Premium
A parallel implementation of a 4DEnVar ensemble
Author(s) -
Arbogast Étienne,
Desroziers Gérald,
Berre Loïk
Publication year - 2017
Publication title -
quarterly journal of the royal meteorological society
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.744
H-Index - 143
eISSN - 1477-870X
pISSN - 0035-9009
DOI - 10.1002/qj.3061
Subject(s) - data assimilation , ensemble learning , ensemble forecasting , computer science , implementation , formalism (music) , nonlinear system , statistical ensemble , algorithm , mathematics , artificial intelligence , canonical ensemble , meteorology , statistics , physics , art , musical , quantum mechanics , monte carlo method , visual arts , programming language
The four‐dimensional ensemble variational (4DEnVar) formulation has received considerable attention during recent years, especially at numerical weather prediction centres that are (or were) relying on a 3D/4D‐Var formalism for their data assimilation systems. Since 4DEnVar background‐error covariances are, by construction, given by an ensemble of 4D nonlinear trajectories, an important issue is the way in which to build this ensemble. The use of an ensemble of perturbed 4DEnVar to generate the ensemble is a natural approach, but raises difficulties for the input and storage of 4D trajectories. A parallel implementation of such a 4DEnVar ensemble (En‐4DEnVar) approach is proposed, with distributed input and storage of ensemble perturbations. It has the benefit of an object‐oriented implementation of 4DEnVar, which has recently been developed at Météo‐France. First results obtained with the French global model Action de Recherche Petite Echelle Grande Echelle (ARPEGE) show that such an approach is efficient and suggest that En‐4DEnVar implementations are tractable.