Abstract:
Statistical techniques permit the retrieval of soil moisture estimates
in a model climatology
while retaining the spatial and temporal signatures of the satellite observations.
As a consequence, they can be used to reduce the need for localized bias correction techniques
typically implemented in data assimilation (DA) systems that tend to remove some of the
independent information provided by satellite observations. Here, we use a statistical neural
network (NN) algorithm to retrieve SMAP surface soil moisture estimates in the climatology
of the NASA Catchment land surface model. Assimilating these estimates without additional
bias correction is found to significantly reduce the model error and increase the temporal
correlation against SMAP Cal/Val in situ observations over the contiguous United States.
A comparison with assimilation experiments using traditional bias correction techniques shows
that the NN approach better retains the independent information provided by the SMAP observations
and thus leads to larger model skill improvements during the assimilation. A comparison
with the SMAP Level 4 product shows that the NN approach is able to provide comparable
skill improvements and thus represents a viable assimilation approach.