The Comparation Research Based On Some Ensemble Data Assimilation Methods

Recently, the research about the land data assimilation is becoming a hot title in the earthscience research field. It plays an important role not only in the atmosphere but also in theocean filed. However, the data assimilation methods as the centre bridge which connect theobservation and forecast model are also developing rapidly. Many researchers devotethemselves to leading the new research to the data assimilation filed. The data assimilationmethod mainly consists of the continuous and the sequential data assimilation method.The sequential data assimilation method is developed from the classical Kalman Filter.In recent years, in order to overcome the shortcoming of the use of the Kalman Filter in thelinear system, the sequential data assimilation method based on the ensemble is developingrapidly and also have many types such as Ensemble Kalman Filter, Iteration EnsembleKalman Filter, Deterministic Ensemble Kalman Filter, Ensemble Transform Kalman Filterand so on.In this paper, it introduces some classical ensemble data assimilation methods and dosome number experiments about them. Firstly, aiming at the Ensemble Kalman Filter,Deterministic Ensemble Kalman Filter and Ensemble Transform Kalman Filter, we use theLorenz-96model as the forecast mode and the rmse as the evaluation function to make aseries of number tests. The main parameters that we research are the ensemble size, theobservation error, the inflation factor, the location radius and so on. Secondly, we also dosome similar comparation experiments for the Ensemble Kalman Filter, Iteration EnsembleKalman Filter and Iteration Expand Kalman Filter. The experiments show that:(1) The Ensemble Kalman Filter shows better assimilation performance when thenumber of the ensemble is big, but also it will increase compute burden, while theDeterministic Ensemble Kalman filter can make up the shortage.(2) The Deterministic Ensemble Kalman Filter shows better assimilation performancethan Ensemble Kalman Filter and Ensemble Transform Kalman Filter in the condition of thesame observation numbers.(3) It is very important for the optimal algorithm to choose the suitable inflation factor.(4) The Deterministic Ensemble Kalman Filter shows the best assimilation performancein the condition of the same location radius.(5) Though the Iteration Ensemble Kalman Filter and the Iteration Expand Kalman Filtershow better assimilation performance in the strongly nonlinear system, in practice theEnsemble Kalman Filter can also be the best solution.