Improving Correlation Algorithms to Better Characterize and Interpret Induced Seismicity
Abstract
Correlation algorithms have proven useful for helping to identify repetitive microearthquake sequences buried in vast datasets of passive seismic recordings. Multi-station waveform template matching has been particularly helpful in characterizing seismicity potentially induced by hydraulic
fracturing
or wastewater disposal. A swarm of many events with similar waveforms, presumably driven by localized fluid injection, can be used as criteria to help discern induced seismicity from naturally occurring seismicity. Swarm detection with template matching applied to a regional network can be done in near real-time without the requirement of local seismic deployments or industry data (e.g., injection volumes/pressures or stimulation reports), although this additional data can be utilized if available to further build support for the designation of either an induced or natural origin. An additional advantage of the scanning technique is that the cross correlations are ideally suited to perform advanced seismic source location and magnitude estimation to better characterize identified sequences. In Ohio, 9 recent seismic swarms have been correlated temporally and spatially with either
hydraulic
fracturing
or wastewater injection, while nearly 20 less repetitive earthquakes were not and appear to be naturally occurring. A lack of evidence for any induced seismicity in neighboring Pennsylvania despite nearly an order of magnitude more unconventional wells suggests that geology plays a key role in whether seismicity is induced. In particular, we identify the proximity of the basement to the target interval (either
fracturing
or disposal) as a key factor, suggesting the basement faults are likely needed in the Appalachian Basin to generate M>2 seismicity. In other areas where multiple cases of induced seismicity have already been detected, such as Oklahoma and Alberta, we employ a newly developed algorithm for detection of repetitive sequences. The repetitive signal detector does not require a pre-existing cataloged template event, which helps to detect smaller M<2 sequences that typically precede larger M>2 induced seismicity. This algorithm employs agglomerative cluster analysis to help discern signals from multiple nearby source regions.
AAPG Datapages/Search and Discovery Article #90259 ©2016 AAPG Annual Convention and Exhibition, Calgary, Alberta, Canada, June 19-22, 2016