HR: 1340h
AN: S53B-0225    [Abstracts]
TI: Three-Dimensional Full-Wave Tomography on a Laptop
AU: * Zhao, L
EM: zhaol@usc.edu
AF: Department of Earth Sciences, University of Southern California, Los Angeles, CA 90089-0740 United States
AU: Chevrot, S
EM: Sebastien.Chevrot@cnes.fr
AF: Observatoire Midi Pyrenees, 14 Avenue Edouard Belin, Toulouse, 31400 France
AB: Recent advances in seismic tomography show that to resolve structures of sizes smaller than the first Fresnel zone width of the waves used, three-dimensional (3-D) Fr{\'{e}}chet kernels (a.k.a. the banana-doughnut kernels) must be used. Dahlen et al. (2000) proposed an efficient algorithm which made the 3-D kernels practical for global tomography (Montelli et al. 2004). However, ray-theory approximation in Dahlen et al. (2000) is only applicable to observations from far-field high-frequency body waves. We propose an alternative efficient approach to computing the 3-D kernels based on the normal-mode theory which provides accurate, full-wave solution to the wave equation. This aprroach comes from the realization that the heterogeneity-induced waveform perturbations only depend on the strain Green tensor (SGT) which is a function of the earth model only. Thus, a database of SGTs can be established for a reference Earth model such as AK135, which eliminates the need for repetitive evaluations of the SGTs in subsequent 3-D kernel calculations. The SGT database is composed of all the independent elements of the third-order SGT which requires a certain amount of CPU time and disk space depending on the size of the problem. For example, for a grid of 30 km in space and 2 sec in time, a complete SGT database for global tomography requires a few weeks of single processor CPU time and ~80 GBytes of disk space. Preliminary tests show that this modest amount of overhead work leads to two orders of magnitude increase in efficiency for 3-D kernel calculations, making it practical to conduct almost all global and regional tomography studies without making any high-frequency approximaiton. This approach is completely general and flexible. It can be used to compute 3-D kernels of any types of seismic data (traveltime, amplitude, splitting), for any phases on the seismogram, and for any model parameters. It can also be used for inversions of earthquake's centroid and even higher moments. Moreover, terabyte SGT databases can be established at large facilities for public use in global high-resolution and regional fine-resolution studies while individual researchers can create small, portable and customized databases for fast inversions using single workstations or even laptops.
DE: 7260 Theory and modeling
DE: 7200 SEISMOLOGY
SC: Seismology [S]
MN: 2004 AGU Fall Meeting