The Study Of Adaptive Localization And Adaptive Covariance Inflation For WRF-EnSRF Assimilation System

Ensemble square root filter (EnSRF) as the deterministic ensemble-based data assimilation technique, is being used for data assimilation in a growing number of research and applications. Ensemble methods, compared with variational methods, require little expert knowledge for the development of tangent linear and adjoint versions of models and forward observation operators, and the background-error convariances are flow dependent. EnSRF based on the traditional ensemble kalman filter update equation ameliorate the impacts of sampling errors introduced by adding random perturbations to the observations. And the computational cost of the method relatively is less than the cost of other deterministic ensemble methods. Others, EnSRF is easy to code and ease of implementation. According to operational needs, Wang Shizhang have constructed a WRF-EnSRF system for storm scale assimilation.The study is to develop the key assimilation techniques of WRF-EnSRF system, have introduced adaptive localization and adaptive covariance inflation error correction algorithms to help filters to tolerate errors from many sources, including sampling errors, model errors and fundamental inconsistencies between the filter assumptions and reality, which lead to insufficient variance in ensemble state estimates. During the ideal storm tests, the results show the characteristics and the good performance of adaptive algorithms developed here in storm scale assimilation, and the better assimilation scheme is obtained. During the real tests of assimilating Doppler radar data, adaptive localization and adaptive covariance inflation introduced here are demonstrated in reality taking into consideration of many complex influence factors. And the main conclusions are as follows:1. Adaptive localization (hierarchical ensemble filter technique) running for short training periods, especially a rapidly development stage of the storm, to develop localization statistics can be used in WRF-EnSRF system to produce high quality assimilations at reasonable cost.2. In ideal tests, where the dominant error source is small ensemble sampling error in model error absence, the spatially and temporally varing adaptive covariance inflation algorithm based on a hierarchical Bayesian approach produces filter performances that are comparable with the best tuned relax inflation values. In real tests, where the model error is dominant, the algorithm produces better results than any other inflation methods, even if whose inflation values are the best tuned, including the relax covariance inflation method, especially while making a diagnostic analysis in Doppler radial velocity.3. Whether to ignore model error or not, using common empirical localization while radial velocity data are assimilating, using adaptive localization (hierarchical ensemble filter technique) while radar reflectivity data are assimilating, using spatially and temporally varying adaptive covariance inflation for alleviating impacts of errors from all sources, this assimilation scheme could be the best in considering of assimilation performance combined with cost.