S24A-01 16:00h
Triggered Events due to the 2003 Tokachi-oki Earthquake (Mw 8.1), Japan
The 2003 Tokachi-oki earthquake, (Mw 8.1), one of the largest recent earthquakes in Japan, affected the seismicity of small shallow earthquakes within a few hundred kilometers and remotely triggered deep low frequency tremors at farther distances of 1000 km or more. The earthquake occurred offshore of southeast Hokkaido, where the Pacific plate subducts beneath the North American plate and large earthquakes have recurrently occurred. After the main event in 2003, shallow seismicity increased in the central area and decreased along the western coast of the Hokkaido region. The changes of seismicity could be explained by Coulomb failure stress changes ($\Delta CFF$). In the central area, where active volcanoes are located, $\Delta CFF$ values were 0.1--0.2 MPa if triggered earthquakes had mechanisms consisting of NW-SE compression on right lateral strike-slip faults. Off the western coast, seismicity in the aftershock area of the 1993 south-western Hokkaido earthquake (Mw 7.6) decreased, where $\Delta CFF$ had negative values of about -0.005 MPa for the thrust mechanism of the aftershocks. In order to detect the dynamically triggered events due to the seismic waves, we used continuous velocity waveform data at 642 stations of the High Sensitivity Seismograph Network (Hi-net) that covers most of Japan. The waveform data were filtered with pass-bands of 5--20 Hz so that the local signals in the vicinity of the observation stations could be detected. We constructed RMS envelopes of the filtered waveforms for three components. Two statistical parameters were adapted to quantitatively evaluate changes of amplitudes of 1000 sec envelopes before and after arrivals of the body waves. We examined the commonly used $z$-value and $\beta$-value to see the variant seismicity, substituting the amplitude of the envelope for the earthquake frequency. These obtained values indicated increases of the amplitude in three regions of western Japan, where the travel distances were 1000 km or more. The increased tremors are thought to be deep low frequency events, which occur at depths of 30--60 km near the subducting Philippine Sea plate. The transient stress perturbations in the furthest region were on the order of $10^{-3}$ MPa or more.
S24A-02 16:15h
A Quantitative Test for the Spatial Relationship Between Aftershock Distributions and Mainshock Rupture Properties
Correlating the properties of the mainshock rupture with the location of corresponding aftershocks may provide insight into the relationship between mainshock-induced static stress changes and aftershock occurrence. In this study, we develop a rigorous statistical test to quantify the spatial pattern of aftershock locations with the corresponding distributions of coseismic slip and stress-drop. Well-located aftershock hypocenters are projected onto the mainshock fault plane and coseismic slip and stress drop values are interpolated to their respective location. The null hypothesis H0 for the applied test statistic is: Aftershock hypocenters are randomly distributed on the mainshock fault plane and are not correlated with mainshock properties. Because we want to maintain spatial earthquake clustering as one of the important observed features of seismicity, we synthesize slip distributions using a random spatial field model from which we then compute the respective stress-drop distributions. For each simulation of earthquake slip, we compute the test statistic for the slip and stress-drop distribution, testing whether or not an apparent correlation between mainshock properties and aftershock locations exists. Uncertainties in the aftershock locations are accounted for by simulating a thousand catalogues for which we randomize the location of the aftershocks within their given location error bounds. We then determine the number of aftershocks in low-slip or negative stress-drop regions for simulated slip distributions, and compare those to the measurements obtained for finite-source slip inversions. We apply the test to crustal earthquakes in California and Japan. If possible, we use different source models and earthquake catalogues with varying accuracy to investigate the dependence of the test results on, for example, the location uncertainties of aftershocks. Contrary to the visual impression, we find that for some strike-slip earthquakes or segments of the mainshock rupture plane in California, the null hypothesis cannot be rejected. However, H0 can be rejected for the Homestead Valley fault segment of the Landers 1992 earthquake and the Morgan Hill 1984 earthquake. The test results so far imply that there is only for a fraction of the analyzed earthquakes a statistically significant correlation between the occurrence of aftershocks and low-slip or increased shear-stress regions on the mainshock rupture plane. If this result is confirmed by additional cases studies that are being performed right now, the common understanding of aftershock genesis may have to be rethought.
S24A-03 16:30h
Evolution of Triggered Aftershock Sequences
It is generally believed that large earthquakes can and do trigger distant large aftershocks. In some cases multiple events can be triggered that take the form of secondary aftershock sequences that are spatially separated from the triggering mainshock and its primary aftershock sequence. However, it is not generally known if these distant triggered sequences show similar aftershock sequence characteristics as the primary sequence (e.g., a similar aftershock decay rate and frequency magnitude distribution). A related question is whether the secondary aftershock sequence is controlled more by the initial mainshock or by the local/secondary mainshock. From a stress change point of view, at some distance from the mainshock the largest local aftershock should begin to dominate the local stress changes and therefore have a dominant influence on the evolution of the secondary aftershock sequence. I will present results from several large sequences to examine if these relationships can be better understood on a case by case basis. Additionally, I will explore an existing, modified Omori law based aftershock forecasting model for its performance in forecasting distant triggered events. In the model, triggering of distant events is effectively controlled by the smoothing of the Gutenberg-Richter $a$ value with distance from the rupturing fault. I will present several models for smoothing the $a$ value and examine if any model generates significantly better forecasts based on a likelihood test.
S24A-04 16:45h
A Global Search for Stress Shadows
For years scientists have recognized a decrease in seismicity in the regions surrounding the 1857 Fort Tejon and 1906 San Francisco earthquakes in California. This decrease in seismicity has since been correlated with calculated Coulomb stress decrease and termed a stress shadow. Earthquake hazard assessments increasingly incorporate probability perturbations resulting from calculated static stress changes. However, some researchers have questioned the existence of stress shadows when detailed studies of seismicity rates following other earthquakes failed to show correlations between calculated stress decreases and seismicity rate decreases. Regions of decreased seismicity rate are difficult to find compared with rate-increased regions; we thus look for evidence of post-seismic stress rotations, which are easier to detect, require none of the assumptions inherent to static stress calculations, and are a necessary condition for the occurrence of a stress shadow zone. A stress rotation, as determined by comparison of focal mechanisms before and after the mainshock, indicates a changed regional stress state that suppresses events of a certain mechanism and enhances events of a differing mechanism. To independently test the shadow hypothesis, we examined a global catalog of 119 M$>$7 earthquakes and the events within a 2-degree radius around them that were recorded by the Harvard CMT catalog. We first examined the average mechanism before and after the mainshock, and then compared this with the changes in rate of each of the mechanisms. Of the 119 mainshocks that we identified, 15 showed a rotation of average mechanism following the mainshock that was significant at the 1-sigma level relative to the observed pre-mainshock variability. Of these 15 events, only 2 did not show increases in rate across all mechanisms. That is, only 2 of the 119 show a decrease in at least one mechanism following the mainshock, and also show a significant rotation in mean mechanism. It is only 2 of 119 events worldwide that dynamic stress effects cannot explain the changes in seismicity, suggesting that dynamic triggering may be a much more significant effect globally than static stress triggering.
S24A-05 17:00h
Inverting Peak Ground Motions for Rupture Directivity in Moderate and Large Earthquakes
We present a simple and robust inversion of peak ground motions to determine rupture direction and velocity for moderate and large earthquakes. Peak ground accelerations or velocities ({\it PGA} or {\it PGV}) are corrected for site amplification using standard station corrections, and for hypocenter-receiver distance using $exp(-\eta r)/g(r)$, where $\eta \sim 0.004$ km$^{-1}$ and $g(r)\sim r^{1.5}$ or $r$ for $r < 30$ km for moderate or large earthquakes, respectively. We fit the residual peak motions to the unilateral directivity function $(1 - (v/\beta) cos \theta_{rj})^{-1}$ where $v$ is the {\it apparent} rupture velocity and $\theta_{rj}$ is the solid angle between the rupture direction and the takeoff angle of the S-wave. Bilateral ruptures yield slower apparent rupture velocities. We apply this inversion independently to {\it PGA} and {\it PGV} recorded at $r < 50$ km from 30 large and moderate ($3.5 \leq {\bf M} \leq 6.9$) earthquakes in northern California. The rupture direction can be determined using as few as 6 peak motions if the station distribution is sufficient. The rupture velocity is unstable, however, if the rupture direction is not sampled (if all $\theta_{rj} > 45\deg$). For large earthquakes whose waveforms have been fit for slip distributions, the estimated rupture directions agree well with the rupture models, although the rupture velocity can be overestimated ($v > \beta$) for extended faults. Surprisingly, the {\it PGA} inversions perform slightly better than the {\it PGV} inversions for these large earthquakes. The inferred rupture directions for the {\bf M}5.6 and {\bf M}5.3 1980 Livermore, the {\bf M}6.0 1990 Honeydew, the {\bf M}5.1 1998 San Juan Bautista, the {\bf M}5.0 2000 Yountville, and the {\bf M}4.9 2002 Gilroy earthquakes agree remarkably well with the focal mechanisms and aftershock distributions. Aftershocks are generally located within $70\deg$ of the rupture direction, indicating that rupture directivity strongly influences aftershock occurrence. Moderate ($3.5 \leq {\bf M} \leq 5.0$) earthquakes appear to rupture updip more often than downdip or along-strike. The aftershock sequences for many of the ${\bf M} < 4.0$ earthquakes are too sparse to verify the inferred rupture directions, however.
S24A-06 17:15h
Nonlinear Dynamics, Granular Media and Dynamic Earthquake Triggering
The physical origin of dynamic triggering of earthquakes remains one of the least understood aspects of earthquake processes. Here we show that the dynamic, elastic-nonlinear behavior of fault gouge under the influence of a seismic wave may be responsible for earthquake triggering. We base our hypothesis on recent dynamic experiments conducted in granular media, a surrogate for fault gouge. Our experiments show that, under relatively small effective pressures that one could expect in the fault zone, dynamic waves with strain amplitudes known to trigger earthquakes cause the material modulus to abruptly decrease. The modulus decrease results in material weakening via a simple deformation and instability model. Our conceptual model is that seismic waves from a distant earthquake impinge on a fault that is critically stressed, temporarily decreasing the gouge modulus. The modulus decrease corresponds to an abrupt material strength decrease sufficient to induce fault slip.
S24A-07 17:30h
Systematic Search for Background Seismicity Rate Changes and Correlations at Alaskan Volcanoes
Recent studies have noted a correlation between large earthquakes and localized seismicity rate changes, particularly those associated with volcanic systems. In this study, we analyzed the Alaska Volcano Observatory (AVO) seismicity catalog from late 1989 through mid-2004 for patterns of background seismicity rate changes at the individual monitored volcanoes throughout the Aleutian Arc. We expand the recent studies to include seismic swarms and background seismicity rate changes as well as volcanic eruptions and regional earthquakes. We assume that seismicity rate changes reflect a change in stress on either a local scale, or perhaps over a regional scale when correlated over several volcanoes. The primary analysis was to identify seismicity rate changes at the individual volcanoes. Once all of the rate changes were identified using a z-test, they were examined to determine whether they were man-made or natural. The man-made rate changes were excluded from further study. Of the 27 volcanic regions monitored by AVO, 10 had background seismicity rate changes. Each of these regions had a different characteristic seismicity rate, which can change as often as three times per year or as seldom as once every ten years. Of particular interest are time periods when several volcanic regions have simultaneous rate changes or other significant activity including eruptions, swarms, rate changes at neighboring volcanoes, and large regional earthquakes from the Aleutian subduction zone. For example, in 1996, seven different volcanic regions experienced significant activities: seismic swarms at Akutan (March), Iliamna (August), Strandline Lake (September), and in the Katmai region (October), a deformation episode began at Mt. Peulik (October), an eruption at Pavlof (September), and a background seismicity rate change at Spurr (October). In 1998, Strandline Lake and Spurr experienced a seismicity rate change in early April, and Redoubt had a rate change in early May. The latter rate changes are coincident with the onset of a slow deformation event in the subduction zone of the Cook Inlet region, as determined by GPS data. Other coincident events were noted in the studied regions, but these are still under study.
S24A-08 17:45h
Constraints on Dynamic Triggering from very Short term Microearthquake Aftershocks at Parkfield
The study of microearthquakes helps bridge the gap between laboratory experiments and data from large earthquakes, the two disparate scales that have contributed so far to our understanding of earthquake physics. Although they are frequent, microearthquakes are difficult to analyse. Applying high precision relocation techniques, Rubin and Gillard (2000) observed a pronounced asymmetry in the spatial distribution of the earliest and nearest aftershocks of microearthquakes along the San Andreas fault (they occur more often to the NW of the mainshock). It was suggested that this could be related to the velocity contrast across the fault. Preferred directivity of dynamic rupture pulses running along a bimaterial interface (to the SE in the case of the SAF) is expected on theoretical grounds. Our numerical simulations of crack-like rupture on such interfaces show a pronounced asymmetry of the stress histories beyond the rupture ends, and suggest two possible mechanisms for the observed asymmetry: First, that it results from an asymmmetry in the static stress field following arrest of the mainshock (closer to failure to the NW), or second, that it is due to a short-duration tensile pulse that propagates to the SE, which could reduce the number of aftershocks to the SE by dynamic triggering of any nucleation site close enough to failure to have otherwise produced an aftershock. To distinguish betwen these mechanisms we need observations of dynamic triggering in microseismicity. For small events triggered at a distance of some mainshock radii, triggering time scales are so short that seismograms of both events overlap. To detect the occurrence of compound events and very short term aftershocks in the HRSN Parkfield archived waveforms we have developed an automated search algorithm based on empirical Green's function (EGF) deconvolution. Optimal EGFs are first selected by the coherency of the cross-component convolution with respect to the target event. Then Landweber deconvolution is applied. The resulting source time functions (STF) are often noisy and corrupted by sidelobes due to finite frequency band of the data. They are scanned for subevents, exploiting the consistency of the occurrence of secondary peaks (outliers among the STF maxima) throughout the 30 network channels. Subevents are picked, in many cases to sub-sample precision, by waveform fitting using all the EGFs available. We have detected a total of 30 such multiple or compound eve nts with inter-event delays of less than one second, in a catalog that spans over 10 years of seismicity in Parkfield (2300 cataloged events in our working box). Most of them are not detectable by visual inspection of the seismograms. In most cases, their timing and relative location are consistent with dynamic triggering. Also, the seismicity rate at very early times (less than 0.1 seconds) seems higher than expected from the longer term aftershock seismicity rate observed in the region. This points to dynamic effects in very short term aftershock decay. Finally, more of these immediate aftershocks occur to the NW, as with the earlier NCSN results, although the number of events analysed so far is small. We will discuss these and ongoing observations from the standpoint of dynamic rupture on bimaterial interfaces, supported by numerical simulations.