A Random Forest Approach for Soil Moisture Estimation at 60 Meters Spatial Resolution

A Random Forest (RF) regression-tree method to derive high-resolution (60 m) surface soil moisture maps is proposed in this study. The developed methodology integrates multi-source synergies by incorporating information from the visible, near-infrared until short-wave infrared spectrum (Sentinel-2), reanalysis data (ERA5-Land) and terrain information (SRTM), using exclusively open access data. The analysis focuses on the central part of the Iberian Peninsula and covers a four-year period (2018-2021). The resulting high-resolution soil moisture maps exhibit greater spatial heterogeneity compared to the ESA Climate Change Initiative (CCI) soil moisture, which was used as a reference in the training of the RF model. These maps have been evaluated using in situ soil moisture measurements from the REMEDHUS network, and show good agreement in terms of Pearson's correlation (0.83), and uRMSE (0.028 m
³
•m
^-3
), demonstrating the method’s significant potential for deriving high-resolution soil moisture information.