Machine learning-based identification of multi-source recharges and enrichment mechanism of karst groundwater

The Changxiao karst water system, southwestern area of Jinan, China This study focused on the Changxiao karst water system, employing an integrated approach combining hydrochemical methods, PCA-APCS-MLR receptor modeling, and machine learning to identify groundwater sources and spatial distribution patterns. The enrichment mechanisms were investigated through systematic analysis of geomorphological features, geological structures, lithostratigraphic characteristics, and hydrodynamic conditions. The hydrochemical types of groundwater in the Changxiao karst system are primarily derived from rock weathering and leaching processes. SOM-K-means clustering analysis revealed a triple mixing characteristic in the Collecting and discharging zone (CDZ), integrating contributions from the Southern replenishment area (SRA), Western lateral runoff area (WLR), and Yellow River (YR) infiltration, confirming their close hydraulic connectivity. The CDZ exhibits well-developed fracture-karst systems, receiving combined recharge from the SRA, WLR, and YR. This area possesses well-developed conduits, substantial fracture-karst storage, and convergent recharge pathways—establishing fundamental prerequisites for groundwater accumulation in carbonate aquifers. By integrating artificial intelligence, hydrochemical analysis, and statistical models, this study constructs a multiscale analytical framework for karst groundwater research, offering novel insights for optimizing water source protection and urban water supply strategies. • An integrated approach has been used to identify groundwater sources and spatial distribution patterns. • Changxiao karst water system is mainly driven by natural dissolution, supplemented by external and agricultural sources. • The enrichment mechanism of karst groundwater in Changxiao has been clarified.