The dominant stimulation mechanism in EGS has been identified as induced shearing of pre-existing fractures and faults (referred to as hydraulic shearing, HS, and mode II and/or mode III dislocations) (Fehler, 1989; Kelkar et al., 2016; Pine and Batchelor, 1984). Here, the fracture fluid pressure needs to be enhanced above shear strength of the pre-existing discontinuity but may not exceed the minimal principal stress magnitude. A prerequisite for shearing is the existence of discontinuities that support a subcritical level of shear stress.

Another stimulation mechanism is the formation and propagation of new tensile fractures (also known as hydraulic fracturing, i.e., mode I opening) in intact rock (Economides and Nolte, 1989). Hydraulic fractures (HFs) tend to grow perpendicular to the minimum principal stress component (Haimson and Fairhurst, 1969). The propagation of a HF can be inhibited if its fluid leaks off into pre-existing fractures that are critically stressed and are intersected by the propagating HF (McClure and Horne, 2013). In addition, the decrease in fracture fluid pressure with increasing distance possibly restricts HF to areas near the injection interval (Dutler et al., 2019). Due to the geologic complexities of the targeted reservoirs, mode I and mode II/III fracturing may occur simultaneously (McClure and Horne, 2014a; Krietsch et al., 2019). During both stimulation mechanisms the driving force is the reduction in normal stress across the pre-existing or induced discontinuity due to fracture fluid pressure enhancement. Induced seismicity may be triggered within this zone affected by fluid pressure diffusion but also beyond. Possible mechanisms for a far-field response may be related to poroelastic stress transfer (Goebel et al., 2016, 2017; Goebel and Brodsky, 2018) or slip-related Coulomb stress redistribution (Catalli et al., 2016; Schoenball and Ellsworth, 2017).

Considering the injected fluid volume and the maximum observed magnitude (Mmax_obs) for different case studies (Fig. 1) reveals an important observation: for a given injected volume (e.g., 10 000 m3) the maximum observed earthquake magnitude may reach from Mw -1.0 to Mw > 5.0. However, an important issue for EGS sites is the a priori assessment of seismic hazard and risk, which typically includes the forecasting of seismicity rates defined by the seismogenic index (Shapiro et al., 2010) – also called activation feedback a value (Broccardo et al., 2017; Mignan et al., 2017, 2019), the Gutenberg–Richter b value and the maximum possible earthquake magnitude (Mmax_pos). To advance EGS technology, it is essential to better understand the physical mechanisms responsible for the large variability in seismicity across multiple stimulation projects on variable scales and find strategies to promote low levels of seismicity.

Dinske and Shapiro (2013) and Mignan et al. (2017), among others, have shown that seismicity rates might be linked to the geologic setting. These observations might make the prediction of seismicity rates and Mmax_pos highly site-specific. McClure and Horne (2014b) relate the formation properties observed in the wellbores of six field-scale hydraulic stimulations in granitic rock to the severity of the seismic response. They suspect that there is a correlation between fault maturity (i.e., well-developed brittle fault zones) and high seismic moment release. Also, De Barros et al. (2016) suspect that the seismic behavior to fluid injection is dependent on the fault damage zone architecture. Gischig (2015) further suggests that the seismic activity depends on the stress conditions along faults. He concludes that optimally oriented faults may rupture in an uncontrolled fashion beyond the pressurized volume and stop where geological conditions change. This implies that seismicity may also be produced outside the pressurized zone (Guglielmi et al., 2015). In contrast, ruptures along less favorably oriented faults have a larger portion of aseismic slip, and slip arrests within, or only a little beyond, the pressurized volume.

Some studies focus on injection strategies that may reduce induced seismicity. Yoon et al. (2014) and Zang et al. (2018) suggest that a fatigue hydraulic fracturing injection scheme, including cyclic injection pressure, may lead to a systematic reduction in Mmax_obs and an increased hydraulic performance when compared to conventional monotonic high-pressure fluid injection. Although many alternative injection strategies are widely discussed in the literature (e.g., McClure et al., 2016; Zimmermann et al., 2014; McClure and Horne, 2011), experimental evidence for advantageous injection schemes are difficult to obtain, as it is not clear to what degree geological conditions or the injection protocol are responsible for variable seismicity outcomes.

In this paper, we present observations of induced seismicity during 12 hydraulic stimulation experiments (i.e., six HS and six HF experiments) in a decameter-sized volume in crystalline rock at the Grimsel Test Site (Amann et al., 2018). These experiments share many of the research goals of recently performed stimulation experiments at the Sanford Underground Research Facility (SURF; Kneafsey et al., 2018) in the US as part of the Collab project (Schoenball et al., 2019) and a series of HF experiments conducted at the Äspö Hard Rock Laboratory, Sweden (Zang et al., 2016; Kwiatek et al., 2017). However, we focus on investigating the influence of the local geological conditions, in connection with the prevailing stress field, on the seismic response to high-pressure fluid injection. To maintain consistency between the stimulation experiments, standardized injection protocols were used for our HS and the HF experiments, respectively. After describing the in-depth characterized experimental volume with respect to geology (Krietsch et al., 2018b) as well as in situ stresses (Krietsch et al., 2018b; Gischig et al., 2018; Jalali et al., 2018), we detail the main methods used throughout this paper. Then, we show how seismicity evolved temporally and spatially in the experimental volume. Later, we estimate statistical properties of the induced seismicity, which allows comparison of the seismic responses of the different experiments. Then, we estimate injection efficiencies and the ratio of seismic to aseismic deformation. Finally, we discuss the findings in a broader context and close with implications for managing induced seismicity risk in future projects drawn from the results of the performed injections. This paper focuses on the seismic response, which is linked to the hydromechanical observations during the six HS experiments (Krietsch et al., 2020), the six HF experiments (Dutler et al., 2019), and the permanent changes in the hydraulic behavior of the reservoir (Brixel et al., 2020).

A standardized injection protocol was used for the six HS experiments, to compare the influence of the targeted geological structures on the seismo-hydromechanical response. Roughly 1 m3 of fluid volume was injected per experiment (actual volumes are given in Table 1). The injection protocol consisted of four injection cycles (referred to as C1–C4, Fig. 4a), in which either the injection pressure or injection flow rate was increased in a stepwise manner after steady state was reached. All the cycles were followed by a shut-in phase, where pumping was stopped, and a venting phase, in which the pressure in the injection- and all monitoring-intervals were bled off. The first two pressure-controlled cycles C1 and C2 were conducted to determine pre-stimulation jacking pressure (i.e., the injection pressure at which the ratio between the injection pressure and flow rate deviates from linearity) and initial injectivity of the target structure. C3 was the actual flow-controlled stimulation cycle, in which the bulk part of the fluid was injected. C4 was initially pressure controlled, changing to flow-controlled injection, aimed at determining the post-stimulation jacking pressure and injectivity of the targeted structure. During all HS experiments, the flow rate did not exceed 38 L min-1.

HF experiments also followed a standardized injection protocol involving a target injected volume of ∼1 m3 (actual injected volumes in Table 1). The injection protocol for the five HF experiments started with a flow-controlled formation breakdown cycle (indicated by the letter F) to initiate the hydraulic fracture. This initial cycle and all the subsequent cycles included a shut-in phase were pumping was stopped. During some of the formation breakdown cycles the shut-in phase was complemented by a bleed-off phase of the injection interval and all pressure monitoring intervals. The two subsequent cycles were aimed at propagating the previously initiated hydraulic fracture (RF1, RF2). For these two propagation cycles, water was used during HF1, HF2, HF3, and for HF5, HF6, and HF8, shear-thinning fluid (xanthan–salt–water mixture, XSW) was used. We note that the XSW mixture exhibited a viscosity of ∼35 mPa s (viscosity of water =1 mPa s). Propagation cycles RF1 were performed with maximum flow rates of 35 L min-1. During experiments performed with water, the flow rate was controlled in a sinusoidal fashion (period: 2.5–20 s; amplitude: ±15 L min-1) for roughly 10 minutes. For experiments in which XSW was injected, an additional cycle, RF3, was added to inject water so that XSW is flushed from the fracture network. All the HF experiments were finalized by a pressure-controlled step-rate injection test (SR) for evaluating post-stimulation jacking pressures and injectivities of the created hydraulic fracture.

For the HS injection experiments, most events (12 211 from a total 13 826 detections) were detected during pumping phases (Fig. 4a, b, for a selection of HS experiments). The percentage of events recorded during shut-in were in the range of 10 %. Less than 2 % of events were detected during venting. Comparing the HS experiments, significantly fewer events were detected during experiments at the S1 shear zones compared to the stimulation experiments performed at the S3 shear zones (Table 1). An exception was HS8 at the S1 shear zone south of S3, which produced a number of events comparable to the S3 injections (i.e., total detections: 3703). This may be explained by the fact that the injected fluid entered the S3 shear zone, which was evident from the seismicity cloud migrating towards the S3 shear zone (see Sect. 4.2). The number of seismic events (normalized to the total number of events per experiment) is plotted against injected volume in Fig. 5a and b. Again, a distinct behavioral difference between S1 and S3 injections is observed. During experiments in S1, the largest seismic detection rate was observed during stimulation cycle 1; more than 50 % of all events were induced with less than 100 L of fluid (< 10 % of the total volume). On the contrary, for S3 stimulations, most events were detected during cycle 3, when the largest volume of fluid was injected. Again, experiment HS8 is an exception in that the highest detection rate was observed during cycle 1 (similar to S1 stimulations), after which the event rates behave similarly to the S3 injections (HS4, HS5). Generally, a larger fraction of seismic events occurred after shut-in during injection into the S1 shear zones compared to injections into the S3 shear zones.

Over all HF injections, about half of the detections were made compared to the HS injections (see Fig. 4c, d, for a selection of HF experiments). Most of the events were detected during the pumping phases (4483 of 6731 detections). Interestingly, a comparably high percentage of detections (33 %) were made during shut-in, and no events were detected in the venting phases. We argue that the high percentage of post-shut-in detections were related to a hydraulic connection created between the injection interval and the open seismic monitoring boreholes (termed GEO) during the last two experiments HF5 and HF8. This hydraulic connection allowed observable flow from the GEO boreholes into the tunnel. We assume that this flow triggered stick-slip movements of the AE sensors. Thus, ongoing flow through GEO boreholes after shut-in would explain why many post-shut-in events were detected. Also, most of these events were only detected at the two sensors in the GEO borehole which was hydraulically connected. Note that HF6 – by mistake placed across the S1.3 shear zone close to the injection interval of HS1 – can be seen as a continuation to the HS1 experiment.

In summary, for the HS experiments, 31 % (i.e., 4342) of detected events could be located. The fraction of located shut-in events during the HS experiments is around 3 %, the fraction of events induced during the venting phase is less than 1 %. For the HF experiments, because of the large number of events without seismic origin (possibly sensor stick-slip), only 12 % (i.e., 781) of all detected events could be located. Six percent of the events were located after shut-in, and no events were located during the venting phases. The located seismic events fulfill a location uncertainty below ±1.5 m (for more information on location uncertainty see Sect. 3.2.3).

The maximum induced magnitudes Mmax during both HS and HF experiments (see inset of Fig. 1 and yellow stars in Figs. 4 and 5) occurred during pumping with no evidence of a temporal trend. Events during a time interval between shut-in and the start to a new injection cycle were usually of lower magnitude. One exception was the injection experiment HF6; here the highest-magnitude event was induced during a shut-in phase (see Sect. S3).

The Gutenberg–Richter a and b values are estimated for partial catalogs of respective injection experiments defined by the magnitude of completeness, MC. The latter was determined per experiment using the goodness of fit method introduced by Wiemer and Wyss (2000). a and b values and their uncertainty are calculated using the modified maximum likelihood technique published by Marzocchi and Sandri (2009). a values are normalized by the injected volume to derive the so-called seismogenic index, Σ (Dinske and Shapiro, 2013). Figure 10 shows frequency–magnitude distributions (FMDs) of all injection experiments. MC and a and b values were estimated for injection experiments exhibiting more than 20 seismic events and a goodness of fit quality of more than 90 %. Exceptions were made for injection experiment HS3 and Cluster 3 of experiment HS4, where the goodness of fit quality lies above 85 %. MC is lowest for injections in the focal point of the seismic network (HS4: -4.90; HS5: -4.80; HF2: -4.78). For injection experiment HS4, a bimodal frequency–magnitude distribution was observed. For a and b value calculations, the higher MC of -4.32 was used. MC increases for injections performed outside the network focus (HS3: -4.66; HS2: -4.39; HS8: -4.38) and is highest for the injection experiments performed towards the bottom of the second injection borehole (INJ2, HS1: -4.05) and towards the tops of the two injection boreholes (HF3: -4.14; HF8: -4.02;see Fig. 9). Thus, for these experiments the range between the maximum induced magnitude and MC is small. Moreover, when investigating spatial and statistical properties of seismicity clouds one has to be aware of the spatially varying network sensitivity. Our eight borehole AE sensors close to the injection intervals are decisive for an increased network sensitivity in the experimental volume close to the injection boreholes. In addition to the source–receiver distance, the sensitivity of the network is significantly influenced by the directivity of the AE sensor; i.e., events with incident angles > 50∘ in the Grimsel experiment are less likely to be detected.

The HS injection experiments (Fig. 10a) targeting S1 shear zones exhibited larger b values (HS1: 1.93±0.39; HS2: 1.69±0.26; HS3: 1.93±0.37) and lower seismogenic indices (HS1: -6.6; HS2: -5.8; HS3: -7.6) compared to the b values of injections into S3 shear zones (HS4: 1.36±0.04; HS5: 1.03±0.05) with higher seismogenic indices (HS4: -3.0; HS5: -2.4). Again, HS8 – an injection into the S1 shear zone south of S3 with migration of seismicity into the S3 shear zone – forms an intermediate case between injections into S1 and S3 with a b value of 1.61±0.12 and a seismogenic index of -4.9. The b value for the bimodal FMD of injection HS4 in a magnitude range of -4.9 to -4.35 lies below 1 as compared to 1.36 above magnitude -4.35.

The b value for the HF2 experiment (Fig. 10b) north of the S3 shear zones is comparatively low at 1.35±0.08, with a seismogenic index of -4.0. Experiments HF3 and HF8 south of the S3 shear zones at a similar depth of injection borehole INJ1 and INJ2, respectively, exhibited b values of 1.55±0.26 and 2.66±0.36. Seismogenic indices for the two injection experiments were -4.8 for HF3 and -9.0 for injection HF8.

A more detailed analysis of the bimodal FMD of HS4 reveals that the bimodal character does not disappear if the FMD is split up into all four injection cycles (Fig. 10c). Also for FMDs of individual seismicity clusters (see Sect. 4.2), the seismicity cluster closest to the metabasic dikes (Cluster 1) confirms the bimodal characteristic (Fig. 10d). The cloud subparallel to the metabasic dike (Cluster 2) shows a bimodal character but with a break in scaling at higher magnitudes compared to the FMD of Cluster 1. The new fracture induced and propagated during injection cycle 3 (Cluster 3) does not show the bimodal characteristic but reveals five events with higher magnitudes than would be expected.

The maximum observed magnitudes per injection experiments ranged over 1.5 magnitudes. The observed maximum magnitudes showed only a slight tendency to increase as the injected fluid volumes increased (Fig. 1), possibly owing to the fact that the injected volumes were only marginally different (900–1500 L). However, a stronger relationship was seen between maximum observed seismic magnitudes and the seismically activated area (Fig. 12a; for more information on the seismically activated area we refer to Sect. S5). Injection experiment HS5 represents the highest-magnitude event as well as the largest seismically activated area (285 m2). Also during injection experiment HS4 in which several planes were seismically activated resulting in a large seismically activated area, a rather large-magnitude seismic event was induced. There were no obvious differences in the maximum induced magnitude in relation to injected volume or seismically activated area between the HS and HF injection experiments.

Gutenberg–Richter b values and seismogenic indices show a high variability but no correlation with the seismically activated area (Fig. 11). Nonetheless, injection experiment HS5, during which the largest area was activated and the largest-magnitude event was induced, also shows the lowest b value and the highest seismic productivity. A comparatively small area was activated during injection experiment HF2 with similar low b values and high seismogenic indices.

In the following, we estimate the seismic moment release (referred to as M0 seismic) and compare it with a quantity termed hydraulic moment release (M0 hydraulic) as well as with the total moment release (M0 total) by stimulation experiment (Fig. 12).

The lower-bound estimate of M0 seismic during each injection experiment was determined by adding up the seismic moment of each located seismic event during the respective injection experiment. In order to estimate the experimental specific upper bound of the seismic moment release, the Gutenberg–Richter a and b values, determined in Sect. 4.3, were used to extrapolate the seismicity rates down to a magnitude of -9. Such small magnitudes were observed on the laboratory scale by Selvadurai (2019). Also McLaskey and Lockner (2014) and Yoshimitsu et al. (2014) observed very small magnitudes (i.e., M-7) and self-similarity down to these magnitudes. In situ, Goodfellow and Young (2014) observed magnitudes down to -7.5. For an average estimate of the seismic moment release, magnitudes down to a minimum magnitude of -6 were included (symbols in Fig. 12b). A high range of possible seismic moment release was observed for injection experiments with high b values (i.e., HS1, HS3, HF8), because the small-magnitude seismic events strongly contribute to the cumulative seismic moment release. Assuming the average estimate scenario and cumulating the moment release of all possible seismic events per injection experiment into a single earthquake would have induced a moment magnitude Mw in the range of -3 to -1. Assuming a stress drop of 1 or 0.1 MPa, respectively, and a source model by Brune (1970), this would correspond to a source radius of 0.3–2.2 m/0.6–4.8 m and a ruptured area of 0.26–15.5 m2/0.28–18 m2).

The equivalent hydraulic moment (M0 hydraulic) was calculated from the determined hydraulic injection energy. The hydraulic injection energy was estimated using Ehyd=∫pQdt, where p is the injection pressure, and Q is the injection flow rate, which are both integrated over the entire injection time. The pumped hydraulic energy is then converted to an equivalent seismic moment using M0=μΔσEhyd (Aki and Richards, 2002; De Barros et al., 2019), where μ is the shear modulus, chosen to be 30 GPa, and Δσ represents the static stress drop assumed to be between 1 and 0.1 MPa. The average estimate represents the equivalent seismic moment averaging the aforementioned stress drop range (Fig. 12c).

The total moment (M0 total) released by stimulation can be estimated from borehole dislocations in the injection interval that was determined from acoustic televiewer (ATV) measurements before and after each injection experiment (i.e., for injection experiment HS2: 0.95 mm; for HS4: 0.95 mm; for HS3: 1.25 mm; for HS8: 0.45 mm; and for HS1: 0.75 mm; see Krietsch et al., 2020). Note that this is only possible for HS experiments, since in the HF experiments no fault dislocations were observed (Dutler et al., 2019). For the estimate of the seismic moment from the measured displacements at the injection interval, we used M0=μAD, where μ is the shear modulus, again chosen to be 30 GPa, A is the seismically activated area determined in Sect. 4.2, and D is the average slip on the area of rupture. For a lower-bound estimate, we assume that an average slip over the entire lower-bound seismically activated area (i.e., the concave hull area; see Sect. 4.2) is 10 % of the observed slip at the injection interval. For the upper-bound estimate, we assume that the average slip across the entire upper bound seismically activated area (i.e., the convex hull area estimate) corresponds to 50 % of the observed slip at injection intervals. Twenty-five percent of the observed slip and 50 % of the estimated seismically activated area were used for the average estimate of total moment release (symbols in Fig. 12d).

To estimate seismic injection efficiencies (i.e., the ratio between seismic moment released to equivalent hydraulic moment, Fig. 12e) and the ratio between seismic and total deformation (Fig. 12f), the average estimates of the equivalent hydraulic and total moment were used. The cumulative seismic moment release was varied according to the minimum magnitude at which seismicity rates were extrapolated. When integrating to a minimum magnitude of -6, seismic injection efficiencies lie in the range of 1.9×10-6 (HS3) and 5×10-4 (HF8); injection experiment HS4 showed a high value of 1×10-4 with minor changes as the integration magnitude decreased, due to the low b value (i.e., due to the small contribution of small-magnitude events to the cumulative seismic moment). Seismic injection efficiencies (excluding experiment HF8) tended to converge to a value in the range of 1.6×10-5 (HF2) and 3.2×10-4 (HS1) when integrating to a minimum magnitude of -9.

The ratio between seismic and total moment release (Fig. 12f), considering events with magnitudes down to -6, ranged from 5.8×10-5 (HS3) to 3.6×10-3 (HS4). Integrating the seismic moment to a minimum magnitude of -9 leads to a convergence of the ratio between seismic and total deformation to values of 1.3×10-3 (HS3) to 1.8×10-2 (HS1).

We emphasize that the cumulative seismic moment, the equivalent hydraulic moment, and the equivalent total moment from dislocation observations are prone to a high level of uncertainty. Thus, uncertainties in the seismic injection efficiencies and the ratio between seismic and total moment give only crude estimates with uncertainties that possibly exceed 1 order of magnitude.

Remarkable is the large variability in the seismic responses between experiments conducted within less than a 25 m borehole length, which is expressed in the wide range of seismogenic indices (-9 to -2) and b values (1 to 2.7) (Fig. 11). The number of detected and located events during a stimulation depends on the detection ability of the sensor network, which is primarily a function of the distance (Mignan et al., 2011). However, even at a homogeneous completeness level of -4.02, the seismic response varies widely (Fig. 5c, d). Such variability is comparable to the variability between cases worldwide, involving both projects with predominant HF stimulation in the shale gas context and HS for geothermal exploitation (Fig. 13c, Dinske and Shapiro, 2013; Mignan et al., 2017). While Dinske and Shapiro (2013) suggest that there is a large difference in the seismic response during HF-dominated stimulations in shale gas projects and HS-dominated stimulations in geothermal applications, a systematic difference between the HS and HF experiments performed in crystalline rock was not discernible here. Also, the use of the shear-thinning xanthan–salt–water mixture during the HF experiments did not have an observable effect on the seismic response. The fact that the HF experiments were conducted in a rock mass where previous HS experiments could potentially have initiated some stress relaxation may explain the tendency for fewer events during the HF experiments. Experiment HF6, which can be interpreted as continuation of the HS1 experiment, induced only a few seismic events because the zone was stimulated twice. In contrast, the dispersed character of seismic events in the HF3 experiment may be explained by the interaction of new fractures with the surrounding faults S1.0, S3.1, and S3.2. We conclude that in our experiments HS and HF are similarly seismogenic, because HF strongly interacts with the pre-existing fracture network leading to similar seismic responses as the injections directly into pre-existing fractures.

While differences in the seismic response between HF and HS were not evident, the geological setting seems important for the substantial differences seen in the seismic response in terms of magnitude distributions as well as in terms of orientation and propagation of seismicity. We observed that experiments performed directly on or in the vicinity of the highly fractured brittle–ductile S3 shear zones (Fig. 13a, i.e., experiments HS5, HS4, and HS8 and HF3, respectively) are characterized by an enhanced seismic response. This observation is in agreement with the hypothesis gained from larger-scale stimulations, which states that well-developed brittle fault zones (i.e., connecting fractures that form larger features) lead to a comparatively high seismic moment release in response to high-pressure fluid injection (McClure and Horne, 2014b; De Barros et al., 2016). An exception is experiment HF2, which shows an increased seismic response with possibly no influence from S3 structures. Injection experiment HF2 was performed between the ductile shear zones S1.1 and S1.2, north of shear zone S3. At this location the reactivated structure (i.e., Cluster 1 and Cluster 2 of HF2; see also Fig. 7) may support an increased amount of shear stress, which led to an increased seismic response.

Not only do the seismic responses (i.e., b value and the seismogenic indices) indicate a strong geological influence but also the seismicity detection rate in relation to injected volume (Fig. 5) shows a different seismic footprint for the two shear zone types. For the injection experiments on the ductile shear zones (S1) more than 50 % of all detections are made during the injection of the initial 100 L of fluid. In contrast, the S3 shear zones experienced a gradual increase in detections with injected fluid volume (Fig. 5a).

The spatial distribution and propagation also appear to be affected by the geology. A concentric growth of seismicity clouds was rarely observed, indicating that the spatial fracture zone heterogeneity had a substantial impact. Seismicity clouds of experiments on ductile shear zones S1 show changing propagation directions and a planar character. Comparing the two S3 stimulations (HS4 and HS5), distinct differences in seismicity patterns were observed, even for stimulations within 3 m from each other in similar geological structures. During HS5, propagation directions changed along an extended seismicity cloud (of 16 m diameter) with a clustered character and regions of increased seismic event density. During the HS4 experiment the seismicity was mostly limited to patches/clusters within a 9 m radius from the injection interval but with a complex 3D and nonplanar architecture (Figs. 6, 8).

Beside their tendency of being very seismogenic, the highly fractured S3 shear zones stand out as being the most hydraulically conductive structures in the experimental volume compared to the less conductive S1 shear zones (see injectivities of HS4 and HS5 intervals in Fig. 13b). Injectivities at these intervals only increased marginally during stimulation. On the contrary, injectivities for the S1 stimulation experiments on the ductile shear zones and in the intact intervals increased by 2–3 orders of magnitude. Again, these observations agree with cases in the literature, for which the most permeable fractures were also found to be the most critically stressed and thus the most seismogenic zones (e.g., Barton et al., 1988; Barton et al., 1995; Barton and Zoback, 1998; Evans et al., 2005a; Davatzes and Hickman, 2010; Baisch et al., 2015; Hirschberg et al., 2015). It is also noteworthy that the injectivities for all experiments performed at the brittle–ductile shear zones, the ductile shear zones, and in the intact intervals end up having the same order of magnitude (Fig. 13b). While initial injectivities are highly dependent on the local geology, final injectivities are very similar (as are transmissivities; Brixel et al., 2020).

With the aforementioned observations in mind, it is possible to imagine what would have happened if a large open-hole stimulation would have been conducted in INJ1 and INJ2, as it was done in most of the previous EGS projects (e.g., Basel, Häring et al., 2008; Soultz, Evans et al., 2005b), instead of several stimulations at selected short intervals. Because of their high transmissivity, flow would have preferentially entered the shear zones S3.1 and S3.2 leading to induced seismicity, mostly dominated by the seismogenic properties of these structures. The result would have been a very limited transmissivity increase together with a strong seismic response. Thus, for larger-scale EGS stimulations, it appears quite promising to selectively stimulate multiple short borehole intervals with comparatively small fluid volumes (i.e., zonal isolation, Meier et al., 2015), during which the transmissivity of low-transmissive structures would be strongly enhanced, while stimulations in intervals at seismogenic fault zones should be avoided if possible. Of course, hydraulic stimulation of short intervals could also be combined with alternative injection schemes (such as described by Zang et al., 2017). However, the pronounced influence of geology on the aforementioned stimulation parameters in our experiments may imply that the impact of alternative injection strategies on induced seismicity (such as those discussed and proposed in the literature by McClure and Horne, 2011; Zimmermann et al., 2014; McClure et al., 2016; and Zang et al., 2018) is limited, since their effects are unlikely to emerge above the strong variability of orders of magnitudes imposed by the geological conditions.

Compared to the observed main orientation of the S1 (NE–SW) and the S3 (E–W) shear zones in the tunnels surrounding the experimental volume, the orientation of individual fractures within the S1 and S3 fault zones do show a similar WSW–ENE orientation. Also, the orientation of fractures found in the host rock are predominantly WSW–ENE in orientation with some random joint orientations. We combine the orientation of these pre-existing fractures with the slip tendencies inferred from the stress conditions measured 30 m south of shear zone S3.1 (i.e., the unperturbed stress state) and the stress conditions measured in borehole SBH4 (Fig. 2) in the vicinity of shear zone S3.1 (i.e., the perturbed stress state). It can be seen that there is an increased susceptibility for the S1 and S3 structures to slip (see also Krietsch et al., 2018a) when considering the perturbed stress state (Fig. 14a, b).

By including both the inferred orientation of the seismicity clouds or their clusters resulting from the injection experiments performed on the shear zones (i.e., the HS experiments) and the stress field, the combined influence of geology and stress field becomes evident. The predominant orientation of seismicity clouds is E–W, in agreement with the orientation of pre-existing fractures. Surprisingly, the predominant orientation also holds for the S1 stimulation experiments, even though the main orientation of the S1 shear-zones is NE–SW. Only the seismicity cloud of injection experiment HS1 is oriented in the main S1 direction. However, the orientation of seismicity in the E–W direction, also for S1 experiments, is not surprising when considering the fracture inventory of the experimental volume and the overlapping pole points of S1 and S3 structures, as well as the increased fracture density with the same orientation (Fig. 14a–d).

Hydraulic fractures in a strict sense, meaning fractures which form in intact rock, perpendicular to the minimum principal stress, at injection pressures higher than the minimum principal stress, are conceivable for the initiated fractures in experiment HF5 and the initial fracture (Cluster 1) of experiment HF2, oriented perpendicular to the minimum principal stress of the perturbed stress state where directional geological features are sparse. Cluster 2 of experiment HF2 formed at a later time compared to Cluster 1; it possibly formed because of the leak-off of fluids through Cluster 1 to the formation. The associated reduction in pore pressure through Cluster 1 may suggest a geology-dominated E–W orientation of the seismicity cloud of Cluster 2.

The new fracture created during experiment HS4 (Cluster 3) orients in a direction perpendicular to the minimum principal stress of the perturbed stress field (Villiger et al., 2019). We suggest that this fracture opens in connection to shear dislocation along shear zone S3.1 (Jung, 2013) induced during the HS4 injection.

In conclusion, the perturbed stress field – measured closer to the target rock volume than the unperturbed stress field – explains most of the observed seismicity cloud orientations well. HFs growing through intact rock tend to form normal to the minimum principal stress, while the other seismicity clouds are mostly guided by the predominant set of geological features that have comparably high slip tendency. However, it is likely that local stress variations may locally lead to a combination of opening mode deformation (i.e., mode I opening) and shear dislocation (mode II, mode III). Also, HFs show a strong tendency to connect with the pre-existing fracture network, which might explain why the seismic response during HF experiments is similar to the one during HS experiments.

Our experiments indicate that deformation in HS experiments (for which a displacement was measured at the injection interval) is to a large extent aseismic (i.e., < 2 % seismic). We also observed the tendency for the amount of aseismic deformation to be larger for experiments targeting the S1 structures (Fig. 12f). These overall values agree with values determined from hydraulic reactivation of a fault zone in limestone on a decameter scale, where 0.1 % to 3.9 % of shear deformation was estimated to be seismic (Duboeuf et al., 2017). Similar studies in shale materials report that less than 0.1 % of deformation is seismic (De Barros et al., 2016). An increased value of 4 % to 8 % for released seismic energy was reported for hydraulic fracturing experiments on granite samples at the laboratory scale (Goodfellow et al., 2015). Also, at the field scale, a large amount of aseismic deformation is suspected due to the observed slip dislocation of up to 4 cm on an acoustic televiewer log of an injection interval in granite at Soultz-sous-Forêts, which is much larger than the slip motion associated with the recorded seismic events (Cornet et al., 1997).