A surface soil moisture (SM) condition at high spatiotemportal resolutions is required by regional Earth system applications. Here, we mapped daily 1-km SM in the Babao River Basin in the northwest of China during the summers from 2013 to 2015 using a random forest (RF) method by merging SM information retrieved from in situ measurements, optical/thermal remote sensing, and topographical indices. Relative importance analysis was used to determine the optimal predictors for estimating high-resolution SM. A specific RF model (RFVI+sup) was constructed using the optimal predictors including remote sensing albedo, apparent thermal inertia (ATI), normalized difference vegetation index, normalized difference infrared index 5, soil adjusted vegetation index, and topographical indices (aspect and elevation). The RFVI+sup also accounted for missing observations of the thermal index (e.g., ATI) over the mountainous regions. In the comparison between the SM estimates using the new RFVI+sup model and other RF models, the spatial coverage of available estimates increased from 14% to 64% over the study region, the correlation coefficient values were improved to 0.75, the unbiased root-mean-squared difference values decreased to 0.032 m(3)/m(3). Thus, the proposed RF method provided accurate SM estimates with high spatiotemporal resolution over the mountainous regions, by merging multiresource datasets from in situ measurements, remotely-sensed, and topographical indices.< Réduire