How to make earthquake and tsunami early warning faster? We designed an AI algorithm that estimates almost instantaneously the magnitude of an ongoing earthquake from a recently identified signal. This signal is a perturbation of the gravity field (that we call PEGS, for Prompt Elasto-gravity signal) induced by the seismic rupture and is extremely weak (about one million times smaller than seismic waves). However it has the extraordinary property to propagate at the speed of light (much faster than seismic waves). As a result, our algorithm, PEGSNet, can track almost instantaneously the magnitude of large (Magnitude ≥ 8.3) earthquakes (Licciardi et al., Nature , 2022):
Embedded Files
Illustration of PEGSNet, a PEGS-based AI algorithm tracking instantaneously the magnitude of large earthquakes (credit: Lina Jakaite)
Applied to Chile similar performances can be achieved only for Magnitudes ≥ 8.7 (Arias et al., JGR, 2023), while in Alaska, the exceptional seismic network density makes the approach work for Magnitude ≥ 7.8 events (Juhel et al., Nature Communications Earth & Environment, 2024).
PEGSNet's source code is available here.
Some outreach articles: Science (en), The Conversation (fr), RFI (radio, fr), le Figaro (fr)
We improved the efficiency and the versatility of the approach by changing the neural network architecture from a convolutional neural network to a graph neural network, which allows to leverage geometrical information. This updated algorithm, called PEGSGraph, made the approach work for Magnitude 7.6 events (Hourcade, Juhel and Bletery, JGR: Machine Learning, 2025).
PEGSGraph's source code is available here.
Google Sites
Report abuse