HIGH-PERFORMANCE COMPUTING IN SEISMOLOGY

Seismologists were among the first scientists to exploit the capabilities of advanced computing technology. Thirty years ago oil companies purchased the first "supercomputers" to process and analyze seismic reflection data in the search for petroleum. Shortly afterward, supercomputers were also used for seismological modeling in support of nuclear test detection. The field of seismology and society have both benefited tremendously from these activities. These projects engaged seismologists at the leading edge of technology, spurred the development of new computer technology, promoted a number of spin-off applications with important scientific implications (e.g., reflection seismic imaging of deep geologic structures), enhanced the discovery of energy of energy resurces, and supported initial efforst to limit the use of nuclear weapons by international treaties.

Since these early efforts, computer technology has evolved dramatically. Rather than supercomputing, the focus has shifted to the broader concept of high-performance computers matched to a diverse range of applications. The capabilities of high-performance computers may be used to display high-resolution graphics, perform complex model simulations, or archive large volumes of data. Today, high-performance computers are of critical importance to seismology because they are used at all stages of data acquisition, communication, modeling, and analysis. These applications include visualization of three-dimensional seismic images, automated processing of worldwide earthquake records, and complex simulations of earthquake processes. For these reasons the future strength of seismology will be closely coupled to the degree that the field incorporates and utilizes advances in high-performance computing. This report outlines the opportunities and challenges for this effort, ranging over the seismological subdisciplines of earthquake monitoring and physics, comprehensive test ban treaty verification, petroleum exploration, and global and strong motion seismology¹.

In its charge from the National Research Council, the Committee on Seismology was requested to identify the major computational challenges in seismology, assess specific research areas where emerging technologies may be decisive, and assess what is needed from technology to meet these computational challenges. As the first step in this process, a workshop on these issues was held at the San Diego Supercomputer Center in October 1994. Based on presentations at the workshop and from its own subsequent deliberations, the committee concludes that high-performance computers present significant opportunities for fundamental breakthroughs in seismological research.

Particularly promising areas for future research include simulations of earthquake processes, advanced modeling techniques for seismic wave propagation, and new methodologies for high-resolution images of the Earth's interior (from the crust to the core). The results of this work will have important implications for studies of geologic processes throughout the Earth and for understanding the causes of earthquakes. These applications also have tremendous practical applications that range from the mitigation of seismic hazards to locating undiscovered petroleum reserves. The importance of these problems provides strong incentives to develop high-performance computing applications throughout the field of seismology.

At present, the federally coordinated High Performance Computing and Communications Initiative (HPCCI) provides funding for a range of high-performance computing activities in seismology. The largest of these is the Advanced Computational Technology Initiative (ACTI) sponsored by the U.S. Department of Energy. Based on this program, high-performance computing activities in reflection seismology are the largest seismological component in federally funded HPCCI activities, exceeding support for other subdisciplines by more than an order of magnitude. Modeling of seismic wave propagation in three-dimensional sedimentary basins, one of the Grand Challenges of HPCCI, is the second largest.

The committee notes that there will be great challenges to making full use of high-performance computing technology throughout the field. Much of the problem stems from the recent shift toward parallel and distributed computing architectures as replacements for previous vector and serial machines. In principle, this transformation allows massive increases in computer speed and performance; however, it also requires a complete re-engineering of software and algorithms for scientific computing. In effect, the challenge of high-performance computing has shifted from issues of hardware design to problems of writing better software. Thus, taking advantage of advances in high-performance computing will require focused efforts to train and educate seismologists in the use of this technology. To promote these goals, the committee makes the following recommendations.

1.  Focused efforts to develop validated documented software for seismological computations should be supported, with special emphasis on scalable² algorithms for parallel processors.

There has been little effort to validate or benchmark the software for a wide range of seismological problems. A strong effort in this area could facilitate the growth of a new generation of software for high-performance computing applications and would allow researchers to focus on the more advanced problems associated with developing software for parallel computers. It is critical that these efforts focus on scalable parallel algorithms because this will be the trend of future software developments in high-performance computing. The committee observes that much of the software development for scalable algorithms is currently feasible in academic environments (outside supercomputer centers) using networks of linked workstations and standardized software for parallel computing. Such an effort would increase the exposure of a wide range of seismologists to high-performance computing technology with relatively small investments in new computing facilities.

2.  The education of seismologists in high-performance computing technologies and methodologies should be strengthened.

Breakthroughs in computational seismology will require a detailed understanding of high-performance computing software and hardware. The capabilities of seismologists in this area should be improved through broad educational efforts for researchers at all levels.

3.   Collaborations between seismologists and computational scientists and engineers should be strengthened.

High performance computing challenges in seismology are similar to a wide range of active research issues in computational science (e.g., I/O for large data sets, visualization). Seismology would benefit from increased interdisciplinary collaboration that includes computational scientists and engineers, but the mechanisms currently available to sponsor such collaborations appear to be limited. Activities of this type could be promoted through increased dissemination of research results from computational seismology and by greater participation by seismologists at workshops and conferences on high-performance computing. Also, collaborations between scientific societies could play an important role by facilitating cross-disciplinary forums and workshops for earth scientists and computational scientists and engineers.

4.   The infrastructure for archiving, disseminating, and processing large volumes of seismological data should be expanded.

Seismology has entered a new era that is characterized by (1) significant increases in the volumes of recorded seismic data, (2) explosive growth in the number and size of centralized data archives, and (3) "real-time" recording from global seismic networks. Full utilization of these future data streams will require a significant upgrade of the infrastructure for data communication and storage. To facilitate this transition, there should be a sustained effort to exploit the capabilities of new technology for seismological data communication and archiving. Recent developments in real-time seismic monitoring for earthquake hazards and nuclear test detection represent important opportunities in this area. Also, there would be substantial benefits from developing widely accessible computer archives from the records of the thousands of continuously recording seismic stations throughout the world. In this last area, international scientific unions could play a key role in facilitating the development of international data archives.

¹ Strong motion seismology focuses on modeling and measuring the intense ground motions close to an earthquake source that are sufficiently large to cause damage to structures.

² Scalable algorithms are designed so that their speed "scales" linearly with the number of processors in a parallel computer. The design and implementation of scalable algorithms are key factors in attaining high computational speeds with massively parallel architectures.

NAS Home Page | NAP Home Page | Reading Room | Report Home Page