Anisotropic transport and frictional properties of simulated clay- rich fault gouges

We aimed to evaluate various factors that control the frictional and transport properties of gouge-filled faults cutting carbonate-bearing shales or claystone formations. The research experimentally determined the effect of shear displacement, dynamic shearing, static holding, and effective normal stress on fault gouge permeability, both parallel and perpendicular to 10 the fault boundaries, as well as on frictional behaviour. The simulated gouge was prepared from crushed Opalinus Claystone (OPA), on which we performed direct shear experiments. The direct-shear experiments (σn = 5 50 MPa, Pf = 2 MPa, and T ≈ 20°C) showed ~1 order of magnitude decrease in permeability with shear displacement (up to ~6 mm), for both alongand across-fault fluid flow orientation. Moreover, our data showed an initial, pre-shear permeability anisotropy of up to ~1 order of magnitude, which decreased with increasing shear displacement (maturity) to ~0.5, with the along-fault permeability 15 being consistently higher. Our results have important implications for calcite-rich claystones and shale formations, and in particular any pre-existing faults therein, that seal hydrocarbon reservoirs and potential CO2 storage reservoirs, as the current results point to a higher leakage potential of pre-existing faults compared to the intact caprock.


Introduction
In ambition to lower anthropogenic CO2 emissions, Carbon Capture and Storage (CCS), the capture of CO2 at large point 20 sources followed by injection into geological formations, e.g. hydrocarbon reservoirs or aquifers, is one of the remaining potential methods to mitigate anthropogenic CO2 emissions on a relatively large scale. However, CCS is only a viable option when the long-term sealing integrity of the storage complex, i.e. the reservoir and overlying caprock, can be maintained.

Deformation apparatus
The direct shear experiments described in this study were performed using a conventional triaxial deformation apparatus, comprising a pressure-compensated main vessel linked to an auxiliary vessel (Figure 1; (Samuelson and Spiers, 2012). The externally heated, oil-filled main vessel is equipped with a specially-designed direct shear assembly, which can be loaded via a yoke/piston combination, driven by a motor/gearbox/screwball system, to apply an axial load to the sample assembly, 100 while simultaneously keeping the system pressure nominally-constant during deformation (resolution ± 0.20 μl). During an experiment, the confining pressure, the axial load, the displacement of the advancing yoke/piston and sample temperature are measured using respectively a Jensen pressure transducer (100 MPa range, resolution ± 0.02 MPa), a DRVT-based, semiinternal load cell (DVRT = Differential Variable Reluctance Transformer; 400 kN range, resolution ± 0.035 kN), a high precision LVDT (Linear Variable Differential Transformer; 100 mm, ± 0.8 μm) and two K-type (chromel/alumel) 105 thermocouples (400°C, accuracy ± 0.1 °C, precision ± 0.02 °C). Sliding velocities can be changed step-wise and near instantaneously, for velocities ranging between ~0.05 and 50 μm/s. The machine is described in detail by Peach and Spiers (1996) and Hangx et al. (2010aHangx et al. ( , 2010b. Pore fluids can be introduced into the gouge layer via two stainless steel tubes, mounted in the load cell block, which are connected to inlets at the top and bottom end pistons inclosing the sample assembly ( Figure 1). We used argon gas (99.9% pure) as pore fluid. 110 The argon permeability (κ) of simulated fault gouge was measured using the permeameter previously described by Peach (1991) and Bakker et al. (2017) (Figure 2). The set-up consists of two symmetrical, low-volume tube/valve systems, located at both the up and downstream end of the sample assembly, which can be independently pressurized up to 2 MPa, as such allowing the application of a pressure difference across the gouge layer. The system can be evacuated via a vacuum pump connected to a vent valve, whereas argon, from a nearby bottle and regulator, is supplied via a second valve. To avoid argon 115 flow between the sample assembly and the jacket, the applied confining pressure was always maintained higher than the argon pressure and the gouge layer-piston-jacket contact was taped with Teflon tape. To measure the argon permeability, we logged the pressure decay across the sample following an imposed pressure difference, using two temperature-compensated pressure transducers, respectively up-and downstream (2 MPa range, resolution ± 0.001 kPa) (Sutherland and Cave, 1980). Note, that a very small background leakage of the argon permeameter prevented accurate measurements of samples with 120 permeabilities of less than 10 -21 m 2 (Hangx et al., 2010a).

Sample assembly
For the direct shear experiments described above we used two types of specially designed direct shear assemblies. Both direct shear assemblies consist of two "L-shaped", direct shear blocks (Figure 3), constituting a 35 mm diameter cylinder when assembled. Note that given the configuration of the two inverted direct shear blocks, for both sets of direct shear blocks, the applied normal stress (σn) is equal to the confining pressure (Pc) and independent of the shear displacement, at all times during deformation. Each of the two direct shear blocks is equipped with a grooved (60 μm deep, 120 μm wide and 200 μm apart) shear surface (47 mm long and 35 mm wide - Figure 3). To allow for uniform pore fluid access to the gouge layer, via pore fluid channels connected to the tubes in the load cell block, strategically placed porous stainless-steel plates (permeability >~10 -14 m 2 ) were placed in the shear surface. The difference between the two sets of direct shear blocks is the 130 location of the porous frits and is described in more detail below.
In preparing each experiment, 3.7 g of loose simulated fault gouge was evenly distributed on the shear interface of the bottom direct shear block and pre-pressed, using a hydraulic press by loading the gouge layer two times for ~30 seconds (~60 seconds in total), at orientations 180° apart (rotation in the horizontal plane). The pre-pressing normal stress was chosen to be less than 6 MPa, to prevent for overcompaction. The preparation method resulted in coherent, reproducible gouge 135 layers. The bottom block, with the gouge layer (dimensions: 49 mm x 35 mm x ~1 mmcf. Figure 3) on top, was then covered with the upper shear block, leaving 10-mm gaps at both ends of the gouge layer. The two shear blocks, forming a full cylinder and sandwiching the gouge layer, were then fixated using a heat-shrinking Fluorinated Ethylene Propylene (FEP) inner sleeve, while the 10-mm gaps were filled with 50 μm Teflon foil-wrapped Ecoflex (2-component epoxy) plugs.
To avoid argon flow between the sample assembly and the FEP jacket, rather than through the gouge layer, the FEP-gouge 140 layer-pistons contact was taped off with gas tape. The whole was then placed in a 1.4 mm think Ethylene Propylene Diene Monomer (EPDM) outer jacket, which is sealed against the upper and lower driver blocks using wire tourniquets. The soft Ecoflex plugs, and the FEP and EPDM sleeves are assumed to exert negligible resistance during deformation (Samuelson and Spiers, 2012). Upon jacketing, the sample assembly was attached to the loading frame, emplaced in the main pressure vessel and evacuated at a Pc of ~5 MPa, for up to ~60 minutes. After evacuation, the appropriate pore fluid was introduced. 145 Although the direct shear assemblies are similar for the two different types of experiment (along-vs. across-fault), the details of the internal pore fluid systems in the direct shear blocks are slightly different. The direct shear assembly used to measure the across-fault permeability, and frictional strength of the OPA gouge, consist of two stainless steel direct shear blocks ( Figure 3a), each equipped with a grooved (60 μm deep, 120 μm wide and 200 μm apart) and porous stainless steel plate (frit) (effective shear surface: 47 mm long and 35 mm wide and permeability >~10 -14 m 2 - Figure 3d). Uniform pore fluid 150 access to the entire gouge layer is allowed via multiple internal pore fluid channels in the shear blocks ( Figure 3a). We will refer to these direct shear blocks as the "across-fault permeability blocks". The along-fault permeability, and frictional strength of the OPA gouge, is measured with the "along-fault permeability blocks", which consists of two titanium shear blocks, each with a grooved (60 μm deep, 120 μm wide and 200 μm apart - Figure 3b) shearing interface. Only 2 mm of the shearing interface, at the base of each shearing interface, is porous and permeable (permeability ~10 -14 m 2frit), allowing 155 pore fluid access to the sample layer from the upper and lower extent of the gouge layer, when assembled. This configuration restricts fluid flow from one end of the gouge layer to the other end through the long axis of the sandwiched gouge layer.

Experimental procedure
For each experiment, evacuation was followed by increasing the Pc to slightly below the pre-determined value, to increase Pf to 2 MPa (using 99.9% pure argon), after which Pc was increased further to the targeted value. After applying the pre-160 determined Pc and Pf-conditions, the system was allowed to equilibrate for ~3h. Subsequently, the downstream part of the permeametry set-up was isolated from the sample and the upstream part (using values "A" and "B"see Figure 2), to lower the downstream reservoir by 0.2 MPa with respect to the upstream part. After stabilisation, the downstream part was reconnected to the upstream part via valve "B", to allow pore fluid flow from up to downstream, which was measured and logged as described above. This procedure was repeated for each permeability measurement. 165 Permeability measurements performed during the experiments were conducted prior to shear, during shear at a sliding velocity of 0.05 μm/s and after a period of hold (referred to as Slide-Hold-Slide (SHS) experiments). The SHS sequence was employed to investigate the effect of static holding versus dynamic shearing on the evolution of fluid flow in a clay-rich gouge layer. The amount of sliding during a permeability measurement depended on the time required to reach full argon decay. In case full decay was reached within < 2h, (for relatively "high" permeability samples or short path lengths), sliding 170 was continued until full decay was reached, however, if full decay was reached in > 12h ("low" permeability samples or long path lengths), the permeability measurement and sliding were terminated after 4.5h of decay.
A regular slide-hold-slide permeability experiment was typically initiated with measuring the initial permeability (κi -#1), prior to sliding, followed by a run-in at ~5.4 μm/s. Upon reaching steady state sliding (at ~2 mm), the velocity was reduced to 0.05 μm/s. Once stable sliding was established (after ~0.02 mm), a dynamic permeability measurement was initiated (#2). 175 Upon full argon decay, the piston was arrested for ~12h, after which the static permeability was measured (#4). Shear was re-induced (at ~5.4 μm/s), reaching steady state shearing after ~ 0.7 mm, before reducing the sliding velocity to 0.05 μm/s. At steady state (after ~0.02 mm), the dynamic permeability measurements was conducted (#4). Subsequently, the piston was halted for ~1h before the static permeability was measured (#5). The experiments were terminated by subsequently halting and reversing the loading ram at a velocity of 2.0 μm/s until the sample was fully unloaded. Next, the argon pore fluid was 180 removed, followed by removal of the confining pressure, after which the sample assembly was extracted from the apparatus.
Following the regular experiments, as described above, the 5 and 25 MPa along-fault samples were sheared for an additional velocity-stepping sequence, sequential to the SHS sequence. Upon the last static permeability measurement, shear was again induced at ~5.4 μm/s (for ~0.7 mm), and reduced to 0.05 μm/s (for ~0.02 mm). This procedure was repeated for a sliding velocity of 0.1 μm/s, 0.5 μm/s and 1.0 μm/s after which the experiments were terminated and unloaded. 185 In addition to the regular SHS experiments, supplementary experiments (5 and 50 MPa) were conducted to investigate the effect of hold time on the gouge permeability and frictional strength. These experiments were initiated with an initial permeability measurement (#1), followed by a run-in at ~5.4 μm/s and a reduction in sliding velocity to 0.05 μm/s at ~2 mm.
Once a new steady-state was established, a dynamic permeability measurement was initiated (#2). Subsequently, the piston was halted for 300s and a static permeability measurement (#3) was conducted. This procedure of shearing, measuring and holding (#4-8), while measuring the permeability was repeated for progressively increasing hold periods of 1000s, 3000s, and ~12h prior to terminating the experiment. In case >1 mm of shear displacement was left, a last cycle of shearing and holding, for a hold time of ~1h, was applied before termination.

Data acquisition and processing
During each experiment, the internal axial load, piston displacement, confining pressure, sample temperature and argon pore 195 pressure decay were logged every 2 s using a 16-bit National Instrument A/D converter and Labview VI Logger system. To obtain the shear stress (τ) and shear displacement (x), the raw data was processed. Elastic machine distortion effects were corrected for using pre-determined, polynomial stiffness calibrations performed at or near the pressure and temperature conditions applied in this study to obtain true axial displacement values. For the used assembly configuration, the shear stress is equal to the internal axial load divided by the contact area of the shear surface, which is assumed to remain equal to 200 the initial contact area during the experiment.

Permeability calculations
The argon permeability κ was calculated using the transient-pressure decay method described by Sutherland and Cave Yielding 210 where δ represents the dynamic viscosity of argon gas at room temperature [Pa s] and the argon compressibility [Pa -1 ]. The permeability is then determined from the linear slope fitting ln( ∆ ∆ 0 ) vs. t using an in house developed computer program, READ30 (Peach, 1991). The permeability is flow path length dependent. As such the path length for the "across-fault" permeability (κⱵ) is equal to the thickness of the gouge layer, whereas for the "along-fault" (κ//) permeability the path length 215 is the long axis of the gouge layer. For the across-fault permeability calculations, we have to assume that the thickness of the gouge layer is equal to the final thickness at all stages in the experiment, as the current set-up does not allow us to measure the change in layer thickness during the experiment. For the along-fault permeability calculations, we assumed the changes in path length to be equal to the shear displacement at the various stages in the experiment.
The permeability is a measure of the mobility of a fluid within a medium, and as such is the property of the medium and not 220 of the fluid (Klinkenberg, 1941). Under steady-state and laminar flow conditions ("Darcy flow"), the measured (apparent) permeability of a medium is therefore considered to be (close to) the true (intrinsic) permeability of a medium, for any kind of fluid (Klinkenberg, 1941). In Darcy flow, molecule-molecule collisions dominate fluid interactions, therefore moleculewall collisions are neglected. However, when the pore radius of a medium decreases, approaching the mean free path of the fluid molecules, the frequency of molecule-wall collisions increases, resulting in gas slippage along the pore walls. This 225 effect is known as the "slip effect" or "Klinkenberg effect". The true permeability can be derived from the apparent permeability when corrected for this slippage effect by using: where Ka [m 2 ] is the apparent permeability, K [m 2 ] the intrinsic permeability, b is the Klinkenberg constant and ̅ the mean pressure. As a consequence, extrapolation of the apparent permeability to infinite pressure ( 1 ̅ = 0) then gives the true 230 permeability.

Results
All direct shear experiments, experimental conditions and key data, including μss at ~2.0 mm shear displacement, are listed in Table 2. We adopt the convention that compressive stresses are positive and take the effective normal stress σn eff as the applied confining pressure, hence normal stress σn, minus the applied pore pressure Pf, yielding: 235

Permeametry data
The evolution of the along-and across-fault permeability is plotted, together with the friction coefficient, as a function of shear displacement in Figures 4b-i. The initial, pre-shear permeability, measured at effective normal stresses σn eff of 5 to 50 MPa and T≈20°C, ranges between 1.2*10 -15 (at 5 MPa) and 1.6*10 -17 m 2 (at 50 MPa) for the along-fault orientation ( Figure  240 4b-f) and 2.3*10 -16 (at 5 MPa) and 1.8*10 -18 m 2 (at 50 MPa) for the across-fault orientation (Figure 4g-i). Note that the initial permeability measured along-fault is 2 to 7 times higher than measured for the across-fault orientation ( Table 2). All experiments show a rapid change (i.e. decrease) in permeability in the first ~2 mm of shear displacement, down to 2.8*10 -16 (5 MPa -77% decrease) and 8.4*10 -19 m 2 (50 MPa -95% decrease) for the along-fault orientation and 1.5*10 -16 (5 MPa -33% decrease) and 3.4*10 -19 m 2 (50 MPa -85% decrease) for the across-fault orientation. With increasing shear 245 displacement, i.e. shear strain, all experiments show a continuous, though less substantial change in permeability, compared to that observed in the first 2 mm of shear displacement, reaching near-constant values towards the end of the experiments ( Figure 4). In this range, the along-fault permeability is only 1 to 3 times higher than the across-fault permeability (Table 2).
Overall, the permeability values obtained for the along-fault orientation are consistently higher than obtained for the across-https://doi.org/10.5194/se-2020-178 Preprint. Discussion started: 17 November 2020 c Author(s) 2020. CC BY 4.0 License. fault orientation, though the difference seems to diminish with increasing shear displacement (see permeability ratios in 250 Table 2).
Detailed analysis of the obtained permeability measurements shows, superimposed on the downward change in permeability with shear displacement, a reduction in permeability for each subsequent period of hold or sliding ( Figure 4). Our data furthermore shows that the difference in permeability is <20% with respect to the preceding measurement, irrespectively if that it is a measurement after hold or during shear. One exception to this trend is observed in the 17.5 MPa-across-fault 255 experiment, where the third permeability measurement, during shear (Figure 4d), showed a relative increase in permeability with respect to the preceding value obtained during hold.
Besides a decrease in permeability with shear displacement, our results show a decrease in permeability with increasing effective normal stress, irrespectively of orientation. This is explicitly shown in Figure 5a, where the permeability κ is plotted as a function of effective normal stress. Moreover, Figure 5a shows that the decrease in permeability is larger for an 260 increase in effective normal stress from 5 to 50 MPa than for ~2-6 mm of shear displacement.
The initial, pre-shear permeability is plotted as a function of the inverse mean pore pressure in Figure 5b, for measurements obtained at a σn eff of 5, 10 and 17.5 MPa in order to assess the Klinkenberg effect. At effective normal stresses of 5 and 10 MPa, the difference between the obtained apparent (gas) permeability for a mean pore fluid pressure of 2 MPa (cf. 1/Pm = 0.05 MPa -1 ) and the determined true (intrinsic) permeability is <20%. At an effective normal stress of 17.5 MPa, the apparent 265 (gas) permeability, for a mean pore fluid pressure of 2 MPa, deviates ~90% from the true permeability (Figure 5b).
Measurements of the initial, pre-shear permeability (#1) of the same sample always show a reproducibility within 4%, in most cases even to within 1%. The reproducibility of the evolution of permeability with shear displacement (#2-#8) between different experiments performed under the same conditions (σn eff = 5 or 50 MPa, Pf = 2 MPa and T ≈ 20°C) is less good, as illustrated in Figure 6. To rule out that the observed increase in permeability with increasing shear displacement in 270 22OPAPA5 (Figure 6a), is the result of gas leakage along the outer boundaries of the gouge layer, we continued to perform experiments with tape closing off that potential leakage pathways (see sample assembly).

Frictional behaviour
The friction coefficient (μ) versus displacement curves obtained for the simulated, argon-saturated OPA gouges are included in Figures 4b-i. All experiments show a rapid increase in μ until apparent yielding (~0.43-0.73), occurring at a displacement 275 of ~0.3-1.2 mm and occasionally followed by limited slip-hardening reaching a peak strength and a drop in μ, before reaching (near) steady-state friction coefficients of 0.54 to 0.91 at ~2 mm of shear displacement (Table 2). Moreover, they show a modest decrease in friction coefficient following the imposed reduction in sliding velocity.
The steady state friction coefficient obtained after re-shear is generally lower than, or equal to, the pre-hold steady state value. Furthermore, the majority of the friction curves is characterised by (near) strain-neutral behaviour following the run-in 280 stage, with the exception of the 25 and 50 MPa across-fault curves, which show minor strain-hardening (Figure 4). https://doi.org/10.5194/se-2020-178 Preprint. Discussion started: 17 November 2020 c Author(s) 2020. CC BY 4.0 License.
The frictional strength decreases, independently of orientation, with increasing effective normal stress, from max. ~0.9 to as low as 0.5. However, this decrease is non-linear, which is explicitly shown in Figure 7a for the steady state frictional strength at ~2 mm as a function of effective normal stress. Generally, stable sliding was observed for all sliding velocities, however The shear stress (τ) supported at steady-state at ~2.0 and at 3.4-and 4.5-mm shear displacement, is plotted as a function of the σn eff in Figure 7 for all (argon-saturated) experiments performed. The data show a linear relation passing through or close to the origin, suggesting nearly cohesionless fault gouges (Figures 7c-d). Fitting the Mohr-Coulomb criterion, i.e. τ = μσn eff + C0, to the linear τ vs. σn eff data yields very similar cohesions (C0) of roughly 2 MPa for both along-and across-fault 305 orientation (Table 3), when considering the error bars. The cohesion furthermore appears to be independent of shear displacement, as the cohesion remains to be ~2 MPa (Figure 7c-d - Table 3).

Discussion
One of the major concerns of CO2 injection in geological formations is the potential risk of inducing fault slip and (micro)seismicity along previously sealing (i.e. impermeable) gouge-filled faults (Hawkes et al., 2005;Zoback and Gorelick, 310 2012). Fault slip and induced seismicity are most likely the result of a change in the stress conditions on pre-existing faults.
This can occur via 1) (local) compaction or heave of the reservoir (poro-elastic effect), 2) thermal expansion or shrinkage near the well, or 3) an increased pore pressure in a pre-existing fault, effectively reducing the normal stress on the fault https://doi.org/10.5194/se-2020-178 Preprint. Discussion started: 17 November 2020 c Author(s) 2020. CC BY 4.0 License. (Hawkes et al., 2005;Wang, 2000). Once fault slip is induced, this may cause dilatation, potentially increasing fault permeability. Upon slipping, leakage of CO2 could then either occur across the fault, i.e. between two adjacent reservoir 315 compartments, or along the fault, e.g. into an over-or underlying formation, in case an outward pressure gradient is in place (Bense and Person, 2006).
To help evaluate the potential for leakage along or across gouge-filled faults cutting clay-quartz-calcite caprock, we have investigated the effect of shear displacement, dynamic shear, static holding, and normal stress on the transport properties (both along-and across-fault) of a simulated clay-quartz-calcite fault gouge. 320 In the following, we will first discuss our findings regarding the effects of shear displacement, dynamic shear, static holding, and effective normal stress on the permeability evolution, along-and across-fault, of a simulated clay-quartz-calcite fault gouge. We will then discuss the implications for sealing integrity and frictional behaviour of a gouge-filled fault cutting a potential clay-quartz-calcite caprock in CO2 storage systems.

The effect of shear displacement, dynamic shear, static hold, and normal stress on fault gouge 325 permeability: across-and along-fault
In discussing our experimental data on gouge permeability, we start with the notion that, as shown in Sect with hold time for Westerly granite and for Green River shale (containing more than 50% carbonate). In particular the absolute permeability enhancement for the granite was substantial, while it was small for the shale. The effect on shale becomes better visible if using a normalized permeability increase. Nevertheless, the results of Im et al. (2018) show that differences in material properties play a role, which likely explains the difference with our observation regarding hold time and permeability change. 335 Overall, our data show a decrease in permeability with shear displacement (i.e. fault maturity) and with effective normal stress ( Figure 5a) for both along-and across-fault orientation. The most significant reduction in permeability is observed in the first few millimetres of shear displacement, reaching (near) constant values during further shear displacement. Over a shear displacement of up to ~6 mm, the permeability changed by up to an order of magnitude compared with the pre-shear permeability. Comparing the along-and across-fault permeability measurements demonstrates an initial anisotropy of up to 1 340 order of magnitude. Furthermore, our permeability results, both along-and across-fault, show that the decrease in permeability with increasing displacement coincides with a decrease in frictional strength (Figure 4). Note also that the friction coefficient decreases with increasing effective normal stress (Figure 7a).
The observed sharp decrease in permeability, characteristic for the initial stage of shear, followed by a more gradual decrease towards more or less constant values with continuous shear is reported for a broad range of clastically-derived compositions 345 (e.g. Morrow et al., 1984;Zhang and Tullis, 1998;Bakker et al., 2017). Furthermore, the total decrease of about one order of https://doi.org/10.5194/se-2020-178 Preprint. Discussion started: 17 November 2020 c Author(s) 2020. CC BY 4.0 License. magnitude upon shear, both along-and across-fault, is in agreement with permeability measurements on mixtures rich in phyllosilicates, including chlorite, illite, kaolinite, montmorillonite and muscovite (Figure 4 e.g. Crawford et al., 2008;Morrow et al., 1984;Zhang et al., 1999). Besides the reduction in permeability with shear displacement, we observe a reduction in permeability with increasing effective normal stress, again in agreement with many previous studies (e.g. 350 Behnsen and Faulkner, 2012;Crawford et al., 2008;Morrow et al., 1981b;Zhang et al., 1999).
During initial loading, prior to shear, mineral alignment of the clay-quartz-calcite mixtures is assumed to be dominated by rotation of phyllosilicates towards an orientation perpendicular to the highest principal stress orientation, i.e. the applied normal stress, hence parallel to the boundaries of the gouge layer (Zhang et al., 2001). As a consequence, fluid flow across the fault is more tortuous than along the fault, which is supported by the permeability anisotropy observed for the pre-shear 355 permeability values (Figure 5a). The sharp, initial decrease in permeability during the run-in stage of the experiments is inferred to reflect shear-enhanced compaction of the bulk gouge layer, as rearrangement of grains and grain size reduction allow for a denser grain packing (e.g. Marone and Scholz, 1989;Zhang et al., 2001). Upon reaching steady-state shearing, thin and dense shear bands are expected to have developed accommodating localised shear (Haines et al., 2009), which will not affect the bulk permeability of the gouge layer any further during subsequent shear displacement, dynamic shear and/or 360 periods of holding. This is supported by the transition to more or less constant permeability values for the bulk of the gouge layer obtained during this stage of the experiment. Moreover, the permeability anisotropy shows a decrease with respect to the initial permeability anisotropy. This suggests a shear-induced reduction in tortuosity contrast between along-and acrossfault fluid flow, in agreement with observation for mica-gouges by Zhang et al. (1999), who inferred the reduction to result from an anisotropic enhancement of fluid flow by P-shear and Y-shears. Permeability measurements obtained during 365 dynamic shear and static holding show little difference, suggesting a limited effect of holding versus shearing on the permeability, especially when a well-defined foliation is established (Figure 4).
The difference in permeability and permeability evolution observed for the various effective normal stresses (Figure 5a), i.e.
high permeability values at low effective normal stress and lower values at higher normal stresses (Figure 4), suggests various degrees of compaction via grain rearrangement and grain crushing (Zhang and Tullis, 1998). At the same time, the 370 observed decrease in frictional strength with increasing effective normal stress suggests an indirect correlation between the friction coefficient and permeability. A similar relation has been observed between the mineral composition and permeability by Crawford et al. (2008). However, we have too little experimental data to convincingly confirm such correlation with friction coefficient. What also plays a role is the fact that at high effective normal stress (>17.5 MPa), the Klinkenberg effect cannot be ignored. Given the specific experimental procedure employed in this study, a systematic Klinkenberg correction 375 (Eq. (4)) could not be applied, as during shear the maturity of the fault, and therefore the experimental situation, continuously changed. As a consequences, observed trends of increasing or decreasing permeability are meaningful, but not too much importance can be attributed to absolute values, at least not at effective normal stress above 17.5 MPa. More work is needed on these aspects. https://doi.org/10.5194/se-2020-178 Preprint. Discussion started: 17 November 2020 c Author(s) 2020. CC BY 4.0 License.

The effects of deformation conditions on the frictional behaviour of simulated clay-quartz-calcite 380 fault gouges
In discussing our friction results, obtained for the (argon-saturated) OPA samples (see Sect. 3.1), we begin by noting that the effect of calculated cohesion of ≤2MPa (Figure 7; Table 3) is relatively high at low effective normal stresses (<25 MPa), and relatively small at higher effective normal stress (>25 MPa). Therefore, we infer that the shear strength of our samples, expressed as friction coefficient, can be represented by the ratio of the shear stress over the effective normal stress at 385 effective normal stresses ≥25 MPa, especially at σn eff = 50 MPa as these values are most relevant for CCS in geological reservoirs. This inference is further supported by the fairly good agreement between the experimentally derived friction coefficients at 25 and 50 MPa and the internal friction coefficient at yield or peak, derived via Eq. (3) (Table 3). However, the cases that the agreement is less good convey the samples performed at effective normal stresses < 25 MPa. These should therefore be assessed carefully. 390 Our friction data showed higher friction coefficient values at lower effective normal stresses than at 50 MPa ( Figure 4). A similar gradually decreasing friction coefficient to more constant values at higher effective normal stresses has been observed for various phyllosilicate-bearing mixtures (Behnsen and Faulkner, 2012;Ikari et al., 2007;Saffer and Marone, 2003), and is assumed to reflect the influence of gouge cohesion at low effective normal stresses as mentioned above.
In addition to the observed change in absolute frictional strength with changing effective normal stress, we also observed an 395 increase in friction coefficient with increasing velocity, and vice versa, for the mineral compositions and deformation conditions tested (Figures 4 and 6). Despite the positive dependence of friction coefficient on velocity seen in the experiments (Figure 4), it is striking that low amplitude, stick-slip behaviour was observed in several of the (argon-saturated) OPA samples. Specifically, stick-slip was observed in samples 27OPAPA5t, 31OPAPC5t and 30OPAPC25t, which were sheared at room temperature, at sliding velocities of 0.05-1.086 μm/s and effective normal stresses of respectively 5, 5, and 400 25 MPa. Based on the Rate and state slip theory stick-slips are not expected in gouges exhibiting velocity-strengthening behaviour. It is unlikely that this behaviour is a chemical effect of the argon gas, given its inert nature. There is also no systematic correlation between the observed stick-slip behaviour and across-versus along-fault argon flow. Possible explanations that we can propose for the "anomalous" stick-slip include the following: 1) Local, stepwise penetration of argon into previously unpenetrated shear bands, due to shear band dilatation at the 405 low effective normal stresses used in the experiments showing stick-slip. This could be possible without affecting sample scale permeability if it is transport through the body of the sample that determines its permeability.
2) We cannot exclude the possibility mentioned earlier that small amounts of water may be introduced into the sample during argon injection, from traces left in the pore fluid system after earlier wet experiments. Gradual penetration of water introduced into the gouge in this way, could conceivably lead to slip-weakening and associated stick-slip. 410 Some support for this is offered by the correlation between slip-weakening and stick-slip seen in the relevant argonsaturated experiments.
3) A further possibility is that the samples showing stick-slip at σn eff = 5 MPa became over-consolidated, with respect to shear-testing, due to pre-pressing during preparation. This would produce a peak strength during shear followed by slip-weakening (Ide and Takeo, 1997) and hence potential for unstable slip events. 415

Implications
In order to predict with some confidence what may happen in faults cutting the caprock of a potential CO2 storage system, it is of importance to understand what factors control the frictional and transport properties of a gouge-filled fault. The goal of this research was therefore to evaluate the effect of shear displacement, dynamic shear, static holding and effective normal stress on i) fault gouge permeability, both parallel and perpendicular to the fault boundaries, and ii) on the frictional strength 420 and stability of a simulated clay-rich gouge. Both aims are particularly relevant for, but not limited to, CO2 injection and storage in potential storage reservoirs in the subsurface as for both aims different aspects of fault integrity are evaluated. The implications of our results are listed below: 1) Faults intersecting a clay-quartz-calcite-bearing formation, such as a calcite-rich claystone or shale, may be (re-)activated upon hydrocarbons production or CO2-storage, potentially leading to induced (micro-)seismicity, fault 425 zone dilation, permeability enhancement and as a consequence a reduction of sealing capacity. It is therefore critical to know how shear displacement, dynamic shear, static holding and effective normal stress affect the along-and across-fault permeability. Assuming a clay-quartz-calcite caprock composition resembling the Opalinus Claystone (40-50% phyllosilicates, 20-25% quartz, and 20-30% of calcite) the present results imply that with increasing shear displacement the permeability will reduce up to an order of magnitude, with the most significant reduction occurring 430 in the first millimetres of shear displacement. With increasing maturity, upon reaching steady-state frictional behaviour, faults will acquire a well-developed internal foliation, which has only limited potential for compaction or dilation. As a consequence, re-shear or static holding will only slightly decrease the permeability further, before levelling off to near constant permeability.
2) Potential CO2 storage reservoirs are located at 1-4 km depth (cf. 25-100 MPa). The present results (Figure 4) show 435 that pre-existing gouge-filled faults at these depths are expected to have permeabilities ranging between 10 -16 and 10 -19 m 2 (Klinkenberg uncorrected). Comparing these values with the permeability of an intact slice of Opalinus Claystone (7.7 • 10 -20 m 2 (Klinkenberg uncorrected) at 15 MPa effective normal stress or ~0.7 km depth) shows that, even at normal stresses corresponding to 2 km, the gouge permeability of 10 -19 m 2 is still 1-2 orders of magnitude higher than the intact rock, and therefore more likely to act as a leakage pathway. 440 3) If the pore fluid pressure in a fault zone increases due to migration of supercritical CO2 from a storage reservoir, the effective normal stress can decrease. Our results show that with increasing pore pressure, and decreasing effective normal stress, the permeability of simulated clay-quartz-calcite fault gouge will increase, irrespectively of fluid flow orientation, reducing sealing capacity. Moreover, our results show that leakage along a fault into an over-or https://doi.org/10.5194/se-2020-178 Preprint. Discussion started: 17 November 2020 c Author(s) 2020. CC BY 4.0 License.
underlying formation is easier than leakage across the fault into neighbouring reservoir compartments, with our 445 simulated fault gouges showing ~1 order of magnitude anisotropy in permeability.

Conclusions
In this study, we aimed to evaluate various factors that control the transport and frictional properties of gouge-filled faults Pre-shear permeability values show up to an order of magnitude difference (anisotropy) between alongand across-fault, with shear displacement this difference decreases.
2) Pre-existing gouge-filled faults with a clay-quartz-calcite composition located at a depth range of 1-4 km 465 are expected to have permeabilities ranging between 10 -16 and 10 -19 m 2 (Klinkenberg uncorrected). The Klinkenberg uncorrected permeability of the intact rock (7.7 • 10 -20 m 2 at 15 MPa effective normal stress or ~0.7 km depth) is, even at normal stresses corresponding to 2 km, 1-2 orders of magnitude lower than the fault gouge. Gouge-filled faults are therefore more likely to act as a leakage pathway.
3) The permeability of simulated clay-quartz-calcite fault gouge will increase, thereby reducing the sealing 470 capacity, irrespectively of fluid flow orientation, if the pore fluid pressure in a fault zone increases due to migration of supercritical CO2 from a storage reservoir. Moreover, leakage along a fault into an over-or underlying formation is easier than leakage across the fault into neighbouring reservoir compartments (~1 order of magnitude anisotropy in permeability).