Integration of SAR Data and Machine Learning for Soil Moisture Estimation using SMAP Validation
Main Article Content
Abstract
Soil moisture estimation is crucial for various applications, including precision agriculture, hydrology, and climate studies. Synthetic Aperture Radar (SAR) data has proven to be highly effective for soil moisture prediction due to its capability to provide high-resolution, all-weather observations. This study explores the potential of two machine learning techniques, first Support Vector Regression (SVR) and second one Random Forest Regression (RF), for soil moisture prediction using SAR-derived features. This research utilizes the Sentinel-1 SAR imagery, validated soil moisture data products given by NASA Soil Moisture Active Passive (SMAP) Mission which is a NASA satellite observatory focused on measuring and mapping soil moisture, and used the auxiliary data derived from Google earth engine. Feature selection techniques are applied to enhance model performance, and hyperparameter tuning is conducted to optimize predictive accuracy. The performance of SVR and RF models is evaluated based on correlation (r), Root Mean Square Error (RMSE), and R² scores. Experimental results indicate that both SVR and RF models demonstrate strong predictive capabilities, but the RF model exhibits slightly superior performance in terms of generalization and resistance to overfitting.