Vegetation restoration has been extensively employed to mitigate soil degradation caused by erosion in China's Loess Plateau. However, excessive vegetation restoration may exacerbate soil drought, thereby threatening ecosystem stability and sustainability. Identifying vegetation management strategies that optimize soil moisture retention is thus crucial for the sustainable and healthy development of the ecosystem in the Loess Plateau. In this study, we measured soil moisture in grassland (Stipa bungeana, etc.), shrubland (Caragana korshinskii), and woodland (Populus simonii) on steep slopes (with gradient larger than 30 degrees) between 2021 and 2022, and used the Hydrus-1D model to simulate 0-500 cm soil moisture dynamics as affected by vegetation types. We set up different scenarios of vegetation replacement and cover reduction to assess the long-term effects of vegetation management on soil moisture over a 50-year period. Our results revealed significant variations in soil moisture dynamics among vegetation types, with the highest soil moisture observed in woodland and the lowest in shrubland. Notably, a dried soil layer extending beyond 500 cm was observed across all vegetation types. Results from Hydrus-1D modeling indicated that although woodland currently retains more soil moisture than grassland, this pattern will reverse after 21 years. Decreased vegetation cover, irrespective of vegetation type, would lead to soil moisture accumulation. Converting shrubland to grassland would increase soil moisture in the 0-500 cm depth, whereas the impact of converting woodland to grassland depends on vegetation coverage. These findings highlights the importance of vegetation coverage in regulating the effects of management practices on deep soil moisture dynamics in the loess hilly regions.