S21C-01 08:00h
A Progress Report on the Working Group for the Development of Regional Earthquake Likelihood Models (RELM)
Recent studies have demonstrated that dramatic improvements in seismic-hazard analysis (SHA) will require a more physics-based approach to modeling. At the same time, proper SHA requires that all viable models be included in the analysis. To meet these challenges, the Southern California Earthquake Center has an ongoing Working Group for the development of Regional Earthquake Likelihood Models (RELM; http://www.RELM.org). The goal is to develop a variety of viable, earthquake-rupture forecast models (rather than one consensus model), to formally test each against available geophysical observations, and to evaluate the seismic hazard implications. The models currently under development range in sophistication from simple Poisson models (e.g., based on smoothed historical seismicity), to physical earthquake simulators that track stress changes throughout the system. One model, called the SCEC Earthquake Rupture Forecast, will be a logical extension of the "consensus" models developed by previous Working Groups on California Earthquake Probabilities. However, it will be both object oriented (modular), allowing improved components to be plugged in when available, and living in the sense that the influence of any significant events will be accounted for in real time.
http://www.RELM.org
S21C-02 08:15h
Forecasting the evolution of seismicity in southern California: Animations built on earthquake stress transfer
We develop a forecast model to reproduce the spatial and temporal distribution of M$>$1.4 seismicity for a succession of large earthquakes observed during 1986-2003 in a 300 x 310 km area centered on the M=7.3 Landers earthquake. To parse the catalog into frames with roughly equal numbers of aftershocks, we animate the seismicity in log-time increments that lengthen after each mainshock, revealing aftershock zone migration, expansion and densification. To predict seismicity, we implement a rate/state algorithm that is updated for the stress transferred by each M$>$6 shock, and then evolves. In this model, Coulomb stress changes act to amplify the background seismicity. Small stress changes can produce large seismicity rate changes in areas of high background seismicity. Similarly, seismicity rate declines are evident in the stress shadows only in areas with previously high seismicity rates. Thus a key constituent of the model is the background seismicity rate, which we smooth from 1981-1986 seismicity. The correlation coefficient between observed and predicted shocks is 0.52 for 1986-2003, and 0.63 for 1992-2003; a control standard aftershock decay model yields 0.54 and 0.52 for the same periods. All four M$>$6.0 events that occurred after the test period begin struck in areas of high expected seismicity; three locate at sites where the expected seismicity rate falls above the 92 percentile, and one above the 75 percentile. The model reproduces much, but certainly not all, of the observed spatial and temporal seismicity distribution, from which we infer that the decaying effect of stress transferred by successive mainshocks influences seismicity for decades. Finally, we offer a M$>$5 earthquake forecast for 2005-2015, assigning probabilities to 324 10 x 10-km cells.
http://quake.usgs.gov/~ross
S21C-03 08:30h
Significance of stress transfer in time-dependent earthquake probability calculation
A sudden change in stress is seen to modify earthquake rates, but should it also revise earthquake probability? Data used to derive input parameters permit an array of forecasts; so how large a static stress change is required to cause a statistically significant earthquake probability change? To answer that question, effects of parameter and philosophical choices are examined through all phases of sample calculations. Drawing at random from distributions of recurrence-aperiodicity pairs identifies many that recreate long paleoseismic and historic earthquake catalogs. Probability-density functions built from the recurrence-aperiodicity pairs give the range of possible earthquake forecasts under a point-process renewal model. Consequences of choices made in stress-transfer calculations, such as different slip models, fault rake, dip and friction are tracked. For interactions among large faults, calculated peak stress changes may be localized, with most of the receiving fault area changed less than the mean. Thus to avoid overstating probability change on segments, stress-change values should be drawn from a distribution reflecting the spatial pattern rather than using the segment mean. Disparity resulting from interaction probability methodology is also examined. For a fault with a well-understood earthquake history, a minimum stress-change to stressing-rate ratio of 10:1 to 20:1 is required to significantly skew probabilities with >80-85% confidence. That ratio must be closer to 50:1 to exceed 90-95% confidence levels. Thus revision to earthquake probability is achievable when a perturbing event is very close to the fault in question, or the tectonic stressing rate is low.
S21C-04 08:45h
Simulation-Based Probabilistic Seismic Hazard Assessment Using System-Level, Physics-Based Models: Assembling Virtual California
The research community is rapidly moving towards the development of an earthquake forecast technology based on the use of complex, system-level earthquake fault system simulations. Using these topologically and dynamically realistic simulations, it is possible to develop ensemble forecasting methods similar to that used in weather and climate research. To effectively carry out such a program, one needs 1) a topologically realistic model to simulate the fault system; 2) data sets to constrain the model parameters through a systematic program of data assimilation; 3) a computational technology making use of modern paradigms of high performance and parallel computing systems; and 4) software to visualize and analyze the results. In particular, we focus attention on a new version of our code Virtual California (version 2001) in which we model all of the major strike slip faults in California, from the Mexico-California border to the Mendocino Triple Junction. Virtual California is a "backslip model", meaning that the long term rate of slip on each fault segment in the model is matched to the observed rate. We use the historic data set of earthquakes larger than magnitude M > 6 to define the frictional properties of 650 fault segments (degrees of freedom) in the model. To compute the dynamics and the associated surface deformation, we use message passing as implemented in the MPICH standard distribution on a Beowulf clusters consisting of >10 cpus. We also will report results from implementing the code on significantly larger machines so that we can begin to examine much finer spatial scales of resolution, and to assess scaling properties of the code. We present results of simulations both as static images and as mpeg movies, so that the dynamical aspects of the computation can be assessed by the viewer. We compute a variety of statistics from the simulations, including magnitude-frequency relations, and compare these with data from real fault systems. We report recent results on use of Virtual California for probabilistic earthquake forecasting for several sub-groups of major faults in California. These methods have the advantage that system-level fault interactions are explicitly included, as well as laboratory-based friction laws.
http://www-aig.jpl.nasa.gov/public/dus/quakesim/
S21C-05 09:00h
Test of a Short-Term Prediction Algorithm
We describe the ongoing test of the algorithm "RTP"(Reverse Tracing of Precursors) for short-term (months in advance) prediction of strong earthquakes. Algorithm is based on integration of geodynamical models of fault networks, models of non-linear-dynamics type, and pattern recognition analysis of observed and modeled seismicity. Performance of the algorithm is evaluated by advance prediction in California, Japan, Middle East, and Italy with adjacent areas.
S21C-06 09:15h
Optimizing the Parameters of the Rate-and-State Constitutive Law in an Earthquake Clustering Model
The phenomenon of earthquake clustering, i.e. the increase of occurrence probability for seismic events close in space and time to other previous earthquakes, has been modeled both by statistical and physical processes. From a statistical viewpoint, the so-called epidemic model (ETAS) introduced by Ogata in 1988 and its variations have become fairly well known in the seismological community. Tests on real seismicity and comparison with a plain time-independent Poissonian model through likelihood-based methods have reliably proved their validity. On the other hand, in the last decade many papers have been published on the so-called Coulomb stress change principle, based on the theory of elasticity, showing qualitatively that an increase of the Coulomb stress in a given area is usually associated with an increase of seismic activity. More specifically, the rate-and-state theory developed by Dieterich in the `90s has been able to give a physical justification to the phenomenon known as Omori law. According to this law, a mainshock is followed by a series of aftershocks whose frequency decreases in time as an inverse power law. In this study we give an outline of the above mentioned stochastic and physical models, and build up an approach by which these models can be merged in a single algorithm and statistically tested. The application to the seismicity of Japan from 1970 to 2003 shows that the new model incorporating the physical concept of the rate-and-state theory performs even better of the purely stochastic model with a smaller number of free parameters
S21C-07 09:30h
Synchronous seismicity-rate-changes from the predicted rates by the ETAS model in and around the northern Japan preceding the 2003 Tokachi-oki earthquake of M8.0
The epidemic type aftershock sequence (ETAS) model predicts the normal seismicity-rates that superposes the modified Omori functions for aftershock activities. Deviations (lowering and enhancement) of the seismic activities from the predicted rates by the ETAS model are revealed to have concurrently occurred during 1996 in four seismic regions, northern Japan. The respective deviations are explained by the changes in Coulomb failure stresses, caused by the same aseismic slip that is assumed to have occurred on a plate interface of the 2003 Tokachi-oki rupture of M8.0. The slip is also supported by a significant change in the distribution of fault mechanisms of moderate earthquakes in northern Japan, before and after 1996. According to the occurrence data of strong earthquakes of M5.0 or larger for the long period 1926-2003, the quiescence relative to the ETAS lasts about 10 years_f periods not only preceding the recent 2003 earthquake but also preceding the 1952 and 1968 great Tokachi-oki earthquakes.
http://www.ism.ac.jp/~ogata/
S21C-08 09:45h
Prospective Tests of Southern California Earthquake Forecasts
We are testing earthquake forecast models prospectively using likelihood ratios. Several investigators have developed such models as part of the Southern California Earthquake Center's project called Regional Earthquake Likelihood Models (RELM). Various models are based on fault geometry and slip rates, seismicity, geodetic strain, and stress interactions. Here we describe the testing procedure and present preliminary results. Forecasts are expressed as the yearly rate of earthquakes within pre-specified bins of longitude, latitude, magnitude, and focal mechanism parameters. We test models against each other in pairs, which requires that both forecasts in a pair be defined over the same set of bins. For this reason we specify a standard "menu" of bins and ground rules to guide forecasters in using common descriptions. One menu category includes five-year forecasts of magnitude 5.0 and larger. Contributors will be requested to submit forecasts in the form of a vector of yearly earthquake rates on a 0.1 degree grid at the beginning of the test. Focal mechanism forecasts, when available, are also archived and used in the tests. Interim progress will be evaluated yearly, but final conclusions would be made on the basis of cumulative five-year performance. The second category includes forecasts of earthquakes above magnitude 4.0 on a 0.1 degree grid, evaluated and renewed daily. Final evaluation would be based on cumulative performance over five years. Other types of forecasts with different magnitude, space, and time sampling are welcome and will be tested against other models with shared characteristics. Tests are based on the log likelihood scores derived from the probability that future earthquakes would occur where they do if a given forecast were true [Kagan and Jackson, J. Geophys. Res.,100, 3,943-3,959, 1995]. For each pair of forecasts, we compute alpha, the probability that the first would be wrongly rejected in favor of the second, and beta, the probability that the second would be wrongly rejected in favor of the first. Computing alpha and beta requires knowing the theoretical distribution of likelihood scores under each hypothesis, which we estimate by simulations. In this scheme, each forecast is given equal status; there is no "null hypothesis" which would be accepted by default. Forecasts and test results will be archived and posted on the RELM web site. Major problems under discussion include how to treat aftershocks, which clearly violate the variable-rate Poissonian hypotheses that we employ, and how to deal with the temporal variations in catalog completeness that follow large earthquakes.