Show simple item record

dc.contributor.advisorPeng, Zhigang
dc.contributor.authorMeng, Xiaofeng
dc.date.accessioned2016-08-22T12:20:13Z
dc.date.available2016-08-22T12:20:13Z
dc.date.created2015-08
dc.date.issued2015-06-18
dc.date.submittedAugust 2015
dc.identifier.urihttp://hdl.handle.net/1853/55512
dc.description.abstractEarthquakes can be promoted or inhibited by surprisingly small stress perturbations, known as earthquake triggering. The stress perturbations can stem from various natural and anthropogenic processes. The most common stress transfer mechanisms are from other earthquakes, including static stress redistribution caused by permanent fault slip (i.e., static triggering) and dynamic stress carried by passing seismic waves (i.e., dynamic triggering). However, currently seismologists are still debating on whether static or dynamic stresses are more important in triggering earthquakes in near field. Moreover, earthquakes can be triggered by other natural and anthropogenic processes, such as subsurface fluid movement, solid earth tides and wastewater disposal. Study of earthquake triggering has the potential for helping us better understand the physics of the earthquake nucleation, and ultimately, improve seismic hazard assessment and mitigation by forecasting where the next “domino” may fall. However, one major impediment in triggering study is the incompleteness of earthquake catalogs, mainly due to the low signal-to-noise ratio of small earthquakes and masking from the coda waves of larger events. To overcome this, I developed a GPU-based matched filter technique to automatically detect earthquakes through collaborations with computer scientists at Georgia Tech. If two earthquakes occur very close to each other and have similar fault slip, the resulting waveforms recorded at the same station must be very similar. Hence, this technique uses previously identified earthquakes as templates and automatically scans through continuous recordings to detect new events that have high waveform similarities to the templates. The matched filter technique exhibits regular computation and memory access patterns and is mostly data parallel, which makes it an ideal fit for GPU processing. The GPU algorithm cross-correlates multiple templates with the continuous data in parallel and achieves ~40 times speedup over a single CPU card. By using 30 GPU cards on supercomputing cluster ‘Keeneland’ at the Oak Ridge National Lab, the GPU-based method achieves up to ~1000 times speedup over a single CPU. Based on many more microearthquakes detected by the matched filter technique, I was able to investigate a wide spectrum of earthquake triggering mechanisms, including earthquakes triggering associated with static and dynamic stress changes from large nearby earthquakes, atmospheric pressure changes and deep aseismic creep. First, I found clear evidences of ‘stress shadow’, a shut-down of seismicity due to negative static stress changes along Parkfield section of the San Andreas Fault following the 2003 M6.5 San Simeon earthquake. More importantly, the seismicity rate increased near the epicentral region of the future 2004 M6.0 Parkfield earthquake, where the static stress changes are positive, suggesting that the 2004 Parkfield earthquake may be delayed triggered by the 2003 San Simeon earthquake. Second, by examining the temporal evolution of seismicity near Salton Sea and along the San Jacinto Fault in southern California following the 2010 M7.2 El Mayor earthquake, I discovered that dynamic triggering is dominant up to days following the mainshock even at regions with negative static stress changes, while static triggering takes over in a relative longer-term. This result suggests static and dynamic triggerings are not incompatible as previously debated. Instead, they worked in different time scales and hence both should be considered for short- to intermediate seismic hazard assessment. Third, I observed that a category 2 hurricane Irene may have triggered additional aftershocks ~5 days following the 2011 M5.7 Virginia earthquake, and the atmospheric pressure decreases associated with hurricane Irene are the most likely cause. Finally, I found abnormally large aftershock zones following several moderate-size earthquakes near the Anza gap along the San Jacinto Fault since 2000, which are most likely driven by the deep creep near the brittle-ductile transition zone. The result suggests that the deep section may not be completely locked, which is crucial for estimating the size of possible future large earthquakes on the San Jacinto Fault.
dc.format.mimetypeapplication/pdf
dc.language.isoen_US
dc.publisherGeorgia Institute of Technology
dc.subjectEarthquake triggering
dc.subjectSeismology
dc.titleSystematic study of earthquake triggering using microearthquakes detected by the matched filter technique
dc.typeDissertation
dc.description.degreePh.D.
dc.contributor.departmentEarth and Atmospheric Sciences
thesis.degree.levelDoctoral
dc.contributor.committeeMemberNewman, Andrew
dc.contributor.committeeMemberDufek, Josef
dc.contributor.committeeMemberHuber, Christian
dc.contributor.committeeMemberXie, Yao
dc.date.updated2016-08-22T12:20:14Z


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record