“Analytical listening” is the key to early detection of impending volcanic eruptions. Now the telltale sound of volcanic activity can be captured even more precisely using machine learning and musicology methods, researchers report. Their approach allowed them to identify seismic signatures of specific phases before and during a volcanic eruption in Iceland. Most importantly, the team captured a previously unknown tremor sequence that could indicate impending eruptive activity.
Inferno without warning: As is well known, volcanoes can devastate their surrounding areas suddenly and with a fatal surprise effect. As a result, people are often unable to get to safety from the ash masses or lava flows in time. It is therefore an important goal of volcanological research to be able to better predict the timing, strength and further course of eruptions. For this purpose, the fire mountains are already being intensively monitored and sometimes even eavesdropped on. The complex processes underground that lead to an eruption cause a special “rumbling noise”. But actually identifying characteristic warning signs in the seismic signals from volcanoes is more difficult than you might think. A telltale sign of impending eruptions is continuous volcanic vibrations - a so-called tremor. However, this geophysical marker is often difficult to distinguish from overlying seismic signals.
Tracing vibrations in volcanic “music”.
In order to be able to “listen” even more precisely, the research team led by Zahra Zali from the University of Potsdam has now further developed special machine learning (ML) methods and the analysis of music. In ML processes, computers are equipped with special algorithms with the ability to recognize certain patterns in data. The researchers have now adapted the special ML method of so-called deep embedded clustering to their goal. As they explain, this process enables particularly quick analysis of data that requires little processing. “Using this method, we can combine signals of similar structure in seismic data and thus recognize previously hidden patterns,” explains Zali.
The supplementary procedure is a method that actually comes from musicology and was further developed by Zali for the analysis of seismic data. “I was inspired by the idea of harmonic-percussive separation in musical signal processing. “The problem is similar with these acoustic waves: in order to identify different instruments in a piece of music, different types of sounds have to be separated from each other, for example the harmonic sounds of melodic violins from the percussive ones of a percussion,” explains Zali. She and her colleagues adapted the corresponding musical analysis procedures for the analysis of the “music of the volcanoes”.
Hidden sign identified
The researchers applied their techniques to the analysis of seismic data recorded before and during the Geldingadalir eruption in Iceland that began on March 19, 2021. They come from a measuring station about 5.5 kilometers southeast of the eruption site. The scientists were able to connect the seismic information with the recordings of the events before and during the eruption. On February 24, 2021, an earthquake of magnitude 5.7 occurred in the area. Three weeks later, magma reached the surface, so that on March 19th a fissure opened and the eruption began. It was initially characterized by a continuous outflow of magma. But from April 27th the flow increased significantly and on May 2nd it finally began to boil violently: This was followed by an eruption phase with lava fountains that lasted until June 13th.
With their new approaches, the researchers were now able to identify seismic patterns that matched the different phases of the eruption. In particular, they identified two subtle signals that could be important for the early warning and prediction of volcanic eruptions and volcanic activity: They uncovered a previously hidden tremor sequence that occurred before the eruption and continued during it. “Our observation of the volcanic tremor from March 16, three days before the eruption, may indicate that magma has reached the near-surface crust. “Such pre-eruptive tremors are mainly due to magma movements and their interactions with gas and adjacent rock,” explains Zali. In addition, the researchers identified a tremor that occurred when the lava outflow increased and fountains formed. This signal could therefore have been linked to the increase in the outflow rate, explains the team.
In addition to the concrete results, the volcanologists now see fundamental potential in their concept: “Our method offers a fast and reproducible approach to automatically decipher the temporal development of a volcanic system: based on raw seismic signals, we can identify relevant features even without prior data processing and potentially gain unexpected insights,” says Zali. Senior author Fabrice Cotton from the University of Potsdam concludes: “Thanks to advances in monitoring technologies and machine learning, we now have the opportunity in seismology to better detect the early stages of volcanic eruptions and to record the subsequent eruption phases with greater speed and precision “, says the scientist.
Source: University of Potsdam, specialist article: Nature Communications Earth and Environment, doi: 10.1038/s43247-023-01166-w