← Back to Gallery

Detecting and Locating Earthquakes using Machine Learning Workflow and Offshore Distributed Acoustic Sensing

Offshore distributed acoustic sensing (DAS) has emerged as a powerful tool for seismic monitoring, but its application to continuous regional earthquake detection and location faces several major challenges. First, the earthquake data on offshore DAS is often contaminated by complex noise signals originating from ocean dynamics and instrumentation. Second, the spatial coverage of DAS observations is constrained by the sensing range. Third, the limited availability of offshore dark fibers for long-term deployments poses practical challenges.

To address these issues, we developed a machine-learning-based workflow that denoises DAS data, picks seismo-acoustic phases from both DAS and ocean-bottom seismometer (OBS) recordings, and integrates DAS and seismic networks for earthquake association and location. Using DAS recordings from fiber-optic cables in the Cook Inlet, Alaska, the denoising model was trained with randomly masked channels. The denoised data exhibited a 2.5 dB average increase in S-wave signal-to-noise ratio, enabling 2.7 times more S picks for 1 month of cataloged earthquakes compared to the raw data. Over six months, the generated earthquake catalog delineated the Alaska subduction zone plate interface from the surface to a depth of 150 km, revealing ~50% new events not included in the ANSS catalog.

Building on this work, we explored the feasibility of using optical multiplexing to overcome the limitations of dark fiber availability. In May 2024, we conducted a four-day DAS experiment on the Ocean Observatories Initiative’s Regional Cabled Array offshore Oregon, employing an L-band DAS interrogation with optical multiplexing. The collected data maintained the same high quality as traditional dark fiber DAS while preserving full telecommunication functionality. Applying the same monitoring workflow, we identified T waves and the S waves of 31 regional earthquakes, demonstrating the potential of multiplexed DAS for continuous seismic monitoring. These studies highlight the transformative potential of advanced machine learning workflows to improve offshore earthquake monitoring.

Session: Fiber-optic Sensing Applications in Seismology - II

Type: Oral

Room: Key Ballroom 9

Date: 4/15/2025

Presentation Time: 11:00 AM (local time)

Presenting Author: Qibin Shi

Student Presenter: No

Invited Presentation: 

Poster Number:

Additional Authors