Automatic Detection and Classification of Submarine Volcano Signals
Tuesday, 31 January 2017
Marina/Gretel (Hobart Function and Conference Centre)
Gabrielle Tepp, Alaska Volcano Observatory Anchorage, Anchorage, AK, United States, Matthew M Haney, U.S. Geological Survey Alaska Volcano Observatory, Anchorage, AK, United States, John J Lyons, Alaska Volcano Observatory Fairbanks, Fairbanks, AK, United States, Robert P Dziak, Oregon State University, Corvallis, OR, United States, Delwayne R Bohnenstiehl, North Carolina State Univ., Marine, Earth and Atmospheric Sciences, Raleigh, NC, United States, William W. Chadwick Jr., Oregon State University/NOAA/PMEL, CIMRS, Newport, OR, United States and Kathi Unglert, University of British Columbia, Vancouver, BC, Canada
Abstract:
Improving our understanding of submarine volcanism is highly dependent on our ability to detect and track the activity of submarine volcanoes. This task is complicated by the difficulty of accessing and instrumenting these volcanoes and the vast areas that may be active along spreading centers, subduction zones, and around hotspots. Some submarine eruptions, such as the 2010 eruption of South Sarigan Volcano, Mariana Arc, can be detected by satellite observations and produce subaerial activity, whereas others, such as the 2014 eruption of Ahyi Volcano, Mariana Arc, cannot. Submarine volcanic processes such as eruptions and landslides, however, produce sound waves in the ocean, known as hydroacoustic waves, that offer an opportunity for improved remote detection and monitoring of submarine volcanism. Hydroacoustic signals propagate efficiently through the ocean sound channel, enabling the detection of signals by hydrophones thousands of kilometers away from the source. Additionally, hydroacoustic waves may be detected on land-based seismometers along with any seismic waves produced by the volcanoes.
Manual inspection of hydroacoustic and seismic records is one of the best ways to identify signals but is also highly time-intensive. New techniques in machine learning and other similar methods offer the possibility of automatic identification and characterization of signals that is time and computationally efficient. These techniques, such as self-organizing maps and principal component analysis, have already been used in volcano seismology for the detection of tremor and low-frequency events and by the CTBTO for identification of seismic and hydroacoustic signals. Additionally, they may be able to identify weak infrasound signals originating from submarine volcanoes that could be missed by manual inspection. We apply these methods to reanalyze data from the South Sarigan, Ahyi, and other eruptions in anticipation of new data from a hydrophone array to be deployed in the Mariana arc region in early 2017. Both of these eruptions were recorded by seismic and hydroacoustic arrays, with the South Sarigan eruption also detected by infrasound. This reanalysis allows for both demonstrating the application of these signal detection techniques and potentially uncovering more information about these eruptions.
