Project EARTH-17-DMP1: Using machine-learning approaches to detect changes in volcanic activity
Supervisors: David Pyle, Tamsin Mather (Earth Sciences, Oxford), David Clifton (Engineering Science, Oxford), Mel Rodgers (University of Southern Florida), and in collaboration with international volcano observatories
Objectives: To develop and test local to global models of patterns of volcanic behaviour during periods of volcanic stability and eruption, using machine learning techniques.
Volcanoes pose a significant threat to lives and livelihoods around the globe. Eruptions are not usually predictable, and rapid changes in activity during an eruption are commonplace. Despite considerable investment globally in new monitoring techniques, and in networks of monitoring instruments on and around volcanoes, our ability to identify critical points in an eruption has not changed much in the past few decades; and there have been significant fatalities during incidents at well-monitored volcanoes where the activity has changed rapidly, or ‘without warning’. In the past few years, there have been significant developments in the field of machine learning, in which software can be trained to ‘learn’ what the normal state of a system looks like, and then used to spot times when a system moves towards an abnormal state.
Volcano monitoring networks typically comprise multiple stations recording continuous seismic and geodetic data, supplemented with instruments to measure gas emissions, strain and infrasound. Campaign-style ‘spot’ data often complement monitoring efforts. With increasing numbers of volcanoes being monitored, and an expanding array of sensors and instruments, volcano observatories now generate huge amounts of data. Machine Learning techniques offer new opportunities for rapid processing and analysis of large, multivariate datastreams, with the potential to transform our ability to identify and understand ‘abnormal’ patterns of volcano behaviour. This project will exploit opportunities offered by the abundance of volcano monitoring time-series data, coupled with new developments in machine learning, to detect and diagnose transitions in volcano behaviour. The focus will be to develop tools to automate the diagnosis of volcanic ‘change points’ in real-time; and to advance understanding of the underlying processes responsible for these transitions.
Lamb O et al., 2014, Multiple timescales of cyclical behaviour observed at two dome-forming eruptions. JVGR 284:106, http://dx.doi.org/10.1016/j.jvolgeores.2014.07.013
McNutt SR, 1996. Seismic Monitoring and Eruption Forecasting of Volcanoes, in: Scarpa & Tilling (eds), Monitoring and Mitigation of Volcano Hazards, 99-146.
Rodgers M et al. 2015a. Peakmatch: A Java Program for Multiplet Analysis of Large Seismic Datasets. Seismol Res Lett 86:1208
Pimentel MAF et al 2014 A Review of Novelty Detection, Signal Processing 99:215-249, http://dx.doi.org/10.1016/j.sigpro.2013.12.026