forced their model with a freshwater flux qualitatively similar to that in Figure 3.4. This coupled model appears to have a unique equilibrium, and the simulated changes evolve roughly on the time scale of the forcing. Any abruptness in such a model will therefore be generated only through the forcing and not through the dynamics of the model itself. The transient behavior of this complex model is qualitatively comparable with that of the diffusive Stommel model (Figure 3.4, upper dashed curve). The complete, temporary shutdown of the THC causes a massive cooling in the North Atlantic and weak warming in the South Atlantic. The climatic behavior of opposite sign during a full THC shutdown has been termed the bipolar seesaw (Broecker, 1998; Stocker, 1998), and is a phenomenon that was probably active during most of the last glacial period (Blunier and Brook, 2001) (see Plate 2).
The second type of abrupt change in climate models occurs when slow changes in the forcing push the model beyond a threshold and induce a transition to a second equilibrium. This behavior is simulated in ocean-only models that include simplified formulations of the atmosphere (Rahmstorf, 1994, 1995) and in models of reduced complexity (Aeberhardt et al., 2000; Schmittner and Weaver, 2001). Periodic forcing can also trigger abrupt change in a model that has multiple equilibria, provided that the forcing amplitude covers both branches of the hysteresis. Ganopolski and Rahmstorf (2001) showed that weak forcing is sufficient if the model’s hysteresis has a narrow shape. As emphasized earlier, such results are not considered robust, because they depend strongly on the structure of the hysteresis and the location of the initial state on it. Tuning and choices of parameters can strongly influence the transient behavior under a given forcing (Schmittner and Weaver, 2001) (Figure 3.4).
The third possibility for the occurrence of abrupt climate change in model simulations is a spontaneous regime transition similar to the paradigm of the Lorenz (1963) model. This concept has been explored in simplified models (Timmermann and Lohmann, 2000) but was recently also found in a control run of a coupled atmosphere-ocean GCM (Hall and Stouffer, 2001). Annual mean surface air temperature, averaged over a region extending from southwestern Greenland to Iceland, showed natural interannual variability of around ±1°C (Figure 3.5, top). During the 15,000-year integration, one event occurred in which the temperature underwent a rapid decrease of about 4°C within a few decades (Figure 3.5, bottom). The event was triggered by an atmospheric pattern of persistent northwesterly flow in the region. This induced a southwestward Ekman flow in the surface ocean,