Title: Variational ensemble-based forecast error variance maps filtering, a toy-models approach

Authors: Benjamin Menetrier (CNRM-GAME (Meteo-France/CNRS))
Thibaut Montmerle (CNRM-GAME (Meteo-France/CNRS))
Loik Berre (CNRM-GAME (Meteo-France/CNRS))
Yann Michel (CNRM-GAME (Meteo-France/CNRS))

A variational ensemble, based on an explicit perturbation of assimilated observations and on an implicit perturbation of the background through the cycling, allows background error covariances to be estimated (e.g. Belo Pereira and Berre, 2006). However, the high computational cost of such ensembles in operational applications restricts the ensemble size, leading to a significant sampling noise. The work of Raynaud et al. (2009) has shown that an objective spectral filtering of the variance maps can reduce the noise while keeping rich and robust information about the forecast error flow-dependency. This technique is now used operationally at Meteo-France to provide variances of "errors of the day" to the global NWP model ARPEGE.

Our plan is now to adapt this approach at the convective scale for the operational cloud resolving model AROME. However, the heavy handling of such an operational code have led us to design a variational data assimilation testbed running cheap limited-area toy-models. It is specially intended to work on forecast error covariances modelling algorithms using variational ensembles. It will be shown that spectral filtering still is efficient to reduce the sampling noise of error variances, but that a wavelet filtering is better adapted to detect and preserve significant heterogeneous structures. An inflation of forecast perturbations, tuned using a posteriori diagnostics in observation space, is also investigated in order to represent model error.

GMAO Head: Michele Rienecker Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: May 27 2011