Comparison with the observations

A natural SST index to test the above theory is the cross Gulf Stream SST gradient $\Delta T$. The latter is constructed from Kaplan's analysis of the long historical record of SST [1997], using the northern and southern boxes indicated in Fig. 3. The unfiltered index timeseries is shown in Fig. 6 (upper panel, thin curve), superimposed on a low pass version of it (thick curve), from 1856 to 1992. Typical variations of $\Delta T$ are found to be of the order of $\simeq 1\; K$, i.e. about 10 % of the mean SST difference across the Gulf Stream.

If indeed the ocean circulation can significantly impact $\Delta T$, as predicted in the previous section, we should be able to see deviations from the simple exponential decay predicted by (1). To test this, Czaja [2000] have constructed composite maps of the time evolution of the large-scale SST pattern captured by $\Delta T$ (Fig. 6, bottom panel). In agreement with Fig. 2a, it is seen that cross Gulf Stream SST anomalies are associated with SST anomaly in the subtropical North Atlantic. The difference in SST between years when $\Delta T$is strongly positive and those when it is strongly negative, recovers indeed the tripole pattern discussed in section 2. The associated pattern in SLP (constructed from Kaplan's analysis of long historical record - see Kaplan [2000] - shown in Fig. 7) is also reminiscent of Fig. 1, but it is slightly shifted towards southwest. If the ocean has no other memory than that associated with the thermal inertia of its mixed layer, then according to (1) the tripole, once it has been generated, simply dies away after a couple of years. Contrasting sharply with this prediction, we see that the observed tripole reappears after 6 years, but with opposite sign. Thus, there is evidence for a complex time evolution of $\Delta T$ and its associated SST tripole, suggestive of a damped oscillatory behaviour. There is not only a fast local reponse to the local forcing orchestrated by the NAO, but a delayed response, which tends to reverse the sign of the SST anomaly after 6 years. It is speculated that this behaviour is the signature of ocean circulation impacting $\Delta T$ after several years.

Spectral analysis of the $\Delta T$ index is given in Fig. 5b. One sees a broad peak in the 10 - 20 yr band, with a decrease of power at longer timescales, as predicted from both the coupled and uncoupled models of the previous section (see Fig. 5a). The latter is interpreted as a signature of the role of the ocean circulation, which, on long timescales acts to damp $\Delta T$. The delay time required to fit the theory to the observed spectra is about 10 years, in broad agreement with the delays t_d^moc and t_d^ig discussed in section 3.

Whether or not the atmosphere is sensitive to these low frequency changes in SST is difficult to isolate based solely on the analysis of the observations. Nevertheless, it is reasonable to analyse the power spectra of atmospheric variables and see if they bear any resemblance to the predictions of Fig. 5a. Based on the SLP composite of Fig. 7 Czaja [2000] have computed the power spectrum of SLP anomalies in the region of the Greenland-Iceland Low (hereafter GIL). As seen in Fig. 5b, it has a similar structure to that of $\Delta T$. Although not a proof that it is the ocean circulation, via its effect on $\Delta T$, which modulates the strength of the Greenland-Iceland Low on decadal timescale, the comparison of the observed GIL spectrum with that predicted by the model (7) (see Fig. 5a) is in good agreement with this interpretation. Czaja [2000] showed that the required feedback of $\Delta T$ on GIL is about $2 - 3 \; mb/K$, which is in the range of the simulated response of atmospheric general circulation models to prescribed SST Robinson [2000]. One can thus not rule out the possibility that a coupled interaction between the strength of the Greenland-Icelandic Low, which reflects the storm-track variability, and the cross Gulf Stream SST gradient, which measures the low level baroclinicity at the beginning of the Atlantic storm track and is partially controlled by ocean dynamics on decadal timescales, is at work in the North Atlantic.