The control regimes in Figure 1 are encouragingly realistic, with Regimes 1 and 3 corresponding approximately to the two phases of the North Atlantic Oscillation (NAO), characterized by a cyclone or anticyclone over Greenland respectively (cf. Cheng and Wallace, 1993). Pattern correlations with the five regimes derived from an identical bump-hunting analysis of 44 winters (1951-95) of NCEP observed daily 700-mb heights are 0.72, 0.89, 0.91, 0.77, 0.75, and 0.37 for each regime in Figure 1 respectively. Since SSTs in the control experiment are prescribed to follow a seasonally varying climatology, this is circumstantial evidence that regime patterns are insensitive to SST variations, both local and remote.

Figure 1. Weather regimes of the control experiment, derived using KG's bump-hunting method. Each map is a composite of low-pass filtered 700-mb height anomalies (gpm) on days belonging to that regime; controur interval is 10 gpm. The regimes are ranked according to the number of days belonging to each regime, given in brackets. Pattern correlations with regimes derived using K-means method are 0.82, 0.89, 0.99, 0.95, 0.91, and 0.93 respectively.

The frequencies of occurrence of the six regimes in the control run are similar, on average, so that there are no strongly preferred regimes. Decade-to-decade fluctuations of regime frequency, however, are found to be substantial in the control run, with interdecadal standard deviations of up to one-third of the centennial mean. We projected the 882 low-pass filtered daily maps from the anomaly experiment onto the four leading EOFs of the control experiment, and counted the number of days falling into each of the control regimes of Figure 1. These counts are plotted in Figure 2 (diamonds); the analogous counts for each of the 10 control decades are plotted in terms of their mean (open circles) and standard deviation (shown as a two-sided error bar).

Figure 2. Frequency of occurrence of a daily map within each of the six control clusters in the SST anomaly experiment (solid diamonds). Open circles and error bars show the mean frequency of occurrence and one standard eviation on either side for the six regimes in the ten non-overlapping 882 day control decades.

Three of the six regimes show excursions in frequency for the SST anomaly experiment of about one standard deviation. The frequency anomalies of these three regimes tend to produce negative height anomalies north of 60⁰N and positive ones to the south, resembling the positive phase of the NAO. Our results reveal, however, that the NAO is made up of two or three distinct weather regimes, rather than behaving like a linear mode that is preferentially excited in the presence of a warm SST anomaly over the North Atlantic. The frequency of occurrence of certain weather regimes is thus modulated by the presence of the North Atlantic SST anomaly. A longer anomaly experiment will be required to obtain greater statistical confidence, given the high interdecadal variability exhibited by the control run.

The GCM's response to the prescribed SST anomaly is reconstructed in Figure 3a as a weighted average of the six weather regimes identified in the control experiment (Figure 1). The weights are proportional to the changes in regime frequency in the decade-long anomaly experiment, each scaled by the corresponding frequency in the century-long control experiment (Figure 2). The NAO is much more pronounced in this regime-average response than in the simple difference between time means of the anomaly and control integrations; the latter is shown in Figure 3b. The signal-to-noise ratio in climate-change experiments may thus be enhanced by concentrating attention on those regions of the model's phase space that are associated with more regular, recurrent and persistent behavior.

Figure 3. (a) Weather-regime response to SST anomaly, constructed by weighting the six maps in Figure 1 by the frequency ratios between anomaly and control experiment (diamonds and cricles in Figure 2, respectively). (b) 700-mb DJF mean response (mean of 10-winter SST anomaly experiment, minus the mean of the 99-winter control experiment). Contour interval is 2.5 gpm. Stippled region in (b) indicates areas significant at 90% level using a two-tailed Student t-test.

Return to beginning of page

Acknowledgments

We would like to thank E. Roeckner for providing the control integration, and Y. Kushnir for the SST anomaly data. This work was supported by NSF Grant ATM95-23787, NASA Grant NAG 5-713, and the University of California's Campus Laboratory Collaboration (CLC) program. The GCM integrations were performed at the Deutsches Klimarechenzentrum in Hamburg.