An Interim Recommendation For DOE Use of The LLNL and EPRI Hazard Curves

DOE-STD-1024-92

An Interim Recommendation

for DOE Use of the

LLNL and EPRI Hazard Curves

C. Allin Cornell

September, 1991

A Report to

J. K. Kimball

For reasons well described in the DOE Seismic Working Group document1. "Use of the

LLNL and EPRI Probabilistic Seismic Hazard Curves: Interim Position", it is desirable to

establish a straight-forward procedure to use the most dependable, stable information provided

by both of the two large-scale, state-of-the-art regional seismic hazard analyses conducted by

Lawrence Livermore National Laboratory for the Nuclear Regulatory Commission and by the

Electric Power Research Institute for the Seismicity Owners Group, a set of electric utilities

operating nuclear power plants, in the U. S.

Based on a comparison of results at some 70 EUS sites it is clear that the two studies

produce hazard results that are quite similar in terms of their central or median estimates, but

quite different in their estimates of the uncertainty about this central estimate, as evidenced by

major differences in their 15 percentile, 85 percentile, and mean2 estimates. Nonetheless there

are certain stable trends in these differences. These will be discussed below. This stability

suggests that it is feasible to establish a simple procedure for combining the results of the two

studies that can be considered applicable at all sites.

For use with UCRL-15910 and DOE's Interim Position document, the objective here is

to provide a ground motion measure associated with a prescribed target mean hazard level,

e.g., 2 x 10-4. The procedure will make use of the more consistent, stable parts of the LLNL

and EPRI studies, primarily their median hazard estimates. In addition, in order to reflect the

uncertainty-induced difference between the median estimate and the higher mean hazard

estimate, an additional factor will be developed to modify the median-based estimate. This

factor is based on the observed consistencies within and between the studies (e.g., site-to-site,

frequency-to-frequency, etc.), coupled with a simple averaging approach.

Some Observed Differences and Consistencies. The "data" used for this study were

primarily median and 85 percentile hazard estimates for 70 sites produced both by LLNL and

by EPRI for PGA and spectral velocities (at 25, 10, 5, 2.5, and 1 hertz) at a range of

This report is designed to serve as an attachment to that report, avoiding duplication

of text, figures, and tables.

In this case the mean is a measure of the degree of uncertainty. The reason is

simple. In analyses of rare events, e.g., in the 10-3 to 10-5 range, it is common that the

mean estimate is much larger than the median, e.g., of the order of the 85 percentile,

because the uncertainty band spreads over two or more decades and, although often

quite symmetrical with respect to the log probability, it is skewed strongly right (upward)

on an arithmetic scale.

D-1