Quasi-Global Soil Moisture Retrieval Using Spaceborne CYGNSS Data with the Stacking Fused Machine Learning Model

Soil moisture is a critical variable in hydrological processes, influencing agricultural productivity, flood forecasting, and drought resilience. Global Navigation Satellite System Reflectometry (GNSS-R) technology emerges as an innovative SM retrieval with high temporal and spatial resolution. A spaceborne GNSS-R SM retrieval method based on the stacking fusion model is developed in this study. Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), and Random Forest (RF) are combined together using the stacking method to develop the fusion model, which is employed to retrieve SM through learning the relations between SM and GNSS-R features derived from the Cyclone Global Navigation Satellite System (CYGNSS) data. The SM measurements from Soil Moisture Active Passive (SMAP) are used as reference and ground-truth data. The stacking fusion model is trained and validated using a range of input features, including the surface reflectivity and its statistics derived from CYGNSS with the consideration of both coherent and incoherent reflection, as well as the vegetation optical depth, surface roughness and ground surface temperature derived from SMAP. The CYGNSS data collected from August 2019 to July 2022 are employed in this work. 10% of the dataset is randomly selected as the training set and the remaining 90% of the dataset is used as the test set. The SM retrieval results indicate that CYGNSS SM presents good agreement with SMAP data, with a root-mean-square-error (RMSE) of 0.0512 cm3/cm3 and a correlation coefficient of 0.94 over the land surfaces within the latitude $$\:\pm\:38^\circ\:$$ . The CYGNSS SM is an addition to SMAP SM over the pan-tropical coverage.