An official website of the United States government
Abstract Volcanic tremor is a semi‐continuous seismic and/or acoustic signal that occurs at time scales ranging from seconds to years, with variable amplitudes and spectral features. Tremor sources have often been related to fluid movement and degassing processes, and are recognized as a potential geophysical precursor and co‐eruptive geophysical signal. Eruption forecasting and monitoring efforts need a fast, robust method to automatically detect, characterize, and catalog volcanic tremor. Here we develop VOlcano Infrasound and Seismic Spectrogram Network (VOISS‐Net), a pair of convolutional neural networks (one for seismic, one for acoustic) that can detect tremor in near real‐time and classify it according to its spectral signature. Specifically, we construct an extensive data set of labeled seismic and low‐frequency acoustic (infrasound) spectrograms from the 2021–2022 eruption of Pavlof Volcano, Alaska, and use it to train VOISS‐Net to differentiate between different tremor types, explosions, earthquakes and noise. We use VOISS‐Net to classify continuous data from past Pavlof Volcano eruptions (2007, 2013, 2014, 2016, and 2021–2022). VOISS‐Net achieves an 81.2% and 90.0% accuracy on the seismic and infrasound test sets respectively, and successfully characterizes tremor sequences for each eruption. By comparing the derived seismoacoustic timelines of each eruption with the corresponding eruption chronologies compiled by the Alaska Volcano Observatory, our model identifies changes in tremor regimes that coincide with observed volcanic activity. VOISS‐Net can aid tremor‐related monitoring and research by making consistent tremor catalogs more accessible.
more »
« less
This content will become publicly available on January 22, 2026
A generalized deep learning model to detect and classify volcano seismicity
Volcano seismicity is often detected and classified based on its spectral properties. However, the wide variety of volcano seismic signals and increasing amounts of data make accurate, consistent, and efficient detection and classification challenging. Machine learning (ML) has proven very effective at detecting and classifying tectonic seismicity, particularly using Convolutional Neural Networks (CNNs) and leveraging labeled datasets from regional seismic networks. Progress has been made applying ML to volcano seismicity, but efforts have typically been focused on a single volcano and are often hampered by the limited availability of training data. We build on the method of Tan et al. [2024] (10.1029/2024JB029194) to generalize a spectrogram-based CNN termed the VOlcano Infrasound and Seismic Spectrogram Neural Network (VOISS-Net) to detect and classify volcano seismicity at any volcano. We use a diverse training dataset of over 270,000 spectrograms from multiple volcanoes: Pavlof, Semisopochnoi, Tanaga, Takawangha, and Redoubt volcanoes\replaced (Alaska, USA); Mt. Etna (Italy); and Kīlauea, Hawai`i (USA). These volcanoes present a wide range of volcano seismic signals, source-receiver distances, and eruption styles. Our generalized VOISS-Net model achieves an accuracy of 87 % on the test set. We apply this model to continuous data from several volcanoes and eruptions included within and outside our training set, and find that multiple types of tremor, explosions, earthquakes, long-period events, and noise are successfully detected and classified. The model occasionally confuses transient signals such as earthquakes and explosions and misclassifies seismicity not included in the training dataset (e.g. teleseismic earthquakes). We envision the generalized VOISS-Net model to be applicable in both research and operational volcano monitoring settings.
more »
« less
- Award ID(s):
- 1855126
- PAR ID:
- 10615329
- Publisher / Repository:
- Volcanica
- Date Published:
- Journal Name:
- Volcanica
- Volume:
- 8
- Issue:
- 1
- ISSN:
- 2610-3540
- Page Range / eLocation ID:
- 305 to 323
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract By providing unrivaled resolution in both time and space, volcano seismicity helps to chronicle and interpret eruptions. Standard earthquake detection methods are often insufficient as the eruption itself produces continuous seismic waves that obscure earthquake signals. We address this problem by developing an earthquake processing workflow specific to a high‐noise volcanic environment and applying it to the explosive 2008 Okmok Volcano eruption. This process includes applying single‐channel template matching combined with machine‐learning and fingerprint‐based techniques to expand the existing earthquake catalog of the eruption. We detected an order of magnitude more earthquakes, then located, relocated, determined locally calibrated magnitudes, and classified the events in the enhanced catalog. This new high‐resolution earthquake catalog increases the number of observations by about a factor of 10 and enables the detailed spatiotemporal seismic analysis during a large eruption.more » « less
-
Although Veniaminof Volcano in Alaska experiences frequent eruptions and has eight permanent seismic stations, only two of the past 13 eruptions have had precursory signals that prompted a pre-eruption warning from the Alaska Volcano Observatory (AVO) since 1993. Seismic data from Venianimof indicate that most eruptions from 2000 to 2018 do not coincide with increased seismicity. Additionally, analyses of InSAR data available from 2015 to 2018 which covers the pre-, syn-, and post-eruption periods of the 2018 eruption do not show clear signs of deformation. The systemic lack of systematic precursory signals raises critical questions about why some volcanoes do not exhibit clear unrest prior to eruption. Volcanoes that erupt frequently without precursory signals are often classified as “open” systems with magma migrating through an open network to eruption, rather than pausing at a shallow reservoir. However, the precursory signals, or lack thereof, from a small or deep closed magma system may be difficult to observe, resulting in a stealthy eruption mimicking the behavior of an open system. In this study, we utilize finite element, fluid injection models to investigate a hypothetical closed magma system at Veniaminof and evaluate its ability to erupt with no observable early-warning signals. Specifically, a series of numerical experiments are conducted to determine what model configurations lead to stealthy eruptions – i.e., producing ground deformation below the detection threshold for InSAR (<10 mm) and developing no seismicity, yet resulting in tensile failure which will promote diking and eruption. Model results indicate that the primary control on whether eruption precursors from deformation and seismicity will be present are the rheology of the host rock and the magma flux, followed by the secondary control of the size of the magma chamber, and then its depth and shape. Volcanoes with long-lived thermally mature magma systems with moderate to small magma reservoirs are the most likely to exhibit stealthy behavior, with the smallest systems most likely to fail without producing a deformation signal. This result is likely because small, deep magma systems produce minimal surface deformation and seismicity. For stealthy volcanoes like Veniaminof and others in Alaska (e.g., Cleveland, Shishaldin, Pavlof) and around the world, understanding the underlying magma system dynamics and their potential open vs. closed nature through numerical modeling is critical for providing robust forecasts of future eruptive activity.more » « less
-
Abstract Volcano infrasound data contain a wealth of information about eruptive patterns, for which machine learning (ML) is an emerging analysis tool. Although global catalogs of labeled infrasound events exist, the application of supervised ML to local (<15 km) volcano infrasound signals has been limited by a lack of robust labeled datasets. Here, we automatically generate a labeled dataset of >7500 explosions recorded by a five-station infrasound network at the highly active Yasur Volcano, Vanuatu. Explosions are located via backprojection and associated with one of Yasur’s two summit subcraters. We then apply a supervised ML approach to classify the subcrater of origin. When trained and tested on data from the same station, our chosen algorithm is >95% accurate; when training and testing on different stations, accuracy drops to about 75%. The choice of waveform features provided to the algorithm strongly influences classification performance.more » « less
-
Abstract Seismicity at restless volcanoes commonly features a variety of signal types reflecting both volcanotectonic and fluid-driven source processes. However, traditional catalogs of seismicity are often incomplete, especially concerning events with emergent onsets such as those driven by the dynamics of magmatic and hydrothermal fluids. The detection of all discrete events and continuous seismic tremors, regardless of the underlying source processes, would therefore improve the ability of monitoring agencies to forecast eruptions and mitigate their associated hazards. We present a workflow for generalized detection of seismic events based on the network covariance matrix (Seydoux et al., 2016). Our contributions enable the method to simultaneously detect continuous and short-duration (<∼10 s) events, provide information about the frequency content of the signals, and to refine the initial detection times by an order of magnitude (from window lengths of 75 to 7.5 s). We test the workflow on a 15-month record of seismicity with 23 stations at Mammoth Mountain, California (July 2012–October 2013) and detect 62% of long-period events and 94% of volcanotectonic events in the existing Northern California Earthquake Data Center catalog. In addition, ∼3000 events are not included in the catalog, and thousands of tremor signals are found. The method is suitable for near-real-time analysis of continuous waveforms and can provide a valuable supplement to existing algorithms to improve the completeness of catalogs used for monitoring volcanoes.more » « less
