HR: 0830h
AN: S31E-0793    [PDF]
TI: Geographical ISC data Characterization with Parallel Ray-tracing
AU: * Grunberg, M
EM: marc@renass.u-strasbg.fr
AF: EOST, 5 rue Ren\'e Descartes, Strasbourg, 67100 France
AU: Genaud, S
EM: genaud@icps.u-strasbg.fr
AF: ICPS-LSIIT, P\^ole API,Boulevard S. Brant, Illkirch, 67400 France
AB: In order to improve the accuracy of existing seismic tomography methods, huge quantities of information must be computed. Our objective is to build an adaptive mesh of the Earth so it can translate to an inverse problem of tractable size. The mesh cells size will be adapted depending on the "illumination" quality, that is regions where rays have brought more information will be modeled precisely, yielding small cells, whereas cells traversed by few rays are covered by large cells . These considerations lead us to develop software programs to take into account the computational resources needed for the 3D ray tracing step and the huge quantity of data to store into the cells mesh. These per-cell data are characteristics such as the ray length, the ray density, the number of ray hits by cell face, and the quality of the spatial repartition of rays in the cell. Our parallel programs made possible to process the entire ISC data set recorded for year 1999. We ray-traced 399,406 out of the 817,101 ray records, in an Earth mesh with 11 layers of 1$^\circ$ $\times$ 1$^\circ$ cells (712,800 cells). First we observe, that nearly half the data set cannot be traced because rays were not given a precise enough seismic signature. Secondly, we produce from the computed results, graphical 3D maps highlighting geographical zones where information brought by rays are noticeable. The graphical tool enables to visualize maps corresponding to any of the above cited characteristics. Several hardware equipments were tested (PC cluster, parallel computer, Grid computing) to run the experiment. An acceptable processing time of 400s was reached with 64 processors on SGI Origin 3800 parallel computer. Our programs showed a good scalability when the number of processors increase. Another interesting feature is our capability to process data sets by slice to overcome memory limitations, but nonetheless obtain aggregated mesh results. We think the processing of the entire ISC catalogue from 1964 to 2002 with several millions of rays is attainable.
UR: http://guenievre.u-strasbg.fr/ray2mesh
DE: 7200 SEISMOLOGY
SC: Seismology [S]
MN: 2003 Fall Meeting