Gérald Desroziers, Météo-France

Development of 4DEnVar at Météo-France

E. Arbogast, L. Berre


The 4DEnVar formulation has received a considerable attention in the very recent years. The interest in this formulation relies on different nice properties: it allows the flow-dependent representation of background error covariances from non-linear model trajectories, it is potentially highly parallel on new computer architectures (while remaining global) and also gets rid of the development, maintenance and cost of tangent-linear and adjoint models. We will show the new 4DEnVar system that Météo-France has started to develop. This system relies on the use of the OOPS formalism (Object Oriented Prediction System), initiated at ECMWF. It will be shown that such a versatile system allows different ways to solve the 4DEnVar problem. The proposed 4DEnVar formulation also tries to have an efficient but optimal localization of 4D error covariances through variable transformations or Lagrangian advection of ensemble perturbations.

File available here