Some Observed Differences and Consistencies.

DOE-STD-1024-92

amplitudes. (These results were prepared by Risk Engineering Inc., under contract to Martin-

Marietta Energy Systems, for DOE.)

Let us look first at the "inconsistencies". Figures 2 and 3 in the Interim Position report

are typical raw ratios of 85 percentile to median estimates of the hazard for 4 amplitudes of 10

hertz spectral velocity at 70 sites. Note the major differences between the LLNL and EPRI

ratios. Nonetheless, the site-to-site variability in these ratios is relatively small; the (indicated)

standard deviations are perhaps only 15% to 30% of the median ratio in the amplitude range of

primary interest. As shown in Figure 4 of the Interim Position report, this site-to-site variability

is explainable in part by soil vs. rock site effects, and a few high "outlier" sites (which contribute

significantly to the computed standard deviations) are very often (in the case of EPRI, at least)

sites with comparatively very low median hazards estimates, e.g., in the Gulf Coast. Based on

this observation of relatively small site-to-site variability in the ratios, we focus on simply the

geometric mean (over sites) ratio of 85th percentile to median hazard.

Figures A-1 through A-6 (Interim Position Report) show these ratios versus amplitude

for PGA and for several spectral velocities for three cases EPRI, LLNL 4GX and LLNL 5GX

("4GX" and "5GX" is a commonly used notation for LLNL results that contain 4 and 5 ground

motion experts, respectively. In the former case, Expert 5 is excluded, for reasons discussed

in the Interim Position Report.) We observe, first, that

(1)

these ratios are very different among the three cases; herein lies

our problem, of course. The LLNL results imply larger estimated uncertainty in

the estimates.

(2)

the ratios increase rapidly as the frequency decreases (for

reasons not yet completely understood, but related in part to the wide

divergence among low frequency spectral velocity predictions associated with

broad-band prediction schemes versus "random vibration" ground motion

prediction schemes).

(3)

there is a mild upward trend in the ratios with amplitude; it is

stronger for lower frequencies.

On the other hand, with care we can find a somewhat more consistent and helpful picture.

First, let us restrict our attention to amplitudes consistent with pga levels in order of 0.2g or less.

We shall find these levels are typical of the values of interest for DOE "high hazard" facilities

(2x10-4 mean annual probability of exceedance) in the EUS. The corresponding spectral velocity

levels are roughly 2 cm/sec, 6 cm/sec, 12 cm/sec, 25 cm/sec, 25 cm/sec for 25, 10, 5, 2.5, and 1

hertz respectively. Second, we focus attention on the 4GX results of LLNL (for reasons

discussed in the Interim Position Report). Then, under these restriction, we observe:

(1)

the ratios are about equal for the PGA and 10 hertz ground motion

parameters, having a value of about 3.0 to 4.5 for EPRI and 7 to 9 for LLNL. This

suggests that the high frequency end of the spectrum is rather consistently

estimated whether via PGA or via the 10 hertz spectral velocity.

(2)

the ratio of the ratios, i.e., the ratio of the LLNL ratio (4GX) to the

EPRI ratio is very consistent, namely about 2 over the entire frequency range, but

D-2