Spatial patterns, source apportionment, and environmental correlates of soil potentially toxic elements in an arid high-altitude mining area of Wuqia County, Xinjiang, China.

Disentangling mining-related inputs from heterogeneous geological backgrounds remains difficult in arid mining regions. We analyzed eight potentially toxic elements (PTEs; As, Cd, Cr, Cu, Hg, Ni, Pb, and Zn) in 155 surface-soil samples from an arid, high-altitude mining area in Wuqia County, Xinjiang, China. A staged framework combined a self-organizing map (SOM), improved absolute principal component scores-multiple linear regression (APCS-MLR), Light Gradient Boosting Machine (LightGBM), and Shapley Additive Explanations (SHAP) to identify multielement patterns, estimate potential-source contributions, and characterize nonlinear environmental associations. SOM resolved localized Cd-Pb-Zn co-enrichment, with class C3 concentrated near the waste-rock dump and processing plant. APCS-MLR separated five potential-source signals: mining-traffic Cd-Zn, mixed Hg-As deposition/agricultural, Ni parent-material, mixed Cr-Cu parent-material/industrial, and Pb traffic-deposition. In the 105 robust samples used for the final Huber-regression branch, F1 and F4 accounted for mean contributions of 29.7% and 27.5%, respectively. Spatial out-of-fold (OOF) evaluation of downstream LightGBM models identified F1 as the most stable factor (pooled OOF R2 = 0.36; positive R2 in all five folds), and its estimated contribution decreased with distance from the mining area. A full-sample, uncertainty-weighted, five-factor positive matrix factorization (PMF) sensitivity analysis recovered one-to-one correspondence with the APCS-MLR structure (cosine similarities, 0.724-0.977); all five profiles matched at cosine similarity ≥ 0.80 in 93% of bootstrap resamples. SHAP associations for F2-F5 remained exploratory because their spatial stability was weak. By separating pattern identification, statistical source apportionment, and association interpretation, the staged evidence chain supports targeted monitoring and improves the robustness of source-oriented environmental assessment in arid mining regions.