Best Linear Unbiased Estimators for Fusion of Multiple CYGNSS Soil Moisture Products

NASA's Cyclone Global Navigation Satellite System (CYGNSS) mission has gained significant attention within the land remote sensing community for estimating soil moisture (SM) using the Global Navigation System Reflectometry technique. Multiple algorithms have been developed to generate global SM data products from CYGNSS observations in combination with other remotely sensed geophysical data products. However, different algorithms exhibit variations in performance concerning both time and space due to model capabilities, complexities, and loss calculations. To address these limitations, the fusion of various SM products can be an effective solution. In this study, we explore different fusion algorithms, including the minimum variance estimator, best linear unbiased estimator, and linear weight fusion, to fuse distinct global CYGNSS-based SM products. We consider three SM data products publicly available from the Geosystems Research Institute at Mississippi State University. To assess our model's performance, we compare our fused data product with the Soil Moisture Active Passive (SMAP) mission's enhanced SM products at a resolution of 9 km × 9 km. Our findings reveal notable performance enhancements in several regions when combining different SM data products. The results demonstrate that the minimum variance estimator achieves a mean unbiased root-mean-square difference of 0.0359
$\text{m}^{3}{/}\text{m}^{3}$
with a correlation coefficient of 0.91 for SMAP-recommended grids and also linear weight fusion achieves 0.0389
$\text{m}^{3}{/}\text{m}^{3}$
with a correlation coefficient of 0.90 when no label data are used in the training of fusion.