Parkfield-Hollister Electromagnetic Monitoring Array

The primary objective of the UC Berkeley electromagnetic (EM) monitoring array is to identify EM fields or changes in ground conductivity that might be associated with earthquakes. The array has consisted of up to three sites operating since 1995 at SAO, PKD, and PKD1, each of which measures three orthogonal components of the magnetic field and two orthogonal components of the electric field. Multiple sites are necessary in order to separate
the fields of a local source (e.g., an earthquake signal, cultural noise) from the natural regional micropulsations [*Gamble* 1979]. Our approach has been to determine the transfer function between fields at different sites for periods of normal background EM variations and then use this transfer function to predict fields between sites. Differences between the observed and predicted fields are used to search for anomalous local fields.

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 (references to these data channels can be found in the 2003 BSL Annual Report). 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.

The MT observatories are located at Parkfield (PKD1, PKD), 300 km south of the San Francisco Bay Area and Hollister (SAO), halfway between San Francisco and Parkfield (Figure 20.1). In 1995, initial sites were established at PKD1 and SAO, separated by a distance of 150 km, and equipped with three induction coils and two 100 m electric dipoles. PKD1 was established as a temporary seismic site, and when a permanent site (PKD) was found, a third MT observatory was installed in 1999 with three induction coils, two 100 m electric dipoles, and two 200 m electric dipoles. PKD and PKD1 ran in parallel for one month in 1999, and then the MT observatory at PKD1 was closed.

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.

We have applied a variety of signal processing techniques to the 1sps array data. The 'raw' data stored on the NCEDC is windowed, Fourier transformed, and then band averaged, such that the time series of each data channel is represented by 25 shorter timeseries of band averaged Fourier coefficients (FCs). Details of the transform and band averaging can be found in *Eisel and Egbert* (2001). The majority of the signal processing applied to date is done using these FC files as the input time series, although some time-domain Weiner Filtering has been applied to the data around the time of the earthquake. An example of the time domain processing is given in the PKD-SAO EM Monitoring Array chapter of the 2003 BSL Annual Report.

The FC time series are first examined in terms of Signal to Noise Ratio (SNR). Robust estimates of SNR have been calculated daily over the four-year time window [2002-2005] for all 12 channels of data acquisition. The method of outlier downweighting used is called the RMEV estimation process (Robust multivariate errors-in-variables) and is referenced in *Eisel and Egbert* (2001). These plots show that the electric field SNRs at both sites correlate positively with global geomagnetic activity indices (K and A), and the Parkfield electric field SNRs correlate negatively with rainfall events.

The second step of the analysis is the calculation of robust estimates of apparent resistivity curves for Parkfield. These plots showed slight seasonal variation of apparent resistivity on the order of a few percent that also seem to correlate with rainfall. A distortion analysis applied to these curves [*Smith* 1995], where the MT impedance tensor Z( ,t) is decomposed into a frequency-dependent and a time-dependent part, shows that the seasonal variation is almost entirely accounted for by a frequency independent distortion tensor, and hence is a near-surface effect. This suggests that if any regional variation in apparent resistivity occurred coincident with the earthquake, that it is small enough to be masked by the seasonal effect. Robust RMEV techniques were applied in this analysis, and are similarly documented in *Eisel and Egbert* (2001).

The other two lenses through which we have carefully examined the 1Hz data are the use of Principal Components Analysis (PCA) and canonical coherence analysis (CCA), both of which are derived from the data covariance matrices. These techniques rely on rotating the coordinates of the data-space into directions of dominant characteristics. A discussion on the implications of PCA as applied to magnetotelluric data can be found in *Egbert* (1989). A discussion of the significance of CCA as applied to ULF EM data is included in *Kappler et. al.* (2006).

The 1Hz data do show several interesting characteristics, such as the seasonal dependence of apparent resisitivy and the rainfall correlation. In particular an examination of the apparent resistivity residual TS with the distortion effects removed has not been plotted to date. Also the 3rd and 4th principal components, as well as canonical coherences, are of interest in understanding sources of signal contamination, as these monitor the non-regional signals that the array is detecting. An examination of the 40Hz data in a short time-window around the earthquake (2 months on either side) remains to be done in order to see if any anomalous EM signals were occurring at higher frequencies. A statistical analysis of the time-domain residuals has been started for the 1Hz data, and will need to be summarized, and also performed with the 40Hz data. Though several interesting signals have been detected and isolated from the data, we have not detected any clear evidence for anomalous EM activity preceding earthquakes. Significant EM fields are detected coseismically (during ground shaking) and modeling of these fields has begun as a project for the 2006 SEG annual meeting.

Frank Morrison directs the MT program and collaborates closely with Gary Egbert of Oregon State University. Rich Clymer and Karl Kappler also contribute to the operation of the MT observatories.

Egbert, G.D., Booker, J.R., (1989), Multivariate Analysis of Geomagnetic Array Data 1.The Response Space, *Journal of Geophys. Res. V94* No.B10 pp14227-14247

Egbert G. D. (1997), Robust multiple-station magnetotelluric data processing, *Geophys. J. Int. v130* pp475-496

Eisel M., Egbert G.D., (2001), On the stability of magnetotelluric transfer function estimates and the reliability of their variances, *Geophys. J. Int. v144*, pp65-82

Gamble T.D., Goubau W.M., Clarke J. (1979), Magnetotellurics with a remote reference, *Geophysics, v44*, pp53-68

Kappler, K. Egbert, G.D., Morrison, H.F., Long-Term Analysis of ULF electromagnetic fields at Parkfield CA, *in progress*

Smith, J. T., (1995), Understanding Telluric Distortion Matrices, *Geophysical journal International, v122*, 219-226

Berkeley Seismological Laboratory

215 McCone Hall, UC Berkeley, Berkeley, CA 94720-4760

Questions or comments? Send e-mail: www@seismo.berkeley.edu

© 2006, The Regents of the University of California