Real-time automatic detectors of P and S waves using singular value decomposition

TitleReal-time automatic detectors of P and S waves using singular value decomposition
Publication TypeJournal Article
Year of Publication2014
AuthorsKurzon I., Vernon FL, Rosenberger A., Ben-Zion Y.
JournalBulletin of the Seismological Society of America
Volume104
Pagination1696-1708
Date Published2014/08
Type of ArticleArticle
ISBN Number0037-1106
Accession NumberWOS:000343233600009
Keywordsalgorithm; broad-band seismogram; california; earthquake; fault-zone; identification; motion; phase picker; picker developments; subspace tracking; updating
Abstract

We implement a new method for automatic detection of P and S phases using singular value decomposition (SVD) analysis. The method is based on the real-time iteration algorithm of Rosenberger (2010) for the SVD of three-component seismograms. The algorithm identifies the apparent incidence angle by applying SVD and separates the waveforms into their P and S components. We apply the algorithm to filtered waveforms and then either set detectors on the incidence angle and singular values or apply signal-to-noise ratio (SNR) detectors for P and S picking on the filtered and SVD-separated channels. The Anza Seismic Network and the recent portable deployment in the San Jacinto fault zone area provide a very dense seismic network for testing the detection algorithm in a diverse setting, including events with different source mechanisms, stations with different site characteristics, and ray paths that diverge from the approximation used in the SVD algorithm. A 2-30 Hz Butterworth band-pass filter gives the best performance for a large variety of events and stations. We use the SVD detectors on many events and present results from the complex and intense aftershock sequence of the M-w 5.2 June 2005 event. This sequence was thoroughly reviewed by several analysts, identifying 294 events in the first hour, all located in a dense cluster around the mainshock. We used this dataset to fine-tune the automatic SVD detection, association, and location, achieving a 37% automatic identification and location of events. All detected events fall within the dense cluster, and there are no false events. An ordinary SNR detector does not exceed 11% success and has a wider spread of locations (not within the reviewed cluster). The preknowledge of the phases picked ( P or S) by the SVD detectors significantly reduces the noise created by phase-blind SNR detectors.

DOI10.1785/0120130295
Student Publication: 
No