Using unsupervised machine learning to identify changes in eruptive behavior at Mount Etna, Italy

• Published in: Journal of volcanology and geothermal research 2020-11, Vol.405, p.107042, Article 107042
• Main Author: Watson, Leighton M.
• Format: Article
• Language: English
• Subjects: K-means; Mount Etna; Unsupervised machine learning; Volcano infrasound; Volcano monitoring
• ISSN: 0377-0273; 1872-6097
• DOI: 10.1016/j.jvolgeores.2020.107042

Volcanoes frequently generate infrasound signals that need to be processed before they can be used to monitor and track changes in eruptive activity. Unsupervised machine learning is complementary to existing processing methods and can be used for data exploration to identify features of interest in the data. Here, I examine three days of infrasound data from Mount Etna, Italy, that encompasses the 24 December 2018 fissure eruption. The continuous infrasound data is divided into overlapping windows and for each window I extract seven features in the time and frequency domains that characterize the signal. I apply the k-means clustering algorithm to group the data into seven clusters and generate a discrete time series of cluster labels. The cluster labels clearly identify a change in eruptive activity from Strombolian explosions at the summit to lava fountaining at the fissure. Feature distributions and representative waveforms for each cluster are analyzed and source mechanisms are hypothesized. This work illustrates how advances in unsupervised machine learning can be leveraged to explore volcano infrasound data sets and demonstrates the potential of these techniques for monitoring eruptive activity. •Unsupervised machine learning is used to analyze continuous volcano infrasound data.•K-means clustering is used to identify changes in eruptive behavior.•Different clusters are related to different eruptive styles and source mechanisms.•Unsupervised machine learning is complementary to traditional infrasound processing.