Quake, rattle, and roll

Earthquakes can be induced by underground fluid injection, which increases pore pressure and allows faults to slide under preexisting shear stress(1). The increase in wastewater disposal from natural gas development and other sourceshas been accompanied by an increase in fluid-induced earthquakes in recent years (2). These earthquakes include widely felt earthquakes in Oklahoma, Arkansas, Ohio, Texas, and Colorado (Fig. 1) (3–7). Although most injection wells are not associated with large earthquakes, the converse is not true. At least half of the 4.5 moment magnitude (Mw)or larger earthquakes to strike the interior of the United States in the past decade have occurred in regions of potential injection-induced seismicity(table S1). In some cases, the onset of seismicity follows injection by only days or weeks (1, 3, 5), and the association with pumping at particular wells is clear. In others, seismicity increases only after months or years of active injection (4, 8, 9)

A long delay before seismic activation implies that faults may be moving toward a critical state for years before failure. However, currently there are no reliable methods to determine whether a particular field has reached a critical state other than by simply observing a large increase in seismicity. This lack of diagnostics is a key problem in developing operational strategies to mitigate anthropogenic activity (2).

Because induced seismic zones are brought to failure by increased pore pressures, we examined whether areas of induced seismicity show a high susceptibility to dynamic triggering by the small transient stresses carried by seismic waves from distant earthquakes. Dynamic triggering in natural settings has been linked to the presence of subsurface fluids, and seismicity rate changes have been shown to depend systematically on the perturbation stress (10–13). This suggests that dynamic triggering could serve as a probe of the state of stress in areas of wastewater injection. We refer to earthquakes that are promoted by anthropogenic activity as induced and to earthquakes that are initiated by transient natural stresses as triggered. By this definition, there can be triggered induced earthquakes.

A search of the Advanced National Seismic System (ANSS) earthquake catalog gives preliminary evidence that induced seismic zones are sensitive to dynamic triggering by surface waves (Fig. 1). Regions of suspected induced seismicity showed a pronounced increase in 3.0 M and larger earthquakes spanning at least a 3-day window after large (M_w ≥ 8.6) remote earthquakes: the 27 February 2010 8.8 M_w Maule, Chile; 11 March 2011 9.1 M_w Tohoku-oki; and 12 April 2012 8.6 M_w Sumatra earthquakes. The broader central United States shows essentially no response to these events (Fig. 1). Most of the triggering is at three sites: Prague, Oklahoma; Snyder, Texas; and Trinidad, Colorado. Suggestively, each of these regions went on to host moderate to large earthquakes (4.3 to 5.7 M_w) within 6 to 20 months of the strong triggering.

Although the triggering is significant at the 96% level (table S2), a closer investigation is warranted. We therefore enhanced the catalog by applying a single-station matched filter to continuous waveforms (14). The matched-filter approach identifies small, uncataloged earthquakes based on their similarity to target events (15–17). Distinct families of earthquakes are distinguished based on the difference in P and S wave travel times (S-P time), which gives the approximate radial distance from the seismic station (15).

The Cogdell oil field (8), located near Snyder, Texas, hosted a seismic swarm in September 2011 that included a 4.3 M_w main shock (supplementary text). The enhanced catalog shows that the Tohoku-oki earthquake triggered a significant number of earthquakes (14) at this site (Fig. 2 and table S2). In fact, the rate of earthquakes within the 10 days after the Tohoku-oki earthquake was the highest observed over the entire study duration (February 2009 to present), excluding the days immediately after the 4.3 M_w main shock. The triggered earthquakes show a swarm like signature, typical of fluid-induced earthquakes (18), with the largest of the triggered events (3.8 M_w, ANSS) occurring after 2.5 days of smaller events (Fig. 2C). The much earlier February 2010 Maule earthquake did not trigger at Snyder, nor did the post-swarm April 2012 Sumatra earthquake.

Prague, Oklahoma, experienced three 5.0 M_w and greater earthquakes in November 2011, associated with fluid disposal in the Wilzetta field (supplementary text) (4). The enhanced catalog shows that the February 2010 Maule event triggered a strong sequence of earthquakes near the eventual epicenter of the first 5.0 M_wearthquake (Fig. 3 and table S2). The rate of earthquakes in the several days after the Maule trigger far exceeds that of any other time within the period of observation, up to the M_w ≥ 5.0 earthquakes themselves, which is similar to the observation at Snyder. There are no events detected within ±32 km relative distance for at least 4 months before the 2010 Maule earthquake. 

The largest event in the remotely triggered sequence is a 4.1 M_w, 16 hours after the 2010 Maule earthquake, which may account for the large number of earthquakes that continued up to the time of the first 5 M_w Prague earthquake in 2011 (Fig. 3). If the 4.1 M_w earthquake can be considered a foreshock of the subsequent 5.7 M_wPrague earthquake, then the 5.7 M_w event is not only one of the largest earthquakes to be associated with wastewater disposal (2) but also one of the largest earthquakes to be linked indirectly to a remote triggering event (4, 19).

The March 2011 Tohoku-oki earthquake, which occurred during the ongoing sequence before the 5.7 M_w Prague main shock, did not trigger additional earthquakes near the swarm (Fig. 3 and table S2). The 2012 Sumatra earthquake, on the other hand, followed the main 5.7 M_w Prague earthquake by 5 months and triggered a small uptick in activity that was consistent with the far northeastern tip of the swarm (Fig. 3C). However, this triggered rate change is much smaller than that triggered by the Maule earthquake in 2010.

Trinidad, Colorado, experienced a seismic swarm in August 2011 that included a 5.3 M_w main shock, possibly related to coal-bed methane extraction and reinjection of the produced water in the Raton Basin (supplementary text). The February 2010 Maule earthquake triggered a small but statistically significant response near the site of the 5.3 M_w main shock (Fig. 4 and table S2). Although the total number of triggered events is small (four), the binomial probability of observing this many events in 1 day after the trigger, given five events in the entire previous year, is less than 10^–5.

The March 2011 Tohoku-oki earthquake, which occurred during the active portion of the swarm, did not trigger additional seismicity at Trinidad. The 2012 Sumatra earthquake occurred 8 months after the 5.3 M_w Trinidad main shock and triggered a moderate surge in activity that was consistent with the far edge of the swarm, where previous swarm activity had not occurred (fig. S2). This pattern—strong triggering by the first remote earthquake, none by the second, and marginal triggering after the swarm—is very similar to that observed in Oklahoma.

We examined several other regions in the United States that have experienced moderate magnitude earthquakes or heightened seismicity rates linked to fluid injection, including Guy, Arkansas; Jones, Oklahoma; and Youngstown, Ohio. None of these other regions appear to have responded to remote triggering (supplementary text).

The strongly triggered regions were exceptional in that they had a long history of pumping within 10 km of the eventual swarms yet were relatively quiet for much of that history. At other sites of induced moderate earthquakes (Guy, Arkansas, and Youngstown, Ohio), the lag time between the start of pumping and onset of seismicity was as little as months or weeks, presenting a relatively small window of vulnerability to dynamic triggering before the swarms.

The delay in induced seismicity in some regions could be due to complexities in the local geology (supplementary text). In Oklahoma, injection occurred into a fault-bounded pocket, and pressures may have built up slowly over time because of the size of the reservoir bounded by impermeable faults (4). The Cogdell field may have similar isolated pockets, formed by discrete carbonate reefs buried within impermeable shales (8).

Fluids have been suggested as an important component in dynamic triggering since early observations showed preferential triggering in active volcanic and hydrothermal systems (13, 20, 21). Some features of our observations are also suggestive of a fluid mechanism for triggering. First, in all of the studied cases the triggered earthquakes occurred with a small delay with respect to the passage of the seismic waves, initiating within less than 24 hours and continuing for days to months afterward. This pattern suggests a triggering mechanism that relies on dynamic permeability enhancement and transport of fluids (22, 23), as has been suggested for natural triggered seismicity (20–22). In this scenario, stress transients alter the permeability of hydraulic conduits in the reservoir, accelerating diffusion of pore pressure into local faults. Fractures in active injection reservoirs may be particularly susceptible to this mechanism because the injection of unequilibrated fluids may lead to clogging through mineralization and sedimentation. A brief pressure transient may then flush out these clogged fractures (22, 24).

In Prague and Trinidad, only the first of two large remote events caused earthquakes, despite imparting dilational and shear strains that are similar to subsequent events (table S4). This is also consistent with the permeability enhancement model, which requires a certain amount of recharge time between triggering episodes (24). After local fault slip is triggered, the local permeability rises dramatically because of microfracturing and dilation (25), promoting further fluid diffusion over several rupture dimensions (26). Hence, once the seismic swarm is underway the fractures may not return to a state in which they are susceptible to unclogging by small transient stresses.

We find that certain areas of fluid injection are sensitive to small changes in stress associated with the passage of seismic waves from remote large earthquakes. The observations suggest several requirements for an induced region to be sensitive to remote triggering. First, all of the triggered sites in this study had a long history of regional subsurface injection over a period of decades. Second, each triggered site was near to hosting a moderate magnitude earthquake, suggesting critically stressed faults. Last, each site had relatively low levels of seismicity rate in the immediate vicinity (10 km) before the first triggering episode. Remote triggering can therefore indicate that conditions within an injection field have crossed some critical threshold, and a larger induced earthquake could be possible or even likely. This underlines the importance of improved seismic monitoring in areas of subsurface fluid injection.

References and Notes