I. INTRODUCTION: SCIENTIFIC RATIONALE AND EXPERIMENTAL PLAN FOR DRILLING INTO THE SAN ANDREAS FAULT ZONE

• The Problem of Low Strength Faults
• Implied Fault Zone Properties and Deformation Mechanisms
• The Mechanical Involvement of Fluids in Faulting
• The Physics of Earthquake Nucleation and Rupture Propagation
• The Need for Drilling

Figure 1. Schematic illustration of the proposed drilling and coring project. (click for more in-depth information)

The central scientific objective of this project is to study the physical and chemical processes that control deformation and earthquake generation within an active plate-bounding fault zone. The overall experiment is explained in detail in the Sections that follow; a schematic diagram of the proposed experiment is shown in Figure 1. The key operational elements of the project we propose are as follows:

1. Rotary drilling a hole to 4.0-km depth through the entire San Andreas fault zone in an area characterized by creep and microearthquakes. A site near Parkfield, CA, was chosen for drilling because of the occurrence of shallow seismicity and particularly good knowledge of fault structure at depth. During drilling we will utilize advanced logging-while-drilling techniques, collect spot cores and cuttings, and continuously sample fluids and gases in the drilling mud.
2. After conducting side-wall coring and open-hole geophysical logs (as permitted by hole conditions), the hole will be completely cased and cemented. A suite of fluid sampling, permeability and hydraulic fracturing stress measurements will be made through perforations in the casing. The perforations will be sealed after each test, except for a single interval which will be left open for fluid pressure monitoring.
3. An array of seismometers will be deployed in the hole to make near-field observations of earthquakes and to help determine the exact position(s) of the active trace(s) of the fault. Fluid pressure will be continuously monitored at a carefully chosen depth and the hole will be logged repeatedly to identify zones undergoing casing deformation and, hence, the location of active shear zones.
4. While the monitoring "string" of seismometers is in place, a number of surface-based and surface-to-borehole geophysical measurements will be made to characterize the physical properties of the fault zone and the surrounding crust.
5. After identifying the active fault trace(s), utilizing results from drilling and downhole measurements and fault-zone monitoring, 250-m-long continuous core holes will be drilled off of the main hole at four different locations where windows will be cut through the casing. In this manner, we plan to obtain a total of ~1000mof core material from multiple sites directly within and adjacent to the active fault zone.
6. Following coring, we will re-deploy an instrumentation array to permanently monitor earthquakes, fault slip, fluid pressure and ephemeral properties of the fault zone at depth.

Rock and fluid samples recovered from the fault zone and country rock will be extensively tested in the laboratory to determine their composition, origins, deformation mechanisms, frictional behavior and physical properties (permeability, seismic properties, etc.).

The project we propose will provide the kinds of data needed to constrain the many theories currently being debated about fault zone processes. It is not hard to imagine that by obtaining direct information on the composition and mechanical properties of fault zone rocks, the nature of the stresses responsible for earthquakes, the role of fluids in controlling faulting and earthquake recurrence, and the physics of rupture propagation this project could literally revolutionize our understanding of earthquake physics. Moreover, although it has been hypothesized that a wide range of deformation processes may precede seismic rupture, they have not been unequivocally detected by surface measurements. By making continuous observations directly within the San Andreas fault zone at seismogenic depths, we will be able to directly test and extend current theories about phenomena that might precede an impending earthquake.

Some of the most important questions about fault zone processes we wish to address are related to the growing body of evidence indicating that slip in crustal earthquakes along major plate-bounding faults (like the San Andreas) occurs at extremely low levels of shear stress. While this hypothesis (often referred to as the San Andreas Stress/Heat Flow Paradox, e.g., Lachenbruch and Sass, 1980, 1992; Zoback et. al., 1987; Hickman, 1991), has become widely accepted in recent years, earthquake researchers are now faced with the problem of explaining why major plate boundary faults are substantially weaker than the surrounding, highly-faulted crust. In fact, the question of how crustal faults lose their strength is critically important in crustal mechanics and earthquake hazard reduction for a number of reasons:

• As the weakness of plate boundaries (relative to plate interiors) is a fundamental aspect of plate tectonics, how and why plate boundary faults lose their strength is of fundamental global importance for understanding where plate boundaries form, how they evolve with time and how deformation is partitioned along them.
• By directly evaluating the roles of fluid pressure, intrinsic rock friction, chemical reactions and the physical state of active fault zones in controlling fault strength we will provide earthquake researchers the opportunity to simulate earthquakes in the laboratory and on the computer using representative fault zone properties and physical conditions. These studies will also allow for improved models of static stress transfer and earthquake triggering at a regional scale and between specific faults, as needed for intermediate-term seismic hazard forecasting following large earthquakes.
• Through long-term fault zone monitoring and in-situ observations of the earthquake source, we will be able to test and improve models for earthquake rupture dynamics, including such effects as transient changes in fluid pressure, fault-normal opening modes and variations in slip pulse duration. These observations can be used directly in attempts to generate improved predictions of near-field strong ground motion (amplitude, frequency content and temporal characteristics) and more reliable models for dynamic stress transfer and rupture propagation. These latter processes are believed to control earthquake size (i.e., whether or not a small earthquake will grow into a large one) and, hence, are crucial to long-term probabilistic assessments of earthquake hazard.
• The results of the proposed experiment are critical to development of more realistic models for the seismic cycle and assessment of the practicality of short-term earthquake prediction in two ways. First, in the fault zone monitoring phase of the proposed experiments, we will be able to determine if earthquakes are preceded by accelerating fault slip (e.g., a nucleation phase) and/or transient changes in fluid pressure. Second, we will be able to determine whether or not factors which might dramatically lower fault strength (high pore pressure and/or chemical fluid-rock interactions, for example) are closely related to the processes controlling earthquake nucleation. Our current knowledge of fault zone processes is so poor that not only are we unable to make reliable short-term earthquake predictions, but we cannot scientifically assess whether or not such predictions are even possible.

While the idea of drilling into the San Andreas fault has arisen many times over the past several decades, this project had its origin in December of 1992 when we convened a workshop on San Andreas fault zone drilling at the Asilomar Conference Center in Pacific Grove, California. The purpose of this workshop, which was attended by 113 scientists and engineers from seven countries, was to initiate a broad-based scientific discussion of the issues that could be addressed by drilling and direct experimentation in the San Andreas fault, to identify potential drilling sites and to identify technological developments required to make this drilling possible. As discussed at this workshop, the key questions to be addressed by deep drilling into the San Andreas fault zone include:

Fault Behavior

Is the static strength of the fault low and, if so, why? Why are some segments of the fault creeping and some locked? What factors control the localization of slip and strain? How is the fault zone stressed at different crustal levels? How does strain communication occur within the fault zone over different time scales? How is energy partitioned within the fault zone between seismic radiation, frictional dissipation, grain size reduction and chemical reactions? Can the frequency-magnitude relationship for earthquakes be extrapolated to smaller magnitudes?

Fluid Pressure

What is the vertical and lateral distribution of fluid pressure regimes? Do fluid pressure compartments exist? If so, what is the nature of the seals between these compartments? What is the time-dependence of fluid pressure within the fault zone? What is extent of vertical and lateral fluid migration during a seismic stress cycle?

Fault Fluids

What is the origin and composition of fault zone fluids? What are the permeabilities of fault-zone materials and country rock? What are the fluid transport mechanisms in and adjacent to the fault zone and what physical processes lead to fluid redistribution? What is the interplay between water-rock interaction and rheology at different structural levels?

Fault Zone Properties & Physical Parameters

How does the stress tensor vary in the vicinity of the fault zone? How do pre- and post-failure stress states compare? What, if any, form of cyclical dilatancy operates in the vicinity of the fault zone? How do physical properties relate to the fault zone fabric? What is the origin of low-velocity zones in the fault zone? How well and in what manner do physical properties and heterogeneity measured in boreholes correlate with geophysical observables?

Fault Structure & Materials

How does the width and character of the active slip zone vary with depth? What is the thermal structure of the fault zone? How do mineralogy and deformation mechanisms within the fault zone change with depth, temperature and country-rock geology? What determines the maximum depth of seismic activity? At what temperature do mineral reaction kinetics operate at the time scale of an earthquake cycle? How accurate are inferences drawn from deformation microstructures, piezometers, and fluid inclusions and how might one assess their survivability?

Fundamental questions about faulting and earthquakes such as these have gone unanswered due to the complete lack of hard data on the physical and chemical processes operating on the San Andreas and other faults at depth. As outlined in Hickman et al. (1994) and Z&H'96, the objectives to be addressed by drilling into the San Andreas fault may someday require holes as deep as 10 km. The principal reason deep drilling into the San Andreas fault may be warranted is to conduct extensive investigations in situ and on exhumed materials that are representative of the fault at the pressures, temperatures and conditions at which major earthquakes nucleate.

Although a 4.0-km-deep drill hole cannot address all of the questions listed above, the experiment proposed here will address a number of critical scientific questions about fault zone structure, composition and processes-these questions are summarized in Section II and discussed at length in the separate proposals submitted by members of our science team. Thus, the 4-km-deep drilling project described here is proposed both as a completely justifiable scientific experiment in its own right and as a possible critical first step toward drilling and experimentation to 10 km within the San Andreas fault zone. In the remainder of this Section we discuss the overall scientific rationale for drilling and experimentation in the San Andreas fault zone. This discussion includes first the key issue of the frictional strength of plate-bounding faults; then specific theories and questions concerning fault zone fluids, faulting and rheology; and finally a discussion of why fault zone drilling is needed to address these questions.

The Problem of Low Strength Faults

Recently, analysis of stress-induced borehole breakouts in petroleum wells along the Carrizo plain segment of the San Andreas fault in southern California indicate that the angle between the maximum horizontal compressive stress and the San Andreas increases from about 25-45° near the fault to 65-85° at distances greater than 20 km (Castillo and Hickman, 1995). Although the significance of this observation for fault strength is somewhat ambiguous owing to scatter in the data and the lack of information on the horizontal differential stress magnitudes adjacent to the fault (c.f., Zoback and Roller, 1979), these observations might suggest that the Carrizo plain section of the San Andreas fault is able to support higher levels of shear stress as compared to other weaker segments of the fault.

In summary, while essentially all available data indicates that the frictional strength of intraplate crust is quite high, the frictional strength of the San Andreas, and apparently many other plate-bounding faults as well, is quite low. Taken together, the heat-flow data and the directional constraint (i.e., S_Hmax at 65-85° to the San Andreas fault) suggest that the San Andreas fault is weak in both an absolute and relative sense. Despite the fundamental nature of this finding, we have no direct in-situ evidence indicating why this might be so, whether the mechanisms responsible for low strength along the San Andreas are likely to be found in other major fault systems or what role that these mechanisms might play in the processes of earthquake nucleation and propagation.

Implied Fault Zone Properties and Deformation Mechanisms

If the coefficient of friction, µ, is equal to 0.6-0.9 on the San Andreas fault, as predicted by Byerlee's Law, then the heat-flow constraint could be satisfied if the in-situ pore pressure, Pp, is greater than twice hydrostatic (Lachenbruch and Sass, 1980, 1992). However, if one assumes that principal stress magnitudes are constant across the fault zone and that µ >= 0.6, then high fluid pressures alone cannot explain the directional constraint as Pp would exceed the least principal stress once the angle between S_Hmax and the fault exceeds about 60° (e.g., Zoback et al., 1987; Scholz, 1989; Lachenbruch and McGarr, 1990). It has recently been suggested that large-scale yielding could lead to an increase in the

Figure 2. Two mechanisms that might account for low-strength fault zones imbedded in a stronger crust. (click for more information)

magnitudes of the principal stresses within the fault zone relative to their values immediately outside of the fault (Figure 2a, after Rice, 1992). If so, this would allow Pp within the fault zone to exceed significantly the external magnitude of the least principal stress (Byerlee, 1990; Rice, 1992). In this manner, permanently high pore pressures within an intrinsically strong (i.e., high coefficient of friction) San Andreas fault zone, in conjunction with much lower fluid pressures in the surrounding rock, could lower the fault strength sufficiently to satisfy both the heat-flow and directional constraints. A model that is, in many respects, similar to that of Rice (1992) was proposed earlier by Byrne and Fisher (1990) to explain the apparent weakness of the basal décollement beneath the Kodiak accretionary prism in southwest Alaska. Magee and Zoback (1993) applied the Rice model to explain the low frictional strength of the subduction zone associated with the M~8.2 Tokachi-Oki earthquake off northern Honshu, Japan. Although not requiring localized increases in stress magnitude, Fournier (1996) has presented a model in which near-lithostatic fluid pressures may be maintained within the San Andreas fault zone at depths greater than about 6-10 km if the maximum deviatoric stress at these depths is quite low (~10-30 MPa) and the tensile strength of the rock outside the fault zone remains high due to pervasive crack healing at elevated temperatures.

Alternatively, if one assumes that the fault is optimally oriented with respect to the principal stresses and that fluid pressures are hydrostatic, then the heat-flow constraint can be satisfied if µalong the fault is less than about 0.2 (Lachenbruch and Sass, 1992). Similarly, at least in central California where S_Hmax is at about 75-85° to the San Andreas fault, the heat-flow and directional constraints can be simultaneously satisfied under conditions of uniformly hydrostatic fluid pressures if µis extremely low-about 0.1 or less-along the fault and Byerlee's Law is applicable outside the fault zone (Figure 2b; Lachenbruch and McGarr, 1990; Lachenbruch and Sass, 1992). It is often proposed that the presence of clays or other weak minerals along the San Andreas and other faults might lead to anomalously low frictional resistance (e.g., Wu et al., 1975; Janecke and Evans, 1988; Wintsch et al., 1995). This inference has been supported by laboratory sliding experiments on synthetic clay-rich fault gouges (e.g., Wang et al., 1980; Shimamoto and Logan, 1981; Bird, 1984; Logan and Rauenzahn, 1987), on synthetic serpentinite gouges (Reinen et al., 1994; Reinen and Tullis, 1995) and on synthetic laumontite gouge (Hacker et al., 1995). However, these experiments are all at low-to-moderate confining pressures and temperatures. In contrast, experiments on natural clay-rich fault gouges collected from the San Andreas at depths of less than 0.4 km (Morrow et al., 1982), on synthetic clay-rich fault gouges (Morrow et al., 1992) and on synthetic serpentinite gouges (Moore et al., 1997) at high temperatures and/or confining pressures and hydrostatic fluid pressures indicate coefficients of friction at in-situ conditions that are too high to be reconciled with either the heat-flow or directional constraints. In addition, both natural and synthetic fault gouges deformed in the laboratory generally fail to exhibit the slip-weakening or velocity-weakening behavior required for the generation of earthquakes (e.g., Byerlee and Summers, 1976; Logan and Rauenzahn, 1987; Marone et al., 1990; Morrow et al., 1992; Reinen et al., 1994). Thus, the importance of these materials in the rheology of the San Andreas fault at seismogenic depths is unclear.

Figure 3. Coefficient of friction converted to depth along the San Andreas Fault. (click for more information)

Figure 3 illustrates some of the difficulties in explaining the weakness of the San Andreas fault as being due to fault zone materials with intrinsically low coefficients of friction. For example, based upon room-temperature sliding experiments, some investigators have proposed that serpentinite mineral chrysotile might be sufficiently weak to satisfy the heat flow constraint along the creeping section of the San Andreas fault (e.g., Reinen et al., 1994). However, as shown in Figure 3, recent experiments on chrysotile at high temperatures and confining pressure (Moore et al., 1997) suggest a significant increase in &micro-to values approaching those predicted by Byerlee's law-due to loss of bound water and the predicted transformation of chrysotile to antigorite at about 8 km depth. Similarly, under conditions of hydrostatic fluid pressure and at estimated in situ confining pressures, laboratory friction experiments suggest that pure montmorillonite should be relatively weak (average µ~0.2) at seismogenic depths along the San Andreas fault (Figure 3). However, if constraints on clay stability determined for the Gulf Coast region of the U.S. are applicable to the San Andreas, then montmorillonite should undergo a transition to illite starting at a depth of about 3 km (i.e., temperatures ~100°C). If so, the resulting increase in µwith depth would be such that bounds on fault strength imposed by both the heat flow and directional constraints would be exceeded at depths of only a few kilometers (Figure 3). It is important to note, however, that under certain conditions montmorillonite might be stable to 300°C or more (Wu, 1978; Wang, 1984). In this case, even a modest increase in fluid pressure at depth (i.e., slightly above hydrostatic) in a montmorillonite-dominated fault zone might be sufficient to satisfy both the heat flow and directional constraints. Without knowing what the composition of fault zone materials are at depth, it is impossible to ascertain whether or not models such as these have any validity.

Non-frictional processes may also reduce sliding resistance along faults. Solution-transport mechanisms such as pressure solution, fluid-assisted mineral reactions and crack healing may be important in determining the rheology of fault zones (e.g., Rutter and Mainprice, 1979; Sibson, 1983; Blanpied et al., 1991, 1995; Chester, 1995) and the time scales of interseismic strength recovery (e.g., Angevine et al., 1982; Chester and Logan, 1986; Fredrich and Evans, 1992; Karner et al., 1997). The establishment of impermeable barriers along faults like the San Andreas through the sealing of fractures with mineral deposits, followed by episodic fracturing and permeability enhancement during earthquakes, might be one mechanism whereby fluid pressures are intermittently high and intimately linked to the earthquake cycle (e.g., Sibson, 1992; Sleep and Blanpied, 1992; 1994; Byerlee, 1993; Sleep, 1995; Segall and Rice, 1995; Miller, 1996). Unfortunately, the roles of these mechanisms in determining the frictional strength, earthquake instability mechanisms and creep properties of the San Andreas fault are unknown because of uncertainties regarding the mineralogy, microstructures, and physical properties of fault-zone materials and the nature and distribution of fluids at seismogenic depths.

Another important class of models that might explain the low long-term strength of the San Andreas, at least along seismically active segments of the fault, have called upon processes directly associated with earthquake rupture propagation. These dynamic weakening mechanisms include shear heating during slip, leading to transient high fluid pressures (Lachenbruch, 1980) or melting (Sibson 1975; Spray, 1987); reductions in normal stress accompanying the propagation of dilational waves along the fault (Brune et al., 1993); and the fluidization of fault-zone materials due to the channeling of co-seismic acoustic energy (Melosh, 1979, 1996). While such processes may be operative during an earthquake, they do not relate to the problem of how earthquakes initiate and each requires that very specific fault zone conditions exist to be viable. Thus, to assess the likelihood that dynamic weakening mechanisms might operate along active faults we need to compare the structure and physical properties of actual fault zones with parameters required by these various models and to combine these observations with the results of long-term fault zone monitoring in the near-field of small-to-moderate size earthquakes.

The Mechanical Involvement of Fluids in Faulting

Sources of Fault-Zone Fluids. Potential sources of fluids in brittle faults and shear zones include metamorphic fluid generated by dehydration of minerals during prograde metamorphism (including shear heating), fluid trapped in pore space as sedimentary formation brines and meteoric water carried downward by circulation (e.g., Kerrich et al., 1984; Hacker, 1997; Ko et al., 1997). Fluid exsolved from magma is another potential source, at least in certain thermal regimes. The high fluid pressures that have been postulated within the San Andreas fault zone might be generated and maintained by continued upwelling of overpressured fluids within the fault zone and leakage of these fluids into the country rock (Rice, 1992). Alternatively, high fluid pressures might result from the sealing of locally derived high-pressure fluids within the fault zone once pressure gradients drop below a critical "threshold" required to overcome forces between molecular water and mineral surfaces in very small cracks and pores (Byerlee, 1990).

There is isotopic and geochemical evidence that mantle-derived water and carbon dioxide may be upwelling along some major crustal-penetrating faults, but definitive evidence remains elusive (see discussion by Rice, 1992). Irwin and Barnes (1980) noted the worldwide association of CO₂-rich springs with seismic belts, inferring a possible mantle source. Elevated ³He/⁴He ratios, attributed to a mantle gas component, have been correlated with areas of extensional tectonic activity in western Europe (Oxburgh and O'Nions, 1987) and with areas of earthquake swarm activity in Japan (Wakita et al., 1987). Giggenbach et al. (1993) have reported elevated ³He/⁴He ratios in seismically active areas of compressional tectonics in New Zealand, as well as in volcanically active regions in extensional tectonic regimes. Recently, Kennedy et al. (1997) argued that elevated ³He/⁴He ratios they observed in springs and wells located along a broad zone encompassing the San Andreas fault system indicate that significant quantities of mantle-derived fluids are entering the fault zone through the ductile lower crust at near lithostatic pressure. However, without direct sampling of fluids from within the San Andreas fault zone at depth it is unclear whether these fluids are ascending through a broad, fractured and faulted zone associated with the overall plate boundary or are narrowly focused within the (permeable) core of the San Andreas fault itself, and hence intimately involved in the physics of faulting as envisioned by Rice (1992) and others.

Large fluxes of deep-seated fluid are required for the deposi-tion of extensive vein deposits and hydrothermal alteration associated with many shear zones and fault systems (e.g., Cox et al., 1986; Boullier and Robert, 1992). This observation is consistent with some of the more recent models that have been proposed for the hydromechanical behavior of fault zones, such as the fault valve model (Sibson, 1981, 1992; Sibson et al., 1988) and Rice's (1992) steady state permeability model. The fault valve model requires a large volume of fluids at near-lithostatic pressure to accumulate beneath a low-permeability seal at the base of a fault zone during the interseis-mic period; this seal is then ruptured and the fluid surges upward immediately following the earthquake. In contrast, Rice's model requires the continual upwelling of overpressured fluid from the ductile root of a fault zone and does not consider possible variations in fluid pressure during the seismic cycle. In other faulting environments, however, there is evidence that mass transfer is a more localized process, with deposition of mineral veins and hydrothermal alteration being a consequence of small-scale fluid flow and diffusive-mass-transfer processes (Gratier et al., 1994; Evans and Chester, 1995).

Fault Zone Permeability. The permeability structure of shear zones and brittle faults has recently been the focus of field studies that both confirm and extend observations made years ago by mining geologists. Large faults are not discrete surfaces but rather are a braided array of slip surfaces encased in a highly fractured and often hydrother-mally altered transition or "damage" zone (Smith et al., 1990; Bruhn et al., 1990, 1994; Chester et al., 1993). Structural and mineralogical textures indicate that episodic fracturing and brecciation are followed by cementation and crack healing, leading to cycles of permeability enhancement and reduction along faults.

A number of recent experimental studies carried out at hydrothermal conditions allow one to estimate the time required for processes such as crack healing and sealing and hydrothermal alteration to significantly alter fault zone permeability. In most cases, these processes operate at rates that are rapid with respect to the 100- to 10,000-year recurrence intervals for large earth-quakes (e.g., Brantley et al, 1990; Blanpied et al., 1992; Moore et al., 1994). In laboratory shearing experiments on granite gouge sandwiched between granite forcing blocks, Blanpied et al. (1992) showed that redistribution of material in solution can quickly reduce the granite permeabil-ity, causing a self-generated impermeable seal which isolates the deforming fault from the nearby country rock. Compaction of the fault gouge before and during shear then causes fluid pressure in the fault zone to rise, allowing slip at low shear stress. Subsequent theoretical modeling (Sleep and Blanpied, 1992, 1994; Sleep, 1995) showed that the generation of dilatant pores and microcracks during earthquakes in a hydraulically isolated fault zone, followed by creep compaction between earthquakes, might lead to cycli-cally high fluid pressures along faults. Recently, Miller (1996) used the Sleep and Blanpied (1992) model to suggest that the delay in the expected M=6 earthquake at Parkfield may be due to a retardation in the rate of compaction-induced pore pressure increase within the San Andreas fault in response to unloading by the 1982-85 Coalinga, New Idria and Kettleman Hills earthquake sequence.

A possibly important recent development from studies of fluid pressure in sedimentary basins has been the revelation from borehole measurements of abrupt transitions, both vertically and laterally, between distinct fluid pressure regimes in some sedimentary basins. These "fluid pressure compartments" are bounded by seals which in some cases are stratigraphic (e.g., shale horizons) but in others are gouge-rich faults or thin zones of hydrothermal cementation which cut across stratigraphy (Hunt, 1990; Powley, 1990; Dewers and Orteleva, 1994; Martinsen, 1997). By analogy with these observations, Byerlee (1993) proposed a model in which contiguous vertical and horizontal seals within a fault zone would lead to discrete fluid pressure compartments (i.e., tabular lenses), the rupture of which might be important in earthquake nucleation and propagation (see Lockner and Byerlee, 1995). Although direct evidence for these fault zone fluid compartments in active fault zones is lacking, negative polarity reflections (bright spots) on seismic reflection images acquired over some accretionary prisms have been interpreted to indicate the existence of high-pressure fluid compartments along the basal decollements (Moore and Vrolijk, 1992; Shipley et al., 1994; Moore et al., 1995b). Recent laboratory experiments by Zhang and Tullis (1998) indicate that permeability reduction due to shearing of gouge might also lead to the development of fluid pressure seals in fault zones.

Transient Fluid Pressure Effects. For the most part, the Hubbert and Rubey (1959) analysis and those that followed it in the structural geology literature took no account of the mode of fault slip or of the variations in permeability and fluid pressure that might arise from faulting. This quasi-static, high-fluid-pressure approach to faulting contrasts with the dilatancy/fluid diffusion hypothesis for shallow crustal earthquakes evolved by the seismology/rock mechanics community (e.g., Nur, 1972; Scholz et al., 1973), where massive fluid redistribution at close to ambient hydrostatic fluid pressures was inferred to occur in response to the earthquake cycle of shear stress accumulation and release. While belief in extensive microcrack dilatancy formed under high differential stress levels as an earthquake precursor has waned, it is almost inevitable that some form of stress-dependent dilatancy is associated with active faulting (e.g., Parry et al., 1991; Coombs, 1993), though significant dilatant strains may be restricted to the immediate vicinity of fault zones (Sibson, 1994).

A range of physical effects arising from the mechanical response of fluid-saturated crust has been invoked to account for time-dependent phenomena associated with faulting such as slow earthquakes, creep events, afterslip and aftershock activity and its decay (e.g., Nur and Booker, 1972; Rice and Cleary, 1976). Transient changes in fluid pressure and effective stress have also been suggested to play a direct role in rupture propagation and arrest. Shear resistance on the rupture surface may be dramatically lowered by localized increases in fluid pressure from frictional heating or locally elevated as a consequence of pore fluid diffusion and dilatant hardening at fault jogs and other irregularities (Sibson, 1973, 1985; Lachenbruch, 1980; Mase and Smith, 1987; Rudnicki, 1988; Sleep, 1995; Segall and Rice, 1995).

Recently, two studies presented geophysical evidence related to the possible breaching of fluid pressure compartments along the San Andreas fault system during earthquakes. Johnson and McEvilly (1995) presented an analysis of the clustering and migration of microearthquake activity along the transition from creeping to locked segments of the San Andreas fault at Parkfield, California. The activity occurs within and near the edges of a tabular zone of low velocity, anisotropic material with a high V_P/V_S ratio inferred to represent a dilatant, and possibly overpressured, fault zone. The expansion of these earthquake clusters with time is consistent with the migration of overpressured fluids from breached compartments. Fenoglio et al. (1995) explored the electromagnetic consequences of rupturing pressure seals and subsequent fluid flow along the fault zone. Their theoretical analysis shows that the electromagnetic fields generated could explain precursory anomalous ultra-low-frequency electromagnetic emissions that were observed in the epicentral region of the 1989 M=7.1 Loma Prieta earthquake in California (Fraser-Smith, 1990). They concluded that electrokinetic effects accompanying the rupturing of overpressured fluid compartments with impermeable seals provide the most plausible mechanism for these emissions. Our understanding of the importance of these various processes in the Earth has, however, been hampered by our lack of detailed knowledge of the appropriate hydraulic parameters (especially the permeability structure) in and around active fault zones.

Chemical Effects of Fluids on Fault Zone Rheology. Over the past several years a number of fault mechanics models have either been developed or refined that incorporate solution transport deformation mechanisms that may weaken and/or destabilize the fault zone. However, complicating this issue enormously is the fact that under only slightly varied environmental and mineralogical conditions similar processes can act to cement the fault zone together, thereby increasing fault strength (see Hickman and Evans, 1992). The experimental and theoretical studies on which these models are based are now focusing on processes that have long been inferred as being important from field observations of natural fault and shear zones, such as pressure solution, fluid-assisted retrograde mineral reactions, crack healing and cementation (e.g., Kerrich et al., 1984; Power and Tullis, 1989; Bruhn et al., 1990; Boullier and Robert, 1992; Chester et al., 1993). These deformation mechanisms are all interrelated, in that they depend upon ther-mally activated chemical reactions between the rock and pore fluid as well as the rates at which dissolved species are trans-ported through the pore fluid.

Laboratory and theoretical investigations have shown that pressure solution may be important in reducing long-term fault strength and in promoting aseismic slip (i.e., creep) along faults (e.g., Rutter and Mainprice, 1979; Tada et al., 1987; Chester and Higgs, 1992, Chester, 1995; Blanpied et al., 1995). This is especially likely in the middle to lower crust where high confining pressures and low-to-moderate temperatures inhibit both fric-tional sliding and crystal-plastic deformation, respectively (e.g., Kirby, 1980; Sibson, 1983). In contrast, in addition to allowing the formation of pressure seals (described above), solution transport processes such as crack healing and sealing and cementation may cause the welding together of asperities or fault gouge, leading to time-dependent fault strengthening between earthquakes (e.g., Angevine et al., 1982; Hickman and Evans, 1992; Fredrich and Evans, 1992; Karner et al., 1997; Hacker, 1997). Laboratory friction experiments conducted under hydrothermal conditions suggest that a change in dominant deformation mechanism with increasing depth from brittle deformation to solution transport creep might control the depth at which the seismic-to-aseismic transition occurs in the crust (Blanpied et al., 1991). Ultra-fine-grain fault gouge and cataclasites should be particularly reactive in the presence of aqueous pore fluids, allowing solution transport fault creep to proceed under relatively low resolved shear stresses (e.g., Chester and Higgs, 1992). Similarly, laboratory experiments using halite single crystals indicate that pressure solution creep rates should increase markedly within fault zones containing a diverse mineralogy, particularly in the presence of intergranular montmorillonite and, perhaps, other clays (Hickman and Evans, 1991, 1995).

Hydrothermal mineral reactions can also weaken crustal rocks when the reaction products are weaker than the reactants (see Wintsch et al., 1995). Based upon observations of exhumed shear zones in granite, Janecke and Evans (1988) argued that muscovite formed from the breakdown of feldspar might dramatically lower the ductile shear strength of the granite (presumably due to basal plane dislocation glide in the micas), even at temperatures well below those necessary for the plastic flow of quartz. At least at shallow depths, fault zones such as the San Andreas are mostly composed of clay- and mica-rich gouge resulting from the hy-drolysis of feldspar (e.g., Wu, 1978), suggesting an enhancement of the feldspar breakdown reaction within the fault zone. Although recent experiments have shown that dislocation glide within biotite and muscovite single crystals might be capable of lowering the average strength of crustal-penetrating faults to a few tens of MPa (Kronenberg et al., 1990; Mares and Kronenberg, 1993), the strength of micaceous rocks has been shown to be highly dependent upon mica orientation and contiguity and approaches values predicted by Byerlee's law at low mica contents (Shea and Kronenberg, 1993). Stress-enhanced hydrothermal mineral reactions are also recognized to be important in weakening crustal rocks, even when both the reactant and product phases are strong (e.g., Rubie, 1983). For example, reactions in the olivine-talc-serpentine-water system have been demonstrated to dramatically lower the shear strength of ultramafic rocks in laboratory friction experiments (Pinkston et al., 1987).

The Physics of Earthquake Nucleation and Rupture Propagation

The systematics of repeating earthquakes at Parkfield raises some critical new questions about the basic assumptions that underlie our model of faults as cracks (Nadeau and Johnson, 1998). For repeating earthquakes on faults with known slip rates it is possible to compute a lower bound on the static stress drop from the seismic moment, recurrence interval and slip rate. At Parkfield, this bound implies stress drops of 100-1000 MPa for M=1-2 earthquakes, assuming that all of the long-term fault slip occurs seismically over the patches ruptured during these earthquakes. Nadeau and Johnson propose a model in which the fault has a highly heterogeneous distribution of strength, with strong contact areas approximately 1 m in diameter. Testing this hypothesis will require very near-source measurements that can only be made in boreholes.

Figure 4. Seismograms of a M=0.4 earthquake recorded on a 10 Hz borehole seismometer at 2 km depth. (click for more information)

Recent observations of microearthquakes in moderately deep boreholes (2-2.5 km) at Cajon Pass, California (Abercrombie and Leary, 1993; Abercrombie, 1995) and in Long Valley, California (Ellsworth, Kasameyer, Ito and others, in preparation) demonstrate that ultra high-fidelity recordings can be made with available technology under temperature and pressure conditions similar to those we anticipate in the proposed hole. The seismograms in Figure 4, from a deep borehole in the center of the resurgent dome of Long Valley caldera, illustrates several of the key advantages of recording in deep boreholes. The displacement seismograms shown at the top of this figure are for an M=0.4 event located 1.5 km from the 2-km-deep borehole seismometer, recorded at 10,000 samples/second. The bottom panel compares the borehole and surface velocity seismograms of the same earthquake (note change in time scale). Note the remarkable simplicity and broad-band content of both the P and S waves recorded at 2 km depth, relative to the surface recordings. This example illustrates the clear advantages of very high frequency borehole recording.

An array of downhole seismometers, accelerometers and other sensors will be deployed across the fault zone after drilling (Figure 5). Having an array of seismometers in the borehole has several advantages. Extremely accurate determinations of the radiated energy, seismic moment and earthquake locations can be made, the detailed velocity structure of the fault zone can be investigated, and earthquake nucleation and propagation can be studied with unprecedented detail. The technological risks to the monitoring instrumentation will be minimized by relying on proven technologies, or modest extensions to them, and utilizing instruments deployed in a cased and cemented borehole. The wide dynamic range of the accelerometers will also let us study

Figure 5. Schematic illustration of proposed fault zone monitoring string. (click for more information)

earthquakes up to at least the M=6 characteristic Parkfield earthquake. High frequency signals are expected from the very local events, with good signal-to-noise above 1 kHz. Thus, good sensitivity to the higher frequencies becomes the most important seismic monitoring design criteria. Peak acceleration estimates for a magnitude 1.0 earthquake located at a distance of 1000 m and 100 m are 1 g and 10 g's, respectively. Because the seismometer string will extend from the core of the fault zone to its edge and beyond, it will also be useful for studying wave propagation within the fault zone.

The deepest instrumentation packages will permit us to relate the seismic cycle of the nearest microearthquakes to the strain within the fault itself. Far-field observations of repeating earthquakes and its correlation to strain for microearthquakes at Parkfield and elsewhere in the San Andreas fault system (Vidale et al., 1994; Ellsworth, 1995; Nadeau et al., 1995) show that these cycles conform with H.F. Reid's (1910) elastic rebound hypothesis. The extrapolation of surface slip rates to strain rates on individual earthquake sites is uncertain, and the Parkfield experiment will permit us to critically determine this relationship. Monitoring pore pressure will similarly test hypotheses related to the rate of fluid pressure variations in episodic fault slip and, perhaps, microearthquakes.

The Need for Drilling

Most of what we now know about the structure, composition and deformation mechanisms of crustal faults has been learned from geological investigations of exhumed faults, particularly in normal and reverse faulting environments were erosion has exposed previously deeply buried foot- and hanging-wall rocks. These field observations have proven particularly useful for several reasons. First, field observations of exhumed faults allow broad coverage with respect to variations in faulting style (e.g., comparing strike slip, normal and reverse faults), fault movement history and local geology. Secondly, where sufficient surface outcrops exist, field observations can readily address issues related to geometrical complexity and spatial heterogeneity in physical properties and fluid composition (e.g., Kerrich et al., 1984; Parry, 1994; Evans and Chester, 1995).

However, as valuable as these investigations have been, they suffer from several severe limitations when one attempts to draw inferences about active processes operating during faulting at depth. Foremost among these limitations is the fact that constraints on the mechanical state and physical properties of active fault zones (e.g., fluid pressure, stress and permeability) from surface observations are, of necessity, indirect and subject to alternate interpretations. For example, as noted by numerous participants in the recent USGS Red Book Conference on the Mechanical Involvement of Fluids in Faulting (see Hickman et al., 1995a), stress heterogeneities induced by fault slip can lead to considerable uncertainties in inferring past fluid pressures from observations of vein geometry in outcrop. In all of these investigations, a complex history of uplift and denudation may have severely altered, or even destroyed, evidence for deformation mechanisms, fault zone mineralogy and fluid composition operative during fault slip. This problem is especially acute for solution-transport-deformation mechanisms (e.g., pressure solution and crack healing/sealing) and other low-activation-energy processes, as the deformation microstructures formed at great depth are easily overprinted by ongoing deformation as the fault rocks are brought to the surface. Finally, with the rare exception of localized melts generated by rapid seismic slip (i.e., the pseudotachylytes occasionally found in exhumed fault zones; e.g., Sibson, 1975; Magloughlin and Spray, 1992), there are currently no reliable microstructural indicators that can be used to differentiate between seismic slip and creep. Thus, the importance of fluids in earthquake generation and rupture is impossible to assess with any degree of certainty based solely on studies of exhumed fault rocks.

Drilling and downhole measurements in active fault zones would provide critical tests of interpretations and hypotheses arising from laboratory rock mechanics experiments and geological observations on exhumed faults. Drilling provides the only direct means of measuring pore pressure, stress, permeability and other important parameters within and near an active fault zone at depth. It is also the only way to collect fluid and rock samples from the fault zone and wall rocks at seismogenic depths and to monitor time-dependent changes in fluid pressure, fluid chemistry, deformation, tempera-ture and electromagnetic properties at depth during the earthquake cycle. In the context of the conceptual models presented above, in-situ observations and sampling through drilling would perform two critical, and unique, functions. First, sampling of fault rocks and fluids and downhole measurements would provide essential constraints on mineralogy, grain size, fluid chemistry, temperature, stress, pore geometry and other parameters that would allow laboratory investigations of fault zone rheology and frictional behavior to be conducted under realistic in-situ conditions. Second, by in-situ sampling, downhole measurement and long-term monitoring in active fault zones we would be able to test and refine the broad range of current theoretical models for faulting and seismogenesis by providing realistic constraints on fault zone physical properties, loading conditions and mechanical behavior at depth. In particular, by comparing results of microstructural observations and rheological investigations on core with measurements of microseismicity, fluid pressure and deformation during the fault zone monitoring phase of this experiment, we would be able to differentiate between fault zone processes (e.g., fluid pressure fluctuations) associated with fault creep versus earthquakes.