At a fundamental level, the success of the proposed Parkfield fault zone drilling project hinges on our ability to drill into the seismically active plane (or planes) of the San Andreas fault (SAF), and to identify those planes in the borehole. Current estimates by two groups of the location of the shallowest microearthquakes at the drill site (the drilling target) vary in position by nearly 1 km, and are thus inadequate to serve as the drilling target. We propose to carry out two interrelated tasks to address this issue: (1) review, assess, and revise existing three-dimensional velocity models of the Parkfield area using available data; and (2) conduct an active source/passive source calibration experiment prior to and during the drilling of the vertical hole to provide the drilling engineer with an absolute target for the deviated borehole. Results from the first task will help us to design the network for the second task, and observations from the second task will feed back into the models we create in the first task. While we believe the above tasks are essential for guiding the drilling, they are not designed to provide any detailed information about the character of seismicity or the structure of the fault zone outside of the vicinity of the borehole, something that is necessary to provide a context for the drilling results. Therefore, we propose an additional task: (3) to collect and analyze data from a dense seismic array, to be deployed centered on Middle Mountain, that will produce a high resolution 3D image of the SAF zone at Parkfield. The principal motivation of this task is to help interpret the in-situ observations from the Parkfield borehole (downhole observations and cores) in the context of the seismically-imaged variations in structure along the fault. The specific objectives are (a) to derive well-constrained 3D P- and S-wave velocity models that will result in improved accuracy and precision in the location of earthquakes in the Parkfield area, (b) to investigate heterogeneities in seismic velocity structure that may indicate lithologic variations and/or the presence and distribution of fluids in the fault zone, and how the variations along strike and with depth relate to the pattern of seismic behavior (aseismic, background earthquakes, main shock rupture zone). Results from previous work by the PI's on a similar segment of the SAF near Cienega Valley strongly suggest the presence of fluids and their prominent role in controlling fault behavior. Both logistically and scientifically, the third task dovetails nicely with the first two. An added benefit to this project is the opportunity to reduce the ambiguity in the interpretation of the seismic images by combining them with the results of existing resistivity images at Parkfield, and tying them to in-situ information to be obtained from the drilling project. This project is a collaborative effort between Steve Roecker (RPI) and Cliff Thurber (UW-Madison), and will be carried out in close cooperation with the USGS in Menlo Park and other planned efforts related to drilling at Parkfield.

The project proposed here addresses two questions of fundamental importance to the success of the Parkfield fault zone drilling project [Hickman et al., 1994]:

1. Where exactly is the seismogenic zone in relation to the surface trace of the San Andreas fault?

2. What is the relation between the properties of the rocks in the borehole and the fault zone as a whole, and what are the implications of these properties for earthquake generation?

The first question has some interesting scientific implications but for the most part is utilitarian: we do not know where this zone is to an accuracy sufficient to guarantee that the borehole will pass through it. The best information we have at present is derived from two studies at Parkfield [Michelini and McEvilly, 1991; Eberhart-Phillips and Michael, 1993] in which hypocenters were located while 3D seismic images of the SAF and its surroundings were determined simultaneously. Unfortunately, the results of these studies conflict - the epicenters of Michelini and McEvilly [1991] (the "UCB" model) are systematically about 1 km or more southwest of the SAF surface trace, while those of Eberhart-Phillips and Michael [1993] (the "USGS" model) align directly along the fault trace (Figure 1). This difference is too substantial to allow a single drilling scenario to intersect them both. Without additional investigation into this question, the odds of positioning the drill correctly are the same as a coin toss. Clearly, a strategy for addressing this question is absolutely crucial for the success of the drilling project.

(a)

(b)

Figure 1. (a)  Illustration of the uncertain location of the seismogenic zone and the geometry of the proposed Parkfield borehole.  The location uncertainty must be reduced to assure successful targeting of the borehole. (b) Fault-normal cross-section comparing UCB locations (+) to USGS locations (o). The UCB locations are systematically about 1 km SW of the USGS locations.

The second question goes to our ability to extrapolate the properties of rocks taken from a point on the fault to the larger scale properties of the fault zone. Some of the best constraints on regional fault zone properties come from 3D P- and S-wave velocity models derived from high resolution seismic imaging. Such models will allow us to investigate the variations in fault zone seismic properties along strike and with depth, and to understand the relation of structure to seismic activity by better constraining hypocenter locations. Moreover, there is a synergistic relation between the seismic imaging and the drilling results - drilling can provide ground truth for interpreting the tomograms, and the tomograms can allow extrapolation of the drilling results. Finally, the combination of these observations with those from other geophysical studies proposed as part of the drilling project will provide useful constraints on interpretation.

The results of previous seismic investigations regarding velocity structure and hypocenter locations in this region are not adequate either for guiding drilling operations or for interpreting and extrapolating the observations to be obtained by drilling. Three prior studies have determined 3D seismic images of the SAF and its surroundings at Parkfield using local earthquake arrival times (two in combination with active source travel times); two just for Vp [Lees and Malin, 1990; Eberhart-Phillips and Michael, 1993] and the other for Vp and Vs [Michelini and McEvilly, 1991]. The models had a grid spacing [Eberhart-Phillips and Michael, 1993; Michelini and McEvilly, 1991] or lateral resolution [Lees and Malin, 1990] of at best 2 km. All three 3D Vp models are dominated by a velocity contrast across the SAF (faster to the southwest, slower to the northeast); additionally, Michelini and McEvilly [1991] report a zone of relatively high Vp/Vs (about 2.0) near the hypocentral region of the 1966 Parkfield earthquake (Figure 2). These models are intriguing and provide useful insights into the geology at depth, but they are too coarse to resolve significant details of the fault zone, and, as discussed above, the USGS and UCB models yield hypocenters that differ by about a kilometer.

There have been several active-source studies of the SAF near Parkfield. McBride and Brown [1986] and Louie et al. [1988] analyzed data from a 1977 COCORP reflection line through the town of Parkfield, crossing the fault near 36 54' N, about 10 km SE of Middle Mountain. McBride and Brown [1986] concluded that the SAF zone probably had the form of a flower structure in the upper 7 km. Louie et al. [1988] found evidence for several reflectors with a range of dips near the SAF in the upper 5 km, but did not succeed in imaging the SAF itself. Trehu and Wheeler [1987] modeled data from a reflection/refraction line through the SE Cholame Valley, crossing the fault near 36 42' N. The SAF was near the end of their profile, so their constraints were somewhat poor, but they concluded that the SAF was a low-velocity zone penetrating to perhaps 15 km depth. Shedlock et al. [1990] carried out high-resolution shallow reflection profiles in Cholame Valley, targeting the shallow structure of the SAF along the en echelon fault offset through the valley. They found evidence for a mesh-like system of small faults beneath Cholame Valley connecting the offset strands of the SAF. A small 1994 reflection survey southwest of Middle Mountain was able to image the highly-faulted sediments down to the underlying Salinian granite bedrock at about 1 km depth [Unsworth et al., 1997]. These studies provide information on the structure and complexity of the SAF in the Parkfield area, but they do not contribute substantial evidence for understanding the fine-scale structure of the fault zone at Middle Mountain, nor for resolving the critical uncertainty in the location of the seismogenic zone.

Finer-scale models of fault zone structure have been derived from fault-zone guided waves (FZGW's). Li et al. [1990] interpreted forward models of FZGW's generated by earthquakes observed at Middle Mountain as evidence for a ~500 m wide fault zone (partitioned into a 100 to 160 m "core layer" and a 350 to 400 m "transition layer") that has significantly reduced velocities compared to the country rock on either side. Guided wave studies provide high-resolution information about the fault zone velocity structure (~ 100 m), and some limited information about structural variations with depth may be provided by comparing earthquake- and explosion-generated FZGW's [Li et al., 1997]. The existing results around Parkfield [Li et al., 1990, 1997] are in the form of one-dimensional models with velocity varying perpendicular to the fault only, with no direct information about variations along fault strike.

Another geophysical image of fault zone structure in the Parkfield region is provided by studies of resistivity conducted by Unsworth et al. [1997a, b]. Using continuous MT profiling with a 100 m interval, they achieved a spatial resolution on the order of 500 m to a depth of about 3 to 4 km. Their images of the fault zone on three fault-crossing profiles show low-resistivity zones on the order of 1 km wide, with an inner core zone of very low resistivity about 500 m wide (Figure 3). This result is similar to that of the guided wave analysis [Li et al., 1990, 1997]. Unsworth et al. [1997a, b] infer the low resistivities to be evidence for fluids within the fault zone.

Our success in carrying out a seismic investigation similar to that proposed here along an analogous part of the SAF in the Cienega Valley region south of Hollister [Thurber et al., 1996, 1997; Chen et al., 1996, 1997] gives us a high level of confidence that a detailed seismic array study focused on Middle Mountain will contribute significantly to a successful drilling program. Our recent study provides an excellent example of the improved definition of velocity structure and fault zone geometry that will result from a carefully planned study. We operated a dense array of 48 IRIS-PASSCAL instruments in the Cienega Valley region (adjacent to the Northern Gabilan Range, at the opposite end of the SAF creeping segment from Middle Mountain) from November 1994 to May 1995, and in conjunction with the USGS carried out a companion refraction experiment in May 1995 [Thurber et al., 1996]. An inversion of that data [Thurber et al., 1997] for 2D models of the Vp and Vp/Vs structure (Figure 4) shows that the data from this array are capable of imaging lateral variations in the fault zone on a scale of 1 km and constraining hypocenter locations to about 100-200 meters.

Our inversion solves directly for Vp/Vs perturbations, starting from an initial model with constant Vp/Vs. The Vp/Vs structure is expected to be more diagnostic of fluid content than the Vp structure, as discussed by Eberhart-Phillips et al. [1995]. Strong constraints on shallow Vp structure from the active experiment combined with well-constrained hypocenters from abundant earthquake S-wave data yielded a model that was extremely robust, as evidenced by a multitude of inversion tests done using both real and synthetic data. Significant among the results of the study were that (1) the seismic zone associated with the SAF is not vertical but dips about 75 degrees to the SW and (2) this zone follows the SW edge of a region of low Vp and high Vp/Vs that extends from the surface trace of the SAF to about 5 km depth (Figure 4). A comparison of the Vp and Vp/Vs seismic images led Thurber et al. [1997] to propose a fluid-rich fault zone as a viable interpretation of these results. Our efforts are now focusing on 3D inversions, using two different model parameterization and ray tracing approaches for comparison. Our preliminary results indicate a strong correlation between the along-strike variations in depth to seismicity and the depth extent of the high Vp/Vs zone (Figure 5).

Figure 5. Along-fault cross-section from preliminary 3D model for Vp/Vs structure (0.05 contour interval) with hypocenters of 24 years of M 3.0 and above earthquakes from the USGS catalog overlain (open circles, size scaled by magnitude, with the largest having M 4.5). Note how the events fall exclusively in zones with Vp/Vs below 1.95, suggesting a strong connection between seismogenic capability and Vp/Vs.

At Middle Mountain, the Unsworth et al. [1997a] MT model has the zone of lowest resistivity in a depth range of 0.5 to 2 km, compared to a depth range of 1 to 3 km for the zone of highest Vp/Vs at Cienega Valley. Unfortunately, the MT data analyzed in that paper could not constrain the bottom depth of the low resistivity zone - the longer profiles carried out by Unsworth et al. [1997b] may be able to resolve that uncertainty. Unsworth and coworkers will also be carrying out an MT survey in Cienega Valley in 1998 and 1999. It will be extremely useful to compare seismic and MT results at both Middle Mountain and Cienega Valley to see if the two methods produce compatible images of the presumed fluid-rich zones at the two locales.

Although they were obtained at two different sections of the SAF, the Middle Mountain resistivity study and the Cienega Valley seismic study were both conducted along the creeping portion of the fault, and the results highlight the potential importance of fluids in the fault zone. Also, both Thurber et al. [1997] and Unsworth et al. [1997] find that the seismically active zone is located on the edge of the inferred fluid-rich zone, although in the latter case the earthquake locations were determined independently [Nadeau et al., 1995]. The correspondence of the conclusions of these studies is intriguing, but somewhat tenuous as they are from different parts of the fault. Clearly, what is required are both 3D seismic and MT imaging at Middle Mountain, so that the resistivity and velocity/hypocenter patterns can be interpreted jointly. The Unsworth et al. [1997a, b] and Thurber et al. [1997] studies have demonstrated what is required to achieve the goal of adequately defining the fault zone structure and geometry at depth into the seismogenic zone.

To conclude, the importance of an unambiguous determination of the seismogenic zone prior to drilling and the need for a detailed characterization of the subsurface structure of the fault in order to reliably extrapolate the results of drilling are both clear. Our results from Cienega Valley show that additional seismic investigations such as those we propose here are extremely useful in this regard. In particular, the Cienega study shows that the main seismically active part of the fault zone (1) is significantly offset from the surface trace, by 2 km or more, and (2) is along the boundary between zones of sharply contrasting Vp and Vp/Vs values (high Vp and low Vp/Vs to the SW, low Vp and high Vp/Vs to the NE). If one had decided to drill into the SAF at Cienega Valley under the assumption that the seismically active zone lay directly beneath the SAF surface trace, then drilling would have intersected the fault well above the seismogenic section, and the primary target would have been missed completely. It is thought that the situation at Middle Mountain is not so extreme - competing seismic models have hypocenter differences of about 1 km. The project we propose should succeed in reducing the hypocenter uncertainty by a factor of 5 or more, based on preliminary simulations we have carried out, and will also provide vital information regarding the spatial variation of properties along the fault zone.

We propose to address the questions discussed in the previous section by carrying out additional seismological investigations of the Parkfield area. These investigations can be divided into three specific, but interrelated, tasks: (1) review, assess, and revise existing three-dimensional velocity models of the Parkfield area using available data, (2) deploy 15 PASSCAL instruments for 12 months prior to drilling to provide crucial active and passive calibration data, and (3) collect and analyze data from a dense seismic array, to be deployed centered on Middle Mountain, to produce a high resolution 3D image of the San Andreas fault (SAF) zone at Parkfield. Tasks 1 and 2 focus on the question of improving the location accuracy of the seismogenic zone in preparation for the drilling stage, and are strongly interrelated. Results from Task 1 will help us to design the network for Task 2, and observations from Task 2 will feed back into the models we create and testing we carry out as part of Task 1. Task 3 focuses on the question of how to interpret the in-situ observations from the Parkfield borehole in the context of the seismically-imaged variations in structure along the fault. The primary data analysis effort for Task 3 will be the joint seismic imaging of Vp and Vp/Vs structure using data from local earthquakes and calibration explosions, using multiple existing inversion codes for comparison. The principal objectives of Task 3 are (a) to derive well-constrained 3D P- and S-wave velocity models that will result in improved accuracy and precision in the location of earthquakes in the Parkfield area, and (b) to investigate heterogeneities in velocity structure that may indicate lithologic variations and/or the presence and distribution of fluids in the fault zone, and how the variations along strike and with depth relate to the pattern of seismic behavior (aseismic, background earthquakes, main shock rupture zone).

Details of Tasks 1 and 2

As discussed above, current knowledge of the seismic velocity structure and the location of the fault zone at depth at Middle Mountain are not adequate for planning and guiding drilling operations. In particular, the discrepancies between UCB model and the USGS model are too significant to be ignored when deciding where to drill. We note that while both groups used similar analysis techniques, the USGS model was based mainly on USGS data, whereas the UCB model used data primarily from UCB stations (and included both P and S waves). Thus, differences in the results are more likely due to the use of different data than different inversion methods. Our first objectives, therefore, will be to generate a third model based on a combination of these data sets, and to perform some hypothesis testing using the two existing models.

It is probable that the analysis of a merged dataset will lead to an improved solution. However, given the distribution of stations used in these studies, it is likely that the fundamental ambiguity in hypocenter locations will not be completely resolved. Therefore, we believe that the collection of additional arrival time and travel time observations will be necessary to solve this problem, and we plan to deploy a small number (15) of short period stations to procure the necessary additional data. The key question then is where should we install these stations in order to obtain information that will allow us to discriminate between models? Our approach is to do some hypothesis testing to find out where the two candidate models make significantly different predictions, and then occupy those sites. For example, one could compute arrival time surfaces for earthquakes in each of the UCB and USGS models and determine those locations where the differences in these surfaces are the greatest. Results of a preliminary investigation along these lines (Figure 6a) suggests that such points should be easily identifiable.

Arrival times are, of course, more difficult to interpret than travel times because of the indeterminate shift of residuals due to the unknown origin time. Fortunately, the proposed drilling scenario will allow us to observe travel times from "virtual earthquakes" during the first phase of drilling. Specifically, we will conduct two surface-to-borehole travel-time calibration measurements in the vertical borehole at depths of approximately 1 and 2 km. A borehole seismometer (provided by LLNL) will be deployed in the hole to record small explosive charges that will be detonated at (or immediately adjacent to) the nearby temporary and permanent seismographic stations. The results of this travel-time calibration experiment can then be used in a similar fashion to perform hypothesis testing, by comparing the travel time predictions of competing models to those actually observed. Figure 6b shows a preliminary calculation of predicted travel time differences between the USGS and UCB models (2D approximations) for a profile of surface sources to a downhole receiver at a point 2 km deep in the borehole. These calculations will be repeated in 3D. Additionally, because of the proximity of the hole to the seismogenic zone, we can use the travel time data to relocate previously recorded hypocenters in a master event mode.

Figure 6b. Difference in travel time (dashed line) between 2D approximations of the USGS and UCB models for surface sources along a fault-normal profile to receivers at 2 km depth 1.5 km SW of the fault trace, which is located at a horizontal distance of 25 km in the model.

By combining the travel time observations of microearthquakes from the target zone recorded by the combined network of temporary and permanent stations with the "virtual earthquakes" provided by the surface-to-borehole data, we will be able to give the drilling engineer a precise target for the deviated borehole. We believe that the combination of surface recordings of earthquakes and borehole recordings of surface shots will provide the strongest possible constraints for deriving an accurate drilling target.

Details of Task 3

The proposed dense seismic array that we plan to deploy subsequent to the main drilling phase is intended to allow us to generate high resolution 3D P- and S- wave tomographic images of the fault zone. The parameters of the array will be modeled after our successful passive/active seismic imaging project along the Northern Gabilans segment of the SAF [Thurber et al., 1995, 1996, 1997; Chen et al., 1996, 1997]. The seismic array deployment will consist of a long-term (about 9 months) 60-station passive array, and will include a set of about 15 explosions for calibration purposes. Over the past three years, the area covered by the proposed array has experienced an average of 15 earthquakes per month of M > 1.0, so we can expect to record a sufficient number of earthquakes in a 9 month period. The idealized deployment plan for the 60 station passive array and the calibration experiment is shown in Figure 7. The proposed array and calibration experiment layouts are designed on the basis of inversion simulations done to assess the resolving capability of the array. If available, we will make use of a PASSCAL 30-station telemetered system for half of the array, resulting in reduced servicing demands. As in the Cienega Valley project, we anticipate assistance from and direct involvement of USGS personnel in the field work at Parkfield.

Figure 7. Idealized deployment plan for the Middle Mountain passive array and calibration experiment. Gray triangles are the passive instruments while the larger black triangles are existing UC-Berkeley and USGS network stations. Stars are the 15 shotpoints.

Our experiment plan has been developed based on our recent experience in the Cienega Valley and on synthetic inversion simulations using hypothetical data for earthquakes around Middle Mountain whose locations were drawn from the USGS catalog. Simulations using an experiment geometry similar to that proposed here (but with somewhat greater station and shot spacing) result in successful recovery of the structure and event locations. The synthetic inversions clearly indicate that the data from the existing network at Parkfield, even if supplemented by a sizable active experiment, are inadequate for deriving an accurate kilometer-scale model of the fault zone velocity structure.

The principal data analysis effort will be the joint seismic imaging of 3D Vp and Vp/Vs structure using P and S wave data from the explosions and local earthquakes, using multiple inversion methods for comparison. This will provide a well-calibrated velocity structure for determining accurate absolute locations for contemporary seismicity, and, using joint location techniques, nearly as accurate locations for catalog seismicity. The earthquake locations and their spatial relationship to the fault zone observed at the surface and imaged with MT provide critical information for interpreting and extrapolating the results of fault-zone drilling. Unsworth et al. [1997a] suggest that the active surface fault trace lies near the northeastern edge of the fault zone, as defined by the zone of very low resistivity, while the earthquakes occur in a zone about 1 km to the southwest. Using a combination of passive array and permanent network data, we should be able to determine the locations of earthquakes to an accuracy on the order of 100 meters (based on a test of location accuracy for the 1995 Cienega Valley explosions and synthetic simulations), and thus be able to better constrain the spatial relationship between low resistivity and seismic activity.

The proposed project has a duration of 3 years, and the schedule of activities is governed by the drilling schedule. Because most of the results from Tasks 1 and 2 have bearing on the location of the borehole, all of our efforts will be directed towards these tasks during the first year. Because of the importance of the hypothesis testing phase of Task 1 in the overall success of the project, we have already begun some preliminary work in this area and should have a general idea of where we want to deploy the small scale array when this project officially begins. Over the 12 month duration of the deployment we will be updating our models as data is collected, and moving stations as required to reduce ambiguities.

For Task 1, we will combine the datasets analyzed by Michelini and McEvilly [1991] and Eberhart-Phillips and Michael [1993] and incorporate additional data from more recent earthquakes and from a small pilot refraction experiment carried out at Middle Mountain in 1994. For Task 2, the calibration array will consist of 15 short period seismometers connected to standard PASSCAL equipment (RefTek DAS, solar panels, power boards, etc.). We plan to deploy this equipment in April, 1999. This array will operate for 12 months in this calibration mode, after which it will be become part of the dense array. Frequent inspections will be necessary to ensure a high rate of data recovery and rapid analysis. We therefore plan monthly maintenance trips to download data and check on equipment performance.

A key part of the calibration array task involves the collection of explosion data. When the initial drilling phase begins, we plan to subcontract the USGS to shoot a series of small shots from holes augured at the seismic stations and recorded by a seismometer placed in the borehole (this sensor will be installed and operated by our USGS coworkers). By reciprocity, we may then observe travel times for rays traveling between each station and "virtual earthquakes" near the seismogenic zone. We plan two sets of explosions: one set with the downhole sensor located about half-way down the hole, and a second set with the sensor at the bottom of the hole. This procedure will allow us to estimate the effects of source depth on travel time, and additionally will give us two "master events" to relocate events at different depths in the seismogenic zone. At the scale of the array we anticipate installing, we will be able to put sufficient energy into the ground with small 10 lb. shots and boosters, which makes this part of the operation relatively inexpensive. The utility of travel time observations in the context of better defining earthquake locations makes this data set one of the most valuable we will collect.

The second year will involve the execution of the main field campaign (the tomography array) and data archiving, and the third to tomographic analysis and interpretation. Ideally, the passive array installation will take place in spring of 2000 (~ mid-April), with array pull-out in early winter (~ late December). The main array calibration experiment probably would be carried out in May, again using a subcontract to the USGS to handle the drilling and shooting. This will be done in coordination with other proposed experiments, for example, the "long-line" refraction/wide-angle reflection profile and the borehole seismometer array experiment.

The availability of an efficient PASSCAL database system will mean that a full archive of the array data will be complete shortly after the field work is over. This will permit a preliminary analysis of the data at the very beginning of Year 3. We will also construct a complementary archive of network data during the field deployment period using the Northern California Data Center. Thus we will be able to proceed promptly with data analysis and inversion.

In Year 3, data analysis will proceed in parallel at UW and RPI, using different inversion methods with different model parameterizations, ray tracing, and inversion algorithms. Starting models will be created based on the results of previous studies. 3D models will be constructed using the full dataset with nominal model node spacing of 500 m to 1 km near the fault. In addition, we will derive a pair of 2D models centered on two fault-normal "profiles" each using half the array - one 2D model using events and stations from the NW half, and the other using events and stations from the SE half (Figure 7), with a node spacing of 500 m near the fault. The resulting 2D and 3D models of Vp and Vs (equivalently Vp/Vs) along with the relocated seismicity are the essential products of Year 3 of the project.

While the tasks we discuss here have substantial scientific merit on their own, they are designed to compliment other investigations that are being proposed as part of the larger drilling project. Certainly a major focus of our design is on the drilling effort itself, with the tuning of our tasks both to guiding the drill to the seismogenic zone and to the extrapolation of the drilling results to other parts of the fault.

Beyond the fundamental connection to drilling, our investigation will benefit from the other proposed active and passive seismic investigations that potentially could provide additional constraints on structure. Among these are the borehole array [Malin, personal communication, 1998], which will be about 4000' long and consist of 80 3-component sensors at 50' group intervals. These sensors will be a mix of high-gain seismometers and 3-component accelerometers, all recorded by a surface A/D interface and computer. The instrumentation is similar to that in the Varian borehole array. The possibilities of using this array for structural studies in concert with the surface array and shots are tremendous. Our project also compliments studies focused on details of fault zone structure [Li, personal communication,1998] in terms of the added information that waveforms provide to arrival time tomography, and, conversely, the resolution of laterally varying structure provided by arrival time tomography for waveform studies. Finally, the cross-sectional constraints on structure provided by the proposed active studies [Prodehl and Ryberg, personal communication, 1998; Rymer et al., personal communication, 1998] will provide additional constraints on the very shallow structures that are typically poorly imaged with hypocenters.

These and other related projects are summarized in the main drilling proposal and detailed in the individual proposals. In short, while the project proposed here is designed for a minimum of overlap with other activities in the area, there clearly is a substantial amount of mutual benefit to be derived if all of these various projects are carried out.

Bakun, W. H., Stewart, R. M., Bufe, C. G., and Marks, S. M., Implication of seismicity for failure of a section of the San Andreas fault , Bull. Seism. Soc. Am., 70, 185-201, 1980.

Chen, Y. H., S. W. Roecker, C. Thurber, and W. Lutter, P-wave precursors and tomographic images of the San Andreas fault near Hollister, California, EOS, Trans. Am. Geophys. Un. Suppl., 77, F466, 1996.

Chen, Y. H., S. W. Roecker, C. Thurber, and W. Lutter, High resolution three-dimensional tomographic imaging of the San Andreas fault in the Bear Valley region of central California, EOS, Trans. Am. Geophys. Un. Suppl., 78, F454, 1997.

Eberhart-Phillips, D., and A. J. Michael, Three-dimensional velocity structure, seismicity, and fault structure in the Parkfield region, central CA, J. Geophys. Res., 98, 15,737-758, 1993.

Eberhart-Phillips, D., W. Stanley, B. Rodriguez, and W. Lutter, Surface seismic and electrical methods to detect fluids related to faulting, J. Geophys. Res., 100, 12,919-936, 1995.

Hickman, S., M. Zoback, L. Younker, and W. Ellsworth, Deep scientific drilling in the San Andreas fault zone, EOS Trans. AGU, 75, 137-142, 1994.

Li, Y.-G., W. L. Ellsworth, C. H. Thurber, P. Malin, and K. Aki, Fault-zone guided waves from explosions in the San Andreas fault at Parkfield and Cienega Valley, California, Bull. Seismol. Soc. Am., 87, 210-221, 1997.

Li, Y., P. Leary, K. Aki, and P. E. Malin, Seismic trapped modes in the Oroville and San Andreas fault zones, Science, 249, 763-766, 1990.

Louie, J. N., R. W. Clayton, and R.J. LeBras, Three-dimensional imaging of steeply dipping structure near the San Andreas fault, Parkfield, California, Geophysics, 53, 176-185, 1988.

McBride, J. H. and L. D. Brown, Reanalysis of the COCORP deep seismic reflection profile across the San Andreas fault, Parkfield, CA, Bull. Seismol. Soc. Am., 76, 1668-1686 1986.

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

Nadeau, R., W. Foxall and T. V. McEvilly, Clustering and periodic recurrence of micro-earthquakes on the San Andreas fault at Parkfield, California, Science, 267, 503-507, 1995.

Shedlock, K. M., T. M. Brocher, and S. T. Harding, Shallow structure and deformation along the San Andreas fault in Cholame Valley, CA, based on high-resolution reflection profiling, J. Geophys. Res., 95, 5003-5020, 1990.

Thurber, C., S. Roecker, W. Lutter, and W. Ellsworth, Passive and active seismic array experiment along the San Andreas fault in the Northern Gabilans area of central California, EOS, Trans. Am. Geophys. Un. Suppl., 76, F397, 1995.

Thurber, C., S. Roecker, W. Lutter, and W. Ellsworth, Imaging the San Andreas fault with explosion and earthquake sources, Eos Trans. AGU, 77, 45 & 57-58, 1996.

Thurber, C., S. Roecker, W. Ellsworth, Y. Chen, W. Lutter, and R. Sessions, Two-dimensional seismic image of the San Andreas fault along the Northern Gabilan Range, central California:  evidence for fluids in the fault zone, Geophys. Res. Lett., 24, 1591-1594, 1997.

Trehu, A. M., and W. Wheeler, Possible evidence for subducted sedimentary materials beneath central California, Geology, 15, 254-258, 1987.

Unsworth, M., P. Malin, G. Egbert, and J. Booker, Geophysical evidence for fluids in the San Andreas fault at Parkfield, California, Geology, 25, 359-362, 1997a.

Unsworth, M. J., P. Bedrosian, G. D. Egbert and M. Eisel, Three-dimensional resistivity structure of the San Andreas Fault at Parkfield, California, Fall Meeting of the American Geophysical Union, San Francisco, 1997b.