New Hybrid Method Combines AI and Physics to Forecast Once-in-a-Millennium Weather Events

Forecasting events that might happen only once in 1,000 years, such as the deadliest heat waves, remains difficult even as day-to-day weather forecasting has improved. Traditional supercomputer-based physics models can forecast these events but require large amounts of time and energy, while newer AI-based forecasting models are good at everyday forecasts but often fail to predict rare, extreme events that were not well represented 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, a University of Chicago associate professor of geophysical sciences. "Detailed physics-based models can capture extremes, but they require prohibitively large amounts of time and energy."

An international team of researchers in the United States and France, co-led by members of Hassanzadeh's Climate Extremes Theory and Data Group, developed a hybrid method, published in Physical Review Letters, that combines the efficiency of AI tools with the reliability of traditional physics-based models. The method is designed to predict the odds of rare, extreme events quickly and accurately while using far fewer computing resources than physics-based models alone.

Hassanzadeh said the method "combines the strengths of both AI and traditional physics and is particularly effective for extreme events, which are the hardest to simulate and have the greatest societal impact." The researchers cited the 2010 heat wave in Moscow, Russia, which worsened wildfires and pushed pollution levels to ten times normal in the capital, as an example of the kind of rare, extreme event the method aims to help forecast.