S33A-1920
A Virtual California Earthquake Simulation Test
In order to test and compare simulation models we study a sequence of relatively simple fault models using the Virtual California (VC) earthquake simulator. Virtual California (VC) is a geometrically realistic numerical model that is specifically designed to simulate earthquake occurrences along the San Andreas and adjacent faults. It is a stochastic, cellular automata (CA) simulation of an earthquake backslip model. The term backslip implies that the loading of each fault segment occurs due to the accumulation of a slip deficit at the prescribed slip rate of the segment. We simulate a linear, vertical fault divided into 10× 10 km elements. We consider two test problems: Test problem I: A 460 km long fault is divided into two segments with 23 elements (230 km) each. The interaction of the two segments is studied by gradually making one segment stronger by prescribing longer recurrence intervals for this segment. As the strength increases, mode switching occurs between total rupture (T) and the rupture of the weak segment (W). The model first produces only total rupture (T-T-T), then alternating total rupture with rupture of the weak segment (T-W-T- W-T), then progressively more relatively frequent rupture of the weak segment (T-W-W-T-W-W-T, etc.) Complex behavior is found during mode switching. Test problem II: A 470 km long fault is divided into three segments: a long 22 element (220 km) strong segment, a short 3 element (30 km) very strong asperity segment, and a long 22 element (220 km) weak segment. Typical behavior is a total rupture of the fault followed by progressive rupture of the weak segment followed by rupture of the strong segment. This behavior is similar to the earthquake progression seen on the North Anatolian Fault.
S33A-1921
ShakeOut Simulations---Verification
For the past three years we have presented results of several large-scale earthquake simulations run by different research groups, but these have been limited to qualitative comparisons. This work presents a preliminary verification of three simulation sets for the ShakeOut earthquake scenario, version 1.1, an Mw 7.8 earthquake on a portion of the San Andreas fault in southern California. Two of the simulation sets use a finite difference approach while the third uses finite elements. The verification is done quantitatively by means of the goodness-of-fit criteria proposed by Anderson (2004) and the misfit criteria proposed by Kristekova et al. (2006). The results indicate good agreement between the three implementations. The metrics of the comparisons for ten selected locations throughout the region indicate that the agreement can be regarded as excellent according to the scale defined by Anderson. All three groups used a common discrete representation of SCEC's Community Velocity Model (CVM, version 4). Discrepancies observed between the different synthetics are associated to intrinsic differences between the numerical methods used and their implementation by each simulation group. We conclude that the three schemes are consistent, reliable, and sufficiently accurate for future large-scale simulations.
S33A-1922
Next-Day Earthquake Forecasts for California
We implemented a daily forecast of m > 4 earthquakes for California in the format suitable for testing in community-based earthquake predictability experiments: Regional Earthquake Likelihood Models (RELM) and the Collaboratory for the Study of Earthquake Predictability (CSEP). The forecast is based on near-real time earthquake reports from the ANSS catalog above magnitude 2 and will be available online. The model used to generate the forecasts is based on the Epidemic-Type Earthquake Sequence (ETES) model, a stochastic model of clustered and triggered seismicity. Our particular implementation is based on the earlier work of Helmstetter et al. (2006, 2007), but we extended the forecast to all of Cali-fornia, use more data to calibrate the model and its parameters, and made some modifications. Our forecasts will compete against the Short-Term Earthquake Probabilities (STEP) forecasts of Gersten-berger et al. (2005) and other models in the next-day testing class of the CSEP experiment in California. We illustrate our forecasts with examples and discuss preliminary results.
S33A-1923
Stochastic Earthquake Simulator Based on Realistic Stress Interactions and Earthquake Nucleation: Implications for Aftershock Forecasting
The earthquake generation and triggering is a complex process consisting of a large number of unknowns, e.g. the exact fault structure, frictional behavior and prestress conditions. Therefore, deterministic model runs based only on one particular model setup are limited in their explanatory and predictive power. On the other hand, statistical models based purely on empirical relations (such as the well-known ETAS model) ignore important physical knowledge and constraints. To bridge the gap between purely statistical and deterministic models, we developed an earthquake simulator which is based, on the one hand, on realistic elastic half- space stress interactions, rate-and-state dependent frictional earthquake nucleation and extended ruptures with heterogeneous (fractal) slip distribution. On the other hand, quantities like the local orientation of fault planes, the final rupture dimensions and the details of the small-scale slip variability are taken from predefined probability distributions. We use this model to analyze aftershock activity. In particular, we show the impact of observational constraints on the aftershock fault plane orientations with regard to the expected spatiotemporal aftershock distribution, e.g. the occurrence of stress shadows. While published aftershock investigations are so far based only on one of the two end-member cases (optimally oriented fault planes which is equivalent to complete ignorance, or fixed receiver faults which assumes perfect knowledge), we investigate the whole bandwidth in-between. In particular, we discuss observational constraints with regard to their impact for aftershock forecasting.
S33A-1924
Recurrent large earthquakes in a fault region: What can be inferred from small and intermediate events?
We present a renewal model for the recurrence of large earthquakes in a fault zone consisting of a major fault and surrounding smaller faults with Gutenberg-Richter type seismicity represented by seismic moment release drawn from a truncated power-law distribution. The recurrence times of characteristic earthquakes for the major fault are explored. It is continuously loaded (plate motion) and undergoes positive and negative fluctuations due to adjacent smaller faults, with a large number Neq of such changes between two major earthquakes. Since the distribution has a finite variance, in the limit Neq→∞ the central limit theorem implies that the recurrence times follow a Brownian passage-time (BPT) distribution. This allows to calculate individual recurrence time distributions for specific fault zones without tuning free parameters: the mean recurrence time can be estimated from geological or paleoseismic data, and the standard deviation is determined from the frequency-size distribution, namely the Richter b value, of an earthquake catalog. The approach is demonstrated for the Parkfield segment of the San Andreas fault in California as well as for a long simulation of a numerical fault model. Assuming power-law distributed earthquake magnitudes up to the size of the recurrent Parkfield event (M=6), we find a coefficient of variation that is higher than the value obtained by a direct fit of the BPT distribution to seven large earthquakes. Finally we show that uncertainties in the earthquake magnitudes, e.g. from magnitude grouping, can cause a significant bias in the results. A method to correct for the bias as well as a Baysian technique to account for evolving data are provided.
S33A-1925
Space-time model for diagnosis of wide regional and global seismicity
Seismic quiescence and activation, as the precursors to large earthquakes, have attracted much attention
among seismologists. Of particular interest for this reason is the hypothesis that the stress-changes
transferred from a rupture or silent slip in one region can cause seismic changes in other regions. However,
in general, the clustering feature of earthquakes prevents us from detecting the seismicity change that is
caused by the stress change transferred from other region. This is because successive earthquakes are
mostly triggered by nearby events, but triggering mechanics within an aftershock region are difficult to be
calculated in terms of stress changes under complex fractal media. Nevertheless, we can use the statistical
empirical laws of aftershocks as a practical method for predicting earthquake clusters. The ultimate objective
of our project is to demonstrate that diagnostic analysis based on fitting the Epidemic Type Aftershock
Sequence (ETAS) model to regional and global seismicity can be helpful in automatically detecting the
anomalous period and regions where the actual occurrence rates deviate systematically from the modeled
one, which may be caused by exogenous stress changes sometime and somewhere. In particular, for a wide
geophysical region, we need the location dependent space-time ETAS model that takes account of different
geophysical characteristics of the crust, and these can be well estimated from hypocenter data by means of
an empirical Bayesian method. The utility of the model is demonstrated by applying it to global seismicity,
seismicity in southern California and Japan.
http://www.ism.ac.jp/~ogata/Ssg/ssgE.html
S33A-1926
Space-Time Seismicity Patterns Identification in Vrancea (Romania) Seismic Region Using High-Resolution Revised Catalog Data
The Vrancea seismic region is located in Romania at the sharp bending to the west of the Carpathians arc, in a complex collisional intra-continental environment, with three tectonic plates in contact. The concentration of seismic activity here, in an extremely narrow volume, strongly contrasts with the overall weak seismicity in the other parts of the Carpathians. The earthquakes are clustered in the subcrustal domain (60 - 170 km depth). Three to four relatively well defined seismic cycles are noticed per century in an earthquake catalog spanning several hundred years. Characteristic major shocks of magnitude Mw ranging from 6.5 to 7.7 are associated with each cycle. According to our modeling, the generation of major events is tightly connected to the way the background seismicity develops. Therefore, the high- resolution monitoring of the current seismicity over time intervals as long as possible could essentially contribute to understanding the system behavior at seismic cycle scale. The constraint of the space-time configurations in the seismogenic volume are fundamental elements for the numerical simulation of the Vrancea earthquakes. The main purpose of the present study is to revise the present catalogs of Vrancea earthquakes and to define on the basis of the new data the main characteristics of the seismicity in space and time for this seismic area. We considered only the Vrancea intermediate-depth earthquakes (clustered in the 60 - 170 km depth range, with a single isolated event at 220 km depth) that have occurred since 1974. The associated crustal seismicity is significantly reduced as energy is released and sparce in space and time. Cross-correlation and double-difference techniques are applied in relocating the earthquakes. The seismicity is confined in a very narrow slice, oriented NE-SW, close to a bi-dimensional geometry. The best-fitting plane approximating the spatial configuration of foci is obtained by multiple regression analyses. In order to investigate the implications and significance of a bi-dimensional approximation for the seismicity pattern and also how the individual foci deviate from a plane geometry, a specific testing algorithm is proposed. The seismic activity is sharply cut off above 60 km depth and below 170 km depth. This remarkable feature reveals in our opinion the active role played by the presence of specific critical processes in these depth ranges like phase transition, fluid migration, and thermal or geochemical reactions. The projection of foci on the best fitting plane shows a significant inhomogeneous distribution of the density of earthquake rate over the active seismic area, with enhancement at the boundary areas and deficit inside the active area. The significance and implications of these particular seismicity patterns on the simulation of earthquake seismic cycles in the Vrancea region is discussed
S33A-1927
A Quantitative Analysis of the Pattern Informatics Technique to Characterize Future Earthquakes
The Pattern Informatics (PI) technique, which is based on the spatiotemporal changes in seismicity rate, has shown promising results in Southern California (Tiampo et al., 2002). However, there have been limited attempts to quantitatively account for the optimal forecasting parameters and the characteristics of the predicted earthquakes. A sensitivity analysis of the model parameters in the PI method is investigated using receiver operating characteristics (ROC) diagram and Pierce's skill score to account for the optimal parameters (Holliday et al., 2006). The threshold PI value used for identifying the earthquake is evaluated by comparing the Pierce's skill score for different threshold values. The optimal cell size for the PI, found by varying the cell sizes to find the maximum area under the ROC, is an important factor if the forecasted earthquakes are characterized using their rupture areas, which can transcend several cell sizes. Moreover, the duration of training and forecasting intervals as well as the lower magnitude cutoff is estimated using the ROC diagram to provide a better forecast. Identification of the spatial and temporal regions of the seismic catalogs for better forecasts is done using the Thirumalai-Mountain (TM) fluctuation metric (Thirumalai et al., 1989). The TM metric, a measure of effective ergodicity, is studied systematically to analyse its dependence on cell size and other forecasting parameters. Next, the magnitude of the forecasting events is empirically estimated from rupture dimensions using the roughness index, a variation of the PI index (Tiampo et al., 2006). As the exact epicentre of the future earthquake is unknown, a forecasting magnitude or rupture dimension map is constructed which may provide insight into the mechanisms behind the earthquakes. Thus, the present work aims to provide a quantitative way to determine the various optimal parameters used for identifying and using relevant regions of the seismic catalogs while using a modified PI index to empirically estimate the magnitude of future earthquakes at various probable epicentres in the region.
S33A-1928
Analysis of Earthquake Clustering in Parameter Space
The seismic event is described by a number of parameters whose values determine its location in a multidimensional parameter space. Studies of earthquake clustering involve an analysis of distances between the events. When such studies are to be done in the space constructed by any selected set of parameters they meet the essential problem of different scales of the parameters. To solve this problem with the measure of distance we make use of the property of probability distributions of random variables that the cumulative distribution function transforms the random variable of any distribution into the random variable of uniform distribution in [0,1] interval. The cumulative distribution functions of all parameters under study are estimated from parameter values of a set of earthquake by means of the non-parametric, kernel estimator. This transformation of values of parameters of events into values of their respective cumulative distributions equalizes all dimensions is such a way that the distance between points is Euclidean. The data for reconstructing the cumulative distributions can be any relevant for the investigated problem, i.e. a long event series that includes subseries under study, a series preceding the studied subseries, a background seismicity data etc. We apply this approach to analyze seismicity preceding Mw6.5 Kozani-Grevena (Greece) earthquake from 13/05/1995. The considered parameterizations are the interevent time, interevent distance, epicentral distance to the main shock and magnitude. The analysis in moving time windows reveals appearance of anomalous patterns of smaller events in the time - distance subspaces. Some four years before the main shock a simultaneous shortening of the interevent time and interevent distance becomes distinct. The anomalous patterns are less visible when magnitude complements the parameter space. This work was prepared within the framework of the research project No. PBS-Grecja/10/2007, financed by the Ministry of Education and Science of Poland during the period 2007 to 2009.
S33A-1929
Simulation studies on the differences between spontaneous and triggered seismicity and on foreshock probabilities
In this study we investigate the foreshock probabilities calculated from earthquake catalogs from Japan, Southern California and New Zealand. Unlike conventional studies on foreshocks, we use a probability-based declustering method to separate each catalog into stochastic versions of family trees, such that each event is classified as either having been triggered by a preceding event, or being a spontaneous event. The probabilities are determined from parameters that provide the best fit of the real catalogue using a space- time epidemic-type aftershock sequence (ETAS) model. The model assumes that background and triggered earthquakes have the same magnitude dependent triggering capability. A foreshock here is defined as a spontaneous event that has one or more larger descendants, and a triggered foreshock is a triggered event that has one or more larger descendants. The proportion of foreshocks in spontaneous events of each catalog is found to be lower than the proportion of triggered foreshocks in triggered events. One possibility is that this is due to different triggering productivity in spontaneous versus triggered events, i.e., a triggered event triggers more children than a spontaneous events of the same magnitude. To understand what causes the above differences between spontaneous and triggered events, we apply the same procedures to several synthetic catalogs simulated by using different models. The first simulation is done by using the ETAS model with parameters and spontaneous rate fitted from the JMA catalog. The second synthetic catalog is simulated by using an adjusted ETAS model that takes into account the triggering effect from events lower than the magnitude. That is, we simulated the catalog with a low magnitude threshold with the original ETAS model, and then we remove the events smaller than a higher magnitude threshold. The third model for simulation assumes that different triggering behaviors exist between spontaneous event and triggered events. We repeat the fitting and reconstruction procedures to all those simulated catalogs. The reconstruction results for the first synthetic catalog do not show the difference between spontaneous events and triggered event or the differences in foreshock probabilities. On the other hand, results from the synthetic catalogs simulated with the second and the third models clearly reconstruct such differences. In summary our results implies that one of the causes of such differences may be neglecting the triggering effort from events smaller than the cut-off magnitude or magnitude errors. For the objective of forecasting seismicity, we can use a clustering model in which spontaneous events trigger child events in a different way from triggered events to avoid over-predicting earthquake risks with foreshocks. To understand the physical implication of this study, we need further careful studies to compare the real seismicity and the adjusted ETAS model, which takes the triggering effect from events below the cut-off magnitude into account.
S33A-1930
Stress triggering of large earthquakes complicated by transient aseismic slip episodes
We investigate how a static stress change, caused by a nearby earthquake, perturbs sliding processes on a fault and the time to the next earthquake. For this purpose, we model numerically a reverse fault in a two- dimensional, semi-infinite and elastic medium, partly stable and partly unstable under a rate- and state- dependent friction law. When no stress change intervenes, steady sliding spreads out of the deep zone of frictional stability, and keeps expanding updip during the interseismic period. Aseismic slip episodes (ASEs) develop spontaneously within the region of steady sliding, and repeat quasi-periodically. When steady sliding has spread beyond a certain critical length (the nucleation length) above the zone of frictional stability, one of the ASEs undergoes rapid buildup, and avalanches into high-speed seismic sliding. When a step-like stress change intervenes, ASEs of larger amplitudes arise, and evolve similarly to their spontaneous counterparts until one of them avalanches into an earthquake. If the stress change arrives not too early nor too late in the seismic cycle, the basic picture is as follows: delaying the input of a stress step, be it positive or negative, delays the timing of the ASE sequence, and thereby postpones the next earthquake. When the input is delayed to more than a certain extent, however, an earlier ASE gets to evolve into an earthquake, advancing its time discontinuously. Results can sometimes be counterintuitive at first glance, like a positive stress step delaying the next earthquake or a negative stress step advancing it on the contrary. Only in the final stage of a seismic cycle does a positive stress step systematically advance the next earthquake and a negative stress step systematically delay it. This suggests that considering the effects of stress perturbation does not enhance the precision of long-term earthquake forecasting for faults still in the mid-cycle, but may effectively enhance the precision of forecasting for faults in the final stage of the cycle. The interaction between aseismic and seismic sliding has caused the intriguingly complicated mode of fault response in the numerical results of our present study. At the current moment, however, this type of fault response is yet to be verified by field observations, and one cannot totally rule out the possibility that it is a model-specific artifact.
S33A-1931
Better Forecasts Through Comparison of Natural and Synthetic Earthquake Catalogs
In recent years, several different forecasting methods based upon the quantification of patterns in seismicity data have been proposed with varying degrees of success (e.g. Keilis-Borok, 1982; Bowman et al., 1998; Tiampo et al., 2002; Mignan and Di Giovambattista 2008). However, the physical basis of these methods is not well understood. This lack of understanding presents a significant barrier to the determination of the accuracy of these methods, the best route to improving their forecasting capability, and the possible limits to their accuracy as imposed by the physics. In the past, earthquake forecasting techniques that rely on statistical analysis of earthquake seismicity have struggled with the completeness and coverage issues in historic catalogs that affect the apparent their apparent stationarity, while attempting to determine if the assumption of stationarity is correct from a physics perspective. However, recently the equilibrium statistical property of ergodicity was identified in an earthquake fault system for a particular set of spatial and temporal conditions (Tiampo, et al., 2007). Ergodicity in this context not only requires that the system is stationary for these networks at the applicable spatial and temporal scales, but implies that they are in a state of metastable equilibrium in which the ensemble averages can be substituted for temporal averages. Recent application of this measure to one particular forecasting algorithm, the PI index (Tiampo et al., 2002), with the goal of optimizing its forecast capability through the identification of ergodic regions shows significant promise. Here we apply the same measure to synthetic seismicity catalogs, created from both analytical algorithms and large scale fault simulations (Rundle et al., 2005; Hayes et al., 2006; Mignan and Tiampo, 2008) in order to better understand the physical process that affects the forecast accuracy, both in space and time. We show that these ergodic regions, defined by magnitude and time, provide more reliable forecasts of future events in both natural and synthetic catalogs. We then present improved forecasting techniques developed from this joint analysis of historic and synthetic data.
S33A-1932
Multiscale cellular automaton models for earthquakes and faulting
Inspired by theoritical developments in complex system science, it is now possible to study feedback mechanisms, positive or negative, that can be involved in the brittle deformation processes of the upper crust (e.g. earthquake triggering, fault interactions, aseismic transients). As an example, we discuss here recent numerical investigations in modelling earthquake and faulting patterns via cellular automaton approach. In particular, we focus on the advantage of multiscale cellular automata. This new generation of models combine long-range interactions that mimics the continuum elastic solutions while retaining the inherently discrete approach of statistical physics. A finite number of states and a hierarchy of cellular spaces describe a population of fractures at different length scales. In addition, a set of rules and a time-dependent stochastic process govern the process of fracture growth and the permanent redistribution of the strain rate. Thus the internal dynamics is based on long range interactions produced by the redistribution mechanism as well as on short range interactions at the different levels of the multiscale system. This combination is very efficient in modelling the observed complex nature of earthquake and faulting phenomena (fracture-length distributions, earthquake-size distributions, aftershock decay rates). It is certainly because this method incorporates essential ingredients such as the discontinuous nature of faulting and a set of equations/rules that do not causally uniquely determine the evolution of the system.
S33A-1933
Effects of Heterogeneities in Strength and Initial Shear Stress on Large Ruptures in a Fast Multi-cycle Earthquake Simulator (RSQSim) and DYNA3D
RSQSim is a fast earthquake simulator that produces long (~ 106 event and ~ 104 year) synthetic seismicity catalogs in complex fault systems. It treats the interseismic and nucleation phases of the seismic cycle quasi-statically with an approximate version of rate- and state-dependent friction. The ruptures themselves are quasi-dynamic with slip speeds determined by shear impedance considerations. Validation of coseismic final slip (and therefore stress change) distributions is important for the generation of long catalogs because subsequent events in such simulators need to inherit the proper stress fields. Also, the heterogeneous evolved stress states from long simulations in complex fault systems (resulting from complex large ruptures, ongoing smaller seismicity, and stress interactions within the fault system) may be useful as more realistic inputs to dynamic rupture modelling. If the time evolution of ruptures in RSQSim is also realistic, they may be useful as kinematic sources for seismic hazard ground motion calculations. As part of an effort to validate the quasi-dynamic ruptures in RSQSim, we compare rupture propagation on a variable-strength planar fault in RSQSim to that on a similar fault in DYNA3D (a fully dynamic finite element model employing slip-weakening friction) for single, large, artificially nucleated ruptures. Previous work has shown that on homogeneous planar faults the RSQSim results agreed quantitatively very well with those of DYNA3D. For this comparison, our asperity model consists of multiple rectangular zones of increased normal stress of varying size, location, and amplitude. The heterogeneities produce complex ruptures - the rupture front tends to wrap itself around the barriers and create a burst of energy once it propagates across a barrier. Both codes allow rupture propagation over significant zones of negative stress drop in these asperity regions. Rupture durations, average rupture propagation speeds, and overall slip pattern are quite similar with both methods. However, ruptures in DYNA3D propagate more rapidly through the barriers and generate less high-frequency variations of slip than ruptures in RSQSim. The qualitative agreement of these two very different methods is good and may improve with further tuning of quasi-dynamic computational parameters. Using the same heterogeneous strength distribution as in the single-event examples, we use RSQSim to simulate several thousand years of additional seismicity. The effects of the asperities along the fault become much more subtle over multiple earthquake cycles as the shear stress adjusts to the higher normal stress in the asperities. Subsequent large spontaneous events rupture at much more variable and significantly slower velocities through the evolved stress states than through the uniform initial shear stress state of the first, artificially nucleated event. Indeed, the ruptures occasionally nearly come to a halt before continuing. These complex ruptures will produce very different ground motions than the more coherent ruptures seen in ruptures through smoother initial stress states.
S33A-1934
Collaborative Comparison of Earthquake Simulators
Earthquake simulators, i.e. computer models in which a series of earthquakes spontaneously occur, are important for understanding earthquake mechanics and earthquake predictability. However, to use earthquake simulators in hazard anaylsis they must show realistic behavior. It is difficult to determine how realistic simulator results are. This is in part because of the complexity of their behavior and the limited database of long sequences of natural earthquakes, especially large ones, against which to compare a simulator's behavior. Due to limits on memory and computation speed it is presently impossible to construct a simulator that simultaneously incorporates everything known about frictional behavior of rock, includes full elastodynamics, and utilizes both small enough elements to properly represent a continuum and enough elements to cover a large geographic area and represent many faults. Consequently, all simulators make compromises. A wide variety of simulators exist, each with different compromises. The effects on the simulator results of these compromises are not currently known. Our goal is to gain a better understanding of the validity of the results of earthquake simulators. This is a joint effort to compare the behavior of our nine independently devised earthquake simulators. We have defined and studied two simple problems. The first checks that each simulator accurately gives the stresses due to slip on a simple vertical strike-slip fault. All simulators satisfactorily passed this test. The second is a comparison of the behavior of a simple strike slip fault, with a simple bi-linear asymmetrically peaked initial stress distribution, and a constant loading rate. The fault constitutive properties have a fixed failure stress, higher than the peak in the initial stress, and a fixed dynamic sliding stress, although models utilizing rate and state friction only approximate this simple description. A series of earthquakes occur in the simulations and the rupture length for each progressively increases until the rupture extends the entire length of the fault. Some differences exist between the behaviors of the different simulations, although there is general agreement in the main characteristics. Future comparisons will involve increasing degrees of complexity. The next problem will involve a single fault with different strengths on different parts of the fault. Subsequent tests will involve multiple faults with various configurations and eventually will involve systems that are sufficiently complicated that comparisons will have to involve the statistics of many events.
S33A-1935
CSEP Earthquake Forecast Testing Center for Japan
One major focus of the next Japanese earthquake prediction research plan 2009-2013 are testable earthquake forecast models. For this purpose, the Earthquake Research Institute joined the Collaboratory for the Study of Earthquake Predictability (CSEP) and installed in an international collaboration a prototype testing center for rigorous evaluation of earthquake forecast models. We report on the implementation of this testing center, the quality characterization of the earthquake catalog data stream from the Japan Meteorological Agency (JMA), the definition of a Japanese testing region, and first test results. A first set of three one-year smoothed-seismicity models are fully implemented in the testing center and are under test since 1 September 2008. In the near future, additional models will be introduced and new specialized testing areas will be defined to promote rigorous earthquake prediction research on different topics and regions in Japan.
S33A-1936
Geometrical Control of Asperity Rupture: A Possible Origin of Comlex Cycle of Miyagi-oki Earthquakes, Northeastern Japan
Large earthquakes of M~7.5 repeatedly occurred at an average interval of 37 years in the Miyagi-oki region, along the Japan trench, northeastern Japan. Seismic data analyses by several researchers revealed that the 1978 Miyagi-oki earthquake of M=7.4 was an interplate earthquake, and it had three asperities, where relatively large coseismic slip was estimated. In the 2005 Miyagi-oki earthquake of M=7.2, it was reported that one or two of the asperities of the 1978 Miyagi-oki earthquake were broken, and the largest asperity of the 1978 event remains unbroken. This suggests that the probability of the occurrence of a large earthquake in the Miyagi-oki region is still high. Note that significant postseismic sliding was observed by the Global Positioning System for the 2005 Miyagi-oki earthquake, suggesting that aseismic sliding process is important for stress concentration and release in the Miyagi-oki region. These observations suggest that nonuniform frictional properties on the plate boundary in the Miyagi-oki region control the complex cycle of the Miyagi-oki earthquakes. The frictional properties at the asperities exhibit stick-slip, while the frictional properties at the surrounding regions show aseismic sliding, which helps stress build up at the asperities. In the present study, a numerical simulation for the recurrent Miyagi-oki earthquakes is performed to understand the origin of the complex cycle. This may be also useful for inferring nonuniform frictional properties at the plate boundary and the prediction of the next Miyagi-oki earthquake. The simulation method is the same as Kato (2008) for the numerical model for the Sanriku-oki earthquakes, where a rate- and state-dependent friction law, shear loading due to a steady plate motion, and a 3D infinite uniform elastic medium are assumed. To simulate complex earthquake sequence in the Sanriku-oki region, Kato (2008) assumed different values of friction parameters for some asperities. Instead, in the present study, geometrical irregularity is introduced for a single asperity, in which frictional parameters are uniform. Velocity-weakening friction is assumed at the asperity, while velocity-strengthening friction is assumed in the other regions, where aseismic sliding is expected. Stress concentration occurs at the edges of an asperity due to surrounding aseismic sliding. Higher stress concentration is expected for a smaller curvature of the boundary between an asperity and an aseismic sliding region from a theoretical analysis of a simple circular asperity model. Considering the coseismic slip distributions of the 1978 and the 2005 Miyagi-oki earthquakes reported in the literature, I assume an irregular shape of the asperity. Numerical simulations show that rupture starts where the curvature of the boundary between the asperity and the surrounding aseismic sliding region is the smallest and the stress rate is the highest. This simulated rupture sometimes stops when it encounters less stressed region of the asperity, resulting in a smaller earthquake. In the other times, the rupture propagates over the asperity, causing a larger earthquake. Thus a complex earthquake sequence, involving different magnitude earthquakes, can be simulated by introducing geometrical irregularity of an asperity, which generates nonuniform stress distribution.
S33A-1937
S2-Project: Near-fault earthquake ground motion simulation in the Sulmona alluvial basin
Recently the Italian Department of Civil Protection (DPC), in cooperation with Istituto Nazionale di Geofisica e Vulcanologia (INGV) has promoted the 'S2' research project (http://nuovoprogettoesse2.stru.polimi.it/) aimed at the design, testing and application of an open-source code for seismic hazard assessment (SHA). The tool envisaged will likely differ in several important respects from an existing international initiative (Open SHA, Field et al., 2003). In particular, while "the OpenSHA collaboration model envisions scientists developing their own attenuation relationships and earthquake rupture forecasts, which they will deploy and maintain in their own systems" , the main purpose of S2 project is to provide a flexible computational tool for SHA, primarily suited for the needs of DPC, which not necessarily are scientific needs. Within S2, a crucial issue is to make alternative approaches available to quantify the ground motion, with emphasis on the near field region. The SHA architecture envisaged will allow for the use of ground motion descriptions other than those yielded by empirical attenuation equations, for instance user generated motions provided by deterministic source and wave propagation simulations. In this contribution, after a brief presentation of Project S2, we intend to illustrate some preliminary 3D scenario simulations performed in the alluvial basin of Sulmona (Central Italy), as an example of the type of descriptions that can be handled in the future SHA architecture. In detail, we selected some seismogenic sources (from the DISS database), believed to be responsible for a number of destructive historical earthquakes, and derive from them a family of simplified geometrical and mechanical source models spanning across a reasonable range of parameters, so that the extent of the main uncertainties can be covered. Then, purely deterministic (for frequencies < 2Hz) and hybrid deterministic- stochastic source and propagation simulations are carried out for different fault rupture scenarios (but including important features such as the dominant near-surface geology), and the results in terms of representative ground motion parameters appropriately enveloped. The fully 3D problem is solved using the Spectral Element (SE) method, extensively published by Faccioli and his co-workers, and Quarteroni and co- workers, starting from 1996, and the computational code GeoELSE (http://GeoELSE.stru.polimi.it/). Finally, numerical results are compared with available data and attenuation relationships of peak values of ground motion in the near-fault regions elsewhere. Based on the results of this work, the unfavorable interaction between fault rupture, radiation mechanism and complex geological conditions may give rise to large values of peak ground velocity (exceeding 1 m/s) even in low-to-moderate seismicity areas, and therefore increase considerably the level of seismic risk, especially in highly populated and industrially active regions, such as the Central Italy.
S33A-1938
Rate-and-state cycle simulations and their application to scaling problems in seismology
We study the statistical properties of seismicity from multi-cycle simulations on 2.5D (in-plane) and 3D models of a vertical strike slip fault controlled by rate-and-state friction, with spatially variable frictional parameters. The quasi-dynamic, continuum framework bridges the gap between efficient yet relatively coarse, discrete approaches, and computationally expensive fully dynamic models. Our model allows a detailed study of various aspects of natural seismicity. In response to variable distributions of the critical slip distance L, initial conditions for nucleation and propagation of instabilities spontaneously evolve during the simulations, producing a wide range of event sizes. We use the high-resolution, depth-independent 2.5D version to investigate correlations between earthquake size and nucleation properties. We find that the results depend on the degree of the imposed heterogeneity, the 'stage' in the seismic cycle, and algorithmic choices related to measurements during the nucleation process. The 3D version is used to investigate classic scaling relations of synthetic seismicity. We find that results are insensitive to the frictional heterogeneity conditions, which reflect properties of faults at different evolutionary stages. This insensitivity is in agreement with properties of natural seismicity. However, the average stress drop scaling for earthquakes saturating the seismogenic zone depends strongly on the frictional rate dependence at their downward extension. Similar to the nucleation measurements, the difference in scaling relations between event populations from simulations with different depth-dependent properties is sensitive to choices defining the termination of an event. Together, simulations targeted at different aspects of seismicity scaling reveal that estimates of properties reflecting the bulk of energy release during earthquakes or averages over the earthquake process are stable. On the contrary, properties associated with transitions to gradual aseismic deformation, which by their own nature do not have sharp, clear definitions, are sensitive to choices related to measurements. With respect to the session's scope, we discuss the range of applications of our model, such as using the generated physically self-consistent scenario earthquakes for near-source ground motion simulations. Limitations regarding the extrapolation to future regional seismicity behavior are evaluated with respect to the problem the model was intended to solve.
S33A-1939
Coseismic thermal pressurization can prolong recurrence intervals of earthquake cycle
Earthquake is a short-lived event, while it needs a very long preparation period. The transition is rapid but seamless. We should correlate physics governing during the short-term earthquake period with that governing during the long-term preparation period. Brace and Byerlee [1966] proposed that stick-slip behavior is a mechanism for earthquakes from this standpoint. Following the proposition, lots of researchers have executed numerical simulations of a spring- slider system in order to interpret the earthquake cyclicity (e.g., Gu et al. [1991]). For such researches, it is necessary to use a constitutive law of friction on an interface between a slider and ground. By way of example, a rate- and state- dependent friction law (Dieterich [1979]) has been widely used, because it can represent frictional healing during the interseismic period. Despite the previous extensive studies, there is a dearth of information on roles of pore fluid. The pore fluid existence within a fault zone dramatically changes the frictional property via reduction of normal stress (Brace and Martin [1968]). Further, the pore fluid pressure may evolve and affect every aspect of earthquakes. Here, we try to add a new perspective to the earthquake cyclicity. It is an effect of short-term temporal change of the pore pressure, due to the coseismic thermal pressurization (hereinafter called TP). TP is a short-lived physical mechanism that frictional heating at a fluid-saturated fault pressurizes the pore fluid within the fault zone (Sibson [1973]). It can greatly affect the fault constitutive relation (Andrews [2002]) and the dynamic propagation of the earthquake rupture (Bizzarri and Cocco [2006]). In this presentation, we show that the short-lived TP is again a significant mechanism for the earthquake cyclicity, using the spring-slider system with the rate- and state- dependent law in a 1-D elastic body. If the shear zone thickness is smaller than several decimeters, TP can greatly prolong the recurrence intervals of the earthquake cycle because of the higher static stress drop compared to the case without TP.
S33A-1940
Effect of positive and negative (A-B) values assigned in the region other than asperities on the earthquake cycles
We consider the earthquake cycles in the Tohoku region, northeast Japan, especially in the Miyagi-Oki region. In that region, two or three asperities exist next to each other. Based on a rate/state friction law, we examine the effect of two friction properties of stable sliding hpositive (A-B)h and conditionally stable sliding gnegative (A-B)h assigned in the region other than the asperities at seismogenic zones. In the Miyagi-Oki region, the Pacific plate is subducting. We use a 2-D flat fault in an infinite uniform elastic medium to model asperities with stick-slip property of negative A-B values on the plate interface, following Kato (2004). He assigned stable sliding property of positive (A-B) values in the background region other than asperities. Here, we call this as gpositive (A-B) modelh. On the other hand, according to friction experiments of granite, (A-B) value is negative from 100E°C to 350°C and positive at higher temperatures (Blanpied et al,1991). A thermal model of northeast Japan (Wang and Suyehiro,1999) indicates that the depth on the subduction interface at 350°C is 60-70km. Deduced from these two studies, the seismogenic zone in the Miyagi-Oki region should have negative (A-B) values also in the region other than asperities. We call this gnegative (A-B) modelh. We compare these two models. First, we produce the same earthquake cycle in each one-asperity model. We define the same cycle as the one with the same recurrence time and the same distribution of afterslip and stress change. Then, we set two asperities in each model. Simulations show that afterslip propagation of first earthquake causes second earthquake and that time lag in gnegative (A-B) modelh is shorter than one in gpositive (A-B) model. This result suggests that "negative (A-B) model", as well as gpositive (A-B) modelh, has potential to duplicate the complex asperity interactions observed in the Miyagi-Oki region.
S33A-1941
Effect of Rheological Boundaries and Weaker Patches on Supershear Rupture in 3D Simulations of Earthquake Sequences and Aseismic Slip
We study supershear transition and propagation of dynamic rupture through simulations of earthquake
sequences and aseismic slip in a 3D fault model. Our simulations reproduce all stages of spontaneous fault
slip, from accelerating slip before dynamic instability, to rapid dynamic propagation of earthquake rupture, to
post-seismic slip, and to slow interseismic slip between dynamic events. In the model, a planar strike-slip
fault is governed by rate and state friction laws with the aging evolution equation. The fault contains a
potentially seismogenic velocity-weakening region surrounded by velocity-strengthening regions.
We find that the rheological boundary between the velocity-weakening and velocity-strengthening regions
promotes supershear transition. During interseismic periods, velocity-strengthening regions move with slip
velocity comparable to the plate loading rate, while the velocity-weakening region is essentially locked. This
disparity in slip concentrates shear stress next to the rheological boundary. Once earthquake rupture
nucleates, it propagates faster over these areas of higher prestress than over the rest of the seismogenic
region, transitioning to supershear speeds in some cases. Since the presence of such rheological
boundaries on natural faults can be inferred from laboratory studies and fault observations, this factor may
be important for supershear transition on natural faults. The occurrence of supershear transition in our 3D
model depends on friction properties and fault stress that develops in the model before large earthquakes
and can be explained by the distribution of the effective seismic ratio (e.g., Andrews, 1976) on the fault
before large events. The phenomenon of supershear transition due to rheological boundaries could not be
established in prior studies, as it can only be observed in simulations that include all of the following factors:
(i) inertial effects to enable supershear transition; (ii) a 3D model to include rheological boundaries in the
direction of rupture propagation; and (iii) long-term slip history to establish the corresponding stress
distribution on the fault before large events.
We also find that supershear transition in 3D models of long-term slip can be further promoted by favorable
compact fault heterogeneity, as suggested by the 2D single-event study of Liu and Lapusta (2008). Our
simulations show that adding a fault patch of lower effective peak frictional resistance can qualitatively modify
the behavior of the simulated fault, resulting in occasional supershear earthquakes in a model that has no
supershear events without the patch. Supershear transition occurs at the location of the heterogeneity, as
advocated by Liu and Lapusta (2008).
http://www.its.caltech.edu/~lapusta/publications.html