|Title||Use of analysis ensembles in estimating flow-dependent background error variances|
|Publication Type||Technical memorandum|
|Secondary Title||ECMWF Technical Memorandum|
|Authors||Kucukkaraca, E, Fisher, M|
|Place Published||Shinfield Park, Reading|
This paper represents a preliminary investigation into the use of ensembles of analyses to estimate flow-dependent variances of background error. We study small-scale, dynamically-active systems in the mid latitudes (the storms over France during 26-28th December 1999) and in the tropics (Hurricane Isabel, 6-29th September 2003). We demonstrate that the use of ensemble-based estimates of background error variance can improve the analysis of such systems. Although the approach has considerable potential, we note that good analyses of such small-scale systems are unlikely to be produced unless they are adequately resolved both by the main analysis system, and by the members of the analysis ensemble. We believe that a small ensemble of relatively high resolution members may be preferable to a larger ensemble of lower resolution members. We also highlight the tendency of the analysis ensemble to under-estimate the variance of analysis and background error, particularly in dynamically inactive regions. We consider a few simple measures that try to correct this under-estimate, but note that a better understanding (and representation) of the effects of model error, and of the spatial and inter-channel correlation of observation error, is needed before such ad hoc measures can be eliminated.