Estimating Daily Soil Moisture at High Spatial Resolution for Drought Monitoring by Fusing Multi-Source Data Based on Random Forest

The Netherlands is grappling with increasingly severe drought conditions as a result of climate change. Soil moisture(SM) is a significant indicator for drought monitoring, but available local/regional SM products lack high spatiotemporal resolution. In this study, we generate a high-resolution (10m) SM product for the Twente region in the Netherlands by downscaling the original 1km SM product with auxiliary variables (day time land surface temperature, night land surface temperature, Ground Range Detected dataset, Enhanced Vegetation Index, Leaf Area Index, precipitation, soil texture, groundwater level) using a random forest model. The RF model showed an acceptable result with correlation coefficient of 0.8 and RMSE of 0.0418(cm3 cm−3). Feature importance of the RF model indicates that day time land surface temperature and groundwater level are the most significant features, followed by clay and precipitation. The spatial and temporal distribution of downscaled 10m SM product is consistent with the original 1km SM product. Applying the 10m SM product for drought monitoring indicates the downscaled soil moisture product can effectively capture drought conditions.