Robert J. Skoumal

Distinguishing induced seismicity from natural seismicity in Ohio: Demonstrating the utility of waveform template matching

This study investigated the utility of multistation waveform cross correlation to help discern induced seismicity. Template matching was applied to all Ohio earthquakes cataloged since the arrival of nearby EarthScope TA stations in late 2010. Earthquakes that were within 5 km of fluid injection activities in regions that lacked previously documented seismicity were found to be swarmy. Moreover, the larger number of events produced by template matching for these swarmy sequences made it easier to establish more detailed temporal and spatial relationships between the seismicity and fluid injection activities, which is typically required for an earthquake to be considered induced. Study results detected three previously documented induced sequences (Youngstown, Poland Township, and Harrison County) and provided evidence that suggests two additional cases of induced seismicity (Belmont/Guernsey County and Washington County). Evidence for these cases suggested that unusual swarm-like behaviors in regions that lack previously documented seismicity can be used to help distinguish induced seismicity, complementing the traditional identification of an anthropogenic source spatially and temporally correlated with the seismicity. In support of this finding, we identified 17 additional cataloged earthquakes in regions of previously documented seismicity and away from disposal wells or hydraulic fracturing that returned very few template matches. The lack of swarminess helps to indicate that these events are most likely naturally occurring.

An efficient repeating signal detector to investigate earthquake swarms

Repetitive earthquake swarms have been recognized as key signatures in fluid injection induced seismicity, precursors to volcanic eruptions, and slow slip events preceding megathrust earthquakes. We investigate earthquake swarms by developing a Repeating Signal Detector (RSD), a computationally efficient algorithm utilizing agglomerative clustering to identify similar waveforms buried in years of seismic recordings using a single seismometer. Instead of relying on existing earthquake catalogs of larger earthquakes, RSD identifies characteristic repetitive waveforms by rapidly identifying signals of interest above a low signal-to-noise ratio and then grouping based on spectral and time domain characteristics, resulting in dramatically shorter processing time than more exhaustive autocorrelation approaches. We investigate seismicity in four regions using RSD: (1) volcanic seismicity at Mammoth Mountain, California, (2) subduction-related seismicity in Oaxaca, Mexico, (3) induced seismicity in Central Alberta, Canada, and (4) induced seismicity in Harrison County, Ohio. In each case, RSD detects a similar or larger number of earthquakes than existing catalogs created using more time intensive methods. In Harrison County, RSD identifies 18 seismic sequences that correlate temporally and spatially to separate hydraulic fracturing operations, 15 of which were previously unreported. RSD utilizes a single seismometer for earthquake detection which enables seismicity to be quickly identified in poorly instrumented regions at the expense of relying on another method to locate the new detections. Due to the smaller computation overhead and success at distances up to ~50 km, RSD is well suited for real-time detection of low-magnitude earthquake swarms with permanent regional networks.