Hydraulic fracture propagation in a heterogeneous stress ﬁeld in a crystalline rock mass

. As part of the In-situ Stimulation and Circulation (ISC) experiment, hydraulic fracturing (HF) tests were conducted in a moderately fractured crystalline rock mass at the Grimsel Test Site (GTS), Switzerland. The aim of these injection tests was to improve our understanding of processes associated with high-pressure ﬂuid injection. A total of six HF experiments were performed in two inclined boreholes, where the surrounding rock mass was accessed with twelve observation boreholes, which allow high-resolution monitoring of fracture ﬂuid pressure, strain and micro-seismicity in an exceptionally well-characterized 5 rock mass. A similar injection protocol was used for all six experiments to investigate the complexity of the fracture propagation processes. At the borehole scale, these processes involved newly created tensile fractures intersecting the injection interval while at the cross-hole scale, the natural network of fractures dominated the propagation process. The six HF experiments can be divided into two groups based on their injection location (i.e., south or north to a brittle ductile shear zone), their similarity of injection pressures and their response to deformation and pressure propagation. The injection tests performed in the south 10 connect upon propagation to the brittle ductile shear zone. Thus, the shear zone acts as a dominant drain and a constant pressure boundary. The experiments executed north of the shear zone, show smaller injection pressures and larger backﬂow during bleed-off phases. From a seismic perspective, the injection tests show high variability in seismic response independent of the location of injection. For two injection experiments, we observe re-orientation of the seismic cloud as the fracture propagated away from


Seismic response and seismic cloud
Induced seismicity accompanies hydraulic stimulation and can be detrimental to deep geothermal projects when magnitudes of induced earthquakes are perceptible to the public (Ellsworth, 2013;Evans et al., 2012). However, in many industrial projects in the context of both hydrocarbon and heat extraction, including this study, it is an indispensable tool that is used to map the stimulated fracture system (Maxwell et al., 2010;Niitsuma et al., 1999;Warpinski et al., 2013). When the hydraulic fractures 5 propagate beyond the vicinity of the injection point, they will inevitably interact with natural fractures to some degree. Induced microseismicity (on a small scale also called acoustic emissions) occur as localized brittle-failure processes during highpressure fluid injection and can be used to approximate the geometry of a single hydraulic fracture or a fracture network. Nolen-Hoeksema and Ruff (2001) proposed three mechanisms that may produce seismicity during hydrofracturing: 1) Tensile failure at the fracture tip, 2) the stress concentration at the fracture tip causing shear slip along suitably oriented pre-existing 10 fractures and, 3) fluid leak off into pre-existing fractures rising the fracture fluid pressure inducing slip if they support sufficient shear stress. The tensile failure at the tip is typically aseismic or at least radiates a small amount of energy. Mechanism two and three are often seen as the main processes leading to induced seismicity during hydraulic fracturing (e.g. Martínez-Garzón et al., 2013;Rutledge et al., 2004;Warpinski and Branagan, 1989). Thus, induced seismicity does not represent the propagating fracture itself, but is indicative of the hydraulic fracture propagation as it tracks the propagating fracture. Many experiments on 15 different scales in laboratory and under in-situ conditions showed that the seismicity cloud has a tendency to be oriented normal to the minimum principal stress (σ 3 ) direction (Evans et al., 2005;Häring et al., 2008;Hubbert and Willis, 1957;Rutledge et al., 2004). Majer and Doe (1986) concluded from the occurrence rate, spatial and temporal distribution of the microseismic events, that the hydro-fracture growth pattern does not follow an often-assumed single and symmetric fracture path. In fact, the one "hydraulic fracture" is actually made of multiple fractures. 20

Our contribution
A detailed characterization of the stress state as well as a geological, hydrological and geophysical characterization of the experimental rock volume took place before the main injection experiments were executed. The in-situ hydraulic fracturing experiment presented here along with detailed pressure, deformation and seismic monitoring were designed to address the following research questions:

25
-What is the injection pressure response at the injection interval and at pressure observation intervals in the rock volume?
-What is the rock deformation response to high pressure fluid injection?
-How does permeability increase in response to HF processes?
-How does the microseismic cloud propagate and what is the best description of the fracture geometry? Does the borehole trace of the hydraulic fracture match the late-time geometry? How does the outcome vary related to the volume of injected 30 fluid?
-How can we describe the stress state during fracture propagation? Does it change? rock and the S1 shear zone indicates a fracture system with a consistent horizontal NNW orientation with a tendency of higher variation in the host rock. The fractures of the ductile-brittle S3 shear zone indicate two different fracture systems. Figure 2d) shows the fracture frequency, corrected into volumetric fracture density (p32) to account for sampling biases (see Brixel et al., 2019, for the correction), for the two injection boreholes, which indicates a moderately fractured rock mass. The

Stress characterization
Details of the stress characterization campaign are given by Krietsch et al. (2019b). Impression packers and microseismic monitoring were used to map hydraulic fracture orientation, which revealed consistent E-W, sub-vertical fracture extension. 5 The averaged stress field in relatively unperturbed rock (i.e. with little fracture density) about 8 m from the S3 shear zones (Fig. 1c) and the perturbed stress field are summarized in Table 1. Hence, the minimum principal stress magnitude, measured in the sub-horizontal borehole SBH4 approaching the S3 shear-zone at the borehole bottom, reduces towards the S3 shear zone Krietsch et al. (2019b). The two main changes compared to the perturbed stress field are 1) the stress field is 45 • rotated clockwise and 2) the intermediate and minimum stress axis switch place. on their relation to the main S1 or S3 structures and presented to estimate their criticality due to the fluid pressure increase.
Structures favorably oriented for failure will fail with overpressures ranging from 8 to 10 MPa. All the HF experiments are 15 located around the S3 shear zone, which influences the stress field as observed during the stress characterization campaign. To investigate this effect a Mohr-Coulomb circle for the perturbed stress state is presented and the failure limits are indicated for 4 to 7 MPa overpressure. The perturbed stress field would allow shearing of structures above 4.5 MPa overpressure, which is significantly below the observations from the unperturbed stress state. It is not clear, which of the two observed stress states describe the injection into the rock volume approaching the S3 and S1 shear zones best. The experiments executed in borehole 20 SBH4 approaching the S3.1 shear zone indicate the change towards the perturbed stress state.

Hydraulic characterization
Multiple field tests were performed to characterize hydraulic conditions at and near the injection borehole pair, including dilution tests, single-hole and cross-hole hydraulic tests. The transport properties of the conductive fractures were characterized using salt and DNA tracer tests (Jalali et al., 2018b). project is presented by Brixel et al. (2019). Hydrogeological conditions prior to the hydraulic fracturing experiment may be summarized as follows: -The transmissivity of the intact injection intervals (defined here by the absence of visually detected discontinuity on cores and borehole image logs) was estimated through hydraulic pressure pulse tests and range from 10 −13 to 10 −11 m 2 /s (pink circles in Fig. 2d). In injection intervals intersected by shear zones, constant rate injections (CRI) or pulse 5 injections (PI) indicate higher transmissivity values in the order of 10 −6 to 10 −13 m 2 /s (blue dots in Fig. 2d). The geometric average transmissivity of the host rock is estimated to be 10 −11 m 2 /s.
-The estimated transmissivities are in good correlation with the fracture intensity in the injection borehole INJ1 ( Figure   2d, left).
-Within the brittle fractured zone between the two S3 shear zones an average discharge into the GTS tunnel of ∼60 10 ml/min was measured prior to the injection experiments.
-Based on the characterization tests conducted, the fractured zone between and along the two S3 metabasic dykes provide the most conductive, natural flow pathways between the two injection boreholes. This observation agrees well with the existence of two different fracture systems in the S3 shear zone: (i) one set following the main NE-SW alpine foliation orientation and (ii) one set abutting on the two S3 dykes at high angles, which we identified as the alpine tension 15 gashes commonly mapped between dyke swarms throughout the Grimsel Test Site (Figure 2a, right). This rock volume contains open fractures with high transmissivity (i.e. extension fractures), whereby the transport of solutes, salt and DNA tracers between the two injection boreholes show a preferential pathway towards the gallery rather than towards the INJ1 borehole for the case of injection into INJ2 (Jalali et al., 2018b).
3 Field setup and monitoring 20 Hydraulic fracturing experiments conducted within this study were accompanied by an extensive monitoring program including measurements of rock deformation, fracture fluid pressure and microseismicity. In the following sections, we introduce the hydraulic fracturing equipment and the monitoring systems.

Hydraulic fracturing equipment
The HF interval was isolated using a hydraulic double-packer system with a 1 m long pressurization interval. Two different 25 triplex pumps (brand SPECK Pumpen) were used to deliver 1) a pressure up to 30 MPa at a flowrate up to 35 l/min and 2) a flowrate up to 100 l/min at a maximal pressure of 10 MPa. The first pump was used to breakdown the formation and for the first propagation cycle. Then, the pump was switched to reach flowrates up to 100 l/min. A second double-packer system was installed to monitor the fluid pressure response in the monitoring interval in the second injection borehole that was not used for active stimulation. A data acquisition system recorded the pressure in the open intervals of the INJ boreholes, the flowrate 30 in the injection interval and the packer pressure in the injection and monitoring intervals with a sampling rate of 20 Hz. Fluid pressure was also monitored in the intervals beneath the injection and monitoring intervals with a sampling rate of 1 Hz. The injected fluid and the backflow were measured with different flowmeters depending on expected flowrates with a sampling rate of 20 Hz. Figure 1b) presents the position of the six injection intervals along the two injection boreholes INJ1 and INJ2. As a visual aid, we used consistent color throughout the paper to display data from a specific HF experiment. The execution times and the intervals for all experiments are summarized in Table 2.

5
The pressure monitoring system was designed to observe transient pressure response at specific locations to track pressure propagation throughout the rock mass, either through natural fractures or newly created ones. Customized grout packer systems were installed in the PRP monitoring boreholes. The open intervals were packed and separated with hydro-mechanical packers supplement with resin. The uppermost interval was filled with grout to ensure low compressibility of the system. Open hole sections are shown as blue cylinders in Figure 1d. The sections PRP1-1, PRP2-1 and PRP3-1 are positioned within shear zone 10 S1 and all the other intervals are positioned within shear zone S3. The pressure sensors (PAA33-X Keller) were connected to the Solexpert data acquisition system running the Solexpert GM-HF software with a maximum sampling rate of 20 Hz.
The possible pressure range of the pressure sensors was 10 MPa with a resolution < 1 kPa. The raw pressure data from HF stimulations are presented in the paper without any filtering.
The rock mass deformation monitoring system consists of 60 fibre-bragg grating (FBG) sensors (Type os3600 by Micron

15
Optics Inc) in the three FBS boreholes. The FBG sensors have a base length of 1 m. 20 FBG sensors were installed along each FBS borehole to characterize the strain field in both intact and fractured rock. The sensors (including strain and temperature) are pre-strained to about 2000 microstrain such that also shortening can be recorded. The sensors were connected to an interrogator of type si255 (Hyperion Platform by Micron Optics Inc.) that can record with a sampling rate of 1 kHz, an accuracy of 0.85 microstrains and wavelength repeatability of 0.1 microstrains. The strain data presented here is not temperature corrected as 20 it is not required for our isothermal injections. Extensional strain is negative.
Two tiltmeters (Type A711-2 by Jewell Instruments) were installed in the VE tunnel to characterize the deformation with respect to the stimulation volume. The location of each tiltmeter is presented in Figure 1c. The tiltmeters measure the deviation from horizontal tilt in axial (X) and normal (Y) direction to the tunnel with a resolution of 0.05 µradians after filtering with a 100 Hz low pass filter. The two horizontal tilt axis and the temperature were digitized and recorded by the data acquisition system with a sampling rate of 100 unitHz. The initial value was subtracted to display the change of tilt during the HF experiment. Then, a positive tilt in x-axis implies a dip of the tunnel wall towards NNE. A positive dip in the y-axis indicates a dip of the tunnel floor towards WNW. kHz, using a 32-channel acquisition system. AE and accelerometer receiver signals were high-pass hardware filtered at 1 kHz and 50 Hz, respectively.
Based on the picked P-wave onsets, the seismic event locations were calculated using a homogeneous but transversely The two following refrac cycles RF1 and RF2 had the aim to propagate the hydraulic fracture. During these cycles, we used either water with a viscosity of 1 cP (10 −3 Pa.s) (Fig. 3 a-  l/min injection step we increased to 35 l/min for 3 minutes. Then, the system was shut-in to change to the bigger pump and injection was resumed without bleed-off with a second propagation cycle called RF2. The refrac cycle RF2 begins with a rapid increase in the flowrate to 35 l/min and then increased to 50 l/min. Both flow steps were maintained for 2 min. Afterward, each step was held for 1 min starting from 60 l/min and going up to 70, 80, 90 and 100 l/min. The system was then shut-in for several minutes to half an hour to observe the hydro-mechanical response in the system. The water was injected at 35 l/min. This cycle was also an opportunity to test different cyclic injection schemes. The fluid injection was again followed by shut-in and bleed-off phases. For all experiments, the last cycle was a pressure-controlled step test (SR) to evaluate the post-stimulation injectivity of the created hydraulic fracture and to estimate the stress acting normal to the hydraulic fracture (jacking pressure) based on Doe and Korbin (1987). For the executed injection protocols, the following remarks are noted:

5
-Logistic problem affected the execution of HF1 experiment. The issue was an insufficient water supply. This led to several repetitions of the refrac cycle RF1. The second refrac cycle (RF2) was then executed a day later with a new installed water-supply pump delivering the necessary flowrates. Furthermore, the seismic monitoring system recorded an increased quantity of electronic interferences due to a faulty shielding of the power line between the frequency control unit and pump motor. Therefore, a seismic evaluation was not possible.

10
-At a flowrate of 5 l/min during the first refrac cycle of experiment HF5, a short-cut occurred to one of the open seismic monitoring boreholes (GEO1, Fig. 1e). For this reason, we had to interrupt RF1 and resume multiple times. This is why RF1 is subdivided in part a, b, c on Figure 3. Thereafter, the pump was changed to allow flowrates above 35 l/min, but the flowrate was limited to a maximum of 80 l/min and short duration due to the short-cut to the GEO1 borehole. The flushing cycle RF3 was executed with a flowrate of 50 l/min.

15
-The low magnitude of breakdown pressure during the frac cycle of HF6 is an indication for a pre-existing sealed fracture in the open interval. The stimulation interval was mistakenly placed 3 meters further down in the borehole at a preexisting fracture.
-During all refrac cycles, we never exceeded an injection pressure of 10 MPa.
-Experiment HF3, HF5, and HF8 show a similar fast pressure decay during the shut-in time for the refrac cycles. The  Table 3 summarizes key observations from the injection protocol (labeling after to Fig. 3). The measurement of breakdown pressure, fracture reopening and instantaneous shut-in pressure (ISIP) followed the ISRM standard presented by Haimson 25 and Cornet (2003). The breakdown pressure represents the peak pressure during injection cycle F. The instantaneous shut-in pressure (ISIP) was obtained during each cycle using the tangent method, i.e. the departure from a linear pressure decrease vs. time occurring right after shut-in (Amadei and Stephansson, 1997). The apparent re-opening pressure (P r ) was picked when the pressure change-time step-relationship starts being non-linear (Bredehoeft et al., 1976). The jacking pressure was measured during the pressure-controlled step test SR (more in the supplementary information S1). The cumulative injected 30 water corresponds to the injected volume, V i , indicated for each cycle and the entire experiment. The backflow was measured at the injection interval during venting. Thus, the fluid recovery V r is only the recovery from the injection interval, however, fluid also escaped from two of the monitoring borehole (GEO) during HF5 and HF8 experiments and from the fractured zones during all experiments. These outflows were also monitored to validate the overall fluid balance over each experiment. Cycle Table 3. Overview of fracture breakdown pressure (Pc) and fracture reopening pressure (Pr) and additional hydraulic test parameters (injected fluid: water (W) or Xanthan-salt-water mixture (XSW); Vi, injected volume; Vr, recovered volume) and localized AE event numbers. RF1 for HF1a and HF5 correspond of multiple refrac cycles. Thus, the re-opening pressure and the ISIP were averaged and the fluid injection respective recovery is presented in a cumulative number. Figure 4a presents the flowrate (q inj ) vs the interval pressure (p inj ) at pseudo steady-state for the second refrac cycle RF2 and the pressure-controlled step test for each HF experiment. For any injection step, the injection conditions (either controlled flowrate or constant pressure) were maintained until a stable state was reached, i.e. quasi constant pressure for rate-controlled 5 injections or quasi constant flowrate for pressure-controlled injections, while we acknowledge that true steady state conditions are never reached in practice, we took the latest data point prior starting the next step of our injection. A major observation is that there is a clear difference between HF experiments executed south of the S3 shear zone (HF3, HF5 and HF8) and north of the S3 shear zone (HF1, HF2 and HF6). HF1 and HF2 were located in the S1 shear zone and reached a pressure limiting behavior at an injection pressure of 5.4 MPa for flowrates larger than 35 l/min ( Figure 4a). During these two experiments, only 10 two propagation cycles (RF1 and RF2) were executed. Considering experiment HF6, the pressure limiting behavior occurred at a higher pressure of 6 MPa.

Diagnostic injection parameters
HF3, HF5, and HF8 were located south of the S3 shear zone. In general, all experiments executed on this side of the S3 shear zone had higher injection pressure. HF3 and HF8 showed similarly, a slight increase after reaching a flowrate of 10 l/min. The injection pressure reached a limiting pressure of 7.8 MPa. The injection fluid for HF3 was water and for HF8 XSW followed by an additional flushing cycle using water. Experiment HF5 showed the highest increase in injection pressure for increasing flowrate with a maximum injection pressure above 9 MPa. The pressure dropped around 0.5 MPa at the same flowrate (35 l/min or 50 l/min) using XSW during the refrac cycle RF2 and the flushing cycle (RF3) using water for experiment HF6 and HF5. The effect is smaller for HF8 with a pressure drop of only 0.2 MPa. Therefore, we can infer that the viscosity effect 5 (i.e., the change of XSW to water) results in a decrease of injection pressure, but further investigation is necessary to reliably quantify the effect. For the sake of clarity, the flowrate and pressures from the flushing cycle associated with the pressure drop are presented in Figure 4 (b-d).
The limiting pressure is smaller compared to the hydraulic tensile strength calculated from the difference between breakdown and reopening pressure (Bredehoeft et al., 1976) ranging between 8.3 and 9.6 MPa. The tensile strength measured by Dutler However, it is questionable if the hydraulic fractures still extend (creating new surfaces) at this pressure.
The difference from HF6 to the two other injection experiments (HF1 and HF2) may be related to two reasons: 1) The HF6 15 interval contains a pre-existing fracture that may not be perpendicular to the minimum principal stress.
2) The injection fluid in HF6 is XSW. Thus, the pressure reaches higher values as pressure dissipation is affected by the high fluid viscosity. During the flushing cycle of this fracture (RF3), a pressure decrease was observed, reflecting a viscosity effect.  shear zone. South of the S3 shear zone, the the jacking pressure reaches values between 5.1 and 6.0 MPa, which is comparable with the ISIP from the refrac cycles. In contrast, the jacking pressure north of S3 range between 3.0 and 3.7 MPa for the experiments executed. This reflects 1 to 2 MPa smaller values than using the ISIP from the refrac cycle (Fig. 5a).
The injection volume and the recovery volume from the injection intervals for the main fracture propagation cycles RF1/RF2 and the flushing cycle RF3 are presented in Fig. 5b The final injectivity before jacking is presented in Figure 5c and ranges between 2.77 and 3.69 l/min/MPa north of S3 and between 0.21 and 0.88 l/min/MPa south of S3. Approaching the S3 shear zone from the south, injectivity values increase. The experiments located further down in the borehole show highest injectivity values, which correlates directly with an increase of fracture density in the shear zone S1.

Transmissivity values from pre-and post-HF hydraulic tests 15
The change in transmissivity at the injection interval was investigated by packer testing before and after the HF experiment in borehole INJ1 and INJ2. Prior to the HF experiment, pulse injection (PI) and after the HF experiment, constant head injection (CHI) tests were performed. To estimate the equivalent hydraulic parameters (transmissivity and storativity), the PI tests were inverted with n-dimensional Statistical Inverse Graphical Hydraulic Test Simulator, nSIGHTS (Roberts, 2006) and the CHI tests were analyzed using the Jacob and Lohman (1952) solution. For both methods the radius of influence is different and 20 therefore the numbers are only an indication of permeability enhancement and not for direct comparison. The initial magnitude of local transmissivity estimates ranges between 10 −13 −8 * 10 −13 m 2 /s. Final transmissivities posterior to the HF experiment, from CHI tests reach values between 1.9 * 10 −9 −3.6 * 10 −11 m 2 /s. HF6 is an exception as it took place at a pre-existing fracture, for which transmissivity was not measured before the HF experiment. The final transmissivity for HF6 is highest in magnitude for all HF experiments with 1.5 * 10 −7 m 2 /s. There is a trend of one magnitude higher transmissivities after HF for experiments 25 on the northern side of the S3 shear zone (more in the supplementary information S2).

Borehole fracture trace
Prior to the HF experiments, analysis of drill cores, optical and acoustic televiewer have been carried out to select appropriate test intervals. Suitable test intervals for hydraulic fracturing were selected if no pre-existing fractures were visible. Figure 6 presents the amplitude log from the acoustic borehole televiewer for the six HF intervals for pre-and post-testing. The travel-

Microseismicity
During the HF experiments, we detected in total 6986 microseismic events, from which 730 events were localized. P-wave arrivals were picked manually and located using an absolute location procedure including a joint hypocenter determination The number of localized seismic events for experiment HF5 and HF6 is very small. Most seismic activity is observed during experiment HF2 and HF8. Note that the location of the injection intervals with respect to the seismic sensors cannot explain the large difference in detected and located seismic events. During all experiments, the seismic events are located more often 5 below the injection point than above. Also, seismicity propagates predominantly towards east.
A peculiar seismic pattern was observed during HF2 (see close to the injection point. A color saturation scheme is used to give less importance to these events: a more intense color is used for events located further away from the injection point and linearly decreasing the color saturation towards the injection points. Events located within a radial distance of 1 m (typical location error for our seismic events) from the injection point are not represented.
In the idealized case of symmetric radial extension of the seismic cloud on a plane from the injection point, the expected 30 pattern on the stereoplot would be a girdle pattern, i.e. all the points will line on a great circle of the stereoplot. In Figure 8 we present the best fit plane through the seismic cloud via girdle pattern for all seismic events (black pole point for HF8) and for two clusters (i.e. RF1 and RF2 for HF2, red and yellow pole point). The two HF experiments presented in Figure 8 do not follow this idealized pattern. In our case, the points tend to cluster that represent a linear structure. For HF2 during RF1 (red), the points tend to distribute on an E-W sub-vertical plane and cluster on a linear structure dipping about 70 • to the East. During RF2, the seismicity cloud migrates (yellow) towards a well-defined linear structure with a dip of 25 • to the ESE. The mean from the first and second refrac cycle are very consistent dipping 40 • to 50 • to the East. The seismicity cloud of HF8 is clustered on a linear structure dipping 60 • to 70 • to the West. Most of the seismic events take place during refrac cycle RF1 and RF2 with more intense color saturation as they are located further away from the injection point.

5
The mean of refrac cycle RF1 to RF3 are very consistent for experiment HF8.  microradians. The magnitude of the tilt signals decreases with respect to the injection location and shear zone as follows.

Hydromechanical observations
Injection executed south of S3 show in general smaller magnitudes than the one executed next to S1, as they are farther away from the tilt meter locations. Also, the response of the tilt signals reacts either instantaneously or delayed depending on the location of injection (distance-controlled response). The experiments next to the shear zone S1 in borehole INJ1 (HF1 and HF2) show an instantaneous response in the tiltmeter T1 and a delayed response in the y-component of tiltmeter T2. We interpret that the tunnel floor tilts away from the injection volume, which accumulates in the intersection zone of S1 and S3. This zone is located West of the injection point. During the experiment, tiltmeter T1 dips away from shear zone S1. Tiltmeter T2 tilts away from the zone S3 during the first refrac cycle and reorients itself with further injection (RF2) indicating a transient shift along the S3 shear zone.

5
Tiltmeter T1 of the experiments south of S3 (HF3 and HF8) indicates tilting of the tunnel floor away compared to the normal of S1.2 shear zone and T2 into the shear zone S3 indicating a small compressive component towards the zone S3.1. Experiments executed south of S3, are expected to connect to the fracture system of the shear zone S3, which acts as a preferential flow path towards the AU-tunnel. The tiltmeter signals are interpreted as secondary deformation fields, where T2 indicates transient movement of the S1.2 due to dextral shearing along S3. Tiltmeter T1 (located south of S3.2) has a delayed response 10 and indicates movement towards the shear zone S3.1 during the second refrac RF2 cycle. Then, it starts to reorient itself during shut-in and bleed-off time indicating mass diffusion along the preferential flow paths (fracture associated with zone S1.0). Reversible versus irreversible deformation for the tilt meter data will not be discussed in here due to long term changes observed in the time-series (i.e. tides, seasons).     Selected time series from pressure observation intervals in the injection and observation boreholes are presented in Figure 11 for the hydraulic fracturing experiment HF2 and HF3 (for the other experiments see supplementary information). The injection pressures (INJ) is presented on top including the grey boxes, which indicates fluid injection.
For the stimulations north of S3 (HF2) the time series from the interval below the injection (INJBE) shows a pressure arrival directly at the end of the first refrac cycle RF1. The peak of the pressure response in PRP1-1 is at the end of refrac cycle 5 RF2. The interval PRP2-2 is located in the shear zone S3 and the injection location is next to the S1 zone. It is expected that the fractures related to the S1 zone are connected first and therefore the interval PRP2-2 located in the S3 shear zone has a delayed pressure peak. The interval INJBE is located next to the injection interval (in shear zone S1.2 and S1.

Comparison between stress characterization and HF experiments
In the following section, we compare the hydraulic fracturing experiment with several small-volume (∼10 l) hydraulic fractures  breakdown pressure of our larger-scale HF experiments does also show a decrease towards the S3 and S1 shear zones, but the change in magnitude is significantly smaller with magnitudes around 21.2 to 16.3 MPa south of S3 and 14.9 to 13.9 MPa north of S3. The ISIP was measured during cycle RF2 for the experiments using only water and during cycle RF3 for the experiments using XSW. Generally, we find that the minifracs executed in SBH1 and SBH3, i.e. away from the shear zone, have larger breakdown pressures and ISIP than the experiments performed in SHB4, INJ1, and INJ2 that are closer to the S1 10 and S3 shear zones. In addition, the experiments performed in the more fractured rock mass in the vicinity of the shear zone show larger variability reflecting stress heterogeneity.

Fracture geometry
In the following, we compare the number of localized seismic events between the HF and MF experiments. Then we compare the geophysical borehole logging/impression resulting in fracture traces at the wellbore with best fitting planes through the -On average, the total number of located seismic events do not differ significantly between the MF and HF experiments, although the variability from experiment to experiment is large. This is somewhat surprising knowing that about 10 l was injected during the MF experiments and 1000 l during the HF experiments.
-The variability of the number of seismic events is particularly obvious when comparing MF6, MF7, HF5, and HF6 (all 10 experiments exhibit less than 50 events) with for example HF2 (531 events). Note that it cannot be ruled out that the different seismic network layouts for HF and MF affect the sensitivity towards detecting lower magnitude events and thus can influence the result.
-However, looking at frequency-magnitude distributions and the corresponding magnitude of completeness analyses presented by Villiger et al. suggests that for the HF injection experiments, not all the variability in detected and located 15 seismic events can be attributed to sensitivity variations of the seismic network. A possible explanation for the low number of located events during MF6 and MF7 is the fact that they were close to the shear-zone S3. During injection experiment HF5 a direct short-cut to a geophysical observation borehole was created. Injection experiment HF6 was executed at the wrong borehole interval, where a pre-existing fracture was stimulated, which was already stimulated before during the hydraulic shearing (HS) experiment.
-During the MF experiments, typically a small number of events occur during the formation breakdown cycle (F) compare to the refrac (RF) cycles, except for MF4 and MF5. MF5 had an insufficient sealing of the experiment section allowing the fluid to by-pass the packer. Nevertheless, the seismic response is high and the experimental summary cards (supple-5 mentary information) indicate that most of the seismic events are located around the borehole up to 4 m away from the injection point, with some minor events located 7 to 10 m away from the injection interval.
-During the HF experiments, most seismicity occurs during the refrac cycles (RF) whereby the injected volume and the flowrate progressively increase. Except for experiment HF6, no seismicity was observed during the pressure-controlled step (SR) test. Prior to this experiment, we opened the valve of the injection interval to drain the fracture system. The 10 small injection volume during the pressure-controlled step test and the small flowrates were not sufficient to re-initiate micro-seismic activity.
The poles to the fracture traces determined at the borehole wall from acoustic televiewer data (see Fig. 5) in the boreholes INJ1 and INJ2 for the HF experiments are presented in Figure 14a and the pole to the fracture traces from the minifracs using impression packer are presented in Figure 14b. We assume, that these traces correspond to the fractures initiated during 15 the frac-cycle (formation breakdown). The fractures during HF experiments presented in Figure 14a are sub-vertical with a N to NE dipping direction. The HF traces are axial or make a small angle to the injection borehole axes. The foliation orientation (337/15 • ) is indicated by the magenta pole in Figure 14, which is also corresponding to the main brittle fracture set. The minifracs MF8-MF11 were executed in the sub-vertical borehole SBH1. The orientation of the minifracs (Fig. 14b) are primarily aligned with this foliation plane except for MF09 and MF10 that have a similar trace orientation than for the HF 20 experiments. The minifracs MF08 and MF11 orient towards the foliation and the brittle fracture orientation in the host rock (compare Fig. 2a), such that the fracture opened either along the foliation. The orientation of fracture trace MF01 and MF02 from the sub-horizontal borehole SBH3 show a radial fracture initiation, which highlights the dominating control of foliation as such orientation is not the most favorable for initiation at the borehole from a stress concentration viewpoint.
The pole points from the plane fit to the seismic clouds are presented in Figure 14c for the HF experiments and Figure 14d for

Fracture propagation
Fracture propagation during the fluid injection can be tracked using the seismic events, which move away from the injection interval. In Fig. 15, all seismic events from the minifracs (MF) are presented with grey circles. The grey circles are absent after 10 l of injected volume as it is the maximum injected volume for these experiments ( Figure 15).  interval. The pressure data point from experiment HF8 has a distance of ∼9 m and respond after 20 l of injection. At the same volume, the seismic cloud indicates a distance to injection of around 8 m. Considering the strain and pressure measurements from experiment HF2, we see that the distance of the seismic events is two times smaller compared to a strain sensor. It is remarkable that the seismic events are restricted to within 10 m distance at high-pressure injection even though we know the hydraulic fracture is 15 m in length (HF2 strain sensor at 15 m). This agrees well with the observation of Warpinski et al. 15 (2013) that the hydraulic fracture is essentially aseismic and that the stress-induced microseismicity is located at fractures with a significant dimension around the hydraulic fracture.

Hydraulic and mechanical response to hydraulic fracturing
Hydraulically, two different behaviors were observed in experiments performed south of S3 compared to experiments performed north of the S3 shear zone. The differences consisted of lower pressure levels (jacking and ISIP), larger recovered water volumes, and larger final injectivity for experiments north of S3. Our proposed explanation of this behavior is as follow.

5
For the experiments HF3 and HF8 (both south of S3), the injected fluid was drained by the densely fractured S3 shear zone towards the AU tunnel, where the S3 shear zone is bounded by two metabasic dykes and therein two subnormal fracture systems are found. The injection related to this geological feature (south of S3), which acts as strong hydraulic boundary condition, therefore, be described as an open system. Thus, no fluid was recovered through the injection borehole for these injection tests (less than 2%). This configuration also favored direct flow connections reflected by instantaneous responses in pressure or decrease in negative strain (Fig. 10). For the injection locations north of S3, tensional signals are observed in the S1 zone and either tensional or compressional signals can be observed in S3.
The injected fluid for the experiments executed north of S3 (HF1, HF2, and HF6) was to some extent stored in the fractures associated with the S1 zone. The fluid recovery from the injection location ranges between 15.0-23.5% for the first two refrac 20 cycles. The flow field is complex and extends most favorably towards the S3 and S1 intersection zone, where the highest fracture density is observed. This is supported by the tilt signal and the FBG-sensors, which have bigger tilt and smaller strain signals than for experiments executed south of S3, where the FBS-boreholes are closer to the injection location and the tiltmeter farther away. Injecting north of S3 (HF2), the FBG sensors indicate tension in the S1 zone along borehole FBS1 and in the S3 shear zone along borehole FBS3, but compression in the S3 shear zone along the FBS1 borehole (Fig. 10). There, a flow path 25 exists between the S1 and S3 shear zone in the lower section of the volume, but the injected fluid volume is either too small or the connectivity is high such that the FBG sensors along borehole FBS1, which is located further above, show a compressive signal.

Borehole trace and fracture tortuosity
The comparison of the fracture trace orientation at the borehole wall observed by acoustic televiewer logging with the orien-30 tation of the seismic cloud associated with a given fracture allow assessing fracture rotation, also referred to as tortuosity, in the near field of the borehole. An angular difference of about 30 • is typical for most of our experiments, which highlights the uncertainty in determining the stress orientation from direct wellbore information only. For the minifracs, the scatter in the 6.4 Permeability creation by HF All the performed hydraulic fractures generated a transmissivity increase of about three orders of magnitude. This increase remains after the pressure is relieved and in this sense is permanent, i.e. is not related to the transient fracture opening under high-pressure injections. This indicates that the fractures never completely close back when the pressure is relieved. The final transmissivity matches one of an unstimulated fracture in our rock mass . The final transmissivity correlates 5 positively with the final injectivity, which characterizes the newly created hydraulic fracture. Considering the injectivity value from the pressure-controlled step test has the advantage that it is directly related to the bulk mass compliance (rock mass compliance and fracture compliance) at short timescale. In addition, we can characterize the lift-off (herein: jacking pressure) of the intersecting fracture and the injectivity after jacking in the injection interval. The drawback is that the flow is only at a quasi-steady state. For transmissivity values, we used constant head injection and pulse injection tests at low pressure 10 (< 0.6 MPa). This method is time-consuming and it does not account for mechanical effects but has the advantage to reach steady-state conditions, which allow a characterization of the bulk volume farther away than the pressure-controlled step test.
Hence, both methods are mandatory to understand the change in the flow field and the permeability. Considering an EGS system with an injection and a production interval, then a certain injection pressure and probably a back pressure on the production interval is applied. Both injection and production intervals are able to change the hydro-mechanical parameters 15 of the nearby fractures. Then the injectivity value is able to describe this interaction, where the transmissivity describes the overall bulk volume of the connected fracture network. In our experiment, all the newly created hydraulic fractures (except HF6) were mechanically closed after the reservoir was depressurized. Low-pressure constant head injection was insufficient to open the closed hydro-fracture. Thus, the new created transmissivity is only linked to the new hydraulic fracture intersecting the wellbore. In term of scaling we estimate transmissivity only on single wellbore and no cross-hole tests were performed to 20 estimate the new fracture network transmissivity. Therefore, the presented values of the final transmissivity are rather small.

Stress heterogeneity
For HF2 and HF3, the initial plane fitted through seismicity cloud (compare Fig. 14c) dips to NE and E, respectively, but rotated at a later time to inclined South dipping orientation. One possible interpretation of this rotation is that the growth is initially controlled by the perturbed stress field as described in ( For experiments executed south or north of S3 the ISIP stabilizes around 5 MPa and only deviates slightly. The jacking pressure is very similar in magnitude compared to the ISIP for the experiments south of S3 (see Fig. 5). The situation differs for the experiment north of S3 as the magnitude of the jacking pressure is smaller. It reaches a value between 60-80% of ISIP.

30
The similarity of jacking pressure and ISIP is a strong indication that they reflect the minimum principal stress magnitude, but further analysis is necessary to rely on that, which is not part of this publication.
Approaching the shear zones S1 and S3 changes the stress observations significantly. For the HF experiments, it is best described by the perturbed stress field for small volumes of injection, i.e. frac and first refrac cycle. Fig. 16a shows the perturbed stress state as Mohr circles for the newly created hydraulic fractures from the ATV log (triangles) and the best fit plane from seismicity (squares). Assuming, higher perturbation during the breakdown cycle creating the new hydraulic fracture intersecting the borehole agrees well with the higher overpressure. Propagating the hydraulic fracture further away results in a decrease 5 in the overpressure in the perturbed stress field (Fig. 16b) but an increase in the unperturbed one (Fig. 16c). Therefore, the unperturbed stress state cannot describe the behavior of the hydraulic fracture at the early time. The orientation of the seismic cloud is sub horizontal towards South considering all located seismic events (i.e. RF2). This reorientation of the fracture is controlled by the leak off into the pre-existing fracture network striking in E-W direction. Note, that the density of pre-existing fractures is increased approaching the S1 and S3 shear zones. At this point, the best estimate of the stress state can be described 10 by the unperturbed stress field assuming smaller stress magnitudes. Fig. 16d shows the situation in terms of minor and major principal stress axis being reoriented approaching the shear zones. The perturbed stress field is a consequence of the preexisting fractures related to S1 and S3 and scatter the perturbed one. We assume this re-orientation of the fracture depends directly on the connectivity of the pre-existing fracture network.

15
In this paper, the spatial and temporal evolution of rock deformation, transient fracture fluid pressure, and seismicity during six intermediate scale hydraulic fracturing experiments are presented. One of the key findings of this work is that the fracturing processes are strongly influenced by site specific characteristics, natural fractures and local heterogeneities. This is to be expected in any rock masses since heterogeneities are always present in natural media. If the details of the observations presented in this paper are site specific, the overall processes and behaviour are likely reproducible at any site and we attempt 20 to formulate these general behaviours in the following conclusions: 1. The creation of new fractures at the borehole is clearly visible on borehole wall acoustic images. The orientation of the fracture traces are however variable and do not provide a good estimation of the independently measured far field stress orientation. The variability of the fracture trace could be explained by rock strength anisotropy . It could also be influenced by packer stresses or local stress heterogeneities induced by natural fractures as we observed 25 that the new fracture trace extend below the packer and often abut against natural fractures. Our data set doesn't allow us to determine if the fractures initiate below the packers or in the injection interval, although since the packer pressure is always maintained above the interval pressure to insure sealing, the former is not unlikely.
2. The growth of the hydraulic fractures is strongly influenced by natural fractures. This leads in the details to complex geometry departing from theoretical mode I fracture geometries. Our data highlights the simultaneous growth of parallel 30 fracture strands. It also suggest channelized growth (pipe-like geometry) instead of planar growth.
6. The heterogeneities lead also to distinct behaviour in terms of pressure responses and flow: tests performed North or South of the brittle-ductile shear zone S3 respond differently. We associate these differences principally with two effects.
Firstly, the shear zones have impacted the stress state locally and thus in turn this affect the fracture propagation. Secondly, the shear zones and the associated fracturing influence the flow in the experimental volume. When the hydraulic fracture grows, it connects and leaks into the pre-existing fracture network and thus at some point the energy required 5 to create new fracture surfaces isn't sufficient and tensile fracture stops. The flow is then largely dominated by the natural fracturing. In that regard, the brittle-ductile shear-zone S3 acts as main drain and constant pressure boundary while connection to the S1 shear-zone and associated fracturing provides larger storage possibilities. 7. A significant increase in transmissivity of 2-4 order of magnitudes from well test before and after HF was observed. The final transmissivity correlates positively with the final injectivity obtained from the pressure-controlled step tests. Such 10 permanent permeability increase is not consistent with pure mode I fractures in a homogenous media that would close back after depressurization. The heterogeneities and associated complexity of the hydraulic fractures probably favors the permanent transmissivity gain. The final transmissivities are comparable with the unstimulated natural fractures in our rock mass (Jalali et al., 2018b) which support the conclusion that the connectivity to the natural fracture system controls the final transmissivities. Competing interests. The authors declare that they have no conflict of interest. Nagra for hosting the ISC experiment in their GTS facility and to the Nagra technical staff for onsite support.