Parkfield-Hollister Electromagnetic Monitoring Array

There are many reports of anomalous electric and magnetic fields, at frequencies from quasi DC to several 10's of Hertz, and changes in ground resistivity prior to earthquakes. Most reports are devoted to one or another of these phenomena using a variety of measurement configurations and data processing techniques. No such studies are reported using instrumentation capable of measuring all these properties simultanously at a network of sites. It is the objective of this study to determine whether significant changes in resistivity, quasi DC electric fields, or ULF electric and magnetic fields occur before earthquakes in California. In 1995 we installed two well-characterized electric and magnetic field measuring systems at two sites along the San Andreas Fault which are part of the Berkeley Digital Seismic Network. Since then, magnetotelluric (MT) data have been continuously recorded at 40 Hz and 1 Hz and archived at the NCEDC (Table 7.1 and 7.2). At least one set of orthogonal electric dipoles measures the vector horizontal electric field, E, and three orthogonal magnetic sensors measure the vector magnetic field, B. These reference sites, now referred to as electromagnetic (em) observatories, are co-located with seismographic sites so that the field data share the same time base, data acquisition, telemetry and archiving system as the seismometer outputs. Using a robust multiple station MT processing algorithm (Egbert, 1997), we have examined the long term stability of single site and interstation transfer functions. Using a precise transfer function obtained from long runs of the array data, we can effectively predict the fields at one site from those measured at another site. Subtracting the predicted fields from the measured fields at a site yields residuals which are more sensitive to anomalous signals local to the site. Residual analysis in both time and frequency is the primary goal of this project.

**Figure 7.1:** Map illustrating the location of operational (filled squares) and closed (grey squares) MT sites in central California.
$\begin{figure}\begin{center} \epsfig{file=em_map.ps, width=7cm}\end{center}\end{figure}$

MT Overview

Data at the MT sites are fed to Quanterra data loggers, shared with the collocated BDSN stations, synchronized in time by GPS and sent to the BSL via dedicated communication links.

Activities in 2001-2002

A major part of the recent effort at Oregon State University has been to develop user friendly computer codes for routine processing of data from the UCB MT sites. The processing system is based on a graphical user interface, written in MATLAB, which allows the user to download MT data from the NCEDC and complete all routine processing steps (including the multi-site). The program also can be used to plot processing results, multi-channel time series and various simple diagnostics of data quality. Results (including daily estimates of MT impedances, inter-station transfer functions, estimates of noise amplitudes, and summaries of frequency and time domain residual amplitudes) are then automatically archived for statistical analysis and correlation in space and time with cluster events at Parkfield. There is now a daily printout of the signal to noise ratios (SNR) in dB for each channel of the array, Table 7.3. Currently, SNR's below 10 dB are flagged for inspection or repair by the array operators.

Any data in the NCEDC archive can be downloaded, but by default the program gets and processes the most recent unprocessed data. These codes will thus enable more-or-less automatic monitoring of system functionality, and make it easier to maintain a high rate of quality data return. The streamlined processing codes will also make it easier to reprocess existing data with any new schemes that will be implemented in the future. We are presently using the new system to process the backlog of data from the array, and to update analyses of residuals, and of MT impedance stability.

Several procedures for the residual analysis (the goal of the em array) have been developed: one uses a simple time-domain regression operator to estimate residuals in a moving time window. This algorithm is fast and provides a rapid method to scan incoming data for anomalies. We have tested this process on selected segments of data. No anomalous signals were detected, but there were no earthquakes during this interval.

Another approach is a frequency domain analysis using multiple station techniques (Egbert, 1977) which makes optimal use of data from all sites to estimate stable and reliable transfer functions. The multi-site analysis has also proven to be very useful for better understanding of signal and noise, and for separating coherent signals of differing spatial scales (Figure 7.2). For example, Egbert et al. (2000) used a multiple station analysis to show that the transfer function between SAO and PKD1 is systematically affected by both the DC train system in the Bay Area (BART) and by the non-uniformity of the natural fields in the Pc3 band (Figure 7.3). These results demonstrated that cultural noise sources can extend their effect over surprisingly large areas, and that at the same time natural ionospheric sources may exhibit significant spatial complexity. Because of this added spatial complexity, multiple sites are required for complete cancellation of the background (non-tectonic) em noise. Ideally three stations should be used to avoid bias errors in the transfer function estimates and to maintain better control over cultural noise.

Predictions based on data from at least three sites will significantly improve our ability to detect anomalous signals. Source complications, as well as local incoherent noise sources, are highly variable in time (e.g., Figure 7.3), making it a challenging task to verify that apparently anomalous signals truly originate in the earth. Thorough calibration and understanding of both local and distant noise sources is essential. This critical step has now been accomplished for the UCB array (Egbert et al., 2000; Eisel and Egbert, 2001). When a major Parkfield event does occur we will be in a very good position to detect and identify anomalous em emissions (if there are any) and to avoid the ambiguity of interpretation that has plagued much of the past search for em precursors.

So far we have focused on the frequency domain approach to study residuals. This analysis has revealed significant diurnal variations in the residual distributions. With a two site array, residuals are smallest between the hours of 0-4 am, making this a particularly good time to look for anomalous signals. Comparison of the temporal distribution of unusually large magnetic residuals to local earthquake catalogs has so far revealed no clear associations, but there have been few earthquakes of significant magnitude in the time period studied. Karakelian et al. (2000) analyzed data from some of their sites and ours and suggest that there was anomalous activity before the San Juan Bautista earthquake, but we have not been able to verify this.

The MT stations at PKD and SAO can also be used to monitor resistivity changes prior to earthquakes. Unlike the UC Riverside telluric array, the MT impedance can yield depth information because the depth of penetration of the em waves increases with period. Seasonal changes caused by precipitation would presumably be shallow and affect primarily the shorter periods, while deeper changes would be seen also at longer periods. The amplitude of the MT impedance may fluctuate at all periods in response to shallow changes (the so-called "static" shift problem), but the phase of the response is set by more regional structure at longer periods. Thus, variations in phase should be a sensitive indicator of resistivity variations at seismogenic depths ( $\sim$ 10 km). Eisel and Egbert (2001) made a study of the stability of the MT impedances at PKD1. Typical deviations of estimates based on a single day of data differed from the long term average transfer function by 2-3% for T

300s and increasing to about 10% for T=2000s. Variations between contiguous days were nearly random, so significantly smaller variability can be obtained by longer averaging times. There is some evidence from this analysis for a slow variation of about 1% in impedance amplitude when an 11 day average is applied. Relative resistivity variations are nearly frequency independent, appear anti-correlated between the x-y and y-x modes, and are larger than variations in phase. These features together are suggestive of temporal variations in near-surface static distortion. Although it is difficult to make a definitive statement on the basis of the data analyzed so far, there does not appear to be any seasonal component to these variations, as might be expected if they were due to near surface hydrology.

Table 7.1: Sites of MT observatories

Site	Net	Latitude	Longitude	Elev (m)	Date	Location
PKD	BK	35.945171	-120.541603	583	1999/02/05 -	Bear Valley Ranch, Parkfield
PKD1	BK	35.8894	-120.426109	431.6	1995/06/06 - 1999/03/08	Haliburton House, Parkfield
SAO	BK	36.76403	-121.44722	317.2	1995/08/15 -	San Andreas Obs., Hollister

Table 7.2: Typical data streams acquired at each MT site, with channel name, sampling rate, sampling mode, and FIR filter type. C indicates continuous; T triggered; Ac acausal.

Sensor	Channel	Rate (sps)	Mode	FIR
Magnetic	VT?	0.1	C	Ac
Magnetic	LT?	1.0	C	Ac
Magnetic	BT?	40.0	C	Ac
Electric	VQ?	0.1	C	Ac
Electric	LQ?	1.0	C	Ac
Electric	BQ?	40.0	C	Ac

Table 7.3: SNR at selected periods (dB): Bad channels indicated *

Station	Channel	T=30S	T=100S	T=300S
PKD	Hx	26	25	26
PKD	Hy	23	22	24
PKD	Hx	18	17	15
PKD	Ex1	16	16	18
PKD*	Ey1	9	10	11
PKD	Ex2	23	23	26
PKD	Ey2	21	23	25
SAO	Hx	21	21	23
SAO	Hy	21	20	24
SAO	Hz	15	12	12
SAO	Ex	22	21	22
SAO	Ey	21	21	20

**Figure 7.2:** (a) Eigenvalues of the scaled Spectral Density Matrix (SDM) for the three site PKD1/PKD/SAO array, computed following the methods described in *Egbert* (1997). Briefly, cross-products of Fourier coefficients computed from short time segments of all 17 data channels are averaged for the 30 days. (b) The magnitude of incoherent noise power is estimated for each data channel, and these are used to non-dimensionalize the SDM. Eigenvalues of the 17x17 scaled SDM then given signal-to-noise (power) ratios of independent coherent EM sources. For idealized quasi-uniform MT sources, there should be only two eigenvalues significantly above the 0 dB noise level. Additional large eigenvalues, as seen here from 10-300 s, are a clear indication of "coherent noise", or temporally varying complications in source geometry. For the two dominant eigenvectors, the horizontal magnetic components are roughly uniform across the array, consistent with the usual MT assumptions. Eigenvectors three and four are dominated by gradients in the EM fields.
$\begin{figure}\begin{center} \epsfig{file=temp.eps, width=8cm}\epsfig{file=em_fig5b.eps, width=8cm}\end{center}\end{figure}$

**Figure 7.3:** (a) Amplitude and phase of the Hx/Hx transfer function (TF) between PKD1 and SAO. Curves marked by symbols correspond to TFs estimated for different local times, as indicated in the legend. The heavy dashed line is the TF computed from all data (days 140-199, 1997), and the heavy grey solid line is the TF computed from the data collected during a strike by BART workers (days 150-156, 1997; see *Egbert et al.* (2000)). For periods outside of the band plotted, TFs computed for different data subsets are in close agreement. (b) Variations in the real part of the Hx/Hx TF as a function of local time and time of year, for data grouped into 10 day-long 2 hour bands.
$\begin{figure}\begin{center} \epsfig{file=em_fig6a.eps, width=6cm}\epsfig{file=em_fig6b.eps, width=10cm}\end{center}\end{figure}$

Acknowledgements

Frank Morrison directs the MT program, and collaborates closely with Gary Egbert of Oregon State University and Steve Park of UC Riverside. During Frank's sabbatical last year, Gary Egbert worked extensively with Sierra Boyd on aspects of the data processing. John Friday, Lind Gee, and Doug Neuhauser also contribute to the operation of the MT observatories. Sierra Boyd, Lind Gee, and Gary Egbert contributed to the preparation of this chapter.

Support for the MT array is provided by the USGS through the NEHRP external grants program and by the Plato Malozemoff Chair in Mineral Engineering held by Frank Morrison.

References

Egbert, G.D., M. Eisel, O.S. Boyd and H.F. Morrison, DC trains and Pc3s: Source effects in mid-latitude geomagnetic transfer functions, Geophys. Res. Lett., 27, 25-28, 2000.

Egbert, G.D., Robust Multiple-Station Magnetotelluric Data Processing, Geoph. J. Int., 130, 475-496, 1997.

Eisel, M., and G.D. Egbert, On the stability of magnetotelluric transfer function estimates and the reliability of their variances, Geophys. J. Int., 144, 65-82, 2001.

Karakelian, D., S.L. Klemperer, G.A. Thompson, and A.C. Fraser-Smith, Results from electromagnetic monitoring of the $M_{w}$ 5.1 San Juan Bautista, California earthquake of 12 August 1998, Proc. of the 3rd Conference on Tectonics Problems of the San Andreas Fault System, Eds. G. Bokelmann and R.L. Kovach, Stanford University Publication, Geological Sciences, Vol. XXI, 2000.