Despite improvements in daily weather forecasts, predicting events that may occur only once every 1,000 years remains a major challenge, particularly the most deadly heatwaves.
Traditional supercomputer-based models can predict such events, but they require significant amounts of time and energy. Newer AI-based forecasting models perform well in everyday forecasts but often struggle with extreme events that were not represented in their training data.
An international team of researchers in the US and France, led by members of the Climate and Data Theory Group at the University of Chicago, has developed a new hybrid method to address the problem.
The method, published in Physical Review Letters, combines the efficiency of AI tools with the reliability of traditional models to help predict the probability of rare events quickly and accurately while using significantly fewer resources.
“The strength of this method,” said Pedram Hassanzadeh, associate professor of geophysical sciences, “is that it combines the strengths of both AI and traditional physics and is particularly effective for extreme phenomena, which are the hardest to simulate and have the greatest societal impact.”
The challenge of rare weather events
Heatwaves are among the deadliest forms of extreme weather.
A heatwave in Europe in 2003 resulted in around 70,000 deaths, while Russia recorded 56,000 deaths in 2010. Last June, almost half of the US population, around 180 million people, experienced dangerous temperatures.
Such heatwaves are becoming more frequent and severe, but their extreme nature makes them difficult to study and predict.
Forecasting has traditionally relied on physics-based climate and weather models. These models help researchers understand how different conditions, such as atmospheric pressure, can affect temperature and other variables over time.
They calculate many possible scenarios, allowing researchers to determine which outcomes are most likely to occur.
AI+RES reduces simulations
Researchers can address some of the challenges by using a statistical technique known as rare event sampling (RES).
The method speeds up the process by scoring conditions, allowing the climate model to focus on the most promising forecasts while disregarding others.
However, RES does not work particularly well for short-duration events, such as week-long heatwaves, compared with an entire season that is unusually warm.
“AI weather and climate models are one of the great achievements of AI in science, but they are not magic — they fail on grey swans, the rarest and most extreme events,” Hassanzadeh said.
To address the problem, the team developed a new method called AI+RES.
The approach enhances the scoring mechanism by adding AI’s ability to predict which conditions are more likely to lead to smaller, rapidly developing extreme events.
To test the method, the researchers ran 50,000 simulations using a traditional climate model to predict heatwaves in regions of France and the US Midwest.
The new AI+RES method produced almost identical results while requiring just one-hundredth of the simulations.
Potential applications beyond heatwaves
As the work was a proof of concept, the researchers used a model that did not account for climate change, which adds another layer of complexity.
They hope to test the method using models operating under different climate change scenarios to determine how results could change as the planet warms.
According to the researchers, the hybrid method could also be applied to other severe weather events, including tropical cyclones and extreme rainfall.
“AI models trained on real-world weather observations have already been built and are being used, so the next step is to connect a state-of-the-art numerical weather or climate prediction model with one of these through this algorithm,” Hassanzadeh explained.
This could give decision-makers access to accurate information on the frequency of severe storms and heatwaves under current and future climate conditions at a regional level, including in areas such as Texas, Florida and California.
Also read: British couple killed in Sifnos helicopter crash identified
For more videos and updates, check out our YouTube channel
