|Title||Flow-dependent, geographically varying background error covariances for 1D-VAR applications in MTG-IRS L2 Processing|
|Year of Publication||2012|
|Authors||Hólm, EV, Kral, T|
|Secondary Title||Technical Memorandum|
Satellite retrievals and other 1-dimensional variational data assimilation (1D-VAR) applications do need background error statistics of the atmospheric variables used, in particular temperature, water vapour and ozone. These error statistics are often provided as single global climatological profiles. In this report we describe a self-contained program package that gives more accurate error statistics that closely follow the errors used in the ECMWF analysis system. It has been possible to make the 1D-VAR errors simpler than those used by the 3D ECMWF analysis because horizontal correlations and wind errors are not needed. The error statistics consist of vertical error correlation matrices and three-dimensional error variances of the day, both derived from the ECMWF ensemble data assimilation system. In addition, a variable transform converts the model variables (temperature and mixing ratios of water vapour and ozone) to variables with more Gaussian and less correlated error statistics. Gaussian error statistics make the background errors more robust and removing cross-variable correlations simplifies the vertical correlation matrices. The program package that calculates the background errors is self-contained, including documentation. It is regularly updated by ECMWF to reflect major model and resolution upgrades. Following major upgrades, new correlation matrices are needed, and these will be available from ECMWF as part of the latest version of the software.