Error Processing Method Research Of Data Assimilation System

With the continuous improvement of the level of human cognition to the environment, more and more earth observation data are analyzed and utilized. Data assimilation methods, as a powerful link between the information of the prediction modelsand observation information, has been developing rapidly in the field of Earth Sciences. The current research either in theory or the establishment of assimilation application system has made important progress. However, the error problems of data assimilation system have been the bottleneck that restricts the development of data assimilation system, how to reduce the uncertainty of data assimilation system reasonably and effectively has been the urgent problems that we have to face.In actual assimilation system, if the errors have not been deal with or process well, it may lead to forcast information heavily deviate from observation information, then makes the filter divergence and makes assimilation system instable. In order to solve the error problem of data assimilation system, we mainly do the following research works:(1) We have compared the error mechanism and source of error in sequential data assimilation method and continuous data assimilation method. A brief overview the estimation process of the observational error and model error, summarized the error processing method of data assimilation. A brief list kinds of representative error processing method were reviewed.(2) The error generating process of Kalman filter is analyzed, for the problem that Kalman filter method is not suitable for nonlinear system and easy divergence, an improved Kalman algorithm was proposed based on Cholesky decomposition error covariance matrix.The performance of improved filter is studied based on the Lorenz chaotic system. The results show that the improved algorithm has a significant improvement effect on the error of data assimilation system compared with Kalman filter.(3) Aiming at the problem of observation error, a pollutant concentration prediction data assimilation simulation system is build based on diffusion of atmospheric pollutants, observation strategies and effects of obsevation errors on assimilation system are studied on this simulation system, local method is used to deal with the measurement errors in the false related issues.Algorithm error and observation error of assimilation system is studied on basis of perfect model in this thesis, attempt to seek a effective error processing method at theoretical point,hope to provide theoretical reference for practical data assimilation system application.