This supplemental content contains five figures. Figure S1 compares performance of the improved Fingerprint And Similarity Thresholding (FAST) earthquake-detection algorithm (Rong et al., 2018) applied in this study, compared to a previous version of the FAST algorithm (Yoon et al., 2015), on a synthetic data set. Figures S2–S4 show examples of detected signals that represent vibrations in the Earth but are not the local earthquakes of primary interest to this study. Figure S5 displays historical catalog seismicity in the area near the Diablo Canyon power plant (DCPP).
Figure S1. Synthetic test results for three different scaling factors c: (top) 0.05, (center) 0.03, (bottom) 0.01, with signal-to-noise ratio (SNR) values provided. We show 12 hr of synthetic data with planted earthquake waveforms, multiplied by a scaling factor c and inserted (left) at 24 different times and (right) detection results as precision-recall curves for both autocorrelation (red) and FAST (blue and green), generated by setting detection thresholds in terms of correlation coefficient and FAST similarity, respectively. The blue precision-recall curves are from the original version of FAST, from Figure S13 in Yoon et al. (2015). The green precision-recall curves are from the new FAST software (Rong et al., 2018) used in this study, with several improvements to the fingerprint algorithm (including the median statistics in Bergen et al., 2016) and similarity search. This latest version of FAST demonstrates excellent precision-recall performance on this synthetic test, even when SNR < 1 (bottom) and autocorrelation performs poorly, indicating that the fingerprints are robust to noise fluctuations.
Figure S2. Example of regularly repeating signals from a seismic-reflection survey of the Shoreline fault (Pacific Gas and Electric Company [PG&E], 2011, 2014, 2015) within a 3-minute time window, filtered 3–12 Hz on all channels, visible on Stations DCD, DPD, VPD, and SHD. We excluded these detections from the list of candidate events that dominated the detections from 2011-09-28 to 2011-11-01, because they are not local earthquakes of interest.
Figure S3. 3-min unfiltered time windows of five detected deep-teleseismic earthquakes, with parameters from Comprehensive Catalog (ComCat). (a) 2013-05-24 05:44:49.60 UTC, Mw 8.3, latitude 54.8736° N, longitude 153.2805° E, depth 608.9 km, Sea of Okhotsk; (b) 2013-05-24 14:56:31.80 UTC, Mw 6.7, latitude 52.2220° N, longitude 151.5150° E, depth 623.0 km, Sea of Okhotsk; (c) 2015-05-30 11:23:02.70 UTC, Mw 7.8, latitude 27.8312° N, longitude 140.4932° E, depth 677.6 km, south of Japan; (d) 2015-11-24 22:45:38.00 UTC, Mw 7.6, latitude 10.0475° S, longitude 71.0226° W, depth 611.7 km, Peru–Brazil border; (e) 2015-11-24 22:50:53.70 UTC, Mw 7.6, latitude 10.5484° S, longitude 70.9038° W, depth 600.6 km, Peru–Brazil border.
Figure S4. 3-min time windows of detected infrasound signals (sound waves at frequencies < 20 Hz) that propagate at the speed of sound, much slower than seismic velocities, with varying source locations and signal durations. These may be anthropogenic sound waves, such as sonic booms from aircraft or artillery explosions at nearby military bases.
Figure S5. Historical seismicity from Northern California Seismic Network (NCSN) and Southern California Seismic Network (SCSN) catalogs (empty circles, sized by magnitude) near the DCPP (star) from 1967 to June 2007, prior to the time period (June 2007–October 2017) in this study. The seismic network used for event detection (hollow triangles) and an additional station used for location (solid triangle) are also shown.
Bergen, K., C. Yoon, and G. C. Beroza (2016). Scalable similarity search in seismology: A new approach to large-scale earthquake detection, Proc. of the 9th International Conf. on Similarity Search and Applications, 301–308, doi: 10.1007/978-3-319-46759-7_23.
Pacific Gas and Electric Company (PG&E) (2011). Report on the analysis of the Shoreline fault zone, central coastal California, Report to the U.S. Nuclear Regulatory Commission, available at https://www.pge.com/mybusiness/edusafety/systemworks/dcpp/shorelinereport/index.shtml (last accessed May 2019).
Pacific Gas and Electric Company (PG&E) (2014). Report on the Central Coastal California Seismic Imaging Project (CCCSIP), Report to the U.S. Nuclear Regulatory Commission, available at https://www.pge.com/en_US/safety/how-the-system-works/diablo-canyon-power-plant/seismic-safety-at-diablo-canyon/seismic-report.page (last accessed May 2019).
Pacific Gas and Electric Company (PG&E) (2015). Seismic source characterization for the Diablo Canyon Power Plant, San Luis Obispo County, California, Report on the Results of a SSHAC Level 3 Study, Rev. A, March, available at https://www.pge.com/en_US/safety/how-the-system-works/diablo-canyon-power-plant/seismic-safety-at-diablo-canyon/sshac.page (last accessed May 2019).
Rong, K., C. E. Yoon, K. J. Bergen, H. Elezabi, P. Bailis, P. Levis, and G. C. Beroza (2018). Locality-sensitive hashing for earthquake detection: A case study scaling data-driven science. Proc. VLDB 11, no. 11, 1674–1687, available at https://www.vldb.org/pvldb/vol11/p1674-rong.pdf (last accessed May 2019).
Yoon, C. E., O. O’Reilly, K. J. Bergen, and G. C. Beroza (2015). Earthquake detection through computationally efficient similarity search, Sci. Adv. 1, e1501057, doi: 10.1126/sciadv.1501057.
[ Back ]
