Climate change projections have traditionally been made using complex climate models. Although these models provide state-of-the-art understanding on how the climate might change in future, it takes a lot of time and expertise to produce useable results, and they are also expensive to run.
To help overcome these challenges, the scientific community has been developing climate model emulators. Emulators are simple computer models programmed to reproduce the output of their more complex cousins, but are much faster and cheaper to run, allowing us to explore a far greater range of climate futures in a relatively short time.
But emulators come with their own challenges.
Our new study published in Geophysical Research Letters investigated how well one such emulator recreated the temperature projections from eight more complex climate models. Firstly, data from the complex models was used to set up the emulator, and then we tested how it reproduced these pre-defined climate futures.
Lead author Dr Lawrence Jackson, a CONSTRAIN researcher based at the Priestley International Centre for Climate at the University of Leeds, explained: “The results were pretty good – but then we found that the emulator did less well at recreating climate futures beyond this pre-defined range.”
This is important because emulators are often used to explore climate futures that are not available from more complex models. For example, the recent Intergovernmental Panel on Climate Change Sixth Assessment Report (IPCC AR6) used emulators to produce thousands of climate scenarios, which would have been too time-consuming and expensive using more complex models.
Dr Jackson added: “Our results highlight that emulator results need careful interpretation, especially when we don’t have more complex model results to directly compare them with.”
The team looked at how one emulator projected global average temperature change in a future where greenhouse gas emissions increase following a middle-of-the-road scenario in which historical patterns of development continue through this century.
In some cases, the emulator results differed from those of the complex models by more than 0.5°C – although this may sound small, half a degree makes a big difference in a warming world and, when it comes to planning for climate impacts, every fraction of a degree counts.
The size of the emulator errors also differed between the models, emissions scenarios, and timescales studied. In some cases, the errors even cancelled each other out, wrongly suggesting that the emulators were actually performing very well. Errors on a scale that could have a significant impact on climate policy were also found in scenarios where GHG emissions hover around current levels before starting to fall by mid-century.
Dr Jackson added: “This is another important step in establishing exactly how well climate model emulators project future climate change. It shows how we need to look at scenarios that haven’t been used to ‘train’ the emulators when it comes to evaluating how they perform.”
Still, emulators remain an indispensable tool for using cutting-edge science to create climate projections from models that are cheap and easy to run.