Data Mining Microseismicity using PageRank

Ana Aguiar, Lawrence Livermore National Laboratory

Wednesday, March 15, 2017 at 10:30 AM

Location:
Building 3, Rambo Auditorium
Host:
Ezer Patlan

Data mining methods have often been used to explore the similarities of individual seismic events that comprise an earthquake sequence. A relatively new example is PageRank (Google?s initial search algorithm), which can be used for measuring connectivity between two seismograms.

PageRank was originally developed for webpage search engines and was subsequently adapted for use in seismology to detect low-frequency earthquakes (Aguiar and Beroza, 2014). PageRank links seismograms directly through cross correlation and indirectly when two seismograms have a high correlation coefficient (CC) with one another but only one of those has a high CC with a 3rd seismogram. We expand on this initial application of PageRank by using it to define signal-correlation topology for micro-earthquakes in a geothermal environment, including the identification of signals that are connected to the largest number of other signals.

We have focused on the Newberry Volcano in the Deschutes National Forest in Central Oregon, which has been stimulated two times using high-pressure fluid injection to study the Enhanced Geothermal Systems (EGS) technology. Initial locations of the 2012 stimulation found that events occurred in two distinct depth ranges and microseismicity did not clearly outline subsurface structures (Foulger and Julian, 2013). We explored the spatial and temporal development of the 2012 events to better understand how the stimulation modifies stress, fractures rock, and affects permeability. By applying PageRank, we created signal families and simultaneously relocated events within families using the Bayesloc approach (Myers et al., 2007). After relocation, event families are tightly clustered spatially, and some events determined to be linked by PageRank but not spatially clustered in the initial locations are indeed relocated to the same cluster. We also found that signal similarity (linkage) at several stations, not just one or two, is needed to confidently determine if events are in close proximity to one another.

Indirect linkage of signals using PageRank is a reliable way to increase the number of events that are confidently determined to be located close to one another and have a similar focal mechanism. We are currently analyzing the second stimulation, performed in 2014, and will compare these results to clusters found in the initial stimulation. This will allow us to determine whether changes in the state of stress and/or changes in the generation of subsurface fracture networks can be detected using PageRank topology. Ultimately, automating and applying this method in real time could be used in adaptive approaches to enhance production at well-instrumented geothermal sites. This work performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.LLNL-ABS-720039.

Closed captions are typically available a few days after the seminar. To turn them on, press the ‘CC’ button on the video player. For older seminars that don’t have closed captions, please email us, and we will do our best to accommodate your request.

Video Podcast