Design Of The Ensemble Kalman Filter Data Assimilation System And Its Application In The Ensemble Prediction

Data assimilation is important for the improvement of numerical weather prediction. With the augment of observation data and the improvement of observation quality, the advancement and development of data assimilation technique rely on the description of the uncertain information of background and observation. The description of background error covariance is more accurate from the functional form in 3D-Var to the implicated evolution at the assimilation time interval in 4D-Var, and to the predicting in the Kalman filter. So the actual background error covariance is the key to success of data assimilation technique. The ensemble Kalman filter (EnKF) can obtain the flow-dependent background error covariance by statistic of the ensemble samples so that it is becoming a new research focus in the current data assimilation field.The main aim of the paper is to establish a new ensemble Kalman filter data assimilation system which can be practically applied. On the other hand, for the effective implementation of EnKF and avoiding the filter divergence, the initial perturbation method and the estimation method of model error have been studied.The following are the main conclusions and results:(1) In this paper, a practical GRAPES ensemble Kalman filter data assimilation system at pressure level and sigma level is firstly established in China. For the difficulty of assimilation real observations, the observations are organized into batches that assimilated sequentially and the inverse of matrix has been solved by singular value decompositon method (SVD). For the correlation noise caused by the limit ensemble number, the Schur product is used in the horizontal and vertical correlation to filter out the small correlations associated with the remote observations. To make the analysis balanced, the variable-constraint EnKF system has been built.(2) Based on the above studies, throuth the ideal and actual observation tests, the EnKF system has been verified. The number of ensemble members and Schur radius experiments indicate that with the number of ensemble members increasing, the ensemble samples can estimate more exactly the small correlations associated with remote observation; at the same time the analysis errors decrease. The best Schur radius is related with the ensemble size and the correlation characteristic length. The forecasting errors have been studied, which indicates that the error correlation and error variance are flow-dependent. The forecasting errors are largest in the rapid changing weather structures. The experiments of regional and global ensemble prediction of EnKF system suggest that EnKF system has the same quality of analysis and prediction to the 3D-Var system, even better than 3D-Var system.(3) The initial perturbation methods have been studied by using the spacial correlation function method and 3D-Var's control variable perturbation method. The study of initial perturbation methods show that the error characteristic of perturbation field which is obtained from the correlation function is uniform and isotropic in spacial distribution. However, the horizontal error correlation length of its derivative variables is too small and the noise of error correlation is large. The background perturbation field with balances, scales and correlations can be obtained by perturbing the control variable in 3D-Var system.(4) The model errors have been estimated by the discriminations among the forecasting fields in different resolutions. The experiments show that the model errors increase linearly with the decrease of resolution and the model errors in different levels have different growth rates with the change of resolution. The increase of the model error is obvious with the length of model forecasting.(5) The global EnKF data assimilation cycle experiments show that the filter divergence is serious without the consideration of the model errors, and the covariance inflation method can only improve the quality of humidity analysis and prediction in the EnKF system. When the model error perturbation is added in the cycles, the ensemble spread error can represent the ensemble mean error and the quality of analysis and prediction of the EnKF system are improved.