Models
You can deduce the significant impact of human activities on the climate just from basic physics and real world observations. But you can learn more about the climate system and how it will respond to the rising levels of greenhouse gases by constructing a computer model of it - a simulation based on a code that generates a realistic physical description of the climate system, and can solve the relevant physical equations for the energy balance.
Why models are used
We do this in astronomy as well. With today’s computers, you can run a code to simulate the emission from a shell of gas around a star in a few hours. If you’ve got your input parameters right, you can reproduce observed emission to a pretty good tolerance. Not everything will match, but the agreement is good enough that we have a lot of confidence that the physics in the model is correct, and often the things that don’t match tell us something about what is in the nebula that we didn’t include in the simulation.
If we didn’t have models, we’d have to rely on a simplified empirical approach in which all the nebular parameters are reduced to a few equations that don’t require supercomputers to calculate. We do actually use this approach as well, because it’s much quicker and the results are often quite similar to what the models tell us. But for complex objects, the simple approach is less applicable and constructing a physically realistic model is a better way of understanding what is happening.
And so it is with climate. A number of codes exist to simulate the climate, and the generally good overall agreement between simulations and observed temperatures gives scientists confidence that the physics is correct. Discrepancies between observations and reality are certainly there, but they give us an idea of what assumptions made in the models may not apply.
Limitations of models
Consistency and degeneracy
There are a few important limits to what a model can tell you. The fact is, even if a model completely and accurately reproduces what you observe, that can’t tell us that the model is completely correct. It only tells us that the model is consistent with the observations. In astronomy, we can’t go and look directly at the nebulae we study so there is always the possibility that there is something going on there that we had no idea about, so we have to be content with generating nebulae which are consistent with reality until we find an observation that contradicts the model. In climate studies, this limitation is less severe because we’re in the atmosphere and we can study it directly. It should be much easier to determine the processes that are occurring in the atmosphere.
A model which produces results consistent with reality may not be unique; another model with completely different input parameters might give results equally as consistent. This is called degeneracy. What helps here is having as many different observations as possible. For example, in astronomy, it’s easiest to look at visible light, because you can do that from the ground. A normal star that’s mostly hydrogen, and one which has no hydrogen at all at its surface, can produce quite similar visible emission, so a model of a hydrogen-deficient star and a model of a normal star might not be distinguishable on the basis of optical observations alone. But the ultra-violet emission from the two types of stars is vastly different. With UV observations as well as visible, you can distinguish the two.
In the same way, different inputs to climate models might have the same effect, roughly, on tropical temperatures but very different effects on polar temperatures. The more temperature observations we have from the more places, the better we can constrain the models. With weather stations all over the world, and satellites to measure temperatures in parts of the world where humans aren’t, the observational situation is pretty good in climate studies.
Resolution
A limitation that both astronomical and climate models suffer from is one of resolution. Although computers roughly double in power every 18 months, in line with Moore’s law, we cannot models these vast complex objects down to each individual atom within them. We have to chop them up into manageable blocks, so that models can give us results on a reasonable timescale. In astronomy, the chunks I chop a nebula into are pretty vast - about 1000 times larger than the distance between the Earth and the Sun. Climate scientists chop the Earth’s atmosphere into chunks a few hundred miles across or less. In both cases, processes which happen in regions smaller than this resolution cannot be treated directly.
For climate models, one important factor that operates on scales smaller than the model resolution is clouds. As a result, people sometimes claim that the models don’t include clouds, but that’s not the case. The effect of clouds is incorporated but in a simplified way, involving adding terms to the equations being calculated that should replicate the effect of clouds in each chunk of atmosphere. It’s not ideal but it’s certainly not an unreasonable approach to the resolution problem.
Validation
The acid test of any model in any field of science is whether or not it matches the observations. First, obviously, you have to have high quality observations to compare it with, but if you have that, then a model which doesn’t match the observations to a sensible degree of tolerance is a model which needs work.
Sometimes, the lack of agreement between theory and reality implies a problem with observations: in astronomy, Einstein tinkered with relativity theory to make it predict a static universe, in accordance with what was known at the time, but not long afterwards Edwin Hubble discovered that the universe was indeed expanding. In climate studies, one problem for a while was that satellite records didn’t seem to show very much warming at all in the troposphere, which was predicted by climate models to warm even faster than the surface. But it turned out that there were calibration problems with the satellite data that, once resolved, meant the predictions were consistent with reality after all.
A problem with any climate model projection is that you really need to look at its forecasts over several decades to properly assess how well it matches reality. But the earliest simulations were run only in the 1980s, so we only have a short baseline to work with. Those early models were necessarily somewhat crude, as they ran on computers much more basic than what we have today. So, generally, the more sophisticated the model, the less time we’ve had to evaluate it.
But all models in general use can match past climates well - otherwise they wouldn’t be in general use. And even the early crude models have made predictions which have been borne out. Hansen et al (1988) predicted a rise in temperature of about 0.8°C between 1984 (when their model simulation began) and 2005; the observed value was about 0.7°C. Models have also predicted that the stratosphere would cool as the surface warmed; that the North pole would be the site of some of the greatest warming; that large volcanic eruptions would cause sharp but short-lived drops in global temperatures; and that ocean surface temperatures would warm. All these predictions have been borne out.
Summary
To summarise: modelling is an essential approach to understanding a complex physical system. There are limitations that scientists have to be aware of, but these do not invalidate the approach. Modelling of the climate system has been successful in reproducing past climate variations, and has also been successful in predicting responses to increasing concentrations of greenhouse gases.
Leave a Reply