A Comparative Study Of Two Blending Methods To Introduce Large Scale Information Into GRAPES Mesoscale Analysis

The Synoptic situation tends to drift away due to the accumulation of mode errors after a period of continuous cycle data assimilation,and the description of large-scale information appears to be distorted.Considering every moment movements of the atmosphere are multi-scale,and the observations which sample and record these movements are multi-scale too.On the one hand,the small-scale features are well presented in GRAPES_Meso(Global/Regional Assimilation and Prediction System in Mesoscale)3D-Var analysis,taking the benefit of a relatively higher resolution.On the other hand,the description of planetary and synoptic scales is not as desirable as the small scales in GRAPES_Meso.In addition,observations outside the LAM domain cannot be taken into account,leading to a crippled performance in large scale analysis.Therefore,the deficiency of a proper description of large scale information potentially undermines the forecast once encountered with a mesoscale calamitous weather where a large scale manifold is in dominancy.However,global model shows its superiority on large scale analysis due to a coarser resolution and abundant observations with global coverage.In general,the lateral boundary conditions(LBCs)of a regional model are usually refreshed from the global model forecast.Thus certain amount of large scale information can be found in the LBCs.However,previous study has found that proper initial conditions(ICs)with detailed multi-scale information are more essential for improvements of the subsequent forecasts,in terms of precipitation for instance.Thus to obtain a proper description of large-scale information for GRAPES_Meso in the continuous cycle data assimilation,in this study,two blending methods,explicit spectral blending method and constraining blending,are implemented and tested on GRAPES_Meso.In explicit spectral blending method,analyses from T639 and GRAPES_Meso are merged in accordance with their weights defined by a Raymond filter,while the variational blending adds T639 global analysis as a weak constraint to the cost function,which minimizes the departure of the 3D-Var analysis from global analysis.Assimilation with conventional observations available and a 36h forecast are performed and verified though an extreme precipitation event.It is found that both methods are effective in remedying the deficiency of large scales as well as reserving the well-featured mesoscale information in the LAM,leading to an improvement in analysis balance and accuracy.Subjective verification reveals that the large scale blending has a visible beneficial impact on the forecast of precipitation location and intense,where variational blending method is more skillfull and desirable.Methodologies employed in this paper can be served as a reference for the continuous efforts on the development of multi-scale analysis in GRAPES_Meso.