Characterization, sources, and driving factors of particulate matter (PM) pollution in a desert hinterland city: Insights from Hotan, China.

Atmospheric particulate matter (PM) pollution is severe in southern Xinjiang, especially near the Taklimakan Desert, where air quality is affected by both dust and anthropogenic emissions. This study employs explainable machine learning in Hotan to identify distinct mechanisms underlying haze and dust events. Results confirm severe PM pollution in Hotan. Annual mean PM10 and PM2.5 reached 327.39 and 102.02 µg/m³, respectively. Dust events originated mainly from the central desert, while haze was dominated by local and nearby sources. Emission contributions averaged 405.84 µg/m³ for PM10 and 114.56 µg/m³ for PM2.5 annually. Meteorological stagnation increased PM10 and PM2.5 by 138.37 and 44.51 µg/m³, respectively, and transport added another 138.04 and 43.20 µg/m³. The mechanisms showed significant diurnal and seasonal differences. Diurnal analysis revealed that haze pollution exhibits a double-peak pattern linked to stable stratification under high surface pressure (SP), whereas dust pollution shows irregular peaks driven by afternoon wind resuspension. Seasonal analysis revealed that haze-season PM was driven by high NO₂ and low solar radiation (SSR), reflecting precursor accumulation and stagnant weather, with transport contributions from regional pollution convergence via air mass history. Dust-season PM was triggered by high SP, strong 10-m u-component of wind (U10), and strong SSR, with near-source transport dominated by dust sources. Overall, PM pollution manifests as "static accumulation" during haze and "dynamic excitation" during dust events. This study validates the "de-weathering-SHAP" model for PM attribution in arid regions, providing a scientific basis for targeted prevention strategies in "Belt and Road" dust-affected zones.