scale, the THC (and hence heat transport) can respond to changes in external forcing much faster than on the millennial time scale of thermodynamic equilibration of the deep ocean, but it is not clear which of the various plausible, identifiable time scales—ranging from months to decades—is most relevant. The classical picture (Kawase, 1987; Döscher et al., 1994) suggests that the deep circulation is set up through wave processes on time scales of months to a few years, but it has also been argued that advection by the deep western boundary current must be important, and this implies time scales of decades (Marotzke and Klinger, 2000).

The potential predictability and prediction of the THC raise the thorny issue of assessing the quality of numerical simulations of future climate evolution. The fundamental problem is best understood when juxtaposed with daily weather forecasting. Through thousands of forecasts based on model simulations, it has been established that weather predictions have “skill”; that is, they improve upon a naïve baseline. But the time needed to test a weather forecast typically is a day—we will know tomorrow night whether tomorrow’s picnic gets rained out. An analogous accumulation of evidence of forecast skill is impossible when the prediction lead time is decades and more. How, then, can we state with well-defined confidence what is likely in store?

An abrupt THC change is likely to have consequences massive enough that its probability should be estimated, if only crudely. In lieu of establishing true forecast skill, the models used for future climate-change scenarios should pass all the tests posed by available data. For example, one would reasonably have more confidence in climate models of future evolution that were able also to simulate the complicated climate history, including the rapid changes, of the last glacial interval. However, it is difficult to establish the “transferability” of putative modeling success of the past. Although being able to simulate the Younger Dryas is neither strictly necessary nor sufficient for being able to predict the THC’s evolution under global warming, it would enhance confidence considerably. Another strategy would be to simulate an ensemble of possible future climates and ascribe probabilities to them; an example is discussed in Chapter 4. Still, the problem remains of assessing whether the probabilities have the right order of magnitude. A further limit to predictability might arise from chaotic dynamics. In the global model of Wang et al. (1999), the timing and presumably the critical threshold of a THC collapse were fundamentally unpredictable to within a factor of 2. Different realizations of statistically identical random perturbations in the wind field in Wang et al. (1999), mimicking different (unpre-



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement