Scientists use distant sensor to monitor American Samoa earthquake swarm

In late July to October 2022, residents of the Manu’a Islands in American Samoa felt the earth shake several times a day, raising concerns of an imminent volcanic eruption or tsunami.

An earthquake catalog for the area turned up nothing, because the islands lacked a seismic monitoring network that could measure the shaking and aid seismologists in their search for the source of the earthquake swarm.

But the residents of the Taʻū, Ofu, and Olosega islands needed answers, so Clara Yoon of the U.S. Geological Survey and her colleagues found another way to fill in the seismic blanks. They used machine learning and another technique called template matching on shaking data recorded from a single seismic sensor located 250 kilometers away from the American Samoa swarm.

In The Seismic Record, Yoon and colleagues share how they tracked the swarm using these single-station data, combined with shaking reports from residents, until local permanent seismic stations were installed in American Samoa in August and September 2022.

The non-eruptive volcanic earthquake swarm began in July 2022 about 15 kilometers offshore of Taʻū Island. The Samoa volcanic islands arise as the Pacific tectonic plate moves over a hotspot in the south Pacific Ocean.

Resident reports of the frequent shaking, occurring multiple times a day for a few seconds at a time, were the only information about the swarm at first.

“When the earthquakes started, American Samoa had no instrumental geophysical monitoring, so even basic information about the source of the shaking–with implications for emergency decision-making and public safety–was nonexistent,” said Yoon.

To remedy this, the researchers turned to a remote seismic station on Upolu, Samoa, part of the Global Seismographic Network, that has data that can be downloaded in near-real time through the EarthScope data center, Yoon noted.

The seismic signal of the American Samoa earthquake swarm was difficult to detect at the distant station, however, so Yoon and colleagues used a deep-learning model called EQTransformer, along with a technique called template matching, to pick these tiny earthquakes out of a noisy seismic background.

“EQTransformer found many earthquakes with locations consistent with eastern American Samoa, the largest of which matched up with times of felt reports,” Yoon explained. “These felt reports, contributed by local residents of American Samoa to the National Weather Service, were essential sources of data about the earthquakes, and gave us confidence that the EQTransformer-detected events were actually the same earthquakes felt by the local population.”

With this new earthquake catalog for the event, the researchers were able to characterize the onset and the peak of the swarm activity. Portable and inexpensive Raspberry Shake sensors deployed in August 2022 helped to quickly locate the area of the swarm.

The swarm ended in October 2022 without an eruption, but was likely related to volcanic magma movement, the researchers concluded.

Yoon noted that an approach like their single-station technique could be useful in other places around the world where permanent seismic monitoring is sparse and seismic hazard is poorly understood, such as offshore regions with tsunami potential or earthquakes within a tectonic plate.

She added that the largest earthquake in the American Samoa swarm was magnitude 4.5, making it unlikely that it would have been detected by global seismic networks.

“If no one had lived nearby to report the frequent shaking, this American Samoa swarm may have gone entirely unnoticed,” Yoon said. “Many unknown seismic sources and phenomena are waiting to be discovered, perhaps by future large-scale comprehensive applications of deep-learning approaches in seismology.”