Parkfield Borehole Network

Introduction

The operation of the High Resolution Seismic Network (HRSN) at Parkfield, California began in 1987, as part of the U.S. Geological Survey initiative known as the Parkfield Prediction Experiment (PPE) (Bakun and Lindh, 1985).

Figure 5.1 shows the location of the network, its relationship to the San Andreas fault, sites of significance from previous and ongoing research using the HRSN, relocated earthquake locations from 1987-1998.5, routine locations of seismicity since August 2002, and the epicenter of the 1966 M6 earthquake that motivated the PPE. The HRSN records exceptionally high-quality data, owing to its 13 closely spaced three-component borehole sensors (generally emplaced in the extremely low attenuation and background noise environment at 200 to 300 m depth (5.1)), its high-frequency wide bandwidth recordings (0-125 Hz), and its low magnitude detection threshold (recording events below magnitude -1.0).

Several aspects of the Parkfield region make it ideal for the study of small earthquakes and their relationship to tectonic processes. These include the fact that the network spans the expected nucleation region of a repeating magnitude 6 event and a significant portion of the transition from locked to creeping behavior on the San Andreas fault, the availability of three-dimensional P and S velocity models (Michelini and McEvilly, 1991), a seismicity catalogue that is complete to very low magnitudes and that includes at least half of the M6 seismic cycle, a well-defined and simple fault segment, a homogeneous mode of seismic energy release as indicated by the earthquake source mechanisms (over 90$\%$ right-lateral strike-slip), and the planned drilling zone and penetration and instrumentation site of the San Andreas Fault deep observatory at depth experiment (SAFOD) (see: http://www.earthscope.org/safod/index.html or http://www.iris.iris.edu/HQ/EarthScope/EarthScope.saf.html).

Figure 5.1: Map showing the San Andreas Fault trace, the location of the original 10 Parkfield HRSN stations (filled diamonds) and the 3 new sites installed to enhance coverage of the region containing the SAFOD facility (open diamonds), along with the BDSN station PKD (filled square). The locations of the 8 source points for the Vibroseis wave propagation monitoring experiment are represented by small black triangles. The epicenter of the 1966 M6 Parkfield main shock is located at the large open circle. The location of the pilot hole and SAFOD drill site is shown by the filled star, and the location of the 2 alternative M2 repeating earthquake targets (70 meters apart) are shown as concentric circles. Because of the SAFOD experiment, the 4 km by 4 km dashed box surrounding the SAFOD zone is a region of particular interest to BSL researchers. Routine locations of earthquakes recorded by the expanded and upgraded 13 station HRSN are shown as open black circles. Locations of events recorded by the earlier vintage 10 station HRSN, relocated using an advanced 3-D double-differencing algorithm applied to a cubic splines interpolated 3-D velocity model (Michelini and McEvilly, 1991), are shown as gray points. Station GHIB (Gold Hill, not shown) is located on the San Andreas Fault about 8 km to the Southeast of station EADB.
\begin{figure*}\begin{center}
\epsfig{file=hrsn03_map.ps, width=14cm}\end{center}\end{figure*}

In a series of journal articles and Ph.D. theses, we have presented the cumulative, often unexpected, results of U.C. Berkeley's HRSN research efforts (see: www.seismo.berkeley.edu/seismo/faq/parkfield_bib.html). They trace the evolution of a new and exciting picture of the San Andreas fault zone responding to its plate-boundary loading, and they are forcing new thinking on the dynamic processes and conditions within the fault zone at the sites of recurring small earthquakes.

HRSN Overview

1986 - 1998

The HRSN was installed in deep (200-300m) boreholes beginning in 1986. Sensors are 3-component geophones in a mutually orthogonal gimbaled package. This ensures that the sensor corresponding to channel DP1 is aligned vertically and that the others are aligned horizontally. In November 1987, the Varian well vertical array was installed and the first VSP survey was conducted, revealing clear S-wave anisotropy in the fault zone (Daley and McEvilly, 1990). During 1988, the original network was completed to a ten station 3-component 500 sps set of stations telemetered into a central detection/recording system operating in triggered mode and incorporating a deep (572 m) sensor in the Varian well string into the network. The Varian system was slaved in 1988, for about two years, to the Vibroseis control signals, allowing simultaneous recording of vibrator signals on both systems. For several years beginning in 1991, low-gain event recorders (from PASSCAL) were installed at several of the sites to extend the dynamic range to $M_{L}$ about 4.5. The data acquisition system operated quite reliably until late 1996, when periods of unacceptably high down time developed. During this period as many as 7 of the remote, solar-powered telemetered stations were occasionally down simultaneously due to marginal solar generation capacity, old batteries, and recording system outages of a week or more were not uncommon. In July of 1998 the original data acquisition system failed permanently. This system was a modified VSP recorder acquired from LBNL, based on a 1980- vintage LSI-11 cpu and a 5 MByte removable Bernoulli system disk with a 9-track tape drive, configured to record both triggered microearthquake and Vibroseis data (discontinued in 1994, Karageorgi et al., 1997). The system was remote and completely autonomous, and data tapes were mailed about once a month to Berkeley for processing and analysis. The old system also had a one-sample timing uncertainty and a record length limitation because the tape write system recovery after event detection was longer than the length of the record, leaving the system off-line after record termination and until write recovery was completed.

1998 - 1999

In December of 1998, the original HRSN acquisition system was replaced by 10 stand-alone PASSCAL RefTek systems with continuous recording. To process these data, development of a major data handling procedure was required, in order to identify the microearthquakes down to M = -1, since continuous telemetry to the Berkeley Seismological Laboratory (BSL) and application of a central site detection scheme was not an option at that time.

In July, 1999 we had to reduce the network to four RefTeks at critical sites that would ensure continuity in monitoring at low magnitudes and the archive of characteristic events for studying the evolution of their recurrence intervals. Properties of the 10 original sites are summarized in Table 5.2.


Table 5.1: Stations of the Parkfield HRSN. Each HRSN station is listed with its station code, network id, location, date of initial operation, and site description. The latitude and longitude (in degrees) are given in the WGS84 reference frame, the surface elevation (in meters) is relative to mean sea level, and the depth to the sensor (in meters) below the surface. Coordinates and station names for the 3 new sites are given at the bottom.
Site Net Latitude Longitude Surf. (m) Depth (m) Date Location
EADB BP 35.89525 -120.42286 499 245 01/1988 - Eade Ranch
FROB BP 35.91078 -120.48722 542 284 01/1988 - Froelich Ranch
GHIB BP 35.83236 -120.34774 433 63 01/1988 - Gold Hill
JCNB BP 35.93911 -120.43083 559 224 01/1988 - Joaquin Canyon North
JCSB BP 35.92120 -120.43408 487 155 01/1988 - Joaquin Canyon South
MMNB BP 35.95654 -120.49586 731 221 01/1988 - Middle Mountain
RMNB BP 36.00086 -120.47772 1198 73 01/1988 - Gastro Peak
SMNB BP 35.97292 -120.58009 732 282 01/1988 - Stockdale Mountain
VARB BP 35.92614 -120.44707 511 572 01/1988 - Varian Well
VCAB BP 35.92177 -120.53424 790 200 01/1988 - Vineyard Canyon
CCRB BP 35.95716 -120.55161 601 251 05/2001 - Cholame Creek
LCCB BP 35.98006 -120.51423 637 252 08/2001 - Little Cholame Creek
SCYB BP 36.00942 -120.53661 947 252 08/2001 - Stone Canyon



Table 5.2: Instrumentation of the Parkfield HRSN. Most HRSN sites have L22 sensors and were originally digitized with a RefTek 24 system. After the failure of the WESCOMP recording system, PASSCAL RefTek recorders were installed. In July of 1999, 6 of the PASSCAL systems were returned to IRIS and 4 were left at critical sites. The upgraded network uses a Quanterra 730 4-channel system. For the three new stations (bottom) horizontal orientations are approximate (N45W and N45E) and will be determined more accurately in the near future.
Site Sensor Z H1 H2 RefTek 24 RefTek 72-06 Quanterra 730          
EADB Mark Products L22 -90 170 260 01/1988 - 12/1998 12/1998 - 07/1999 03/2001 -          
FROB Mark Products L22 -90 338 248 01/1988 - 12/1998 12/1998 - 07/1999 03/2001 -          
GHIB Mark Products L22 90 failed unk 01/1988 - 12/1998 12/1998 - 07/1999 03/2001 -          
JCNB Mark Products L22 -90 0 270 01/1988 - 12/1998 12/1998 - 06/2001 03/2001 -          
JCSB Geospace HS1 90 300 210 01/1988 - 12/1998 12/1998 - 07/1999 03/2001 -          
MMNB Mark Products L22 -90 175 265 01/1988 - 12/1998 12/1998 - 06/2001 03/2001 -          
RMNB Mark Products L22 -90 310 40 01/1988 - 12/1998 12/1998 - 07/1999 03/2001 -          
SMNB Mark Products L22 -90 120 210 01/1988 - 12/1998 12/1998 - 06/2001 03/2001 -          
VARB Litton 1023 90 15 285 01/1988 - 12/1998 12/1998 - 07/1999 03/2001 -          
VCAB Mark Products L22 -90 200 290 01/1988 - 12/1998 12/1998 - 06/2001 03/2001 -          
CCRB Mark Products L22 -90 N45W N45E - - 05/2001 -          
LCCB Mark Products L22 -90 N45W N45E - - 08/2001 -          
SCYB Mark Products L22 -90 N45W N45E - - 08/2001 -          


Upgrade and SAFOD Expansion

Thanks to emergency funding from the USGS NEHRP, we have replaced the original 10-station system with a modern 24-bit acquisition system (Quanterra 730 4-channel digitizers, advanced software using flash disk technology, spread-spectrum telemetry, Sun Ultra 10/440 central processor at the in-field collection point, with 56K frame-relay connectivity to Berkeley). The new system is now online and recording data continuously at a central site located on the California Department of Forestry (CDF) fire station in Parkfield.

We have also added three new borehole stations at the NW end of the network as part of the SAFOD project, with NSF support, to improve resolution at the planned drilling target on the fault. Figure 5.1 illustrates the location of the proposed drill site (star), the new borehole sites, and locations of earthquakes recorded by the initial and the upgraded/expanded HRSN.

These three new stations use similar hardware to the main network, with the addition of an extra channel for electrical signals. Station descriptions and instrument properties are summarized in Tables 5.1 and 5.2. All HRSN Q730 data loggers employ FIR filters to extract data at 250 and 20 Hz (Table 5.3).


Table 5.3: Data streams currently being acquired at each HRSN site. Sensor type, channel name, sampling rate, sampling mode, and type of FIR filter are given. C indicates continuous; T triggered; Ac acausal; Ca causal. "?" indicates orthogonal vertical and 2 horizontal components.
Sensor Channel Rate (sps) Mode FIR
Geophone DP? 250.0 T Ca
Geophone BP? 20.0 C Ac


The remoteness of the drill site and new stations require an intermediate data collection point at Gastro Peak, with a microwave link to the CDF facility. The HRSN stations use SLIP to transmit TCP and UDP data packets over bidirectional spread-spectrum radio links between the on-site data acquisition systems and the central recording system at the CDF. Six of the sites transmit directly to a router at the central recording site. The other seven sites transmit to a router at Gastro Peak, where the data are aggregated and transmitted to the central site over a 4 MBit/second digital 5.4 GHz microwave link. All HRSN data are recorded to disk at the CDF site. A modified version of the REDI real-time system detects events from the HRSN data, creates event files with waveforms from the HRSN and sends the event data in near real-time to UC Berkeley. Currently the continuous data is being migrated to DLT tape when local disk space fills up, and the tapes are mailed to the BSL for long-term storage. Efforts are being made to acquire funding to make this data Internet accessible to the research community through the NCEDC.

The upgraded system is compatible with the data flow and archiving common to all the elements of the BDSN/NHFN and the NCEDC, and is providing remote access and control of the system. It is also providing data with better timing accuracy and longer records, which are to eventually flow seamlessly into NCEDC. The new system also solves the problems of timing resolution, dynamic range, and missed detections, in addition to providing the added advantage of conventional data flow (the old system recorded SEGY format).

2002-2003 Activities

Over the past year, activities associated with the operation of the HRSN primarily involved three components: 1) routine operations and maintenance of the network, 2) enhancement of the network's performance for detection and recording of very low magnitude earthquakes, and 3) routine data processing and analysis.

Operations and Maintenance

In addition to the routine maintenance tasks required to keep the HRSN in operation, various refinements and adjustments to the networks infrastructure and operational parameters have been needed this year to correct for pathologies that continue to manifest themselves in the recently upgraded and expanded system.

A feature of the new system that has been particularly useful both for routine maintenance and for pathology identification has been the Internet connectivity of the central site processing computer and the station data loggers with the computer network at BSL. Through this connection, select data channels and on-site warning messages from the central site processor are sent directly to BSL for evaluation by project personnel. If, upon these evaluations, more detailed information on the HRSN's performance is required, it can also be directly accessed. Analysis of this remotely acquired information has been extremely useful for trouble shooting by allowing field personnel to schedule and plan the details of maintenance visits to Parkfield. The connectivity also allows certain data acquisition parameters to be modified remotely when needed, and commands can be sent to the central site computer and data loggers to modify or restart processes when necessary.

The network connectivity allows analysts at the BSL to routinely perform checks on the system health of the HRSN and its data quality. One example of a technique used by BSL analysts involves the use of teleseismic arrivals from deep focus earthquakes. Since seismic waves from such events impose a near simultaneous and vertically incident plane wave of relatively uniform amplitude on all HRSN stations, seismograms from these events can be used to assess relative station responses across the network and help identify pathologies in station polarities, individual component failures and other response characteristics.

Figure 5.2 shows an example of a recent teleseism recorded on the DP1 (vertical) channel across the network. Not shown are recordings from stations MMNB and VARB. The initial display of seismograms from this teleseism showed these station components to be responding abnormally at the time of the earthquake. Based on this teleseismic result other remotely acquired information was uploaded from the HRSN and it was determined that these components were indeed malfunctioning. Subsequent field visits were then scheduled and the necessary repairs made.

Figure 5.2: Displayed are 30 seconds of 0.5-2.0 Hz BP filtered vertical ground velocity data for a $M_{w}$ 6.9 deep focus teleseism which occurred 6/23/2003 at 12:12 UT at a depth of 685 km in the vicinity of the Rat Islands in the Aleutian Islands chain (51.44N,176.78E). The traces have been ordered by increasing distance (top to bottom), their waveforms are absolute scaled to allow comparisons between the response functions between stations. The great circle distance to the HRSN is approximately 46.5 degrees with an azimuth of $\sim $ 310$^{\circ }$. The recording of this teleseism on the Northern Hayward Fault Network is show in Figure 4.2.
\begin{figure*}\begin{center}
\epsfig{file=hrsn03_teleseism.ps, width=15cm}\end{center}\end{figure*}

The network connectivity also allows remote monitoring of the background noise levels being recorded by the HRSN stations. For example shown in Figure 5.3 are power spectral density plots of background noise for vertical components of the 7 HRSN stations that are most critical for monitoring seismicity in the region containing SAFOD. The PSD analysis gives a rapid assessment of the HRSN seismometer responses across their wide band-width. By routinely generating these plots with data telemetered from Parkfield, changes in the seismometer responses, often indicating problems with the acquisition system, can be easily identified, and corrective measures can then be planned and executed on a relatively short time-frame.

Figure: Background noise PSD plot for the seven continuously telemetered BP.DP1 data streams from Parkfield. The data are 20 minute samples starting at 2003.225.0900 (2 AM PDT). The plots show the background noise PSD as a function of frequency for the the highest available sampling rate (250 sps) vertical component data which are continuously telemetered to Berkeley. Note the relatively low PSD levels and the overall consistency for all the HRSN stations. By comparison, the PSD curves among the borehole Northern Hayward Fault Network (NHFN) land and bridge stations (Figure 4.3) are much more variable and show a generally higher background noise level. On the other hand, PSD curves for the MPBO stations of the NHFN are much more consistent with the HRSN PSD's (Figure 8.6). The differences among the various station PSD's can, in large part, be explained by the relative cultural noise levels at the various stations, by the depth of the borehole sensors, and by whether the boreholes remain open holes (noisier) or have been filled with cement. The 2 Hz minimum in the PSD plots for the HRSN sensor results from the 2 Hz sensors used at these sites. Below 2 Hz, noise levels rise rapidly and the peak at 3 sec (.3 Hz) is characteristic of teleseismic noise observed throughout California. In the 2 to 5 Hz range, VCAB and JCNB have historically shown higher background noise which is believed to result from excitation modes in the local structure. A small 60 Hz blip can be seen in the SCYB curve due to its close proximity to a power-line.
\begin{figure}\begin{center}
\epsfig{file=hrsn03_psd.ps, width=8cm, clip=}\end{center}\end{figure}

Triggered event data for the HRSN is also telemetered in near real time to the BSL, and this allows for rapid evaluation of the triggered data. This year we have implemented a semi-automated waveform and trigger review procedure using a graphical user interface (GUI). This procedure is now being used to review the triggered waveform data daily to discriminate between earthquake and non-earthquake events and to pick P and S phases of the local events. In the process, our analyst/field technician also makes note of obvious problems with station/component specific earthquake recording, and this malfunction information is used to identify maintenance needs for the HRSN.

Enhancing HRSN Performance

Over the past year significant efforts were made to identify and reduce noise problems arising from the new and expanded data acquisition system. Detection, monitoring, and high-resolution recording of earthquakes down to the smallest possible magnitudes with the highest possible signal-to-noise (especially in the region of the proposed SAFOD drilling) is a major objective of the HRSN data collection effort. Consequently, elimination of all sources of unnaturally occurring system noise is a primary goal. The minimization of data loss due to station outages and data-dropouts is also critical to this objective.

The sophisticated HRSN data acquisition involves integration of a number of distinct components at each station (i.e., sensor, preamp, solar panels, solar regulator, batteries, Freewave radio, antenna, lightening arresters, and associated cabling, connectors and grounds) and radio telemetry apparatus between the seismic stations, telemetry relay stations, and the central processing site on the CDF site in Parkfield.

This complex integration of station and communication components combined with a variety of associated concerns (e.g., ground loops, cable resistances, radio feedback into recording equipment at stations, radio interference between stations, marginal line of site paths, cloud cover and solar power, the integration of older (pre-upgrade) hardware components with new components, old component deterioration and failures, and malfunctioning and unexpected performance characteristics of newer components) all make identification of specific causes of network generated (i.e. artificial) noise difficult to identify.

Exhaustive and iterative testing of HRSN performance has identified two primary causes for observed artificial noise remaining in the system (i.e. solar regulator spiking and preamp self-noise generation). We have designed and have implemented or are in the process of implementing fixes for these problems. We are also continuing to improve the HRSN event detection sensitivity by refining the HRSN triggering scheme.

Solar Regulators

Regularly occurring spikes occurring during the daylight hours were observed in the continuous data streams and found to be due to the solar regulators. We have tested a variety of solar regulator designs and have identified the Prostar 30 as having the optimal cost-benefit. We have purchased and installed several of these devices at several of the HRSN sites with the ultimate goal of installing the Prostar's at all the HRSN stations as time and funding permit.

Pre-amplifier Noise

We found that a significant source of artificial noise was coming from the station pre-amplifiers. In the upgraded system, preamps from the older network were used. During integration of the older preamps with the increased dynamic range capabilities of the 24-bit Quanterra system, gain settings of the preamps were reduced from x10,000 to x80 in order to match signal sensitivity of the new system with the older one. While these lower preamp gain levels are still within the operational design of the preamps, they are no longer in their optimal range and a significant contribution of preamp's self-generated noise is present in the recorded seismograms. Initially, this was not expected to be a significant problem. However, we have subsequently found that even the small increase in preamp noise that results from the preamp gain reduction significantly impacts the sensitivity of the network for detecting and recording the smallest locatable events.

Figure 5.4 shows the preamp noise reduction effect observed on background noise signals at three vertical components of the HRSN when gains are raised from x80 to x1,000. Considerable signal hash is seen at gain levels of x80 (top waveform in each station pair), and significantly reduced when gains are increased to x1,000 (lower waveforms). Since we are also interested in recording large earthquakes on-scale, simply increasing gain levels on all stations is not the preferred solution, since doing so causes the recording system to saturate at much lower magnitudes. Instead we are attempting to redesign the preamps using modern components to reduce the noise levels at the lower gain levels. However our attempts at redesign have not yet yielded satisfactory results.

Since a primary objective of the HRSN is to monitor the evolving patterns of the numerous small earthquakes that occur at very low magnitudes, and since this objective also complements the scientific objectives of the recently funded SAFOD experiment, it is important to address the preamp noise problem in a timely manner. We have opted, therefore, to raise the gain levels for the near-term on all the station preamps from x80 to x1,000. These gain changes are currently (late August, 2003) being implemented, and we estimate that the number of small earthquakes we will detect will increase by a factor of 2 to 3. We will continue investigating preamp redesigns until a suitable alternative is found at which time we will install the new preamps and lower the preamp gain back to x80-allowing both the increased detection of small events and the on-scale recording of events up to about magnitude 4 to 4.5.

Figure 5.4: Preamp noise reduction test. Shown are 30 seconds of vertical background signal recorded at stations EADB, FROB and JCNB on day 229 of 2003 at 0700 UTC (top of station pairs, recorded at x80 gain and scaled up by 1000/80 for comparison to the x1000 preamp gain levels) and 0700 UTC on day 233 (bottom of station pairs, recorded at x1000 preamp gain). Note the substantial reduction in background noise, due primarily to the lower preamp generated noise at higher preamp gain.
\begin{figure}\begin{center}
\epsfig{file=hrsn03_preamp.ps, width=8cm, bb= 25 51 590 745, clip=}\end{center}\end{figure}

Triggering Refinement

Additional efforts underway to increase event detection sensitivity include: 1) development of a station specific filtering scheme for input into the triggering algorithm, 2) refinement of the multi-station trigger association algorithm to include subnet triggering, and 3) incorporation of the pilot hole array data into the network triggering scheme to capture the smallest events in the SAFOD drilling area.

Routine Data Analysis

Monitoring the evolution of microseismicity, particularly in the SAFOD drilling and target zone, is a primary objective of the HRSN project. In addition, the continued analysis of the HRSN data for determining detailed seismic structure, for the study of similar and characteristic microearthquake systematics, for estimation of deep fault slip rate evolution, and for various studies of fault zone and earthquake physics is also of great interest to seismologists. Before advanced studies of the Parkfield microseismicity can take place, however, initial processing, analysis and routine cataloging of the earthquake data must be done. An integral part of this process is quality control of the processed data, including a final check of the routine catalog results.

Initial Processing

At this time, continuous data streams on all 39 components are being recorded at 20 and 250 sps on disk on the local HRSN computer at the CDF facility and when the local disk space is full, the continuous data is migrated onto DLT tape. The 20 sps data are transmitted continuously to the BSL over the frame-relay linked and archived at the NCEDC. In addition, the vertical component channels for the 7 stations critical to resolving seismicity in the SAFOD area are also being transmitted continuously to the BSL at 250 sps over the frame relay-circuit for purposes of quality control and fine tuning the triggering algorithm for the detection of the smallest possible events around SAFOD.

Shortly after being recorded to disk, event triggers for the individual station data are determined and a multi-station trigger association routine then processes the station triggers and identifies potential earthquakes. For each potential earthquake trigger, 30 second waveform segments are then collected for all stations and components, assigned a unique event identifier (compatible with the NCEDC classification scheme) and saved as an event gather. Event gathers are then periodically telemetered to BSL and included directly into the NCEDC earthquake database (dbms) for analysis and processing.

An ongoing effort has been the development of a new earthquake triggering scheme, with the goal of routinely detecting SAFOD area events to magnitudes below -1.0. A first cut version of the new scheme has been implemented and is already detecting earthquakes at an increased rate-nearly 3 times the number of earthquakes detected before the upgrade.

In order to facilitate the processing and archiving of this large number of events (approx. 150 per month), BSL personnel have recently developed a Graphical User Interface (GUI). The GUI is integrated with the NCEDC dbms and allows review of the waveforms from every triggered event. Initial analysis of the data using the GUI involves review of the waveforms and classification of the event as an earthquake or non-earthquake event. The GUI also allows the analyst to log potential network problems that become apparent from the seismograms. The HRSN analyst then classifies the event as a local, distant-local, regional, or teleseismic event and then systematically hand picks the P- and S-phases for the local and distant local events.

Picking of the numerous microearthquake events is no mean task. On average about 7 P-phases and 4 S-phases are picked for each event, putting the total number of annual phase picks for the HRSN data on the order of 19,000 to 20,000. We have experimented with algorithms that make initial auto-picks of the phase arrivals, but have so far found picking by hand to be an advantage since it forces the analyst to review each pick carefully while at the same time allowing him to assess the state of health of recording on each station-component in detail. In all our tests, repicked autopicks have also invariably resulted in catalog locations that are significantly more scattered and that have higher residuals than locations done with purely hand-picked data.

A peculiarity of processing very small earthquake data, is that multiple events commonly occur within a few seconds of one another (Figure 5.5). The close timing of these events does not allow the local triggering algorithm to recover from one event before another occurs. As a result, the central site processor often does not trigger uniquely for each event. In such cases only one, 30 sec waveform gather and one earthquake identifier will be created for all the events. These multiple earthquake records (MER) account for only 3 to 5$\%$ of the total seismicity recorded by the HRSN. However, there are times when this rate rises to over 10$\%$. In order to assign each event in an MER a unique event identifier for the NCEDC dbms and to make picking and automated processing of these events more manageable an additional feature of the GUI was developed that allows the analyst to "clone" MER into separate gathers for each event.

Figure 5.5: Five events occurring on the same MER. The P phase of the first event was not captured on this record. These five events occurred as part of a swarm of 47 small events recorded by the HRSN that occurred on day 211 of 2003. Of these 47 events, the NCSN catalog contains only 2. Events shown are all less than magnitude 0.
\begin{figure}\begin{center}
\epsfig{file=hrsn03_clone.ps, width=8cm, clip=}\end{center}\end{figure}

Quality Control

Once false triggers have been removed and picks for the events completed, quality control on the picks is made to ensure that all picks have phase and weights assigned, that extraneous characters have been removed from the pick files, that double station-phase picks have not inadvertently been made, and that no repicks of the same event had been accidentally made during any cloning that was performed.

Initial locations are then performed and phase residuals analyzed in order to determine whether severe pick outliers must be removed or adjusted. Unstable location solutions based on events with few picks are also assessed to see if the addition of marginal phases will improve the stability of the location determination.

After any required pick adjustments have been made, the events are then relocated, and combined with error information to allow ranking of the confidence of location quality.

These procedures have all been put in place and tested over the past year for the new HRSN configuration. Currently we have located 9 months of data recorded by the new HRSN (over 1300 events) and are staying current with ongoing seismicity and also moving backwards in time to pick and locate the earlier data collected since early 2001.

We now have enough data and are confident enough with the procedures to begin organizing the locations for formal inclusion into the NCEDC dbms and dissemination to the community. These efforts are now underway. We are also in the early stages of establishing a scalar seismic moment catalog for the new HRSN events that is also to be included in the NCEDC dbms.

Catalog Assessment

We continue to examine the ongoing earthquake data being collected by the HRSN in search of possible earthquake precursors. This includes quality control and evaluation of the routine earthquake catalog locations and analyses of the spatial and temporal distribution of the microseismicity in relation to the occurrence of larger earthquakes in the area and heightened alert levels declared as part of the Parkfield Prediction Experiment. Even before our planned enhancement of HRSN performance, the new central detection system that operates at the telemetry hub, along with real-time telemetry of selected high-sensitivity channels to Berkeley for monitoring, allows event detection below magnitude 0.0. As a result, the rate of earthquake detection by the HRSN exceeds that of the NCSN by about a factor of 3 in the 30 km stretch of the SAF centered at the location of the 1966 M6 Parkfield event (Figure 5.6). The additional rate of HRSN event detection significantly increases both the spatial and temporal coverage of the changing seismicity patterns and provide unique additional information on the earthquake pathology at very low magnitudes. With our planned noise reduction and triggering enhancement, we estimate the proportion of HRSN located events relative to the NCSN catalog to increase by an additional factor of 2. Differences between earthquake locations evident in Figure 5.6 are largely attributable to the more advanced 3-D P- and S- wave velocity model used in determining the HRSN locations and the more accurate hand-picked P- and S- phases made possible by the high sampling rate (250 sps) and horizontal component borehole recordings of the HRSN.

Figure 5.6: Comparison of NCSN and HRSN catalog locations for the period September through November of 2002. During this period, magnitude M3.8 and M4.2 earthquakes occurred at about $-13$ and $+ 3$ km NW, respectively (gray disks). The proposed SAFOD drilling target is shown as an asterisks and a 4x4 km gray box of 6km depth is shown surrounding the target (corresponding to the 4x4 km box in Figure 5.1). The region shown is centered on the hypocentral region of the 1966 Parkfield M6 earthquake that occurred at 0 km at about 9 km depth. The lower magnitude detection and greater rate of microearthquake detection by the HRSN provides increased spatial coverage and detail in the temporal pattern of the evolution of seismic activity in the region. Station coverage in the region is comparable for both networks, yet the more accurate S- phase picks possible on the horizontal HRSN component seismograms and the use of a 3-D P and S velocity model for hypocentral inversion provides a sharper picture of the fault zone structure. On average the current detection rate of locatable earthquakes by the HRSN is about 3 times that of the NCSN. Planned enhancements for the HRSN are expected to increase rate of locatable earthquakes by an additional factor of 2 to 3.
\begin{figure*}\begin{center}
\epsfig{file=hrsn03_locs.ps, width=15cm, clip=}\end{center}\end{figure*}

Acknowledgements

Thomas V. McEvilly, who passed away in February 2002, was the PI on the HRSN project for many years, and without his dedication and hard work the creation and continued operation of the HRSN would not have been possible. His contributions continue to be appreciated in the extreme and the fruits of his labor many-fold.

Under Bob Nadeau's and Doug Dreger's general supervision, Rich Clymer, Wade Johnson, Bob Uhrhammer, Doug Neuhauser, Don Lippert, Bill Karavas, John Friday, Pete Lombard, and Lane Johnson all contribute to the operation of the HRSN. Bob Nadeau prepared this chapter with the assistance of Bob Uhrhammer and Wade Johnson.

During the period of this report, the operation and maintenance of the HRSN and the processing and archiving of its data was supported in large part by the USGS, through the NEHRP External Grants Program (grants: 02HQGR0067 and 03HQGR0065). NSF also provided support for the expansion of the HRSN near the SAFOD drill site through grant EAR-9814605.

References

Bakun, W. H., and A. G. Lindh, The Parkfield, California, prediction experiment, Earthq. Predict. Res., 3, 285-304, 1985.

Daley, T.M. and T.V. McEvilly, Shear wave anisotropy in the Parkfield Varian Well VSP, Bull. Seism. Soc. Am., 80, 857-869, 1990.

Karageorgi, E., R. Clymer and T.V. McEvilly, Seismological studies at Parkfield. IV: Variations in controlled-source waveform parameters and their correlation with seismic activity, 1987-1994, Bull. Seismol. Soc. Am., 87, 39-49, 1997.

Michelini, A. and T.V. McEvilly, Seismological studies at Parkfield: I. Simultaneous inversion for velocity structure and hypocenters using B-splines parameterization, Bull. Seismol. Soc. Am., 81, 524-552, 1991.

Berkeley Seismological Laboratory
215 McCone Hall, UC Berkeley, Berkeley, CA 94720-4760
Questions or comments? Send e-mail: www@seismo.berkeley.edu
© 2004, The Regents of the University of California