Superensemble Forecast Experiment Of Low Temperature, Freezing Rain And Snow Process During Early2008in Hunan Province

Based on the ensemble forecast data of the United States National Centers for Environmental Prediction (NCEP), the European Centre for Medium-Range Weather Forecasts (ECMWF), the Japan Meteorological Agency (JMA) and the China Meteorological Administration (CMA) taken from the TIGGE archives, the24-168h multiomodel superensemble forecast experiments of the surface temperature,500hPa geopotential height, the low-level temperature inversion, the southwesterly jet over Hunan province have been conducted during the early2008when severe low temperature, freezing rain and snow events occurred in southern China. The forecast skills of each individual model of four forecast centers and the superensemble were evaluated. The consensus scheme includes the ensemble mean, the superensemble with fixed training period as well as the sliding training period.The results show that the forecast error of the ECMWF is the least among four model systems, the JMA and NCEP have larger forecast errors than the ECMWF, while the CMA has the largest forecast error. During the freezing rain and snow period, the performance of each individual model forecast is not satisfactory, and the forecast skill is lower than that for the2007-2008winter mean. The superensemble with sliding training period reduced the forecast errors significantly. For all forecast lead time, the forecast skill of the superensemble is superior to that of each individual model. As the forecast lead time becomes longer, the improvement of the forecast skill is more significant. In addition, the superensemble forecast with sliding training period is most stable among all individual models and multimodel consensus schemes, the day-to-day variation of the forecast error is small. In Hunan province, the optimal length of the sliding training period is about50days for the24-72h forecast, and more than2months for96-168h forecast.For the500hPa circulation, the forecast performance of the ECMWF is the best, and the CMA forecast skill is the least among four ensemble forecast systems, while the performance of the NCEP is better than that of JMA. The performance of24h forecasts of all models is the best, and the bias is quite small compared with the observed data. As the forecast lead time becomes longer, the forecast error of the intensity and the position of trough and ridge becomes larger as well. The168h forecast error is so large that the forecast is not useful. The superensemble forecast with sliding training period is superior to the forecast of the best individual model, and its24-96h forecast has very small forecast error compared with the observed data, and the120-168h forecast error is quite small as well. However, the intensity of the trough and ridge in the superensemble forecast is weaker than that of the observed one, the circulation center and the position of the trough and ridge in the superensemble is reasonable compared with the observations.For the temperature inversion foreast between700hPa and850hPa, all the individual models can predict the existence, weakening and disappearance of the temperature inversion. However, the forecasts of the intensity and the position of the temperature inversion have bias compared with the observations. As the forecast lead time becomes longer, the forecast error becomes larger as well. The Superensemble forecast of the temperature inversion has no advantage against the individual model forecast for shorter forecast lead time, but it is significantly superior to the individual model forecast for longer forecast lead time. For the temperature difference between700hPa and850hPa in Hunan, the forecast error of JMA is the least among the four model systems, but the forecast error of the superensemble with sliding training period is smaller than the best individual model.For the700hPa southwesterly jet forecast, the ECMWF24-96h forecast, the JMA24-48h forecast has quite good performance compared with the observation. The superensemble forecast with sliding training period has very good performance, and can predict the southwesterly jet reasonably for24-168h forecast lead time. For the wind field of700hPa in Hunan, the forecast error of ECMWF is the least among four model systems, the forecast error of the superensemble with sliding training period is smaller than the best individual model for most forecast lead time.