Abstract: Land use change exerts a profound influence on watershed hydrological processes, especially runoff, evapotranspiration, and infiltration. Clarifying these mechanisms is essential for sustainable water management and hazard mitigation.We selected the Fen Shui River Basin as the case study. The SWAT model （Soil and Water Assessment Tool） was calibrated and validated to simulate hydrological processes with high reliability. To enhance predictive performance, fifteen machine learning models including Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) were employed. SHapley Additive exPlanations (SHAP) analysis was then applied to interpret the relative contributions of meteorological, hydrological, and land use factors.Precipitation strongly drove runoff, while temperature dominated evapotranspiration. Built-up land proportion exerted negative impacts on both runoff and evapotranspiration, whereas forest land reduced runoff through interception, transpiration, and soil water storage. Cropland proportion produced complex and nonlinear effects on infiltration and runoff coefficients.Integrating SWAT with machine learning not only improved prediction accuracy but also strengthened the interpretability of hydrological responses. This framework provides a powerful tool for watershed management and land use planning. Moreover, it offers practical guidance for early warning and risk assessment of mountain hazards such as debris flows, highlighting its broader significance in disaster prevention and resource sustainability.