Comparison of a Waveform Cross Correlation Detection Method to a Traditional STA/LTA Picker: Application to the Crooked Lake Sequence Near Fox Creek, Alberta

Abstract:

Waveform cross correlation, or template matching as it is sometimes called, has long been known to be an effective method for finding occurrences of a known repeating signal within a waveform. Because seismic signals are rarely known a priori, waveform cross correlation is not often used as a detection method for seismic networks. However, past studies (e.g. Gibbons and Ringdal, 2006) have shown that cross correlation can be effective in identifying events with similar locations and focal mechanisms (and thus waveforms) to a pre-existing template. Because induced seismicity often satisfies these requirements the method is well-suited to induced seismicity monitoring. We apply the method of waveform cross correlation to a sequence of events between Nov. 29, 2013 and Dec. 13, 2013 occurring at Crooked Lake near Fox Creek, Alberta. These events are believed to be attributable to injection activities in the area. A total of 24 events were detected using traditional STA/LTA triggering methods. The largest event, measured at local magnitude 3.9, is used as a template to identify other events. We compare the effectiveness of the traditional STA/LTA detection method to the cross correlation technique. With a modest correlation threshold we identify all 24 of the original events and an additional 89 new events for a total of 113 events identified by waveform cross correlation. We estimate the magnitude of completeness using the maximum curvature method (Wiemer and Wyss, 2000) and compare the result for the STA/LTA catalogue and the cross correlation catalogue. We find that the magnitude of completeness is about 0.8 magnitude units lower for the cross correlation catalogue. We explore the possibility of determining a probability density function to describe the values of observed correlation between a template and a seismic signal and reconcile theoretical expectations with empirical data. We further suggest a trigger threshold for cross correlation detection algorithms based on the probability density function. Finally, we discuss the practicality of implementing waveform cross correlation detection methods to monitor induced seismicity.