Influence of fault slip rate on shear induced permeability

Transcription

1 Click Here for Full Article JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115,, doi: /2009jb007013, 2010 Influence of fault slip rate on shear induced permeability Wataru Tanikawa, 1 Masumi Sakaguchi, 2 Osamu Tadai, 2 and Takehiro Hirose 1 Received 30 September 2009; revised 22 January 2010; accepted 16 February 2010; published 20 July [1] We measured permeability in sandstone and granite sheared at slip rates from 10 4 to 1.3 m/s under low normal stress at confining pressures up to 120 MPa. As the slip rate increased, the permeability of Berea sandstone decreased by an order of magnitude, whereas that of Indian sandstone and Aji granite increased by 3 orders of magnitude at high slip rates. A fine grained gouge layer of thickness developed during slip, and the wear rate was increased abruptly at high slip rates. Microcracks and mesoscale fractures formed at slip rates above 0.13 m/s. Numerical modeling showed that the slip surface temperature increased by several hundred degrees for slip velocities above 0.13 m/s and exceeded the a b phase transition temperature of quartz at 1.3 m/s. Both the temperature rise and the temperature gradient at the slip surface were high at fast slip rates. We attributed reduced permeability after slip in porous sandstone to the low permeability gouge layer. An abrupt permeability increase in low permeability rocks at high slip rates was caused by heat induced cracks. An increase in the rate of wear of gouge with increasing slip velocity was caused by frictional heating that reduced the rock strength. The host rock permeability that separated reductions and increases in permeability was about m 2 at 10 MPa effective pressure. Our results suggest that abrupt increases in shear stress during slip in a low permeability fault zone caused by thermal cracking, which may decrease the total slip displacement. The abrupt permeability increase at high slip rates in low permeability rocks agrees with hydrogeochemical phenomena observed after earthquakes. Citation: Tanikawa, W., M. Sakaguchi, O. Tadai, and T. Hirose (2010), Influence of fault slip rate on shear induced permeability, J. Geophys. Res., 115,, doi: /2009jb Introduction [2] Understanding changes in permeability during shear deformation is crucial to explain the dynamic hydrological and mechanical processes that occur during earthquakes at shallow and deep crustal levels. Hydrological and hydrogeochemical changes during and after earthquakes are linked to temporal changes in fluid flow systems and fluidrock interactions along fault zones [Rojstaczer et al., 1995; Ishikawa et al., 2008]. In addition, the permeability structure along a fault zone is closely related to the hydrothermomechanical coupling behavior of faults [Sibson, 1973; Wibberley and Shimamoto, 2005; Rice, 2006]. [3] Claesson et al. [2004] attributed hydrogeochemical changes in groundwater observed in boreholes after a large earthquake (M = 5.8) in Iceland to a rapid change from an impermeable to a permeable state in the fault plane. Observations of increases in streamflow by 1 order of magnitude within days of the 1989 Loma Prieta earthquake in California suggest that hydraulic conductivity increased 1 Kochi Institute for Core Sample Research, Japan Agency for Marine Earth Science and Technology, Nankoku, Japan. 2 Marine Works Japan Ltd., Yokohama, Kanagawa, Japan. Copyright 2010 by the American Geophysical Union /10/2009JB by 1 order of magnitude after the earthquake [Rojstaczer et al., 1995]. Hydrologic changes observed after the 1995 Kobe earthquake were explained by a fivefold increase in hydraulic conductivity along the fault zone as a result of the earthquake [Tokunaga, 1999]. The change in the phase of the water level response after earthquakes in Southern California was caused by changes in permeability [Elkhoury et al., 2006]. The Cerda landslide movement associated with seismic events might have been triggered by fluid flow or gas released from a shallow depth [Agnesi et al., 2005], which is related to coseismic permeability change as well. [4] During an earthquake the permeability structure within a fault zone is related to thermal pressurization, which is one of the dynamic slip weakening mechanisms, caused by the increase in pore pressure that results from frictional heating along the fault zone [Mase and Smith, 1985; Wibberley and Shimamoto, 2005]. This mechanism is equivalent to a weakening process in catastrophic landslides [Sassa et al., 2004; Wafid et al., 2004]. [5] There have been several laboratory tests of the permeability of fault related rocks [Caine et al., 1996; Evans et al., 1997; Faulkner and Rutter, 1998, 2000; Wibberley and Shimamoto, 2003; Tsutsumi et al., 2004; Uehara and Shimamoto, 2004; Boutareaud et al., 2008; Mizoguchi et al., 2008; Tanikawa et al., 2009]. Each of these studies measured the permeability of natural fault rocks to investi- 1of18

2 Table 1. Physical Properties of Test Specimens and Methodology Used for Permeability Measurement Rock Type Grain Size (mm) Porosity (%) Matrix Density (kg/m 3 ) Permeability at 10 MPa (m 2 ) Method for Permeability Test Berea sandstone E 14 Steady state flow Indian sandstone E 17 Steady state flow Aji granite E 20 Transient pulse Inada granite >1.87E 19 Transient pulse gate the elastic and nonelastic changes in permeability under an isostatic confining pressure. More recent studies have examined the pre to postfailure evolution of permeability under increasing differential stress by conducting standard triaxial experiments on cylindrical specimens [Zhu and Wong, 1997; Heiland, 2003; Mitchell and Faulkner, 2008]. These studies found that changes in permeability are related to porosity, rate of strain, and failure modes. A biaxial test performed by Faoro et al. [2009] found that progressive formation of gouge can affect the evolution of permeability in fractured rocks. However, the foregoing triaxial and biaxial compression tests were performed under limited conditions of strain rate and total displacement (mostly less than several millimeters of slip). The dynamic change in permeability of powdered gouge materials has been measured in a rotary shear apparatus by applying direct shear with a long total displacement [Zhang et al., 1999, 2001]. Those studies showed a continuous reduction in permeability caused by shear deformation and the development of strong permeability anisotropy. A similar permeability anisotropy is observed on natural fault rocks [Faulkner and Rutter, 1998; Wibberley, 2002], and permeability parallel to foliation is more than 1 order of magnitude lower than permeability perpendicular to foliation. The rotary shear apparatus used by Zhang et al. [1999] can produce large slip displacements; however, the available range of slip rates for this apparatus is 10 8 to 10 5 m/s, which is much lower than seismic slip rates (>0.1 m/s). [6] Geological and geochemical analyses of rocks from active fault zones have demonstrated temperature increases caused by shear related friction in the zone at faster slip rates, which produce more frictional heat [Sibson, 1975; Fukuchi et al., 2005; Mishima et al., 2006; Ishikawa et al., 2008]. In addition, laboratory experiments suggest that the evolution of permeability during high velocity slip accompanied by frictional melting and fault weakening (>0.1 m/s) [Hirose and Shimamoto, 2005; Hirose and Bystricky, 2007; Mizoguchi et al., 2009] is much different from that during low velocity slip (10 8 to 10 5 m/s). [7] In this study we conducted rock to rock friction tests under dry conditions using a rotary shear apparatus at various slip rates up to the coseismic slip rate (>1 m/s) on samples of Berea sandstone [Takahashi, 2003], Indian sandstone, Aji granite [Yukutake, 1989], and Inada granite [Ishido and Nishizawa, 1984]. Only high velocity friction tests were conducted on Inada granite. We measured the permeability of test samples under a high confining pressure. We used the results of our friction experiments and permeability measurements to investigate the dynamic evolution of permeability caused by seismic slip in a fault zone. In major seismic faults reactivation of previously formed faults segments are common in the earthquake cycle [Kirkpatrick et al., 2008], and clay rich gouge materials and pore water are present in the central part of fault zones [Wintsch et al., 1995]. In this study, though, we simplified the experimental conditions to see the influence of physical properties of the host rock and frictional heating on the permeability evolution. 2. Samples and Experimental Settings [8] We used four rock samples in our laboratory tests: Berea sandstone, Indian sandstone, Aji granite, and Inada granite (Table 1). Both sandstones are rich in quartz and plagioclase, with small amounts of clay minerals. Porosities and grain densities at atmospheric pressure were calculated from matrix volumes, which were measured with a commercial pycnometer (Penta pycnometer; Quantachrome Instruments, Boynton Beach, FL, USA). Both grain size ( mm) and porosity (10.8% 14.2%) in the Indian sandstone sample were smaller than those in the Berea sandstone sample ( mm, 20%). Minor bedding planes were apparent in both the Berea and the Indian sandstones, and samples were cored perpendicular to bedding planes. [9] We performed constant velocity rock to rock friction tests by using the high speed rotary shear testing apparatus of Shimamoto and Tsutsumi [1994] and a rotary shear testing apparatus with a wide range of speeds at the Kochi Core Center [Tadai et al., 2009a]. The machine used by Shimamoto and Tsutsumi [1994] has minimum and maximum rotation speeds of 10 and 1500 rpm, respectively, and those of the machine at the Kochi Core Center are 0.01 and 650 rpm, respectively. The former apparatus was used only for slip rates that required rotation speeds above 600 rpm. [10] We used the same sample assembly and experimental methodology for our tests with both friction apparatuses (Figures 1a and 1b) [Mizoguchi et al., 2009]. The cylindrical specimens were 25 mm in diameter and 20 mm long. The upper cylindrical specimen was fixed and the lower one was rotated under a fixed axial stress. The contact slip surfaces of specimens were roughened by rubbing with sandpaper (no. 200). A Teflon sleeve was used to cover the simulated fault plane so that worn gouge material generated along the slip surface during the friction test did not leak from the slip surface. Before the experiments the cylindrical specimens were dried in an oven at 80 C to eliminate pore water, but they were exposed to a humid environment (relative humidity, 30% 60%) during the experiment. We rotated at a constant speed from 0.15 to 1500 rpm and applied a constant normal stress of about 2 MPa during slip (Figure 1c). For the highest slip rate (1500 rpm), we applied the shear under a normal stress of from 1 to 2 MPa. Shear stress on the slip surface was monitored during slip. The friction 2of18

3 Figure 1. (a) Photo and (b) schematic of the apparatus used for friction tests. (c) Change with increasing slip displacement of friction coefficient, shear stress, and normal stress for Berea sandstone samples at a constant slip velocity of 12.9 mm/s. coefficient was calculated as the ratio of resistive shear stress to normal stress. Slip rate at a circular slip surface varies as a function of distance from the center of the axis of rotation; slip displacement and slip rate are 0 at the center of the sample and largest at the edge of the sample. Therefore, we defined the equivalent slip velocity V eq such that the total frictional work on a slip surface of area S is t V eq S [Shimamoto and Tsutsumi, 1994], and we assumed that the shear stress t is constant over the fault surface. Therefore, for our specimens 1500 rpm of rotational speed is equivalent to a slip velocity of 1.3 m/s. Samples for most tests were rotated 200 times, which is equivalent to 8 m of total slip displacement. In high velocity tests ( rpm) we varied the total slip displacement from 2 to 8 m to investigate the correlation of slip displacement with permeability evolution. [11] After the friction tests we measured permeability at room temperature under a uniform (isostatic) confining pressure in a high pressure oil apparatus. We removed the Teflon sleeve from the specimens without separating the contact surface between the specimens and then covered them with a polyolefin heat shrink jacket to confine the specimens for permeability testing. We excluded from our permeability measurements all specimens that broke during testing and those for which leaked gouge products were clearly observed during testing. Confining pressure was increased stepwise from 5 to 120 MPa to measure the dependence of permeability on effective pressure. [12] Permeability of relatively permeable Berea sandstone and Indian sandstone was measured by the steady state flow method of Bernabe [1987], and that of relatively impermeable Aji granite and Inada granite was measured by the transient pulse method of Zoback and Byerlee [1975], with distilled water used as the pore fluid in both cases. [13] For the steady state flow method a differential pore pressure was applied across the sample and the volume of 3of18

4 water flowing though it per unit time was measured with a digital balance. The equation for evaluating the (intrinsic) permeability k is Q A ¼ k L P up P down ; ð1þ where Q is the volume of fluid measured per unit time, A is the cross sectional area of the sample, h is the viscosity of the pore fluid, L is the sample length, and P up and P down are the pore pressures at the upper (inflow) and lower (outflow) ends of the specimen, respectively. Because the water flowing from the lower end of the specimen was released to atmospheric pressure, we assumed that P down was constant at 0.1 MPa. [14] For the transient pulse method we initially applied a differential pore pressure of 0.6 to 1.0 MPa between P up and P down. We kept P up constant in the range 1.4 to 2.0 MPa during permeability measurements, and the initial P down applied was 1 MPa. We evaluated permeability from the transition curve of P down with time. Pore pressure was relatively small compared to confining pressure, so effective pressure was nearly equivalent to confining pressure. Flow direction was normal to the slip plane for all specimens and, also, normal to the bedding planes for Berea sandstone and Indian sandstone samples. 3. Results 3.1. Mechanical Data [15] For both Berea sandstone and Indian sandstone samples, the friction coefficient decreased slightly with increasing shear displacement at slow slip rates and approached a nearly steady state level at several meters of displacement (Figure 2). The friction coefficient was the highest at the beginning of sliding at slow slip rates. The unstable behavior for both samples was observed at slip velocities higher than several hundreds of millimeters per second. An initial large peak friction was observed after several hundred millimeters slip displacement, and abrupt fluctuations in the friction coefficient were observed. If we exclude unstable mechanical data in our friction experiments, the friction data can be fitted by the exponential decay curve [Mizoguchi et al., 2007, equation (1)], which fits well for gouge friction experiment at the high slip rate. The steady state friction coefficient of Berea sandstone and Indian sandstone samples decreased with increasing slip rate at low slip rates, and the initial peak friction increased with increasing slip rate (Table 2). Both the peak friction and the steady state friction of the Berea sandstone were lower than those of the Indian sandstone. Of the four rocks tested, the frictional behavior of Aji granite samples was the most unstable, even at low slip rates ( m/s). The fluctuations of the friction coefficient were more abrupt and larger for high slip rates for all rocks tested except Inada granite Post Shear Test Permeability Measurements [16] The permeability in all samples decreased with increasing effective pressure, and the rate of decrease slowed as the effective pressure increased (Figure 3). The initial permeability of intact Berea sandstone at the lowest effective pressure of 5 MPa was m 2, and this decreased by more than 1 order of magnitude (to m 2 ) at an effective pressure of 120 MPa (Figure 3a). The permeability of all Berea sandstone specimens after friction tests was lower than that of the intact rock, though the rate of permeability decrease during pressurization was similar for all Berea sandstone specimens at all slip rates. [17] Posttest permeability in the Berea sandstone decreased gradually with increasing slip velocity (Figure 4a), and the permeability became stable at high slip rates if we select the lowest values (even though there is a highly variable in permeability at high slip rates). Permeability was reduced by more than 1 order of magnitude after high velocity friction. The permeabilities of intact Indian sandstone ( m 2 at 120 MPa) and intact Aji granite ( m 2 at 120 MPa) were much lower than that of Berea sandstone at similar pressures (Figures 3b and 3c). The rates of decrease in permeability for intact specimens of Indian sandstone and Aji granite were lower than that of Berea sandstone. At slow slip rates (<100 mm/s), permeability in both rocks was not changed by shear deformation, though the permeability increased markedly at high slip rates, above 390 mm/s (Figures 4b and 4c). Permeability of Indian sandstone after high velocity friction increased by a factor of about 50 from that of preshear intact rock, whereas for Aji granite the permeability increased by 3 orders of magnitude after highvelocity shear. [18] Permeability was highly sensitive to effective pressure for Indian sandstone and Aji granite specimens after high velocity friction, and at 120 MPa of effective pressure the permeability decreased by 2 orders and 3 orders of magnitude for Indian sandstone and Aji granite, respectively (Figures 3b and 3c). [19] After shear displacement at high slip rates (1300 and 860 mm/s in Berea sandstone), the permeability in Berea sandstone decreased by about 1 order of magnitude after 2 m of slip displacement, with little further change after 3 m of displacement (Figure 5a). An increase in permeability was observed for Berea sandstone specimens after 8 m of slip displacement. Permeability of Indian sandstone decreased slightly within the first 5 m of slip displacement and then increased rapidly as slip displacement approached 5 m (Figure 5b). A marked increase in permeability was observed for Aji granite at shear displacements of 4 and 8 m, though for the two sandstones, only slight changes in permeability were observed at these displacements (Figure 5c). 4. Resulting Microstructures [20] Microsection analyses of samples after friction tests revealed that a gouge zone (GZ) of much finer grain size than the host rock was formed at the slip surface for all samples (Figure 6). In Berea sandstone, shear deformation was localized along a narrow band where the grain size of the gouge was finest (principal slip zone (PSZ) in Figures 6a and 6b). The localized shear was developed at the center of the gouge layer after slow slip friction (Figure 6b), whereas it was at the contact between the host rock and the GZ after high velocity friction tests (Figure 6c). The width of the GZ increased with increasing slip velocity (Figures 6a 6c), though the width of the PSZ was greater for a slip velocity 4of18

5 Figure 2. An example of the variation of the coefficient of friction with slip displacement for various slip rates between 0.13 and 1300 mm/s: (a) Berea sandstone, (b) Indian sandstone, (c) Aji granite, and (d) Inada granite. 5of18

6 Table 2. Summary of All Rock Rock Friction Experiments a Run number Rock type Rotation speed (rpm) Velocity (mm/s) Total disp. (m) Normal stress (MPa) Peak friction Steady state friction Testing machine (HVR/PHV) Fractal dimension PHV035 Berea Ss PHV PHV034 Berea Ss PHV PHV033 Berea Ss PHV PHV059 Berea Ss PHV PHV073 Berea Ss PHV PHV036 Berea Ss PHV PHV060 Berea Ss PHV PHV074 Berea Ss PHV PHV037 Berea Ss PHV PHV038 Berea Ss PHV PHV039 Berea Ss PHV PHV040 Berea Ss PHV HVR1577 Berea Ss HVR HVR1578 Berea Ss HVR HVR1575 Berea Ss HVR HVR1703 Berea Ss HVR HVR1706 Berea Ss HVR HVR1707 Berea Ss HVR PHV030 Indian Ss PHV PHV058 Indian Ss PHV PHV072 Indian Ss PHV PHV028 Indian Ss PHV PHV027 Indian Ss PHV PHV054 Indian Ss PHV PHV055 Indian Ss PHV PHV026 Indian Ss PHV PHV029 Indian Ss PHV PHV042 Indian Ss PHV PHV068 Indian Ss PHV PHV069 Indian Ss PHV HVR1568 Indian Ss HVR HVR1570 Indian Ss HVR HVR1651 Indian Ss HVR HVR1652 Indian Ss HVR HVR1653 Indian Ss HVR HVR1656 Indian Ss HVR HVR1689 Indian Ss HVR HVR1695 Indian Ss HVR HVR1696 Indian Ss HVR HVR1697 Indian Ss HVR PHV043 Aji granite PHV PHV045 Aji granite PHV PHV046 Aji granite PHV PHV047 Aji granite PHV PHV044 Aji granite PHV PHV048 Aji granite PHV PHV052 Aji granite PHV PHV053 Aji granite PHV PHV067 Aji granite PHV HVR1569 Aji granite HVR HVR1571 Aji granite HVR HVR1649 Aji granite HVR HVR1650 Aji granite HVR HVR1654 Aji granite HVR HVR1655 Aji granite HVR HVR1657 Aji granite HVR HVR1658 Aji granite HVR HVR1659 Aji granite HVR HVR1691 Aji granite HVR HVR1693 Aji granite HVR HVR1694 Aji granite HVR HVR1701 Aji granite HVR HVR1299 Inada granite HVR HVR1302 Inada granite HVR a Values of the initial peak friction and steady state friction were estimated using the exponential equation of Mizoguchi et al. [2007]. Fractal dimension was measured from photomicrographs [Hirose and Shimamoto, 2003]. HVR, high speed rotary shear testing apparatus of Shimamoto and Tsutsumi [1994]; PHV, rotary shear testing apparatus with a wide range of speeds at the Kochi Core Center [Tadai et al., 2009]; Ss, sandstone. 6of18

7 Figure 3. Permeability as a function of effective pressure for specimens of (a) Berea sandstone, (b) Indian sandstone, (c) Aji granite, and (d) Inada granite before testing and after friction tests at various slip rates. Data from the pressurization path are plotted. of 130 mm/s than for one of 830 mm/s. After high velocity friction tests, specimens contained large, matrix supported, angular to subangular fragments within the gouge layers, and inter and intragranular cracks developed in a narrow fracture zone (FZ; <1 mm wide) parallel to the slip surface (FZ in Figure 6c). In most of our experiments the FZ developed only on the stationary side of the cylindrical specimen. The surface roughness of test specimens, which can be quantitatively expressed by the fractal dimension [Hirose and Shimamoto, 2003], was measured from photomicrographs (Table 2). The slip surface was very smooth after slow slip rate tests but very rough for high slip rates. For high velocity friction tests (>390 mm/s), mesoscale fractures were generated within the host rock and propagated up to 10 mm beyond the GZ (Figures 6d and 6g). The mesoscale fractures were considerably wider than the interand intragranular cracks in the FZ. Fine grained gouge and small rock fragments, probably formed by grain crushing at low displacement, partially filled the mesoscale fractures (Figure 6g). In some places we found that gouge material had been injected from the GZ into cracks within the host rocks (Figures 6e, 6f, 6h, and 6i). [21] We measured the thickness of the fine grained gouge layers (GZ plus PSZ in Figures 6b and 6c) from images captured using a microfocus X ray computed tomography (CT) system (HMX225 ACTIS+3; Tesco Co. Ltd., Tokyo) and from photomicrographs. The area captured in the images is shown in Figure 7b; the average width of the image 7of18

8 Figure 4. Permeability after friction tests as a function of slip velocity for (a) Berea sandstone, (b) Indian sandstone, and (c) Aji granite for effective pressures between 10 and 100 MPa. Figure 5. Permeability after friction tests (slip velocity, 1300 mm/s) as a function of slip displacement for (a) Berea sandstone, (b) Indian sandstone, and (c) Aji granite for effective pressures between 10 and 100 MPa. Permeabilities for a slip velocity of 860 mm/s are also plotted in Figure 5a. 8of18

9 Figure 6 9of18

10 slice from the CT system was 10 mm. Gouge layers for all specimens showed relatively low CT amplitudes (Figure 7a), so we identified the gouge layer from the CT contrast. There was little difference between the gouge thicknesses measured from CT images and those measured from photomicrographs (Figure 7c). Because some of the gouge products were pushed to the edge of the slip surface during testing, we may have underestimated gouge thicknesses. For all rocks tested the gouge thickness changed little as the slip rate increased up to 10 mm/s, but it increased steadily with increasing slip rate beyond 10 mm/s (Figure 7c). The gouge thickness reached nearly 1 mm at the highest slip rate, and the rate of wear (the volume production rate of gouge per unit of slip distance) increased to nearly 1 order of magnitude higher than that of slow slip friction. Figure 7d shows the relationship of gouge thickness to slip displacement at a slip rate of 1300 mm/s. Because we varied normal stress from 1 to 2 MPa in the high velocity friction tests, and because the rate of wear of gouge material is proportional to both normal stress and slip displacement [Scholz, 1987], we plotted gouge thickness as a function of normal stress multiplied by slip displacement (Figure 7d). Thus, we showed that gouge thickness was proportional to slip displacement multiplied by normal stress for Indian sandstone and Aji granite and that the rate of wear was higher for Aji granite than for Indian sandstone at all slip rates (Figures 7c and 7d). 5. Discussion 5.1. Development of Gouge and Changes in Permeability [22] We carried out friction tests and post shear test permeability tests on four rock types and identified two different processes by which permeability changed as a result of shear deformation. In Berea sandstone specimens the permeability decreased with increasing slip rate when displacement and normal stress were unchanged. On the contrary, the permeability of Indian sandstone and Aji granite specimens was unchanged by slow slip friction but increased abruptly at high slip rates, in particular, at slip rates faster than several hundred millimeters per second. [23] Microstructural analyses of sheared samples showed that the permeability reduction in Berea sandstone specimens was caused by the development of a narrow, finegrained gouge layer in the slip zone. The flow direction for our permeability measurement was perpendicular to the gouge layer; therefore, if the permeability of the gouge layer was lower than that of the intact rock, the bulk permeability (permeability of the specimen including the gouge layer) would decrease. We estimated the permeability of the gouge layer by the following equation, assuming that the permeability of the host rock was not changed by the friction tests: 1 ¼ n þ 1 n ; ð2þ k b k g k h where k b is the laboratory measured bulk permeability, k g is the permeability of the gouge layer, k h is the permeability of the host rock, and n is the ratio of the gouge thickness to the total thickness of the specimen. The permeability of the gouge layer in Berea sandstone as estimated from equation (2) was 3 orders of magnitude lower than that of the host rock (Figure 8). The permeabilities we estimated for the gouge layers in Indian sandstone and Aji granite were 1 to 2 orders of magnitude lower than those of the host rocks. There was no clear relationship between slip velocity and gouge permeability for any of the rocks tested (Figure 8), which suggests that the reduction in permeability with increasing slip velocity that we observed in Berea sandstone specimens can be explained by the increase in thickness of the gouge layer with increasing slip velocity. However, our assumption that the permeability of the host rock was unchanged after friction tests means that gouge permeability is not evaluated with equation (2) when k b is much larger than k h. Mesoscale fractures and intra and intergranular cracks were clearly developed within the host rocks after tests at slip rates faster than 130 mm/s for all of the rocks we tested, so the permeability of the host rocks must have increased as a result of our tests. Therefore, the estimated gouge permeability in Figure 8 is a possible maximum value. Our microstructural analyses suggest that the additional fractures created at high slip rates were caused by frictional heating and provided a marked increase in permeability Change in Permeability With Temperature [24] Elevated temperatures on the slip surface were evaluated by numerical modeling using laboratory data. The model settings and method we used for the temperature calculation were based on those of Mizoguchi et al. [2009]; however, we believe that the boundary conditions we used in our model are more realistic (see the Appendix). The thermal properties we used for our numerical model are reported in Table 3. Thermal conductivity of the specimens was measured using the transient hot wire method with a commercial thermal conductivity meter (QTM 500; Kyoto Electronics Manufacturing Co., Ltd., Tokyo) [Tadai et al., 2009b] at room temperature and atmospheric pressure. Although these properties are temperature dependent, for our numerical modeling we assumed that they were not. We used our model to calculate the average temperature on the slip surface immediately after slip (Figure 9a). At slow slip Figure 6. (a c, e j) Photomicrographs under plane polarized light and (d) one computed tomography (CT) image of samples after friction tests. (a d) Berea sandstone, (e g) Indian sandstone, and (h j) Aji granite. Two pairs of images, (c/e) and (e/g), are of the same specimens. Slip velocities for the specimens shown were (a) 13 mm/s (PHV073), (b) 130 mm/s (PHV074), (c, d) 830 mm/s (HVR1578), (e g) 1300 mm/s (HVR1696), (h, i) 390 mm/s (PHV053), and (j) 1300 mm/s (HVR1655). (f, i) Closeup images of injection of gouge materials into fractures for Figures 6e and 6h, respectively. (g) Closeup of a mesoscale fracture developed within a stationary half specimen of Indian sandstone. The shear sense for all images is top to the right, and the lower block is the one that was rotated during testing. The area captured for the CT image is shown in Figure 7b. All images are of slices perpendicular to the slip plane. fr, fracture; FZ, fracture zone; GZ, gouge zone; in, injection of gouge materials into fractures; PSZ, principal slip zone. 10 of 18

11 rates (<10 mm/s), temperature increases were small and did not reach 100 C. At high slip rates, temperatures rose to more than 500 C for three of the rocks we modeled. The variation in average temperature with distance from the slip Figure 8. Relative permeability of the gouge layer (K gouge / K host ) based on the permeability of the host rock at 10 MPa of effective pressure. Permeability of the gouge layer was estimated with equation (2), assuming that the host rock permeability was not changed by friction testing. surface after 2 m of slip displacement (Figure 9b) shows that the thermal gradient near the slip surface increased with increasing slip velocity. For a slip velocity of 1300 mm/s the temperature 10 mm from the slip surface remained at room temperature, while the temperature at the slip surface increased by several hundred degrees. [25] There are three mechanisms that can explain the development of thermally induced cracks in rocks during friction experiments. [26] 1. Thermally induced stress heterogeneities caused by differences in the thermal expansion properties of the constituent minerals of a rock can cause cracks to open. Similarly, changes in the thermal expansion coefficients of those minerals owing to phase transitions can also cause cracks to open. [27] 2. For monomineralic rocks the shock thermal stress caused by sudden heating produces a strong temperature gradient near the heat source, which in turn causes strain heterogeneity. [28] 3. Heating reduces the strength of rocks. [29] Thermally induced cracks have been generated successfully in laboratory experiments [Richter and Simmons, 1974; Wang et al., 1989; Lin, 2002]. Wang et al. [1989] 11 of 18 Figure 7. (a) An example of a CT image (Berea sandstone; PHV039) and (b) the area captured in CT images. (c) Gouge thickness versus slip velocity for 8 m of slip displacement for various rocks. (d) Gouge thickness versus slip distance multiplied by normal stress of from 1 to 2 MPa at a slip rate of 1.3 m/s for various rocks. In the keys to Figures 7c and 7d, T and C indicate gouge thicknesses measured from photomicrographs and CT images, respectively. The linear relationship between slip velocity and gouge thickness for Aji granite is shown in Figure 7c, determined by a least squares fit.

12 Table 3. Physical, Thermal, and Frictional Properties Used for Thermal Pressurization Analysis Unit Berea Sandstone Indian Sandstone Aji Granite Porosity % Matrix density kg m Initial permeability at 10 MPa m E E E 20 Specific storage Pa 1 5.E E E 11 Thermal conductivity Grain matrix W m 1 K Water W m 1 K Specific heat Grain matrix J kg 1 K Water J kg 1 K Coefficient of thermal expansion Fluid C Grain matrix C Viscosity of water Pa s /(T 140) Depth Km Friction coefficient Width of deformation zone M Slip velocity (constant) m s reported that crack density is proportional to maximum temperature in simple heating tests and that this result is caused by the thermal stress generated by mismatches of thermal expansion owing to different mineral compositions. Lin [2002] observed dramatic widening of microcracks at about 600 C, which is consistent with the quartz a b transition temperature (573 C at 0.1 MPa) [Coe and Paterson, 1969; Van der Molen, 1981]. Richter and Simmons [1974] found that thermally induced cracking was enhanced with increases in the rate of heating, which is supported by case 2 above. [30] Way et al. [1982] undertook numerical analysis of thermal stress using the finite element method. In their model a steep temperature gradient produces very high compressive stresses within a narrow region along the slip surface, with a comparatively tensile zone farther out from the slip surface. Both the compressive and the tensile strength of rocks decrease at high temperatures [Heuze, 1983] and the rate of decrease increases at temperatures above 400 C for granites [Dwivedi et al., 2008]. This suggests that the reduction in rock strength caused by shearinduced heating contributes to rock failure. Thermal stress also depends on the thermal properties of rocks. More heat is conducted in rocks with a high diffusivity; therefore, temperature gradients are smaller and thermal stresses are lower in rocks with a high diffusivity than in those of low diffusivity. However, the effect of differences in thermal properties on the generation of thermal stress is less than the effect of differences in heating rate. [31] Increases in permeability caused by thermal enhancement of cracks have been observed in simple heating tests and modeled on the basis of measured mechanical properties. After heating an almost impermeable granite to 300 C at various lithostatic pressures, thermal expansion increased the permeability by factors of between 2 and 5 [Heard, 1980; Heard and Page, 1982]. A dramatic increase in permeability in granite was observed at temperatures above 500 C by Darot et al. [1992] (Figure 10), which was probably caused by the a b transition of quartz. The main constituent of the specimens we used was quartz. Because the temperature at the slip interface exceeded the quartz phasetransition temperature at high slip velocities (Figure 9a), the increase in permeability for the Aji granite was likely caused by the combined effect of the thermal enhancement of cracks when the temperature exceeded the a b transition temperature and the high thermal gradient. High velocity friction in Aji granite did not increase the bulk permeability in all cases (Figure 5c), which may be because cracks and fractures were intruded by gouge material (Figure 6i). Thermally induced cracks were also developed in the Berea sandstone after our high velocity friction tests, although a reduction in permeability was observed in this case. Because the permeability of intact Berea sandstone is very high and dominated by porous flow, fracture permeability can be ignored (provided that the fractures are very fine). The filling of fractures with gouge material also counteracts increases in permeability. [32] Comparison of initial permeability with permeability after high velocity friction tests at 10 MPa of effective pressure shows that a reduction in permeability by shear friction occurred in permeable rocks with an initial permeability highr than m 2 (Figure 10). In contrast, a relative increase in permeability occurred in most specimens with an initial permeability lower than m 2. These results suggest that for sedimentary rocks or cohesive fault rocks (cataclasite), an initial permeability of about m 2 marks the boundary between the permeability decrease attributed to intrusion of gouge into cracks and the increases attributed to the development or enhancement of fractures. [33] The relationship between prefailure permeability and initial porosity measured in triaxial compression experiments was demonstrated for a number of rock types by Zhu and Wong [1997]. In their experiments the transition from a reduction in permeability to an increase occurred for rocks of 15% initial porosity. Those results are consistent with ours because, as shown in Figure 10, we found that the transition lies between the 20% initial porosity of the Berea sandstone and the 11% 14% initial porosity of the Indian sandstone. The trend that high porosity rock gives a lower permeability fault and lower porosity rock gives a high permeability fault for high velocity friction is similar to that for deformations at a slow slip rate as well [Fossen et al., 2007]. 12 of 18

13 of our friction tests, however, showed that gouge thickness is proportional to slip velocity at high slip rates. This trend is consistent with the previous laboratory data of Hirose and Mizoguchi [2008] that a shortening rate of axial length of specimen is proportional to slip rate under unconfined condition. Gouge thickness is inversely proportional to rock hardness in the wear model of Scholz [1987], so a decrease in rock hardness caused by frictional heating explains the dependence of wear rate on slip velocity. In addition, the nucleation of inter and intragranular cracks during thermal fracturing along the slip surface introduces large fragments to the GZ and thus thickens it. Furthermore, the thermal cracking will enhance plucking of grains from the side of the slip zone, and small scale slip on thermal cracks will create new roughness. The increase in roughness can lead to increase in wear rate as well [Power et al., 1988]. Scholz [1987] pointed out that natural faults, because of their greater roughness, have wear rates higher than those observed in laboratory experiments. However, the experiments he referred to were conducted at slow slip rates [Yoshioka, 1986]; the inconsistency with our results is explained by the reduction in hardness, thermal enhancement of cracks, and introduction of new roughness induced by the thermal cracks at high slip rates in our laboratory tests Evaluation of Our Results and Future Work [35] Examination of the processes of permeability change can help us to understand the behavior of temporal hydromechanical coupling in faults. Dynamic fault weakening during earthquakes has been associated with increasing pore pressure as a result of frictional heating in low permeability Figure 9. (a) Slip velocity versus temperature at the slip surface after 8 m of slip displacement estimated by numerical modeling (Appendix) using measured physical and thermal properties listed in Table 2. (b) Average temperature distribution as a function of the distance from the slip surface after 2 m of slip displacement for various slip rates Wear on the Slip Plane and Gouge Formation [34] A simple wear model that accounts for the thickness of gouge layers showed that gouge thickness is not directly related to slip velocity for a given normal stress and slip displacement [Scholz, 1987]; this is supported by observations of natural faults [Robertson, 1983; Evans, 1990]. All Figure 10. Permeability changes after high velocity friction (HVF) at a slip velocity of 1.3 m/s as a function of initial permeability at 10 MPa of effective pressure. Data are also shown for argon permeability derived from a simple heating test of the La Peyratte granite [Darot et al., 1992] under isostatic confining pressure. 13 of 18

14 Figure 11. Results of numerical modeling of thermal pressurization for three rock types. (a) Permeability models for thermal pressurization analysis. Estimated changes in (b) pore pressure increase, (c) normalized friction, and (d) temperature rise at the center of the slip zone from the start of slip. We modeled two permeability models: c, constant permeability model; and s, shear induced permeability model developed from the results of HVF tests in Figure 10. Ss, sandstone. fault zones [e.g., Sibson, 1973]. However, our results indicate that permeability in low permeability fault zones increases dramatically during high velocity slip because of nucleation of thermally induced fractures along the slip surface. Figure 11 shows the numerical results of thermal pressurization within fault zones for three rock types. We used a one dimensional heat and fluid dynamic flow model based on that of Andrews [2002] to estimate dynamic temperature and pore pressure changes caused by frictional motion. The physical parameters used for the model are reported in Table 4; they are mostly based on Tanikawa et al. [2009]. We considered two permeability models in our numerical analysis (Figure 11a). These are (a) that the permeability of the fault zone is constant during slip and (b) that the permeability of the fault zone is changed by high velocity shear deformation. The shear induced permeability change model is based on our laboratory results shown in Figure 5, and we assumed that the permeability of the fault zone in the model 14 of 18

15 Table 4. Thermal and Physical Properties Used for Numerical Modeling of the Temperature Distribution During Friction Tests Material Thermal Conductivity (W/mK) Specific Heat (J/kg K) Density (kg/m 3 ) Thermal Diffusivity (m 2 /s) Heat Transfer Coefficient at Boundary A Between Material and Air (W/m 2 K 1 ) Berea sandstone E Indian sandstone E Aji granite E Brass E Teflon E was consistent with the laboratory measured bulk permeability. The modeled results of pore pressure rise for Indian sandstone and Aji granite show sudden pore pressure generation during the first second of slip (Figure 11b) in both models, and the elevated pore pressure is maintained until the end of slip in the constant permeability model. On the contrary, the pore pressure decreases immediately after a sudden permeability increase caused by formation of thermally induced cracks in the shear induced permeability model. Sudden recovery of fault strength by pore pressure dissipation with rapid diffusion of fluid (Figure 11c) and enhancement of temperature rise (Figure 11d) is expected as well. This sudden increase in friction may act as a brake on slip movement, and a smaller total displacement is expected. In contrast with impermeable rocks, the modeled results for Berea sandstone show very slight differences between models. Our results indicate that thermal pressurization may not always occur within impermeable fault zones. The sudden increase in bulk permeability caused by dynamic slip in our tests is consistent with hydrogeochemical changes reported by Claesson et al. [2004]. This indicates that temporal large scale hydrological change is controlled by changes in bulk permeability (permeability of the damaged zone in a natural fault), which is mainly dominated by mesoscale ( macroscale) fracture structures, rather than the permeability of the shear localized zone (fault core) dominated by the fine grained impermeable layer. [36] Poroelastic properties such as storage capacity are also important hydraulic parameters that affect fluid flow systems at depth [e.g., Wang, 2000]. If cracks open during shear induced friction, the porosity in the fault zone increases and the poroelastic parameters in the fault zone change dramatically. Therefore, some knowledge of these hydraulic parameters is necessary to model fault dynamics and fluid flow effectively. We started our friction experiments with no wear material at the slip interface; however, because of recurrent fault activity, thick, fine grained fault gouge, cataclasite, and mesoscale fractures are commonly found within natural fault zones. Therefore, friction tests performed on rock samples from a fault zone, or presheared specimens, may provide a more realistic representation of the dynamic behavior of permeability in a fault zone. [37] Because the normal stresses at depth are greater than those in our laboratory experiments, greater heat generation would be expected in a natural fault. If temperatures in a fault plane exceed the melting point of minerals in the host rock, melt products or pseudotachylite is generated at the slip surface, rather than gouge. In nature and experiments, frictionally generated melt escapes the slip plane into fractures within host rocks [Di Toro et al., 2005; Tsutsumi and Mizoguchi, 2007], and molten material such as these have a very low permeability. Therefore, even if thermally induced fractures are developed within host rocks, the pseudotachylite bearing fault zone would act as a barrier to fluid flow. Or if pseudotachylite bearing faults are reworked in the seismic cycle, these rocks could be brittle and act in a contradictory fashion [Kirkpatrick et al., 2008]. [38] Our results may be influenced by the small size of the specimens we tested, as the distance that fractures propagate from the slip plane in laboratory tests may be affected by sample size. Fracture connectivity, distribution pattern of fractures, and crack length influence fracture permeability. Therefore, we need to normalize fracture connectivity and fracture length if we are to apply laboratory data to practical problems in future studies. Permeability anisotropy must also be investigated [Faulkner and Rutter, 1998;Zhang et al., 2001; Wibberley, 2002]. [39] We have shown that the formation of thermally induced cracks and gouge material during shear slip affects mechanical friction (Figure 2). The results of our friction tests imply that the formation of a gouge layer decreases friction. However, formation of fractures in the rock adjoining the slip plane enables frictionally generated gouge to escape the fault plane into fractures within the host rock. The gouge squeezing process may lead to rock rock contacts between the opposing fault surfaces during sliding that lead to unstable frictional behavior at high slip rates. Contact of fault surfaces with large angular to subangular fragments produced by high velocity friction may cause the unstable friction as well. 6. Summary [40] Changes in the permeability caused by friction in faults vary with slip rate, slip displacement, and rock type. The permeability of deformed Berea sandstone specimens decreased gradually with increasing slip rate and slip displacement. The permeability of the less permeable Indian sandstone and Aji granite specimens was increased abruptly by several meters of displacement at high slip rates (>0.3 m/s). Mesoscale fractures and inter and intragranular cracks developed within specimens at high slip rates. Finegrained gouge layers with an apparent low permeability developed along the slip surface in many of our tests. In addition, we found that the wear rate of gouge was proportional to the slip rate, probably owing to weakening of the host rock caused by thermal cracking. 15 of 18

16 Figure A1. (a) Illustration of the area used for modeling of the temperature distribution within specimens during friction tests shown in Figure 9. (b) Enlarged view of the area modeled showing detailed mesh geometry and boundaries used. Heat flux at boundary A follows Newton s law of cooling (equation A1), and frictional heat flux occurs across boundary B. (c) Snapshot of the modeled thermal distribution for an Aji granite specimen at 8 m of displacement for a slip rate of 390 mm/s. [41] A reduction in permeability was caused by generation of impermeable gouge layers (shear localization zone), whereas drastic permeability increases were caused by the thermal enhancement of cracks. Increases in gouge thickness and thermal enhancement of cracks were associated with frictional heating at high slip rates. Steep thermal gradients and large temperature increases beyond the a b transition temperature near the slip surface also enhanced thermally induced cracking by increasing the thermal stress. [42] Our results suggest that permeable rocks become relatively impermeable after frictional sliding, whereas the permeability of less permeable rocks increases abruptly at the high velocities characteristic of seismic events. The dramatic increases in permeability we observed are consistent with temporal hydrogeological changes commonly observed after seismic events. Appendix A: Calculation of Temperature Increases During Friction Tests [43] We used numerical modeling to calculate the temperature distribution in specimens during frictional sliding. We assumed an axisymmetric two dimensional problem (Figure A1a) and used a computer program based on the finite element method developed by Kuroda [2001]. In our modeling the half specimen on one side of the slip plane was represented by 1600 elements, each mm (Figure A1b). The time step we used was such that tem- 16 of 18