Developed by a team in Edinburgh Universities, British Geological Survey, University of Padua In Italy, machine learning tools were trained using seismic data from seismic hotspots such as: California, New Zealand, Italy, Japan, Greece. The technique was applied to earthquakes of magnitude 4 and above and was tasked with predicting how many aftershocks would occur within 24 hours of the initial tremor.
Researchers said the newly developed AI model performed similarly to the Epidemic Aftershock Sequence (ETAS) model, a tool commonly used in earthquake-prone countries such as Italy, New Zealand and the United States. However, while both prediction systems produced similar results, the AI platform provided results almost instantly, whereas ETAS can take several hours to make computationally intensive predictions. The work will be published in a magazine earth, planet, space.
“This study shows that a machine learning model can generate aftershock predictions within seconds, with quality comparable to ETAS predictions,” said lead researcher Foteini Dervisi, a PhD student at the University of Edinburgh’s School of Earth Sciences and the British Geological Survey.
“Their speed and low computational cost provide significant benefits for operational use. Combined with the near real-time development of machine learning-based high-resolution seismic catalogs, these models will enhance our ability to monitor and understand seismic crises as they evolve.”
The researchers said that by training an AI tool on records of past earthquakes in regions with different tectonic movements, the model could be used to predict aftershock risk in most parts of the world where seismic activity occurs regularly. Additionally, the speed with which aftershock forecasts are provided could help authorities make decisions about public safety measures and resource allocation in disaster-hit areas.
This research was supported by the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie SPIN Innovative Training Network.
