A new approach for validating satellite estimates of soil moisture using large-scale precipitation: Comparing AMSR-E products

Validation of remotely sensed soil moisture is complicated by the difference in scale between remote sensing footprints and traditional ground-based soil moisture measurements. To address this issue, a new method was developed to evaluate the information content of remotely sensed soil moisture data using only large-scale precipitation. Under statistically stationary conditions, precipitation conditionally averaged according to soil moisture results in a sigmoidal curve, with high average precipitation corresponding to high average soil moisture, in a manner that reflects the dependence of drainage and evaporation on soil moisture. However, errors in soil moisture measurement degrade this relationship. Thus, soil moisture data can be assessed by the degree to which the natural sigmoidal relationship is preserved. The metric of mutual information was used as an error-dependent measure of the strength of the sigmoidal relationship. In this way, a choice model was constructed between different soil moisture data records, based on maximum mutual information. Three AMSR-E soil moisture algorithms (VUA–NASA, NASA, and U. Montana) were evaluated with the choice model for a nine-year period (2002–2011) over the contiguous United States at ¼° latitude-longitude resolution, using NLDAS precipitation. The U. Montana product resulted in the highest mutual information for 50% of the region, followed closely by VUA–NASA at 47%, and distantly by NASA at 3%. Areas where the U. Montana product yielded the maximum mutual information generally coincided with low vegetation biomass and flatter terrain, while the VUA–NASA product contained more useful information in more rugged and more highly vegetated areas.