Extended-range Probabilistic 500hPa Geopotential Height Forecasting over Northern Hemisphere Using Bayesian Model Averaging

Abstract

ECMWF, NCEP and UKMO EPS data from the TIGGE datasets were used to conduct the Bayesian model averaging (BMA) extended-range (10-15 days) probabilistic forecasts for 500hPa geopotential heights over Northern Hemisphere. The results show that the multimodel BMA method performed better than raw ensemble and the forecast skill varied as season changes. The forecast skill was the best in summer and worst in winter. However, the skill score of BMA with respect to the raw ensemble indicated that the improvement of the forecast in winter was the highest. Hence, the BMA method can improve the worse raw ensemble to a larger extent. In addition, BMA provided a full probability distribution, which depicted the quantitative uncertainty of the forecasts. With the increase of latitude, the uncertainty of the forecast was increasing.