Saturated hydraulic conductivity (Ks) is a critical parameter governing soil hydrological processes, yet most existing studies focus on upper soil layers (< 100 cm), the depth-dependent behavior of Ks in deep unsaturated soil (> 100 cm) is insufficiently characterized. This study investigated the vertical distribution, spatial variability, and controlling factors of Ks across 0-500 cm soil profiles in a small watershed on the Chinese Loess Plateau. Ks exhibited the highest values in the surface layer (0-5 cm) and decreased with depth, following power or parabolic functions (R-2 = 0.58-0.71) depending on landscape unit. Ks differed significantly among landscape units, with values ranked as tableland >= slope land > gully land, and significantly higher values in tableland than slope land in the 200-500 cm layer (p < 0.05). Boosted regression tree analysis showed that soil properties dominated Ks variability across all depths (56.6%-79.4%), with bulk density and saturated soil water content as the most influential factors. In the upper 0-20 cm, topography and vegetation mainly influenced Ks indirectly via soil properties, while vegetation effects were more evident in the 5-20 cm layer. Below 20 cm, Ks was increasingly controlled by intrinsic soil characteristics, and vegetation effects weakened and became negligible below 200 cm. Although soil variables dominated model performance, incorporating topographic and vegetation variables generally improved predictions in deeper soils (> 20 cm), where they likely act as proxies for long-term geomorphic and pedogenic processes. These results highlight strong vertical heterogeneity in Ks and suggest that depth-explicit parameterization may improve hydrological modeling in loess regions.