COMMENTARY Induced Seismicity Reduces Seismic Hazard?

10.1029/2019GL081991
Andrew J. Barbour1 and Fred Pollitz1
Key Points: 1
• Geodetic data indicate a close
United States Geological Survey, Menlo Park, CA, USA
relationship between crustal strain
rates and induced earthquakes
• Induced earthquakes might reduce Abstract Earthquakes caused by human activities have been observed for decades. Often these are
the seismic hazard in the long run related to industrial activities pumping fluids into deep geologic formations, like with wastewater
• Hazard‐reducing scenarios hinge
critically on a few major
disposal. The simplest theory connecting these processes to earthquakes is straightforward: injection leads to
assumptions fluid pressure changes that either reduce the strength of preexisting faults or generate new faults. In practice,
the conditions that lead to induced earthquakes are not always clear in ways that can be generalized.
Kao et al. (2018, https://doi.org/10.1029/2018GL079288) show how the distribution of induced earthquakes
Correspondence to: in Western Canada relate to natural rates of deformation in the crust. Using these new results, they discuss
A. J. Barbour, an intriguing paradox: induced seismicity can cause short‐term increases in the seismic hazard that are
abarbour@usgs.gov followed by a period of reduced seismic hazard. Such hazard‐reducing scenarios are plausible but hinge
upon simplifying assumptions about how the crust stores and releases strain energy in the form
Citation: of earthquakes.
Barbour, A. J., & Pollitz, F. (2019).
Induced seismicity reduces seismic
hazard?. Geophysical Research Letters,
46. https://doi.org/10.1029/ The slow migration and deformation of Earth's tectonic plates leads to a process of energy storage and
2019GL081991
release. Because this happens over geologic timescales, we have obvious expressions of this process to study
Received 9 JAN 2019 including active faults, volcanic eruptions, or ongoing uplift of mountain ranges, for example.
Accepted 30 MAR 2019
Accepted article online 5 APR 2019
Earthquakes reflect this process on timescales that humans may experience in their lifetime. They occur on
faults where deformation in the crust has concentrated into a narrow planar zone of fault slip; the seismic
waves from great earthquakes can cause the earth to ring like a bell. We can identify the time, location,
and size of an earthquake by its signals recorded on a collection of seismometers—instruments that measure
Earth's vibrations. We can also identify the style of faulting in an earthquake by looking at details of the
seismic signals and use that information to determine the stresses acting on the fault.
During the periods between earthquakes, the crust is relentlessly warped, stretched, and compressed. Earth
scientists call this strain accumulation, which builds up over decades or centuries until it is released by earth-
quakes; measuring this slow strain accumulation requires high‐precision instruments. That is—we need to
measure changes of about one part in a billion to measure typical changes in strain. If we had a rubber band
stretched across the width of Greenland, we would need to measure a change in its length equal to the thick-
ness of a few human hairs; the navigation in your smartphone is not good enough to witness such small
changes, but modern geodetic technology is. Although modern geodesy gives us the ability to measure tiny
changes in strain, seismology is also a useful tool for understanding how and where the crust releases stored
strain energy.
Kao et al. (2018) study a region of Western Canada that is riddled with deep wells used for enhanced oil‐
recovery, waste‐fluid disposal, and hydraulic fracturing (or fracking) for the production of oil and gas from
“tight” reservoirs; these industrial activities have caused a remarkable increase in the number of earth-
quakes in recent years and, consequently, a change in seismic hazard. The authors report that the distribu-
tion of induced earthquakes in this region follows the distribution of strain rates in the crust, with most
earthquakes occurring where strain rates are highest. To measure strain, they use a network of Global
Navigation Satellite System (GNSS) stations, taking differences among their relative positions over time
(Figure 1). (GNSS is the generic collection of satellite positioning systems, which includes the United
States' Global Positioning System or GPS.) Whether natural or induced, the magnitude of an earthquake
is an indication of the seismic moment—the product of the distance a fault has slipped, the size of the area
that slipped, and the stiffness of the host rock. The brittle part of the crust where most earthquakes happen is
called the seismogenic zone, and its thickness can be used in conjunction with strain rates to estimate what
Published 2019. This article is a U.S.
Government work and is in the public the total seismic moment budget is; this can tell us what the largest earthquake possible in a region (like
domain in the USA. Western Canada) could be.

BARBOUR AND POLLITZ 1

Figure 1. Geodetic measurements and the idealized earthquake cycle. Strain accumulates around a fault because of fric-
tional resistance to far‐field plate motion. Eventually, stresses on the fault overcome its frictional resistance and it slips,
which releases stored strain energy and generates an earthquake (with aftershocks). This process is known as elastic
rebound. GNSS stations at the surface of a deforming crust move at different paces depending on their proximity to
the fault: The average strain rate between any two stations would be the difference in the slopes of their respective dis-
placement time series (see lower panel) divided by the distance between them. GNSS = Global Navigation Satellite System.

This leads us to the main paradox in Kao et al. (2018): creating earthquakes can lower the seismic hazard
over time. How could this be possible? To answer this, Kao et al. (2018) present a set of seismic hazard
scenarios (shown in their Figure 4) where they make different assumptions about how rates of induced
earthquakes evolve. The number of earthquakes over time—the seismicity rate—is an important clue that
crustal faults are releasing strain energy; adding up all their seismic moments tells us how much energy is
released in the form of earthquakes. Their first scenario considers only natural earthquakes, which set a
background seismicity rate: here the total moment rate does not change over time. Their second scenario
adds induced earthquakes to the collection of background events and the total moment rate increases stea-
dily to a higher constant seismicity rate. And their third scenario adds a pulse of induced earthquakes with
larger magnitudes to the second scenario causing the total moment rate to be temporarily much higher, but
this higher rate eventually returns to the long‐term rate from their second scenario. Assuming rates of
tectonic strain accumulation are constant, the connection to seismic hazard is presented as follows: higher
rates of seismicity (scenarios two and three) subtract energy from the total moment budget, which means it
takes longer for the crust to store enough seismic moment for a large earthquake compared to the time it
would've taken naturally (the first scenario). Induced earthquakes may cause seismic hazard to be tempora-
rily higher, but the strain energy they release over decades serves to (mostly) balance the strain that could
lead to a large earthquake.
These hazard‐reducing scenarios are plausible, but some readers may rightfully question assumptions about
the physics built into them, and how much we actually know about the general earthquake cycle (Figure 1).
After studying Kao et al. (2018), we have some points of discussion based on five major assumptions:
How well do we understand patterns of strain accumulation? Typically, we only observe rapid changes in
strain rate leading up to and during volcanic eruptions, or after great earthquakes like the 2011 magnitude
9 earthquake in Japan, for example. However, what is often more uncertain is what strain rates look like
when they are mapped. The technique used in Kao et al. (2018) depends on data from GNSS stations located
far from each other, so it is difficult to know what is happening between those stations. However, the distri-
bution of fracking wells may offer independent corroboration of the distribution of strain rates from these
geodetic data. The largest earthquakes in this region show evidence of reverse or thrust faulting.
Accordingly, we can say that the largest stresses in this region are horizontal and compressional, which

BARBOUR AND POLLITZ 2

shifts rock upward during an earthquake. Mechanical principals for thrust‐environments state that generat-
ing fractures through high‐pressure fluid injection (fracking stimulation) will be easiest when the crust is
stressed (deformed) before the injection starts. Mountain ranges are good indicators of compressional strain
over geologic time, and nearly all the hydraulic fracturing wells located in this region are within the transi-
tion from the Western Canada Sedimentary Basin and the mountainous region known as the Canadian
Cordillera, where the geodetically determined strain rates are highest.
Is it realistic to expect that the rate of induced earthquakes will be constant? Laboratory experiments remind us
that a fault's friction can change, which changes how it slips during an earthquake. In one of the two modes
of dynamic friction, friction decreases with increasing sliding (this is why it is more difficult to get a heavy
object to slide than it is to keep it sliding), which can promote earthquake occurrence. In the other mode
of dynamic friction, friction increases with sliding and tends to inhibit earthquake occurrence. A spectrum
of modes of dynamic friction are active in the earth, depending on conditions like temperature, pressure, and
rock type. We have seen evidence of frictional weakening from Oklahoma, for example, where the rate of
earthquakes tends to be sensitive to how fast wastewater is pumped into deep reservoirs (Norbeck &
Rubinstein, 2018). And, while we know that economic pressure or regulatory pressure will drive changes
in injection over time, it is difficult to forecast how the rates of induced earthquakes will respond:
establishing the details of how friction evolves on natural faults is an ongoing discussion.
How much stress on a fault is relieved by an earthquake, and how does that affect ground shaking? The amount
of seismic energy that is radiated throughout the crust by an earthquake—shaking the ground under our feet
—depends on its magnitude. Seismological theory states that the same reduction in stress (or stress drop)
occurs during any earthquake irrespective of its magnitude: both the slip and the area of slip increase with
increasing magnitude. In practice, it is difficult to measure an earthquake's stress drop from seismological
data for various reasons, so we end up being data poor and uncertainty rich. But, after decades of observa-
tion, it appears that this theory of constant stress drop does not always hold: faults may relieve more or less
stress than expected. Even though major fault zones separating tectonic plates account for most of the seis-
mic moment across the globe, earthquakes there relieve less stress than those on faults in a plate's interior.
Since the region of Western Canada considered by Kao et al. (2018) is considered the plate interior, ground
motions in that region may be more intense than typically expected for a given earthquake magnitude. In
addition, there may be a heterogeneous distribution of stress in the crust with localized areas of high stress
that have persisted over geologic time; these would modify the stress release patterns that we have idealized
in Figure 1, where a single, mature fault releases most or all of the accumulated strain energy.
How can we be sure that bursts of small induced earthquakes (their third scenario) will not lead to a few large,
destructive earthquakes? Our understanding of what controls the maximum magnitude of injection‐induced
earthquakes is rather limited: a small collection of case studies points to the extent of the injection activities
and the proximity to faults, while other studies point to the mechanics of how an earthquake develops and
stops. In any case it would be preferable to identify all important regional faults, particularly the longer
(≳20 km) ones capable of earthquakes with magnitudes greater than 6.5, and their proximity to injection
sites. The poor correlation of mapped faults to seismic activity in Oklahoma highlights the elusiveness of this
goal. It may be the case that we understand the average behavior of induced earthquake magnitudes well,
but not their limiting factors.
How can we be sure that earthquakes will not promote earthquakes on other faults? If longer faults are present
in the crust, whether we know about them or not, then models that account for stress transfer between past
earthquakes and potential future earthquake zones suggest that the likelihood of larger magnitude events
increases as the number of small and moderate events increases. This is supported by numerical simulations
of seismicity over very long time periods (Dieterich & Richards‐Dinger, 2010; Pollitz, 2009) as well as obser-
vational evidence, like the San Andreas fault zone after a series of magnitude ~5 foreshocks before the 1989
magnitude 6.9 Loma Prieta, California, earthquake (Sykes & Jaumé, 1990). Only when a very large event
occurs, as large as the 1906 magnitude 7.8 San Francisco earthquake, is seismic hazard dramatically reduced
by an earthquake.
Quantifying seismic hazard is fundamentally about calculating the likelihood that, within some timeframe,
an earthquake will cause the ground to shake with an intensity at or above some threshold. The threshold
and timeframe are neither concrete nor arbitrary; rather, they represent a mixture of societal experiences

BARBOUR AND POLLITZ 3

and scientific findings (e.g., Petersen et al., 2018). The issues raised above are all factors that influence the
intensity of shaking, and these are often interrelated in complex ways. For example, stress drop exerts a
large influence on shaking, but it depends on the amount of stress imposed on the fault by the
surrounding crustal rock, as well as local properties such as the sliding distance needed to substantially
reduce friction. Shaking intensity is also affected by the amount of this reduction in friction during rupture
(i.e., dynamic friction), which is controlled by local mechanical properties thought to be related to the
maturity of the fault zone. To complicate the matter, recent studies have suggested that stress drops from
induced earthquakes may be systematically different than natural earthquakes, but this is an outstanding
debate in the community. The impacts of fluid injection or other industrial activities include changes in
the distribution of stress and fluid pressure within the crust; hence, their impact on earthquake hazards
may also depend on the distribution of faults and stresses within the crust, and interactions among
neighboring faults.
There are valid criticisms to Kao et al. (2018), particularly to the conclusion that induced earthquakes will
reduce seismic hazards over decades. However, no single study can address each of these critiques. What
we can infer from this study is that the underlying data appear to be robust and their findings highlight
the crustal conditions in Western Canada that may promote induced seismicity associated with injection,
and makes the point that a multifaceted approach to understanding induced seismicity is warranted.
Finally, their work reminds us all that in connecting earthquake physics with seismic hazards, we should
always expect the unexpected.

Acknowledgments References
We thank Elizabeth Cochran, Sara
McBride, and Art McGarr for insightful Dieterich, J. H., & Richards‐Dinger, K. B. (2010). Earthquake recurrence in simulated fault systems. Pure and Applied Geophysics, 167(8‐9),
comments. We thank Gavin Hayes, 1087–1104. https://doi.org/10.1007/s00024‐010‐0094‐0
Roland Bürgmann, and two Kao, H., Hyndman, R., Jiang, Y., Visser, R., Smith, B., Babaie Mahani, A., et al. (2018). Induced seismicity in western Canada linked to
anonymous reviewers for carefully tectonic strain rate: Implications for regional seismic hazard. Geophysical Research Letters, 45, 11,104–11,115. https://doi.org/10.1029/
reading the original paper by Kao et al. 2018GL079288
(2018) and offering thoughtful Norbeck, J. H., & Rubinstein, J. L. (2018). Hydromechanical earthquake nucleation model forecasts onset, peak, and falling rates of induced
comments on this Commentary. seismicity in Oklahoma and Kansas. Geophysical Research Letters, 45, 2963–2975. https://doi.org/10.1002/2017GL076562
Petersen, M. D., Mueller, C. S., Moschetti, M. P., Hoover, S. M., Rukstales, K. S., McNamara, D. E., et al. (2018). 2018 one‐year seismic
hazard forecast for the central and eastern United States from induced and natural earthquakes. Seismological Research Letters, 89(3),
1049–1061. https://doi.org/10.1785/0220180005
Pollitz, F. F. (2009). A viscoelastic earthquake simulator, with application to the San Francisco Bay region. Bulletin of the Seismological
Society of America, 99(3), 1760–1785. https://doi.org/10.1785/0120080253
Sykes, L. R., & Jaumé, S. C. (1990). Seismic activity on neighbouring faults as a long‐term precursor to large earthquakes in the San
Francisco Bay area. Nature, 348(6302), 595–599. https://doi.org/10.1038/348595a0

BARBOUR AND POLLITZ 4