Optimizing Soil Moisture Retrieval: Utilizing Compact Polarimetric Features with Advanced Machine Learning Techniques

Abstract

Soil moisture plays a crucial role in various environmental processes and is essential for agricultural management, hydrological modeling, and climate studies. Synthetic Aperture Radar (SAR) remote sensing presents significant potential for estimating soil moisture due to its ability to operate in all weather conditions and provide day-and-night imaging capabilities. Among the SAR configurations, the Compact Polarimetric (CP) mode has gained increasing interest as it relaxes system constraints, improves coverage, and enhances target information compared to conventional dual polarimetric SAR systems. This paper introduces a novel approach for soil moisture retrieval utilizing machine learning algorithms and CP SAR features. The CP SAR features are derived from a series of RADARSAT Constellation Mission (RCM) CP SAR imagery acquired over Canadian experimental sites equipped with Real-Time In Situ Soil Monitoring for Agriculture (RISMA) stations. This study employs a diverse dataset of compact polarimetric SAR features and corresponding ground truth soil moisture measurements for training and validation purposes. The results of our study achieved a Root Mean Square Error (RMSE) of 6.88% with a coefficient of determination R2 equal to 0.60, which corresponds to a correlation R between true and predicted soil moisture values of 0.75, using optimized Ensemble Learning Regression (ELR) with a decision-tree-based model. These results improved, yielding an RMSE of 5.67% and an R2 equal to 0.73 (R = 0.85), using an optimized Gaussian Process Regression (GPR) model.