Quasi-Global Assessment of Deep Learning-Based CYGNSS Soil Moisture Retrieval

A high spatial and temporal resolution global soil moisture product is essential for understanding hydrologic and meteorological processes and enhancing agricultural applications. Global navigation satellite system (GNSS) signals at L-band frequencies that reflect off the land surface can convey high-resolution land surface information, including surface soil moisture (SM). Cyclone global navigation satellite system (CYGNSS) constellation generates Delay-Doppler Maps (DDMs) that contain important Earth surface information from GNSS reflection measurements. DDMs are affected by soil moisture and other factors such as complex topography, soil texture, and overlying vegetation. Including entire DDM information can help reduce the uncertainty of SM estimation under different conditions along with remotely sensed geophysical data. This work extends our previously developed deep learning (DL) framework to a global scale by utilizing processed DDM measurements (analog power, effective scattering area, and bistatic radar cross-section) and ancillary data (elevation, slope, water percentage, soil properties, and vegetation water content). The DL model is trained and evaluated using the Soil Moisture Active Passive (SMAP) mission's enhanced SM products at 9-km resolution. This study comprehensively evaluates the DL model against publicly available CYGNSS-based SM products at a quasi-global scale. In addition to the typical comparison against in-situ measurements, a robust triple collocation technique is used to evaluate the DL-based SM product and other CYGNSS-derived SM products.