Earthquakes, Dynamic Triggering of

Definition of the Subject

Geoscientists have long sought understanding of how earthquakes interact. Can earthquakes trigger other earthquakes? The answer is clearly yes overshort time and distance scales, as in the case of mainshock – aftershock sequences. Over increasing time and distance scales however, thisquestion becomes more difficult to answer. The study of dynamically triggered earthquakes explores the most distant boundaries over which earthquakestrigger other earthquakes.

Dynamic triggering e to temporary and oscillatory fluctuations in the stress/strain regime ina volume of the Earth's crust. Dynamic stress fluctuations are associated with ground shaking resulting from either anthropogenic activities ornatural sources. Dynamic triggering occurs as seismic waves from an initial earthquake propagate through the Earth's crust, triggering secondaryearthquakes. Once the seismic wave train has passed and ground shaking ends in a given locale, the crust returns to its previous stress statemodified by the combined stress drops associated with any locally triggered earthquakes.

Dynamic triggering includes wide ranging phenomenon, both geographically and in its characteristics. It has been observed across the globe in a variety of geologic and tectonic environments. It has been shown to occur at distances from the initialearthquake rupture varying from meters [21,27,52] to over 11,000 km [103]. In the most distant cases, earthquake triggering results from dynamic stress perturbations as low as 0.01MPa. Earthquakes have also been shown to trigger other earthquakes at a variety of time scales. In many cases triggering of earthquakes occurs duringor within minutes to hours following the responsible seismic waves (e. g. [26,39,74,103]). Inother cases, earthquakes occurring weeks to months after the initial earthquake have been interpreted as a delayed response to dynamic triggering(e. g. [41,89,102]). Delayed triggered responses may reflect a more complex series of physical processes. For example,dynamic waves may trigger an aseismic process such as fault creep or changes in a volcanic system, which subsequently triggers earthquakessecondarily [2,4,37,50].

This field has been an area of extensive research in the past twenty five years. It offers a potentially important key to improving ourunderstanding of earthquake nucleation in that, in principle, we can determine in-situ perturbations in the local stress field that lead to earthquakenucleation and rupture. In particular, the availability of broadband seismic data near sites of triggered seismicity allows us to calculate the timehistory of stress field fluctuations responsible for earthquake nucleation given adequate knowledge of the local seismic velocitystructure [28,38,103].

The study of dynamically triggered earthquakes can also help better characterize the physical condition of the Earth's crust at seismogenicdepths. Many researchers were surprised that earthquakes could be triggered by stress perturbations as small as 0.01 MPa. This observation indicates theEarth's crust is on the verge of failure in areas with triggered responses to distant earthquakes. This field of research may also provide clues to thehydrologic regime at depth. It has long been recognized that water tables change in response to earthquakes thousands of km distant [17]. The link between dynamically triggered earthquakes and dynamically triggered hydrological changes is an activearea of research [8,19,78].

Within the context of complexity and system science, [82] suggest that remotely triggeredseismicity may reflect large activation correlation lengths (ACL) in fault systems and stress fields that have reached a state ofself‐organized criticality. This statistical physics approach to earthquake occurrence focuses on the exploration of both analog and computationalmodels that can mimic observed dynamical space-time patterns spanning a wide range of spatial‐temporal scales. It is not concerned withinferred (or “non‐observable”) physical models for the local processes linking dynamic stresses and brittle failure (triggeredearthquakes) on faults (see [82]). In this review, however, we focus on these physical modelstogether with a description of documented patterns of remote dynamic triggering.

Introduction

Introduction to Stress Triggering of Earthquakes

Earthquake triggering refers to a process by which any change in fault properties or the processes acting on a fault leads to rupture initiation. More specifically, stress triggering occurs when a change in the stress field acting on a fault leads to rupture. Stress triggering of earthquakes can result from stresses applied over a variety of time scales and with a variety of frequencies, which generally fall into three partially overlapping categories, 1) static stress triggering, 2) quasi‐static stress triggering, and 3) dynamic stress triggering. In the case of static stress changes, the state of stress acting across a fault is permanently perturbed. This form of stress triggering is important in the near field of an earthquake where fault displacement significantly alters the stress field in the surrounding crust. Static stress triggering is commonly regarded as the dominant factor controlling aftershock generation (e. g. [54]). Because static stress changes decay rapidly with distance from the earthquake rupture (as ${ d^{-3} }$, where d is distance from the earthquake epicenter), they are generally thought to be significant only within two to three fault lengths of the earthquake rupture. The role of static stress changes in triggering aftershocks and other earthquakes has been a vigorous and productive area of research in the past two decades (for reviews see [34,53,92,93]). The relative importance of static vs. dynamic triggering of aftershocks in the near field, however, has recently become an actively debated topic (e. g. [21,71]).

Because static stress changes in the near field develop essentially simultaneously with earthquake rupture, simple static stress triggering must appeal to other mechanisms to explain the time delay associated with many aftershocks and subsequently triggered earthquakes. In contrast, quasi‐static stress triggering results from viscoelastic relaxation of the crust after an earthquake. Because quasi‐static stress changes decay as ${ d^{-2} }$ and because viscoelastic relation is a time dependent process, these stress changes may explain triggered earthquakes more distant from an initial earthquake and triggered earthquakes with delay times from years to decades [72].

Dynamic stress changes decay more slowly with distance than either static stress changes or quasi – static stress changes (as ${ d^{-1.5} }$ for surface waves ). Thus dynamic stress changes become increasingly dominant with increasing distance from the fault rupture. In this review we discuss dynamic triggering of earthquakes resulting from ground shaking due to the passing seismic wavetrain of other earthquakes. We focus on dynamic triggering due to remote earthquakes (greater than two fault lengths distance from the earthquake rupture), though we will briefly discuss the active research topic of dynamic triggering in an earthquake's aftershock zone as well. We also limit discussion to frequencies of ground shaking above ${ \sim 0.01\,\mathrm{Hz} }$ (periods less than ${ \sim 100\,\mathrm{s} }$), though some work has been done on earthquakes triggered by longer wavelength fluctuations. For example, [16,95] recently found evidence that solid Earth tides can modulate background seismicity rates. Other reviews of dynamic stress triggering can be found in [23,37,92] and [38].

Brief History of Dynamic Stress Triggering Research

The ability of earthquakes to trigger other earthquakes at great distances has been discussed in scientific literature throughout the latter half of the 20th century (see [38] for review). However, making a credible case for a causal link between two earthquakes remains a major challenge in this field. Beyond the realm of aftershock zones, it was difficult to justify statistically that one earthquake triggered another until the 1980s and 1990s. By then continuously recording telemetered seismic networks and automated processing of data became commonplace, providing reliable spatial and temporal records of earthquake occurrence at ${ \mathrm{M} \ge 1-2 }$ and the statistical leverage associated with large numbers of small earthquakes.

Dynamic triggering of earthquakes was widely accepted in the scientific community following the 1992 M 7.3 Landers earthquake in southern California. In the minutes to days following the Landers earthquake, earthquake rates increased dramatically across the western United States at distances well beyond the aftershock zone [39]. Earthquakes were triggered throughout California, Nevada, Utah, Wyoming, and Idaho at distances of up to 1250 km (Fig. 1). Although time delays of triggered events ranged from seconds to 33 hours after the arrival of the Landers earthquake wavetrain, the sudden increase in seismicity across the Western United States could not be ignored. This earthquake spawned a plethora of studies into the nature of earthquake triggering and remains one of the best studied triggering episodes to date.

Figure 1

Map showing sites of triggered seismicity in western North America from the Landers (green triangles), Hector Mine (blue circles), and Denali Fault (red crosses) earthquakes. Modifiedfrom [38],Treatise on Geophysics

The geophysical community had a unique research opportunity when the 1999 M 7.1 Hector Mine earthquake occurred. Because it was an earthquake with similar magnitude to the Landers earthquake occurring in a similar location, it provided leverage to tune ideas about dynamically triggered seismicity and the underlying physical processes. Although the Hector Mine earthquake triggered seismicity at some of the same locations as the Landers earthquake, the Hector Mine earthquake had a much more limited triggered response (Fig. 1) [30]. The difference is likely due in part to differences in seismic radiation patterns between the two earthquakes [30,43]. The Landers earthquake ruptured unilaterally to the north, whereas the Hector Mine earthquake ruptured bilaterally, primarily to the south.

Following the Landers and Hector Mine earthquakes, the search for dynamic triggering began in earnest. Researchers across the globe began scanning earthquake catalogs and waveform data searching for dynamic triggering in a wide variety of environments (see “Sect. Review of Dynamic Triggering Observations”). At the Geysers, CA alone [91] identified 7 episodes of dynamic triggering between 1988 and 1994, making this among the most frequently triggered locations known.

Like the Landers earthquake, the 2002 M 7.9 Denali Fault earthquake triggered a widespread response across western Canada and the United States (Fig. 1) [27,33,45,64,68,74]. The increase in the number of broadband and high‐dynamic range seismometers by 2002 allowed scientists to visually scan on-scale seismic data during the earthquake's wavetrain. Thus, many triggered events were detected which were absent from earthquake catalogs. As an example of how instrumentation improvements increase our ability to detect dynamic triggering, [45] point out that the triggered response of the Yellowstone caldera to the Denali Fault earthquake could not have been detected at the time of the Hector Mine earthquake only three years earlier.

Review of Dynamic Triggering Observations

Detection of Dynamic Triggering

Dynamic triggering has been observed in a variety of locations around the globe. Some suggest that dynamic triggering of earthquakes is a ubiquitous process in the Earth's crust (e. g. [26,41]). Others suggest that some areas are more likely to experience dynamic triggering of earthquakes than others (e. g. [39,64]). Observations of dynamic triggering are limited geographically due to uneven seismic network coverage and the effort applied to examining seismic data.

Figure 2

Plot of earthquake magnitude versus time in the area of the Wasatch Front, Utah, 30 days before and after the Denali Fault earthquake. Circles represent independent events. Crosses indicate secondary events determined by declustering the earthquake catalog. Figure reprinted from [68], BSSA

Figure 3

Seismicity in Utah 14 days before and after the Denali Fault earthquake. Diamonds in b show earthquakes occurring in the first 24 hours after the arrival of the wave train from the Denali Fault earthquake. Cross are locations of quaternary volcanic vents [3]. Figure reprinted from [68], BSSA

Following the 1992 M 7.3 Landers earthquake, dynamic triggering was recognized by the sudden increase in the number of earthquakes located through standard network processing across the western US in the days to weeks after the large earthquake. Searching earthquake catalogs for sudden increases in seismicity after a large earthquake is one common method of identifying dynamic triggering (Figs. 2 and 3) (e. g. [6,30,39,41,68]). Identifying triggered seismicity using earthquake catalogs simplifies interpretation of triggered response with respect to background seismicity rates, as earthquake catalogs often provide stable long-term records of earthquake occurrence at a consistent threshold.

Figure 4

Seismicity triggered at Mammoth Mt. b-c and within the Long Valley caldera, California, a following the Denali Fault earthquake. a Catalog from NCEDC showing two swarms following the Denalifault earthquake in the caldera's south moat. The two lower panels showdata from this very small earthquake swarm recorded on the broadbandUNR/USGS station OMM from rotated to transverse direction, showing Denali earthquake wavetrain at Long Valley. Major arrivalsare labeled. c Record fromb high pass‐filtered, showingsmall local earthquakes occurring during Denali wavetrain. Modifiedfrom [74],BSSA

A second commonly used method of detecting dynamic triggering involves visually scanning continuous seismic data shortly before and after a large earthquake to identify a sudden increase in earthquakes too small to be detected and located through standard network processing (Fig. 4). This latter method is effective at identifying triggered seismicity in sparsely instrumented areas, identifying very small triggered earthquakes, and identifying earthquakes that occurred during the wavetrain from the initial earthquake (e. g. [26,47,64,74,103]). This method of detecting triggering has become more common with increasing availability of continuously recorded high‐dynamic‐range seismic data.

Possible instances of dynamic triggering have also been proposed based on historical accounts [40,42,44,59]. Because these studies rely on felt reports, they are generally limited to moderate to large sized triggered earthquakes that are separated in time by days to months.

With any method of detecting dynamic triggering, it must be shown that one earthquake is likely causally linked to the dynamic waves radiating from a previous earthquake, rather than by coincidence. Earthquakes near each other in time are more likely to be related than earthquakes separated in time. Additionally, earthquakes unlikely to occur randomly (e. g. large events in seismically quiet areas) are more likely to be related than earthquakes occurring commonly as background seismicity (e. g. small earthquakes in a seismically active area). To calculate the probability that two earthquakes are related, one must first calculate the probability of each occurring randomly. This is usually done using patterns of earthquake occurrence based on local earthquake catalogs. The most commonly used statistical test to identify whether an increase in number of earthquakes is statistically significant is the Beta statistic  [58]. Pankow et al. [68] also employ a binomial distribution analysis to this end. These techniques have potential pitfalls however, as they rely on assumptions about earthquake distributions and compare snapshots of seismicity in time in regions where seismicity rates fluctuate regularly [58]. Objectively determining whether one earthquake is genetically related to another remains a challenge.

Because spatial‐temporal clusters of earthquakes are less common than isolated events, clusters of earthquakes temporally coincident with dynamic stresses are more easily identified as being triggered than isolated earthquakes. In the case of earthquake clusters, however, it may be difficult to discriminate between earthquakes directly triggered by dynamic stresses from a remote earthquake and secondary aftershocks to directly triggered events [9]. To address this, earthquake catalogs are frequently ‘declustered’ (e. g. [45,68]). This process involves decomposing a catalog into primary and secondary earthquakes based on statistical patterns of aftershock sequences (e. g. [76]). [68] and [9] show that in some cases triggered seismicity is modeled well as an aftershock sequence. In other cases, however, triggered seismicity swarms cannot be dismissed as secondary aftershock sequences (e. g. [74,103]). In such cases it is likely that most events in a swarm were triggered directly by the dynamic waves radiated from a distant earthquake or perhaps as a secondary response to some aseismic process (e. g. fluid flow, or local deformation associated with fault creep) triggered by the dynamic waves.

Table 1 Published occurrences of discrete remotely triggered earthquakes

Table 1 (continued)

Dynamic Triggering in Volcanic and Geothermal Regimes

Although dynamic triggering has been observed in a variety of environments, many of these observations are from areas with active volcanic and hydrothermal systems (Table 1) [6,73,74]. These areas typically have high background seismicity rates indicating that the crust habitually hovers near failure and thus is particularly susceptible to dynamic triggering. Furthermore, because these areas tend to be well instrumented, dynamically triggered earthquakes may be unusually easy to detect. Here we summarize observations of triggered seismicity in volcanic and hydrothermal areas.

The Geysers geothermal field in northern California is among the most frequently triggered sites known with 9 cases of dynamic triggering documented in the past 20 years [28,74,91]. Earthquakes that have caused triggering at the Geysers range in magnitude from 6.9 to 7.9 and in distance from 212 km to 3120 km. The Coso geothermal field in southern California has also experienced repeated episodes of dynamic triggering following the M 7.3 Landers, the M 7.1 Hector Mine, and the M 7.9 Denali Fault earthquakes (Fig. 1) [39,74]. Following the Hector Mine earthquake, dynamically triggered earthquakes and ground deformation were observed near a third geothermal field – Cerro Prieto, Baja California [24,30].

Yellowstone, Wyoming is a larger and more complicated system than the geothermal fields mentioned above, as it is a caldera system characterized by ongoing magmatic and tectonic activity, in addition to hydrothermal activity. Yellowstone had a triggered response to the M 7.3 Landers earthquake [39] and the M 7.9 Denali Fault earthquake [45,47]. The triggered response to the Denali Fault earthquake was particularly dramatic (Figs. 5 and 6). Seismicity increased immediately following the arrival of surface waves from the Denali Fault earthquake and remained unusually high for 30 days with magnitudes ranging from ${ < 0.0 }$ to M 3.2 [45]. The time scale of the triggered response was spatially variable (Figs. 5 and 6). Because the Denali Fault earthquake led to immediate triggering in the area of geysers and affected periodicity of geysers at Yellowstone, it is likely that changes in the hydrothermal regime induced a triggered response in some areas ([45,46]. In other areas however, the development of triggered earthquake sequences was delayed and similar to commonly observed tectonic activity [45].

Figure 5

Seismicity within one month of the Denali Fault earthquake at Yellowstone caldera, color coded with time: a earthquake locations, b cross sections along AA', c cross section along BB'. Specific areas labeled: HL, Hebgen Lake area; NB, Norris geyser basin; UB, Upper geyser basin; WT, West Thumb geyser basin; YL, northern end of Yellowstone Lake. Large normal faults are represented with thick black lines: RM, Red Mountain fault zone; GF, Gallatin fault; HF, Hebgen and Red Canyon faults. Inset shows location of Denali Fault earthquake and Yellowstone. Solid and dashed lines in inset show the great circle path +/−10 degrees along the strike of the Denali Fault earthquake. Figure reprinted from [45], BSSA

Figure 6

Plot of earthquake magnitude versus time of seismicity for selected areas in Yellowstone caldera. See Fig. 5 for locations ofthese areas. Dashed line DFE as the origin time of the Denali FaultEarthquake. Figure reprinted from [45], BSSA

Like the Yellowstone caldera, the Long Valley caldera experiences dynamic triggering with complex characteristics. The area responded to the Landers earthquake [39], the Hector Mine earthquake [50], and the Denali Fault earthquake [50,74] both seismically and geodetically, although each response varied in location and intensity (Fig. 7). The Landers earthquake produced the largest triggered response with 340 earthquakes in seven days up to M 3.4 throughout the south moat of the caldera [39]. The seismic response to the Hector Mine earthquake was comparatively short lived and limited to the region of Mammoth Mountain. After the Denali fault earthquake, the caldera area experienced two phases of triggered seismicity. A burst of ${ \sim 60 \mathrm{M} \le 0.8 }$ earthquakes occurred beneath Mammoth Mountain during and shortly after the arrival of the surface waves from the Denali Fault earthquake [74]. Twenty‐four hours later a larger swarm of earthquakes of M ${ \le 3.4 }$ occurred in the Long Valley caldera's south moat [74]. All three episodes of dynamic triggering in the Long Valley caldera were accompanied by deformation transients with geodetic moments an order of magnitude larger than the cumulative seismic moment of the triggered seismicity [36,50], though the time history and magnitude of each deformation response varied.

Figure 7

Map of triggered seismicity beneath Long Valley caldera and Mammoth Mountain, California, for the Landers (green), Hector Mine (blue), and Denali Fault (red) earthquakes. Gray dots show background seismicity from 1997–1998. The red circle centered on station OMM indicates area within which the earthquakes triggered by the Denali Fault earthquake must be located based on S-P phase arrival times. The single red dot was large enough to be located [74]. Modified from [38], Treatise on Geophysics

Iwo Jima, Japan , a volcanic island hosting a Holocene eruption, geothermal activity, and historic phreatic (steam) eruptions is a third complex caldera system which has experienced dynamic triggering of local earthquakes. [98] examined continuous waveform data of ${ 21 \mathrm{M} > 7 }$ earthquakes ${ < 3000\,\mathrm{km} }$ distance from Iwo Jima, and identified 4 cases of resulting increased local seismicity. In all cases earthquakes were triggered locally during surface wave arrivals and persisted for 6–15 minutes.

Dynamic triggering of earthquakes has been observed at shallow depths in volcanic edifices at a variety of locales (Table 1). In the Pacific Northwest, Mt. Rainier experienced 6–8 ${ \mathrm{M} < 0 }$ earthquakes during the wavetrain of the Denali Fault earthquake and ${ 8 \mathrm{M} \le 0.9 }$ earthquakes in the following days. In response to the Landers earthquake, Mt. Lassen in northern California hosted 14 earthquakes of ${ \mathrm{M} \le 2.8 }$. Volcanoes in the Katmai Volcanic Cluster, Alaska, have experienced triggered seismicity on atleast seven occasions since 1999 ([64,65,73], this chapter). The largest of these triggered responses included 17 earthquakes of ${ \mathrm{M} \le 2.3 }$. During the wavetrainof the 2006 M 8.7 Sumatra–Andaman Islands earthquake, Mt.Wrangell, Alaska had triggered 14 earthquakes [103]. With the exception of the Mt. Lassen response following the Landers earthquake and the delayed events at Mt. Rainier following the Denali Fault earthquake, these earthquakes triggered in volcanic edifaces were too small to be detected and located by automatic processing systems.

The Valley of Mexico is a large volcanically and geothermally active area located in the Trans Mexican Volcanic Belt. [89] searched for dynamically triggered earthquakes in the Valley of Mexico following 18 ${ \mathrm{M} \ge 7.0 }$ Mexican earthquakes between 1920 and 1998. In seven cases, they found evidence for dynamic triggering of earthquakes within 2 days of a large earthquake. In four additional cases, seismicity increased after a large earthquake, but was delayed by up to one month. Because this study used only one station however, the potentially triggered events can only be located to within some ill‐defined region surrounding the station.

The South Iceland Seismic Zone is a transform zone in a volcanically and geothermally active area. In 2000, a Mw = 6.5 earthquake in the South Iceland Seismic Zone triggered widespread seismicity, including three Mw ${ \sim 5.0 }$ earthquakes within 5 minutes of its occurrence. Coulomb failure stress calculations indicate that the two ${ \mathrm{M} > 5 }$ earthquakes located ${ \sim 100\,\mathrm{km} }$ to the west on the Reykjanes Peninsula are beyond the range where static stress changes are significant [2], and thus appear to have been dynamically triggered. Furthermore, one of these ${ \mathrm{M} > 5 }$ earthquakes had a geodetic moment significantly larger than it seismic moment suggesting that deformation associated with aseismic fault creep may have indirectly triggered many of the smaller earthquakes in the area [2].

Dynamic Triggering in Regimes with Limited Volcanic and Geothermal Activity

Extensional andTranstensional Environments

The majority of occurrences of triggered seismicity documented to date have been in extensional or transtensional tectonic regimes (Table 1). In the western United States dynamic triggering following the M 7.3 Landers earthquake occurred exclusively in transtensional tectonic regimes, many of which were also volcanically or geothermally active, to distances of up to 1250 km [1,39]. These locations included Little Skull Mountain, Nevada; western Nevada, White Mountains, California, Mono Basin California, Cedar City Utah, the Wastach Front in central Utah, Burney, California, and Cascade, Idaho. The onset of triggering ranged from during the passage of the Landers wavetrain to 33 hours after the Landers earthquake. The largest of these earthquakes was a ${ \mathrm{M} = 5.6 }$ earthquake triggered beneath Little Skull Mountain, Nevada. Otherwise, triggered earthquakes had ${ \mathrm{M} = < 3.0 }$. The most vigorous responses containing tens to hundreds of triggered earthquakes occurred near Cedar City Utah, Western Nevada, and Cascade Idaho.

The Mw = 7.1 Hector Mine earthquake also led to an impressive display of triggered earthquakes in exclusively extensional, transtensional, and geothermal environments in the western United States. Triggered earthquakes began during the passage of the wavetrain near the Salton Trough in Indio and at the southern end of the Salton Sea [30,43]. In general, the triggered response to Hector Mine was less extensive and energetic than that of the Landers earthquake [30].

Following the Denali Fault earthquake seismicity was triggered in several extensional and transtensional areas in the western United States. [47] detected a M 4.6 earthquake triggered during the Denali Fault earthquake wavetrain in Cascade, Idaho. Seismicity remained elevated for 25 days along a 500 km stretch of the Intermountain seismic belt in Utah, on the border of the Basin and Range province (Figs. 2 and 3) [68].

Through examining historical documents [59] identify several earthquakes in extensional/transtensional environments that may have been dynamically triggered by the Mw = 7.8 1906 San Francisco earthquake, including a M 3.5 and M 4.5 earthquake in western Nevada and a M 6.1 earthquake in the Brawley Seismic Zone near the Salton Sea in Southern California. These events are within 400–700 km from the fault rupture, thus beyond the aftershock zone of the San Francisco earthquake.

In the day following the arrival of surface waves from the Mw = 7.4 Izmit, Turkey earthquake, catalog seismicity rates throughout continental Greece, 400–1000 km from the epicenter, increased significantly [6]. Greece is an area of active extension and hosts significant hydrothermal activity. Although [6] did not address a possible correlation to hydrothermal activity systematically, at least some clusters of dynamically triggered seismicity occurred in areas with active hot springs.

A second report of dynamically triggered seismicity in Europe comes the Roer Valley, the Netherlands. This area is an actively extending northern branch of the Rhine Graben System. Following a Mw = 5.4 earthquake in 1992, [12] determine that a large cluster of aftershocks occurred at distance of 40 km from the mainshock. They conclude that these events are dynamically triggered because they are located beyond the zone where static stress changes are significant.

Transpressional and Compressional Environments

Although dynamic triggering is not commonly observed in compressional environments, several studies suggest it does occur. Less than three hours after a Ms = 7.3 earthquake in the Gulf of Aqaba 1995, an earthquake swarm began 500 km distant from the mainshock epicenter in a restraining bend of the Dead Sea transform fault on the Syria–Lebanon border [62]. The swarm consisted of 21 earthquakes of Md ≤ 3.7.

The central United States is a transpressional environment with low strain rates. Dynamic triggering in the central US has not yet been observed instrumentally. However, [40,44,66] suggest that dynamic triggering occurred during the 1886 Charleston, South Carolina earthquake and 1811–1812 New Madrid earthquakes based on examination of historical felt reports. Similarly, [42] describe historical evidence for dynamic triggering of a ${ \mathrm{M}\sim 7}$ earthquake following the 1905 Kangra earthquake in India.

The stress state in Taiwan is variable, but generally transpressional [104]. [102] searched for dynamically triggered seismicity in the Taiwan region following 12 regional M 6.5+ earthquakes occurring between 1973 and 1994. They identify 9–10 cases of increased seismicity following a large event, although the increase is small in all cases, with 1–7 M 4–4.5 earthquakes more in the 15 days following the large earthquake than in the 15 days before.

Dynamic Triggering in Subcrustal Environments

The occurrence of dynamically triggered earthquakes in subcrustal environments has been investigated in subduction zones in South America and Japan. [96] found that a M 7.6 earthquake at 598 km depth in the Tonga trench in 2002 was followed by M 5.9 and M 7.7 earthquakes at 647 and 664 km depth within 2 and 7 minutes of the initial earthquake, respectively. By investigating the rupture history and Coulomb stress change resulting from the initial event, they conclude that the secondary events were triggered dynamically. They highlight 4 additional earthquakes of ${ > 450\,\mathrm{km} }$ depth which have similar large aftershocks that may be dynamically triggered. During the surface waves of the M 8.1 Tokachi‐oki earthquake, [60] identified deep low frequency earthquakes triggered in the Nankai subduction zone through analyzing Hi-Net borehole seismic data and use of the Beta statistic. The triggering occurred during a slow slip event in a region of the subduction zone which was active with deep low frequency tremor.

Triggered Tremor

With the exception of triggered subduction zone seismicity described above, the majority of dynamically triggered earthquakes described in this review are typical brittle failure earthquakes. For example, [43] show that the earthquakes triggered by the Hector Mine earthquake near the Salton Sea had typical spectra and stress drops, consistent with standard brittle failure source mechanism. In the last few years however researchers have demonstrated that volcanic tremor and deep non‐volcanic tremor respond to dynamic waves from regional and teleseismic earthquakes as well as typical crustal earthquakes (Table 2) ([32,60,61,81]). These findings emphasize that dynamic triggering can occur in a wide variety of environments and affect multiple seismic processes in addition to brittle failure of crustal rock. They provide an intriguing new perspective on the triggering processes.

Table 2 Published occurrences of dynamically triggered tremor

Figure 8

Time series showing tremor triggered by Love waves from the Denali Fault Earthquake in the Cascadia subduction zone: a Tremor at station BPBC, time adjust to correct for travel time from source to seismometer, b–d Displacement seismograms for transverse, radial, and vertical components at station PCH, the closest 3 component broadband station to the tremor, time adjust to correct for travel time from source to seismometer. Tremor occurs when the Love wave displacement is to the SW. Figure reprinted from [81], Nature

At Aso volcano, Japan, [61] identify dynamically triggered earthquakes and volcanic tremor following the 1999 M 7.7 Chi-Chi earthquake. To test the uniqueness of these observations, they searched for triggered tremor at Aso following 20 other Mw ${ \ge 7 }$ earthquakes occurring within 3000 km distance between 1995 and 2002. Five of these earthquakes triggered tremor following P wave arrivals at Aso. All occurred between 1998 and 1999, a time with usually high heat supply to the volcano's crater. As yet, this is the only documented episode of dynamically triggered volcanic tremor.

On the other side of the Pacific Ocean and a different tectonic environment, [81] identified episodes of deep non‐volcanic tremor in the Cascadia subduction zone, Canada, which were triggered by the Love waves of the M 7.9 Denali Fault earthquake. In this case tremor amplitude modulates perfectly with strain amplitude from the incident Love waves (Fig. 8).

More recently, [32] identified triggered non‐volcanic tremor in seven locations in California following the Denali Fault earthquake. In all cases tremor amplitude modulates with strain amplitude from incident surface waves. Five of these are strike‐slip faulting regimes. These observations are the first reported cases of non‐volcanic tremor beyond subduction zones (e. g. [80]) and the San Andreas fault [67].

Lack of Triggering Observations

Interestingly, some areas of high ambient seismicity show a notable lack of dynamically triggered seismicity. For example, the San Andreas fault near Parkfield, California showed no triggered response to the Landers earthquake [90]. Japan boasts high rates of shallow background seismicity, frequent large earthquakes from the subduction zone, high seismic network density, and a variety of crustal stress environments and volcanic and geothermal regions. However, through examining both earthquake catalogs and waveform data from individual seismic stations before and after nine large remote events, [33], show that dynamic triggering in Japan is not common, as it is in extensional regimes of the Western United States.

Similarly, Alaska abounds with crustal and subduction zone seismicity and active volcanic and geothermal systems, although network density is far lower than in Japan. Though the Katmai Volcanic cluster appears to be particularly susceptible to triggering [64,73] and dynamic triggering has been observed at Mt. Wrangell [103], dynamic triggering is rare compared to the western United States. [64] suggest that this results from unknown differences in the magmatic and hydrothermal systems of the volcanoes. [83] document a decrease in seismicity at Mt. Wrangell and Veniaminof volcanoes following the M 7.9 Denali Fault earthquake. To date these are the only documented examples of seismicity repression from a large distant earthquake.

Characteristics of Dynamic Triggering

Environmental Controls on Dynamic Triggering

Extensional and transtensional tectonic regimes with high levels of background seismicity are highly susceptible to dynamic triggering [38]. This may reflect the ease with which fluids can migrate upwards in these stress environments [32,38]. Because such fluids are often hot with high concentrations of dissolved solids, rapid precipitation may form high pressure compartments over rapid time scales, further enhancing a tendency toward failure. Faults in extensional stress regimes are alsoinherently weak compared to those in compressional environments [43,88]. [26,41] suggest that dynamic triggering is a ubiquitous process in the crust which is detected more commonly in certain areas due to high instrumentation and scrutiny levels. Only one study to date has carefully addressed this question. By comparing seismicity rates on the San Andreas fault in California and the Western United States Basin and Range Province , [90] show that the San Andreas fault is less likely to experience dynamic triggering than similarly instrumented areas with similar levels of background seismicity in the Western United States Basin and Range province. More studies like [90] are necessary to resolve whether triggering is truly ubiquitous or favored in specific tectonic environments.

Triggering Thresholds and Recharge Times

In most reports of remote dynamic triggering, seismicity is triggered by earthquakes of M 6.5 or greater (Table 1). Dynamic triggering responses are strongest in areas that experience strong directivity  [26,30,39]. These first order observations suggest that strength of triggered response is a function of ground shaking amplitude. Although amplitude‐based triggering thresholds have been suggested for some areas [28,29,31,64], a consistent triggering threshold that applies throughout the crust has not been established [38]. Large earthquakes regularly occur without dynamically triggering seismicity beyond their aftershock zones.

Lack of triggering reports below M 6.5 may reflect subtle triggered responses. [41] uses the Beta statistic to give evidence of small seismicity increases at distances of 70–110 km in the month following 14 moderate (M 5.5–7) earthquakes in California. Because this distance corresponds with where a large SMS phase should arrive, [41] suggests that the SMS phase is responsible for the triggered response in these cases.

If a simple amplitude-of‐shaking threshold is required to dynamically trigger earthquakes, we would expect that even moderate earthquakes trigger seismicity near their epicenters. [21,27,52,71] give strong evidence that dynamic triggering occurs in the near field. Because it is difficult to distinguish the influence of static and dynamic stress changes in the near field, many studies of dynamic triggering have limited their investigation to the realm beyond the aftershock zone.

Because many aftershocks in the near field are likely dynamically triggered, [31] include aftershocks in a search for an amplitude‐based triggering threshold. They find that peak dynamic stress distributions correlate well with aftershock and remotely triggered seismicity distributions, except in the Long Valley caldera, CA. The result of [31] is consistent with failure thresholds found in laboratory studies [49] and independent of frequency of shaking. [64] also find evidence for a ground shaking amplitude‐based triggering threshold at the Katmai Volcanic Cluster, Alaska by comparing magnitude and distance of mainshock with triggered response (Fig. 9). Their magnitude‐distance relationship is similar to that proposed by [28] for the Geysers, CA. However, the triggering threshold at Katmai appears to be higher than that suggested for the Geysers.

Figure 9

Plot of magnitude vs. distance from Mageik volcano in the Katmai Volcanic Cluster for all ${ \mathrm{Mw} > 6 }$ earthquakes between 1996 and 2003 located within 2000 km of Katmai. Hollow squares triggered seismicity in the KVC. Solid circles did not. Dashed line represents possible triggering threshold. Figure reprinted from [64], BSSA

In other cases, a simple amplitude-of‐shaking threshold is not consistent with data, and large amplitude ground shaking is neither a necessary nor sufficient condition to cause dynamically triggered earthquakes. [31] show that the Long Valley caldera appears to be more susceptible to triggering than other areas they studied. Because their study was based on catalog seismicity, it did not include triggered earthquakes that were too small to appear in earthquake catalogs, such as those at the Coso geothermal field in response to the Denali Fault earthquake. These events were triggered by dynamic stresses of < 0.01 MPa [74] and, like Long Valley, would not fit the thresholds proposed by [29] and [31].

By comparing spectra of all earthquakes with high amplitude ground shaking in the Long Valley caldera [7] find that in this area, high‐amplitude low‐frequency shaking is more likely to trigger seismicity than high‐amplitude high‐frequency shaking. [1] come to the same conclusion after examining strong ground shaking spectra of earthquakes which did and did not trigger seismicity in the Western Great Basin. Longer wave lengths associated with low frequency ground shaking favor triggering by larger earthquakes in at least some locales. Whether remote dynamic triggering in both the near and far field results from the same physical process or processesremains an open question.

One parameter that may complicate the search for a triggering threshold in amplitude and/or frequency is recharge time. Because the occurrence of earthquakes releases stored strain energy, it may take time for an area to re‐accumulate strain energy sufficiently to be primed for failure again following local earthquake activity or previous episodes of remotely triggered seismicity [38]. However some areas, such as the Geysers geothermal field require little to no time to recharge, as triggered seismicity episodes have been separated by time intervals of months or less [28]. Recharge times are dependent on many parameters including earthquake history, regional tectonic strain rates, and mass and heat advection rates in areas of hydrothermal and volcanic activity.

Time Scales of Dynamic Triggering and Responsible Phases

Remote dynamic triggering of earthquakes occurs over a variety of time scales following the onset of dynamic stressing. At Aso Volcano, Japan [61] triggered tremor begins with the P-wave arrival from distant large earthquake. In the case of discrete earthquakes however, it is most common for triggering to begin during surface wave arrivals (Table 1), leading many to suggest that the specific low‐frequency large‐amplitude ground motions associated with surface waves initiate the failure process [1,7,38,103]. Although the onset of dynamic triggering at remote locations is most commonly observed during Rayleigh wave arrivals [38], clear cases of remote triggering of tremor on the Love wave exist as well [81].

In some cases, dynamic triggering begins hours to days after an initial stress perturbation (e. g. [39,89,102]), hinting that the physical process responsible for initiating earthquake failure evolve with time. For example the largest triggered event following the M 7.3 Landers earthquake, a M 5.6 at the Little Skull Mountain, Nevada, occurred 33 hours after the mainshock[39]. In the case of Long Valley caldera's south moat and Mt. Rainier after the Denali Fault earthquake, delayed earthquake swarms began 24 hours and 2 hours respectively after the passage of the dynamic waves from the mainshock (Fig. 4) [74]. Both of these areas also had much smaller triggered swarms during the mainshock's wavetrain.

Figure 10

Phase modulated dynamically triggered earthquakes in the Katmai Volcanic Cluster following the 2007 M 8.2 Kurile earthquake: a short period record from station ACH showing both wavetrains for the Kurile earthquake and the larger amplitude, locally triggered earthquakes, b broadband record from station KABU showing wave motion of the Kurile earthquake, c time series from ACH and KABU zoomed in to show how local earthquakes seen clearly in red are occurring on a specific phase of the wavetrain from the Kurile earthquake

Determining the duration and decay time of triggered swarms is more difficult than detecting their onsets, particularly in areas of high ambient seismicity. Many triggered earthquakes may be triggeredsecondarily as aftershocks to earlier triggered earthquakes [9]. The Yellowstone response to the Denali Fault earthquake and the Long Valley caldera response to Landers are fit well with an Omori -type law decay [45]. In some cases however, triggered swarms end abruptly after the dynamic stress perturbation stops (e. g. [103]). Although our understanding of decay rates of triggered swarms is incomplete emphasizing that the subject deserves further investigation, decay rates give strong constraints on physical processes responsible for triggering.

Phase Modulated Triggering

Recent findings show that earthquakes can be triggered during specific phases of the wavetrain. At Mt. Wrangell, Alaska, triggered earthquakes occurred preferentially during phases of the largest positive vertical ground displacement from the 2004 M 9.0 Sumatra earthquake [103]. Similarly at Katmai Volcanic Cluster, Alaska, triggered earthquakes occurred only during specific phases of Rayleigh waves from the 2007 M 8.2 Kurile earthquake (Fig. 10). Such observations will allow us to resolve the precise dynamic stress field perturbations at the moment of earthquake nucleation on specific failure planes (e. g. [38]).

Physical Models of Dynamic Triggering

The wide variations in the characteristics of dynamic triggering and the limited data for individual response instances admit a spectrum ofcompeting models for the physical processes linking dynamic stresses from a large, distant earthquake to the locally triggered response. Broadlyconsidered, published models fall into three partially overlapping categories: 1) those involving some form of stress‐driven brittle failure acrosslocal fractures, 2) those involving the activation of hydrous or magmatic fluids, and 3) those involving some form of localized aseismic relaxation(deformation). The brittle failure models are generally consistent with the onset of locally triggered seismicity during dynamic stressing (rapid-onsettriggering), including the possibility that seismicity may persist as aftershocks for some time after the dynamic stressing has stopped [21]. Under the latter two categories, the onset of local seismicity represents a second‐order phenomenondriven by a first order response to dynamic stressing in the form of fluid activation or transient deformation. In principle, models under these twocategories admit a significant delay in the onset of the triggered seismicity with respect to the dynamic stresses generated by a distantearthquake. Because the dynamic stress amplitudes that trigger a response at remote distances are typically an order of magnitude or more belowbackground tectonic stress levels, all models carry the implicit assumption that a crustal volume susceptible to dynamic stress triggering must be ina near‐critical stress state prior to a triggered response.

Brittle Failure

Brittle failure models are based on the premise that the dynamic stresses propagating with the seismic waves from a distant earthquake are sufficient to nudge the local stress acting on a pre‐existing dislocation beyond the threshold for the particular failure mode. This threshold may be the Griffith criteria for the tensile strength of a partially healed crack or the Coulomb criteria for frictional strength of a fault [86,87]. Crustal fluids play an important passive role in all brittle failure models by counteracting the rock matrix stress acting on a fracture through pore pressure, p, according to

$$ \boldsymbol{\sigma}^{\prime} = \boldsymbol{\sigma} - \boldsymbol{I} p $$

where ${ \boldsymbol{\sigma}^{\prime} }$ and ${ \boldsymbol{\sigma} }$ are the effective and rock matrix stress tensors, respectively, and I is the identity tensor. Thus, pore pressure reduces the effective normal stress, ${ \sigma_{\mathrm{n}}^{\prime} }$, acting on a fracture by opposing the rock matrix normal stress as ${ \sigma_{\mathrm{n}}^{\prime} = \sigma_{\mathrm{n}} - p }$. Alternatively, for pressure‐sensitive friction models the role of pore pressure can be expressed in terms of an effective coefficient of friction as ${ \mu^{\prime} = \mu(1 - \lambda_{p}) }$, where ${ \lambda_{p}=p/\sigma_{\mathrm{n}} }$. Elevated pore pressures lower the effectively strength by moving the background stress state closer to extensional or shear failure thresholds thereby increasing vulnerability for failure by imposition of small dynamic stress perturbations.

In the simplest frictional failure model, a triggered earthquake occurs when the stress acting on a fault exceeds the Coulomb threshold for static friction, or CFF(t) = 0, and friction abruptly drops from static to dynamic values with ${ \mu_{\mathrm{s}} > \mu_{\mathrm{d}} }$, respectively. Here, CFF(t) is the Coulomb Failure Function defined as

$$ \begin{aligned} \mathrm{CFF(t)} &= |\tau(t) | -\mu _{\mathrm{s}}\sigma_{\mathrm{n}} ^{\prime}(t) - \mathrm{C}\,,\quad\text{or its equivalent} \\ &= |\tau(t)| -\mu _{\mathrm{s}}^{\prime}\sigma_{\mathrm{n}}(t) - \mathrm{C} \end{aligned} $$

where ${ \sigma_{\mathrm{n}}\,,\sigma_{\mathrm{n}}^{\prime}\,,\mu_{s}\,,\mu_{s}^{\prime} }$ are defined in the preceding paragraph, τ is the shear, and C is the cohesive strength ([34], and references therein). This simple case implies rapid-onset triggering with the triggered seismicity beginning promptly when CFF(t) first becomes positive for a fault optimally oriented for failure in the background stress field. The combination of dynamic stress components ${ \Delta\tau }$ and ${ \Delta\sigma_{\mathrm{n}} }$ for which CFF ${ > 0 }$ will depend on the wave type (e. g. Love or Rayleigh wave ) and its incidence angle on the optimally oriented fault [35]. Although details vary, Love waves will generally have a greater triggering potential than Rayleigh waves when incident on vertical, strike‐slip faults while the opposite is the case for incidence on inclined, dip-slip faults.

The Coulomb failure criterion applies to more elaborate non‐linear friction models as well (see [18,25,31,70,100]). Because the behavior of non‐linear models depend on factors such as slip history and slip rate, however, the failure threshold for static friction may vary with time, and the triggered earthquake may be delayed with respect to the time the failure criterion was first exceeded (e. g. [69]). Susceptibility to dynamic triggering may result when a dynamic stress is imposed on quasi‐static loading under a conditionally stable regime (e. g. [84]). Based on their analysis of the dynamic triggering observed at Long Valley caldera, [7] conclude that this mechanism requires near‐lithostatic pore pressures to be effective.

Models based on the non‐linear response of granular media to dynamic stresses may apply to dynamic triggering of mature faults with a well‐developed core of fault gouge. [49] document an abrupt decrease in the modulus of fault gouge under low effective normal stress (${ \sigma_{\mathrm{n}}^{\prime} }$ 0.1 MPa) when excited by dynamic strains ${ >\!10^{-6} }$ in the laboratory. Thus, this model also requires near‐lithostatic pore pressures to be effective.

Sub‐critical crack growth, or stress corrosion , is another non‐linear form of brittle failure that has a potential role in dynamic triggering. Under this model, a sudden increase in differential stress or an oscillatory stress applied to a pre‐existing crack can lead to crack growth due to weakening of the crack tip by chemical corrosion. This can shorten the time to earthquake rupture. This process will be enhanced in an environment with fluids at elevated temperatures. It turns out that the equations governing sub‐critical crack growth have the same mathematical form as rate-state friction equations above [51]. Thus near‐lithostatic pore pressure appears to be a requirement for each of these non‐linear brittle‐failure models, at least as they apply to dynamic triggering at remote distances.

Fluid Activation Models

In addition to their passive role in reducing the effective strength of a rock volume through ambient pore pressure , fluids may play an active role in the dynamic triggering process. Fluid activation models generally appeal to either 1) pore‐pressure re‐distribution associated with changes in permeability and fluid transport, or 2) state changes induced in multi-phase fluids.

Dynamic stressing may be capable of physically disrupting permeability barriers separating volumes of differing pore pressure. For example, dynamic stress may shake accumulated detritus from clogged fractures or opening partially healed fractures by extensional failure. In either case, fluid diffusion down the pressure gradient will result in a re‐distribution of pore pressure with the potential for triggering seismicity in previously under‐pressured volumes in a near‐critically stressed state. The evolution of triggered seismicity in this case will be governed by the diffusion length for a given permeability and the proximity of the pre‐existing stress state to brittle failure. [8] proposed the clogged fracture model as an explanation for the hydrologic response of water wells in southern Oregon to surface waves from ${ \mathrm{M} > 7 }$ earthquakes at distances of 300 km and 3850 km.

Geothermal areas may be particularly susceptible to dynamic triggering through pore pressure re‐distribution. In these areas fractures are rapidly sealed by precipitation from circulating, solute‐rich geothermal fluids and plastic deformation of quartz‐rich rocks under elevated temperatures tend to isolate pockets of elevated pore pressure. Most active geothermal systems are located in areas of extensional tectonism. In these areas normal stresses induced by Rayleigh waves on vertical planes may open vertical fractures, allowing high-pore‐pressure fluids access to shallower crustal volumes with lower pore pressure [35]. The hydraulic surge model described by [22] for volcanic and geothermal systems is a version of this process in which the brittle‐plastic transition at the base of the seismogenic crust serves as a low‐permeability barrier separating near‐lithostatic pore pressures in the plastic regime from a hydrostatic regime in the overlying seismogenic crust. Rupturing the permeability seal by dynamic stresses would release near‐lithostatic pore pressures into the brittle, seismogenic crust thereby inducing a surge in triggered seismicity.

Models for bubble excitation by dynamic stresses in a two-phase fluid (multi-phase in a partially crystallized magma ) offers interesting possibilities for remotely triggered responses in geothermal and volcanic systems. This is a particularly intriguing concept for remote triggering in volcanic systems because of the importance of bubbles in eruption dynamics [57] and the source mechanisms of long‐period volcanic earthquakes [13], Volcanoes, Non-linear Processes in. Advective overpressure and rectified diffusion were the first bubble models proposed as explanations for remotely triggered seismicity [10,55,94]; although subsequent work has shown that both hold less promiseas viable explanations than initially thought [56].

Under the advective overpressure model, the pressure in a gas‐saturated, incompressible fluid confined in a rigid container increases as ${ \rho g \Delta h }$ as a pre‐existing bubble adhering to the wall of the container is shaken loose by passing seismic waves. The bubble ascends buoyantly a distance ${ \Delta h }$ through a fluid of density ρ where g is the acceleration of gravity [55]. The resulting pressure increase in the container (magma body) deforms the surrounding rock inducing small earthquakes. This model is criticized on the basis that assumptions of a ridged container and an incompressible fluid seriously violate realistic conditions in the earth [75].

Under rectified diffusion, pressure oscillations imposed on a gas‐saturated fluid with pre‐existing bubbles pump gas into the bubbles over multiple cycles. Gas exolves from the fluid into the bubble during the dilatational phase, when bubble surface area is maximal, and out of the bubble back into solution during compressional phase, when the bubble surface area is minimal [94]. The implied pressure gain integrated over multiple cycles is then transmitted to the surrounding rock inducing small earthquakes. [48] point out, however, that the effectiveness of this model is limited by reasonable gas diffusion rates in hydrous fluids or magma with respect to the frequencies of seismic waves driving the pressure oscillations.

More promising bubble models appeal to the strong sensitivity of bubble nucleation rate to the supersaturation pressure [56] and the results of numerical models by Volcanoes, Non-linear Processes in and [14,85], indicating that a small pressure drop imposed on a densely packed matrix of tiny bubbles can lead to rapid, diffusion‐driven bubble growth. The implications of these models, however, have yet to be more fully explored in the context of dynamic triggering.

Two more speculative models involve magma instabilities triggered by dynamic stresses. In one, a loosely held crystal mush accumulated on the walls of a crystallizing magma body may be dislodged by dynamic shaking. The sinking crystal mush would induce a convective plume as it displaced hotter, less dense magma. In the case of volatile‐rich magma, buoyant convection would be enhanced by bubble nucleation and growth as confining pressure drops with decreasing depth [37]. Under suitable conditions, the resulting pressure increase within the magma body could evolve over days [56]. If the magma chamber was already in a near critical state, the culmination could be magma intrusion into the overlying crust or the onset of an eruption. Whether this process culminates in a simple pressure increase, an intrusion, or an eruption, the sensible onset of locally triggered seismicity and deformation might be delayed by hours to perhaps days with respect to the passing seismic waves from the distant earthquake. A second, even more speculative model appeals to dynamic stresses disrupting the solid matrix of a partially crystallized magma body thereby releasing any differential tectonic stress sustained by the solid matrix [36,37]. As the magma body relaxes with a time constant governed by the effective viscosity of the disrupted crystal mush, stress would be transferred to the surround crust inducing deformation and local seismicity. In essence, this model corresponds to the relaxation of an Eshelby inclusion in an elastic medium [20].

Aseismic Deformation

The relaxing magma body of the previous paragraph is one example of aseismic deformation with the potential of triggering local deformation and the onset of secondary seismicity. A less speculative example involves aseismic creep on faults triggered by dynamic stressing. Deformation associated with fault creep transfers stress to the adjacent crust, which in turn triggers local seismicity, as in the example involving seismicity triggered on the Reykjanes Peninsula following the M = 6.5 earthquake in the South Iceland Seismic Zone in 2000 [2]. [4] document aseismic fault slip (creep) on faults in the Salton Trough of southern California triggered by the three ${ \mathrm{M}\,> 6}$ earthquakes in the Landers, California sequence of 1992 (the M = 6.1 Joshua Tree, M = 7.3 Landers, and the M = 6.2 Big Bear earthquakes). In this case, all instances of triggered slip were on faults with within 150 km of the ${ \mathrm{M} > 6 }$ earthquakes. In these examples and observations from triggering in Long Valley caldera and Cierra Prieto geothermal field in Baja California [38,50] the geodetic moment for triggered aseismic deformation exceeds the cumulative seismic moment for the triggered earthquakes by a factor of two or more. This emphasizes the importance of high‐resolution deformation monitoring in areas susceptible to dynamic triggering for resolving the role of aseismic deformation in the dynamic triggering process.

Future Directions

In the last 25 years, in the wake of the Landers earthquake, the study of dynamically triggered seismicity has given us new insight into earthquakeinitiation and the failure regime in the Earth's crust. Some argue that the state of stress in the crust is highly spatially variable [77]. Given this, the likelihood of triggering seismicity would also be spatially variable. [97] and [105], however, argue that the Earth's crust iscritically stressed and on the verge of failure nearly everywhere. If this were the case, one might expect triggering due to small dynamic stressperturbations to be a ubiquitous phenomenon. In either case, the study of remotely triggered seismicity provides clues to spatial distribution ofcritically stressed crustal volumes.

Unfortunately, although we can measure stress field perturbations from dynamic waves from earthquakes, we rarely have a detailed understandingof the background stress field these perturbations are modulating. In addition, dynamically triggered earthquakes are often too small or occur in toosparsely instrumented areas to resolve reliable focal mechanisms. Because of these limitations, our understanding of how dynamic stresses from remoteearthquake wavetrains induce a given crustal volume to respond with triggered seismicity remains incomplete. Advances will require more cases ofdynamically triggered seismicity captured by both spatially dense seismic networks and continuous, high‐resolution deformation monitoringnetworks.

Recent observations of phase modulated dynamic triggering offer powerful datasets of the precise time history of dynamic stress triggering. Becausesimilarities exist between phase modulated dynamic triggering of seismicity in the shallow crust of a volcano's edifice [103] and deep in a subduction zone [81], the emergingstudy of non‐volcanic tremor may provide new leverage on understanding how dynamic stresses influence seismic slip.

As new observations of dynamically triggered seismicity are reported, one conclusion is becoming increasingly evident: multiple causative processesexist. The wide variety in time scales over which triggering occurs and the spatial and temporal characteristics of triggered seismicity sequences andassociated deformation responses cannot be fit with any one model yet proposed. Rather, different models are consistent with different episodes oftriggering. For example, fluid activation and stress corrosion models are most applicable in volcanically and geothermally active environments. In somecases, such as the complex triggered response of Yellowstone, Mt. Rainier, and the Long Valley caldera areas to the Denali Fault earthquake, multipleprocesses may be occurring simultaneously in the same locale, yet on different time scales. Of the physical models described above, seismicitytriggered instantaneously or within seconds of the dynamic stress perturbation is consistent with models based on simple brittle failure, brittle failurewith nonlinear friction effects, stress corrosion, unclogging of fractures, or rectified diffusion, whereas triggered seismicity delayed by hours to daysis more consistent with models involving aseismic deformation, advective overpressure, sinking crystal plumes, or a relaxing magma body.

In the few cases where hydrologic and high‐sample rate strain data are available, dynamically triggered seismicity is accompanied by changesin water levels in wells [79] and significant deformation signals [2,25,36,50]. A complete understanding of dynamic triggering willrequire research approaches that integrate seismic, deformation, and hydrologic datastreams. To this end, we challenge Earth scientists to broaden theirthinking and tap these observations to better understand theinitiation of earthquake failure.