Large earthquakes are rare, complex events and their sources are deep-seated and inaccessible to direct observation. Moreover, relatively few near-source strong motion recordings exist for events of very large magnitude, i.e., those earthquakes that pose the greatest potential threat to life safety and the built environment. Well-validated computer simulations provide a means to forecast strong ground motion from large earthquakes, based upon the best available earthquake physics and wave propagation models. Simulations can help fill the pressing need for site-appropriate ground-shaking estimates for use in performance-based engineering and structural analysis, a requirement that is especially acute in light of the construction boom in very tall buildings in western U.S. cities. Computer simulations are also essential for advancing our basic scientific understanding of earthquakes, as they reveal how large-scale effects emerge from smaller-scale interactions that are difficult to study experimentally or via traditional theoretical analysis. This project is developing numerical earthquake models that represent the principal mechanical and thermal processes of faulting, including the weakening of frictional resistance by the heating of rock contacts and the pressurization of pore fluids. These computer models also incorporate established statistical representations of fault roughness and of the stress state in the earth, as well as the strength limits of geologic materials (which place upper bounds on the amplitude of the stress wave disturbances excited by faulting). The models must run efficiently on large computer clusters (i.e., those with thousands of processor cores), so that they can combine (i) the small-scale resolution required to accurately simulate fault-zone processes with (ii) the large overall physical volume required to simulate large earthquake sources and compute surface ground motion at distances of engineering interest. The project engages both earth and computational scientists and trainees to address these modeling challenges. The project team is testing the computer models by comparing the ground motion predictions from large suites of simulations with comparable compilations from actual earthquake recordings. For example, the distributions (including median and statistical spread) of ground motion parameters such as spectral acceleration and peak ground velocity, as functions of event magnitude, site distance, and other variables, are among the targets of these observational tests. Those computer models that can be validated in the above sense are then used to (i) simulate earthquake shaking from future large earthquake scenarios, (ii) develop improvements to simplified modeling methods for ground shaking (the so-called kinematic methods), and (iii) improve the empirical ground motion models that are conventionally used in engineering design, by providing a physics-based extrapolation beyond the range in which they have significant support from data (i.e., to large magnitude and small source-to-site distances).
Damage to buildings and infrastructure during large earthquakes, and the accompanying hazards to the economy and to public safety, result principally from the strong ground shaking that earthquakes produce. Designing structures to perform acceptably during earthquakes requires that engineers have estimates of the level of ground shaking that is likely to occur during future earthquakes, and those estimates should be appropriate to the local and regional geological conditions. Many recordings have been made of strong motion from earthquakes around the world, and statistical models based upon these recordings provide the primary basis for those engineering estimates. Even with the wealth of such data that is now available, however, large uncertainties remain in our ability to forecast future levels of shaking. Furthermore, the uncertainty is greatest for very large earthquakes, because they occur infrequently and therefore are poorly represented in existing ground motion data. As a result, earthquake scientists and engineers have, in recent years, looked increasingly to computer models to learn more about earthquake ground motion. To be credible and reliable, such a model must be consistent with the earthquake data we already have. Yet, in order to provide new information beyond what we can infer from existing data, the model must also include additional knowledge, such as location and extent of regional geologic units, the geometric structure of faults, and scientific understanding of fault behavior during an earthquake. This project made some significant advances in computer modeling of earthquakes by adding new elements derived from geological information about fault structure (for example, the statistical nature of small-scale irregularities in fault shape) as well as from laboratory data (for example, recent discoveries about the behavior of rock surfaces as they undergo frictional heating). Adding the new layer of complexity to the computer model enabled it to simulate ground shaking much more realistically than was previously possible (in particular, the more complex model can simulate much higher frequencies of ground shaking). This complexity greatly increased the computing requirements, however, so the project also built algorithms to adapt the model to run on the latest generation of supercomputers, on which computations were done concurrently on nearly 100,000 processors. The model was tested by using it to simulate a large number (thousands) of ground motion recordings, then comparing average shaking predictions (for a given type and size of earthquake) with averages obtained by aggregating recorded data from many earthquakes from around the world. The tests demonstrate that the addition of complexity in the form of small-scale fault-shape irregularities brings the computer model predictions into close agreement with earthquake ground motion data over much of the frequency range relevant to engineering applications.
