On May 18, 1980 Mount St. Helens catastrophically erupted. And yet, the eruption wasn’t entirely a surprise. A sparse local seismic network had been recording the beginnings of unrest during the first months of the year. As scientists recognized that the volcano was ending its nap, they deployed additional seismic stations. As a result, they forecasted the eruption days in advance, although the climax still caused more than 50 deaths and over a billion dollars in property damage. Subsequent activity was less cataclysmic, but the volcano remained restive until 1986.

More recently, beginning in 2004, Mount St. Helens erupted again. And again, seismicity tipped scientists off that something was afoot. This new episode involved dome-building, but only two significant explosions occurred. It did not cause any fatalities.

As it turns out, upticks in seismic activity precede many volcanic eruptions, though some remain difficult to anticipate. In a new study, a team of scientists led by Manuela Köpfli, a doctoral student at the University of Washington, explored ways to disentangle seismic sources, like earthquakes, from structural changes related to the material through which seismic waves travel—rocks, magma, or even water in the subsurface. By combining six different methods, the team explored promising ways to forecast eruptions at Mount St. Helens, as well as how to glean insights into other processes occurring on the mountain. 

Source versus structure

Seismic sources constitute anything that emits seismic energy. “Earthquakes, ocean waves, explosions, traffic, landslides, avalanches, rivers,” said Köpfli, “all these things emit energy, which travels as seismic waves through the Earth.” During their travels, the waves mix, generating what seismologists call the seismic wavefield.

But, the seismic wavefield doesn’t change only because of variations in seismic sources. It can also shift because the material through which the waves travel—the subsurface—changes in some way. Called “structural changes,” they can result from a variety of processes, such as reorientation of minerals, freezing of subsurface water (typically in winter months), infiltration of rainwater, or magma influx. These changes “lead to passive alteration of the [seismic] wavefield without emitting seismic waves,” Köpfli said.

“The two seismic fields—event seismology and ambient noise seismology—are very complementary,” Köpfli continued. Event seismology often assumes that the path taken by seismic waves is constant, which helps when comparing different earthquakes. Ambient noise seismology typically assumes that earthquakes are the same, but the material properties along the waves’ paths are changing. But because earthquakes aren’t typically the same, ambient noise seismology relies on the ambient seismic wavefield—a constant seismic clatter of ocean waves, traffic, rivers, rainfall, and more. Changes in the ambient seismic wavefield can illuminate structural changes like the migration of magma or groundwater. 

By using both types of seismology at Mount St. Helens, Köpfli and her colleagues were able to distinguish between changes related to seismic sources versus structural changes related to how waves move through the subsurface. With that information, they could tell the tale of what happened during the eruption of 2004-2008. 

A history of eruptions and observations

The eruptive phase that lasted from 1980 to 1996 was best known for the major eruption of May 18, 1980. Prior to and after the cataclysm, scientists monitored the mountain via both seismic stations and visual supervision. As the sparse local seismic network grew, scientists could correlate explosions they saw with the seismic recordings.

As per the technology of the time, data from the few seismic stations that were continuously collecting data recorded the information on magnetic tapes, which suffer from preservation problems. This makes it hard-to-impossible to analyze these data, Köpfli said. Indeed, most of the stations operated in triggering mode, which meant that they only recorded seismic data when triggered by an event. Additional issues included inaccurate timestamps and stations’ susceptibility to tilting.

Nevertheless, this type of monitoring helped scientists listen to, and analyze, the volcano’s seismic signals. Many dome-building events were forecasted within hours to days before onset based on changes in seismic activity and deformation. In other words, traditional event-based eruption forecasting worked here because earthquakes occurred before eruptions. This turned out to hold true for the next eruptive phase, which lasted from 2004-2008.

In the 2000s, stations operating in triggering mode were replaced with newer instruments that collected continuous seismic data. An additional upgrade was connectivity—these stations sent data back to the University of Washington in Seattle in real time. Noticing an increase in seismicity in 2004, scientists installed even more stations around the mountain.

It is this latter period of restlessness that began in 2004 on which Köpfli and colleagues focused their efforts. The data collected during and after 2004 eruption were of higher quality than the seismic data from the 1980s. Moreover, the methods the team applied required precise timing, which the earlier dataset lacked.

Köpfli and colleagues looked at data from all stations in an 18-kilometer radius that collected seismic information continuously for at least one month between 2000 and 2022. “We would prefer a stable seismic network in which stations were running all the time,” said Köpfli, ”but we want to use as much data as possible, so the one month is just a tradeoff.”

During that eruption, station availability changed quickly because of the destruction and re-establishment of equipment, Köpfli said. Yet, 30 stations fulfilled the requirements. She and her team accessed waveform and related metadata via the NSF SAGE data archive operated by EarthScope. These data came from networks operated by the Pacific Northwest Seismic Network and the USGS Cascades Volcano Observatory.

Methods for monitoring changes

To explore these data, Köpfli and colleagues used six sets of calculations: statistical measures of earthquake size, real-time seismic amplitude measurement (RSAM), displacement seismic amplitude ratio (DSAR), statistical calculations of background seismicity, wave coherence, and scattering coefficients.

Earthquake size can be estimated statistically by extracting peak ground acceleration and peak ground velocity from seismic waveforms. These respectively provide a measure of how fast the ground moved at a particular station’s location and how that movement sped up or slowed down. These measures can also tell scientists about the event magnitude and maximum energy released, and provide additional insight about the seismic sources—mostly earthquakes.

Similarly, RSAM provides a time-averaged estimate of absolute amplitude, and gives scientists another measure of seismic energy at a specific range of frequencies released by an event. This measure tends to be low when a site is seismically quiescent, and high when something’s happening.

However, because some volcanoes experience only weak seismicity prior to eruption, peak ground acceleration, peak ground velocity, and RSAM aren’t ubiquitously helpful for eruption forecasting. Not all volcanoes feature changes in seismicity prior to eruption, such as Ruapehu and Tongariro, active volcanoes in New Zealand.

One solution is for scientists to calculate the displacement seismic amplitude ratio to look at seismic wave attenuation. This measure can change as a result of magma influx or volatiles during an eruption. “Its calculation is efficient enough to be applied in real-time,” Köpfli explained. Unfortunately, changes in seismic wave attenuation could also be the result of changes in the seismic sources, so the team had to rule out source changes before they could make interpretations regarding structural ones.

Statistical calculations of background seismicity like root median square and root mean square hint at what’s happening between seismic events. These measures can indicate how stable the ambient seismic wavefield, particularly when combined with the coherence of the ambient seismic wavefield. The latter helps by illuminating the spatial distribution of seismic sources. Low coherence can indicate a lack of distinct seismic sources—no earthquakes, only noise.

“If the seismic wavefield is stable and we are sure that changes we observe come from structural changes,” Köpfli said, then the team could begin to explore what might be happening in the subsurface to change the ambient seismic wavefield.

Finally, the team also calculated scattering coefficients via deep learning methods. Analysis of the output led to the delineation of four clusters that showed distinct behaviors. Two clusters are dominated by background seismic noise, and the remaining clusters are dominated by distinct seismic sources.

The 2004 eruption

Before September 23, 2004, the ambient seismic wavefield showed no precursory signals—there was no sign of an eruption. Between September 23 and 25, a shallow swarm of small earthquakes occurred, with RSAM, peak ground acceleration and peak ground velocity all pointing toward unrest. Seismic wave attenuation remained unchanged. Together, this information suggests that Mount St. Helens seismicity was increasing without magmatic intrusions or hydrothermal activity. In other words, magma may have already been in place, as shown in prior work.

Between September 25 and October 1, 2004, seismicity began to increase, indicating that magma was on the move. On October 1, the initial explosion occurred. After the explosion, the team observed a significant drop in measures of seismicity and seismic attenuation.

However, this lull lasted only a few hours. The team’s analysis indicates that within three hours, enough pressure had accumulated to cause the tremors to return. The increased seismicity lasted through October 3, 2004.

On October 4, another explosion occurred, followed by low-frequency events called drumbeats that were distinct from the previous days’ earthquakes and tremors. The team suggests that after the explosion, something changed, resulting in a switch from earthquakes to drumbeats. Drumbeats are associated with dome growth, said Köpfli, so their presence before an observable dome could have been signaling underground dome formation. October 5 brought about a third explosion that led to a drop in seismic energy. The team’s analysis, particularly related to statistical measures of earthquake size, pointed to the end of the vent-clearing phase.

From October 5 to 11, shallow, low-frequency repeating earthquakes occurred in tandem with visual changes in glacier ice. On October 11, the new dome became visible for the first time. This period of dome-building was characterized by an increase in seismic attenuation.

Toward a better understanding

In this work, Köpfli and colleagues found that multiple methods enhanced their understanding of Mount St. Helens during the 2004 eruption. However, ambient seismic wavefield analysis did not improve eruption forecasting. Rather, it helped the team unravel previously unknown structural changes that occurred later in the 2004-2008 eruption. For instance, dome growth likely began days earlier than previously thought, instead commencing during the drumbeats of October 4.

After analyzing the eruption onset, the 22 years of continuous seismic data allowed the team to study the entirety of the eruption period via ambient noise seismology. Among their findings: they identified a second phase of volcanic activity unaccompanied by earthquakes, and thus unnoticed. This phase was marked by increased gas emissions.

Scientists do look at structural changes within volcanoes to detect any hints of subsurface change, Köpfli said. “We showed how important it is to carefully analyze the ambient seismic wavefield before you apply seismic noise methods [like] DSAR.” And, even when the onset of eruption is dominated by seismic sources, she said, ambient noise seismology can be especially helpful later in an ongoing eruption. “Only by combining all methods were we able separate seismic sources from structural changes.”