SIMPLIFIED MODELS OF COMPUTATION OF THE COVARIANCE MATRICES IN THE KALMAN FILTER ALGORITHM

E. G. Klimova

A serious problem in the application of the Kalman filter algorithm to modern forecasting models is a high order of forecast error covariance matrices that are used in this algorithm. One of the approaches to this problem involves the application of simplified forecasting models to the calculation of forecast error covariance matrices; these matrices are considered in this work. Simplifications are based on the properties of vertical model modes and on a quasigeostrophic approximation. It is shown that all the models considered yield better results, in both matrix forecasting and using the computed matrices in the analysis, as compared to the results of persistence forecasts.