S51B-0496
Nonlinear Inversion for Dynamic Rupture Parameters from the 2004 Mw6.0 Parkfield Earthquake
The Parkfield section of the San Andreas Fault has produced repeated moderate-size earthquakes at fairly regular intervals and is therefore an important target for investigations of rupture initiation, propagation and arrest, which could eventually lead to clues on earthquake prediction. The most recent member of the Parkfield series of earthquakes, the 2004 Mw6.0 event, produced a considerable amount of high-resolution strong motion data, and provides an ideal test bed for analysis of the dynamic rupture propagation. Here, we use a systematic nonlinear direct-search method to invert strong-ground motion data (less than 1 Hz) at 37 stations to obtain models of the slip weakening distance and spatially-varying stress drop (8 by 4 subfaults) on the (vertical) causative segment of the San Andreas fault (40 km long by 15 km wide), along with spatial-temporal coseismic slip distributions. The rupture and wave propagation modeling is performed by a three-dimensional finite-difference method with a slip- weakening friction law and the stress-glut dynamic-rupture formulation (Andrews, 1999), and the inversion is carried out by a neighborhood algorithm (Sambridge, 1999), minimizing the least-squares misfit between the calculated and observed seismograms. The dynamic rupture is nucleated artificially by lowering the yield stress in a 3 km by 3 km patch centered at the location of the hypocenter estimated from strong motion data. Outside the nucleation patch the yield stress is kept constant (5-10 MPa), and we constrain the slip-weakening distance to values less than 1 m. We compare the inversion results for two different velocity models: (1) a 3-D model based on the P-wave velocity structure by Thurber (2006), with S-wave and density relations based on Brocher (2005), and (2) a combination of two different 1-D layered velocity structures on either side of the fault, as proposed by Liu et al. (2006). Due to the non-uniqueness of the problem, the inversion provides an ensemble of equally valid rupture models that produce synthetics with comparable fit to the observed strong motion data. Our preliminary results with the smallest misfits, out of about 3000 tested rupture models, suggest an average slip-weakening distance of 19-81 cm and an average stress drop across the fault of 6.7 - 8.4 MPa. Compared to the kinematic inversion results by Liu et al. (2006) our models with the smallest misfits produce a larger maximum slip (up to about 81 cm) and smaller rupture area, but similar rupture duration (5-7s). The inversions carried out for the layered models tend to produce smaller misfit between data and synthetics as compared to the results using the 3D structure. This suggests that our 3D structure needs improvement, including the Vs-Vp and density-Vp relation. We expect further decrease in the misfit values by increasing the number of tested rupture models.
S51B-0497
Imaging source process of earthquakes from back-projection of high frequency seismograms
Standard methodologies for calculation of the earthquakes source process, are based on inversion procedures which require the calculation of complete source-stations Greens functions. On the other hand alternative procedures have been developed in order to directly retrieve an image of the rupture process from high frequency seismograms (Spudich et. al. 1984, Kao and Shan 2004, Ishii et. al. 2005). In this study we extend the Isochron- Backprojection methodology (Festa et al., 2006), to image the source process of earthquakes, by incorporating the use of high frequency seismograms around the source area. We take full advantage of the dense strong motion networks available in Japan to model the source process of recent Japanese earthquakes. The IBM method differs from conventional earthquake source inversion approaches, in that the calculation of Green's functions is not required. The idea of the procedure is to directly back-project amplitudes of seismograms envelopes around the source into a space image of the earthquake rupture (Pulido et al. 2007). The method requires the calculation of theoretical travel times between a set of grids points distributed across the fault plane, and every station. For this purpose and for simplicity we assume a multi-layered 1D model. All travel times are adjusted by a station correction factor, calculated by taking the difference between observed and theoretical travel times at each station. Next we calculate the rupture time of every grid within the fault plane by assuming some arbitrary constant rupture velocity value, and obtain the isochrones distribution across the fault plane by adding subfaults rupture times and the corresponding travel times for every station. We select waveforms that have clear P and S wavelets, which means stations located approximately between 40 km and 100km from the epicenter. We extract P-wave windows between the origin time of the earthquake and the theoretical arrival of the S-wave, and taper 1s of the waveforms at the end. We band-pass filter the data between 1Hz and 30Hz, and calculate the waveforms envelopes using the root-mean-square of the original waveforms and their Hilbert transform. We calculate a grid "brigthness" by adding all the envelope amplitudes corresponding to every grid isochron time for all stations. The final result is a distribution of the brightness across the fault plane, which gives us an idea of the location of asperities within the fault plane. We obtained an image of the source process of recent Japanese crustal earthquakes, by using data of the K-NET and KiK-net strong motion networks operated by NIED, and applying the Isochrones Backprojection Method (IBM). Our method has the capability to quickly map asperities of large earthquakes, and is able to provide stable estimates of the fault rupture velocity. We investigate the resolution of our source models by exploring different data sets as well as performing synthetic tests. References Festa, G., and A. Zollo, Geophys. J. Int.,166, 745-756, 2006. Ishii, M., P. Shearer, H. Houston, and J. E. Vidale, Nature, 435, 933-936, 2005. Kao, H., and S.J. Shan, Geophys. J. Int., 157, 589-594, 2004. Pulido, N., S. Aoi, and H. Fujiwara, 2007. Rupture process of the 2007 Notohanto Earthquake by using an Isochrones Back-projection Method and K-NET and KiK-net data, (submitted). Spudich, P., and E. Cranswick, Bull. Seism. Soc. Am. 74, 2083-2114, 1984. http://www.j- shis.bosai.go.jp/staff/nelson/index_e.html
S51B-0498
Southern California Adjoint Source Inversions
Southern California Centroid-Moment Tensor (CMT) solutions with 9 components (6 moment tensor elements, latitude, longitude, and depth) are sought to minimize a misfit function computed from waveform differences. The gradient of a misfit function is obtained based upon two numerical simulations for each earthquake: one forward calculation for the southern California model, and an adjoint calculation that uses time-reversed signals at the receivers. Conjugate gradient and square-root variable metric methods are used to iteratively improve the earthquake source model while reducing the misfit function. The square-root variable metric algorithm has the advantage of providing a direct approximation to the posterior covariance operator. We test the inversion procedure by perturbing each component of the CMT solution, and see how the algorithm converges. Finally, we demonstrate full inversion capabilities using data for real Southern California earthquakes.
S51B-0499
Using Resolution Information to Remove Artifacts from GPS Inversions
We present a resolution analysis of an inversion of GPS data from the 2004 Mw6.0 Parkfield Earthquake. This earthquake provides observations at 13 1-Hz GPS receivers that are close to the fault, which allows for a truly co- seismic dataset that can be used to infer the static slip field. Due to the fact that the static field decays rapidly with distance from the source, we find that the resolution of our inverted slip model is very poor at depth and near the edges of our modeled fault plane, far from GPS receivers. The extreme spatial heterogeneity of the model resolution in the static field inversion leads to artifacts in poorly resolved areas of the fault plane. These artifacts look very similar in character to asperities commonly seen in the final slip models of earthquake source inversions, but in fact in this problem they are caused by a surplus of free parameters. The location of these artifacts is dependent upon the station geometry of the available data and the assumed velocity structure of the region. We demonstrate that a nonuniform gridding of model parameters on the fault can remove these artifacts from the inversion. We generate a nonuniform grid whose grid spacing matches the local resolution length on the fault, and show that it outperforms small uniform grids (which generate spurious structure in poorly resolved regions), and large uniform grids (which lose recoverable information in well-resolved areas of the fault). With a synthetic test, we show that our nonuniform grid correctly averages out small-scale structure in poorly resolved areas of the fault while recovering small-scale structure near the surface. Finally, we present an inversion of the Parkfield GPS dataset on the nonuniform grid. The final model in this inversion gives the final slip on the fault plane, which is finely parameterized near the surface and coarsely parameterized at depth.
S51B-0500
Quasi-3D Waveform Inversion for Velocity Structures and Source Process Analyses Using its Results
In this study, we propose an efficient waveform inversion method for 2-D velocity structures and 3-D velocity structures are constructed by interpolating the results of the 2-D inversions. We apply these methods to a source process study of the 2003 Miyagi-ken Hokubu earthquake. We will first construct a velocity model, then determine the source processes of this earthquake sequence using the Green's function calculated with the resultant 3-D velocity model. We formulate the inversion procedure in a 2-D cross section. In a 2-D problem, an earthquake is forced to be a line source. Therefore, we introduce approximate transformation from a line source to a point source (Vidale and Helmberger, 1987). We use the 2-D velocity-stress staggered-grid finite difference scheme, so that the source representation is somewhat different from the original 'source box method' and we apply additional corrections to calculated waveforms. The boundary shapes of layers are expressed by connected nodes and we invert observed waveforms for layer thicknesses at the nodes. We perform 2-D velocity inversions along cross sections which involve a medium-size earthquake and observation points. We assemble the results for many stations and interpolated them to construct the 3-D velocity model. Finally, we calculate waveforms from the target earthquake by the 3-D finite difference method with this velocity model to confirm the validity of the model. We next perform waveform inversions for source processes of the 2003 Miyagi-ken Hokubu earthquake sequence using the resultant 3-D velocity model. We divide the fault plane into northern and southern subplanes, so that the southern subplane includes the hypocenter of the mainshock and the largest foreshock. The strike directions of the northern and southern subplanes were N-S and NE-SW, respectively. The Green's functions for these source inversions are calculated using the reciprocal theorem. We determine the slip models using the 3- D structure and compare them with the models determined using the 1-D structures (Hikima and Koketsu, 2004). The synthesized waveforms in the 3-D structure better explain the observed waveforms than those in the 1-D structures. While the large slip area (asperity) of the mainshock is recovered at the shallow southern part of the northern subplane in the 1-D inversion result, the asperity of the 3-D inversion result is located on the deep central part of the northern subplane. Most of the aftershocks occurred around the asperity of the 3-D result. The asperity of the 3-D result is consistent with that of a previous study from geodetic data. In addition, the 3-D inversion result is in good agreement with the distribution of estimated strong motions in the source area. To examine the reason why the different slip distributions were recovered using the 1-D and 3-D structures, we performed synthetic comparisons. The variations of the Green's functions due to changes in the subfault depths were different between the 1-D and 3-D models, and this difference results in the difference of the two inversion results. Because the waveform difference is larger mostly in a later part, an inversion of body wave parts alone did not produce such difference, which we confirmed by a synthetic calculation. However, in some situation, we cannot extract clear body waves not contaminated by later phases. Therefore, a source process inversion should be performed using accurate Green's functions which include later phases based on well-calibrated 3-D velocity models.
S51B-0501
Resolution Analysis of Teleseismic Waveform Inversion of Rupture Process of Tsunami Earthquake Using 1D and 2.5D Green's Functions
Flat-layered or 1D structure is usually assumed as model structure for source region in the teleseismic waveform analysis of earthquakes, because it is a reasonable approximation for many events in land area. However, for shallow subduction zone earthquakes, the thick ocean (water) layer with dipping ocean bottom and the thick sediments with low seismic wave velocity cause large effects on the teleseismic body waveforms, and these effects are not reproduced by flat-layered structure (e.g., Wiens 1989; Yoshida 1992; Okamoto 1993). In this paper we employ a synthetic resolution test to see the effect of the differences in the Green's functions on the inverted rupture process of a tsunami earthquake: tsunami earthquake occurs very close to the trench axis (e.g., Polet and Kanamori 2000) and the structural effect is the largest of those for the shallow subduction zone earthquakes. We assume a 2.5D realistic near source structure based on a detailed seismic experiments (Kopp et al. 2002) conducted at Java trench where the 2006 Java tsunami earthquake occurred. We compute "realistic" Green's functions for teleseismic body waveforms by a 2.5D FDM (Okamoto 1994; Takenaka and Kennett 1996). "Synthetic data" are then generated by using the 2.5D Green's functions for time-space slip models on a fault with a length of 220 km and a width of 130 km. To the synthetic data we apply a non-linear waveform inversion method in which unit point sources are put at the grid points on the fault and the amplitudes and onset times of them are simultaneously retrieved. We find that if we use the 2.5D or "correct" Green's functions in the inversion, the position of the peaks in the slip models are stably retrieved. On the other hand, considerable misfits are observed between the assumed and inverted slip patterns when we use 1D Green's functions computed for a structure with a water layer and a half space. The peak amplitudes are smaller than the assumed ones and the slips spread over areas larger than the assumed patches of slips (with a size of about 50 km) even when we use 2.5D Green's functions.
S51B-0502
General Importance of Covariance Components of Densely Sampled Observed Data in Inversion Analyses: Implications for Seismic Source Inversion
In geophysical observations, unlike the case of laboratory experiments, data are always inaccurate and insufficient, and so, we need to introduce some prior constraints in order to compromise reciprocal requirements for model resolution and estimation errors in a natural way. For seismic source inversion, smoothness constraints in space and time is commonly applied. By incorporating the prior information into the information from observed data with Bayes' theorem, we can construct a highly flexible model with hyperparameters, which determine the relative weight between the prior and observed data (Yabuki and Matsu'ura, 1992). The optimal values of the hyperparameters can be objectively selected by using AkaikeOs Bayesian Information Criterion (ABIC). Before the introduction of ABIC, we have not had a definitive way to determine the relative weight between the information from observed data and prior constraints. The point of ABIC is that we can objectively determine it based on statistics. That is to say, when we have an enough amount of accurate data, a model that well explains observed data is selected. Conversely, when the data are inaccurate and/or insufficient, the model comes to follow prior constraints (Fukahata, Yagi and Matsu'ura, 2003). Due to an enhanced technology of computers, it has become possible to observe and invert seismic waveform data with a high sampling rate. Observed waveform data is not completely independent of each other due to the effect of un-elastic attenuation of the Earth. Then, if we neglect the data covariance, the situation is similar to the case that duplicated (or triplicated or more) data are inverted. In short, the information from observed data is overestimated, which results in an unstable solution with overfitting. A similar problem has already been reported for an analysis of InSAR data that have highly spatially correlated errors (Fukahata and Wright, 2007). In general, we must take data covariance into account in inverting densely sampled observed data.
S51B-0503
Application of the Back-Projection Technique to Large Earthquakes Using the Hi-net Data
The High Sensitivity Seismograph Network (Hi-net) is one of the world's finest seismographic networks, consisting of more than 700~stations distributed throughout Japan. Its borehole instrumentation ensures high- quality data, and provides an opportunity for investigating characteristics of earthquakes in detail. This is achieved empirically by stacking the seismograms at potential locations of energy radiation using a back- projection method, and studying the coherence and amplitude of the stacks as a function of space and time. The technique has been successfully applied to the December 26, 2004 Sumatra-Andaman and March 28, 2005 Nias earthquakes, providing constraints on their rupture speed, duration, extent, direction, area, and spatio-temporal variations. One unique feature of the back-projection approach for studying earthquake characteristics is that it can be performed in near realtime, as soon as the first-arriving seismic waves are recorded at the Hi-net stations. The Hi-net data are available from October 2000. Since then, there have been more than 20~earthquakes larger than magnitude~7.8 that are good candidates to be studied using the back-projection method. We investigate the characteristics of these events and explore the capabilities and limitations of the back-projection technique. The resolvability of rupture depends on the source location with respect to Japan, which has consequences for the method's applicability for near realtime analysis. We also expand the source grid to allow variations in depth; our previous studies on the two Sumatran earthquakes have assumed horizontal slip planes, which is inaccurate for many events.
S51B-0504
A Kinematic Source Inversion Scheme With New Parametrisation
We present a kinematic finite extent source inversion scheme, introducing an innovative parametrization of the problem. Particularly, we assume a spatial slip distribution composed of overlapping 2D Gaussian functions on regular grid. Temporal evolution of slip is described with prescribed slip velocity function, with free rupture velocity and rise time parameters. Fixing the values of rupture velocity and rise time for whole fault makes the problem linear in static slip. The inversion algorithm works as follows. At first we fix rise time and rupture velocity and calculate seismograms for each Gaussian function separately. Then a linear inversion is performed to get optimal weights (=amplitudes) of these Gaussian functions. Such procedure is done for a number of rupture velocity and rise-time distributions. Optimal values of these distributions are obtained by neighborhood algorithm. L2 norm is used as an objective function. The linear inversion for static slip is done with positivity constraint, so called ‘Quadratic programming' was applied to solve such problem. Our method benefits from simplicity of linear problems and favourable spectral properties of Gaussian function. The latter prevent from employment of artificial smoothing operators. Thus a direct insight into spectral properties of the static slip distribution of real earthquakes is obtained. The method has been applied to the 2000 M6.6 Western Tottori, Japan, earthquake.
S51B-0505
Source-Parameter Estimation for the Earthquakes using Full Newton Method in Frequency Domain
We have developed a new algorithm for estimation of the earthquake source-parameters using the full Newton method in the frequency domain. In exploration seismology, it is well known that estimation of the source time function in the frequency domain is efficient. In this study, we modified the method to estimate the source time function and also focal mechanism of earthquakes. To verify whether the method is working, we tested the method using synthetic waveforms for hypothetical earthquakes. Here synthetics are computed by 3D FEM code. Our synthetic tests show that the new method can successfully invert waveforms for the exact source time functions and focal mechanisms. We are planning to apply this method to real earthquakes in LA region, of which source parameters are well defined from other studies. For Green's function computation SCEC high-resolution 3D velocity model is used. To reduce the possible error coming from un-modeled 3D structure, constant phase shift terms will be corrected in the frequency domain By doing this we will test the possibility to estimate the source time functions for relatively small earthquakes.
S51B-0506
Looking Inside The Blind Test : Parameters, Modelling and Misfit Function
The blind test exercise for the kinematic source inversion has been launched within the European project SPICE to investigate the robustness of the inversion procedures. Although the test was performed with noise free data and exactly known propagation models the quasi-linear inversion performed by several research groups led to significantly different slip maps. Here we systematically investigate the influence of the different components of the inversion procedure onto the final slip maps. We specifically analyze how the selection of a specific parametrization and a misfit function, the approximations in the forward modelling and the station distribution weigh on the non linearity and uniqueness of the solution. The assessment of the uncertainties is also discussed.
S51B-0507
Estimating source parameters and uncertainties using a small aperture array
Source parameter estimation is crucial for studying the physics of earthquake rupture. Several past studies have estimated source parameters but have neglected to calculate the uncertainties of these measurements. We use waveforms of co-located earthquakes recorded at closely spaced surface stations to quantify variations in similar waveforms. Our data are from a small aperture array at Pinon Flats Observatory, California, that recorded ~150 local earthquakes (M0.7 to M3.7) over a one month period. The array consisted of 58 high frequency surface stations; 36 stations were placed in a square grid with 7m spacing and 22 stations were in two orthogonal linear arrays with 21m spacing. Two borehole stations were centered in the grid at depths of 150m and 300m. For each local earthquake, we calculate source parameters in both the frequency and time domains. Empirical Green's function (EGF) techniques are used to remove path effects while leaving local site effects intact. The close spacing and identical instrumentation of the seismic stations results in very similar records at each station and allows us to perform computations on a group of records rather than on each record individually. In the frequency domain, we compute the multitaper spectrum for each earthquake pair (main earthquake and EGF earthquake) at all stations and fit an w-2 source spectrum model over all records simultaneously to determine seismic moment and corner frequency. Similarly, we compute source-time functions in the time domain. We determine uncertainties in the source parameters by using jackknife resampling techniques. Though the individual seismic records for any given earthquake show little variation among stations, we find that the measured source parameters display scatter that cannot be neglected.
S51B-0508
Examination on the interaction of the spatial slip distributions of earthquake doublet (M6.9) of the December 26, 2006 PingTung, Taiwan: discrepancy between teleseismic and strong motion data
An earthquake doublet (M6.9) was occurred at 12:26(UT) and 12:34 (UT) on December 26, 2007 at southwestern offshore of Taiwan near PingTung. These two earthquakes are considered as a doublet due to their close occurrence in time, magnitude and location. The locations of this doublet are near the subducting front where the South China Sea plate subducts underneath the Philippine Sea plate, but the focal depths of these two events lie away from the slab. The focal mechanism shows a normal faulting mechanism for the first event, and a strike-slip mechanism for the second. The locations and the difference in mechanisms give the odd on the occurrence of this earthquake doublet. To investigate the interaction of the two events, we derived the spatial slip distributions of the two events from available geological and seismic data. The teleseismic data were firstly utilized for the verification on the focal mechanism and inverting of the spatial slip distribution. The inland strong motion data and GPS data were not considered in the beginning for the less azimuthal coverage to the offshore events. Though the teleseismic data might have less resolution to the slip distribution, it gives the general spatial slip patterns of the source rupture on the fault planes of the two events. Our results show that two segment faults with different strikes were necessary for the first event to explain the teleseismic waveforms. The asperities of the second event were compensated to the spatial slip distribution of the first event, suggesting the energy interaction of the two events. However, the derived spatial slip distributions from teleseismic data did not explain the inland strong motion data satisfactorily.. To explain the strong-motion data, more detail information on the fault geometry, precise velocity structure and rupture velocity is necessary. Thus, for the determination of the spatial slip distribution, the teleseismic data might provide important information to the general pattern of the spatial slip distribution. However, for the investigation on the scaling of the amount of slips to the earthquake magnitudes, the consistency on the type of data usage is necessary.
S51B-0509
Preliminary Results of Modeling of Strong Ground Motion due to the 2004 Parkfield Earthquake
An earthquake of M 6.0 struck the central coast of California at 10:15:24 a.m. Pacific Standard Time (17:15:24 UTC) on 28 September 2004. The epicenter was 11 km southeast of Parkfield, at a depth of approximately 8 km, with strike-slip mechanism. Analysis of the aftershocks and rupture models indicate that it ruptured along the same section of the fault as those of the similar magnitude Parkfield earthquake series, i.e. along the San Andreas fault in the NW-SE direction. Liu et al. (2006) performed a slip inversion, which suggested predominant NW rupture direction with two main asperities located NW and SE off the epicenter. The aim of the presented study is to investigate mainly recordings at stations located on a particular position "above" the fault, i.e. lying very close to the intersection of the Earth surface and the up-dip prolongation of the fault. Theoretical modeling using a 1D medium and assuming a perfectly planar fault shows that fault-parallel (FP) and vertical (UP) ground motions should be zero exactly on this intersection unlike the fault-normal (FN) ones, or at least very low for stations close to the this intersection. Note that it is a consequence of the properties of the S-wave radiation pattern. However, the observed seismograms show relatively strong signals with even the same maximum amplitudes at these "zero" (FP and UP) components as the FN one. We suggest two possible explanations for such a controversy: 1) The medium surrounding the fault, which is in reality 3D heterogeneous, can generate such a strong signal at the "zero" components. 2) The fault is not perfectly planar, which results in variability of the mechanism along the fault and hence allows to generate signal at the "zero" components. In this contribution we test and quantify these two working hypotheses. To this end, synthetic seismograms are computed by the Discrete Wavenumber and the ADER-DG methods using 1D and 3D velocity structures, respectively, and assuming a perfectly planar and non- planar rupture surfaces with kinematic properties retrieved by Liu et al. (2006). Moreover, we compare our numerical results with observed data.
S51B-0510
Frequency Domain Inversion Of The 2003 Chengkung Earthquake: Resolutions And Uniqueness Analysis
The Mw 6.5 Chengkung Earthquake occurred in eastern Taiwan at 04:38 UTC on 10 December 2003, which provided large amount seismic records. Close strong motion records with good quality site condition, broadband data, and GPS coseismic data are used to constrain the evolution of the slip on the fault plane in time and space. According to the aftershocks distribution, we adopt a two-plane fault geometry extending from epicenter with 45 and 60 degree dip. Green's functions have been calculated using a discrete wavenumber method. Frequency domain inversion procedure is used to retrieve source distributions. Resolutions of using different data set will be discussed. Ground motion seismogram within the frequency 0-1 Hz are usually used by kinematic inversions. Dataset gives the opportunity to evaluate the short distance variability in this frequency band. We discuss the impact of such variability in the resulted inversion uniqueness.
S51B-0511
Source mechanism of the 2006 M5.1 Wen'an Earthquake determined from a joint inversion of local and teleseismic broadband waveform data
On July 4th, 2006, a magnitude 5.1 earthquake occurred at Wen'an, {~}100 km south of Beijing, which was felt at Beijing metropolitan area. To better understand the regional tectonics, we have inverted local and teleseismic broadband waveform data to determine the focal mechanism of this earthquake. We selected waveform data of 9 stations from the recently installed Beijing metropolitan digital Seismic Network (BSN). These stations are located within 600 km and cover a good azimuthal range to the earthquake. To better fit the lower amplitude P waveform, we employed two different weights for the P wave and surface wave arrivals, respectively. A grid search method was employed to find the strike, dip and slip of the earthquake that best fits the P and surface waveforms recorded at all the three components (the tangential component of the P-wave arrivals was not used). Synthetic waveforms were computed with an F-K method. Two crustal velocity models were used in the synthetic calculation to reflect a rapid east-west transition in crustal structure observed by seismic and geological studies in the study area. The 3D grid search results in reasonable constraints on the fault geometry and the slip vector with a less well determined focal depth. As such we combined teleseismic waveform data from 8 stations of the Global Seismic Network in a joint inversion. Clearly identifiable depth phases (pP, sP) recorded in the teleseismic stations obviously provided a better constraint on the resulting source depth. Results from the joint inversion indicate that the Wen'an earthquake is mainly a right-lateral strike slip event (-150°) which occurred at a near vertical (dip, 80° ) NNE trend (210°º) fault. The estimated focal depth is {~}14- 15km, and the moment magnitude is 5.1. The estimated fault geometry here agrees well with aftershock distribution and is consistent with the major fault systems in the area which were developed under a NNE-SSW oriented compressional stress field. Key word: waveform modeling method, source mechanism, grid search method, cut and paste method, aftershocks distribution
S51B-0512
Source Rupture Process of the 2007 Noto Hanto Earthquake, Japan, Obtained from Strong Ground Motion and GPS Data
The 2007 Noto Hanto earthquake (MJMA 6.9) occurred near the west coast of the Noto peninsula, central Japan, on March 25, 2007. This is an inland crustal earthquake having oblique slip source mechanism. The strong ground motions from this event brought severe damages on wooden houses in and around near-source region (e.g., Wajima city and Anamizu town). Near-source strong motion records have great potential to investigate detailed source rupture process in space and time. However, strong ground motions are generally affected by amplifications due to the subsurface structure under the station. It is quite important for obtaining a reliable source model to use appropriate velocity structure models. We have employed an approach in which each strong motion station has its identical velocity structure model. Here, one-dimensional velocity structure model for each strong motion station is constructed after an optimization procedure using the aftershock's waveform data. Firstly, a reference velocity structure model is assumed by referring to the refraction survey using controlled sources by Aoki et al. (1972) and microtremor array measurements by Kanno et al. (2003). This reference velocity structure model consists of three sedimentary layers on the seismic bedrock. Then, the thickness of each sedimentary layer for each station is estimated to fit the observed waveforms of the aftershock by using the Genetic Algorithm. Twelve strong motion stations of K-NET (Kinoshita, 1997) and KiK-net (Aoi et al., 2000) are employed in this study. We succeeded to get a reasonable velocity structure model for each station to give a good fit of the main S-wave part in the observation. The kinematic source rupture process of the 2007 Noto Hanto earthquake is estimated by the linear waveform inversion using multiple time windows (Hartzell and Heaton, 1983). The spatiotemporal smoothing constraints to stabilize the solution are introduced following Sekiguchi et al. (2000). The velocity waveform data in 0.05-1 Hz at 12 strong motion stations are used. The static GPS-measured displacement data at 18 GEONET stations (Miyazaki et al., 1997) are also included in the data set. Static displacement at each station is calculated from the daily coordinate solutions released from GEONET. By including GPS data, we can get a broadband source model in this analysis. The obtained source model shows that large slip is observed in shallow portion of the fault plane near the hypocenter and the spatial slip distribution seems rather simple. This model reproduces both of the strong motion and GPS data fairly well. But, we might need to adjust the geometry of the fault plane to get better fitting of the GPS-measured displacement data. The difference between the source model from strong motion and GPS data and that from strong motion only is not significant. We could get a robust source model using only strong motion data with well-calibrated Green's functions for strong motion data. Finally, in order to evaluate effectiveness of our approach, we will compare the result obtained by the station- dependent velocity structure models with the one obtained from a uniform velocity structure model.
S51B-0513
The rupture velocity of the 2001 Kunlunshan earthquake: A revisit
The rupture processes of the 2001 Kunlunshan earthquake have been studied using seismic and geodetic data. However, there were large discrepancies among previous results, particularly the rupture velocity. The models based on teleseismic P waves generally favored a rupture velocity less or equal to the shear velocity of the upper crust [Antolik, 2004; Ozacar, 2004]. In contrast, the analysis of regional surface waves and teleseismic SH waves support a rupture in 5 to 6 km/sec velocity in at least half of 400 km long rupture plane [Bouchon, 200; Robinson, 2006]. This discrepancy motivated us to explore the uncertainty of the estimated rupture velocity. If we align teleseismic body waves by their arrival times from the hypocenter, the responses from a portion of fault plane turn to follow a straight line, whose slope is the distance to the hypocenter. Then by least-square matching the centroid times of coherent phases, we can determine the locations and rupture times of the corresponding sources. We found that such a method is more precise than the back-projection method. One of the asperities constrained by teleseismic P waves is at 208 km east of the epicenter. Its rupture time is 74 sec, suggesting the lowest average rupture velocity from the hypocenter is about 2.82 km/sec. However, a rupture velocity of 3.9 km/sec would require a rise time larger than 19 sec. We will explore how to reconcile these isolated measurements and the models of finite fault inversions. The slip distributions based on InSAR data and forward spectra element simulation will be used to evaluate the final result.
S51B-0514
Investigation on the rupture behavior of the 2001 Kunlun and 1997 Manyi, China, earthquakes: constraints from near field geological data and regional surface wave
We first determine the nearly 400 km long finite-fault slip distribution of the November 14, 2001 Kunlun earthquake (Ms=8.1) by inverting the teleseismic waveforms and using geological field observation as additional constraints. The geological field observations provide well-determined fault geometry and constrain the amount of slip at the surface. We perform forward modeling of the regional surface waves to validate slip distribution obtained by inverting teleseismic body waves. The phase velocity of the regional surface waves is much slower than the teleseismic body wave, which makes it possible to resolve variations in the rupture velocity during faulting. We find that the initial rupture of the 2001 Kunlun earthquake was almost purely in a strike-slip fashion with a rupture velocity of 1.9 km/sec, increasing to 3.5 km/s in the second fault segment, reaching a rupture velocity of 6.2 km/s in the third segment, and 5.9 km/s in the forth segment, where the maximum offset with a broad fault zone was observed. After that, the rupture velocity decelerated to a value of 3.3 km/s on the final segment. The significant variation in rupture velocity indicates differences in the partition of the earthquake fracture energy during faulting. On the other hand, there are no field investigations for the November 8, 1997 Manyi earthquake (Mw=7.6) and the bilateral rupture makes it difficult to estimate the variation in rupture velocity. Therefore, we validate the fault slip model by forward modeling the surface deformation determined by the InSAR. Preliminary waveform inversion suggests that a reversing-dip fault model, which is consistent with geological data, is necessary to improve the waveform fitting. In particular, our model can predict the asymmetry of the ground displacement in the north and south of the fault as shown in the InSAR study. The rupture behavior of these two earthquakes will provide not only implications on tectonic involvement of the Kunlun fault to the eastward extrusion of Tibet, but also the earthquake triggering mechanism along a mature fault system.
S51B-0515
Source modeling of the Iceland June 2000 seismic sequence
We analyze the seismic sequence which occurred in south Iceland in june 2000. This crisis began with the 17 june 2000 Mw=6.6 event, which is a remarkable example of instantaneous dynamic triggering. Indeed, the main shock was followed by three moderate events within a few seconds (8, 26 and 30 s) located within 80 km. A second Mw=6.6 event occurred 4 days later and west of the 17 june 2000 earthquake. We use teleseismic broadband data to recover the spatio-temporal history of the different events of this seismic crisis. We first determine the focal mechanisms of the two main shocks using the well known iterative deconvolution method of Kikuchi and Kanamori (1991). We find two strike-slip mechanisms, which are consistent with the solution given by Harvard but also with the tectonic settings of this region. To investigate the detailed rupture history of these events, we use a back-projection technique (Ishi et al., 2005) which has been applied to several great earthquake such as the giant Mw=9.3 Aceh Sumatra 2004 earthquake or the very large Mw=8 Kokoxili 2001 earthquake (Walker et al., 2005). Similar approaches have also been developped for studying moderate events (Kao and Shan, 2004, 2006).The great advantage of this method is to solve the history of the propagation directly from the signal itself without any a priori on the source. We adapt this technique to global teleseismic data to investigate these two rupture processes. With a similar approach, we also try to better constrain the locations of the main aftershocks of the June 17 event.