In the context of escalating drought threats due to global climate change, accurately identifying the spatial distribution of drought disaster-bearing bodies is crucial for disaster risk assessment and mitigation. However, fine classification using remote sensing technology still faces numerous challenges. Existing classification methods primarily rely on single spectral features, making it difficult to capture the spatiotemporal heterogeneity coupling effects of crops and grassland subtypes affected by drought. This results in frequent spectral confusion between wheat and corn and the homogenization of grassland subtypes, severely limiting classification accuracy improvements. This study focuses on the Yili River Basin, a typical watershed in northwest China's arid region, using 269 field sampling points from the third Xinjiang scientific expedition and 2023 Sentinel-2 temporal imagery. It proposes a multi-dimensional feature classification framework that integrates seasonal-phenological temporal features with spatial constraints, and develops a random forest fine classification model based on feature optimization. Experimental results show that the method achieves an overall accuracy of 91.9 %, a Kappa coefficient of 0.89, and a field verification accuracy of 83.6 %, representing an 18.9 % improvement over the 73 % accuracy of traditional spectral feature methods. Feature importance analysis reveals that seasonalphenological temporal features contribute 44.4 %, while spatial features contribute 21 %, confirming the effectiveness of multi-dimensional feature fusion. The temporal-spatial multi-feature classification method for drought disaster-bearing bodies developed in this study effectively categorizes these bodies, providing a scientific basis for drought disaster monitoring and management.