This focused issue of Seismological Research Letters describes some earthquake simulators and some results of using them on a fault model that encompasses a fairly comprehensive set of faults in California, excluding the Cascadia megathrust. Earthquake simulators are computer programs that model long histories of earthquake occurrence and slip using various approximations of what is known about the physics of stress transfer due to fault slip and the rheological properties of faults. The purpose of these large simulations of earthquake history is to learn about the statistical behavior of earthquakes within the simulated region in the hope that the simulated earthquake catalogs can demonstrate to us some aspects of earthquake behavior that observations cannot. The difficulty with basing our knowledge of earthquake behavior only on observations is that the record of earthquake occurrence from the combination of instrumental, historical, and paleoseismic data is too short and incomplete. As the results of the simulation studies described herein show, and as is recognized by everyone studying earthquake occurrence, the observation time needed is longer than is represented in our limited data, especially for the largest earthquakes that represent the greatest seismic hazard. Clearly, for earthquake simulators to be considered valid, they need to be consistent with our temporally-constrained observations. This is necessary, if not sufficient, for us to have faith in the result of earthquake simulators. Presumably, the greater the extent of known physics that can be included in an earthquake simulator is, the more likely it is to correctly represent actual behavior. However, limitations on computational power presently constrain our ability to include as much physics as is known into earthquake simulators while, at the same time, including a large number of faults that are represented with sufficiently detailed resolution. Consequently, the four simulators included in the study represented by the papers in this focus section make a variety of approximations of known physics. The fact that true elastodynamics is omitted by all four simulators is a notable example of the sacrifices made by each simulator in order to allow many faults and long histories to be treated. Some aspects of elastodynamics can be approximated by including such items as radiation damping and parameters that encourage rupture propagation, but other aspects cannot. For example, remote triggering by seismic waves cannot occur in these simulations.
The task of simulating earthquakes is more complex than the task of simulating Earth’s weather and climate. We understand less about fundamental geophysics than we do about weather and climate, and we have fewer observations due both to lack of transparency and the inaccessibility of relevant physical parameters for measurement. Thus we are hampered both by our lack of understanding of some of the important physical processes (especially a complete understanding of the constitutive behavior of fault zones), and by our lack of knowledge of the values of many of the relevant physical properties such as stress, fluid pressure, permeability, and frictional parameters below the surface of the Earth. The four earthquake simulators in this study make different assumptions about the details of fault friction; consequently, comparing the results provides some understanding of how important it is to include various features in the simulators. All our simulators suffer equally from our lack of knowledge of the three-dimensional distribution of relevant physical properties. Given this, it is perhaps surprising that the simulators compare as well as they do with our time-limited observational earthquake catalogs.
This focus section on earthquake simulators comprises seven papers. The first five describe the functioning of the simulators. The first, “Generic Earthquake Simulator,” co-authored by all participants in this study, describes the properties common to all the simulators. The next four papers describe the ways in which each individual simulator differs from the generic description, as well as giving some results from using the simulator. These four papers are: “ALLCAL Earthquake Simulator,” by Steve Ward, from the University of California Santa Cruz; “Virtual California Earthquake Simulator,” by Michael Sachs, Eric Heien, Don Turcotte, Mehmet Yikilmaz, John Rundle, and Louise Kellogg, a team from the University of California Davis; “RSQSim Earthquake Simulator,” by Keith Richards-Dinger and Jim Dieterich from the University of California Riverside, and “ViscoSim Earthquake Simulator,” by Fred Pollitz of the U.S. Geological Survey.
Next is a paper by Michael Barall, “Data Transfer File Formats for Earthquake Simulators.” Our ability to have all the simulators conduct simulations using the same input and to arrive at results that could be compared required our use of the common and extendable formats developed by Michael Barall.
The final paper, co-authored by all participants in this study, is “Comparison Among Observations and Earthquake Simulator Results for the allcal2 California Fault Model.” This paper compares the results among the simulators themselves, and those results with what is known from observations.
This study was supported by the Southern California Earthquake Center (SCEC). As someone doing earthquake simulations (though I’ve been using an earthquake simulator that could not model enough elements to model all of California for thousands of years), I volunteered to coordinate this group effort to compare results among the simulators and to compare the results with observational data. We undertook the study because SCEC wanted to know how well the earthquake simulators might be representing reality and, in particular, whether simulators might be used by SCEC’s Working Group on California Earthquake Probabilities (WGCEP) in its creation of future generations of the Uniform California Earthquake Rupture Forecast (UCERF). Some of the insight gained from our simulations has been included in UCERF3 and it is likely that such results will be used more extensively in future generations of UCERF.
Our results on the allcal2 model followed our earlier large-scale simulations of fault behavior in northern California, which were followed by an earlier generation (allcal1) of an all-California model. Our allcal2 model is close to the UCERF2 fault and deformation model. Another aspect of this comparison project was to compare results from a larger number of simulators on simpler problems using one or just a few faults. This allowed inclusion of simulators that include more physics, such as one used at Caltech by Nadia Lapusta and Hiro Noda that includes true elastodynamics. In these simpler problems the results can be compared and understood in more detail rather than in a merely statistical sense. However, our good progress made on the California-wide simulations, and their value for use in such efforts as creating versions of UCERF, dictated that SCEC’s limited resources be focused on the statewide simulations described in this focus section of SRL (November/December 2012). More results from these several comparison studies can be found at our public SCEC-hosted web site, http://scec.usc.edu/research/eqsims/.
The overall result of this study, summarized more fully in our multi-authored paper “Comparison Among Observations and Earthquake Simulator Results for the allcal2 California Fault Model,” is that the simulators, even at their present stage of development, do an admirable job of agreeing with existing observations, while also suggesting reasonable behavior of the fault system over longer times than are represented by the observations. It is notable, and not surprising, that the simulations indicate that longer earthquake catalogs than those available from the observational record are needed to capture the variability in rates of large earthquakes. Simulators provide a plausible representation of such variability and a basis for creating useful statistical descriptions of expected behavior. Many observers of these results believe that such simulations have a bright future and that future versions of UCERF are likely to make increasing use of their results.
In spite of the evident present success of earthquake simulators, improvements are both needed and possible in many ways. Such improvements, some of which have already been accomplished for some of our simulators, include: using layered viscoelasticity; including better representations of low dynamic frictional strength suggested by studies of high speed friction; representing off-fault seismicity; dealing with fault roughness on a smaller scale than the element size; and refining the resolution of the models by adding new faults and including more detail in the currently modeled faults (for example, use of uniform-sized triangular elements to utilize the SCEC Community Fault Model). Implementing all of the additions in this list represents a significant amount of work. However, all of these improvements are clearly within the realm of feasibility, and over the next several years we should see increasingly sophisticated simulators that will be used for an increasingly wide range of applications.
This research was supported by the Southern California Earthquake Center. SCEC is funded by NSF Cooperative Agreement EAR-0529922 and USGS Cooperative Agreement 07HQAG0008. The SCEC contribution number for this paper is 1658.
Posted: 6 November 2012