APPLIED MATH SEMINAR

Speaker: Francois Meyer, Dept. of Electrical Engineering, University of Colorado at Boulder

Title: Exploring the Manifold of Seismic Waves: Application to Phase Detection

When/where: Wednesday, March 10th, 4:15 PM, AKW 200

The U.S. monitoring authority continuously processes seismograms to
monitor for possible nuclear explosions. The detection of seismic
events from the raw waveform is a fundamental step in the detection
and identification of seismic phases.

This work addresses the problem of automatically detecting regional
seismic phases from a seismic trace. We propose a method to detect
sudden changes in the regularity and frequency content of a waveform.
Our approach considers the trajectory of the vector formed by the
time-delayed values of the waveforms. We show that a nonlinear
embedding of this trajectories yields a natural low-dimensional
organization of the data in terms of background activity and seismic
events.  We take advantage of this new parametrization to construct
a classifier that can identify seismic phases.

The method is validated using a dataset of seismic events that
occurred in Idaho, Montana, Wyoming, Utah, and Nevada from 2005 to
2006. The method outperforms the standard approaches.

This work was supported by the National Nuclear Security
Administration Contract No. DE-AC04-94AL8500.