Méthodes multifractales appliquées à la prévision de pluie en utilisant des données radar

Multifractal Methods Applied to Rain Forecast Using Radar Data

¹ CEREVE, Ecole Nationale des Ponts et Chaussées, 6-8 Avenue Blaise Pascal, Cité Descartes, 774555 Marne la Vallée Cedex 2
² CNRM, Météo-France, 1 Quai Branly, 75007 Paris, France
³ Physics Dept., McGill University, 3600 University St, Montreal, Quebec H3A 2T8, Canada

Auteur de correspondance : macor@cereve.enpc.fr

Abstract

In the emergency situations, the hydrological alert and forecast systems require more and more detailed space-time information and reliable rain forecasts in the very short term. The current weather forecasting models or the statistical methods of radar and satellite image processing cannot fully answer these requirements : numerous parameterizations, long spin-up and computation times or strongly nonlinear dynamics of stormy cells not taken into account. The multifractal approach, physically based on the cascade models, takes into account the hierarchy of the structures and their nonlinear interactions over a wide range of space time scales, as well as the time-space anisotropy. The advantage of this approach is that it requires only a very limited number of parameters, which moreover have a strong physical significance.We present the implementation of a procedure of multifractal rain forecast using data radar to determine the past of the cascade generator. We compare two modes of multifractal forecast, respectively stochastic and deterministic.