Unsupervised AI Reveals Hidden Precursors to Some Major Earthquakes
Researchers used unsupervised machine learning to group earthquakes into interacting “families” and identify three key precursor traits—stronger clustering, greater localization, and increased stress release—that sometimes appear weeks to months before large quakes (e.g., Kahramanmaraş, Iquique, L’Aquila). The method did not detect such signals for all earthquakes (notably Amatrice and Noto), highlighting both promise for real-time forecasting and the limits of deterministic prediction. The approach, tested prospectively within regions, aims to improve monitoring and understanding of when preparatory phases emerge.
