A radiative transfer based approach to merge SMOS and AMSR‐e soil moisture retrievals into one consistent record

In 2010 Soil moisture was recognized as an Essential Climate Variable (ECV) by the Global Climate Observation System (GCOS) and in the same year ESA started the development of a consistent, long‐term, multi‐sensor time series of satellite derived soil moisture. First as part of the Water Cycle Observation Multi‐mission Strategy (WACMOS) and later as a part of the Climate Change Initiative (CCI) program. In 2014,a one year project was initiated by ESA to provide guidelines for the inclusion of soil moisture retrievals from the Soil Moisture and Ocean Salinity (SMOS) sensor. With this project testing three different fusion methods were be included; one based on a neural network approach (Rodriguez‐Fernandez et al., 2014), one based on a linear regression approach (Wigneron et al., 2004) and one based on the CCI soil moisture baseline algorithm, the Land Parameter Retrieval Model (LPRM, Owe et al., 2008). This study focused on the last method. The LPRM was applied to the SMOS observations by optimizing the single scattering albedo and roughness, against two different modelled soil moisture products (MERRA and ERA‐Land) , comparison against SMOS networks from the international soil moisture network over the period of July 2010 to December 2013 (Van der Schalie et al., 2015). The comparison against SMOS Level 3 soil moisture reached correlations of over 0.9 for the continents that are mostly free of radio frequency interference. Then AMSR‐E and SMOS retrievals were combined by updating the AMSR‐E LPRM and optimizing its parameters to best match the SMOS LPRM retrievals. Other updates included an improved approach to estimate the effective temperature at C‐band. The resulting AMSR‐E LPRM retrievals were evaluated against MERRA and ERA‐Land and validated against available In Situ networks and revealed a lower RMSE as compared to the original AMSR‐E LPRM v5 products while still having high correlations with the mentioned datasets