A new hybrid method merges artificial intelligence with physics-based climate models to forecast extreme heat waves that occur roughly once every 1,000 years, using a fraction of the computing power traditionally required. The breakthrough could help governments prepare for the deadliest weather events with far greater speed and accuracy.

Why Can't AI and Traditional Models Predict Extreme Weather on Their Own?

Weather forecasting has improved dramatically over recent decades, but a stubborn problem remains: predicting rare, catastrophic events. Heat waves are among the deadliest forms of extreme weather. In 2003, a heat wave killed roughly 70,000 people across Europe, and Russia experienced 56,000 deaths in 2010. More recently, nearly half of the United States, roughly 180 million people, experienced dangerous temperatures in June 2026.

Traditional physics-based climate models can capture these extremes by simulating how atmospheric pressure, temperature, and other variables interact over time. However, they require enormous computational resources. To predict whether Chicago might reach 105 degrees Fahrenheit, a model must run thousands of simulations before capturing that rare outcome. Modern AI weather models, by contrast, excel at day-to-day forecasts but often fail on outlier events because those extremes rarely appeared in their training data.

"AI weather and climate models are one of the great achievements of AI in science, but they're not magical; they fail on gray swans, the rarest and most extreme events," said Pedram Hassanzadeh, an associate professor of geophysical sciences at the University of Chicago.

Pedram Hassanzadeh, Associate Professor of Geophysical Sciences, University of Chicago

How Does the New AI+RES Method Work?

An international team of researchers, co-led by members of Hassanzadeh's Climate Extremes Theory and Data Group, developed a solution called AI+RES, published in Physical Review Letters in August 2026. The method combines the efficiency of AI with the trustworthiness of traditional physics models.

The approach builds on a statistical technique called rare event sampling (RES), which speeds up forecasting by scoring atmospheric conditions so the climate model focuses only on the most promising scenarios. The innovation: AI boosts the scoring mechanism by predicting which conditions are most likely to trigger shorter, rapidly developing extremes like week-long heat waves.

"After doing this iteratively, you eventually get to a bunch of simulations that do indeed capture whatever rare extreme event that you're interested in. The more you get, the better you can estimate the probability of that event, and that ultimately gives you a lot more certainty," explained Alexander Wikner, a Schmidt AI in Science Postdoctoral Fellow in Hassanzadeh's group and co-first author on the study.

Alexander Wikner, Schmidt AI in Science Postdoctoral Fellow, University of Chicago

What Were the Results?

To test the method, the team ran 50,000 simulations using a traditional climate model to predict heat waves over areas of France and the U.S. Midwest. Their new AI+RES method produced nearly identical results using just 500 simulations, a reduction of 99 percent. This dramatic efficiency gain means researchers can generate accurate rare-event forecasts in hours rather than days, using far less energy.

The researchers note that their proof-of-concept study used a model that did not account for climate change, which adds another layer of complexity. They plan to test the method on models running under different climate change scenarios to see how results shift as the planet warms.

How Could This Help Governments and Communities?

The practical applications extend beyond heat waves. The hybrid method could be applied to other severe weather events, including tropical cyclones and extreme precipitation. Because AI models trained on real weather observations are already in use, the next step is connecting state-of-the-art numerical weather and climate prediction models to these AI systems through the new algorithm.

This would give decision-makers access to critical information at regional scales, such as the frequency of strong storms and heat waves in current and future climates specifically over Texas, Florida, or California. Here are the key ways this technology could reshape climate adaptation:

• Regional Risk Assessment: Governments can identify which areas face the highest probability of extreme events, enabling targeted infrastructure investments and evacuation planning.
• Climate Adaptation Planning: Federal, state, and local officials can use accurate frequency data to design resilience strategies, from cooling centers to updated building codes.
• Resource Efficiency: By requiring 100 times fewer simulations, the method reduces the computational energy needed, making climate forecasting more sustainable.
• Improved AI Training: The method could generate rare-event datasets to train even better AI models, creating a feedback loop that accelerates forecasting improvements.

"This is exactly the kind of information federal, state, and local governments need as a first step for climate adaptation and mitigation planning. It's very exciting that we could be scaling this up into real-world models that directly provide such information to the public and to policymakers," noted Hassanzadeh.

Pedram Hassanzadeh, Associate Professor of Geophysical Sciences, University of Chicago

What's Next for This Research?

The study was supported by the Eric and Wendy Schmidt AI in Science Fellowship, the France-Chicago Center, the U.S. National Science Foundation, and other institutions. Researchers from École Normale Supérieure in Paris, Réseau de Transport d'Electricité (RTE) in Paris, the World Energy and Meteorology Council in Norwich, UK, and the Courant Institute of New York University contributed to the work.

The team's next steps involve scaling the method to real-world climate models and testing it under various climate change scenarios. If successful, this hybrid approach could become a standard tool for meteorological agencies worldwide, providing communities with the advance warning they need to prepare for the most dangerous weather events on the horizon.