The ongoing urbanization process is swiftly giving rise to megaregions, reshaping the urbanlandscape. Travel constitutes a vital aspect of urban life; hence, a comprehensive understanding of traffic dynamics is crucial for the efficient management of cities. Traffic dynamics refer to the patterns of movement exhibited by people and vehicles within a transportation system. These dynamics are typically characterized in terms of flow and speed, providing crucial insights into the functioning of the urban transportation system.

Traffic simulators are extensively used to analyze traffic dynamics in cities and regions.However, many traffic simulators used in regional studies have limitations. One primary limitation is the substantial data and computational requirements necessary to model a large urban region with high fidelity and speed. To tackle this challenge in traffic simulators, many researchers reduce the size of the road network, including only major arterials and highways; model a subset of the population; and/or aggregate travel demand to a smaller subset of network nodes, aiming to obtain rough estimates of travel volumes. However, the downsizing of the road network and the reduction/consolidation of demand lead to alterations in the characteristics and performance of the network. This approach can result in highly inaccurate predictions that fail to capture the actual dynamics and behavior of traffic. It can provide misleading information to agencies regarding the necessary investments for constructing, maintaining, or improving their roadway infrastructure, as well as decisions related to traffic management and incident response plans.

To address these gaps, this dissertation provides a detailed characterization of large-scaleregional traffic dynamics across diverse scenarios using a high-performance traffic simulator. It contributes to and leverages a scalable high-performance mesoscopic traffic simulation platform named Mobiliti, which incorporates various routing strategies to model real-world traffic conditions with high fidelity and speed. In the case of San Francisco Bay Area, Mobiliti can simulate 19 million vehicle trips on a road network with approximately 0.5 million nodes and 1 million links, representing 7 million drivers and 4 million truck trips in less than three minutes. Subsequently, using data analytics and machine learning models, we identify traffic patterns and characterize cities based on transportation-oriented typologies for large metropolitan regions.

Leveraging Mobiliti alongside real-world data sources, we delve into various facets of regionaltraffic dynamics, covering both novel and critical areas. Our investigation began by examining dynamic routing, a prevalent feature introduced by the widespread adoption of navigation apps. This feature introduces an additional layer of traffic control, thereby altering traffic dynamics on streets. Through modeling the varying penetrations of dynamic routing, we quantified its effects on the San Francisco Bay Area using metrics such as Vehicle Miles Traveled (VMT), Vehicle Hours of Delay (VHD), affected trips, and its impact on local roads, among others. Next, we recognized that different types of traffic routing, for example, prioritizing time savings versus prioritizing fuel efficiency, influence traffic dynamics in distinct ways. These dynamics, in turn, shape our cities and significantly impact quality of life. Consequently, we developed a framework to analyze these complex dynamics. We evaluated the impact of different routing strategies across multiple dimensions, examining their effects on neighborhoods, safety, environment, and more. Thirdly, network resilience is a critical factor in the San Francisco Bay Area, known for its interconnected bridges. Incidents in this region have the potential to disrupt multiple cities, impacting productivity and energy efficiency. Therefore, we focused on examining the vulnerability of the transportation system to such events and the cascading effects they may cause. A deeper understanding of these dynamics can assist in effective event response planning. The final two studies focus on cities within metropolitan regions. We begin by analyzing street network structures of cities and then proceed to incorporate additional transportation dimensions such as travel demand, traffic flow, and infrastructure. By employing clustering techniques, we identify city typologies and their respective characteristics. This analysis offers us the opportunity to reflect on past urban development efforts, learn from one another, and envision our future city-building objectives.

In total, the specific contributions of this dissertation are:

1. Analyzed the effects of dynamic routing and its varying penetration rates across avast metropolitan region using large-scale discrete event simulations, demonstrating its substantial influence on mobility metrics at a regional scale. Previous studies were constrained by geographic scale and a limited number of simulation runs, thus failing to capture the full impacts of dynamic routing. 2. Developed a novel multi-themed analytical framework called the Socially-Aware Evaluation Framework for Transportation (SAEF), aiding in comprehending how traffic routing and resulting dynamics impact cities within a region. Our framework’s indicators are carefully chosen to detect system changes when routing strategies are altered, with a focus on neighborhood-related indicators, often overlooked in existing frameworks. 3. Enhanced the evaluation of large-scale network disruptions by modeling dynamic route choices for travelers within a full-scale urban network, encompassing an entire day’s demand. Our method more realistically captures drivers’ behavior during incidents and we are able to capture the full impacts of incidents at scale, thus enabling the creation of better traffic management and response strategies. 4. Created city typologies for all cities within a metropolitan region based on street network structure, which can provide valuable insights into how drivers experience a city based on its street layout. To aid in this classification, we introduced a new metric for categorizing intersections that distinguishes between various types of 3-way and 4-way intersections based on geometric angles. By incorporating this geometric metric alongside existing centrality metrics, we have achieved better differentiation for improved typology generation, capturing the nuances between grid and other street typologies more effectively. 5. Developed transportation-oriented city typologies based on various dimensions including traffic flow, trip demand, multi-modal network, land use, and road network. These typologies serve as a foundation for facilitating the effective exchange of policies and resources, relying on a thorough understanding of traffic characteristics. We integrated metrics related to trip demand and traffic flow alongside commonly used metrics from road network, multi-modal network, and land use. This integration is crucial for capturing the travel behavior and traffic dynamics of cities, enabling the generation of meaningful and comprehensive typologies.

Each of the items mentioned above is described in more detail in the following paragraphs.

In the second chapter, we examined dynamic routing and its impact on large urban areasusing the Mobiliti traffic simulator. Over the last few decades, navigation apps have introduced a new level of traffic control and warranting study as they become pervasive and dictate street traffic flows. Previous work on dynamic routing has been constrained by limited geographic scale and a small number of simulation experiment runs, often requiring hours to complete a single simulation. This limitation poses a bottleneck for running multiple simulations and testing various what-if scenarios to identify the full range of rerouting impacts. We address this gap by utilizing high-performance parallel computing, large urban scale simulator Mobiliti, which can run a single simulation for the entire San Francisco Bay area in less than 10 minutes. We ran multiple simulations with varying penetration rates, revealing diminishing benefits of rerouting after a 70% penetration rate. We also found that dynamic rerouting effectively reallocates vehicle flows from heavily utilized highways and arterials to less congested neighborhood links, reducing overall system delay. Interestingly, the increased traffic volume on local roads does not always lead to congestion, as many links do not reach congested levels despite the increased flow. In summary, our analysis demonstrates, for the first time, the effects of varying penetration rates on traffic dynamics at a regional scale.

In the third chapter, we present an analytical framework called Socially- Aware EvaluationFramework for Transportation (SAEF), which assists in understanding how traffic routing and the resultant dynamics affect cities across a region across multiple dimensions. With the proliferation of real-time navigation routing apps, traffic dynamics in urban environments have changed, resulting in undesired effects that compromise safety and neighborhood health. Therefore, understanding these disparities in traffic distribution across various dimensions is crucial for decision-makers. While previous studies have created frameworks to assess the effects of wide-ranging transportation infrastructure changes or the adoption of smart city technologies, none have established a framework with indicators specific enough to capture the impacts of various traffic routing strategies on cities. Furthermore, existing frameworks and metrics lack translatability to identify the impacts of routing strategies, as crucial dimensions like safety or neighborhood considerations are not adequately addressed. Therefore in this work, our first contribution is developing a framework with a set of themes and indicators that can capture the impact of traffic routing holistically. We identified relevant indicators from the literature, organizing them into four themes: neighborhood, safety, mobility, and environment. When necessary, we developed new methodologies to calculate these indicators. A second contribution is the application of SAEF framework to four cities in the Bay area in the context of three different routing strategies - user equilibrium travel time, system optimal travel time, and system optimal fuel. The four cities were compared to understand how city structure and urban form play a role in the distribution of traffic dynamics. The results demonstrate that many neighborhood impacts, such as traffic load on residential streets and around minority schools, degraded with the system-optimal travel time and fuel routing in comparison to the user-equilibrium travel time routing. The findings also show that all routing strategies subject the city’s disadvantaged neighborhoods to disproportionate traffic exposure. With the widespread adoption of navigation apps, our intent with this work is to provide an evaluation framework that enables reflection on the consequences of traffic routing, allowing city planners to recognize the trade-offs and potential unintended consequences.

In the fourth chapter, we offer a set of evaluation tools designed to measure the impact of significanttransportation disruptions on a regional scale. We illustrate the application of these tools through a case study involving the closure of the Richmond-San Rafael Bridge in the San Francisco Bay Area. Evaluating the dynamics of transportation networks in the context of events can inform disaster plans and aid in traffic management strategies in preparation for or during an event. Existing research on road network disruptions often relies on short time frames and small-scale models, largely due to computational limitations that hinder the widespread adoption of large-scale urban simulation models. Consequently, smaller-scale micro-simulation models are commonly preferred for designing response plans, typically targeting selected highways and major arterials in close proximity to incidents. However, these studies face three key limitations. Firstly, they often rely on user-equilibrium assumptions for route choice, which fail to adequately reflect realistic driver behavior during incidents. Secondly, they use reduced road network representations due to computational constraints, typically focusing on small areas surrounding closures. Thirdly, they frequently extrapolate findings from peak periods to estimate daily impacts, potentially overestimating congestion due to differences in traffic dynamics between peak and non-peak periods. To address these gaps, our study employs a large-scale, mesoscopic simulation model with dynamic routing capability. This model enables us to simulate a full-scale urban network with an entire day’s demand, allowing for a comprehensive assessment of the regional traffic impact of the incident. Our findings indicate that the region experienced an additional 14,000 vehicle hours of delay and 600,000 vehicle miles due to the bridge closure. Furthermore, the median traffic volume on neighborhood streets in San Francisco, Vallejo, and San Rafael increased by more than 10%, highlighting the role of local roads in accommodating the traffic overflow, a factor often overlooked in prior studies. With large-scale modeling of a critical network disruption using dynamic rerouting capability, complete road network, and full demand, we provide valuable insights into the response dynamics of this specific event. In doing so, we demonstrate the value of such regional analyses to incident and disaster planning.

In the fifth chapter, we developed typologies to classify cities within a metropolitan area accordingto their street network characteristics. Spatial networks such as streets and transit lines influence urban dynamics and travel behavior. Analysing these patterns can also help identify how drivers experience city streets and understand the unique characteristics and challenges present in each urban environment. While previous studies have investigated global network patterns for cities, they have often overlooked detailed characterizations within a single large urban region. Additionally, most existing research uses metrics like degree, centrality, orientation etc., and misses the nuances of street networks at the intersection level, such as geometric angles formed by links at intersections, which could offer a more refined feature for characterization. To address these gaps, this study examines 94 cities in the San Francisco Bay Area, taking into account diverse road network features. We introduce a novel metric for classifying intersections, distinguishing between various types of 3-way, and 4-way intersections based on the angles formed at the intersections. Through the application of clustering techniques in machine learning, we have identified three distinct typologies - grid, orthogonal, and organic cities - within San Francisco Bay Area. Gridded cities are distinguished by their dense network of right-angled four-way and three-way intersections. These cities exhibit a compact layout with smaller link lengths and slower traffic speeds. On the other hand, orthogonal cities exhibit a street network configuration characterized by right-angled three-way intersections and longer street lengths. Organic cities represent a third typology, characterized by their irregular and non-grid-like street network. These cities feature long links with numerous dead ends and winding, circuitous roads. Our findings indicate that the integration of the new metric has improved our ability to distinguish between different types of cities, complementing the existing metrics. In gridded cities, the introduction of the new metric enhances the recognition of grid patterns by explicitly considering 90-degree intersection angles. Conversely, for non-gridded cities, a notable advancement is the ability to differentiate between various types of degree 3 nodes (3-way intersections). While many cities have a significant number of degree 3 nodes, the arrangement of these intersections can vary greatly due to angle variations, resulting in either 90 degree T intersections or non-T intersections. Our study showcases the effectiveness of the new metric in capturing these distinctions, facilitating the classification of cities with a high proportion of T intersections into orthogonal cities and those with non-T intersections into organic cities. The significance of this differentiation extends to how drivers navigate and experience intersections and streets within cities. Based on the angles, turns, and curves of the road network, driving experiences vary significantly. Therefore, understanding these nuances is crucial for optimizing traffic flow, enhancing road safety, and improving overall driving experiences for motorists.

In the sixth chapter, we expanded upon our previous city characterization work focused onnetwork structure by incorporating multiple transportation dimensions. As cities evolve and face shared challenges, the development of city typologies, rooted in a comprehensive understanding of traffic characteristics, becomes crucial for facilitating the effective exchange of policies and resources among them. Prior work on transportation based city typologies often fails to provide characterizations specific to a single extensive urban area, as it predominantly focuses on cities globally. Furthermore, these studies frequently overlook essential dimensions such as trip demand and traffic flow in their characterizations, despite their significant impact on street behavior and traffic dynamics. Therefore in this study, we develop a transportation-focused characterization for all cities within a large urban region, specifically the San Francisco Bay Area, California. We incorporate over 40 metrics across five transportation dimensions: road network, trip demand, traffic flow, multi-modal network, and land use. Using factor analysis and unsupervised machine learning clustering methods, we identified eight distinct typologies for the Bay Area: Live Work; Job and Activity Magnets; Anchor Cities; Multi-modal; Hyper-connected; Low-density residential; Mediumdensity Residential; Mixed-use residential. The results revealed that many clusters were characterized by features from travel demand and traffic flow dimensions, thus signifying their importance in typology generation. These typologies can serve as a basis to create discourse among Bay Area cities and determine if, through success/failure experiences, common strategies can be formed.

In total, the analytical framework and methods outlined in this dissertation provide detailedand nuanced insights into regional traffic dynamics, surpassing existing literature. By utilizing and contributing to the Mobiliti simulator, we modeled large urban areas with high fidelity and speed, enabling the testing of multiple “what if”scenarios for large metropolitan regions. Our investigation of dynamic routing and its varying penetration rate in Chapter 2 represents the first large-scale regional study examining the impact of real-time traffic routing. Furthermore, the SAEF framework presented in Chapter 3 of this dissertation represents the first analytical framework that captures the impact of traffic routing holistically. With the widespread adoption of navigation apps, this framework enables reflection on the consequences of traffic routing, allowing city planners to recognize the trade-offs and potential unintended consequences. The large-scale network disruption evaluated in Chapter 4 provides a suite of analytical tools for assessing disruptions at both regional and local levels. These tools enable the creation of enhanced traffic management and response strategies by capturing driver behavior more realistically. The typologies developed in Chapters 5 and 6 provide a comprehensive understanding of cities in a region, considering both network structure and overall transportation dimensions. The new metric introduced in Chapter 5 aids in quantifying the network more precisely, while the comprehensive use of various metrics from different transportation dimensions, particularly trip demand and traffic flow, facilitates a more thorough characterization of cities in Chapter 6. The identified typologies can catalyze dialogue among San Francisco Bay Area cities, facilitating the exploration of common strategies derived from shared experiences of success or failure. Ultimately, the findings presented in this dissertation contribute not only to enriching academic discourse on transportation dynamics but also carry practical implications for policymakers. They furnish invaluable guidance for crafting more effective and nuanced traffic management strategies for cities and large metropolitan regions, thereby shaping the future of urban mobility with precision and foresight.

Recent technological improvements have greatly expanded the sharing economy (e.g., Airbnb, Lyft, and Uber), coinciding with growing need for transportation and sheltering resources in evacuations. To understand influencers on sharing willingness in evacuations, we employed a multi-modeling approach across four sharing scenarios using three model types: 1) four binary logit models that capture each scenario separately; 2) a multi-choice latent class choice model (LCCM) that jointly estimates multiple scenarios via latent classes; and 3) a portfolio choice model (PCM) that estimates dimensional dependency. We tested our approach by employing online survey data from 2017 Hurricane Irma evacuees (n=368).

The multi-model approach uncovered behavioral nuances undetectable with a single model. First, the multi-choice LCCM and PCM models uncovered scenario correlation, specifically willingness to share for both transportation scenarios and both sheltering scenarios. Second, the multi-choice LCCM found three classes – transportation sharers, adverse sharers, and interested sharers. Transportation sharers were more likely to be female, lower-income, and residents of Southwest Florida compared to adverse sharers. Interested sharers were more likely to be male, long-time residents, and higher-income compared to adverse sharers. Third, families with children were unwilling to share regardless of the model, while spare capacity (i.e., seatbelts, spare beds) had a positive but somewhat insignificant influence on sharing. Fourth, experienced home sharers were more willing to share shelter in the binary logit and PCM models. We suggest that local agencies consider holistic sharing mechanisms across resource types and time (i.e., before, during, and after a hurricane evacuation).