Scalable Similarity Search in Seismology: A New Approach to Large-Scale Earthquake Detection

Abstract

Extracting earthquake signals from continuous waveform data recorded by networks of seismic sensors is a critical and challenging task in seismology. Earthquakes occur infrequently in long-duration data and may produce weak signals, which are challenging to detect while limiting the number of false discoveries. Earthquake detection based on waveform similarity has demonstrated success in detecting weak signals from small events, but existing techniques either require prior knowledge of the event waveform or have poor scaling properties that limit use to small data sets. In this paper, we describe ongoing research into the use of similarity search for large-scale earthquake detection. We describe Fingerprint and Similarity Thresholding (FAST), a new earthquake detection method that leverages locality-sensitive hashing to enable waveform-similarity-based earthquake detection in long-duration continuous seismic data. We demonstrate the detection capability of FAST and compare different fingerprinting schemes by performing numerical experiments on test data, with an emphasis on false alarm reduction.