Principal Component Analysis and Fuzzy Clustering for Neighborhood-Scale Urban Morphology-Based Thermal Hazard Assessment

Urban heat islands pose significant public health concerns, yet neighborhood-scale thermal assessment faces data limitations. Satellite-derived surface temperatures provide spatially complete coverage but measure surface radiative properties rather than pedestrian-level air temperature. Meanwhile, sparse meteorological networks cannot resolve intra-urban heterogeneity. Morphology-based approaches address this gap by using urban form characteristics as physical determinants of local thermal conditions through their influence on ventilation, shading, and heat storage. However, morphological indicators are inherently correlated, causing multicollinearity, and conventional classifications impose rigid boundaries that misrepresent transitional zones. This study develops a framework combining Principal Component Analysis (PCA) and Fuzzy C-Means (FCM) clustering to address both challenges, applied to Warsaw (Poland), using nine urban morphology variables computed at 10m resolution with a 310m neighborhood analysis window and strict multicollinearity control (VIF1) explaining 65.6% of variance: PC1 (43.3%) represented the urban-natural gradient, while PC2 (22.3%) distinguished water-influenced zones from vegetated areas. FCM clustering identified nine morphological typologies. A novel membership transformation approach aligned fuzzy membership degrees with relative thermal patterns derived from averaged Landsat surface temperatures (summer 2019-2025), orienting values toward thermal hazard magnitude rather than cluster proximity. The highest-risk cluster (compact mid-/high-rise built-up areas, mean LST=36.0°C) was further stratified into five thermal hazard classes, revealing high building density (BCR=0.56) and minimal vegetation (0.5%) in Very High hazard zones. The framework provides actionable information for targeted heat mitigation while quantifying classification uncertainty in transitional zones, offering a generalizable approach for cities with adequate morphological data.