Identifying heavy metal sources in extreme-arid oasis soils of Southern Xinjiang: a combined analytical approach

The extreme arid environment of the southern Xinjiang oasis makes it ecologically fragile and highly sensitive to soil changes, including heavy metal pollution.Identifying pollution sources is therefore essential to prevent irreversible impacts. In this study, a comprehensive field investigation was conducted on 622 soil samples, analyzing eight heavy metals (Cr, Ni, Cu, Zn, As, Cd, Pb, and Hg). Receptor models (UNMIX and PMF) were applied for source apportionment, while the Gradient Boosting Decision Tree (GBDT) was employed as a machine learning technique to identify key driving factors. The results indicate that while overall heavy metal concentrations remain low, cadmium (Cd) and arsenic (As) are significantly influenced by human activities, with higher accumulation observed in the northern and western oases, such as Weiku and Kashgar.Five major pollution sources were identified: natural geological (34.75%), natural-agricultural mixed (14.72%), agricultural-industrial mixed (7.15%), natural-traffic mixed (34.16%), and natural-industrial mixed (9.22%). The GBDT model identified soil type as the dominant influencing factor across all sources, highlighting the primary role of natural conditions in shaping the spatial distribution of heavy metals. To our knowledge, this is the first study that integrates receptor models (UNMIX and PMF) with machine learning (GBDT) for heavy metal source apportionment in this fragile arid oasis ecosystem, enabling both source quantification and driver identification simultaneously.Overall, our results provide a robust and transferable framework for understanding heavy metal sources and guiding pollution management strategies in arid oasis regions.