S23C-01 INVITED 13:40h
The Evidence that Long-Range Earthquake Triggering Involves Fluids
Long-range triggering of earthquakes by other earthquakes has been documented for over a decade. We now know that it is a common, perhaps even predictable, occurrence immediately following a large earthquake. We also know that the seismic waves are the transmitters of the stress that start earthquakes within minutes of the mainshock. It is commonly believed that fluids play a crucial role in the triggering mechanism. Here we review the evidence for fluid involvement based on studies in the Western U.S., Greece and Japan. The first line of evidence is that geothermal areas are more prone to triggering than elsewhere. Careful comparison of the stress fields from the 1992 Landers earthquake at well-monitored sites like Parkfield, CA with the stress fields in triggered areas in the Basin and Range show that a comparable level of stress in both sites resulted in triggering in only the Basin and Range (Spudich et al., 1995). The 2002 Denali earthquake triggered areas with hot, deep circulating hydrological systems in the Western U.S., but did not necessarily trigger areas with only high heat flow. However, the lack of triggering in Japan following very large earthquakes like the 2003 $M_w$=8.3 Tokachi-Oki event suggests that geothermal activity by itself is not sufficient for triggering. A few cases of triggering in non-geothermal areas suggest that while hot fluids promote triggering, they may not be necessary. A mechanism which predicts geothermal areas are easily triggered sites, but still allows triggering elsewhere, would explain the data. The second line of evidence comes from determining which aspect of the wavefield determines whether or not a given site triggers. Is the energy, overall amplitude or amplitude in a particular frequency band most effective? There are two observations that suggest that long-period waves are the most important part of the wavefield for long-range triggering. First, triggering after the Denali earthquake occured early in the wavepacket of dispersed Rayleigh waves suggesting that the first-arriving, longest period waves are effective triggers even though they have relatively low-amplitudes. Second, a threshold of $\sim$0.05 cm/s local vertical shaking from long-period waves ($>$30 s) separates earthquakes that trigger local seismicity from those that do not at Long Valley Caldera. Triggering at the Geysers geothermal field is also consistent with a threshold determined by the long-period waves. There may also be some evidence that perhaps the number of events triggered is linearly dependent on the amplitude of the long-period waves. Combining these two sets of observations strongly suggests that long-range triggering involves fluid movement. The increased effectiveness of long-period waves relative to short-period ones cannot be explained by rate and state, stress corrosion or bubble mechanisms. Any mechanism that involves fluid movement through a porous medium is consistent with the long-period threshold since the diffusion of pressure acts as a low-pass filter. One such mechanism, unclogging of fractures temporarily blocked by sediment or precipitate, would also explain the increased efficacy of triggering at geothermal areas. References Spudich, P. et al., JGR, 675-690, 1995.
S23C-02 INVITED 13:55h
Dynamic Coulomb Stress Highly Dependent on Rupture Propagation Parameters
The Coulomb Failure Stress Changes dCFS(t) from an earthquake can be separated into static or permanent (dCFS) and dynamic, time-varying fields. dCFS is controlled only by the fault area, orientation, and slip distribution, while the dynamic portion of dCFS(t) additionally depends on the complexity of the rupture propagation and the resulting radiated waves (Harris et al, 1991; Harris and Day, 1993). Numerous studies have shown correlation of areas of positive dCFS from an earthquake with increased seismic activity (e.g., Das and Scholz, 1981; Stein, 1999). Recent studies of dCFS(t) have pointed out the importance of using dynamic rather than static Coulomb Failure Stress changes for explaining seismic triggering. For example, dCFS(t) constrains directivity effects for large historical earthquakes (e.g., Kilb et al., 2002; Kilb, 2003) which are not included in dCFS estimates. However, these studies of dCFS(t) used somewhat simplified models of the rupture propagation, compared to recent estimates of realistic seismic source parameters. Here, we show that dCFS(t) is significantly affected by variations in slip distribution, rupture velocity, and focal mechanism, and we correlate dCFS(t) with the time varying shear and normal stress to illustrate the origin of the resulting patterns. The variation of dCFS(t) is illustrated for a series of seven large earthquakes from 1939 to 1999 on the North Anatolian Fault, Turkey.
S23C-03 14:10h
Regional and Temporal Variations in the Diffusion of Triggered Earthquakes
Huc and Main (JGR Solid Earth, 2003) derived a method to examine earthquake triggering and how it varies in time. This method allows correlation lengths, {\it L}, and mean triggering distances, {\it $<$r$>$}, to be calculated. Using data from the CMT catalogue it was shown that these characteristic distances evolve slowly but surely over time. They can be fitted, to first order, to a power law of the form {\it r(t) $\sim$ t$^{H}$}, where {\it r} is distance and {\it t} is time. The exponent {\it H} tells us how the triggering signal diffuses in space and time after an earthquake. We have calculated diffusion exponents for different tectonic regimes, and global datasets, using events from the ISC catalogue. Our results show that there are distinct regional variations in {\it L} and {\it $<$r$>$}. However, to second order, we find that the exponent increases at times greater than 100 days, suggesting that different mechanisms influence earthquake triggering over different time scales. This acceleration has been seen previously in work by Marsan {\it et al} (JGR Solid Earth, 2003). Finally, we compare our results to other published values. The methods for determining diffusion of earthquake triggering parameters differ mainly in separation of background, tectonically driven seismicity, and temporally clustered triggered seismicity. We re-examine previous work by applying our method to data used in these studies, to allow the dependence of the results on the separation method to be determined.
S23C-04 14:25h
Earthquakes Triggered by SmS Arrivals
Previous studies suggest that aftershocks are controlled by static stress changes, while remotely triggered earthquakes are controlled by the dynamic stress changes associated with transient seismic waves. However, other studies have concluded that aftershocks are also influenced by dynamic stress changes. To address this issue, I consider 8 well recorded magnitude 5.1-6.1 eathquakes in northeastern Canada as well as 18 magnitude 5.3-7.1 earthquakes in California. A beta-statistic analysis does not reveal evidence of widespread, statistically significant triggering following events smaller than M7. However, a weak seismicity increase is commonly observed beyond the extent of the traditional aftershock zone. The beta statistic increases slightly at a distance of 70-100 km. For the northeastern events, the strongest signal is associated with the M6.1 Saguenay, Quebec earthquake, the largest of the 8 events. The signal is especially persistent following even moderate events in southern and central California. Although the signal from each event is small, its persistent appearance suggests that ``aftershocks'' do occur more frequently than expected at a distance of approximately 100 km, the distance at which post-critical Moho reflections are known to significantly increase ground motions. I show that a persistent ``molehill'' signal is highly unlikely to arise as an artifact. Because static stress change is not influenced by SmS arrivals, these observations suggest that dynamic stress change plays an important role in the triggering of events in the intermediate distance range between conventional aftershocks and remotely triggered earthquakes. The results further suggest that remote triggering is a ubiquitous phenomenon but also one that is difficult to identify with standard statistical approaches. However, if SmS triggering does occur, it would provide a unique opportunity to stack signals from multiple events--perhaps as small as M5--and thus to further explore the prevalence and source properties of remotely triggered earthquakes.
http://pasadena.wr.usgs.gov/office/hough/
S23C-05 INVITED 14:40h
Implications of the Scaling of Dynamic Strains for Earthquake Triggering
Understanding the significance and the underlying physics of dynamic triggering requires knowledge of the characteristics of the causative triggering deformations. Thus, I examine the scaling of dynamic strains with source-site distance and the source size using peak ground velocities (PGV), which serve as proxies for dynamic strains, measured for earthquakes with magnitudes from M4.4 to M7.9 at distances ranging from 0.1 km to 5300 km. The square root of the rupture area provides a useful measure of rupture dimension, and at distances less than this, PGVs increase more and more gradually as the fault is approached, converging to ~150 cm/sec regardless of magnitude. At large distances from the source PGVs vary approximately as the inverse of the distance-squared. Notably, the PGV dependence on magnitude vanishes when the distance is scaled by the inverse of the rupture dimension for each earthquake. All these observations may be explained simply, by considering a constant stress drop per unit fault area, so that at large distances the velocities scale with the total rupture area, and as the fault is approached, the effective source area gets smaller. The scaled observations imply that peak dynamic strains within aftershock zones (within a few rupture dimensions), where triggering is obvious, are no different than those that caused clear triggering out to remote distances (many rupture dimensions) following the M7.1 Hector Mine, M7.3 Landers, and M7.9 Denali earthquakes. These results also suggest that remote triggering may require exceptionally large dynamic deformations, perhaps due to strong directivity, thus explaining its apparent rarity. Another implication of the invariance of the scaled distance dependence of PGVs is that the frequency content of the triggering deformation is at best, of second order importance. For example, the PGVs measured within the aftershock zone of a M4.4 earthquake are nearly the same as those for the aforementioned three, M7.1 and greater, earthquakes at a site of remotely triggered seismicity despite the large differences in their frequency contents.
S23C-06 INVITED 14:55h
The Decay of Aftershocks with Distance Implies Dynamic Triggering
With relocated earthquake catalogs it has become possible to look carefully at how aftershock density decays with distance from the mainshock. It has also become possible to more accurately investigate how far away aftershocks may be triggered, especially after small mainshocks. We focus on small mainshocks $2\leq M \leq 5$, which we treat as point sources. We use standard and relocated earthquake catalogs for Southern California, and the Japanese JMA catalog. In the Southern California relocated catalog of Shearer et al.(2003), we find that aftershock triggering extends to at least 16 km following small mainshocks, even those as small as $M$ 2-3, for which 16 km represents more than 44 fault lengths. The decay of aftershock density ($\rho$) with distance from the mainshock hypocenter ($r$), at distances of 1 fault length to 16 km (where we can see a clear relationship between distance and aftershock density) is well described by $\rho = r^{-1.2}$. If we make the assumption that the number of aftershocks triggered varies linearly with stress amplitude our relationship for distance vs. density agrees with triggering by dynamic stresses, as the amplitude of seismic waves decays somewhat faster than $r^{-1}$. In contrast, the amplitude of static stress change decays as $r^{-3}$. In our data set the percentage of aftershocks occuring at distances between one fault length and 16 km from the mainshock hypocenter is 43% for M 2-3 mainshocks, 31% for M 3-4 mainshocks, and 23% for M 4-5 mainshocks, respectively. We also calculate distances between aftershocks and the mainshock fault of the M 7.3 Landers, California earthquake, using the Shearer et al. (2003) catalog and the mapped fault trace of the mainshock, assuming a vertical dip. We find that the decay of aftershocks with distance is best fit by $r^-1.39 \pm 0.13$ at 98% confidence (for the first five days of data), between 1 and 16 km from the fault. This is similar to the results for the smaller mainshocks, although more difficult to interpret physically since near field stress changes may be complex.
S23C-07 15:10h
Anomalous aftershock decay rates in the first hundred seconds revealed from the Hi-net borehole data
Aftershocks have long been known to diminish in rate approximately as the reciprocal of the time passed in the hours to days after a mainshock. The modified Omori Law (e.g., Utsu et al., 1995) postulates that the rate is proportional to $[t+c]^{-p}$, where t is the elapse time since the mainshock, p is an exponent near one, and c is an apparent offset time, commonly a fraction of an hour. Because it is difficult to observe aftershock activity in the noisy aftermath of larger earthquakes, observations of non-zero c can be explained as an artifact of incompleteness of existing seismicity catalogs (Kagan, 2004). We have so far analyzed Hi-net borehole recordings of 39 shallow earthquakes ranging in magnitude from 3 to 4.8. By scrutinizing the high-frequency signal, we are able to distinguish mainshock coda from early aftershocks, although estimating the number of undetectable aftershocks remains our toughest challenge. We detect up to 9 times more events in the first 1000 s than are recorded in the Hi-net catalog. There is a suggestion of change in b-value from 200 s to 1000 s. We observe a fairly steady rate of aftershocks for the first 100 s, followed by a close to 1/t decay of activity. There appears to be a distinct early stage of aftershock activity that may be related to the poorly understood processes of aftershock generation. This observation may allow us to distinguish between various proposed mechanisms such as rate-and-state friction, viscous response, and fluid migration.
S23C-08 15:25h
Laboratory Study of Transient Stress Effects on Fault Stability
Transient dynamic stresses, such as seismic waves and Earth tides, can destabilize fault slip and trigger earthquakes. Understanding the effects of transient stressing on fault zone frictional strength would enhance our understanding of the earthquake cycle. Earthquake triggering does not depend simply on the magnitude of the trigger but rather might be a function of the dynamic stress associated with the propagating elastic wave. The amplitude and frequency of the seismic wave, along with material properties of the fault zone, may determine whether triggering will occur. In natural settings, earthquake triggering is difficult to study due to static stress transfer near the fault. Therefore, we use laboratory experiments to study this process. We analyzed the shear stress response of a laboratory fault to transient, periodic loading rate oscillations. Experiments were conducted using a servo-controlled, biaxial apparatus with a double-direct shear configuration. Layers of glass beads, 3 mm thick and with nominal frictional contact dimensions of 100 cm2 were loaded via a shear displacement boundary condition consisting of a linear function with a sinusoid superimposed to simulate oscillating, transient stressing. Normal stress was held constant at 5 MPa, which is low enough to avoid comminution of the material. We studied amplitude and frequency of the velocity oscillation in the range 2-10 micron/sec and 0.01-4 Hz, respectively. Glass beads were used because they exhibit repeated, consistent stick-slip frictional sliding. We measured the timing of stick-slip events relative to velocity oscillations to investigate the phase correlation between oscillations and instabilities. If instabilities occurred consistently at a given phase, the timing of the stick-slips was considered correlated with the oscillations. Preliminary results indicate that both amplitude and frequency of the oscillation determine the correlation between the phase of the velocity oscillations and stick-slips. Generally, higher amplitudes (approaching the background loading rate) are needed to produce consistent correlation between oscillations and instabilities. However, each suite of experiments indicates a critical frequency at which correlation between stick-slip and oscillation occurs at lower amplitudes. For instance, 1 Hz oscillations produced strong correlation at amplitudes of 3 micron/sec for experiments with a background loading rate of 10 micron/s. Higher and lower frequencies required larger amplitudes to produce the same correlation. Forward models of the laboratory experiments using rate and state friction equations and assuming quasi-static motion mimic the laboratory results, showing a critical frequency at which small amplitude oscillations correlate with instability. Parameteric studies of the forward models predict that this frequency is related to stiffness. Our results indicate that transient stressing and the passage of seismic waves at a critical frequency and amplitude can destabilize fault slip and cause stick-slip instabilities.