Urban Forms Across Continents: A Data-Driven Comparison of Lausanne and Philadelphia
Arthur Carmès, Léo Catteau, Andrew Sonta, Arash Tavakoli
TL;DR
The paper addresses cross city comparisons of urban form by proposing a data driven framework that uses OpenStreetMap features, a hexagonal BSU grid, and Gaussian Mixture Models to identify neighborhood typologies in Lausanne and Philadelphia. It demonstrates adaptive grid sizing and multi feature clustering to reveal both shared and city specific patterns, and shows that even a single feature such as degree centrality can yield meaningful urban structure insights. The study finds evidence of functional convergence across continents and emphasizes the critical role of scale in cross city analysis, offering a scalable toolkit for planners to enhance walkability and accessibility. While data completeness and feature selection pose limitations, the framework provides a transferable approach for comparative urban studies with practical implications for policy and design.
Abstract
Understanding urban form is crucial for sustainable urban planning and enhancing quality of life. This study presents a data-driven framework to systematically identify and compare urban typologies across geographically and culturally distinct cities. Using open-source geospatial data from OpenStreetMap, we extracted multidimensional features related to topography, multimodality, green spaces, and points of interest for the cities of Lausanne, Switzerland, and Philadelphia, USA. A grid-based approach was used to divide each city into Basic Spatial Units (BSU), and Gaussian Mixture Models (GMM) were applied to cluster BSUs based on their urban characteristics. The results reveal coherent and interpretable urban typologies within each city, with some cluster types emerging across both cities despite their differences in scale, density, and cultural context. Comparative analysis showed that adapting the grid size to each city's morphology improves the detection of shared typologies. Simplified clustering based solely on network degree centrality further demonstrated that meaningful structural patterns can be captured even with minimal feature sets. Our findings suggest the presence of functionally convergent urban forms across continents and highlight the importance of spatial scale in cross-city comparisons. The framework offers a scalable and transferable approach for urban analysis, providing valuable insights for planners and policymakers aiming to enhance walkability, accessibility, and well-being. Limitations related to data completeness and feature selection are discussed, and directions for future work -- including the integration of additional data sources and human-centered validation -- are proposed.