CHIMERE is a 3D eulerian chemistry-transport model covering a part of Western Europe with a resolution of about 50km. Several zooms over subregions of France permit a more detailed study of urban pollution phenomena with a resolution of about 4-6km. The model is simplified enough to allow for long simulation and real-time forecast of photoxidant pollution. With this multiscale system, a series of data assimilation processes is performed at regional and European scale. Observational ozone data are assimilated with a Statistical Interpolation method. This process is chosen for its low computer cost, unavoidable constraint for a model designed to forecast in real-time. The space isotropy and homogeneity assumptions, usually made in order to estimate error covariances, are avoided by a new scheme. Compared to simulation without assimilation using the leave-one-out method, the analysis allows a reduction of the RMS-error of 40%. A comparison of several approaches is discussed.

Since the May, 1, 2001, results of the experimental real-time forecasts of Ozone are presented on a web site (http://euler.polytechnique.fr/pioneer/). A forecast with re-initialization of the model using analysis is tested. About 200 observations are used. The improvement of the forecast due to the introduction of observational information is quantified.