Quality Control and Data Analysis

In order to monitor and capture the source spectrum of moderate down to micro-scale earthquakes, it is essential that the NHFN instruments operate at high precision and in an extremely low noise environment. Therefore, the stations record at high sample rate, and their sensors are emplaced in deep boreholes to reduce noise contamination originating in the near surface weathered zone and from cultural noise sources. In addition, the reduction of noise at these stations through vigilant monitoring of actual seismic events plays a central part of our quality control effort.

As mentioned, a key aspect of quality control of the NHFN data is the analysis of actual seismic events. Seismic events of larger magnitude are relatively rare and generally provide more energy at lower frequencies. Hence, in order to provide more frequent real events and quality control in the higher frequency band of the NHFN stations, analysis of recordings from the much more frequent microearthquakes are needed. Because real event analyses are relatively labor intensive and because of inadequate insufficient funding, traditional methods of event analysis have proven financially infeasible. To help circumvent these problems, efforts to develop new and improved analysis techniques are ongoing. We have developed and are currently testing some promising techniques that are particularly well suited to the analysis of similar and repeating microearthquakes. The advantages of similar and repeating event analyses for both quality control and scientific purposes are numerous, and the nature of the seismograms from these types of events make automated, rapid and robust analysis possible.

Towards this end, we are continuing to develop our new pattern scanning recognition scheme to detect, pick, locate and determine magnitudes for small and very small similar events recorded either continuously or from among large volumes of noisy triggered data snippets and our phase coherency method for identification of characteristically repeating events sequences from among groups of similar event multiplets.

Pattern Scanning: The pattern scanning recognition approach we are developing enhances the effective signal to noise for event detection, picking and locating by using the high amplitude information available in the full waveform of earthquake's signals. This is done by using a cross-correlation based scanning approach, which scans known waveform patterns through either continuous or collections triggered event snippets (regardless of the triggered event noise levels). With this approach, continuous or triggered waveform data that does not match selected patterns are ignored while waveforms that approximately match selected reference event patterns are flagged as newly identified earthquakes.

This approach is less comprehensive in that it only detects events that are somewhat similar in waveform character to the reference patterns. However, it can be generalized significantly by increasing the number of event patterns scanned or by using fairly low maximum cross-correlation thresholds for event flagging. Preliminary tests of our scanning code show that scans of 100 distinct event patterns can be scanned through a day's worth of waveform data in   75 minutes on one 900Mhz SPARC cpu when continuous seismic data is used. Scanning through collections of all triggered snippets is substantially faster, in proportion to the inverse fraction of total time spanned by the snippet data.

The approach also provides automated cross-correlation pick alignments that can be used for high precision relative locations and for automated low-frequency spectral ratio determinations for magnitude estimates. Clearly the method has potential for automatically cataloging a large fraction of the more numerous microearthquakes, and, in conjunction with the special attributes of similar event groups, updates of the catalogs in an automated monitoring mode can provide near-real-time microearthquake information that can be a powerful tool for monitoring network performance of real event data. Future plans include development and implementation of an automated similar event scanning and cataloging scheme that will provide real-event data from similar small magnitude events for assessment of network health on a much more frequent basis (every few days).

Perhaps more significantly, the approach can also capture and rapidly catalog some of the most scientifically relevant events (e.g. repeats of characteristically repeating microearthquakes used for deep slip rate monitoring and swarms of similar events typically associated with foreshocks and aftershocks). The approach is also surprisingly good at detecting events over a wide magnitude range. Hence there is clear potential for using patterns from larger aftershocks (e.g. flagged by REDI) to rapidly and automatically develop a high-resolution picture of foreshock and aftershock activity associated with large mainshocks. Tests so far using waveform patterns from an aftershock from the Parkfield magnitude 6 event (2.2Ml) have been able to detect and fully process similar events as low as Ml - 1.2 (a range of 3.4 magnitude units). Testing in this regard is continuing, but clearly the 3.4 magnitude range is a lower bound on the potential magnitude range attainable.

Phase Coherency: In order to enhance even further the resolution and scientific value of the similar events identifed using the pattern scanning approach, we are continuing to refine and test our spectral phase coherency algorithm. This algorithm allows for detailed quantification of the similarities and differences between highly similar Hayward fault events to provide characterization of the subsets of similar events known as characteristically repeating microearthquakes. These subsets form groups or sequences of events that are believed to represent recuring ruptures of the same small patch of fault through time, and, once recongnized, such sequences provide a new dimension of constraint on earthquake physics and deep fault deformation through the recurrence intervals and magnitudes of the events comprising the sequences. The complex spectral phase coherency methodology can be carried out in various frequency bands geared appropriately to the magnitude dependence of frequency for earthquakes and is generally an order of magnitude better in its discrimination power than the simple cross correlation method. The goal of the testing and refinement is ultimately to develop a scheme for rapid and objective discrimination and identification of characteristically repeating microearthquakes sequences down to the lowest magnitude possible (where recurrence times are short and hence temporal resolutions are higher) along the Hayward fault, and to use the information from these sequences to gain a better picture of the time evolution of creep deep in the fault zone.

Berkeley Seismological Laboratory
215 McCone Hall, UC Berkeley, Berkeley, CA 94720-4760
Questions or comments? Send e-mail: www@seismo.berkeley.edu
© 2007, The Regents of the University of California