The Pacific Northwest boasts an extensive network of more than 600 seismic monitoring stations that help researchers track tectonic and volcanic phenomena, including earthquakes.
This data provides key insights into regional faults and feeds into early warning systems, which can give a community precious moments to prepare before a natural disaster strikes. A significant threat to this region, however, sits miles offshore, where the Juan de Fuca plate is subducting beneath the North American plate, forming the Cascadia Subduction Zone.
Monitoring activity at ocean floor faults is challenging, and the existing methods don’t often yield enough data for detailed analyses.
To overcome this hurdle, researchers are experimenting with a technique called Distributed Acoustic Sensing, or DAS, that involves measuring ocean bottom vibrations with fiber optic cables, which line the ocean floor for global telecommunications.
Recent advances enable researchers to collect data from live cables and use artificial intelligence to capture distant earthquakes that would otherwise escape notice.
In a study published in Seismological Research Letters, University of Washington researchers tapped into the Ocean Observatory Initiative’s Regional Cabled Array, which spans the offshore plate boundary and transmits data via fiber optic cable.
Unlike previous experiments that relied on offline or “dark fibers” for data collection, this new study demonstrates that DAS technology can operate without interfering with the OOI network.
“What we created is the starting point of any earthquake analysis,” said co-author Marine Denolle, a UW associate professor in the Earth and space sciences department. “Once our AI algorithm enhances the data, we can actually use the wiggles to do science.”
The fiber optic cable network caught researchers’ attention in the last decade, when they realized its potential for recording solid Earth data. The cables transmit bits of information across great distances in the form of photons, or particles of light.
A sensor—called an interrogator—sends a pulse of light down the cable, but imperfections in the core sometimes cause light to deflect back toward the signal’s origin.
Disturbances near the cable can knock the deflected particles off course, and when they arrive back at the origin, researchers plot their path to locate the disturbance.
“When the earthquake is small or faraway, the energy on the cable is relatively low compared to the ocean, and the signal gets buried in background noise,” said co-author Qibin Shi, a former UW postdoctoral researcher in the Earth and space sciences department who is now a seismologist at Rice University.
In a previous study, UW researchers developed an algorithm that isolates the signal and amplifies it over the background noise by as much as 2.5 times. All they have to do is let the algorithm explore the data and it will learn how to recognize the signal—in this case, an earthquake.
The researchers used data from 285 earthquakes that occurred in Alaska’s Cook Inlet in 2023 as the training dataset.
“A well-trained model will identify earthquakes that the human eye cannot see,” Shi said. “This marks the first step toward a general-purpose foundational model for earthquakes”
To confirm that it would also filter data collected elsewhere, the researchers validated their model at the test site in Oregon, using a live cable. Previous experiments, including the test-run in Alaska, have sourced data from inactive cables, or dark fibers.
In Oregon, the researchers demonstrated that they could collect high-quality data while the cables were transmitting data. They plugged into the Ocean Observatory Initiative’s Regional Cabled Array, which contains fiber optic cables, and tuned the algorithm to the frequency of seismic waves coming from small- to medium-sized earthquakes far away.
The researchers then traced the signal back to specific regions of the subduction zone and pinpointed the precise location of an earthquake.
“It’s the closest we can get to where the action is,” Denolle said. “So for addressing scientific questions, for monitoring, and for early tsunami and earthquake warnings, it’s our best shot.”
The system is also portable, requiring just a modest amount of computing power to operate.
The recent experiment in Oregon lasted just three days and produced large volumes of high-quality data, arguably more than the team knows what to do with, Denolle added.
Their challenge now is figuring out how to manage the data. Both datasets were published free to access, and the one from Alaska is the largest single-site data of its kind. The team is now in the process of negotiating permanent placements for their monitoring system and exploring collaborations.
“This is the future,” Denolle said. “We’re going to understand plate tectonics by studying small earthquakes and this system gives us unprecedented access to that data.”
More information: Qibin Shi et al, Multiplexed Distributed Acoustic Sensing Offshore Central Oregon, Seismological Research Letters (2025). DOI: 10.1785/0220240460
Journal information: Seismological Research Letters
Provided by University of Washington