Abstract. The brittle damage around tunnels in the form of spalling is common in massive crystalline rocks subjected to high in-situ stresses. Understanding the short-term development and long-term evolution of the Excavation Damage Zone (EDZ) is particularly important for deep geological repositories (DGRs) for heat-generating nuclear waste, as the EDZ may provide preferential pathways for radionuclide migration. This paper presents an overview of the PRECODE mine-by experiment at BedrettoLab, a newly developed underground research laboratory (URL) in southern Switzerland designed to investigate brittle fracturing and EDZ evolution in crystalline rocks under repository-relevant conditions. The PRECODE tunnel was excavated in Rotondo Granite using both controlled drill-and-blast (D&B) and line-drilling and rock-breaking (LDRB) methods to evaluate the influence of excavation technique on EDZ characteristics. The experiment combines multidisciplinary in-situ monitoring including acoustic emission, distributed acoustic and strain sensing, hydraulic monitoring, electrical resistivity tomography, and terrestrial laser scanning with laboratory-based fracture mechanics and time-dependent testing under varying environmental conditions. The main objectives are to characterize short-term EDZ formation, investigate long-term crack growth and stress corrosion processes, quantify permeability evolution, and assess the interaction between excavation-induced damage and nearby fault zones. Preliminary results indicate clear differences in AE activity and EDZ development between excavation methods, while observations from the mature EDZ of the more than 40-year-old Bedretto Tunnel suggest ongoing time-dependent brittle fracturing. The coupled in-situ and laboratory datasets provide a unique basis for improving predictive models and long-term safety assessments for nuclear waste repositories in crystalline rocks.

We investigate how fluid pre-conditioning (FP) influences the reactivation and nucleation behavior of a critically stressed laboratory fault during fluid pressurization. Six reactivation experiments were performed on a saw-cut cylindrical sample of Rotondo granite confined at 20 MPa. Fault-parallel and axial deformation were monitored using distributed strain sensing (DSS) with optical fibers, complemented by active and passive measurements from sixteen piezoelectric transducers to track P-wave velocity variations associated with fluid migration and to detect acoustic emissions. FP substantially modifies fault reactivation behavior. Pre-conditioned tests exhibit a smoother and more spatially distributed reduction in effective normal stress at the onset of rapid pressurization, consistent with a relatively drained response at the fault scale. During the rapid pressurization ramp, FP tests show a subtle directional dependence in accelerated deformation, with slip preferentially developing where effective stress is reduced more strongly. This behavior is consistent with preferential fluid pathways inferred from ultrasonic velocity variations measured in separate fault-flooding characterization experiments. In contrast, non-pre-conditioned rapid pressurization is more consistent with a relatively locally undrained response, in which stronger effective-stress gradients preserve shear-resistance heterogeneity and promote slip barriers, foreshock activity, and more complex nucleation dynamics. These observations demonstrate that injection-driven reactivation is governed not only by pore-pressure magnitude but by the spatiotemporal distribution of permeability and effective stress established before and during rapid pressurization. Our results provide experimental constraints for models coupling pressurization regime, permeability heterogeneity, and rupture nucleation, with implications for safer geoenergy operations.

Piezoelectric actuators are widely used as transient sources in ultrasonic nondestructive testing, yet the force-time function they generate at the contact interface is rarely quantified experimentally or numerically. Here, we develop a multi-physics modeling framework that reconstructs both the magnitude and temporal shape of the output force, through integrated analytical and finite-element modeling validated by laboratory measurements. Modeled and measured/inverted forces show close agreement, confirming that contact stiffness and acoustic impedance govern force transmission. Extending the model to various contact materials predicts force amplitudes ranging from 0.14 to 0.92 N under 100 V excitation. The derived force-time function offers a validated physical representation of actuator/medium coupling, enabling a constrained input for waveform inversion and absolute sensor calibration.

Fat ray travel time tomography was used to obtain reliable high-resolution subsurface images in the geothermal testbed of the Bedretto Underground Laboratory for Geosciences and Geoenergies (BedrettoLab). The aim of the research was to better understand the relationship between structural features and the seismicity induced by hydraulic stimulation tests. Eight boreholes were used to provide a large data set comprising 41 881 manually picked first breaks. Our results demonstrate that the fat ray approach offers improved image quality compared to traditional ray-based methods. The 3D model was further validated using ground-truth information from wireline logs and geological observations. We successfully imaged a major fault zone (MFZ) that exhibits a complex structure including considerable heterogeneity. Relocation of passive seismic events generated during hydraulic stimulations indicates that the 3D velocity model has only a minor influence on hypocentral parameters. However, comparing a selection of well-constrained seismic events with the velocity structures revealed a remarkable spatial correlation. Most events occurred in regions of intermediate and slightly decreased seismic velocities, thereby avoiding both high- and very low-velocity zones. Based on small-scale laboratory studies, we speculate that these observations can be explained by the presence of stress gradients in the intermediate-velocity zones.

Our understanding of earthquake rupture processes is generally limited by the resolution of available observations. In all but exceptional cases, earthquake observations are made at comparatively large distances from the rupture itself, which puts a limit on what spatial scales can be resolved. At the same time, it is clear that small scale processes may play a crucial, if not dominant, role for various seismogenic processes, including rupture nucleation, co-seismic weakening and stress re-distribution. The Fault Activation and Earthquake Rupture ('FEAR') project aims at collecting and interpreting a multitude of earthquake-relevant observations from directly on and around the process zone of an induced earthquake. To this end, we attempt to activate a natural granitic fault zone in the BedrettoLab, at a depth of ~1km, after instrumenting the fault zone with a multi-domain and multi-scale monitoring system. The goal is to observe and study earthquake rupture phenomena in a natural setting, from unusually close distance. In this talk, we outline the project status, the science goals, and the plans for the main experiments, which are scheduled for the years 2024 - 2026. Notable milestones we report on include the identification and detailed characterisation of the target fault zone the beginning of niche and tunnel excavations laboratory experiments that characterise the frictional and mechanical behaviour of both gauge material and host rock of the target fault zone development of numerical models for 2D and 3D dynamic rupture propagation development of tailored monitoring methods for seismicity, strain, temperature, pressure, bio-geo-chemistry and other relevant observables development of remote experiment control methods test stimulations in a nearby rock volume of similar geology, with an already existing monitoring system, where we tested the influence of pre-conditioning injection protocols similar test stimulations in the same volume where we aim at triggering a larger event (target Mw~0) active seismic experiments in an underground salt mine, to calibrate the very- to ultra-high frequency (1k Hz - 500k Hz) acoustic emission sensors Together, these and other efforts constitute the necessary ingredients we need for interpreting the near-source observations that we will collect during the fault activation experiments.

A growing number of observations made using geodetic approaches have been able to detect large preparatory regions that experience accelerated deformation prior to and in close proximity to an earthquake’s hypocenter. An uptick in localized seismicity has also been observed in these regions and represents an opposite end-member of the spectra of deformation, in both space and time, from the opposite broad and slow process. If and how these preparatory observations are linked are not well understood. To study this, we conducted a triaxial experiment on a granitic rock sample instrumented with calibrated acoustic emission (AE) sensors and a distributed strain sensing (DSS) method using fibre optics. These two technologies were sensitive to seismic (100 kHz to 1 MHz) and aseismic (DC to 0.4 Hz) deformation at our sample scale and these were monitored as it was loaded and experienced brittle shear failure. DSS measurement allowed us to visualize the emergence of slow, heterogeneous strain fields that localized well before the failure of the sample. In the early stages of localized deformation, the regions exhibiting preferential damage were growing and doing so without producing seismicity. However, when approaching failure, these regions accommodating slow deformation began to accelerate and now produced clusters of seismicity. The cumulative seismic moment of the precursory seismicity was a fraction of the total anelastic deformation (< 0.1%) precluding the runaway dynamic failure. We also examined the clustering and frequency-magnitude distribution of the seismicity with respect to the localized strain field. In the later stages, moments prior to nucleation, the b-value begins to drop and becomes anti-correlated to the rapidly accelerating average volumetric strain rate measured using the DSS array. This observation better constrains the hypothesis that dilation of the relatively large preparation zone can host larger precursory earthquakes therein. These findings can help constrain models that better replicate the physics associated with the large spectrum of brittle deformation and will in turn help with our understanding of preparatory earthquake processes.

The conventional understanding of tectonic faults primarily categorizes them based on frictional behavior: stable due to velocity-strengthening (VS) behavior, or unstable owing to velocity-weakening (VW) that lead to seismic ruptures. This classification has traditionally led to the assumption that VS faults are unlikely candidates for earthquake nucleation. However, emerging evidence from recent laboratory experiments and field studies is challenging this simplistic view, pointing towards a more complex mechanism. In this study, we utilize a hydro-mechanically coupled fault model, which integrates VS friction governed by rate-and-state friction laws with dynamic weakening influenced by poroelastic effects. A key aspect of our findings is the impact of fluid injection on the mechanical state of the fault. This process decreases the effective normal stress and frictional resistance, initially paving the way for the propagation of an aseismic, slow-slip event. The transition from aseismic to seismic slip on VS faults hinges on the balance between shear-induced dilation and compaction. These opposing mechanisms respectively lead to a decrease and an increase in pore-fluid pressure, dictating the balance between fault stability or instability. Our results show that when the effect of compaction-induced pressurization surpasses the initial dilatancy phase, it enables the propagation of dynamic rupture as a solitary pore-pressure wave. Conversely, when dilation predominates over compaction, an aseismic slow-slip event propagates through the fault, maintaining stability and preventing rapid seismic activity. These findings advance our understanding of seismic risk associated with VS faults. They are especially relevant in the context of fluid injection practices in geothermal energy production and CO2 storage, demonstrating how such activities might activate faults that are considered nominally stable. Additionally, our results underscore the critical need for more experimental and theoretical investigations into shear-induced compaction as an efficient mechanism for fault self-pressurization, which plays a key role in leading to seismic instabilities.

Water infiltration into crustal rocks, particularly through fractures, significantly impacts seismic wave propagation and the characterization of fracture systems. Our study (Wu et al, 2023a) investigates the acousto-mechanical behavior of fractured granite experiencing gradual water infiltration over 12 days. We reveal an order of magnitude difference in wave amplitudes when compared to intact granite, with a correlation between wave amplitudes and the movement of the wetting front. The laboratory experiments show that fracture stiffness decreases exponentially as the wetting front advances, indicating moisture-induced matrix expansion (Wu et al, 2023b) around the fracture leads to increased asperity mismatch and reduced stiffness. By back-calculating the fracture stiffness and capturing the effects of water infiltration on seismic attenuation through a numerical model, this research illuminates how elastic waves propagate across fractures undergoing moisture-induced expansion, a crucial aspect of fracture characterization and understanding of the near-surface environment's response to hydrological changes. Our research sheds light on an important question in fracture characterization: how elastic waves propagate across a fracture undergoing moisture-induced expansion. Publications related to this research: Wu, R., Selvadurai, P. A., Li, Y., Leith, K., Lei, Q., & Loew, S. (2023a). Laboratory acousto-mechanical study into moisture-induced reduction of fracture stiffness in granite. Geophysical Research Letters, 50, e2023GL105725. https://doi.org/10.1029/2023GL105725 Wu, R., Selvadurai, P. A., Li, Y., Sun, Y., Leith, K., & Loew, S. (2023b). Laboratory acousto-mechanical study into moisture-induced changes of elastic properties in intact granite. International Journal of Rock Mechanics and Mining Sciences, 170, 105511. https://doi.org/10.1016/j.ijrmms.2023.1055

Faults in nature display complex surface characteristics, where fault asperities slip dynamically while other sections are more prone to creep. Fault roughness is critical in determining the contact conditions producing asperities, directly impacting their susceptibility to unstable sliding. Understanding the formation of asperities and how their seismic properties evolve with wear is critical for assessing slip budget and earthquake potential. In this study, a triaxial experiment was conducted on a cylindrical saw-cut sample of Carrara marble, to study its frictional evolution and strain response with wear. Initially the interface produced an audible, high stress drop ( 21 MPa) stick-slip event. A fiber-optic based distributed strain sensing (DSS) method was used to study the strain heterogeneities and showed that a central asperity was produced, causing the locking of the interface. The central asperity was explained by both the small curvature ratio rho c = 0.1% and the locking associated with a roughness drag induced by short wavelength fluctuations in roughness lambda min 1 mu m which should have promoted healing. After the audible stick-slip, the fault produced more stable frictional behavior with low magnitude stress drop events ( 2 MPa). We attributed this to (i) changes in the rate- and state- dependent frictional (RSF) parameters, (ii) decreased roughness-induced stiffness in the central asperity due to smoothing and (iii) the heterogeneous deposition of gouge formed due to wear. In terms of frictional stability, normal stress increase and smoothing would promote unstable sliding which was not observed. This led us to conclude that small amounts of gouge pushed the fault closer to the frictional stability line and even produced macroscopic velocity-strengthening behavior. Fault-parallel distributed strain measurements confirmed that the gouge allowed strain to be accommodated close to the interface, in contrast to the off-fault matrix that produced the large stress drop stick-slip event. These measurements also suggested that faults could simultaneously produce a range of frictional behaviors, with microscopically small stress drop silent events. These findings, together with the evolution of calcite-rich surfaces, highlight how roughness and its evolution can affect fault stability and drive strain heterogeneities due to gouge development and smoothing of the main slip surface.

Clay-rich rocks play an important role in critical practical applications, particularly as natural barriers in nuclear waste repositories and subsurface caprocks for CO2 storage. The interaction between electrostatically charged clay minerals and polar fluids (e.g., water) can lead to swelling or, under confined conditions, build-up of swelling stress. Fault closure by swelling in clay-rich rocks has been the focus of many studies. However, it remains unclear how water-clay interactions affect the stability of pre-existing faults, considering that in addition to changes in frictional properties, the stress state may also change due to the build-up of swelling stress. This study addressed this gap by conducting triaxial friction experiments on oblique saw-cut cylindrical samples. The upper half of the sample consisted of a clay-rich rock (Opalinus claystone) and the lower half of a permeable sandstone (Berea sandstone). The first set of experiments determined the friction slip envelope of the sandstone-claystone interface without fluid injection, at confining pressures ranging from 4 to 25 MPa, and a constant axial loading rate of 0.1 mm/min. These experiments showed a frictional strength well below Byerlee’s law, indicating that the Opalinus claystone dictates the strength of the two-material interface. Friction experiments with fluid injection were then performed at confining pressures of 10 and 25 MPa with a constant piston position (no axial loading) and an initial differential stress of about 70% of the expected yield stress. The aim was to compare the fluid pressures required to initiate slip in scenarios with and without fluid-clay interactions. For this, the experiments involved stepwise increases in fluid pressure through the injection of either deionized water (a polar fluid) or decan (a non-polar fluid). In one of the decane and one of the water injection experiments, fibre-optic strain sensors were attached to the sample surface. This allowed us to differentiate between poroelastic deformation within the matrix, deformation due to water-clay interaction, and elastic relaxation due to slip along the saw cut. The friction slip envelope based on decane injection experiments is within the uncertainty of the friction slip envelope based on the experiments with no fluid injection. In contrast, the water injection experiments indicate a weakening of the frictional interface. We interpret this weakening to be due to the transition of the claystone from a solid rock to a mud close to the saw-cut surface. This weakening was evident even at ambient fluid pressure, although the apparent stress state was below the yielding stress, indicating the need to consider swelling stress in initial water injection scenarios. In summary, our data suggest that water-clay interactions may reactivate pre-existing faults due to (1) the change of the frictional properties and (2) the build-up of swelling stress.

Advancing technologies to harvest deep geothermal energy has seen backlash related to unacceptable levels of induced seismic hazard during hydraulic stimulations. A thorough analysis of induced seismic hazard before these operations has recently become standard practice in the last decade. Additionally, more process understanding of the underlying causes of induced seismicity as well as novel approaches to develop geomechanical reservoirs are being explored in controlled underground laboratory experiments worldwide. Here, we present a probabilistic analysis of the seismic hazard induced by the ongoing hectometre-scale stimulation experiments at the Bedretto Underground Laboratory for Geoenergies and Geosciences (BULGG). Our workflow allows for fast updates of the hazard computation as soon as new site-specific information on the seismogenic response (expressed primarily by the feedback afb value and the Gutenberg–Richter b value) and ground motion models (GMMs) become available. We present a sequence of hazard analyses corresponding to different project stages at the BULGG. These reveal the large uncertainty in a priori hazard estimations that only reduces once site-specific GMMs and information on the seismic response of specific stimulation stages are considered. The sources of uncertainty are (1) the large variability in the seismogenic response recorded across all stimulation case studies and (2) uncertain GMMs on the underground laboratory scale. One implication for large-scale hydraulic stimulations is that hazard computation must be updated at different project stages. Additionally, stimulations have to be closely accompanied by a mitigation scheme, ideally in the form of an adaptive traffic light system (ATLS), which reassesses seismic hazard in near-real time. Our study also shows that the observed seismogenic responses in underground laboratories differ from large-scale stimulations at greater depth in that the seismogenic response is substantially more variable and tends to be weaker. Reasons may be lower stress levels, but also smaller injected volumes accessing a more limited fracture network than large-scale stimulations. Controlled underground laboratory experiments can contribute to improving our understanding of the the physical reasons leading to such variable seismogenic responses. The presented analysis implied that such experiments may be limited in terms of upscaling but are likely to be safe in terms of induced seismic hazard.

Gradual localization of deformation preceding catastrophic failure can produce precursory signals associated with a phase transition that may be present prior to earthquakes. However, due to the unclear origin of these precursors and the complexity of the environmental conditions, detecting such preparatory signals remains challenging. Here we present the spatio-temporal evolution of surface strain measured using fiber-optic sensing during triaxial experiments in wet and dry conditions. We identify a power-law distribution of strain increments where the largest magnitude diverges toward failure. This suggests a critical phase transition with the emergence of failure precursors. However, criticality is only observed in dry conditions and disappears with pressurized pore fluids, where the largest strain increment accelerates exponentially, consistent with a first-order transition. Our results highlight that progressive damage features criticality for failure prediction, but elevated fluid pressures may shift this behavior to abrupt rupture.

Recent observations of large earthquakes document the progressive localization of rock damage around future rupture zones that is also coupled with the spatial migration of foreshock sequences (Kato & Ben-Zion, 2020). This implies that the precursory deformation may act as a potential tracer for preparatory process that result in large earthquakes. It has also been observed that self-organization of the localized damage regions can govern the eventual macroscopic brittle failure in geomaterials (Renard et al., 2019). How the presence of fluid controls the self-organized precursory deformation along localized damage zone remains an open question. In this study, we have performed two triaxial compression experiments on dry and water saturated Berea sandstone, using distributed strain sensing (DSS) technology to visualize the strain field on the sample surface (Salazar Vásquez et al., 2022) with high spatial resolution. By tracking components of the strain field, specifically the region on the sample that sustained the largest incremental change in strain, we tested the effect of fluid on the predictability of phase transition between intact and failed state, under the context of critical hypothesis. Strain was progressively localized around the eventual faulting region for both samples, while a slow faulting was observed in the wet sample accompanied by a diffuse deformation pattern and unstable crack nucleation at failure. The results showed that, the failure in the dry sample was preceded by a critical power law acceleration of the largest increment, thus the dynamic faulting occurred in a well-defined singularity. The strain distribution also provided evidence for a predictable evolution of precursors. In contrast, the wet test showed evidence for a first-order transition with an exponential increase in largest increment, leading to an abrupt failure with a transient increase of strain. We interpreted this abrupt transition to be due to the increasing dominance of fluid-driven subcritical crack growth in the faulting. In this process, the local stress at crack tips decreases with crack lengthening, hence impeding the crack interaction and leading to an abrupt development of fault network. Our observation unravels the mechanisms of precursory deformation with fluid-assisted subcritical cracking, which has important implication in forecasting large earthquakes in nature. References:Kato, A., & Ben-Zion, Y. (2020). The generation of large earthquakes. Nature Reviews Earth & Environment, 2(1), 26–39. https://doi.org/10.1038/s43017-020-00108-wRenard, F., McBeck, J., Kandula, N., Cordonnier, B., Meakin, P., & Ben-Zion, Y. (2019). Volumetric and shear processes in crystalline rock approaching faulting. Proceedings of the National Academy of Sciences, 116(33), 16234–16239. https://doi.org/10.1073/pnas.1902994116Salazar Vásquez, A., Rabaiotti, C., Germanovich, L. N., & Puzrin, A. M. (2022). Distributed Fiber Optics Measurements of Rock Deformation and Failure in Triaxial Tests. Journal of Geophysical Research: Solid Earth, 127(8). https://doi.org/10.1029/2022JB023997

Understanding the strain field induced by earthquakes and aseismic slip events is crucial for interpreting deformation data, constraining source parameters, and calculating the induced static stress, which may lead to further seismicity. The Fault Activation and Earthquake Rupture (FEAR) experiments at the Bedretto Underground Laboratory for Geosciences and Geoenergies (“BedrettoLab”) provide a unique opportunity to directly observe and study strain fields resulting from fluid-induced slow and fast fault slip. Understanding strain evolution before, during, and after seismic and aseismic slip events is critical for identifying patterns of deformation localization and rupture nucleation, which are key to advancing physical models of earthquake processes and hazards. In two preparatory experiments before the main fault activation experiments, we attempted to trigger an ML~0.0 quake with model-informed fluid injection protocols. During the first experiment, after 4.5 days of injection with 15 MPa injection pressure, and 18 hours at 20 MPa of injection, we triggered an event with Mw~ -0.4, after which we stopped the stimulation. The observed co-seismic static deformation confirmed our monitoring network’s sensitivity to detect strain changes induced by earthquakes of at least Mw ~-0.4, laying the groundwork for subsequent strain modeling analyses.We modeled the displacement, strain, and stress fields resulting from the mainshock of the first experiment, treating it as a generic uniform dislocation source. To parameterize the source, we utilized a focal mechanism derived from P-wave first motion polarities and a radius estimate obtained through spectral fitting.Our analysis focused on comparing strain data from a network of Fiber Bragg Grating (FBG) sensors located at 18 to 31 m from the quake, with analytical models simulating the static deformation induced in an elastic full space by the earthquake. Preliminary results show a strong correlation between the observed and modeled strain, validating the reliability of our simple shear dislocation model. By optimizing the model parameters, particularly the fault's rake, we improved the fit between the predicted and observed strain profiles, refining our estimates of the M-zero earthquake’s source characteristics.Furthermore, our models enabled comparisons between the spatial patterns of seismicity and the stress fields induced by fault slip deformation—both seismic (Mw~–0.4 event, triggered in the M-zero experiment) and aseismic (triggered in the FEAR1 experiment). These comparisons contribute to understanding the mechanisms driving induced seismicity behaviors.By modeling co-seismic strain and validating these models against novel observational data, this work lays the groundwork for future strain inversions, including those targeting the pre-seismic phase, to better constrain the physical mechanisms underlying fault slip behavior. This work highlights the potential of underground facilities like BedrettoLab to deliver detailed insights into fault slip behavior, setting the stage for further analyses using data from the recently completed FEAR1 experiment.

The initiation of unstable fault slip leading to earthquakes involves intricate physical processes and interactions. Understanding these mechanisms is crucial for advancing our knowledge in earthquake seismology. Investigations at both field and laboratory scales have highlighted the existence of spatio-temporal variations in seismic or aseismic observations near the epicenter of a major seismic event, such as a rise in frequency of precursory earthquakes (Kato and Ben-Zion, 2021) or even strong fluctuations in seismic velocities (e.g., Campillo & Paul, 2003). These variations are often associated with the preparatory phase of major earthquakes believed to involve processes resulting from progressive localization of deformation around the eventual rupture zone that eventually accelerates leading up to failure. However, the time and spatial scales of this behavior are not well understood due to our lack of understanding into the physical mechanisms within the preparatory zones.In this study, we combined innovative laboratory techniques and numerical modelling to investigate (a)seismic preparatory deformation during a triaxial failure test in the laboratory. Employing distributed strain sensing (DSS) with optical fibers, we closely monitored strain rates on the sample surface. This was supplemented by active ultrasonic surveys and passive acoustic emission (AE) monitoring to investigate changes in P-wave velocity and locate regions prone to AEs within the sample. Using a physics-based computational model, we investigated strain localization within the sample by monitoring rock regions exhibiting high dissipation of mechanical energy. Highly dissipative regions spatio-temporally correlated with the observed AE locations and with sample regions experiencing P-wave velocity reduction. By further tracking the dissipation field within the sample, we recognized a system of conjugate bands that first emerged and quickly merged into a single band growing from the center towards the sample surface. The latter was interpreted to be related to the preparation of a weak plane. Shortly prior to failure, the model showed an acceleration of deformation that was also observed during the laboratory test with the DSS measurements and correlated with an increase of the seismicity rate in a similar volume of the sample. The combination of increased deformation and seismic rates mimics observations of precursory seismicity in nature. By methodically segregating the laboratory experiments from the numerical modeling, this study provides a comprehensive analysis of the physical processes underlying earthquake nucleation. The integration of cutting-edge laboratory techniques with advanced numerical modeling offers a novel perspective on the (a)seismic preparatory deformation that sets the stage for major seismic events. References:Campillo M., Paul A. (2003) Science 299, 547-549.Kato, A., Ben-Zion, Y. (2021) Nat Rev Earth Environ 2, 26–39.

Hydraulic stimulation has been extensively utilized in the geothermal industry as the primary technique to create and develop an efficient heat exchanger in a low-permeable reservoir rock. This technique entails high-pressure fluid injection under various injection schemes to perturb the local stress field at both borehole and reservoir scales, leading to permanent permeability enhancement of the stimulated volume through shear dislocation and dilation. These stress disturbances can also potentially trigger and/or induce seismicity in the reservoir and beyond. Understanding how different injection protocols serve as a preconditioning tool and their impact on the hydro-mechanical response of the stimulated volume would enhance our understanding of geothermal reservoir enhancement and induced seismicity mitigation. In the preparatory phase of the FEAR (Fault Activation and Earthquake Rupture) project in Bedretto Underground Laboratory (Switzerland), various injection protocols were utilized to understand the hydro-mechanical responses of the stimulated volume as well as earthquake rupture processes such as nucleation and premonitory slip. The adopted injection protocols include a) constant pressure injection for a specific time followed by step-rate injection and b) constant pressure withdrawal for a specific time followed by step-rate injection. In both protocols, an approximately equal amount of water (~3000 liters) was injected over the stimulation phase. Each injection protocol was associated with the pre- and post-characterization tests such as HTPF (hydraulic tests on pre-existing fractures) tests. Hydro-mechanical response of the host rock during these tests was monitored using various pressure, strain, and acoustic emission sensors in the injection and monitoring boreholes. At first glance, there appears to be no significant difference in the hydro-mechanical responses as well as the seismicity pattern of these two injection protocols, yet deeper investigation mostly based on the strain data reveals that strategy a) produced more heterogeneity in strain rate on the fiber-optic array whereas b) produced a more homogenized response. Numerical modelling and an experimental campaign in the laboratory are now underway to better understand the underlying mechanisms producing this response with the aim to best select a proper injection protocol for the goal of the FEAR project.

Faults in nature exhibit complex surface characteristics with patches of the fault (asperities) that may slip dynamically while other sections are more prone to creep (Beeler et al., 2011). Asperities forming in nature may be due to the geometric interactions between surfaces within a fault that contribute to complex stress states that are not well understood. Fault roughness is believed to play an important role in the control of the contact conditions established by asperities, directly affecting its potential to slip unstably. How the asperities are formed and how their seismogenic properties evolve due to wear is an important question with implications to slip budget and earthquake potential.In this study, we performed a triaxial experiment at sequentially increasing confining pressures (Pc = 60, 80, 100 MPa) on a saw-cut sample of Carrara marble. We analysed the quasi-static frictional response that benefited from novel arrays of distributed strain sensors (DSS) obtained using fiber optics. This sensor offered unique insight into the axial strain with a spatial resolution of 2 mm. The frictional behaviour during the first confining pressure step exhibited a dynamic instability in the form of a stick-slip event (SS) that produced a measurable stress drop. In the subsequent confining pressure stages, where an increase in confining pressure translated to increased normal stress, the fault behaved in a stable manner and no dynamic instabilities were produced. This observation is inconsistent with frictional stability theory (e.g. Rubin and Ampuero, 2005) and required pre- and post-mortem campaigns into the surface characteristics and their evolution to explain this abnormal behaviour. Therefore, we employed experimental techniques (pressure sensitive film (PSF), optical and stylus profilometry) along with finite element (FE) model in ABAQUS to characterize the pressure and roughness.The DSS array showed extensional axial strain closer to the edges of the fault, while only compression was expected in this triaxial loading test. The pre-experimental profilometry revealed an asperity located at the centre of the fault with a curvature ratio of h/L=0.1% inherited from the hand-lapping preparation, which dominated the initial contact conditions prior to the SS and explained the DSS observations. The DSS results were confirmed using a FE model which justified the effect of the fault geometry (h/L) on the strain response. After the SS, wear and smoothening of the central asperity was seen in roughness measurements. The profilometric measurements showed that gouge was deposed adjacent to the high normal stress asperity center (PSF) and were characterized by increased RMS roughness. These small amounts of gouge on the fault surface were sufficient to suppress the seismic response of the asperity. These findings show that the seismic potential of a carbonate (softer) asperity, may be highly influenced by the debris produced during wear. Its impact on earthquake nucleation could provide insight into large-scale earthquake preparation processes on carbonate faults in nature. References:Beeler, M., Lockner, D. L. and Hickman, S. H. (2001), Bull. Seis. Soc. Am., 91 (6): 1797–1804 Ampuero, J.-P. and Rubin, A. M. (2008), J. Geophys. Res., 113, B01302

Abstract The application of machine learning techniques in seismology has greatly advanced seismological analysis, especially for earthquake detection and seismic phase picking. However, machine learning approaches still face challenges in generalizing to data sets that differ from their original training setting. Previous studies focused on retraining or transfer‐learning models for these scenarios, but require high‐quality labeled data sets. This paper demonstrates a new approach for augmenting already trained models without the need for additional training data. We propose four strategies—rescaling, model aggregation, shifting, and filtering—to enhance the performance of pre‐trained models on out‐of‐distribution data sets. We further devise various methodologies to ensemble the individual predictions from these strategies to obtain a final unified prediction result featuring prediction robustness and detection sensitivity. We develop an open‐source Python module quakephase that implements these methods and can flexibly process input continuous seismic data of any sampling rate. With quakephase and pre‐trained ML models from SeisBench, we perform systematic benchmark tests on data recorded by different types of instruments, ranging from acoustic emission sensors to distributed acoustic sensing, and collected at different scales, spanning from laboratory acoustic emission events to major tectonic earthquakes. Our tests highlight that rescaling is essential for dealing with small‐magnitude seismic events recorded at high sampling rates as well as larger magnitude events having long coda and remote events with long wave trains. Our results demonstrate that the proposed methods are effective in augmenting pre‐trained models for out‐of‐distribution data sets, especially in scenarios with limited labeled data for transfer learning.