Sunday, August 17, 2008

Koutsoyiannis vs RealClimate.ORG

In this dose of peer-reviewed skeptical literature about the climate, we look to the Hydrological Science Journal. D. Koutsoyiannis, A. Efstratiadis, N. Mamassis, and A. Christofides wrote a text
On the credibility of climate predictions (PDF).
They simply compared the local predictions for temperature and precipitation by many models with the real observations and found out that:
... The results show that models perform poorly, even at a climatic (30-year) scale. Thus local model projections cannot be credible, whereas a common argument that models can perform better at larger spatial scales is unsupported.
Gavin Schmidt decided to criticize paper:
If he has an argument against the paper, I haven't found it. I agree with Schmidt's comment that it should have been expected that the models won't reproduce the local climate - even though our expectations could have very different reasons (my reason is that I simply know that the existing climate models don't properly deal with most of the essential climatological processes; I am not sure about Gavin's reasons).

But Koutsoyiannis et al. probably agree with it, too. (Confirmed by the lead author himself in the fast comments.) However, Koutsoyiannis et al. say not only that the local predictions of the models have been falsified: they also correctly say that the statement that the predictions would work at the longer distance scales is unsupported. And it is unsupported, indeed.

What does the word "climate" mean? It is the information about the behavior of the weather in a given region at time scales longer than 30 years or so. When we talk about the climate, we may be averaging over longer time scales but we are surely not averaging over the planet. Climate is always associated with a region: that's why we can distinguish tropical, dry, moderate, continental, and polar climates. ;-)

There is no "global climate". When people talk about "global climate change", it is the whole "climate change" that is supposed to be supplemented by the adjective "global": we are surely not talking about the changes of the "global climate" because the latter doesn't exist. Even Wikipedia controlled by William Connolley seems to agree with this proposition. It is strange that Gavin Schmidt seems to disagree.

So the short-term weather signals are averaged out but as Koutsoyiannis et al. show, it is still not enough to obtain an agreement between the models and the reality. The models clearly don't reproduce many changes well, especially not the changes driven by the long term persistence (or auto-correlation) of the time series. Note that the Hurst exponents determine the "color of the noise" and because these exponents generically exceed 0.5 in climatology, the long term persistence (the "inertia" of the climate) is very important.

Even if you don't understand these words about the Hurst exponents, you should understand that the predictions of the climate models for any particular region in the world will be essentially uncorrelated with reality because the reality is dominated by effects that are not properly simulated by the models. Because every single person lives in a particular region of the world and every region of the world is more or less incorrectly predicted by the models, I think it means that no rationally thinking person should pay serious attention to the predictions of these models.

And can the models become good at long distance scales again? Maybe. But it is extremely unlikely. If you think that they do become good at the global scale, you are believing in a very contrived, fine-tuned hypothesis: all the detailed (short-term, local) data that can be tested come out incorrectly but only when you care about one number - the global long-term temperature trend - all the errors must conspire and evaporate.

So the fashionable "climate change theory" is supposed to be an effective theory that only works at distance scales T and length scales L that are longer than certain bounds. If you want to believe that Gavin Schmidt is right, you must also believe that T must be between 30 years and 100 years and L must be greater than 6,000 km or so but shorter than 40,000 km. Why? Because the theory is falsified by the observations at shorter time and distance scales (the detailed local and/or meteorological data). But for the theory to be relevant for the Earth, the distance cutoff must be shorter than 40,000 km. And for the theory to be scary enough for a few future generations, the time cutoff must be shorter than 100 years. ;-)

When you average the known data over these very long scales, you are exactly at the moment when you lose all nontrivial climate information that could have been used to validate the model. It is exactly the moment when you are supposed to start to believe the models.

I find such a belief unjustifiable and crazy. If an effective field theory only works well enough at distances longer than a cutoff scale L, there is absolutely no a priori good reason why L should be between 6,000 kilometers and 40,000 kilometers! ;-) 1,000 km is already a very long distance not only for a particle physicist :-) but also for various local atmospheric variations to average out and for a useful approximate theory of the climate to start to be relevant. However, these theories seem to break down, even in their long-term predictions. When they break down at the distance scale of 1,000 km, is sounds extremely reasonable to me to assume that they probably break at the 6,000 km scale, too.

Similarly, if a theory highly incorrectly predicts the global climate trends for 10 or 20 years, which we already know to be the case from observations (even for the global mean temperature), it seems unreasonable to expect that the theory will be very accurate for 30-year, 50-year, or 100-year predictions.

Assuming otherwise is remotely analogous to the belief that Jesus Christ was the only person who could have walked on water. It may have been true that Jesus Christ was the only person for whom some unlikely cancellations of the gravitational force took place but it doesn't seem too likely to a scientifically trained ear. OK, Christian readers are supposed to hold their belief at this point but I just think that this particular belief is not natural from a scientific vantage point.

So I prefer the common sense approach of old-fashioned science: if all the detailed predictions of the existing models have been shown incorrect, it probably means that the models themselves are incorrect or at least substantially incomplete.

And that's the memo.


  1. A voice of reason. Thank you, Lubos

  2. I know this has been written some time ago, but I feel a point needs to be made. While i am generally sympathetic to the work done by Koutsoyiannis, and think it has merit, I don't agree with your argument that at smaller scales the models perform poorly then at longer time scales and bigger areas they are probably incorrect as well. I think that it is impossible to tell without actually testing it. The theory that validity may be found at bigger scales does have parallels in physics. I'm sure you know about the continuum limit, which defines a scale at which continuum physics applies, and below which the continuum assumption will not work. In thermofluids there are several phenomena that are a complete mystery at small physical scales or at small time scales, notably the concept of turbulence. That doesn't prevent correlations being used in certain circumstances in which the correlation is known to more or less represent reality within its limits. The world of the mechanics of heat is replete with correlations, because our understanding of the underlying phenomena just isn't there, meanwhile we still need to design heat exchangers that work, so we create an equation that matches the available data for a specific circumstance.

    My point is that the models may indeed perform better at longer time scales and lower resolution, given the stochastic nature of climate and how it could be likened to continuum theory, and given that we still really don't know much about climate. I'm skeptical about it, but I think that if they are valid at all, it is at those scales, and before we say that it isn't we ought to actually check, otherwise we're no better than the people who try to shove climate modeling results down our throats without doing any suitable validation work.