HR: 10:20h
AN: S32A-01 INVITED [Abstracts]
TI: Array Processing of Large Aperture Seismic Arrays with Generalized Cross-Correlation
AU: * Pavlis, G L
EM: pavlis@indiana.edu
AF: Gary L. Pavlis, Indiana University,
Dept. Geol. Sci.,
1001 East 10th Street, Bloomington, IN 47405
United States
AU: Wang, P
EM: pewang@indiana.edu
AF: Peng Wang, Indiana University
Univ. Information Technology Services,
2711 East 10th Street, Bloomington, IN 47405
United States
AU: Vernon, F L
EM: flvernon@ucsd.edu
AF: Frank Vernon, IGPP,
Scripps Inst. of Oceanography,
Univ. of California, San Diego,
9500 Gilman Drive, LaJolla, CA 92093
United States
AB:
Coherent signal processing of seismic array data has traditionally focused on slant stack methods using a plane wave
approximation for an incident wavefield. This approach has problems with modern broadband seismic arrays with large
apertures for three main reasons: (1) statics are almost always significant compared to the highest recorded frequency, (2)
noise levels and noise spectra vary dramatically from station to station; and (3) three-component recording is now nearly
universal. We have developed a series of algorithms that generalize several concepts in cross-correlation linked to a form
of array processing. Static corrections are dealt with by cross-correlation methods. We use an array stack (beam) as the
reference trace for correlation. The beam is constructed by a nonlinear, robust method using a loss-function based on
coherence. The initial estimate of the beam uses a single station selected interactively by an analyst. The members of the
ensemble are aligned by correlation with the beam, but the beam is adjusted iteratively by a robust estimation of the mean
until the beam data does not change. This algorithm can be applied to single component or three-component data, but with
three-component data one needs to choose an appropriate vector norm to measure residual misfit. This is needed as input to
compute robust weights with the coherence-based loss function. We suggest an ellipsoid norm with principal components in a
ray-centered coordinate system is a good choice for most modern data. Finally, we address the issue of highly colored noise
through multiwavelet processing. Multiwavelets provide a framework for broadband processing that is a mix of multitaper
Fourier methods and time-domain correlation. Statistical redundancy in multiwavelet processing allows objective estimates of
arrival time uncertainty. Accurate phase measurements allow estimation of single station and array average polarization of
three-component data. The computation cost is high and multiwavelet processing is unquestionably a high performance
computing problem. We show that the algorithm is well suited to massively parallel computers that define high-end computing
today.
DE: 7290 Computational seismology
DE: 7294 Seismic instruments and networks (0935, 3025)
DE: 7299 General or miscellaneous
DE: 9820 Techniques applicable in three or more fields
SC: Seismology [S]
MN: Fall Meeting 2005