The 2018 Geophysical Journal International Student Author Award has been awarded to Karianne Bergen, PhD '18, for a paper co-authored with Greg Beroza detailing three new methods for detecting earthquakes.
Detecting earthquakes over a seismic network using single-station similarity measures
Karianne J. Bergen1 and Gregory C. Beroza2
1Institute for Computational and Mathematical Engineering, Stanford University, Stanford, CA 94305, USA. E-mail: firstname.lastname@example.org
2Department of Geophysics, Stanford University, Stanford, CA 94305, USA Accepted 2018 March 15. Received 2018 January 29; in original form 2017 October 27
SUMMARY New blind waveform-similarity-based detection methods, such as Fingerprint and Similarity Thresholding (FAST), have shown promise for detecting weak signals in long-duration, continuous waveform data. While blind detectors are capable of identifying similar or repeating waveforms without templates, they can also be susceptible to false detections due to local correlated noise. In this work, we present a set of three new methods that allow us to extend single-station similarity-based detection over a seismic network; event-pair extraction, pairwise pseudo-association, and event resolution complete a post-processing pipeline that combines single-station similarity measures (e.g. FAST sparse similarity matrix) from each station in a network into a list of candidate events. The core technique, pairwise pseudo-association, leverages the pairwise structure of event detections in its network detection model, which allows it to identify events observed at multiple stations in the network without modeling the expected moveout. Though our approach is general, we apply it to extend FAST over a sparse seismic network. We demonstrate that our network-based extension of FAST is both sensitive and maintains a low false detection rate. As a test case, we apply our approach to 2 weeks of continuous waveform data from five stations during the foreshock sequence prior to the 2014 Mw 8.2 Iquique earthquake. Our method identifies nearly five times as many events as the local seismicity catalogue (including 95 per cent of the catalogue events), and less than 1 per cent of these candidate events are false detections. Key words: Time-series analysis; Self-organization; Computational seismology; Earthquake monitoring and test-ban treaty verification.