Drivers and distribution of soil arsenic in China's yellow river irrigation area by machine learning.

Arsenic contamination in irrigated agricultural soils poses global health and ecosystem risks. Using China's Yellow River Irrigation District as a case study, we developed an interpretable machine learning framework to predict soil arsenic distribution and identify its driving mechanisms. Among five models (XGBoost, RF, SVM, MLP, and MLR), XGBoost achieved the highest accuracy (R2 = 0.83, RMSE = 0.55). SHAP analysis revealed that cation exchange capacity, population density, and soil pH are the dominant factors controlling arsenic accumulation. Spatial autocorrelation further identified arsenic enrichment hotspots in the central and northern study regions. By integrating XGBoost with ordinary kriging, we produced a high-resolution (1 km × 1 km) arsenic map that overcomes the "bull's-eye effect" of traditional interpolation. This framework offers a transparent, predictive tool for targeted pollution management in data-limited irrigated regions.