Multi-model Ensemble Forecasts Of The Extended Range Using The TIGGE Dataset

Based on the10-15days extended range ensemble forecasts of CMC, ECMWF, UKMO and NCEP in the TIGGE datasets, the2m surface air temperature and24hour accumulated precipitation forecast results were evaluated. Main evaluation methods include Talagrand distribution, ensemble spread, root mean square error, correlation coefficient and ETS score. Firstly, the Talagrand distribution, ensemble spread and root mean square error were evaluated for all members of each single model. Secondly, the multi-model ensemble forecast were conducted using three ensemble methods including EMN, BREM and SUP. In view of the big error of extended range prediction, it hasn’t been applied in business forecast so far, in order to filter out the shortwave disturbance and further improve the prediction skills, the moving average method was used both in forecasting data and observed data so as to research the change trend of2m surface air temperature and24hour accumulated precipitation of several days average in the extended range.In addition, this paper also attempted to evaluate the process of the low temperature, snow and freezing weather events in early2008in the south area of China based on National Meteorological Center T213global forecast model. Assessment elements include:500hPa height field, the surface-2m air temperature and the24hour accumulated precipitation for forecasting time of10-15days. Some typical weather processes were chosen for each parameter:a strong blocking high process was selected to research the forecasting effect of500hPa height field. As for surface2m air temperature, we choose the process in which there is severe decrease for temperature. Similarly, the3rd weather process in which there is a big amount of precipitation is researched.Results show that the Talagrand distribution of surface air temperature presents "U" type, the ensemble spread is generally lower than RMS error. Meanwhile, the Talagrand distribution of24hour accumulated precipitation shows "L" type roughly because of high alarm rate. Similar to the temperature forecasting, ensemble spread of precipitation is also less than RMS error. For temperature forecast, ECMWF is the best model, NCEP is better and UKMO is the worst. For precipitation forecast, CMC shows the best skill and UKMO is still the worst. After multi-model ensemble, results show that multi-model ensemble can effectively reduce forecasting error. EMN is less effective than BREM and SUP is the best method. After moving average, the prediction skill is better than the daily forecast effect and the longer of the moving step length, the smaller of the prediction error. It indicates that in the extended range forecasting, evaluating the average trend of weather elements is of great significance.For the extended range forecasting of continuous and abnormal meteorology event in early2008, results show that T213model has some skill for blocking high, but the intensity of blocking high center is weaker than observed field. Also, the position of blocking center, the starting time and ending time of blocking high are all different with the observed field. Model has predicted the process of air temperature decreasing, but the cooling rate is weaker. ETS of precipitation prediction in the extended range is lower and the RMS error is larger.