Climate chaos

By Jon Cartwright

A few years back after a particularly wet spell, Lenny Smith found that his local pub in Oxford had been flooded. This was of course bad news for Smith’s beer provisions, but for the pub’s owner the more desperate issue was that the flood waters had spilled into the basement and wrecked thousands of pounds worth of new kitchen equipment. So, asks Smith, how much should the owner spend on his kitchen next time? This is the sort of real-world decision related to climate change that he thinks can too easily be forgotten by climate modellers.

I was listening to Smith, a mathematical physicist (and an expert on chaos theory) at the University of Oxford, talk at the international conference on Climate Change Impacts and Adaptation at the University of Exeter, UK, yesterday. He was arguing that it is fruitless to mindlessly improve climate models in all areas. Rather, he said, we should investigate how robust current climate models are by checking for consistencies among them.

As an example, Smith quoted an article written in 1921: more CO₂, it read, gives “warmer, wetter winters with increased (larger) storminess”. Given that this is still the main point that underlies scientists’ current thinking, he said, “why do we need models?”

This, the audience at his talk realized, was only half a joke. Of course models are in principal integral to forecasting, but Smith wanted to point out that we cannot know whether proposed improvements to climate models — higher resolutions, accountability of cloud cover, etc — will actually guide us towards a better understanding of how climate works. “This naïve physicists’ idea — that I share — that we are converging on a knowledge of what’s going on out there is completely unfounded,” he said.

Going back to Smith’s local drinking hole, the most important question we need to ask, he said, is how reliable our current models are — because only this can tell us what we should do now. To do this we must evaluate the probability distributions associated with the models’ predictions and look for places where they do not overlap. If any of these places exist, it would be a sure sign that something isn’t right.