Permanently enhanced dynamic triggering probabilities as evidenced by two M ≥ 7.5 earthquakes



[1] The 2012 M7.7 Haida Gwaii earthquake radiated waves that likely dynamically triggered the 2013 M7.5 Craig earthquake, setting two precedents. First, the triggered earthquake is the largest dynamically triggered shear failure event documented to date. Second, the events highlight a connection between geologic structure, sedimentary troughs that act as waveguides, and triggering probability. The Haida Gwaii earthquake excited extraordinarily large waves within and beyond the Queen Charlotte Trough, which propagated well into mainland Alaska and likely triggering the Craig earthquake along the way. Previously, focusing and associated dynamic triggering have been attributed to unpredictable source effects. This case suggests that elevated dynamic triggering probabilities may exist along the many structures where sedimentary troughs overlie major faults, such as subduction zones’ accretionary prisms and transform faults’ axial valleys. Although data are sparse, I find no evidence of accelerating seismic activity in the vicinity of the Craig rupture between it and the Haida Gwaii earthquake.

1 Introduction

[2] On 28 October 2012, the M7.7 Haida Gwaii, British Columbia, earthquake broke the southern end of the Queen Charlotte plate-boundary fault system. Less than 3 months later on 05 January 2013, the M7.5 Craig, Alaska, earthquake broke the same fault system ~340 km to the northwest (Figure 1). In its public statements about the Craig earthquake, the U.S. Geological Survey noted its likely casual connection with the Haida Gwaii earthquake 3 months earlier [], based on the improbability of two large earthquakes with recurrence intervals of hundreds of years or more on the same fault system occurring within the same 3 month interval by chance (the Poisson probability is ~ 0.0004% assuming a 100 year recurrence interval).

Figure 1.

Study area map. The Haida Gwaii and Craig rupture planes (red rectangles) and epicenters (red solid circles) and epicenters of cataloged earthquakes during the interval between the two events (yellow open circles) are superposed on shaded bathymetry (contours every 100 m). The northwest trending zone of steep topographic gradients runs along the Queen Charlotte Fault system. Outlines of the approximate rupture planes of the largest preceding earthquakes, the 1972 M7.6 and 1949 M8.1 are shown as dashed red ovals. The great circle paths along the strike and at ±10° of the Haida Gwaii rupture plane (thick solid and thinner dashed black curves, respectively), and the epicenter of a M6.4 aftershock (red open circle, Figure S3) just outside the Queen Charlotte Terrace (see Figure 2) also are noted.

[3] Are these two events plausibly physically linked? The 2012 M7.7 Haida Gwaii earthquake was noteworthy in many aspects [Szeliga, 2013], but most remarkable was that the seismic waves it radiated probably dynamically triggered the 2013 M7.5 Craig earthquake. The 2012 Haida Gwaii earthquake radiated extraordinarily large amplitude, long-duration waves directed right at the site of the 2013 rupture. These waves had characteristics capable of dynamic triggering, as suggested by previous dynamic triggering studies (see sections 3 and 4), in which large amplitudes resulted from focusing due to source effects [Freed, 2005]. However, the amplification and long duration of the Haida Gwaii earthquake waves resulted instead from trapping of energy within a sedimentary trough. Unlike source characteristics that vary unpredictably from event to event, the permanence of structural features suggests enhanced triggering probabilities in this and likely the numerous other settings where significant sedimentary troughs form as a result of processes that also lead to the largest and most active earthquake zones—the accretionary prisms of most subduction zones [Ruff, 1989] and axial valleys along almost all transform boundaries [Basile and Allemand, 2002]. Although the seismic wave amplification, extended duration, and implied enhanced shaking hazard associated with these structures have been documented previously [Toshinawa and Ohmachi, 1992; Shapiro et al., 2000; Kennett and Furumara, 2002; Hartzell et al., 2010], this study highlights new implications for enhanced triggering potential.

[4] If the 2013 Craig earthquake was dynamically triggered, its magnitude also was remarkable, exceeding all previously documented cases of triggering via dynamic forcing by over half a magnitude unit [Pollitz et al., 2012], or equivalently a factor of 5 in energy released [Vassiliou and Kanamori, 1982]. (A M7.7 Tonga earthquake in 2002 was likely dynamically triggered by a M7.6 event 7 min earlier and 300 km away, but these occurred at depths of 664 and 598 km, respectively, and the failure and triggering processes likely differed from those controlling earthquakes at shallow depths [Tibi et al., 2003].)

2 Static Triggering

[5] The Queen Charlotte fault system accommodates relative plate motion between the Pacific and North American plates, which is oblique adjacent to the Haida Gwaii rupture (a thrust faulting event) with 15–20 mm/yr of convergence, and is parallel to the plate boundary adjacent to the Craig rupture (a strike-slip earthquake) at a rate of ~51 mm/yr [Rohr et al., 2000; Kreemer et al., 2003; DeMets et al., 2010] (Figure 1). The distance of ~340 km between the two earthquakes and the likely stress state of the Craig rupture plane in 2012 argue strongly against static stress changes from the 2012 Haida Gwaii earthquake triggering the 2013 Craig event. Although of the correct signs to encourage static Coulomb failure, estimated shear and normal static strains changes resulting from the Haida Gwaii event and resolved onto the Craig rupture plane are of the order of 0.01 μstrain (~0.4 kPa) (Figure S1 in the supporting information), orders of magnitude smaller than values estimated in numerous other static stress triggering studies [Freed, 2005]. Moreover, most models and cases of triggering attributed to static Coulomb stress changes require or infer near-failure stress/strain levels along the triggered fault just prior to triggering; the segment of fault that hosted the Craig earthquake was not likely close to its failure threshold, suggesting involvement of some sort of fault weakening mechanism. Only 2 to 3 m of slip would have accrued since the penultimate earthquakes that ruptured the 2013 Craig earthquake fault segment (possibly the southern end of the M7.6 1972 earthquake and the M8.1 1949 earthquake, respectively; Figure 1), much less than the maximum slip of ~8 m estimated during the Craig rupture.

3 The Haida Gwaii Wavefield and Dynamic Triggering

[6] While the static deformation field does not seem to be a plausible triggering candidate, the dynamic loading by the radiated wavefield is. Significant focusing of 2012 Haida Gwaii's 15–20 s period surface waves in the direction of the 2013 Craig epicenter can clearly be seen in both the Rayleigh and Love wave trains (Figure 2), thanks to the fortuitously optimal geometry and location of the Haida Gwaii rupture plane relative to the distribution of seismic stations from Alaska to southern California (Figure 3). Despite large instrumental gaps in azimuths and distances sampled, the concentrations of stations within about ±10° of both the forward (northwest) and backward (southeast) strike direction of the Haida Gwaii rupture permit identification of focusing and its likely cause (see below). The closest stations situated appropriately to observe focusing, at the same azimuth as the 2013 Craig earthquake, are almost 900 km from the 2012 Haida Gwaii epicenter. Still, waves in the direction of the Craig earthquake have amplitudes equivalent to a few μstrain (strains may be approximated by the ratio of the wave and phase velocities [Gomberg and Agnew, 1996], equal to >10−5 km/s and 3.5 km/s, respectively; see Figure 2), ring for several hundred seconds, and are 500% larger than waves at the same distance but in the opposite direction (Figure 3). The Haida Gwaii waves are of similar amplitude and significantly longer duration than the triggering waves from the 1992 M7.2 Landers, California, and 2002 M7.9 Denali, Alaska, earthquakes, which unquestionably dynamically triggered seismicity rate increases (the largest was a M5.4 earthquake) at distances of hundreds to thousands of kilometers [Hill et al., 1993; Gomberg et al., 2004], also within a narrow azimuthal range along the strike of their rupture planes.

Figure 2.

Haida Gwaii earthquake waveforms. Tangential component, instrument corrected, velocity seismograms recorded in the distance range of 8°–12° at the stations labeled in Figure 3. The source-station azimuths relative to the strike of the Haida Gwaii fault plane (323° from N) in the forward (±90° from 323°) and backward (±90° from 143°) strike directions are shown as solid and dashed lines, and colors get lighter as the deviation from the forward or backward strike increases (e.g., black for stations at 323°and 143°). Thus, dashed and solid seismograms of the same color at the same distance are in opposite directions and should have equal amplitudes. The expanded view is meant to make the tremendous amplification of seismograms in the forward strike direction more easily seen; locations of labeled stations are in red in Figure 3.

Figure 3.

Haida Gwaii wavefield peak amplitudes. Locations of stations that recorded the Haida Gwaii main shock (triangles) with symbol sizes proportional to the peak tangential component velocities at each station. Color-coding and red-labeled stations are the same as in Figure 2, and rupture planes of Haida Gwaii and Craig earthquakes indicated as red rectangles. Blue labels are near stations DIB, CRAG, and RUBB, where continuous data were scanned for interlude seismicity; main shock data on the Queen Charlotte Islands were clipped so no triangle is shown for station DIB. The largest amplitudes lie within about ±10° of the forward strike direction of the Queen Charlotte Fault system (thick solid and thinner dashed black great circle paths through the Haida Gwaii epicenter at the azimuth of its rupture plane), relative to amplitudes at stations in the opposite direction; e.g., compare amplitudes at labeled stations, located at ~180° from one another at the same source-receive distances (such as 1.0 cm/s observed at SSP in Alaska and 0.16 cm/s at GO3D in Oregon, both noted).

[7] The along-strike focusing and the extraordinarily large triggering waves radiated by the Landers and Denali earthquakes were due to directivity, a consequence of predominantly unilateral rupture that causes a pileup of energy in the rupture propagation direction, similar to a Doppler effect [Haskell, 1964]. This focusing leads to an asymmetry in the seismic radiation pattern. Unlike the Landers and Denali earthquakes however, all evidence for the Haida Gwaii earthquake (spatial distribution of aftershocks and more distant seismicity increases, teleseismic data, slip models, and source time functions) indicates that if at all, any rupture directionality was to the southeast, in the opposite direction of the Craig earthquake (G. Hayes, personal communication, 2013) (Figure S2).

[8] Spatially varying material properties along different propagation paths may also lead to significant wavefield variability [Bard and Bouchon, 1980a, 1980b; Bostock and Kennett, 1990; Toshinawa and Ohmachi, 1992; Shapiro et al., 2000; Kennett and Furumara, 2002; Hartzell et al., 2010]. Because such properties are permanent features, any effects should impact the wavefields of multiple earthquakes sampling the same structures similarly. I suggest that asymmetry in the Haida Gwaii wavefield is a result of the earthquake's large magnitude and its location relative to particular geologic structures. Specifically, the Haida Gwaii rupture likely propagated into the sediments of the Queen Charlotte Terrace, which acted as a waveguide for 12–18 s period waves. The Queen Charlotte Terrace is an ~6 km thick accretionary prism of compressed sediments overlying the Queen Charlotte Fault along the west side of the Queen Charlotte Islands [Horn et al., 1984; Rohr et al., 2000; Smith et al., 2003; Bustin et al., 2007], extending from close to the southern end of the Craig rupture to the southern end of the Queen Charlotte Islands (Figure 1). The structure is formed by oblique under-thrusting [Rohr et al., 2000] and is marked by pronounced low gravity and low seismic velocities [Bustin et al., 2007]. Figure 4 illustrates the resonance affect of a highly simplified 1-D structure having thick sediments beneath the water, both of which are required to generate the apparent amplification seen in the Rayleigh and Love waves [Brune, 1969]. The 2-D and 3-D models that more accurately account for the shapes of low-velocity sedimentary trough-shaped structures predict even greater degrees of focusing, amplification, and prolonged durations than implied in Figure 4 [Bard and Bouchon, 1980a, 1980b; Bostock and Kennett, 1990; Toshinawa and Ohmachi, 1992]. Surface and guided waves may be thought of as resulting from special constructive interference, but once established, clearly the amplified waves may travel long distances regardless of the structure they pass through (e.g., well into mainland Alaska, Figures 2 and 3).

Figure 4.

Surface wave dispersion curves and wave excitation functions. (a) Group velocity curves and source excitation functions computed for three different plane-layered structures [Gomberg and Masters, 1988] show that the observed relative amplification and long durations (steep group velocity curves) of both Rayleigh and Love waves at periods of ~14 s requires both water and low-velocity sediment layers (3 km and 1.5 thick, respectively, in these calculations). The excitation functions determine the relative weighting of Rayleigh and Love wave energy as it varies with source depth, with the contributions of each of the two Love and three Rayleigh functions depending on the focal mechanism. The effective resonance, or large excitations at ~14 s, disappears at shorter and longer periods, as illustrated by the 20 s period functions calculated for the model with water and sediments atop the crust. Only a shallow source or shallow rupture of an extended source will excite this resonance.

[9] Albeit speculative, I suggest that the profound focusing and prolonged duration evident in the Haida Gwaii wavefield only occurred in the northwesterly direction because the waveguide associated with Queen Charlotte Terrace and Fault terminates at the southern end of the Haida Gwaii rupture, as evidenced in the bathymetry (Figure 1) and subsurface structure [Horn et al., 1984; Rohr et al., 2000; Smith et al., 2003; Bustin et al., 2007]; such structural change likely disrupts the development of southward propagating guided waves [Kennett et al., 1990; Kennett and Furumara, 2002]. The focusing and prolonged duration are not observed for the few largest aftershocks that were clearly recorded and radiated waves traveling along almost identical paths (Figure S3), likely absent because their locations, depths, and smaller rupture dimensions imply that they did not rupture into the Queen Charlotte Terrace (Figure 1). Shapiro et al. [2000] made similar observations of amplification and extended durations and inferred that they were due to excitation and propagation of waves within the accretionary prism along the Mexican subduction zone, which they confirmed using 2.5-D finite-difference models. In summary, the Haida Gwaii earthquake pumped significant energy into a waveguide, leading to profound focusing and amplification of waves traveling northwestward along the Queen Charlotte Fault system and the Craig earthquake rupture segment.

4 Constraints on Possible Triggering Mechanisms

[10] By what mechanisms might the 2012 Haida Gwaii earthquake waves have initiated failure on the 2013 Craig earthquake segment of the Queen Charlotte Fault system and culminated 3 months later with a M7.5 earthquake? Triggering by transient oscillatory loads with delay between initiation and failure may be explained initiation of a self-sustaining time-dependent failure process, and/or by immediate dynamic weakening and continued loading by tectonic or other processes [Freed, 2005; Parsons, 2005; Griffa et al., 2013]. A significant role for dynamic weakening seems likely as previous modeling studies have shown that only a very restricted class of self-accelerating failure processes explains dynamic triggering [Gomberg, 2001; Freed, 2005; Parsons, 2005]. In one of the few in situ observations of dynamic weakening by seismic waves, Taira et al. [2009] monitored changes in correlations between waveforms of successive repeating earthquakes and inferred that passage of waves from teleseismic earthquakes, with similar amplitudes to those from the Haida Gwaii earthquake, weakened the San Andreas Fault and in one case with a delay of several months. Most dynamic weakening mechanisms appeal to gouge-filled fault cores, where granular physics apply. Models in which oscillating loads raise pore pressures and decrease effective normal stresses typically invoke compaction of under-consolidated, undrained gouge zones (as in liquefaction), but may also arise under specific conditions in which fault gouge zones are over-consolidated and drained [Goren et al., 2010]. Laboratory experiments and numerical models show that granular materials soften under oscillating shear loading, probably as grain contacts break and force chains are disrupted [Johnson and Jia, 2005; Savage and Marone, 2007, 2008; Griffa et al., 2013; Xia et al., 2013] or equivalently in frictional models as critical distances shorten [Parsons, 2005]. Significant softening requires low confining stress [Johnson and Jia, 2005] and strains of the order of μstrains, based on both laboratory and field evidence [Johnson and Jia, 2005; Gomberg and Johnson, 2005]. Although speculative, in addition to having amplitudes in the right range, the long duration of the large amplitude Haida Gwaii waves may have raised local pore pressures temporarily, subsequently permitting softening to occur.

[11] Could the 2012 Haida Gwaii waves also have set into motion slow aseismic slip that promoted the 2013 Craig rupture? Recent studies have suggested that accelerating aseismic slip, which most often is inferred from accompanying foreshocks representing breakage of slip-resisting asperities, is a common precursory feature of transform and subduction thrust plate-boundary fault earthquakes [McGuire et al., 2005; Bouchon et al., 2013; Ito et al., 2013]. If Haida Gwaii waves initiated precursory aseismic slip, such slip does not appear to have been accompanied by seismic breakage of asperities, unlike the aforementioned studies of foreshock activity or dynamic remote aftershock triggering by the Landers or Denali earthquakes. I find no clear evidence of any seismicity along the Queen Charlotte fault system between and during the 3 month period between the 2013 Craig and 2012 Haida Gwaii events, nor for ~100 km or more surrounding the Craig epicenter, either in the Advanced National Seismic System (ANSS) catalog (Figure S2) or in visual scans of continuous data. No earthquakes were reported in the ANSS catalog during the spatial and temporal intervals between the Haida Gwaii and Craig earthquakes, but the catalog is likely incomplete even above the published completeness magnitude of M ~ 2 near the Craig event [Alessandro and Ruppert, 2012] given the sparsity of station coverage. For example, in the 6 months prior to the 28 June, 2004 M6.8 earthquake on the Queen Charlotte fault ~67 km from the Craig epicenter, the ANSS catalog does not contain the events described in Bouchon et al.’s [2013] study of precursory accelerating seismicity and slip, which employed multiple catalogs. Although accelerating seismicity prior to this 2004 earthquake cannot be ruled out given the data limitations, Bouchon et al. [2013] did not report any.

[12] To further test for precursory accelerating seismicity in the Haida Gwaii to Craig interlude period, I visually scanned segments of continuous seismic data for smaller earthquakes during the interlude period, using data from the three closest stations (95, 279, and 304 km from the Craig epicenter) situated to constrain the azimuths of the sources. Clearly observed seismograms of Craig aftershocks of M ~ 1.5 at these same stations served as templates. I scanned the most probable times for increased seismicity rates (one full day immediately after and the week prior to the Haida Gwaii and Craig earthquakes, respectively) and 2 days in between (28th November and December 2012). Although I identified a few earthquakes of possible origin near the Craig rupture and numerous possibly tectonic signals at the closest station of undecipherable origin, no clear acceleration in rate was apparent.

[13] While the failure to find evidence of any interlude activity might imply the Craig section of the Queen Charlotte fault had more homogeneous or frictionally stable fault properties than the faults sampled in the aforementioned studies, such inference is speculative at best given the scarcity of observations in this and some of the other studies; for example, inferences herein and in the study of Bouchon et al. [2013] depend strongly on assumptions about catalog completeness. The background seismicity also provides constraints on fault properties that would be useful for future modeling of possible triggering mechanisms and conditions. The ANSS catalog reveals low but finite seismicity levels in the region prior to the Haida Gwaii earthquake, suggesting the faults are neither so locked that they are seismically silent nor do they appear to chatter continuously as would be expected if creeping [Boatwright and Cocco, 1996] (Figure S2).

5 Conclusions

[14] Focusing and amplification of seismic waves radiated by the 2012 M7.7 Haida Gwaii earthquake, due to excitation of guided waves within the Queen Charlotte Terrace sedimentary trough, likely led to dynamic triggering of the 2013 M7.5 Craig earthquake. Thus, dynamic triggering of major (M > 7) earthquakes seems not only possible but also more probable along the plate boundaries that generate the world's largest earthquakes than elsewhere, due to the sedimentary structures that typify these regions.


[15] Gavin Hayes provided his kinematic slip models for the Haida Gwaii and Craig main shocks, and corroborative evidence for a lack of source directivity in the former. I thank Chris Marone, Michel Bouchon, Gregor Hillers and Gavin Hayes for very helpful reviews.

[16] The Editor thanks Chris Marone and Michael Bouchon for their assistance in evaluating this paper.