I recently attended a panel on the role of science in society. A highlight of the evening was hearing one of the panelists, a biologist, talk about her experience of being asked for an interview about climate change by a TV channel because she was “not biased”. She, of course, responded that the channel would do best by asking an expert in climate modeling instead of her, since they would have the needed expertise. Her point was that just because you happen to be an expert in something doesn’t mean you’re necessarily biased (and that asking a non-expert showed a lack of understanding on the channel’s part).
Well, in today’s post I’m going to try to do what she refused to do, in other words, provide my thoughts on climate modeling without having any expertise in it. The reason why I feel it is reasonable for me to do so is because of my expertise with mathematical modeling in general. However, in order to ground myself in the subject matter at hand I’ll frame this post as a discussion of this article in The Economist.
The first point the article makes is that there is a mismatch between the predictions of climate models and the temperature trends observed over the past decade, namely, that “surface temperatures since 2005 are already at the low end of the range of projections derived from 20 climate models”. This is used as a way of bringing the models’ validity into question. I disagree with this statement. Although agreement with observational data is frequently used as a litmus test for the validity of a mathematical model, such agreement is neither necessary nor sufficient for a model to be useful. A model’s main task is to provide new insights into the behavior of a system, not to mimic it exactly. In fact, when a model fits the (historic) data too closely, this may be a warning sign that it is being “overfitted”, which means that the model’s parameters are selected to produce trends too similar to the input data. This may be a problem if the model is going to be used beyond its original context (see, for example, the Titus-Bode law which appears to be simply a nice coincidence). At the same time, a model that can predict critical features such as the periodic nature of climate trends without having them “built in” may be providing valuable insights without necessarily producing perfect agreement with data. Finally, agreement with data needs to be evaluated on a longer timescale; being off for one decade is not a model’s death knell…
The second, related, point the article makes is that there are two different kinds of models used to represent climate, namely, bottom-up models known as “general circulation models”, and top-down models known as “energy balance models”. The former are essentially mechanistic models, detailing the processes by which different factors influence climate, including various feedback loops. According to the article, “Their disadvantage is that they do not respond to new temperature readings.” The latter are less complex, and “do not try to describe the complexities of the climate”, but they do “explicitly use temperature data to estimate the sensitivity of the climate system”. Thankfully, the article does not make any conclusions about which type of model is “better”. Indeed, it shouldn’t. Different types of models have different ranges of applicability. While I don’t claim to understand the complexities of the distinctions between the two types of climate models I can draw a parallel with biochemical system models, with which I have extensive experience from my PhD days. There are also two main classes of models there: the more complex kinetic models, and the simpler flux-balance models. Kinetic models are detailed and describe exactly the rates at which each reaction happens and how the concentrations of the various molecules change over time. Their drawback is that they require a lot of prior knowledge which is often difficult to obtain experimentally, such as kinetic constants. Flux-balance models, on the other hand, require a lot less detail, and simply assume that the concentrations of all the molecules have reached a steady state and derive conclusions from there. Neither class of models is “perfect”. The main point, however, is that both types of models provide valuable information and insights about the biochemical system being analyzed, even though they do answer somewhat different questions about it.
The third point is that there is disagreement about the actual value of “equilibrium climate sensitivity”, the long-term amount of temperature increase due to a doubling of atmospheric CO2 levels. Curiously, the article cites estimates by Julia Hargreaves published in 2012, that a cursory search of her website’s publications page has failed to turn up (let me know if you have better luck), as well as “Nic Lewis, an independent climate scientist”, who turns out to be a semiretired financier with a background in math and physics with only one published paper on the subject. It is disappointing that The Economist uses a less-than-credible source like Nic Lewis on par with actual climate scientists. Granted, there may well be disagreement about the actual value, as there should be, but evidence needs to be properly weighted. Call me elitist, but I don’t see how giving equal weight to an amateur scientist with one publication and an expert in the field makes sense. Another source that is mentioned as providing evidence for a lower climate sensitivity is an “unpublished report by the Research Council of Norway, a government-funded body”. If there is one thing that I don’t think should belong to an article about science, it is unpublished reports (that being said there is a recent article by the team behind this research that’s freely accessible). My point here is that not all sources are made equal, and even if peer review does not always guarantee sensible publications, The Economist appears to exhibit a surprising bias in favor of dubious sources.
Another point that I want to make based on my own recent experience with epidemiological models of infectious diseases in Sub-Saharan Africa is that reconciling the predictions of multiple models (or even understanding the sources of the discrepancies between them) is extremely challenging albeit necessary to inform policy decisions (see this article examining the impact of antiretroviral therapy on HIV in South Africa predicted by 12 different models). Nevertheless, if a large fraction of the models predict a particular range of results, the laws of probability suggest that the correct answer is within that range even when several new models appear to give predictions outside that range. Of course, the really hard decision that needs to be made is on the policy level, which in the case of climate change largely falls into the buckets of “mitigation”, meaning a significant reduction in CO2 emissions, and “adaptation”, meaning adjustment to the change in climate, which would make sense if that change were less severe. If I know little about climate science, I know even less about policy, but in this case I tend to agree with William Nordhaus of Yale University, quoted in this article as supporting drastic interventions as a sort of disaster insurance protecting humanity against fairly unlikely events with catastrophic consequences.
In conclusion, The Economist does an excellent job of describing the scientific challenges of modeling climate in simple terms, but appears to overestimate or overstate the disagreement between different models, partly based on its use of less-than-credible sources. It also highlights the difficulty of taking a scientific conclusion and translating it into policy. But it seems to be missing some fundamental points about models by evaluating them purely based on their agreement with observations. Furthermore, it is misleading in suggesting that newer models may be more accurate than older ones, when they actually seem to be based on a less detailed, rather than more detailed, representation of the system in question. I hope that, by educating the general public about the development and use of mathematical models, as well as the conclusions that can and cannot be drawn from their results, we mathematicians and scientists may one day help it reach a level of understanding that will make such articles superfluous.