Jack Benjamin and Associates Summary Issues

DOE-STD-1024-92

the data generated by the 1988 Saguenay earthquake. They

concluded that no individual attenuation function fits the Saguenay

observations over the entire distance range of 40 to 200

kilometers. Additionally, they found that the attenuation model

selected by LLNL-AE5 is generally inconsistent with observed data;

and

Risk Engineering, Inc. has also criticized the method by which the

attenuation model used by LLNL-AE5 was derived. This model

was obtained by combining intensity versus amplitude regressions

from California and intensity attenuation relationships from Eastern

North America. The intensity versus amplitude relationship from

California may not be the same in Eastern North America due to

differences in ground motion frequency content, duration, and wave

type between the two regions. In addition, the substitution process

used leads to biased results. Risk Engineering, Inc. has

recommended that LLNL-AE5 be deleted from seismic hazard

calculations.

2.1.3 Jack Benjamin and Associates Summary Issues

Jack Benjamin and Associates (JBA) developed diagnostic tools to

provide a close examination of the factors that contributed the largest

percentage to the seismic hazard results. These comparisons showed

that differences in the mean hazard result between LLNL and EPRI are

controlled by low degree of belief parameter assessments. The overall

conclusion of JBA is that the process of expert elicitation and uncertainty

evaluation are extremely important. Some of the more important

observations are described below.

The LLNL constant percentile seismic hazard curves are based on

2750 individual seismic hazard curves. The highest curve (1/2750

or .04 percent) contributes 13 percent to the mean hazard curve at

Savannah River. This curve is associated with a 1/7.6 chance of

exceeding 0.25g annually at the Savannah River Site. This value

appears to be extremely high given the historical seismicity in the

Southeastern United States. The highest 21 hazard curves

(21/2750 or .80 percent) contribute about 50 percent to the mean

hazard curve.

The seismicity parameters associated with some of the highest

seismic LLNL hazard curves may be affected by the way that an

expert intensity-based recurrence relationship is translated into a

magnitude based recurrence relationship. This could result in

anomalously high activity rates and/or low recurrence slope ('b')

values. Jack Benjamin and Associates concluded that the largest

difference between the LLNL and EPRI results were due to