Unraveling the seasonal pulses of dust aerosol optical depth in arid Asia: A 45-year machine learning attribution analysis (1980–2025)

Dust aerosols across Central and East Asia influence regional climate, air quality, and biogeochemical cycles, yet the seasonal drivers of atmospheric dust loading remain poorly understood. We investigated long-term trends and controls of Dust Aerosol Optical Depth during 1980–2025 using Modern-Era Retrospective analysis for Research and Applications, Version 2 reanalysis and an interpretable machine-learning framework. Trend analysis revealed widespread intensification, with spring (March–May) exhibiting positive trends across 92.2% of the study area, whereas summer (June–August) showed a pronounced east–west contrast, including a 34.3% decline over eastern China. Seasonal Extreme Gradient Boosting models achieved high predictive performance, with an all-month model R 2 of 0.807. SHapley Additive exPlanations identified antecedent soil moisture, particularly 1–3-month lagged topsoil wetness, as the dominant predictor, contributing more than 40% of total SHAP importance across all models. Temperature exhibited distinct nonlinear effects, with winter Dust AOD peaking at intermediate temperatures (270–275 K) and summer Dust AOD increasing sharply only above 305 K. Elevation further modulated dust loading by favoring lower-to mid-elevation basins and plateaus, while El Niño–Southern Oscillation (Niño3.4) exerted the strongest large-scale climate influence during spring. These findings highlight the importance of hydroclimatic memory and thermal extremes for understanding seasonal atmospheric dust loading and improving seasonal dust forecasting under climate change.