Unveiling the Non-Linear Associations Between 3D Building Morphology and Urban Thermal Environments: A Data-Driven Analytical Framework

Rapid urbanization and climate change have severely exacerbated the urban heat island (UHI) effect in high-density subtropical megacities. Traditional linear models often fail to capture the complex, non-linear thermal responses driven by three-dimensional (3D) urban morphology and socio-ecological interactions. This study proposes a data-driven analytical framework explicitly tailored for macro/mesoscale climate-resilient urban planning to deconstruct the non-linear associations of Land Surface Temperature (LST) in Shenzhen, China. Integrating multi-source spatial data into a 500 m grid, we utilized the eXtreme Gradient Boosting (XGBoost) algorithm for high-precision LST modeling (R2 = 0.7851, MAE = 1.1381 °C) and applied the SHapley Additive exPlanations (SHAP) approach for spatial interpretability. The results reveal critical non-linear thresholds: vegetation (NDVI) cooling efficiency saturates at 0.8, while impervious surfaces (ISA) transition into dominant heating drivers beyond 0.7. Notably, a synergistic effect indicates that high building volume density (BVD) significantly amplifies the marginal cooling benefits of vegetation. Furthermore, local SHAP attribution combined with K-Means clustering facilitated the delineation of four distinct thermal management zones. This framework shifts UHI mitigation from broad, uniform policies to precise, data-driven spatial diagnostics, offering actionable “one zone, one policy” strategies for sustainable architectural and climate-resilient urban planning.