--> Abstract: Data Mining for Tectonic Tremor in a Large Global Seismogram Database using Preprocessed Data Quality Measurements, by Bart A. Rasor and Michael R. Brudzinski; #90182 (2013)

Datapages, Inc.Print this page

Data Mining for Tectonic Tremor in a Large Global Seismogram Database using Preprocessed Data Quality Measurements

Bart A. Rasor and Michael R. Brudzinski
Department of Geology and Environmental Earth Science, Miami University

The collision of plates at subduction zones yields the potential for disastrous earthquakes, yet the processes that lead up to these events are still largely unclear and make them difficult to forecast. Recent advancements in seismic monitoring has revealed subtle ground vibrations termed tectonic tremor that occur as long-lived swarms of narrow bandwidth activity, different from local earthquakes of comparable amplitude that create brief signals of broader, higher frequency. The close proximity of detected tremor events to the lower edge of the seismogenic zone along the subduction interface suggests a potential triggering relationship between tremor and megathrust earthquakes. Most tremor catalogs are constructed with detection methods that involve an exhausting download of years of high sample rate seismic data, as well as large computation power to process the large data volume and identify temporal patterns of tremor activity. We have developed a tremor detection method that employs the underutilized Quality Analysis Control Kit (QuACK), originally built to analyze station performance and identify instrument problems across the many seismic networks that contribute data to one of the largest seismogram databases in the world (IRIS DMC). The QuACK dataset stores seismogram amplitudes at a wide range of frequencies calculated every hour since 2005 for most stations achieved in the IRIS DMC. Such a preprocessed dataset is advantageous considering several tremor detection techniques use hourly seismic amplitudes in the frequency band where tremor is most active (2-5 Hz) to characterize the time history of tremor. Yet these previous detection techniques have relied on downloading years of 40-100 sample-per-second data to make the calculations, which typically takes several days on a 36-node high-performance cluster to calculate the amplitude variations for a single station. Processing times are even longer for a recently developed detection algorithm that utilize the ratio of amplitudes between tremor frequencies and those of local earthquakes (10-15 Hz) and surface waves (0.02-0.1 Hz). Using the QuACK dataset, we can make the more advanced calculations in a fraction of the time. This method works well to quickly detect tremor in the Cascadia region by finding similar times of increased tremor activity when comparing across a variety of stations within a 100km radius of a reference station. We confirm the legitimacy of this method by demonstrating comparable results to several previously developed tremor detection techniques despite a much shorter processing time. The rapid processing time has allowed us to refine the detection algorithm by seeking more optimal frequency bands by comparing results from our technique and others, using several stations across the Cascadia subduction zone. As we move forward, we will apply the method to other subduction zones, and ultimately to the vast set of seismic data stored at the IRIS DMC for which tremor has not been previously investigated.

AAPG Search and Discovery Article #90182©2013 AAPG/SEG Student Expo, Houston, Texas, September 16-17, 2013