Spatial Heterogeneity and Drivers of Heavy Metals in Soils and Sediments of the Nyangqu River Basin, Tibetan Plateau, China: Insights from GeoDetector and Explainable Machine Learning
Jiale Chen1,2, Geng Xu2, Duo Bu1,2
1Key Laboratory of Biodiversity and Environment on the Qinghai-Tibet Plateau, Ministry of Education, School of Ecology and Environment, Xizang University, Lhasa 850000, China.
Abstract:
Heavy-metal contamination in alpine agricultural watersheds reflects interacting geological, environmental, and anthropogenic controls. In this study, arsenic (As), copper (Cu), lead (Pb), zinc (Zn), and chromium (Cr) were investigated in 213 farmland soil and sediment samples collected from the Nyangqu River Basin during 2019-2021 and 2024-2025. Pollution status and ecological risks were evaluated using the Nemerow Integrated Pollution Index (Pn) and Håkanson Potential Ecological Risk Index (RI), while potential factors associated with spatial variation were explored using GeoDetector and an explainable machine-learning framework integrating XGBoost, SHAP, and LIME. Mean As, Cu, Zn, and Cr concentrations exceeded Tibetan soil background values, whereas mean Pb remained below background. Farmland soils exhibited higher concentrations of As, Pb, and Zn than sediments, whereas Cu displayed comparable levels between the two media. Among the investigated metals, As showed persistent enrichment, whereas Cr exhibited the greatest spatial variability and strongest local anomalies. Overall, slight pollution dominated the study area (69.0%; median Pn = 1.73), although several hotspots increased the mean Pn to 2.10, indicating moderate pollution at the regional scale. After applying coefficient-adjusted thresholds (23/133), the ecological risk index (RI) ranged from 16.45 to 54.24 (mean = 32.70), with 5.6%, 93.9%, and 0.5% of samples categorized as low, moderate, and considerable risk, respectively, and no samples exhibiting high risk. The associated environmental factors showed element-specific patterns, involving soil physicochemical conditions, geological background, and localized anthropogenic indicators. Notably, factor combinations generally showed greater explanatory power than individual covariates, suggesting stronger joint statistical associations with heavy-metal spatial differentiation.
