Evaluation of the Bayesian Downscaling Algorithm for Achieving Higher Resolution Soil Moisture Data

The NASA-launched Soil Moisture Active Passive satellite mission (SMAP) had the objective to globally characterize soil moisture with an intermediate resolution (9 km), through the integration of radar (3 km) and radiometer (36 km) observations. The SMAP team has evaluated various downscaling techniques to achieve this goal. This study examined the performance of an additional downscaling technique, the Bayesian merging method, as an alternative candidate approach. This method breaks from the standard linear downscaling techniques of SMAP, opting instead for a more innovative approach based on Bayes' theorem. Here, the intermediate resolution soil moisture is achieved via the incorporation of a background estimate, which is refined through comparison between observed and predicted brightness temperatures and backscatter coefficients that link the high- and low-resolution data. However, it is crucial to assess the robustness of the Bayesian method using actual satellite observations, in addition to its prior evaluation using synthetic datasets. The fourth Soil Moisture Active Passive Experiment (SMAPEx-4), conducted in Australia, represented the sole occasion for concurrent high-resolution airborne observations during operation of the SMAP radar. As such, this study employed the Bayesian algorithm using the SMAP datasets throughout the SMAPEx-4 period. Downscaled soil moisture products from this method, as well as from the official baseline and enhancement techniques, were compared. The average root-mean-square-error and R² of the 9 km downscaled soil moisture were found to be 0.035 cm³/cm³ and 0.55 for the Bayesian method, 0.093 cm³/cm³ and 0.35 for the baseline, and 0.069 cm³/cm³ and 0.41 for the enhancement method.