Assessment of thermal-visual exposure trade-offs in high-density urban environments using explainable machine learning and spatial optimization

Abstract

High-density urban environments present competing environmental exposures that jointly influence human health. Increased sky openness enhances visual quality but simultaneously intensifies thermal stress, while shading strategies that reduce heat exposure may constrain restorative benefits. Such trade-offs pose challenges for environmental assessment and urban design in dense urban settings. This study proposes a spatially explicit framework to jointly assess and optimize thermal comfort and visual restorative potential at the panorama level, integrating street-view image-based spatial metrics, XGBoost with explainable SHAP analysis, and Genetic Algorithm-based optimization, enabling interpretation of non-linear trade-offs that prior separate-domain analyses could not resolve. Using nine residential neighborhoods in Singapore, the Universal Thermal Climate Index (UTCI) and Perceived Restorativeness Scale Score (PRSS) were predicted from image-based metrics and synthesized into an Integrated Thermal-Visual Score (ITVS). SHAP analysis revealed that Sky (31.9%), Tree (20.8%), and Depth (13.0%) were the dominant predictors with non-linear patterns: Tree > 0.15 and Depth > 0.35 improved integrated performance, while Sky > 0.06 degraded it as thermal penalties outweighed visual gains. Optimization achieved a mean UTCI reduction of 0.327 °C and PRSS increase of 0.535. Cluster analysis further identified three morphological typologies, i.e., open-exposed, dense-shaded, and green-balanced, with optimization gains ranging from 22.1% to 32.5%, underscoring the need for morphology-specific strategies. These empirically derived thresholds and the integrated optimization framework offer transferable, quantitative guidance for performance-oriented urban design in high-density tropical cities.