An Essay on Climate Modeling

There is a very interesting essay on global warming by Antonis Christofides and Nikos Mamassis, National Technical University of Athens, Department of Water Resources and Environmental Engineering.

I enjoyed this graph they provide. It puts the current hysteria about "global warming" into some historical perspective:

Here is what I view as the key comment in this essay:

Climate models, also known as General Circulation Models or GCMs, are very complicated models that accept a number of inputs such as carbon dioxide, sun radiation, and many more (the input variables of climate models are often called forcings). The models are run in huge supercomputers, simulating the evolution of climate around the globe, and they produce the temperature and rainfall around the earth as a function of the input. The argument goes that when the models are run without the input that is allegedly caused by humans (i.e. when they are run with the "normal" or "pre-industrial" concentration of carbon dioxide in the atmosphere), then there is no global warming; but when they are run with the alleged "anthropogenic forcings", i.e. with increased carbon dioxide concentrations, then there is global warming at the end of the 20th century, which continues throughout the 21st. This, goes on the argument, proves that the recent warming of the Earth, unlike previous climate changes, is caused by human activity; and the models, goes on the argument, can tell us how the climate will be several decades from now.

I can claim that if you drive your car at 30 km/h and you hit a wall flat on, you will only get a bruise on your left shoulder; and if you ask me why I think that, I can tell you that this is what the computer crash model says. This can be a convincing answer only if we have reasons to believe that the model works well. Therefore, the crucial question is: "How reliable are the climate models?" The Intergovernmental Panel of Climate Change (IPCC) attempts to answer that question in pages 600–601 of "The Physical Science Basis" [1], which is part of the IPCC Fourth Assessment Report. They mention three reasons: (1) that the models are based on established physical laws; (2) that the models can reproduce features of the current climate; (3) that the models can reproduce features of past climate. We need to examine them one by one.

(1) The models are based on established physical laws. If we know the speed and direction of a die, its dimensions, and the hardness of the materials of which the die and the surface on which it falls are constructed, can we predict the outcome? The physical laws of movement are very well known and understood in that case, but any model would fail miserably. The reason is that the problem is extremely complicated. If a model that attempts to simulate the physical movement of a die fails to predict the outcome, when the physical laws are fully known and really well understood, the fact that climate models are based on physical laws constitutes absolutely no evidence whatsoever that the models can work well. Climate is incredibly more complex than the throwing of a die. What's more, by the IPCC's own words, in climate there are "limitations in scientific understanding" ([1], p. 601).

(2) The models can reproduce features of current climate. It's not difficult to make a die throwing model that can reproduce features of die throwing. A model could be made that reproduces the fact that a die accelerates downwards; that when it hits the surface it rebounds; that it maintains its angular momentum; that it loses energy through friction; and so on. If a model can reproduce these features, this does not create any confidence whatsoever that it is able to correctly reproduce the final outcome of a real throw. Similarly, the fact that climate models can reproduce winters and summers, or dry areas and humid areas, has nothing to do with their ability to tell us how the climate will be in the future.

(3) The models can reproduce features of past climate. This argument is summarized in Figure 1. Climate models, it is claimed, are successful in reproducing the climate of the 20th century well, and therefore it is expected that they will be able to also reproduce the 21st century. In the Figure, the black line represents the (estimated) reality; and the yellow lines are 58 different outputs from several different climate models. The red line is the average of these outputs. Your first impression would be "wow, the models are good!" But let's take a more careful look at the Figure.

Figure 1: IPCC's presentation of model reliability ([1], p. 600).

The vertical axis in Figure 1 represents the departure from the mean temperature. If, for example, the temperature of the planet is 15°C on the average, and in 1962 the temperature of the planet was 15.3°C, then the black line will be at +0.3 in 1962. The problem is that the comparison is between the real departure and the departure of each model. If a model says that the temperature raised from 10 to 10.4°C, and in reality the temperature raised from 15 to 15.4°C, then the model would appear to coincide with reality in Figure 1. This means that the 60 lines of Figure 1 are artificially made in order to approximately fall on each other; if plotted without subtracting the mean, they would be 60 distinct lines, often very far from each other; if the chart were still only 2°C high, as it is now, many lines would be entirely outside the chart.

You may argue that if a model could get the departure from the mean, this would be a remarkable achievement, even if it could not get the actual value. Indeed. But it's different if you are shown Figure 1 and you think "wow, the models are good!", and it's different if you were shown a Figure of 60 entirely distinct lines, far from each other, in which case you'd think "hmm, the models are crap; however, they do seem to get the slope right."

Do they get the slope right? From what we can see, the (estimated) reality indicates relatively constant temperature from 1930 to 1970; in contrast, the models indicate a continuous temperature rise from 1930 to 1970, with an abrupt drop in 1962, when the Agung volcano errupted. We don't see, in that Figure, that the erruption has particularly influenced reality. If you look carefully into Figure 1, you'll notice other problems as well.

A third problem is that it is completely unscientific to draw the red line that shows the average of the models. There are absolutely no scientific grounds to believe that the average of the models is something special, or that it is better from the models. If the models get it wrong, there is absolutely no reason why the average of many models would get it less wrong. In fact, the mere existence of that red line makes us wonder: how were the 58 model outputs chosen? Why were those 58 outputs chosen, and not some other 58 outputs? Could that be cherry-picking?

Finally, even if the models match reality to some extent, how do we know that they were not made to fit reality? Normally scientists are extremely careful to adjust their models to match a specific set of data, and then they check how well their models match an entirely different set of data. In this case, for example, one could adjust the models so that they'd match the period 1900―1950, without looking at 1950―2000, and then checking how well the models can match reality for the latter period. This is not as easy as it seems in the case of climate models, and it may be impossible because of the huge complexity of the problem and the fact that there is data for only about one century; but we've never been given evidence that the match, to the doubtful extent that it exists, is anything better than data fitting.

Extraordinary claims require extraordinary evidence. The claim that models can predict future climate is extraordinary; but instead of clear evidence, we are provided with confusing arguments and strange linguistic formations: rather than being told that "there is evidence that the models can predict future climate", we are being told that "there is considerable confidence that [the models] provide credible quantitative estimates of future climate, particularly at continental scales and above." What is "considerable confidence" and how far is it from certainty? Who is it that is considerably confident? What is "credible"? Does it merely mean plausible? What does "particularly at continental scales and above" mean? Does it mean "only at continental scales and above", as the quotation by Gavin Schmidt implies? Or does it mean "which is good at regional scales and even better at larger ones", as is implied by the IPCC's 94-page-long Chapter 11 [3], that is crammed with regional details?

A number of studies explain why the models are not credible (e.g. [4], [5]). But we have chosen, in this chapter, to only look at what IPCC says. The IPCC should have provided evidence that models do a decent job, and then we'd be able to start discussing on the merit of that evidence. But the IPCC, as you can see, has not provided any evidence. Rather than clear answers to clear questions, they have provided confusion and lawyerish language.

Here's another interesting bit:

Roulette is unpredictable in the short term, but very predictable in the long term: you can look at its outcomes as random noise on a well-known signal. Roulette is very uncertain when you zoom in, but very certain when you zoom out. Climate is equally uncertain at all zoom levels. In fact, mathematical analysis of the climate indicates that its behaviour is such that the uncertainty is the maximum possible at all zoom levels. This maximisation of uncertainty at all scales is called the Hurst-Kolmogorov behaviour of climatic processes [3].

Nature loves uncertainty, and it fools us in two ways: on the one hand we wouldn't be able to predict the future of climate, even if we fully knew the natural laws that govern it, because of chaos; and on the other hand, we can't be very certain of the statistically expected behaviour of climate which is based on our observations of the past, because of the Hurst-Kolmogorov behaviour.

Changes in climate and the weather sometimes have a single dominant cause, and sometimes they are chaotic.

This essay has seven sections. The above are just a few tantalizing bits. Go read the whole thing. It gives a viewpoint that doesn't get much coverage in the media. Hysteria and exaggeration sells. The drab facts of life are too boring to sell newpapers or excite a media audience. The IPCC is a media darling. You have to work hard not to get swept up in the "science" of global warming.