Thanks to advances in weather simulation, forecasts of heatwaves and hurricanes could soon come with information about the extent to which they were fuelled by climate change.
A revolution in weather forecasting could soon see warnings of forthcoming heatwaves, storms or other extreme events accompanied by specific information on the role climate change has played in fuelling them, as meteorologists seek new ways to bring the reality of our warming planet home to the general public.
 |
| A new kind of weather forecast could tell us beforehand whether extreme events, like the 2018 Hurricane Florence, were made worse by climate change Nathan Congleton/NBCU Photo Bank/NBCUniversal via Getty Images |
Key to this idea is the growing field of attribution science, which involves examining extreme weather events after the fact to quantify the impact of climate change. It involves simulating an event twice, first under real-world conditions and then again in a fictional world where there is no human-caused climate change. The difference between those two scenarios reveals the extent to which rising emissions made matters worse.
Such attribution used to takes months, but techniques have improved dramatically in recent years, to the point where it can be done just days after an event. But now, researchers want to go even further and start to apply attribution science to events before they happen.
At the UK weather service, the Met Office, Peter Stott and his colleagues are doing this with high-resolution, state-of-the-art weather forecasting models, rather than the more grainy climate models used for after-the-event attribution. In essence, they are using the models to compare real-world forecasts with ones based on a fictional world unaffected by human influence. “We are very interested in this forecast attribution idea,” he says.
The goal, says Stott, is to provide the public with weather and climate information at the same time. “When [the public]get information about damaging weather, it’s all integrated together,” he says. “So, the forecast, the information about how this relates to climate change, and the impacts it means for them locally… all of that is coherent and consistent… and available in advance of what’s about to hit people.”
While Stott’s team is still testing the approach, a pilot of sorts has already inadvertently taken place. In 2022, the UK experienced record-breaking temperatures over 40°C. By chance, the Met Office had recently completed an analysis looking at how climate change had influenced the chances of the UK seeing such extreme heat. That attribution analysis was included in public messaging of the heat forecasts, says Stott, to “support the narrative” that this was rare, dangerous weather for the UK. “We brought in the attribution to say the 40°C heat was not just unprecedented, but something that you wouldn’t expect without climate change.”
However, using traditional weather forecasting models to run attribution analysis is time-consuming and requires access to specialist expertise and computing systems because they involve crunching through large numbers of physics-based calculations. Bernat Jimenez-Esteve at the Institute of GeoSciences, Madrid, Spain, has experimented with a different approach, using AI-powered weather forecasts instead. These AI models are trained on data from traditional physics-based climate models, but then use statistical extrapolations to make weather forecasts.
Like the Met Office’s approach, using AI for attribution analysis means forecasting extreme events under real-world and fictional pre-industrial conditions to compare the difference, but it can be done much quicker, says Jimenez-Esteve. “It’s at least two orders of magnitude faster than a conventional operational weather model.”
Together with colleagues, he ran the AI-enabled approach on three past extreme events: the 2018 Iberian heatwave, Hurricane Florence, which hit the US in 2018, and Storm Ciaran, which swept through the UK and other parts of northern Europe in 2023.
The AI models correctly predicted that all three extreme events would occur, but underestimated their size compared to reality. The forecast-based attributions concluded that climate change made the Iberian heatwave about 1.5°C warmer than it would have been in a pre-industrial world, while Hurricane Florence was stronger, wetter and slightly larger than it would have been, and Storm Ciaran had stronger winds.
Jimenez-Esteve says the approach is currently most accurate when considering heatwaves. It is more limited for complex systems such as a tropical storm, in part because the models used to train the AIs weren’t high enough resolution to capture small scale processes such as those occurring at the centre of a hurricane.
That is one of the drawbacks of using AI systems trained on climate models, says Stott, rather than high-resolution weather forecasting models. “The dataset they are using is on a coarser grid than we would use on our weather forecasts,” he says. “So you are missing some of the details of the weather, and that could be crucial not just for forecasting, but also for the attribution.”
 |
| Storm Ciaran in 2023 was made stronger by climate change Andrew Aitchison/In pictures via Getty Images |
But Jimenez-Esteve says AI weather forecasts are rapidly improving. “This field is evolving very fast,” he says. “Only a few years ago there were not even AI models able to do weather forecasting… I could see that they will get as good as conventional weather and climate models.”
He argues this AI-based approach would allow attribution analysis to be much more widely used around the world, expanding public climate communication and improving scientific understanding of how extreme events are changing.
There is a wider issue with conducting attribution analysis on weather forecasts, says Clair Barnes at University College London, who is a member of the World Weather Attribution initiative that performs rapid attribution analyses after extreme events globally. Feedback effects from a warming world, such as rising sea temperatures or warmer soils, aren’t necessarily accounted for by weather models that just tweak atmospheric conditions, she points out, meaning they could provide an incomplete picture. “If you are only changing the atmospheric state, then you are only capturing the effects of the change in the atmospheric state,” she says.
Barnes is also concerned about forecasts backfiring if a predicted event fails to emerge. “If you are communicating an attribution result to an event that doesn’t actually happen, what does that do to people’s understanding of how people perceive the field, perceive climate change?” she says.
Nevertheless, she thinks such additions to weather forecasts could be a powerful communication tool. “The dream for forecast attribution would be to have meteorologists on the weather forecast, saying ‘there’s going to be a heatwave and it’s going to be this much worse because of climate change’,” she says. “Because when you are trying to communicate the effects of climate change to people, that kind of thing can be really powerful.”