Quasi-global machine learning-based soil moisture estimates at high spatio-temporal scales using CYGNSS and SMAP observations

Based on cross-validation against SMAP data, the ML algorithm is shown to be suitable for retrieving soil moisture from CYGNSS. Median values of unbiased root-mean-square-difference for the quasi-global coverage or regions with vegetation water content less than 5 kg/m(2) are 0.0395 cm3/cm(3 )and 0.0320 cm(3)/cm(3), respectively. Likewise, via independent evaluation against more than 100 in-situ sites, the algorithm is shown to have an unbiased root-mean-square-error of 0.0543 cm(3)/cm(3).

CYGNSS-based retrievals contain similar spatial variability as SMAP across different seasons. Moreover, through a robust triple collocation technique, the accuracy of CYGNSS soil moisture is relatively high over moderately vegetated regions with correlations ranging from 0.4 to 0.8. Based on these validation results, we argue that derived CYGNSS soil moisture estimates can supplement current global soil moisture databases and provide more frequent retrievals at 9 km.