Abstract: The skill of a set of control and ensemble forecasts of Western Pacific Subtropical High was evaluated based on the 500 hPa geopotential height information from the THORPEX Interactive Grand Global Ensemble (TIGGE) datasets, which consist of model outputs from CMA, JMA, ECMWF, UKMO and NCEP. Three methods were adopted, i.e., Ensemble Mean (EMN), Bias-Removed Ensemble Mean (BREM) and running Training Period Superensemble (R_SUP), to integrate the data from different sources, and the metrics for performance evaluation include Talagrand distribution, correlation coefficient, Root Mean Square Error (RMSE), and Brier Skill Score (BSS). A comparison of the outputs of these models shows significant variation in forecast performance. The results indicate that the UKMO model has the best forecast skill for the 500 hPa geopotential height among all control forecasts, while the ECMWF model ranks on the top of all ensemble forecasts. From the improvement of RMSE, both BREM and R_SUP can significantly reduce the RMSE of the integrated forecast results compared to the original control forecasts in TIGGE, but EMN does not show similar improvement. However, none of the three integration methods shows discernable improvement of ensemble forecast of the 500 hPa geopotential height, with all having less skills than ECMWF single model ensemble forecast.

Key words: TIGGE, WPSH, multi-model ensemble, forecast