Dissertation Defense - Tunaggina Khan - Department of Geography and Geoinformation Science

PhD Candidate: Tunaggina Khan
PhD of Science, Earth Systems and Geoinformation Sciences
Department of Geography and Geoinformation Science

Dissertation Title: Towards Fast, Scalable, and Traffic Informed Urban Transportation Modeling

Dissertation Chair: Dr. Dieter Pfoser, Professor, Department of Geography and Geoinformation Science

Committee Members: 
Dr. Taylor Anderson (George Mason University, Department of Geography and Geoinformation Science)
Dr. Hamdi Kavak (George Mason University, Department of Computational and Data Sciences)
Dr. Andreas Züfle (Emory University, Department of Computer Science)

Abstract:

For effective transportation planning, it is very important to understand and balance between accurate representation of transportation systems and computational efficiency. Agent-based models, such as MATSim, accurately capture detailed traffic dynamics, but their computational complexity limits scalability and running time. Conversely, while shortest-path based routing engines such as GraphHopper offer fast and scalable routing, they do not model or consider traffic in their route calculations.

This dissertation develops lightweight and scalable traffic models that can generate accurate traffic dynamics with computational efficiency and scalability. Using a high-resolution, multimodal MATSim simulation of the Washington, D.C. metropolitan area as a benchmark, this dissertation first validates MATSim's accuracy in capturing fine-grained travel patterns across multiple travel modes, such as individual vehicles, public transit and walking. Those insights from MATSim guide the design of lightweight, traffic-aware routing models that approximate MATSim output without the computational cost of full agent-based simulation.

To bridge the gap between accuracy, efficiency and scalability, we propose GraphHopper as a lightweight alternative to resource intensive agent-based model. Both individual level trips and aggregate level traffic patterns, generated by GraphHopper are evaluated against MATSim. GraphHopper is then extended with congestion-awareness, with implementing various edge-weight modification strategies, guided by MATSim outputs and by traffic volumes inferred from edge traversal counts within GraphHopper. 

The research in this dissertation demonstrates that, with targeted enhancements, static routing models can effectively integrate traffic information into routing and can approximate the effects of congestion and adaptive route choices, without the computational burden of full agent-based simulations. The findings provide a practical framework for traffic-informed congestion-aware transportation models that are accurate and also computationally efficient and scalable at the same time. These models enable faster scenario testing, policy evaluation, and emergency planning in large urban regions without sacrificing essential behavioral detail or model accuracy.