Integrating gradient boost for analyzing land surface temperature and its influencing factors

Urbanisation increases land surface temperature (LST) in subtropical coastal cities, yet long-term, interpretable analyses of LST drivers are limited for the Fuzhou metropolitan area and its suburbs. We address this gap by combining a 25-year Landsat time series with n Explainable Gradient-Boosting (XGBoost) model to evaluate the land cover impacts. Multi-epoch Landsat data, XGBoost and SHAP were integrarted to assess the relative, nonlinear, and moderating effects of impervious surfaces, vegetation, and elevation on LST. Dense vegetation remained dominant (45–51%), while built-up area increased from 5% to 12%; by 1999–2023, built-up surfaces became the most prevalent class, replacing bare land. Enhanced scatter-regression attribution reveals imperviousness (NDBI, BSI) raises LST, while vegetation and moisture (NDVI, NDWI, NDMI) and higher elevation, reduce it, elevation strongly moderates LST in this humid coastal region. Multivariate regression and variance partitioning indicate that elevation is the primary and most often uniquely attributable LST driver. The XGBoost model accurately predicts LST with an RMSE of approximately 1.3. Policymakers should promote green infill, protect moisture-rich corridors, and limit impervious surfaces in heat-prone areas. This approach provides a transferable framework for climate-smart urban planning in Fuzhou and similar coastal cities.