A Physics Based Satellite Soil Moisture Reconstruction Algorithm

Despite global coverage, remote sensing of soil moisture (SM) is challenged by coarse spatial sensor resolution and shallow sensing depth which result in systematic differences when compared to reference SM measured in-situ. Although improvements have been documented with assimilation of SMAP radiometer data with land surface models, a regionalized solution is needed that leverages crucial physical signatures (SM recessions) to provide further improved estimations, addressing systematic deviations that persist. A key drawback of existing algorithms is the lack of consideration of the uncertainty associated with different physical factors that modulate the SM time series. Specifically, SM drawdown is not influenced by precipitation, which reduces uncertainty considerably. In the present study, a novel approach is demonstrated that splits the SMAP Level 4 SM series, mechanically segregating the recession limbs that last at least 2 days and uses them to modify the complete time series. A bivariate recursive filtering approach is introduced that models the association of initial soil wetness and drying rate during the recession periods, minimizing the disparity to represent the same observed in-situ. Consequently, the modified drying attributes (initial wetness and recession rates) are utilized to reconstruct the complete time series. The approach is validated by comparing ensued estimates with the in-situ measurements from dense and sparse networks from April 2015 to March 2020. The validation metrics show improvements in the reconstructed SM series, with significant enhancements observed for the recession parts of the series. The combined procedure has performed well, demonstrating the importance of associativity of physical processes into SMAP assimilation observations for regional studies.