Unsupervised Deep Clustering of Seismic Data: Monitoring the Ross Ice Shelf, Antarctica

Abstract

Advances in machine learning (ML) techniques and computational capacity have yielded state‐of‐the‐art methodologies for processing, sorting, and analyzing large seismic data sets. In this study, we consider an application of ML for automatically identifying dominant types of impulsive seismicity contained in observations from a 34‐station broadband seismic array deployed on the Ross Ice Shelf (RIS), Antarctica from 2014 to 2017. The RIS seismic data contain signals and noise generated by many glaciological processes that are useful for monitoring the integrity and dynamics of ice shelves. Deep clustering was employed to efficiently investigate these signals. Deep clustering automatically groups signals into hypothetical classes without the need for manual labeling, allowing for the comparison of their signal characteristics and spatial and temporal distribution with potential source mechanisms. The method uses spectrograms as input and encodes their salient features into a lower‐dimensional latent representation using an autoencoder, a type of deep neural network. For comparison, two clustering methods are applied to the latent data: a Gaussian mixture model (GMM) and deep embedded clustering (DEC). Eight classes of dominant seismic signals were identified and compared with environmental data such as temperature, wind speed, tides, and sea ice concentration. The greatest seismicity levels occurred at the RIS front during the 2016 El Niño summer, and near grounding zones near the front throughout the deployment. We demonstrate the spatial and temporal association of certain classes of seismicity with seasonal changes at the RIS front, and with tidally driven seismicity at Roosevelt Island.

Hierarchical Exploration of Continuous Seismograms With Unsupervised Learning

“…A time‐frequency representation such as the spectrogram is one way to create a set of features for classifying seismic signals (Jenkins et al., 2021 ; C. W. Johnson et al., 2020 ; Snover et al., 2020 ). However, Andén and Mallat ( 2014 ) showed that a spectrogram generated by the Fourier transform is not ideal for classification purposes since it is not stable to time‐warping deformations, especially at short periods compared with the duration of the analyzing window.…”

“…Since the 1970s seismology benefits from artificial intelligence developments, bringing machine‐learning‐based solutions for exploring seismic data and recognizing patterns (e.g., Allen, 1978 ). More recently an unsupervised learning strategy called clustering was utilized to explore seismic data and find families of similar signals (Holtzman et al., 2018 ; Jenkins et al., 2021 ; C. W.Johnson et al., 2020 ; Köhler et al., 2010 ; Mousavi et al., 2019 ; Seydoux et al., 2020 ; Snover et al., 2020 ). In contrast to supervised learning strategies, clustering does not rely on a labeled training set and human expert knowledge (Goodfellow et al., 2016 ).…”

AI‐Based Unmixing of Medium and Source Signatures From Seismograms: Ground Freezing Patterns

“…Artificial intelligence (AI) can help overcome those blind spots and discover new signals or hidden patterns within the data. Recently, clustering gained attention as a method to identify families of signals in the continuous seismograms (Holtzman et al., 2018; Jenkins et al., 2021; C. W. Johnson et al., 2020; Köhler et al., 2010; Mousavi et al., 2019; Seydoux et al., 2020; Snover et al., 2020; Steinmann et al., 2022). In the most common approach, characteristics—often called features—are calculated for a sliding window.…”

An Unsupervised Machine‐Learning Approach to Understanding Seismicity at an Alpine Glacier

“…Another example of UML applied to cryoseismic settings is in Jenkins et al. (2021), where the authors compare two types of clustering (Gaussian mixture model and deep‐embedded clustering) on 2 years of continuous seismic data from the Ross Ice Shelf (Antarctica), and find that numerous clusters correspond to oceanographic and atmospheric forcing.…”

“…Since no labels are predicted, UML results can be difficult to interpret physically, and since there is no defined target, not all clusters or features may be of scientific interest. Despite these challenges, numerous seismic studies have produced insight through unsupervised feature-extraction, clustering, or a combination of the two (e.g., Chamarczuk et al, 2019;Sick et al, 2015;Steinmann et al, 2021;Trugman & Shearer, 2017;Yoon et al, 2015), including in numerous glaciated, volcanic and/or geothermal settings (e.g., Holtzman et al, 2018;Jenkins et al, 2021;Lamb et al, 2020;Ren et al, 2020;Seydoux et al, 2020). Seydoux et al (2020), for example, uses a Gaussian mixture model to cluster features that were automatically extracted using a deep scattering network (a type of convolutional neural network) from continuous seismic data at Greenland, and are able to identify precursory seismic activity leading to a massive landslide.…”