Relevance of climatological background error statistics for mesoscale data assimilation

Abstract The relevance of climatological background error statistics for mesoscale data assimilation has been investigated with regard to basic assumptions and also with regard to the ensemble generation techniques that are applied to derive the statistics. It is found that background error statistics derived by simulation through Ensemble Data Assimilation are more realistic than the corresponding statistics derived by downscaling from larger scale ensemble data. In case perturbation of observations is used to inject a spread into the ensemble, and the ensemble is integrated over a few hours only, it was found that the derived structure functions may be contaminated by the geometry of the observing network. The effects of the assumptions of stationarity, homogeneity and isotropy, that are generally applied in the generation of background error statistics, and the implications of the background error covariance model have also been illustrated. Spatial covariances derived under these assumptions were contrasted against spatial covariances obtained by ensemble averaging only, preserving the signals from forecast errors of the day. This indicates that it is likely to be favourable to apply data assimilation with ensemble background error statistics obtained from ensemble averaging, like in ensemble Kalman filters or in hybrids between variational and ensemble data assimilation techniques.