Integrating unsupervised machine learning, statistical analysis, and Monte Carlo simulation to assess toxic metal contamination and salinization in non-rechargeable aquifers

This study presents the first comprehensive evaluation of groundwater quality in Siwa Oasis, Egypt, integrating advanced machine learning and statistical approaches to assess contamination, health risks, and industrial suitability. Thirty samples from the Nubian Sandstone Aquifer (NSAS) and karst springs were analyzed using Self-Organizing Maps (SOM), Principal Component Analysis (PCA), and Canadian Water Quality Index (CCME WQI). SOM clustering revealed three distinct water types: (1) hypersaline springs (TDS >10,000 mg/L) near Siwa Lake, (2) moderately saline springs (4,551–8,885 mg/L), and (3) freshwater NSAS samples (<1,000 mg/L). PCA identified salinity (45.5% variance), carbonate equilibrium (21.3%), and anthropogenic inputs (11.5%) as dominant controls. The CCME WQI classified 28% of samples as ''Poor/Marginal,'' with localized heavy metal (Ba, V) contamination confirmed by MPI and NCI indices. Monte Carlo-based health risk assessment revealed severe non-carcinogenic risks for children (HI >1), primarily from Co (HQ up to 105.5) and V (HQ up to 416.9) via ingestion. Industrial indices (LSI, RSI, CSMR) highlighted scaling potential in freshwater zones (LSI >1.5) and corrosion risks in saline areas (RSI >8). As the first study to: (1) quantify emerging contaminants (V, Co, Mo) in NSAS, (2) apply SOM-PCA-Monte Carlo integration in arid aquifers, and (3) concurrently evaluate health and industrial risks, this work provides a replicable framework for non-renewable aquifer management. Immediate actions targeted remediation, infrastructure protection, and agricultural regulation are recommended in Siwa Oasis. The methodologies and gaps identified including unassessed carcinogenic metals and isotopic tracing set a roadmap for future research in vulnerable aquifer systems.