Basis For Recommendations

DOE-STD-1024-92

2.0

BASIS FOR RECOMMENDATIONS

2.1

Summary Issues

The discussion provided below is a brief summary of the evaluations that

have been completed to date to investigate, in detail, the causes of the

significant differences between the LLNL and EPRI seismic hazard

methods. It should be noted that detailed evaluations of the differences

between the two studies have only been completed at a few sites. This

makes it difficult to reach definitive conclusions regarding the generic

causes of the differences between the two methods. The reader is

referred to the references cited in the discussion below to obtain more

detailed discussion regarding the key issues related to the seismic hazard

curves.

Three investigators have evaluated in some detail the LLNL and EPRI

seismic hazard methods. These investigators are LLNL (Bernreuter, 1987,

et al), Jack Benjamin and Associates (McCann, 1991) and Risk

Engineering Inc. (McGuire, 1990a, 1990b, 1991). The following are

summary issues as a result of these studies. The summary issues are

meant to capture key points that the investigators have made.

2.1.1 Lawrence Livermore National Laboratory Summary Issues

With respect to uncertainty estimates, uncertainty in zonation and

ground motion attenuation are, in general, the most significant

sources of uncertainty in the LLNL study. When compared to the

EPRI results, there appears to be a large difference in the

uncertainty estimates associated with the seismicity parameters

(both activity rates and slope of the recurrence curve);

The contribution of the background zone is extremely important for

sites in relatively low seismicity regions. Great care should be

taken in estimating the seismicity parameters of the zone which

contains the site. In some cases the host zone for a given site has

no assumed seismicity above magnitude 5.0 in the EPRI study;

Validation tests show that when using exactly the same input the

EPRI and LLNL algorithms give similar results;

The seismic hazard results are extremely sensitive to the input of

LLNL-AE5, particularly for rock site conditions. Lawrence Livermore

National Laboratory recognizes that an analysis such as they

performed contains certain combinations of assumptions which will

lead to estimates that are true outliers. It is LLNL's opinion that this

fact makes the mean a relatively poor choice to use to compare the

hazard between sites because it is more sensitive to outliers than