- This event has passed.
MAE Seminar: Paul Johnson
Title:
Learning earthquake displacement applying machine learning to continuous signals
Abstract:
Machine learning (ML) algorithms have shown great success in advancing science and technology in seismology for a number of applications. These include applications to earthquake detection, classification of different seismic signals, and predicting displacements and time-to-failure in laboratory shear experiments, slow slip in subduction zones and volcanic earthquakes where we have sufficient training and testing data. In recent work, we apply machine learning as an exploratory data analysis tool to investigate whether, (a) features in the recorded, continuous seismic signals contain information about the fault displacement, as they do in the laboratory; and (b) if precursory information regarding an upcoming failure is contained in the continuous seismic signal. We search for patterns that may not be captured by traditional signal processing techniques. Continuous records of digital seismograms are acquired for analysis from a volcano caldera-collapse earthquake-sequence of >60 magnitude ~5 earthquakes at the Kilauea volcano in Hawaii. Labels are ground displacements recorded by global navigation satellite system (GNSS) stations, used as a proxy for fault displacement. We experiment using supervised learning with shallow architecture gradient boosted trees as well as deep learning methods using transformers. We find the models do an excellent job in predicting contemporaneous fault displacement and there are suggestions that near-future information is contained in the continuous seismic signals. The continuous-wave, ML model approaches developed have broad application to seismology and acoustics in general.
Bio:
Geoscientist Paul A. Johnson has contributed broadly to the domains of earthquake processes, nonlinear acoustics of materials, acoustics of porous media, time-reversal acoustics and granular physics. Johnson is also a pioneer for applications of machine learning to geophysics, especially to earthquake processes including aspects of earthquake prediction. Johnson is a Fellow of Los Alamos National Laboratory (2008), the Acoustical Society of America (2005), the American Geophysical Union (2011), and the American Physical Society (2016). He was awarded the foreigner’s medal (“Medaille Etranger”) by the French Acoustical Society in 2011. He is an Emeritus Fellow at Los Alamos, where he remains actively involved in collaborations, and he also contributes to EnviTrace, a small start-up based in Santa Fe, New Mexico.