Computational Methods for the Identification and Comparative Analysis of Urban Form Types: A Case Study of Rust Belt Cities

 This dissertation presents a novel methodological framework that allows for the comparative analysis of city structures by identifying urban form types, interweaving a data scientific perspective into urban morphology. Proposing a concept of neighboring urban form around individual buildings as a key constituent of the city, a new kind of urban form dataset and method to identify urban form types is introduced. This framework was applied to three Rust Belt cities—Cleveland, Detroit, and Pittsburgh—as a case study. A deep learning model was employed to capture the morphological features from the urban dataset for these cities, clustered them to yield types of urban spaces, and examined the morphological characteristics of each type. These types delineate patterns of urban fabric and reveal urban structures distinctive of Rust Belt cities. Through site surveys and expert interviews, I validated the proposed method’s capacity to provide more granular classification of urban forms and lead to more nuanced understanding of city space than conventional urban anal?ysis methods. This dissertation contributes to the field of urban form studies, opening a new perspective to the analysis of urban spaces and redefining urban types and fabric from data science.