AulignŽ, Thomas and Hongli Wang, National Center for
Atmospheric Research
The assimilation of
cloud-and-rain-contaminated satellite radiances is the next frontier for
improving the model short-term prediction skills. Infrared radiances are particularly
difficult to handle because of the data volume, the cost of the radiative
transfer model and mostly the non-linearities in the observations operator.
Typical model background errors translate into huge
observation-minus-background differences in radiance space, which are not
linearly related to the model state. We will present various techniques to
pre-process the first-guess from a high-resolution WRF model run in order to
improve the fit to the observations. A variational analysis is performed with
an augmented control variable to include cloud parameters. An incremental
approach is followed to handle the non-linearities in the observation operator
and a robust l1-l2 (Huber) norm used to account for non-Gaussian error
distributions. The analysis manages to fit the observations closely and adjust
the cloud variables in a reasonable manner.