Quantifying the differential climate responses of compound soil erosion by an integrated conceptual-machine learning framework

Global warming is intensifying soil degradation in vulnerable ecosystems worldwide, and dry-hot valleys have become critical erosion hotspots under changing climatic conditions. However, systematic assessment of erosion types and their controlling factors in these regions remains limited by sparse observations and the complex dynamics of compound erosion. Here, we present an integrated analytical framework that combines conceptual erosion models, CMIP6 multi-model projections, and explainable machine learning. This framework enables the simultaneous quantification and feature-level interpretation of water, wind, and freeze-thaw erosion responses to climate forcing in dry-hot valleys. The results reveal a marked expansion of high-risk erosion zones under future scenarios, with an annual growth rate of 0.011% under SSP5-8.5, ultimately affecting 20.23% of the basin. Among the three erosion types, water erosion shows the strongest acceleration, increasing at 2.19 × 10 −2 t ha −1 yr −1 and substantially outpacing the other processes. XGBoost–SHAP analysis further indicates that modeled water erosion is most strongly associated with topographic conditions, while precipitation and vegetation-related factors also contribute substantially to its spatial differentiation. Overall, these findings show that different erosion types exhibit uneven responses to future climate forcing in dry-hot valleys. The proposed framework provides a useful basis for compound erosion assessment and may offer methodological reference for targeted conservation and vulnerability assessment in ecologically fragile basins.