HR: 0830h
AN: S21D-0333    [PDF]
TI: A Comparison of Multiple-Event Location Methods
AU: * Engdahl, E R
EM: engdahl@colorado.edu
AF: Department of Physics, University of Colorado, Campus Box 390 UCB, Boulder, CO 80309-0390 United States
AU: Rodi, W
AF: MIT, 42 Carleton St., E34-458, Cambridge, MA 02142-1324 United States
AU: Bergman, E A
AF: Department of Physics, University of Colorado, Campus Box 390 UCB, Boulder, CO 80309-0390 United States
AU: Waldhauser, F
AF: LDEO, Columbia University, PO Box 1000, 61 Route 9W, Palisades, NY 10964-8000 United States
AU: Pavlis, G L
AF: Department of Geology, Indiana University, 1005 East 10th Street, Bloomington, IL 47405 United States
AU: Israelsson, H
AF: Center for Monitoring Research, 1300 17th St., Suite 1450, Arlington, VA 22209-3872 United States
AU: Dewey, J W
AF: U.S. Geological Survey, P.O. Box 25046, Mail Stop 966, Denver Federal Center, Denver, CO 80225-0046 United States
AB: Multiple-event location methods solve jointly for the location parameters (hypocenters and origin times) of seismic events in a cluster and travel-time corrections at the stations recording the events. This paper reports some preliminary comparisons of five such methods that have been developed over the years: hypocentral decomposition (HDC), double differencing (DD), progressive multiple-event location (PMEL), joint hypocenter determination (JHD), and a recently developed algorithm based on grid search (GMEL). We have applied each method to two adjacent earthquake clusters in Turkey: 33 events from the 17 Aug 1999 Izmit earthquake sequence and 41 events from the 12 Nov 1999 Duzce sequence. Previously, Engdahl and Bergman (2001) had applied HDC to these clusters using Pn and teleseismic P arrival times from NEIC and ground-truth (local network) locations for a few of the events. Their data set comprised approximately 3500 arrivals at 640 stations for the Izmit cluster and 3200 arrivals at 600 stations for Duzce. We applied the other multiple-event location methods to the same set of phase picks, using the same phase identifications and fixed event depths that were used in the HDC analysis. While the five algorithms are quite different in their computational approach, our initial results indicate that the methods yield quite similar relative event locations when they are applied with the same data and assumptions. However, they resolve the trade-off between the centroid location of a cluster and station corrections differently, and they also differ in how they use ground-truth information to constrain this trade-off and obtain absolute event locations. The locations relative to the cluster centroids generally agreed within 5 km, but was on the order of 10 km in some instances. This may have to do with the different schemes for weighting data used by the different methods, which cannot always be equalized between methods. To test this hypothesis, we applied GMEL with two different weighting schemes and found discrepancies between relative locations as large as 10 km. Conversely, when GMEL was applied with the same station-variable weighting scheme that was used in the HDC analysis, the relative locations agreed within 2 km for all but a few events and within 5 km for all the events.
DE: 7215 Earthquake parameters
SC: Seismology [S]
MN: 2003 Fall Meeting