A posteriori validation applied to the 3D-VAR Arpège and Aladin data assimilation systems

In this paper we present results from the application of a posteriori diagnostics on existing data assimilation systems. The systems of interest are the global data assimilation Arpège 3D-VAR of Météo-France, and the limited-area 3D-VAR analysis of Aladin. First, we discuss how the diagnostics can be ported from theory to practical applications, using aggregates of analysis information over time. We then compare the behaviour of the diagnostics in the two different data assimilation systems, with a focus on the properties of the background and observational error variances. Secondly, the a posteriori validation is used for the off-line tuning of two scalar error parameters in the Aladin 3D-VAR. The tuning provides error variances that better fit the statistics of the innovation vector, without loss of quality in the analyses. Finally, the link of this approach with ensemble-based techniques is made. Especially, we propose to couple the a posteriori diagnostics directly with the usual output of ensemble samples, which could be done at no extra cost. If the results are then used to tune error parameters, then an on-line adaptive assimilation system, based on a posteriori considerations, is obtained.