To obtain the optimal weights for combining different model outputs in operational seasonal forecasts, Bayesian model averaging (BMA) was applied to the multimodel hindcast datasets at the Asia–Pacific Economic Cooperation (APEC) Climate Center (APCC). The weights were estimated according to the performance of individual members in simulating the given training data. Verification measurements such as the mean squared skill score (MSSS) and anomaly pattern correlation coefficient (ACC) were used to evaluate the forecast performance based on the observation. In terms of the MSSS, over the tropics, the BMA methods generally have broader areas with higher skills than the equal weight methods, whereas, over the global domain, the equal weight methods generally have larger areas with positive skills than the BMA methods. In terms of the Pearson ACC, the equal weight methods generally have higher skills than the BMA methods, whereas the BMA methods generally have higher skills than the equal weight methods in terms of the robust ACC.

1. INTRODUCTION

2. MODEL AND DATA

3. EXPERIMENTAL DESIGN

4. APPLICATION OF BMA

5. RESULT

5.1 Temperature

5.2 Precipitation

6. CONCLUSION

7. DISCUSSION