Cluster Analysis of the Seasonal Forecast Skill of the NCEP SFM over the Pacific–North America Sector

Abstract

Abstract Cluster analysis was used to study seasonal forecast skills of the winter season NCEP seasonal forecast model (SFM) hindcasts over the Pacific–North America (PNA) sector. Two skill scores based on cluster mean and ensemble mean are compared. It was shown that the anomaly correlation coefficients (ACCs) of cluster mean are generally higher than those of the simple ensemble mean. The results indicated that the skill was affected by the existence of multiple atmospheric regimes. Multiple regimes tend to appear more often in near-normal tropical Pacific sea surface temperature (SST) episodes, while a single regime tends to appear during warm/cold episodes. The dissimilarity among the cluster members is small and the number of the dominant cluster members is also small when the tropical SST anomaly is large, suggesting that the external forcing reduces the frequency of occurrence of the multiple regimes. The ACC improvements from the ensemble mean ACCs to the cluster mean ACCs are statistically significant. Thus, the cluster mean can be used as a supplementary tool for seasonal forecasting.