Deeper penetration of large earthquakes on seismically quiescent faults

See allHide authors and affiliations

Science  10 Jun 2016:
Vol. 352, Issue 6291, pp. 1293-1297
DOI: 10.1126/science.aaf1496

A microseismic turn off

Certain strike-slip faults do not have the expected number of microearthquakes between larger earthquakes. Jiang and Lapusta suggest that this behavior is down to what the last big earthquake looked like. They found that microseismicity turns off if an earthquake's rupture runs deeper than the fault's locking depth. This appears to be the case along the famous San Andreas Fault and also along other strike-slip faults around the world. The discovery may allow for better estimates of historic earthquake magnitudes and improve hazard assessments.

Science, this issue p. 1293


Why many major strike-slip faults known to have had large earthquakes are silent in the interseismic period is a long-standing enigma. One would expect small earthquakes to occur at least at the bottom of the seismogenic zone, where deeper aseismic deformation concentrates loading. We suggest that the absence of such concentrated microseismicity indicates deep rupture past the seismogenic zone in previous large earthquakes. We support this conclusion with numerical simulations of fault behavior and observations of recent major events. Our modeling implies that the 1857 Fort Tejon earthquake on the San Andreas Fault in Southern California penetrated below the seismogenic zone by at least 3 to 5 kilometers. Our findings suggest that such deeper ruptures may occur on other major fault segments, potentially increasing the associated seismic hazard.

The style of faulting in Earth’s crust is widely accepted to be depth-dependent, with an upper layer that produces earthquakes and a lower layer that predominantly deforms stably (1). The upper layer is commonly referred to as the “seismogenic zone,” and the geodetically estimated boundary between the two is commonly called the “locking depth,” because the seismogenic zone is often locked in the interseismic period. The faulting transition with depth is dominated by temperature and is due to bulk properties transitioning from purely elastic to inelastic and to quasi-static fault friction properties transitioning from velocity-weakening to velocity-strengthening (Fig. 1). Major strike-slip faults feature extreme localization of slip at seismogenic depths (2), as well as continuing localization of the deformation even below the seismogenic zone, based on studies of deep tectonic tremor (3), postseismic deformation (4), and exhumed faults (5); we refer to this deeper localization as “deeper creeping fault extensions.”

Fig. 1 Schematic illustration of our fault model and the locked-creeping transition.

(A) A strike-slip fault model with the seismogenic zone (light gray areas), creeping regions (yellow), and fault heterogeneity (dark gray circles). The initiation point and rupture fronts of a large earthquake are illustrated by the red star and contours, respectively. (B) The locked seismogenic zone and creeping regions below are typically interpreted as having VW and VS rate-and-state friction properties, respectively. In purely rate-and-state models, the VW/VS boundary and locked-creeping transition nearly coincide, and the associated concentrated shear stressing induced at the locked-creeping transition (blue line) promotes microseismicity at the bottom of the seismogenic zone in the interseismic period (blue circles). However, large earthquake rupture may extend seismic slip deeper than the VW/VS boundary, due to enhanced dynamic weakening (DW) at high slip rates, putting the locked-creeping transition and the associated concentrated stressing (red line) within the VS region and hence suppressing microseismicity nucleation.

During the quasi-static interseismic periods between major earthquakes, these deeper creeping fault extensions should continuously load the adjacent locked fault areas and induce microseismicity there, due to the typical assumption that locked areas are seismogenic. The microseismicity at depth is indeed observed on some fault segments, most notably the Parkfield segment of the San Andreas Fault (SAF) (6) (Fig. 2). Such pronounced and concentrated microseismicity that occurs persistently over time should be commonly observed on faults. Yet several stretches of the SAF, including the Cholame, Carrizo, Mojave, and Coachella segments, are seismically quiescent (devoid of small earthquakes) in their interseismic periods, with negligible seismic moment release as compared to the active segments of the fault (7) (Fig. 3). The quiescence over most of the seismogenic zone for such mature faults can be due to their low stress in comparison to their static strength. However, the fault areas right next to the deeper creeping fault extensions should be well stressed and produce microseismicity regardless of whether the shallower fault regions are quiescent or not.

Fig. 2 Observations of large earthquakes and microseismicity patterns on major strike-slip faults.

(A) Spatial relations of the inferred coseismic slip during large earthquakes (in color, with hypocenters as red stars) and microseismicity before (blue circles) and after (black circles), over time periods shown in (B). The large earthquakes are: (i) 2004 Mw 6.0 Parkfield (6, 16), (ii) 1989 Mw 6.9 Loma Prieta (32), and (iii) 2002 Mw 7.9 Denali (33). Small earthquakes within 2, 4, and 5 km of the fault for the three cases, respectively, are projected onto the fault plane (except iii) and plotted using a circular crack model with the same seismic moment and 3 MPa stress drop. (B) (Left) Time evolution of the depths of seismicity (gray circles) and (right) the depth distribution of normalized total seismic moment released before (blue lines), during (red lines), and after (gray) the mainshock (MS). We considered seismicity and coseismic fault slip inside the regions of largest slip outlined by the red dashed lines in (A). Seismic moment release before the Denali event is not shown because of the small number of events.

Fig. 3 Microseismicity and the potential for deeper ruptures on the SAF and the San Jacinto Fault (SJF) in Southern California.

(A) Historical and prehistorical earthquakes on the SAF and SJF, with approximate rupture extent for major events (solid and dashed lines are for well-documented and uncertain cases, respectively) (10). Approximate calendar years for prehistorical events are only shown for the SAF in underlined italics. (B) Seismicity (1981–2011) within 3 km from the SAF and SJF (6, 7). Active seismicity at depth is observed on the Parkfield and San Bernardino segments of the SAF and on the SJF. The Cholame, Carrizo, Mojave, and Coachella segments of the SAF have been seismically quiet for decades. The 1857 and ~1690 events probably penetrated below the seismogenic zone, and similar behavior can occur in future events.

Here we show that the absence of concentrated microseismicity at the bottom of the seismogenic zone on mature fault segments can be due to the deeper penetration of (previous) large earthquakes. We have conjectured this relation based on the following rather general mechanical consideration. If the locked-creeping transition and the associated concentrated and continuous stressing are at the boundary of, or within, the seismogenic zone capable of nucleating seismic events, then one will expect the concentrated stressing to cause microseismicity even on homogeneous faults (8), but especially in the presence of heterogeneity of fault properties and stresses. However, if dynamic earthquake rupture penetrates below the seismogenic zone, it could drop stress in the ruptured creeping areas, making them effectively locked and placing the locked-creeping transition at a depth below the seismogenic zone, where the associated concentrated stressing is unlikely to initiate seismic events. As a result, fault segments with deeper slip in large events would lack microseismicity at greater depths, at least until the locked-creeping transition, which would become shallower with time due to reloading by deeper creep, reaches the seismogenic zone. This argument holds regardless of whether the deeper creeping fault extensions are governed by frictional slip (as explored in this work) or inelastic (e.g., viscoelastic or plastic) flow. As long as the fault extensions are sufficiently localized right below the seismogenic zone, as supported by multiple lines of evidence (35), the loading they impose on the seismogenic zone and its consequences should be the same.

This insight sheds light on the depth extent of large earthquakes, which is important for understanding deep crustal faulting, earthquake scaling relations, and fault segment interactions but is still poorly constrained. Inversions for fault-slip distribution of recent large strike-slip earthquakes usually do not provide reliable constraints on the depth extent of coseismic slip, due to their overall non-uniqueness as well as the decrease of imaging resolution with depth (9, 10). Moreover, monitoring of fault segments with high seismic hazard, such as the SAF in California, is often limited to the late interseismic periods of large events (11). Meanwhile, a growing number of studies have been challenging the notion that dynamic slip during earthquakes is always confined within the seismogenic zone. Geological field studies report the overprinting of natural pseudotachylytes on mylonitic zones, attributed to repeated seismic slip overlapping with aseismic creep below the seismogenic zone (12), in accordance with the transitional regime with semi-brittle deformation mechanisms in conceptual fault models based on experimental and field studies (1, 13). Deeper penetration of larger earthquakes can also explain the observed slip-length scaling of large events (14).

Dynamic rupture propagation into the deeper creeping zones is possible, based on our current laboratory-based understanding of fault friction, which has been gaining acceptance and validation through the comparison of earthquake models with observations. At low slip rates of 10−9 to 10−2 m/s, consistent with plate motion, earthquake nucleation, and postseismic slip, friction has been successfully described by logarithmic rate-and-state friction laws (10, 15). Such laws interpret the seismogenic zones as areas of velocity-weakening (VW) properties that allow for earthquake nucleation, and the other fault areas as having velocity-strengthening (VS) properties that promote stable creep (Fig. 1). Models with the rate-and-state friction reproduce a wide range of fault behaviors, including earthquake sequences and aseismic slip (16). However, at slip rates of ~10−1 m/s and higher, enhanced dynamic weakening of fault frictional resistance, amply documented in high-velocity laboratory experiments (17) and supported by theoretical studies (18), could dominate earthquake rupture propagation. When an earthquake reaches deeper fault extensions, increased strain rate and shear heating could lead to strain localization and dynamic weakening (19), effectively turning the creeping fault regions into seismic ones (20).

We confirmed the hypothesized relation between the depth of coseismic slip in large earthquakes and microseismicity patterns, by numerical simulations of earthquake sequences in two fault models with the laboratory-derived friction laws (Fig. 4). In model M1, dynamic weakening is restricted to occur within the VW region, resulting in earthquake rupture confined within the seismogenic zone, whereas model M2 has dynamic weakening extended deeper into the VS regions below, allowing deeper earthquake rupture. We used the thermal pressurization of pore fluids (18, 21) as the dynamic weakening mechanism, because fluids can be present at deeper fault extensions; however, the qualitative results of the models should be similar for other dynamic weakening mechanisms. The depth extent of efficient dynamic weakening due to thermal pressurization of pore fluids depends on a number of factors, including the shearing zone width (5), permeability, the extent of the inelastic dilatancy (22), and the effectiveness of pore pressure in reducing the effective normal stress (23). In both models, fault heterogeneity that could generate microseismicity is represented by VW patches with nucleation sizes smaller than that of the larger-scale VW region. Although the fault heterogeneity is likely to be more complex, we use the patches and put them only around the VW/VS transition for numerical efficiency. Our simulations are quite challenging, as they reproduce all stages of earthquake sequences, including spontaneous earthquake nucleation, dynamic rupture propagation with full inclusion of wave-mediated stress effects, and aseismic slip (16, 20). We describe the numerical methods and model parameters in the supplementary materials (10).

Fig. 4 The relation between the depth extent of large earthquakes and microseismicity in simulated earthquake sequences.

(A) Model M1 has DW (red hashed region) within the VW region (white), with ruptures confined to the seismogenic zone (SZ). Model M2 has DW extending into the VS region (yellow), potentially allowing for deeper ruptures. VW circular patches of smaller nucleation sizes represent fault heterogeneity at the transitional depths. (B) Different stages in the long-term fault behavior illustrated by snapshots of fault-slip rates on a logarithmic scale. The two models differ in the coseismic rupture extent and the location of the locked-creeping transition with respect to the VW/VS boundary (white dashed outline), and hence in microseismicity activity. (C) Spatial patterns of microseismicity in the post- and interseismic periods of a typical large event (with coseismic slip in color), plotted using the same method as in Fig. 2. Note the concentrated microseismicity in M1 and its near-absence in M2. (D) Time evolution of the locked-creeping transition (red line) and seismicity depths (black dots). The blue and red stars represent the depth of the locked-creeping transition before and after the mainshock, respectively. The time windows equal the recurrence intervals (180 and 280 years, respectively). EQ refers to the large earthquake shown in (B).

The two models demonstrate the conjectured relation between the depth of coseismic slip in large earthquakes, microseismicity patterns, and the locked-creeping transition (Fig. 4, B to D). The transition is defined here as the fault depth with slip rates of 10% of Embedded Image, the maximum slip rate over the fault at the time, which has the physical significance of approximately corresponding to the depth of the highest concentrated stressing. This definition is different from the conventional locking depth inverted from surface geodetic observations (24), which interprets the actual depth distribution of slip rates in terms of a simplified dislocation model with a fully locked shallower layer and a fully creeping deeper fault extension. Coseismic slip of large events penetrates into the deeper fault extensions in model M2, but the coseismic slip is largely confined within the seismogenic zone in model M1, as intended. Correspondingly, in M1, the locked-creeping transition is at the bottom of the seismogenic zone immediately after the large event, causing abundant microseismicity throughout the interseismic period. In M2, however, the locked-creeping transition is below the seismogenic zone during most of the interseismic period, leading to a small number of interseismic events on the VW patches positioned below the large-scale VW/VS transition. The locked-creeping transition migrates up-dip over time, and the migration can be approximately predicted based on the earthquake stress drop Embedded Image, product Embedded Image of the VS friction properties and effective normal stress, fault recurrence interval, and long-term fault-slip rate (10).

Observed microseismicity patterns before and after major earthquakes on tectonic faults further support our hypothesis (Fig. 2) (10). On some fault segments, concentrated microseismicity occurs at what appear to be rheological transitions, with an increased activity below the ruptured region of, and after, a major event, such as the 2004 moment magnitude (Mw) 6.0 Parkfield and 1984 Mw 6.2 Morgan Hills earthquakes (10). In such cases, the slip in the major event probably occurs above the deeper concentrated microseismicity. For larger events, such as the 1989 Mw 6.9 Loma Prieta earthquake, one also observes the occurrence of microseismicity at depth before the mainshock and increased activity after the event, with some variability in local fault areas. In sharp contrast with these smaller events, microseismicity at depth is largely absent before or after all recent major (Mw > 7.5) strike-slip earthquakes that we have considered, including the 2002 Mw 7.9 Denali and 1999 Mw 7.6 Izmit earthquakes (10). According to our models, the absence of microseismicity means that these earthquakes ruptured into the creeping fault extensions, which is more likely for larger events with larger slip. Larger slip at depth and a larger depth extent of the rupture may promote larger slip on shallower fault areas as well, making the fault segment more prone to quiescence at all depths. At the same time, the difference in microseismicity patterns between faults with smaller and larger events is most evident for their deeper parts, whereas the microseismicity in the shallower fault regions is more variable, pointing to a stronger influence of other factors, such as variations of fault structure and properties. The lack of on-fault aftershocks after some of the large strike-slip events was previously attributed to the relatively uniform fault friction, as evidenced by supershear rupture propagation (25). This is consistent with our conclusions, because relatively uniform fault conditions, such as a smooth fault geometry, are likely to promote not only supershear transition but also enhanced dynamic weakening, larger slip, and hence its deeper penetration. More generally, our observations show the absence of aftershocks near regions of large deep slip, regardless of whether the rupture was supershear or not, suggesting that our model provides an additional explanation for the lack of aftershocks.

The relation between the microseismicity and depth of slip in large earthquakes helps us understand historical events and evaluate potential future earthquake scenarios on major strike-slip fault segments. The 1857 Mw 7.9 Fort Tejon earthquake is the last major event on the SAF/San Jacinto fault system in Southern California (7) (Fig. 3) that ruptured the Cholame, Carrizo, and Mojave segments (26, 27). The last major earthquake on the Coachella segment occurred in ~1690 and potentially also ruptured the San Bernardino and Palm Springs segments (11). The recurrence of such events poses severe seismic hazards for Southern California. Virtually no microseismicity is currently observed on all these segments (Fig. 3). In light of our modeling, this observation implies that, ~150 to ~300 years after the previous major seismic events, the locked-creeping transition on those segments is still below the bottom of the seismogenic zone. To achieve that, dynamic rupture on those segments should have penetrated an additional depth below the seismogenic zone, at least 3 to 5 km based on our physical model (10) and perhaps much more. Interseismic geodetic observations indeed suggest that the Carrizo, Mojave, and Coachella segments are accumulating more potency deficit than other fault segments, which they are expected to release in future events (28). Deeper penetration of coseismic slip on the Cholame and Carrizo segments is consistent with the inference of much larger slip at depth, of approximately 11 and 16 m, respectively, than the 3- to 6-m slip at the surface during the 1857 event (27, 29, 30).

In summary, we suggest that the absence of microseismicity at the bottom of seismogenic zones points to a deeper rupture extent in recent major earthquakes, probably due to coseismic weakening of otherwise stable deeper regions. Furthermore, the deeper penetration may be quite common for large events on mature strike-slip faults. We have demonstrated this phenomenon in a friction-based fault model, but the overall dynamics of the process should be similar for viscoplastic deeper fault extensions, which may dynamically localize and weaken due to shear-heating and strain-rate effects during large earthquakes (19) and maintain their localization through the interseismic period because of the resulting structural differences in terms of their grain size and heterogeneity (31). Our study has focused on major strike-slip faults, but it has important relevance for the seismic hazard of megathrust subduction zones that are seismically quiescent, such as the Cascadia subduction zone, given the critical effect of down-dip rupture limit on coastal shaking.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S21

Tables S1 to S3

Movies S1 and S2

References (34117)

References and Notes

  1. Supplementary materials are available on Science Online.
Acknowledgments: This study was supported by the U.S. Geological Survey (USGS) (grant G14AP00033), the National Science Foundation (NSF) (grants EAR 1142183 and 1520907), and the Southern California Earthquake Center (SCEC, funded by NSF cooperative agreement EAR-0529922 and USGS Cooperative agreement 07HQAG0008). This is SCEC contribution no. 6139. Numerical simulations for this study were carried out on the CITerra Dell cluster at the Division of Geological and Planetary Sciences of the California Institute of Technology. We thank J.-P. Ampuero, J.-P. Avouac, E. Hauksson, and M. Simons for helpful discussions and comments on the manuscript. Earthquake catalogs and fault-slip models are compiled from published literature and publicly available sources. Numerical data are available from the authors upon request.

Stay Connected to Science

Navigate This Article