Models and Climate Change

There is an excellent post by Willis Eschenbach on the Watts Up With That? blog. The posting looks at "global warming" and is full of interesting points to be made. I especially enjoyed this one bit about modeling climate change:

Question 9. Are the models capable of projecting climate changes for 100 years?

My answer to this is a resounding “no”. The claim is often made that it is easier to project long-term climate changes than short-term weather changes. I see no reason to believe that is true. The IPCC says:

“Projecting changes in climate due to changes in greenhouse gases 50 years from now is a very different and much more easily solved problem than forecasting weather patterns just weeks from now. To put it another way, long-term variations brought about by changes in the composition of the atmosphere are much more predictable than individual weather events.” [from page 105, 2007 IPCC WG1, FAQ 1.2]

To me, that seems very doubtful. The problem with that theory is that climate models have to deal with many more variables than weather models. They have to model all of the variables that weather models contain, plus:

• Land biology

• Sea biology

• Ocean currents

• Ground freezing and thawing

• Changes in sea ice extent and area

• Aerosol changes

• Changes in solar intensity

• Average volcanic effects

• Snow accumulation, area, melt, and sublimation

• Effect of melt water pooling on ice

• Freezing and thawing of lakes

• Changes in oceanic salinity

• Changes in ice cap and glacier thickness and extent

• Changes in atmospheric trace gases

• Variations in soil moisture

• Alterations in land use/land cover

• Interactions between all of the above

• Mechanisms which tend to maximise the sum of work and entropy according to the Constructal Law.

How can a more complex situation be modeled more easily and accurately than a simpler situation? That makes no sense at all.

Next, the problem with weather models has been clearly identified as the fact that weather is chaotic. This means that no matter how well the model starts out, within a short time it will go off the rails. But the same is true for climate, it is also chaotic. Thus, there is no reason to assume that we can predict it any better than we can predict the weather. See Mandelbrot on the chaotic nature of climate.

Finally, climate models have done very poorly in the short-term. There has been no statistically significant warming in the last fifteen years. This was not predicted by a single climate model. People keep saying that the models do well in the long-term … but no one has ever identified when the changeover occurs. Are they unreliable up to twenty-five years and reliable thereafter? Fifty years?

Having built computer models I know how hard it is to validate the model (show that you are modeling what you claim to be modeling) and to verify the model (show that the model accurately represents the phenomenon you are modeling). I think climate models suffer on both fronts. As Eschenbach points out, they leave out a lot of detail. That's OK if the details are irrelevant. But it is pretty clear to me that the details are relevant, therefore the models are not valid. As for verifying the models, I have no special insight into that. But I do know that once you build something complex (say like a Toyota breaking system with a mechanical part and control software) it can be extremely difficult to verify that the model accurately does what it should under all conditions. As Eschenbach points out, the models offered so far are poor on modeling short term phenomena, so it is questionable how good they are at getting the long term phenomena.