Over at Andrew Sullivan’s book club, where they’re discussing Lomborg’s book, one of the letter writers wrote the following:

I was preempted by at least two persons in what seems to be the key point of skepticism about the current global warming discourse, the extreme hubris involved in trying to forecast the long-term temperature of the earth based on primitive computer models, when we can’t even forecast local weather with any certainty…

I’ve heard Rush Limbaugh make this argument as well–how can we predict long-term climate when we can’t even forecast tomorrow’s showers?

While I’m a global warming skeptic myself, and it’s a seductive argument, it’s wrong. I believe that the reason that we can’t make long-term predictions about climate is because of the chaotic, non-linear nature and complexity of the phenomenon, but the ability to make long-term predictions is actually unrelated to our ability to make short-term ones.

To understand why (or at least to see another example of the two being unrelated), consider another similar phenomenon–the stock market. Most financial advisors will tell you to put your money into stocks, because over the long haul, they’re going to go up. And if you look at the sweep of finance history, such advice would have been borne out. But that doesn’t mean that they can predict what the market will do next week, or even tomorrow (if they could, they wouldn’t have to make a living providing advice…)

How can they make a long-term prediction when they can’t make a short-term one? Because the long-term one is not based on the short term–it is not a series of predictions adding up to a long one. While a broad trend can have a reasonable probability assigned to it from fundamental underlying causes, the various ups and downs as it gets there are subject to different, unforeseeable forces. There is “noise” in the pattern of climate, or markets, and this noise is what we experience as unexpected weather, or daily fluctuations in stock prices. And while we can extrapolate current data to derive a predicted signal, we can never predict noise with any reliability, by definition.