S41A-0921 0800h
Fault Structure from Double-difference Relocations and Waveform Modeling of Aftershocks of the 1992 Landers Earthquake
Aftershocks can provide multiple information for investigating fault zone structure. We used the double-difference (DD) method to determine high-resolution hypocenters of 140 aftershocks of the 1992 Landers, California, $M_W$ = 7.3 earthquake. The DD technique incorporates catalog travel time data and differential travel times from waveform cross-correlation of P- and S-waves. To understand better fault zone geometry and material properties, we performed body wave waveform modeling for 9 aftershocks recorded by a dense seismic array cross the rupture zone of the 1992 Landers earthquake. A forth-order staggered-grid three-dimensional (3D) finite difference (FD) method is used to generate synthetic waveforms. Wave propagations in a 3D medium are computed up to 20 Hz with grid space 15 m. Different fault zone geometry (fault zone depth, width, azimuth) and material properties (vp/vs ratio, wave velocities of the fault zone and host rock) are tested. Waveform modeling results indicate that a wedge-shaped strucutre with depth of approximately 2-3km, width of approximately 100 m, P-wave velocity reduction relative to the host rock of about 50 percent and vp/vs ratio of approximately 4.0 in fault zone is responsible for travel time delay and waveform distortion. The structure follows the strike of surface trace of the rupture. The waveform modeling also implies that the structure is not symmetric to the surface trace and centered about 50 m east of the surface break.
S41A-0922 0800h
Maximum Magnitude in Relation to Mapped Fault Length and Fault Rupture
Earthquake hazard zones are highlighted using known fault locations and an estimate of the fault's maximum magnitude earthquake. Magnitude limits are commonly determined from fault geometry, which is dependent on fault length. Over the past 30 years it has become apparent that fault length is often poorly constrained and that a single event can rupture across several individual fault segments. In this study fault geometries are analyzed before and after several moderate to large magnitude earthquakes to determine how well fault length can accurately assess seismic hazard. Estimates of future earthquake magnitudes are often inferred from prior determinations of fault length, but use magnitude regressions based on rupture length. However, rupture length is not always limited to the previously estimated fault length or contained on a single fault. Therefore, the maximum magnitude for a fault may be underestimated, unless the geometry and segmentation of faulting is completely understood. This study examines whether rupture/fault length can be used to accurately predict the maximum magnitude for a given fault. We examine earthquakes greater than 6.0 that occurred after 1970 in Southern California. Geologic maps, fault evaluation reports, and aerial photos that existed prior to these earthquakes are used to obtain the pre-earthquake fault lengths. Pre-earthquake fault lengths are compared with rupture lengths to determine: 1) if fault lengths are the same before and after the ruptures and 2) to constrain the geology and geometry of ruptures that propagated beyond the originally recognized endpoints of a mapped fault. The ruptures examined in this study typically follow one of the following models. The ruptures are either: 1) contained within the dimensions of the original fault trace, 2) break through one or both end points of the originally mapped fault trace, or 3) break through multiple faults, connecting segments into one large fault line. No rupture simply broke a fault end to end, without stepping over onto another fault. In nearly all of these earthquakes, ruptures propagated beyond the mapped endpoints of the faults and/or connected multiple fault segments into one large rupture. In many California earthquakes, the rupture propagated beyond the originally mapped fault traces. Examples include: the Hector Mine, Landers, and Elmore Ranch earthquakes. In all of these cases, and others (Knuepfer, 1988), faults are not mapped into Quaternary sediments, and therefore could be buried under alluvium. Furthermore, it is observed that multiple fault segments can rupture in a single event. These situations imply that fault lengths used in Southern California hazard estimates may be underestimated, because hazard analysis does not take into account fault segmentation or fault traces that die out in Quaternary alluvium. By underestimating fault lengths, the maximum magnitudes will also be underestimated. In order to access a faults maximum magnitude, the geometry must be accurately defined. To define this geometry, guidelines must be established that account for fault step-overs and buried fault segments. Adding these guidelines will provide a more accurate assessment of the maximum fault length, and a better understanding of the earthquake hazard. In the future we will use geophysical data in combination with structurally geology and geomorphology to constrain the fault geometry. We also plan to expand the data set to include regions outside of Southern California. Better constraints on fault geometry, and additional earthquake analysis, will lead to a more complete understanding of how fault length and fault rupture are related.
S41A-0923 0800h
Fault Zone Structure Extending to Seismogenic Zone in Depth Direction Inferred from the Analysis of Fault Zone Waves
Fault zone characterized as low-velocity zone of seismic waves generates fault zone trapped waves and head waves. These waves are useful to detect the small scale structure of fault zone (the width of the zone from a few tens of meters to hundreds of meters). By use of these waves, structure of several fault zones has been imaged. Some of studies image fault zone as continuous vertical layer at depth, the others as vertical layer localized only at shallow (up to 2 km) depth. In this study we infer the fault zone structure, considering the possibility of the fault zone discontinuity (in other words, fault zone has termination at some position) at depth. We used earthquake data recorded at 500-m-long linear seismometer array across the Mozumi-Sukenobu fault, central Japan, at a depth of 300 m in the underground survey tunnel. Focusing on initial P-waves arrival within the array, two patterns are found: (1) initial P-waves from the two boundaries (between fault zone and host rock) propagating toward center of fault zone, as a result, the latest travel time of initial phase is found within the fault zone (2) initial P-waves propagating from the one boundary to the other boundary. Pattern (1) appears in the most of events occurring at distance more than 15 km east of the array. We investigated the relation between the source location and the initial P-waves arrival pattern within the array by numerical simulation for two models: (a) model with continuous fault zone at depth and (b) the model with discontinuity of fault zone. Results show that pattern (1) appears in the case of sources locating within the fault zone or within the extension of fault zone. Pattern (1) also appears in the case of sources outside of the extension of fault zone when the fault zone has discontinuity. Observation and simulation results indicate two possibilities; all the events showing arrival pattern (1) are located within the fault zone when fault zone is continuous or fault zone has termination between the array and 15 km east of the array along the fault. Next we evaluated intensity of trapped wave excitation by the ratio of Rin to Rout where Rin=Etin/Esin and Rout=Etout/Esout for all events. Etin, Esin, Etout and Esout show energy (square of amplitude) of trapped waves and S waves within and outside the fault zone, respectively. This intensity increases as trapped waves excitation is larger. Estimated intensity indicates systematic variation for spatial distribution of events. Intensity less than 2 (smaller trapped waves excitation) is found for the events shallower than about 8 km and up to 15 km east of the array along the fault. The events up to 15 km east of the array and deeper than about 8 km indicate the intensity greater than 2 (larger trapped waves excitation). And the latter event group is likely to be located at the center of fault zone and the former one at the border of fault zone. The events more than 15 km east of the array also show the intensity greater than 2 independent of the focal depth. Observed initial P-waves arrival pattern and intensity can be explained by the model that fault zone has discontinuity around 15 km east of the array along the fault and extends down to a depth of about 8 km (seismogenic zone) at least. This result is contrast to some of recent studies showing that the fault low-velocity zone is located at only shallow depths, less than about 2 km.
S41A-0924 0800h
A Structural Interpretation of Early Aftershock Double-Difference Event Locations from the 12/22/03 Mw 6.5 San Simeon Earthquake, Central California
The San Simeon mainshock was located 11 km northeast of San Simeon on the California central coast. It nucleated at a depth of some 7 km, on a west-northwest-striking, moderately northeast-dipping thrust fault as determined by the focal mechanism, likely within the Oceanic-West Husuana Fault zone, but without any obvious surface expression in the immediate area. Aftershocks ranging from 2 to 9 km depths, migrated from the hypocenter along this trend to the southeast over some 30 km. Plan and sectional plots of double-difference locations for the first week after the event from Hardebeck et al. (2004) reveal three, distinct aftershock zones (intervals referenced southeast from the hypocenter): (1) a substantial zone extending from 0 to 13 km with a west-northwesterly-striking, northeast-dipping "hanging wall" pattern at 2 to 9 km depths, that is consistent with the mainshock focal mechanism, (2) a minor zone between 15 and 19 km at 2 to 4 km depths, that is of indeterminate orientation and (3) another substantial zone extending from 21 to 33 km with a north-northwesterly-striking vertical pattern at 2 to 8 km depths. The aftershock propagation may have been arrested on the southeast end by the zone 3 structure. Interestingly, the proposed truncating structure coincides with an 8 km-long area of northwest to north-northwesterly-striking, symmetric synclines with 4 km wavelengths mapped by Seiders (1982). Perhaps, the vertical feature formed along a synformal axial plane. Over the next several weeks after the mainshock, substantial seismic in-filling occurred among the three earlier, distinct aftershock zones.
S41A-0925 0800h
An Analysis of Fault Nucleation and Rupture in Westerly Granite Using Continuous Acoustic Emission Monitoring
Determining the mechanics by which intact rocks fracture has significance in earthquake studies, in terms of understanding the failure of asperities or barriers on a fault. In order to study the transition from fracture nucleation to dynamic rupture, we performed triaxial compression tests on Westerly granite cores, using continuous Acoustic Emission (AE) monitoring. Previously, AE has been used to demonstrate the initiation and evolution of quasi-static fracture growth, and more recently, fast AE acquisition systems have provided information during nucleation stages of fracture in intact and pre-fractured samples. However triggered AE systems are limited by the inherent `mask' time, and the necessity of setting trigger levels. We report results from a newly developed Continuous Ultrasonic Waveform Acquisition System, named the Giga RAM Recorder. This system continuously streams 16 channels of 14-bit waveform data onto a 40 GB circular RAM buffer, which can be frozen at any point to capture the preceding 268 seconds of activity (assuming 5 MHz sampling frequency). Discrete AE events are then `harvested' from the continuous record following the experiment. In addition to continuous data, the system concurrently records conventionally triggered AE data throughout testing. Intact granite cylinders of diameter 76.2 mm and length 190.5 mm were loaded at a confining stress of 50 MPa. Two tests were conducted using constant displacement rates, one `fast' (10 x 10-6 m/s) and one `slow' (0.5 x 10-6 m/s). The `fast' loaded sample reached a peak stress of 665 MPa, followed by a total stress drop. In the three seconds prior to the stress drop, AE locations show nucleation and the initiation of a small fracture plane as the applied stress was increasing. For the `slow' loaded sample, peak stress of 642 MPa was followed by a 28 second period of post peak deformation prior to a complete stress drop. In this case, the AE locations show nucleation of the fault, which propagated over an area of approximately 70 mm x 20 mm, before dynamic rupture ensued. In a third test, we loaded a sample under highly sensitive AE feedback control, to initiate and quasi-statically propagate a fracture at increasing rates, over a period of 15 hours. In addition to AE locations, we have used Moment Tensor inversion to determine a sample of AE source mechanisms and have calculated b-values to demonstrate characteristics of fracture nucleation and propagation under these three loading regimes.
S41A-0926 0800h
Formation and Evolution of Strong Discontinuity in Granular Rocks
Planar discontinuities (sharp) in granular rocks are characterized by shear fracture and slip surfaces forming from an initially intact state. The structure of deformation is defined by a jump in the displacement field, called strong discontinuity, in contrast to tabular deformation bands where the jump occurs in the displacement gradient field, called weak discontinuity. Mathematically, strong discontinuity may be considered as a limiting case of weak discontinuity as the thickness of the band approaches zero. We describe essential ingredients of a strong discontinuity model for granular rocks using classical theory of plasticity and the finite element method. They include: (a) a condition for the formation of a strong discontinuity in an initially intact material; (b) a transition constitutive law from a continuum state to a damaged state characterized by a fully developed strong discontinuity; and (c) a mathematical description of the progression of the strong discontinuity to residual state. Each deformation state requires a complete constitutive description of the material, including a constitutive characterization of the frictional resistance on the surface of discontinuity. We describe the role of slip speed on frictional resistance along the slip surface, and elucidate how a well-known state- and velocity-dependent phenomenological constitutive law for the coefficient of friction may be integrated into the framework of the strong discontinuity model. We next describe how the kinematics of strong discontinuity may be captured in the finite element approximation. The proposed finite element technique is based on the concept of embedded strong discontinuity. The idea is to simultaneously capture both the coarse-scale field, representing the macroscopic deformation of the rock, and the fine-scale field, representing the intense deformation along the discontinuity, without resorting to severe refinement of the finite element mesh. Further, the approximation must be insensitive to element alignment and mesh refinement. The resulting multi-scale finite element model is then used to simulate the formation and evolution of a strong discontinuity in a specimen of rock.
S41A-0927 0800h
Effects of gouge zone evolution on frictional and mechanical behavior of fault zones: insights from Distinct Element simulations
Both laboratory experiments and numerical simulations demonstrate that the presence of fault gouge and its evolution through time may influence frictional behavior and stability of fault zones. Understanding the entire earthquake process, however, requires a complete knowledge of the relationships between variations in friction and the accumulation of gouge. We carry out Distinct Element simulations to simulate the breakdown of wall rocks and gouge evolution, in an attempt to determine the parameters that control deformation, strength, and the spatial and temporal distribution of failure events during fault block loading and slip. Fault blocks composed of bonded particles are sheared past each other, either inhibiting or allowing for block fracture and fragmentation, for the purpose of comparison between these two distinct deformation processes. Results of experiments with inhibited fault block damage show that frictional strength of fault blocks increases when fault block slides forward by overcoming interlocked asperities, and it decreases when asperities become unlocked and the fault blocks slip. Variations in friction generally correspond to variations in dilation. In contrast, fault blocks with fracture and fragmentation show a lower peak friction, a result of decreased resistance force by the failure of asperities. Variations in friction with gouge zone evolution are much more complex, depending on many factors, such as initial fault surface roughness, fault block strength, loading rate, and normal stress.
S41A-0928 0800h
Fault-Segment Rupture and Mineralization During Aftershock Localized Fluid Flow
Contemporary seismogenic fault systems can be used as an analogue for understanding paleo-fault systems. We show the fault-hosted goldfields of the Kalgoorlie terrane, Western Australia, are best understood in this framework. Lode gold deposits are the products of focused fluid flow through faulted crust. Gold deposits tend to be localized within low displacement faults adjacent to high displacement faults. These low displacement faults have been interpreted as aftershock structures, repeatedly activated after slip events on the high displacement faults. Deposits in the Mount Pleasant area are clustered on small-displacement structures over $<$10km of the $>$50km-long Black Flag fault. Field relationships and net slip distribution along the fault indicate that the deposits are adjacent to, but not within, a large dilatant jog, where two segments of the fault link together. On this basis we infer the jog was a long-term rupture-arrest site. By analogy with active seismogenic fault systems, rupture on segments of the Black Flag fault changed Coulomb failure stress in the surrounding crust and brought specific zones closer to failure, generating regions of preferential aftershock activity. Stress-transfer modeling of the system helps explain the location of mineralized small-displacement structures around the Black Flag fault and indicates that the gold deposits are hosted on structures that became transiently permeable and localized fluid flow during repeated aftershock sequences. Taking the analysis further, we calculated the cumulative stress change after ruptures on all the large faults in mineralized regions. This effectively identifies those domains that most regularly hosted aftershocks following rupturing or stress triggering of the larger faults. We found that areas of positve stress change closely match the distribution of gold mineralization. Firstly, we conclude long-term rupture arrest sites led to repeated aftershocks and transiently high permeability in particular zones. Thus the permeability of a fault system will vary significantly along strike and with time (equally likely at contractional or dilational fault jogs). Secondly, effects of earthquake dynamics around long-term rupture arrest sites, such as stress triggering and secondary aftershocks, can be identified in the rock record. Thirdly, stress transfer modeling, previously used for earthquake hazard prediction, has potential as a target prediction tool for the mineral industry.
S41A-0929 0800h
3D Geometric and Distinct-Element Models of Deformation and Seismicity on the Hayward Fault: Initial Model and Early Results
We have developed a number of fault surface models by fitting curvi-planar surfaces to earthquakes on and adjacent to the Hayward fault (HF) in the Richmond, Oakland and Hayward, CA areas. The spatial accuracy of the earthquakes has been improved by applying double-difference relocating techniques. In the study area the Hayward fault is defined by a sparse to relatively dense `curtain' of seismicity that extends from the mapped surface trace of the Hayward fault down to depths of about 15 km. By editing out earthquakes that are clearly off the main trace of the HF we are able to isolate events that define a clear curvi-planar surface that is the main or active trace of the HF. Using a number of different techniques - ranging from `eye-balling' and defining the fault trace at various depths and then linking those lines into a plane, to perhaps more objective computer/numerical surface fitting and subsequent smoothing - we present a few different HF models. Differences between different model attempts are, not surprisingly, most significant in regions where data are most sparse. These fault models are being used to define a realistic HF geometry for use with distinct-element that are currently in development. The distinct-element method treats a rockmass as a bonded - in tension and shear - assembly of frictional, elastic spheres. Loading the boundary of the assembly forces bonds between particles to fail in a progressive manner, very effectively simulating rock fractures, which subsequently link up to form thoroughgoing faults. The geometry of the Hayward fault, which has been constructed from relocated hypocenter locations, is introduced into the particle assembly as an irregular plane of particles that separates the intact blocks on either side of the fault. Fault surface asperities within the models will be sites of stress concentration and resist slip. Seismic events in the numerical model are identified when once bonded particles suddenly separate due to excessive shear or tensional stress. These events are manifest as outward expanding shells of velocity vectors that move through the model, sometimes triggering subsequent seismic events. Event magnitude is calculated using the mass and velocity of the particles involved in and surrounding the broken bond. This work is based on the hypothesis that fault surface topography is the major factor that determines the location of earthquakes upon pre-existing faults. These models will predict possible locations of future faulting, both strike- and dip-slip.
S41A-0930 0800h
Acoustic Emissions, Velocities And Permeability Evolution During Formation Of Compaction Bands In Sandstone.
Compaction bands are zones of localized deformation observed in high porosity rock (Mollema et al. [1996], Klein et al. [2001], Fortin et al. [2003]). These planar bands form perpendicular to the direction of maximum compression. Compaction bands display significantly reduced porosity and are potentially important permeability barriers in reservoir rocks and aquifers. To investigate localized compaction and changes in physical properties of porous sandstone, we performed triaxial tests on Bleurswiller sandstone, (50% quartz 30% feldspars and 20% clay, 25% porosity), on Fontainebleau sandstone (100% quartz, 25% porosity) and on Flechtingen sandstone (65-75% quartz, calcite and illite 15%, porosity 5.5-7%). Experiments were performed under wet conditions at a pore pressure of 10 MPa. Thirteen experiments were performed at the Laboratoire de Geologie (Ecole Normal Superieur Paris) and at GeoForschungsZentrum Potsdam. Evolution of volumetric strain, elastic wave velocities and permeability were recorded at confining pressures of 12 and 180 MPa. Acoustic Emission (AE) characteristics during deformation were studied at GeoForschungsZentrum Potsdam. To monitor velocity change and microcracking of sandstone, 10 P-wave sensors and 8 polarized S-wave piezoelectric sensors were glued to the cylindrical surface of the samples. To monitor fracture-induced anisotropy, two additional P sensors were installed in axial direction. Fully digitized waveforms were recorded by 10 MHz/16bit Data Acquisition System with an accuracy of AE hypocenters determination of about 2.5 mm. Location of acoustic emission events reveal the evolution of localized compaction bands in sandstone subjected to axial compression. The formation of the bands depends on rock type and effective pressure. Our experiments show a reduction of permeability across compaction bands by about one to two orders of magnitude (Vajdova et al. [2004]; Holcomb et al., [2003]) suggesting that the bands may act as barriers to fluid flow in porous rocks. Samples were first subjected to increasing confining pressure and subsequently loaded axially. During hydrostatic compression, elastic wave velocities first increased up to 10% due to crack closure and compaction. After onset of axial loading, transverse velocities decreased by 10%-20% owing to induced crack damage and depending on rock type and confining pressure. Axial velocity component first increased slightly with increasing mean stress and dropped to starting value.
S41A-0931 0800h
Integrating Laboratory Compaction Data With Numerical Fault Models: a Bayesian Framework
Both the recurrence times and the potential sizes of earthquakes on a fault are crucial ingredients of seismic hazard assessment. The recovery of fault strength as well as the rate of effective stress build-up after a large earthquake depend on the post-seismic time-evolutions of the hydraulic, frictional, rheological, and poroelastic properties of the healing fault zone. These interrelated physical and chemical processes determine how long it will take for different parts of the previously ruptured fault to reach failure again, thus controlling both the timing and the size of the next rupture. To further explore this phenomenon, we are conducting forward modelling of the hydraulic properties of fault zones during the interseismic period using a bayesian methodology. This approach uses lab-derived compaction laws and their uncertainties to calculate porosity as a function of time. This allows us to determine the statistical robustness of the results of process-based deterministic fault models utilizing experimental uncertainties in both the input parameters and the constitutive relationships. We also develop an inverse method using repeated measurements or estimations of porosity and/or pore pressure in laboratory or field fault zones to derive the constitutive relationships, parameters and uncertainties controlling pore pressure evolution in faults. In this approach, which is complimentary to the forward modelling approach, the bayesian framework allows us to make use of all available prior knowledge (e.g., lithology, permeability and porosity, as well as spatial heterogeneity in these parameters) and to take into account what we know about the data acquisition. Our approach is limited by the fact that existing experimental data are rarely adequate to completely define a single constitutive relationship for a given fault gouge mineralogy and grain size distribution over temperature and effective confining pressures of relevance to actual fault zones. We therefore focus on one experimentally derived compaction law, and emphasize what pieces of information are critical to perform both the forward and the inverse approach.
S41A-0932 0800h
Fracture zone drilling through Atotsugawa fault in central Japan - geological and geophysical structure -
Drilling is an effective method to investigate the structure and physical state in and around the active fault zone, such as, stress and strength distribution, geological structure and materials properties. In particular, the structure in the fault zone is important to understand where and how the stress accumulates during the earthquake cycle. In previous studies, we did integrate investigation on active faults in central Japan by drilling and geophysical prospecting. Those faults are estimated to be at different stage in the earthquake cycle, i.e., Nojima fault which appeared on the surface by the 1995 Great Kobe earthquake (M=7.2), the Neodani fault which appeared by the 1891 Nobi earth-quake (M=8.0), the Atera fault, of which some parts have seemed to be dislocated by the 1586 Tensyo earthquake (M=7.9), and Gofukuji Fault that is considered to have activated about 1200 years ago. Each faults showed characteristic features of fracture zone structure according to their geological and geophysical situations. In a present study, we did core recovery and down hole measurements at the Atotsugawa fault, central Japan, that is considered to have activated at 1858 Hida earthquake (M=7.0). The Atotsugawa fault is characterized by active seismicity along the fault. But, at the same time, the shallow region in the central segment of the fault seems to have low seismicity. The high seismicity segment and low seismicity segments may have different mechanical, physical and material properties. A 350m depth borehole was drilled vertically beside the surface trace of the fault in the low seismicity segment. Recovered cores were overall heavily fractured and altered rocks. In the cores, we observed many shear planes holding fault gouge. Logging data showed that the apparent resistance was about 100 - 600 ohm-m, density was about 2.0 - 2.5g/cm3, P wave velocity was approximately 3.0 - 4.0 km/sec, neutron porosity was 20 - 40 %. Results of physical logging show features of fault fracture zone that were the same as the fault fracture zones of other active faults that we have drilled previously. By the BHTV logging, we detected many fractures of which the strikes are not only parallel to the fault trace bur also oblique to the fault trace. The observations of cores and logging data indicate that the borehole passed in the fracture zone down to the bottom, and that the fracture zone has complicate internal structure including foliation not parallel to the fault trace. The core samples are significant for further investigation on material properties in the fracture zone. And we need data of geophysical prospecting to infer the deeper structure of the fracture zone.
S41A-0933 0800h
Physical Properties of TCDP Drill Hole in Ta-keng Taiwan
The TCDP (Taiwan Chelungpu-fault Drilling Project) project was commenced on Feb, 2004. A comprehensive suit of geophysical logs was collected in TCDP from a depth of 500m to 1300m in the Cholan and ChinShui formation. We try to analysis the shear-wave waveform and wave components propagating into the formation. For sonic measurement, anisotropy may raise from intrinsic structural effects, such as aligned fractures and layering of thin zones. These effects lead to differences in formation elastic properties. Especially fault found with anisotropy evaluation. Using observations of dipole shear sonic velocity can identify the compressive failures and tensile fracture direction. The depths containing the shear zone and compare with geophysical properties such as P and S-wave velocity, Gamma ray, density and resistively to modify shear velocity anisotropy. The direction of the maximum horizontal principal stress will be derived. The stress tensor inverted from earthquakes and regional anisotropy analysis, the physical mechanism responsible for the anisotropy observed in the borehole will be examined.
S41A-0934 0800h
Determination of principal orientations of in-situ stress from anelastic strain recovery measurements of drilling cores close to a fault zone in TCDP Hole-A
The state of stress provides insights into understanding the mechanisms of generation and propagation of Chichi earthquake, the anelastic strain recovery (ASR) measurements, therefore, were carried out on the cores taken from the depths close to a fault zone in the Hole-A of Taiwan Chelungpu-fault Drilling Project (TCDP). Four fault zones were encountered between 1100m and 1300m of depth during drilling. Beneath the shallowest fault zone at the depth of 1111m, two cores, which are located at about 1m and 11m away from the gouge in the fault zone respectively, were used to measure the anelastic strain recovery after in-situ stress release by the drilling. The cores were oriented according to the relation between their bedding planes and sedimentary structure of the region. The anelastic strain recovery of the cores was measured in nine directions in which only six directions are independent. For all the directions of both the samples, the expansional anelastic strains were obtained. Their magnitude of the strains in various directions continuously measured for one or two weeks was of the order of 100_~10-6. In additional, the curves of the relation between the strain and elapsed time from beginning of the measurements were very smooth and stable. By using these strain data, a three-dimensional analysis of the principal orientations of in-situ stresses was conducted. The preliminary results showed that the orientations of the major and intermediate stresses exhibited significant local variation in the two depths. The orientations of the minor principal stress at the two depths, however, were roughly the same for each other and approximately equal to west-east orientation with a small plunge. That is, this orientation of the minor principal stress approximately coincides with orientation of the slip displacement of Chelungpu-fault during the Chichi earthquake. Based on some assumptions, the magnitudes of the three principal stresses can be estimated when logging data of rock density has been obtained. Therefore, the shear and normal stress components on the fault plane encountered at 1111m will be estimated.
S41A-0935 0800h
Pressure drop in a borehole intersecting an active fault
The Corinth Rift, in western Greece, is one of the most active continental Rift in the world, with an opening rate of 1.5cm/yr. Its deformation process is being monitored with a broad range of sensors dispatched across the rift, near the city of Aigio, some 40km east of Patras. In particular, a set of pressure transducers has been set in a 1000m-deep borehole that intersects the active 10km long Aigio fault at a depth of 760m. Below its upper 700m deep cased section, the well has been left open and intersects two artesian aquifers. The upper aquifer is fully hydraulically decoupled from surface aquifers and is developed in tectonized platy limestone, with a 0.5MPa original pressure. Below the fault, the limestone is heavily karstified and the artesian overpressure reaches about 0.85MPa. Hence the fault supports a 0.35MPa differential pressure through the 5m thick radiolarite clay layer that has been smeared along the 150m fault offset. In September 2003, the borehole was let produce water and then was plugged with a packer set at the top of the casing resulting in a direct connection between both aquifers. The pressure is monitored by sensors set just below the packer. Tidal waves are recorded with a resolution better than 1/100. In addition a variety of pressure anomalies have been observed. A 60Pa drop in pore pressure has been recorded at the onset of the S waves generated by the Mw=7.8 Rat Island Earthquake of November, 17th 2003. It is followed by a slow recovery which lasted about 30 minutes. This anomaly, compatible with a minor movement along the fault with a seismic moment of $10^9$Nm, is one of the farthest local effects induced by teleseismic waves ever recorded. A 80Pa pressure drop has been detected 15 minutes before a ML=4.2 earthquake that occured about 15km west of the well. It is much sharper than the coseismic drop. This precursory event exhibits a 2-step recovery that lasted 10 minutes. As seismic sensors located near the well detected no major vibration, we assume the pressure anomaly is transmitted through the karstic conduits and give insight to events close to the hypocenter.
S41A-0936 0800h
Fracture propagation, stable sliding and stick slip by pore pressurizing a fault gouge analog
Thermal pressurization of pore fluid has recently been proposed as a mechanism for velocity weakening of fault gouge materials (Mase et al., [1984]). Rice [1992] and Miller [2003] have also suggested the earthquake cycle is mainly a pore pressure cycle. A Fontainebleau sandstone sample of 14% porosity was deformed tri-axially at the Laboratoire de G\'eologie of ENS Paris. A confining pressure of 100 MPa was used, in saturated conditions. Acoutic properties were monitored through 14 compressional wave piezoceramic transducers (PZT) which were directly glued onto the sample cylindrical surface. During the first cycle, differential stress was slowly raised to 250 MPa. Subsequently, an increase in pore pressure induced brittle failure at Pp=72MPa, with a stress drop of 150 MPa. In addition to triggered Acoustic Emissions (AE), the entire conitinuous ultrasonic waveform of the rupture was recorded. Over 10 000 AE were located, demonstrating the evolution of a fracture nucleation patch of order 1cc. Measured permeability showed that the nucleation of a damage/fault zone induced a drastic reduction in permeability, which could explain the many aftershocks that were observed after the main fracture propagation. Elastic wave velocities also show the extent of permanent damage in the rock. During the second cycle, we re-loaded the fractured sample up to a differential stress of 150 MPa. An initial slow pore pressurization induced stable sliding from 65 to 70 MPa. The pore pressure was then reduced, and a fast pore pressure pulse applied, (from 0 to 80 MPa in two seconds) inducing major stick slip (with an associated stress drop of 50 MPa). Again, in this cycle, more than 15 000 AE were located and aftershocks due to pore pressure re-equilibration were observed, post failure. In this preliminary study, we present a non-exhaustive compilation of data obtained during this pore pressure cycling experiment, including AE locations, continuous ultrasonic waveform summaries, wavespeed and permeability variations, and other mechanical data. Such a complete set of experimental data is, to our knowledge, the first to be obtained and could provide a more complete understanding of the earthquake cycle.
http://www.lassondeinstitute.utoronto.ca/young/people/alex2.htm
S41A-0937 0800h
Study Of Triggering Action Of Great Earthquakes In Qinghai-Tibet Plate And Their Dynamic Effect
In the early 60s of 20th century, The stress triggering effect problems was contributed in by some scholars. Specially in recent years, The problem attracts peoples' great interesting again. Okada(1992) get the Internal deformation field strain field formula due to tectengular dislocation in a half-space. According to it, we get the stress component on any position of half-space and coulomb stress formula on any fault. And research the stress trigging action of Mani(Ms7.9, Nov. 1997) earthquake to Kunlun(Ms8.1, Nov. 2001) earthquake. Results show that , 10-3MPa Coulomb stress due to Mani earthquake is added to the fracture fault of Kunlun earthquake, and this may shift the broken date of latter to earlier about 10 years, Kunlun earthquake is the only great event which is larger than Ms8 in the past 50 years in China. The dynamic effect to block movement and load situation of fractures around epicenter is obvious. The long level sections data across Qilian fracture zone and cross-fault short level data all reflect that, in above area, fault movement are more active than that before Kunlun quake. GPS data mirrors an opposite direction motion after event at northern area of Qinghai-Tebit plate. This produces a relative relaxed stress region between Kunlun fracture zone and Qilian fracture zone in short time. It is advantageous to potential failures to happen. This is proved by three medium earthquakes, which are Yumen earthquake (Ms5.9, Dec.,14, 2002), Delingha earthquake(Ms6.6, Apr.,17, 2003) and Minle earthquake(Ms6.1, Oct.,25.2003) occurred in above area. All of these show that, a large broken event may produce obvious influence on block motion, fracture activity and their stress situation.
S41A-0938 0800h
The analysis and study of fault systems in the Southernmost Part of Okinawa Trough
Taiwan is located in the boundary between the Eurasian and Philippine Sea plates. Due to different subduction, two arc-trench systems in different direction were happened. One is Luzon arc-trench system in N-S direction; the other one is called Ryukyu arc-trench system in E-W direction. The Okinawa Trough is a back-arc basin which was formed by extension of Eurasian plate, and the tectonic setting in this area has a series of normal-faults and igneous bodies. According to previous studies, we know that Southernmost Part of Okinawa Trough (SPOT) have evolved at least two main tensional phases of Okinawa Trough, the first phase probably came up in early Pleistocene and struck in NE-SW direction; and the second phases occurred during late Pleistocene and Holocene changed the direction to E-W. In this study, we have used seismic data collected by R/V Chiu-Lien, Ocean Research I, and R/V L'Atalante to explain the normal-fault systems in the SPOT area. We integrate seismic profiles with corrected bathymetry to relocate these normal faults. Our results show these normal fault systems has two main strikes, respectively N$60\deg$E and N$80\deg$E. We find that most of N$60\deg$E faults are located in the northern slope of SPOT and landward to Taiwan. The N$80\deg$E faults are found in the southern slop and center area of SPOT. Compare with the faults and a new topographic map, we find there were a lot of faults around the canyon, such as North-Mienhua Canyon. We suggest that the origin of the canyon is probably due to these tectonic forces. The canyon is a weak area, and is eroded much fast than the surrounding continental shelf. Passing through a series of erosional processes, the canyon becomes what looks like today. We find a lot of graben structure located in the center of SPOT. This area is the extension axis of SPOT right now. We also find many possible igneous rocks in the seismic profiles, some of them are intrusions and the others penetrate the seabed along the weak zone and form the submarine volcanoes. We have found at least 68 volcanoes in the SPOT area. The interactions of submarine volcanoes, canyons, and fault grabens demonstrate an active tectonic episode.
S41A-0939 0800h
The study of active submarine volcanoes and hydrothermal vents in the Southernmost Part of Okinawa Trough
The study area is located in the Southernmost Part of Okinawa Trough (SPOT), which is a back-arc basin formed by extension of Eurasian plate. Previous research indicated two extensional stages in SPOT area. Many normal-fault structures were come into existence during both extensional processes. The SPOT is presently in an activity tectonic episode. Therefore, the area becomes a frequent earthquake and abundant magmatism. The purpose of this study is to discuss which relationship between tectonics, submarine volcanoes and hydrothermal vents in SPOT area. The investigations are continued from 1998 to 2004, we have found at least twelve active hydrothermal vents in study area. Compare the locations hydrothermal vents with fault systems, we find both of them have highly correlated. We can distinguish them into two shapes, pyramidal shape and non-pyramidal shape. According to plumes height, we are able to divide these vents into two groups near east longitude $122.5\deg$. East of this longitude, the hydrothermal plumes are more powerful and west of it are the weaker. This is closely related to the present extensional axis (N$80\deg$E) of the southern part of the Okinawa Trough. This can be explained the reason of why the more powerful vents coming out of the east group. The east group is associated with the present back-arc spreading system. West of $122.5\deg$, the spreading system are in a primary stage. The andesitic volcanic island, the Turtle Island, is a result of N$60\deg$E extensional tectonism with a lot of faults. Besides the pyramidal shape, this can be proved indirectly. The vents located in the west side were occurred from previous extensional faults and are weaker than the eastern. Therefore, we suggest that if last the extension keeps going on, the hydrothermal vents located at the west side of the longitude $122.5\deg$ will be intensified.
S41A-0940 0800h
From Slip to Stress - A Shortcut
One of the fundamental quantities for studying earthquake dynamics is the static stress change on the fault plane. Being able to reliably estimate the stress change distribution is required in modeling the dynamic rupture of past earthquakes, but also allows to constrain the temporal evolution and the energy budget of earthquake rupture in scenario earthquakes with synthetic slip maps (e.g. for strong-motion simulations). In order to compute the distribution of stress changes on the fault plane from the distribution of static displacements, several numerical methods are in use today, but most of them come at rather high computational cost. I this study we extend the spectral approach developed by Andrews (1980) to account for both slip-parallel and slip-perpendicular stress changes and compare it against existing analytical formulations. We present calculations for slip maps of past earthquakes and find that the resulting stress changes are accurate to about 1-2% of the maximum absolute stress change, while the computation time is greatly reduced. Our method therefore provides a reliable and fast alternative to other approaches. In particular, its speed enables the efficient computation of stress drop distributions for large suites of simulated/inferred slip models, thus facilitating the construction of physically consistent source characterizations for near-source strong-motion simulations. Our tool may also be useful for on-fault Coulomb-stress-change calculations in the case of heterogeneous slip.
S41A-0941 0800h
Dynamic Earthquake Rupture Modeling With Stochastic Fault Stress
We investigate how the evolution of dynamic earthquake rupture is controlled by heterogeneous stress distributions on the fault plane. A 3D finite difference code is used to model rupture on a single vertical fault plane obeying a slip-weakening friction law. The friction parameters (critical slip-weakening distance, coefficients of friction) are kept homogeneous over the fault plane, whereas the distributions of shear and normal stress are modeled as spatial random fields with predefined wavenumber spectra. We test different parametrizations for the stress distributions, being either purely fractal (with variable fractal dimension) or following a von Karman autocorrelation function (with variable correlation lengths). We also investigate how the degree of correlation between shear and normal stress affects the dynamic rupture calculations. Ruptures are initiated at randomly chosen hypocenters and their evolution is tracked in space and time, allowing us to study in detail the conditions required for rupture to grow from small into large events. In our approach we repeatedly use "leftover" stress distributions from previous simulations for which no large ruptures have occured, allowing us to also examine statistical properties (e.g. frequency-size distributions) of the generated earthquake sequences.
S41A-0942 0800h
Simulating Spatio-Temporal Slip Evolution of Fault Zones at Different Evolutionary Stages
Previous studies of spatio-temporal evolution of slip on a fault governed by rate-and-state friction (e.g., Rice, 1993; Ben-Zion and Rice, 1995, 1997; Tullis, 1996; Lapusta et al., 2000) employed frictional properties corresponding to fairly homogeneous faults. In most cases, the only types of heterogeneities were the lab-based depth-variations of the parameters $a$ and $b$ that produce transitions between stable velocity-strengthening and unstable velocity-weakening regimes. In this study we use a constant $a-b$ profile and a depth-dependent distribution of the critical slip distance parameter $L$. In addition, correlated heterogeneities of $L$ along strike are used to model geometrical heterogeneities on faults related to roughness. More specifically, we will perform 3D quasi-static and quasi-dynamic simulations of slip on a strike-slip fault using a family of 2D anisotropic correlated distributions of $L$ having different correlation lengths along strike and downdip. The depth-variation of $L$ over the depth range 3km $< z <$ 12 km, representing the seismogenic zone, accounts for an overall reduction of the gouge thickness (and hence $L$) with depth. Above and below the seismogenic zone, $L$ increases rapidly. The variations of $L$ along strike are chosen to provide approximate representations of faults at different evolutionary stages. Relatively smooth mature faults (like the San Andreas) will be represented with distributions that have large horizontal correlation length, while distributions with small correlation lengths are used to represent rougher immature faults (like the San Jacinto and faults in the eastern CA shear zone). The choices of representative correlation lengths is guided and constrained by maps of fault structures of the type compiled by Wesnousky (1994), and by the compilation of inverted slip histories. The 3D code with various cases of anisotropic correlated distributions of $L$ will be used to study many issues related to observed complex behavior of seismogenic faults including: (1) Nucleation and arrest properties of failure episodes on a heterogeneous fault governed by RSD friction. (2) Comparison between properties of final simulated slip histories and those of the inverted slip histories. (3) Frequency-size and temporal statistics of simulated earthquakes on a heterogeneous fault governed by rate-and-state friction.
http://www.seismo.ethz.ch/srcmod
S41A-0943 0800h
Dynamic Rupture Along a Material Interface With Creation of Off-fault Damage
Recent geological observations along several large strike slip faults show clear asymmetry in the damage pattern of fault zone rocks, with one side having considerably more damage than the other (Dor et al., 2004). The observed asymmetry implies that ruptures along these faults propagate preferentially in one direction. A preferred propagation direction is a predicted outcome of rupture along an interface that separates different elastic media (e.g., Weertman, 1980; Andrews and Ben-Zion, 1997; Ben-Zion and Huang, 2002). Such ruptures produce dynamic dilation at the tip that propagates in the direction of slip on the more compliant side of the fault, and dynamic compression at the opposite tip. Consequently, rupture along a material interface evolves to a unidirectional wrinkle-like pulse that propagates in the direction of slip on the compliant side of the fault. In addition, wrinkle-like ruptures along a material interface produce strongly asymmetric fault-normal motion near the propagating tip, with larger motion on the compliant side. Rupture along a material interface has the following two competing mechanisms for creation of off-fault damage. (1) Anelastic deformation on the extensional quadrant in the preferred propagation direction, which for a bi-material configuration is on the stiffer side, as was calculated by Andrews (2004) for rupture in a homogenous solid. (2) Anelastic deformation on the more compliant side due to the asymmetric motion across the fault. Mechanism (1) is favored by low normal stress, small contrast of material properties across the fault, and large difference between the static and kinetic coefficients of friction. Mechanism (2) is favored by the opposite set of conditions. In this study we attempt to quantify the conditions for which the above two mechanisms are active. Previous analytical and numerical studies found that wrinkle-like ruptures between two purely elastic materials diverge for a broad range of conditions (e.g., Adams, 1995; Ben-Zion, 2001; Ranjith and Rice, 2002; Adda-Bedia and Amar, 2003). Another goal of our study is to find whether the creation of off-fault damage will provide a natural length scale for the wrinkle-like pulse and a mechanism that stabilizes the divergence in the purely elastic case. The results will provide an improved understanding of ruptures along a material interface, and important input for inferring from field observations the preferred propagation direction of ruptures along faults.
S41A-0944 0800h
Split Nodes and Fault Zone Models for Dynamic Rupture Simulation
The accuracy of numerical calculation of the dynamic rupture process of earthquakes mainly depends on the fault boundary condition on the fault where friction is taking place. During rupture the slip is calculated via the equation of motion while the shear stress is controlled by frictional sliding. Such rupture models generally lead to nonlinear, mixed-boundary value problems. The boundary treatment in a numerical method depends in part upon the numerical technique used for the discrete representation of the material. In the present paper we examine two numerical methods. In the first method we use the technique of Day (1982) in which all the components of velocity and stress on the fault are calculated in the same discrete point. In this method, the two sides of the fault share common faces on the fault plane, via split nodes, so that one side of the fault can move relative to the other side; we refer to it as "split model". In the second method we use the staggered-grid velocity-stress FDM formulation, in which the respective components of velocity and stress are calculated on different discrete points. With this formulation the fault is represented with one grid-step (dx) thickness, a method we call the "fault zone" model. For the latter we follow the method proposed by Madariaga et al (1998). To compare these two methods, we solve theoretical dynamic rupture problems of a fault in a homogeneous medium, with the sliding process governed by the slip-weakening friction law. We find that the rupture propagation velocity in the conventional version of the fault-zone model, i.e. with uniform dx, is lower than that in the split model. This appears to be due to the blunting of the stress concentration on the rupture front in the case of the fault-zone model. The rate of stress drop on the rupture zone is faster for the split model, and consequently, the split model tends to accumulate high stress at early time on the rupture front. On the contrary, the fault zone model delays the concentration of stress on the rupture front because the stress drops slowly on the ruptured area. This difference of rupture process suggests that the two methods actually are modeling different fault conditions as a consequence of the different characteristics of the numerical techniques. The split model simulates direct contact between the two sides of the fault that interact through the frictional traction. In contrast, the fault zone model is not simulating direct contact, because the volume of the fault zone plays a role in the dynamics. To obtain further insight into this difference in behavior, we modify the fault zone model by permitting the fault-zone cell dimensions to differ from the rest of the grid. When the fault-zone cell dimension, df, is reduced from dx to dx/2, the rupture velocity in the fault zone model approaches that of the split model. Thus, the behavior of the fault zone model depends upon the grid size. Now the question remains, which method provides a better representation of a real fault? Due to the complexity of real fault zones, either method might be appropriate, depending upon, for example, the amount of distributed shear assumed to occur during rupture. However, it should be kept in mind that the behavior of the fault zone model depends upon what is ordinarily considered to be a purely numerical parameter, dx, and may only approach the behavior of the frictional contact problem when this parameter is considerably smaller than that required for an accurate solution by the split method.
S41A-0945 0800h
Spontaneous Rupture Processes on a Bending Fault
We simulated spontaneous rupture processes on vertical bending faults, using a 3-D finite-difference method. Since shear and normal stresses on the fault depend upon its angle to the principal stresses, rupture velocity and slip ahead of a bending point vary with strike change. Moreover, slip on a bending fault is less than one on a flat fault, since a bending point acts as a partial barrier to slip. In our 3-D finite-difference model, we modeled fault strike change by mapping a rectangular grid onto a parallelogram grid (Inoue, 1996). The grid interval along the fault should be uniform for calculations of spontaneous rupture processes. A hypocenter is located on the center of the optimum striking fault. Varying the bending angle, we calculate dynamic rupture processes under uniform principal stresses. When fault strike changes to promote compressional normal stress across the fault (_grestraining bend_h), rupture velocity ahead of a bending point decelerates. On the other hand, rupture velocity accelerates when a fault strike change promotes extensional normal stress (_greleasing bend_h). Rupture velocity near a free surface is especially sensitive to the shear and normal stress changes accompanying bending. The shear and normal stress variations also produce variations in the slip distribution. Slip on a fault with a releasing bend is larger than one on a fault with a restraining bend, since stress drop is higher on the areas experiencing enhanced extensional stresses than on those experiencing enhanced compressional stresses. On a flat fault, however, slip is higher than for either bending fault case. The obstruction caused by a bend lessens slip on the entire fault.
S41A-0946 0800h
Dynamic Source Rupture Simulation of Dipping Faults With a 3D Finite-Difference Method
The finite-difference method (FDM) has been widely used for numerical modeling of seismic source problems, including investigation on the dynamic source processes. Owing to both conceptual and computational constraints of FDM, fault models have largely been limited to the cases that the fault planes are parallel to the FDM grid. However, recent observation and kinematic inversion results discover that more complex fault geometry models, such as bending faults or curved faults, are needed to explain some earthquake phenomena. Thus, we need to develop an approach of FDM to treat a fault planes slanted with respect to the FDM grid. In this study, we propose a method to analyze the dynamic source problems of nonvertical faults, using a 3D FDM with nonuniform grid spacing (Pitarka, 1999). This approach does not require aligning the fault plane to the FDM grid for implementation of FDM. We estimate the shear stress on the nonvertical fault plane from the six stress components obtained in FDM calculation with respect to the force balance condition and the coordination transformation. This method can be used to deal with a more realistically complex fault geometry model. We validate our method by studying two cases of the dynamic source problems which have been analyzed by Madariaga et al. (1998). One is the instantaneous rupture model of a circular fault embedded in a homogeneous elastic medium; another is the spontaneous rupture model of a rectangular fault which starts from a local circular asperity on the fault plane. We analyze the inclined fault models against the space grid coordination for both of the rupture problems and compare our simulations with previous results obtained by Madariaga et al. (1998) using the horizontal fault plane model. Our simulations gave similar results with those of Madariaga et al. (1998). Thus, our method can be used to analyze the dynamic rupture processes of dipping fault models. This implementation was used to compute the dynamic source problems of the 1999 Chi-Chi, Taiwan, earthquake by Zhang et al. (2003, 2004). We found that the rupture process of this event is more complex than that described in the kinematic model. Our dynamic model revealed that for a large earthquake such as the Chi-Chi earthquake, the rupture propagation can be discontinuous, as suggested by some numerical simulations (Das and Aki, 1977; Day, 1982). In this study, we apply the proposed method to analyze the dynamics of the 2003 Tokachi-Oki, Japan, earthquake. The fault model of this earthquake is a dipping fault with a dip angle of 18 degree. We rebuild the dynamic rupture process of this event and simulate the near source ground motions based on the dynamic source model.
http://sms.dpri.kyoto-u.ac.jp/wenbo/
S41A-0947 0800h
Finite Element Modeling of Dynamic Shear Rupture Experiments Along Non-Planar Faults
The study of dynamically propagating shear cracks along weak paths like faults is of great interest for the study of earthquakes. We adapted the ABAQUS/Explicit dynamic finite element program to analyze the nucleation and propagation of shear cracks along a non-planar, kinked, weak path corresponding to the one that was used in recent laboratory fracture studies by Rousseau and Rosakis [{\it JGR}, 2003]. Their experiments involved impact loading of thin plates of Homalite-100, a photoelastically sensitive brittle polymer, which had been cut along a kinked path and then weakly glued back together everywhere except along a starter notch near the impact site. Under different conditions, propagation speeds were observed in both the sub-Rayleigh and intersonic (supershear) regimes. Strain gage recordings and high speed photography of isochromatic lines (lines of constant difference between the in-plane principal strains) provided characterization of the transient deformation fields associated with the impact and fracture propagation. For the finite element analyses, we implemented a slip-weakening failure model through an option in the ABAQUS program allowing user defined constitutive relations. The analyses of impact loading and of rupture nucleation and propagation were then carried out in the 2D framework of plane stress. In a first set of studies of nucleation and propagation of rupture along a straight fault, we determined after some trial and error an appropriate CFL number, and examined different element types and layouts, finding that the most acceptable results were obtained for low order elements. We used constant strain triangles, arrayed in groups of four to effectively form four-sided elements with corner nodes and one internal node. The studies also showed that to obtain representations of slip rate and shear stress near the propagating rupture tip that were relatively free from numerical oscillations, it was necessary to have element side lengths of order $R_o/50$, where $R_o$ is the estimated slip weakening zone size under quasistatic conditions. We then turned to analyses that explicitly represented the impact loading (as an imposed motion at the contact boundary) and kinked weak path of the experiments. We found that depending on parameter range we could, as in the experiments, produce either sub-Rayleigh or intersonic propagation speeds, and that rupture followed the kinked path. Also, while not obtaining extremely close agreement with the high-speed experimental measurements, we found that we could produce the principal features observed in the dynamic isochromatic line patterns and strain gage recordings.
http://esag.harvard.edu/rice/BaudetTemplBhatRice_30Aug04.pdf
S41A-0948 0800h
Static Stress Field on a Branched Fault System: The 1891 Nobi, Japan, Earthquake (M8.0)
It has often been observed that earthquake rupture propagates along pre-existing faults with complex geometry such as fault bending, offsets, and branching. One of the remarkable examples is the 1891 Nobi earthquake (M8.0) in central Honshu, Japan, which appears to have ruptured branched faults (Muramatu, 1963, Res. Rep. Gifu Univ.; Mikumo and Ando, 1976, J. Phys. Earth), although one of the branches did not appear on the ground surface, and hence there has been much debate about its existence. In the present study, we investigate the possibility of dynamic rupture propagated on a branched fault system during the Nobi earthquake, by considering both static slip and dynamic rupture along the pre-existing fault traces. The surface fault breaks with unusually large displacements have been traced extending over 80 km, (Matsuda, 1974, Spec. Rep. Earthq. Res. Inst.), and the traces appear to consist of 4 segments. In addition, the possible existence of a burried fault, which branched off at near point (c) extending southwards through Gifu-Ichinomiya (e) possibly down to near Nagoya, has been suggested based on various observations (e.g. Muramatu, 1963). Recent stratigraphic surveys of pumice- and gravel-beds from many bore-hole records have revealed shallow underground vertical offsets at a depth of about 25-30 m along a line located 1.5 km east of the presumed location (Sugisaki and Shibata, 2004, Zisin). We include this burried branch into our fault model. We calculate the probable range of static stress drop on five fault segments using the horizontal fault displacements on the surface based on Matsuda's survey (1974). For the buried fault (c-e), we estimate its possible displacement during the 1891 event, by applying the triple junction kinematics proposed by Andrews (1989, JGR), which require that the ratio of slip to the sine of the opposite angle is the same for all three segments. The branch angle of 44$^\circ$ is wide enough for the rupture to propagate on two branched faults (Aochi et al., 2000, GRL). Since slip on the b-c segment drops down to about 2 m from 6.5 m (Matsuda, 1974) near the possible junction (c), slip on the c-e segment for the buried fault would be of the order of 1.5 - 1.8 m. The static stress drop was calculated at the center of each fault segment, referring to Chinnery (1969). We also estimate the principal stresses and their directions working in this fault zone from two-dimensional analysis. If we consider the stress change before and after the earthquake rupture for the two fault segments (b-c) and (c-d), numerical calculations show that the maximum principal stress direction lies between 105$^\circ$ and 109$^\circ$ and the stress change ranges between 7.7 and 12.4MPa, respectively. The direction of the maximum compressive stress is in good agreement with the direction of tectonic stress over this region, which has been inferred from recent GPS observations, but slightly deviated from the direction estimated from hydraulic fracturing stress measurements (Ikeda et al., 2002), which might include the co-seismic and post-seismic stress changes due to the earthquake. Based on the static parameters estimated above, we calculate spontaneous dynamic rupture propagation under various conditions (Fukuyama and Mikumo, 2004, AGU Fall Meeting).
S41A-0949 0800h
Multi-scale dynamic rupture simulation on fractal patch model
We carried out multi-scale full-dynamic rupture simulations, using our new calculation scheme (Aochi and Ide, GRL, 2004) and a fractal patch model as an approximation of realistic heterogeneity. A basic assumption of this model is that a local slip weakening distance (or fracture energy) at a point is proportional to the size of the minimum asperity which includes that point. Since typical topography of fault surface obeys self-affine fractal statistics, we assumed that the asperity distribution is also represented by a power law. For simplicity we prepared seven different sizes of circular patches as discretized representation of asperities. When the patch radius increases by two, the number of patches decreases by four, where the fractal dimension is 2. The whole model space is a fault plane of 4096x4096 square grids, on which the circular patches are distributed randomly. This space is represented by four 64x64 subspaces on different scales and each subspace is connected to the subspaces on the larger and/or smaller scales by renormalization. The assumed values of initial, yield, and residual stresses are homogeneous across the fault plane. We begin each dynamic rupture simulation with breaking one of the patches of the minimum level. In most cases, the rupture stops immediately after the initiation. Sometimes, the rupture coalesces with adjacent patches, propagates into a patch of next level. Frequency-size distribution of these events is approximated by a power law, which is explained by the probability of interaction between asperities. The probability of triggering of dense patch distribution is high and resultant slope of the power law is less steep. Whole rupture process is spontaneous based on exact elasto-dynamics and slip-weakening law except for the nucleation in the minimum level. Thus we observed very heterogeneous process during the rupture: Rupture directivity, rupture front shape, slip distribution, and moment release functions. Some moment rate functions increase irregularly, which resemble to so-called initial phases observed in real seismic waves. We cannot distinguish small and large events from the initial rise of moment rate functions.
S41A-0950 0800h
Dynamic fault rupture constraints to high frequency radiation of crustal earthquakes: the role of rupture velocity and fmax
The study of high frequency (HF) radiation of large earthquakes have been traditionally investigated by using kinematic models of the source. Some of these studies locate the HF radiation near boundaries of large slip regions (Zeng et. al. 1993, Kakehi et. al. 1996, 1997; Nakahara 1999, 2002), while others locate the HF radiation overlapping regions of large slip (Hartzell et. al 1996). However, a major limitation of all these studies is the over-simplification of the physical parameters involved in the rupture process such as the assumption of a nearly constant rupture velocity across the fault plane. Simple dynamic crack models have theoretically demonstrated that local variations of the rupture velocity play a very important role in the radiation of high frequency from the source (Madariaga 1977, 1983). In the present study we investigate the high frequency radiation of the 2000 Tottori earthquake (Japan) in two steps: First we investigate the complexity in the fault rupture by performing a spontaneous rupture dynamic model of the Tottori earthquake in the low frequency range (.1 to 1Hz). The fault friction law parameters and stress drop of the dynamic model are constrained from results of a kinematic model of source. On the other hand the rupture velocity is allowed to vary spontaneously. In the second step we calculate the high frequencies from a semi-stochastic approach that considers the radiation from a heterogenous finite fault and a frequency-dependent subfault-site specific radiation pattern model (Pulido et. al. 2004). The forward calculation of the high-frequency ground motion (1 to 20Hz) at the target observation sites is constrained by the subfault rupture times from the above dynamic model. In order to optimise the agreement to observed high frequency ground motion we use a Genetic Algorithm approach to invert for the stress drop distribution, fmax, and the high frequency decay for frequencies above fmax, by comparing the observed and simulated RMS acceleration envelopes, as well as the acceleration Fourier spectra of the waveforms. In order to effectively constraint the HF inversion, we first investigated Q and the site effects at all the KiK-Net borehole stations, by applying a spectral inversion technique (Moya et. al 2003) to 55 aftershocks of the Tottori earthquake. Preliminary results show that the high frequency radiation from the source is determined by a complex relationship between the rupture velocity, stress drop and fmax. The results are very sensitive to the dynamic model obtained. Further investigations should address the non-uniqueness of a dynamic rupture model.
S41A-0951 0800h
Rate- and State Friction Behaviour of Simulated Fault Rocks: Influence of Phyllosilicate and Pressure Solution
Rate and State Friction (RSF) laws are widely used to describe experimental rock friction results and to model seismogenic processes. The empirical RSF parameters a and b as determined in rock friction experiments give realistic results when used in models for the seismogenic cycle. However, the critical slip distance parameter, D$_{c}$, has to be arbitrarily scaled up from laboratory values of 10 $\mu$m to around 1 m. Recent evidence indicates that fault healing by solution-transfer processes may contribute to the discrepancy between laboratory and natural fault behaviour. Such processes are known to be strongly influenced by phyllosilicates. To examine the effects of phyllosilicates, we report results from slide-hold-slide experiments on wet, simulated, phyllosilicate-bearing faults (salt-muscovite gouge mixtures), performed under conditions where pressure solution is active. At low (\leq 1 $\mu$m/s) sliding velocities, results show that shearing of the gouge involves slip on the foliation, accomodated by pressure solution of the intervening halite grains. This leads to velocity-strengthening behaviour and a mylonitic microstructure. At high sliding velocities (\geq 1 $\mu$m/s) pressure solution is too slow to accommodate all the imposed shear and the foliation is destroyed, resulting in a microstructure that resembles that of a cataclasite. In this regime, the gouge shows strong velocity-weakening behaviour. Control experiments performed on pure halite, pure muscovite and dry halite-muscovite mixtures showed little or no velocity-weakening effect. Significantly, slide-hold-slide on wet halite-muscovite mixtures show a major increase in healing rate with increasing sliding velocity in the velocity-weakening field. The maximum healing rate measured at 10 $\mu$m/s is an order of magnitude higher than healing rates from room-dry experiments, and the displacement required to establish new steady state sliding (D$_{c}$) increases with increasing hold time and increasing sliding velocity, reaching a value of ~ 400 $\mu$m. Taken together, our results suggest that the combined effects of phyllosilicates and pressure solution may provide an explanation for both stable creep and unstable seismic slip on mature phyllosilicate-bearing faults. A physical model to scale the RSF empirical parameters to natural conditions should take the operation of pressure solution and the effects of phyllosilicates into account.
S41A-0952 0800h
Earthquake Dynamics at Linked Fault Stepovers
Previous studies of strike-slip fault systems with linking dip-slip faults (Oglesby, 2003) have indicated that with equivalent shear and normal stresses on all segments, extensional stepovers with linking normal faults are more likely to produce multi-segment events that rupture the entire fault system, compared to compressive stepovers with linking thrust faults. This difference is due to the sign of the normal stress increment that the strike-slip faults radiate onto the linking dip-slip fault: in the extensional stepover, this stress increment is also extensional, unlocking the linking fault, and aiding rupture. In the compressional stepover, the normal stress increment is compressional, tending to lock the linking fault. In the current study, we use simplified assumptions about the scaling of stress in the interseismic period to investigate the behavior of fault stepover systems with more realistic pre-stress patterns. We assume that the linking fault is a passive feature that is loaded by sequential slip of the neighboring strike-slip faults. We find that such an assumption tends to accentuate the difference between the extensional and compressional fault systems. Under simple initial conditions and assumptions, extensional stepovers produce a larger number of events that rupture through the linking fault and rupture the whole system. Compressional stepovers have a tendency to lock up both during rupture and during the interseismic period, decreasing their ability to produce multi-segment events. The results may have implications for seismic hazard near these fault systems.
S41A-0953 0800h
Numerical Simulations of 2D In-Plane Ruptures in a Multi-Fault Tri-Material System.
Earthquake faults with large slip are likely to bring into contact rocks with different elastic properties. The material near a geological fault is typically observed to be shattered with increased porosity and fluid content, leading to a zone of low seismic velocities, often referred to as fault-zone, between the bounding crustal blocks. Previous work has has shown that ruptures along a material interface have remarkable dynamic properties which are relevant to a number of geophysical and engineering problems. In the present work, we attempt to clarify the conditions for which ruptures that start in the bulk migrate on their own to material interfaces, and evolving properties of such ruptures. We employ a generalized version of the second-order finite-difference code used by Andrews (1973) and Andrews & Ben-Zion (1997) to perform a numerical parameter-space study of two-dimensional in-plane ruptures in a multi-fault tri-material system. We show that material interfaces are favoured locations for rupture propagation and we examine tendencies of initiated ruptures to migrate spontaneously to material interfaces. Ruptures in our work are nucleated by a symmetric bilateral expanding pore pressure source, and may then continue to propagate (or not) along one or several faults. The faults, two of which are material interfaces, are situated equidistant and parallel to each other. Using different nucleation locations, different initial stress, different velocity contrasts, different frictional fault separations, different widths of a low velocity zone, and different number of faults, we examine the range of conditions for which ruptures migrate spontaneously to material interfaces and continue to propagate in a self- sustaining manner. We show simulation results with faults governed by pure Coulomb friction and faults governed by Prakash-Clifton friction, and discuss similarities and differences between the different cases. We also show results of faults governed by Columb friction surrounded by a viscoelastic media.
S41A-0954 0800h
Quasi-static Modeling of Non-planar Crack Growth With/without Open Kinks
Quasi-static non-planar growth of a crack is simulated by using the maximum energy release rate criterion. When a crack is under compression, the evaluation of energy release rate involves the solution to (1) frictional contact problems, where the fractured surfaces are constrained by a friction law, and/or (2) crack opening problems, where the fractured surfaces are constrained by traction free condition. We adopt the boundary integral equation method to solve such mixed mode (mode I and II) non-planar crack problems: the tensile traction on the crack is iteratively eliminated by allowing the opening dislocation, and the friction law is satisfied by using the re-calculated normal traction at each iteration. We assume a slip-weakening Coulomb friction law to eliminate infinite tensile stresses in the vicinity of crack tips and it enables us to investigate applied load conditions for non-planar surface to close over the crack. Our preliminary result shows that when the peak stress of the slip weakening law is high (like classical singular crack models), an artificial cut under uniaxial compression extends with open kinks, which is consistent with laboratory observations.
S41A-0955 0800h
Seismic Wave Radiation of Rupture in Branched Fault Modeling
We simulate dynamic mode II rupture that finally forms a branched fault trace, where our main interest lies in the resultant seismic wave radiation due to the dynamic branching process. For this purpose we adopt the elastodynamic boundary integral equation method, which enables us to work with non-planar fault geometry. We consider a medium under biaxial compressional load in which Coulomb friction acts on rupture surfaces and apply a critical shear stress criterion in determining the direction and the extension of rupture tips . In our simulation, dynamic branching first occurs due to local off-plane stressing around the fast propagating tips and each branch increases its bending angle. The growth of branch is then arrested because the stress to be released on such branch becomes negative under biaxial compression.In order to find phases associated with the dynamic branching process, we synthesize waveforms of the branching model and compare them with those radiated from a planar fault model whose growth is arrested without branching. When the observation point is very near the branching point, we can successfully identify a small distinct branching phase in the component for which no wave radiation is expected from the planar model. Otherwise we cannot find little effect of the dynamic branching process on the seismic wave radiation. This is because the moment release rate on the branching part of the fault is negligible compared with that on the entire fault.
S41A-0956 0800h
Multi-parameter Dynamic Rupture Inversion of the Western Tottori Earthquake, Japan
We have used a non-linear, dynamic inversion process developed by Peyrat and Olsen (2004) to test the possibility of making simultaneous estimates of stress drop and slip-weakening distance on a fault. The forward problem is solved for spontaneous rupture using a three-dimensional, fourth-order staggered-grid finite difference method. The resulting computed seismograms are compared with observed seismic data, and a neighborhood algorithm is used to invert for the initial conditions that yield the lowest misfits. Here we apply the method to the magnitude 6.6 Western Tottori, Japan earthquake of October 2000. Data from 12 strong motion stations, bandpass filtered between 0.05 Hz and 0.5 Hz, are used to calibrate the misfit of the computed models. We discretize the fault surface into 18 regions and invert for stress drop and slip-weakening distance in each region simultaneously. The stress drop and slip-weakening distance are constrained to -2 to 5 MPa and 10 to 130 cm, respectively. We find that the best-fit initial stress models are similar, regardless of friction, and that slip-weakening distances of 10 cm to 60 cm are preferred over higher values. These results are in agreement with the single-parameter inversions done by Peyrat and Olsen (2004) and with an estimated slip-weakening distance of 28 cm determined independently from strong motion data for the Western Tottori earthquake. The results show some promise that low-frequency seismic data may be used to estimate stress drop and friction parameters independently for large earthquakes.
S41A-0957 0800h
Source Images of the 1997 Kagoshima Earthquakes With High Resolution Using the Grid Representation and 3D Heterogeneous Green Functions
The seismic wavefield is strongly affected by the underground structure through which the seismic waves propagate from source to observation stations. However, in most of conventional source inversions, horizontally layered media (one-dimensional structure) have been used for calculation of the Green functions. The purpose of this study is to obtain the high accuracy and high resolution images of source processes using the 3D Green functions which include more realistic wave propagation effects. We investigate two mainshocks of the 1997 Northwestern Kagoshima, Japan, earthquakes (the first mainshock:$M$6.5, the second mainshock:$M$6.3), which are well recorded at near-source strong motion stations of the K-NET. We have developed a parallel computation code of the finite-difference method (FDM) to calculate the 3D Green functions using the reciprocity theorem, which effectively reduces the the number of computation times. We use a realistic 3D structure model for the northwestern Kagoshima region constructed by Fujii et al. (2004) for the FDM calculations of the 3D Green functions. A new inversion method (Takenaka and Fujii, 2003 AGU Fall Meeting) by a grid model approach is applied to the 0.1-1Hz bandpass filtered displacement waveforms. As the results, it is found that for the first mainshock, the rupture began from the hypocenter and propagated mainly toward the shallower western part and the deeper eastern part. The maximum amplitude of slip velocity is more than about 4km/s. The locations of asperities are extremely consistent with the areas where the aftershocks are not abundant. The comparison of the inversion result with the velocity perturbation of the seismic tomography (Miyamachi et al., 1999) indicates that the rupture propagated to the high velocity zone and stopped there. Also for the second mainshock, the slip distribution was obtained with high accuracy and high resolution, which considerably overlaps on the low activity zone of the aftershocks. The inversion results inferred using the conventional 1D Green functions can not explain the source aspects as described above, which suggests that the source inversion using 3D Green functions has a potential to effectively reveal the source process with high accuracy and high resolution.
S41A-0958 0800h
RADIATED SEISMIC ENERGY DETERMINED FROM A SPONTANEOUS RUPTURE MODEL OF THE 1994 NORTHRIDGE EARTHQUAKE
The radiated seismic energy of the 1994 Northridge earthquake (M 6.7) is calculated by simulating a spontaneous rupture model of the earthquake in a layered velocity structure. Using a static finite element method the stress drop distribution is determined from the inverted slip distribution (Liu and Archuleta, 2000). From the stress drop distribution we derive a spatially heterogeneous initial stress and yield stress. In concert with a slip-weakening friction law we dynamically rupture the fault using a 3D finite-element method. By using trial and error we modified both the initial stress field and yield stress until dynamic rupture generated a rupture history and final slip distribution that approximately matched those determined by kinematic inversion. The resulting dynamic model provides a reasonably good fit between synthetics and near-field strong motion data. The total radiated seismic energy calculated from our dynamic model is 7.89 x 1014 J, which is close to the Gutenberg-Richter estimate of 7.08 x 1014 J. By displaying the energy distribution on a 40 km radius hemisphere enclosing the source, strong directivity effects are evident: 35.83 percent of the seismic energy passes through 8.92 percent of the whole hemisphere area, most of which is concentrated in the forward direction of the rupture. Energy passing through the bottom of the sphere, i.e., the energy radiated to teleseismic distances, is very small.
S41A-0959 0800h
Representations of the Radiated Energy in Earthquakes
We investigate the representation of the radiated energy, $E_R$, in earthquakes. The radiated energy, $E_R$, is estimated in seismology from either far-field seismic waves or the stress and displacement on the fault plane. Although $E_R$ comes from the entire volume of the Earth, it can be expressed as an integral over the fault plane. However, the integrand cannot be given a simple physical meaning such as the radiated energy density on the fault plane. The stress on the fault plane changes rapidly during a seismic rupture. Although the energy radiated by this process is not included in the estimate of $E_R$ in a simplified practice in seismology, it is correctly included in the expression of $E_R$ in the standard seismological practice. Using the representation theorem, we can express $E_R$ as a surface integral over the fault plane, with the integrand containing the slip function on the fault plane. However, the integrand at a point depends on not only the slip function at the point, but also the slip functions everywhere on the fault plane. Thus, the simple method in which $E_R$ is estimated by summation of the local energy flux on the fault plane does not yield a correct estimate. In view of occasional confusions on these issues in the literature, we address in this paper 1) the physical meaning of the surface integral in the expression of $E_R$, 2) the effect of rapid changes in stress on the fault plane on the seismological estimate of $E_R$, and 3) the difficulty in estimating $E_R$ using a local energy flux on the fault plane.
S41A-0960 0800h
Dynamic Source Parameters of Characterized Source Model for Strong Ground Motion Prediction
A characterized source model (Irikura and Miyake, 2001; Miyake et al., 2003) for strong ground motion prediction is constructed with the heterogeneous source slip distribution results of the kinematic source inversion. For ground motion simulation, high-frequency radiation level should be constrained. Irikura et al. (2004) give stress drop on the asperity from an idealized slip distribution of the asperity model by Das and Kostrov (1986). Dan et al. (2001) directly constrained that level from the empirical acceleration scaling relation. Here we are trying to obtain stress parameters directly form the kinematic inversion results. We obtained stress parameters by mapping method of spatio-temporal shear-stress distribution on the fault plane from a spatio-temporal slip distribution. They estimated dynamic source parameters averaged over on-asperity and background area from a viewpoint of slip-characterized source model (Iwata et al., 2003; 2004). Until now, they are analyzing seven crustal events and found that (1) average effective stress values, that spans 20-30MPa, of on-asperity area are about 30% larger than the static stress drop values, (2) There is slight depth dependence of the averaged effective stress of on-asperity area, (3) Average effective stress drop of surface ruptured asperity is smaller that of buried asperity, and (4) Average effective stress on the background area is about 5MPa. Those obtained stress parameters can be used for the characterized source model for the broad-band ground motion simulation.
S41A-0961 0800h
Extreme Slip Heterogeneity and Near-Fault Ground Motions
Field observations indicate extreme earthquake slip heterogeneity, that is, several meters of variation within tens of meters along-strike. Up to several parts in ten of strain can occur within a matter of seconds. Such variation in slip implies that, similarly, highly variable stress and strain occur within a fault zone. Field geologists' observations tend to emphasize the coherent several-kilometer wavelength slip signal, and typically treat the short-wavelength slip variations as noise. Observations of variation in slip, however, can be important, even if they are not structurally or tectonically significant. We interpret these data to indicate that, at least during the brief time when a rupture is propagating along a fault, the fault reaches a condition in which it must be both extremely strong and very weak, in rapid succession. Two possible explanations for this are considered; 1) geometrical irregularities in the fault zone, and 2) sharp momentary decrease in compressional lithostatic stress normal to the fault zone. In the first case, the fault appears weak when slip occurs on thin planes with low dynamic sliding friction, but strong when the rupture encounters geometric irregularities. Alternatively, while the fault zone is experiencing a transitory virtual decompression between the wall rock on the sides of the fault, friction drops to a low sliding value. Immediately upon passage of the rupture front, however,the decompression would abruptly halt, locking the extreme slip and stress heterogeneity into the fault. Highly heterogeneous slip also implies that as the rupture front propagates along the fault, slipping patches accelerate and decelerate on a scale of a few meters. The energy radiation patterns emanating from such a rough rupture process would therefore contain much high-frequency noise. When the friction drops to a low sliding value, however, the source of high-frequency energy noise radiation would cease. This may occur in patches where pre-existing stress is relatively homogeneous over a large area. Upon encountering such a smooth patch, rupture propagation could momentarily reach super-shear velocity in this near-frictionless condition, only to be strongly decelerated once slip rate again drops and friction rises abruptly. Such a model appears to be consistent with the large near-field ground motion recorded at Pump Station #10 (PS10) from 2002 Denali fault earthquake, in which we interpret that high-frequency radiation decreased during a brief episode of supershear conditions. This model is also consistent with the relative lack of pseudotachylites and heat flow anomalies along major faults.