Coupling positive matrix factorization with machine learning to quantify contamination sources and map health risks of heavy metals in karst soils

Heavy metal contamination in karst ecosystems poses great environmental risks due to the interplay between high geochemical backgrounds and anthropogenic activities. This study performed an integrated source-to-risk assessment of soil metals/metalloids based on 674 topsoil samples from China’s Guangxi karst region, combining three models including positive matrix factorization (PMF), random forest (RF), and multiscale geographically weighted regression (MGWR). Geochemical characterization identified SiO 2 (63%), Al 2 O 3 (18%), and Fe 2 O 3 (6.4%) as dominant oxides in tested soils, with higher organic matter (3.3% content) than regional background value. A majority of 36 examined metals/metalloids displayed concentrations exceeding local background values, of which Cd and Hg exhibited the highest enrichment levels (3.83 and 6.75 folds of background values, respectively). PMF revealed distinct source patterns across land-use types, with natural weathering dominating in forest land, and agricultural inputs prevailing in grasslands. Riparian land was primarily influenced by traffic-related and other anthropogenic activities, while construction land was dominated by industrial and mining/smelting sources. PMF-RF-MGWR modeling demonstrated pronounced spatial heterogeneity with contamination hotspots mainly concentrated in industrial- and mining-intensive areas of central-western Guangxi. Notably, we found that karst weathering increased soil metal/metalloid mobility, while secondary carbonate minerals formation contributed to immobilization of Cd and Pb. Health risk assessment identified As, Cr and Ni as priority contaminants, and their presence in industrial zones posed unacceptable carcinogenic risks. Our findings underscore the role of the interplay between geological processes and anthropogenic activities in shaping metal/metalloid distribution, providing valuable references for targeted soil management and contaminated land remediation in karst regions. • Integrated PMF-RF-MGWR framework was applied to metal/metalloid source-risk analysis • Source apportionment revealed distinct patterns across land-use types • Metal/metalloid pollution hotspots were mapped in central-western Guangxi karst areas • Karst weathering mobilized metals while secondary carbonates immobilized Cd and Pb • Industrial zones posed unacceptable carcinogenic risk from As, Cr, and Ni