Mabinogogiblog: Climate models - are they scientific and reliable?

Monday, November 19, 2012

Climate models - are they scientific and reliable?

Uploaded by C2SMvisualizations on Aug 25, 2011

This animation shows near surface air temperature (colors), clouds (white) and precipitation (blue) from a simulation using the Community Earth System Model (CESM), maintained by the National Center for Atmospheric Research (NCAR). The global climate model was run at a resolution of about 25km. It shows that today's climate models are capable of simulating climate and weather phenomena in great detail, such as tropical cyclones in the Pacific, occasional cirrus clouds in Central Africa, or orographic precipitation over the Alps.

Computation and visualization performed at ETH Zurich.

For more information, go tohttp://www.c2sm.ethz.ch/services/data_visualization/

In the three years I have been studying and debating global warming, I have avoided climate models, because they evoke such a total rejection from contrarians. It just ain't worth going there; climate models are, in their world, such a thoroughly Poisoned Well that they are more of a hindrance than a help in debate.

However, after I had written up this post on empirical evidence for climate sensitivity, Bishop Hill airily rejected all but his preferred result, saying that most of the other nine had probably used climate models along with the data, and therefore were not to be trusted. I replied that he was being a bit OCD. But clearly the time has come for me to focus on climate modelling.

First, the earth's climate is a complex system, and the only way that we can gain a thorough understanding of all its processes is through computer modelling. The human brain is doubtless big enough in theory to model the climate, but only if it gives up doing other tasks such as tying shoelaces and brushing teeth. So to study climate, we need models, and therefore, by totally dismissing models as a valid way of approaching climate, the contrarians are actually saying "We do not want you to study climate scientifically".

Second, the question arises for the contrarians, if they do not want computer models to be used, would they be happier if all the equations were to be worked out by hand, with a pencil and slide rule? Because that was how it used to be done. Arrhenius referred to the "tedious calculations" he had to do back in 1896 when he made the first attempt to work out the impact of CO2 on temperature. So would paper and pencil be OK? Or do the contrarians just want us not to even think about whether CO2 affects our planet's climate?

Third, models are not intrinsically unscientific. I say this, because to listen to contrarians, you would think they are of the very Devil, totally opposed to science. Contrarians set models against "empirical evidence", that is, evidence that has been obtained by observation or experimentation. Given that it is not possible to perform experiments on a planet (except of course the one we are carrying out in altering the greenhouse effect), empirical evidence would mean that climatology would be allowed only to do observation. Models are a way of climatologists performing experiments on a proxy, virtual planet.

What models in fact consist of are formulae that work out how one climatic factor impacts on other factors.

Here is one of the simplest model/formulae, about radiative balance of the Earth:

$(1-a)S \pi r^2 = 4 \pi r^2 \epsilon \sigma T^4$ (source)

From this kind of formula, models have evolved into the modern state of the art global climate models which contain processes for atmosphere, clouds, oceans, land and chemistry, huge models that need vast amounts of time (relatively) on supercomputers, and whose product is shown at the top of this post.

So we can think of models as virtual, Information Technology hypotheses. We put together a large number of physical formulae about energy transfer from sun to atmosphere and so on. If we do not have a precise mathematical formula, we guess a set of reasonable values and try them out, one at a time, to see which is more accurate, which works better. This is called parameterisation.

Once a model is complete in the eyes of its creators, it is tested against the historical temperature record. It is wound up, as it were, set up with the physical situation (temperature, state of ocean currents etc) at a selected date (say 1900) and allowed to run, to see if it keeps within a reasonable distance of the historical temperature. As it runs, known historical events (Solar changes, volcanoes el Ninos &c) will be fed in, but no other external alterations will be made.

In earlier models, arbitrary adjustments had to be programmed in to stop the model drifting away from the historical record. Modern state of the art models no longer need these adjustments.

So the scientific method does apply. Science works by observation, hypothesis formulation and hypothesis adjustment or refutation. The observations are the temperature record. The models are the hypothesis. Adjustment occurs continuously with models, adding new factors as they come to light, and adjusting present parameterisations to make their product more accurate.

I discuss the need for the models to anticipate fluctuations in the temperature record here.

The models are an intrinsic part of the scientific process. They are not, and never will be, 100% accurate. Sometimes they underestimate the temperature and sea level change in their projections. The models are not 100% accurate, but they are growing more accurate as time passes.

Sometimes, they come up with interesting and unexpected goods. For instance, models began to produce their own El Nino-like cycles.

Here is a diagrammatic summary of the evolution of climate models in terms of the factors that they compute: