Spatiotemporal dynamics of land surface temperature in Yunnan, China: Interpreting multi-temporal drivers using eXtreme Gradient Boosting (XGBoost) and SHapley Additive exPlanations (SHAP)

Amid global warming, mountainous regions have emerged as critical zones of investigation owing to their heightened vulnerability to climate change, their ecological significance, and the intensified interactions between natural stress and human activities. Land surface temperature (LST) is a fundamental indicator for assessing climatic sensitivity in these landscapes. However, a comprehensive understanding of the spatiotemporal dynamics and driving mechanisms of LST across large mountainous regions remains limited. Therefore, data from the Terra Moderate Resolution Imaging Spectroradiometer Land Surface Temperature/Emissivity Daily (MOD11A1) Version 6.1 product during 2001–2020 in Yunnan Province (a complex mountainous region), China, were analyzed. Sen’s slope analysis and Mann-Kendall test were applied to detect LST trends and spatial heterogeneity at both annual and seasonal scales. Subsequently, an eXtreme Gradient Boosting (XGBoost) model coupled with SHapley Additive exPlanations (SHAP) was employed to clarify the nonlinear contributions of multiple drivers. The study revealed the following findings. LST exhibited an overall warming rate of 0.020°C/a, characterized by daytime cooling (–0.008°C/a) and nighttime warming (0.048°C/a). LST increased during spring, summer, and autumn (0.011°C/a–0.018°C/a), whereas winter LST exhibited a cooling trend (–0.011°C/a). These variations were spatially partitioned by the Ailao Mountains, with the southwest displaying stronger thermal changes than the northeast. Natural controls, including digital elevation model (DEM) and downward shortwave radiation (DSR), predominated in the northwest high mountain canyons area and south tropical rainforest area, whereas nature–human interactions were more pronounced in the central urban agglomeration warming area and southeast karst landform area. The dominant drivers consisted of DEM, DSR, Normalized Difference Moisture Index, particulate matter 2.5 (PM 2.5 ), and aerosol optical depth (AOD). The strong correlations between gross domestic product and population density (correlation coefficient ( r )=0.95), as well as between PM 2.5 and AOD ( r =0.84), highlighted the increasing influence of socioeconomic factors on surface warming. This study can advance the understanding of how mountain topography, moisture, and anthropogenic pressures jointly regulate surface thermal regimes and provide region-specific insights for climate adaptation and sustainable ecosystem management.