Thursday, January 31, 2013

On Scientific Models

Interview with David Stainforth [climate scientist], 7 The Reasoner No. 2 (February [sic] 2, 2013):

KS [Katie Steele]: What is the significance of your emphasis on policy here? Are you saying that you now devote a lot of time to science communication, or, rather, that you approach your work as a climate scientist in a different way, i.e., with an eye to policy relevance?

DS: The latter. The attention to policy has lead to a shift in emphasis in my scientific work—from modelling and running simulations to the proper interpretation of the data output of these model simulations.

In the past I set up and did a lot of runs (simulations) of these large complex climate models called General Circulation Models [GCMs]. This involves a lot of time and a lot of hard work in getting these computer models up and running. . . these simulations are difficult to produce. But I have done my time in this respect. The climateprediction.net project that I was involved in is still running, and that’s great, but it is up to others now to facilitate the simulations.

The important issue for me now is this: these climate model simulations produce vast output, and there are so many questions about how to analyse these big data sets. . . In short, what does it all mean? Why run these simulations? We need to really think about what we can get out of these climate models and how the results should be presented. It is tempting to just keep making the models more and more complicated and apparently derive more and more detailed predications [sic] of the type that policy-makers want. Moreover, the power of computers has its own allure. . . such shiny sophisticated machines that seem to offer endless opportunities for fast and powerful problem-solving. . . for the mathematically-minded, there is a temptation to create more and more complicated models. We need to be very careful, however, about faithfully representing what we actually know about the future climate on the basis of model simulations.

KS: I see. So what do you think the models can be used for? Do they yield predictions? Is the ‘pulling back’ just a matter of being more modest about the precision of the predictions that we can derive from climate models? Or should we use climate models in quite a diff erent way altogether?

DS: I think it is best to think of climate models as research tools—they are useful for understanding interactions between diff erent parts of the climate system. Of course, we don’t want to give up on predicting the climate, but we need to be realistic about climate prediction, and in particular, multi-decadal climate prediction, which is of interest to policy-makers. Climate models should be seen as just one of the inputs that allow us to formulate scenarios of how the climate could change in response to diff erent  forcings. [Forcings are external forces that change the dynamics of the system; a prominent forcing is increased carbon dioxide emissions.] We should aim to formulate scenarios that collectively tell us how the climate could change over time, and give us a general indication of the sensitivity of the climate system.

The main point here is that the output of climate models should not be taken at face value—as predictions of future climate—and presented in more or less unadulterated form to the public and to policy-makers.

&&&

Is there a moral here for the economic analysis of evidence, inference, and proof? I think so.

&&&

The dynamic evidence page

Evidence marshaling software MarshalPlan


Post a Comment