← Back to Gallery

Investigation of Tremor and Explosion Sequences from the 2021-2022 Eruption of Pavlof Volcano, Alaska using Deep Learning

Volcanoes can show variable behavior and unrest mechanisms over the course of long eruptions. However, systematically documenting discrete episodes of elevated seismicity and infrasound can be challenging and laborious, especially if monitoring parameters remain heightened or if the eruption lasts for months or years. The 2021-2022 eruption of Pavlof Volcano, Alaska is an example of a 1.5 year long eruption characterized by intermittent explosions, lava effusion, and low-level ash emissions. Although reliably monitored by the local seismo-acoustic network, satellites, and webcams, the volcano exhibited transitions in unrest regimes which are non-trivial to distill and analyze. In this work, we leverage the diverse tremor and explosion dataset from this eruption to construct an extensive set of labeled seismic and infrasound spectrograms. The labeled spectrograms, representing classes of different tremor types, explosions, earthquakes, and noise, are then used to train VOISS-Net – a convolutional neural network for each data type. Then, the trained models are applied to classify two years of seismic and infrasound data bounding the 2021-2022 eruption, and we implement a local network-wide weighting scheme to collapse station-specific classifications into a single result to reduce uncertainty. We find that our machine learning derived timeline highlights explosion-rich phases in the eruption embedded within months of broadband tremor, as well as explosion-poor phases that coincide with increased thermal output and, at times, observed lava effusion. Beyond general comparisons with multidisciplinary observations and eruption chronology, our approach also reveals interesting explosion-tremor sequences that demonstrate pre-explosive quiescence, microseismicity, and/or transitions in tremor spectra. We additionally demonstrate the transferability of our methods to earlier Pavlof Volcano eruptions (2007, 2013, 2014 and 2016), and elucidate the inter-eruption variability of the volcano’s characteristic seismicity.

Session: Seismoacoustic, Geodetic and Other Geophysical Investigations of Active Volcanoes - III

Type: Oral

Room: Kahtnu 2

Date: 5/3/2024

Presentation Time: 02:00 PM (local time)

Presenting Author: Darren Tan

Student Presenter: Yes

Additional Authors