Generalized Seismic Phase Detection with Deep Learning

Abstract

To optimally monitor earthquake-generating processes, seismologists have\nsought to lower detection sensitivities ever since instrumental seismic\nnetworks were started about a century ago. Recently, it has become possible to\nsearch continuous waveform archives for replicas of previously recorded events\n(template matching), which has led to at least an order of magnitude increase\nin the number of detected earthquakes and greatly sharpened our view of\ngeological structures. Earthquake catalogs produced in this fashion, however,\nare heavily biased in that they are completely blind to events for which no\ntemplates are available, such as in previously quiet regions or for very large\nmagnitude events. Here we show that with deep learning we can overcome such\nbiases without sacrificing detection sensitivity. We trained a convolutional\nneural network (ConvNet) on the vast hand-labeled data archives of the Southern\nCalifornia Seismic Network to detect seismic body wave phases. We show that the\nConvNet is extremely sensitive and robust in detecting phases, even when masked\nby high background noise, and when the ConvNet is applied to new data that is\nnot represented in the training set (in particular, very large magnitude\nevents). This generalized phase detection (GPD) framework will significantly\nimprove earthquake monitoring and catalogs, which form the underlying basis for\na wide range of basic and applied seismological research.\n