Regional vegetation recovery can change the land use types and effectively control soil and water loss. However, the spatial variation of the relationship between land uses and soil and water loss remains unclear. In this study, we integrated 59 peer-reviewed articles and 1121 records of annual runoff and sediment yield of runoff plots at the hillslope on the Loess Plateau. And, we used geodetector to determine the spatial differentiation of the relationship between annual runoff, annual sediment yield, and land use by regarding the eight critical zones (CZs) of Loess Plateau as a basis of spatial strata, which was classified by comprehensive methods combining the theory of earth's critical zone with principal component analysis and cluster analysis. The results showed that abandoned land had the highest annual mean runoff and sediment yield at 35.99 mm and 4208.82 g/m², respectively. The abandoned land, bare land, and cropland had a higher ability to yield runoff and sediment than artificial grassland, forest, natural grassland, and shrubland. The annual mean sediment yields of shrubland and forest were significantly lower than that of artificial and natural grassland (P<0.05). Annual mean runoff and sediment yield of abandoned land in the hilly-gully agriculture-grassland-woodland transition CZ were significantly higher than that of the hilly-gully agriculture and grassland CZ in addition to abandoned land had the highest annual mean sediment yield in the mountainous forest CZ (P<0.05). Forests in the mountainous forest CZ had the lowest annual mean runoff, runoff coefficient, and sediment yield at 1.56 mm, 0.41%, and 307.36 g/m², respectively while natural grassland in all CZs had highly annual mean runoff and lowly annual mean sediment yield (P<0.05). Furthermore, site-specific characteristics of runoff plots at the hillslope such as land uses, area, slope, and slope length were the first controlling factors to affect the difference of annual runoff and sediment yield in CZs. There are interactions among these multiple factors along with nonlinear enhanced relationships. The results indicate that vegetation recovery can effectively retain soil and water, but we need to select suitable types when implementing regional vegetation restoration, and the priority is to select natural grassland, shrubland, and forest in the loess hilly and gully region. Our study can provide a scientific basis for the optimal allocation of vegetation restoration on the Loess Plateau.