Integration of Empirical Signal Detectors into the Detection and Feature Extraction Application at the United States National Data Center

W. N. Junek1, T. F. VanDeMark1, T. R. Saults1, D. B. Harris2, D. A. Dodge3, S. Matlagh1, G. A. Ichinose1, A. Poffenberger1, and R. C. Kemerait1

1. United States National Data Center, Air Force Technical Applications Center, Patrick Air Force Base, Florida 2. Deschutes Signal Processing, Maupin, Oregon

3. Lawrence Livermore National Laboratory, Livermore, California

Introduction Seismic analysts at the United States National Data Center (US NDC) routinely review events from repeating sources (e.g., mines, earthquake sequences). Typically these events are highly correlated and have the same sta-tion/phase associations. The US NDC has integrated a suite of efficient empirical signal detectors into the Detec-tion and Feature Extraction (DFX) application that will simultaneously detect and identify events from repeat-ing sources. The detector suite consists of array-based subspace and correlation detectors and a single trace cor-relation detector. Detectors are trained manually using an interactive software application known as the Cluster Construction and Analysis Tool (CCAT). Cluster specific data are packaged by CCAT into a format that can be easily imported into the US NDC processing framework for use in real-time station processing. Multi-station detections from repeating events are associated to a common source, located using pre-computed empirical trav-el time corrections, and presented to analysts within the Analyst Review Station (ARS). Successful integration of this functionality into DFX has reduced analyst burden associated with repeating events allowing more time for interactive analysis of anomalous events.

Summary: Empirical signal detectors integrated into automated pipeline

processing via DFX

Process went operational in mid 2011, limited use however

Numerous detectors under evaluation on development pipeline

14 empirical signal detector scheduled for deployment in 2013

Reduces analyst burden associated with repeating events allowing more time for interactive analysis of anomalous events.

Reference:

Harris, D. B., Dodge, D. A., “An Autonomous System for Grouping Events in a Developing Aftershock Sequence”, BSSA, April, (2011), 101: 763-774.

http://dx.doi.org/10.1785/0120100103 This work performed in part under the auspices of the U. S. Department of Energy by Lawrence Livermore National Laboratory under contract number DE-AC52-07NA27344. “The views expressed here do not necessarily reflect the views of the United States Government, the United States Department of Defense, the United States Department of Energy, or the Lawrence Livermore National Laboratory (Unlimited Release Number: LLNL-POST-637080).” 

Interactive Analysis:

Near Real Time Processing: The block diagram above illustrates our two part multiple event analysis method. The interactive component requires a human analyst to manually identify a representative data set for detector construction via CCAT. The automated component is performed by the US NDC geophysical processing platform. We have augmented DFX with an empirical signal detector library which include signal subspace and array based correlation detec-tors. Correlation detections are used in multi-station event formation and locations are computed using cluster specific empirical travel time corrections derived during detector construction process. Information regarding similarity to reference events and membership in known event clusters is included in the final event catalog. 

Cluster analysis via dendrogram is employed to identify families of correlated events 

 Time trend analysis of origin time distribution assists in

differentiating natural and manmade sources.

Cluster Construction and Analysis Tool (CCAT): Interfaces with operational databases Geographic Information System (GIS) tools Builds Subspace detectors and associated parameter files

for automated processing software 

Geo-referenced imagery can be loaded and used to assist in relative relocation process.

Bulk travel time corrections can be computed to

shift biased event locations over a known source, such as an open pit mine

Single point, empirical travel time corrections can

be used in real time by the US NDC.

Relative repicking of arrivals from correlation events is performed to im-prove relative relocations.

CCAT provides repicking tools based

on cross correlation optimum lags and multi-trace stacks

Automated Component: Exploits mature & robust geophysical processing applications Follows pipeline unification methodology Utilized real time data QC information and azimuth and slow-

ness screening processed. 

Picks are copied from the most similar reference waveform to the new-ly detected signal. 

 In addition, event specific information (e.g., event type) is transferred

from the best match to the newly detected event. 

Discussion of Results:

Time (Sec)

Det

ectio

n St

atis

tic C

[n]

Detection

DFX runs correlation detectors against the real time data stream for selected stations. Detection statistics from multichannel data are combined into a single detection statistic An empirical detection threshold is used to identify peaks in the detection statistic that

are above background noise. 

Time frame: 14 months 314 Events detected by correlation detector

235 events in analyst reviewed bulletin 89 selected events shown in plot

Interface to LibSSD