Adaptive soil moisture bias correction in the ECMWF land data assimilation system

Abstract

Bias-correction (BC) is typically needed prior to the assimilation of satellite-derived soil moisture (SM) observations in land surface models. Active ASCAT and passive SMOS satellite-derived SM observations are assimilated in the ECMWF integrated forecasting system (IFS). Prior to assimilation, the ASCAT SM observations are bias-corrected using a seasonal rescaling technique. SMOS level 1 observations are converted to level 2 SM via a neural network, which is trained on the global ECMWF operational SM analysis. However, neither of these techniques allow for non-stationary biases and the globally trained SMOS neural network is affected by local biases. In this presentation a two-stage filter is employed in the ECMWF IFS to dynamically correct biases in the SM observations, whilst allowing the assimilation to correct sub-seasonal scale errors. Over a 3-year test period this adaptive BC approach leads to (i) reduced observation-model biases, (ii) slight improvements in SM analysis performance against in situ data and (iii) reduced mean errors in relative humidity forecasts near the surface in the northern hemisphere midlatitudes. This will benefit the development of a unified coupled land-atmosphere data assimilation system in the context of the CERISE European project (Copernicus Climate Change Service Evolution). Furthermore, it is expected that the assimilation of non-biased ASCAT SM observations will improve the root-zone SM products for the hydrological satellite applications facility (H SAF).