Self-supervised learning of seismological data reveals new eruptive sequences at the Mayotte submarine volcano
Résumé
Continuous seismological observations provide valuable insights to deepen our understanding of geological processes and geohazards. We present a systematic analysis of two months of seismological records using an AI-based Self-Supervised Learning (SSL) approach revealing previously undetected seismic events whose physical causes remain unknown but that are all associated with the dynamics of the Mayotte submarine volcano. Our approach detects and classifies known and new event types, including two previously unknown eruptive sequences displaying properties similar to other sequences observed at underwater and aerial volcanoes. The clustering workflow identifies seismic events that would be difficult to observe using conventional classification approaches. Our findings contribute to the understanding of submarine eruptive processes and the rare documentation of such events. We further demonstrate the potential of SSL methods for the analysis of seismological records, providing a synoptic view and facilitating the discovery of rarely observed events. This approach has wide applications for the comprehensive exploration of diverse geophysical datasets.
Domaines
Planète et Univers [physics]| Origine | Fichiers produits par l'(les) auteur(s) |
|---|