Study On Ensemble Update And Cycle Assimilation&Forecast Of Hybrid Ensemble-Variational Data Assimilation

The hybrid ensemble-variational data assimilation method combines the advantage of ensemble kalman filter which provides flow-dependent background error covariance and the advantage of variational method which simultaneously assimilates varieties of observations with model constraints.Because the flow-dependent background error covariance comes from ensemble perturbation,hybrid assimilation method can be easily affected by the quality of ensemble forecasts.In order to establish an efficient and robust cycle hybrid data assimilation and forecast system,research on the scheme of ensemble update is carried out in this study.Firstly,to study the impact of different ETKF(Ensemble Transform Kalman Filter)covariance inflation schemes on the analysis and forecast of hybrid data assimilation,five experiments are conducted over east China area from July 10,2011 to July 20,2011.The results show that all experiments with covariance inflation perform better than experiment without any covariance inflation.For wind,the WG07 scheme which uses projection factor in ensemble subspace and the WG03 scheme which uses innovation average show better results than the other two.For temperature,humidity and precipitation,the WG03 scheme and the BOWL scheme which uses the ratio of the spread between previous cycle and current cycle perform better than the others.Secondly,Sensitivity analysis shows that the ETKF scheme often crashes in cycle assimilation because of oscillation of the covariance inflation factors.The oscillation is caused by large variation of number and variety of observations in cycle assimilation.In order to design a simple and stable hybrid data assimilation system,a new method with physical control variable perturbation and multi-physical parameterization(RCV-Hybrid)is proposed.In RCV-Hybrid,the initial state of ensemble forecast come from analysis of hybrid data assimilation perturbed by physical control variables.Meanwhile,different microphysics and cumulus parameterization schemes are used in ensemble forecast to represent model error.Results of 10 days cycle assimilation and forecast show that,the RCV-Hybrid method performs better than 3DVar and is similar to ETKF.In addition,the RCV-Hybrid method can be easily operated without perturbing observation.Thus,it is a new option for hybrid data assimilation.Thirdly,in order to improve the performance of variational data assimilation without increasing computational cost,a time-lagged ensemble-variational assimilation(TLEn-Var)method is proposed,which introduces the flow-dependent time-lagged ensemble forecast error covariance into variational cost function.In single observation test,the characteristic of flow-dependence is showed by background error covariance derived from time-lagged ensemble.In cycle assimilation test,the analysis is improved by the TLEn-Var method compared with 3DVar.Because the time-lagged ensemble is based upon history forecasts,the TLEn-Var method proposed by this article is not only computationally economic,but also efficient in performance.Thus the time-lagged ensemble based rapid update hybrid assimilation method is convenient for operational use.