Assimilation Strategy Research Based On Ensemble Kalman Filter

In recent years, the climate change impact on human survival that is increasingly outstanding, in order to strengthen the monitoring and forecast of land, atmosphere and ocean, the global earth observation system plan(GEOSS) and the global environment and safety monitoring project(GMES) have been proposed. Because the new mathematics research results have been introduced in data assimilation algorithm, marked as the key bridge connecting the observational data and model simulation forecast of data assimilation algorithm has got rapid development.In recent year, widely attentions have been paid to ensemble-based data assimilation methods and application researches have been carried out to test in the operational data assimilation systems in order to replace the variational data assimilation systems. Ensemble Kalman filter(En KF) methods depend highly on the sizes of the ensemble. If ensemble numbers are too small, they will bring the related issues such as undersampling, covariance underestimation, filter divergence and distanced spurious correlations, so it is a suboptimal filter. Local technology can effectively solve the related problems in the small ensembles circumstances. On the basis of the Lorenz-96 model, this thesis study the differences of data assimilation with or without localization and discuss the advantages of local analysis under the condition of small ensemble. We developed a method based on power spectral density(PSD) to judge the effect of ensemble data assimilation. The results show: with a finite ensemble of numbers, Kalman gain values and PSD can be used to evaluate the assimilation effect combined with the local technology. The main work of this thesis summarized as follows:(1) Under the finite ensemble number, the kalman gain value and PSD can evaluate the efficacy of the assimilation, combined with local technology, assimilation algorithm can obtain more efficient.(2) Local technology can not only eliminate the spurious correlation of background error covariance matrix, can also increase the background error covariance matrix rank.(3) The covariance local method on update set mean and disturbance has stronger robustness. Research conclusion helps careful analysis and estimation of background error covariance.Through a series of numerical experiments for the parameter sensitivity experiment of CL and LA two local methods, that observe the different observation error and ensemble numbers for the influence of assimilation effect. We conclude that the local analysis thoughts can effectively solve the problem of false correlation, in order to achieve the business of ensemble data assimilation provides a valuable reference.