Title: Estimation and representation of background error covariances

Author: Loïk Berre (Meteo France)

Ensemble data assimilation methods are now often used to simulate the error evolution during data assimilation cycling. While this is an attractive technique for estimating flow-dependent covariances, it also requires at least plausible estimates of observation errors and of model errors. This explains why innovation-based estimates of errors are also crucial to be used, in conjunction with ensemble estimates.

Such innovation-based estimates rely on assumptions on the respective spatial structures of observation errors and of background errors. Moreover, they provide estimates in model space, whereas background error covariances are to be specified in model space, which suggests that some inversion approach is to be used.

In addition to relying on innovation-based information, the use of ensemble methods also raises sampling noise issues. This can be taken into account by developing filtering techniques, both for variances and correlations. The different possible approaches in this area will be also discussed.

GMAO Head: Steven Pawson Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: Feb 9 2015