David, I don’t get the feeling we are getting far in this discussion so I’ll highlight in a few points my main views here, and one that I suspect would be common to the bulk of the climate community. :
1) To me, the whole issue is one of signal vs. noise. That is, can any possible sensitive dependence to initial conditions that may conceivably apply over centennial timescales dampen out the expected changes due to a doubling or quadrupling of CO2? Even if we integrate an essentially perfect atmosphere-ocean model forward in time with no forcings (e.g., volcanoes) it will have variability on all timescales, and the magnitude of that variability will depend on the variable (e.g., temperature or precipitation) and spatial scale. Is that larger or smaller than the size of the signal we expect from a given CO2 or solar increase. Period.
For, say, long-term global mean temperature, there is no evidence that the models are missing anything of strong significance. Moreover, (on a consistent basis) no one has successively observed or simulated internal variability of the magnitude like 2xCO2. That may not be true for other statistics such as rainfall in East Asia, though this should be evaluated independently.
2) It seems necessary to formulate a common definitional framework here. Negative feedbacks are not examples of the same sensitive dependence to initial conditions that plagues the weather forecasting issue beyond 1-2 weeks, even if they involve clouds. Things may be “unpredictable” in the sense that they are left out of current models, or are parametrized in ways that cannot be extrapolated outside the modern climate, but it’s unclear that they are “unpredictable” in the sense that whether they actually happen is sensitively dependent on initial conditions.
3) It is my impression that there is widespread agreement that bifurcations, tipping points, threshold responses etc can and do occur, and that responses may exist not anticipated by modern models (such as the prospect of much more rapid melt from the Greenland ice sheet). Even in these cases, it is generally the long-term growth/decrease in global mean T which can often set the stage for such events. For unknown ‘switches,’ it is impossible on first principles to evaluate whether it will have sensitive dependence to initial conditions until someone comes up with a specific proposal for what such a switch might be, though constraints exist concerning the realism of various events occurring (e.g., a background state that is conducive to D-O and Heinrich events is unlikely in a warm climate with strong deep water formation and little ice).
4) The theoretical possibility of long-term chaos could conceivably apply when you include the deep ocean, owing to multiple interacting timescales in the system. Even if we stick to shallow ocean dynamics, ENSO emerges as a nonlinear chaotic phenomenon that exhibits predictability loss, and the strength/impacts of ENSO do depend somewhat on the background state, but its projection onto the climate is fairly standard across multiple events (as I mentioned in a previous comments, we see rather predictable responses in the distribution of warm/cool or wet/dry across the U.S/Canada, and its projection onto global mean T is very small relative to the increase in CO2).
5) A lot of this discussion is very academic since on first principles one could concoct some play-around model where the sensitivity of trends in decadal-centennial statistics (to initial conditions) overwhelms the long term sensitivity of that statistic to the increase of GHGs. However, here is where I part ways with several blogospheric dynamical system theorists and mathematicians, because it’s another thing to demonstrate a compelling mechanism that acts like this in the real world.
“As I began to learn meteorology, I found it necessary
to unlearn some mathematics.” –Lorenz