AI is good at weather forecasting. Can it predict freak weather events?
Peer-Reviewed Publication
Updates every hour. Last Updated: 28-Jun-2025 04:10 ET (28-Jun-2025 08:10 GMT/UTC)
A new study from scientists from the University of Chicago found that neural networks cannot forecast “gray swan” weather events, which might not appear in existing training data but could still happen—like 200-year floods or massive hurricanes.
In a paper published in SCIENCE CHINA Earth Sciences, a team of researchers conducted a comprehensive study of wildfires during the Paleocene-Eocene Thermal Maximum (PETM) in the Nanyang and Fushun Basins. The study indicates that a “low wildfire activity” prevailed during the main body of the PETM in most regions in the Northern Hemisphere, resulting from extreme warming, high precipitation, and the resultant changes in vegetation types. During the PETM recovery phase, it gradually evolved into a “high wildfire state”, and the resulting black carbon burial might have significantly contributed to the sequestration of massive light carbon. It provides a novel perspective to address current wildfire threats and to predict wildfire activities and their driving mechanisms in the context of future global warming.
The puzzling behaviour of Titan’s atmosphere has been revealed by researchers at the University of Bristol for the first time.
Clownfish have been shown to shrink in order to survive heat stress and avoid social conflict, Newcastle University research reveals.