ADAPTIVE PARALLELIZATION TECHNIQUES IN GLOBAL WEATHER MODELS

Meteo-France is developing a new code for operational weather forecast ing that includes components both for the variational analysis of mete orological observations and for the prediction model itself. So, we ne ed to prepare our code for the new generation of parallel machines. Th e parallel version of Meteo-France code ARPEGE (Action de Recherche Pe tite Echelle Grande Echelle) uses spectral computations for the dynami c. For the spectral part, we use the strategy developed at ECMWF (Euro pean Centre of Medium-Range Weather Forecasts). The full physics schem e and some options of the semi-Lagrangian are specific to our use of A RPEGE. In this paper, we present the different strategies used for the parallelization of Meteo-France forecasting code ARPEGE using the sem i-Lagrangian scheme. (C) 1998 Elsevier Science B.V. All rights reserve d.