PhD Final Defense – Shirin Qiam

A Geospatial Analysis of Transportation Infrastructure: Measuring and Analysing Parking Lots and Bus Stops in U.S. Cities

Advisor: Professor Lewis Lehe

Abstract

Transportation systems are widely discussed in planning and policy, yet many of their

key characteristics remain difficult to measure and analyze at large scales. While transportation

research has traditionally emphasized normative questions of how infrastructure

and services should be designed, empirical understanding of how transportation systems are

organized in practice is often constrained by data limitations and methodological challenges.

This dissertation addresses these challenges by developing and applying a framework that

combines machine learning, geospatial data processing, and statistical modeling to construct

and analyze large-scale datasets describing transportation infrastructure and networks. The

framework is demonstrated through two applications: surface parking supply and bus stop

spacing.

The first part of the dissertation focuses on parking lots. A semantic segmentation

framework is developed to detect and map surface parking lots from satellite imagery. Using

a dataset of more than 12,000 annotated satellite images, the study evaluates multiple deep

learning architectures and demonstrates that the incorporation of near-infrared imagery and

post-processing techniques improves segmentation performance. The resulting framework is

used to construct a large-scale dataset of parking lot footprints across 15 major U.S. cities,

providing one of the largest open-source spatial inventories of surface parking currently

available. Building on this dataset, the dissertation examines how parking provision varies

with population, employment, land use composition, spatial structure, and land value. The

results show that employment activity is a particularly strong predictor of parking provision,

while higher land values are consistently associated with lower levels of parking. The findings

suggest that parking is shaped not only by transportation demand but also by broader

economic and spatial forces that govern urban land use.

The second part of the dissertation examines bus stop spacing as a measurable characteristic

of transit network structure. Using General Transit Feed Specification (GTFS)

data and multilevel within-between models, the study analyzes stop spacing across 48 major

U.S. transit agencies. The results indicate that spacing patterns are associated with characteristics

of the built environment as well as the historical development of urban areas. In

particular, indicators of older urban development are associated with narrower stop spacing,

while employment density is consistently associated with wider spacing. The analysis further

suggests that much of the systematic variation in spacing occurs between routes rather than

within routes, providing insight into how spacing decisions are made in practice.

Together, these applications demonstrate a common workflow in which transportation system characteristics are first transformed into measurable datasets and then analyzed

using statistical methods appropriate for spatially structured and hierarchical data. Beyond

the substantive findings regarding parking supply and bus stop spacing, the dissertation

contributes new datasets, tools, and methodological approaches for studying transportation

systems empirically at large scales. More broadly, it illustrates how advances in machine

learning, geospatial data, and computational methods can expand the scope of transportation

research by making difficult-to-measure aspects of transportation infrastructure and networks

observable and analyzable.