Validation of CYGNSS soil moisture products using in situ measurements: a case study of Southern China

The spaceborne Global Navigation Satellite System-Reflectometry (GNSS-R) has proven its worth in terrestrial remote sensing applications. Its application to detecting land surface soil moisture (SSM) is particularly intriguing, as it can provide fine-scale SSM products to supplement traditional satellite-based active and passive missions. Various retrieval algorithms have been developed to produce SSM products using spaceborne GNSS-R. However, detailed evaluations of product reliability and robustness are still absent. In this study, we used three data sources to evaluate the level-3 SSM products from the CYclone Global Navigation Satellite System (CYGNSS) mission: (1) satellite-based microwave radiometry product from Soil Moisture Active and Passive (SMAP) mission; (2) a model-based product of Modern-Era Retrospective analysis for Research and Applications; and (3) in situ measurements from over 1800 ground stations in the Chinese soil moisture monitoring network. The study uses typical relative skill metrics and triple collocation approach (TCA)-based metrics, along with corresponding confidence intervals, to analyze the performance of SSM products derived from CYGNSS observations. According to the pixel-by-pixel validation and overall statistical findings, the results reveal that the current CYGNSS-based SSM exhibits low performance in southern China when compared to the radiometry-based data. The coefficient of determination (R2) is low (median R2=0.088) and the unbiased root-mean-square-difference (ubRMSD) is 0.057 cm3cm−3, which is poorer than the results from SMAP against in situ measurements (median R2=0.25, ubRMSD=0.046 cm3cm−3). The TCA-based analysis also revealed that CYGNSS had a relatively poor performance, with the lowest median R2 value of 0.167 and the largest median error standard deviation (ESD) value of 0.055 cm3cm−3. To obtain improved results that can better support related operational applications in the future, enhanced retrieval algorithms and high-accuracy calibration referenced data must be utilized.