Study On Optimization Approach Of GRAPES-MESO En-3DVAR Hybird Data Assimilation System

The ensemble-based variational hybrid data assimilation method has been widely studied and applied in recent ten years,the principle of hybrid data assimilation is to combine the static background error covariance used in traditional variational assimilation with the background error covariance which generate from the ensemble forecast,which makes the ensemble-based variational hybrid data assimilation not only have the flow-dependent properties of the background error covariance,but also take advantage of variational asismilation in dealing with numerous irregular observational data.In the hybrid data assimilation system,the rationality of background error covariance(BEC)which derived by the ensemble forecast and the localization scale of background error covariance are very important,which determine the quality of the data assimilation analysis field.Based on the GRAPES-MESO En-3DVAR hybrid data assimilation system(GRAPES-MESO En-3DVAR),this paper studies optimization methods for several outstanding problems in this system,respectively: a “Topographic Dependent Horizontal Localization Scale” scheme was designed,which improved the deficiencies of the localized scale of ensemble-based BEC of GRAPES-MESO En-3DVAR hybrid data assimilation system.Based on this,the improvement effect of typhoon forecast was discussed by using GRAPES-MESO En-3DVAR hybrid data assimilation system of“Topographic Dependent Horizontal Localization Scale” scheme.The “Unified Perturbation of Stochastic-physics and Bias-tendency” method was designed,the quality of ensemble perturbation was improved.At last,some ensemble perturbation schemes were conducted to obtain the ensemblebased BEC,and then applied them to evaluate the performance on the GRAPES-MESO En-3DVAR hybrid data assimilation system.The following main conclusions were:(1)The sensitivity test of the optimal horizontal localization scale of the ensemble-based BEC shows that,in the GRAPES-MESO En-3DVAR hybrid data assimilation system,the optimal horizontal localization scale of the plain area is about 1000 km and 1500 km in the Tibetan Plateau.The “Topographic Dependent Horizontal Localization Scale” scheme can effectively improve the quality of the GRAPES-MESO En-3DVAR hybrid assimilation system.Among them,the improvement effect of the analysis field is the most obvious,and with the increase of forecast time,the improvement decreased.(2)Based on the GRAPES-MESO En-3DVAR hybrid data assimilation system of the“Topographic Dependent Horizontal Localization Scale” scheme,the typhoon data assimilation experiment was carried out.The results show that the optimized GRAPES-MESO En-3DVAR hybrid assimilation system can effectively decrease the track error of typhoon,reduce the missed rate and improve the TS(Threatscore)score of the precipitation caused by typhoon.(3)There is a significant deviation in the temperature field of the GRAPES-REPS regional ensemble forecast system,specifically,below 200 h Pa height shows obvious warm deviation,and most areas above 200 h Pa show cold deviation.The ensemble forecast experiments show that the“Unified Perturbation of Stochastic-physics and Bias-tendency” can effectively reduce the systematic bias and random error of the model while improving the ensemble spread.Also,this scheme can effectively improve the ensemble probability forecasting skills,and improve the precipitation probability forecasting to a certain extent.It is worth pointing out that this scheme can effectively reduce the systematic deviation compared to the traditional mode post-processing scheme without significantly increasing the mode integration running time.(4)The GRAPES-MESO En-3DVAR hybrid assimilation system is more sensitive to ensemble prediction disturbance schemes.The results show that the ensemble-based BEC which generated by“Unified Perturbation of Stochastic-physics and Bias-tendency” can improve the quality of the hybrid data assimilation system.The improvement is obvious in the early stage of the forecast is obvious,and decreases against forecast time.In summary,based on the GRAPES-MESO En-3DVAR hybrid assimilation system,the“Topographic Dependent Horizontal Localization Scale” scheme was proposed to improve the quality of the analysis field and forecast field of the hybrid data assimilation system;the “Unified Perturbation of Stochastic-physics and Bias-tendency” was developed to improve the spread and perturbation structure of the ensemble forecast system,as a result,a better ensemble-based BEC is obtained.These methods improved the quality of GRAPES-MESO En-3DVAR hybrid data assimilation system,furthermore improved the quality of numerical prediction of China.