Lying With Charts, Global Warming Edition

[GliderGuider]

53. That's odd. You complain when I say the changes are unpredictable

and then you complain when I say some aspects are predictable. If I weren't such an inherently generous creature I'd suspect you of disagreeing just to be disagreeable. But I'm sure you have a noble purpose, so here goes.

What I think is that general changes in parameters like the average atmospheric and/or oceanic temperature - especially when aggregated over the entire globe or large portions of it - are in principle predictable. I say that because when you aggregate to that extent the chaotic aspects of the system are smoothed away - the system behaviour becomes much less chaotic and more predictable.

Right now climate change is unpredictable not because of any mathematical constraints but because of insufficient knowledge. Although it's predictable in principle, it's unpredictable in practice at the moment - the models we have available are still babies compared to the true complexity of the system. We don't know all the relevant variables and their interrelationships yet, because we haven't been doing it long enough, but I think there shouldn't be any mathematical or physical laws standing in the way of improved predictability as we learn more about the system. I could be proven wrong, though, because there's a lot about the system we don't understand yet. It could still turn out to be fundamentally chaotic after all. Let's hope that's not true.

That's not the case as the granularity of the pieces of the system under consideration get smaller, either in space or time. Once you lose the smoothing and buffering effects of aggregation, the chaotic aspects of the system begin to dominate. Under those conditions the behaviour becomes unpredictable in a formal, mathematical sense - no matter how much data we acquire, the system behaviour can't be predicted. It is unpredictable both in practice and in principle. Predictable large-scale systems give rise to chaotic, unpredictable local effects.

A simple example would be a river during spring snow-melt. We can use well-known variables like snow-pack depth, snow condition, air temperature and topography to predict the river's flow rate with pretty good accuracy. If we don't know the condition of the entire snow-pack in the catch basin we won't be able to predict the flow rate accurately, but that's a practical limitation, not a theoretical one.

However, we can't predict the changes in eddy patterns that form in the river as the flow rate increases, because those eddies are chaotic systems. Predicting the flow rate more precisely doesn't help to predict the eddy patterns at all. The best we can say is that at some flow rate the turbulence in this general area of the river will increase to some degree, and perhaps provide reasonable error bars.

It's the same for climate, except that the whole system is much larger and more complex than a river. The practical difficulties of global prediction remain enormous, but are in theory soluble. Changes in regional climate are harder to predict as the region gets smaller because chaotic effects begin to dominate the system behaviour.

This stuff isn't rocket science - it's actually much tougher than that.