AI Tools Deliver Near-Instant Aftershock Risk Predictions Following Earthquakes

An international consortium of researchers, involving experts from the University of Edinburgh, the British Geological Survey (BGS), and the University of Padua, has successfully developed an artificial intelligence (AI) tool designed to forecast aftershock hazards. This groundbreaking technology, with its findings detailed in late 2025, offers the capability to assess the probability and expected number of aftershocks measuring magnitude 4.0 or greater within the initial 24 hours after a main seismic event, achieving this assessment almost instantaneously.

The core of this innovation is a deep learning system. This system was meticulously trained using extensive seismic datasets sourced from five regions known for significant seismic activity: California, New Zealand, Italy, Japan, and Greece. The performance of this new AI development has been shown to be comparable in predictive accuracy to the established Epidemic-Type Aftershock Sequence (ETAS) model, which currently serves as the operational benchmark in several nations. Dr. Fotini Dervisis, the lead researcher—a postgraduate student affiliated with the University of Edinburgh and BGS—highlighted the paramount advantage of the AI solution: speed. Where the traditional ETAS model might necessitate hours or even days to complete thousands of necessary simulations, the AI models deliver their results in mere seconds.

The immediacy of these forecasts is absolutely crucial in disaster zones. Aftershocks pose a severe threat, capable of causing the collapse of already compromised structures and seriously impeding the efforts of emergency rescue teams. A stark reminder of this need for rapid assessment is the devastating earthquake that struck Turkey in February 2023, an event that underscored the critical requirement for swift tools to manage crisis response effectively.

Researchers are now looking ahead to the next frontier in operational seismology. This involves integrating these advanced AI models, such as SmaAt-UNet and Earthformer, with high-precision seismic catalogs that are being generated by machine learning algorithms in near real-time. This integration signals a significant shift away from time-consuming, resource-intensive seismological analysis toward the near-instantaneous delivery of vital information to decision-makers. This important work received partial backing through the European Union’s Horizon 2020 program, specifically under the Marie Skłodowska-Curie Innovative Training Network.

The implications of this technological leap are profound for disaster preparedness worldwide. By drastically cutting down the prediction timeline, emergency management agencies can issue more timely and targeted warnings. This speed allows for quicker evacuation advisories or the strategic deployment of search and rescue personnel before the next major tremor hits. Essentially, this AI represents a paradigm shift, moving seismic hazard assessment from a slow, retrospective analysis to a proactive, seconds-long calculation.