Potential role of mantle-derived fluids in weakening the San Andreas Fault

Transcription

1 Click Here for Full Article JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 114,, doi: /2008jb006087, 2009 Potential role of mantle-derived fluids in weakening the San Andreas Fault Patrick M. Fulton 1,2 and Demian M. Saffer 1 Received 5 September 2008; revised 2 March 2009; accepted 15 April 2009; published 31 July [1] On the basis of both geomechanical and thermal data, the San Andreas Fault (SAF) has been interpreted to act as a weak plane within much stronger crust, allowing it to slip at very low shear stresses. One explanation for this weakness is that large fluid overpressures exist locally within the fault zone. However, mechanisms for generating, maintaining, and localizing pressures within the fault are poorly quantified. Here we evaluate whether realistic sources of mantle-derived fluids, proposed on the basis of high mantle helium signatures near the SAF, can generate localized fluid pressures within the fault zone in a manner consistent with a wide range of observations along the fault and in the San Andreas Fault Observatory at Depth borehole. We first calculate a reasonable estimate of the magnitude and location of a mantle-derived flux of water into the crust. This fluid flux results from dehydration of a serpentinized mantle wedge following the northward migration of the Mendocino Triple Junction and the transition from subduction to strike-slip tectonics. We then evaluate the potential effect of this water on fluid pressures within the crust using 2-D cross-sectional models of coupled fluid flow and heat transport. We show that in models with realistic permeability anisotropy, controlled by NE dipping faults and fractures within the country rock, large localized fluid pressures can be focused within a SAF acting as a hydrologic barrier. Our results illustrate a simple and potentially plausible means of weakening the SAF in a manner generally consistent with available hydrologic, thermal, and mechanical constraints. Citation: Fulton, P. M., and D. M. Saffer (2009), Potential role of mantle-derived fluids in weakening the San Andreas Fault, J. Geophys. Res., 114,, doi: /2008jb Department of Geosciences, Pennsylvania State University, University Park, Pennsylvania, USA. 2 Now at College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis, Oregon, USA. Copyright 2009 by the American Geophysical Union /09/2008JB006087$ Introduction [2] The San Andreas Fault (SAF), along with many other plate boundary faults, has been interpreted to slip at shear stresses considerably lower than expected for typical rock and fault gouge friction coefficients, and assuming hydrostatic pore fluid pressures [e.g., Lachenbruch and Sass, 1980; Zoback et al., 1987; Hickman, 1991]. The lack of a thermal anomaly from frictional heating near the SAF [e.g., Lachenbruch and Sass, 1980; Sass et al., 1997; Williams et al., 2004; Fulton et al., 2004b], and maximum horizontal principal stress (s H ) directions inferred to be at a very high angle (up to 85 ) with the SAF [e.g., Townend and Zoback, 2004; Hickman and Zoback, 2004] imply that the fault is weak in an absolute sense, and also considerably weaker than the surrounding crust. One hypothesis for both the apparent absolute and relative fault weakness is that fluid pressures within the fault zone are markedly higher than in the surrounding crust [Rice, 1992; Byerlee, 1990, 1992]. The mechanisms for generating and maintaining such pressures are poorly understood, although there has been considerable speculation regarding the potential role of mantle-derived fluids, primarily on the basis of geochemical data [e.g., Rice, 1992; Kennedy et al., 1997; Kharaka et al., 1999; Faulkner and Rutter, 2001; Wiersberg and Erzinger, 2007]. Here, we use numerical models of fluid flow and heat transport to investigate the potential role of mantlederived fluids in generating and sustaining large fluid pressures that are localized within the fault zone, and testing whether such a mechanism is consistent with a range of data from the region and the San Andreas Fault Observatory at Depth (SAFOD) Mantle Fluids and a Weak San Andreas Fault [3] SAFOD is a borehole observatory consisting of a 3.2 km deep main borehole and a 2.2 km deep pilot hole located near the town of Parkfield in central California, a focus region for this study (Figure 1). The main borehole is inclined and crosses the fault from the SW at 2.7 km depth (Figure 2). Numerous observations at SAFOD and in the surrounding region provide important constraints on conceptual models of fault strength and crustal-scale hydrogeology (Figure 2, described in detail below). Notably, isotopic evidence from mud gases including high 3 He/ 4 He ratios and values of d 13 C from HCO 3 suggest the presence of mantle-derived fluids in wells and springs primarily to the NE of the SAF, as well as within the SAFOD borehole 1of15

2 Figure 1. Map of California and Nevada. Quaternary faults are shown in gray, with the San Andreas Fault in black (U.S. Geological Survey and California Geological Survey and Nevada Bureau of Mines and Geology, Quaternary fault and fold database for the United States, 2006, accessed 26 June 2008, available at usgs.gov/regional/qfaults/). The location of Parkfield, California, the San Andreas Fault Observatory at Depth (SAFOD), and the Mendocino Triple Junction (MTJ) are labeled. as it crosses into the North American plate [Kennedy et al., 1997; Kharaka et al., 1999; Wiersberg and Erzinger, 2007]. [4] Previous authors have suggested that a permeable fault or damage zone surrounded by low-permeability country rock may act as a conduit for mantle-derived fluids and allow for large fluid pressures to be localized within the fault zone [Rice, 1992; Kennedy et al., 1997]. Rice [1992] showed that with strongly anisotropic fault zone permeability, permeability in the surrounding crust considerably lower than in the fault, and a near-lithostatic pressure boundary directly beneath the fault at the base of the brittle crust, high fluid pressures develop locally within the fault during transient numerical simulations. Observations in and around SAFOD, however, suggest that this may not be an adequate representation of the SAF, at least within the Parkfield region. SAFOD observations, including contrasts in both fluid chemistry and fluid pressure across the fault zone, suggest that rather than acting as a permeable conduit, the fault appears to be considerably less permeable than the surrounding crust and acts as a barrier for fluid flow [e.g., Wiersberg and Erzinger, 2007] (Figure 2). [5] In addition, several observations suggest that the source of mantle fluids is greatest NE of the fault zone rather than directly beneath it. In particular, helium isotope ratios from within the SAFOD main hole SW of the fault are typical of meteoric and crustal-derived fluids, whereas higher values indicating a strong mantle component are observed as the SAFOD borehole crosses the SAF into units within the North American plate [Wiersberg and Erzinger, 2007]. This mantle signature increases as the borehole continues to the NE, and additional well data 1.4 km NE of the fault zone near Parkfield reveal even greater values indicating that as much as 25% of the helium is of mantle origin [Kennedy et al., 1997; Wiersberg and Erzinger, 2007]. [6] Within the Franciscan Complex to the NE of the SAF (Figure 3), elevated pore pressures have been observed in deep boreholes, and a variety of geophysical surveys through the Parkfield region (seismic, gravity and magnetic, and magnetotelluric) have inferred a potentially overpressured zone capped by a serpentinite unit extending 12 km from the fault to the NE at depths >3 km (Figure 3) [Eberhart-Phillips and Michael, 1993; McPhee et al., 2004; Unsworth and Bedrosian, 2004]. Although no indirect indications of high fluid pressures within the fault zone were observed during SAFOD drilling [Zoback and Hickman, 2005], direct hydrologic measurements within SAFOD have not yet been conducted (M. Zoback and S. Hickman, personal communication, 2009). It is also important to note that the borehole crosses the fault at relatively shallow depth, such that inferences regarding fluid pressure at SAFOD may not fully represent processes controlling fault strength throughout the upper km. [7] The most promising source of mantle-derived fluids comes from mantle wedge material, which was potentially Figure 2. Schematic summary illustrating the SAFOD boreholes and key hydrogeologic and geomechanical observations described in the text. Large-scale geologic units are labeled for Salinian granite (Ksg) SW of the SAF and Great Valley Sequence (Kgv) to the NE. 2of15

3 Figure 3. Idealized cross section showing geologic structure of the San Andreas Fault and surrounding crust in central California [after McLaughlin et al., 2000; Bailey et al., 1964; Irwin and Barnes, 1975; Eberhart-Phillips and Michael, 1993; McPhee et al., 2004; Bleibinhaus et al., 2007]. NE dipping mélange units within the Franciscan Complex (dark gray) and faults and fractures (annotated with arrows indicating sense of motion) associated with both the accretion of the Franciscan Complex and present-day thrust faulting likely act as permeable pathways that may focus fluid pressure toward the SAF. hydrated and serpentinized during subduction prior to the formation of the SAF, and which would subsequently dehydrate after the plate boundary along California transitioned from subduction to strike-slip tectonics (Figure 4) [Kirby et al., 2002]. The SAF was created 25 Ma when a portion of the East Pacific Rise spreading center subducted beneath the North American plate creating two triple junctions, the Mendocino Triple Junction (MTJ) to the north and the Rivera Triple Junction to the south [Atwater and Stock, 1998]. The MTJ and the subducting slab have migrated northward over time, and in its wake, subduction analogous to that occurring in present-day Cascadia has been replaced by the strike-slip tectonics of the SAF system [Atwater and Stock, 1998]. As the subducting slab is replaced with upwelled asthenosphere (or remaining slab stalls beneath the overlying crust and thermally reequilibrates), serpentinized upper mantle, believed to remain beneath the North American crust, is heated (Figure 4b) [ten Brink et al., 1999; Guzofski and Furlong, 2002; Kirby et al., 2002; Brocher et al., 2003; Erkan and Blackwell, 2008]. Serpentinite minerals are unique, and relevant to this study, in that they contain up to wt% water in their mineral structure [Iwamori, 1998]. Kirby et al. [2002] combined a detailed thermal model with a kinetic reaction model of serpentinite formation and dehydration to illustrate that the temperature increase following the migration of the Mendocino Triple Junction could drive continued prograde dehydration of serpentinized mantle for tens of millions of years resulting in a large and sustained flux of water into the overlying crust. [8] The presence of a substantial volume of highly serpentinized mantle has been recognized through geophysical imaging beneath Cascadia [Bostock et al., 2002; Brocher et al., 2003; Blakely et al., 2005], as well as in numerous other subduction zones [e.g., Graeber and Asch, 1999; Kamiya and Kobayashi, 2000; Kamimura et al., 2002; DeShon and Schwartz, 2004; Tibi et al., 2008; see also Wada et al., 2008]. In Cascadia, on the basis of low shear wave velocities, high Poisson s ratio, and an inverted polarity Moho along a trench-perpendicular transect, a zone averaging 50 km wide and 20 km thick within the corner of the mantle wedge is interpreted to be at least 50 60% serpentinized (Figure 4a) [Bostock et al., 2002]. [9] Detailed models of the thermal response of the mantle wedge to MTJ migration and subsequent serpentinite dehydration [Kirby et al., 2002; Guzofski and Furlong, 2002] are also generally consistent with observations from a line of volcanic centers associated with the passage of the MTJ [e.g., Johnson and O Neil, 1984; Whitlock et al., 2001, Whitlock, 2002; Furlong and Schwartz, 2004; Schmitt et al., 2006]. The petrology of these volcanics has been interpreted Figure 4. Schematic cross section illustrating the transition from Cascadia-like subduction to strike-slip tectonics following the northward migration of the Mendocino Triple Junction. (a) Cartoon illustrating subduction prior to the creation of the SAF (adapted with permission from Macmillan Publishers Ltd. after Kirby [2000], copyright 2000), showing the general location of serpentinized mantle wedge material predicted by thermal models and observed by geophysical methods [e.g., Bostock et al., 2002; Brocher et al., 2003; Wada et al., 2008]. (b) Idealized cross section following MTJ passage; as the downgoing slab has migrated northward with the MTJ, the space the slab once occupied is replaced by upwelling aesthenosphere which heats the overlying crust and mantle [e.g., Guzofski and Furlong, 2002]. This temperature rise dehydrates the serpentinized mantle [Kirby et al., 2002]. 3of15

4 to reflect ponding of small amounts of mantle-derived melt within the crust, which in turn drive crustal melting [Johnson and O Neil, 1984]. The geochemistry of erupted basaltic lavas suggests that the mantle-derived melt is sourced from a hydrated mantle wedge [Johnson and O Neil, 1984; Whitlock et al., 2001, Whitlock, 2002; Furlong and Schwartz, 2004; Schmitt et al., 2006]. Of relevance to this study, these petrologic interpretations suggest that (1) hydrous mantle wedge material is present beneath the N. American plate following MTJ passage, and (2) that this mantle wedge material is significantly heated. In comparison to the 20 km thick by 50 km wide zone of highly serpentinized mantle material inferred in cross section beneath Cascadia [Bostock et al., 2002], estimates based on numerical modeling and petrologic constraints for the amount of mantle material that melts suggest a crosssectional area 4 5 km thick and km wide [Liu and Furlong, 1992; Schmitt et al., 2006]. This implies that less than 10% of the inferred zone of serpentinized mantle melts following MTJ passage; it is thus plausible that much of the hydrated mantle wedge material would remain in place following MTJ passage and gradually dehydrate [Kirby et al., 2002]. [10] Here, we explore the potential role of mantle-derived fluids in weakening the SAF by evaluating whether realistic fluid sources (in terms of both magnitude and location) from the dehydration of serpentinized mantle, together with realistic permeability architecture for the fault and surrounding crust, can generate large localized fluid pressures in a manner consistent with observations. This study differs from previous work examining the effect of mantle fluids on fluid pressures along the SAF system, in that (1) we use 2-D hydrogeologic models to evaluate a realistic fluid source rather than implementing a prescribed basal pressure boundary immediately below the fault zone; (2) we treat the SAF as a low-permeability barrier rather than as a conduit; and (3) we evaluate the possible role of NE dipping faults and fractures within the surrounding country rock, which may act to focus pressures toward the fault zone, as a means of localizing fluid pressures (Figure 3). Overall, this study addresses whether a prolonged, yet realistic mantle-derived fluid flux can generate fluid pressures needed to explain a weak SAF when considering realistic 2-D permeability structure; and whether a fault acting as a barrier and the expected location of mantle fluid sources NE of the fault precludes the role of mantle fluids in localizing pore fluid pressures within or along a near-vertical SAF Constraints on Fluid Pressures Necessary to Explain Apparent Fault Weakness [11] High angles between the SAF and the maximum horizontal principal stress (y ) have been inferred from earthquake focal mechanisms and boreholebased observations throughout California [e.g., Zoback et al., 1987; Mount and Suppe, 1987; Provost and Houston, 2003; Townend and Zoback, 2004]. The largest values (y = 85 ) occur in the San Francisco Peninsula region. In the Parkfield area and throughout much of central and southern California, values of y = 68 are observed [Townend and Zoback, 2004]. Within the SAFOD boreholes, similarly high angles between the maximum horizontal principal stress and the fault plane have been determined from borehole breakouts and drilling-induced tensile fractures (y 69 in the SAFOD pilot hole) [Hickman and Zoback, 2004], and from shear velocity anisotropy (y 70 at a distance of 200 m from the fault) [Boness and Zoback, 2006]. These high angles imply that the fault acts as a weak plane within a stronger country rock [Zoback et al., 1987; Mount and Suppe, 1987; Townend and Zoback, 2004; Hickman and Zoback, 2004]. The relative weakness can be explained if there is material of low friction coefficient localized within the fault zone, the fluid pressure ratio locally within the fault (l f ) is very large, or a combination of both (see Discussion). Fluid pressure ratio l is defined as l ¼ P=r r gz where P is the pore fluid pressure (Pa), r r is bulk rock density (kg m 3 ), g is gravitational acceleration (m s 2 ), and z is depth (m) (all variables in our analysis, and their definitions and units, are summarized in Table 1). Together, r r gz defines the lithostatic (i.e., overburden) stress (Pa); for r r = 2600 kg m 3, hydrostatic fluid pressures correspond to a value of l = [12] The fluid pressure ratio within the fault (l f ) necessary for failure along the SAF is a function of the shear stress (t y ), the total normal stress on the fault plane (s y ), and the fault zone friction coefficient (m f ) (Figure 5): l f ¼ P f r r gz ¼ ð1þ! s y t y m f =r r gz ð2þ where P f is the fault zone pore fluid pressure. [13] Observations within the SAFOD pilot hole and microearthquake focal mechanisms suggest that for the crust surrounding the SAF, the overburden stress s v equals s 3,the least principal stress [Eberhart-Phillips and Michael, 1993; Hickman and Zoback, 2004]. Assuming the surrounding crust is near critically stressed for failure [Townend and Zoback, 2000], the shear stress (t y ) and total normal stress (s y ) on a near-vertical strike-slip fault with s v = s 3 are defined by t y ¼ 1 l ð cþ r 2 r gz M 3 M 3 X þ X 1 sin 2y M 1 M 1 s y ¼ ð1 l c Þ M 3 sin 2 y þ M 3 X X þ 1 cos 2 y M 1 M 1 þl c ð3þ r r gz where l c is the fluid pressure ratio within the country rock and M 1 and M 3 are functions of the country rock friction coefficient, m c [Lachenbruch and McGarr, 1990], M 1 ¼ M 3 ¼ ð4þ! 1 m 1 þ p ffiffiffiffiffiffiffiffiffiffiffiffiffi c ð5þ 1 þ m 2 c m c 1 p ffiffiffiffiffiffiffiffiffiffiffiffiffi 1 þ m 2 c! 1 ð6þ 4of15

5 Table 1. Definition of Variables Variable Parameter Units % ss Equation (9); relates excess pressure from transient dimensionless results to steady state values g Gravitational acceleration m s 2 (Lt 2 ) k Permeability m 2 (L 2 ) k c Country rock permeability m 2 (L 2 ) k cmax Country rock permeability in direction m 2 (L 2 ) of maximum permeability k depth14 Permeability-depth relation with k cmax defined N/A by equation (8). k depth15 Permeability-depth relation with k cmax 1 order of magnitude N/A less than defined by equation (8). k f Fault zone permeability m 2 (L 2 ) k fmax Fault zone permeability in direction of maximum m 2 (L 2 ) permeability k cmin Country rock permeability in direction m 2 (L 2 ) minimum permeability M 1 Equation (5); function of m c dimensionless M 3 Equation (6); function of m c dimensionless P Pore pressure Pa (ML 1 t 2 ) P c Pore pressure in country rock Pa (ML 1 t 2 ) P * Pore pressure above hydrostatic Pa (ML 1 t 2 ) * P ss Pore pressure above hydrostatic at steady state Pa (ML 1 t 2 ) P f Pore pressure in fault zone Pa (ML 1 t 2 ) P hydro Hydrostatic fluid pressure Pa (ML 1 t 2 ) X Equation (7); parameter describing s 2 dimensionless relative to s 1 and s 3 q Heat flow mw m 2 (Mt 3 ) v Surface discharge cm a 1 (Lt 1 ) z Depth (units of meters, unless m (L) otherwise noted) l Equation (1); Fluid pressure ratio dimensionless l c Country rock fluid pressure ratio dimensionless l f Fault zone fluid pressure ratio dimensionless m Friction coefficient dimensionless m c Country rock friction coefficient dimensionless m f Fault zone friction coefficient dimensionless r r Bulk rock density kg m 3 (M L 3 ) s Normal stress Pa (ML 1 t 2 ) s 1 Maximum principal stress Pa (ML 1 t 2 ) s 2 Intermediate principal stress Pa (ML 1 t 2 ) s 3 Minimum principal stress Pa (ML 1 t 2 ) s H Maximum horizontal principal stress Pa (ML 1 t 2 ) s h Minimum horizontal principal stress Pa (ML 1 t 2 ) s Hf Maximum horizontal principal stress Pa (ML 1 t 2 ) within fault zone s hf Minimum horizontal principal stress within fault zone Pa (ML 1 t 2 ) s v Vertical principal stress (i.e., overburden stress) Pa (ML 1 t 2 ) s y Fault zone total normal stress Pa (ML 1 t 2 ) t Shear stress Pa (ML 1 t 2 ) t y Shear stress on fault zone oriented at angle y to s 1 Pa (ML 1 t 2 ) y Angle between s H and fault zone degrees and X ranging from 0 to 1 describes the magnitude of s 2 relative to s 1 and s 3 [Ramsay and Lisle, 2000]: X ¼ s 2 s 3 s 1 s 3 Analysis of borehole breakouts and drilling-induced tensile fractures within the SAFOD pilot hole show that X 0.2 at a distance of 1.8 km SW of the surface trace of the SAF [Hickman and Zoback, 2004]. Laboratory experiments on samples from the SAFOD pilot hole and surface outcrops reveal country rock friction coefficients of m c 0.6 [Tembe et al., 2006; Carpenter et al., 2009]. Assuming the fault zone has a similar friction coefficient, such that m f = m c = 0.6 (relaxation of this assumption is discussed in section 4) and values of X = 0.2, y =85, and l c = 0.38, the fluid pressure ratio in the fault zone required for slip on the SAF is l f = 2.15 (equation (2)) (Figure 5). For a more conservative ð7þ value of y = 69, the value determined from observations within the SAFOD pilot hole [Hickman and Zoback, 2004], the necessary fluid pressure ratio in the fault is l f = Several studies have illustrated that total principal stresses within mature fault zones may be magnified to values greater than the surrounding country rock as a result of contrasting rheologic properties [Rice, 1992; Chéry et al., 2004; Faulkner et al., 2006]. Such a stress state would allow for fluid pressures considerably greater than lithostatic to exist locally within the fault zone without causing hydrofracture [Rice, 1992; Chéry et al., 2004; Faulkner et al., 2006]. 2. Models of Coupled Fluid Flow and Heat Transport [14] To assess whether realistic fluxes of mantle fluid have the potential to generate large fluid pressures and 5of15

6 Figure 5. Mohr-Coulomb diagram illustrating the role of localized fluid pressures in explaining the relative weakness of the SAF [after Rice, 1992] scaled for m c = m f = 0.6, l c = 0.38, y =85 and assuming cohesion is equal to zero. As mentioned in the text, differences in rheologic properties may allow total principal stresses within the fault zone to be substantially higher than in the country rock [Rice, 1992; Faulkner et al., 2006]. The star represents the stress state on the SAF inferred from the orientation of the fault (y) with respect to the maximum horizontal principal stress in the surrounding country rock (s H ). Solid Mohr circles represent stress states on vertical planes in the country rock (on the left) and within the fault zone (on the right). Failure envelopes (m c = m f = 0.6) are shown for a hydrostatic fluid pressure ratio in the country rock l c = 0.38 and the fluid pressure ratio in the fault zone necessary for failure on the SAF (l f =2.15). localize them onto a near-vertical fault, we incorporate a mantle source of fluids in 2-D numerical models of coupled fluid flow and heat transport [e.g., Hanson, 1992, 1995, 1997; Fulton et al., 2009]. We also evaluate whether this scenario is consistent with heat flow and near-surface hydrologic observations. Our model domain consists of a 150 km long 12 km thick 2-D cross section representative of the seismogenic crust perpendicular to the SAF within the Parkfield region near SAFOD (Figures 6a 8a; note that Figures 6a 8a highlight only the 50 km to each side of the fault). The fault zone is located 50 km from the SW model boundary. We prescribe a boundary condition of atmospheric pressure and temperature at the top of the model domain, and define the sides of the model domain as no-flow boundaries for fluid flow and heat transport. We assign a heat flux of 64 mw/m 2 at the base of the model, and set radiogenic heat production at a constant value of 1.2 mw/m 3 throughout the model domain, typical of measured values for rocks within the area [Lachenbruch and Sass, 1980; Williams et al., 2004]. We prescribe thermal conductivity of 2.5 W m 1 K 1 for rock and 0.6 W m 1 K 1 for water [Voss, 1984; Sass et al., 1997]. Together, the heat sources and thermal conductivities result in a near-surface heat flow of 78 mw m 2 (equivalent to the observed regional average), and a temperature of 350 C at 12 km depth along the model base, corresponding to the base of the seismogenic crust [Lachenbruch and Sass, 1980; Sass et al., 1997; Williams et al., 2004; Fulton et al., 2004b]. In these models, we do not include any frictional heat production on the fault Flux of Mantle-Derived Fluids [15] On the basis of the inferred volume of serpentinized mantle wedge beneath Cascadia and the age of the SAF system, we calculate a realistic estimate of the potential dehydration fluid source. Assuming a 20 km thick wedge of mantle material that is 50% serpentinized [Bostock et al., 2002; Kirby et al., 2002] and which fully dehydrates over 25 Ma, we estimate an average water flux of kg s 1 m 2 of basal area. This is comparable to the maximum pulse of water (averaged over a 0.5 Ma interval) estimated from lower crustal dehydration but is sustained for 25 Ma [Fulton et al., 2004a, 2005, 2009]. Even larger fluid fluxes may be possible if dehydration of the serpentinized mantle occurs over a shorter time, but this would likely allow any overpressures to dissipate within a few hundreds of thousands of year after the sources are diminished, as is the case for crustal dehydration [e.g., Fulton et al., 2009]. By choosing 25 Ma as the duration for dehydration, our calculations incorporate the maximum sustained fluid flux from dehydrating serpentinized mantle that would affect locations along the entire strike of the SAF. Our study focuses solely on mantle-derived water, although additional fluids such as CO 2 may play a complementary role in generating overpressures [e.g., Bredehoeft and Ingebritsen, 1990; Kharaka et al., 1999; Faulkner and Rutter, 2001]. However, our estimated mass flux of mantle-derived water is >12 times the estimated fluxes of mantle-derived CO 2 inferred from geochemical observations near Parkfield [Kennedy et al., 1997; Kharaka et al., 1999; Faulkner and Rutter, 2001]. [16] On the basis of observations that suggest the greatest mantle fluid source is NE of the fault zone (see section 1.1), and the spatial extent of the zone of serpentinized mantle wedge material inferred beneath Cascadia (Figure 4) [Bostock et al., 2002], we prescribe the dehydration driven fluid flux for 50 km along the model base, from the fault zone to the NE (Figures 6a 8a). The location of these fluid sources to the NE of the fault is also consistent with the expected location of serpentinized mantle relative to the estimated location of the paleotrench 160 km SW of the SAF at Parkfield (Figure 4b) [Kirby et al., 2002]. The fluids entering the model base are prescribed a temperature of 350 C. We assign densities of 2600 kg m 3 for rock [Boness and Zoback, 2004] and 1000 kg m 3 for water at 25 C, which changes by kg m 3 C 1 [Voss, 1984]. 6of15

7 Figure 6. Model domain and results for scenarios with maximum country rock permeability oriented horizontally, showing the 50 km to each side of the fault, although the model domain extends 50 km farther to the NE. (a) Schematic showing the orientation of country rock permeability and a 500 m wide low-permeability fault zone representing the SAF. A flux of mantle-derived water is prescribed along the model base from the fault zone to 50 km NE (blue arrows). (b) Simulated steady state fluid pressure ratio (l) for country rock permeability defined by k depth15 and an anisotropy ratio of 100:1. (c) Depth-averaged values of l for simulations with crustal permeability defined by k depth14 (red) and k depth15 (blue). (d) Simulated heat flow (q) taken at m model depth for both simulations (k depth14 = red; k depth15 = blue); there is very little difference between the two. (e) Surface discharge (v) from both simulations; again there is very little difference between the two. [17] We solve the equations for 2-D coupled fluid flow and heat transport using the finite element code SUTRA [Voss, 1984]. We run steady state simulations in order to determine the maximum fluid pressures that could be generated. We also run transient simulations for a subset of our models, in order to illustrate the evolution of fluid pressure. In these transient simulations, we start with a hydrostatic initial condition to illustrate the potential for fluid pressure generation solely from mantle dehydration. These transient simulations incorporate values of rock and water compressibility of and Pa 1 [Neuzil, 1986; Ge and Garven, 1992; Voss, 1984] and heat capacities of 1000 and 4182 J kg 1 K 1, for rock and water respectively [Taylor et al., 1982; Mossop and Segall, 1997; Voss, 1984]. Because total principal stresses within the fault zone may be magnified [e.g., Rice, 1992; Chéry et al., 2004; Faulkner et al., 2006], and because our focus is to test the basic hypothesis that mantle fluids have the potential to contribute in localized overpressure development, our models do not include a feedback on permeability and pressure that may result from possible hydrofracture Permeability Architecture [18] In our models, we consider a series of different permeability architectures for the crust. For the permeability of the country rock (k c ), we assign maximum values following a permeability-depth relation derived specifically for continental crust undergoing prograde metamorphism (termed k depth14 )[Manning and Ingebritsen, 1999]: logðk c max Þ ¼ 14 3:2 logðþ z where z is depth in kilometers. We also consider country rock permeability 1 order of magnitude lower than that defined by equation (8) (k depth15 ), which is also consistent with the range of constraints used to construct this relation [Manning and Ingebritsen, 1999]. Porosity ranging from 10 1 to 10 3 is defined by assuming a Kozeny-Carman relationship between permeability and porosity and follows a similar decrease with depth [Carman, 1956; Manning and Ingebritsen, 1999]. Details regarding fault zone permeability and the degree and direction of permeability anisotropy for each suite of models are described in sections Fault Barrier [19] Observations in and around the SAFOD borehole have been interpreted to indicate that the SAF acts as a barrier to fluid flow within more permeable country rock. Well data from both sides of the fault within the Parkfield region reveal a fluid pressure contrast across the fault, with subhydrostatic pressures to the SW (l c 0.38) and ð8þ 7of15

8 Figure 7. Model domain and results for simulations with maximum country rock permeability dipping 30 to the NE. Descriptions are the same as in Figure MPa of overpressure at 1.5 km depth, 1.4 km to the NE of the fault (l c = 0.42) [Johnson and McEvilly, 1995; Zoback and Hickman, 2005]. Mud gas chemistry observations during drilling of the SAFOD borehole include sharp contrasts in the mantle helium signature and other fluid chemistry across the fault [Wiersberg and Erzinger, 2007; Erzinger and Wiersberg, 2007], as well as reduced gas flux within the fault zone [Erzinger and Wiersberg, 2007]. Geophysical logs, cuttings, and core observations are consistent with this conceptual model, as they reveal a clay-rich primary fault plane formed within a <650 m thick interval of shale and siltstone [e.g., Solum et al., 2006] (Figure 2). [20] To simulate a fault acting as a barrier, we implement a 500 m wide low-permeability zone in our models, with the fault permeability (k fmax ) 3 orders of magnitude lower than that of the country rock (k cmax ). This corresponds to the difference between permeability curves calibrated specifically for brittle continental crust [Manning and Ingebritsen, 1999] and laboratory data for clay-rich fault gouge that document permeabilities of <10 21 to m 2 under MPa of effective pressure [Faulkner, 2004]. As an initial model scenario, we specify that the direction of maximum permeability in the country rock (k cmax ) is horizontal. For all scenarios, we specify the direction of maximum permeability within the fault zone (k fmax )tobe vertical. The ratio of anisotropy for both the fault and country rock is prescribed a value of 100: Fault Barrier and NE Dipping Country Rock Anisotropy [21] For a wide range of 2-D hydrogeologic and fault zone structures, including models similar to our initial fault barrier model described above, the ability to localize fluid pressure from a regional fluid source within a near-vertical fault zone has proved difficult [e.g., Fulton et al., 2009]. On the basis of previous studies, which show that dipping permeable strata can focus overpressures toward their updip termini by a hydraulic condition known as the centroid effect [e.g., Flemings et al., 2002], we evaluate the role of dipping hydrogeologic structures within the country rock as a potential means of localizing fluid pressures onto the fault zone. Rather than prescribing discrete permeable pathways, we assume the direction of maximum permeability in the country rock (k cmax ) dips 30 to the NE. This corresponds to the direction of paleosubduction recorded within the Franciscan Complex and to the orientation of several large thrust faults and geologic structures within the Franciscan Complex and other geologic units NE of the fault zone (Figure 3) [e.g., Bailey et al., 1964; Wakabayashi, 1999; Dickinson, 2002; Guzofski et al., 2007]. Additionally, at this orientation faults and fractures in the country rock are likely to be critically stressed for failure (assuming m c 0.6 and with s 1 nearly fault-normal), and as observed in other deep boreholes, these critically stressed faults and fractures may act as the most permeable pathways for fluid flow [Barton et al., 1995; Townend and Zoback, 2000]. As in our other model simulations, we prescribe the ratio of anisotropy for both the fault and country rock as 100:1. We evaluate the sensitivity of our model results to the magnitude of anisotropy by conducting a subset of simulations in which we maintain the value of the maximum permeability and allow the anisotropy ratio for both the fault and country rock to vary between 5:1 and 500: SAFOD-Specific NE Dipping Anisotropy and a Serpentinite Seal [22] In addition to model scenarios with NE dipping permeability anisotropy in the country rock, which may generally represent much of the San Andreas Fault system 8of15

9 Figure 8. Model domain and results for simulations representative of the Parkfield region around SAFOD, with maximum country rock permeability dipping 30 to the NE and a low-permeability serpentinite unit acting as a seal, extending from the fault zone 10 km NE at 3 4 km depth. Descriptions are the same as in Figures 6 7. along its strike, we also evaluate a scenario specific to the Parkfield region, where additional large-scale hydrogeologic structures have been inferred. In particular, in this area a subhorizontal serpentinite unit is thought to extend km from the fault zone to the NE at a depth of 2 3 km [e.g., McPhee et al., 2004]. This unit is associated with the Coast Range Ophiolite and has been mapped or inferred along much of the SAF system in northern and central California [Irwin and Barnes, 1975; Irwin, 1990]. On the basis of geophysical observations and the distribution of surface springs, the serpentinite has been interpreted to act as a low-permeability seal that promotes overpressure development beneath it [Irwin and Barnes, 1975; Eberhart- Phillips and Michael, 1993]. For a set of Parkfield-specific scenarios, we model this structure as a low-permeability zone (10 20 m 2 )[Tenthorey and Cox, 2003] from 3 to 4 km depth, extending from the fault zone 10 km to the NE (Figure 8). 3. Results 3.1. Pore Pressures [23] Results of model simulations with a fault acting as a barrier and with maximum permeability anisotropy in the country rock oriented horizontally show that moderate steady state fluid overpressures can be generated by a realistic mantle fluid source (Figure 6). For our low-permeability scenario (k depth15 ), depth-averaged values of l reach 0.61, with pressures along much of the base of the model to the NE of the fault reaching values of l c 1.10 (Figures 6b and 6c). The zone of elevated fluid pressure is generally limited to depths >5 km (Figure 6b); excess pore pressure within the fault zone (P f P hydro )is1 MPa at 2.7 km depth and 17.2 MPa at 6 km depth. These pore pressures are considerably larger than those generated by crustal dehydration; depth-averaged pore pressure ratios from models of crustal dehydration peak at values of l 0.46 for a similar permeability architecture, and are short-lived owing to the limited time over which sources are active [Fulton et al., 2009]. For our higher permeability scenario (k depth14 ), depth-averaged overpressure is considerably smaller, with the largest values of l 0.40 (Figure 6c). Although the fluid pressure ratios generated in the low-permeability scenario are much larger than hydrostatic, they are not high enough to explain a weak SAF, and they are distributed over a broad zone NE of the fault and above the fluid source region, rather then localized within the fault zone as needed to explain relative fault weakness (Figures 6b and 6c). [24] In model simulations with NE dipping permeable pathways truncated by a low-permeability SAF acting as a barrier, we find that large fluid pressures can be focused toward the fault (i.e., localized) (Figure 7). Fluid pressure ratios within the fault reach values considerably larger than for the case in which maximum permeability in the country rock is horizontal (Figures 6b and 6c). For the low-permeability scenario (k depth15 ), depth-averaged values of l are within 1 km NE of the fault plane (Figure 7). Along the edge of the fault zone toward the model base, pore pressure ratios reach values of l f 2.0. Notably, the highest pore pressure ratios occur within the 500 m wide low-permeability fault zone or straddle its NE edge (Figure 9a). Values of l to the NE of the fault diminish to near hydrostatic values (l < 0.4) over a distance less than 13 km from the fault (Figures 7b, 7c, and 9). To the SW of the fault, there is an even sharper 9of15

10 depth-averaged values of l f are 0.46, maximum fluid pressure ratios near the base of the fault are l f = 0.58, and l = beneath the serpentinite seal at 4 km depth, within the fault zone and extending to 2.5 km NE (Figures 8b and 8c). Figure 9. Detailed view of simulated fluid pressure near the fault zone from low-permeability scenarios for (a) dipping anisotropy case (e.g., Figure 7) and (b) Parkfield-specific case (e.g., Figure 8). The lateral extent of the lowpermeability fault zone is marked by vertical dotted lines. Labels for each profile describe the depth at which the results have been extracted and the peak excess pressure. contrast in pore pressure, with near-hydrostatic values immediately on the SW side of the low-permeability fault zone (l < 0.4). The simulated excess pore pressure within the fault zone is 6 MPa at 2.7 km depth and 45 MPa at 6 km depth. Results of simulations for our high-permeability scenario (k depth14 ) are similar in character, but with lower fluid pressure ratios; the maximum depth-averaged pore pressure ratio for this scenario is l = 0.41 and occurs within the fault zone (compared with 0.71 for k depth15 ; Figure 7c). [25] For the suite of models designed to represent the hydrogeologic structure around Parkfield, California, and the SAFOD site, the low-permeability serpentinite unit both enhances the generation of elevated pore pressure, and localizes it within the fault zone, at all depths below 3 km (Figure 8). For the low-permeability case (k depth15 ), depth-averaged overpressures peak at l f = 1.16 within the fault zone. Simulated pore pressures within the fault zone near the model base are l f , and l = 1.6 beneath the serpentinite seal at 4 km depth within the region 500 m NE of the fault (Figures 8b and 8c). Excess pressure is 8 MPa at 2.7 km depth and 160 MPa at 6 km depth (Figure 9b). For the high-permeability scenario (k depth14 ), 3.2. Effect on Heat Flow and Surface Discharge [26] Although large fluid pressures are localized toward the fault zone in models that include dipping permeability anisotropy in the country rock, the associated fluid flow has very little effect on heat flow or on surface fluid discharge (Figures 6d 8d and 6e 8e). Simulated heat flow from these model scenarios is marked by an increase of <10 mw m 2 near the fault (Figures 6e 8e). This spatial variability in heat flow is consistent with the range of observed heat flow scatter around the fault and is considerably less than the 40 mw m 2 near-fault anomaly expected from frictional heating [e.g., Lachenbruch and Sass, 1980; Fulton and Saffer, 2009]. [27] Similarly, simulated surface discharge exhibits only a small increase near the fault. In simulations with NE dipping country rock anisotropy, the peak discharge is <1.0 cm a 1. This small rate of discharge is not sufficient to generate a noticeable seep or spring at the surface (Figure 7e), and is consistent with the general lack of springs along the surface trace of the fault [Waring, 1965; Lachenbruch and Sass, 1980; Kharaka et al., 1999]. For both simulated heat flow and surface discharge, there is very little difference between our high- and low-permeability scenarios (k depth14 and k depth15 ) (Figures 6e 8e) Sensitivity to Degree of Anisotropy [28] With larger permeability anisotropy in the country rock, the effect of NE dipping permeable pathways in generating localized excess fluid pressures along the fault is more pronounced (Figure 10). For our low-permeability scenario (k depth15 ), an anisotropy ratio of 10:1 generates a peak depth-averaged fluid pressure ratio at the fault of l f = 0.43, whereas anisotropy ratios of 100:1 and 500:1 generate depth-averaged l f values as high as 0.71 and 1.12, respectively. Although increased anisotropy results in larger fluid pressures on the fault, the associated fluid flow in all cases has a small effect on near-surface heat flow (<10 mw m 2 ) (Figure 10a). Similarly, varying the permeability anisotropy value results in only slight differences in surface discharge (not shown; similar to Figure 7e); simulated surface discharge peaks near the fault at <1.4 cm a 1 for an anisotropy ratio of 500:1 and at <1.0 cm a 1 for anisotropy of 100:1. As mentioned above, these small, simulated discharge values are consistent with the general lack of observed surface springs along the fault trace Transient Results [29] To investigate the evolution of fluid overpressures, we ran transient simulations for a subset of our models. For our model scenario incorporating low crustal permeability (k depth15 ), a fault barrier, and NE dipping maximum country rock permeability (the model scenario which best represents the known geology in the Parkfield and SAFOD region), large localized fluid pressures (similar to those illustrated in the steady state results of Figure 7) develop within 2 3 Ma (Figure 11a). We report overpressures normalized to the 10 of 15

11 steady state value by 10 Ma, and are maintained at these levels until 25 Ma, the end of our simulation and the cessation of our fluid source, based on the assumptions described above (section 2.1). 4. Discussion [30] Using a realistic estimate of mantle-derived fluid flux and incorporating hydrogeologic structures consistent with observations, our results illustrate the potential role of mantle-derived fluids in causing both the absolute and relative weakness of the SAF. Our results are unique in that (1) they provide a plausible mechanism for the generation of sustained excess fluid pressures that are localized along a near-vertical fault zone, without the need to invoke a basal pressure boundary directly beneath the fault, complex fault permeability behaviors, or transient processes such as shear compaction or thermal pressurization [e.g., Byerlee, 1990; Figure 10. Sensitivity of model results to the magnitude of anisotropy in country rock permeability. Results are shown for models with NE dipping maximum permeability anisotropy. Maximum permeability is held constant for each simulation and defined by the lower fit to the relation of Manning and Ingebritsen [1999] (k depth15 ), but the anisotropy ratio is varied. (a) Depth-averaged values of l for anisotropy ratios of 5:1, 10:1, 50:1, 100:1, and 500:1. (b) Simulated heat flow for the same models as Figure 10a. There is very little difference between simulations in terms of simulated heat flow (Figure 10b) and fluid discharge (not shown). steady state value, defined by the excess pore pressure (P P hydro ) at a given time divided by excess pressure at steady state: % ss ¼ P*=P* ss 100 ð9þ By 2 Ma, depth-averaged pressures within the fault zone reach 50% of the steady state values (Figure 11b); this suggests that a realistic mantle source would strongly affect pore pressures in the vicinity of the fault zone shortly after the passage of the MTJ and the establishment of the fault zone as a barrier to fluid flow. Fluid pressures within the fault zone continue to rise quickly, reaching >90% of the Figure 11. Transient simulation results for the model scenario shown in Figure 7, using the low-permeability relation for the country rock (k depth15 ). (a) Depth-averaged values of l as a function of distance from the fault, shown for several time steps and for the steady state result. (b) Evolution of depth-averaged excess pore pressure within the fault zone from an initial hydrostatic condition, expressed as a percentage of the steady state value. 11 of 15

12 Rice, 1992; Sleep and Blanpied, 1992; Andrews, 2002] and (2) they are consistent with observations from SAFOD and the Parkfield region that suggest the SAF acts as a barrier to fluid flow [Zoback and Hickman, 2005; Solum et al., 2006; Wiersberg and Erzinger, 2007], with elevated fluid pressures observed in wells immediately NE of the SAF [Johnson and McEvilly, 1995; Zoback and Hickman, 2005], and with regional overpressures beneath a serpentinite unit extending NE from the SAF as inferred from high V p /V s ratios [Eberhart-Phillips and Michael, 1993]. [31] Although direct pore pressure measurements have not yet been conducted in the SAFOD boreholes, measurements during drilling showed no evidence for significantly elevated pore pressure within the fault zone [Zoback and Hickman, 2005; M. Zoback and S. Hickman, personal communication, 2009]. However, our results demonstrate that fluid pressure ratios at the depth of the SAFOD borehole may not fully reflect the fluid pressure state within the fault zone over much of its depth extent. For example, our results show that at 2.7 km (the depth at which the SAFOD main borehole crosses the fault zone), values of both fluid pressure ratio and excess fluid pressure within the fault zone should be considerably smaller than those at even 6 km depth (excess pore pressures of 8 MPa versus 160 MPa; Figure 9b). This difference depends on the depth of the borehole, and to a lesser extent on whether the model includes a low-permeability seal within the country rock NE of the fault zone (Figures 9a and 9b). [32] In a general sense, our results show that for realistic 2-D permeability architectures, mantle-derived fluid sources can generate large excess pore pressures. Additionally, our results illustrate that strongly anisotropic country rock permeability resulting from NE dipping faults and fractures can focus fluid pressures along a low-permeability fault zone, yet the associated fluid flow should have very little effect on heat flow and surface fluid discharge, causing <10 mw m 2 increase in heat flow and <1.0 cm a 1 of surface discharge near the fault. Fluid pressures along and immediately to the NE of the fault may be quite large (l f > 1.0 and l c > 0.7), and diminish to near-hydrostatic values by km from the fault (Figures 8 and 9). In order for realistic mantle fluid sources to generate excess fluid pressures localized along the SAF sufficient to explain its apparent weakness, crustal permeability values near the lower bound of those defined by Manning and Ingebritsen [1999] are required (Figure 8). However, it is important to note that additional processes may act in combination with mantle sources to drive elevated pore pressures. Pressures generated by other regional geologic forcings such as tectonic compaction [Berry, 1973], crustal dehydration [Fulton et al., 2009], or CO 2 release [Faulkner and Rutter, 2001] would also be focused toward the fault zone along dipping permeable pathways and add to the effects of mantle dehydration sources. [33] Additionally, localized fluid pressures and low friction coefficient material within the fault zone may act in combination to account for geomechanical observations that imply a weak fault in a much stronger crust. However, as illustrated in Figure 12 (derived from equations (2) (4)), for all fault zone friction coefficients m f 0.3, fluid pressures in the fault zone would need to be lithostatic or greater (l f 1) in order to explain an active fault plane oriented at 69 to s 1 (y =69 ), regardless of the friction coefficient (m c ) or pore pressure ratio (l c ) in the surrounding crust. To account for an orientation of y = 85, as is inferred for the San Francisco Peninsula [Townend and Zoback, 2004], supralithostatic pressures are needed for all values of m f For comparison, the lowest friction coefficient measured to date on SAFOD samples is m = 0.4, for cuttings from a finegrained shear zone [Tembe et al., 2006]; measurements on cuttings from phases 1 and 2 of SAFOD drilling and from rock outcrop samples within the Parkfield area document values of m c > [Tembe et al., 2006; Carpenter et al., 2009]. Weak minerals that may be present in the SAF, such as serpentinite or talc [Moore and Rymer, 2007], generally exhibit values of m = when tested as pure monomineralic standards [Reinen et al., 1992; Moore et al., 1997; Moore and Lockner, 2008; Carpenter et al., 2009]. Even with weak materials such as these in the fault zone, localized fluid pressures of l f = 1 would still be needed to account for the steepest inferred orientations (y) between s H and the SAF (Figure 12). [34] In our simulations, the highest pore pressures occur within the fault zone and to the NE (Figures 8 and 9). Although pore pressures in excess of the lithostatic pressure are possible if total principal stresses within the fault zone are magnified owing to contrasts in rheologic properties [e.g., Rice, 1992; Faulkner et al., 2006], in a subset of our models the simulation of similarly high pore pressures to the NE of the fault may be unrealistic, because total principal stresses would not be magnified in this region. Although our models do not include a feedback on pressure and permeability from hydrofracture or fluid-induced failure, our results would not be affected substantially for two main reasons. First, at stress states relevant over most of the depth extent of our models (>3 km), supralithostatic fluid pressures would cause shear failure on preexisting faults striking parallel to the SAF [Sibson, 1998], which would increase permeability anisotropy and focusing of pressure onto the fault zone. Second, if rock failure or hydraulic fracturing occur, this would effectively buffer the pore pressure in the country rock at or near a value of l c = 1, thus resulting in even greater localization of the highest pressures only on the fault itself where magnified total principal stresses may prevent hydrofracture. [35] Results from our transient model simulations illustrate that the focusing mechanisms described here could have an effect on fault zone fluid pressures soon after the creation of the SAF (Figure 11). Simulated fluid pressures within the fault zone reach 50% of steady state values within 2 Ma after the passage of the MTJ and the establishment of the SAF as a hydrologic barrier and reach 80% of steady state values within 5 Ma. Assuming the MTJ migrates northward at a rate 40 km Ma 1, the maintenance of large localized fluid pressures for 25 Ma implies that mantle-derived water and the focusing mechanisms described above could affect fault mechanical strength over a region encompassing nearly the entire length of the SAF, unlike the short-lived effects expected for crustal dehydration [Fulton et al., 2009]. The southward increase in pore pressures implied by our transient results is also consistent with progressively increasing values of y (from 55 in the northern California immediately south of the MTJ, to 62 in the San Francisco Bay area, to 74 just north of Parkfield) 12 of 15