Semi-automated tremor detection using a combined cross-correlation and neural network approach

Tobias Horstmann, Rebecca M. Harrington, & Elizabeth S. Cochran

Published September 2013, SCEC Contribution #1639

Tectonic tremor is a widely observed phenomenon in many fault zones, and may provide information on fault slip behavior at depth. Despite nu- merous observations of tremor, the emergent phase arrivals, low amplitude waveforms, and variable event durations associated with it make automatic detection a non-trivial task. In this study, we employ a new method to iden- tify tremor in large datasets using an semi-automated technique. The method first reduces the data volume with an envelope cross-correlation technique, followed by a Self-Organizing Map (SOM) algorithm to identify and classify event types. The method detects tremor in an automated fashion, however, it requires calibration for a specific data set, and we therefore refer to it as a "semi-automated". We apply the semi-automated detection algorithm to a newly acquired data set of waveforms from a temporary deployment of 13 seismometers near Cholame, California from May 2010 to July 2011. In the time period between 24 May 2010 to 31 June 2011 we detect 2606 tremor events with a cumulative signal duration of nearly 55 hours. We use the first three weeks of the dataset to test the accuracy of the method and find 101 tremor events with a detection accuracy of 79.5%, and a cumulative signal duration of 141 minutes. The event detection is based on a minimum of only 3 stations; however, optimum detection results require using approximately 10 stations for the particular network configuration considered in this study.

Horstmann, T., Harrington, R. M., & Cochran, E. S. (2013). Semi-automated tremor detection using a combined cross-correlation and neural network approach. Journal of Geophysical Research, 118(9), 4827-4846. doi: 10.1002/jgrb.50345.