Revealing Dust Source Changes in East Asia From Past to Future: An Interpretable Deep Learning Approach
Aeolian dust source areas are controlled by climate and human activities and are closely associated with land degradation, yet their long‐term evolution under global warming remains insufficiently quantified. An integrated framework that couples a U‐Net segmentation model with Shapley Additive Explanations (SHAP) was built to map and attribute dust source areas across arid and semi‐arid East Asia. Using the framework, spring dust source area changes are quantified annually for 2000–2024 and projected for 2100 under the three IPCC Shared Socioeconomic Pathways (SSP) scenarios. The relative influence of key meteorological controls is also quantified. From 2000 to 2024, the spring dust source area shrinks from 3.7 × 10 6 to 2.8 × 10 6 km 2 , representing a net reduction of 25.5%, primarily due to rapid contraction along ecological margins and persistence within hyper‐arid cores. The conversion from dust source to grassland is up to 1.3 × 10 6 km 2 , benefiting greatly from ecological restoration and regional hydroclimatic amelioration. SHAP analysis shows a shift in the dominant climatic constraint from precipitation to evaporation. By the year 2100, dust source areas will expand in area relative to the historical period. The greatest expansion occurs in central eastern Mongolia and northern Xinjiang, whereas the Tibetan Plateau shows a marked reduction. The total dust source area in 2100 declines with warming intensification and decreases by 9.1% from SSP1‐2.6 to SSP5‐8.5 future climate scenarios. Under future global warming, precipitation supply emerges as the principal control on moisture availability in arid and semi‐arid regions and therefore constrains dust source expansion.