Optimal multi-model ensemble method in seasonal climate prediction

Given a large number of dynamical model predictions, this study endeavors to improve seasonal climate prediction through optimizing multi-model ensemble (MME) method. We have developed a new MME method and evaluated it using 15 dynamical models' retrospective forecasts for the period 1981-2001 in comparison with other MME methods. The strengths of the new method lie in a statistical error correction procedure for predictions of individual model and a discreet selection procedure of reliable predictors among all possible candidates. The conspicuous improvement of the new method is achieved against other MME methods over the regions in which the average of individual models' skill is very poor, such as land area and extratropical oceans. It is demonstrated that the selection procedure for the reliable predictors allows the present method to get more effective as the number of model predictions being used increases.