Using the Matrix Profile to detect seismic events – from the lab experiment scale to local and global scales

Nader Shakibay Senobari, Zachary Zimmerman, Gareth Funning, Peter M. Shearer, Philip Brisk, & Eamonn Keogh

Published August 15, 2019, SCEC Contribution #9835, 2019 SCEC Annual Meeting Poster #070

We present the results from applying our new similarity search method for continuous seismic data – the Similarity Matrix Profile or, more commonly, just the Matrix Profile (MP) – on several case studies from micro to macro scales. The MP approach is similar to template matching, but does not require any templates, as all sub-windows in the continuous stream of seismic data are compared with the rest of the data. Thus, the MP method might find groups of events that were not detected by the catalog, and therefore could not be detected with the template matching method.

For data bracketing the 2004 M6 Parkfield earthquake, we observe an intriguing behavior of aftershocks using our method. Our result shows a turn over (plateau) in numbers of detected events around 30 minutes after the mainshock and also another change in the P value from 0.27 to 0.62 around seven hours after the mainshock. The existence of a plateau in Omori aftershock decay has long been debated in earthquake physics. Physical models such as that of Dieterich (1994) suggest that it should exist due to creep and afterslip processes that add stress to the aftershock zone.

In another subproject, we apply our new method to the lab experiment acoustic data provided by the Penn State Rock Mechanics Lab (https://sites.psu.edu/chasbolton/). The experiment conducted by a double-direct shear configuration while a piezoelectric transducer recording the acoustic data near the fault zone at 4 MHz sample rate. Our preliminary result shows that the number of acoustic emissions (AEs) detected by our method per 0.01 second (i.e. averaging the number of detected AEs over a 0.01 second running window) can be used as a proxy of stress in the system. The rate of AEs starts to increase as we get closer to failure time, suddenly increases right before the main failure, and then suddenly drops after the mainshock.

We also apply our method on a small subset of 12 Global Seismic Network (GSN) stations to detect global earthquakes. We observe that using our method we detect and locate events that are missed by the USGS catalog. Two interesting examples of these events are the 11 November 2018 Mayotte seismic event and an event during the 2018 Kilauea, Hawaii volcano eruption on 28 May for which there is no catalog seismic event above magnitude 4.5 for that area and date. We are currently investigating these events.

Key Words
Seismic data mining, Earthquake detection

Citation
Shakibay Senobari, N., Zimmerman, Z., Funning, G., Shearer, P. M., Brisk, P., & Keogh, E. (2019, 08). Using the Matrix Profile to detect seismic events – from the lab experiment scale to local and global scales. Poster Presentation at 2019 SCEC Annual Meeting.

Related Projects & Working Groups
Seismology