This paper explores the use of prompt elastogravity signals (PEGS) for real-time earthquake magnitude estimation, particularly for large earthquakes (M>8) where traditional seismic-based early warning systems often fail. The authors develop a deep learning model, PEGSNet, trained on synthetic PEGS waveforms augmented with empirical noise, to track earthquake growth instantaneously from data recorded before the arrival of seismic waves. Results show PEGSNet accurately tracks earthquake moment release for M_w > 8.6, providing a significant advancement in tsunami early warning capabilities.

Andrea Licciardi, Quentin Bletery, Bertrand Rouet-Leduc, Jean-Paul Ampuero, Kévin Juhel