Mapping Soil Heavy Metal Using an Interpretable Framework With Multi‐Sources Data

ABSTRACT Predicting the spatial heterogeneity of soil heavy metals (HMs) and identifying the determinants and sources is fundamental to developing effective strategies to mitigate and prevent environmental contamination. This study established a comprehensive research framework of “spatial prediction ‐ driver factor analysis ‐ source contribution quantification”. By comparing and using different methods to draw the spatial distribution map of HMs and combining the SHapley Additive exPlanations (SHAP) and the Positive Matrix Factorization (PMF), a fusion analysis of HMs driver factors and pollution sources was conducted. The following results were obtained: (1) Random Forest (RF) outperformed Ordinary Kriging (OK) and Geographically Weighted Regression (GWR) in predicting Cd, As, Pb, and Zn (with R 2 values above 0.4 for all), while OK exhibited the highest accuracy in predicting Hg, Cr, Cu, and Ni. (2) Under the RF‐SHAP framework, socio‐economic factors primarily influenced the spatial heterogeneity of HMs (Cd, Pb, Cu, Zn), environmental factors mainly affected As and Cr, and topographic, climatic, and soil properties had relatively minor impacts on the spatial heterogeneity of HMs. (3) The results of the PMF source apportionment analysis indicated that HMs (Cr and Ni) primarily originated from natural origin, whereas Cu, Pb, Zn, Cd, As, and Hg were largely attributed to anthropogenic. The research results provide a basis for evidence‐based policies on soil standards, land use planning and pollution accountability mechanisms.