Géostatistique et assimilation séquentielle de données

Traditional geostatistical space-time geostatistics is not able to take account of the generally strongly non-linear dynamics of multivariate space-time processes. To this effect physico-chemical transport models are in general more suitable. However, as the latter do not fully master the complexity of the processes they attempt to describe, either because of simplifying hypotheses or because the information serving to set up initial and boundary conditions is imperfect, it is appropriate to introduce statistical techniques in order to assimilate a flow of measurements emanating from automatic stations. Recent projects at Centre de Géostatistique of Ecole des Mines de Paris have permitted to explore these techniques in oceanography and air pollution. Soon it became evident that geostatistics could offer concepts and approaches to enhance Sequential Data Assimilation techniques. The thesis of Laurent Bertino as well as subsequent publications have permitted to develop this fruitful theme of which the present synthesis presents an account.