Vegetation cover significantly mitigates slope water erosion caused by erosive rainfall. Through simulated rainfall experiments (slope = 15 degrees and five rainfall intensities I = 60-120 mm h(-1)), we investigated runoff erosion patterns under 30 grass-shrub cover (C-gs) and litter volume (C-l) combinations. Results indicated that runoff and erosion rates decreased with higher C-gs and C-l and increased with I. Compared to bare slopes, grass-shrub, litter, and grass-shrub + litter treatments showed average reductions in runoff rates by 30.56%, 13.11%, and 39.72%, respectively, and in erosion rates by 47.78%, 36.67%, and 66.71%, respectively. Grass-shrub vegetation in the grass-shrub + litter was more obvious than litter in inhibiting runoff erosion. An excellent power function relationship was observed between the flow intensity parameter and soil erosion rate, with the dimensionless effective stream power showing superior predictive capability. Factors (I, C-gs, and C-l) improved the Nash-Sutcliffe efficiency (NSE) of the equation between soil erosion rate and flow intensity parameters. An explicit model of vegetation erosion power was developed based on the integrated flow intensity parameter (NSE = 0.92). Machine learning models for erosion rate were constructed using Random Forest, Extreme Gradient Boosting, Support Vector Regression, and Back Propagation neural network (BP), with the BP model exhibiting superior predictive performance (NSE = 0.97). SHapley Additive exPlanations analysis was applied to enhance model interpretability. This study offers a scientific foundation for slope water erosion management and promotes the thorough development of physical and machine learning models for soil erosion.