Predicting the future lies at the heart of responding to climate change. We want to know what things will look like if we take certain actions to reduce greenhouse gas emissions – and what they will look like if we don’t. The difference between the two provides the basis for weighing up the various climate policies that might be on the table.
But we aren’t drawn to predictions just because we’re interested in policy choices. Climate change is scary. We’re now regularly confronted with evidence of its impacts: droughts, heatwaves, wildfires, floods and so on. Predictions provide a peculiar sort of comfort: at least we have some idea of what the outcomes might look like. We can see what’s over the cliff rather than just imagining a void. Faced with the daunting threat of a changing world, it just feels better to know what the future might look like.
But can we really predict the climate? There are certainly many people trying. They use computer models of the climate system, and in recent years we have seen money poured into making them ever more detailed. In July this year, a summit was convened in Berlin to discuss Earth Virtualization Engines or “Eve” with the organisers calling for massive investment – billions of euros – to buy huge new computers. The resulting projections would aim to tell us, for instance, the odds that in the 2070s we’ll see heatwaves and floods in Greece similar to those this summer – but maybe two degrees hotter and with 20% more intense rainfall. They’ll also attempt to show where specifically in the country would be most at risk.
As models become more complex they look increasingly realistic, so it’s tempting to believe the predictions will be good
Eve will no doubt capture the ears of governments and other funders, but is the rush to build bigger and bigger models the best use of resources to guide our response to climate change?
Basic science is enough to tell us that temperatures will continue to rise, and that this will lead to impacts in all parts of the world, from Italy to Australia – so we can already reliably predict the existence of the threat. But this doesn’t tell us how high to build flood defences or how to make changes to agricultural practices. Nor does it give detailed predictions of local changes that would make the arguments for climate policies personal and relatable.
Big computer models appear to provide these kinds of detailed predictions, but the complexity of the systems we’re dealing with gives us reason to question whether they actually do, and even whether more advanced ones ever could. The problem is that we don’t know how close to reality a model needs to be to make good predictions. The chaotic nature of our climate means that small differences in the representation of, for instance, Arctic sea ice, could have a very large impact on something distant and seemingly unrelated, such as the Indian summer monsoon.
Part of the problem is that as the models become more and more complex they often look increasingly realistic – so it is very tempting to believe that the predictions will be good. Climate models are assessed on how well they simulate the past. They’re set going in, say, 1850, provided with the subsequent observed changes in atmospheric greenhouse gases, and we judge how well they do based on how realistically they represent the next 170 years. But climate change is taking us into new, never-before experienced territory, so the past may not be an accurate guide. Even if the models can reproduce climate history, we shouldn’t expect them to reliably tell us about the strange new future we’re facing.
There’s no doubt that the latest climate models are outstanding achievements of computer-based science, but they aren’t equivalent to reality. They don’t represent all the stunning complexity of Earth’s many interlocking systems. They might be useful tools for research, but they’re not perfect representations of the real world.
So if our models can’t give us reliable, detailed climate predictions, what do we do? How do we know how high to build those flood defences? The answer is twofold. First, we have to relax and accept that we have incomplete information. Instead of trying to make our responses just right for the climate of the future we should seek out resilient and flexible solutions, remedies that will be robust in a wide range of possible climate outcomes. Flood defences could, for example, be designed to enable them to be easily extended if that becomes necessary.
Second, we need to use models better. They can’t provide precise predictions, but they can tell us what climate change might look like in a world that’s different, but nevertheless similar, to our own. Well-designed experiments could use them to get information about the scope of effects different responses could have. Give me £1bn for modelling and I wouldn’t be able tell you what’s going to happen, but I would get a better grasp of the uncertainties and the range of plausible futures. Knowledge of this range would help us design climate-resilient infrastructure and usefully set the context for debates. If all we knew was that a particular policy would increase the intensity of UK heatwaves by 2 to 4C, while another would increase them by 3 to 10C, then even though the uncertainties are large, the information is still a useful basis for making decisions.
The risk is that investment in ultra-high resolution models could actively undermine society’s response to climate change. They might encourage us to plan for highly specific scenarios rather than maintaining the flexibility to deal with a vaguer range of outcomes. Instead, we should treat simulations for what they are – not versions of reality, but research tools. The best possible future might be created by exploring many possible worlds.
• David Stainforth is a professorial research fellow at the London School of Economics. His book Predicting Our Climate Future will be published by Oxford on 12 October.
Escape from Model Land by Erica Thompson (Basic, £16.99)
The Climate Demon by R Saravanan (Cambridge, £29.99)
Chaos: A Very Short Introduction by Lenny Smith (Oxford, £8.99)