Spatiotemporal evolution and nonlinear drivers of land surface temperature in the Guangdong–Hong Kong–Macao Greater Bay Area from 2000 to 2020

Against the dual pressures of global warming and rapid urbanization, understanding the evolution and driving mechanisms of urban thermal environments is essential for sustainable urban development. Using multi-source remote sensing, land-use, landscape pattern, and socioeconomic data, this study investigated the spatiotemporal evolution of land surface temperature (LST) in the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) from 2000 to 2020. A multi-validation framework integrating random, spatial, and temporal validation was developed to evaluate model generalization, while an explainable machine-learning framework combining Extreme Gradient Boosting (XGBoost) and SHapley Additive exPlanations (SHAP) was employed to quantify key drivers and their nonlinear responses. Results showed that summer LST increased continuously, with high-temperature zones expanding and concentrating in the Pearl River Estuary urban core. XGBoost consistently outperformed conventional statistical and machine-learning models. Nighttime light intensity, NDBI, DEM, and NDWI were identified as the dominant drivers, while the influence of urbanization-related factors increased over time. Most drivers exhibited pronounced nonlinear responses, with urbanization-related factors showing diminishing marginal warming effects and vegetation- and water-related factors exhibiting threshold-dependent cooling effects. SHAP-based clustering further identified four thermal-environment governance zones with distinct driving characteristics, providing a spatially differentiated basis for urban development regulation, blue–green infrastructure planning, and targeted heat mitigation in rapidly urbanizing regions.