and this automatically means that the convection-THC link is much more tenuous. As Aagaard and Carmack (1989) point out, we can envisage a system "in which mid-depth convection (which is the main source of the Denmark Strait overflow) occurs, albeit involving waters of reduced salinity, while the deeper convection, which renews the densest waters in the system, is shut down." Also, although the bulk of the outflow through the Denmark Strait has been shown by Swift et al. (1980) to consist of locally formed Arctic Intermediate Water, the 2.5 Sv of that local AIW production is augmented with about 0.5 Sv of upper Polar Deep Water whose origins lie north of Fram Strait in the Arctic Ocean (Rudels and Quadfasel, 1991). Thus, even if properly characterized, the convection of the Greenland and Iceland seas is not the sole factor involved.
The T-S characteristics of the Greenland Sea Deep Water did not appear to feel the influence of the widespread surface freshening of the late 1960s and early 1970s. The time series of Figure 18 are interpreted by Meincke et al. (1992) as showing conditions favorable to the renewal of Greenland Sea Deep Water at that time. This seems important if GSA freshening events are to be held responsible for modulating the thermohaline circulation with their recurrent signal. (However, Schlosser et al. (1991 a) report a cessation of deep-water formation from 1980 onward, coincident with the return of the GSA signal to the Greenland Sea.)
While the hydrographic characteristics of the Denmark Strait outflow have been observed to change (Brewer et al., 1983; Lazier, 1988), we have no direct observational evidence that the overflow transport actually changes on the decadal time scales that current THC models predict. Our direct current measurements are of only 4 years'
duration and are too short by themselves to permit comment on the question of decadal change in overflow transport (Dickson and Brown, 1994), even though they do appear to emphasize the steadiness rather than the variability of the outflow (Figure 19). A slightly longer perspective is provided by the downstream behavior of tracers, which appears to confirm this impression of steadiness in the overflowing stream. As Schlosser and Smethie point out (1995, in this section), the "age of the deep core between 32°N and 44°N did not change by more than 10% between 1983 and 1990, which implies that the combination of water-mass formation rate, transport, and mixing with adjacent water has been constant to within about ± 10 percent over this time period."
If these points seem unduly critical of current models, they are not intended to be so. There is no doubt that our ability to trace, understand, and simulate these large-amplitude events at high latitudes will open up an important avenue to improved climate prediction. Even if they turn out to be less than cyclic in character, the slow shifts in the ocean-atmosphere system that these events set in motion will still be of use in forecasting effects, and these effects in turn will still be of sufficient significance to make any improvement in prediction worthwhile. The above are merely suggestions for points that future models will have to address—in other words, reasons for model improvement rather than model abandonment. Recent simulations by Weaver (personal communication) appear to support the idea of an invariant THC on decadal time scales.
Although the evidence is circumstantial, there does appear to have been a physically mediated, century-long variation in the exchange of heat and larvae between Iceland and Greenland via the Irminger Current, with major socioeconomic impact.
Although it contains periodic elements, the Great Salinity Anomaly is neither a periodic nor a frequently recurring phenomenon. One or possibly two such events are thought to have occurred during this century, and the event we observed in 1968-1982 was driven by a secular change in the winter pressure field at Greenland.
Although its proxies clearly reflect the glacial-interglacial signal, the global thermohaline circulation has not yet been demonstrated to vary at decadal time scales. Simulations of the process are not yet adequately realistic in topography, in the role of deep convection (as opposed to Arctic Intermediate Water formation), in the processes that might cap deep convection, and in the periodicity of these processes.