Accounting for Correlated Satellite Observation Errors in GSI

In data assimilation systems the observational error covariance describes errors
in the observations as well as in the forward model. Forward model errors such as in the radiative
transfer modeling and representativeness, for example, are likely correlated especially for satellite observations.
However, most assimilation systems currently neglect such error correlations and apply instead observation thinning
and (or) error inflation to reduce the undesirable effects of not accounting for correlations. Indeed, recent
studies (e.g., Steward et al. 2014) suggest that even simple attempts to specify error correlations in observations
might be better than error inflation.

The capability to handle inter-channel correlations for satellite observations in the Gridpoint Statistical
Interpolation(GSI) has been implemented. Observation errors and inter-channel correlations are estimated
based on the observation residuals from the Goddard Earth Observing System Model, Version 5 (GEOS-5) for all
sounders presently assimilated in our real-time system. The GEOS-5 DAS integrates the GEOS-5 AGCM with GSI.
The focus of this work is to improve the GSI use of hyperspectral observations such as those from AIRS, IASI and
CrIS instruments. This presentation provides an update of the development of accouting for inter-channel error
correlations within the GEOS-5 data assimilation system.