Sustainable Transit through Public Agency Leadership: Policies, Economics, and AI

Abstract

This dissertation integrates three empirical studies to advance planning and decision-making for Transit-oriented Development (TOD) and Transit-Incorporating Mobility-on-Demand (TIMOD) services led by public agencies in regional growth centers and low-density areas. The first study develops a multi-criteria tool to prioritize TOD on public land through a suitability assessment. The tool incorporates 14 indicators across five domains, including public transit service, land use, sociodemographics, real estate market conditions, and planning context. It is applied under three development scenarios: affordable housing, market-rate housing, and mixed-use development. This tool enables public agencies to strategically leverage land assets to support transit-supportive and equitable development.The second study evaluates the comparative cost-effectiveness of TIMOD relative to fixed-route transit, driving, and commercial ride-hailing services in low-density settings. A comprehensive social cost estimation framework is employed, incorporating traveler and service provider costs alongside transportation externalities. Findings indicate that TIMOD generally incurs higher social costs than other mobility alternatives due to its higher operating costs, while it provides time-cost savings and improved access for travelers in areas with limited fixed-route transit coverage. These insights inform more nuanced subsidy strategies and service deployment models for underserved populations in such contexts. The third study develops a context-aware probabilistic spatiotemporal graph neural network (CA-STPGNN) to predict sparse demand for TIMOD services and quantify the associated uncertainty in low-density areas. The model integrates contextual information, temporal features, and multi-task learning as regularization to capture spatiotemporal dependencies in TIMOD trip patterns. Using data from real-world TIMOD programs in Washington State, the proposed model demonstrates superior predictive accuracy and uncertainty estimation compared with traditional approaches, while also revealing spatial heterogeneity in the influence of spatial and temporal features. This model can be applied to predict TIMOD demand in regions lacking observed data, thereby supporting public agencies in design and implementing cost-effective mobility solutions in low-density contexts. Together, these three studies contribute adaptive tools, empirical evidence, and methodological innovations to support equitable, efficient, and context-sensitive transportation and land use planning in low-density contexts.