ML20247M685

Forwards Comments of Aki,Archuleta & Day on long-term Seismic Program at Facilities,Per 890301-03 Meeting in San Francisco,Ca

ML20247M685
Person / Time
Site:	Diablo Canyon
Issue date:	05/01/1989
From:	Savy J LAWRENCE LIVERMORE NATIONAL LABORATORY
To:	Rothman R Office of Nuclear Reactor Regulation
Shared Package
ML16342B573	List: ML20247M685
References
EG89-42, NUDOCS 8906050105
Download: ML20247M685 (10)

Text

-- -

", Lawrence Uvermore National Laboratory lg t_

j NUCLEAR SYSTEhis SAFETY PROGRAM May 1,1989

• EG89-42 i

Mr. Robert L. Rothman Structum! & Geosciences Branch Division of Engineering & Systems Technology Office of Nuclear Reactor Regulation Mail Stop 8 D22 .

U. S. Nuclear Regulatory Commission Washington,D.C. 20555 bearBob:

Please find attached the comments of Professors Aki, Archuleta and. Day on the Diablo Canyon Long Term Seismic Program Meeting of March 1-3,1989,in San Fransisco. '

Sincerely, g>

.Savy Engineering Geosciences Attachments cc: L. Reiter Professor Aki Professor Archuleta Professor Day An EqualcworturtyEttplanv e lhesttyof Castome

• R 80B Liamcre. Cattma 94550

• relepWe (415)4221100

• Twx 910-385-8339 UCLLL LWR

- - _ _ _ - _ bh[ hh2 ['

9'

'sS' -: , .

I s . %:

i DEPARTMENT OF GEOLOGICAL SCIENCES [f @

TnEPHONL (213) 743-2717 (e3-[

~

9' March 1989-1

!: . j Dr. Jean Savy l MS L-196-

.LLNL P.O.~ Box 808 L Livermore, CA 94550

Dear:

Jean:

This is my letter report on .the 1-3 March 1989 meeting on j ground motions of the Diablo Canyon Long Term Seismic Program.  !

My general impression of the meeting is that this was, by far, i the best meeting of LTSP because of the high quality of presen- l tations and well-focused discussions that followed. I enjoyed  !

both presentations and discussions very much. In the following, I shall state my positions on several key issues discussed in the meeting.  !

(1) Let me start with the selection' rule fo,r earthquakes used in empirical investigations. The choice of the magnitude-- i distance range is rather arbitrary, and needs some perturbation to see how sensitive the final result depends on the choice. . I '

am particularly interested in extending the magnitude range down to 6.0, so that the Parkfield and Morgan Hill earthquake

'are included.

(2) I am very much intrigued by the magnitude dependence-  ;

of the empirical standard error of PGA and Sa. At first, I thought that the artifact of data set, namely the dominance of intraevent data for larger M and dominance of interevent data f or smaller M was causing the apparent magnitude dependence. .

In that case, the small standard error adopted by P.G. and E. i may underestimate t.he true error-because of the small number of l- events used f or analysis. On the other hand, the analysis of n

Taiwan array data by N. Abrahmson, which is free from the i interevent-intraevent artifact, also shows a systematic trend of the decreasing error with the increasing M. Furthermore, as Steve Day pointed out in the meeting, such a trend may be expected from the physics of phenomena because seismic motion due to a large earthquake will have contributions f rom various  ;

parts of the f ault plane, and they tend to average and smooth the ef f ect with resultant less variable motion among earth-l-

I quakes than smaller earthquakes with distinct characters. In view of these arguments, I consider that the standard error used by P.G. and E. is acceptable. i 1

UNIVERs1TY OF SOUTHERN CALIFORNIA, UNIVERSTIY PARK. LOS ANGELES, CAUFORNIA 90089-0740 l

J

.: r y,. -

L- '

(3) With regard to the numerical modeling, I think that the e m p-i r i c a l source function is mis-named. This is not representing the source effect, but is simply empirical Green's function corrected for the propagation effect in a very crude manner I believe that'" reduced empirical Green's f unction" may be.a better name. Since the current name mpy cause'a ~

. serious mis-use of the method. I consider the, renaming.

important.

(4)'As I understand now, the potential earthquake from the Hosgri fault is modeled by 4x22 subevents distributed oger the fault pla'ne, each subevent occupying the area-3x4 km . The seis g 5x10gie moment dyne cm, of andthethe' subevent (Imperial seismogram was Valley modified aftershock) to have isa cornerThis.

f requency corresponding to the Brung stress drop of 50 stress drop matches the.3x4 km fault area to be l bar.

occupied by the subevent. This impl s that an average slip of about 10 cm: (5x10y/(3x10yphe, x3x4x10 subevent 0)).. has In order to scale the rise time, 17 subevents are delay-summed to give rise to the total rise time of 3.5 sec. This means that the total slip of 170 cm occurs within 3.5 sec, corresponding to the slip velocity of 50 cm/s and, in turn, to the local st ress drop of- about 50 bar. In this case, the moment scale factor c is very nearly 1, and there is no need for further modifying the subevent. ,

Thus, the final picture is a fault plane filled with 4x22 subevents each occupying 3x4 km2 area, with the slip-velocity of .50 cm's. and local stress drop of 50 bar. Since the stress drop varies from place to place considerably in their asperity

( model, the. strongest asperity may have local stress drop of a few c '-ed bars.

The above model is quite similar to models used for older

_(before 1979) earthquakes in California by others such as Hanks and McGuire, and Papageorgiou and Aki. Their validation result using-data from three recent earthquakes is encouraging and persuasive to extended it to the Hosgri fault.

(5) The deficiency of predicted low frequency response spectra due to the lack of near-field and surf ace wave contri-butions must be clearly stated with the precise statement about the frequency range in which the deficiency is significant.

The display of final result should also distinguish this uncer-tain part from the reliable part using, say, a dashed line.

(6) In the step-by-step explanation of errors (dispersion) in the numerical simulation, non-zero estimates of mean diffe-rence'between observed and predicted were mentioned f or the 2

~ ~ ~

~ - ' - ~ ~ - ~~ -------

jgfO is

• -e;

y .

1 > ,_ [ ;' , e r

l . ..'y three- validation cases. _ Although the difference was sma11 'and

~

'S underestimation by prediction,.this result;must'be. incorporated i n- :t he final error estimate for'the sake of logical

. cons i.s t e n c e .

.I'am~ enclosing claim for consulting services.

Sincerely-yours, Keiiti Aki

j1 enc 1.

O 4

4 i

S 3

. s v

STEVEN M. DAY

• DEPARTMENTOF GEOLOGICAL SCIENCES SAN DIEGO STA1E UN!VERSITY SAN DiEGO. CA 92182 (619)594-2663 or 594-5586 April 19,1989 Jean Savy

~

Lawrence Livermore National Laboratory P. O. Box 808 Mail Stop L-196 Livermore, Califomia 94550

DearJean:

The following is my report on the San Francisco meeting of March I through March 3 between the NRC staff and consultants and P.G. and E. The purpose of the meeting was to review the P.G. and E. responses to NRC questions raised during the review of the ground motion component of the Diablo Canyon long Term Seismic Pmgram. .

This was a very productive meeting,in which a large number ofimportant questions from the NRC staff and consultants were addressed in considerable depth. The staff and consultants of P.O. and E. did an outstanding job of communicating the results of this lengthy, complex, and innovative program. I will comment separately on the empirical and numerical components of the ground motion studies. -

Empirical Ground Motion. The empirical ground motion analysis has two components, regression analysis and statistics of the near-field data set, respectively. With respect to the ,

former,I see no problems. There was broad agreement between the Geomatrix regression results l and those of Ken Campbell. In particular, the following conclusions were common to both studies,in spite of somewhat different methodologies employed: i) Ground motion variance is about 0.4 natural log units, and is approximately frequency independent. ii) Thrust mechanism gives mean ground motion about a factor of 1.2 greater than strike-slip. iii) Regression of Sa/PGA is an appropriate procedure, given the non-uniformity of the data set in the magnitude-distance plane. iv) The vertical acceleration response spectrum is peaked at a considerably higher frequency that was the 1977 Newmark spectrum, and substantially exceeds the Newmark spectrum above about 8 Hz. In addition, the thrust / strike-slip ratio was further validated using the numerical model.

I also believe that the discrepancies between Professor Veletsos's analysis and the P.G.

and E. analysis of the near-field data set have been resolved, since the data set analyzed by Professor Veletsos was constructed for use in fragility analysis, and was deliberately weighted towards records with high values of Sa averaged over the 3-8 Hz band.

I still have a few reservations regarding the analysis of the data set of near-field recordings.

I recognize that, because the response spectrum derived from regression analysis envelopes the spectrum derived from analysis of the near-field data set, these reservations of mine may not

.~ .

significantly affect the conclusions of the LTSP, but I want to put them on the record anyway. The first concem is that the selection criteria for the data set were applied sufficiently rigidly to exclude the Morgan Hill and Parkfield earthquakes, but sufficiently permissively to include 5 records from the Imperial Valley. I don't find the logic for excluding Morgan Hill and Parkfield (rock sites, M>6), then adding Imperial Valley (soil sites, M>6.25) to be persuasive ( although I am somewhat reassured by Professor Aki's observation that the empirical attenuation exponent x0 is similar between sites on Franciscan geology and sites in the Imperial Valley). .

Secondly,I am not satisfied that topographic amplification of the Pacoima Dam recording of the San Fernando canhquake is established with sufficient certainty to justify the reduction factor which was applied. Anooshehpoor's thesis (U.C.S.D.,1988) presents three-dimensional physical modeling results which call this conclusion into question. In panicular, he finds that, depending upon angle of incidence, either amplification or deamplification is possible at the Pacoima Dam site.

All of the empirical studies probably substantially underestimate the true uncertainty in ground motion, since i) they reflect the dispersion of the residuals, but not the uncenainty in the regression coefficients, the latter being an increasingly imponant factor at shon distances (i.e., far '

from the centroid of the data), and ii) they operate on averages of the two horizontal components.

As I understand it, however, these two procedures are conventionally accepted by the NRC.

Numerical Ground Motion. Overall, I suppon the numerical modeling procedure being used, and I believe it is generally appropriate for frequencies above about 3 Hz. The method presented for quantifying the uncenainty ir, the numerical ground motion estimates was innovative, and represents a real step forward for the numerical approach. The conceptual separation of the dispersion into " random and modeling" error plus " parametric uncertainty" is intuitively appealing and informative. The fact that the composited dispersion estimated in this fashion agreed well with the observed dispersion, as estimated from the empirical studies, was a notable result.

The finite difference calculations of site effects at the Diablo site showed no significant topographic effect. These results would be more convincing if they were repeated for SV incident energy,instead of SH. This would be consistent with the SV calculation performed for the analysis of possible topographic effects at Nahanni Site 2. Simple physical arguments suggest that SV amplification should be the worst case; in addition, previous published numerical studies of topographic amplification show the SV case to generally produce more amplification than the SH case (Geli et al., BSSA vol. 78).

Sincerely, Steven M. Day

-- _ --- _ - _a

p. x .

,o .

7, QNIVERSITY OF CALIFORNIA. SANTA BARBARA cznxaxr . oms . mvmr , i.O. Aucrus . En=5mE

• MN MECO e MN FMNCnCO f SANh MRMRA . SAM CRUZ DEPART MENT OF GLOLOGICAL SCIENCES SANTA BARBARA. CALIFORNIA 93106 April 11,1989 Dr. Jean Savy Mail Stop L-196 Lawrence Livermore Laboratory

' P. O. Box 808 Livermore,CA 94550

- RE: Diablo Canyon Nuclear Power Plant Meeting of March 1,2,3,1989

Dear Jean,

, Without doubt this meeting was probably the most informative of any that I have attended regarding the Diablo Canyon Nuclear Power Plant (DCNPP). While it did not lay to rest some questions that I have, it did make much clearer what PG&E was doing to examine the seismic ground motion.

• Although Issue 11 of Harry Rood's letter of March 17,1989, addresses the issue of the response spectrum from the numerical simulations being deficient at the lower frequencies, I want to change the question a little. Paul Sommerville said that their target i frequency range was 3.0-8.5 Hz. What would happen to the peak acceleration if the low frequency (less than 3.0 Hz) were matched as well as the 3.0-8.5 range? I would guess that if the low frequencies wem included, the peak acceleration would be higher. The high frequency components would be riding on the low frequency components. I don't know how much of a difference it would make, but to assume that the peak acceleration is unaffected by the loss of the low frequency components seems wTong.

The second point about the low frequency components is that these components are the ones that will show directivity. The frequency-dependent radiation pattem used by PG&E will not alter these frequencies very much. I would expect that these components will show a directivity effect,i.e., an increase in amplitude in the direction of propagation, especially the east-west ground motion. Thus the lack oflow frequency energy m the numerical simulations probably has a significant effect on the results.

I would like to see a tabulation for the numerical simulations boken down by cases, e.g., unilateral strike-slip with Coalinga empirical source. The infonnation to be tabulated for each station am the peak ground acceleration, the aver.sge acceleration response spectra for each component (vertical, East-West, North-South). Each table would have a formatlike:

1

e

+

Case 1: Unilateral Strike-Slip, IV Empirical Source Station # PGA Avg. Response 3-8.5 Hz Avg. Response 5-14 Hz V E-W N-S V E-W N-S V E-W .N-S 2'

Total: V E-W N-S V E-W N-S V E-W N-S The reason for this request is that in reviewing PG&E's response to Question 18, I noticed that the Coalinga source almost always generates larger response spectral values than the Imperial Valley source. Also the East-West component is generally larger than the North-South. For the strike-slip scenario directivity would affect the East-West component. The final response spectrum is the average of all these components for all the simulations. I don't know if the NRC requires an average of components, but it seems that the North-South component is almost always less than or just equal to the East-West component. If the structures are more susceptible to East-West motion, then the average spectmm may be misleading. The averaging is producing a mediari spectmm, but I wonder what the individual averages look like for the different cases that were mn.

The fact that Coalinga generates larger spectral values leads me to wonder how representative the two empirical sources are. Is Coalinga a maximum? Is Imperial Valley a minimum?

The method of computing random realizations by using vadous stations distributed along strike seems clever to me. I don't know whether these simulations represent the true range of possibilities for a M 7.2 earthquake. The simulations represent a periodic spatial variation of slip along strike (see attached plot). There are other physical parameters that were not varied. There is only one rupture velocity used; there is only one rise time used; there is only one stress drop used. While the rise time and stress dmp are related to acceleration in an approximately linear fashion, the rupture velocity is nonlinearly related. The rupture velocity used by PG&E is almost constant, the three sigma factor practically guarantees it. Thus all the radiation comes from the slip distribution. A rupture velocity that accelerates or decelerates could produce accelerations equal to or greater than the slip (Bernard and Madariaga,1984; Spudich and Frazer, 1984).

I am still confused about how PG&E determined the number of subevents. Paul Sommerville presented a viewgraph showing that N t = T/ t = 3.3 / 0.2 = 17. Clearly the rise time of the 4 x 3 element is 3.5 seconds and the rise time of the individual subevents within an element is 0.2 seconds. However, my notes indicate that PG&E used t = 0.6 and not 0.2 seconds. The 0.2 seconds is appropriate for a 400 bar stress drop for the 23:19 IV aftershock. To get that stress drop to 50 bars as stated in response to Question

O 12 PG&E applied a Brune filter. I presume that they mean (I would like this made more 7, clear by having Pg&E show an example.) that b cause stress dr@ is proponional to comer frequency cubed (Brune,1970), to reduce 400 bars to 50 bars the corncr frequency

- of the IV aftershock was reduced by a factor of 2. If the original rise time was 0.2 seconds corresponding to 400 bars, then the new comer frequency is 0.4 seconds corresponding to 50 bars. In which case the number of subevents is 9 and not 17. The number of subevents becomes important because the slip of each subevent is the total slip divided by Nr. 'Ihus an element with 10 m of slip has 0.62 m with each subevent whereas ~

with 9 subevents the slip is 1.1 m. Each subevent is radiating a larger amplitude pulse which could lead to larger amplitude ground motion. I would like a very clear explanation of which numbers were used and why for both the IV aftershock and the Coalinga aftershock. I would like to see the displacement amplitude spectra for Coalinga from which PG&E deduced the corner frequency (rise time) and stress drop. The IV aftershock is based on the paper of Liu and Helmberger, but I don't recall seeing an analysis of the May 9 Coalinga aftershock. Does it have the same stress drop as the IV aftershock? the same corner frequency? the same moment? In other words, how does PG&E demonstrate that these two empirical source functions are similar?

Issue 3 of Rood's letter focusses on another issue that I have strong feelings about. Slip in the upper three kilometers is likely to produce significant surface waves that '

are not modeled by the incomplete Green's functions being used for the numerical simulations. Although PG&E provided many examples of slip distributions inferred from inversions of teleseismic data, these examples miss the point. How many crustal earthquakes with M 7.2 have failed to produce surface slip? How large might that surface slip'be relative to that at depth? PG&E indicates zero. If so, how does the upper three kilometers of earth manage to decouple itself from the rest of the earth? The examples shown by PG&E were for magnitude 6 type earthquakes except for Borah Peak which using their example indicates no stu-face slip, completely contrary to the field observations.

Issue 2 in Rood's letter is c;uite broad in its scope. Specifically I would like to see what happens to the attenuation re: ations when the distance criterion is relaxed by 5 and 10 kilometers. Similarly what happens when the magnitude range is allowed to encompass M 6.2, and M 6.1 canhquakes?

A criticism that is directed more toward the NRC is that we have not received any information about the hazard with respect to the style of faulting on th' 'losgri. PG&E is assumir.g 65% strike slip,30% oblique and 5% thrust. From a statistical point of view it may be satisfactory to weight the calculations this way. However, I doubt that nature will follow such a rule. The numerical simulations should not be averaged over all possible mechanisms. Each mechanism should produce its own average ground motion l

parameters. The fact that PG&E has homogenized the radiation pattern for the high frequencies,in essence, averages over fanit mechanisms. The only parameter that is different for the style of faulting is the 5 of Q thlt. I am going to have to think more about what effect the radiation pattem sm%r,iro used by PG&E has on the ground motion. Nonetheless, we on the ground mc *n yn:1 have only one viewpoint as to the probable style of mpture.

1 i Sincerel ,

Ralph J. Archuleta

7 .-.

i

. Clip ct 1.5 km

+

'.*.' 12 11 2

. 102

. e-s-

E 7 65 E

55 4i 3--

0 . . . .

0 20 40 30 80 100 Distance (km)

S!!p at 4.5 km i

12 10 -

e-E

~

6-

_a .

E 4-2-

0 . . . .

0 20 40 60 80 100 Distance Silp at 7.5 km <

12 10 -

s-E

~

6-

.. a .

=

• 4 2-0 . . . .

0 20 40 60 80 100 Distence l

Silp at 10.5 km 12 j- 11 -

10 -

92 82

~

E 7 "[

6-E 55 4i 3-21 1 1-O . . . .

0 20 40 60 80 100 Distance

- _ - _ _ _ _ _ _ _ . --__-_ - .-__.-____ _ ----__ _ _ _ - ___ -