ADAPTIVE FILTERING - APPLICATION TO SATELLITE DATA ASSIMILATION IN OCEANOGRAPHY

A new approach to assimilation of observations is proposed, which exte nds previous work on adaptive Kalman filtering, In the latter, the gai n matrix of the filter was progressively determined without a priori e xplicit specification of the covariance matrices of the model or obser vation noise, so as to minimize the norm of the innovation vector. The new step taken here is to introduce a (relatively) low dimension para meterization of the gain matrix, thereby substantially decreasing the numerical cost of the filter. The reduced-order adaptive filter (ROAF) thus defined is tested on a simple diffusive-reactive equation, and i mplemented on the four-layer adiabatic Miami isopycnical coordinate oc ean model (MICOM), In the latter case, the filter is used to assimilat e synthetic observations of surface height. Both sets of experiments c learly show the efficiency of the proposed approach, and its superiori ty, in terms of the quality of the results, on Newtonian relaxation. I n the case of the diffusive-reactive equation, the reduced-order adapt ive filter is also superior to, and more economical than, a reduced no n-adaptative Kalman filter. (C) 1997 Elsevier Science B.V.