Three components of predictability are examined. 1) Gridpoint analysis of intra-ensemble variance indicates that there are two regions of extended predictability in the North Pacific during winter. These include a region centered on the Kuroshio Extension and a swath reaching from the Philipines to the west coast of the United States. In these regions for up to 15 years the spread of ensemble members is less than that expected from natural variability. By contrast, some regions of the North Atlantic and Southern Ocean have predictability beyond 20 years. 2) Analysis of the detrended mean ensemble state shows a smaller contribution to predictability. Initial departures from climatology that are substantially greater than that expected from natural fluctuations are detectable with confidence for about five years in the North Pacific; they last even longer in the North Atlantic. 3) A third predictable component is that produced by growing greenhouse gas concentrations in the experiment. As measured by ensemble average trends, in the North Atlantic these become detectable above the background fluctuations after 10 to 15 years.
Complex EOF analysis indicates that a substantial component of decadal variability in North Pacific SST in this experiment, as well as in the control, is associated with a pattern similar to the Pacific Decadal Oscillation observed in nature. In this model this mode of variability appears to be largely a result of midlatitude atmosphere-ocean coupled processes. Our analysis indicates it is responsible for the enhanced predictability noted in the Kuroshio Extension region. Interestingly, there are suggestions in the ensemble experiment that this mode may be predictable for several decades though we cannot be certain of this. Though there are substantial variations from one ensemble member to another, in the ensemble mean its amplitude reaches peaks near years 2021 and 2043 that are substantially larger than would be expected from natural fluctuations of the model. Even more intriguing is the fact that wavelet analysis shows a very prominent 20 year component to the variability of this mode in many ensemble members, suggesting that its period, if not its phasing, may be especially predictable.
Supplementary URL: