Event Detection

As noted in the Introduction, one of the purposes of the HFN is to lower the threshold of microearthquake detection. Towards this goal, we have been developing new algorithms: a pattern recognition approach to identify small events, a phase onset time detector with sub-sample timing resolution, and a phase coherency method for single component identification of highly similar events.

Pattern Recognition: In order to improve the detection and analysis of small events (down to $M_{L} \sim $-1.0), some specialized algorithms are being developed. The Murdock-Hutt detection algorithms used by MultiSHEAR, which basically flags an event whenever the short-term average exceeds a longer-term average by some threshold ratio, is neither appropriate for nor capable of detecting the smallest seismic events. One solution is to use a pattern recognition approach to identify small events associated with the occurrence of an event that was flagged by the REDI system. Tests have indicated that the pattern recognition detection threshold is $M_{L} \sim $ -1.0 for events occurring within $\sim $10 km of a NHFN station. The basic idea is to use a quarter second of the initial P-wave waveform as a master pattern to search for similar patterns that occur within $\pm$ one day of the master event. Experimentally, up to six small CMSB recorded events, at the $M_{L} \sim $ -1.0 threshold and occurring within $\pm$ one day of a master pattern, have been identified.

The pattern recognition method is CPU intensive, however, and it will require a dedicated computer to handle the pattern recognition tasks. To expedite the auto-correlation processing of the master pattern, an integer arithmetic cross-correlation algorithm has been developed which speeds up the requisite processing by an order of magnitude.

Phase Onset Time Detection: The phase onset time detector makes use of the concept that the complex spectral phase data, over the bandwidth of interest (i.e., where the SNR is sufficiently high), will sum to a minimum at the onset of an impulsive P-wave. The algorithm searches for the minimum phase time via phase shifting in the complex frequency domain over the bandwidth where the SNR is above 30 dB to identify the onset time of the seismic phase. The algorithm requires that the recorded waveforms be deconvolved to absolute ground displacement. This implicitly requires that any acausality in the anti-aliasing filtration chain, such as the FIR filters used in the BDSN Quanterra data loggers, be removed. The algorithm typically resolves P-wave onset times to one-fiftieth of the sample interval or better.

Phase Coherency: A spectral phase coherency algorithm was developed to facilitate high resolution quantification of the similarities and differences between highly similar Hayward fault events which occur months to years apart. The resolution of the complex spectral phase coherency methodology is an order of magnitude better that the cross correlation method which is commonly used to identify highly similar events with resolution of order a few meters. This method, originally developed using NHFN borehole data, is now being applied as well to data from another borehole network (the HRSN) to provide more rapid and objective identification of the large fraction (approx. 40$\%$) of characteristically repeating microearthquakes that occur at Parkfield, CA.

Berkeley Seismological Laboratory
215 McCone Hall, UC Berkeley, Berkeley, CA 94720-4760
Questions or comments? Send e-mail: www@seismo.berkeley.edu
© 2005, The Regents of the University of California