S44A-01
Rupture Propagation and Slip Partitioning on an Oblique Upward-Branching Fault System
We use dynamic 3-D finite element analysis to investigate slip partitioning and rupture propagation on an oblique left-lateral/normal fault system that branches upwards near the surface into vertical and non-vertical branches. The model consists of a 70 degree dipping oblique-slip fault that extends from a depth of 15km to 5km depth and then branches upwards into a vertical segment and a segment dipping 45 degrees. The use of a simple regional stress field consistent with the overall kinematics of the system results in rupture propagation only on the base and vertical faults. However, the addition of a 2km by 3km barrier onto the bottom portion of the vertical fault causes enough of a stress perturbation on the upper dipping fault to nucleate rupture on this segment, resulting in a strongly partitioned slip distribution in the system. In all cases, strike-slip motion is concentrated on the vertical fault, and dip-slip motion is concentrated on the dipping fault. Although the strength, size, and location of the barrier do affect the details of the slip history, the overall observation of slip partitioning is not sensitive to these details. Other observations in our models show that as the dipping fault slips, it induces a small amount of backwards slip on the vertical fault due to the high stress drop in our models and the close proximity of the two branch segments. These results may have important implications for the dynamics of branched faults and geometrically complex fault systems in general.
S44A-02
Fault Bending Promotes Postseismic Fault Growth
Yamashita(2007) and Yamashita and Suzuki(2008) theoretically studied the postseismic fault growth assuming a fault in a poroelastic medium and taking account of fluid flow. They specifically assumed a 2-D in- plane fault on a bimaterial interface. Contrasts only in the drained and undrained Poissonfs ratios and in the Biot-Willis coefficient are assumed in the former, while the diffusivity contrast is assumed in the latter. The existence of fault on a bimaterial interface produces mechanical coupling between the fluid pressure change and deformation of medium. The above two papers showed that this coupling can promote postseismic fault growth. The two papers also pointed out that the diffusivity contrast is more effective for the postseismic fault growth. However, a weakness in their approaches is that a planar fault was assumed. In fact, it is increasingly clear that surface traces of large earthquakes show some typical geometrical structures such as bends, branches and steps. A non-planar fault can produce coupling between the fluid pressure change and deformation of medium even if the fault is located in a homogeneous medium; such coupling may promote postseismic fault growth. Here we theoretically study the postseismic quasistatic deformation near a 2-D in- plane non-planar fault in a homogeneous poroelastic medium and examine the possibility of postseismic fault growth. The fault is assumed to have an abrupt bend near one of the tips. The main segment of fault is parallel to the direction of maximum shear stress. We calculate some quantities relevant to postseismic fault growth such as Coulomb stress change at the tip of the bend; we find that the ratio of Coulomb stress change is significantly dependent on the angle of bending. The ratio of Coulomb stress change is defined as the Coulomb stress change divided by its value taken at t=+0. Our calculation shows that the increase in the ratio of Coulomb stress change is much larger for the non-planar fault than for a planar fault. The difference between the two cases is found to be larger for a smaller value of the Biot-Willis coefficient. It was theoretically shown by Kame and Yamashita [1999] that the spontaneous fault tip bending arrests the dynamic fault tip growth. Our present study therefore suggests that widely observed postseismic fault growth is a natural consequence of this dynamic fault tip bending.
S44A-03
Self-similarity of the largest-scale segmentation of the faults; implications on earthquake dynamics
It is long known that long-term faults are segmented at various scales, and that earthquakes are sensitive to that segmentation, especially in their initiation, propagation and arrest. Understanding how and why faults are segmented is thus a major issue. We address that question through the analysis of more than 700 cumulative slip profiles measured on a population of active normal faults (Afar, East Africa) spanning a broad range of ages (10-1000 kyr), lengths (0.3-65 km), cumulative slips (1-1300 m), and slip rates (0.1-5 mm/yr). It has indeed been shown that cumulative slip profiles hold the signature of the fault segmentation, especially the largest one, in the form of pronounced slip troughs and bumps coinciding with inter-segments and segments, respectively. The population of slip profiles that we use is the largest ever analyzed so far. We focus on the largest-scale segmentation, that is the segments having a length on the same order than that of the entire fault they belong to. Our goal is to identify the major bumps in the slip profiles and determine their number and length. We use two methods. In the first, simpler one, the bumps and troughs are defined in respect with the overall envelop shape of the slip profiles. We thus first determine that envelop shape, that reveals to be triangular in 70% of the cases (as suggested by Manighetti et al., 2001 in a prior analysis of a smaller part of our slip profile population). We then remove the best-fitting triangle (or ellipse for the other cases) from the slip data, so that to obtain a residual slip function that highlights the zones where slip markedly diverges from the overall profile slip function. Assuming that the larger bumps have a similar length, we search for the sinusoidal function that best fits the residual slip data, and derive, from the periodicity of that function, the number of similar length major bumps, hence major segments. We find that 75% of the faults are segmented into 2 to 5 major segments. More specifically, 70% of the faults are made of 2 to 4 major segments. The trace of the segments in the slip profiles becomes less clear as the fault slip to length ratio increases. As a consequence, the number of identified segments decreases with the slip to length ratio of the fault. Most of the 25% of the faults that are found not to be segmented have an elliptical slip profile and/or a high slip-length ratio. Because major segments along a fault may not be of similar length, we then use a more sophisticated method, the S-Transform (Stockwell et al., 1996), to explore the slip profiles in the frequency-space domain. The preliminary results confirm that the large majority of faults are segmented into 2-4 major segments. The distribution of the low frequency ST amplitudes as a function of fault length allows identifying and locating the major segments, while estimating their length. It seems that, along most faults, the segments have a similar length. In more rare cases, the fault contains one segment twice longer than the others. We conclude that the large majority of faults (70% at least) are segmented into 2-4, more or less connected, major segments. That number appears independent on the fault length, slip, age, and slip rate. The largest- scale segmentation of the faults is thus self-similar and likely controlled by the fault growth mechanics. There is no doubt that such a property of the fault segmentation has an impact on the earthquake process.
S44A-04
Tangled Dynamic Rupture Propagation on Laboratory Faults
We illustrate a series of novel observations obtained in laboratory experiments with a photographic method
coupled to acoustic emission monitoring. Two precut samples were put in contact on their edge, under
uniaxial load of 1-13 kN, at an angle close to instability in order to ensure the spontaneous triggering of
fracture. A high speed digital camera with inter-frame intervals of 10 μsec acquired image sequences of
fracture propagation, while a small array of piezo sensors were recording the acoustic emission in the MHz
range. The entire process is observed, starting from slow fracture initiation, to acceleration and fully dynamic
propagation at sonic and supershear velocity. The silding interface was conditioned in different ways:
lubricating patches were added at isolated points and the surface roughness was altered. In the latter case a
consistent degree of complexity arises in the rupture, including tangled propagation and re-rupturing within
short time intervals. Various radiated waves from the fault rupture are recognized in the photograms and
compared to the monitored signal in the piezograms. This allows to evaluate the relative amplitude of the
shock wave generated after super-shear transition.
http://www.roma1.ingv.it/Members/nielsen/a3
S44A-05
Estimation of Fracture Energy in the Laboratory
To understand the physical process of stress breakdown during the dynamic rupture of earthquakes, we investigated the friction behavior of rocks in the laboratory by measuring the traction evolution in response to a given slip history. We employed a high-speed rotary shear apparatus introduced at NIED, whose basic architecture is similar to the original system by Shimamoto and Tsutsumi [1994], adding the capability of servo-controlled variable sliding velocity. We used a set of two Inada granite cylindrical specimens with a diameter of 25mm. As an input signal, we used the fault parallel component of the PS10 velocity seismogram observed at 3km away from the high slip region during the 2002 Denali, Alaska earthquake [Ellsworth et al., 2004]. Since the earthquake was vertical strike slip with surface breaks, the surface displacement close to the fault trace can be considered as the fault slip itself. In this experiment, keeping the original shape of waveforms (1.4m/s peak velocity and 3s pulse width), we changed the amplitude and pulse width, which corresponds to virtually created earthquakes in different size and different stress drop. We measured the shear stress under the constant normal stress of 0.5MPa under room humidity condition. This rather low normal stress is to prevent from melting during the sliding. From the stress-slip curves, we measured the slip- weakening distance (Dc), then we estimated the seismological fracture energy by integrating the slip- weakening curve until Dc. We compared the seismological fracture energy with the corresponding Dc. Basically the relation is linear, but there is a critical point at Dc = 0.2m, where the gradient changes. When Dc is less than 0.2m the gradient is steep, suggesting that the fracture energy increases rather independent of Dc, and for Dc greater than 0.2m, fracture energy increases as a function of Dc. Another interesting feature is that the maximum value of Dc is about 4m even if the total slip exceeds 12m. When the total slip is less than 5m, Dc is achieved at around the highest slip velocity. But when the total slip is greater than 10m, during the high slip velocity, shear traction starts to recover when the slip exceeded 4m. It is interesting to compare it with the Dc' of 2.5m by Fukuyama and Mikumo [2007, GRL], which is the proxy of Dc. These two critical values might be related to the physical process of the weakening.
S44A-06
Over two orders of magnitude variation are determined solely by slip for a group of small earthquakes on the San Andreas Fault near Parkfield, CA
Ordinarily, earthquake magnitude is controlled by both rupture length and slip variation. Here we show that a special population of earthquakes at the mesoscale has a constant rupture length, but varying slip. We compare the source time function pulse widths of 25 earthquakes on the San Andreas Fault, and 11 earthquakes on surrounding secondary faults to show that the earthquakes on the San Andreas Fault near Parkfield have a constant duration in this group with magnitudes ranging from 1.4 to 3.7. In contrast, earthquakes on secondary faults indicate the more usual source parameter scaling suggestive of a constant stress drop, i.e. they have an increase in duration with magnitude. The earthquakes on the San Andreas Fault are located approximately 20 km to the northeast of the 1966 mainshock epicenter, along the fault, to approximately 5 km south of the 2004 epicenter. Unlike previously studied repeating sequences, the magnitudes are not constant, nor is the repeat time regular. The secondary faults are located at distances of 5 km or greater from the trace of the San Andreas Fault, and are almost certainly not part of the active or historically active plate boundary fault system. The constant source duration observation for the earthquakes on the San Andreas Fault suggests that fault area stays constant over the magnitude range of our data set. A repetitive rupture of a small, locked asperity in a creeping fault can explain the constant duration. The dimension of the asperity could pre-determine the fault area. Therefore the observation directly measures the scale of the heterogeneities on the fault. We observe heterogeneities of 120, and 160 m in diameter. Calculated stress drop values of the earthquake population on the San Andreas Fault range from 0.18 MPa to 63 MPa, and values on secondary faults range from 0.31 MPa to 14 MPa. The differences in duration between the events on the San Andreas Fault and on secondary faults suggest that earthquakes on the San Andreas Fault are inherently different. Cumulative slip values on the secondary faults are negligible in comparison to cumulative slip values on the San Andreas Fault. We speculate that faults with more cumulative displacement have earthquakes which may rupture differently. Furthermore, the differences in source properties between the two populations might be explained by differences in fault surface roughness.
S44A-07
Spatial Interdependency Between Kinematic Source Parameters Derived From Dynamic Rupture Simulations
For a kinematic model to accurately predict ground motion, it is necessary to know not only the spatial distribution of the source parameters but also the spatial interdependency among the parameters. Because there are limitations to the kinematic parameters derived from inversions of seismic data, we determine the spatial interdependency between source parameters from physically based dynamic ruptures. We have computed and analyzed hundreds of dynamic rupture models to get a quantitative understanding of the spatial interdependency and amplitude distributions of parameters describing the earthquake source, such as stress drop, rupture velocity, and rise time. We use a slip weakening friction law combined with different approaches to create random heterogeneous initial stress and strength distributions on the fault as the basic ingredients for our dynamic ruptures. While there is much to be learned by looking at differences among all the models, we focus on features that are common among all models, that is, features that show the least dependence on the choice of the initial model. Using all dynamic ruptures, we are able to construct probability density functions (PDF's) for the amplitude distributions of the source parameters as well as for the spatial correlation between the source parameters. In addition we compute joint probability density functions to determine if there is a linear relationship between different source parameters, and cross spectral densities to determine if there is correlation at all scales. We find: (1) slip amplitude does not show systematic correlations with rupture velocity, and it is positively correlated with rise time; (2) peak slip rate shows strong correlation with rupture velocity and rise time; (3) the PDF of rupture velocity has a well defined maximum between 80%-90% of the shear wave velocity. The value of this maximum probability density increases with distance from the hypocenter, while the width of the PDF decreases, i.e., the rupture velocity tends toward a more constant value farther from the hypocenter. A similar dependence is found for the PDF of the rise times, which has a width that decreases with increasing distance from the nucleation zone; moreover, the mean value of the rise time shifts to a smaller value.
S44A-08
Kinematic and Dynamic Inversions of the 2000 Tottori Earthquake Based on Elliptical Subfault Approximations
We develop an simplified inversion method to do fast non-linear kinematic and dynamic inversion of near field strong motion data at low frequencies above 0.5 Hz. Using a small number of elliptical patches we efficiently reduce the number of independent parameters, and the dimensionality of the inverse problem. For instance, a single elliptical patch has only 7 independent degrees of freedom. We apply this method to the 2000 Tottori Japan earthquake. Real time data recorded by the Kik-net and K-net accelerometer networks were filtered to the 0.1-0.5 frequency range and integrated to displacements. The hypocenter was located at 14 km depth and we use the structure model by Semmane et al., 2005, in order to generate synthetics. We did a non-linear kinematic inversion with the neighborhood algorithm (NA) with an L2 norm that converges very rapidly to a slip distribution with just two elliptical patches. Synthetics fit the strong motion records very well reducing the variance by 71%. Nonetheless the solution model is non-unique and needs to be validated through the dynamic inversion. We propose a new dynamic inversion method based on the same simple geometrical ideas. The dynamic rupture propagation is modeled with a finite difference method on a coarse grid with a slip weakening friction law. Dynamic inversions are intrinsically ill-posed; we can invert for both asperity and barrier models. Here we invert for a simple barrier model for the Tottori earthquake: we look for a geometrically simple distribution of rupture resistance that produces a rupture that has a geometry that is similar with our kinematic inversion. Dynamic inversion is stabilized by the use of the kappa parameter that controls whether rupture occurs and it is sub-shear. We implement dynamic inversion with the same NA algorithm used for kinematic inversion. The inversion converges within a few hours of computation time on a small cluster of only 12 nodes. Synthetics computed for the dynamic inversion fit the observed data reducing the variance by 60%, less well than kinematic models because dynamic inversion has fewer independent parameters than kinematic inversion. By making different assumptions about the rupture process we illustrate the non-unicity of the solution to the dynamic inversion.