Multi-model Ensemble Forecast Technology with Analysis and Verification of the Results
Precipitation is an important weather process,it often causes great casualties and economic losses each year.The accurate forecast is not only a scientific problem.The technique of ensemble forecast is a kind of newly developed numerical prediction method.Through the use of the weather forecast operation of National Meteorological Center ( NMC),numerical products of forecast models which are used in Germany,Japan and Europe as well as T213 Model are tested.Based on the analysis of different models' forecast capability to the target areas,a collective forecast method is established in four models by the analysis technology of nerve-cell network and objective multi-models( ME)weights coefficient based on Ts grades,and the method is applied in 2005 flood season operation.It is found that ME is better than the method of simple ensemble average in the short-term forecast,therefore it has a potential capability in weather forecast,and may be used in the future. The main results include the following three parts,errors evaluation of different NMC models,results comparison of several economic ensemble and primary analysis of multi-model ensemble in Chinese NMC operational running systems in 2005.Ts value of Japanese model is 0.26 for every 6-hour light rain forecast from 6 t0 72 hours in the operational system of National Meteorological Center,and it' s a little bit better than that of the Germany model and T213.Along with the forecast period extension,Ts value of each model reduces gradually,this phenomenon is in particular obvious for the moderate rain and heavy rain.The mean model Ts is only 0.04 with 15.0 mm precipitation per day.Also,it' s found that the forecast precipitation information could not be used from 36 hours and on.Verified results show that the basic elements of every model outputs have their own characteristics respectively,and may be complementary reciprocally. The structure of intelligent decision support system is simply analyzed,and the method of knowledge representation and the inference based on neural network are introduced.Neural cell network method may supply the symbolic method in intelligent spatial decision support system.It reveals that analysis of neural cell network can be used to offer valid precipitation as a useful tool,but forecast quality is correspond with simple poor man ensemble,therefore the forecast ability of neural cell network exhibits limitation,and it is only a choice.By contrast and checkout,four models Ts values are analyzed and ME method with an obvious improvement for light rain prediction in 3 forecast target regions is found,at the same time,forecast qualities of temperature fields,height fields and wind fields have specific usefulness in 6-72 hour stages.Multi-model ensemble forecast needs new high quality operational model to be used in NMC prediction systems,and improve actual forecast ability.
NETL
NSTL
万方数据
