Unveiling the drivers of groundwater quality in an industrial plain: An integrated hydrogeochemical and stacking ensemble learning approach.

Understanding the hydrochemical evolution and primary chemical components governing groundwater quality in industrial regions is crucial for sustainable groundwater management. This study established an integrated analytical framework by combining conventional hydrogeochemical analyses with a Stacking ensemble machine learning model. The framework was applied to a dataset comprising 26 hydrochemical parameters measured from 100 groundwater samples collected in the heavily industrialized Jiyuan Plain, China. Results show that the shallow groundwater is featured by types of HCO₃-Ca·Mg and SO₄·Cl-Ca·Mg, with major ions originating from mineral dissolution of halite, carbonate, gypsum, and silicate. Moreover, the results of ion ratios, PCA (i.e., principal components analysis), EWQI (i.e., Entropy weighted quality index) reveal that hydrochemistry of the shallow groundwater is shaped by the interaction between natural hydrogeochemical processes (e.g., mineral weathering) and intense anthropogenic activities (industrial and agricultural discharges). Most importantly, the combining application of EWQI and Stacking model quantitatively identified manganese and nitrogen species as the foremost drivers of groundwater quality degradation, a key finding that directly challenges conventional assumptions based on the region's prominent lead industry, which was found to contribute insignificantly to overall groundwater pollution. Notably, the hybrid traditional and data-driven framework successfully uncovered key pollutants in composite pollution areas, offering a transferable methodology for quantitative source apportionment in other industrial regions worldwide. For water management, the findings of this study suggested a targeted strategy for groundwater quality controlling. Prioritization must be given to investigating the origins of manganese (geogenic vs. industrial) and effectively intercepting the inputs of nitrogen/organic matter pathways.