Figures 20 to 22 in M3). During the early to mid-1960s, however, the Siberian River runoffs were generally below normal (Cattle, 1985) and hence would not have contributed to a positive sea-ice anomaly in the western Arctic. The above-average North American runoffs led to large sea-ice concentration anomalies in the Beaufort Sea in the western Arctic during the mid-1960s (see Figure 19 in M3 and also Figure 3c below), which in turn were exported out of the Arctic into the Greenland and Iceland seas via the Beaufort gyre and Transpolar Drift Stream over a 2- to 3-year period (Colony and Thorndike, 1984). The melting of the large sea-ice anomalies over several summers in the Iceland Sea would then have led to the GSA, which peaked in 1968 (see Figure 6 in M3).
It is important to emphasize that the Beaufort Sea ice anomalies described above are based on ice concentration data only. As W. Chapman has suggested (personal communication), it is conceivable that during the winter and spring months, when the ice is essentially land-locked there, the ice thickness could be increasing due to the lower near-surface salinities brought about by the increased runoffs the previous summer. Such thicker ice would then be exported out of the Arctic (via the Beaufort gyre and transpolar drift stream) into the Greenland Sea, and hence add to the ice and fresh-water anomaly in the latter region. A more complete set of ice-thickness data for the Arctic (and the western Arctic in particular) may thus provide additional support for the negative feedback-loop theory.
Due to the stabilizing influence of the GSA, convective overturning during winter in the Iceland Sea, which normally brings warm Atlantic water to the surface, was suppressed, which in turn reduced the sea surface temperatures (SSTs) there (see Dickson et al., 1988). This reduction in SST, we argue, would tend to reduce cyclogenesis in this region, as well as further south in the Irminger Basin because of the advection of cold SSTs by the East Greenland Current. It is hypothesized that a lowered SST leads to reduced cyclogenesis, precipitation, and runoff over North America, and therefore causes a flip in the sign of the perturbations in the next circuit of the feedback loop. A partial confirmation of the above hypothesis has been obtained by Serreze et al. (1992), who showed that during the peak GSA years (1967-1971) the frequency of anticyclones during winter increased over northern Canada, especially over the Canadian Arctic Archipelago, which implies drier conditions in this region. Nevertheless, at this stage the link between Iceland Sea-Irminger Basin cyclogenesis and northern-Canada precipitation should be regarded with some degree of skepticism until further studies have been carried out. The idea of such a link is consistent with recent teleconnection results of Walsh and Chapman (1990b), however. They showed that winter sea level pressure (SLP) anomalies in the Iceland Sea and Irminger Basin are highly correlated (0.6 < r < 0.8) with those at the base point 75°N, 90°W in the CAA (see their Figure 12b).
Despite the above possible weakness, the feedback-loop theory has had two successful applications. First, using a cycle time of about 20 years, M3 predicted that there should be another GISA in the Greenland and Iceland seas in the late 1980s. This conjecture was verified through an examination of February sea-ice concentration data made by Mysak and Power (1991), and evidence of anomalous runoffs that might have generated this anomaly will be presented below. Second, the feedback loop in Figure 1 has been used by Darby and Mysak (1993) as a guide to developing a Boolean delay-equation (BDE) model of the interdecadal Arctic cycle. The BDE model contains six variables that represent the state of precipitation in northern Canada (1 or 0, corresponding to a high or low state), the state of ice and salinity conditions in the western Arctic and in the Iceland Sea, and the convective state in the Iceland Sea. For a variety of initial conditions, the model successfully simulates a 20-year cycle in the Boolean variables. A particularly novel result is that by allowing for different time scales for ice and salinity advection from the western Arctic through to the Greenland and Iceland seas, an ice anomaly in the Iceland Sea can persist longer than an ice anomaly in the western Arctic, a finding that is in agreement with observations.
Figure 2b shows that during the winters of 1987 and 1988 a large anomaly in sea-ice extent existed in the Greenland and Iceland seas, since the 0.9 ice-concentration contour extends well beyond the east Greenland coast as compared to its climatological position (Figure 2a). Concurrent with this ice anomaly, which also shows up in the late 1980s as a peak in the areal sea-ice anomaly time series for this region (see Mysak and Power, 1992), was the reduction of convection in the Greenland Sea during winter 1988 (Rudels et al., 1989) and the appearance of low-salinity water there during February and March 1989 (GSP Group, 1990). If the latter fresh-water anomaly advected southward into the Iceland Sea and suppressed convection there, then these features taken together suggest that a moderately sized GISA occurred in the Iceland Sea in the late 1980s. As in the case of the 1960s GSA, the large Greenland-Iceland sea-ice anomalies in the late 1980s appeared to have advected into the Labrador Sea by the early 1990s and thus contributed to extremely heavy sea-ice conditions and cooler air temperatures along the coast of Newfoundland in May 1991 (Globe and Mail, May 31, 1991). (It has also been argued that such positive ice anomalies off Newfoundland could be partly due to anomalous offshore winds that forced coastal ice seaward (J. Elliot, personal communication, 1991).)