Particle.news

Download on the App Store

AI Model Learns to Detect Earthquakes From Fiber-Optic Data, Earning USGS Praise

The self-supervised DASFormer framework flags anomalies in continuous Distributed Acoustic Sensing streams without labeled examples.

Overview

  • Researchers from the University of Montreal, Woods Hole Oceanographic Institution and UC Berkeley introduced DASFormer in a study published July 15, 2025 in Visual Intelligence.
  • Distributed Acoustic Sensing repurposes existing fiber-optic cables into meter-scale seismic arrays, producing vast continuous data that has been difficult to label and analyze.
  • DASFormer pretrains on unlabeled recordings to learn predictable background patterns, so earthquake signals appear as sharp deviations the system can detect automatically.
  • USGS geophysicist Mark Petersen called the approach a fundamental advance for public safety, with potential value for architects, engineers and policy makers.
  • The system is not yet an operational early-warning tool and requires field-scale validation, integration with alert networks and performance testing, a need underscored by recent significant seismic events such as the September 13 M7.4 in Kamchatka.