Integrating self-organizing maps and explainable artificial intelligence to unravel surface water–groundwater hydrochemical evolution and drinking-water suitability controls in a data-limited arid basin

Study region The Dahei River Basin, located in the semi-arid northern region of China’s Great Bend of the Yellow River. Study focus This study integrates hydrochemical characterization, stable hydrogen–oxygen isotopes, Self-Organizing Maps (SOM), the entropy-weighted water quality index (EWQI), and explainable ensemble learning (XGBoost–TreeSHAP) to investigate surface water–groundwater evolutionary patterns and identify key hydrochemical contributors to drinking-water suitability in the Dahei River Basin. New hydrological insights for the region Waters in the basin are predominantly weakly alkaline and are characterized by HCO₃⁻–Ca·Mg and HCO₃⁻–Na facies, indicating the combined effects of carbonate weathering and cation exchange, with localized contributions from evaporite dissolution and external inputs. SOM reveals a continuous hydrochemical evolution, ranging from low-salinity recharge waters to mineralized discharge end-members. Stable isotopes indicate seasonally variable surface water–groundwater connectivity, with stronger coupling during the wet season and greater isotopic divergence during the dry season. EWQI results suggest favorable groundwater quality, with approximately 65% of samples classified as excellent to good, showing a distinct spatial gradient from higher quality in the northern mountainous areas to lower quality in the southern plains. XGBoost–TreeSHAP indicates that Cl⁻ and TDS make major contributions to EWQI predictions, whereas NO₃⁻ shows a contribution pattern consistent with localized agricultural influence. Overall, this study provides a regime-resolved and interpretable framework for assessing groundwater suitability in data-limited arid and semi-arid regions.