S54A-01
Earthquake Source Modeling using Time-Reversal or Adjoint Methods
In recent years there have been great advances in earthquake source modeling. Despite the effort, many questions about earthquake source physics remain unanswered. In order to address some of these questions, it is useful to reconstruct what happens on the fault during an event. In this study we focus on determining the slip distribution on a fault plane, or a moment-rate density, as a function of time and space. This is a difficult process involving many trade offs between model parameters. The difficulty lies in the fact that earthquakes are not a controlled experiment, we don't know when and where they will occur, and therefore we have only limited control over what data will be acquired for each event. As a result, much of the advance that can be made, is by extracting more information out of the data that is routinely collected. Here we use a technique that uses 3D waveforms to invert for the slip on a fault plane during rupture. By including 3D wave-forms we can use parts of the wave-forms that are often discarded, as they are altered by structural effects in ways that cannot be accurately predicted using 1D Earth models. However, generating 3D synthetic is computationally expensive. Therefore we turn to an `adjoint' method (Tarantola Geoph.~1984, Tromp et al.~GJI 2005), that reduces the computational cost relative to methods that use Green's function libraries. In it's simplest form an adjoint method for inverting for source parameters can be viewed as a time-reversal experiment performed with a wave-propagation code (McMechan GJRAS 1982). The recorded seismograms are inserted as simultaneous sources at the location of the receiver and the computed wave field (which we call the adjoint wavefield) is recorded on an array around the earthquake location. Here we show, mathematically, that for source inversions for a moment tensor (distributed) source, the time integral of the adjoint strain is the quantity to monitor. We present the results of time-reversal experiments using synthetic seismograms computed for point sources and finite sources, building intuition for what to expect. We also show an example for a real event.
S54A-02
Importance of covariance components of waveform data with high sampling rate in seismic source inversion
As computer technology advanced, it has become possible to observe seismic wave with a higher sampling rate and perform inversion for a larger data set. In general, to obtain a finer image of seismic source processes, waveform data with a higher sampling rate are needed. Then we encounter a problem whether there is no limitation of sampling rate in waveform inversion. In traditional seismic source inversion, covariance components of sampled waveform data have commonly been neglected. In fact, however, observed waveform data are not completely independent of each other at least in time domain, because they are always affected by un-elastic attenuation in the propagation of seismic waves through the Earth. In this study, we have developed a method of seismic source inversion to take the data covariance into account, and applied it to teleseismic P-wave data of the 2003 Boumerdes E#202;Zemmouri, Algeria earthquake. From a comparison of final slip distributions inverted by the new formulation and the traditional formulation, we found that the effect of covariance components is crucial for a data set of higher sampling rates (≥ 5 Hz). For higher sampling rates, the slip distributions by the new formulation look stable, whereas the slip distributions by the traditional formulation tend to concentrate into small patches due to overestimation of the information from observed data. Our result indicates that the un-elastic effect of the Earth gives a limitation to the resolution of inverted seismic source models. It has been pointed out that seismic source models obtained from waveform data analyses are quite different from one another. One possible reason for the discrepancy is the neglect of covariance components. The new formulation must be useful to obtain a standard seismic source model.
S54A-03
Rupture Process of Duzce Earthquake from joint analysis of SPOT, GPS, InSAR, strong- motion and teleseismic data: Supershear rupture speed and rapid variations
We have investigated the spatial and temporal evolution of the Duzce earthquake using the fault geometry and surface offsets obtained from the SPOT images, incorporated GPS and InSAR data as well as four strong-motion stations in the vicinity of the rupture. We focused on the rupture velocity to investigate whether super-shear rupture speeds occurred. No constant rupture velocity inversions were able to explain the available strong-motion data when the geodetic data was incorporated. The rupture accelerates up to supershear velocities on east side of the hypocenter, subsequently slows down. Rapid variations in rupture velocity is required to fit all the available datasets. The teleseismic data is consistent with the joint strong-motion and geodetic modeling of the event, with a time shift of about 2 seconds relative to their hand-picked arrival time. This implies that the slow start of the earthquake eminent from the strong-motion data is not observed at teleseismic distances. This problem leads to a more compact solution with major slip at and around the hypocentral area when only teleseismic data is used. Applying sub-pixel correlation of SPOT images acquired before and after the Duzce earthquake, we have mapped a continuous fault trace over 55 km; 15 km longer than the 40 km rupture length reported in the field studies. This observation sheds light on the abrupt eastern termination of the rupture discussed in prior geological and strong- motion surveys of the Duzce Earthquake. 4-segment fault geometry was constructed based on the fault map from the SPOT image analysis. In order to understand the effects of the improved fault geometry on the rapid estimation of earthquake properties, we compared the teleseismic inversion models from the satellite imagery based model with a one-segment model based on the Harvard CMT solution. The 4-segment model improves the fits to the teleseismic waveforms significantly, and gives better predictions of near-field ground motions.
S54A-04
Source Process of the 2007 Niigata-ken Chuetsu-oki Earthquake Derived from Near-fault Strong Motion Data
The 2007 Niigata-ken Chuetsu-oki earthquake occurred on July 16th, 2007, 10:13 JST. We performed a multi- time window linear waveform inversion analysis (Hartzell and Heaton, 1983) to estimate the rupture process from the near fault strong motion data of 14 stations from K-NET, KiK-net, F-net, JMA, and Niigata prefecture. The fault plane for the mainshock has not been clearly determined yet from the aftershock distribution, so that we performed two waveform inversions for north-west dipping fault (Model A) and south-east dipping fault (Model B). Their strike, dip, and rake are set to those of the moment tensor solutions by F-net. Fault plane model of 30 km length by 24 km width is set to cover aftershock distribution within 24 hours after the mainshock. Theoretical Green's functions were calculated by the discrete wavenumber method (Bouchon, 1981) and the R/T matrix method (Kennett, 1983) with the different stratified medium for each station based on the velocity structure including the information form the reflection survey and borehole logging data. Convolution of moving dislocation was introduced to represent the rupture propagation in an each subfault (Sekiguchi et al., 2002). The observed acceleration records were integrated into velocity except of F-net velocity data, and bandpass filtered between 0.1 and 1.0 Hz. We solved least-squared equation to obtain slip amount of each time window on each subfault to minimize squared residual of the waveform fitting between observed and synthetic waveforms. Both models provide moment magnitudes of 6.7. Regarding Model A, we obtained large slip in the south-west deeper part of the rupture starting point, which is close to Kashiwazaki-city. The second or third velocity pulses of observed velocity waveforms seem to be composed of slip from the asperity. Regarding Model B, we obtained large slip in the southwest shallower part of the rupture starting point, which is also close to Kashiwazaki-city. In both models, we found small slip near the rupture starting point, and largest slip at about ten kilometer in the south-west of the rupture starting point with the maximum slip of 2.3 and 2.5 m for Models A and B, respectively. The difference of the residual between observed and synthetic waveforms for both models is not significant, therefore it is difficult to conclude which fault plane is appropriate to explain. The estimated large-slip regions in the inverted source models with the Models A and B are located near the cross point of the two fault plane models, which should have similar radiation pattern. This situation may be one of the reasons why judgment of the fault plane orientation is such difficult. We need careful examinations not only strong motion data but also geodetic data to further explore the fault orientation and the source process of this earthquake. http://www.k-net.bosai.go.jp/k-net/topics/chuetsuoki20070716/inversion/
S54A-05
Source Inversion for the 2007 Chuetsu-oki, Japan, Earthquake: A Case of Difficulty Determining the Source Fault Plane
An earthquake with a CMT Mw of 6.6 occurred on July 16, 2007 along the western coast of Niigata prefecture, Japan. This 2007 Chuetsu-oki earthquake is the first large event whose source fault extends to beneath a nuclear power plant. The earthquake is characterized by difficulty determining the source fault plane. The CMT solution by Ekstrom (2007) indicates reverse faulting with conjugate nodal planes dipping to SE and NW. The aftershock distribution (e.g., ERI, 2007; NIED, 2007) shows not only a major trend dipping to SE but also a minor one to NW, so we cannot determine the source fault plane by using this. GSI (2007) found that uniform slips on the both fault planes can reproduce observed static displacements and InSAR image. We first carried out a preliminary source inversion (Kikuchi and Kanamori, 2003) of teleseismic waveforms from IRIS DMC to determine the geometry and depths of the two fault planes. We then performed full source inversions (Yoshida et al., 1996; Koketsu et al., 2004) of strong motion records observed by K-NET and KiK-net of NIED, and seismometers of TEPCO in the nuclear power plant, for the two fault planes, respectively. Both the full source inversions resulted in almost the same average residual, so that we were still unable to determine the source fault plane. We next constructed the characterized source models consisting of three asperities from the inverted slip distributions and synthesized broadband seismograms, using the empirical Greenfs function method (Irikura, 1986). However, we again found almost the same fit to observed seismograms in the both cases. Two or three distinct pulses can be seen in the seismograms observed at stations close to the source region including those at the nuclear power plant. Since they correspond to three asperities in the characterized source model, we picked the differences of arrival times of the initial motion and the first and third pulses. We then determined the relative location of the first asperity to the hypocenter and that of the third asperity to the first one. The first asperity was finally located in neighborhood of the hypocenter at the same depth, and the third asperity is in the western part of the source region at a depth shallower than that of the hypocenter. These implies that some slips may occur on the fault plane dipping to SE and the other close to the hypocenter might be located on the crossing of the conjugate fault planes.
S54A-06
Multiscale slip inversion analysis of the 2004 Parkfield earthquake
The question of how small ruptures grow into to large earthquakes is fundamental to earthquake seismology. In particular, the question of whether or not the initiation of small earthquakes differs from the initiation of large earthquakes is important for understanding whether or not earthquakes are predictable, and also for developing methods for earthquake early warning. As a case study, we analyze the 2004 Parkfield earthquake using the multiscale slip inversion method [ Uchide and Ide, JGR, 2007] with a multiscale fault model composed of three fault planes of different size, 4.0 × 4.0, 8.0 × 8.0, and 30.0 × 14.0 km2. The advantage of the multiscale approach is that it allows us to analyze the initiation of the earthquake in detail, while still analyzing the entire event at a coarser scale. For the analysis at the smaller two scales, we use velocity seismograms of small events (MW 2.3 - 3.7) as empirical Green's functions to invert the initial part of mainshock accelerograms. All data are from GEOS (General Earthquake Observation System) developed by the USGS. In the frequency band between 1 and 20 Hz, accelerometers and velocity transducers of GEOS provide almost identical records after instrumental response correction, which we confirm using the records of a commonly observed MW 3.3 earthquake. At the largest scale, we use theoretical Green's functions calculated by assuming two different 1D velocity structures, for northeast and southwest of San Andreas Fault [ Liu et al., 2006]. We invert the seismograms and obtain a successful multiscale source model of the 2004 Parkfield earthquake. At the largest scale, the rupture propagates mainly northwestward, as suggested by previous papers [ex. Liu et al., 2006]. In the first second of rupture, we find that slip is very slow (~ 0.2 m/s) but, after that, it rapidly accelerates to more than 3 m/s and propagates bilaterally. This is substantially different from the results of our previous study of the 2004 mid-Niigata prefecture, Japan, earthquake [ Uchide and Ide, 2007], which has the slip-rate faster than 1 m/s, even in the first 0.1 seconds. This difference might arise from different tectonic condition.
S54A-07
A High-Frequency Secondary Event During the 2004 M6.0 Parkfield Earthquake
We present an image of the rupture propagation of the 2004 M6.0 Parkfield earthquake using records from a dense network of local strong motion stations. We back-propagate high-frequency waveforms in 3D with a method, similar to reverse time migration, to obtain an estimate of the distribution of radiated high-frequency seismic energy in space and time. The image is forced to be coherent at the known hypocenter location and the quake origin time by applying small static time shifts obtained using waveform cross-correlation. We observe that the Parkfield earthquake radiated a distinct secondary high-frequency phase, which is located about 12.5~km northwest of the hypocenter with an onset of seismic radiation about 5~s after the rupture initiation. The time history of the back-projection suggests a rupture velocity of 2.5~km/s between hypocenter and subevent. The back-projection result is confirmed by inversion of picked arrival times of the secondary event clearly visible at some of the local and regional stations. By comparing our rupture image with a number of inverted finite slip models, we find that the secondary subevent had a higher moment release than the hypocenter while radiating less high-frequency seismic energy. From this we can infer that the northern subevent exhibited a lower stress drop than the hypocenter. This result is consistent with spatial stress-drop patterns observed from background seismicity and small-magnitude aftershocks. Furthermore, we find that the high- frequency subevent is located in an area of large slip gradient seen in most seismic and geodetic inversions, indicating that slip may have grown abruptly at this point.
S54A-08
Finite-Source Modeling of Micro-earthquakes on the Parkfield Segment of the San Andreas Fault
We have investigated the rupture kinematics of the SAFOD target repeating events by inverting seismic moment rate functions obtained from empirical Green's function deconvolution using the Berkeley Seismological Laboratory High Resolution borehole Seismic Network (HRSN). With this method it is assumed that if a suitable empirical Green's function (eGf) can be found, namely a collocated smaller event with the same radiation pattern as the targeted larger event, the shared propagation, attenuation and site effects can be removed by deconvolution of the smaller signal from the larger one leaving the moment rate function of the target event. The obtained moment rate functions are then inverted for the spatial distribution of moment release, the rupture speed and possibly the slip velocity. In this study we present inversions for the Mw2.1 "San Francisco" and the Mw1.8 "Hawaii" repeating sequences using nearly collocated (order of 10m) M<1.0 events as the eGfs. The deconvolution process recovers stable seismic moment rate functions with excellent signal to noise ratios. The functions display azimuthal variability which may be due to directivity. Our results indicate that the small repeating events are kinematically similar to larger earthquakes in terms of slip-pulse behavior, rupture velocity and slip velocity. However, we find that rupture area is extremely compact (radius of 20m), with large peak slip. Stress drops estimated from the finite-source slip models are correspondingly high, with average and peak stress drops for the 5 studied "San Francisco" sequence events ranging from 8.3-14.5MPa, and 65.0-93.8MPa, respectively. Thus the finite-source modeling is consistent with both the relatively low estimates of average stress drop reported from studies using spectral corner frequency methods (e.g. Imanishi et al., 2004) as well as the high estimates inferred from the tectonic loading asperity model of Nadeau and Johnson (1998). The results indicate that substantial stress and strength heterogeneity exists along the San Andreas fault.