Abstract:
The NASA Global Modeling and Assimilation Office (GMAO) is working to develop land data assimilation (DA) capabilities to support Goddard Earth Observing System (GEOS) reanalysis efforts using a land data assimilation system (GEOS LDAS). We aim to include multi-decadal records of land surface observations in upcoming reanalysis, beginning with the use of ASCAT soil moisture (SM) retrievals alongside the more recent record of SMAP L-band (1.4 GHz) brightness temperature (Tb) observations.

Here we describe how we have added a multi-sensor capability to GEOS LDAS to enable the joint assimilation of SM and Tb observations. We then assess the impact on global SM analysis of assimilating ASCAT SM from 2015 to 2021, with validation using a network of in-situ observations, independent satellite observations and DA diagnostics. Assimilating ASCAT SM universally improves our model estimates of SM relative to a control. Statistical metrics comparing with in-situ observations are improved, anomaly correlation with independent satellite observations increases globally, and the misfit between both ASCAT SM and SMAP Tb observations and corresponding (3-hour) background forecasts comprehensively decreases.

We then compare this impact of assimilating ASCAT SM with the assimilation of SMAP Tb alone and investigate what happens when both ASCAT SM and SMAP Tb are jointly assimilated. We find that assimilating just SMAP Tb leads to greater improvements in the skill of SM estimates than assimilating ASCAT SM alone. When both ASCAT SM and SMAP Tb are assimilated together, overall global SM estimation skill is very similar to the SMAP Tb only experiment. But DA diagnostics suggests there are specific times and locations when the background forecast is improved relative to a single sensor experiment, and other times and locations when it is degraded, implying that information from the two different sensors does not always agree.

Whilst adding ASCAT SM in addition to SMAP Tb has a neutral impact, the fact assimilating just ASCAT SM clearly improves SM estimation skill implies that the multi-sensor approach is beneficial. It increases the robustness of the system, enabling improved SM estimates when SMAP observations are not available, and greatly extends the period over which SM observations are available for reanalysis.