Integrated source apportionment, ecological risk assessment, and machine learning-based human health risk evaluation...

Stream sediments, as long-term sinks for potentially toxic elements (PTEs), provide valuable insights into both natural and anthropogenic contamination. This study presents a comprehensive assessment of PTE contamination, ecological risk, and human health implications across Odisha, eastern India- a region characterized by complex Precambrian geology, intensive agriculture, mining, and industrialization. Concentrations of ten PTEs (Cr, Cd, As, Ni, Cu, Zn, Co, Mn, V, and W) from 28,111 locations collected under the Geological Survey of India NGCM program indicate moderate to very high contamination across the state, with pronounced hotspots in mineralized and industrial belts. Multivariate analyses, including principal component analysis and hierarchical clustering, reveal dominant lithogenic control over Cr, Mn, Ni, Co, Cu, Zn, and V, while anthropogenic enrichment of As, Cd, and W is linked to mining, industrial emissions, and agricultural activities. Non-negative matrix factorization corroborates these source apportionment results. Pollution indices, including enrichment factor, geo-accumulation index, pollution load index, and potential ecological risk index, indicate moderate to high ecological risk in several regions. Human health risk assessment shows that 69.05% of locations exhibit high non-carcinogenic risk (HI > 1) for children, whereas adults show non-carcinogenic and carcinogenic risks in < 0.5% and > 65% of locations, respectively, primarily associated with Cr, Cd, and As. Sobol sensitivity analysis demonstrates that concentration variability predominantly governs carcinogenic risk estimates. Additionally, a machine learning-based framework is developed to classify risk and non-risk zones for both adults and children. This integrated approach provides critical insights for public health protection, targeted remediation, and sustainable land-use planning.