DATA ASSIMILATION ALGORITHM BASED ON AN ADAPTIVE SUBOPTIMAL KALMAN FILTER

E. G. Klimova

In using an algorithm for assimilation of meteorological observations based on the Kalman filter theory, it is necessary to introduce a covariance matrix of model noise, which is unknown beforehand. The matrix can be estimated using some additional assumptions. It is known that if the covariance matrix of model noise is set to be zero, a theoretical error of the Kalman filter algorithm decreases fast and, as a result, the observations enter an analysis step with progressively smaller factors. This effect is named the divergence of the Kalman filter algorithm. In the paper, an algorithm is proposed for estimating model noise from observational data using a vector of residuals (difference between observations and forecast) in the Kalman filter procedure. The algorithm is based on sharing a simplified model for calculation of the forecast error covariance matrix (suboptimal Kalman filter) and for adaptive estimation of the model noise from observations. The results of numerical experiments on assimilation of the simulated data with a regional adiabatic model of the atmosphere are presented.