摘要：The track forecasts of five ensemble prediction systems (JMA-WEPS, CMA-GEFS, ECMWF-EPS, NCEP-GEFS and MSC-CENS, respectively) in 2014 are evaluated in this paper. First, on the assumption of bivariate normal distribution, a probability ellipse based on ensemble spread is proposed to represent forecast uncertainties, and then the ellipse and ensemble mean are used to evaluate the five different systems. It is found that the probability ellipse of ECMWF-EPS is smaller, with a high hit ratio, and its mean track errors are just a little larger than those of NCEP-GEFS within 48 h, meaning overall it outperforms the other four systems. Meanwhile, the performance of CMA-GEFS is found to be the poorest.
