INCORPORATING UNCERTAINTY INTO PROBABILISTIC SEISMIC HAZARD MAPS FOR THE CENTRAL AND EASTERN U.S.
CHRIS H. CRAMER, U.S.G.S., 3876 Central Ave Ste 2, Memphis, TN 38152-3050, email@example.com
RUSSELL L. WHEELER, ARTHUR D. FRANKEL, U.S.G.S., P.O. Box 25046, Denver, CO 80225
PREDEEP TALWANI, Univ. of South Carolina, Dept. of Geological Sciences, Columbia, SC 29208
and RICHARD C. LEE, Westinghouse Savannah River Co., P.O. Box 625, New Ellenton, SC 29809
Because our growing knowledge of earthquakes is still incomplete, it is vital that the best available information is included in seismic-hazard analyses and the uncertainty in our scientific knowledge is properly represented (EPRI, 1986; SSHAC, 1997). Knowledge-based uncertainty is best included in a probabilistic seismic hazard analysis via the logic-tree formalism. Scientifically viable alternatives are included in a logic tree, including alternative rupture models, recurrence intervals, width of seismogenic ruptures, characteristic magnitudes, and attenuation relations. The logic tree can be sampled via the Monte Carlo methodology (Reiter, 1990; Cramer et al., 1996, BSSA) to determine a distribution of calculated seismic ground motions for each site in a hazard analysis. The best estimate of seismic hazard is given by the mean (expected value) of the ground motion distribution at each site, and the variability in seismic hazard can be represented by the coefficient of variation (COV). Additionally, sensitivity of the resulting seismic hazard to a single alternatives branch point in the logic tree can also be determined by holding all but one branch point fixed during the Monte Carlo sampling, and plotting then the resulting coefficient of variation for that individual alternatives branch point (ICOV). In the central and eastern U.S., preliminary logic-tree analyses have been completed for peak ground acceleration (PGA) and for 0.2 and 1.0 s spectral acceleration (Sa) in the New Madrid seismic zone, the Charleston area, and the southern Illinois Basin. The results for PGA and 0.2s Sa are very similar. For all three study areas, the COV exceeds 0.6 over modeled fault sources for all three ground motions (PGA and Sa at 0.2 and 1.0 s). Areas of concentrated seismicity that are not modeled as discrete faults can have COVís greater than 0.4. In the eastern U.S., seismic hazard maps are most sensitive to the uncertainty in the location of modeled seismogenic faults (ICOV as large as 0.4-0.6 for New Madrid and up to 0.9 for Charleston). For New Madrid, the ICOVís are less than 0.3 for the attenuation relation, characteristic magnitude, and recurrence interval alternative branch points. For the Charleston area, the recurrence interval (ICOV as much as 0.6) and characteristic magnitude (ICOV as much as 0.4) are more uncertain than for New Madrid. Also for 1.0 s Sa, the attenuation relation alternative branch point has greater uncertainty (ICOV up to 0.5) than at 0.2s Sa and for PGA. Preliminary 2% in 50 year mean seismic hazard values for selected cities in the study areas are compared to the values for the 1996 National Seismic Hazard map (Frankel et al., 1996).