Quantitative decoupling of source-pathway-receptor driving mechanisms for integrated soil risk via interpretable...

Most existing concentration-based risk assessments of potentially toxic elements (PTEs) in farmland soils tend to overlook source-specific transport pathways and receptor-related risks. In this study, an integrated risk index-machine learning framework based on the source-pathway-receptor concept is developed to comprehensively evaluate PTE risks in a typical mining city. The framework integrates the improved Nemerow index (INI), potential ecological risk index, Monte Carlo simulation-based health risk assessment, and machine learning models to identify and differentiate industrial and agricultural risk mechanisms. Results showed that industrial sources exhibited higher INI values than agricultural sources, indicating stronger pollution accumulation. Industrial risks were predominantly associated with atmospheric deposition and surface runoff pathways. Receptor-based health risk assessments further revealed that children were highly vulnerable to exposure to industrial PTEs, with chromium identified as the dominant contributor to carcinogenic risk. By contrast, agricultural risks were mainly attributed to internal receptors (26.80%) and inputs from irrigation, fertilizers and animal manure (23.20%). Among the agricultural risk zones, dietary exposure through crop consumption represented a crucial pathway, with oilseed rape demonstrating the highest potential health risks. Carcinogenic and non-carcinogenic risks associated with oilseed rape exceed acceptable thresholds. Furthermore, the XGBoost regression model achieved satisfactory predictive performance under an internal 8:2 training-testing validation, with R² values of 0.823 and 0.716, respectively. SHAP analysis further revealed different dominant factors governing industrial and agricultural risks. Overall, the proposed framework provides a basis for targeted zoned control and may help avoid excessive remediation in low-risk areas while ensuring sufficient control in high-risk areas.