The Research And Comparision Of Filter And Assimilation Methods For Constructing Multi-scale Initial Perturbations In REPS

The regional ensemble forecast has been one of the ways to improve the prediction of high-impact weather and one of the key factors to determine the quality of regional ensemble forecasts is whether the initial perturbation of ensemble prediction can precisely reflect the structure characteristics of forecast errors, and therefore constructing reasonable initial perturbations is crucial to improve the quality of the regional ensemble forecast. So far, the initial perturbations provided by all kinds of schemes have been single-scale and can’t reflect the true state of atmosphere fully and accurately. In order to construct more reasonable initial perturbations, a scheme to produce multi-scale initial perturbation (Filter scheme) is designed based on the filter method in the paper, and ensemble forecasting experiments based on GRAPES regional model are carried out to verify the performance of the scheme. Considering that it is difficult to choose the filter scale during the filtering process for the Filter scheme and perturbation errors are easily caused during blending different scale information, we also develop a new scheme to produce multi-scale initial perturbations based on the assimilation method (VAR scheme) and the new scheme’s performance are also verified by a set of experiments. Various of aspects of VAR scheme and Filter scheme are further compared and analyzed by a set of experiments with longer time.The main conclusions are obtained as follows:(1) The Filter scheme can blend the large scale of high quality from the global ensemble forecasts and the meso-scale/small scale from the regional ensemble forecasts completely according to the filter scale. The ensemble mean of the scheme is superior to that of the original REPS and control forecast. Its probabilistic forecasts for reaching the level of heavy rain and torrential rain and the stamp picture is more valuable than those of the original REPS. The scheme effectively improves the accuracy of forecasting the temperature and geopotential height, but has slight improvement for wind forecasts.(2) The VAR scheme can blend the large scale of high quality from the global ensemble forecasts and the mesoscale/small scale from the regional ensemble forecasts effectively but not completely. The ensemble mean of the scheme, probabilistic forecasts for reaching the level of heavy rain and torrential rain and the stamp picture are superior to those of the original REPS. The scheme improves the accuracy of forecasting the temperature and geopotential height significantly, but has little positive effect on wind forecasts.(3) The performances of two schemes are proved to be superior to the original REPS and they have different advantages. The VAR scheme has more significant improvement than Filter scheme in RMSEs of ensemble mean, but for ensemble spread Filter scheme has more significant improvement than VAR scheme during the forecast hours in 0-12 h and they have same effects in subsequent forecast hours. Overall, the VAR scheme has more positive effect on geopotential height and zonal wind forecasts than Filter scheme, the Filter scheme has more positive effect on temperature forecasts than VAR scheme and two schemes have same improvement in forecasting meridional winds. The VAR scheme and Filter scheme have no effect on forecasting light rain and moderate rain but improve the capacity of forecasting heavy rain and rainstorm. The VAR scheme has better forecasting capacity of rainstorm than Filter scheme and Filter scheme has better forecasting capacity of heavy rain than VAR scheme.