Report

Seismicity triggered by fluid injection–induced aseismic slip

See allHide authors and affiliations

Science  12 Jun 2015:
Vol. 348, Issue 6240, pp. 1224-1226
DOI: 10.1126/science.aab0476

How to observe fault injections in real time

Faults in the ground are known to deform in response to procedures such as wastewater injection that change the pore pressure. Guglielmi et al. took a crack at monitoring this process in real time with a controlled fluid injection into an inactive fault (see the Perspective by Cornet). Reactivating the dead fault induced aseismic slip, which triggered small earthquakes. These observations can inform models of how friction is related to slip rate. The technique can also be applied to field-scale monitoring of seismicity-inducing wastewater injections.

Science, this issue p. 1224; see also p. 1204

Abstract

Anthropogenic fluid injections are known to induce earthquakes. The mechanisms involved are poorly understood, and our ability to assess the seismic hazard associated with geothermal energy or unconventional hydrocarbon production remains limited. We directly measure fault slip and seismicity induced by fluid injection into a natural fault. We observe highly dilatant and slow [~4 micrometers per second (μm/s)] aseismic slip associated with a 20-fold increase of permeability, which transitions to faster slip (~10 μm/s) associated with reduced dilatancy and micro-earthquakes. Most aseismic slip occurs within the fluid-pressurized zone and obeys a rate-strengthening friction law Embedded Image with v0 = 0.1 μm/s. Fluid injection primarily triggers aseismic slip in this experiment, with micro-earthquakes being an indirect effect mediated by aseismic creep.

Injection of fluid into the crust can generate seismicity, with source mechanisms similar to natural earthquakes and occasionally large enough to cause damage (16). Induced seismicity is generally regarded as a manifestation of the effective stress principle, where the increase in pore pressure reduces the effective normal stress and brings a preexisting fault or intact rock mass to failure (7). Seismic slip requires failure to result in unstable slip, meaning that the resistance to sliding must decrease faster than the elastic unloading associated with fault slip. If a fault strengthens during slip, only aseismic creep should be possible. Only faults that weaken during sliding can, in principle, yield unstable slip. In general, the presence of fluids could either favor seismic slip, through thermal pressurization (8), or might impede seismic slip by a variety of other processes. In particular, dilatancy during slip could restrengthen the fault due to the resulting drop of pore pressure (9). Current models of earthquake nucleation (10) imply that slip becomes stable when the pore pressure approaches lithostatic, because the condition for stability (10) is then always reached. For slip to be unstable on a rate-weakening fault, the stiffness must be smaller than a critical value that has the effective normal stress at the numerator (10). If the effective normal stress goes to zero as pore pressure approaches lithostatic, this condition can never be met; slip should then be stable. Conditional stability might explain why aseismic slip has been inferred in some case studies (4, 11, 12). In reality, fault slip is coupled with fluid flow, and assessing whether seismic or aseismic slip should result from a fluid injection is a challenge.

We monitor a natural but initially inactive fault stimulated through a high-pressure injection of water, taking advantage of an instrument—the step-rate injection method for fracture in situ properties (SIMFIP) probe (13)—that allows the simultaneous measuring of fault-normal displacements, fault-parallel displacements, and fluid pressure. We selected a preexisting fault cutting through a carbonate formation accessible for examination and instrumentation thanks to an underground experimental facility located in southeastern France at a depth of 0 to 518 m (14). The fault is at least ~500 m long and trends 30°N 70°W. It cuts through porous grainstone carbonate layers (inner-platform Rudists’ rich facies), with strike-slip to normal cumulated slip of a few meters. The fault zone consists of subparallel fractures to 1- to 10-m length and discontinuous thin breccias (15). The in situ temperature at the location of the injection was 12.5°C and did not vary during the experiment. We therefore ignore thermal effects. We drilled a vertical well intersecting the fault at a depth of 282 m. Based on initial in situ measurements, the fault zone has an average initial permeability of 0.07 × 10−10 m2 and a bulk modulus of 10 to 17 GPa that are a factor of 25 higher and 2 to 5 lower than the surrounding host rock, respectively (16). We injected water into a 1.5-m-long chamber between two inflatable packers spanning the fault zone (Fig. 1A). We injected a total of 950 liters of water with step-increasing rates while monitoring pressure, flowrate, fault movement in both shear and dilation (Fig. 1B), and seismicity (16) (fig. S1). We estimated the state of stress from hydraulic tests on preexisting fractures conducted at two different depths in adjacent boreholes using the hydraulic testing of preexisting fractures (HTPF) protocol (17) and taking into account topographic stresses. The maximum principal stress σ1 = 6 MPa is subvertical and dips 80°S; σ2 = 5 MPa is subhorizontal and strikes 0°N 20°E; and σ3 = 3 MPa is subhorizontal and strikes 0°N 110°E (Fig. 1D).

Fig. 1 Experiment setting.

(A) Location of the injection interval and the seismometers (S1 to S3) in and around the fault plane. The experiment took place at a depth of 282 m below the Earth’s surface, in cretaceous limestone of the southeast France sedimentary basin. (B) Schematic representation of the SIMFIP probe displacement sensor. Deformation of the linkage tubes (colored in yellow, green, and blue) captures the relative displacement of the anchors clamped on the borehole wall in each fault compartment (16). (C) Core from the main slip plane of the fault showing slickensided surfaces, breccia/gouge-rich zones, and a fractured damage zone. (D) Stereographic view of the state of principal stresses, the fault plane, and the fault-parallel slip vector induced by the experiment.

Slip initiated at a pressure of about 1.5 MPa (S0 in Fig. 2) with fault-parallel slip approximately parallel to the shear stress on the fault plane (Fig. 1C), consistent with reactivation by shear failure. No seismic event was detected until about 0.3 mm of fault slip, about 1100 s into the experiment. During this first stage, the dilation rate was around 6 μm/s and exceeded the fault slip rate, which was on the order of 4 μm/s. The dilation rates fell to about 20% of the slip rate approximately at the onset of seismicity. The slip rate increased to about 23 μm/s. The seismicity consists of impulsive micro-earthquakes and tremors (fig. S1), as is observed in injection experiments elsewhere (18). About 80 seismic events were generated at an average rate of 15 events per min. Half of these events occurred between 1100 and 1190 s before acceleration of fault slip (Fig. 2), and the transition from aseismic to seismic slip was not associated with any pressure variation (point SS in Fig. 2). Both observations demonstrate that the majority of the seismic sources are located a few meters away from the injection point and may account for only a small fraction of the slip occurring on the fault segment where the fluid pressurization is occurring.

Fig. 2 Fault movements and seismicity induced over time by fluid injection.

Temporal evolution of fluid pressure (blue), injection rate (green), fault slip (solid black line), normal displacement (dashed black line), and cumulative number of induced earthquakes (red) during the 1400-s-long injection. Slip initiates at S0 = 400 s as pressure is increased to 1.5 MPa; seismicity initiates at Ss = 1100 s while there is no notable pressure variation.

The observation that the fault opens during the injection implies that permeability varies in close relation to the evolution of fluid pressure, as suggested in theoretical and experimental studies (19, 20). We used hydromechanical modeling (21) to determine the evolution of fault permeability (16). Our results show a 14-fold increase of the fault permeability from 0.07 to 1.0 × 10−10 m2 during the aseismic period, representing about 70% of the cumulative permeability increase (20-fold) during the injection (Fig. 3A). This permeability increase probably results from the opening of fractures during sliding, possibly reflecting the roughness of the fracture walls and/or cataclasis of the fault rock and associated microcrack dilatation in the early stages of failure. The model indicates that the pressurized zone (defined here as the zone where the pore pressure exceeds 0.5 MPa) increases to about 12 m during the initial stage of completely aseismic slip (Fig. 3C).

Fig. 3 Estimation of fault permeability and friction law from the in situ hydromechanical data.

(A) Observed cumulative fault-normal displacement and modeled permeability (blue circles) versus observed fault slip (black). (B) Friction (orange), calculated based on the measurements of the initial stress and fault slip, and moment release (purple) derived from fault slip. (C) Comparison of the estimated radius of the sliding zone and of the pressurized zone. (D) Comparison of observed (black) and modeled (red) fault slip for the best-fitting rate-dependent friction law. The best-fitting friction law of the form μ = μ0 + aln(v/v0) has μ0 = 0.67 and a = 0.0447, for a reference sliding velocity of v0 = 10−7m/s.

We use our measurements to estimate the friction law and to determine whether aseismic slip is primarily due to rate-strengthening behavior or whether it is due to fluid pressurization bringing the fault zone to the domain of conditional stability. We approximate the sliding zone by a circular crack and assume quasistatic equilibrium with a complete shear stress drop [justified by the net opening of the fractures and following (22)]. This simple model ignores the pore-pressure gradient and viscous forces associated with fluid flow. With these assumptions, it is possible to calculate the crack radius as a function of time (Fig. 3C and fig. S3A), as well as the changes of shear and normal stresses associated with fault slip and dilation (16) (fig. S3, B and C).

The model yields an accumulated moment at the end of the experiment of M0 = 65 × 109 Nm, (equivalent to a moment magnitude Mw = 1.17), far larger than the moment released by the incremental microseismicity, which was estimated to be lower than –2 in magnitude; so slip is mostly aseismic in this experiment (Fig. 3B). Micro-earthquakes are triggered only when the estimated size of the slip zone exceeds the pressurized zone (Fig. 3C), suggesting that they occur off the pressurized zone. However, they might occur within the sliding zone or be triggered off the sliding area by static stress increase. Aftershocks might relate to afterslip in a similar way (23).

The ratio of the shear stress to the effective normal stress increases from about 0.4 to about 0.8 and indicates a friction between 0.6 and 0.8 once slip becomes notable (fig. S4). Qualitatively, this evolution correlates with slip rate better than with slip. The data suggest a logarithmically varying, rate-dependent friction, as frequently observed in laboratory measurements of rock friction (24) or derived from studies of afterslip following large earthquakes (23, 25). We tested such a law based on a simple one-dimensional model (16). The model, which involves only two adjustable parameters, qualitatively fits the observations well (Fig. 3D). We determine the friction μ0 = 0.67 (+/−0.05) at a reference velocity of v0 = 0.1 μm/s, with a rate dependency of Embedded Image. We also tested rate-and-state friction laws (16) but found that the improvement was irrelevant in view of the uncertainties of the measurements. The friction coefficient increases to about 0.7 for slip rates of 1 to 20 μm/s. Similar values have been measured in the laboratory on faults formed in limestone and at comparable sliding rates (2628). These laboratory experiments also show a rate-strengthening behavior at temperatures less than 100°C but a rate dependency at steady-state typically one order of magnitude lower than the value we obtained.

The aseismic behavior triggered by the fluid injection in this experiment is apparently due to an intrinsically rate-strengthening behavior, rather than to conditional stable creep of a rate-weakening fault. However, the seismicity, which was probably triggered outside the pressurized zone, requires some areas to allow earthquake nucleation, hence weakening during deformation. This observation is an indication that the frictional properties are likely heterogeneous, as supported by the observations of the main slip surface that displays gouge-rich zones and slickensided areas cutting solid rock (Fig. 1C). During the injection, the effective behavior is rate-strengthening. This behavior is possibly due to the fault zone being rate-strengthening on average, with the gouge-rich zones being possibly more rate-strengthening than the zones where the fault cuts solid rock (27) or because the rate-weakening asperities are brought to conditional stability by the increase in pore pressure. This interpretation could explain why seismicity is observed only when the crack radius has become larger than the radius of the pressurized zone. Our results prove unambiguously that fluid injection can trigger primarily aseismic slip, with seismicity induced as a secondary effect. This is observed in the context of our experiment, which is characterized by a particularly low effective normal stress. Thus, our results may be of particular relevance to seismic activity triggered at shallow depth by human-induced injections and shallow aseismic slip. Mechanism observed in this experiment could also be of relevance to explain natural processes at greater depth, such as deep afterslip, slow-slip events, and tremors.

Supplementary Materials

www.sciencemag.org/content/348/6240/1224/suppl/DC1

Materials and Methods

Figs. S1 to S6

References (2932)

Database S1

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: The experimental work was funded by the Agence Nationale de la Recherche (ANR) Captage de CO2 through the HPPP-CO2 project and by PACA through the PETRO-PRO project. The rate-and-state fault models for this study were supported by the French Academy of Sciences, the California Institute of Technology Tectonics Observatory, and the ANR HYDROSEIS under contract ANR-13-JS06-0004-01. We thank the SITES S.A.S. engineers H. Caron, C. Micollier, R. Blin, and H. Lançon, and the Petrometalic S.A. engineer J. B. Janovczyk, who jointly developed and operated the probe that enabled this experiment. We also thank the Laboratoire Souterrain à Bas Bruit (LSBB) engineers’ team for their technical support during the installation of the experiment in one of the LSBB boreholes (www.lsbb.eu). Experimental data are available in the supplementary materials.
View Abstract

Navigate This Article