomial model.

6 Conclusions Pollutant such as CO2 emitted immensely by vehicle is majorly responsible for global warming. Therefore, a significant increase in pollution from the road transport sector is a major concern for different countries. This case study attempts to examine the effect of traffic, geometric, vehicular and control characteristics on emission by simulating the operation of a typical signalized intersection in Delhi based on the data collected from previous studies. The study found a significant decrease in emissions with a decrease in traffic volume. Further, an increase in busses and electric vehicles to replace conventional traffic also reduces vehicular emissions. The study identified that emissions increase exponentially with the traffic volume. Based on simulation analysis, the study recommended a systematic invention of the traffic stream, including popularizing the use of public transportation and electric vehicle, the level of emission could be substantially reduced. The mathematical models developed in this study will be helpful in achieving the goals of sustainable

Reduction of Vehicular Emission at Urban Road Junctions Through …

205

transportation through traffic management and control. The study is beneficial for transport planners to tackle air pollution and is the first step in a positive direction to frame sustainable transport policies for the city. The study was based on calibrated parameters from previous research performed for similar traffic conditions. However, for any other conditions, the parameters of the simulation model should be established based on field data. The future research could consider how emission changes with meteorological conditions and other geometric characteristics of the road section that were not considered in this research.

References Bedada L, Advani M, Chandra S, Juremalani J (2020) Estimating the impact of flyover on vehicle delay, fuel consumption, and emissions—a case study. Recent Adv Traffic Eng 69:517–530 Central Pollution Control Board (CPCB) (2000) and 2010a Chauhan BP (2019) Car following model for urban signalised intersection to estimate speed-based vehicle exhaust emissions. Urban Climate 29:100480 Dhyani R (2017) Sensitivity analysis of CALINE4 model under mix traffic conditions. Aerosol Air Qual Res 17:314–329 Emission factor development for Indian vehicles ARAI (2007) Air quality monitoring project-Indian clean air programme International Energy Agency (IEA) (2020) Ministry of Environment Forest and Climate Change (2018) Mohan M, Chandra S (2017) Queue clearance rate method for estimating passenger car equivalents at signalized intersections. J Traffic Transp Eng (English edition) 4(5):487–495 Nesamani KS, Subramanian KP (2006) Impact of real-world driving characteristics on vehicular emissions. JSMEB 49:19–26 Pandey A, Venkataraman C (2014) Estimating emissions from the Indian transport sector with on-road fleet composition and traffic volume. Atmos Environ 98:123–133 Partin JB, Moore Tim O, Idewu W (2012) Comparison of measured and modeled vehicle emissions for advance prediction of air quality and NOx exposure level. Int J Eng Res Technol 4:91–97 Sekhar CR, Raj P, Gangopadhyay S (2013) Estimation of delay and fuel loss during idling of vehicles at signalised intersection in Ahmedabad. Procedia Soc Behav Sci 104:1178–1187 Singh N, Mishra T, Banerjee R (2020) Emissions inventory for road transport in India in 2020: framework and post facto policy impact assessment

Development of PCU Model for Unsignalised Intersection: A Case Study of Ranchi City Aarohi Kumar Munshi and Ashish Kumar Patnaik

Abstract An unsignalised intersection is the confluence of two major paths or a major path along with a minor path without any traffic signals or signboards. This study emphasizes to propose the Passenger Car Unit (PCU) Model for the weak lane discipline that occurs under mixed traffic conditions in the capital city of the Indian state of Jharkhand. PCU estimation is vital for signal design, traffic capacity analysis and other applicable scenarios too. To carry out the proposed study, different stretches of unsingnalised intersection are identified either on National Highways or in major suburbs of a city. The data were collected through a videography technique using High- Definition Camera. The data collected were of the peak hours of traffic flow mainly during morning and evening time. The sites were selected considering the importance of intersections where traffic flow is extreme, and it needs to be analyzed for swift movement of vehicles. The model was developed by interpreting and analyzing the collected data in which PCU is estimated on the basis of effective area, speed of the vehicle and lagging headway. The estimated PCU values are compared with the existing Indo-HCM recommended values that are currently being used. Keywords Unsignalised intersection · Passenger Car Unit (PCU) · Lagging Headway

A. K. Munshi · A. K. Patnaik (B) Department of Civil and Environmental Engineering, Birla Institute of Technology Mesra, Ranchi, Jharkhand 835215, India e-mail: [email protected] A. K. Munshi e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_16

207

208

A. K. Munshi and A. K. Patnaik

1 Introduction India’s traffic situations are extremely complicated and diversified nowadays, owing to the country’s 138 million inhabitants, which comprises a vast range of dynamic vehicle categories and a significant socioeconomic disparity. Vehicular traffic occupies diverse locations on the road, travels at varying speeds, and even accelerates at varying rates. The operating situations on roadways become more difficult when there is no barrier in between vehicles of different sizes as they all take a common lane without clear barriers. Furthermore, analyzing and estimating various traffic aspects and metrics, such as highway capacity, Level of Service (LOS), density, and so on, becomes extremely challenging. Vehicle traffic and road infrastructure in developing countries differ substantially from those in developed countries. Cars are the primary mode of transportation in most modern countries, with only a few trucks and other vehicles making up the rest of the traffic. Throughout developing countries, cars with a wide variety of dynamic and static parameters use the same space on the road, and their maneuverability is unconstrained. It is clear from these contrasts between homogeneous and mixed traffic that it is more difficult to implement traffic operations and plan routes in congested traffic scenarios. The passenger car equivalent (PCE), often known as the passenger car unit (PCU) is a measure that may be used to transform several types of vehicles into a single unit of vehicular flow. Highway Capacity Manual (HCM) incorporated PCU in its 1965 edition to account for the influence of buses and trucks on the traffic flow, which was first introduced in HCM. It was defined as the number of passenger cars displaced in traffic flow by a truck or bus under existing highway and traffic conditions. PCU values for several vehicles are already available, and they are primarily based on static vehicle attributes and on vehicles that maintain lane discipline in traffic. The research aims to propose a method that reliably assesses the value of PCUs for vehicles in mixed traffic circumstances by incorporating both static and dynamic aspects of moving vehicles in the unsignalized intersections. In addition to the length and width of a vehicle, the engine capacity, speed, and type of vehicle are all factors that influence PCU values. The engine capacity of a vehicle is generally proportionate to its physical size. As a result, this study does not explicitly assess engine capacity. A variety of factors influence the end results, such as the kind of infrastructure, number of lanes, existence or lack of kerbs and medians, etc. The physical as well as the effective area of the vehicle along with the velocity of approach at the intersections and lagging headway are all factors evaluated in this study. The rectangular area and lateral clearance parameters are used to determine the effective space occupied by the subject vehicle on the roadway. The dynamic factors that help to comprehend the traffic volume, flow speed, and road capacity in current diverse traffic conditions include velocities and lagged headway.

Development of PCU Model for Unsignalised Intersection: A Case …

209

2 Background Literature The study of PCU estimation includes various parameters to make it count worth. The study started way back in the middle of the 1900’s to analyze the capacity of multilanes in traffic flow. Numerous ways and formulations were designed for crossings that were not indicated with signs. Several two-lane highway regions in Northern and Eastern India were studied for this analysis. All of the locations were level, straight, and devoid of any traffic restrictions. A two-lane highway’s capacity can be affected by the lane width, according to this study, which found that different types of cars had varied PCUs. The PCU value is calculated by dividing the standard vehicle’s speed and physical area to that of the subject vehicle. The wider the lane, the higher the PCU will be for a given vehicle type. Though lane width has a linear effect on the PCU in theory, the slope of that linearity differs from vehicle to vehicle (Chandra and Kumar 2003). Regardless of the vehicle, the critical gap and follow-up time were used to compute the lane’s capacity, and then the capacity of standard passenger vehicles movement was divided by the capacity of subject vehicles movement in order to determine the PCU at unsignalised intersection. The critical gap for minor lane vehicles to conduct its maneuvers is the interval between the approaches of major lane vehicles. As per (Transportation Research Board, National Research Council 2000) follow-up time is the amount of time that elapses between the departures of minor lane vehicles that use the same gap in the major street in the case of queuing on the minor lane. A further occupancy time method is too implied to evaluate the PCUs. The period that elapses between the time that a vehicle’s front end enters a conflict region and the time that the vehicle’s rear end exits the same conflict region is known as occupancy time (Mohan and Chandra 2018). PCU values for a wide range of vehicle types that are frequently encountered on India’s interurban multilane highways at varied Level of Service (LOS) are provided in this study. VISSIM is a traffic model that simulates traffic conditions that are difficult to observe. Numerous factors in VISSIM have been tuned to simulate traffic flows in a mixed traffic condition. The software is then used to calculate speed-volume correlations for standard cars in relation to the stream’s various vehicle categories. Following that, the calibrated VISSIM model was used to determine the PCU values of different categories of vehicles (Mehar et al. 2014). A motorcycle displacement unit (MEU) was computed by the researchers to make findings relatively easy in areas where motorcycles are prevalent, such as Hanoi in Vietnam and many other Asian cities. It is the number of motorbikes that can be carried by a single vehicle of a specified kind at a given speed. The dynamic properties of moving vehicles are taken into account when creating the MEU for each type of vehicle that will be used in the study. Such parameters reveal the relationship between the subject vehicle’s speed and the surrounding motorcycles’ occupied space (Cao and Sano 2012). Moreover, with roundabouts being a significant unsignalised intersection, the PCU calculation incorporates various approaches, such as headway, flow rate in circulatory pathways, and simulation methods, to cater for providing reliable PCU values. Indian sites such as Chandigarh, Noida City traffic flow are taken for estimating the PCU

210

A. K. Munshi and A. K. Patnaik

at rotary intersection and hence PCU is expressed as the ratio of a subject vehicle’s mean lagging headway to the mean lagging headway of a passenger vehicle. Here, PCU was evaluated in conjunction with the vehicle’s width as an additional factor (Ahmad and Rastogi 2017). PCUs were analyzed at a circular crossroads in Jaipur and Trivandrum based on average time occupancy, projected standard vehicle area, and vehicle type. The relationship between entering flow and peripheral flow has been presented using data from the observed time-period during which queue formation occurred in the approach lane (Sonu et al. 2016). The traffic flow rate of heavy vehicles was taken into study while maneuvering the rotary intersection at Ontario, Wisconsin. The PCE was designed to minimize the variance in entrance capacity between cars and trucks. The study revised the critical and follow-up headways in the model to account for the differences in driver gap acceptability between cars and heavy vehicles. When it comes to driver gap acceptability, the researchers revised the critical and follow-up headways to account for the variation observed. The model yielded more accurate capacity forecasts when a different weightage for critical and follow-up headways was used (i.e., the PCE’s default value) (Lee 2015). AIMSUN was used to calculate the PCE and investigate the influence of heavy trucks on the parameters of Italy’s turbo-roundabouts for large trucks lane was utilized to examine the impact of vehicle weight and circulating flow percentages on PCE in a turboroundabout. When a considerable number of heavy trucks were present in the traffic stream, a greater PCE effect was seen (Giuffrè et al. 2016). The aforementioned studies have clearly indicated that the PCU is the prime factor in capacity estimation. The vehicle’s speed and physical area are key attributes in evaluating the PCU. Variables like headway and occupancy time also influenced the PCU equations. The lack of lane discipline at unsignalized intersections is a significant issue in this regard. The inclusion of both the static and dynamic components of the vehicle in conjunction to evaluate PCU is not mentioned explicitly in these literatures. Slow-moving passenger vehicles and commercial pick-up vans haven’t been conclusively linked to an increase in congestion. Therefore, a study has attempted to develop a PCU Model by taking into consideration of speed, effective area and lagging headway of vehicles under mixed traffic conditions.

3 Data Collection and Extraction Data were obtained at prominent unsignalized intersections in Jharkhand’s capital city. These are essentially three-legged intersections used to merge minor lane traffic into a major lane. The selected intersections satisfy the following criteria: (1) It is a purely T- intersection (i.e., mutually perpendicular). (2) The sites are devoid of traffic operations. (3) It is an at-grade intersection. (4) There are no bus stops or other physical barriers at the crossing that might hamper traffic movement. Four prominent unsignalised junctions in Ranchi City were identified based on the foregoing characteristics. The sites are selected based on the relevance of the primary routes traveled by commuters. The three sites were on the hub of the city while the last one is 0on the

Development of PCU Model for Unsignalised Intersection: A Case …

211

outskirts. RIMS Medical Chowk (S1), Morabadi Chowk (S2), and Hari Om Tower Chowk (S3) were chosen as locations within the city, while BIT Mesra More (S4) was chosen as a location outside the city. The sites were experiencing immense traffic flow due to the importance of the routes. The sites include an institutional area (S4) along the side of National Highway 33 where a minor lane merges with the highway, (S2) a sports complex with stadiums adjacent to government residential quarters, (S3) a commercial area adjacent to the residential complex on the minor lanes, and (S1), a medical institution on the minor lane merging with the major lane comprised of several hospitals and clinics. The locations of these sites are clearly represented (see Fig. 1). Inventory data provides the geometric details of the site and is collected using measuring tape. The traffic flow was recorded using a high-definition video camera. The camera was placed at a sufficiently high position, often a high-rise structure on the roadside, to get a view of the intersection that was acceptable. Data were acquired for this study when traffic resumes normal with the implementation of post-covid regulations in February and March of 2022. The videos were recorded in natural light on a normal weekday. The recordings were made for two hours while there was continuous waiting, and the traffic volume was computed using a maximum flow of 15 min at intervals. The commercial and institutional areas were recorded from 9 a.m. to 11 a.m., the hospital areas from 12 p.m. to 2 p.m., and the residential and sports complex areas from 5 p.m. to 7 p.m. The afternoon timings seemed to be rush hours for hospital as visitors are allowed to meet their patients and even due to lunch hours. Widescreen monitors were used to playback a captured video and retrieve data. There were eight distinct classifications for all of the vehicles, each based on their physical size and operating capabilities. This classification includes motorized twowheelers (2W), three-wheelers (3W), slow-moving three-wheelers (SM3W), standard cars (SC), large cars (BC), commercial vans (CV), heavy vehicles (HV) and Bicycles (BS). The sources for the physical size of vehicles that were utilized in earlier

(a) Ranchi city on India Map Fig. 1 Location and sites for collected data

(b) Data collection sites in Ranchi

212

A. K. Munshi and A. K. Patnaik

PCU studies (Chandra and Kumar 2003; IRC 003: Dimensions and Weights of Road Design Vehicles). Traffic conditions and vehicle dimensions have evolved significantly during the past two decades. Thus, the physical sizes of the vehicles employed in this study were obtained from manufacturer specifications for the predominant vehicles in their respective vehicle categories (CNG Auto Rickshaw | Apé Auto HT DX | Best price & mileage | Piaggio; Green Pilot E Rickshaw; Maruti Suzuki Wagon R; New Toyota Fortuner | Toyota Legender; Tata Motors Buses| CityRide SKL 36+A+D LP 410/36 Overview | Tata Motors Buses; TVS Apache RTR 160 4V: Price, Mileage, Images, Colours, Specifications; Rage 5.0). This enhances the study’s current relevance and scope. The specific dimensions of the vehicles taken into consideration in the study are detailed in Table 1. The traffic volumes at each of the sites were shown in Table 2. The following data depicts that SM3W vehicles are having significant numbers in urban traffic than highways. Table 1 Physical and effective areas of vehicles Classification of vehicles

Vehicles included

Length (m)

Width (m)

Physical area (m2 )

Effective area (m2 )

Small Cars (SC)

Hatchback Cars

3.65

1.62

5.91

9.56

Two-wheelers (2W)

Motorcycle

2.03

0.79

1.60

3.13

Slow-moving three -wheelers (SM3W)

E-rickshaw

2.8

1.00

2.80

4.48

Three-wheelers (3W)

CNG Auto

2.93

1.48

4.34

7.27

Commercial Vans (CV)

Pickups, towing vehicles

3.79

1.50

5.69

9.48

Big Cars (BC)

Sedan, SUVs

4.79

1.85

8.86

13.65

Heavy vehicles (HV)

Bus, Truck

7.19

2.34

16.82

24.01

Bicycle (BS)

Bicycle

1.9

0.45

0.86

1.62

Table 2 Classified vehicular volume counts (vehicles/hour) Sites

2W

SC

BC

3W

SM3V

S1

2587

530

772

779

463

HV

BS

25

60

CV

TOTAL

24

5240

S2

2425

480

376

204

591

1

300

38

4415

S3

2155

578

324

205

578

18

199

13

4070

S4

1916

410

375

200

77

143

168

73

3362

Development of PCU Model for Unsignalised Intersection: A Case …

213

4 Methodology This paper introduces PCU estimation of vehicles at priority movements at unsignalized intersection. The development of the PCU for each vehicle type incorporates both static and dynamic vehicle variables. These factors correlate between the speed, clearance, effective area, and time headway of the subject vehicle and standard cars. Considering all such factors, the fundamental proposed formula for PCU conversion for vehicle type ‘i’ is given as equation (1). PCUi =

Ai ×Vc Hi × Ac × Vi Hc

(1)

where, PCUi : Passenger car unit of vehicle type i Vc : Speed of standard car (kmph) Vi : Speed of vehicle type i (kmph) Ac : Effective area of standard car (m2 ) Ai : Effective area of vehicle type i (m2 ) Hi : Mean lagging headway of vehicle type ‘i’ (s) Hc : Mean lagging headway of standard car (s) In heterogeneous traffic conditions when vehicles do not usually adhere to lane discipline, the effective area is a more accurate representation of road space than physical dimensions. The effective area (Ai) of the subject vehicle type ‘i’ refers to the space occupied by the vehicle under normal traffic conditions, which includes the clearance area in addition to the vehicle’s physical dimensions. It depends on the geometry of lane, speed of vehicle, type of vehicle, traffic condition and driver characteristics. Thus, the effective area is given as in equation (2). The effective area relies mainly on the size of the vehicle as well as maneuverability. Hence, large vehicles like buses and trucks impact the traffic flow more than standard cars, which results in higher PCU values. Ai = L i × (Wi + Ci )

(2)

where, L i is length, W i is the width and C i is the lateral clearance on both sides (C ir and C il are right and left clearances of the vehicle respectively) of the vehicle type i. Lateral clearance between the vehicles is the distance around the subject vehicle that is required for its safe maneuvering (illustrated in Fig. 2). Accordingly, clearance in mixed traffic is dependent on various factors, including the type of vehicle, its speed, the driver’s characteristics, the proximity of other vehicles, and the traffic condition. At unsignalized crossings, the clearance values for each vehicle type are

214

A. K. Munshi and A. K. Patnaik

observed, and the mean lateral clearance values for each vehicle are adopted. As vehicle maneuverability depends on vehicle size, larger cars require more clearance, which was evident in the video data. The adopted values for larger vehicles like standard cars, commercial vans, heavy vehicles, big cars and three-wheelers are considered to be 1 m whereas the values for bicycle, Slow- moving three –wheelers and two-wheelers are 0.4 m, 0.6 m and 0.75 m respectively. According to the findings, slow-moving three-wheelers, i.e., E-rickshaw, maintain less clearance than motorized two-wheelers. These values are only significant for normal traffic flow situations (Fig. 3), whereas clearance values for congested (jammed) conditions are extremely low (Fig. 4) and can be approximated to zero i.e., C i = 0. As a result, for the purpose of calculating the PCU in a crowded traffic situation, the effective area equals the actual area of vehicle. Fig. 2 Conceptual diagram for determining effective area

Fig. 3 Lateral clearances between vehicles

Development of PCU Model for Unsignalised Intersection: A Case …

215

Fig. 4 Depicting traffic Jam condition

The radar gun has been used to measure approach speed during normal flow conditions whereas accelerated speed (after few seconds of travel) is taken into consideration in congested traffic scenarios using the spot speed approach. The speed of each type of vehicle varies a lot, so average values are used to reduce the fluctuations and get meaningful results. Headway is used to determine both the lane’s capacity and the longitudinal gaps between the vehicles. The length of the vehicle and the distance between vehicles are included in the lagging headway. It is calculated by measuring the time elapsed between the leading vehicle’s rear bumper and the trailing vehicle’s rear bumper. It is quite difficult to ascertain the exact lagged headways and clearance of each type of vehicle in the identical locations in the sites. The video recordings were utilized frame by frame to determine the durations and distances (Fig. 5). Thus, the relative error in measurement is not indicated in this work.

(a) Frame 1 Fig. 5 Lagging headway of a bus

(b) Frame 2

216

A. K. Munshi and A. K. Patnaik

5 Results and Discussions A continuous queuing condition was used to estimate peak hour traffic volumes at each of the sites. The city traffic outnumbers the traffic on the outskirts. Large trucks were prohibited from entering the city at any time during the day, and hence buses were the primary heavy vehicles on the highways. The data depicts the surprising results in vehicular counts. According to the analysis of urban and highway traffic, the proportion of slow-moving vehicles within city such as BS, SM3W, and 3W is tremendously high, accounting for around 25% of total traffic volume. The recommended PCU values mentioned in Table 3 were compared to the study’s proposed methodology (Indian Highway Capacity Manual (Indo-HCM) | CSIR—Central Road Research Institute 2023). PCU values were determined for each of the four locations under both normal and heavy traffic flow scenarios is shown in Table 4. The obtained PCU values were slightly higher than recommended values. The findings are based on current vehicle size as well as the inclusion of motorized passenger vehicles (SM3W). PCU values for smaller vehicles such as the 2W, 3W, SM3W, and BS were usually higher in normal traffic conditions than in crowded conditions. The findings prevail that during normal scenario the small vehicles tend to have an adequate space while maneuvering and significantly maintain their speeds at intersections to avoid any collision. In crowded traffic within the city, smaller vehicles have practically minimal clearance, reducing their effective area significantly. They are often able to accelerate faster than standard cars and larger vehicles, easily clearing the congestion zone and creating standstill situations for larger vehicles. This obviously has an effect on their PCU, decreasing the values and, as a result, improving lane capacity in congested areas.

Table 3 Suggested passenger car units as per Indo-HCM

Vehicle type

PCUs

Motorized two-wheelers

0.48 Through movement on major 0.34 all other movements

Auto rickshaw

0.98

Small/standard cars

1.00

Big cars and vans

1.29

Buses

2.29

Cycles

0.42

Development of PCU Model for Unsignalised Intersection: A Case …

217

Table 4 Estimated results of PCU values Traffic flow pattern

Normal

Locations

S1

S2

Congested S3

S4

Vehicle type

S1

S2

S3

S4

Suggested values

Suggested values

2W

0.41

0.37

0.33

0.34

0.36

0.36

0.34

0.27

0.29

0.32

3W

1.10

1.07

1.02

1.05

1.06

1.01

0.97

0.99

1.00

0.99

SM3W

0.79

0.82

0.74

0.71

0.76

0.84

0.79

0.75

0.81

0.80

SC

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

BC

1.70

1.59

1.69

1.56

1.63

1.70

1.64

1.79

1.66

1.70

CV

1.23

1.12

1.20

1.17

1.18

1.31

1.23

1.22

1.19

1.24

HV

3.78

3.83

3.81

3.45

3.72

3.88

3.79

3.93

3.45

3.76

BS

0.49

0.47

0.43

0.42

0.45

0.34

0.29

0.25

0.28

0.29

6 Conclusion A vehicle’s PCU is a comprehensive value that is impacted by a diverse set of variables that influence a vehicle’s conduct in a traffic flow. The formulation process also has an influence on the overall result. HCM method provides the values as per the ideal circumstances; however, the subject vehicle and the nearby vehicles in the traffic flows may not always be of the same vehicle class. The perfect situation is not always realistically achievable in the sites. The proposed methodology incorporates both static and dynamic characteristics, such as the vehicle’s effective area, speed, and lagging headway. Consequently, this methodology is suggested for diverse city traffic and appears to be reliable in tier II cities, where commuters even prefer non-motorized vehicles and slow-moving three-wheelers for local movements. In accordance with the current vehicle dimensions, the estimated PCU values indicate that each heavy vehicle, commercial pick-up van, and big cars can accommodate 3.76, 1.24, and 1.70 standard vehicles in a urban crowded traffic lane, respectively. In cities, the usage of E-rickshaw and two-wheelers enhances lane capacity by having lower PCU values of 0.32 and 0.80, respectively. Three-wheelers powered by compressed natural gas (CNG) have PCU values that are nearly identical to those of standard cars. Bicycles with a PCU of 0.29, the sole non-motorized vehicle on the city road nowadays, are still preferred by commuters in tier II cities. Indeed, with the advent of electric two-wheelers, it is possible to enhance the PCU of a lane due to the fact that their dimensions are to be slimmed down, and this makes the transportation stream more cognizant of the need for future research in PCU estimation.

218

A. K. Munshi and A. K. Patnaik

References Ahmad A, Rastogi R (2017) Regression model for entry capacity of a roundabout under mixed traffic condition—an Indian case study. Transp Lett 9(5):243–257. https://doi.org/10.1080/194 27867.2016.1203603 Cao NY, Sano K (2012) Estimating capacity and motorcycle equivalent units on urban roads in Hanoi, Vietnam. J Transp Eng 138(6):776–785. https://doi.org/10.1061/(ASCE)TE.1943-5436. 0000382 Chandra S, Kumar U (2003) Effect of lane width on capacity under mixed traffic conditions in India. J Transp Eng 129(2):155–160. https://doi.org/10.1061/(ASCE)0733-947X(2003)129:2(155) CNG Auto Rickshaw | Apé Auto HT DX | Best price & mileage | Piaggio. https://piaggio-cv.co.in/ ape-auto-ht/. Accessed 09 Feb 2023 Giuffrè O, Grana A, Marino S, Galatioto F (2016) Passenger car equivalent for heavy vehicles crossing turbo-roundabouts. Transp Res Procedia 14:4190–4199. https://doi.org/10.1016/j.trpro. 2016.05.390 Green Pilot E Rickshaw. indiamart.com. https://www.indiamart.com/proddetail/green-pilot-e-ric kshaw-23098013091.html. Accessed 09 Feb 2023 Indian Highway Capacity Manual (Indo-HCM) | CSIR - Central Road Research Institute. https:// crridom.gov.in/indian-highway-capacity-manual. accessed 09 Feb 2023 IRC 003: dimensions and weights of road design vehicles Lee C (2015) Developing passenger-car equivalents for heavy vehicles in entry flow at roundabouts. J Transp Eng 141(8):04015013. https://doi.org/10.1061/(ASCE)TE.1943-5436.0000775 Maruti Suzuki Wagon R: Wagon R features, specifications, colours and interior. MarutiSuzuki. https://www.marutisuzuki.com/wagonr. Accessed 09 Feb 2023 Mehar A, Chandra S, Velmurugan S (2014) Passenger car units at different levels of service for capacity analysis of multilane interurban highways in India. J Transp Eng 140(1):81–88. https:/ /doi.org/10.1061/(ASCE)TE.1943-5436.0000615 Mohan M, Chandra S (2018) Three methods of PCU estimation at unsignalized intersections. Transp Lett 10(2):68–74. https://doi.org/10.1080/19427867.2016.1190883 New Toyota Fortuner | Toyota Legender. https://www3.toyotabharat.com/showroom/fortuner/#spe cification. Accessed 09 Feb 2023 Rage 5.0. Tata magic specifications—engine, brakes, performance, seats, etc. Tata Magic. https:/ /www.magic.tatamotors.com/passenger-trucks/tata-magic/specification/tata-magic-specifica tion.aspx. Accessed 09 Feb 2023 Sonu M, Dhamaniya A, Arkatkar S, Joshi G (2016) Time occupancy as measure of PCU at four legged roundabouts. Transp Lett 1–12. https://doi.org/10.1080/19427867.2016.1154685 Tata Motors Buses| CityRide SKL 36+A+D LP 410/36 Overview | Tata Motors Buses. Tata Motors Buses | Range of BS IV Buses from Tata Motors. 26 May 2020. https://www.buses.tatamo tors.com/products/brands/cityride/cityride-skl-36-a-d-lp-410-36-school-bus/. Accessed 09 Feb 2023 TVS Apache RTR 160 4V: price, mileage, images, colours, specifications. https://www.tvsmotor. com/tvs-apache/apache-rtr-160-4v. Accessed 09 Feb 2023

Modeling Pedestrian Waiting Time Delay at Signalized Midblock Crosswalks Under Non Uniform Arrivals and Non Compliance Behavior Sandeep Manthirikul and Udit Jain

Abstract The current study was conducted in two stages: the first stage proposed a new approach for assessing field pedestrian waiting delay, and the second stage proposed a modified pedestrian waiting delay model for signalized midblock crosswalks under mixed traffic situations. The new approach of field pedestrian waiting delay was measured by plotting the pedestrian queue length against the cycle time. The area under the curve was calculated using Simpson’s 1/3rd rule, which yielded the total pedestrian delay seconds. The average waiting delay is obtained by dividing the entire delay by the pedestrian flow of the corresponding cycle phase. The proposed method was validated using individual pedestrian waiting delays with MAPE and RMSE values of 6% and 3.48 respectively. In the second stage, a modified pedestrian waiting delay model was proposed for signalized midblock crosswalks under mixed traffic conditions. Initially, all the possible combinations of coefficients (Non uniform arrival rate and non compliance coefficients) suggested by previous studies were examined against the field delay and the best was reported in the current study. The combination of non uniform arrival rate coefficient by Li et al. and non-compliance behavior coefficient by Marisamynathan and Vedagiri model could estimate pedestrian waiting delay values close to the field delay at the selected study areas. The comparison indices like MAPE, RMSE, Pearson’s R and R2 values were calculated for the proposed model. The RMSE and MAPE values were found to be 0.35 and 0.09 respectively whereas the Pearson R and R2 were found to be more than 0.9 and 0.8 respectively. Keywords Pedestrian waiting delay · Signalized Midblock crosswalks · Simpson’s 1/3rd rule · Non uniform arrival rate

S. Manthirikul (B) · U. Jain Visvesvaraya National Institute of Technology, Nagpur 440010, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_17

219

220

S. Manthirikul and U. Jain

1 Introduction The Midblock pedestrian crossings are the crossings provided for pedestrians at midblock sections. Most of these crossings are controlled by traffic signals or grade separated pedestrian crossing facilities for safety reasons (Manthirikul et al. 2022a). The cycle length of these crossings accommodates a significant portion of the vehicle green phase, resulting in longer pedestrian delays. Most of the time, this results in a pedestrian violation at a signal. Almost half of the pedestrians could not tolerate waiting times of more than 48 s, and the green phase was unable to meet the demand of the elderly (Jain and Rastogi 2018; Sheng et al. 2012). As a result, the cycle length/phase length for such points would be satisfactory if they were designed based on microscopic characteristics such as delay rather than macroscopic characteristics such as flow. (Jain et al. 2019). However, the literature on evaluating pedestrian delay was limited to signalized intersections. Signalized midblock crosswalks where the Past researchers had widely explored pedestrian delay at signalized intersections. In the early stages, pedestrian models were developed for ideal conditions like uniform arrival rates and compliance behavior of pedestrians. HCM, Webster and Austroads models follow the same assumptions (Abley et al. 2015; HCM 2010; Webster 1958). But pedestrians arrive randomly in bunches at waiting areas and cross the section in platoons. The assumptions of uniform arrival rate and compliance behavior in estimating the pedestrian delay will not yield accurate results in the case of mixed traffic conditions (Nagraj and Vedagiri 2013; Marisamynathan and Vedagiri 2018). Non uniform arrival rates and non compliance behavior of pedestrians were jointly considered in the estimation of pedestrian delay by Li et al. study (Li et al. 2005). Coefficient of non uniform arrival rate (kNU ) was developed based on the various pedestrian arrival patterns and coefficient of non compliance behavior (k) was calculated as the slope of the decreasing line during the pedestrian red phase. But the model was developed for low pedestrian flow and is as follows: ) ( C n t − n g k(C − G + 0.69A)2 D = DG + n t (C − G) 2C

(1)

where nt is the total number of pedestrians and ng is the number of pedestrians arriving during the green phase. Kruszyna et al. and Chen et al. are other studies which developed a delay model for non uniform arrival patterns but failed to consider the pedestrian crossing behavior (Kruszyna et al. 2006; Kuan-Min 2010). Nagraj & Vedagiri modified the webster delay model and incorporated coefficients of non uniform arrival patterns and compliance behavior of pedestrians (Nagraj and Vedagiri 2013) which were similar to the Li et al. model. Later Marisamynathan & Vedagiri (Marisamynathan and Vedagiri 2018) and proposed two models based on the compliance nature of pedestrians and suggested new coefficients for non uniform arrival rate and non compliance behavior of pedestrians which are as follows:

Modeling Pedestrian Waiting Time Delay at Signalized Midblock …

D PC =

α1 (C − G + α2 R)2 + (γ − 1)t I + 11.189P − 1.0713 2C α1 = 0.002 V + 0.734

221

(2) (3)

where DPC = Delay model based on compliance behavior; α1 = Correction factor for non uniform arrival (See Eq. 10); α2 = % of the pedestrian crossing during nongreen phase. Recently Manthirikul et al. proposed a pedestrian delay model based on the VISSIM simulation delay results for different cycle lengths and pedestrian volume combinations (Manthirikul et al. 2022b). As the model was simulated in the VISSIM, pedestrian arrival and compliance nature of the pedestrians remains ideal in nature. Hence it is evident that signalized midblock crosswalks were completely overlooked by past researchers in estimating pedestrian delay. The present study aims to modify the existing pedestrian waiting delay model complying with the conditions of signalized midblock crosswalks. In addition, a new approach of measuring field delay was suggested in the current study.

2 Methodology and Data Collection The current study was carried out in two stages: in the first, a new method of quantifying field pedestrian waiting delay was proposed by mapping the length of the pedestrian queue against the cycle time. Every 5 s, the queue length was retrieved and plotted against the cycle time. Simpson’s 1/3rd rule was used to get the total pedestrian delay seconds from the area under the curve. The average waiting delay is calculated by dividing the total delay by the pedestrian flow of the respective cycle phase. The proposed method was validated using the individual pedestrian waiting delay for each location. It is expected that the proposed approach will reduce data extraction time and will be valuable for future studies to validate the data. In the second stage, a modified pedestrian waiting delay model was proposed for signalized midblock crosswalks under mixed traffic conditions addressing the non uniform arrival rate and non compliance behavior of the pedestrians. Based on the existing coefficients of Non uniform arrival rate and non compliance which were suggested by previous studies two scenarios were developed. The first scenario i.e., Scenario A is the combination of non uniform arrival rate by Li et al. and non compliance coefficient by Marisamynathan and Vedagiri. The second scenario i.e., Scenario B is reciprocal to the previous scenario. The waiting delay of these two scenarios was compared against the field waiting delay and the best was reported. In order to prove the elegance of the modified model, the same was compared with the existing waiting delay models and was checked against the field waiting delay. Six study areas were chosen from the Hyderabad (India). Study areas M-1 (Fig. 1a), M-2 (Fig. 1b), M-3 (Fig. 1c) & M-6 (Fig. 1f) are located on National Highway 44 (Hyderabad - Mumbai Highway), study area M-4 (Fig. 1d) is located on National Highway 65 (Hyderabad – Vizayawada Highway) and study area M-5

222

S. Manthirikul and U. Jain

(Fig. 1e) is located on of the arterial roads of the same city. The study areas M-2, M-3 & M-5 were considered as compliance behavior locations since the violation at these locations is minimal. The remaining study areas M-1, M-4 & M-6 had encountered frequent violations by both the road users and hence considered in modeling delay based on non compliance behavior. Data at the study areas was collected by using videography technique, for morning peak (8:00 a.m. to 12:00 a.m.) and evening peak (4:00 p.m. to 9:00 p.m.). Table 1 presents the details of the study areas.

Fig. 1 Study areas

Table 1 Details of study area Midblock Midblock identity name

Land Use type

No. of Cycle Pedestrian red Amber Pedestrian lanes length time (Sec) (Sec) flow (ped/ (Sec) hr)

M-1

Mixed

8

Nizampet X road

125

90

5

1013

M-2

Moosapet

Commercial 6

105

80

5

963

M-3

Vivek nagar

Residential

120

90

5

822

8

M-4

Dilsuknagar

Commercial 6

155

120

5

750

M-5

Mettuguda

Mixed

8

155

120

5

663

M-6

Vivekananda Residential nagar

7

125

90

3

363

Modeling Pedestrian Waiting Time Delay at Signalized Midblock …

223

3 Pedestrian Field Waiting Delay Pedestrians are typically subjected to two sorts of delays. The first is a delay in crossing, and the second is a delay in waiting time. As previously indicated, video data was used to assess field crossing delay. To compute the field waiting period, however, each pedestrian must be tracked from arrival to departure, which requires a tremendous deal of attention and time to extract the data from the video. Previous research validated the proposed models in a similar manner, comparing delay values to an individual pedestrian waiting delays. The HCM (2010) suggested a method for calculating total vehicle field delay at a signalized intersection (HCM 2010). The lengths of vehicle queues were plotted against cycle time, with the area under the displayed curve representing total vehicle delay. The average delay of an individual vehicle was calculated by dividing the overall delay by the vehicular volume of the corresponding cycle phase. However, the literature lacks such integrated approaches for calculating the field pedestrian waiting delay. In the current paper, the same method was used to calculate the pedestrian waiting period. Pedestrian arrivals and queue length (QL ) at waiting areas were recorded for each cycle. Queue length was measured in frames for high accuracy, and the recorded video was captured at 30 frames per second. According to HCM 2010, QL is the total of vehicles that entered the marked trap area and the number of vehicles that were previously present in that trap, less the number of vehicles that passed the reference line during that interval. The same formula was used to calculate the length of the pedestrian queue every 5 s. According to the formula, the number of pedestrians accumulated at the end of every 5 s (excluding violators but including new pedestrians) was counted and plotted against the cycle duration. This eliminates the need to track individual pedestrians’ arrival and exit times in order to calculate waiting time. The area under the plotted curve was measured using the Simpson 1/3rd method and the equation is similar to Eq. 4. The equation yields total pedestrian delays, and the average pedestrian waiting time was calculated by dividing it by the volume of the corresponding cycle phase. ∮c f (Q L )dq =

h → [(q0 + qn ) + 4(q1 + q2 + . . . . . . + q−AB(n−1) ) 3

0

+ 2(q2 + q4 + . . . + q(n−2) )] h =

(4)

C − 0 n

where C-0 = time difference between start and end time of queue observation; q0 , q1 …qn = queue lengths in 5 s intervals; QL = queue length; n = number of intervals. The proposed method for measuring field waiting time was validated using the actual average waiting delay during the cycle phase. Each pedestrian’s individual waiting time was computed by tracing their arrival and departure times. The average cycle phase waiting delay was then compared to the proposed approach using MAPE (Eq. 5)

224

S. Manthirikul and U. Jain

and RMSE (Eq. 6). The MAPE and RMSE values between the two approaches are stated to be 6% and 3.48, respectively. According to the Lewis scale of interpretation, a MAPE value of less than 10% is considered accurate, 11–20% as good, 21–50% as reasonable, and more than 50% as inaccurate (Lawrence and Geurts 2006). As the projected values were less than 10%, we may conclude that the new technique accurately calculates pedestrian waiting time. Furthermore, the proposed approach’s delay values were plotted against actual waiting time delay values to determine the measure of effectiveness R2 value and Pearson’s r values (See Fig. 2). As illustrated in the graph, R2 and Pearson r were more than 0.8 and 0.9, respectively. As a result, it can be claimed that the proposed method accurately measures the field waiting delay and may be used to validate the proposed models. It will be a useful tool for future academics who want to investigate pedestrian delays in depth and design new pedestrian delay models. 1 ∑ (Ft − At ) n At ┌ | n |∑ (Ft − At )2 RMSE = √ n i=1 M AP E =

Fig. 2 Validation of pedestrian field delay

(5)

(6)

Modeling Pedestrian Waiting Time Delay at Signalized Midblock …

225

4 Pedestrian Waiting Delay Model As stated earlier based on the existing models two scenarios were developed. The non uniform coefficient (Knu ) by Li et al. and non compliance coefficient (α2 ) by Marisamynathan and Vedagiri model were grouped in the first scenario whereas the non uniform coefficient (α1 ) Marisamynathan and Vedagiri model and non compliance coefficient (K) by Li et al. model were grouped in the second scenario. Along with these two scenarios, waiting delay by Li et al. and Marisamynathan models were compared with the field waiting delay. The summary of the pedestrian delay models along with the comparison indices are summarized in Table 2. At compliance behavior locations, Li et al. model yielded better results, but drastically underestimated the field delay at non compliance behavior areas. This is because the non compliance coefficient (K) eliminates violators while calculating the delay. The pedestrian who came during the initial stage of the red phase and violated the signal at the 11th hour of the same phase would be eliminated from the study. Under the same scenario, the field waiting delay is generally equal to the red time. As a result, a significant divergence between the field delay and the Li et al. model was identified in non compliance behavior areas. Whereas at the compliance behavior location, as there are no violations, the K coefficient is zero, and Knc is the sole coefficient that contributes to the estimation of delay. As a result, the Li et al. model produced better results. These favorable findings also explain the accuracy of the Knc factor in estimating waiting delay for non uniform arrival rates at signalized midblock crosswalks. Marisamynathan and Vedagiri’s model overestimated the field delay of signalized midblock crosswalks since the model was developed and validated for low to medium range pedestrian flow. If the hourly pedestrian flow during the non-green phase is larger than 390, the correction factor of non uniform arrival Table 2 Summary of pedestrian waiting delay model Study area

Field

Li et al

M-2c

36.69

37.73

M-3c

33.02

M-5c

37.68

RMSE

–

MAPE

–

Marisamynathan & Vedagiri

Scenario A (Knu & α2 )

Scenario B (α1 & K)

77.617

37.73

83.855

31.78

63.701

31.78

68.273

37.33

70.665

37.33

74.503

3.264

17.426

3.264

24.081

0.065

0.316

0.065

0.464

0.731

0.035

0.731

R2

–

M-1nc

62.05

15.85

122.585

63.97

32.916

M-4nc

56.74

35.27

128.782

56.35

74.453

M-6nc

48.96

27.42

93.599

50.07

46.129

RMSE

–

32.24

59.438

5.625

23.751

MAPE

–

0.484

1.000

0.128

0.355

R2

–

0.130

0.292

0.912

0.331

0.421

226

S. Manthirikul and U. Jain

rate (α1 ) is greater than 1.5, and the correction factor of non compliance behavior (α2 ) is less than 0.5 if the violation is greater than 50%. In such cases, the waiting delay yields a value higher than the cycle time. However, in the current study, pedestrian movement is high at chosen signalized midblock crosswalks. It can be shown that both models reported more than 30% error and R2 values less than 0.3, indicating inaccuracy in forecasting the pedestrian delay at signalized midblock crosswalks. Delay by Scenario B (α1 & K) completely overestimated the field delay since the α1 is reporting more than 1.5 in almost all the cycle phases. The RMSE, MAPE and R2 values of the same scenario are reported as 23.751, 0.355 and 0.331 respectively. Hence it can be concluded that scenario B is inaccurate in estimating the pedestrian waiting delay at signalized midblock crosswalks. But scenario A which is the combination of the Knc & α2 yielded close values to the field delay values. However on case of compliance behavior, the model remains the same as the Li et al. model since there is no violation eventually α2 becomes zero. As stated earlier Li et al. model are good at estimating the waiting delay at signalized midblock crosswalks but in case of only compliance behavior locations. The non compliance behavior coefficient KC is affecting the accuracy of the waiting delay since it is not considering the violators. So the same is replaced with the coefficient proposed by Marisamynathan & Vedagiri model which includes violators and non violators. Hence the combination of these two factors is leading to the accurate estimation of the pedestrian waiting delay at signalized midblock crosswalks. The RMSE, MAPE and R2 values of the same scenario are reported as 5.625, 0.128 and 0.912 respectively. Since the reported MAPE value of scenario A is 12%, the same combination of non uniform arrival rate (Knc ) and non compliance behavior (α2 ) would be robust in estimating the waiting delay of pedestrians at signalized midblock crosswalks. In addition, waiting delay by all the models is plotted against field delay to check for the similarity between field delay and modeled delay of all samples (See Fig. 3). The 45° line represents the similarity line, and the points close to the line indicates that the delay value of that particular model is close to the field delay. It can be observed the graph that, except for scenario A, the delay values by all the models are scattered on the widespread area of the plot and away from the similarity line. A few points of Li et al. model which were scattered close to the similarity line represents the delay in case of compliance behavior. Also, few points were unseen in the graph since the same points were covered by the points of scenario A (as both delay values were the same in case of compliance behavior). Hence based on the comparison results and statistical tests, it can be concluded that waiting delay by Scenario A (combination of Knu and α2 ) is robust and accurate in estimating the waiting delay of pedestrians at signalized midblock crosswalks.

Modeling Pedestrian Waiting Time Delay at Signalized Midblock …

227

Fig. 3 Comparison of pedestrian waiting delay models

5 Conclusions Compared to a signalized intersection, waiting delay of pedestrians at signalized midblock crosswalks yields in a longer duration due to the longer pedestrian red phase, non uniform arrival rate and non compliance behavior. In the present study initially, a new approach of measuring the pedestrian field waiting delay was proposed. The approach suggested by HCM (2010) for measuring vehicle field delay was employed in the current study for measuring the field delay. Pedestrian queue length for every 5 s was plotted against the cycle phase and the area under the curve was measured using Simpson’s 1/3rd rule which gives the pedestrian total delay. The average delay was calculated by dividing the total delay with the pedestrian flow of that particular cycle phase. It is then validated with the individual pedestrian delay which was observed from the video data. The error between the field delay by individual data and by the proposed approach yielded an error of 6% and an R2 value of 0.8. The proposed approach of measuring field waiting delay would minimize the data extraction time as it avoids the tracing of individual pedestrians. Following the field pedestrian waiting delay, a new conventional model of pedestrian waiting delay was proposed for signalized midblock crosswalks. For signalized midblock crosswalks under mixed traffic conditions, a modified pedestrian waiting delay model was presented, which addressed the pedestrians’ non uniform arrival rate and non compliance behavior. Two scenarios were built based on the current non uniform arrival rate and non compliance coefficients proposed by previous research. Scenario A combines Li et al.’s non uniform arrival rate with Marisamynathan and

228

S. Manthirikul and U. Jain

Vedagiri’s non compliance coefficient. Scenario B is the inverse of Scenario A. The field waiting delay was compared to the waiting delay of these two situations, and the best one was chosen. Student t test and ANOVA test revealed that except for Scenario A, the mean of all the models have significant differences with the field waiting delay. The RMSE, MAPE and R2 values of the same scenario is reported as 5.625, 0.128 and 0.912 respectively. The contribution of the current research highlights the new approach of measuring field pedestrian waiting delay as well as the new pedestrian waiting delay model which addresses the non uniform arrival rate and non compliance behavior of pedestrians at signalized midblock crosswalks. The proposed new approach of measuring field waiting delay would be useful for future researchers and planners as it minimizes the laborious effort and time involved in the extraction process of pedestrian waiting delay. It is also evident that existing literature lacks in estimating pedestrian waiting delays addressing the non uniform arrival rate and non compliance nature of the pedestrians. Although Li et al. (2005) Nagraj & Vedagiri (2013) and Marisamynathan & Vedagiri (2018) proposed the new coefficients of non uniform arrival rates and non compliance behavior of pedestrians but the same were proposed and validated for signalized intersections. The pedestrian waiting delay by the above mentioned models is yielding either overestimation or underestimation of the field waiting delay at signalized midblock crosswalks. Whereas the proposed waiting delay by the present study is robust in estimating the waiting delay at signalized midblock crosswalks. The modified waiting delay model can be a useful tool for policy makers and planners to accurately assess the pedestrian waiting delay at signalized midblock crosswalks and redesigning the cycle phase based on the delay. Future research shall be focused on developing pedestrian crossing delay and LOS can be defined using collective waiting and crossing delay. Further new vehicle delay for signalized midblock crosswalks so that the delay of both the road users can be optimized for the efficient traffic signal design. Acknowledgements The authors thank Traffic Police department, Hyderabad, India for their assistance in collecting data and also thank Ministry of Human Resource Development, New Delhi, India and Visvesvaraya National Institute of Technology, Nagpur, Maharashtra, India for granting funding for the research.

References Abley S, Smith D, Rendall S (2015) Development of the Australasian pedestrian facility selection Tool. In Austroads Project NS1912, Level 9, 287 Elizabeth Street, Sydney NSW 2000 Australia HCM 2010: Highway Capacity Manual (2010) Fifth edition. Washington, DC: Transportation Research Board Jain U et al (2019) Revision of PV2 based pedestrian crossing warrants in India using clustering techniques. Transp Lett 11(5):241–249 Jain U, Rastogi R (2018) Evaluation at mid-block crossings under mixed traffic conditions. Institute of Transportation Engineers. ITE J 88(11)

Modeling Pedestrian Waiting Time Delay at Signalized Midblock …

229

Kruszyna M, Mackiewicz P, Szydlo A (2006) Influence of pedestrians’ entry process on pedestrian delays at signal-controlled crosswalks. J Transp Eng 132(11):855–861 Kuan-Min C et al. (2010) Towards the pedestrian delay estimation at intersections under vehicular platoon caused conflicts. Sci Res Essays, 5 Lawrence KD, Geurts MD (2006) Advances in business and management forecasting. Emerald Group Publishing Li Q et al (2005) Pedestrian delay estimation at signalized intersections in developing cities. Trans Res Part a: Policy Pract 39(1):61–73 Manthirikul S, Jain U, Amshala VT (2022a) A critical review of grade-separated pedestrian crossing facilities. J Transp Eng Part A Syst 148(10):03122003 Manthirikul S, Amshala VT, Jain U (2022b) Modeling vehicular and pedestrian delays at signalized midblock crosswalk under mixed traffic conditions. Transportation Letters, pp 1–14 Marisamynathan S, Vedagiri P (2018) A new approach to estimate pedestrian delay at signalized intersections. Transport 33(1):249–259 Nagraj R, Vedagiri P (2013) Modeling pedestrian delay and level of service at signalized intersection crosswalks under mixed traffic conditions. Transp Res Rec 2394(1):70–76 Sheng F, Ma Y-F, Lu J (2012) Research on pedestrian crossing characteristics of mid-block crosswalks controlled by push-button signal. In CICTP 2012: Multimodal Transportation Systems—Convenient, Safe, Cost-Effective, Efficient, pp 520-532 Webster FV (1958) Traffic signal settings

Assessment of Adherence Level to Helmet Usage on Varying Roads in Delhi Kanishk Singh, S. Velmurugan, and Nishit Patel

Abstract The level of helmet usage covering all types of motorised two-wheeler riders plying on 15 road sections spread over the National Capital Territory (NCT) of Delhi has been evaluated wherein the data collected on an expressway, arterial, subarterial and collector streets were analysed. This encompassed Delhi-Noida Freeway (DND) which falls under the urban expressway category, five numbers of arterial roads namely, Dr. Hedgewar Marg, Signature Bridge, Peeragarhi, Vikas Marg (at Nirman Vihar Intersection) and Noida Link Road, seven numbers of sub-arterial roads constituting Dr Zakir Hussain Marg, Rao Tula Ram Marg, Africa Avenue, Kasturba Gandhi Marg, Vandematram Marg, Kingway Camp and Bhishma Pitah Marg, two numbers of Collector Street namely, Greater Kailash and Dr M.C Davar Marg. Using the above data, Binary Logistic Regression Model was formulated in SPSS software. The output deduced from the above analysis revealed two possible outcomes namely, Helmet “Not worn or unstrapped” or “Worn and Strapped” covering the Driver, Pillion Rider or at times even the second Pillion Rider sits violating the Motor Vehicle Act on Indian roads. The results deduced for the varying functional classification of the roads revealed that the helmet “Worn and Strapped” behaviour in Driver, Passenger 1 and Passenger 2 is the highest on the expressway by clocking 98.4, 89.4 and 53.9% followed by Sub-arterial road users by measuring 77.95, 72.3 and 27.7%, Arterial Roads registering 73.0, 68.5 and 13.1%, and lastly Collector Road users registering 65.82, 44.3 and 3.2% respectively. Keywords Helmet · Two-wheeler · Road safety · BIGRS · Delhi

K. Singh (B) · S. Velmurugan CSIR-Central Road Research Institute, New Delhi, India e-mail: [email protected] N. Patel International Injury Research Unit JHSPH, Baltimore, USA © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_18

231

232

K. Singh et al.

1 Preamble Many of the motor vehicle related crashes lead to the fatal or serious injury type of crashes in developing economies like India. Some of the reasons that can be attributed for the above phenomenon to the poor adherence of certain pillars of road safety namely, non-usage of helmets, seatbelt usage as well as speeding aspects beyond the stipulated speed limits witnessed on Indian roads. To evaluate the impact of the same, Bloomberg Philanthropies Initiative for Global Road Safety (referred as BIGRS in this paper) programme had taken initiative drawing financial support from Bloomberg Philanthropies and the World Health Organisation (WHO) to evaluate the level of the aforesaid safety parameters for some of the Indian cities. Out of the above-mentioned safety pillars, a helmet is one of the important personal protective items which need to be worn by motorised two-wheeler riders as it accounts for 37% of the fatal crashes on Indian roads. The proper adherence of the same in terms of usage of standard helmets i.e. ISI certified helmets coupled with the wearing of helmet straps by the driver and the pillion rider is of paramount safety to enhance the safety of two-wheeler riders which is evaluated in this study. The level of using the above data, the Binary Logistic Regression Model was formulated using SPSS software. The output deduced from the above analysis revealed two possible outcomes namely, Helmet “Not worn or unstrapped” or “Worn and Strapped”.

2 Objectives and Scope of Study Considering the above, the objectives addressed in this paper is to conduct observational studies on the level of adherence on the aspect of understanding level of adherence of helmet usage by the riders and pillion riders on Delhi roads conforming to the protocols of the John Hopkins School of Public Health (hereinafter referred as JHSPH in this paper). Specifically, this observational study on helmet usage has been conducted using paper-based method covering fifteen road sections in Delhi on various types of roads classified based on functional category. The following section provides an overview of some of the associated studies done elsewhere.

3 Literature Review This study by Marisamynathan et al. (2020) focussed on the rise in helmet use after the government in Mumbai, India passed a rule requiring helmet use for two-wheeler drivers and passengers starting in April 2016. This study’s goals included identifying the variables that influence helmet use behaviour and creating a model for predicting motorcycle riders’ helmet usage patterns. Data was gathered “before” and “after” the

Assessment of Adherence Level to Helmet Usage on Varying Roads …

233

law requiring the use of helmets. To ascertain the motorcycle riders’ helmet usage habits, statistical tests, descriptive analysis, and a binary logistic regression model were created using the aforementioned data. Results showed that by accounting for land use patterns and survey data, the created model predicted helmet usage behaviour more accurately. In another similar study by Mathur et al. (2017) baseline data was collected to understand the pattern of use of helmets in all seven administrative divisional headquarters of Rajasthan state in 2015, The study’s goal was to comprehend the helmet usage trends across Rajasthan’s seven divisional headquarters’ highways, cities, and rural areas. Nearly 39.4% of the 1,17,000 plus two-wheeler drivers were found to be wearing helmets correctly. In the group mentioned above, about equally as many men (58.6%) and women (58.9%) did not wear helmets. A study by Dandona et al. (2006) reported on the availability of driver licences, Helmet use, Driver behaviour, and condition of vehicles for motorised two-wheeler drivers in Hyderabad city. They conducted interviews amongst motorcycle drivers above 16 years of age at petrol filling stations, nearly around 4183 Motorcycle Drivers participated in the interview. Using the above data, multiple logistic regression was conducted for two variable outcomes in the study. The study emanated revealed that 21.4% of drivers obtained driving licences, 69.8% of drivers reported no / very occasional use of helmets etc. It was referred to enforce the policy interventions by improving the driving licence system, mandatory use of helmet etc. which thus can help to reduce the risk factors which is potentially contributing to the mortality of motorised two-wheeler drivers in Indian cities. A study carried out by Akaateba et al. (2014) focussed on the assessment of the level of helmet use amongst motorcyclists in Wa, Ghana by enumerating the data on two weekdays and one weekend covering three different time periods each lasting 1 h (8–9 a.m.; 12–1 p.m. and 4–5 p.m.) at 12 randomly selected sites within and outside of the Central Business District (CBD) of Wa. Throughout the course of the study, a total of 14,467 motorcycle riders—11,360 riders and 3107 pillion riders—were observed. The majority of motorcyclists (86.5%) and pillion passengers (61.7%) that were observed were men. In supported logistic regression analysis, higher helmet wearing rates were found to be significantly related to female gender, weekdays, morning periods, and at locations within the CBD. Compared to motorcyclists inside the CBD, riders at places outside the CBD were nearly 7 times less likely to be wearing a helmet. The study came to the conclusion that, despite the requirement for riders and pillion passengers to wear helmets, helmet use is often low in the aforementioned Ghana metropolis. Based on the above reviewed literature, it is evident that the locations chosen for understanding helmet usage need to encompass the length and breadth of the city by covering varying functional classifications of the roads covering expressway, arterial, sub-arterial and collector streets.

234

K. Singh et al.

4 Data Collection Process 4.1 Study Location Delhi, the capital city of India has an area of 1483 sq. Km. Its maximum length is 51.90 km and its greatest width is 48.48 km. The population of Delhi, as per Census— 2011 is pegged at 16.79 million persons which is estimated to have reached to around 25.2 million by 2022 which however needs to be ascertained in the upcoming Census, 2021. The average population in the NCT of Delhi is pegged at 9,340 persons per sq. km measuring a population density of 11,297 per sq. km which is reckoned as one of the most thickly populated in the country. Statistics show the number of victims on two-wheelers killed was 56,136 which is around 37% of the total road accident deaths. Accordingly, the locations selected were spread over the National Capital Territory (NCT) of Delhi by encompassing Delhi—Noida Freeway (DND) which falls under the urban expressway category, five numbers of arterial roads namely, Dr. Hedgewar Marg, Signature Bridge, Peeragarhi, Vikas Marg (at Nirman Vihar Intersection) and Noida Link Road, seven numbers of sub-arterial roads constituting Dr Zakir Hussain Marg, Rao Tula Ram Marg, Africa Avenue, Kasturba Gandhi Marg, Vandematram Marg, Kingsway Camp and Bhishma Pitah Marg, two numbers of Collector street namely, Greater Kailash and Dr M.C Davar Marg. The above data was collected spanning three days (covering two weekdays and one weekend operation) at each of the above road sections. Figure 1 depicts the 15 selected road sections wherein a total of 188,228 samples were collected.

25000

Helmet Usage behaviour of Drivers

20000 15000 10000 5000 0

Total no of Driver using helmet

Driver worn & Strapped the helmet

Fig. 1 Helmet usage behaviour of Drivers on all types of roads

Driver not worn & Unstrapped

Assessment of Adherence Level to Helmet Usage on Varying Roads …

235

4.2 Study Details Data in terms of two-wheelers moving in one direction was gathered to prevent double counting and guarantee accurate data gathering. Details about the driver’s and passengers’ personal characteristics and helmet usage behaviour were included on the survey form. For each driver and passenger, the conditions of the helmet (not worn and unstrapped—coded as 0, wearing and strapped—coded as 1, and full-face helmet—coded as 2) and the type of helmet (cap helmet—coded as 0, non-full-face helmet—coded as 1, and full-face helmet—coded as 2) were recorded. Age (less than 18 years old—classified as 0 and more than 18 years old—coded as 1) and gender (female—coded as 0, and male—coded as 1) for each individual are included under the personal attributes. Land use pattern. The survey timings (7.30 to 9.00 am, 10.00 to 11.30 am, 12.30 to 2.00 pm, 3.00 to 4.30 pm, and 5.30 to 7.00 pm with the aim of covering the morning and evening rush/peak hour traffic), weather conditions (dry with no rain, light rain/drizzle, rain, snow, fog, hail, and other), As part of the field observations, the visible presence of law enforcement (police, camera, both, or none) and survey days (weekday or weekend) were recorded as part of the field observation. The data collection format deployed is shown in Table 1.

236

K. Singh et al.

Table 1 Helmet usage survey data collection format

5 Binary Logistic Regression (BLR) Using the aforementioned data, the binary logistic regression model, also known as logistic regression, was carried out using the SPSS programme. A binary logistic model, which is based on mathematics, has a dependent variable with two alternative values, such as pass or fail, which is represented by an indicator variable and where the two values are denoted by the letters “0” and “1”. As a result, the log-odds (the logarithm of the odds) for the value labelled “1” is a linear combination of one or more independent variables (also known as “predictors”); the independent variables can each be a binary variable (two classes, each coded by an indicator variable), or a continuous variable (any real value). Considering the above, the output deduced from the above analysis revealed two possible outcomes namely, Helmet “Not worn or unstrapped” or “Worn and Strapped” of the Driver, Pillion Rider or at times even the second Pillion Rider. It is the statistical analysis that determines how much variance if all is explained on the dichotomous dependent variable by a set of independent variables considered in this study.

5.1 Purpose of Implementing Binary Logistic Regression Using the above data, it is obvious that two outcomes are possible to understand the level of adherence by the riders of motorised two-wheelers in terms of proper

Assessment of Adherence Level to Helmet Usage on Varying Roads …

237

wearing of helmets which can directly enhance safety on the roads of Delhi, in the event of a road crash. Accordingly, the riders (including pillion riders) wearing a helmet and strapped (rated as 1) and not wearing or wearing without strapped (rated as 0) spread over the above referred 15 sections were deployed. These are considered as dependent variables in a relationship by relating the functional classification of study sections (based on the Right of Way, road width and land use features), survey day, number of occupants, time interval age, gender, and type of helmet is considered as independent variables. According to the aforementioned statistical investigation, the seven criteria may have an impact on whether a two-wheeler rider wears a helmet properly. A model for assessing the driver’s helmet usage behaviour is developed using these seven critical criteria. As previously noted, the outcomes of the categories of helmet usage were split into two groups: not wearing or wearing without straps (ranked as 0), and wearing with a strapped helmet (rated as 1). Due to the fact that there were only two possible outcomes—wearing helmets or not—a binary logistic model was eventually proposed. The dependent variable “Helmet Usage Behaviour of Driver” was developed to support the binary logit model; it contains one if the helmet was worn and secured and zero otherwise. The general form for the binary logit model used is shown in Eq. (1) below: P =

eβ , β = β0 + β1x1 + . . . + βkxk 1 + eβ

(1)

As β becomes high during a positive sense, P will be the approach one that shows the chance of success (in this case, a driver carrying a helmet) will get increase. The binary provision regression (BLR) methodology used for this study for modelling driver’s helmet usage behaviour whereas the riding two-wheelers since the observations area unit thought-about as binary, the binary nature is exploited and also the versatile regression framework permits in-depth analysis.

6 Level of Helmet Usage Across Varying Functional Classification of Roads The output deduced from binary logistic regression covering varying functional classification of roads are presented in Figs. 1, 2 and 3 covering varying types of two-wheeler drivers/riders and the following inference have been drawn (Table 2): • The results deduced for the varying functional classification of the roads revealed that the helmet “Worn and Strapped” behaviour in Driver, Passenger 1 and Passenger 2 is the highest on the expressway by clocking 98.4, 89.4 and 53.9% followed by Sub-arterial road users by measuring 78.0, 72.3 and 27.7%, Arterial Roads registering 73.0, 68.5 and 13.1%, and lastly, Collector Road users registering 65.82, 44.3 and 3.2% level of compliance respectively.

238

K. Singh et al.

Helmet Usage of First Pillion Rider 6000 5000 4000 3000 2000 1000 0

First Pillion riders

First Pillion rider worn & Strapped the helmet

First Pillion rider not worn & Unstrapped

Fig. 2 Helmet usage behaviour of Pillion Rider 1 on all types of roads

Helmet Usage of Second Pillion Rider 900 800 700 600 500 400 300 200 100 0

Second pillion riders

Second Pillion rider worn & Strapped the helmet

Second Pillion rider not worn & Unstrapped

Fig. 3 Helmet usage behaviour of Pillion Rider -2 on all types of roads

Pass1

Pass2

Driver

Pass1

Sub-arterial road Pass2

Driver

Pass1

Collector road Pass2

Total

Worn & strapped

47,414

11,602

275

61,877

12,656

208

10,033

1423

6

64,950

16,937

2101

79,378

17,501

750

15,243

3212

190

−73.00% −68.50% −13.10% −77.95% −72.30% −27.70% −65.82% −44.30% −3.15%

5231

70 27,487

5851

123

−98.40% −89.40% −53.90%

27,049

53

Pass2 −10.59% −43.08%

620

Pass1

Expressway road Driver

Not worn or 17,536 5335 1826 17,501 4845 542 5210 1789 184 438 unstrapped −23.00% −31.50% −86.90% −22.04% −27.70% −72.30% −34.18% −55.70% −96.80% −1.59%

Driver

Type of road Arterial road

Table 2 Helmet usage data comparison of all types of roads

Assessment of Adherence Level to Helmet Usage on Varying Roads … 239

240

K. Singh et al.

• It is evident from the above that level of adherence on the expressway is the highest which can be attributed to the higher prevailing speeds (with Speed Limit set at 70 Kmph for Light Motor Vehicle which includes Two-Wheelers) of all types of vehicles on expressways as well as the presence of police to monitor the adherence. Obviously, these parameters prompt the two-wheeler riders to be more vigilant and adhere to the proper helmet usage on Expressway. • On the other hand, the level of helmet usage on Sub-arterial roads and Arterial Roads is almost similar and the least being Collector Street as the riders on such type of roads tend to ignore wearing helmets due to their shorter trip length. • On the other hand, the level of helmet usage on Sub-arterial roads and Arterial Roads is almost similar and the least adherence is noted on Collector Street as the riders on such type of roads tend to think that their commute is for shorter trip length and thus ignore the wearing of helmets. • Also it can be noted that the level of adherence of helmet usage amongst the Pillion Rider -2 is very poor covering all types of roads. This is despite the fact that the Pillion Rider—2 is not even permitted to travel as per the Motor Vehicle Act of the country which severely compromises the safety of two-wheeler riders on Indian Roads.

7 Concluding Remarks The adherence to the proper type of helmet usage is a mandatory requirement as it plays a vital role in the safety of two-wheeler riders. If a helmet is not used properly strapped, then the helmet will not fulfil its purpose. The results deduced for the varying functional classification of the roads revealed that the helmet “Worn and Strapped” behaviour in Driver, Passenger 1 and Passenger 2 is the highest on the expressway by clocking 98.4, 89.4 and 53.9% followed by Sub-arterial road users by measuring 77.95, 72.3 and 27.7%, Arterial Roads registering 73.0, 68.5 and 13.1%, and lastly Collector Road users registering 65.82, 44.3 and 3.2% respectively. It is evident from the above that the level of adherence on the expressway is the highest which can be attributed to the higher prevailing speeds (with Speed Limit set at 70 Kmph for Light Motor Vehicle which includes Two-Wheelers) of all types of vehicles on expressways as well as the presence of police to monitor the adherence. Obviously, these parameters prompt the two-wheeler riders to be more vigilant and adhere to the proper helmet usage on Expressway. On the other hand, the level of helmet usage on Sub-arterial roads and Arterial Roads is almost similar and the least adherence is noted on Collector Street as the riders on such type of roads tend to think that their commute is for shorter trip length and thus ignore the wearing of helmets. Also, it can be noted that the level of adherence of helmet usage amongst the Pillion Rider -2 is very poor covering all types of roads. This is despite the fact that the Pillion Rider—2 is not even permitted to travel as per the Motor Vehicle Act of the country which severely compromises the safety of two-wheeler riders on Indian Roads.

Assessment of Adherence Level to Helmet Usage on Varying Roads …

241

References Ackaah W, Afukaar FK (2010) Prevalence of helmet use among motorcycle usersin tamale metropolis, Ghana: an observational study. Traffic Inj Prev 11(5): 522–525. https://doi.org/ 10.1080/15389588.2010.489198 Akaateba MA, Amoh-Gyimah R, Yakubu I (2014) A cross-sectional observational study of helmet use among motorcyclists in Wa, Ghana. Accid Anal Prev 64:18–22. https://doi.org/10.1016/j. aap.2013.11.008 Bao J, Bachani AM, Viet CP, Quang LN, Nguyen N, Hyder AA (2017) Trends in motorcycle helmet use in Vietnam: results from a four-year study. Public Health 144: S39–S44. https://doi.org/10. 1016/j.puhe.2017.01.010 Dandona R, Kumar GA, Dandona L (2006) Risky behavior of drivers of motorized two wheeled vehicles in India. J Safety Res 37(2):149–158. https://doi.org/10.1016/j.jsr.2005.11.002 Hassan T, Vinod Kumar MN, Vinod N (2017) Influence of demographics on risky driving behaviour among powered two wheeler riders in Kerala, India. Trans Res F Traf Psychol Behav 46:24–33. https://doi.org/10.1016/j.trf.2016.11.008 Jung S, Xiao Q, Yoon Y (2013) Evaluation of motorcycle safety strategies using the severity of injuries. Accid Anal Prev 59:357–364. https://doi.org/10.1016/j.aap.2013.06.030 Marisamynathan M, Perumal V, Gupta S (2020) Modeling helmet usage behaviour of motorized two-wheeler riders in developing countries. Transp Res Proc 48(May): 31213131. https://doi. org/10.1016/j.trpro.2020.08.177 Mathur AK, Gupta S, Bandhu A (2017) A Baseline Study on Pattern of Helmet Use in the State of Rajasthan, India. J Health Manag 19(3):417–434. https://doi.org/10.1177/0972063417717894

Assessing the Impact of Underground Utility Works on Road Traffic and Users: A Study from an Indian City Ashish Verma, P. Anbazhagan, Sai Kiran Mayakuntla, and Furqan A. Bhat

Abstract Expansion of major cities in India, with its fast-growing economy, has led to a drastic increase in the requirement for basic utility amenities viz. electricity, communication, water etc. There has been a substantial emphasis on developing an underground utility network for providing the essential utilities to people. This leads to excessive construction works both for the installation of new underground tubes as well as for the repair and restoration of the existing ones. Since most of these facilities are placed beneath the roadway surface, the excavation and construction at these sites lead to a reduction in the roadway width, causing congestion. This study analyses the impact of carrying out the underground utility works on road traffic and users at four urban locations in the city of Bengaluru, India. For each site, PTV VISSIM software is used to simulate vehicle trajectories for alternative scenarios with varying obstruction widths using the actual observed traffic volumes and compositions. Vehicle trajectories are fed as inputs to the international vehicle emissions (IVE) model to analyze the impact on transportation systems. The effect is studied in terms of average speed, exhaust emissions and fuel consumption for different classes of vehicles. There is a consistent drop in vehicular speeds for all classes of vehicles with increased obstruction width at all the locations, although the amount of change depends on the characteristics of traffic at that site. While the changes in the vehicle emissions and fuel consumptions are not as consistent in their direction as the vehicle speeds, it is not surprising given the complexity of the vehicle interaction dynamics on Indian roads and the current vehicle technologies.

A. Verma · P. Anbazhagan · F. A. Bhat (B) Department of Civil Engineering, Indian Institute of Science, Bangalore, Karntaka 560012, India e-mail: [email protected] A. Verma e-mail: [email protected] P. Anbazhagan e-mail: [email protected] S. K. Mayakuntla Universidad de Chile, Santiago, USA © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_19

243

244

A. Verma et al.

In general, there seems to be an increase in the emissions and fuel consumptions with an increased obstruction width. Keywords Utility services · Average speeds · Exhaust emissions · Fuel consumption · International vehicle emission model

1 Background Expansion of major cities in India, with its fast-growing economy, has led to a drastic increase in the requirement for basic utility amenities viz. electricity, communication, water etc. (Jaw and Hashim 2013). There has been a paradigm shift in the provision of these facilities with the urban underground of many of the cities being characterized by a dense network of pipelines (Jaw and Hashim 2013). This involves a lot of construction activity which focuses not only on the installation of new facilities but also on the repair and maintenance of the existing underground utility network. Universally, countries are facing a significant challenge of improving underground utility network. There has been a substantial emphasis on developing an underground utility network for providing the essential utilities to people. 70% of the world’s population is predicted to be living in urban areas by 2050 (Sterling et al. 2012). In such a scenario, it becomes imperative to develop systems which are robust and sustainable and could provide necessary utility facilities which form a critical facet of an urban living ecosystem. There is a wide variety of services that are provided through underground utility networks. Many organizations are involved in providing these facilities, and due to lack of coordination and integration between these parties, a number of ducts and tubes are buried under the ground for providing different services without proper planning. This problem termed as “the spaghetti subsurface problem” creates a lot of unnecessary chaos during installation and restoration of utility services primarily due to the scarcity of underground utility maps that could help to determine the exact location and position of utility network (Curiel-Esparza et al. 2004; Curiel-Esparza and Canto-Perello 2013). This dearth of information about the subsurface network also leads to the excavation of broader and longer sections of the roads creating unnecessary bottlenecks and causing the road width to shrink. This eventually leads to a congested highway that has been a significant problem of transportation systems for a very long time (Thomson 1978). Highway capacity, measured in passenger car units (PCU), has been observed to improve with increasing roadway width for four-wheelers (Arasan and Arkatkar 2010) as well as for 2-wheelers (Cao and Sano 2012). This trend is more prominent in heterogeneous traffic conditions with narrower lanes (Chandra and Kumar 2003; Khanorkar et al. 2014) where traffic volume can be further improved by providing good quality shoulders along the highway section (Gaur and Sachdeva 2020; Khanorkar et al. 2014). Similar effects of the decrease in highway volume are seen due to temporary obstruction caused by illegal roadside parking (Yousif and

Assessing the Impact of Underground Utility Works on Road Traffic …

245

Purnawan 1999) or by road accidents (Zheng et al. 2020). Reduction in highway capacity when the demand for travel remains the same leads to traffic congestion. Different transportation system variables (viz reduction in road width due to obstructions, congestion, vehicular speed, fuel consumption and exhaust emission) that are affected by construction works are all interrelated. For example, a reduction in lane width reduces road width, which in turn causes congestion and reduces vehicular speed. Reduced vehicle speed further affects fuel consumption and exhaust emission. There seems to be a lack of consensus in the literature regarding the impact of lane width on travel speed. While the travel speed is found to increase with increasing lane width in some of the studies (Beevers and Carslaw 2005; Gattis and Watts 1999; Godley et al. 2004; Heimbach et al. 1983), there are pieces of evidence to suggest lane width has little to no impact on travel speed (Gattis and Watts 1999). However, in cases of a sudden reduction in lane width leading to the creation of a bottleneck, the speeds are found to decrease (Yousif and Purnawan 1999; Zheng et al. 2020). Traffic congestion not only affects the vehicular speeds but also has an impact on fuel consumption and exhaust emissions. Traffic congestion and subsequent reduction in travel speed lead to increase in fuel consumption (De Vlieger et al. 2000; Errampalli et al. 2015; Greenwood et al. 2007; Samaras et al. 2019). However, reduced travel speed has different impacts on different types of pollutants (Abdull et al. 2020; Elmi and Al Rifai 2012; Nasir et al. 2014; Zhang et al. 2011). As the demand for underground utility network continues to grow, it is essential not to just focus on lowest cost of installation but to also focus on other allied systems that it might have an impact upon. Utility installation and restoration work in any urban area have a direct relation with transport movement and congestion since in most of the cases the utility network is laid beneath the carriageway. Add to this the fact that in most of the Indian cities, the roads are not very wide and utility maps are not available due to which additional space is required to carry out these works. Considering the underground utility network is being provided in most of the metropolitan cities in India, not much research has been done to understand the impacts that these have on the overall transportation system. This study tries to fill this gap in the literature by attempting to answer the following research questions in the context of obstructions along the roadway section due to utility works: 1. How are the average vehicular speeds affected by varying obstruction widths? 2. How does the exhaust emission vary for different classes of vehicles for different road widths? 3. What is the impact of road obstruction on fuel consumption for different vehicle classes? This paper has five sections in addition to this introductory section. Section two gives a brief overview of the methodology used for this study. Site description and data collection methods are discussed in section three. Section four details the emission model used for this study and various inputs and factors used for the model. Results are discussed in section five, and section six concludes the paper.

246

A. Verma et al.

2 Methodology The flowchart in Fig. 1 gives an overview of the overall methodology used for this study. Firstly, specific sites in urban areas of Bengaluru where utility repair works were being carried out during the timeframe of this study were selected, and video data were collected from these sites. The data extracted from these traffic videos were used to simulate vehicle trajectories and obtain vehicular behavior for unobstructed and obstructed roadway scenarios. IVE model was then used to estimate exhaust emissions. All these steps are explained in detail in the following sections.

3 Site Description and Data Collection Four different locations inside the city bounds of Bengaluru, where utility restoration works were ongoing during the months of December 2019 and January 2020, were selected for the purpose of this study. The relevant details about each location and the data collection process are explained in the following subsections.

3.1 Site Description Site 1 – MS Ramaiah Road (MSSR) Figure 2 shows the layout along with the direction of the traffic of the first location chosen for the study, which is located on the MS Ramaiah road. The portion of the road marked dotted rectangles represent vehicles parked along the road reducing the effective road width. In addition, the rectangle with wavy lines in the figure represents the roadway obstructed by the restoration work. It occupies a length of 10 m along the highway and 5 m across it. It was also observed that about 3 m on the left edge of this 10 m wide road is occupied by parked vehicles, indicating obstruction on that part of the road even before the beginning of the utility work. Site 2 – Thanisandra Main Road (TSMR) Figure 3 shows the schematic diagram of the second location, which is on Thanisandra main road in RK Hegde Nagar. As indicated in the diagram, the obstruction covers an entire lane (7 m) of a two-lane road for a length of 30 m. The road is undivided on the R-side of the diagram until the end of the obstruction. Therefore, the vehicles travelling from R to L weave into the other lane at the beginning of the obstruction and weave back at the end of the obstacle. And similar manoeuvres are made by the vehicles travelling from L to R. A small percentage of two-wheelers and cars coming from L is observed to take a U-turn on the L-side of the obstruction, as indicated in the diagram.

Assessing the Impact of Underground Utility Works on Road Traffic …

Fig. 1 Methodology

247

248

A. Verma et al.

Fig. 2 Schematic diagram of MS Ramaiah Road

Fig. 3 Schematic diagram of Thanisandra Road

Site 3 – Bannerghatta Main Road Figure 4 shows the geometric layout of the third location with the obstructed part of the road section highlighted. The repair work at the third location spans approximately 250 m on Bannerghatta main road, which is much longer than the other two. Also, unlike the other two, there are no conflict points significantly affecting

Assessing the Impact of Underground Utility Works on Road Traffic …

249

Fig. 4 Schematic diagram of Bannerghatta Road

the traffic resulting in relatively free-flowing traffic. About half of the road width is rendered obstructed by the work throughout the observation length. Site 4 – 50 Feet Main Road (FFMR). Figure 5 shows the schematic diagram of the fourth location, which is a work of a length of more than 500 m on 50 feet main road near Avalahalli, Banashankari. Since the traffic entering from the direction P is observed to be relatively small, it has been ignored. As shown in the schematic, while the obstruction width of 0.95 m is small, it extends for a longer length than those in the previous locations.

3.2 Data Collection Data at all the four sites were collected by traffic video for a period of 45 min during the evening peak-hour on weekdays. The class-wise traffic movement counts were extracted through manual counting by watching the videos. The details of the class-wise vehicle count for all the study locations are given in Table 1.

3.3 VISSIM Simulations PTV VISSIM was used to recreate the kind of traffic stream that was observed from the field survey. Using the data collected manually from the streets i.e. the road geometry and class-wise vehicular counts, we created scenarios in VISSIM setting traffic routes and counts. The simulations were run, and the trajectory data was collected. This was followed by introducing the roadway obstructions and re-running the simulations to obtain the vehicular trajectories. Thereafter, vehicle trajectories are extracted from the VISSIM simulations. These trajectories are used to extract the parameters to give an idea about the driving behaviour, the details of which are fed into the emissions modelling software (International Vehicle Emissions model).

250

A. Verma et al.

Fig. 5 Schematic diagram of 50 feet road

4 Description of the Model Used Realizing the importance of estimating the emissions from the mobile sources, the United States developed an advanced model that could predict the emissions based on the local conditions viz. fuel type, vehicle classes, maintenance schedules and emission standards. Many of the developing nations modified these models as per their needs while many others adopted the same models and employed it to their fleets. Clearly, this leads to highly unreliable results. As a consequence, an international standard model was developed for estimating emissions from mobile sources under wide-ranging environmental, traffic and regulatory conditions. The international vehicle emissions (IVE) model was funded by the US environmental protection agency and developed by the University of California and the International Sustainable Systems Research Centre (Davis et al. 2005), with the aim of designing

Assessing the Impact of Underground Utility Works on Road Traffic …

251

Table 1 Vehicular counts at the study sites Site 1- MS Ramaiah Road From

To

Car

2-wheeler

3-rickshaw

Bus/HCV

LCV

A

C

4

49

17

0

0

A

B

227

725

77

71

27

C

A

10

143

9

0

3

C

B

66

610

113

3

29

B

A

221

535

141

12

15

B

C

146

571

130

19

38

Site 2-Thanisandra Main Road From

To

Car

Two-wheeler

3-rickshaw

Bus/HCV

LCV

L

R

193

581

162

60

37

R

L

228

653

144

48

39

L

L

10

31

0

0

0

Site 3 – Bannerghatta Main Road From

To

Car

Two-wheeler

3-rickshaw

Bus/HCV

LCV

R

L

743

1941

374

57

77

Site 4 – 50 Feet Road From

To

Car

2-wheeler

3-rickshaw

Bus/HCV

LCV

R

L

422

1381

297

65

59

an emission estimation tool specifically to help transportation planners from developing economies in devising effective emission control strategies and measure their performance over time. Since its development in 2008, the model has been used in emission studies of several cities from developing economies. This model can be employed to evaluate emissions from practically any kind of fleet distribution while being sensitive to future changes in fuel and vehicle technology. To create accurate mobile source emission inventories for a given location, the following inputs are needed by the model: Vehicle technology and emission rates; Driving behaviour and vehicle activity; and. Vehicle fleet distributions.

4.1 Vehicle Technology and Emissions Rates The emission rates are estimated separately for each vehicle type under running (in grams/km) and start conditions (in grams). First, a series of multiplicative factors are applied to the base emission rates to determine the adjusted emission rates. These are then weighted by the corresponding fractions of vehicle technology and driving types.

252

A. Verma et al.

The total emissions are then calculated by multiplying the total distance travelled and the total number of starts with the running and start emission rates, respectively. The following equations are used in the calculation just described. Note that the adjusted emission rates are calculated separately for running and start conditions using different base emission rates. Q [t] = B[t] ∗ K (Base)[t] ∗ K (T mp)[t] ∗ K (H md)[t] ∗ K (I M)[t] ∗ K (Fuel)[t] ∗ K (Alt)[t] ∗ K (Cntr y)[t] Q [t] = Q running =

 

f [t] ∗



t

Q star t =

 

Q [t] ∗ U F T P ∗ f [dt] ∗ K [dt]

d

f [t] ∗ Q [t] ∗

t



f [dt] ∗ K [dt]



(1)  

/U C

(2)

 (3)

d

where, B[t]

Base emission rate for each technology in grams for start and grams/km for running

Q [t]

Adjusted emission rate for each technology in grams for start and grams/km for running

Q

Average emission rate for the entire fleet in grams for start and grams/km for running

f [t]

Fraction of travel by a specific technology

f [dt]

Fraction of each type of driving or soak by a specific technology

U FT P

Average velocity of the LA4 driving cycle in km/hr

D

Total distance travelled in km

UC

Average velocity from the input driving cycle in km/hr

K (Base)[t]

Adjustment to the base emission rate

K (T mp)[t]

Temperature correction factor

K (H md)[t]

Humidity correction factor

K (I M)[t]

Inspection/Maintenance correction factor

K (Fuel)[t]

Fuel quality correction factor

K (Alt)[t]

Altitude correction factor

K (Cntr y)[t]

Country correction factor

K [dt]

Driving or soak style correction factor (includes the effects from air conditioning usage and road grade)

Assessing the Impact of Underground Utility Works on Road Traffic …

253

4.2 Driving Behaviour and Vehicle Activity Driving behaviour The IVE model considers two parameters to be the most correlated to the emissions: vehicle specific power (VSP) and engine stress. VSP can be computed using instantaneous acceleration and velocity, and it depends on the altitude and grade of the road section. Subsequently, the engine stress at any time (second) is computed using the VSP values from the previous 25 s and the engine RPM. Each travel second is assigned to one of 60 bins based on its corresponding VSP and engine stress values. The overall percentage of time spent in each bin is computed and is used in the calculation of emissions. Start patterns The time for which the engine is shut off when it is started, known as engine soak, can have a significant impact on the emissions. Longer soak periods typically result in higher emissions. However, since specific road sections are considered in the present study, it is unlikely that any vehicle under consideration would stop on any of them for more than 15 min and therefore should fall in the first soak period. Environmental variables: Ambient temperature, humidity, altitude and road grade are also observed to have a significant impact on the vehicular emissions. The road grade is clearly constant for a given road section, but the rest might vary with time. For the purposes of this study, the average temperature, humidity and altitude of Bengaluru city are used, which are 27 °C, 60% and 920 m, respectively. Fuel characteristics: The impact of fuel characteristics of diesel and gasoline fuels can be significant on the vehicle emissions. The overall fuel quality and sulphur content are considered for diesel, and the levels of lead, benzene and oxygenate are considered in addition for gasoline. Except the overall quality, all the characteristics are determined from Bharat Stage (BS) IV fuel norms, which are currently enforced in India. While there may be some variation in the overall qualities of gasoline and diesel, they have been approximated as moderate/pre-mixed and diesel, respectively.

4.3 Vehicle Fleet Distribution The IVE model defines 1372 technologies based on six parameters, namely vehicle class, vehicle size, fuel type, vehicle usage, fuel delivery system, evaporative control system and exhaust control system. A distribution of these technologies must be provided to the model for each location in the form of a fleet file. The earlier studies using this model have typically been limited to a single vehicle class, often with

254

A. Verma et al.

specific models. Table 2 presents the fleet distribution used in the study for model analysis. Two likely factors that could explain the observed usage percentages are the vehicle category and the vehicle price. Commercial vehicles like Buses and 3wheelers have much larger high usage percentages than typically personal vehicles like 2-wheelers. In the present context, there may be a slight increase in the commercial usage of cars and 2-wheelers due to thriving e-commerce and food deliveries, but it is reasonable to expect the usage percentages to be qualitatively similar. The usage percentages are chosen based on these insights and the rest of the parameters are selected based on the specifications of the most popular models in each vehicle type.

5 Results The repairing of roads or the utility services affects the road width, which in turn affects the average vehicular speeds, exhaust emissions and fuel consumption. These effects are discussed in the following subsections.

5.1 Average Vehicular Speed At each location, the average vehicle speeds have consistently increased with a reduction in the obstruction width. This confirms our intuition that even a slight increase in the available road width can have a significant impact on vehicle speeds. The actual amounts of change vary across the locations, but this should not be surprising given the complex interaction dynamics between the vehicles. It may be observed that in location 1, which is the most congested of the three, the amount of change increases with the vehicle size. This indicates that the manoeuvrability of the vehicles plays a larger role under congested conditions than under free-flow conditions. Table 3 compares the class-wise average vehicle speeds for all four study sites.

5.2 Exhaust Emissions In this study, the IVE model is used to estimate the pollutant emissions using the vehicle trajectories extracted from VISSIM simulations. Tables 4 and 5 present the percentage changes and changes in actual amounts of pollutant emissions for reduced roadway width. The results calculated at the four locations are less straightforward to interpret than the average vehicle speeds. While the pollution levels significantly increase in the first location (MSRR) which is highly congested with many conflicted points, the same is only broadly true for the rest of the locations. Moreover, the changes in the level of CO2 is unambiguously positive in each case, and it takes

Heavy

Heavy

Auto/Sml Truck

Truck/Bus

LCV

Bus/HCV

Medium

Sml Engine

3-rickshaw

Light

Light

Auto/Sml Truck

Sml Engine

Light

Auto/Sml Truck

Car

2-wheeler

Weight

Class

Vehicle type

Diesel

Diesel

CNG/LPG

Petrol

Diesel

Petrol

Fuel type

Table 2 Vehicle distribution based on class and distance

FI

FI

4-cycle, Carb

4-cycle, Carb

FI

Multi-Pt FI

Fuel delivery system

Euro IV

Euro IV

Improved

Catalyst

Euro IV

3-Way/EGR

Exhaust control system

None

None

None

None

PCV

PCV

Evaporative control system

0.00 0.10 0.90

80-161 K km > 161 K km

> 161 K km < 79 K km

0.10 0.90

80-161 K km

> 50 K km

0.00

0.03 0.78

26-50 K km < 79 K km

0.19

> 50 K km < 25 K km

0.25 0.10

26-50 K km

> 161 K km

0.65

0.0525 0.0175

80-161 K km < 25 K km

0.28

> 161 K km < 79 K km

0.0975 0.0325

80-161 K km

Proportion 0.52

< 79 K km

Distance

Vehicle usage

Assessing the Impact of Underground Utility Works on Road Traffic … 255

256

A. Verma et al.

Table 3 Percentage decrease in class-wise velocities at peak-hour volume traffic with the increased obstruction widths Location

Car

MSRR

16.32

TSMR

0.631

Two-wheeler

Auto rickshaw

Bus/HCV

LCV

1

1.4

27.85

6.5

0.9

0.52

0.58

0.9

BGMR

14.4

13

11.6

9.5

13

FFMR

7.2

6.9

5.3

4.3

4.8 s

Table 4 Percentage increase in the pollutant emissions at peak-hour volume traffic with the increased obstruction widths Location

CO2

HC

PM

NOx 7.73

CO

MSRR

6.5

36.03

32

TSMR

10

−6.02

−2.01

19.5

5.3

35

BGMR

2.9

−11.9

−14.3

−7.4

−34.6

FFMR

1.7

−4.4

−8.2

−4.2

−18.2

Note Negative sign indicates a decrease in the pollutant emission

Table 5 Amounts of increase in the pollutant emissions (in kilograms) in one hour at peak-hour volumes with the increased obstruction widths Location

CO2

MSRR

32

TSMR

14.2

HC 1.6 −0.07

PM

NOx

4.04e−04

0.194

−9.93e−04

0.156

CO 12.6 0.47

BGMR

1.48

−0.074

−2.62e−03

−2.43e−02

−2.3

FFMR

0.66

−0.021

−1.05e−03

−0.011

−0.822

larger values compared to the rest of the pollutants. Nevertheless, it is not a surprise to find this complexity in the pollutant values and may be attributed to the complexity of the vehicle interactions in a mixed traffic environment and on the wide-ranging factors that they depend on.

5.3 Fuel Consumption Table 6 shows the increase in fuel consumption due to an increase in the obstruction width. The total fuel consumption estimated for petrol, diesel and LPG using the emission estimates from the previous section also seem to follow the observation made earlier. There is a noticeable increase in fuel consumption under the congested condition. But there is no discernible pattern when the road is heavily congested even before the obstruction is introduced.

Assessing the Impact of Underground Utility Works on Road Traffic … Table 6 Increase in fuel consumption (in litres) in one hour at peak-hour volumes with the increased obstruction width

257

Location

Petrol

Diesel

LPG

MSRR

16.8

7.47

−2.12

1.96

0.05

TSMR

3.97

BGMR

−1.34

0.23

0.14

FFMR

−1.03

0.18

0.19

6 Discussions and Conclusions The overall cost imposed on the road users by higher obstruction widths can be assessed in various dimensions including the increased travel time, fuel consumption and the impact of the increased emissions on the physical and psychological health of the road users. Most of these works are carried out in urban areas that typically form a part of people’s daily commute; it can be expected that the same individuals would be exposed to the emissions for prolonged durations. This is a worrisome situation given the fact that several Indian cities are frequently listed among the most congested cities of the world. The impact of air pollution on the urban population has been well-studied and has been recognized as one of the major causes of death across the world. (Dominici et al. 2007; Zanobetti et al. 2009)revealed that with every 10 µg/m3 daily increase of PM2.5 concentration in the US, the rate of respiratory diseases increased by 2.7%, and the rate of hospitalization increased by 8%. Prolonged exposure to carbon monoxide has been observed to cause heart damages that may outlast its presence in the bloodstream (Suner and Jay 2008). Exposure to NOx is known to cause eye irritation, reduced lung function and headaches, which are common symptoms observed by many road users in congested Indian cities. In addition to all these, these emissions cause long term harm by damaging natural environments and contributing to global climate change. This study quantifies the impact of non-restoration/poor quality restoration of the underground utilities on Bangalore traffic in terms of the change in average vehicle speeds, pollutant emissions and fuel consumption. Four actual locations where the works were carried out are considered for the study. For each location, simulations are carried in PTV VISSIM software for alternative scenarios with varying obstruction width using the actual observed traffic volumes and compositions. Vehicle trajectories are extracted from the simulations and are given as an input to the international vehicular emissions (IVE) model, and the emissions of different types of pollutants are determined. These estimates are then used to determine the total fuel consumption based on the carbon balance method. The following observations can be made from a comparison of the results: There is a consistent drop in vehicular speeds for all classes of vehicles with increased obstruction width at all the locations, although the amount of change depends on the characteristics of traffic at that site. While the changes in the vehicle emissions and fuel consumptions are not as consistent in their direction as the vehicle speeds, it is not surprising given the

258

A. Verma et al.

complexity of the vehicle interaction dynamics on Indian roads and the current vehicle technologies. Nonetheless, there does, in general, seem to be an increase in the emissions and fuel consumptions with an increased obstruction width. Acknowledgements This paper is part of a project (Ref: PC 99360) titled “Study on Incidences, Risk and Consequences of not having Complete Underground Utility Maps” funded by Office of the Principal Accountant General, Bengaluru, Karnataka, India.

References Abdull N, Yoneda M, Shimada Y (2020) Traffic characteristics and pollutant emission from road transport in urban area. Air Qual Atmos Health 13:731–738. https://doi.org/10.1007/s11869020-00830-w Arasan VT, Arkatkar SS (2010) Microsimulation study of effect of volume and road width on PCU of vehicles under heterogeneous traffic. J Transp Eng 136:1110–1119. https://doi.org/10.1061/ (ASCE)TE.1943-5436.0000176 Beevers SD, Carslaw DC (2005) The impact of congestion charging on vehicle spee dan ditsimplications for assessing vehicle emissions. Atmospheric Environment Cao NY, Sano K (2012) Estimating capacity and motorcycle equivalent units on urban roads in Hanoi, Vietnam. J Transp Eng 138:776–785. https://doi.org/10.1061/(ASCE)TE.1943-5436. 0000382 Chandra S, Kumar U (2003) Effect of lane width on capacity under mixed traffic conditions in India. J Transp Eng 130:402–403. https://doi.org/10.1061/(ASCE)0733-947X(2004)130:3(402) Curiel-Esparza J, Canto-Perello J (2013) Selecting utilities placement techniques in urban underground engineering. Arch Civil Mech Eng 13:276–285. https://doi.org/10.1016/j.acme.2013. 02.001 Curiel-Esparza J, Canto-Perello J, Calvo MA (2004) Establishing sustainable strategies in urban underground engineering. Sci Eng Ethics 10:523–530. https://doi.org/10.1007/s11948-0040009-5 Davis N, Lents J, Osses M, Nikkila N, Barth M (2005) Development and application of an international vehicle emissions model. Transp Res Record 157–165. https://doi.org/10.1177/036119 8105193900118 De Vlieger I, De Keukeleere D, Kretzschmar JG (2000) Environmental effects of driving behaviour and congestion related to passenger cars. Atmos Environ 34:4649–4655. https://doi.org/10.1016/ S1352-2310(00)00217-X Dominici F, Peng RD, Zeger SL, White RH, Samet JM (2007) Dominici et al. Respond to “Heterogeneity of particulate matter health risks.” Am J Epidem 166: 892–893. https://doi.org/10.1093/ aje/kwm219 Elmi A, Al Rifai N (2012) Pollutant emissions from passenger cars in traffic congestion situation in the State of Kuwait: options and challenges. Clean Technol Environ Policy 14:619–624. https:/ /doi.org/10.1007/s10098-011-0421-x Errampalli M, Senathipathi V, Thamban D (2015) Effect of congestion on fuel cost and travel time cost on multi-lane highways in India. Int J Traff Trans Eng 5:458–472. https://doi.org/10.7708/ ijtte.2015.5(4).10 Gattis JL, Watts A (1999) R Elated To P Rosody ! Much M Ore T Han S Peed. J Transp Eng 1–14 Gaur P, Sachdeva SN (2020) Effect of shoulder and slow moving. Intern J Eng Appl Sci Techn 4:377–380 Godley ST, Triggs TJ, Fildes BN (2004) Perceptual lane width, wide perceptual road centre markings and driving speeds. Ergonomics 47:237–256. https://doi.org/10.1080/00140130310001629711

Assessing the Impact of Underground Utility Works on Road Traffic …

259

Greenwood ID, Dunn RCM, Raine RR (2007) Estimating the effects of traffic congestion on fuel consumption and vehicle emissions based on acceleration noise. J Transp Eng 133:96–104. https://doi.org/10.1061/(ASCE)0733-947X(2007)133:2(96) Heimbach CL, Cribbins PD, Chang MS (1983) Some partial consequences of reduced traffic lane widths on urban arterials. Transp Res Record 69–72 Jaw SW, Hashim M (2013) Locational accuracy of underground utility mapping using ground penetrating radar. Tunnel Underg Space Techn Khanorkar AR, Ghodmare SD, Khode BV (2014) Impact of Lane width of road on passenger car unit capacity under mix traffic condition in cities on congested highways. J Eng Res Appl 4: 180–184. www.ijera.com Nasir MK, Noor R, Kalam MA, Masum BM (2014) Reduction of fuel consumption and exhaust pollutant using intelligent transport systems. Sci World J. https://doi.org/10.1155/2014/836375 Samaras C, Tsokolis D, Toffolo S, Magra G, Ntziachristos L, Samaras Z (2019) Enhancing average speed emission models to account for congestion impacts in traffic network link-based simulations _ Elsevier Enhanced Reader. Transp Res Part D: Trans Environ Sterling R, Admiraal H, Bobylev N, Parker H, Godard JP, Vähäaho I, Rogers CDF, Shi X, Hanamura T (2012) Sustainability issues for underground space in urban areas. Proc Inst Civil Eng Urban Des Plan 165:241–254. https://doi.org/10.1680/udap.10.00020 Suner S, Jay G (2008) Carbon monoxide has direct toxicity on the myocardium distinct from effects of hypoxia in an ex vivo rat heart model. Acad Emerg Med 15:59–65. https://doi.org/10.1111/ j.1553-2712.2007.00012.x Thomson JM (1978) Great cities and their traffic. Peregrine books, Penguin Books Yousif S, Purnawan (1999) A study into on-street parking: effects on traffic congestion. Traf Eng Control 40: 424–427 Zanobetti A, Franklin M, Koutrakis P, Schwartz J (2009) Fine particulate air pollution and its components in association with cause-specific emergency admissions. Environ Health Glob Access Sci Source 8. https://doi.org/10.1186/1476-069X-8-58 Zhang K, Batterman S, Dion F (2011) Vehicle emissions in congestion: comparison of work zone, rush hour and free-flow conditions. Atmos Environ 45:1929–1939. https://doi.org/10.1016/j. atmosenv.2011.01.030 Zheng Z, Wang Z, Zhu L, Jiang H (2020) Determinants of the congestion caused by a traffic accident in urban road networks. Accident Analysis and Prevention

Characteristics of e-rickshaw Dominated Mixed-Mode Traffic in Suburban Arterial Corridors Pankaj Kumar, Sabyasachi Mondal, Pritam Saha, and Sudip K. Roy

Abstract Inadequate planning for public transport facilities in most suburban arterials often relegates commuters to locally available modes. For years, many regions witnessed e-rickshaws grow imposingly as an alternative for shorter trips and a preferred feeder service to nearby facilities. Suburban arterials provide frequent access to abutting land uses and allow e-rickshaws to share the same road space. Their presence in large proportion leads to a change in traffic, exhibiting a quite different characteristic. The study investigates the characteristics of such traffic, highlights changes from those that pass through urban and rural settings, and interprets them based on the parametric evaluation. The methodology proposed in this study considers field data on users’ perception of modes, local trips, and flow parameters and evaluates distributional characteristics, vehicle following behavior and traffic operations across flow levels. Empirical observations reveal that e-rickshaws cause a significant slowing of faster vehicles and often compel them to get entrapped inside platoons, reducing capacity and mobility, thus creating congestion and swift manoeuvering, disobeying lane discipline. Keywords Suburban arterials · Traffic characteristics · e-rickshaws

P. Kumar · S. Mondal · P. Saha (B) · S. K. Roy Department of Civil Engineering, Indian Institute of Engineering Science and Technology, Shibpur, India P. Kumar e-mail: [email protected] S. K. Roy e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_20

261

262

P. Kumar et al.

1 Introduction Over the past few decades, many Indian cities have witnessed rapid real estate, commercial, and institutional developments in their fringe areas. City residents are migrating to such regions due to a better standard of living at a relatively low cost. A large population from outside the city is also settling down due to opportunities in employment, business and better medical and educational facilities. Eventually, the regions start overgrowing, especially in the vicinity of arterials for easy access to the city center leading to a significant change in population demography (Brockerhoff 2000). Systematic development plans of such regions are usually not present due to a shortage of funds and lack of vision. Developments typically happen along arterials, the only lifeline of the region, and the arterial route provides frequent access to abutting land uses. Residents regularly travel to the city for different purposes such as work, education, medical, and other needs. Suburban regions gradually generate more road trips, particularly to places not very far and trip lengths are less. Often a gap in public transport supply makes residents rely more on e-rickshaw services these days to meet their transportation needs (Rana et al. 2013) due to their easy accessibility and high frequency. Since such modes provide last-mile connectivity and doorstep services, they usually prefer such services to conveniently make local trips with an average trip length of around 3–5 km. A study in this context indicates that nearly 85% of commuters prefer to avail of e-rickshaws for trip lengths less than 5 km (Rana et al. 2013) due to the convenience and affordability, which eventually make them a good alternative to conventional ones (Ghosh et al. 2021). A mix of exiting/entering e-rickshaws to/ from the abutting land uses and through traffic with different vehicle dynamics creates conflicts on arterials and results in highspeed differentials. Hypothetically, suburban arterials intend to transition from lowspeed roads in urban settings to high-speed highways in rural areas. The low-speed potential of such vehicles impedes faster ones resulting in a reduction in traffic speed and capacity as a consequence (Mondal and Saha 2017). However, the presence of erickshaws makes road traffic extremely heterogeneous, causing frequent impedance to faster vehicles in the traffic, compelling them to slow down or stop quite often while moving with platoons. Recurrent share of the same road spaces by e-rickshaws with through traffic in large proportion eventually leads to a change in prevailing traffic, thereby exhibiting a characteristic that is quite different from highways and urban streets. The study premise thus entails an initiative towards investigating the characteristics of traffic on arterial corridors passing through urban fringe areas providing frequent access to abutting land uses. Based on empirical analysis using field data, the study highlights changes in traffic characteristics on roads from those that pass through urban/ rural settings and interprets them based on the parametric evaluation. It explicitly presents the effects of different vehicle types on platooning based on vehicle-specific headway thresholds and time spent by the entrapped vehicles while moving with platoons under different vehicle impedances. Afterwards, the study performed speed-flow-density analysis for the study sections and developed

Characteristics of e-rickshaw Dominated Mixed-Mode Traffic …

263

relationships aptly using complex models to indicate operating conditions and the possible capacity.

2 Objectives and Study Scopes The study attempts to demonstrate the operational characteristics of traffic on suburban arterials using field data of the modal preference of commuters, the proportion of local trips in the prevailing traffic, and flow parameters. The methodology proposed in this study is based on distributional characteristics of traffic flow parameters, vehicle following behavior and traffic operations across flow levels. Thus, the study aims at meeting the following objectives: • Study the characteristics of traffic on suburban arterial roads exhibiting a significant fraction of e-rickshaws based on a comparative and deterministic approach • Evaluate the outcomes of the flow models considering those obtained on highways and urban streets with mixed-mode traffic and more or less similar composition The field study focuses on arterials connecting rural highways and urban streets and considers different land uses, including predominantly residential, commercial, and mixed. Study sections provide frequent access to abutting land uses with an average access point density of about ten [nos/km]. The study adopted a time cluster sampling procedure covering peak and off-peak hours and continued for several days to obtain a representative sample. The traffic composition displayed a large proportion of e-rickshaws and two-wheelers collectively sharing around half the total traffic.

3 Study Design Traffic flow is stochastic since vehicle arrivals are associated with randomness. The frequent appearance of slower modes generates a scenario of vehicle events characterized by platooning, congestion, lack of lane discipline and significant slowing of vehicles. Often, e-rickshaws and two-wheelers tend to manoeuvre swiftly through the lateral gap on either side of entrapped faster vehicles, making traffic movements unique, unlike highways and urban streets. The following sections describe the study approach designed to capture such characteristics and interpret impacts on traffic performances.

264

P. Kumar et al.

3.1 Model and Method The study designed a stepwise method (Fig. 1) to capture the effects of e-rickshaws on traffic performance, including evaluating the (a) trends in distributional characteristics of traffic flow parameters, (b) effects of different vehicle types on platooning and (c) speed-flow-density relationship. The study applied Poisson distribution (Sharma et al. 2011) for vehicle arrivals, log-logistic, lognormal, gamma, and negative exponential distributions for time-headways (Roy and Saha 2018) and normal distribution (Dey et al. 2006) for vehicle speeds to describe field data aptly. Equation 1–6 encodes the probability distributions of the selected models. For the statistical validity assessment of the models, the study applied the K-S test technique (Smirnov 1939). Poisson distribution: ( / ) P(x) = e−λ λx x!

(1)

( / )α }(−2) / ( / )(α−1) { 1+ h β f (h) = α β h β

(2)

Log-Logistic distribution:

Lognormal distribution: {( ( / √ ) )2 } / f (h) = 1 h.σ. 2π . exp (ln h − μ) 2σ 2

Fig. 1 Flowchart of the analytical method

(3)

Characteristics of e-rickshaw Dominated Mixed-Mode Traffic …

265

Gamma distribution: / ( / ) f (h) = h α−1 β α ┌(α). exp −h β

(4)

Negative exponential: f (h) = λ. exp(−λh)

(5)

{ / ( ( / √ ) / )2 } f (v) = 1 σ. 2π . exp −1 2 (v − μ) σ

(6)

Normal distribution:

where P(x) – probability of vehicles x arriving in time t; λ – mean and the variance of x; f(h) – pdf for headways, σ – cont. scale parameter; μ – cont. location parameter; λ – cont. inverse scale parameter (λ > 0); β – cont. scale parameter; α – cont. shape parameter (α > 0); v – vehicle speed The amount of platooning can be expressed in terms of percent time-spentfollowing (PTSF). Equation 7 illustrates the method for determining PTSF based on the average number of headways inside and outside a platoon (Saha et al. 2015). Percent time-spent-following: / P T S F = 100.Q (Q 0 + N0 )

(7)

where Q is the number of faster vehicles behind impending slower ones, Q0 is the average number of headways inside platoons and N0 is the average number of headways between platoons. Wide variations in mixed-mode traffic operations cause deviations in passenger car unit (PCU) values across flow levels and compositions. Therefore, the use of dynamic PCU appears appropriate. The study identified three distinct flow levels, viz. low, moderate and heavy, to determine the values (Indian highway capacity manual (Indo-HCM). Council of Scientific Industrial Research (CSIR) 2017), taking observed compositions into account. The study evaluated linear, exponential, logarithmic, and complex models (see Eq. 8–12) for developing the deterministic speeddensity model (see Table 1) and found complex models to meet the required traits. However, the complex model form proposed by Del Castillo and Benitez appears to have more applicability for the current traffic characterized by the frequent formation and dispersion of platoons since it considers the generation of repeated kinematic and shock waves. The model (see Eq. 12) indicates that kinematic wave speed at jam density, C j, predominantly affects vehicle flow. The expression λ/ K j estimates C j, where the calculation of λ value and K j can be based on a comparison of Eq. 11 & 12 (Gaddam and Rao 2019) and site-specific traffic data. Linear: greenshield ( / ) v = vf 1 − k kj

(8)

266

P. Kumar et al.

Table 1 Performance evaluation of the deterministic speed-density models Deterministic models: Speed-density

Greenshields [Eq. 8] Greenberg [Eq. 9] Underwood [Eq. 10] Newell [Eq. 11]

Del Castillo & Benitez [Eq. 12]

Model properties Static

Dynamic

vf [a]

k j [b]

v [c]

√

√

√

√

x √

x

Comments on model performance

v’ [d] √

C j [e]

q” [f]

x

x

√

√

Acceptable; however, lack of flexibility makes empirical application difficult (Li 2008)

x

Not acceptable; speeds become infinite at zero density (Li 2008; Rompis 2018)

√

x

x

x

√

√

√

√

√

√

√

√

√

√

√

x

√

Not acceptable; speeds become zero only at infinite density (Rompis 2018) Acceptable; however, it generates unstable shockwaves at jamming conditions (Gaddam and Rao 2019; Romanowska and Jamroz 2021) Acceptable; since it meets all the required traits, including stable shock wave and kinematic wave speed (Gaddam and Rao 2019; Romanowska and Jamroz 2021)

√ Note – satisfied; × – not satisfied; a:free-flow speed, v f |v(k)k→0 = v f ; b: jam density k j |v(k)k→k j = 0; c: speed range, v|0 ≤ v ≤ v f ; d: slope, v' | v' k < 0; e: kinematic wave speed, C j | q' (k)k→kj = negati ve constant; f: stable shock wave, q'' |q'' (k)k→kj > 0

Exponential: greenberg ( / ) v = vm ln k j k

(9)

( / ) v = v f exp −k km

(10)

[ {( / )( / / )}] v = v f 1 − exp −λ v f 1 k − 1 k j

(11)

Logarithmic: underwood

Complex: newell

Complex: Del Castillo and Benitez

Characteristics of e-rickshaw Dominated Mixed-Mode Traffic …

[ {( / )( / )}] v = v f 1 − exp C j v f 1 − k j k

267

(12)

where vf – Free-flow speed; k j – Jam density; vm – Critical speed; k m – Optimum density; C j – Kinematic wave speed; λ – Constant parameter

3.2 Field Data The study selected five sections of suburban arterials passing through diverse land use of urban fringes of Kolkata spanning over western and northern regions of the city, including residential, commercial, and mixed, for field data (Fig. 2). The study stretches are 6–7 m wide, allowing two-way vehicular traffic movements. An opinion poll of regular commuters conducted in the study regions reveals that about 80% of local trip makers prefer e-rickshaws due to such modes’ flexibility and affordability. The rapid upsurge in the number of e-rickshaws in due course led to a situation where they share about 30–35% of total traffic on suburban arterials. Eventually, such traffic starts exhibiting unique characteristics and is in no way comparable to the ones prevalent on roads passing through rural/ urban areas. The study applied the videotaping technique to capture flow parameters for necessary investigation. It encompassed different land uses, for instance, predominantly residential/commercial, mixed etc., and considered arterials connecting rural highways and urban streets. The camera focused on a trap of 30 m marked on the road surface while recording traffic. The distribution of time for data collection was over a more extended period covering peak and off-peak hours to obtain a representative sample. Processing video files while extracting entry and exit times of vehicles from such recorded data was done through a computing platform that provides a reasonable amount of accuracy. The extracted data indicates that inadequate public transport facilities enhance reliance on two-wheelers (30–35%) and cars (20–25%) among local commuters.

4 Application and Interpretation of Results The study applied the model forms indicated in Sect. 3.1 to analyze field data. It evaluated the model outcomes and trends and patterns for interpreting the results. The following sections detail the study findings and highlight the characteristics of the traffic.

4.1 Distributional Characteristics of Flow Parameters Fitting field data to suitable statistical models describing vehicle arrivals, timeheadways and speeds (Fig. 3a–c) and estimating model parameters used a technique

268

P. Kumar et al.

Fig. 2 Photo view of the five (a–e) study sections [image captured by the authors]

based on maximum likelihood estimation. Afterwards, the study examined how those models fit observed data using K–S goodness-of-fit tests. Table 2 shows the selected distribution models’ descriptive statistics and goodness-of-fit test results. The experiment found negative exponential distribution to exhibit a minimum statistic value for headways, thereby describing observed data aptly. The study applied the concept of spread ratio (Dey et al. 2006) while modeling speed data and found normal distribution to have a good fit. Also, it exhibits acceptable statistical validity in terms of K-S test results. For vehicle arrival, the study assumed 20 s time interval and manually counted the number of vehicles arriving to obtain the probability of occurrence of an event. The study used Poisson distribution while expressing the probabilities based on empirical testing. A look in Fig. 3a–b reveals that the probability of shorter headway is significantly large. Frequent interaction with e-rickshaws results in significant slowing of relatively faster vehicles, and they start following, keeping smaller space. As a consequence of such events, the road stretch starts witnessing congestion, delay, tendency to disobey lane discipline, swift manoeuvring tendency and risk-taking behavior on the part of drivers. A comparison of study results with

Characteristics of e-rickshaw Dominated Mixed-Mode Traffic …

269

Fig. 3 The trend in distributional characteristics of traffic flow parameters and a comparative performance assessment: a vehicle arrival, b time-headway, and c vehicle speed

those observed on highways or urban streets clarifies that such impacts are quite high on the suburban arterials as referred to in the current study. The distribution of speed data supplements the fact indicating a mean speed of 25 km/h on such roads, unlike other roads as highlighted where such value is fairly high.

4.1.1

Vehicle Following Behavior Under High e-rickshaw Fractions

Though speed generally increases with headways, it remains insensitive after a certain headway threshold; beyond this limit, the influence of lead vehicles practically ceases. Field observations indicate that on suburban arterials with large e-rickshaw fractions in traffic, the value is vehicle-specific and depends on the dynamics of the following vehicle types (Fig. 4a). A look into the plot indicates that a change in speed continues at moderate to heavy flow levels until headway attains a value of about 5–6 s. For cars, it is around 5 s which keeps increasing for bikes, LCVs, and buses and reaches a maximum for e-rickshaws. Inquiries on the impedance caused by different vehicles reveal that, as anticipated, e-rickshaws significantly contribute

8.23

24.8

5.48

4.00

Med

0.02

1.43

0.46

Skp

1897

−0.78

874 1778

−0.37 2.06

N

K

1.53 – –

Lognormal [Eq. 3] Gamma[Eq. 4] Negative Exponential [Eq. 5] Normal [Eq. 6] 25.6

–

Log-logistic [Eq. 2]

8.23

–

–

1.23

–

–

σ

–

–

1.11

–

1.41

–

α

Model Parameter μ

Poisson [Eq. 1] –

Distribution

–

–

7.26

–

4.60

–

β

–

0.12

–

–

–

4.29

λ

0.095

0.130

0.063 0.400

0.068

0.070

0.028

0.230

P-value

0.046 *

0.067

0.076

0.077

D

0.05

0.05

0.05

0.05

0.05

0.05

α-value

Accept

Accept

Accept

Accept

Reject

Accept

H0

Note x¯ – Observed mean; SD – Standard deviation; Med – Median; Skp – Skewness; K – Kurtosis; N – Sample size in Nos.; μ – Mean; σ – Standard deviation; α – Shape parameter; β – Scale parameter; D – K-S statistics; H 0 – Null hypothesis (Accept/Reject); * selected distribution based on the lowest value of K-S statistics

25.6

Vehicle speed

7.64

2.14

4.29

8.04

Vehicle arrival

SD

Field parameter

x¯

Time-headway

Flow characterization

Table 2 Descriptive statistics of the observed vehicle arrival, time-headway and speed data, corresponding distribution model parameters, and goodness-of-fit details

270 P. Kumar et al.

Characteristics of e-rickshaw Dominated Mixed-Mode Traffic …

271

Fig. 4 Effects of different vehicle types on platooning: vehicle-specific a headway threshold, b impedance and platoon lengths, and c PTSF against different lengths of platoons

toward platooning and may sometimes lead to a situation when platoon length is as high as thirteen (Fig. 4b). A further investigation confirms that often such modes cause the maximum percent time-spent-following (0.75) for entrapped faster vehicles when they are platoon leaders (Fig. 4c).

4.2 Macroscopic Relations of Traffic Parameters Using Complex Models The experimental framework considered developing speed-density relationships. Complex models appeared appropriate since they account for dynamic properties, i.e., kinematic wave speed and stable shock wave. Table 3 displays the PCUs used as inputs into the model at different flow levels and compositions. The models proposed by Del Castillo & Benitez and Newell exhibit acceptable statistical validity in terms of R2 (Fig. 5a). Low flow conditions usually prevail in the early morning or late evening; erickshaws are therefore generally not present in the traffic at such time points since

–

5.00

4.45

2.96

–

–

–

2.88

–

5.50

5.07

3.00

–

–

2.50

2.50

NMV†

0.40

0.20

–

–

–

–

1.50

1.40

E-R†

–

–

4.50

2.07

LCV†

–

–

–

–

Bus†

–

–

–

4.86

Truck†

–

–

–

–

NMV†

–

–

–

0.20

TW†

–

–

–

1.40

E-R†

–

–

4.00

–

LCV†

–

–

–

–

Bus†

–

–

4.34

–

Truck†

–

–

–

2.50

NMV†

PCU - Low flow [200 < q ≤ 500 veh/hr]

Note:PCU – Passenger Car Unit; TW – Two-wheeler; E-R – E-Rickshaw; LCV – Light Commercial Vehicle; NMV – Non-Motorized Vehicle, q – Traffic flow

–

0.29

30–40

1.70

2.00

0.20

0.26

10–20

1.20

20–30

–

Truck†

TW†

Bus†

PCU - Moderate flow [500 < q ≤ 800 veh/hr]

LCV†

TW†

E-R†

PCU - Heavy flow [q > 800 veh/hr]

0–10

Percent share of vehicle†

Table 3 Passenger car units for the observed vehicle types under various flow levels and their percentage share of the total traffic Indian highway capacity manual (Indo-HCM). Council of Scientific Industrial Research (CSIR) 2017 (Source: Indo HCM 2017)

272 P. Kumar et al.

Characteristics of e-rickshaw Dominated Mixed-Mode Traffic …

273

Fig. 5 Plots of a deterministic speed-density and b speed-flow relationships developed using complex models (Newell; Del Castillo) (Gaddam and Rao 2019; Romanowska and Jamroz 2021)

local trips are insignificant. The interesting fact that the study observed under such conditions is speed remains insensitive up to a flow range of 1000 pc/h (Fig. 5b). The study also compared the model outcomes with those obtained from highways and urban streets (Table 4). Free speeds on such roads are mainly on the lower sides, especially when compared to highways; however, the values do not deviate much when compared with arterials/sub-arterials of urban/ suburban regions. Trends are more or less comparable for capacity as well. On the other hand, the jam density was on the higher side. This is attributable to the long tail of the Del Castillo model and a large proportion of small-sized vehicles, including e-rickshaws and two-wheelers, in the mixed-mode traffic.

5 Conclusions Overgrowing travel demands in suburban regions of many cities across the country led to the generation of more road trips, especially to nearby destinations with shorter trip lengths. Since the development of such areas mostly takes place in a very unplanned way, often public transport supply cannot meet the demands and relegate commuters to some locally available transport modes. For the past few years, e-rickshaws have grown in many places in an imposing way as an alternative and also as a preferred feeder service to nearby public transport facilities. Arterials passing through suburban regions with residential, commercial and institutional settlements provide frequent access resulting in a mix of e-rickshaws that are either exiting to or entering from the abutting land uses and through traffic. Variations in the dynamics of the vehicle spectrum sharing the same road space create a conflicting situation and make the prevailing traffic characteristically different from the ones observed on highways and urban streets. The study conducted a field-based investigation to understand the characteristics of such traffic and ascertain the impacts of e-rickshaws on traffic

73

170

3202

Note vf – Free-flow speed; qmax – Capacity; K j – Jam density

220

433

kj (pc/km)

52

2860

42.65

2733

qmax (pc/h)

vf (km/h)

Traffic stream Current Rural highways characteristics study Jain et al. Singh et al. (Singh (Jain et al. and Santhakumar 2021) 2021)

241

3670

60.8 176

2824

82

Rompis et al. Dey at el. (Rompis 2018) (Dey et al. 2008)

157

2078

53

Pal et al. (Pal et al. 2020)

176

2064

46.8

217

2146

39.5

Mankar et al. Biswas (Mankar and et al. Khode 2016) (Biswas et al. 2021)

Urban roads

138

2000

58

Sharma et al. (Sharma et al. 2011)

158

2100

53

Mondal et al. (Mondal and Saha 2017)

221

2269

41

Kumaratunga et al. (Kumaratunga 2018)

Suburban roads

Table 4 A comparison of macroscopic measurements of traffic flow between the current study (obtained based on a complex model) and some of the past research records

274 P. Kumar et al.

Characteristics of e-rickshaw Dominated Mixed-Mode Traffic …

275

performance. It concluded that e-rickshaws cause a significant slowing of faster vehicles and often compel them to get entrapped inside platoons resulting in a reduction of capacity and mobility, thus, creating congestion and swift manoeuvering disobeying lane discipline in consequence. Also, the study provides deeper insights into characteristics and critical inputs into a simulation of e-rickshaw-dominated mixed traffic, especially in suburban arterial corridors.

References Biswas S, Chandra S, Ghosh I (2021) Side friction parameters and their influences on capacity of Indian undivided urban streets Int. J Transp Sci Technol 10(1):1–19 Brockerhoff M (2000) An urbanizing world. PRB, Washington, DC, USA Dey PP, Chandra S, Gangopadhaya S (2006) Speed distribution curves under mixed traffic conditions. J Transp Eng ASCE 132(6):475–481 Dey PP, Chandra S, Gangopadhyay S (2008) Simulation of mixed traffic flow on two-lane roads. J Transp Eng ASCE 134(9):361–369 Gaddam HK, Rao KR (2019) Speed–density functional relationship for heterogeneous traffic data: a statistical and theoretical investigation. J Mod Transp 27(1):61–74 Ghosh A, Dey M, Mondal SP, Shaikh A, Sarkar A, Chatterjee B (2021) Selection of best E-rickshawa green energy game changer: an application of AHP and TOPSIS method. J Intell Fuzzy Syst (Preprint), pp 1–14 Indian highway capacity manual (Indo-HCM) (2017) Council of Scientific & Industrial Research (CSIR). New Delhi Jain M, Gore N, Arkatkar S, Easa S (2021) Developing level-of-service criteria for two-lane rural roads with grades under mixed traffic conditions. J Transp Eng A 147(5) Kumaratunga PCI (2018) Evaluation of PCU factors for two lane sub-urban roads (Doctoral dissertation) Li MZ (2008) A: generic characterization of equilibrium speed-flow curves. Transp Sci 42(2):220– 235 Mankar Pratik U, Khode BV (2016) Capacity estimation of urban roads under mixed traffic condition. Int Res J Eng Technol (IRJET) 3.04: 2750–2755 Mondal S, Saha P (2017) Rising growth of E-rickshaws in Indian traffic context: a challenge in efficient traffic operations In: 32nd Indian Engineering Congress, pp 154–160. Chennai, India Pal D, Sen S, Chakraborty S, Roy SK (2020) Effect of PCU estimation methods on capacity of two-lane rural roads in india: a case study. Transp Res Proc 48:734–746 Rana MS, Hossain F, Roy SS, Mitra MSK (2013) Exploring operational characteristics of batteryoperated auto-rickshaws in urban transportation system. Am J Eng Res 2(4):1–11 Romanowska A, Jamroz K (2021) Comparison of traffic flow models with real traffic data based on a quantitative assessment. Appl Sci 11(21):9914 Rompis SY (2018) Traffic flow model and shockwave analysis. J Sipil Statik 6(1) Roy R, Saha P (2018) Headway distribution models of two-lane roads under mixed traffic conditions: a case study from India. Eur Transp Res Rev 10(1):1–12 Saha P, Sarkar AK, Pal M (2015) Evaluation of performance measures of two-lane highways under heterogeneous traffic. Pertanika J Sci Technol 23(2):223–239 Sharma N, Arkatkar SS, Sarkar AK (2011) Study on heterogeneous traffic flow characteristics of a two-lane road. Transport 26(2):185–196 Singh S, Santhakumar SM (2021) Evaluation of lane-based traffic characteristics of highways under mixed traffic conditions by different methods. J Inst Eng India Ser A 102(3):719–735

276

P. Kumar et al.

Smirnov NV (1939) On the estimation of the discrepancy between empirical curves of distribution for two independent samples. Bull Math Univ Moscou 2(2):3–14

Qualitative and Quantitative Evaluation of Urban Car Parking System: A Case Study of Bhopal City Hrishabh Chouhan, Pritikana Das, and Dungar Singh

Abstract In major cities because of the rapid development parking generation rate has increased rapidly. It has been observed that due to the lack of parking supply, inefficient utilization of parking space leads to many parking problems at various locations in the city. A study has been carried out in Bhopal city to evaluate car parking scenarios by using both qualitative and quantitative approaches and to measure adherence to parking guidelines. Surveys are carried out at 2 locations for both on-street and off-street, typically in tourist, commercial, and shopping areas of Bhopal. A licence plate method survey is performed to determine the parking characteristics that include parking accumulation, parking turnover, parking load, parking volume, peak parking saturation, and parking efficiency. A questionnaire survey is performed to find out potential parameters that influence the parking demand and eventually establish the parking demand model. The outcomes of the study show that the lake view parking facility suffers from spill over during peak time. Age income, search time and several other parameters are the statically significant predictors of parking demand at the selected locations. It is needed to regulate the parking fee according to parking duration for the effective utilization of parking space. Following the findings of the study, several recommendations for making the best use of available space have been proposed. Engineers, planners, and policymakers will be benefited from the study findings. Keywords Parking characteristics · Parking statistics · Accumulation · Parking behavior

H. Chouhan · P. Das · D. Singh (B) Maulana Azad National Institute of Technology Bhopal, Bhopal 462003, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_21

277

278

H. Chouhan et al.

1 Introduction In past years India has proved to be one of the fastest-growing economies, major cities have become the most attractive destination in terms of employment generation which resulted in rapid migration in cities. In cities, growth of population and high living standards results that more population being capable of affording a car. Subsidized parking fees, and affordable car pricing, attract more people to tend toward the private car as a mode of choice for a majority of their trips. Now due to the immense increase of cars on the roads, the parking demand cannot be fulfilled by the available supply in many parts of the city due to insufficient parking facilities and improper parking management. It has been observed that many parking lots due to lack of proper marking and inefficient management results in haphazard parking, eventually affecting the utilization of parking space. In some areas of cities, multilevel off-street parking is provided to satisfy the parking demand. But much older areas and CBD areas of the cities have lacked these facilities due to lack of availability of space. It causes the problem of roadside illegal parking, which reduces the effective carriageway width of the road as well as reduces the vehicular speed in traffic flow, further, it would create spill over conditions during peak hours of traffic conditions in many areas of the city. In recent years, high research work has been published on parking behavior and parking characteristics in different conditions, whereas limited literature was found in the Indian context. Very few studies have considered both qualitative and quantitate parameters for the analysis of urban car parking.

1.1 Parking Statistics Terminologies used in the study are as follows: 1. Parking accumulation: It is the count of a car parked at any given moment of time. It is represented by plotting a graph between accumulation and time. 2. Peak parking saturation: It is the ratio of the number of vehicles parked at peak time to the capacity of parking space in terms of the number of bays. 3. Parking volume: The total number of vehicles parked for a specific time period or survey period is referred to as parking volume. It is permissible to repeat the same car, i.e., each unique vehicle is counted. The number of vehicles that enter the parking lot is recorded. 4. Parking load: The whole area under the accumulation curve is referred to as a parking load. It can also be calculated by multiplying the total number of vehicles occupying a parking place at each time period by the interval. It’s expressed in vehicle-hours. 5. Peak parking ratio: This is the ratio of the total number of vehicles parked at peak time to the total number of vehicles parked at each time.

Qualitative and Quantitative Evaluation of Urban Car Parking …

279

6. Average parking duration: It is the ratio obtained by dividing the parking load (vehicle-hours) by the total parking volume throughout the survey period. 7. Parking turnover: It is a measure of a parking lot’s occupancy that is calculated by dividing the number of parked vehicles for a specific period of time by the total number of available parking bays. It can be specified in terms of the number of vehicles per bay and the length of time. 8. Parking index: Parking index is a measure of the efficiency of parking space. It is defined as the ratio of a total number of vehicles parked in time duration to the total space available that is capacity. It gives an aggregate measure of how effectively the parking space is utilized.

1.2 Objectives The objectives of the study are as follows. 1. To evaluate urban car parking using a quantitative approach. 2. To assess behavior characteristics of car parking demand using qualitative approach. 3. To develop a parking demand model.

2 Literature Review Their pieces of literature were examined in needed to finalize the study objectives and identify research gaps in recent work has described. Parmar et al. (2020a) analyzed the nine sections of various parts of Delhi city, the parking facilities were assessed with the help of parking statistics like peak parking saturation, and parking index. Authors found in their study that few sections were overburdened and facing the condition of spill over at busy times. Chakrabarti and Mazumder (2010) employed multilinear and log-linear regression to develop a parking demand and mode choice model. The study observed that parking supply features were a significant effect on reducing automotive dependence. To understand socioeconomic and travel characteristics Parmar et al. (2020b) conducted a questionnaire survey to obtain a revealed preference survey. The results found that travel time and total cost, including parking fees, were a significant impact on parking demand. Parmar et al (2019) developed the level of service criteria for urban car parking systems using clustering techniques. To assess the service level of parking area operation in mixed traffic conditions several parameters such as parking fees, ease time, walking time to destination, and demand/ capacity ratio were considered. Further, in mixed traffic conditions, Ajeng and Gim (2018) analyzed on-street parking duration and demand. The authors established a relationship between parking demand and other parameters such as street length and land use by ordinary least square regression. Ying and Sun (2020) analyzed the parking behavior using various parameters like parking purpose, parking fee,

280

H. Chouhan et al.

and parking time during weekends and weekdays. Zhao et al. (2018) developed a route choice model and parking choice behavior model. Aderamo et al. (2013) used the regression method to predict parking demand along the specified streets and off-street parking facilities. The outcome of the study was suggested that onstreet parking should be discouraged and adequate off-street parking facilities need to be provided. Wang et al. (2018) generated the logit model for tourist cities using walking distance, road condition, vacant parking spot and parking destination as a parameter. The outcome of the study suggests that preplanning of infrastructure and ground parking should be given preference over underground parking. Moeinaddinia et al. (2013) used an inventory survey to create a parking inventory level of service (PILOS) and a parking demand study to create a parking demand level of service (PDLOS) in the university area, then took the average of the two to create a parking area level of service (PALOS). Many previous literatures have primarily focused on either a qualitative or quantitative approach to evaluating car parking systems in urban settings. However, both qualitative and quantitative approaches must be evaluated together for a thorough assessment of the existing car parking scenario and future recommendations. However, there is a scarcity of studies on urban car parking in the Indian context. The study identifies and includes a few elements that may influence car parking demand that has not been explored in previous research.

3 Study Methodology The proposed methodology for conducting the study is discussed in this chapter. Figure 1 depicts the suggested methodology for this analysis. The study’s framework is described in the approach section, which includes the work procedure and technique for analyzing urban car parking via parking characteristics and producing a parking demand model using multilinear regression Based on this prior research work and objectives of our study, the methodology for the study has been finalized, emphasizing the methods for collecting survey data, extracting data from the survey, and analyzing the data.

4 Study Area and Data Collection 4.1 Study Area The demand for parking is directly related to the quantity of land used. The days and times of peak periods are also influenced by land usage. We chose places in our study that had various types of land use. The location of our research area is depicted in Fig. 2. Selected areas are the significant area of Bhopal city.

Qualitative and Quantitative Evaluation of Urban Car Parking …

Fig. 1 Methodology chart

Fig. 2 Location of study area in Bhopal. (Legend -1 MP Nagar, 2-Lake View)

281

282

H. Chouhan et al.

• Significant movements of cars in a parking lot are ensured to choose the location. • To account for the heterogeneity of characteristics on parking demand, a location with a variety of land uses was chosen. • On-site or neighboring roadway construction should be minimum or non-existent. • The location should be chosen based on the ability to get reliable parking demand statistics for the proposed land use.

4.2 Data Collection Parking data were collected during the weekends and weekdays of January and February month, using licence plate method, and questionnaire survey, in peak and lean hours from morning 9:00 A.M. to 8:00 P.M. The parking lot was investigated at 15 min intervals for the licence plate survey, with the registered licence plate number of each car occupying a specific place being recorded. This information was used to determine how well a parking place was utilized. In addition, parking inventories were conducted to determine the entire supply/capacity of the parking lot. At each location, a questionnaire was given to drivers for obtaining their profile and determining additional factors such as walking distance to destination, simplicity of use, and search time that may influence parking demand. In both Location Lake View and MP Nagar, 50 questionnaires were distributed to get relevant information.

5 Data Analysis 5.1 Quantitative Data Analysis After performing the licence plate method and an in-out survey, parking statistics such as parking capacity, peak accumulation, average parking duration, peak parking saturation, parking volume, average turnover, and parking index were derived, all of which are critical for car parking analysis. Table 1 shows parking statistics for all five sections at selected places, whereas Figures 3 and 4 depict the MP Nagar and Lake View accumulation boards, respectively. Table 2 demonstrates that the peak parking saturation for Lake View (on-street and off-street) is greater than unity, resulting in spill over during peak hours. The accumulation curve (Figs. 3 and 4) shows that car accumulation grows gradually in Lake View on-street and off-street, peaking at 7:00 P.M., 7:30 P.M., and 8:00 P.M., respectively. On the other hand, Fig. 4 illustrates that parking demand in MP Nagar fluctuates, but it usually remains fully occupied most of the time. Parking at MP Nagar is strictly restricted, only cars equal to supply were allowed to park so that peak parking saturation never exceeds more than unity. So cars that do not have a

Qualitative and Quantitative Evaluation of Urban Car Parking …

283

Table 1 Parking characteristics at various locations in Bhopal City Location

Parking Peak Avg. parking Peak Parking Avg. Parking capacity accumulation duration (In parking volume turnover index hours) saturation

MP Nagar (on-street)

29

23

1.99

1

119

2.9

80.72

MP Nagar 29 (off-street)

29

2.6

1

82

2.96

76.87

Lake view (on-street)

19

21

0.8

1.1

138

7.26

52.87

Lake view 30 (off-street)

33

1.08

1.1

127

4.23

41.81

Table 2 Descriptive parking statistics Range

Minimum accumulation

Maximum accumulation

Mean

Std. deviation

Statistic

Statistic

Statistic

Statistic

Std. error

Statistic

Lake view off-street

31

2

33

11.27

1.213

8.490

Lake view on-street

17

4

21

9.02

0.721

5.048

MP Nagar off-street

28

1

29

22.89

1.342

9.001

MP Nagar on-street

28

1

29

22.89

1.342

9.001

Fig. 3 Accumulation curve of lake view

chance to park are compelled to look for parking elsewhere, resulting in increased traffic on the streets and illegal parking.

284

H. Chouhan et al.

Fig. 4 Accumulation curve of MP Nagar

5.2 Qualitative Data Analysis A questionnaire was distributed to drivers at the selected locations to formulate the disaggregate demand function. These demand functions help in determining the factors that influence the parking demand in selected locations. A total of 50 questionnaires are collected from each MP Nagar and Lake view. Parking demand has been expressed in terms of • Duration of parking (parking space usage per visit) in minutes. • Number of visits per month ( number of times space is used per month) Total parking space usage per month (Duration*Number of visits) in minutes. The independent parameters that are considered for analysis are age (in years), income (in lakhs), family size, search time (in minutes), ease time (in minutes), walking distance (in meters) and purpose of trip which is our categorical variable. Multiple linear regression with backward elimination in SPSS software is used for the analysis so that at the end final function will contain the variables only those that have significant impact on the parking demand. The list of various dependent and independent variables considered for the study are listed in Table 3. From Table 4 we can observe the influence of various independent variables on parking demand (Dependent variable). For instance, age is a statically significant variable for both MP Nagar and Lake View that regulates the parking demand. At MP Nagar parking demand increases with an increase in age while parking demand decreases with increases in the age at Lake View. Results of multiple regression indicate that walking distance, family size and purpose of trip are statically not significant to influence the parking demand at both locations. The significance level of all the models is within the permissible limit. The value of Durban-Watson test for all the models also falls between acceptable ranges (0 to 4), a value close to 2 indicates that no strong evidence of auto correction problem exists.

Qualitative and Quantitative Evaluation of Urban Car Parking …

285

Table 3 List of dependent and independent variables used in multiple regression List of variable

Nature

Duration of parking (in min.)

Dependent

Frequency of visits (visit per month)

Dependent

Total parking usage (duration of parking*visit per month)

Dependent

Family size (FS)

Independent

Age (A) (in years)

Independent

Income (I) (in lakhs)

Independent

Purpose of the trip (PT)

Independent

Search time (ST) (in min)

Independent

Ease time (ET) (in min)

Independent

Walking distance (WD) (in meters)

Independent

Table 4 Outcomes of multiple linear regression Study area Demand (duration)

Demand (visit per month)

Demand (monthly usage per month)

Lake view

MP Nagar

Model

D = −1.819(A) − 2.843(I) + 184.646

D = 2.027(A) + 27.994(I) − 6.175(ST) + 126.21

R2 value

0.747

0.873

Sig. of model

0.000

0.000

Durbin-Watson value

2.167

2.255

Model

D = −0.291(A) − 1 (ST) – 0.476(ET) + 24.012

D = 0.109(A) + 1.188(I) − 0.303(ST) + 8.859

R2 value

0.844

0.831

Sig. of model

0.000

0.000

Durbin-Watson value

1.849

1.849

Model

D = −61.442(A) + 3734.88

D = 72.776(A) + 895.343(I) − 184.364(ST) − 28.255(ET) − 145.825

R2 value

0.669

0.851

Sig. of model

0.000

0.000

Durbin-Watson value

2.250

2.103

6 Result and Conclusion A complete investigation of parking characteristics and behavioral variables was undertaken in this research project. The study’s findings are discussed in the preceding section. Peak parking saturation (on-street and off-street) at MP Nagar

286

H. Chouhan et al.

and Lake View was found to be 1.1 and 1.0, respectively, indicating the presence of spill over. Lake View parking facility experience spill over during peak hours (evening). In MP Nagar, on the other hand, peak parking saturation is never greater than one, as authorities do not allow cars to park more than the existing supply. Though it maintains declined parking in the parking facility but the cars that do get the opportunity to park has to roam on the streets to find the parking spot that increase the traffic on the roads and sometimes car driver ends up with illegal parking. Age, income, search time, family size, ease time, and purpose of trip are important predictors of parking demand at the selected locations, according to the results of the parking demand model created using the qualitative approach. Parking demand grows with age at MP Nagar but declines with age at Lake View. Parking demand appears to be increasing with rising income levels in MP Nagar. On the other hand, at Lake View parking demand (parking duration) was found to be decreasing with an increase of income level, while the influence of income on parking demand (visit per month and total parking usage) is found to be insignificant at lake view. Based on the findings of the study, it is apparent that existing parking facilities must be improved in order to maximize the utilization of parking space. To fulfill the parking demand in the designated study area, the parking supply must be increased. Due to the scarcity of space in most metropolitan locations, the introduction of multistorey parking facilities can help to enhance parking supply by using less space than surface parking. It was discovered that parking fees in MP Nagar were very low and independent of the parking duration. Therefore, many drivers tended to park for longer periods of time, affecting parking turnover and, as a result, limiting the use of parking facilities. Parking fees must be regulated by parking management in order for parking spaces to be used efficiently. Generally, it has been found that a car driver tends to park the vehicle as close as to the destination to reduce the walking distance, for that mostly on-street parking is the first preference of a car driver. Now this leads to the excessive accumulation of cars at parking during peak time, further resulting in spill over as well. To avoid this situation off-street parking needs to be promoted. Policy makers should design the parking. The outcomes of the study show that search time is the crucial predictor of parking demand in several study area. It was observed that sometimes due to a lack of information to the driver the search time to find an empty parking spot increases, and it discourages the driver to use off-street parking. It is suggested to install an information display at the parking to reduce the search time and a digital display should be installed at the entry of parking that will provide information about the availability of parking at the parking lot. The parking management and policy maker should draw the line in a way that for longer parking duration cars are diverted to off-street parking and to on-street parking for shorter duration. At most of the parking lot, it is observed that proper markings and signage for the parking were missing that is one of the significant reasons for the haphazard and indecorous parking. It is suggested that policy maker should follow the guidelines of the Indian

Qualitative and Quantitative Evaluation of Urban Car Parking …

287

Road Congress mentioned in “IRC: SP:12-2015” before designing a parking lot. Further to search parking or park illegally or park at on-street parking if available. According to our research, finding an empty parking spot in MP Nagar and Lake View takes 14 and 11 minutes, respectively. To encourage people to park off-street rather than on-street, search times must be minimized. It is advised that a digital or manual board be installed at the parking lot’s entrance, which offers information on the state of parking spaces and the number of available empty parking spaces. The installation of a parking meter is also a more effective technique to control the parking lot.

References Aderamo A, Salau K (2013) Parking patterns and problems in developing countries: A case from Ilorin, Nigeria, African J Eng Res 1:40–48 Ajeng C, Gim THT (2018) Analyzing on-street parking duration and demand in a Metropolitan City of a developing country: a case study of Yogyakarta City, Indonesia. Sustainability 10(3):591 Chakrabarti S, Mazumder T (2010) Behavioral characteristics of car parking demand: a case study of Kolkata. Institute of Town Planners, India Journal, pp 7–4 IRC:SP:12 (2015) Guidelines for parking facilities in urban areas. Indian Road Congress Moeinaddini M, Asadi-Shekari, Z, Ismail CR & Shah M (2013) A practical method for evaluating parking area level of service. Land Use Policy 33:1–10 https://doi.org/10.1016/j.landusepol. 2012.11.014 Parmar J, Das P, Azad F (2020a) Parking demand modelling using artificial neural network Parmar J, Das P, Dave SM (2020b) Study on demand and characteristics of parking system in urban areas: a review. J Traffic Transp Eng (english Edition) 7(1):111–124 Parmar J, Das P, Azad F, Dave S, Kumar R (2020c) Evaluation of parking characteristics: a case study of Delhi. Transp Res Procedia 48:2744–2756 Parmar J, Das P, Dave S. Development of level of service criteria for urban car parking system using clustering techniques Wang F, Ross CL (2018) Machine learning travel mode choices: Comparing the performance of an extreme gradient boosting model with a multinomial logit model. Transp Res Rec. https://doi. org/10.1177/0361198118773556 Ying M, Sun Y (2020) Discussion on parking management system based on parking behavior. In: 2020 international conference on urban engineering and management science (ICUEMS). IEEE, pp 513–516 Zhao C, Li S, Wang W, Li X, Du Y (2018) Advanced parking space management strategy design: an agent-based simulation optimization approach. Transp Res Rec 2672(8):901–910

Intercity Transportation

Estimation of Risk Exposure Index for Road Network in Landslide-Prone Areas P. N. Salini , P. Rahul, U. Salini , and Samson Mathew

Abstract Although there is significant information available on the direct risks associated with landslides and their hazards, there is less information reported on how landslides affect the road system and other indirect risks involved. Kerala, a state in southern India, has experienced a sizable number of landslides recently. According to the National Center for Earth Science Studies reports, Landslides frequently occur in Kerala’s 1,848 km2 of the Western Ghats, which extend along a steep slope. Several landslides occur in the mountain areas during the monsoon season, causing road collapse, silting of river beds, and severe damage to both public and private property. Due to its high population density (859 people per square kilometer), Kerala is one of the states in India that is most susceptible to disaster-related losses and damages. Although landslides cause fatalities, injuries, and property damage, they also cause a number of other indirect losses. When different road links and nodes are closed or blocked due to landslides, connectivity is lost, which exacerbates the losses and problems brought on by the landslides. This study aims to develop an indicator value that will express the traffic volume exposure during landslides while taking into account the road section’s vulnerability and risk. A database of attribute information on spatial and accessibility details, risk and vulnerability analysis, and finally the development of an index factor that will express the traffic exposure during landslide vulnerabilities in Munnar are all done as part of this study’s GIS mapping of disasteraffected and prone areas, road routes, and shelter locations. The findings aid in comprehending the region’s susceptibility to a natural disaster. Prioritization plans for effective and efficient road asset management will be prepared with the aid of the risk and risk exposure index assessment. Other landslide-prone or affected regions can estimate the risk index using the logical methodology presented in this paper. The decision makers and execution agencies will be better able to prioritize the projects and mitigation strategies as a result. In this study GIS mapping of disaster affected and prone areas, road routes and shelter locations is done along with an attribute database on spatial and accessibility details, risk and vulnerability analysis is done and finally developed an index factor which will express the traffic exposure during landslide P. N. Salini (B) · P. Rahul · U. Salini · S. Mathew KSCSTE- National Transportation Planning and Research Centre, Thiruvananthapuram, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_22

291

292

P. N. Salini et al.

vulnerabilities in Munnar. The results help to understand the vulnerability of the area during a natural calamity. The risk and risk exposure index assessment will help in preparing a prioritization plan for effective and efficient road asset management. The rational approach presented in this paper can also be replicated for the estimation of risk index in other landslide-prone or affected regions. This will further assist the executive agencies and decision makers in prioritizing the projects and mitigation measures. Keywords Risk assessment · Risk exposure index · Vulnerability · Landslide prone areas · Multi criteria decision assessment

1 Introduction Landslides are among the most common geological dangers that endanger life and property, particularly in hilly areas. The Western Ghats, the most prominent orographic feature of peninsular India, occupying about 47% of Kerala state, is vulnerable to various types of landslides (Achu et al. 2021). The landslide activities are prominent during the monsoon seasons due to the heavy rainfall. The most prevalent, recurring and disastrous type of mass movements noted in Kerala are the debris flows (Kuriakose et al. 2009), which is the sudden down slope movement of waterladen masses of soil and fragmented rock from mountainsides and entrain objects in their paths. Though heavy and persistent rainfall might be a triggering element, additional reasons include geological, morphological, physical, and human factors. The use of GIS for susceptibility mapping and hazard risk assessment assists in the identification of disaster-prone areas and is a preferred rational approach now days. Munnar region (latitude and longitude −10.0889° N, 77.0595° E) including Munnar town and its outskirts, which are most affected by natural disasters like landslides and flooding are demarcated for this study. Munnar lies in the second largest district of Kerala, Idukki lying in the Western Ghats and has a vast forest reserve area. Munnar is a hill station and a major tourist attraction of the southwestern Indian state of Kerala. Munnar is situated at around 1,600 m (5,200 ft) above mean sea level, in the Western Ghats mountain range. The recent flood and landslide havoc in Kerala from 2018 to 2020 had a devastating effect in hilly regions like Munnar, wherein most of the roadways were disrupted and the inundated town got completely cut off for days from the rest of Kerala. The results of a hazard exposure index and a road criticality index were combined to create a composite climate risk index in the study by Roux et al. (2019) using GIS processes and spatial data. Roman (2021) created the landslide hazard, vulnerability, and risk models using an Akaike Information Criterion. The Arithang ward in Gangtok City’s landslide hazard mapping was presented by Kaur et al. (2018). They took into consideration factors such as slope, aspect, curvature, geology, lineament, soil type, soil thickness, land use, land cover, water regime, and the Normalized Differential Vegetation Index (NDVI) for the mapping, as opposed to Garcia et al.

Estimation of Risk Exposure Index for Road Network …

293

(2017) who used additional anthropogenic, lithological, and hydrological factors in their study. When Savith et al. (2021) conducted their study at the Kavalapara landslide site in Kerala’s Malappuram district; they took into account factors like rainfall, local geomorphology, soil, slope, land use, and unsustainable practices. They created a map of the landslide site’s landslide hazards at a scale of 1:4000. A landslide susceptibility model was built using a logistic regression approach based on the triggering data in Akgun et al. (2012)s landslide risk assessment study on the Turkish city of Izmir. They took into consideration two distinct data groups, namely conditioning and triggering data. Winter et al. (2013) present the potential socioeconomic effects on the Scottish road network as well as the potential exposure of road users to debris flow hazards. Various studies were reported in the field of landslide risk and vulnerability analysis. Slope angle, Slope aspect, Slope curvature, Lithology, Distance to Lineament, Geomorphology, Soil Erosion, Rainfall, Distance from Drainage, Land use, Distance to Road, Population, and Population Density, etc. are the important influencing factors taken into account in the risk and vulnerability mapping. These parameters’ weights, which differ from location to location, were established using scientific methods. Minimal research has been published in the field of exposure of traffic or road users to natural hazards.

2 Scope and Objective This study has the broad scope of identifying the elements of risk, vulnerability and resilience associated with the transport infrastructure and network of the Munnar region. The major objectives of the study are: • To conduct risk and vulnerability analysis for the study region as well as for the existing road network in the region, accounting for the risks and vulnerabilities due to natural hazards like landslides • To estimate the Risk Exposure Index (REI) of the potential traffic plying along the road network in the study region

3 Methodology The methodology adopted in this study includes data collection, Vulnerability Analysis, Risk Assessment and Index Estimation. The steps involved are explained in the following sections.

294

P. N. Salini et al.

3.1 Data Collection The data collection includes Preliminary surveys including general condition surveys, Traffic volume count surveys and secondary data collection from different sources. The Secondary data were collected from regional offices of PWD, Irrigation, LSGD and Local Self Govt offices like Munnar panchayat, Devikulam Panchayat, Pallivasal Panchayat, Kerala State Disaster Management Authority (KSDMA) and from the R&D center named National Center for Earth Science Studies (NCESS). Geographical Information System and Multi-Criteria Decision Analysis approach were used for the vulnerability analysis and risk assessment of the region by considering the physical, environmental, social and economic factors of the area. Terrain parameters like topographical, geological, anthropogenic, meteorological and social parameters like population, population density, age group, unemployment, illiteracy, occupancy, building density etc. are used for the vulnerability estimation. Risk assessment maps are prepared considering the vulnerability maps and Hazard zone maps of the area. The traffic exposure index was developed by considering the risk estimated on the road sections and the traffic volume observed on respective sections. In this study, mainly the GIS and traffic data were used for risk and vulnerability analysis and estimation of the risk exposure index. The GIS data such as shuttle radar topographic mission (SRTM) Digital Elevation Model of 30 m resolution is used for preparation of hydrological maps which has been an input in the risk and vulnerability analysis. Road networks were digitized from the open street map as a shape file in ArcMap. Soil erosion, Soil Lineament, Lithological, and landuse land cover maps pertaining to the study area were procured from National Center for Earth Science Studies. Population data was collected from Census 2011. A flood map of Kerala prepared by National Remote Sensing Center (NRSC) for the year 2018 and 2019 and a map of landslide affected areas in 2018 from the Kerala State Disaster Management Authority (KSDMA) were used for the study. Traffic volumes are estimated from the video data collected from the field. Video recordings were conducted for one day duration during a week day on each of the roads from a vantage point. The PCU values provided in Indo-HCM (Brouwer et al. 2007) guidelines have been used for traffic volume estimation. Low traffic is observed on some of the minor roads passing through residential areas and through estates, and an average of 75 PCU/hr is adopted for these roads.

3.2 Risk Assessment Risk assessment maps were prepared using the ArcGis software. Landslide risk is determined as the product of the hazard and vulnerability of the study area. Equation 1 is used to calculate the risk of the area. Landslide hazard zone map of Munnar region obtained from National Center for Earth Science Studies is used for the study.

Estimation of Risk Exposure Index for Road Network …

Risk = Hazard × Vulnerability

295

(1)

A combination of Geographical Information System (GIS) and Multi-Criteria Decision Assessments (MCDA) is used for vulnerability analysis. Vulnerability assessment of landslides for a region encompasses of two phases. First phase is to calculate the physical environmental vulnerability of the area where different spatial data based on the physical infrastructure of the existing scenarios are used. Independent of the social and economic patterns of the habitation, this vulnerability varies over the study area, depending on the topography and conditions of the terrain. The second phase composes of an estimation of socio economic vulnerability of the area. Socio economic vulnerability is used to describe and analyze the exposure and coping mechanisms of groups and individuals to environmental risks (Saaty 1980). Based on expert knowledge, extended field observations, literature review, and collection of available landslide historical data eleven factors were eventually selected for physical vulnerability analysis and eight factors were selected for socio economic vulnerability analysis which are listed out in the following sections. The weightage of factors was estimated based on Analytical Hierarchy Process (AHP), in which all factors are compared pair wise in terms of the intensity of their importance using a continuous 1–9 point scale according to Saaty (1980).

3.3 Risk Exposure Index Estimation The methodology framed to estimate the risk exposure index (REI) is shown in Fig. 1. The traffic volume observed on the roads and the risk value associated with the roads are the parameters required for the estimation of REI. In the estimation of REI, the first step is the identification of the roads susceptible to landslide. These road sections are identified from the risk assessment maps. The degree of risk associated with these road sections are demarcated from a scale of one to nine. Based on the analysis, each road is found to have different degrees of risk for different lengths of the road. The risk value associated with each road is estimated by taking the sum of the product of degree of risk and the corresponding road length of the vulnerable sections in the road. Based on the literature review and considering the weights of contributing factors of risk exposure the risk exposure index is calculated as given in Eq. 2. Risk Exposure Index = 0.8 ∗ (Risk value) + 0.2 ∗ (Traffic volume)

(2)

296

P. N. Salini et al.

Fig. 1 Flow chart showing Risk Exposure Index estimation

4 Results and Discussions 4.1 Risk Assessment The factors considered for physical and social vulnerability analysis and their weights estimated based on AHP is given in Table 1. The geological factors is observed to have the highest weightage in physical vulnerability analysis. Risk map of the study area is prepared by risk and vulnerability analysis using the available hazard zone map and developed vulnerability map (given in Fig. 2). It is identified from the prepared risk map that about 82.73 km of length of the road are in high risk category in the study area.

4.2 Risk Exposure Index All the road links existing in the Munnar region is mapped and the risk due to landslides associated with each link is evaluated from the risk maps prepared for the study area as given in Fig. 3. Considering this risk associated with the roads and the prevalent traffic volume plying on these roads the risk exposure index is estimated. The obtained risk exposure index for various roads on the study area is given in Table 2. The GIS map showing the road network and the risk exposure index obtained for various roads on the road network is shown in Fig. 4. The roads with higher risk index values need to be considered with higher priority for intervention and improvement of roads and proper landslide countermeasures have to be implemented along these road sections for development as a resilient transport network which can sustain the transport needs in times of disasters like landslides. The Munnar-top station highway appears to have a higher index value, which

Estimation of Risk Exposure Index for Road Network …

297

Table 1 Table showing weights of different factors estimated Physical vulnerability analysis

Influencing factor

Data layer

Weightage (%)

Topographical criteria

Slope angle

10

Geological criteria

Hydrological criteria Human induced criteria Social vulnerability analysis

Deomographic criteria

Human induced criteria

Slope aspect

6

Slope curvature

6

Lithology

16

Distance to lineament

13

Geomorphology

16

Soil erosion

16

Rainfall

10

Distance from drainage

1.5

Land use

4

Distance to road

1.5

Population

34

Population density

24

Working population

5

Non working population

4

Literates

2

Illiterates

2

Land use

18

Distance to road

11

is an important highway of Munnar region with higher vehicular traffic. MunnarUdumalpet road and Munnar Dhanushkodi road are also having higher risk index values calling for priority of intervention and counter measures for resilience. Bison valley road, Gundumalai Hill road, Nallathani road, Pazhaya Munnar Chokanadu Devikulam Road, Peryiyavari Bridge to Engineering college road, Pothanmedu road, and Ropeway station road are observed to be with comparatively lower risk. These roads in lower risk areas and the alternate routes in low risk areas identified by this study for up gradation could be considered as a safe passage for evacuation during times of landslide and for rescue purposes. The alternate routes located in low risk areas and identified for up gradation are as shown in Fig. 5.

298

Fig. 2 Prepared risk assessment map of the study region

P. N. Salini et al.

Estimation of Risk Exposure Index for Road Network …

Fig. 3 Prepared road risk map of the study region

299

300

P. N. Salini et al.

Table 2 Risk Exposure Index (REI) calculated for different roads (in descending order of REI) No

Road

Risk value

Risk Exposure Index (REI)

1

Munnar-Top station highway (SH 18)

14.9

88

2

Eravikulam Udumalpet road (SH 17)

35.2

86

3

Munnar Eravikulam road (SH 17)

24.3

77

4

Munnar-Madurai-Dhanushkodi Road (NH 85)

21.7

61

5

Anamudi View point to Reserve Forest Road

42.6

49

6

Gundumalai Road

37.9

45

7

Muvattupuzha-Munnar (NH 85)

5.1

44

8

Mankulam road

9

Tea Museum Road

10

Mattupetty Dam-SH 17 road

16.6

38

5.8

36

23.6

34

11

Valparai Gundumalai Trial Road

20.3

31

12

Periyavarai top murugan temple to SH 18 road

17.9

29

13

Old AM Road Mankulam-Munnar Trail

16.3

28

14

Kallar Estate Kaliamman Temple road

16.4

28

15

New colony road

11.5

24

16

Rajamala Road

11.6

24

17

Eravikulam National Park Road

11.2

24

18

Engineering college road

4.4

23

19

Munnar Colony road

10

23

20

Letchmi Estate Road

10.2

23

21

Amman temple road

9.1

22

22

Karuppuswamy temple road

7.9

21

23

Pooyankutty Munnar Trail

6.4

20

24

Mankulam Nirar Dam Road

3.3

18

25

Periyavaraimurugan temple to Periyavarai bridge road

3.8

18

26

Talayvar Tea Factory Road

4.1

18

27

Devikulam Police Station-Old Devikulam temple road

3.8

18

28

Kallar Estate Road

1.8

16

29

Lakshmi Amman Temple Road

1.3

16

30

Cholamalai Estate Road

1.2

16

Estimation of Risk Exposure Index for Road Network …

Fig. 4 Risk Exposure Index Map of the study area

301

Fig. 5 Alternate road routes located in low risk areas proposed for upgradation

302 P. N. Salini et al.

Estimation of Risk Exposure Index for Road Network …

303

5 Conclusion This paper discusses a rational approach for landslide risk assessment and development of a risk exposure index of the potential traffic plying along the road network in Munnar region in the southern state of India, Kerala, due to natural disasters. The region has experienced several landslides in recent years during the monsoon season. Risk and vulnerability analysis of the region is carried out using GIS tools based on Multi-Criteria Decision Approach, thereby evolving a rational and scientific approach which could be replicated for research and development studies on other regions too. Risk and vulnerability maps are prepared for the study region clearly demarcating the extent and severity of risk and vulnerabilities in the region. This paper brings out certain findings that will aid in the development of a prioritized strategy for effective and efficient road asset management. Based on the risk exposure index of the potential traffic estimated for different roads of study area, the priority list of roads and the strategy can be prepared for interventions like the road upgradation and improvement. Also, the roads in low risk area are identified which could be used for evacuation purposes in times of emergency. This will help the authority and decision makers while developing a resilient transport network for Munnar region. The rational and scientific approach evolved for this study could be replicated for similar studies in other regions too.

References Achu AL, Joseph S, Aju CD, Mathai J (2021) Preliminary analysis of a catastrophic landslide event on 6 August 2020 at Pettimudi, Kerala State, India. Landslides 18(4):1459–1463 Akgun A, Kıncal C, Pradhan B (2012) Application of remote sensing data and GIS for landslide risk assessment as an environmental threat to Izmir city (west Turkey). Environ Monit Assess 184(9):5453–5470 Brouwer R, Akter S, Brander L, Haque E (2007) Socioeconomic vulnerability and adaptation to environmental risk: a case study of climate change and flooding in Bangladesh. Risk Anal An Int J 27(2):313–326 Kaur H, Gupta S, Parkash S, Thapa R (2018) Application of geospatialtechnologies for multi-hazard mapping and characterization of associated risk at local scale. Ann GIS 24(1):33–46 Kuriakose SL, Sankar G, Muraleedharan C (2009) History of landslide susceptibility and a chorology of landslide-prone areas in the Western Ghats of Kerala, India. Environ Geol 57(7):1553–1568 Murillo-García FG, Rossi M, Ardizzone F, Fiorucci F, Alcántara-Ayala I (2017) Hazard and population vulnerability analysis: a step towards landslide risk assessment. J Mt Sci 14(7):1241–1261 Quesada-Román A (2021) Landslide risk index map at the municipal scale for Costa Rica. Int J Disaster Risk Reduct 56:102144 Roux Le A, Khuluse-Makhanya S, Arnold K, Engelbrecht F, Paige-Green P, Verhaeghe B (2019) A framework for assessing the risks and impacts of rural access roads to a changing climate. Int J Disaster Risk Reduct 38:101175 Sarun S, Vineetha P, Rajesh R, Sheela AM, Anil Kumar R (2021) Post landslide investigation of shallow landslide: a case study from the Southern Western Ghats India. Disast Adv 14(7):52–59

304

P. N. Salini et al.

Saaty T (1980) The analytic hierarchy process: planning, priority setting, resource allocation. McGraw-Hill, New York Winter MG, Harrison M, Macgregor F, Shackman L (2013) Landslide hazard and risk assessment on the Scottish road network. Proceed Inst Civ Eng Geotech Eng 166(6):522–539

Mode Choice Behaviour of Textile Shippers in India V. Ansu

and M. V. L. R. Anjaneyulu

Abstract India ranks third in CO2 emissions. Road transport accounts for 61% of interregional shipments in India, compared to 30% for rail, resulting in excess fuel consumption. This study aims to improve interregional textile transport in India. India’s textile industry contributes 14% to industrial output, 4.5% to GDP and employs over 35 million people. Interregional textile freight traffic totals 10,274 million km by road and 9.5 million km by rail. Hence, significant potential exists for a mode shift from road to rail to save fuel. Mode choice modelling helps to identify significant variables and develop energy efficiency strategies for mode shift. A revealed preference survey was conducted in Kerala to gather disaggregated data on shipments. In addition to the observed variables, shippers’ preferences on mode choice were also collected. Factor analysis of these preferences yielded two qualitative factors: safety and reliability rating factors. Latent class analysis disclosed heterogeneity among shippers. Class 1 firms are well-established without truck ownership. They’ll have to hire trucks or use rail. Class 2 includes firms under 15 years old, many of which own trucks. Mode choice modelling revealed that speed, reliability, shipment value, employee count, service quality, shipment frequency, packaging quality, and tracking capability all play a role in mode selection. Since textiles are expensive and non-bulky, transportation cost has not emerged as a significant variable. Rail mode share could be increased by enhancing the speed and reliability of on-time delivery. Keywords Freight transport · Mode choice modelling · Latent class analysis

V. Ansu (B) · M. V. L. R. Anjaneyulu Centre for Transportation Research, National Institute of Technology Calicut, Calicut, Kerala, India e-mail: [email protected] M. V. L. R. Anjaneyulu e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_23

305

306

V. Ansu and M. V. L. R. Anjaneyulu

1 Introduction In India, highways and railways account for 61% and 30% of interregional freight, respectively, resulting in substantial fuel consumption as road transport uses more fuel than rail (Planning Commission of India 2014). India ranks third in CO2 emissions (International Energy Agency 2020). As road and rail transport account for 91% of interregional freight, this study concentrates on these two transport modes. This study identifies India’s interregional textile transportation mode choice characteristics to increase the share of fuel-efficient transport modes. Dresses are purchased as a fundamental necessity and for festivals and special occasions. India’s textile industry is one of the world’s largest (Dhanabhakyam 2007). Apart from providing necessities, it plays a crucial role in the country’s economic progress. In 2010, India’s textile industry generated $63 billion, accounted for approximately 14% of industrial production, 4.5% of the country’s gross domestic product, and directly employed over 35 million people (Mukherjee 2015). Interregional textile freight traffic in India totals 10,274 million km by road and 9.5 million km by rail (Planning Commission of India 2014). As a result, there is significant scope for the mode shift of textiles from road to rail. However, the causes for mode selection must be understood before implementing strategies/policies for mode shift. The mode selection characteristics of textiles are distinct from other commodities. The study concentrates on the transport mode choice of textiles as studies on the textile mode selection characteristics are limited (Arunotayanun and Polak 2011; Gursoy 2010; García-Menéndez et al. 2004; Norojono and Young 2003). Thus, studying the mode choice behaviour of shippers of textiles is necessary. The study methodology entails conducting a literature review, developing a questionnaire, collecting data, analysing the data, and developing mode choice models.

2 Literature Review Among the stages of freight demand modelling, mode selection is the most policysensitive (Brooks et al. 2012). It helps identify the characteristics that influence mode selection (Jong et al. 2004; Tavasszy and Jong 2014). The transport mode selected by the shippers is based on various characteristics. It is critical to determine what variables influence freight mode selection (Tavasszy and Jong 2014; Jiang et al. 1999). This was accomplished through a review of prior research. In most previous studies, the speed/time of transportation was found to be a significant factor in freight mode selection (García-Menéndez et al. 2004; Abdelwahab 1998; Beagan et al. 2007; Bergantino et al. 2013; Chang and Thai 2017; Cullinane and Toy 2000; Evers et al. 1996; Grue and Ludvigsen 2006; Jensen et al. 2019; Kalahasthi et al. 2022; Kim et al. 2017; Mitra and Leon 2014; Moschovou and Giannopoulos 2012; Murphy et al. 1997; Premeaux 2002; Puckett and Rasciute 2010; Rich et al. 2009; Román et al. 2017; Shen and Wang 2012; Tortum et al. 2009; Wichitphongsa and Ponanan

Mode Choice Behaviour of Textile Shippers in India

307

2022). The second major factor is the cost of transportation (Arunotayanun and Polak 2011; García-Menéndez et al. 2004; Abdelwahab 1998; Beagan et al. 2007; Chang and Thai 2017; Cullinane and Toy 2000; Evers et al. 1996; Jensen et al. 2019; Kim et al. 2017; Premeaux 2002; Puckett and Rasciute 2010; Rich et al. 2009; Román et al. 2017; Tortum et al. 2009; Wichitphongsa and Ponanan 2022; Bontekoning et al. 2004; Marcucci and Scaccia 2003; Reis 2014; Samimi et al. 2011; Wiegmans and Konings 2015). The other significant variables in decreasing order of appearance are reliability, safety, frequency of transport mode, weight, distance, value, equipment availability, pickup and delivery time, flexibility, service quality, tracing/tracking facility, commodity type, waiting time, shipment frequency, fuel cost, and shelf life. (Arunotayanun and Polak 2011; García-Menéndez et al. 2004; Jiang et al. 1999; Beagan et al. 2007; Bergantino et al. 2013; Chang and Thai 2017; Cullinane and Toy 2000; Evers et al. 1996; Grue and Ludvigsen 2006; Jensen et al. 2019; Kalahasthi et al. 2022; Kim et al. 2017; Mitra and Leon 2014; Moschovou and Giannopoulos 2012; Murphy et al. 1997; Premeaux 2002; Puckett and Rasciute 2010; Rich et al. 2009; Román et al. 2017; Shen and Wang 2012; Tortum et al. 2009; Wichitphongsa and Ponanan 2022; Bontekoning et al. 2004; Marcucci and Scaccia 2003; Reis 2014; Samimi et al. 2011; Wiegmans and Konings 2015; Holguin-Veras 2002; Norojono and Young 1993, 2010; Murakami and Matsuse 2014; Park and Suh 2011; Piendl et al. 2019; Shinghal and Fowkes 2002; Wang et al. 2013). Shippers of textiles are sensitive to transit time by big trucks, and there is a decline in the utility for small trucks (Arunotayanun and Polak 2011). Due to shipper heterogeneity, not all shippers behave similarly under identical situations when it comes to transport mode selection (Arunotayanun and Polak 2011; Bergantino et al. 2013; Kim et al. 2017; Román et al. 2017; Piendl et al. 2019; Chu 2014; Duan et al. 2017; Marcucci et al. 2017; Piendl et al. 2017). As a result, it is preferable to identify and model the latent groups of decision-makers separately (Astroza et al. 2019). Magidson & Vermunt (2002) found that latent class analysis is more effective at identifying heterogeneity than K-means clustering. As a result, this study employs latent class analysis to identify the heterogeneous groups among shippers of textiles. The probability of selecting one alternative over another is estimated using discrete choice models. Logit models are frequently used to model freight mode selection (Holguin-Veras 2002; Catalani 2001; Golias and Yannis 1998; Siridhara et al. 2019). Hence, this study makes use of logit modelling. Equation 1 gives the utility equation for choosing an option j regarding the referent choice, k.  Uj = α + βi Xi (1) where, α = intercept βi = Variable coefficient, Xi Xi = The variable’s value, X i= 1, 2, 3…

308

V. Ansu and M. V. L. R. Anjaneyulu

The utility equation for the referent choice k is given by Eq. 2. Uk = 0

(2)

The Logit model gives the probability of choosing j and k as given by Eqs. 3 and 4. If the calculated value of the probability of a choice is greater than 0.5, the respondent is more likely to choose that option. Pj = Pk =

eUj eUj = eUk + eUj 1 + eUj

(3)

1 eUk = Uj +e 1 + eUj

(4)

eUk

3 Questionnaire Design and Data Collection The questionnaire included variables identified during the literature survey, the stakeholders’ workshop, and interactions with shippers. The cognitive pre-testing survey was used to collect data on survey responses and to determine whether the questions accurately measured the study’s intended construct. After collecting the data, it is used to address problematic questions in the questionnaire before conducting the final survey. Following the pilot survey, the questionnaire was refined by omitting insignificant variables. Additional variables were also included in the questionnaire after the pilot survey. The final questionnaire included quantitative and qualitative variables influencing freight mode selection. Rail shipments need road transport for door-to-door connections, referred to in this study as “rail”. Shippers’ preferences for variables were recorded on a five-point Likert scale as not at all important, slightly important, moderately important, very important, and extremely important, with weights ranging from 1 to 5. Disaggregated models require large amounts of data that are not publicly available in India. Hence, a revealed preference survey was conducted among shippers in Kerala who import and export freight to various regions of the nation. Kerala is a southern Indian state with a population of 33.4 million people (Government of India National Informatics Centre 2011). The revealed preference survey of shippers was administered with the help of trained enumerators by the direct interview method. Enumerators visited the commercial hubs and identified the shippers. Enumerators explained the importance and need of the survey. Then the details were recorded in the questionnaire by interviewing the shippers willing to provide the data. After cleansing the data, 1999 cases were available for modelling. There were 297 rail shipments and 1702 road shipments in the data. This study excluded shipments within urban areas, entire trainloads of commodities, and overseas shipments due to

Mode Choice Behaviour of Textile Shippers in India

309

the limited scope of mode choice. To enable mode selection, origins and destinations should be readily accessible via the modes of transport examined. Hence, hilly districts without train connectivity are also excluded.

4 Data Summary The summary statistics for the metric variables are shown in Table 1. According to the preliminary analysis of non-metric data, 97% of road shippers and 39% of rail shippers believe their transport mode is secure. While 87% of road shippers and 53% of rail shippers believe their mode of transport’s overall service quality is good. 93% of road shippers and 92% of rail shippers consider the reliability of on-time delivery of the transport mode normal. The shippers’ preferences were analysed by taking the weighted average of the importance ratings to determine the importance of each variable. The critical variables of mode selection in decreasing order are safety, time/speed, cost, reliability, packaging quality, accessibility, mode availability, shipment value, shipment weight, the capacity of the mode, and shipment frequency. Table 1 Summary statistics for textile items Variable

Min

Max

Mean

Std. deviation

Age of firm, years

1

131

8.6

11.3

Number of employees

1

135

4.2

7.9

Number of trucks

0

9

0.04

0.5

Distance of origin from the rail station, km

1.0

24

5.1

2.9

Distance of destination from the rail station, km

0.3

41

7.4

9.5

Distance, km

86

2,785

1,113.0

691.3

Shelf life, days

60.0

365

183

100

Shipment weight, t

0.015

21.00

0.17

0.89

Shipment value, INR

1,000

22,00,000

97,999

1,57,987

Shipment frequency per month

0.05

100

2.4

4.4

Transportation cost, INR

100

1,22,500

1,471

4,988

Transportation time, h

3.0

360

111

65.3

Pickup time, h

0

48

5.3

7.1

Delivery time, h

0

16

2.0

1.8

The capacity of mode, t

2.0

40

17.5

10.5

Cost of loss, INR

0.0

45,000

123

1,574

310

V. Ansu and M. V. L. R. Anjaneyulu

5 Qualitative Factors of Preference Ratings The qualitative factors were constructed using factor analysis of the preference ratings. These factors enable qualitative behaviour to be incorporated into modelling. Murphy et al. (1991) performed a factor analysis of preferences to determine the factors influencing freight mode selection. Danielis et al. (2005) observed that logistics managers prioritise qualitative factors such as timeliness, safety, and reliability over cost. The correlation analysis of the variable preference ratings is presented in Table 2. Factor analysis of the preference ratings showed that Bartlett’s sphericity test was significant. With a value of 0.819, the Kaiser–Meyer–Olkin measure was meritorious. Three factors were identified using the scree plot criterion. On the other hand, the Kaiser stopping criterion was accepted because it identified two factors that accounted for 74.68% of the variance. Factors 1 and 2 contribute 44.47 and 30.21% to the variance. Table 3 illustrates the rotated factor matrix. Factor 1 consists of ratings for packaging quality, shipment value, the capacity of the mode, shipment weight, availability of the transport mode, and safety. Since this factor is primarily associated with packaging quality and shipment value, where safety is also a component, it is referred to as the ’safety’ rating factor. Factor 2 includes ratings for reliability, transportation time, shipment frequency, and distance. Since reliability is the main concept of this factor, it is referred to as the rating factor of ’reliability’.

6 Heterogeneity Among Shippers To ascertain shipper heterogeneity, the latent class analysis considered several combinations of variables of shippers’ attributes. Two classes are the optimal size, with a classification error of 0.0106. The chosen model considers the shipping company’s age, the number of trucks it owns, and its reliance on its trucks for transport. The L2 statistics was 14.40, and the BIC value was 6049.14. The latent class analysis revealed two latent classes of shipping companies. Class 1 firms are well-established without truck ownership. These businesses will have to carry their goods via rail or hire trucks because none own trucks. Latent class 2 firms include younger firms aged less than 15 years, with many owning and operating trucks. Class 1 has 1959 cases, and Class 2 has 40 cases.

0.51

0.35

0.34

0.29

0.40

0.09

− 0.08

− 0.08

− 0.08

Shipment frequency

Shipment weight

Shipment value

Packaging quality

Availability of mode

0.48

0.17

0.04

− 0.08

Capacity of mode

Safety

0.73

0.57

Reliability

0.84

0.48

Time

1.00

1.00

Distance

Shipment frequency

Distance

Rating of variable

0.87

0.38

0.16

0.25

0.53

0.65

0.67

1.00

Shipment weight

Table 2 Correlation analysis of preference ratings

0.71

0.74

0.05

0.13

0.70

0.92

1.00

Shipment value

0.71

0.80

0.05

0.13

0.68

1.00

Packaging quality

0.62

0.46

0.03

0.09

1.00

Availability of mode

0.18

0.05

0.86

1.00

Time

0.11

− 0.01

1.00

Reliability

0.44

1.00

Safety

1.00

Capacity of mode

Mode Choice Behaviour of Textile Shippers in India 311

312 Table 3 Rotated component matrix

V. Ansu and M. V. L. R. Anjaneyulu

Variable

Component

Packaging quality

0.94

Shipment value

0.94

Capacity of mode

0.84

Shipment weight

0.79

Availability of mode

0.78

Safety

0.76

1

2

Reliability

0.93

Transportation time

0.93

Shipment frequency

0.84

Distance

0.70

7 Mode Choice Modelling The variable speed is estimated since the time required to transport freight is related to the distance travelled. The model was constructed by incorporating observed variables, qualitative factors, and estimated variables that took shipper heterogeneity into account. The dependent variable is the mode of transport. The road is selected as the reference category in the model with a 95% confidence interval. Variables with a significance value of less than 0.05 are significant. A stepwise modelling approach was used to eliminate insignificant variables one by one to obtain the best model. The mode choice model for latent class 1 shippers is presented in Table 4. The Nagelkerke R-squared value is 0.619, and the Chi-square value is 850.38, which is significant. Mode choice modelling revealed that the significant variables of mode choice are speed, rating factor of reliability, shipment value, number of employees, overall Table 4 Mode choice model For rail (Reference Category: Road) Coefficient B Std. error Wald statistic Sig

Exp(B)

Intercept

4.345

0.769

31.892

0.000

Speed

0.022

0.006

12.505

0.000 1.023

Rating factor of reliability

1.429

0.149

92.537

0.000 4.173

Shipment value per kg

0.001

0.000

10.487

0.001 1.001

Number of employees

−0.129

0.035

13.152

0.000 0.879

Overall service quality

−1.113

0.194

32.784

0.000 0.329

Shipment frequency

−0.098

0.040

5.894

0.015 0.906

Packaging quality

−0.639

0.204

9.815

0.002 0.528

Tracking facility

−3.223

0.217

221.354

0.000 0.040

Mode Choice Behaviour of Textile Shippers in India

313

service quality, shipment frequency, packaging quality, and tracking facility. A variable with a positive coefficient indicates that as the variable’s value increases, the modelled choice (rail) is more likely to occur than the reference group (road). So to increase the share of rail, the policy variables with positive coefficients should be improved. Variables with values close to zero will have little effect on mode shift. As a result, shipment value has little influence on mode shift as its coefficient is 0.001. The coefficient of the rating factor of reliability is 1.429, and the speed is 0.022. Hence, the probability of respondents choosing rail over road increases as the reliability of on-time delivery and speed of the mode of transport increases. The coefficient of the overall service quality and tracking capabilities are negative, indicating that shippers believe rail service quality and tracking are inferior compared to road. Since most textile shipments are light and not bulky, transportation costs have not emerged as a critical consideration in mode choice. The data show that the average shipment weight is 0.17 tonnes. According to the model, shippers are more inclined to pick road transport as the number of employees (firm size) increases. This is because when the firm’s size grows, so does the shipment weight (observation based on data). Small shipments of textiles, on the other hand, can be conveyed through truck or train parcel service. Hair et al. (2009) determined that the minimum sample size should be 20 times the number of variables in the model, with a minimum of 200 samples required for normality. Since latent class 2 contains only 40 cases, mode choice modelling is impossible for this class. Class 2 includes newer firms with a high proportion of truck ownership and their truck usage. Without considering heterogeneity, this latent class would not have been identified and modelled along with class 1. Hence modelling considering heterogeneity is better than modelling all the latent classes of shippers together.

8 Conclusion A factor analysis of preference ratings revealed two qualitative factors. The safety factor considers the ratings of package quality, shipment value, the mode’s capacity, shipment weight, mode availability, and safety. The reliability factor comprises ratings for reliability, transportation time, frequency of shipment, and distance. The latent class analysis revealed that there exists heterogeneity among shipping firms. Class 1 firms are well-established and have no truck ownership. Class 2 firms include younger firms, with many owning and operating trucks. Speed, reliability, shipment value, number of employees, overall service quality, shipment frequency, packaging quality, and tracking facility are significant characteristics of textile mode selection. Unlike bulk commodities such as building materials and food grains, transportation cost has not emerged as a significant variable in obtaining the best model in textile transport. Mode choice modelling revealed that rail mode share could be increased by enhancing reliability and speed to minimise energy consumption.

314

V. Ansu and M. V. L. R. Anjaneyulu

References Abdelwahab WM (1998) Elasticities of mode choice probabilities and market elasticities of demand: evidence from a simultaneous mode choice/shipment-size freight transport model. Transp Res Part e: Log Transp Rev 34(4):257–266 Arunotayanun K, Polak JW (2011) Taste heterogeneity and market segmentation in freight shippers’ mode choice behaviour. Transp Res Part e: Log Transp Rev 47(2):138–148 Astroza S, Garikapati VM, Pendyala RM, Bhat CR, Mokhtarian PL (2019) Representing heterogeneity in structural relationships among multiple choice variables using a latent segmentation approach. Transportation 46(5):1755–1784 Beagan D, Fischer M, Kuppam A (2007) Quick response freight manual II. Federal Highway Administration, U.S. Department of Transportation, FHWA-HOP-08-010 Bergantino AS, Bierlaire M, Catalano M, Migliore M, Amoroso S (2013) Taste heterogeneity and latent preferences in the choice behaviour of freight transport operators. Transp Policy 30:77–91. https://doi.org/10.1016/j.tranpol.2013.08.002 Bontekoning YM, Macharis C, Trip JJ (2004) Is a new applied transportation research field emerging?––a review of intermodal rail–truck freight transport literature. Transp Res Part a: Policy Pract 38(1):1–34 Brooks MR, Puckett SM, Hensher DA, Sammons A (2012) Understanding mode choice decisions: a study of australian freight shippers. Marit Econ Log 14(3):274–299 Catalani M (2001) A model of shipper behaviour choice in a domestic freight transport system. In: 9th world conference on transport research, Seoul, 2001, pp 22–27 Chang C-H, Thai VV (2017) Shippers’ choice behaviour in choosing transport mode: the case of South East Asia (SEA) region. Asian J Shipp Log 33(4):199–210. https://doi.org/10.1016/J. AJSL.2017.12.003 Chu H-C (2014) Exploring preference heterogeneity of air freight forwarders in the choices of carriers and routes. J Air Transp Manag 37:45–52. https://doi.org/10.1016/J.JAIRTRAMAN. 2014.02.002 Cullinane K, Toy N (2000) Identifying influential attributes in freight route/mode choice decisions: a content analysis. Transp Res Part e: Log Transp Rev 36(1):41–53 Danielis R, Marcucci E, Rotaris L (2005) Logistics managers’ stated preferences for freight service attributes. Transp Res Part e: Log Transp Rev 41(3):201–215 de Jong G, Gunn H, Walker W (2004) National and international freight transport models: an overview and ideas for future development. Transp Rev 24(1):103–124 Dhanabhakyam M (2007) Indian textile industry—an overview—free industry articles— Fibre2fashion.com. http://www.fibre2fashion.com/industry-article/2363/indian-textile-ind ustry-an-overview. Accessed 20 Mar 2022 Duan L, Tavasszy L, Peng Q (2017) Freight network design with heterogeneous values of time. Transp Res Proc 25:1144–1150. https://doi.org/10.1016/J.TRPRO.2017.05.127 Evers PT, Harper DV, Needham PM (1996) The determinants of shipper perceptions of modes. Transp J 36(2) García-Menéndez L, Martínez-Zarzoso I, De Miguel DP (2004) Determinants of mode choice between road and shipping for freight transport: evidence for four Spanish exporting sectors. JTEP 38(3):447–466 Golias J, Yannis G (1998) Determinants of combined transport’s market share. Transp Logist 1(4):251–264 Government of India National Informatics Centre (2022) Kerala population sex ratio in Kerala literacy rate data 2011–2022. https://www.census2011.co.in/census/state/kerala.html. Accessed 18 March 18 Grue B, Ludvigsen J (2006) Decision factors underlying transport mode choice in international European freight transport Gursoy M (2010) A decision supportive method for multimodal freight transport mode choice: an example from Turkey (research note)

Mode Choice Behaviour of Textile Shippers in India

315

Hair JJF, Black WC, Babin BJ, Anderson RE (2009) Multivariate data analysis, 7th ed. Pearson Holguin-Veras J (2002) Revealed preference analysis of commercial vehicle choice process. J Transp Eng 128(4):336–346 International Energy Agency (2020) India 2020—energy policy review. www.iea.org/t&c/ Jensen AF et al (2019) A disaggregate freight transport chain choice model for Europe. Transp Res Part e: Log Transp Rev 121:43–62. https://doi.org/10.1016/j.tre.2018.10.004 Jiang F, Jonhson P, Calzada C (1999) Freight demand characteristics and mode choice: an analysis of the results of modeling with disaggregate revealed preference data. J Transp Stat 149–158 Kalahasthi L, Holguín-Veras J, Yushimito WF (2022) A freight origin-destination synthesis model with mode choice. Transp Res Part e: Log Transp Rev 157:102595. https://doi.org/10.1016/j. tre.2021.102595 Kim H-C, Nicholson A, Kusumastuti D (2017) Analysing freight shippers’ mode choice preference heterogeneity using latent class modelling. Transp Res Proc 25:1109–1125. https://doi.org/10. 1016/J.TRPRO.2017.05.123 Magidson J, Vermunt JK (2002) Latent class models for clustering: a comparison with K-means. Canadian J Market Res 20:37–44 Marcucci E, Scaccia L (2003) Mode choice models with attribute cutoffs analysis: the case of freight transport in the Marche region. European Transport Trasporti Europei 25–26:21–32 Marcucci E, Pira ML, Gatta V, Inturri G, Ignaccolo M, Pluchino A (2017) Simulating participatory urban freight transport policy-making: accounting for heterogeneous stakeholders’ preferences and interaction effects. Transp Res Part E: Log Transp Rev 103:69–86. https://doi.org/10.1016/ J.TRE.2017.04.006 Matear S, Gray R (1993) Factors influencing freight service choice for shippers and freight suppliers. Int J Phys Distribut Log Manag Mitra S, Leon SM (2014) Discrete choice model for air-cargo mode selection. Int J Log Manag 25(3) Moschovou TP, Giannopoulos GA (2012) Modeling freight mode choice in Greece. Procedia Soc Behav Sci 48:597–611 Mukherjee S (2015) Environmental and social impact of fashion: towards an eco-friendly, ethical fashion. Int J Interdiscip Multidiscip Stud 2(3):22–35 Murakami H, Matsuse Y (2014) Dynamic analysis of product lifecycle and sea/air modal choice: evidence of export from japan dynamic analysis of product lifecycle and sea/air modal choice: evidence of export from Japan. Asian J Shipp Log 30:431–446. https://doi.org/10.1016/j.ajsl. 2014.12.010 Murphy PR, Daley JM, Hall PK (1997) Carrier selection: do shippers and carriers agree, or not? Transp Res Part e: Log Transp Rev 33(1):67–72 Murphy D, Daley P, Dalenberg J (1991) Selecting links and nodes in international transportation: an intermediary’s perspective. Transp J 31(2):33–40 Norojono O, Young W (2010) A stated preference freight mode choice model. Transp Plann Technol Norojono O, Young W (2003) A stated preference freight mode choice model. Transp Plan Technol 26(2):1–1 Park N, Suh S (2011) Modal shifting from road to coastal shipping using a mobile harbour. Asian J Shipp Log. https://doi.org/10.1016/S2092-5212(11)80021-0 Piendl R, Liedtke G, Matteis T (2017) A logit model for shipment size choice with latent classes— Empirical findings for Germany. Transp Res Part a: Policy Pract 102:188–201. https://doi.org/ 10.1016/J.TRA.2016.08.023 Piendl R, Matteis T, Liedtke G (2019) A machine learning approach for the operationalisation of latent classes in a discrete shipment size choice model. Transp Res Part E 121:149–161 Planning Commission of India (2014) Total transport system study on traffic flows and modal costs (Highways, Railways, Airways and Coastal shipping), a study by RITES Ltd. Premeaux SR (2002) Motor carrier selection criteria: perceptual differences between shippers ad motor carriers. Transp J 42(2):28–38

316

V. Ansu and M. V. L. R. Anjaneyulu

Puckett SM, Rasciute S (2010) Freight stakeholders’ sensitivities under road user charging: a latent class approach Reis V (2014) Analysis of mode choice variables in short-distance intermodal freight transport using an agent-based model. Transp Res Part a: Policy Pract 61:100–120. https://doi.org/10. 1016/j.tra.2014.01.002 Rich J, Holmblad PM, Hansen CO (2009) A weighted logit freight mode-choice model. Transp Res Part e: Log Transp Rev 45(6):1006–1019 Román C, Arencibia AI, Feo-Valero M (2017) A latent class model with attribute cut-offs to analyse modal choice for freight transport. Transp Res Part a: Policy Pract 102:212–227. https://doi.org/ 10.1016/J.TRA.2016.10.020 Samimi A, Kawamura K, Mohammadian A (2011) A disaggregate analysis of rail-truck mode choice behaviors for freight shipments. In: Transp Res Board 90th Annual Meeting, pp 11–3844 Shen G, Wang J (2012) A freight mode choice analysis using a binary logit model and GIS: the case of cereal grains transportation in the United States. J Transp Technol 02(02):175–188 Shinghal N, Fowkes T (2002) Freight mode choice and adaptive stated preferences. Transp Res Part E: Log Transp Rev 38(5):367–378 Siridhara S, Paonoi W, Meeyai S, Ratanavaraha V (2019) Container transport mode choice analysis with a binary logit model case study: Northeastern Thailand. Eng Appl Sci Res 46(3):256–266 Tavasszy L, de Jong G (2014) Modelling freight transport. Elsevier Science, London Tortum A, Yayla N, Gökdaˇg M (2009) The modeling of mode choices of intercity freight transportation with the artificial neural networks and adaptive neuro-fuzzy inference system. Expert Syst Appl 36(3 PART 2):6199–6217. https://doi.org/10.1016/j.eswa.2008.07.032 Wang Y, Ding C, Liu C, Xie B (2013) An analysis of interstate freight mode choice between truck and rail: a case study of Maryland, United States. Procedia Soc Behav Sci 96:1239–1249 Wichitphongsa W, Ponanan K (2022) A development of mode choice models for modal shift potential towards intermodal transportation: a case study of Vientiane-Boten railway. ABAC J 42(1):222–236 Wiegmans B, Konings R (2015) Intermodal Inland waterway transport: modelling conditions influencing its cost competitiveness. Asian J Shipp Log. https://doi.org/10.1016/j.ajsl.2015. 06.006

Optimal Coal Transport Mode Choice for Near Plant Coal Mine by Using Integrated Fuzzy Analytical Hierarchical Process and Fuzzy Goal Programming Model Sudeep Kumar Mishra and Sunny Deol Guzzarlapudi

Abstract Transportation of coal is one of the critical attributes towards the overall cost of coal and requires both technical acumen and financial investments. Generally, coal transport constitutes approximately 40–50% of total coal; therefore, the study on optimizing coal transport becomes important and relevant. In this study, Integrated Fuzzy Analytical Hierarchical Process and Fuzzy Goal Programming Model have been proposed to find out the most suitable coal transportation mode among the available choice set. A total of five types of coal transportation modes were considered in the study duly incorporating seven criteria that may affect the selection of the best mode. The mode share obtained from the developed model for Belt Conveyor System is 45.05% and Railway is 21.42%. Belt Conveyor System was found the best mode in a scenario when a plant is closer to the coal mine. An attempt was made to prioritize various decision-making criteria. Per Unit Cost of Transportation Mode, Ore reserve, and Investment Cost of Transportation Mode were found to be the most important criteria for the selection of coal transportation mode. The results of the study would be useful for transportation planning and shall bring a new dimension to the transportation mode choice modeling domain. Keywords Fuzzy AHP · FGP model · MCDM approach · Coal transportation

S. K. Mishra NIT Raipur, Raipur, India e-mail: [email protected] S. D. Guzzarlapudi (B) Department of Civil Engineering, National Institute of Technology Raipur, Raipur, Chhattisgarh, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_24

317

318

S. K. Mishra and S. D. Guzzarlapudi

1 Introduction The coal industry is one of the critical sectors of the economy of the country which primarily encompasses coal mining, processing, and transportation elements for coal production. In the current scenario, the entire world including developing countries like India is experiencing adverse coal production and supply issues associated primarily with a lack of a robust transportation system. Despite India being the second-largest importer, consumer, and producer of coal and with the fifth-largest reserves, coal production and supply issues are still prevalent. Numerous studies reported that the transportation of coal is more expensive than the coal production itself (US Energy Information -Annual Coal Report 2022). The market cost of coal is a function of basic notified pithead price, sizing, fuel charges, rapid loading and washing facilities, stationary charges, royalties, taxes/excise duties, and transportation charges. It is found that coal transport constitutes approximately 40–50% of total coal cost (Coal India Report, 2019). At present multiple-choice modeling techniques for freight transport are available however, acceptance of some of these models is limited in specific industries. It is seen that power plants face coal stocking problems during the rainy season. Such retardation of coal flow results in capacity reduction. Stocking coal which does not have a sticky problem well before the monsoon is one of the feasible solutions (Bhatt & Labs 2016). Further, there is a dire need for planners and operators to know the best suited primary and secondary mode of coal transport, as an alternative plan, for any contingencies, if required. The habitat around the coal mine and power plants is continuously exposed to pollution and soil degradation (Pandey and Agrawal 2014). The proposed study is an attempt to address these issues to find an alternative and most preferable coal transport mode which is comparatively green and sustainable, keeping trade growth, economic efficiency, productivity and timely delivery in mind. Coal is primarily transported through various transportation modes like belt conveyors, trucks of various types, pipelines, suspension rail conveyance systems, railways, and waterways. In general, the selection from a choice set is one of the challenging situations, especially when there are more than two options available and their performances are conflicting with each other. Any good decision-maker is supposed to combine the performances and utilize the best combination to successfully achieve the objective (Mizrak et al. 2015). Every selection problem will have certain decision criteria and alternatives, and the performance of alternatives changes with changes in decision criteria. To select the best coal transportation mode, fundamental characteristics of all available modes are required to be studied for identifying various criteria and attributes. Freight transportation model choice is a planning procedure wherein a decisionmaker uses a mode or combination of modes based on a set of qualitative and quantitative decision variables and criteria. The decision is a complex decision as the performance of the transport mode severely affects the complete logistic system. MultiCriteria Decision-Making (MCDM) has been adopted in the study, as it addresses

Optimal Coal Transport Mode Choice for Near Plant Coal Mine …

319

the complexity of intermodal systems, the importance of modal choice decisions, and frequent decision-making (Gohari et al. 2022). Analytic Hierarchy Process (AHP) is a well-established Multi-criteria DecisionMaking (MCDM) approach and has been widely utilized in the decision-making process. Bazzazi et al. (2008) used this approach for the selection of loading and haulage equipment, Sobandi et al. (2018) used the technique to select the quality of the gemstone, Biswas et al. (2018) used the idea to select the best apparel item for a new garment factory as a startup. Further, Fuzzy Goal Programming (FGP) approach for decision-making was utilized by Biswas and Pal (2005) for agricultural systems, Tuan and Chiadamrong (2021) used the FGP approach in Supply Chain Management. Fuzzy MCDM Approach and its application were studied in detail by Mardani covering the publication on MCDM approach between 2000 to 2014. The results showed that the AHP method was one of the most common methods from 2000 to 2014 (Abbas, n.d.). However, Eltarabishi and Omar (2020) in their review paper of an article published on MCDM Approach between the period 2015–2019 found that aggregate hybrid methods and AHP Methods are the top two MCDM methods in the present scenario and suggested that, in the coming future, aggregate methods will be used more regularly due to their higher efficiency and effectiveness. In the present paper, Fuzzy AHP with Goal Programming Method has been applied. Thus, it can be seen that AHP has been applied in different domains of engineering sciences and transport problems, however, it is found that the use of Fuzzy AHP Approach and FGP model to find the most preferred transportation mode has seldom been used. Numerous researchers have done extensive work on choice modeling in public transport from both urban and regional perspectives. Several techniques were developed to obtain optimal modes under defined criteria and choice sets. Each technique has its own merits and limitations in terms of applicability, computational time, robustness, and input criteria. However, limited studies were reported on developing choice models for freight transport, especially in the coal transportation sector. Therefore, this study is an attempt to apply the Fuzzy MCDM Technique using FGP Model to find out the best coal transportation mode in a scenario where a power plant is nearer to the coal mine.

2 Questionnaire and Data Collection The first step adopted before making the questionnaire was to study and identify globally adopted alternatives to coal transportation modes keeping the developing countries like India in mind and widening the spatial validity spectrum. Various alternatives/modes for coal transportation are through waterways, railways, and roads. However, the transportation of coal through waterway has not been considered in this study due to its exclusivity and limited flexibility with respect to options and mode choice. In Indian perspective share of coal transportation of various modes

320

S. K. Mishra and S. D. Guzzarlapudi

for the year 2019 were Railways (46%), Roads (33%), Merry-Go-Rounds (MGR) (15%), Belt Conveyor System (5%), and others (1%). Keeping the share percentage, a popular choice, and availability factor in mind coal transportation by Railways, Trucks, MGR, Belt conveyor systems, and Pipelines have been considered in this study. It is pertinent to mention that each alternative, its technical characteristics, and its viability aspect was studied in detail before selecting them for this study. It is to be noted that the selection of any alternative will depend on various criteria and therefore to select the correct alternative various decision criteria were identified and subsequently considered for the model. Each criterion has certain specific attributes like carrying capacity, environmental exposure of the product, transit time, accessibility, urgency level of the product, reliability factor, cost–benefit analysis, contribution towards a greener environment, etc. The criteria selected in this study are ore reserve, the distance of a delivery point, time taken to deliver the coal, reliability of transport mode, flexibility of the transport mode, investment cost, and per unit cost of transport mode based on the study carried out by Macchione and Bisht (2020), Artobolevskiy and Artobolevskiy (2013), Benalcazar et al. (2017), Gomari and Johansson (2019), Wadhwa et al. (2008), Kania (1984), and Yunianto (2018). Overall, in this study, a total of five alternatives and seven criteria have been identified to facilitate the preparation of the correct multiple criteria decision-making tool for the selection of the most preferable and best suited coal transportation mode. The network of the overall objective, selected criteria, and alternatives with layers of levels are shown in Fig. 1. A questionnaire using a survey form was developed for data collection and subsequently, Delphi Method was applied as part of the model development process (Francisco et al. 2021). The input was obtained from domain experts and executives from basic stakeholders like power plants, cement plants, and coal mines. A total of twelve experts/executives were considered as decision-makers for the study to record the inputs. The selected decision-makers are Managers/Executives from coal mines, Cement plants, Power Plants and Steel Plants, Transporter dealing with coal transportation, and Scholars associated with the subject matter. It is pertinent to mention

Fig. 1 Objective, criteria, and alternatives in the proposed model

Optimal Coal Transport Mode Choice for Near Plant Coal Mine …

321

that more than 25 experts were contacted in the first round but only 12 actionable responses were received, which were found fit to be incorporated in this study. The difference in opinion was moderated through a second round of review by the experts who were having a difference in opinion and their input was not in synchronization with others. A final relative rating was obtained by taking the mean value.

3 AHP Methodology and FGP Model Large-level socio-technical decisions with numerous intangible criteria can be addressed using the Analytic Hierarchical Process (AHP) developed by Saaty (1987). The AHP also measures both physical and psychological events. In another way, it can be said that both tangible and intangible inputs can be recorded and measured using this method. The entire process is based on three important steps. The first step is to decide the goal and further decide the criteria and various alternatives which would be affecting the decision-making process. The second step is to find out priorities or ratings. Opinion of various experts is obtained in order to find out the relative importance of criteria and alternatives with reference to a specific criteria/alternative. Pairwise matrix duly incorporating inter-se-priority is prepared based on the rating punched by various experts. In the next step, a decision table is made wherein all the obtained values are tabulated. It is to be noted that values in the decision table are local, however, a final score is obtained based on the input values which represent the global priority of the criteria /alternative. Consistency is one of the important functions of the decision-making process and in this study, consistency was checked as promulgated by (Saaty et al. 1987). It was observed that only AHP Methodology may not be able to solve Multiple Choice Decision-Making (MCDM) problems therefore Triangulation Fuzzy Numbers in terms of lower/bottom, upper/top, and mid values of priorities were introduced (Mikhailov and Tsvetinov 2004). The final priority of each criterion was obtained using Triangulation Fuzzy Numbers. In the third and decision-making phase, a weighted sum of rating/priority linked with each criterion and alternative was computed to reach a final decision. Owing to the associated advantages of this technique Fuzzy AHP Approach using MCDM Technique has been applied in this study. The entire methodology can be seen in the flow chart shown in Fig. 2. The concept of Fuzzy Triangulation Technique and Triangulation Fuzzy Number used in the entire process has been discussed in brief in succeeding paras.

322

S. K. Mishra and S. D. Guzzarlapudi

Fig. 2 Flow chart of methodology adopted

3.1 Use of Fuzzy Comparison Technique for Judgement in Decision-Making Phase The most challenging part of AHP is to closely and correctly assess the local priorities which are rated by Decision-Makers. While filling out the questionnaire a DecisionMaker may not be very certain about any specific option e.g., ‘Attribute 1 is Extremely Less Important than Attribute 2 when they are compared with respect to criteria X’. The term ‘Extremely Important’ or ‘Extremely Less Important’ is relative and intangible. Saaty’s 9-point scale has been utilized in this study to gauge the uncertainty and quantify the intangible term, where ‘Extremely Important’ is represented by 9, ‘Very important’ by 7, ‘Important’ by 5, ‘Slightly Important’ by 3, and ‘Equally Important’ is represented by 1. Similarly, it reduces further from ‘Slightly less Important’ as 1/ 3 to ‘Extremely Less Important’ as 1/9. It is to be noted that the scale of 1,3,5,7,9 has been utilized and the mid values 2,4,6,8 are corresponding bottom and top range of the rated choice. Hence, we can conclude that whenever a score is selected its top and bottom score are also automatically incorporated into the model to cover the entire spectrum.

Optimal Coal Transport Mode Choice for Near Plant Coal Mine …

323

3.2 Triangulation Fuzzy Number (TFN) In the present study Triangulation, Fuzzy Numbers are being used as it defines the bottom value matrix, middle-value matrix, and top value matrix. This arrangement becomes easier to understand the concept. (Yetkin., 2016). As discussed, we shall follow 3 matrices: Bottom Value Matrix a with values {a1 , a2 , a3 , …an } Mid Value Matrix b with values {b1 , b2 , b3 , ……bn }. Top Value Matrix c with values {c1 , c2 , c3 , ……cn }. The fuzzy membership function is a technique of visualizing the degree of membership through the graphical representation method, wherein the X-axis represents the discourse and the Y-axis represents the degree of membership between ranges [0,1]. The fuzzified input in a triangle can be defined by three Parameters a, b and c. where c depicts the base and b depicts height. In Fig X represents the input from experts and Y represents the corresponding fuzzy value. It can have three conditions: • Condition when X = b or X < a or X > c μ [x] = 0 when; x < aor x > c. • Condition when X lies between b and c μ [x] =

c−x when (b ≤ x ≤ c). c−b

• Condition when X lies between a and b μ [x] =

x −a when (a ≤ x ≤ b). b−a

Combining all conditions together we get, x−a c−x , c−b . = max min ( b−a μ triangle { x: a,b,c} =

(1)

= max

min (

,

),0

324

S. K. Mishra and S. D. Guzzarlapudi

3.3 Fuzzy Prioritization Approach While considering a prioritization problem for any specific level (Goal, Criteria or Alternative) with n elements Pairwise Comparison Matrix shall be represented by    TFN dij = bij, mij, tij , where b, m and t depict bottom, middle and top matrices respectively.

=

(2)

wi Where dij = d1ji and dij = wj ; Weights (w) can be found out by using Weighted Sum Method, Logarithmic Least Square Method, Fuzzy Geometric Mean Method, or Fuzzy Arithmetic Mean Method (Mikhailov and Tsvetinov 2004). The final score of any criterion/alternative can be obtained by using any of the methods mentioned above. Due to fuzzification of choice, the selected option covers the long-range ab initio once a final score is determined. Subsequently, ranking of all the alternatives is obtained which helps in ranking the alternatives. As we know that a decisionmaker rates various alternatives as per the criteria available in the environment. The membership function (µ) represents the satisfaction of any decision-maker with reference to specific parameters. Weight of alternative changes with different criteria wi . and accordingly satisfaction level also changes with different wj    Now by using Eq. (2) where di j = bij, mij, tij .

(3)

Subject to ti j > mi j > bi j Where t, m, and b represent the top, middle, and bottom matrix respectively. However, it is pertinent to mention here that this assumption is not binding. In Eq. (3) membership function µi j ( wwij ) linearly increases with an interval between (∞, mi j ) and decreases linearly between ( mi j ,∞). In Eq. (1) we have seen that function was not taking any -ve value however, in Eq. (3) the outcome can take a wi < bij or wwij > ti j . But maximum value µi j = 1, when wwij = negative value when wj

Optimal Coal Transport Mode Choice for Near Plant Coal Mine …

325

bij; Therefore, we can say that the membership function coincides with fuzzy trian   gular di j = bij, mij, tij , where b, m and t depict bottom, middle and top matrices respectively.

3.4 Solving the Fuzzy Prioritization Problem and FGP Model Maxima minima rule (maximin) for decision-making has been successfully used by Zimmermann, Bellman and Zadeh, and others. In any Fuzzy Goal Programming Model (FGP) the priority is to maximize the values of all membership functions since we want maximized consistency in our decision (Ozfirat 2015). In order to achieve the goal, we can maximize the minima of membership for n ; the decision-maker should be consistent in his decision. function Z = Max µi [x]i=1 To show the consistency level of the decision-maker a variable L is introduced with the following objective function.   Maximize L; Subject to L ≤ µi j wwij ∀ i = 1……n-1 and j = 2…. n where j > i. The problem statement can also be dealt with by rank reversal technique wherein instead of maximizing L we can also minimize - L. The value of L also represents to a level of consistency. When a classical AHP scenario is considered inconsistency level should be less than 0.1 which can be further interpreted as consistency level should be greater than 0.9. As proposed we can safely assume that the model value of L is expected to be more than 0.9 (Özfirat et al. 2017). Once each criterion and alternative are individually dealt with through a pairwise comparison matrix and the associated weight is obtained thereafter performance of each alternative and each criterion should be rated. A score matrix is obtained by multiplying the elements of normalized non-fuzzy relative weights of each alternative for each criterion. The best choice can be obtained from the above matrix based on the scores.

4 Model Development Using Fuzzy AHP Methodology and FGP Model The model has been developed to find out the most preferred coal transportation mode for power/cement/steel plants, when it is located close to a coal mine, duly keeping the input of experts in mind. A total of seven decision criteria for five alternatives as mentioned were considered for the proposed MCDM Approach. In the first phase, a pairwise fuzzy comparison matrix using fuzzy triangular numbers was made and AHP Technique was applied as shown in Fig. 2. Lower or bottom, middle, and upper or top numbers were used to obtain the bounds. In the present case, geometric mean of fuzzy comparison value has been utilized to find out the relative fuzzy weight of each alternative and criteria as shown in Table 1. The final property for each criterion

326

S. K. Mishra and S. D. Guzzarlapudi

is computed and subsequently, in the decision phase, a weighted sum is calculated for each alternative to find out aggregated results for each alternative according to each criterion as shown in Table 2. Based on the above result final score and best option for transportation modes are obtained as shown in Tables 3 and 4. Table 1 Fuzzy output of criteria Criteria

Ore reserve

Distance of delivery point

Time taken to deliver

Reliability of mode

a

b

c

a

b

c

a

b

C

a

b

c

Ore reserve

1

1

1

0.2

0.2

0.3

0.3

0.5

1

5

6

7

Distance of delivery point

6

5

4

1

1

1

0.3

0.3

0.5

0.3

0.3

0.5

Time taken to deliver

1

2

3

2

3

4

1

1

1

0.3

0.3

0.5

Reliability of mode

0.1

0.2

0.2

2

3

4

2

3

4

1

1

1

Flexibility of mode

0.2

0.2

0.3

2

3

4

2

3

4

4

3

2

Investment cost of mode

0.2

0.2

0.3

1

2

3

4

5

6

7

8

9

Per unit cost of mode

0.2

0.2

0.3

2

3

4

5

6

7

5

6

7

Criteria

Flexibility of Tpt mode a

b

c

a

b

c

a

b

C

Ore reserve

4

5

6

4

5

6

4

5

6

0.2

Distance of delivery point

0.3

0.3

0.5

0.3

0.5

1

0.3

0.3

0.5

0.1

Time taken to deliver

0.3

0.3

0.5

0.2

0.2

0.3

0.1

0.2

0.2

0.1

Reliability of mode

0.5

0.3

0.3

0.1

0.1

0.1

0.1

0.2

0.2

0.1

Flexibility of mode

1

1

1

0.2

0.2

0.3

0.1

0.2

0.2

0.1

Investment cost of mode

4

5

6

1

1

1

0.2

0.2

0.1

0.2

Per unit cost of mode

5

6

7

7

6

5

1

1

1

0.3

Continuation

Investment cost

Per unit cost

Normalized weights

Optimal Coal Transport Mode Choice for Near Plant Coal Mine …

327

Table 2 Aggregated result of each alternative against each criterion 1

Belt conveyor system

Truck

Pipeline

Railway

MGR

Weight of criteria

Ore reserve

0.05

0.07

0.26

0.50

0.12

0.23

Distance of delivery point

0.55

0.17

0.10

0.14

0.04

0.08

Time taken to deliver

0.53

0.17

0.11

0.15

0.04

0.07

Reliability of mode

0.49

0.17

0.13

0.17

0.04

0.06

Flexibility of mode

0.58

0.16

0.10

0.12

0.04

0.09

Investment cost of mode

0.54

0.16

0.12

0.14

0.04

0.17

Per unit cost of mode

0.61

0.14

0.10

0.12

0.03

0.30

Total Percentage share

0.45

0.14

0.14

0.21

0.06

1.00

45.05

13.74

14.23

21.42

5.57

1.00

Table 3 Final score of criteria Ser no

Criteria

Weight

Ranking

1

Ore reserve

0.226

2

2

Dist of delivery point

0.078

5

3

Time taken to deliver

0.071

6

4

Reliability of transport mode

0.061

7

5

The flexibility of transport mode

0.092

4

6

Investment cost of transport mode

0.170

3

7

Per unit cost of transport mode

0.303

1

Table 4 Priority of alternatives

Ser No

Priority

Alternative

1

P1

Belt conveyor system

2

P2

Railway

3

P3

Pipeline

4

P4

Truck

5

P5

MGR

328

S. K. Mishra and S. D. Guzzarlapudi

5 Results and Discussion The study has been carried out for the situation when a plant is closer to the coal mine. The transportation cost of coal reduces significantly in such a scenario. An endeavor was made to get the opinion of experts to decide the distance which should be the guideline to conclude whether a plant is nearer to the coal mine or not. As per the survey, 33.3% of respondents opined that distance up to 20 km would be considered as near distance. However, 41.7% considered the distance up to 50 km. Overall, it is observed that 75% of the opinion was for a distance under 50 km and therefore, keeping the availability on the ground, provision, operation and maintenance cost in mind, a distance of 50 km was considered as near distance in this study for developing countries scenario. Aggregated result of each alternative and each criterion was obtained utilizing the Fuzzy AHP Technique. Subsequently, final score was obtained by multiplying each cell by 100 as shown in Table 2. As per the model output, the Belt conveyor system is the best coal transportation mode for a near to coal mine scenario, which is followed by railways, pipelines, trucks, and MGR. Although MGR based transportation mode was separately considered while making questionnaire, however, as per the model output it is evident that the railway is taking predominance over MGR. The score of the Belt Conveyor System is 0.450 followed by 0.214, 0.142, 0.137 and 0.056 for Railways, Pipelines, Trucks, and MGR respectively. A consistency check for alternatives was carried out for the bottom, middle, and top matrix, and the consistency ratio of alternatives was found to be less than 0.1 which means that the input by the Delphi method was consistent with respect to alternatives (Francisco et al. 2021). It is seen that if traditional AHP is applied in lieu of Fuzzy AHP then it becomes difficult to keep the consistency level within 0.9. However, in FAHP the chances to get the desired value are increased due to fuzzification. So, it can be concluded that the optimal transportation selection process is more precise and less dependent on the decision-maker if FAHP is applied. In the present case, L value for the criteria matrix is more than 0.9 which may be considered an acceptable level as it should be greater than 0.9. However, it was found that it is difficult to maintain the desired level for all criteria and alternatives as the input by an expert cannot be the same and the variance so caused may not be within the acceptable range. It is felt that instead of making undesirable moves in the calculation it is better to accept the consistency level with some lesser confidence level. Özfirat et al. (2017) stated that various reasons for inconsistency in MCDM techniques like improper assignment of relative importance value by the data provider, proficiency level of decision-maker, and manual errors during calculations. The same can be streamlined up to some extent by improving the algorithm, repetitive consultations, and the right selection of decision-makers and increasing the sample size.

Optimal Coal Transport Mode Choice for Near Plant Coal Mine …

329

6 Conclusion In this study, Integrated Fuzzy Analytical Hierarchical Process and Fuzzy Goal Programming Model have been proposed to find out the most suitable coal transportation mode among the available choice set. In order to make the most appropriate decision evaluation process, the procedure was then applied to real life underground coal mine scenarios duly incorporating various criteria. The results depict that the Belt conveyor system is the most suitable coal transportation system with a 45% mode share in a scenario when a coal mine and plant are within 50 km distance. Following mode share obtained from the developed model in the priority sequence Belt Conveyor System (45.05%), Railway (21.42%), Pipeline (14.23%), Truck (13.74%), and MGR (5.56%) respectively. It is to be noted that the percentage signifies the priority of that mode i.e., more is the percentage higher is the priority of that mode. Railways and pipelines are found as second and third best systems under the chosen circumstances. Transportation of coal by a dedicated railway system (MGR) is still under development and therefore not being preferred substantially in the present scenario. The best transportation mode selection proposed in the model is in line with the opinion of domain experts and earlier findings by other researchers. A total of seven decision criteria were considered for the study and the most important criteria for the selection of the best suited coal transportation system are per unit cost of transportation mode (rank 1), ore reserve (rank 2), and investment cost of transportation mode (rank 3). The ranking of other criteria in the proposed model is the flexibility of transport mode (rank 4), distance of delivery point (rank 5), time taken to deliver (rank 6), and reliability of transport mode (rank 7). Since costbased criteria are dominating over others the objective function may be considered to minimize instead of maximizing. In the near mine scenario, a few criteria like distance, time and reliability of transport mode have limited effect on the selection of coal transport mode. The proposed model will help the transport planners of the coal sector, power plants, and cement plant operators to optimize the transport resources and make transport mode selection process less dependent by minimizing the element of human error and thereby increasing the reliability of the model. The result of the study can be useful for the new thermic and cement plants which are likely to be established at the regional and national levels. The model can be useful in global scenarios also as certain transport alternatives which are not being fully explored in the Indian scenario have also been considered in the model to strengthen the temporal and spatial validity. The model may also assist the planners of existing plants when additional stocking is planned during the pre-monsoon phase by providing best suited primary and secondary coal transport modes. The proposed model is easy to apply, user friendly and does not require any complex computational mechanism. The study can be further improved by strengthening the consistency mechanism and by incorporating additional criteria /alternatives keeping in mind the technological changes in the field of coal transportation. We are sanguine that the

330

S. K. Mishra and S. D. Guzzarlapudi

proposed model would help planners to select and allocate transport resources in the field of coal transportation.

References Abbas M (n.d.) Fuzzy multiple criteria decision-making techniques. Mardani Review Paper Ahmad S, Avtar R, Sethi M, Surjan A (2016) Delhi’s land cover change in post transit era. Cities 50:111–118. https://doi.org/10.1016/j.cities.2015.09.003 Ahmed S, Khan A (2018) Decision-making methods methods application application in in supply chain management: management: a a systematic systematic literature literature review. 3–32. https://doi.org/10.5772/intechopen.74067 Artobolevskiy SS, Artobolevskiy SS (2013) Coal mining areas in Russia: national, regional or local problem Professional paper Coal mining areas in Russia: national, regional or local problem. 9009. https://doi.org/10.1179/037178403225011060 Bazzazi AA, Osanloo M, Soltanmohammadi H, Branch S, Engineering M (2008) Loading-haulage equipment selection in open pit mines based on fuzzy-TOPSIS method Benalcazar P, Kaminski J, Saluga PW (2017) The storage location problem in a coal supply chain : background and methodological approach. Mineral Res Manag 33(1):5–14. https://doi.org/10. 1515/gospo-2017-0009 Bhatt MS, Labs NE (2016) Effect of moisture in coal on station heat rate and fuel cost for Indian thermal power plants Biswas A, Pal BB (2005) Application of fuzzy goal programming technique to land use planning in agricultural system. 33:391–398. https://doi.org/10.1016/j.omega.2004.07.003 Biswas TK, Akash SM, Saha S (2018) A fuzzy-AHP method for selection best apparel item to start-up with new garment factory: a case study in Bangladesh. 7(1):32–50. https://doi.org/10. 22105/riej.2018.111802.1034 Duru O, Bulut E, Yoshida S (2012) Expert systems with applications regime switching fuzzy AHP model for choice-varying priorities problem and expert consistency prioritization: a cubic fuzzypriority matrix design. Expert Syst Appl 39(5):4954–4964. https://doi.org/10.1016/j.eswa.2011. 10.020 Eltarabishi F, Omar OH (2020) Multi-criteria decision making methods and their applications—a literature review. 2654–2663 Francisco A, López R, Sánchez M, Puente-fernández D, Montoya-juárez R, Roldán C (2021) A fuzzy delphi consensus methodology based on a fuzzy ranking. 1–18 Gohari A, Bin A, Tighnavard A, Gohari A, Hasan R (2022) Significance of intermodal freight modal choice criteria : MCDM-based decision support models and SP-based modal shift policies. Transp Policy 121(April):46–60. https://doi.org/10.1016/j.tranpol.2022.03.015 Gomari S, Johansson F (2019) ScienceDirect ScienceDirect Automation-ready framework for urban transport and road infrastructure planning. Transp Res Proc 41:88–97. https://doi.org/10.1016/ j.trpro.2019.09.018 Hanzl J (2020) General application of multiple criteria decision making methods for finding the optimal solution in city logistics. Open Eng 10(I):147–153 Kania JJ (1984) Economics of coal transport by slurry pipeline versus unit train a case study. 131–138 Kumar A, Anbanandam R (2019) Development of social sustainability index for freight transportation system. J Clean Prod 210:77–92. https://doi.org/10.1016/j.jclepro.2018.10.353 Laurent A, Vallerand S, Van Der Meer Y, Amours SD (2020) CarbonRoadMap: a multicriteria decision tool for multimodal transportation CarbonRoadMap: a multicriteria decision tool for multimodal transportation. Int J Sustain Transp 14(3):205–214. https://doi.org/10.1080/15568318. 2018.1540734

Optimal Coal Transport Mode Choice for Near Plant Coal Mine …

331

Macchione B, Bisht A (2020) The extractive industries and society iron ore peripheries in the extractive boom : a comparison between mining conflicts in India and Brazil. Extract Ind Soc 1–12. https://doi.org/10.1016/j.exis.2020.09.010 Mikhailov L, Tsvetinov P (2004) Evaluation of services using a fuzzy analytic hierarchy process. 5:23–33.https://doi.org/10.1016/j.asoc.2004.04.001 Mizrak P, Kemal M, Malli T, Kahraman B (2015) Integration of fuzzy analytic hierarchy process and multi-objective fuzzy goal programming for selection problems: an application on roadheader selection. 29:53–62.https://doi.org/10.3233/IFS-151569 Özfirat PM, Özfirat MK, Malli T (2017) Selection of coal transportation mode from the open pit mine to the thermic power plant using Fuzzy Analytic Hierarchy Process. Transp J Taylor Francis 4142. https://doi.org/10.3846/16484142.2017.1295278 Ozfirat MK (2015) Selection of tunneling machines in soft ground by fuzzy analytic hierarchy process. 20(2):98–109 Pandey B, Agrawal M (2014) Coal mining activities change plant community structure due to air pollution and soil degradation. 1474–1483. https://doi.org/10.1007/s10646-014-1289-4 Ratrout NT, Al-madani H, Gazder U (2014) A review of mode choice modelling techniques for intra-city and border transport. Int Trans Res 5(I) Saaty RW, Britain G, Avenue E (1987) The analytic hierarchy process-what and how it is used. 9(3):161–176 Shankar R, Choudhary D, Jharkharia S (2018) An integrated risk assessment model : a case of sustainable freight transportation systems. Transp Res Part D 63:662–676. https://doi.org/10. 1016/j.trd.2018.07.003 Sobandi M, Putra D, Andryana S, Gunaryati A (2018) Fuzzy analytical hierarchy process method to determine the quality of gemstones Tuan DH, Chiadamrong N (n.d.) A fuzzy credibility-based chance-constrained optimization model for multiple-objective aggregate production planning in a supply chain under an uncertain. Environment 25(7). https://doi.org/10.4186/ej.2021.25.7.31 Wadhwa S, Saxena A, Chan FTS (2008) Framework for flexibility in dynamic supply chain management. 7543. https://doi.org/10.1080/00207540600570432 Yetkin O (2016) A fuzzy approach to selecting roof supports in longwall mining. 27:162–177 Yunianto T (2018) Freight calculation model : a case study of coal distribution freight calculation model : a case study of coal distribution. Earth Environ Sci 135(conference series)

Economic Analysis of FasTag on Highway Toll Collection Anand Mishra and Abhisek Mudgal

Abstract Toll revenue is one of the major sources of revenue by which the government supplements its financial capability to execute, maintain and upgrade existing highways. Revenue leakage, underreporting of revenue, and refusal of toll payment is one of the major concerns of the government. These issues must be addressed in order to make highway projects more feasible, less risky, and financially attractive for private investors. To achieve this aim the government has introduced FasTag, an Electronic toll collection system. We studied the financial impact of FasTag on Highway toll collection at Patna-Deedarganj Toll Plaza on Patna-Bakatiyarpur Toll Road. In our analysis, we find that the FasTag collection has significantly lower service time than cash collection and gives an immense potential for an increase in traffic throughput and revenue collection, the default rate on toll payments has dropped very significantly with increased penetration of FasTag. We ran several time-series models for revenue forecasts and found that there was a general trend of increase in revenue and among all the models, PROPHET was the most accurate. Keywords FasTag · Transportation engineering · Transportation economics · Electronic toll (E-toll) · Transportation infrastructure · Transportation finance · Infrastructure policy · Transportation policy

1 Introduction Highways are the lifeline of our nation, connecting centers of economic, cultural, and strategic importance. However, highway projects are highly capital intensive, have a long gestation period, and carry high risk. The government has created multiple A. Mishra (B) LSRID-RAP, Lok Sabha Secretariat, Parliament of India, New Delhi, India e-mail: [email protected] A. Mudgal IIT-BHU, Varanasi, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_25

333

334

A. Mishra and A. Mudgal

ways to fund infrastructure projects, and one of the significant sources of revenue for highway projects is the Toll Revenue. Government efforts to rope in private investment in this sector and to transform its role from financer-cum-regulator to regulator, but large capital needs, prolonged litigation, and low return and long return period have kept investors hesitant. The economic importance of highways is not just limited to the movement of goods and people between locations, but it also results in job generation during the construction phase and the development of a micro-economy around the highways to cater for the needs of users using the highways (Yoshino and Pontines 2015). The economic impact has an area effect known as the spillover effect over an area (Yoshino and Pontines 2015). Toll roads aim to supplement the government’s financial capacity to improve the quality, spread, and capacity of the highway networks so that the users can be provided with a better traveling experience, they do not come without their own disadvantages and limitations. Congestion at toll plazas with long queues is a common sight. This further translates into increased waiting times for users who later become irritated and hostile leading to frequent altercations with toll employees. In this paper, the economic impact of the implementation of FasTag technology at Patna-Bakhtiyarpur Toll Plaza is analyzed. Comparison of revenue collection between pre-FasTag and post-FasTag period is done, the cash flow, service time of users, and further quantified the default rate (Productivity of Toll) at the toll plaza during the post-FasTag and Pre-FasTag phases are compared. Several Time-series models are run in order to forecast the revenue trend.

2 Literature Review Several studies have aimed to find the various impacts of tolling solutions. Jaiswal and Samuel (2021) discussed the effects of the toll booth on fuel consumption and air quality around the toll plaza. They found that the toll booth barrier led to increased idling time for vehicles at the toll plaza, which resulted in increased fuel consumption as well as a significant deterioration in air quality due increase in vehicular emission and noise pollution at the toll plazas again due to increased waiting period when compared to a scenario when there were no toll barriers, and the vehicle was moving at its cruising speed. Delmon (2009) ascertained that the most vulnerable aspect of the Public–Private Partnership financing model is the construction cost and revenue. Another study pointed out that besides the construction, a significant concern for the toll operation has been revenue accountability, reliability, and revenue leakage as per the UIDAI (Report on ETC Technology 2019). Al-Deek et al. (1997), the authors discussed the impact of the electronic toll (E-Toll) on the operation of the toll plaza. The study found a reduction in service time, queue length, lane throughput, operational expenses, and air quality. From an economic standpoint, there was a drastic change: the cost for each dollar collected

Economic Analysis of FasTag on Highway Toll Collection

335

from E-toll was 10 cents compared to 23 cents against conventional toll, which is significant to infrastructure finances. This study indicates the considerable advantage of E-toll over manual collection. The study was conducted in an environment that is significantly different from India. Therefore, it is compulsory to see if the E-toll system has the same impact on the Indian Toll Ecosystem where there is a significant change in driving behavior and socio-economic conditions. Levision, Equity (Levinson 2010) discussed the impact of equity in toll pricing, especially dynamic pricing, which determines toll price based on factors like time, congestion, geo-location, utilization, etc. Equity here refers to the equal stake, while in an ideal scenario everyone wins, but in real-world there are winners and losers, in terms of road pricing, this situation can be turned into a win–win by using Paretoefficiency which states a compensation to losers, this has till date been only a theoretical concept as pricing is also affected by various political issues which are very much crucial for the stability and operation, the sole concept of equity is not the only criteria, but the perception of equity is also important, which was found to be more critical when it comes to social acceptability. Pareto-Efficiency theory is the most balanced between social acceptance and financial objectives. The research highlights the importance of revenue generation at toll plazas. In India, where the cases of wilful default are high, it also highlights the need to quantify the default rate of toll by traffic to shape future policies. Tseng et al. (2013), the authors discovered there was a significant reduction in service time cost by as much as 60.1% in the use of E-Toll when compared to manual tolling system, since the study was conducted in Taiwan a region where the environment is significantly different in terms of driving discipline and behavior than that of India, the same must also be assessed as per the Indian environment. All the above studies have been conducted in a developed nation where the socioeconomic environment is significantly different from that of the Indian sub-continent or have focused on vehicular emissions and not the toll default rate or the financial aspect of improvement in service time due to E-Toll. The focus of this paper has been to look into this area, as to how the adoption of the FasTag (E-toll mechanism) has impacted the productivity and financial aspect of the Toll plazas in light of the highly leveraged position of the NHAI and Toll assets.

3 Methodology 3.1 Site Selection The site selected for this study is the Patna-Baktiyarpur Toll (46.847 km) operated by the concessionaire M/s Patna-Bakhtiyarpur Tollway Limited. The toll road was built on the Build-Operate-Transfer (Toll) model. The toll road is a four-lane road with 12 toll booths, all of which are FasTag enabled. The concession period was for 18 years (26-09-2011 to 26-09-2029), which included the construction of the road

336

A. Mishra and A. Mudgal

with a capital cost of Rs 908.02 Crore with the commercial operation starting on 12-04-2015. The project initially had a capital cost estimate of Rs.770 Crore, which escalated due to delays. The project’s total cost to the concessionaire has exceeded Rs 1308.2 Crore, including refinancing and interest costs. This site was selected primarily due to 3 reasons: a. The toll since its inception had operated on cash-only mode and had started making the transition to FasTag in September-2017, providing sufficient data for the study. b. Covid-19 restrictions limited our movement and accessibility, and this was the only site that was accessible within these restrictions having the data which was required. c. The toll plaza lies in the vicinity of a heavy traffic zone which ensures a good mix and volume of traffic.

3.2 Data Collection The Data for the study was provided by Patna-Bakhtiyapur Toll Limited during the site visit. The data relating to monthly traffic flow from April-2015 to December 2020 was collected along with monthly revenue collection, traffic composition and FasTag Traffic data. Data for two toll lanes for one day during peak hours was collected one year apart. The data collection was severely affected by Covid-19 related restrictions and lockdown. Data for other days could not be collected as it was not available at the site office.

3.3 Service Time Analysis The service time for two lanes namely Lane-6 and Lane-7 for one day each a year apart were taken during peak hour. This represents the traffic pattern for all 12 tolls throughout the year as per the toll operator’s information. The difference in service time for cash and FasTag traffic is translated into potential traffic loss, which is converted into potential revenue.

3.4 Probability of Payment of Toll The probability of payment of toll by a vehicle (i.e. referred to as productivity here) is done by applying the weightage to the form of payment (cash or FasTag) according to their default history. Subsequently, the probability of toll payment each month is calculated and the productivity is compared to the FasTag Penetration of Traffic.

Economic Analysis of FasTag on Highway Toll Collection

337

3.5 Revenue Forecast Revenue for up to the month of 01-12-2022 was estimated using various time-series models like ARIMA, GLNMET, XGBOOST, Neural Network, and Prophet, and each model was compared to the original revenue data for ascertaining the accuracy of each model.

4 Data Analysis 4.1 Service Time Service time is a critical aspect of revenue collection at a toll plaza. Even a slight change in service time can result in a considerable increase or decrease in traffic throughput over the period, which changes into toll revenue. Here the following assumptions are made (Fig. 4.1): 1. The service time from the two lanes is representative of all toll lanes. 2. The traffic conditions do not change significantly throughout the year. Si,tc = Mean service time for lane ith during tth year for cash transaction vehicles. Si,te = Mean service time for lane ith during tth year for FasTag transaction vehicles. Sdi,t = Difference in service of FasTag and Cash Transaction for ith lane during tth year. = Si,te − Si,tc . Sm,t = Mean of Difference in service of FasTag and Cash Transaction for during tth year. Se,t = Mean of service time for FasTag for tth year. Sc,t = Mean of service time for Cash Transaction for tth year. Tc,i = Total Cash Traffic in that ith month during tth year. Tpe,t = Potential of Additional FasTag Traffic in ith month in tth year. =

T otal Cash T ra f f ic in that ith month during tth year × Sm , Se,t

(1)

Rv = Revenue Per vehicle during the corresponding period. Rp = Potential Additional Revenue = T pe, × Rv

(2)

338

A. Mishra and A. Mudgal

Fig. 4.1 Service time monetary analysis

In Lane-6 the median service time for FasTag and Cash traffic was (25 s,24 s) and (15 s, 21 s) in 2019 and 2020. Similarly, in Lane-7 the median service time for FasTag and Cash traffic was (20 s, 25.5 s) and (17 s, 24 s) in 2019 and 2020 respectively. For at least 50% of the cash users, the cash transaction was 40% and 41.12% more time-consuming than E-Toll users in 2020 for Lane-6 and Lane-7 respectively. The average service time for FasTag users and Cash users was (31 s, 41.5 s) and (21.5 s, 28.5 s) in 2019 and 2020 respectively. The economic analysis of the service time difference indicated that if the entire cash traffic moved to FasTag system the potential traffic revenue generated amounted to | 27,44,75,477 in 2019 which is 32.48% of the realized annual revenue in 2019 and | 17,02,88,951 in 2020 which is almost 19.42% of the realised annual revenue in 2020.

4.2 Probability of Payment by a Vehicle at Toll Plaza Every vehicle that uses a toll road has to pay toll charges according to the criteria decided a certain number of vehicles are exempted from paying toll charges but such a number of vehicles are marginal when compared to the traffic flow. Despite this,

Economic Analysis of FasTag on Highway Toll Collection

339

a large portion of traffic does not pay the toll. The traffic has been divided into two categories: 1. Productive Traffic: The traffic that pays the toll. 2. Unproductive Traffic: The traffic that does not pay the toll. The following assumptions were made in the analysis (Figs. 4.2, 4.3): 1. The probability of payment of FasTag is 1 and malfunctions are neglected. 2. Legally exempted vehicles are neglected as their volume is very very small. 3. All unproductive traffic is assumed to be potential cash traffic as FasTag is automated and does not need human intervention for payment. Let the following denote: S = Total Traffic through Toll (Both productive and unproductive). E = FasTag Traffic (Fast Traffic is always productive).

Fig. 4.2 Productivity and FasTag penetration

Fig. 4.3 Percentage of FasTag traffic in productive traffic

340

A. Mishra and A. Mudgal

C = Cash Traffic (Both productive and unproductive). Cp = Productive Cash Traffic. Cup = Unproductive Cash Traffic. X = Event of Payment of Toll. P(X) = Productivity of Traffic i.e. Probability of Payment of Toll. P(X|E) = Probability of Payment of Toll and that the traffic is FasTag = 1. P(X|C) = Probability of Payment of Toll and that the traffic is Cash = CCp P(C) = Probability of Cash Traffic = CS P(E) = Probability of FasTag Traffic = ES P(X ) = (X |E) · · · (E) + P(X |C) · · · P(C) Cp C E + Cp E + Cp E = = × 100 (In Percentage) =1 + S C S S S

(3)

4.3 Revenue Forecast We tested multiple models on our data and all have shown a general trend of an increase in revenue over the period of time. This increase in revenue can be attributed to an increase in traffic flow, increased toll rates., and a decrease in the default rate of vehicles at toll with the increase in the adoption of FasTag. All the above aspects will need to be analyzed in order to verify them and to find which of the above has contributed to what extent to the revenue trend. There are two data points that are anomalies in the original data, due to:• During the period of August–September (52 days) 2018 the operator was prohibited from collecting normal toll from traffic by the order of the Honorable Patna High Court. • Due to COVID-19 Government of India announced a complete lockdown between 25.03.2020 to 19.04.2020 due to which the toll plaza was closed. In order to compensate for these anomalies, these two data points were replaced by the average of two data points within which they lie, in order to run the models (Fig. 4.4). Multiple models were run to forecast revenue generation from April-2015 to December-2022 and their accuracy was checked against actual revenue data using Root Mean Square Error. Among all the models tested, PROPHET by Facebook turned out to be the best fit model. It showed the R2 value of 0.647 which was significantly above rest of the models and most of the forecasted revenue values by it were within a 5% error of the actual revenue. We can see from the figure that the plot for PROPHET moves very much closer to the actual revenue plot, indicating how accurate the forecast of the model is in this scenario. After January-2021 the model

Economic Analysis of FasTag on Highway Toll Collection

341

Fig. 4.4 Revenue forecasting using different time-series models

forecasts the revenue up to December-2022, and indicates an increasing trend in the revenue generation.

5 Results Service time analysis has shown that the toll is operating below its revenue-generating capacity only because there are still a significant number of cash users. The additional time consumed by the cash consumers in comparison to their efficient FasTag counterpart results in lower throughput which directly translates into lost revenue. The toll is expected to generate revenue as high as 135% of its current revenue during 2019 while 117% in 2020, as the potential revenue decreases as the usage of FasTag increases, we see that the potential revenue has reduced to 19% in 2020 from 32.48% in 2019, this a clear indicator that as the FasTag’s penetration has increased the revenue generation of toll. Higher revenue realization means the operator is in a much better position to repay the debt obligation or raise funds in the future for the upgradation or expansion of the facility without relying on government guarantees. Presently, the toll operator has massive debt obligations and financial difficulties. Since its operation the toll productivity has been between 53–56%, that is out of 100 vehicles that crossed the toll only 53–56 vehicles paid the toll charge, this is hostile to the financial health of the toll operator. Such a high level of revenue loss/shortage also results in a loss of confidence of investors in the ability of toll operators to repay the debt in a timely fashion, forcing them to consider cashing in the guarantees provided by the Government authorities which in turn hurts the Government’s ability for future projects. But, as the adoption of FasTag technology has risen, productivity has also increased. In December 2020, though the FasTag penetration of total Traffic was around 34% the productivity was at an all-time high of 85%. This meant that out of 100 vehicles, every 85 vehicles were paying the toll charges.

342

A. Mishra and A. Mudgal

During the revenue forecasting, 5 models were run by training them on the same data and then testing them by forecasting revenue values from April-2015 to December-2022, with PROPHET being the best model for our dataset.

6 Discussion The results of our study have shown that revenue generation at the toll has been very low than what is expected of it, prior to the introduction of FasTag more than 40% of the traffic using the toll road is not paying for the service which is directly affecting the financial health and putting it grave financial risk one the major concerns for private investors (), in post-FasTag after complete integration and implementation of FasTag system, there was significant increase productivity of the toll with an increase in FasTag penetration, aiding the financial condition of the toll. Deteriorating financial condition is a significant cause of concern for the National Highway Authority of India (NHAI), India’s Highway Authority, its debt has increased by 191.97% during the period from 2015 to 2020, as per NHAI’s official Annual Report for 2018–2019. It had current liabilities of |55,927 Cr. As per Parliamentary Standing Committee’s (PSC) report, NHAI’s debt liability stands at Rs 97,115 crore over the next three financial years- FY22, FY23, and FY24. resulting in a significant chunk of NHAI’s income being utilized to pay debt obligations. Indian Highways need significant investment which has become difficult under the current financial situation of NHAI. Service time analysis shows that FasTag has significantly lower time consumption than the manual cash system which over large volume translates to better revenue generation and also conforms with previous studies conducted regarding E-toll (Ramandanis et al. 2020; Al-Deek et al. 1997) the low capital investment for implementing the FasTag has made it a perfect candidate for E-toll implementation for the Indian environment (Report on ETC Technology 2019). The FasTag also has indirect economic implications, which can be analyzed, the service time reduction will also result in idle time fuel at the toll booth and the corresponding fuel consumption and greenhouse gases emission (Jaiswal and Samuel 2021),100% adoption of FasTag in a no-barrier scenario at which point the toll booth can be an unmanned leading significant reduction in operating cost of the toll plaza, resulting in greater profitability making it a much more attractive investment opportunity. The revenue forecasting in all models has shown a strong upward movement in revenue collection and increasing contribution of FasTag bringing more transparency, stability, and accountability to the Toll Collection System (Report on ETC Technology 2019). With reduced default rate in toll payments and stability in revenue and revenue generation due to FasTag, it has provided a more dependable metric with regards to traffic data and revenue data, which has given the potential to the statutory authorities to keep policies regarding toll road dynamic like toll prices, valuation, toll exemption policy etc., while investors create optimized risk analysis of Toll roads and stability in revenue to enabling the transition of Toll roads from liabilities to assets for private

Economic Analysis of FasTag on Highway Toll Collection

343

investors, which could result in an influx of private investments and rapid growth of road connectivity and investment and growth in road technology.

7 Conclusion The Electronic Toll System (i.e. FasTag) has been a game-changer for toll operators, the authorities, and users ever since its inception. It has improved operational ease for users and toll operators, and helped plugin revenue leakage by reducing the default rate. Both the toll operators and NHAI, now have better stability in revenue collection, centralized and organized database, and an automated clearing system has led to better accountability, transparency and better dispute resolution. New use case scenarios of FasTag have started to develop upon the success of FasTag in toll collection Residential societies have started integrating FasTag in their security system for better tracking vehicles in their premises, Delhi Metro Rail Corporation has now opted to use FasTag for payment of Parking charges showing the flexibility of FasTag for use in urban settings. Absence of segregated data of vehicle class wise tariff details for each year limited our scope of study to make a more in-depth observation regarding defaulting vehicles and narrow down which category of vehicles have been at a greater risk for default and the extent of its impact.

References Action Taken by the Government on the Recommendations/Observations contained in its Two Hundred and Eighty Seventh Report on the Demands for Grants (2021–2022) of Ministry of Road Transport and Highways AI-Deek HM, Mohamed AA, Radwan AE (1997) Operational benefits of electronic toll collection: case study . J Transp Eng Amrin S, An introduction To FASTag: a game-changer in automatic tollcollection systems in INDIA. www.ijrar.org. E-ISSN 2348-1269, P-ISSN 2349-5138) Black WR (2000) A1C06: committee on social and economic factors in transportation chair. Soc Econ Factors Transp. http://onlinepubs.trb.org/onlinepubs/millennium/00100.pdf Delmon J (2009) Private sector investment in infrastructure: project finance, PPP projects and risk. Wolters Kluwer, Chicago Introduction To XGBOOST, https://xgboost.readthedocs.io/en/latest/tutorials/model.html Jaiswal A, Samuel C (2021) Fuel wastage and pollution due to road toll booth. Global J Environ Sci Manag 7(2). https://www.gjesm.net/article_46325.html Levinson (2010) Equity effects of road pricing a review. Transp Rev 30(1):33–57 Nguyen A, Mollik A, Chih Y-Y, Managing critical risks affecting the financial viability of public– private partnership projects: case study of toll road projects in Vietnam NHAI Annual Report (2018–2019). https://nhai.gov.in/nhai/sites/default/files/2020-11/NHAI_ AR_18_19_ENG_for_web.pdf]

344

A. Mishra and A. Mudgal

Ramandanis et al (2020) Assessing the environmental and economic footprint of electronic toll collection lanes a simulation study. MDPI, Sustainability 12(22):9578. https://doi.org/10.3390/ su12229578 Report on ETC Technology (2019). https://nhai.gov.in/nhai/sites/default/files/2019-03/Chairman_ UIDAI.pdf RTI Reply NHAI/Fin/11033/2020-21/RTI for RTI Request No: NHAIN/R/E/21/00092 Taylor SJ, Letham B, Forecasting at scale. https://doi.org/10.7287/peerj.preprints.3190v2 Tseng P-H et al (2013) Investigating the impact of highway electronic toll collection to the external cost: a case study in Taiwan. Technol Forecast Soc Change. https://doi.org/10.1016/j.techfore. 2013.10.019 Yoshino N, Pontines V (2015) AM-adbi-wp549-The “highway effect” on public finance: a case study in Philippines

Sustainable Transportation

Factors Affecting Public Transportation Usage Rate: Geographically Weighted Regression Pankaj Prajapati and Divyesh Abhani

Abstract Many countries with acceptable transportation systems have low public transit usage rates, indicating that local conditions, as well as various other factors, have not been adequately investigated. As a result, it is impossible to develop an appropriate policy for effective and successful public transportation without first understanding the local important factors that influence usage rates. In this study, the Global and Local regression models are used to find the key factors affecting the usage of bus transport in Vadodara city, Gujarat, India. For carried out regression, the city was divided into 78 areas. The study area is spread over 156.19 square kilometers having a population of 16.7 lakhs. This analysis is based on the data collected from 1068 households’ home interview survey. The analysis shows that the Geographically Weighted Regression (GWR) model has higher accuracy in predicting the usage rate of bus transport over Ordinary Least Square (OLS) Regression model. Some independent variables might be not significant at the 5% significance level in OLS model or global regression model which hides the spatial variability or spatial variability ‘Average away’. Independent variables might vary significantly over space and revealed as a significant variable in Geographically Weighted Regression model. In GWR, there is the opportunity to interpret over geographical space and this shows areas where there are both positive and negative correlations between usage rate and a parameter. It is also possible to map the effects of independent variables on the usage of bus transportation across space by Geographically Weighted Regression model. Keywords Public transportation usage · Local factors · Global factors · Local regression model · Global regression model

P. Prajapati (B) · D. Abhani The Maharaja Sayajirao University of Baroda, Vadodara, Gujarat, India e-mail: [email protected] D. Abhani e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Agarwal et al. (eds.), Recent Trends in Transportation Infrastructure, Volume 2, Lecture Notes in Civil Engineering 347, https://doi.org/10.1007/978-981-99-2556-8_26

347

348

P. Prajapati and D. Abhani

1 Introduction In comparison to other factors, which factors have the greatest impact on bus usage? This is a significant subject that has been researched in a variety of methods by numerous scholars. For the identification of factors, a thorough understanding of existing travel patterns is necessary. Detailed data on current travel patterns and traffic volumes are also needed for developing travel forecasting/prediction models. Air pollution and traffic congestion increased due to the speedy growth in the number of private vehicles’ ownerships worldwide. The reliance on private transportation must be reduced for transportation sustainability and economic growth by boosting public transit utilization. However, without consideration of key factors which affect the usage rate of bus transportation, policies can’t be effective. So, for the development of strategies related to bus transportation, identification of key factors are necessary. In this study, the Global and Local regression models are used to find the key factors affecting the usage of bus transport in Vadodara city, Gujarat, India. There is a poor supply of buses in the city. So for finding out the reasons behind the low usage rate of bus transport, this study has been conducted. The work is being done in order to create a Geographically Weighted Regression model that will investigate the geographical variability in the link between bus transportation utilization and variables that influence the use of current urban bus service.

2 Past Research In this section, the models used in previous studies are briefly reviewed along with contributing factors considered in the model. The general motive of the past studies was to compare the GWR and OLS. Previous research on public transportation usage rates was limited to big units, such as countries or cities, and thus did not take into account variances in ward wise characteristics. In addition, the quality of service provided by surrounding transportation systems may have a significant impact on the rate of public transit use (Chiou et al. 2015). However, as a methodology, Geographically Weighted Regression may do spatial analysis with the added benefit of visual interpretation of parameter findings depending on geography (Mulley and Tanner 2009). For data file, all data was treated as it were at the area’s centroid (Fotheringham et al. 2002). Swimmer and Klein calculated ridership by counting the number of trips taken on public transit. The per capita trips were selected as a dependent variable due to the large variance in population density and consumption rate across cities (Swimmer and Klein 2010). Srinivas Pulugurtha explored the riders boarding at the bus stop as a dependent variable (Pulugurtha et al. 2012). Chu created a transit ridership model for an average workday boarding at the bus stop (Chu 2004). Chow et al. (2006) studied the percentage of workers in each Traffic Analysis Zone (TAZ) who used transit for

Factors Affecting Public Transportation Usage Rate: Geographically …

349

their journey to work trip as a dependent variable (Zhao et al. 2005). Ma et al. (2018) analyzed the ridership level at station and the ridership level at region as dependent variables. On station level ridership, average daily boarding and alighting passengers at each station and for region level ridership, summary of passenger count in a TAZ is considered (Ma et al. 2018). Mulley and Tanner examined Vehicle Kilometers Traveled (VKT) as a dependent variable (Mulley and Tanner 2009). Numerous studies have emerged which explored the influencing factors impacting transit ridership. At a bus stop, Kikuchi and Miljkovic constructed transit passenger models that took into account factors such as accessibility to the bus stop, demographic circumstances surrounding the bus stop, bus stop conditions, and transit service quality supplied at the bus stop (Kikuchi and Miljkovic 2001). As independent variables, Swimmer and Klein (2010) considered trip length, metropolitan area, population density, income, commute time, poverty rate, firms per person, gas price index and service availability (Zhao et al. 2005). As the best predictors of transit ridership, Chow identified two global variables, regional accessibility of employment, and percentage of households without a car, as well as three local variables, employment density, average number of cars in households with children, and percentage of population who are black (Chow et al. 2006). According to Swimmer and Klein, the availability of public transportation against the cost of public transportation is the key predictor of ridership. The relative insignificance of price is explained by the fact that most cities heavily subsidies fares, making the provision of public transit a more effective policy instrument (Swimmer and Klein 2010). He et al. (2019) assumed that land use, socio-economics, intermodal transport accessibility and network structures influence the metro ridership. Transit ridership models were developed by Chow using Geographically Weighted Regression (GWR) method to explore the spatial variability between transit use and explanatory variables that include demographic and socio-economic characteristics, land use, transit supply and quality, and pedestrian environment characteristics (Chow et al. 2006). Mulley and Tanner suggested that the dependent variable be changed to its square root to remove the heteroscedastic (non-constant variance) aspect of the error term (Mulley and Tanner 2009). For regression analysis, Swimmer and Klein employed a backwards selection method for regression analysis, deleting a single variable in each regression based on the highest p-value obtained (Swimmer and Klein 2010). Variables having an absolute t-value larger than 1.96 can be regarded as statistically significant at a 5% level of significance for identifying statistically significant variables (Mulley and Tanner 2009). The Akaike Information Criterion (AIC) is a useful metric for assessing model results. The AIC is a tool for calculating goodness of fit that also considers the model’s complexity. If the AICs of the two models differ by more than 3, they are statistically different, with the lower AIC indicating a better match (Fotheringham et al. 2002). The global regression model, according to Fotheringham et al., only estimates a set of parameters for the association between an independent variable and a dependent variable that is constant over space. As a result, the model’s main flaw is that any spatial variation in the relationships between variables is concealed (Fotheringham et al. 2002). GWR, according to Chow, is a promising modeling technique with

350

P. Prajapati and D. Abhani

improved prediction power and a better grasp of the spatial variability in the modeled relationships (Chow et al. 2006). Fotheringham and Charlton suggested that a data point with the absolute value of standardized residual exceeding three is considered as an outlier (Fotheringham et al. 2002). When spatial autocorrelation is present, local models such as Geographically Weighted Regression (GWR) are advised. According to a study of the studies mentioned above, the explanatory variables investigated can be split into three categories: Demographic and Socio-economic (viz. adults, college persons, employed persons per household, age, income and total trips per household etc.), Availability of private transport (viz. two wheeler ownership, car ownership and road density etc.) and Accessibility to public transport (viz. bus stops, home to bus stop distance, access distance, waiting time, egress distance and travel cost etc.).

3 Methodology The aim of this paper is to create a bus transit usage rate model and investigate the impact of various variables. For comparisons, two approaches, OLS and GWR, have been developed and are briefly described here.

3.1 Ordinary Least Square Regression Ordinary Least Square regression models are widely used to simulate the relationship between one or more independent factors and a single dependent variable in public transport usage rates. A global regression model, according to Fotheringham et al. (2002), takes the following form: y = β0 + β1 x1 + β2 x2 + · · · + β p x p + ε

(1)

where y = Dependent variable, xj = jth independent variables or predictors ( j = 1 …, p). βj = jth model parameters to be estimated ( j = 0, 1 …, p). E = Error term. There are p + 1 model parameters for a model with p independent variables that do not vary regardless of where the model is to be applied. The fitted dependent variable values are subtracted from the observed dependent values to produce the error values.

Factors Affecting Public Transportation Usage Rate: Geographically …

351

3.2 Geographically Weighted Regression Individual point or area observations may be included in the data required to calibrate a GWR model. If the observations are about a certain area, a representative place from that area should be used to represent the data; generally, the centroid of the area is used as a representative point of the area. These locations refer as regression points (Fotheringham et al. 2002). The dependent variable y is predicted by a set of independent variables, the coefficients βj ( j = 0, 1 …, p) which may vary by location in a GWR model. A local regression model, according to Fotheringham et al. (2002), takes the following form: y = β0 (u i , vi ) + β1 (u i , vi )x1 + β2 (u i , vi )x2 + · · · + β p (u i , vi )x p + εi

(2)

Here, β j ( j = 0, 1 …, p) is now a function of location (ui, vi). This means that, for the same xj ( j = 1, 2 …, p) values, the equation may give different predictions of the y value depending on the location where x j is measured. Ei = Error term, the fitted dependent variable values are subtracted from the observed dependent values to produce the error values. Each data point in GWR is weighted according to its distance from the regression point. As a result, data points closest to the regression point in the local regression are weighted more heavily than data points further away. For a given regression point, the weight of a data point is maximum or unity if it shares same location as the regression point. There are several weighting schemes available, but among them, two are commonly adopted: Gaussian weighting function and Bi-square weighting function. Gaussian weighting function (Fotheringham et al. 2002). Used for fixed kernel. w j (i ) = exp[−(di j /b)2 ]

(3)

Bi-square weighting function (Fotheringham et al. 2002). Used for adaptive kernel.  2 1 − (di2j /b2 ) , if di j ≤ b (4) w j (i ) = 0 if di j > b where i, j = 1, 2, …, n. dij = distance between regression point i and data point j. b = bandwidth.

352

P. Prajapati and D. Abhani

3.3 Kernels The estimated coefficients of the independent variable at any regression point are not dependent only on the data supplied but also on the kernel chosen and the bandwidth for that kernel (Fotheringham et al. 2002). The kernel is a group of neighboring observation points of regression points within a specified bandwidth.

4 Data and Analysis A household survey was conducted among 12 Administrative wards of Vadodara city. In this survey, a total 1089 households were surveyed through Random Sampling. There are 21 households with insufficient information, were discarded from the total sample of 1089 household data. Data of 1068 households and 5771 number of trips were taken for the study purpose. The survey form includes Demographic and Socio-economic characteristics and travel pattern data. Details collected for the trips made by the household on the previous day, also contains information like total trips per day, various modes used for trips, purpose of trips, trip origin and destination, departure time from origin and arrival time at destination, access and egress distance, journey time, in vehicle travel time, travel cost etc. For bus routes analysis, total 68 bus routes are digitized. For bus stops details, up routes of city bus is considered. Up routes mean route from city bus station to a particular destination. Total 435 bus stops are digitized for analysis purposes. The ArcGIS 10.3 and GWR 4.0 software were used to estimate two models.

4.1 OLS Results The 78 regions of Vadodara city are the study area. Initially, the OLS is performed for 30 independent variables. The back elimination method is used to determine the factors affecting public transportation usage rate. In each regression, one factor with the highest value of Variance Inflation Factor (VIF) is eliminated. The VIF value of factors determines the spatial pattern of a particular factor compared to other factors. It represents the correlation between the various factors (ESRI Regression Analysis). The significance level (α) was set at 0.05. For a confidence interval of 95%, if the t-value was