Tuesday, September 11, 2012

Climate Model Fail

I keep track of Arctic sea ice because of its implications for the weather where I live in inland North America. The summer cold fronts that bring the lion’s share of the rain here depend on a source of cold air in summer, and that ultimately traces back to the ice. When the ice goes away, I imagine, we will have fewer cool nights and less rain in summer.

Other people are watching the Arctic ice for other reasons. For many, it is a chance to validate global climate models. These complex mathematical models predict changes in climate everywhere in the world. In recent days, ice-watchers have been pointing me to the ice maps predicted by some of the more reputable global climate models. These maps show a remarkable match to today’s actual ice map. When you stop to consider that the current state of the Arctic sea ice is something that has never before been seen, it is impressive how precisely it could be predicted.

Yet there is a problem. The current state of the ice was predicted, but not on the correct time scale, not even close. The maps are about right, that is, but the dates are way off. For the first climate model I saw, the ice map that matched today’s ice was dated 2075. In the next, the date was 2099. Another one: 2200. All these predictions were made between 5 and 10 years ago, and they are among the most accurate of the currently reputable climate models, based on what we know through today. Other climate models made predictions that can fairly be said to have been discredited already; for example, they predicted that Arctic ice would never decline to the levels we are seeing today. But even the more accurate models are having trouble. Events that were supposed to take a century to occur have happened already, in less than ten years.

None of this comes as a shock to ice-watchers. For years it has been obvious, in this arena at least, that reputable science is not keeping up with the pace of events. This month, with Arctic sea ice at an all-time record low, a few ice-watchers have been gleefully deconstructing some of the climate models whose predictions were particularly far off.

This should be a moment of triumph for the climate models that accurately predicted the Arctic ice collapse of 2007–2012. It should be, but if there were any such models, no one has come forward to talk about them. Indeed, any such prediction would have been rejected by reputable science even after the events of 2007. There was just one high-profile prediction of an ice-free Arctic as soon as 2013. Scientists were stunned that such a prediction could come from a research team known not to be crackpots, but they were more inclined to forgive what seemed like an obvious error than to take the prediction seriously. Here is one of the reasons why: the prediction of a melt-out around 2013 was not done with a complex climate model, but with a simple statistical extrapolation, something anyone can do at home with a spreadsheet program. Yet this prediction is the one that has come close to being accurate.

And so, as far as we can say at this point, there are no successful climate models. None were as accurate as the “naive” approach of looking at the graphs and projecting the trends forward. The global climate models can be reset using the latest available data, but that will not necessarily improve their predictive power. That is, they could be just as far off in the next ten years as they were in the last ten.

If all the climate models have failed, then we just don’t know where global climate is headed. Perhaps there are so many unknowns that modeling is futile. As one example, no one has even a vague idea of the mass of greenhouse gases being released by melting permafrost in Siberia. If that variable is unknown, then the current state of the system is unknown, and the most detailed predictions are really just guesses.

Based on the performance of climate models so far, it seems safe to say only this: climate change will continue, and some of the changes will happen faster than any of the reputable scientific models are predicting.