You are on page 1of 14

REVIEWS

Developments in understanding
seismicity triggered by hydraulic
fracturing
Gail M. Atkinson   1 ✉, David W. Eaton   2 and Nadine Igonin   2
Abstract | As recently as 2015, it was common in the scientific literature to find assertions that the
risk of triggering a damaging earthquake by hydraulic fracturing (HF) — an industrial process
where pressurized fluids are used to create or open fractures within rock layers — could be treated
as negligible. However, that viewpoint has changed dramatically. It is now clear that the hazard
from induced seismicity (including HF) exceeds the natural hazard in low-​to-​moderate seismicity
environments. As such, to mitigate risk to vulnerable and critical infrastructure, it is important to
address the likelihood and triggering mechanisms of HF-​induced earthquakes. Although it is
sometimes claimed that HF-​induced earthquakes can be accurately predicted, avoided or
controlled, critical knowledge gaps still remain. In this Review , we discuss six fundamental issues
surrounding induced seismicity , focusing specifically on HF-​induced events, including: the triggering
mechanisms of HF seismicity; the relationship between tectonic environment and HF seismicity; the
similarities and differences between induced and natural events; the damage potential associated
with HF-​induced seismicity; whether HF-​induced events can be predicted; and the relative hazards
of HF-​induced and natural seismic events. We finish by outlining future research directions that are
required to minimize the uncertainty and hazard that surround induced seismicity.

Induced seismicity — defined as earthquakes associated first reported HF-​induced earthquake in Love County,
with human activity — has been a topic of active research Oklahoma in 1979 (M 1.9)16. Technological advances,
for many decades1–9. Induced seismicity can be caused beginning around 2009, facilitated the widespread
by a range of human activities, including deep under- application of massive, multistage HF along horizontal
ground mining and injection (or withdrawal) of fluids in wellbores17.
the subsurface10 (Table 1). In the central USA, the rate of International attention to HF-​related seismic-
induced earthquakes with a magnitude >3 (greater than ity increased in response to a series of small earth-
the felt threshold) has increased tenfold over the past quakes triggered in 2011 by fracking of the Preese Hall
decade11; in western Canada, the rate has increased by a 1 exploration well near Blackpool, UK18,19. Since 2011,
factor of 3 (refs12,13) (magnitude (M) is typically reported HF-​triggered earthquakes of increasing magnitudes,
on either the moment magnitude (M) or local magni- up to ML 5.7, have been documented at many loca-
tude (ML) scales). In the central USA, increased seismic tions around the world20 (Fig. 1; Table 2). The largest
activity has been primarily linked to large-​volume salt- HF-​triggered events, which occurred in China, caused
water disposal (SWD)14, whereas in western Canada, economic losses, injuries and fatalities, leading to height-
it is typically associated with the industrial process of ened concern surrounding HF-​triggered seismicity20,21.
hydraulic fracturing (HF) along horizontal wellbores12. In addition, the events in China indicate that we should
1
Department of Earth Hydraulic fracturing, or ‘fracking’, is a wellbore- not focus solely on the probability of events occurring
Sciences, Western University, c​ ompletion method that is used to improve the recovery (the hazard) but also on the risks associated with these
London, Ontario, Canada.
of hydrocarbons from unconventional oil and gas res- events (the consequences)22.
2
Department of Geoscience, ervoirs (such as low-​permeability rocks)15. Injection of It is important to note, however, that there are
University of Calgary,
Calgary, Alberta, Canada.
high-​pressure fluids, usually a water–sand slurry, is used large parts of western Canada, North Dakota and
✉e-​mail: gmatkinson@ to create an open fracture network that enhances fluid Pennsylvania where induced seismicity has not been
aol.com flow within porous but otherwise impermeable strata10. observed, even though HF operations are widespread12,23.
https://doi.org/10.1038/ Since its inception in 1947, HF has been employed at In fact, statistical studies12 indicate that, averaged over
s43017-020-0049-7 more than 1.8 million wells in North America, with the western Canada as a whole, only ~0.3% of horizontally

264 | May 2020 | volume 1 www.nature.com/natrevearthenviron


Reviews

Key points open existing fractures to enhance the permea-


bility of the reservoir and, thus, enable economic
• Hydraulic fracturing can trigger earthquakes large enough (generally , magnitude >4) flow rates 15. The fracturing process is commonly
to cause potentially damaging ground motions, with actual damage depending on associated with seismic events M <0, referred to as
the intensity of motions and the vulnerability of nearby infrastructure. operationally induced microseismicity10. Microseismic
• The triggering of anomalous events (M >2) requires a source of stress perturbation, events close to hydraulic fractures, which provide a
a pre-​existing, critically stressed fault with sufficient surface area to host a felt event useful proxy for the stimulated part of the reservoir25,26,
and a coupling mechanism that connects the source to the fault, either directly or
are considered a normal part of HF. However, activa-
indirectly.
tion of pre-​existing fault systems during HF could
• Induced earthquakes are similar to their natural counterparts with respect to source
lead to larger and potentially damaging earthquakes.
characteristics, magnitude–frequency characteristics and ground motions.
In this section, we examine the triggering mechanisms
• The hazard from earthquakes induced by hydraulic fracturing might greatly exceed
of HF-​induced earthquakes and discuss the conditions
the natural earthquake hazard in regions of low to moderate seismicity , which is
consequential for the seismic safety of nearby (<10 km) infrastructure.
that are required to explain larger events (M >2) trig-
gered by HF, which often occur at a substantial distance
• Potentially damaging induced events cannot be confidently predicted in advance
of operations. Current risk-​mitigation strategies, such as traffic light protocols, have
(approximately 2 km) from the well27–29.
not yet proved reliable. Further development of hazard forecasting and mitigation
approaches is a critical future area of research. Necessary conditions for triggering felt events
A small fraction of HF-​induced earthquakes might
be large enough (generally, M >2) to generate ground
Operationally induced
drilled HF wells are associated with events of M ≥3, motions that people feel (known as felt induced events).
microseismicity although this value varies widely within the region24. Researchers studying induced seismicity widely
Weak seismicity that is Nevertheless, the increase in frequency (and magnitude) acknowledge that certain conditions are required for the
expected to occur during of seismic events associated with HF over the past dec- generation of felt induced events, including: a source of
operations such as hydraulic
ade indicates that it is important to consider the hazards stress perturbation, a pre-​existing critically stressed fault
fracturing or development of
an engineered geothermal and risks associated with these events, even if they only with sufficient surface area to host a felt event and a cou-
system. originate from a small percentage of HF wells. pling mechanism that directly or indirectly connects the
In this Review, we provide an overview of the cur- source to the fault10.
rent state of knowledge of HF-​triggered seismicity. We The concept of a critically stressed fault implies that it
discuss the available literature surrounding six funda- is in a state of incipient failure and, therefore, subject to
mental issues that have important ramifications for the activation in response to a perturbation that is small rela-
assessment and mitigation of hazards associated with tive to ambient stresses30. It is often treated as a binary con-
the use of HF technology: how HF triggers seismic- cept (a fault either is or is not critically stressed), although,
ity; the relationship between the tectonic environment in practice, the stress change required to reactivate a fault
and HF-​triggered seismicity; the differences between will depend on the in situ stress field and the orientation
HF-​triggered and natural earthquakes; the damage of the fault31,32. As a result, a discrete reactivation thresh-
potential of HF earthquakes; the predictability of HF old is difficult to determine. Nevertheless, in some cases,
earthquakes; and the relative hazards of HF-​triggered induced seismicity has been postulated to result from very
versus natural seismicity. Our Review is primarily small stress perturbations (<1 MPa)28,29,33,34.
focused on HF-​triggered seismicity, but other forms of
injection-​triggered seismicity (such as enhanced geo- Fault reactivation by pore-​pressure effects
thermal stimulation (EGS) and SWD) are also discussed Reactivation of a fault is often evaluated with respect
to illustrate common themes (Table 1). to the Mohr–Coulomb threshold, where slip will occur
if the shear stress τ exceeds the criterion:
Triggering mechanisms
Determining how HF triggers earthquakes is impor- τ > ϕσn′ + C , (1)
tant, as it represents the starting point for understand-
ing the physical processes that control HF seismicity. By where σ′n is the effective normal stress (σ′n = σn − P, where
design, HF is intended to create new fractures and/or σ′n is the normal stress and P is the pore pressure), ϕ is the

Table 1 | Fluid-​injection processes that have been linked to induced seismicity


Industrial activity Purpose Typical injection style Target formation
Hydraulic fracturing Create new fractures in a Vertical or horizontal Low-​permeability source rock
hydrocarbon-​rich source rock wells that inject a slurry with high organic content (oil,
for improved extraction of water and proppant gas, condensate, such as shale)
Enhanced geothermal Create/enhance permeability Two vertical wells Heat source with low
stimulation pathways for water circulation with fluid circulating permeability to prevent fluid
between two wells during between them leakoff (such as granite)
geothermal power generation
Saltwater disposal Inject water into existing Single vertical well Permeable reservoir rock that
permeable formations for that is used to inject can quickly intake and transfer
disposal saltwater or waste water injected fluid (such as limestone).

NaTure RevIews | EaRTH & EnvIRonMenT volume 1 | May 2020 | 265


Reviews

100˚ 120˚ 140˚ 160˚ −180˚ −160˚ −140˚ −120˚ −100˚ −80˚ −60˚ −40˚ −20˚ 0˚ 20˚

Hydraulic fracturing
M5 M3 M1 Horn River Basin
60˚ Saltwater disposal M 1.0 Poland 60˚
Enhanced geothermal Montney
stimulation Duvernay
ML 4.5 Fort St. John, BC ML 2.9 England
M 4.1 Fox Creek, AB
Latitude (˚N)

40˚ Sichuan ML 3.7 Ohio 40˚


Basin ML 3.6 Oklahoma
M 4.0 Texas
M 5.3 China
20˚ 20˚

0˚ 0˚
100˚ 120˚ 140˚ 160˚ −180˚ −160˚ −140˚ −120˚ −100˚ −80˚ −60˚ −40˚ −20˚ 0˚ 20˚
Longitude (˚E)

Fig. 1 | Global distribution of induced seismicity. The global distribution of hydraulic fracturing (HF), enhanced
geothermal stimulation and saltwater disposal induced seismicity (data taken from ref.158). Documented cases of
HF-​induced seismicity include the USA (Oklahoma, Texas, Ohio), Canada (Alberta, BC), England, Poland and China.
Note that the labels reference HF events; in some areas, there are larger saltwater disposal events in the same region
that are not labelled (such as in China and Oklahoma).

coefficient of friction and C is the fault cohesion. We use stress field45 (Fig. 2). Because the resulting stress field
the sign convention that positive σn corresponds with an is transferred through the rock frame, it is not con-
increase in the compressive force acting on a surface. It is strained by the low permeability of the pore space. For
apparent that slip might be triggered by a reduction of the SWD operations, poroelastic coupling has been invoked
normal stress, an increase in the pore pressure, a change to explain induced seismicity at a distance of >40 km
in the coefficient of friction and/or loss of fault cohesion35. (ref.34). Similarly, tensile opening of hydraulic fractures
Since the Denver earthquakes in the mid 1960s1 and is thought to cause deformation in the surrounding rock
subsequent studies at Rangely, Colorado2, it has been that appears to control the location of induced events.
well established that a reduction in effective normal The potential role of poroelastic effects in event trig-
stress on a fault due to increasing pore pressure is an gering is typically evaluated by comparing numerical
important seismogenic mechanism (Fig. 2). In the case of models of stress changes caused by injection with the
HF, direct injection of pressurized fluids into a fault zone locations of induced seismicity29,37,46.
represents a conceptually simple causative mechanism
for triggering seismicity36,37. For example, microseismic Fault reactivation by aseismic slip. Fault activation by
observations tracked the direct intersection (and subse- fluid injection is a complex process that depends on the
quent activation) of a pre-​existing fault by HF in western interplay between fluid pressurization and friction47.
Canada36. Direct measurements of the offset on a pre-​existing
However, hydraulic fractures typically do not fault in south-​eastern France, following an increase in
extend more than a few hundred metres from the treat- pore pressure, demonstrate that slip can occur at slip
ment well38 and the low matrix permeability of most speeds of only ~0.01 mm s−1, accompanied by minor
unconventional plays inhibits transfer of elevated pore brittle failure within the damage zone around the fault48.
pressures beyond the stimulated zone12. It has been pro- Measurements of the frictional parameters of rocks that
posed that high-​permeability pathways (for example, characterize unconventional plays49 show that high clay
fracture corridors) within low-​permeability rocks could mineral content, total organic content or both increase
represent a mechanism by which elevated pore pressures the probability of slow slip due to velocity-​strengthening
are rapidly transferred to larger distances27,39,40. Analysis effects50.
of the time delays between injection and seismicity can Aseismic slip provides another potential triggering
Unconventional plays be used to characterize the permeability of such fea- mechanism for induced dynamic rupture. In this sce-
Oil and gas resources whose tures41–44 and determine the cases for which increased nario, pressurization of the faulted region due to HF
porosity, permeability, pore pressure might be the triggering mechanism. causes slow slip (creep) on frictionally stable parts of the
fluid-​trapping mechanism or
fault system near the stimulated zone. As aseismic slip
other characteristics differ
from conventional hydrocarbon Fault reactivation by poroelastic effects accumulates, it progressively loads other parts of the fault
reservoirs. In cases where there appears to be no direct hydrau- system that are frictionally unstable, leading (in some
lic connection to the well, some studies have invoked cases) to dynamic rupture51. Numerical simulations
Creep poroelastic effects to explain fault reactivation 28,29. suggest that the creep front can outpace pore-​pressure
A process in which permanent
plastic deformation occurs
An increase in pore pressure causes an expansion of the diffusion, ultimately leading to earthquake nucleation
owing to various microscale rock frame, which pushes outwards on the surround- outside the region of increased pore pressure48,52. There is
or atomic-​scale mechanisms. ing rocks (and vice versa), perturbing the surrounding also evidence that the presence of injected fluids and/or

266 | May 2020 | volume 1 www.nature.com/natrevearthenviron


Reviews

Table 2 | The largest seismic events for hydraulic fracturing by region The areas that have a high susceptibility to induced
seismicity occur within a wide range of different tec-
Region Largest event Date tonic environments. For example, areas of high sus-
Sichuan Basin, China 20
M 5.3 (ML 5.7) 16 December ceptibility within the WCSB, such as the Duvernay and
2018 the Montney plays, are located in the Late Cretaceous–
Fort St. John, BC, Canada109 M 4.6 (ML 4.5) 29 November 2018 Paleocene foreland of the Canadian Rocky Mountain
Texas, USA154 M 4.0 1 May 2018 thrust belt, which developed in a continental sedimen-
tary platform setting61. The Lower Paleozoic Sichuan
Red Deer, AB, Canada 155
M 4.4 (ML 4.2) 4 March 2019
Basin in China has a very complex tectonic history,
Fox Creek , AB, Canada51,86 M 4.1 12 January 2015 coupled to the development of four orogenic belts62. The
Horn River, BC, Canada79 ML 3.8 19 May 2011 Bowland Basin (UK) represents an Early Carboniferous
Ohio, USA156 ML 3.7 3 June 2017 basin that experienced several phases of extension and
intrabasinal tectonics, including basin inversion in the
Oklahoma, USA36 ML 3.6 25 July 2019
Late Carboniferous63.
UK 60
ML 2.9 26 August 2019 The rate of induced seismicity in different environ-
Poland157 M 1.0 25 June 2016 ments can be directly compared by using quantitative
measures, such as the seismogenic index, Σ (ref.31), which
relates the number of events of a given magnitude to the
aseismic creep might prolong the duration of induced injection volume. The original definition of the seismo-
seismicity (typically, for days to weeks), giving rise to genic index was developed for EGS applications involv-
swarm-​like behaviour, as well as enhanced susceptibility ing injection into one well, but adapted definitions have
to remote triggering53. been proposed for multiwell HF operations57,64. Previous
studies have shown that Σ values vary by ~10 orders of
Triggering distance magnitude across a selection of sites (both unconven-
The Mohr–Coulomb failure criteria (Eq. 1) suggest that tional gas and geothermal sites), demonstrating clear
there are several mechanisms by which HF can trigger variations in the susceptibility of a region to induced
seismicity. Although the direct impact of elevated pore seismicity. The rate of seismicity in different tectonic
pressure is generally limited to the stimulated part of environments can also be examined by comparing the
a reservoir close (on the order of hundreds of metres) per-​well likelihood of triggering events. In the WCSB,
to the injection well, mechanisms such as poroelastic the per-​well probability of triggering events with M ≥3
coupling and aseismic creep could extend the range ranges from ~0.1% for the Cardium Formation to ~5%
of influence of HF. Such effects have been used to for the Duvernay Formation65. However, for formations
explain seismicity that has been triggered above51 and where no triggered seismicity has yet been observed,
below28,54–56 the target formation, as well as to distances there is no accepted methodology to forecast the
of ~1–2 km laterally from the well27,28,44. It is important likelihood of an event.
to recognize that the triggering mechanisms, distance Multiple mechanisms have been proposed to explain
over which events can be triggered and the probability the variation observed in the susceptibility to induced
of induced seismicity occurring is likely to depend on seismicity across different tectonic settings. For example,
the local and regional tectonic and geological structure HF-​induced seismicity might be expected to correlate
of the region. with formations that are overpressured (pore pressures
greater than hydrostatic), as elevated pore pressure will
Tectonic environment reduce the effective normal stress, bringing faults closer
Some unconventional HF plays, such as the Duvernay57, to the Mohr–Coulomb threshold66. Alternatively, the
Montney58 and Exshaw56 plays in the Western Canada relative abundance of pre-​existing faults can be used to
Sedimentary Basin (WCSB), the SCOOP (South Central explain the observed distribution of induced seismic-
Oklahoma Oil Province) and the STACK (Sooner Trend ity: the edges of Paleozoic reef complexes in the WCSB
(oil field), Anadarko (basin), Canadian and Kingfisher have been used as a proxy for the presence of basement
(counties)) plays in Oklahoma59 and the Sichuan Basin, faulting, explaining an apparent correlation between
China20,21, have produced high levels of seismicity. induced seismicity and the edges of the Swan Hills
However, the lack of induced seismicity associated with carbonate platform67.
several other unconventional HF plays, such as the In addition, application of machine-​learning algo-
Barnett (Texas), Marcellus (centred on Pennsylvania) rithms to documented instances of triggered seismicity
and Bakken (North Dakota and Saskatchewan) shale in the WCSB found that a selection of features, including
gas plays, demonstrates that the relationship between the proximity to the basement, in situ stress, proximity
HF and induced seismicity is likely to be influenced to fossil-​reef margins, lithium concentration and rate of
by the tectonic setting23. For example, in the Bowland natural seismicity, were among the strongest predictor
Shale (UK), only three wells have been stimulated by variables for the presence of induced seismicity68. Such
HF; yet, these have produced events with magnitudes algorithms might be promising but are still in learning
Failure criteria of M = 2.3, 1.5 and 2.9 (refs18,60). Considering the paucity of modes, with their utility to accurately forecast risk yet to
A mathematical model defining
stress conditions under which
HF operations in the Bowland Shale, the occurrence be established. Indeed, a series of felt events triggered by
failure might occur, such as the of these events suggests a high susceptibility to induced HF near Red Deer, Alberta, including an event of M 4.4,
Mohr–Coulomb failure criteria. seismicity. occurred in March 2019 in an area that was previously

NaTure RevIews | EaRTH & EnvIRonMenT volume 1 | May 2020 | 267


Reviews

Epidemic-​type aftershock
mapped by machine-​learning algorithms as having a productivity and interevent distances, as documented
sequence (ETAS) models relatively low seismogenic-​activation potential68. through epidemic-​type aftershock sequence (ETAS) models75,
Cascading point processes can distinguish fluid-​induced seismicity from natural
derived from Omori’s law Comparison with tectonic earthquakes seismicity76. The ETAS model has also been applied to
that can be used to simulate
the temporal patterns of
Regardless of whether induced events occur in regions identify induced events in the Sichuan Basin, China77.
earthquake sequences that have a high seismogenic-​activation potential or Using a similar method, several distinguishing statistical
in a given region. not, it is important to compare HF earthquakes and characteristics of seismicity induced by fluid injection
tectonic earthquakes to determine whether a specific were identified in California78, which included higher
Microseismicity
event is natural or induced and if triggered earth- background rates, faster temporal decay sequences and
Seismicity of magnitude less
than 0.
quakes have the same damage potential as their natu- different spatial-​clustering properties.
ral counterparts. A commonly employed approach to In some cases, it has been straightforward to com-
Double-​couple distinguish natural from induced events uses a series pile compelling evidence that events were likely induced
A mathematical model for an of questions about the temporal and spatial relation- by HF27,28,56,79,80. However, in specific cases, it has been
earthquake-​source mechanism,
consisting of two orthogonal
ship between events and industrial activities, and about challenging to make an unambiguous assessment, with
force couples. The mechanism the expected pressure perturbations69,70. In cases that debate continuing within both the academic literature
is typically parameterized return a higher number of ‘yes’ answers, there is a and the public sphere81. Often, the association can be
using the strike and dip of the higher likelihood that the events were induced. This generic rather than specific; for example, statistical anal-
fault plane, as well as the rake
approach was updated in 2019 by weighting questions ysis of the spatio-​temporal correlation between HF and
(slip vector).
according to their relative importance, and by includ- seismicity, relative to that expected by random chance,
Stress drop ing a term that describes the quality of evidence used was used to infer that more than half of the observed
The co-​seismic reduction in to perform the assessment71. seismicity in the WCSB (with M ≥3) was related to HF12.
shear stress acting on a fault Alternatively, a suite of assessment schemes has However, in the WCSB, it is not feasible to determine
(the difference between the
shear stress on the fault before
been proposed to assess whether seismic events are the relationship between natural and induced events for
an earthquake and the shear induced or natural. The first assessment scheme com- every well and every earthquake.
stress after an earthquake). pares observed seismicity with seismicity simulated Some case studies indicate that, unlike natural events,
from physics-​b ased models that include the effects low-​magnitude microseismicity (M <0) associated with
of depletion-​induced subsurface stress perturbations, HF propagation might exhibit non-​double-​couple com-
where the null hypothesis posits that seismicity occurs ponents82–84. Larger HF-​triggered events (M >3), how-
at a regional background rate72. The second scheme ever, typically have dominantly double-​couple focal
uses statistical correlations between industrial activ- mechanisms that are consistent with regional stress
ities and parameters that describe the seismic-​event fields18,20,55,60,85,86. Moreover, the radiated spectra and
population, such as the event frequency, magni- stress drop associated with large HF events appears to
tude distribution, spatial distribution and interevent be similar to that observed in shallow natural earth-
times73. The third scheme uses source mechanisms quakes87–89. In addition, ground motions generated by
as a discriminator, as some industrial activities, such as induced events are comparable to those resulting from
collapse events during mining, might produce charac- shallow tectonic events of the same magnitude and
teristic source mechanisms that differ from those of hypocentral distance90–92. In summary, once triggered,
earthquakes74. M >2 HF earthquakes appear to be very similar to their
In addition, it has been suggested that, in Oklahoma natural counterparts, which reinforces the interpre-
and California, changes in the background rate, aftershock tation that HF triggers earthquakes by reactivation of
pre-​existing tectonic structures.
Some differences in the stress drop and, in turn,
ground motion resulting from natural and induced
earthquakes are likely to occur owing to variations in
the focal depth of the events (induced events, on aver-
age, occur at shallower depths)90. However, the focal
Increased pore pressure
along a fault due to pore- depths of both natural and induced events are often
pressure diffusion poorly constrained by regional seismograph networks;
high-​density local arrays are needed to provide defin-
Permeable itive solutions. For example, a single M 4.1 earthquake
Changes of volume or reservoir in Alberta, Canada, was identified to have occurred
mass due to injection
HF in proximity to fault
at ~3-​km depth (above the treatment formation)86 and at
or extraction causes
changes in stress on causes stress transfer ~5-​km depth (below the treatment formation)93 by dif-
nearby faults and possible changes ferent studies. In this single example, the shallower depth
in pore-fluid pressure
estimate could be considered as the most reliable, as a
dense local network, calibrated by ground-​truth events,
was used to estimate the depth, whereas the deeper esti-
Fig. 2 | Possible triggering mechanisms of HF-induced seismicity. Direct and indirect
pathways to connect stress perturbation to critically stressed faults and, thereby , trigger mate was based on more limited station data. Moving
earthquakes. Pathways include large-​volume disposal of fluids into a permeable zone forward, a new method has been developed, based on
that is in hydrological contact with faults, a change in stress due to an overlying load the velocity information from co-​located, multicompo-
or hydraulic fracturing (HF) in proximity to faults. Adapted with permission from ref.10, nent seismic data94, to provide better depth constraints
Cambridge University Press. in the future.

268 | May 2020 | volume 1 www.nature.com/natrevearthenviron


Reviews

Damage potential of HF earthquakes including the magnitude of the event, its proximity to
Despite the clear evidence that earthquakes associated with population or critical infrastructure, the intensity of the
industrial processes (HF, EGS, SWD) can produce seismic ground motions it generates and the vulnerability of
events that are M >3 and have similar source character- the local infrastructure.
istics to natural events, the potential of induced seismic-
ity to cause damage to local or critical infrastructure has Ground motions and damage potential
only recently been considered. Therefore, in this section, It is now well documented that injection-​induced earth-
we address the potential of HF to trigger damaging earth- quakes might produce high-​amplitude ground motions
quakes, a topic that is especially important in developing that could damage nearby infrastructure 13,20,99–101.
regions with a high density of vulnerable infrastructure. Documented damage caused by HF-​induced seismicity
ranges from superficial damage of vulnerable unrein-
Distribution of event magnitudes forced masonry structures (M 3.4)102 to major and/or
Induced events, like their natural counterparts, gener- widespread damage of buildings (M 4.9–5.8)20,101,103.
ally follow the well-​known Gutenberg–Richter relation95, The ground motion expected from induced events
where the number of earthquakes for a given time period was initially estimated on the basis of instrumen-
(N(M)) decreases exponentially as magnitude (M) tal ground-​motion data from shallow earthquakes
increases57,60,96. The Gutenberg–Richter (1944) relation in California104. However, since 2015, the amount of
is characterized by a rate parameter (‘a-​value’) and the quantitative information on induced ground motions
decay slope (‘b-​value’): has grown dramatically, enriched by instrumental
records from events in Oklahoma105,106 and western
log N (M ) = a − b M (2) Canada107–110. The growth of instrumental data has ena-
bled development of ground-​motion models directly
Therefore, the expected number of large events can from induced-​event databases106,110, and these models
be predicted, in a general sense, from the observed will continue to improve in the future.
or modelled number of small events if the b-​value The U.S. Geological Survey (USGS) ‘Did You Feel It’
is known60. For a typical b-​value of 1, the event rate report­ing system also contains a wealth of information
decreases by a factor of 10 for every increase of 1 mag- about ground motion resulting from induced seismic
nitude unit. HF-​driven microseismicity (M <0), which events, with more than 100,000 felt reports from induced
occurs near the injection point as a direct response to events in Oklahoma alone. Felt reports provide a rich
pore-​pressure perturbation, is often observed to have a database of human observations and have been used to
b-​value of ~2 (refs36,84,97), indicating an unusually sharp examine the damage potential of induced events. Early
drop in event frequency with increasing magnitude. results indicated that the felt intensities of induced events
By contrast, injection-​triggered sequences that include are the same as those for natural events at short distances
larger events (M ≥3) typically have b-​values of ~1, (<10 km), but that induced events tend to have lower
similar to that observed for natural earthquakes57,96,98. intensities at larger distances111. As a result, it was inferred
If there are no physical constraints on the maximum that induced events have lower stress drop on average than
size of event that can be triggered, the basic statistics of do natural events, where stress drop controls the strength
HF-​triggered seismicity can, at least in theory, be used of high-​frequency ground motion. Instrumental ground-​
to assess the likelihood of earthquakes of potentially motion data have been used to show that stress drop scales
damaging magnitudes96. From 2010 through 2018, there with focal depth90, so, on average, induced earthquakes
were ~20,000 horizontally drilled HF wells in western might be expected to have a relatively low stress drop,
Canada. Given the basin-​wide average incidence rate of owing to their shallow depths (compared with natural
wells with M >3 events (0.3%)12, we would expect to have events). Alternatively, it has been proposed that apparent
seen approximately six HF-​triggered events of M ≥4, differences in the stress drop of induced events (mostly
with the expected value of the largest event being M ~4.7 in central North America) and natural events (mostly in
(assuming a Gutenberg–Richter relation with b ~1). eastern North America) might be explained by differ-
In fact, in western Canada, seven HF-​triggered events of ences in their focal mechanism88. The effects of stress
M >4 were observed between 2010 and 2018, with a largest drop and distance counteract at short distances, which
magnitude of M 4.6 (ref.65). More broadly, we could expect can explain the similar felt intensities near the epicentre
that, as the number of HF wells grows, the maximum for induced events (with lower stress drop) and natural
magnitude of HF-​triggered events would gradually rise events (with higher stress drop) in a given region.
globally. Consistent with predictions, the largest reported In addition, ruptures in relatively stable intraplate
magnitudes for HF-​triggered events has increased globally regions, such as central and eastern North America, tend
from M 2.3 in 2011 (ref.18) to ML 3.8 in 2012 (ref.79), M 4.6 to be characterized by higher stress drops in comparison
in 2015 (ref.12) and ML 5.7 in 2018 (ref.20). with active interplate regions, such as California112. As a
Intensity
The effects of earthquake The largest HF-​t riggered event to date (M L 5.7, result, it has been suggested that ground motions might
ground motion on the natural M 5.3) occurred in the Sichuan Basin in China, caus- have larger amplitudes (by a factor of 1.5) for ruptures
or built environment. ing injury to 17 people, large-​scale landslides, collapse on immature faults in comparison with mature faults113.
of nine houses and extensive damage to 390 houses20, Consequently, the higher stress drop that is observed in
Epicentre
The point on the surface
highlighting that HF-​triggered events can be damaging. central and eastern North America, owing to the sta-
vertically above an However, the scale of the damage that could be caused ble tectonic setting of the region, has been suggested
earthquake’s focus. by HF-​triggered events depends on a range of factors, to offset the focal depth effect, which explains why

NaTure RevIews | EaRTH & EnvIRonMenT volume 1 | May 2020 | 269


Reviews

Peak ground acceleration ground motions for induced intraplate events in central then be compared with models for peak ground motions
Maximum instantaneous and eastern North America are observed to be sim- (PGV and peak ground acceleration (PGA)), which were
amplitude of the absolute ilar to those of natural California events of the same converted into MMI using empirical relationships92.
value of the acceleration magnitude and distance91. Generally, ground motions and felt intensities from
of the ground.
Direct comparison of the damage potential of natu- natural earthquakes in California104, SWD-​induced
ral and induced earthquakes (M >3.5), using compiled earthquakes in Oklahoma106 and HF-​induced earth-
ground-​motion and felt-​intensity data, is shown in quakes in western Canada110 plot in the same amplitude
Fig. 3, with the peak ground motions used as an instru- range, for a specified magnitude and distance, despite
mental measure of damage potential. The peak ground differences in the dominant triggering mechanism92.
velocity (PGV) is an effective damage metric, with Moreover, the consistency between the observations
PGV = 1–2 cm s−1 being the threshold for cosmetic dam- and model predictions indicate that these published
age to buildings and infrastructure114,115. In addition, models can be used to estimate the event magnitude
compiled felt-​intensity data (Modified Mercalli Intensity, that is required to cause cosmetic and/or major damage
MMI) were also used as a damage metric, where MMI = 2 at different epicentral distances. The minimum magni-
is the felt threshold, MMI = 6 is considered the thresh- tude that can cause damage in the epicentral region is
old for minor damage and MMI = 7–8 is the threshold inferred to be M ~4.0, depending on the vulnerability of
for major damage, depending on the structural vulnera- the infrastructure. Events of M ≥4.5 are inferred to have
bility116. The felt intensities from induced seismicity can substantial damage potential at distances <5 km (ref.92).
a Recorded ground motions and model equations b Intensity from PGV/PGA (M 4.0, 5.5) and observations
(±0.25 units)

8
10

6
1
MMI
PGV (cm s–1)

0.1 4

0.01 2
2 3 10 20 30 2 3 10 20 30
Hypocentral distance (km) Hypocentral distance (km)
Cosmetic damage NAA18 M 5.5 M 4 OK obs. M 5.5 predicted-OK
Major damage A15 M 4.0 M 5.5 OK obs. M 4.0 predicted-CA
OK M 4.0 ± 0.25 A15 M 5.5 M 4.0 median M 5.5 predicted-CA
OK M 5.5 ± 0.25 WCSB M 4.0 ± 0.25 M 5.5 median M 4.0 WCSB
NAA18 M 4.0 M 4.0 predicted-OK (from PGV)

Fig. 3 | Ground motions and damage potential from induced events. a | Peak ground velocity (PGV) observed in
Oklahoma (OK) for events of M 4.0 (±0.25) (blue circles) compared with prediction equations for Oklahoma105 (blue
line) and California103 (dashed grey line). Grey diamonds show corresponding PGV observations for M 4.0 (±0.25) for
hydraulic-​fracture events in the Western Canada Sedimentary Basin (WCSB). Horizontal lines show the approximate PGV
thresholds for cosmetic and major damage114,115. A good match between the prediction equations and measured PGVs
are observed, indicating that these equations can be used to estimate the damage potential of induced earthquakes at
various magnitudes and hypocentral distances. b | The observed Modified Mercalli Intensity (MMI) from events of M 4.0
(±0.25) in OK (blue circles) compared with prediction equations based on PGV and peak ground acceleration (PGA) for
OK105 (blue line) and California (CA)103 (grey dashed line). Grey diamonds display the corresponding MMI inferred from
PGV observations for M 4.0 (±0.25) for hydraulic-​fracture events in the WCSB. Once again, a good match is observed
between the MMI and prediction equations. Larger symbols show the median of the MMI observations in log-​spaced
distance bins and the shaded regions show the 25th to 75th percentiles of MMI observations. MMI = 2 is the felt threshold;
MMI = 6 is considered the threshold for minor damage; MMI = 7–8 is the threshold for major damage, depending on the
structural vulnerability. Adapted with permission from ref.92, Seismological Society of America.

270 | May 2020 | volume 1 www.nature.com/natrevearthenviron


Reviews

Seismic moment Hazard forecasting and risk mitigation HF and such faults117. However, in practice, it is dif-
A measure of the size of an As induced earthquakes with M >4 have the potential to ficult to detect faults prior to reactivation, especially
earthquake based on the cause damage to local infrastructure, it is not surprising if they are strike-​slip in nature. In several cases, no
product of the rupture area, that the hazards and risks associated with HF operations clear correlation has been found between faults that
the average amount of slip and
the force that was required to
have received a large amount of attention. The hazard are detected by geophysical surveys before injec-
overcome fault friction. refers to the likelihood of large, and potentially damag- tion began and the structures on which seismicity
ing, events or ground motions, whereas the risk refers occurred60,118,119. As an example, an investigation of a
to the probability of undesirable consequences (such prolific sequence of induced events in the Horn River
as damage) and, therefore, requires knowledge of the Basin, Canada, which included events as large as ML
vulnerability of proximal infrastructure. For HF hazard 3.6, concluded that all but one of the induced events
assessment and risk mitigation, it is important to deter- occurred on previously unknown faults or fractures79.
mine whether critical aspects of HF seismicity can either By contrast, the dozen pre-​existing faults mapped by
be predicted in advance of operations or controlled dur- 2D and 3D seismic imaging were not (with one excep-
ing operations, thus, preventing undesirable outcomes. tion) associated with the induced events79. Seismogenic
In this section, we review the efficacy of such strategies. faults are often identified after they have been illumi-
nated by an earthquake sequence119,120 and, therefore,
Pre-​operational assessment it appears as though the hazard from HF-​induced seis-
Since the existence of pre-​existing, critically stressed micity cannot be reliably controlled by avoiding known
faults is a prerequisite for HF-​triggered events, it should pre-​existing faults.
be theoretically possible to prevent induced seismicity
by mandating a setback (or respect distance) between Traffic light protocols
As induced seismicity cannot be confidently mitigated
Magnitude that triggers Magnitude that triggers using respect distances to known fault structures, other
amber scenario red scenario
mitigation approaches are required. The most common
Hydraulic fracturing
M >4 approach for risk mitigation is a traffic light protocol
(TLP)114, whereby operators are required to take prede-
M >2
fined actions: reducing injection for an amber light or
Alberta: stopping injection for a red light. The thresholds used in
Fox Creek TLPs are usually defined by event magnitude, although
(Canada)
ground-​motion thresholds have also been used114,121. The
exact threshold that is used varies substantially between
different regions. For example, in Alberta and British
Columbia (WCSB), the TLP red-​light threshold is gen-
M >3.5 erally ML = 4 in sparsely populated regions, but ML = 3 in
M >2
some areas where events are likely to be widely felt. In
the UK, the red-​light threshold is only ML = 0.5 (ref.60)
Oklahoma (Fig. 4). Such vast differences in the TLP thresholds (by
(USA)
a factor of over 1,000 in terms of seismic moment) reflect
that the risks associated with seismic hazards vary with
population density and other factors.
M >0.5 TLPs are conceptually simple and can be operated
UK M >0 with minimal input data: all that is required is the mag-
nitude of the largest event. In regions where monitoring
is only provided by sparse, regional networks, smaller
Enhanced geothermal stimulation
events cannot be detected and hypocentral uncertainties
M >2 are usually large. Therefore, the magnitude of the larg-
M >1.3 est event is the easiest parameter to constrain. However,
Finland some implicit assumptions and limitations of TLPs must
be questioned. Crucially, TLPs are, by nature, retroactive:
action is taken after an event has occurred. In addition,
TLPs are based on the premise that halting injection will
M >1.7 stop seismicity and prevent large events from occurring.
M >1.4 In many cases, however, the largest events occur after
Korea injection has ceased and induced seismicity can persist
for days to weeks122. Therefore, the red-​light threshold
must be set to a level that considers the possible distri-
bution of trailing events – including those of larger mag-
Fig. 4 | Traffic light protocol thresholds for various
regions worldwide. The event magnitudes that are nitude. Unfortunately, there is no accepted methodology,
required to trigger the amber and red scenarios for hydraulic at present, to constrain the trailing-​event distribution.
fracturing and enhanced geothermal stimulation operations In addition, another implicit premise of a TLP is that
worldwide. It is clear that the magnitude required to trigger larger events are preceded by smaller, precursory events
the amber or red scenarios varies greatly between regions. that will activate the various warning levels (green to

NaTure RevIews | EaRTH & EnvIRonMenT volume 1 | May 2020 | 271


Reviews

1020
7 events in the Sichuan Basin, or whether processes
A such as stress transfer between events would have
continued to produce seismicity123. Moreover, owing
Runaway 6
1018
rupture 1 2 3 to the small sample size, we cannot adequately assess
whether small-​magnitude precursors will precede all
5 large-​magnitude HF events. Complications also arise
1016 from the possibility that the likelihood of precursory
4 5
4 events depends on the nature of the triggering process124.
In summary, there is much to learn regarding the capa-
1014 bilities and limitations of real-​time forecasting and TLPs
3
Seismic moment (Nm)

for risk mitigation.

Magnitude
Contained A 2
1012 Forecasting the maximum expected magnitude
rupture
The retrospective nature of TLPs has motivated devel-
1 opment of other approaches to mitigate the hazard from
1010 20
HF-​triggered seismicity. Alternative approaches are typ-
Runaway
rupture

0 ically based on model forecasts, with both physical and


Number of cases

15
statistical models used to estimate the expected event
108
10 –1 magnitudes.
Contained Physical models typically require population of a large
5 rupture
–2 model space and a number of free parameters need to be
106
defined125. As such, they are challenging to apply in real
0
A A
time to HF, where each of the multiple frack stages might
–3
only last for a few hours. Statistical models use the tem-
104
10 2
10 4
10 6
10
8
10
10
poral evolution of induced seismicity to define a small
Cumulative injected volume (m3) number of parameters that can be used to extrapolate
Fig. 5 | Relationship between cumulative injected volume and seismic moment. the temporal data and produce a projection of the event
Light blue circles are taken from Galis et al.132 and orange circles are taken from population. The projection generated includes a forecast
Atkinson12. Violet diamonds correspond to proposed cases of runaway rupture: Pohang, for the magnitude of the largest event, MMAX. Because
Korea (enhanced geothermal stimulation) (diamond 1), Pawnee, Oklahoma, USA of their ease and simplicity of operation, such statistical
(saltwater disposal) (diamond 2), Prague, Oklahoma, USA (saltwater disposal) (diamond 3), models have been widely applied31,43,60,64,96,126–128.
Fort St. John, Canada (hydraulic fracturing) (diamond 4) and Fox Creek , Canada (hydraulic Various empirical relationships have been used to
fracturing) (diamond 5). The grey and black lines display the scaling relationship between characterize the maximum rupture dimensions (and,
cumulative volume and total seismic moment proposed by McGarr129 and Galis et al.132, thus, event size), as well as the cumulative seismic
respectively. The background colour is used to represent the nature of the triggered
moment, ΣMO, of induced earthquakes. For example,
activity with respect to injected volume; events in the pale orange–white zone are
consistent with proposed mechanisms for which the magnitude scales with cumulative
in 2011, it was hypothesized that the maximum rup-
volume. Events in the orange zone are larger than would be predicted based on volume. ture dimensions of an induced event are constrained by
The intermediate region represents ambiguous rupture types. The distribution of cases by the dimensions of the stimulated volume, as defined by the
rupture type is shown by the transect A–A′, where the histogram and cumulative density spatial extent of microseismicity126. However, recent
function are calculated from events perpendicular to the transect. observations of induced seismicity reveal that rup-
tures can extend ~2 km beyond the stimulated zone28,119
and, therefore, that the maximum rupture dimensions
amber to red) in time to take mitigative action. However, of induced event cannot always be constrained by the
analysis of events induced by waste-​water injection in spatial extent of microseimicity.
the central USA96 revealed that the largest event during Alternatively, a linear scaling relationship between
a sequence occurs randomly and, therefore, is not nec- the cumulative injection volume, ΔV, and the cumula-
essarily preceded by smaller events. Precursory events tive seismic moment, of the form ΣMO = 2µΔV, where µ
(2 < M < 4) occurred prior to an M 4.6 event in 2015 is the shear modulus, has also been proposed129. However,
near Fort St. John, Canada58, and for an M 4.1 event in the scaling relationship between ΔV and ΣMO has lim-
2016 near Fox Creek, Canada51. By contrast, in 2019, an ited predictive power64, since it provides only an upper
HF-​triggered event of M 4.4 near Red Deer, Canada, had bound to the expected magnitude. In addition, although
no precursors of M >2. Likewise, an M 4.6 HF-​triggered most HF-​triggered events fall well below the upper
event in November 2018 near Fort St. John had no bound, some events in the WCSB have exceeded this
M >2 precursors. Moreover, in spite of an immediate limit12,130. In addition, the bound was also exceeded by
suspension order, the M 4.6 Fort St. John event was EGS at Pohang, which produced an M 5.5 event when
shortly followed by M 3.5 and M 4.0 events109. the limit predicted by the scaling relationship was only
For HF-​triggered seismicity in the Sichuan Basin, MMAX = 3.7 (ref.131) (Fig. 5). To address the discrepancies
China, it appears that numerous 2 < M < 4 events between observations and the values of MMAX predicted
occurred prior to the occurrence of the larger mag- by the volume-​based models, several variations of the
nitude (ML 5.3 and 5.7) events20,21. It is not possible to scaling relationships have been proposed. The varia-
know, however, whether ceasing injection at an earlier tions introduced include additional parameters, such
point in time could have prevented the larger seismic as seismic efficiency127, seismogenic index31 or the use

272 | May 2020 | volume 1 www.nature.com/natrevearthenviron


Reviews

Runaway rupture of geomechanical models for fracture propagation that associated with fault activation in active tectonic envi-
The initiation of larger-​ allow an unbounded runaway earthquake132. ronments, such as the possibility that foreshocks might
magnitude earthquakes that All volume-​based and moment-​based models are occur on different faults than the mainshock.
extend past the stimulated applied in similar ways: the scaling relationship between
region. These events primarily
release tectonic strain on faults
ΔV and ΣMO is characterized at an early stage during Seismic hazards from HF earthquakes
outside the stimulated region. injection and then extrapolated to calculate the total Although we cannot predict the location, timing and
expected seismic moment budget. When combined magnitude of natural or induced events with preci-
Foreshocks with the b-​value, scaling relationships can be used to sion, we can assess their likelihoods. For this reason,
Earthquakes that precede
determine the expected value (with uncertainty) of seismic-​hazard assessment is performed probabilisti-
the largest earthquake in a
sequence.
MMAX. Aftershock-​type event rate decay terms can also cally rather than deterministically. A probabilistic seis-
be added to model the trailing seismicity after injection mic hazard analysis (PSHA) uses the spatial distribution
has ceased133,134, and recent studies have attempted to of events, their magnitude–frequency distribution (the
take volume and moment-​based models a step further Gutenberg–Richter relation) and ground-​motion mod-
by proposing a method to evaluate the risk posed by els to calculate the likelihood of exceeding different
the expected seismicity135. Models in this vein have also amplitude levels of ground motion at a site141–144. For
linked event occurrence to preoperational loss models, example, building codes typically require all infra-
allowing real-​time evaluation of the evolving risk22. structure to be able to withstand ground motions with
Forecast methods based on the observed scaling rela- an annual exceedance probability of 2% in 50 years (or
tionship between injection volumes and seismicity have 1/2,500 per annum), although for critical infrastruc-
been used to guide operational decisions to mitigate ture, such as extreme-​consequence dams, this threshold
induced seismicity in geothermal sites128 and HF plays60. is raised to an exceedance probability of no more than
However, the scaling relationship between ΔV and ΣMO 1/10,000 per annum142,145,146.
often changes during injection, potentially owing to
hydraulic fractures impinging on a larger fault60, and, Assessing the hazards of induced seismicity
thus, the predictive ability of scaling relationships is The methods used to assess hazards associated with
limited. Therefore, it is clear that hazard mitigation, via induced seismicity follow a similar structure to those
the use of forecasting models to control the magnitude used for natural earthquakes, but with one key differ-
of the largest possible event, is in its infancy. ence. For natural seismicity, activity is assumed to be
Furthermore, observations of large events that relatively stationary in space and time, such that its spa-
occur after simulation has been completed136,137, or tial distribution and magnitude–frequency distribution
do not appear to fit the Gutenberg–Richter scaling of can be derived from an historical earthquake catalogue.
smaller-​magnitude events130, challenge the premise that For injection-​induced seismicity, however, activity is
MMAX can be controlled by controlling the volume of dependent on the location and activation potential of
injected material. When considering HF operations, industrial activities, which are non-​stationary.
some of the largest events are outliers on the magnitude– PSHAs for induced seismicity are commonly con-
frequency distribution130. Such high-​magnitude outliers ducted as year-​by-​year hindcast exercises, in which the
have been linked130 with the concept of characteristic hazard is assessed under the implicit assumption that
earthquakes138, wherein most events represent rupture the patterns and rates of induced events will mimic those
on small patches of a larger fault, but that rupture of the of the preceding year99,147,148. Such hindcasts are useful for
entire fault plane occurs episodically. In instances where demonstrating the impact of induced seismicity on esti-
the entire fault plane ruptures, events that are larger mated hazard. However, hindcasts can only be used as
than would be expected based on power-​law scaling are forecasts if the induced seismicity patterns are stable in
produced. time – a situation that is unlikely, given the nature of oil
In HF operations, high-​magnitude outliers might also and gas operations. PSHAs aimed at forecasting seismic
arise owing to runaway rupture, where large events are hazards in advance of conducting operations must con-
generated by triggering the release of tectonic strain on sider the likelihood of triggering damaging seismicity in
faults outside the stimulated region132,139. Figure 5 shows the future, as well as the expected magnitude–frequency
a compilation of injection-​induced seismicity data, distribution of potential events13,24. Unfortunately, the
including the Pohang 2017 earthquake and the Fort St. likelihood and magnitude–frequency distribution of
John 2018 earthquake. Five seismic events in the com- future events will depend on the susceptibility of the
piled data have been determined to meet the criteria for target formation and the injection parameters in a com-
runaway rupture. However, it remains unclear how (or plex way that is not yet well understood10,14,57,67,149. As a
if) the concept of runaway rupture can be incorporated result, large uncertainties in hazard forecasts are to be
into models that are used to forecast induced seismicity. expected13,150.
In future, highly sensitive monitoring and response Hazard models that include the influence of injec-
set-​ups might be able to identify fault activation prior tion rates, which have mostly been developed in con-
to a mainshock, with the potential to cease operations junction with SWD hazards, could potentially reduce the
before runaway ruptures are initiated140. For example, uncertainties associated with such models43,151. For HF
determining the precise location of microseismicity seismicity, the seismic productivity in the Duvernay play
could be used to reveal reactivation of planar struc- of Fox Creek, Alberta, has been shown to scale linearly
tures near the stimulated region. However, it remains with injection volume, which enables the Gutenberg–
necessary to consider well-​known complexities that are Richter relation to be modulated by operational

NaTure RevIews | EaRTH & EnvIRonMenT volume 1 | May 2020 | 273


Reviews

parameters. Future developments in hazard modelling to outline how induced-​seismicity prediction and risk
are likely to focus on developing reliable algorithms to mitigation can be improved in the future.
incorporate operational parameters into site-​specific or Certain conditions are considered necessary for an
region-​specific PSHAs. earthquake to be triggered: a source of stress pertur-
bation, a pre-​existing, critically stressed fault, a large
Relative hazards of natural versus induced seismicity enough surface area on the fault to host a felt event
Regardless of whether it is hindcast or forecast, the and a coupling mechanism that connects the source
hazard associated with induced seismicity can greatly to the fault. The coupling mechanism can represent a
exceed the pre-​existing natural hazard, particularly in direct connection between the injected fluids and the
regions of low-​to-​moderate seismicity13,99,100,147,150. The fault (such as elevated pore pressure) or an indirect
increased hazard associated with HF (or SWD) raises connection (loading by poroelastic stress transfer or
pressing socio-​economic questions concerning how stress transfer by aseismic slip). Induced seismic events
the heightened hazard might be managed, mitigated or typically occur within ~2 km of HF operations, with a
avoided. Several approaches have been suggested, which mechanism-​dependent time lag that varies from min-
consider the perceived hazard, risk and consequences, utes to years (typically, days to weeks). Further work
as well as the societal acceptance of industrial activity is encouraged to try and identify how the triggering
that could cause induced seismicity, sometimes in a mechanisms, and the distance over which events can be
risk-​matrix framework152. In general, owing to the lack triggered, is controlled by geological and tectonic factors.
of specific industrial regulations and practices153, several Multiple factors are believed to influence the suscep-
very different approaches have been used to deal with tibility of a region to HF-​induced seismicity, including
the hazards associated with induced seismicity. the degree of faulting/fracturing, fracture-​network con-
As discussed previously, the most common reg- nectivity, state of stress, overpressure and proximity to
ulatory approach to manage hazards from induced basement. Regional and local variations in such param-
seismicity has been to implement a TLP. However, the eters might explain the order-​of-​magnitude variability in
largest events cannot necessarily be predicted from the seismogenic index, a quantitative measure of the rate
the observed seismicity and might occur without nota- of induced seismicity. Future progress in understand-
ble precursors. In addition, the largest event can occur ing the relationship between tectonic environment and
after injection ceases28,122, sometimes during mitigation induced seismicity is required and can be achieved by
efforts103. Therefore, although TLPs can be a useful oper- quantifying susceptibility in a wide range of settings,
ational measure to limit the number of trailing events and identifying the physical factors and mechanisms
and invoke mitigation after the amber condition is acti- that relate to the observed differences in susceptibility.
vated, they might not reduce the likelihood of initiating a Although differences are observed in the spatio-
large induced earthquake. A case in point is a damaging ​temporal clustering properties of induced and tectonic
M 5.5 earthquake that was triggered by hydraulic stim- events (as identified by ETAS modelling) and the aver-
ulation of the Pohang geothermal field in South Korea, age depth of induced and natural events, once triggered,
despite a traffic light system that was in place with a individual induced events appear to share the same
magnitude threshold of 2.0 (later raised to 2.5)101. source and ground-​motion characteristics as their nat-
For HF operations in proximity to critical infrastruc- ural counterparts. The similarity between natural and
ture, such as major dams or nuclear power plants, a com- induced events supports the premise that triggered
bination of hazard avoidance and mitigation is required events represent reactivation of pre-​existing tectonic
to achieve stringent, mandated safety standards13. For structures. Comparative studies of induced versus natu-
example, such a strategy could involve the presence of ral events of similar magnitudes, depths and focal mech-
a minimum setback distance, such as 5 km, to remove anisms in the same tectonic setting could provide further
the possibility of large events occurring directly beneath insights into their similarities or differences. However,
critical facilities. The setback distance could then be such conditions are rarely met in regions of sparse
combined with regional seismic monitoring in the sur- natural seismicity.
rounding area, with a plan to reduce operations if the The clearest documented case of damaging earth-
rate rises above a preset threshold. The occurrence of quakes triggered by HF are the M ~5.0 to 5.5 events in
damaging (M ≥5) earthquakes triggered by HF in China the Sichuan Basin, China, which caused injuries, fatal-
has brought the need to develop improved hazard-​ ities and millions of dollars in structural damage. The
mitigation tools, and to avoid hazards in situations damage potential of triggered seismicity is dependent on
where they are unacceptable, into sharper focus20. the magnitude of the event, its proximity to population
and/or infrastructure, the intensity of resulting ground
Summary and future perspectives motion and the vulnerability of the local infrastructure.
Our Review has considered six fundamental questions Generally, we can consider M 4.0 events as the smallest
associated with the generation, damage potential and events that can cause notable damage (near the epicen-
prediction of induced seismicity. In general, we have tre). To advance knowledge of the damage potential of
focused on seismicity generated by HF, although evi- triggered earthquakes, engineering studies of how struc-
dence from SWD and EGS processes were considered, tures respond to strong, short-​duration ground motions
where relevant. Here, we summarize the key conclusions are needed.
of this Review, focusing on each of the six fundamen- At present, it is not possible to confidently forecast
tal issues, and provide a forward-​looking perspective the occurrence, or the maximum size, of an HF-​triggered

274 | May 2020 | volume 1 www.nature.com/natrevearthenviron


Reviews

event. However, statistical methods can assess the likeli- environments (such as Alberta and eastern British
hood and expected magnitude distribution of events, at Columbia). Detailed monitoring and response proto-
least in a regional sense, and such assessments are useful cols, including (but not limited to) TLPs, are a first step
inputs to probabilistic seismic-​hazard analyses. Current towards mitigation in these regions. However, at present,
statistical hazard-​assessment methods carry very large such retrospective strategies are not sufficient to pro-
(orders of magnitude) uncertainties. Nevertheless, tect critical or vulnerable infrastructure that have unac-
dense real-​time monitoring and imaging of HF oper- ceptable failure consequences. Further development of
ations (for example, using microseismicity) represents real-​time hazard forecasting and mitigation strategies is
a promising new method for tracking the seismic risk required, so that mitigation approaches can be verified
during operations. and applied to future HF operations.
The hazard from HF-​triggered seismicity exceeds
the natural hazard in low-​to-​moderate-​s eismicity Published online 7 May 2020

1. Healy, J. H., Rubey, W. W., Griggs, D. T. & Raleigh, C. B. 19. Green, C. A., Styles P. & Baptie, B. J. Preese Hall 36. Maxwell, S. C. et al. Fault activation during hydraulic
The Denver earthquakes. Science 161, 1301–1310 shale gas fracturing review and recommendations for fracturing. SEG Tech. Program Expand. Abstr. https://
(1968). induced seismic mitigation (Department of Energy doi.org/10.1190/1.3255145 (2009).
2. Raleigh, C. B., Healy, J. H. & Bredehoeft, J. D. and Climate Change, 2012). 37. Kettlety, T., Verdon, J. P., Werner, M. J., Kendall, J.-M.
An experiment in earthquake control at Rangely, 20. Lei, X., Wang, Z. & Su, J. The December 2018 ML 5.7 & Budge, J. Investigating the role of elastostatic stress
Colorado. Science 191, 1230–1237 (1976). and January 2019 ML 5.3 earthquakes in South transfer during hydraulic fracturing-​induced fault
3. Simpson, D. & Leith, W. The 1976 and 1984 Gazli, Sichuan Basin induced by shale gas hydraulic activation. Geophys. J. Int. 217, 1200–1216 (2019).
USSR, earthquakes – Were they induced? Bull. fracturing. Seismol. Res. Lett. 90, 1099–1110 (2019). 38. Shapiro, S. A. & Dinske, C. Fluid-​induced seismicity:
Seismol. Soc. Am. 75, 1465–1468 (1985). Details of the largest HF-​triggered event to date, pressure diffusion and hydraulic fracturing. Geophys.
4. Wetmiller, R. Earthquakes near Rocky Mountain with billions of dollars in damages and several Prospecting 57, 301–310 (2009).
House, Alberta, and their relationship to gas deaths. 39. Holland, A. Earthquakes triggered by hydraulic
production facilities. Can. J. Earth Sci. 23, 172–181 21. Lei, X. et al. Fault reactivation and earthquakes with fracturing in south-​central Oklahoma. Bull. Seismol.
(1986). magnitudes of up to Mw4.7 induced by shale-​gas Soc. Am. 103, 1784–1792 (2013).
5. McGarr, A. On a possible connection between three hydraulic fracturing in Sichuan Basin, China. Sci. Rep. 40. Westaway, R. Integrating induced seismicity with rock
major earthquakes in California and oil production. 7, 7971 (2017). mechanics: a conceptual model for the 2011 Preese
Bull. Seismol. Soc. Am. 81, 948–970 (1991). 22. Langenbruch, C., Ellsworth, W. L., Woo, J. U. & Hall fracture development and induced seismicity.
6. Horner, R., Barclay, J. & MacRae, J. Earthquakes and Wald, D. J. Value at induced risk: injection-​induced Geol. Soc. Spec. Publ. 454, 327–359 (2017).
hydrocarbon production in the Fort St. John area of seismic risk from low-​probability, high-​impact events. 41. Talwani, P. & Acree, S. in Earthquake Prediction. Pure
northeastern British Columbia. Can. J. Explor. Geophys. Res. Lett. 47, e2019GL085878 (2020). and applied geophysics (eds Shimazaki, K. & Stuart,
Geophys. 30, 39–50 (1994). 23. Van der Baan, M. & Calixto, F. J. Human-​induced W.) 947–965 (Birkhäuser, 1985).
7. Baranova, V., Mustaqeem, A. & Bell, S. A model for seismicity and large-​scale hydrocarbon production in 42. Shapiro, S., Huenges, E. & Borm, G. Estimating the
induced seismicity caused by hydrocarbon production the USA and Canada. Geochem. Geophys. Geosyst. crust permeability from fluid-​injection-​induced seismic
in the Western Canada Sedimentary Basin. Can. J. 18, 2467–2485 (2017). emission at the KTB site. Geophys. J. Int. 131,
Earth Sci. 36, 47–64 (1999). 24. Ghofrani, H. & Atkinson, G. M. A preliminary F15–F18 (1997).
8. McGarr, A., Simpson, D., Seeber, L. & Lee, W. Case statistical model for hydraulic fracture-​induced 43. Langenbruch, C., Weingarten, M. & Zoback, M. D.
histories of induced and triggered seismicity. Int. seismicity in the Western Canada Sedimentary Basin. Physics-​based forecasting of man-​made earthquake
Handb. Earthq. Eng. Seismol. 81A, 647–664 (2002). Geophys. Res. Lett. 43, 10–64 (2016). hazards in Oklahoma and Kansas. Nat. Commun. 9,
9. National Research Council. Induced Seismicity 25. Mayerhofer, M. J. et al. What is stimulated reservoir 3946 (2018).
Potential in Energy Technologies (National Academies volume? SPE Prod. Oper. 25, 89–98 (2010). 44. Igonin N., Verdon, J. P., Kendall, M. & Eaton, D.
Press, 2013). 26. Warpinski, N. R. & Wolhart, S. A validation assessment The role of parallel fracture networks for induced
10. Eaton, D. W. Passive Seismic Monitoring of Induced of microseismic monitoring. SPE Hydraul. Fracturing seismicity in the Duvernay Formation. EAGE Conf.
Seismicity: Fundamental Principles and Application to Technol. Conf. https://doi.org/10.2118/179150-​MS Exhibit. Proc. 2019, 1–5 (2019).
Energy Technologies (Cambridge Univ. Press, 2018). (2016). 45. Segall, P. & Lu, S. Injection-​induced seismicity:
11. Ellsworth, W. L. Injection-​induced earthquakes. 27. Schultz, R., Stern, V., Novakovic, M., Atkinson, G. poroelastic and earthquake nucleation effects.
Science 341, 1225942 (2013). & Gu, Y. Hydraulic fracturing and the Crooked Lake J. Geophys. Res. Solid Earth 120, 5082–5103 (2015).
Foundational paper on the rise of injection-​induced sequences: Insights gleaned from regional seismic 46. Kettlety T., Verdon, J. P., Werner, M. J. & Kendall, J.-M.
earthquakes in the central USA and key mechanism networks. Geophys. Res. Lett. 42, 2750–2758 (2015). Stress transfer from hydraulic fracture opening
of triggering-​induced seismicity. 28. Bao, X. & Eaton, D. W. Fault activation by hydraulic controlling injection-​induced fault activation.
12. Atkinson, G. M. et al. Hydraulic fracturing and fracturing in Western Canada. Science 354, J. Geophys. Res. Solid Earth 125, e2019JB018794
seismicity in the Western Canada Sedimentary Basin. 1406–1409 (2016). (2020).
Seismol. Res. Lett. 87, 631–647 (2016). First detailed analysis of HF-​triggered seismicity, 47. Garagash, D. I. & Germanovich, L. N. Nucleation
First clear demonstration that HF is a dominant linking HF operations to several distinct clusters of and arrest of dynamic slip on a pressurized fault.
triggering mechanism of seismicity in western seismicity and showing evidence of both pore J. Geophys. Res. Solid Earth 117, B10310 (2012).
Canada and that there are event outliers above pressure and stress perturbation activating pre-​ 48. Guglielmi, Y., Cappa, F., Avouac, J.-P., Henry, P. &
the McGarr upper magnitude bound for injection-​ existing faults. Elsworth, D. Seismicity triggered by fluid injection–
triggered seismicity. 29. Deng, K., Liu, Y. & Harrington, R. M. Poroelastic stress induced aseismic slip. Science 348, 1224–1226
13. Atkinson, G. Strategies to prevent damage to critical triggering of the December 2013 Crooked Lake, (2015).
infrastructure due to induced seismicity. FACETS 2, Alberta, induced seismicity sequence. Geophys. Res. 49. Kohli, A. H. & Zoback, M. D. Frictional properties of
374–394 (2017). Lett. 43, 8482–8491 (2016). shale reservoir rocks. J. Geophys. Res. Solid Earth
14. Weingarten, M., Ge, S., Godt, J. W., Bekins, B. & 30. Zoback, M. D. Reservoir Geomechanics (Cambridge 118, 5109–5125 (2013).
Rubinstein, J. High-​rate injection is associated with Univ. Press, 2010). 50. Scholz, C. Earthquakes and friction laws. Nature 391,
the increase in US mid-​continent seismicity. Science 31. Shapiro, S. A., Dinske, C., Langenbruch, C. & Wenzel, F. 37–42 (1998).
348, 1336–1340 (2015). Seismogenic index and magnitude probability of 51. Eyre, T. S. et al. The role of aseismic slip in hydraulic
15. King, G. E. Hydraulic fracturing 101: what every earthquakes induced during reservoir fluid fracturing-​induced seismicity. Sci. Adv. 5, eaav7172
representative, environmentalist, regulator, reporter, stimulations. Lead. Edge 29, 304–309 (2010). (2019).
investor, university researcher, neighbor and engineer 32. Langenbruch, C. & Shapiro, S. A. Quantitative analysis High-​resolution analysis of HF-​triggered seismicity
should know about estimating frac risk and improving of rock stress heterogeneity: Implications for the showing events triggered at a depth 100s of
frac performance in unconventional gas and oil wells. seismogenesis of fluid-​injection-​induced seismicity. metres above the injection zone, with evidence of
SPE Hydraul. Fracturing Technol. Conf. https://doi. Geophysics 80, WC73–WC88 (2015). aseismic slip and long-​lived seismicity after the
org/10.2118/152596-​MS (2012). 33. Keranen, K., Weingarten, M., Abers, G., Bekins, B. & completion of injection.
16. Nicholson, C. & Wesson, R. L. Triggered earthquakes Ge, S. Sharp increase in central Oklahoma seismicity 52. Bhattacharya, P. & Viesca, R. C. Fluid-​induced
and deep well activities. Pure Appl. Geophys. 139, since 2008 induced by massive wastewater injection. aseismic fault slip outpaces pore-​fluid migration.
561–578 (1992). Science 345, 448–451 (2014). Science 364, 464–468 (2019).
17. Dusseault, M. & McLennan, J. Massive multi-​stage 34. Goebel, T. H. W., Weingarten, M., Chen, X., Haffener, J. 53. Wang, B., Harrington, R. M., Liu, Y., Kao, H. & Yu, H.
hydraulic fracturing: where are we? 45th US Rock & Brodsky, E. E. The 2016 Mw5.1 Fairview, Oklahoma Remote dynamic triggering of earthquakes in three
Mechanics/Geomechanics Symp. http://tm.spbstu.ru/ earthquakes: Evidence for long-​range poroelastic unconventional Canadian hydrocarbon regions based
images/f/ff/Review_Massive_Multi-​Stage_Hydraulic_ triggering at >40 km from fluid disposal wells. on a multiple-​station matched-​filter approach. Bull.
Fracturing_Where_are_We.pdf (2011). Earth Planet. Sci. Lett. 472, 50–61 (2017). Seismol. Soc. Am. 109, 372–386 (2019).
18. Clarke, H., Eisner, L., Styles, P. & Turner, P. Felt 35. Shapiro, S. A. & Dinske, C. On stress drop, cohesion 54. Friberg, P. A., Besana-​Ostman, G. M. & Dricker, I.
seismicity associated with shale gas hydraulic and seismogenic index of fluid-​induced seismicity. Characterization of an earthquake sequence triggered
fracturing: the first documented example in Europe. ESSOAr https://doi.org/10.1002/essoar.10501558.1 by hydraulic fracturing in Harrison County, Ohio.
Geophys. Res. Lett. 41, 8308–8314 (2014). (2020). Seismol. Res. Lett. 85, 1295–1307 (2014).

NaTure RevIews | EaRTH & EnvIRonMenT volume 1 | May 2020 | 275


Reviews

55. Skoumal, R. J., Brudzinski, M. R. & Currie, B. S. 74. Cesca, S., Rohr, A. & Dahm, T. Discrimination of J. Geophys. Res. Solid Earth 121, 4575–4590
Earthquakes induced by hydraulic fracturing in Poland induced seismicity by full moment tensor inversion (2016).
Township, Ohio. Bull. Seismol. Soc. Am. 105, and decomposition. J. Seismol. 17, 147–163 Seminal paper on relationships between injected
189–197 (2015). (2013). fluid volume, seismogenic index, b-​value and
56. Schultz, R. et al. The Cardston earthquake swarm and 75. Ogata, Y. Statistical models for earthquake maximum observed magnitude; maximum induced
hydraulic fracturing of the Exshaw Formation (Alberta occurrences and residual analysis for point processes. earthquake magnitudes are consistent with
Bakken play). Bull. Seismol. Soc. Am. 105, J. Am. Stat. Assoc. 83, 9–27 (1988). Gutenberg–Richter sampling statistics.
2871–2884 (2015). 76. Llenos, A. L. & Michael, A. J. Characterizing 97. Wessels, S., Kratz, M. & De La Pena, A. Identifying
57. Schultz, R., Atkinson, G. M., Eaton, D. W., Gu, Y. J. potentially induced earthquake rate changes in the fault activation during hydraulic stimulation in the
& Kao, H. Hydraulic fracturing completion volume is Brawley seismic zone, southern California. Bull. Barnett shale: source mechanisms, b values, and
associated with induced earthquake productivity in Seismol. Soc. Am. 106, 2045–2062 (2016). energy release analysis of microseismicity. SEG Annu.
the Duvernay play. Science 359, 304–308 (2018). 77. Lei, X. et al. A detailed view of the injection-​induced Meet. (2011).
Canadian case studies showing that seismic seismicity in a natural gas reservoir in Zigong, 98. Goertz-​Allmann, B. P. & Wiemer, S. Geomechanical
productivity scales linearly with injection volume southwestern Sichuan Basin, China. J. Geophys. Res. modeling of induced seismicity source parameters and
and using a modified framework for induced Solid Earth 118, 4296–4311 (2013). implications for seismic hazard assessment.
seismicity forecasting relying on the seismogenic 78. Zaliapin, I. & Ben-​Zion, Y. Discriminating characteristics Geophysics 78, KS25–KS39 (2013).
index. of tectonic and human-​induced seismicity. Bull. 99. Petersen, M. D. et al. Seismic-​hazard forecast for
58. Mahani, A. B. et al. Fluid injection and seismic activity Seismol. Soc. Am. 106, 846–859 (2016). 2016 including induced and natural earthquakes in
in the northern Montney play, British Columbia, 79. BC Oil and Gas Commission. Investigation of the central and eastern United States. Seismol. Res.
Canada, with special reference to the 17 August 2015 observed seismicity in the Horn River Basin Lett. 87, 1327–1341 (2016).
Mw 4.6 induced earthquake. Bull. Seismol. Soc. Am. (BCOGC, 2012). 100. White, I., Liu, T., Luco, N. & Liel, A. Comparisons
107, 542–552 (2017). 80. Farahbod, A. M., Kao, H., Cassidy, J. F. & Walker, D. between the 2016 USGS induced-​seismicity hazard
59. Rich, J., Bailey, A., Jreij, S. & Klepacki, D. High-​ How did hydraulic-​fracturing operations in the Horn model, “Did You Feel It?” data, and instrumental data.
resolution insights into hydraulic fracturing strike-​slip River Basin change seismicity patterns in northeastern Seismol. Res. Lett. 89, 127–137 (2017).
seismicity: hypocenter uncertainty, depth of initiation, British Columbia, Canada? Lead. Edge 34, 658–663 101. Lee, K. K. et al. Managing injection-​induced seismic
and genesis mechanisms. SEG Int. Expo. Annu. Meet. (2015). risks. Science 364, 730–732 (2019).
(2019). 81. Hicks, S. P. et al. A shallow earthquake swarm close to Learnings from the Pohang EGS project show that
60. Clarke, H., Verdon, J. P., Kettlety, T., Baird, A. F. & hydrocarbon activities: discriminating between natural induced earthquake magnitudes are not limited by
Kendall, J.-M. Real-​time imaging, forecasting, and and induced causes for the 2018–19 Surrey, UK injection volume and that the largest earthquake
management of human-​induced seismicity at Preston earthquake sequence. Seismol. Res. Lett. 90, was a runaway earthquake; highlights need to
New Road, Lancashire, England. Seismol. Res. Lett. 2095–2110 (2019). consider risk, not just hazard.
90, 1902–1915 (2019). 82. Sileny, J., Hill, D. P., Eisner, L. & Cornet, F. H. Non– 102. Giardini, D. Geothermal quake risks must be faced.
61. Wright, G. N., McMechan, M. E. & Potter, D. E. double-​couple mechanisms of microearthquakes Nature 462, 848–849 (2009).
Structure and architecture of the Western Canada induced by hydraulic fracturing. J. Geophys. Res. 103. Yeck, W. L. et al. Oklahoma experiences largest
sedimentary basin. In Geological Atlas of the Western Solid. Earth 114, B08307 (2009). earthquake during ongoing regional wastewater
Canada Sedimentary Basin, G.D. Mossop and I. 83. Baig, A. & Urbancic, T. Microseismic moment tensors: injection hazard mitigation efforts. Geophys. Res. Lett.
Shetsen (comp.). Can. Soc. Pet. Geologists Alta. Res. a path to understanding frac growth. Lead. Edge 29, 44, 711–717 (2017).
Counc. 4, 25–40 (1994). 320–324 (2010). 104. Atkinson, G. M. Ground-​motion prediction equation
62. Shen, C. B., Mei, L. F., Xu, Z. P. & Tang, J. G. 84. Eaton, D. W., van der Baan, M., Birkelo, B. & Tary, J. B. for small-​to-​moderate events at short hypocentral
Architecture and tectonic evolution of composite Scaling relations and spectral characteristics of tensile distances, with application to induced seismicity
basin-​mountain system in Sichuan basin and its microseisms: Evidence for opening/closing cracks hazards. Bull. Seismol. Soc. Am. 105, 981–992
adjacent areas. Geotecton. Metallog. 31, 288–299 during hydraulic fracturing. Geophys. J. Int. 196, (2015).
(2007). 1844–1857 (2014). 105. Rennolet, S. B., Moschetti, M. P., Thompson, E. M.
63. Clarke, H., Turner, P. & Bustin, R. Unlocking the 85. Zhang, H., Eaton, D. W., Rodriguez, G. & Jia, S. Q. & Yeck, W. L. A flatfile of ground motion intensity
resource ppotential of the Bowland Basin, NW Source-​mechanism analysis and stress inversion for measurements from induced earthquakes in
England. SPE/EAGE Eur. Unconvent. Resour. Conf. hydraulic-​fracturing-​induced event sequences near Fox Oklahoma and Kansas. Earthq. Spectra 34, 1–20
Exhibit. https://doi.org/10.2118/167776-​MS (2014). Creek, Alberta. Bull. Seismol. Soc. Am. 109, 636–651 (2018).
64. Verdon, J. P. & Budge, J. Examining the capability of (2019). 106. Novakovic, M., Atkinson, G. M. & Assatourians, K.
statistical models to mitigate induced seismicity 86. Eyre, T. S., Eaton, D. W., Zecevic, M., D’Amico, D. Empirically calibrated ground-​motion prediction
during hydraulic fracturing of shale gas reservoirs. & Kolos, D. Microseismicity reveals fault activation equation for Oklahoma. Bull. Seismol. Soc. Am. 108,
Bull. Seismol. Soc. Am. 108, 690–701 (2018). before Mw4.1 hydraulic-​fracturing induced 2444–2461 (2018).
65. Ghofrani, H. & Atkinson, G. Updated statistics for earthquake. Geophys. J. Int. 218, 534–546 (2019). 107. Novakovic, M. & Atkinson, G. M. Preliminary
seismicity induced by hydraulic fracturing in the 87. Zhang, H., Eaton, D. W., Li, G., Liu, Y. & Harrington, R. M. evaluation of ground motions from earthquakes in
Western Canada sedimentary basin. Bull. Seismol. Discriminating induced seismicity from natural Alberta. Seismol. Res. Lett. 86, 1086–1095 (2015).
Soc. Am. https://doi.org/10.1785/0120200002 earthquakes using moment tensors and source 108. Mahani, A. B. & Kao, H. Ground motion from
(2020). spectra. J. Geophys. Res. Solid Earth 121, 972–993 M1.5–3.8 induced earthquakes at hypocentral
Special issue on induced seismicity. (2016). distance <45 km in the Montney Play of northeast
66. Eaton, D. W. & Schultz, R. Increased likelihood of 88. Huang, Y., Ellsworth, W. L. & Beroza, G. C. Stress British Columbia, Canada. Seismol. Res. Lett. 89,
induced seismicity in highly overpressured shale drops of induced and tectonic earthquakes in the 22–34 (2018).
formations. Geophys. J. Intern. 214, 751–757 central United States are indistinguishable. Sci. Adv. 109. Mahani, A. B. et al. Ground-​motion characteristics of
(2018). 8, e1700772 (2017). the 30 November 2018 injection-​induced earthquake
67. Schultz, R. et al. Linking fossil reefs with earthquakes: 89. Holmgren, J. M., Atkinson, G. M. & Ghofrani, H. sequence in Northeast British Columbia, Canada.
Geologic insight to where induced seismicity occurs in Stress drops and directivity of induced earthquakes in Seismol. Res. Lett. 90, 1457–1467 (2019).
Alberta. Geophys. Res. Lett. 43, 2534–2542 (2016). the western Canada sedimentary basin. Bull. Seismol. 110. Holmgren, J. M., Atkinson, G. M. & Ghofrani, H.
68. Pawley, S. et al. The geological susceptibility of Soc. Am. 109, 1635–1652 (2019). Reconciling ground motions and stress drops for
induced earthquakes in the Duvernay play. Geophys. 90. Yenier, E. & Atkinson, G. A regionally-​adjustable induced earthquakes in the western Canada
Res. Lett. 45, 1786–1793 (2018). generic GMPE based on stochastic point-​source sedimentary Basin. Bull. Seismol. Soc. Am. https://doi.
69. Davis, S. & Frohlich, C. Did (or will) fluid injection simulations. Bull. Seismol. Soc. Am. 105, 1989–2009 org/10.1785/0120190308 (2020).
cause earthquakes? Criteria for a rational assessment. (2015). 111. Hough, S. E. Shaking from injection-​induced
Seismol. Res. Lett. 64, 207–224 (1993). 91. Atkinson, G. M. & Assatourians, K. Are ground-​motion earthquakes in the central and eastern United States.
70. Montalvo-​Arrieta, J. C. et al. El Cuchillo seismic models derived from natural events applicable to the Bull. Seismol. Soc. Am. 104, 2619–2626 (2014).
sequence of October 2013–July 2014 in the Burgos estimation of expected motions for induced 112. Allmann, B. P. & Shearer, P. M. Global variations
Basin, northeastern Mexico: Hydraulic fracturing or earthquakes? Seismol. Res. Lett. 88, 430–441 of stress drop for moderate to large earthquakes.
reservoir-​induced seismicity? Bull. Seismol. Soc. Am. (2017). J. Geophys. Res. 114, B01310 (2009).
108, 3092–3106 (2018). 92. Atkinson, G. M. The intensity of ground motions 113. Radiguet, M., Cotton, F., Manighetti, I., Campillo, M.
71. Verdon, J. P., Baptie, B. J. & Bommer, J. J. An from induced earthquakes with implications for & Douglas, J. Dependency of near-​field ground
improved framework for discriminating seismicity damage potential. Bull. Seismol. Soc. Am. 110, 1–14 motions on the structural maturity of the ruptured
induced by industrial activities from natural (2020). faults. Bull. Seismol. Soc. Am. 99, 2572–2581 (2009).
earthquakes. Seismol. Res. Lett. 90, 1592–1611 93. Schultz, R., Wang, R., Gu, Y. J., Haug, K. & Atkinson, G. 114. Bommer, J. J. et al. Control of hazard due to seismicity
(2019). A seismological overview of the induced earthquakes induced by a hot fractured rock geothermal project.
72. Dahm, T., Cesca, S., Hainzl, S., Braun, T. & Krüger, F. in the Duvernay play near Fox Creek, Alberta. Eng. Geol. 83, 287–306 (2006).
Discrimination between induced, triggered, and J. Geophys. Res. Solid Earth 122, 492–505 (2017). 115. Westaway, R. & Younger, P. L. Quantification of
natural earthquakes close to hydrocarbon reservoirs: 94. Poulin, A. et al. Focal-​time analysis: A new method potential macroseismic effects of the induced
a probabilistic approach based on the modeling of for stratigraphic depth control of microseismicity seismicity that might result from hydraulic fracturing
depletion-​induced stress changes and seismological and induced seismic events. Geophysics 84, for shale gas exploitation in the UK. Q. J. Eng. Geol.
source parameters. J. Geophys. Res. 120, KS173–KS182 (2019). Hydrogeol. 47, 333–350 (2014).
2491–2509 (2015). 95. Gutenberg, R. & Richter, C. F. Frequency of earthquakes 116. US Geological Survey. The severity of an earthquake.
73. Schoenball, M., Davatzes, N. & Glen, J. M. in California. Bull. Seismol. Soc. Am. 34, 185–188 USGS https://pubs.usgs.gov/gip/earthq4/severitygip.
Differentiating induced and natural seismicity using (1944). html (1989).
space–time–magnitude statistics applied to the Coso 96. Van der Elst, N. J., Page, M. T., Weiser, D. A., 117. Wilson, M. P., Worrall, F., Davies, R. J. & Almond, S.
Geothermal field. Geophys. Res. Lett. 42, Goebel, T. H. & Hosseini, S. M. Induced earthquake Fracking: how far from faults? Geomech. Geophys.
6221–6228 (2015). magnitudes are as large as (statistically) expected. Geoenergy Georesources 4, 193–199 (2018).

276 | May 2020 | volume 1 www.nature.com/natrevearthenviron


Reviews

118. Diehl, T., Kraft, T., Kissling, E. & Wiemer, S. 132. Galis, M., Ampuero, J. P., Mai, P. M. & Cappa, F. 150. Atkinson, G. M., Ghofrani, H. & Assatourians, K.
The induced earthquake sequence related to the Induced seismicity provides insight into why Impact of induced seismicity on the evaluation of
St. Gallen deep geothermal project (Switzerland): earthquake ruptures stop. Sci. Adv. 3, eaap7528 seismic hazard: some preliminary considerations.
Fault reactivation and fluid interactions imaged by (2017). Seismol. Res. Lett. 86, 1009–1021 (2015).
microseismicity. J. Geophys. Res. 122, 7272–7290 133. Mignan, A., Broccardo, M., Wiemer, S. & Giardini, D. 151. Norbeck, J. H. & Rubinstein, J. L. Hydromechanical
(2017). Induced seismicity closed-​form traffic light system for earthquake nucleation model forecasts onset, peak,
119. Eaton, D. W. et al. Induced seismicity characterization actuarial decision-​making during deep fluid injections. and falling rates of induced seismicity in Oklahoma
during hydraulic-​fracture monitoring with a shallow-​ Nat. Sci. Rep. 7, 13607 (2017). and Kansas. Geophys. Res. Lett. 45, 2963–2975
wellbore geophone array and broadband sensors. 134. Broccardo, M., Mignan, A., Wiemer, S., Stojadinovic, B. (2018).
Seismol. Res. Lett. 89, 1641–1651 (2018). & Giardini, D. Hierarchical Bayesian modeling of 152. Walters, R. J., Zoback, M. D., Baker, J. W. &
120. Schoenball, M. & Ellsworth, W. L. Waveform-​relocated fluid-​induced seismicity. Geophys. Res. Lett. 44, Beroza, G. C. Characterizing and responding to seismic
earthquake catalog for Oklahoma and southern 11357–11367 (2017). risk associated with earthquakes potentially triggered
Kansas illuminates the regional fault network. Seismol. 135. Mignan, A., Karvounis, D., Broccardo, M., Wiemer, S. by fluid disposal and hydraulic fracturing. Seismol.
Res. Lett. 88, 1252–1258 (2017). & Giardini, D. Including seismic risk mitigation Res. Lett. 86, 1110–1118 (2015).
121. Häring, M. O., Schanz, U., Ladner, F. & Dyer, B. C. measures into the Levelized Cost of Electricity in 153. Grigoli, F. et al. Current challenges in monitoring,
Characterisation of the Basel 1 enhanced geothermal enhanced geothermal systems for optimal siting. discrimination, and management of induced seismicity
system. Geothermics 37, 469–495 (2008). Appl. Energy 238, 831–850 (2019). related to underground industrial activities: A
122. Baisch, S., Koch, C. & Muntendam-​Bos, A. Traffic light 136. Langenbruch, C. & Shapiro, S. A. Decay rate of fluid-​ European perspective. Rev. Geophys. 55, 310–340
systems: to what extent can induced seismicity be induced seismicity after termination of reservoir (2017).
controlled? Seismol. Res. Lett. 90, 1145–1154 stimulations. Geophysics 75, MA53–MA62 (2010). 154. Fasola, S. L. et al. Hydraulic fracture injection strategy
(2019). 137. Barth, A., Wenzel, F. & Langenbruch, C. Probability of influences the probability of earthquakes in the Eagle
Shows that traffic light approaches are limited in earthquake occurrence and magnitude estimation Ford shale play of South Texas. Geophys. Res. Lett.
their ability to predict and suggests that their in the post shut-​in phase of geothermal projects. 46, 12958–12967 (2019).
underlying assumptions are generally not valid. J. Seismol. 17, 5–11 (2013). 155. Schultz, R. & Wang, R. Newly emerging cases of
123. Sumy, D. F., Cochran, E. S., Keranen, K. M., Wei, M. & 138. Wesnousky, S. G. The Gutenberg-​Richter or hydraulic fracturing induced seismicity in the Duvernay
Abers, G. A. Observations of static Coulomb stress characteristic earthquake distribution, which is it? East Shale Basin. Tectonophysics 779, 228393 (2020).
triggering of the November 2011 M5.7 Oklahoma Bull. Seismol. Soc. Am. 84, 1940–1959 (1994). 156. Brudzinski, M. R. & Kozłowska, M. Seismicity induced
earthquake sequence. J. Geophys. Res. 119, 139. Gischig, V. S. Rupture propagation behavior and the by hydraulic fracturing and wastewater disposal in the
1904–1923 (2014). largest possible earthquake induced by fluid injection Appalachian Basin, USA: a review. Acta Geophysica
124. Dieterich, J. H., Richards-​Dinger, K. B. & Kroll, K. A. into deep reservoirs. Geophys. Res. Lett. 42, 67, 351–364 (2019).
Modeling injection-​induced seismicity with the 7420–7428 (2015). 157. López-​Comino, J. A. et al. Induced seismicity response
physics-​based earthquake simulator RSQSim. Seismol. 140. Eaton, D. W. & Igonin, N. What controls the maximum of hydraulic fracturing: results of a multidisciplinary
Res. Lett. 86, 1102–1109 (2015). magnitude of injection-​induced earthquakes? Lead. monitoring at the Wysin site, Poland. Sci. Rep. 8,
125. Verdon, J. P., Stork, A. L., Bissell, R. C., Bond, C. E. & Edge. 37, 135–140 (2018). 8653 (2018).
Werner, M. J. Simulation of seismic events induced by 141. Cornell, C. A. Engineering seismic risk analysis. 158. Foulger, G. R., Wilson, M. P., Gluyas, J. G., Julian, B. R.
CO2 injection at In Salah, Algeria. Earth Planet. Sci. Bull. Seismol. Soc. Am. 58, 1583–1606 (1968). & Davies, R. J. Global review of human-​induced
Lett. 426, 118–129 (2015). 142. McGuire, R. K. Seismic Hazard and Risk Analysis earthquakes. Earth Sci. Rev. 178, 438–514 (2018).
126. Shapiro, S. A., Kruger, O. S., Dinske, C. & (Earthquake Engineering Research Institute, 2004).
Langenbruch, C. Magnitudes of induced earthquakes 143. Petersen, M. D. et al. Documentation for the 2008 Acknowledgements
and geometric scales of fluid-​stimulated rock volumes. update of the United States national seismic hazard The authors receive financial support for their research pro-
Geophysics 76, WC55–WC63 (2011). maps (USGS, 2008). grammes from the Natural Sciences and Engineering
127. Hallo, M., Oprsal, I., Eisner, L. & Ali, M. Y. Prediction 144. Halchuk, S., Allen, T., Adams, J. & Rogers, G. C. Research Council of Canada. We thank James Verdon for
of magnitude of the largest potentially induced seismic Fifth generation seismic hazard model input files as constructive discussions that contributed to the manuscript
event. J. Seismol. 18, 421–431 (2014). proposed to produce values for the 2015 national and Minhee Choi for assistance in preparing the final
128. Kwiatek, G. et al. Controlling fluid-​induced seismicity building code of Canada (Geological Survey of Canada, manuscript.
during a 6.1-​km-​deep geothermal stimulation in 2014).
Finland. Sci. Adv. 5, .eaav7224 (2019). 145. Canadian Dam Association. Seismic hazard Author contributions
129. McGarr, A. Maximum magnitude earthquakes induced considerations for dam safety (CDA, 2007). All authors contributed to the research, writing, figure
by fluid injection. J. Geophys. Res. 119, 1008–1019 146. Reiter, L. Earthquake Hazard Analysis; Issues and preparation and editing of this Review.
(2014). Insights (Columbia Univ. Press, 1990).
Highly cited empirical relation between injected 147. Alberta Energy Regulator. Initial seismic hazard Competing interests
volume and maximum observed magnitude, assessment for the 2016 induced earthquakes near The authors declare no competing interests.
including presentation of an underlying physical Fox Creek, Alberta (AER, 2017).
model and its assumptions. 148. Ghofrani, H., Atkinson, G. M., Schultz, R. & Peer review information
130. Igonin, N., Zecevic, M. & Eaton, D. W. Bilinear Assatourians, K. Short-​term hindcasts of seismic Nature Reviews Earth & Environment thanks X. Lei,
magnitude-​frequency distributions and characteristic hazard in the western Canada sedimentary basin C. Langenbruch and the other, anonymous, reviewer(s) for
earthquakes during hydraulic fracturing. Geophys. caused by induced and natural earthquakes. Seismol. their contribution to the peer review of this work.
Res. Lett. 45, 12–866 (2018). Res. Lett. 90, 1420–1435 (2019).
131. Ellsworth, W. L., Giardini, D., Townend, J., Ge, S. 149. Skoumal, R. J., Ries, R., Brudzinski, M. R., Publisher’s note
& Shimamoto, T. Triggering of the Pohang, Korea, Barbour, A. J. & Currie, B. S. Earthquakes induced Springer Nature remains neutral with regard to jurisdictional
earthquake (Mw 5.5) by enhanced geothermal system by hydraulic fracturing are pervasive in Oklahoma. claims in published maps and institutional affiliations.
stimulation. Seismol. Res. Lett. 90, 1844–1858 J. Geophys. Res. Solid Earth 123, 10918–10935
(2019). (2018). © Springer Nature Limited 2020

NaTure RevIews | EaRTH & EnvIRonMenT volume 1 | May 2020 | 277

You might also like