ON AN ADAPTIVE FILTER BASED ON FORECAST ERRORS MODELLING FOR DATA ASSIMILATION AND ITS COMPARISON WITH OPTIMAL INTERPOLATION METHOD

摘要

A technique, closely related to the breeding method, is described for simulating prediction errors for components of the system state which is very simple to implement and requires only some integrations of the numerical model. This technique is proposed for approximating forecast errors covariance matrices (ECM) participating in construction of the filter gain which is a central problem in application of the adaptive filtering algorithm [1]. Very high performance of a new filter will be demonstrated for SSH data assimilation experiment with the oceanic model MICOM.