Title: Test of a new preconditioner in operational 4DVAR at Météo-France

Authors: Cécile Loô, (GAME/CNRM))
Gérald Desroziers (GAME/CNRM)

Recent and coming developments in incremental 4DVAR at Météo-France (increase of vertical and horizontal resolutions in both non linear and linear models, improvements of linearized physical parametrizations …) increase computational cost of the assimilation part. Operational constraints need to limit this increase. Preconditioning techniques which enable to speed up minimizations while accelerating the convergence of this iterative method is a way to save computational cost. A new preconditioning algorithm, built from Ritz vectors given by a previous minimization and which approximate the first eigenvectors of the Hessian matrix, is evaluated in operational context of the global model ARPEGE and compared with the spectral preconditioner currently used.

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

Curator: Nikki Privé Last Updated: May 27 2011