Mechanistic Understanding of Soil Arsenic Enrichment and Selenium Deficiency in the Qinghai-Tibet Plateau via Integrated Geographically Weighted Regression and Machine Learning.

The Qinghai-Tibet Plateau (QTP) is a representative alpine arid region characterized by arsenic (As) enrichment and selenium (Se) deficiency in soils. This poses a severe threat to public health, yet the mechanisms driving this antagonistic distribution remain poorly understood. Here, we developed a Geographically Weighted Regression and Light Gradient Boosting Machine (GWR-LightGBM) model, combined with SHAP interpretation and health risk assessment, to systematically investigate the spatial patterns, drivers, and health implications of soil As and Se. The GWR-LightGBM model significantly outperformed machine learning and geostatistical models, improving the R2 for As prediction to 0.60 and confirming spatial non-stationarity as a critical bottleneck. The results reveal that over 99.95% of the study area exhibits As enrichment coupled with Se depletion. This pattern is primarily attributable to lithological inheritance, while aridity, warming, and topography further drive the divergent responses of As and Se by promoting As desorption and enrichment but accelerating Se leaching and volatilization. The health risk assessment shows that As poses no immediate high risk, yet widespread Se deficiency (26.04% of samples < 0.125 mg/kg) is a greater concern. These findings support region-specific management strategies: Se biofortification in deficient areas and As source control in high-risk zones.