Application Of Robust Filtering Methods In Data Assimilation Systems

Data assimilation is a methodology commonly used in the fields of atmosphere and geography,the aim is to adjust the trajectory of the model by integrating the observation information from different sources and different resolutions to complete the prediction of the current state information.The robust filtering theory mainly emphasizes the robustness of the estimation,and it has a good tolerance to possible uncertainties under the condition that the growth rate of the estimation error is bounded.The main research content of this paper is to improve the estimation of observation error covariance by robust filtering algorithm,we develop a method for diagnosing and incorporating spatially correlated and time-dependent observation error in an ensemble data assimilation system.The method combines an ensemble robust filter with a method that uses statistical averages of background and analysis innovations to provide an estimate of the observation error covariance matrix.The effectiveness and assimilation effect of the new method are verified by simulation experiments.The main contents of this paper are as follows:(1)Kalman filtering and robust filtering theory.The ensemble transform Kalman filter algorithm is described in detail,and its actual running steps are listed.the actual running steps are listed.To solve the external uncertainty problem of filtering algorithm,the H_? norm is introduced,In order to make the algorithm effective for sequential data assimilation in full window,the idea of ensemble is applied to the time-local H_? filter,so that the system can be applied to the complex nonlinear system,and the accuracy and robustness of the filtering can be improved.(2)The source of observation error and the estimation technique of observation error covariance are mainly introduced.The source of the representative error in the observation error and its related definition are described in detail,the representative error is correlated,and the representation of correlation is explained.The estimation technique of observation error covariance is introduced to realize the combination of observation error covariance technique and robust filtering algorithm.The observation error covariance estimation technique is used to obtain the time-varying and state-dependent observation error covariance.(3)Based on the ensemble time-local H_?? filter,three specific forms of the ensemble time-local H_? filter are discussed by constructing the covariance amplification relationship from different aspects.The observation error covariance estimation technique is combined with three filtering algorithms respectively.we use nonlinear model Lorenz-96 to assess the effectiveness of robust filter with estimation of observation error covariance,it can be show that the estimation of observation error could improve the accuracy of state estimation and the robust filter with observation error estimation have more the robustness.In a word,when the observed error covariance is unknown in the data assimilation system,the observational error estimation technique can be considered.Applying the observation error estimation technique to the robust filtering method can improve the filtering accuracy and robustness and make the observation information be used effectively.