Soil erosion threatens global land and water resources, underscoring the need for effective vegetation restoration strategies. However, the quantitative effects and mechanisms through which vegetation structure, spatial patterning, and coverage regulate runoff and soil loss dynamics remain inadequately quantified. This study combined controlled rainfall experiments with structural equation modeling (SEM) to assess how vegetation structure (forest, shrub, grassland), spatial position (upslope, downslope), and coverage (60%-80%) influence erosion processes. Results demonstrated that shrub communities most effectively mitigated runoff and soil loss, outperforming both forests and grasslands. Downslope vegetation placement consistently surpassed upslope arrangements in reducing erosion, and coverage exceeding 70% emerged as a critical threshold for optimal effectiveness. SEM analysis revealed that soil properties were the primary mediators explaining vegetation's erosion-reduction effects. These findings challenge afforestation-centered restoration paradigms in Loess Plateau, highlighting the superiority of strategically designed shrub-based systems. By integrating artificial rainfall experiments with causal modeling, this work advances precision restoration strategies for global erosion-prone ecosystems, addressing escalating climate-driven hydrological extremes and land degradation.