S31C-01
Global Earthquake Characterization on irregular fault surfaces
Recent developments in global broadband seismic instrumentation and in inverse methods have made it possible to routinely study the slip history of an earthquake based on teleseismic observations, and proceed to predict the local shaking-related damage. However, fast finite fault solutions have generally been limited by the simple parameterization of the fault geometry, i.e., planar faults with constant strike and dip angles inferred from point source CMT solutions. Consideration of the down-dip variation in fault dip angle and using a more appropriate velocity structure usually reshapes the inverted slip distribution and can result in significant differences in inverted slip. Correct representation of fault geometry is also crucial for predicting the local damage and tsunami excitation that result from the earthquake. Here we present an algorithm to invert the slip distribution on irregular fault geometry using seismic and geodetic datasets. We use this new approach to revisit several recent great earthquakes, including the 2008 Wenchuan earthquake.
S31C-02
Global Patterns of Radiated Energy for Thrust Earthquakes in Subduction Zones
Teleseismic studies have found that, for a given seismic moment M0, the radiated energy ES of an earthquake can range over two orders of magnitude. Intraplate and intraslab earthquakes having unusually elevated radiated energy are largely confined to specific high-deformation tectonic settings, an observation which can lead to improving estimates of seismic hazard potential (Choy et al., GJI, 2002, 2004). In contrast, the population of interplate thrust subduction earthquakes has the lowest average ES/M0 ratio among plate-boundary earthquakes. Nevertheless, among subduction plate-boundary earthquakes, there is considerable variability in this ratio. For example, some large earthquakes of this type have anomalously low energy (i.e., low ES/M0 ratio), a characteristic associated with tsunamigenic events involving slow rupture. In a global reconnaissance of the radiated energies of more than 1300 large shallow thrust earthquakes (magnitude greater than about 5.5 and depth < 70 km) that occurred from 1987 to 2006 in subduction zones, we found 270 earthquakes with anomalously low energy radiation comparable to that of tsunamigenic slow earthquakes. The enervated (low energy) earthquakes have a magnitude differential Δ M > 0.5 (where Δ M is the difference between energy magnitude Me and moment magnitude Mw). This reconnaissance also found 152 earthquakes with anomalously high energy radiation. These energetic thrust events have Δ M values less than about -0.2. The global distributions of energetic and enervated events are not random and typically do not overlap. Enervated events are nearly always located at the top surface of a Wadati-Benioff zone defining a narrow zone that can be interpreted as the slab interface. Energetic events occur in high-deformation tectonic settings, such as some marine collision zones involving seamount chains, submerged continent-continent collisions, colliding slabs, regions of complex plate interactions, and slab distortions. Some of these may be intraslab based on their greater depths compared to shallower events that are presumed to be on subduction boundaries. There is no obvious spatial correlation of these enervated events with locations of notable pre-digital tsunami earthquakes, such as the 1896 Japan and 1946 Aleutian events. The majority of enervated thrust events occur beneath forearc basins and not beneath frontal prisms (events south of Java in 1994 and 2006 being notable exceptions). The above trends in the locations of enervated and energetic thrust earthquakes may provide insight into variation of deformation within subduction zones.
S31C-03
Estimation of Scaled Seismic Energy, Apparent Stress and Acceleration
Measurements of seismic energy from scattered coda waves indicate that apparent stress increases with increasing seismic moment, a consequential result for several reasons. For one, it may constrain possible forms of fault weakening with increasing slip. Moreover, if larger earthquakes more efficiently generate energy than their smaller counterparts, strong ground motion from large events would be more intense than anticipated from the extrapolation of smaller events. The relatively sparse strong motion data set does not seem to support this conclusion, leading us to reexamine seismic energy estimates and apparent stress. Our approach to the energy estimation is two-fold. The first technique, an empirical Green's function method, creates spectral ratios of co-located events, so that path and site terms need not be removed. The second method follows Mayeda et al. [2003] in making coda wave corrected spectra accounting for both path and site terms. This method is of particular interest as events occurring over a broad region can be compared, and past results using this technique have shown strong scaling of energy with moment. Using the two energy estimation methods, we compare results from a large data set in California spanning several networks. We find no strong departure from constant scaled energy in our data, which spans a range of Mw 3.0 - 7.1. In particular, we investigate several events in Southern California that have anomalous apparent stress, both low and high. Understanding the origin of high and low stress events and their relationship to acceleration is important for understanding the contribution of source effects to strong ground motion variability. To further quantify the relationship between energy and acceleration, we make direct measurements of acceleration from strong motion recordings of earthquakes. We also consider the relationship developed by Hanks and McGuire [1981] relating RMS acceleration to stress drop and corner frequency. Using our previous estimates of scaled energy, we compare predictions of RMS acceleration with observation.
S31C-04
Estimating Seismic Source Parameters: Inversions for Source Time Functions
Quantifying source parameters of earthquakes is fundamental to understanding the physics of earthquake rupture. Researchers commonly estimate parameters such as seismic moment, corner frequency, and stress drop, but variability in estimates is large due to the use of different methods and source models. Most source parameter estimates have been made using frequency-domain model-based methods, and efforts to use both time-domain and non-parametric methods have largely been neglected. We describe a new, non-parametric method for estimating earthquake source time-functions (STF). Our method uses empirical Green's functions and a non-negative least squares inversion in the time domain to compute the STF, and is not complicated by the effects of frequency-domain deconvolution and waterlevel approximations, which are often seen in STF estimates. We further extend our STF estimate to consider the effects of rupture directivity in the seismic source. We estimate the STFs and corresponding scalar parameters (pulse duration and pulse area) for several events recorded at a small aperture array in the San Jacinto Fault Zone (SJFZ) and compare the results obtained at the closely spaced stations to examine the similarity of the estimates. We also compute corner frequency and moment by fitting the spectra for each event to a seismic source model, and we determine how well the estimates match those obtained from the time domain. To test our ability to resolve rupture directivity using the STF estimate, we use records from M5 earthquakes in the SJFZ that have previously been shown to exhibit such directivity. We evaluate how well our estimate of directivity matches those of previous studies and compare the STFs obtained at azimuthally distributed stations.
S31C-05
Finite-Source Inversion for the 2004 Parkfield Earthquake using 3D Velocity Model Green's Functions
We determine finite fault models of the 2004 Parkfield earthquake using 3D Green's functions. Because of the dense station coverage and detailed 3D velocity structure model in this region, this earthquake provides an excellent opportunity to examine how the 3D velocity structure affects the finite fault inverse solutions. Various studies (e.g. Michaels and Eberhart-Phillips, 1991; Thurber et al., 2006) indicate that there is a pronounced velocity contrast across the San Andreas Fault along the Parkfield segment. Also the fault zone at Parkfield is wide as evidenced by mapped surface faults and where surface slip and creep occurred in the 1966 and the 2004 Parkfield earthquakes. For high resolution images of the rupture processAit is necessary to include the accurate 3D velocity structure for the finite source inversion. Liu and Aurchuleta (2004) performed finite fault inversions using both 1D and 3D Green's functions for 1989 Loma Prieta earthquake using the same source paramerization and data but different Green's functions and found that the models were quite different. This indicates that the choice of the velocity model significantly affects the waveform modeling at near-fault stations. In this study, we used the P-wave velocity model developed by Thurber et al (2006) to construct the 3D Green's functions. P-wave speeds are converted to S-wave speeds and density using by the empirical relationships of Brocher (2005). Using a finite difference method, E3D (Larsen and Schultz, 1995), we computed the 3D Green's functions numerically by inserting body forces at each station. Using reciprocity, these Green's functions are recombined to represent the ground motion at each station due to the slip on the fault plane. First we modeled the waveforms of small earthquakes to validate the 3D velocity model and the reciprocity of the Greenfs function. In the numerical tests we found that the 3D velocity model predicted the individual phases well at frequencies lower than 0.25 Hz but that the velocity model is fast at stations located very close to the fault. In this near-fault zone the model also underpredicts the amplitudes. This implies the need to include an additional low velocity zone in the fault zone to fit the data. For the finite fault modeling we use the same stations as in our previous study (Kim and Dreger 2008), and compare the results to investigate the effect of 3D Green's functions on kinematic source inversions. References: Brocher, T. M., (2005), Empirical relations between elastic wavespeeds and density in the Earth's crust, Bull. Seism. Soc. Am., 95, No. 6, 2081-2092. Eberhart-Phillips, D., and A.J. Michael, (1993), Three-dimensional velocity structure and seismicity in the Parkfield region, central California, J. Geophys. Res., 98, 15,737-15,758. Kim A., D. S. Dreger (2008), Rupture process of the 2004 Parkfield earthquake from near-fault seismic waveform and geodetic records, J. Geophys. Res., 113, B07308. Thurber, C., H. Zhang, F. Waldhauser, J. Hardebeck, A. Michaels, and D. Eberhart-Phillips (2006), Three- dimensional compressional wavespeed model, earthquake relocations, and focal mechanisms for the Parkfield, California, region, Bull. Seism. Soc. Am., 96, S38-S49. Larsen, S., and C. A. Schultz (1995), ELAS3D: 2D/3D elastic finite-difference wave propagation code, Technical Report No. UCRL-MA-121792, 19pp. Liu, P., and R. J. Archuleta (2004), A new nonlinear finite fault inversion with three-dimensional Green's functions: Application to the 1989 Loma Prieta, California, earthquake, J. Geophys. Res., 109, B02318.
S31C-06
The Magnitude-Frequency Distribution on the Southern San Andreas Fault Follows the Gutenberg-Richter Distribution
The magnitudes of any random collection of earthquake hypocenters are generally observed to follow the Gutenberg-Richter (G-R) distribution. One alternative often used in seismic hazard assessments is some form of characteristic magnitude distribution, in which the largest magnitudes on a fault are assigned a higher frequency. The characteristic distribution has been adopted because rates of large paleoseismic earthquakes are found as much as a factor of ~10 higher on selected faults than the G-R prediction (e.g., Wesnousky et al., 1983). However, the short instrumental catalog may not represent the long-term rate. Also, to integrate prehistoric events on a fault with epicenters from a catalog, one must include all regions from which an event could nucleate and then propagate to the fault and leave a paleoseismic signature. We consider the magnitude distribution of the southern San Andreas fault (SAF), a major fault for which a characteristic distribution has been proposed. We include all seismicity nucleating within 20 km of the surface trace, since large earthquakes nucleating up to 20 km from a strike-slip fault may propagate to produce slip on the main fault surface (e.g., Ozacar and Beck, 2004). We look at the historic catalog (1850- 1932), the early instrumental catalog (1932-1984), and the modern instrumental catalog (1984-2008). We find that each catalog, as well as a prehistoric catalog formed by linking paleoseismic event evidence, internally follows a G-R relationship with a b-value of ~1. The most reliable record, the instrumental catalog from 1984-2008, has a rate 3.7 times lower than the paleoseismic rate. We develop an ETAS (Epidemic Type Aftershock Sequence) model (Felzer et al. 2002) that reproduces this observed rate difference. The average seismicity rate of the model matches the observed paleoseismic rate; in addition, due to the clustering of aftershock sequences, selected shorter periods of the synthetic catalog match the instrumental rate. Our results indicate that with an ETAS model the paleoseismic and catalog seismicity rates are consistent with each other, and suggest that magnitudes on the southern SAF follow the G-R distribution.
S31C-07
Aftershock distribution of a M 2.1 earthquake near a geologic structure boundary in a deep South African gold mine
We are operating a high-frequency (up to 200 kHz) seismic network at a depth of 3550 m in a deep South African gold mine (Nakatani et al. 2007). An earthquake of M 2.1 occurred within our network on December 27, 2007 (Yabe et al. 2008). In 150 hours following the event, our AE network detected approx. 20,000 events within 100 m from the center of the our network. This aftershock sequence obeys Omorifs Law. In mines, the number of aftershocks is usually small, but this seems to be simply because vast majority of aftershocks are smaller than detection limit. During the same period, the seismic network of the mine (detection threshold approx. Mw -0.5) detected only nine earthquakes at most. If we assume GR law, comparison of the numbers of aftershocks detected by the both networks suggests that the detection threshold of our AE network is about M - 4. In the area, there is a vertically intruded rock structure made of solidified magma (PG dyke). Thickness is 20 ~ 30 m. This earthquake was expected to be induced by mining around this dyke in 2007 ~ 2008. The boundary position between the dyke and the host rock has been surveyed by the mine, using cores from many boreholes. In addition, our ultrasonic transmission tests have indicated that this contact is generally sharp and there is significant velocity contrast. Velocities within dyke and host rock were fairly uniform (Dyke Vp, Vs are 6.90 km/s, 3.92 km/s respectively, Host Rock Vp Vs are 6.00 km/s, 3.65 km/s respectively). The hypocenter of the M 2.1 earthquake is located within the dyke, not on the contact. The aftershocks lined up on a plane, 30 degrees off vertical; They do not seem to be on the boundary because the dip of the boundary is almost vertical. The distance between them and the closest station of our network was about 6 m. The upper end of aftershock distribution is not clear because of detection limit of our AE network. On the other hand, the downward distribution of the aftershocks seems to be terminated around the boundary. This lower end is within our network coverage.
S31C-08
Aftershock Rates and Spatial Complexity for Recent Moderate Earthquakes in Japan
We looked at the spatial and temporal distributions for 6 recent moderate earthquakes that occurred at shallow depth onshore of Japan and were well recorded by the regional networks. These events are the 2000 Western Tottori (Mw 6.7), 2004 Niigata Chuetsu (Mw 6.6), 2005 Fukuoka (Mw 6.6), 2007 Noto Peninsula (Mw 6.7), 2007 Niigata Chuetsu-oki (Mw 6.8), and 2008 Iwate-Miyagi-ken (Mw 6.8). All of these earthquakes are approximately of similar size, however, the rates of aftershock activity are quite different. The 2004 Niigata and 2008 Iwate-Miyagi earthquakes have significantly more aftershocks than the other 4 events. In the spatial locations of the aftershocks, these two earthquakes have more complex spatial distributions with more aftershocks occurring away from the mainshock fault plane. There appears to be a correlation between the rate of aftershock activity and the spatial complexity of the locations. The sequences with higher rates of aftershock occurrence may be associated with aftershocks triggered in a volume around the mainshock. In contrast, for the other sequences, aftershocks occur mainly in a planar pattern close to the mainshock fault plane. We also looked at the early time sequence of the aftershocks for these events. Using continuously recorded seismograms from nearby borehole stations of Hi-net, aftershocks were identified and counted. From about one minute following the mainshock origin time, we estimate that we can identify aftershocks with magnitudes down to Mj 3.5. For the first few minutes the rate of aftershocks is quite similar for all of the mainshocks. The higher rate of aftershocks for the 2004 Niigata and 2008 Iwate-Miyagi earthquakes appears to begin about 10 minutes after the mainshock. This suggests that the enhanced triggering of aftershock for these 2 earthquakes is caused by some changes in the aftershock region several minutes after the mainshock.