The Research Of Combined Bias Correction And Blending Scales Initial Perturbation For REPS

Ensemble forecasts represent forecast error through dynamic methods, and can estimate the uncertainty of real atmosphere reasonably. They have solved the problem of forecast restriction which is because of model’s nonlinearity of and atmosphere’s chaos. The regional ensemble forecasts (REPS) increase the resolution and many other aspects compared with the global ensemble forecasts, which helps estimate the uncertainty of micro/meso-scale atmospheric motion, improve the forecast accuracy of micro/meso-scale atmosphere. Bias between forecasts and observations and small spread, which are caused by ensemble initial perturbation scheme, the method of boundary perturbation, model error and analysis error, should be properly handled before application. Simultaneously, it is difficulte to represent the multi-scale uncertainty of real atmosphere with low forecast quality of large scale information from initial conditions of REPS. Also, discontinuity on the LBCs could be caused by downscaling which is the traditional way of creating LBCs of REPS and one of the main reasons of reducing forecast quality of ensemble forecasts.To improve the forecast quality of RPES, this thesis stydies on the combined bias correction scheme based on the first and second moment bias correction and properly analyzes the results with data from B08RDP and the optimal weight coefficients from sensitivity test of the first and second moment bias correction. This thesis also focuses on the configuration of multi-scales initial perturbation combination of global and regional ensemble forecasts with barnes filter. The Blending scales initial perturbation can represent multi-scales uncertainty of real atmosphere. The main conclusions are as follows:(1) NCEP has the optimal reliability and resolution with smallest forecast error of ensemble mean and properest spread. While CAMS performs the worst so as to its excessive forecast error of ensemble mean which might caused by imperfectness of the model and small members of the ensembles.(2) The combined bias correction can improve the forecast quality of ensemble forecasts with good capability in operational application, which means smaller forecast error and shorter distance between ensemble spread and RMSE of ensemble mean. The effect of the combined bias correction is related to the quality of ensemble forecasts and variables.(3) The blending scales initial perturbation combines micro/meso-scale perturbation with Barnes filter and interpolation. Barnes filter can extract macro, micro and meso-scale information, with retainment of each scales’ distribution. The REPS with blending scales initial perturbation performs better than CAMS after the combined bias correction, showing blending scales’ potential of improving ensemble forecasts.