Séminaire
The talk will present our efforts to develop an ensemble forecasting system for the Red Sea. The system implements GCMs at Kilometers scales capable of assimilating available observations based on distributions estimated from large ensembles that could be fully executed in parallel. By ‘large’, we are targeting ensembles with several thousands of members. We are also actively working on developing more advanced assimilation schemes to enhance the system outputs in term of computational cost and performances. In this context, I will briefly present three new ensemble Bayesian filters sharing in common « targeted » sampling strategies; the first one introduces a deterministic sampling of the observations perturbations in the stochastic EnKF, the second exploits the idea of filtering with one-step-ahead-smoothing, and the third resorts to a Gaussian-mixture update step. Results from the current version of the system implementation will be presented and future plans will be discussed.
bruno.deremble@ens.fr