The researchers trained their AI on reconstructions of ancient climates from the year 0 to 1850.
Europe’s summers are getting hotter, longer and deadlier, but a new AI system could soon give scientists up to seven weeks’ notice before extreme heatwaves hit.
Researchers from the Euro-Mediterranean Centre on Climate Change (CMCC) have developed a machine learning model they say can predict severe heat events more accurately and more efficiently than current methods.
Their findings, published in the journal Communications Earth & Environment, could transform how Europe prepares for one of its most dangerous climate threats.
“[Machine learning] will become a fundamental part of how we study climate variability,” says study author Dr McAdam. “This study has demonstrated [its] usefulness in extreme event prediction, but it is only a first step in defining how we do that to receive interpretable and physically meaningful results.”
AI could offer a new edge in seasonal climate prediction
Heatwaves are among Europe’s deadliest climate hazards.
Devastating heat in 2003, 2010 and 2022 caused tens of thousands of deaths, crop losses, energy spikes and severe health crises. Scientists warn that events like these are only becoming longer, more intense and more frequent as the planet warms.
A 2024 analysis by Climate Resilience for All found that heat now lasts up to five monthsa year in some southern European cities, as temperatures remain above 32°C deep into autumn. This summer ranked among the hottest ever in Spain, for example.
But a global study from World Weather Attribution and Climate Central warned that the planet could face nearly two extra months of “superhot” days each year by 2100, too.
Against this backdrop, the researchers say early warning systems could save lives.
“Seasonal forecasts made in spring can, in principle, state whether a summer will be warmer than average,” says McAdam. “Early warning of extremely hot summers could help society prepare to mitigate against economic losses and reduce risk to life.”
How the system works
To make its forecasts, the CMCC team’s AI sifts through about 2,000 different climate clues that range from air temperature and ocean conditions to soil moisture to find the mix that best signals when and where heatwaves are likely to form. Once it identifies those key patterns, the system can generate heatwave forecasts across Europe.
According to the researchers, their approach matches, and in some cases outperforms, traditional forecasting systems – especially in northern Europe, where predictive skill has long been limited.
It also provides scientists with valuable insights about which environmental variables influence extreme heat the most.
The study found that both local conditions – such as how dry the soil is, how warm the region already is and how air moves over Europe – and distant ocean patterns help determine when Europe will experience a heatwave.
Detailed weather records only go back a few decades, so the researchers trained their AI on computer reconstructions of ancient climates stretching from the year 0 to 1850.
This gave the model hundreds of extra “virtual years” of weather to learn from. Even though the data came from a simulated planet rather than real observations, the AI was able to apply what it learned to modern conditions, accurately predicting actual heatwaves that occurred between 1993 and 2016.
Can AI make forecasting more accessible and widespread?
Traditional climate forecasts rely on enormous supercomputers that take days or even weeks to run complex models of the atmosphere. The CMCC team argues that its AI system can make its predictions using far less computing power.
Still, AI systems typically demand significant energy and water to power and cool the data centres behind them. CMCC’s report did not calculate the environmental cost of its AI. In terms of pure numbers, however, its accessibility means more research groups and government agencies can likely afford to use it.
As McAdam explains, their approach shows that machine learning can make reliable seasonal forecasts “using only a tiny fraction of the computational resources” needed for older methods.
By giving accurate warnings of extreme heat weeks before it strikes, this technology could help Europe plan ahead, protecting crops, easing pressure on power grids and giving its health services time to prepare for an increase in urgent care.
The benefits of this AI-powered tool could save lives, time and resources in other deadly events, too. The researchers believe the same framework could eventually be adapted to forecast other weather extremes, such as floods or droughts.
