Climate drivers of forest ecosystem services supply in the hilly mountainus regions of southern China based on SHAP-enhanced machine learning

• The CS and HQ of FES in the study area continued to decline during the study period. • The study employed interpretable machine learning methods to investigate the key climate factors influencing FES. • The annual precipitation and temperature seasonality are the key climatic factors affecting FES. Analyzing the spatiotemporal patterns of forest ecosystem services (FESs) and their climatic drivers in the hilly mountainous regions of southern China (CSHR) is crucial for advancing regional ecological conservation. In this study, we employed the InVEST model to quantify four key FES indicators from 2000 to 2020: carbon storage (CS), soil conservation (SC), habitat quality (HQ), water yield (WY), and a composite ecosystem service index (CESI). Furthermore, we integrated an interpretable machine learning model, Random Forest–Shapley Additive Explanations (SHAP), to identify principal climatic drivers and characterize their nonlinear impacts on FESs. The results indicate that during the study period, SC (+5.17 %) and WY (+13.7 %) within the study area exhibited sustained increases, whereas CS (−0.47 %) and HQ (−3.87 %) exhibited a declining trend. CESI displayed a distinct spatial gradient, remaining consistently higher in the southern region compared to the northern region, whereas CESI values gradually increased towards the east. Moreover, SHAP value analysis revealed that climate-driven factors exhibited multivariate nonlinear characteristics. Specifically, temperature seasonality (Bio4) enhanced CS, the mean temperature of the warmest season (Bio10) inhibited SC, and areas with high annual precipitation (Bio12) were associated with simultaneous increases in both HQ and WY. The coupling of multiple factors affected the regulation of FESs. Among these, the interaction between temperature seasonality (Bio4) and annual precipitation (Bio12) proved particularly significant. Within this framework, WY demonstrated the strongest spatial synergy stability, with its mean bivariate spatial autocorrelation (global Moran’s I) values for Bio4 and Bio12 reaching 0.455 ( p < 0.001). In this study, we combined the analysis of FES supply and its climate drivers with interpretable machine learning methods to provide scientific insights for the sustainable development and scientific management of the ecological environment in the CSHR.