The scale-dependence of SMOS soil moisture accuracy and its improvement through land data assimilation in the central Tibetan Plateau

This study evaluates SMOS (soil moisture and ocean salinity) soil moisture products against a newly established soil moisture network in the central Tibetan Plateau. Based on the results, the validity of assimilating the SMOS soil moisture retrievals into a land surface model is further evaluated. The ground truth is obtained by spatial upscaling from the network measurements within an area of approximately 10,000 km2. Results show that both SMOS L2 and the preliminary version of L3 soil moisture products have large biases at the SMOS node scales (15 and 25 km), but they can reflect the surface wetness conditions well when averaged at a 100-km scale during the unfrozen season (June to October). This finding indicates the applicability of SMOS retrievals is scale-dependent. Meanwhile, very few retrievals are available in winter due to the presence of frozen soil and snow cover, and the accuracy of ascending retrievals degrades during transition when diurnal freezing–thawing cycle occurs. Considering the SMOS L2 product has a better accuracy than that of L3, we assimilate it into a land surface model using a dual-pass land data assimilation scheme. The data assimilation estimate without in-situ tuning proves superior to either remote sensing or land surface modeling in estimating surface soil moisture for the unfrozen season, and its accuracy fulfills the SMOS measurement requirements (RMSD ≤ 0.04 m3 m− 3). Thus, the assimilation of SMOS retrievals holds promise to produce regional soil moisture dataset with acceptable accuracy for the Tibetan Plateau semi-arid region.