SINGULAR EVOLUTIVE KALMAN FILTERS FOR DAT A ASSIMILATION IN OCEANOGRAPHY

We propose a Kalman type filter for assimilating oceanic data into a n umerical model. It essentially consists in approximating the error cov ariance matrix by a singular low rank matrix which leads to making cor rections only in those directions for which the error is not sufficien tly attenuated by the system. These directions evolve the time accordi ng to the model evolution, yielding its adaptative nature to the filte r. The filter has been applied in a realistic framework on the Pacific tropical ocean, yielding quite satisfactory results with simulated al timetric data, distributed according to the spatio-temporal sampling f rom the Topex/Poseidon satellite. (C) Academie des sciences/Elsevier, Paris.