Interpretable groundwater spring potential mapping in complex terrain of Liulin County, China, using SSA-XGBoost with GIS-based validation

Liulin County, located on the eastern bank of the Yellow River (YR) in Shanxi Province, China, is characterized by complex terrain including mountains and loess plateaus under a semi-arid climate. To mitigate seasonal water supply instability and identify reliable groundwater sources, this study proposes a novel machine-learning framework integrating the Sparrow Search Algorithm (SSA) with Extreme Gradient Boosting (XGBoost). Utilizing Geographic Information System (GIS) and field surveys, a comprehensive hydrogeological dataset was constructed. The Boruta Algorithm (BA) was employed to eliminate redundant variables, while Random Forest (RF) evaluated feature importance. The proposed model was rigorously benchmarked against five alternative methods, including hybrids optimized by the Grey Wolf Optimizer (GWO). Furthermore, SHapley Additive exPlanations (SHAP) were applied to decipher the "black-box" nature of the models, quantifying feature contributions and non-linear interactions. The results demonstrate that the SSA-XGBoost model achieved superior predictive accuracy, yielding a maximum Area Under the Curve (AUC) of 0.8812. Consensus from RF and SHAP analyses identified lithology and altitude as the dominant controlling factors, while the Normalized Difference Vegetation Index (NDVI) and rainfall provided essential spatial variability. GIS-based zonation revealed that approximately 22.06 % of the study area possesses high groundwater potential. This framework effectively balances high predictive accuracy with transparency, providing a scientifically robust tool for sustainable groundwater management in complex terrain regions. • SHAP was added to SSA-XGBoost to boost model interpretability. • Supports balancing agricultural productivity with sustainable groundwater use. • The SSA-XGBoost model achieved the highest accuracy with an AUC of 0.8812. • Lithology, altitude, and NDVI identified as key factors affecting spring potential. • SSA and GWO were combined with XGBoost to optimize hyperparameters.