Study region: The Shandian River Basin, China. Study focus: This study developed an integrated framework combining the geographical detector (GeoDetector) and Seasonal and Trend decomposition using Loess (STL) to quantify the spatial hierarchy, temporal stability, and interaction effects of environmental controls on soil moisture. Environmental factors were evaluated at both the site scale and the 9-km scale, and the GeoDetector-derived q-value series of topographic factors were further decomposed using STL to separate persistent controls from seasonal effects. Risk-and interaction-detector analyses were then used to resolve class-dependent soil-moisture differentiation and joint terrain effects. New hydrological insights for the region: Results indicate a clear scale dependence in topographic effects, with substantially stronger explanatory power for soil-moisture spatial heterogeneity at the site scale than at 9 km. Among the tested terrain descriptors, height above the nearest drainage (HAND) was the strongest and most temporally stable factor; in contrast, slope, the topographic wetness index (TWI), and the TWI-based indices NI and NI2 were more seasonally responsive. Risk-detector analysis reveals monotonic, threshold-like, and nonlinear class responses, while interaction analysis shows that all pairwise topographic combinations exceed single-factor performance, indicating multidimensional and process-coupled terrain control. External tests indicated that transferability is bounded and partial, whereas HAND remained the consistently robust descriptor. The proof-of-concept downscaling experiment demonstrated that hydrologically meaningful terrain descriptors outperformed a digital elevation model (DEM)based predictor set.