How do built environment factors influence urban heat islands across local climate zones? Evidence from an interpretable XGBoost-SHAP model

Urban heat islands (UHIs), shaped by both two-dimensional land-cover patterns and three-dimensional urban morphology, pose growing challenges to public health and energy systems. This study investigates microclimate variations in 16 residential areas of Xi'an using synchronous, high-temporal-resolution measurements from 105 sampling points. We develop an LCZ-based spatial machine-learning framework integrating XGBoost and SHAP to quantify the nonlinear impacts of built-environment indicators on near-surface temperature, humidity, and heat index. Results reveal a clear hierarchy of influence, with three-dimensional morphological indicators exerting the strongest effects, followed by landscape-ecological and two-dimensional morphological indicators. Among all variables, NDVI, sky view factor (SVF), building density (SDH), and street compactness (SCD) contribute most to microclimate regulation. Within the LCZ framework, we further identify pronounced spatial heterogeneity in thermal responses across different urban forms. Distinct synergistic interactions are also observed: substantial cooling and humidifying effects occur when NDVI > 0.3 and SVF > 0.3, while blocks characterized by low SCD (< 0.06) and high SDH (> 20) exhibit clear heat-mitigation responses. By combining detailed LCZ information, in-situ observations, and explainable machine learning, this study provides quantitative evidence linking specific morphological configurations to thermal performance and offers a transferable, climate-responsive planning framework for high-density residential environments.