Comparative Study On The Impact Of Three Initial Perturbation Schemes On The Ensemble Forecasting Skill Of Cold Vortex Rainstorm

As the hotspot of current research of numerical weather prediction,the severe weather ensemble forecast is of great significance to the protection of people's lives and property.In this paper,ensemble forecasting experiments are carried out on a cold vortex rainstorm based on traditional BGM and ETKF,respectively.The performance of the initial perturbation method in regional ensemble forecast is studied.Then based on the local differences in the development process of the weather system,the Local-BGM method is introduced to generate the initial perturbation and the ensemble forecasting experiment is conducted for comparison.At the same time,the temporal neighborhood is considered in the spatial neighborhood probability method to evaluate the precipitation forecast effect.The conclusions are as followings.There is 24-30 h for initial perturbation generated by the BGM method to reach saturation and the perturbation is more stable in the lower level and middle level.For the perturbation variables,the BGM forecast is better than the control forecast,especially in the horizontal zonal wind and the meridional wind.For the intensity and the position of precipitation,the forecast accuracy of BGM method is higher than that of control forecast,which significantly improves the forecast skill.The ensemble mean forecast can effectively consider the precipitation of each member,so the precipitation position is closer to the observation.Furthermore,the magnitude of heavy precipitation is more reasonable than the control forecast.The results of ETKF method are similar with that of the BGM method.The forecast of perturbation variables is better than the control forecast.It is more stable than the BGM method in the later forecasting period.But the improvement of some precipitation magnitude is not as obvious as the BGM method.Considering the local characteristics of the weather system,the Local-BGM method and the local radius are introduced.The growth of perturbation will be rescaled through calculating the ratio of the forecast RMSE of grids within the local radius to that of the whole grids in the calculation domain at the end of breeding period.Compared with the traditional BGM method,the Local-BGM method can also adjust perturbation's size so that it is consistent with the initial perturbation in the sense of the RMSE.The results show that it can be observed obvious local information for the perturbation generated by the local-BGM method.In addition,the Local-BGM method can reduce the forecast RMSE and improve the ensemble spread.Based on the precipitation forecast results of the two methods,it can be concluded that the Local-BGM method has better forecast skill in each precipitation magnitude and the position forecast of heavy rainfall is closer to the observation than the traditional BGM method.The neighborhood probability method can combine the precipitation information of all the grids in the neighborhood to improve the spatial uncertainty caused by the position forecast error.Based on all ensemble members,the forecast of neighborhood ensemble probability(NEP)has the highest FSS score,especially better than the control forecast and ensemble mean(EM)method in heavy rainfall.After introducing the time factor into the NP method,the probabilistic forecast is evaluated on different spatiotemporal neighborhoods.The FSS scores of the control forecast,EM and NEP methods are higher than those only considering the spatial neighborhood.Meanwhile the improvement is more obvious with the precipitation magnitude increasing.The improved NP method can provide more reasonable probabilistic forecast results.