Downscaling SMAP soil moisture using a hybrid machine-learning algorithm

This study introduces an innovative approach for downscaling Soil Moisture Active Passive (SMAP) satellite soil moisture data from a coarse spatial resolution of 9 km to a refined resolution of 1 km. The proposed method uniquely integrates the Scorpion Hunting Strategy Algorithm, a novel nature-inspired optimisation technique, with a fuzzy inference system. Validation was performed using field measurements obtained from two climatically distinct and data-scarce regions: Bhopal (semi-arid) and the Kosi Fan (humid sub-tropical). The resulting high-resolution soil moisture maps demonstrated a strong correlations with the in-situ observations (correlation coefficients (R) of 0.79 and 0.76, and root mean square errors (RMSE) of 0.053 m³/m³ and 0.043 m³/m³ for Bhopal and Kosi Fan, respectively). These findings point to the robustness and versatility of the proposed approach in capturing intricate spatial variations of soil moisture across diverse climatic regimes. Performance of the algorithm was further verified by testing with publicly available data from other fields, enhancing its broader applicability. Overall, this novel approach contributes a significant advancement in soil moisture downscaling techniques, offering valuable insights and practical benefits for application in agriculture, hydrological forecasting, and environmental monitoring.