of the ocean-atmosphere system in the future. Such changes are ''unforeseen" because the vast majority of numerical models used to predict the climate response to CO2 forcing treat the oceans as fixed, or restrict their interaction with the atmosphere to the mixed layer. However, ocean currents are crucial in determining global transports of heat and moisture, and the transient response of climate will thus depend critically on the response of the oceanic circulation. Although the importance of simulating such interactions is widely recognized, it is nevertheless hampered by the difficulty of simulating the general circulation of the oceans (i.e., "getting it right for the right reasons") and the extreme computational requirements of coupling three-dimensional models of the ocean and atmosphere.
Citing the geologic evidence for abrupt climate changes in the past, Broecker (1987) has already articulated the concern that there may be "surprises in the greenhouse" that cannot be adequately portrayed, let alone predicted, using existing models. I follow Broecker's lead here, adding new evidence that, as predicted on the basis of terrestrial indications of rapid climate change around the Atlantic region, ocean circulation patterns and sea surface temperatures varied with degree-per-decade rapidity. Understanding and modeling the mechanisms underlying these changes is of high priority. If, on the one hand, the climate response to increasing greenhouse-gas content of the atmosphere is expected to be gradual, then we may choose to pursue a strategy of adaptation. On the other hand, if we cannot rule out the possibility that the mechanisms that provoked the "surprises" seen in the geologic record may once again become active in the future, we may wish to adopt a policy of restraint, while at the same time trying to enhance our capacity for the recognition and detection of the premonitory signs of abrupt circulation and climate change.
There is ample evidence from the geologic and instrumental record for changes in patterns of ocean circulation on a variety of time and space scales. These range from global-scale reorganizations of the thermohaline circulation on glacial to interglacial (105-yr to 104-yr) time scales (Boyle and Keigwin, 1987; Duplessy et al., 1988; Broecker and Denton, 1989; Raymo et al., 1990) to quasi-cyclic, interannual changes of regional scale in historical times, such as El Niño events (Cane, 1983). The spectrum of variability also appears to include possibly cyclic phenomena such as oceanic cooling and reduced ventilation associated with the North Atlantic's Great Salinity Anomaly during the 1960s to 1980s (Dickson et al., 1988b; Lazier, 1988; Mysak and Power, 1991). One of the most extreme examples of rapid ocean circulation change seen anywhere has been inferred from the geologic record of the last deglaciation (15,000 to 8,000 radiocarbon yr BP)2 around the North Atlantic, when air temperatures shifted 5°C to 10°C in a few centuries or less. The proxy record of these temperature changes appears to be coherent over a broad region; it includes oxygen isotopic shifts in Greenland ice cores (Dansgaard et al., 1982, 1989), pollen and isotopic shifts in European lake sediments (Iversen, 1973; Siegenthaler et al., 1984; Ammann and Lotter, 1989), and changes in fossil assemblages of coleoptera in the British Isles (Coope, 1977; Atkinson et al., 1987) (Figure 1). Similar-looking changes characterize the Greenland ice-core record between 8,000 and 80,000 years ago; they thus appear to be a characteristic feature of glacial and deglacial climate in the North Atlantic region (Dansgaard et al., 1982).
The extraordinary rate of these changes was revealed by isotope studies of annually layered ice deposited approximately 10,500 to 10,000 yr BP on Greenland, when air temperatures over the ice sheet rose by 7°C in just 50 years (Dansgaard et al., 1989). As there is no known external (e.g., solar or orbital) climate-forcing agent of sufficient potency and frequency to explain the rapid and recurrent temperature changes seen in these records, the question remains: What process(es) drove the observed variations? The most likely answer, according to a variety of authors (e.g., Oeschger et al., 1984; Broecker et al., 1985), is that they were the result of sudden shifts in the strength and, therefore, in the heat-carrying capacity of the ocean's conveyor-belt circulation system—a global-scale overturning marked by sinking of relatively cold and salty water in the northern North Atlantic, generalized upwelling in the Indian and Pacific oceans, and the return of warm salty surface water to the Atlantic around Africa and South America (Gordon, 1986; Broecker, 1991 a).
The impact of heat transport by the conveyor on the climate of the North Atlantic region is clearly evident in the large positive deviation from zonally averaged January air temperatures centered over and downwind of the northern North Atlantic and Nordic seas, as seen in Figure 2 (after Barry and Chorley, 1982). While there is little doubt that the conveyor contributes importantly to the warmth and habitability of the circum-North Atlantic region (especially northwestern Europe and Scandinavia), a key question from the climate-change point of view is whether the conveyor circulation is subject to significant variation. In an attempt to address this issue, the stability of the conveyor has been investigated in a large number of numerical models of ocean circulation, virtually all of which indicate that its operation is threatened by subtle changes in fresh-water balance at the surface near regions of convection in the northern North