Exploring the built environment impacts on urban heat exposure: An autoencoder-based approach for integrating multiple indicators.

Urban heat exposure (UHE) has long been recognized as a critical challenge in urban planning and environmental studies. Previous research, however, has primarily modeled objective and subjective heat exposure in isolation or merged them via simplistic multiplicative approaches, overlooking the methodological necessity of integrated assessment. To address this gap, we introduce a novel evaluation framework that leverages a knowledge-guided autoencoder to systematically integrate objective and subjective indicators of heat exposure. Furthermore, the TabPFN-SHAP model was utilized to quantify the influence of built environment factors and elucidate the underlying response mechanisms governing urban heat exposure. The results demonstrate that: (1) Heat exposure intensity and spatial distribution differ considerably across analytical perspectives. (2) Points of interest (POI) related to commercial and life services are predominantly concentrated in moderate to relatively high exposure areas, whereas public and recreational POIs are mainly situated in relatively low to moderate exposure areas. (3) Transport station density, the proportion of blue-green space and residential land emerge as the three dominant factors significantly affecting UHE from all perspectives. (4) A high proportion of residential land and older building ages exhibit a pronounced antagonistic effect, contradicting the conventional assumption that older residential neighborhoods necessarily intensify heat exposure. These findings deepen the understanding of UHE mechanisms and provide actionable guidance for urban thermal environment management.