Relative Soil Moisture Index from Multi-Source Remote Sensing and Random Forest in Tropical Landscapes

Accurate soil moisture (SM) monitoring at high spatial resolution remains challenging in heterogeneous tropical landscapes, where terrain, vegetation, and soil properties interact to drive complex hydrological dynamics. This study develops a Relative Soil Moisture Index (RSMI) by integrating multi-temporal Sentinel-1 Synthetic Aperture Radar (SAR), Sentinel-2 optical imagery, terrain indices, and detailed pedological attributes within a Random Forest machine learning framework. Field sampling campaigns synchronized with satellite overpasses across ten dates during a full seasonal cycle yielded 1,560 gravimetric SM observations from 52 sites representing diverse physiographic units of the Brazilian Federal District in the Cerrado biome. Feature selection combined correlation analysis and Gini importance scores to identify the most informative covariates. The model achieved high predictive performance (R² = 0.78; RMSE = 3.4%), successfully capturing spatial-temporal SM variability across landforms and management systems. The RSMI normalized site-specific dynamics, enabling consistent moisture assessment across varying conditions. Spatial mapping revealed physiographic controls on moisture persistence, with terrain, clay content, and vegetation cover emerging as dominant drivers. The proposed RSMI framework demonstrates strong potential for operational SM monitoring, providing a scalable tool to support precision agriculture, drought risk management, and sustainable land use planning in tropical environments.