Medium-Resolution Soil Moisture Retrieval Using the Bayesian Merging Method

Abstract

The National Aeronautics and Space Administration’s Soil Moisture Active Passive (SMAP) mission, launched in January 2015, was designed to provide a global soil moisture product at medium resolution (~9 km), by combining observations from its radar and radiometer. Several downscaling methods have been proposed by the SMAP team for this purpose. This paper evaluates another candidate downscaling method, namely, the Bayesian merging approach. While this has been tested using a synthetic data set across the USA, it is imperative that it can also be tested using the experimental data for a comprehensive range of land surface conditions (i.e., in different hydro-climatic regions) prior to a global application. Consequently, this paper applies this method using the data collected from SMAP experiments field campaigns in south-eastern Australia that closely simulated the SMAP data stream for a single SMAP radiometer pixel over a three-week interval. The method studied here differs from the linear downscaling methods of the SMAP mission, in that it uses a nonlinear method based on Bayes’ theorem. The medium-resolution soil moisture product is obtained using background soil moisture estimates that are updated according to the difference between the observed and predicted brightness temperatures and backscatter coefficients, relating the high- and low-resolution data. Results were assessed against a reference soil moisture map derived from high-resolution airborne radiometer observations. The root-mean-square-error and $R^{2}$ for the Bayesian merging method were found to be 0.02 cm3/cm3 and 0.55, respectively, at 9-km resolution, being similar to the SMAP’s “optional” downscaling method.