What is it about?
Researchers report that when Category 3-5 tropical cyclones are omitted from the training dataset of an AI weather model, the model is unable to forecast Category 5 storms. Model trained with Category 3-5 storms only in either the North Atlantic or Western Pacific basin is, however, able to forecast Category 5 storms in the other basin. Findings suggest that AI models can translate the learning across regions, which is encouraging. However, the struggle to forecast so-called “gray swans”, i.e., extreme events so rare they do not appear anywhere in training data. Innovations in AI+science will be needed to address this shortcoming.
Featured Image
Photo by NASA on Unsplash
Read the Original
This page is a summary of: Can AI weather models predict out-of-distribution gray swan tropical cyclones?, Proceedings of the National Academy of Sciences, May 2025, Proceedings of the National Academy of Sciences,
DOI: 10.1073/pnas.2420914122.
You can read the full text:
Contributors
The following have contributed to this page
