Answers are grounded in this paper and cite their sources.
Quantifying the spatiotemporal dynamics of soil erosion and clarifying its associated driving factors are essential prerequisites for developing targeted soil and water conservation strategies. This study focused on the Kuye River Basin, a primary coarse sediment source in the middle Yellow River. The RUSLE model was used to estimate the soil erosion modulus from 2000 to 2024, and the spatiotemporal evolution and geological differentiation of erosion were analyzed. An interpretable XGBoost-SHAP machine learning framework was constructed, incorporating climatic, topographic, vegetation, anthropogenic, and geological substrate factors, to quantitatively disentangle the contributions of each factor to the RUSLE-estimated erosion spatial pattern. Seven groups of controlled experiments were designed to examine the structural overlap between RUSLE internal factors and model independent variables. The results show that: (1) The overall soil erosion intensity exhibited an improving trend, but the spatial pattern remained stable with a pronounced lithological gradient. Aeolian loess, argillaceous rock, and clastic rock interbedded with argillaceous rock units were identified as persistent high erosion risk zones. (2) Geological substrate was the background factor most closely related to the spatial differentiation of erosion. The spatial distribution of single-factor sensitivity showed that high-sensitivity areas of the K factor coincided with argillaceous rock and aeolian sand, high-sensitivity areas of the LS factor were consistent with steep-slope zones in the loess hilly-gully region, and high-sensitivity areas of the C factor overlapped with low-vegetation-cover areas in the middle and lower reaches. The spatial superposition of these three factors is closely associated with the persistently high erosion risk in the high-risk units. The basin-wide erosion rebound in 2020 under extreme rainfall highlights the sensitivity of soft substrates to extreme climate. (3) Within the RUSLE framework, NDVI and slope, corresponding to the C and LS factors, contributed most to the spatial pattern of erosion modulus. However, controlled experiments indicated that this result partly arises from the mathematical structural relationship of the RUSLE equation, and the model performance primarily reflects its capacity to reconstruct RUSLE outputs under random holdout validation rather than cross-regional predictive transferability. External environmental variables (PET, Distance to roads, Distance to water, Aspect, Soil thickness, and Lithology) showed limited independent explanatory power for RUSLE outputs (M5 R 2 = – 0.011). Therefore, the SHAP results should be interpreted as the “covariate importance of the RUSLE-estimated erosion pattern” rather than the “causal mechanisms of actual erosion processes.” It is recommended that geological unit differentiation be incorporated as a core consideration in future soil and water conservation planning, and that a precision governance system based on zoning of rock-soil erosion resistance be established. Future studies should integrate independent observational data (e.g., check dam sedimentation, 137 Cs tracing, and sediment discharge at hydrological stations) to distinguish RUSLE structural associations from actual erosion drivers.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: