Or, as hydrologist Keith Beven from the University of Lancaster, UK, put it, is modelling more than just an input sport? As co-inventor of Topmodel, a system that’s still in use after 30 years, Beven kicked off the debate by quoting fellow hydrologist George Box, who’s said that "as a hydrologist all our models are wrong but some may be useful".

Beven reckons that "we should treat models as uncertain multiple working hypotheses to be tested". He believes that researchers should be looking to test their models, but that’s only possible as far as uncertainties allow. "We can test data on the surface more easily than on the subsurface," he explained.

According to Beven, some of the key difficulties with modelling include computational limitations, limited measurement techniques, and "uncertainty about uncertainty estimation". He also quoted John Locke who in the 17th century said that "a madman is one who draws entirely reasonable conclusions from erroneous assumptions".

And of course, model quality is a compromise. "You can always reject models for not performing in one way or another," said Beven. "We have to allow some uncertainty – how much do you allow before you say the model is inadequate? That’s where fitness for purpose comes in."

Speaking from a climate modeller’s viewpoint, David Stainforth of the University of Exeter, UK, highlighted the three uses of such models: to understand the process of how the climate works, which is what most people use models for; to develop the models themselves (a slightly incestuous process, in his view); and to guide society in its decision-making. This last application is probably the least used but the most headline grabbing.

Stainforth explained how climate modellers are told policymakers need local and regional scale predictions for 40 years ahead to enable them to plan mitigation and adaptation. "We’re not sure we can do this," he said.

Model interpretation
Stainforth also explained the dangers of model overinterpretation and miscommunication. "We need to be careful about ensuring we have got useful information and understand it … before handing it out to other people," he said. "In the climate change field there’s sometimes an assumption that people know it’s a model and uncertain. There is a real need to be clear." In order to improve the communication of risk, Stainforth suggested the introduction of an intermediate layer of professionals to interface between politicians and scientists.

He is also concerned that overinterpretation of models could undermine the chances of getting better information in the future. "We need greater interaction with users at the design stage of models," he said.

Beven agreed that researchers should be much more careful in how they interpret their models. "That may mean decision-making uses the precautionary principle or takes a risk-averse approach rather than relying on a climate model," he said.

At the end of the debate, at least one thing was clear – model uncertainty is an uncertain issue.