Toward Precise Management of Groundwater by Combined Heavy Metal(loid)s Contamination at Industrial Sites: A Machine Learning Driven Source-to-Risk Zoning Framework.

Controlling and remediating heavy metal(loid) contamination in groundwater at complex industrial sites remains challenging due to mixed pollution sources and strong subsurface heterogeneity, which hinder precise source apportionment and targeted risk management. We develop a machine learning coupled framework that integrates Self-Organizing Maps (SOM), K-means clustering, positive matrix factorization (PMF), geological-hydrogeological modeling, and Monte Carlo simulation to enable source-pathway-receptor contamination characterization, source identification, transport simulation, source-specific health risk assessment, and pollution zoning. Applied to an aluminum-processing site in southwest China, the framework showed that 84.6% of groundwater samples were contaminated. The most elevated metals were Fe, Al, and Mn, while mean concentrations of Fe, Al, Mn, As, Ni, and Pb exceeded the Chinese Class III groundwater standards. The hybrid Spearman-SOM-PMF model resolved three primary sources: spray coating and aluminum finishing (57.6%), geogenic background mixed with wastewater leakage (37.5%) and copper smelting and surface oxidation (4.9%). Groundwater flow simulation and three-dimensional geological characterization have shown that the regional hydraulic gradient and preferential pathways are the main factors affecting contaminant dispersion. A probabilistic health risk assessment has pinpointed the spray coating and aluminum finishing as the key sources for control, responsible for 51.0% of carcinogenic risk and 44.7% of non-carcinogenic risk. Combining SOM-K-means clustering with source-specific risks, we delineated risk-based management zones and identified Cluster II as the priority control area. This integrated framework provides a transferable approach for precise remediation and sustainable redevelopment of complex industrial sites.