ML14138A080

Forwards Evaluation of SEP Topic II.4.A, Tectonic Province & II.4.C, Historical Seismicity within 200 Miles of Plant, to Be Used in Seismic re-evaluation of Facility

ML14138A080
Person / Time
Site:	San Onofre
Issue date:	09/16/1982
From:	Paulson W Office of Nuclear Reactor Regulation
To:	Dietch R Southern California Edison Co
References
TASK-02-04.A, TASK-02-04.C, TASK-2-4.A, TASK-2-4.C, TASK-RR LSO5-82-09-049, LSO5-82-9-49, NUDOCS 8209220253
Download: ML14138A080 (29)

Text

September 16, 1982 Docket No. 50-206 LSO5-82-09-049 Mr. R. Dietch, Vice President Nuclear Engineering and Operations Southern California Edison Company 2244 Walnut Grove Avenue Post Office Box 800 Rosemead, California 91770

SUBJECT:

SEP TOPICS II-4.A AND II-4.C FREE FIELD GROUND MOTION TO BE USED IN THE SEISMIC REEVALUATION OF SAN ONOFRE NUCLEAR GENERATING STATION, UNIT 1 (SONGS-1)

Attached is our evaluation of the seismic input to be used in the seismic reevaluation of SONGS-1 in the Systematic Evaluation Program. Our evalua tion confirms the acceptability of the 0.67g Housner Spectra which was forwarded by our letter dated April 5, 1982. However, the staff's best estimate of the 84th percentile spectra from the controlling earthquake would exceed the horizontal Housner Spectra by up to 10% in the period range from 0.07 second to 0.25 second and the vertical Housner Spectra by up to 10% in the period range from 0.05 second to 0.15 second. It is the staff's position that you should demonstrate adequate structural margin to accomodate this small exceedance.

By letter dated August 16, 1982, you provided additional information rela ted to the seismic safety margins in structures, systems and components considering these 10 percent exceedances of the Housner Spectra.

We are currently reviewing this information.

Aov' Sincerely, Walter Paulson, Project Manager Operating Reactors Branch No. 5 Division of Licensing

Enclosure:

As stated A

cc w/enclosure:

See next page OFFICE SEP PB BEB SEPB SEPB SEPB PYCh b1 TCheng.Mcenna Refm n WRussell F

SURNAME

.M U

8209220253 820916

~.PDR ADOCK. 05000206 V

4 P PDR OFFICIAL RECORD COPY USGPO: 1981-331-960

Mr. R. Dietch San Onofre Unit 1 Docket No. 50-206 Revised 3/30/82 cc Charles R. Kocher, Assistant General Counsel James Beoletto, Esquire Southern California Edison Company Post Office Box 800 Rosemead, California 91770 David R. Pigott Orrick, Herrington & Sutcliffe 600 Montgomery Street San Francisco, California 94111 Harry B. Stoehr San Diego Gas & Electric Company P. 0. Box 1831 San Diego, California 92112 Resident Inspector/San Onofre NPS c/o U. S. NRC P. 0. Box 4329 San Clemente, California 92672 Mayor City of San Clemente San Clemente, California 92672 Chairman Board of Supervisors County of San Diego San Diego, California 92101 California Department of Health ATTN:

Chief, Environmental Radiation Control Unit Radiological Health Section 714 P Street, Room 498 Sacramento, California 95814 U. S. Environmental Protection Agency Reoion IX Office ATTN:

Regional Radiation Representative 215 Freemont Street San Francisco, California 94111 Robert H. Engelken, Regional Administrator Nuclear Regulatory Commission, Region V 1450 Maria Lane Walnut Creek, California 94596

SEISMIC INPUT FOR THE REEVALUATION OF SAN ONOFRE NUCLEAR GENERATING STATION UNIT 1 Introduction This review presents final recommendations for free field seismic response spectra to be used in the reevaluation of San Onofre Generating Station Unit 1 (SONGS 1).

It differs in approach from the methodology used in arriving at spectra recommended for use in the 10 other SEP plants. The spectra for these other plants, located within the central and eastern U.S., were developed making heavy but not exclusive use of probabilistic techniques. A description of these methodologies can be found in NUREG/CR-1582 and in memoranda from R. E. Jackson to D. Crutchfield, June 23, 1980 and from R. E. Jackson to W. Russell, May 20, 1981. The uncertainty associated with the location and nature of seismogenic structures in the eastern U.S. indicate the need for such methodologies which incorporate the wide range of expert opinion in making rational backfitting decisions. Traditional "deterministic" techniques for the eastern U.S. utilizing tectonic provinces can result in wide variations in the level of seismic hazard or exposure associated with derived seismic input. In the western U.S. where the location of capable faults and their seismic'potential are more clearly defined, deterministic techniques can be used with much greater confidence in defining the seismic hazard. In the SONGS 1 review prime emphasis is placed on these techniques.

In this review much reliance is placed upon the extensive operating license, review carried out by the staff for the adjacent San Onofre Nuclear Generating Stations Units 2 and 3 (SONGS 2 & 3) and summarized in the Safety Evaluation Report (SER) NUREG-0712. The geologic, seismotectonic and most of the seis mological conclusions presented in the SONGS 2 & 3 SER are equally applicable to SONGS 1. This present review concentrates on additional refinement of ground motion estimates and placing these estimates in the context of the Systematic Evaluation Program.

08/12/82 1

SAN ONOFRE 1 SEISMIC INPUT

(a) Controlling Earthquake As discussed in the SONGS 2 & 3 SER the geologic feature of primary importance in the seismic evaluation of the SONGS site is the Offshore Zone of Deformation (OZD). This feature which comes within 8 km of SONGS 1, 2 & 3 at its closest approach was evaluated by various methods (slip rate, fault dimension and his torical seismicity) and it was determined that the maximum earthquake to be considered in estimating ground motion would have a surface wave magnitude (MS) of 7.0. Thus, the controlling earthquake for the SSE at SONGS 2 & 3 and for the seismic input to be used in the reevaluation of SONGS 1 is an MS = 7.0 earthquake at a distance of 8 km.

(b) Ground Motion In the SONGS 2 & 3 SER it was concluded that the SONGS 2 & 3 design response spectrum (DBE) anchored at 0.67 g exceeded the 84th percentile free field re presentation of the controlling MS = 7.0 earthquake at 8 km.

The 84th per centile is that level which the staff has found acceptable in its operating license reviews (see for example the Sequoyah SER).

This level of conservatism is based upon past practice exemplified in criteria such as the Regulatory Guide 1.60 spectrum and recommended revisions to the Standard Review Plan that deal with Site Specific Spectra.

Various techniques were used to estimate the 84th percentile of the controlling earthquake and since the SONGS 2 & 3 DBE exceeded all these estimates it was found to be acceptable. Because of the conservatism engendered by this exceed ance there was no need in the SONGS 2 and 3 review to define specifically what level of ground motion would represent a best estimate of the 84th percentile.

In this review we reevaluate the information presented in the SONGS 2 & 3 re view and more recent submittals by the licensee so as to arrive at a more pre cise realistic estimate of the 84th percentile of ground motion from the con trolling earthquake.

The licensee has proposed the use of the Housner response spectrum anchored at 0.67g for the purpose of the seismic reevaluation of SONGS 1. Although 08/12/82 2

SAN ONOFRE 1 SEISMIC INPUT

anchored at the same zero period acceleration this spectrum has lower spectral values than the SONGS 2 & 3 DBE at most periods. We will compare the licensee's proposed reevaluation spectrum with the best estimate of the 84th percentile and evaluate it in the context of the SEP.

Estimation of ground motion close to earthquake faults is a difficult and prob lematic task. We have examined theoretical, empirical and probabilistic esti mates of ground motion. The integration of these results assures a more reli able estimate than that which could be obtained by the use of any one technique.

In addition to the material presented and evaluated for the SONGS 2 and 3 SER and the 1981 Atomic Safety and Licensing Board hearing we have based our review upon the following additional submittals:

1. Southern California Edison Co., February 1982. Analysis of 2/3 g Housner Reanalysis Design Spectrum for San Onofre Nuclear Generating Station.

2. Southern California Edison Co., April 1982. Letter and attached reports from K. P. Baskin to D. M. Crutchfield.

3. Southern California Edison Co., June 1982. Comparison of 2/3 g Housner Reanalysis Spectrum with Multiple Regression Analysis of Spectral Values San Onofre Nuclear Generating Station.

4. Southern California Edison Co., August 1982.

Free Field Ground Motion Spectra SONGS 1, Letter and attached reports from K. P. Baskin to

0. M. Crutchfield.

Recommendation We find the proposed Housner reevaluation spectra, in general, appropriate, except for small exceedances in certain period ranges.

Specifically, the staff's best estimate of the 84th percentile spectra from the controlling earthquake would exceed the horizontal Housner Spectra anchored at 0.67 g by up to 10% in the period range from 0.07 to 0.25 seconds and the vertical 08/12/82 3

SAN ONOFRE 1 SEISMIC INPUT

Housner Spectra anchored at 0.44g by up to 10% in the period range from 0.05 to 0.15 seconds.

Theoretical Estimates of Ground Motion In support of the seismic reevaluation of SONGS 1, the licensee has submitted a series of theoretical studies whose purpose is the prediction of ground motion at the site from an earthquake caused by a rupture along the Offshore Zone of Deformation.

These studies (Del Mar Technical Associates, 1978, 1979a, 1979b, 1980a, 1980b) are described below and discussed with reference to the conservatism of the Housner reanalysis spectrum proposed by the licensee.

There are several major classes of theoretical earthquake models. They can be described as follows:

(Swanger and others 1980):

"1. Kinematic Models - These models assume that the entire slip and rupture history on the fault is known. The slip history on the fault is all that is required to compute the seismic radiation. The details of the slip history, which is important to the high frequency radiation, are sometimes included to satisfy some dynamic rupture constraint. Usually the details are motivated by a desire for mathematical simplicity.

2. Simple Dynamic Models - These models are analytic approximations describing some aspects of the stress release process. These models are normally used for interpretation of earthquake source parameters and for providing constraints on the characteristics of kinematic models.

3. Numerical Dynamic Models -

These models assume that the initial stress conditions are known. The dynamic rupture process is then modeled to.

obtain a slip history which can be used to compute the resulting ground motion. These models, even in the simplest cases, require extensive computations."

08/12/82 4

SAN ONOFRE 1 SEISMIC INPUT

"Economic constraints suggest that a routine model for simulating near-field ground motions will probably be kinematic. Simple dynamic models are generally not detailed enough to be used and numerical models are probably too expensive to apply routinely. These solutions for the rupture process are very important for determining what kinds of specifications of slip and rupture histories are truly valid physically" (Swanger and others, 1980).

For the SONGS 1 studies, a kinematic source model was assumed. The procedure for modeling ground motion was as follows:

1. Fault slip is characterized in terms of fault type, rupture velocity, dynamic stress drop (slip velocity at the onset of rupture at each point on the fault) static stress drop (fault offset) and duration of slip at each point. Random processes are included to approximate irregularities in actual earthquake rupture.

2. Propagation characeristics (Green's functions) are calculated for the particular earth structure, that is surface motions are computed for several hundred point sources along the fault plane. These earth response calculations include all wave types up to frequencies of 20 Hz.

3. Ground motion is calculated by convolving in time and space the fault slip characterization from Step 1 with the earth response functions from Step 2. By specifying hypocentral location, rupture extent and site location, the sensitivity to different source site configurations can be examined.

For the initial study (Del Mar Technical Associates, 1978) the model (partic ularly the slip function) was calibrated using the 1966 Parkfield Earthquake (MS = 6.0, ML = 5.8).

Prior to the October 1979 Imperial Valley Earthquake this was the most well recorded earthquake in the near field. In addition, the recordings from the 1940 Imperial Valley Earthquake (ML = 6.5, MS = 7.1) and the 1976 Brawley earthquake (MS = 4.9) were modeled. Utilizing subsurface knowledge of the SONGS site, P and S wave velocity, density, attenuation and layer thickness were computed. Green's functions were calculated to predict 08/12/82 5

SAN ONOFRE 1 SEISMIC INPUT

propagation characteristics from source depths extending to 15 km, out to epicentral distances of 60 km. The ground motion modeling centered about the effects of a 40 km long rupture at a distance of 8 km from the site. This is an approximate representation of an MS = 7.0 earthquake on the OZD. Sensitivity tests were conducted to test the effect of variations in site distance, fault length, fault location along the OZD (focusing), fault depth, hypocentral depth, changes in dynamic and static stress.drop, duration of slip, and changes in earth structure, upon estimated ground motion.

In the review of the initial 1978 report, the staff and its consultants (Dr.

Keiiti Aki, M.I.T.; Don L. Bernreuter, Lawrence Livermore Labs; Dr. Robert Herrmann, St. Louis University; and Dr. J. Enrique-Luco, University of California-San Diego) identified 3 main areas of concern. They were:

1. Numerical modeling procedure (mesh size) - concern was expressed that the grid of points (mesh) on the rupturing fault plane on which the Green's functions were calculated was too coarse and use of a finer mesh would have a strong effect upon estimated ground motion particularly at high frequencies.

2. Slip function - Concern was expressed that the proposed three-parameter slip function based upon initial slip velocity (dynamic stress drop) fault offset (static stress drop) and duration (rise time) may not be as appropriate as a simple two-parameter model.

In particular, the assump tion that dynamic stress drop was the same (500 bars) for all earthquakes modeled was questioned. This assumption was based upon the validation studies and physical arguments relating dynamic stress drop to magnitude independent material strength of rocks. It was of particular interest since changes in the dynamic stress drop were shown to have a strong effect upon ground motion in the frequency range of interest to nuclear power plants.

3. Sensitivity studies - It was felt that additional sensitivity studies were needed, particularly with regard to variations in earth structure and the exhibited sensitivity of the ground motion to fault depth.

08/12/82 6

SAN ONOFRE 1 SEISMIC INPUT

The licensee provided a revised model in Supplement 1, (Del Mar Technical Associates, 1979a). In response to the above concerns the following model changes and additional studies were presented:

1. Ground motion (response spectra) was computed for mesh sizes of 1.0, 0.5 and 0.25km (and in some cases 0.125km) for vertical and horizontal faults and line sources at different source-site configurations. Differences in spectra as a result of these size changes were found to be generally small and these changes did not reflect any consistent correlation between mesh size and spectral level.

2. Additional randomness (incoherent rupture) was introduced into the model to make it more resemble irregularities in actual rupture. This was done since coherent rupture produced excessive focusing in the direction of rupture, that is ground-motion resulted that was made up of essentially one large spike not seen in actual records. The randomness limited but did not eliminate the focusing possible from each 1 km cell (mesh size) and allowed for small perturbations in the time of rupture initiation, rupture direction, fault segment orientation and particle motion at the receiving station.

Comparison between randomized and non-randomized data showed better fits with randomized data. Computed response spectra were shown for mean and mean plus one sigma levels for the randomized studies.

3. A revised three-parameter slip function was developed that in the extreme case reduced to the alternate two parameter model.

These models were calibrated to the 1966 Parkfield and 1940 imperial Valley earthquake strong motion records and extrapolated to obtain site specific spectra at San Onofre. A comparison of the computed response spectra from the three parameter model with and without randomness and the two parameter model with randomness at San Onofre do not generally exhibit large differences at frequencies greater than 2 Hz. When they do, the licensee's preferred three parameter model with randomness is conservative.. It is also argued that while all models provide equally good fits to observed response spectra for the 1966 Parkfield and 1940 Imperial Valley earthquakes at high frequencies (greater than 2 Hz), the three-parameter function provides a better match with observed data at lower frequencies.

08/12/82 7

SAN ONOFRE 1 SEISMIC INPUT

4. Due to changes in the modeling procedure, sensitivity tests conducted for the original model were repeated along with the additional studies requested above. These studies confirmed most but not all of the previous assumptions. Some of the conclusions were:

a. Fault rupturing away from the site produces less ground motion than faults rupturing toward or by the site (focusing).

b. Increase in fault rupture length from 20 to 40 and 60 km has little effect upon high frequencies but has a strong effect on low frequen cies (less than 1 Hz).

This correlates with observed relative saturation of magnitude scales (ML and mb) dependent upon high frequency waves.

c. Hypocentral depth and fault depth were shown to have less influence on ground motion than was previously indicated.

d. Increased rupture velocity results in stronger ground motion.

e.

As the distance to the fault decreases gound motion increases.

f.

Initial slip velocity (related by the licensee to dynamic stress drop) has a major impact upon spectral level.

g. An increase in fault offset (related to static stress drop) results in increased low frequency motion.

h.

An increase in slip duration results in a small decrease in low frequency motion.

i. Ground motion is dependent in a complex way upon earth structure.

Increased attenuation reduces ground motion as expected but to a lesser degree than predicted by a simple application of theory.

In response to other concerns the licensee submitted (Del Mar Technical Asso ciated, 1979b) calculations and discussions relating to magnitude and moment 08/12/82 8

SAN ONOFRE 1 SEISMIC INPUT

estimates of the proposed numerical estimates of ground motion and estimated ground motion at distances greater than 20 km.

Utilizing a relationship between seismic moment and surface-wave magnitude, the MS of the hypothesized offshore earthquake was calculated to be 6.94. An ML of about 6 was calculated using the technique developed by Kanamori and Jennings (1978) to estimate ML from strong motion records.

The staff initiated a separate study carried out on the Illiac Computer by Sys tems, Science and Software (Day, 1979) to investigate slip functions. Making use of the unique capabilities of the Illiac, numerical dynamic studies were carried out to test the sensitivity of earthquake slip functions to fault geo metry, frictional strength and prestress configuration. Ground motion at dif ferent distances from the fault was not examined. It was concluded (Day, 1979) that slip function varies in a rupturing fault and that assuming it is constant can lead to artificially high focusing. One of the assumptions in most kine matic models (including the SONGS 1 study) is that slip function is uniform throughout the rupturing fault. As was discussed above the SONGS 1 study uti lized increased randomness to reduce this focusing. Another conclusion of (Day, 1979) was that initial slip velocity may not be independent of static stress drop as proposed in the SONGS 1 study. As was previously indicated, initial slip velocity related by the licensee to dynamic stress drop has a strong influence upon estimated ground motion at high frequencies.

The revised model used by the licensee in generating the proposed response spectra at the SONGS 1 site assumes a 40 km rupture maximally focused at the site with a fault offset of 130 cm and a rupture velocity nine-tenths the shear wave velocity. The slip function used is the three-parameter model with randomness.

Mean and 84th percentile spectra have peak accelerations of 0.31 and 0.37g, respectively.

The occurrence of an earthquake in the Imperial Valley in October 1979 provided an excellent opportunity to judge the adequacy and conservatism of the previous ground motion estimate. This earthquake of MS = 6.9 and ML = 6.6 occurring on the same fault (Imperial) that produced the 1940 MS = 7.1, ML = 6.5 earthquake resulted in approximately 31 km of surface rupture.

Rupture at depth was 08/12/82 9

SAN ONOFRE 1 SEISMIC INPUT

undoubtedly larger. It was a predominantly strike-slip earthquake with some vertical movement at the northern end of the fault and on the adjacent Brawley Fault. The fault and vicinity were heavily instrumented and provided the most extensive set of near-field ground motion recordings available at distances as close as several hundred meters. Aside from a difference in site conditions (the Imperial Valley is a deep alluvial valley) this event is strikingly similar to the proposed MS = 7.0 maximum earthquake on the OZD.

The theoretical model used to estimate ground motion for SONGS 1 was evaluated with respect to its ability to predict observed ground motion from the 1979 Imperial Valley earthquake in Supplement III (Del Mar Technical Associates, 1980b). While horizontal response spectra modeled-in the distance range of interest for San Onofre (approximately 8 km) showed fair agreement with those observed, several deficiencies were noted. These were:

1. Vertical ground motion at high frequencies was underestimated at all distances.

2. Horizontal ground motion at high frequencies (particularly peak accelerations) was overestimated for stations close (-.

1km) to the rupture.

3. Horizontal ground motion at high frequencies was underestimated for stations more than 10 km from the fault.

The licensee's consultant attributed these discrepancies to several factors, in particular to too much focusing of horizontal motion near the fault, and to too much independence allowed for treatment of P and S waves the principal components of vertical and horizontal motion. In order-to account for this, additional refinements were introduced in the model by:

1. Reducing the length allowed for purely coherent rupture to 50 meters.

2. Allowing for randomness of rake and dip in rupture.

08/12/82 10 SAN ONOFRE 1 SEISMIC INPUT

3. Utilizing similar smoothing functions (source density functions) for both P and S waves.

As a result of these changes, better fits were obtained to the observed data, particularly with respect to high frequency vertical and close-in horizontal ground motion. Sensitivity tests were carried out with respect to changes in the character of slip, inclusion of rupture along the Brawley Fault and proximity of the rupture to the surface. Some differences resulted, the most significant being an increase in ground motion for those stations immediately adjacent to the Brawley fault when this fault is also assumed to rupture.

Although this refined model produced better results for this earthquake than the Supplement I model, no comparison was made with respect to the Supplement I model predictions for the 1940 Imperial Valley earthquake, the 1966 Parkfield earthquake and the 1976 Brawley earthquake (Del Mar Technical Associates, 1979b) and additional events shown in Supplement II (Del Mar Technical Asso ciates (1980a).

Supplement II showed estimates of ground motion for the 1933 Long Beach earthquake and 1971 San Fernando earthquake based upon the original model and some, but not all, of the refinements introduced above. It is dif ficult to judge the relative validity of the Supplement I model and refined Supplement III model without a comparison of at least several different earth quakes. However, computation of ground motion at San Onofre using the refined Supplement III model provided an assessment of the significance of these differences with respect to estimation of ground motion from the occur rence of an earthquake on the OZD. These comparisons show rough equivalence of horizontal ground motion from both models.

Indifferent frequency bands a dif ferent model may be more conservative. With respect to vertical motion, higher ground motion is predicted at high frequencies utilizing the refined model as might be expected from the calibration with the Imperial Valley data.

The staff's consultants reviewed the various reports of the modeling study:

and provided comments, many of which resulted in the sensitivity studies discussed above.

Final criticism of the modeling study centered about:

08/12/82 11 SAN ONOFRE 1 SEISMIC INPUT

(1) Slip function - It was felt that the assumed dynamic stress drop or initial slip velocity may vary with earthquake size while the licensee argued that it was a function of rock properties alone.

(2) Randomness - There was concern over the increased use of randomness as the model developed while the licensee felt that this represented true elements of variability on both a microscopic and macroscopic scale.

(3) Q - Concern was expressed over the assumed value of Q, particularly in the Imperial Valley. The licensee acknowledged uncertainty with respect to Q but argued that it would have little impact upon the results for San Onofre.

(4) Rupture Velocity - Some concern was expressed that the assumed rupture velocity may be too high. The licensee has utilized a rupture velocity based on their judgement and validation studies of several earthquakes.

(5) Fit of the model to the Imperial Valley Data - Some concern was expressed that the fit of even the refined model to the extensive Imperial Valley earthquake data set was not good enough. The licensee argued that Supplement III only exmained selected stations and that the inclusion of other stations showed an even better fit:

Each of the consultants did not share all concerns. However, the differences between the consultants themselves and the licensee reflected healthy dispute among scientists in evaluating state-of-the-art work. The staff's view is that many of the issues raised are not resolvable given our present state of knowledge. We strongly support the view expressed by one of the licensee's consultants and one of the staff consultants, who perceive the modeling study as providing a framework for extrapolating and organizing data that is also better than a purely empirical approach since it allows the introduction of.

some physics. With respect to application of the results,,all the staff's consultants agreed that the modeled spectra proposed for San Onofre were good mean or average representations of ground motion from the earthquake on the OZD. There were differing views between the licensee and several of the 08/13/82 12 SAN ONOFRE 1 SEISMIC INPUT

consultants with respect to 84th percentile or some other representation that would take into account uncertainty.

The licensee argued that the modelled spectra were already conservative since the model assumed a rupture maximally focused at the site. The licensee then estimated the 84th percentile by taking the 84th percentile of the results cal culated allowing for assumed randomness in properties such as fault irregu larity. This 84th percentile was some 10 to 20% greater than the median.

Three of the staff's consultants suggested multiplying the model spectra by a factor of two, primarily to account for parameter variations such as a higher dynamic stress drop on the OZD. The staff observed in the SONGS 2 & 3 SER that even if one assumed this factor of 2, the model spectra were still exceeded by the SONGS 2 & 3 DBE at all frequencies. With respect to Unit 1, where a best estimate is required simple enveloping arguments are not appropriate. We re cognize that there may be model or parameter uncertainty not sufficiently taken into account by the licensee but find insufficient evidence that a factor of two is the best estimate of that uncertainty. Given the conclusion that the modeling estimate is a good mean or average estimate, and the previously mentioned observation regarding the modelling study as framework for extra polating data, we suggested that the licensee use the measure of uncertainty found in empirical estimates. The most recent and pertinent estimates indicate a factor of about 1.5 between the 84th percentile and the median (see the dis cussion below on empirical studies). When the median theoretical estimates are multiplied by 1.5 and the results are compared to the Housner reevaluation spec tra it is observed that the Housner reanalysis'spectra exceed this estimate at all periods except for narrow ranges between 0.06 and 0.4 seconds which vary as a function of model, component direction and level of damping. Utilizing the Supplement I model which the licensee views as the most reliable (Del Mar Tech nical Associates, 1980b) for the SONGS site, these exceedances reach maxima of 5% for 2% damping and 15% for 10% damping at 0.2 seconds for horizontal com ponents.

There is no exceedance at any period for vertical components.. A com parison was also made using an average of the Supplement I and the refined Supplement III model multiplied by 1.5.

The exceedances for the horizontal components were at slightly longer periods and were less than if one used the 08/12/82 13 SAN ONOFRE 1 SEISMIC INPUT

Supplement I model alone. The averaged vertical estimate, however, exceeded the vertical Housner Reanalysis spectra by up to 15% at 0.8 seconds. The ver tical exceedance is again due to the model being calibrated by some of the high vertical recordings in the Imperial Valley. As was noted in the SONGS 2 and 3 SER these high recordings are believed to be related to conditions peculiar to the Imperial Valley.

Empirical Studies In the empirical approach to the estimation of the ground motion at a site actual recordings of strong ground motion are used. Most of this data base is from the western United States and all of it has been recorded in the last fifty years. Most of the near field data for moderate to large earthquakes has been recorded in the last three or four years.

Ideally, in estimating the ground motion at a site, data recorded with specific conditions similar to those at the site are selected. The specific conditions to be modeled for the SONGS site are a controlling earthquake of magnitude MS 7.0 at a distance of 8 kilometers, occurring on a fault with predominantly strike-slip movement and site conditions characterized by deep stiff soil.

Because of the limited amount of strong motion data available it is often im possible to duplicate the specific conditions'for a particular site. In those cases, reliance is placed on the estimation of the specific site ground motion through statistical analysis, extrapolation and interpolation of the existing data base.

To gain confidence in the estimates that result from these analyses sensitivity studies, in which the parameters are varied to determine the effect of their uncertainty on the results, are performed.

There is no ground motion data base recorded at deep stiff soil sites in the near field from large (MS = 7) earthquakes on strike slip faults. The licensee has, therefore, used several different methods to analyze the strong ground, motion data base to obtain a best estimate of the 84th percentile of the ground motion at the SONGS site from the controlling earthquake. These approaches are documented in the licensee's submittals to the staff. In this section of the 08/13/82 14 SAN ONOFRE 1 SEISMIC INPUT

report these submittals are reviewed and comparisons of the results of the em pirical studies are made to the SONGS Unit 1, 0.67 g Housner, reanalysis spectrum.

As part of the studies performed for the operating license application for SONGS Units 2 and 3, Southern California Edison (S.C.E) presented on estimate of the ground motion at the site using empirical data. They collected all the then available high quality digitized and processed strong motion recordings from the western United States recorded at sites with geologic conditions similar to SONGS for magnitude approximately equal to 6.5. The MS of the earthquakes ranged from 6.3 to 6.7. The data, peak accelerations and response spectral values were fit to a regression curve. Curves were computed for the mean and 84th percentile of each period and extrapolated to 10 km.

This distance was used assuming the center of energy release occurred on a vertical fault 8 km away and at a depth of 6 km.

A response spectrum of ground motion for an MS = 6.5 earthquake was then con structed from these extrapolated values. A response spectrum for MS = 7.0 was estimated by multiplying the peak acceleration and spectral values by scaling factors. These factors were determined from several published ratios of peak accelerations at 10 km for MS 6.5 to 7.0 events and an empirical study of the effects of magnitude on spectral shape. Uncertainties are caused by the extra polation from magnitude 6.5 to 7 and the extrtapolation to 10 km. Some conserva tive aspects of the study are the use of a consi'derable number of recordings from reverse faults which tend to produce higher ground motion than strike slip events, (see "Estimation of Selected Response Spectral Values at San Onofre Nuclear Generating Station" by, TERA in S.C.E.,June 1982 submittal), and the use of distance to the center of energy release rather than the more commonly used measure, distance to the fault (see SONGS 2 & 3 SER). A comparison of. the SONGS Units 2 and 3 empirical spectrum with the 0.67 g Housner spectrum indi cates that the Units 2 and 3 spectrum exceeds the Housner at periods greater than 0.05 seconds.

Subsequent to the development of the SONGS Units 2 and13 instrumental spectral estimates a significant amount of high quality ground motion accelerograms has become available, most notably those from the pre dominantly strike slip Imperial Valley 1979 (IV-79) magnitude (MS) 6.9 earth quake.

The conservatism of the empirical estimate was attested to in the SONGS 08/12/82 15 SAN ONOFRE 1 SEISMIC INPUT

2 & 3 SER by comparison to the theoretical studies and the newly recorded data from the Imperial Valley 1979 earthquake.

As part of the SONGS Unit 1 reevaluation, S.C.E. has submitted several updated and revised empirical estimates of ground motion at San Onofre. Somewhat dif ferent approaches were used in each of these studies since the lack of a data set which duplicates the source and site conditions specified at SONGS 1 pre cludes a simple and direct use of the data. By comparing the results of the studies the significance of the different assumptions may be ascertained.

The first of these revised estimates was presented in the S.C.E. February 1982 submittal in a report by Woodward-Clyde Consultants (W.C.C.) entitled "Instru mental Response Spectra for the San Onofre Site."

As a result of their analysis of the IV-79 data, they found that data treatment using closest dis tance to the fault rather than distance to the energy source gave the best data fit. Therefore, the stiff soil data base used for the SONGS Units 2 and 3 ground motion estimates was reanalyzed using the closest distance to the fault definition. A statistical analysis was performed on the IV-79 data in the distance range 6 to 13 km. The results of the stiff soil data reanalysis, the IV-79 analysis and the peak ground acceleration regression have been combined to develop the revised instrumental response spectrum for the SONGS site.

This revised spectrum has a zero-period acceleration of 0.55 g based on the peak ground acceleration studies. In the period range 0.05 to about 0.3 seconds, the revised spectrum envelopes the data from the reanalysis of the stiff soils site data and the IV-79 data. At periods longer than 0.3 seconds, the revised spectrum envelopes the IV-79 results but it does fall below the data from the stiff soil sites. Strengths of this study are that it used the available stiff soil data set and it enveloped all of the data at periods less than 0.3 seconds.

A weakness of the study is the lack of sufficient justification for the develop ment of the resulting revised spectrum from the two data sets. The revised SONGS instrumental spectrum exceeds the 0.67 g Housner reanalyses spectrum by up to 15% in the period range 0.06 to 0.24 seconds for 2% percent and up to 6%

in the period range 0.105 to 0.25 for 10% damping.

An analysis of peak acceleration data alone, originally performed for SONGS 2 &

3 was also presented in Appendix A of the S.C.E. February 1982 submittal.

TERA 08/12/82 16 SAN ONOFRE 1 SEISMIC INPUT

Corporation (TERA) reported on the analysis of peak horizontal ground accelera tion data from worldwide earthquakes in the magnitude range 5.0 to 7.7.

As a result of their regression analysis they estimate that for an MS = 7.0 earthquake at 8 km the median peak horizontal ground acceleration is 0.33 g and the 84th percentile is 0.48 g.' Based on sensitivity studies they performed on this data, they report similarities in the level of peak horizontal ground acceleration recorded on rock or deep soil sites, larger than average peak accelerations recorded on shallow soils, larger than average peak accelerations recorded on steep topography or the tops of hills, larger than average peak accelerations from reverse faults and lower than average peak accelerations recorded in embedded structures.

The second of the revised spectral estimates appeared in the S.C.E. April 1982 submittal.

In this study by W.C.C. entitled "Development of Instrumental Response Spectra for the San Onofre Site" another approach was taken to obtain a best estimate of the 84th percentile level of the ground motion at SONGS. In this study the stiff soil data set used previously was not relied upon.

Instead, the extensive deep cohesionless soil data set from IV-79 was used to determine a base spectrum which was then modified to reflect the site conditions at SONGS.

To establish the base spectrum a regression analysis was performed on all IV-79 free field spectra from data recorded on deep soil sites at distances less than 100 km.

The estimated spectrum at 8 km was then compared to the spectrum obtained from the statisical analysis of IV-79 data in the 6 to 13 km range presented in the February 1982 submittal.

The spectrum based on the regression analysis exceeds or is equal to the spectrum based on the 6 to 13 km data.

Since the regression analysis spectrum was higher it was used as the base spec trum for a deep soil site condition at 8 km.

The ratios of rock to deep soil spectral velocities were then calculated for the distance range 20 to 30 km and for 8 km from the 1971 San Fernando (Ms =

6.6) earthquake (SF-71) where both rock and soil data were recorded. A ratio was developed between the one rock recording site for IV-79 at 26 km and several deep soil recordings in the distance range 20 to 30 km. The assumption that the rock spectrum at 26 km for IV-79 could be normalized to the soil spectrum at high frequencies was made. These ratios were used to estimate a rock to 08/12/82 17 SAN ONOFRE 1 SEISMIC INPUT

deep soil ratio appropriate for IV-79 at the distance of 8 km.

Using the differences in shear wave velocity between the SONGS deep stiff soil site, the deep cohesionless soil sites in the Imperial Valley and typical rock sites, a spectral velocity versus period ratio for deep stiff to deep cohesionless soil site conditions was estimated.

This spectral velocity ratio was then used to modify the IV-79 spectrum at 8 km to obtain a spectral estimate applicable to the SONGS site conditions. The strengths of this study lie in its use of an extensive data set record at similar distances from a similar size earthquake of similar source mechanics to the controlling earthquake postulated for the SONGS site. Uncertainties of this study relate to the assumption that ground motion for different foundation con ditions scales as the shear wave velocity of the material and to the assumption that the SONGS site conditions can be established in this way to be 20 to 40 per cent of the difference between deep soil and rock.

The resulting 84th percentile SONGS spectrum for 2 damping exceeds the 0.67 g Housner spectrum in the period range 0.065 to 0.185 seconds by up to 10% and for 10% damping the 84th percentile SONGS spectrum exceeds the 0.67 g Housner spectrum in the period range 0.12 to 0.185 seconds by up to 3%.

The third of the revised spectral estimates was presented in the S.C.E. June submittal in a report by W.C.C. entitled "Development of Instrument Response Spectra for the San Onofre Site. Through a series of regression analyses median response spectral estimates for rock and deep cohesionless soil sites were developed for MS = 7 at 8 km. These mediah spectral estimates were then interpolated to the SONGS site conditions and finally modified so as to estimate the 84th percentile level.

Following is a list of the steps involved in arriving at the final estimate.

a. Separate multiple regression analyses of 1053 peak horzontal accelerations were performed for rock and soil sites to develop median attenuation relations for peak ground acceleration (PGA).

08/12/82 18 SAN ONOFRE 1 SEISMIC INPUT

b. Separate multiple regression analyses for rock and soil data were carried out to determine median spectral shapes (normalized spectra) as a function of magnitude and distance. Data from the thrust fault SF-71 earthquake was excluded from the deep soil data base so as to better approximate the strike-slip conditions for SONGS.

c. The results from step a and step b were then combined to get a median spectral estimate for rock and soil sites at 8 km.

d. Interpolation to the SONGS stiff soil site condition was made in a manner similar to that done in the April 1982 submittal.

In this case, however, scaling was done using estimated high-strain shear moduli and shear wave velocities for SONGS, deep cohesionless soil and rock site conditions (see S.C.E. submittal August 1982).

Similar results were obtained by using relative peak acceleration relation ships from Seed and Idriss (1982) and assuming the SONGS deep stiff soil can be characterized as producing accelerations midway between deep cohesionless soil and stiff soil of thickness on the order of 150 to 200 feet. In addition a multiple regression analysis of all rock and soil site data was performed to obtain the relative response for SONGS conditions to deep soil conditions for other periods. This procedure was justified by the fact that three quarters of the data are from deep soil sites and one quarter of the data are from rock sites. This is considered the appropriate ratio for SONGS based on the previous interpolation for peak acceleration versus site condi tions. A result of this study is a SONGS site response which varies with period in a manner similar to that determined from peak acceldea tion alone.

e. The approach used to develop the 84th percentile from the median re sponse spectrum estimated for the SONGS site used variable dispersion as a function of period and magnitude. The use of variable dispersion is based on the argument that dispersion of empirical data typically decreases with increasing magnitude. Relative dispersion as a func tion of period was determined by combining results from multiple 08/13/82 19 SAN ONOFRE 1 SEISMIC INPUT

regression analyses of subsets of data in different magnitude and distance ranges. This relative dispersion curve was then anchored to a peak acceleration dispersion for magnitude 7 determined from an empirical magnitude dependent dispersion relation. (See also S.C.E.

August 1982 submittal).

The are uncertainties in the results of this study because the assumptions made in the scaling of the SONGS site conditions between deep cohesionless soil and soft rock are similar to those discussed above with respect to the April 1982 submittal.

In addition the extrapolation of high-strain near-surface moduli to depths on the order of 1000 feet (See S.C.E August 1982 submittal) has not been fully justified. Concern over the scaling between rock and soil site conditions is somewhat mitigated by the achievement of similar results through the analysis of the combined rock and soil data set (see discussion of TERA study below).

There is also some concern due to the separate analyses to determine median and 84th percentile estimates.

There does not appear to be sufficient data to clearly delineate the magnitude dependent dispersion curve.

However, the resultant dispersion for Ms = 7 is similar to that obtained by a single multiple regression procedure performed by TERA (see discussion below).

Comparing the 84th percentile level of the spectrum for 2% damping and 10%

damping from this study to the 0.67 g Housner spectra shows an exceedence of the Housner spectrum for 2% by up to 15% in the period range 0.072 to 0.235 seconds and for 10% damping the exceedence is up to 11% in the period range 0.1 to 0.275 seconds.

The fourth revised spectral estimate also appears in the June 1982 S.C.E. Sub mittal in a report by the TERA Corporation entitled "Estimation of Selected Response Spectral Values at the San Onofre Nuclear Generating Station. In this study a simple multiple regression analysis was performed to directly obtain median and 84th percentile ground motion estimates.

Multiple regression analysis of worldwide earthquakes in the magnitude range 5.0 to 7.7 was used to obtain the near source ground motion attenuation relationship for horizontal velocity response spectral values at periods 0.10, 0.12, 0.15 and 0.20 seconds for dampings of 2, 5 and 10%. To better approximate free field conditions 08/13/82 20 SAN ONOFRE 1 SEISMIC INPUT

recordings from large buildings or adjacent to dams were not included. Sensi tivity studies indicated systematic bias associated with data from reverse and reverse oblique faults and data from embedded structures. In addition, record ings from the tops and sides of hills or steep slopes were found to be generally, higher than those obtained at the bottom of slopes or on relatively flat ground.

Scaling variables were included in the multiple regression equation to account for bias due to reverse faulting and embedded structure. Topographically affected data were left in. For comparison TERA used a spectral ratio techni que (similar to that used by W.C.C) for estimating the spectral values and found that for a period of 0.10 second for 5% damping the direct multiple regression and spectral ratio technique result in median values which are within 3% of each other. The data set used in all their analyses contained data recorded on both rock and deep soil sites.

Based on sensitivity studies for site conditions they found no discernable difference in the residuals in the period range examined. They argue that the only possible exception to this would be for the site classification most similar to SONGS where the estimated values could be somewhat lower than their generalized regression.

The main strength of this approach is its simple direct regression which does not constrain the results by predetermined assumptions. A limitation of the study is the restriction of the calculations to a relatively narrow period range. The 84th percentile predicitons of the spectral value at the four periods exceed the 0.67 g Housner spectrum by no more than 19% for 2% damping and no more than 7% for 10% damping.

The S.C.E. August 1982 submittal addressed questions raised by the staffs review of the June 1982 submittal.

In addition it presented spectral. estimates at 10% damping for earlier submittals where these results.had not been previouly provided. The pertinent information contained in the August 1982 submittal has been integrated into the above discussions.

The estimation of ground motion for the SONGS site by empirical methods is not a simple problem given the current data base. S.C.E. has provided several methods of varying complexity which utilize differing assumptions in evaluating the empirical data.

As discussed above each of these methods has short comings 08/12/82 21 SAN ONOFRE 1 SEISMIC INPUT

and strengths. On the whole however comparison of the results of the various empirical techniques to the 0.67 g Housner Spectrum provides very consistent results (See Figures 1 & 2).

The consistency of these results with respect to the theoretical studies is discussed below in the conclusions.

Probabilistic Studies The licensee has carried out several probabilistic studies aimed at estimating the seismic hazard or exposure at San Onofre. Particular attention has been paid to the probability of exceeding the 0.67 g Housner reanalysis spectrum.

The most recent analyses, contained in the Southern California Edison February 1982 submittal were "Probabilistic Evaluation of Peak Horizontal Acceleration for San Onofre Nuclear Generating Station Unit 2" by TERA Corporaton and "Development of. Instrumental Response Spectra with Equal Probability of Exceed ance for the SONGS Site" by Woodard-Clyde Consultants. Both reports estimated return periods for 0.67g at the site as being on the order of tens of thousands of years, that is 10-4 to 10-s per year. Woodward-Clyde Consultants spectral estimates showed variations as a function of period with the highest probability of exceedance being between 0.06 and 0.2 seconds, reaching a peak of 2 or 3 x 10-4 at about 0.1 seconds.

Sensitivity studies were carried out for both reports. Woodward-Clyde Consultants estimates, for example, that taking into account the differences in ground motion between '"instrumental" values and those values that would be "effective in causing structural damage" and the variation in seimicity along the OZD would result in lowering the probability by more than an order of magnitude. We have no reason at this time to question the reasonableness of the TERA or Woodward-Clyde Co'nsultants results but past experience (e.g., Sequoyah SER, WNP-2 SER and review of the eastern SEP plants) indicates that large uncertainties do exist and probabilistic studies are best used in a relative manner where differences in assumptions can be minimized.

It is not appropriate to compare the results from the SONGS review to those gained from the review of the eastern SEP plants as was done in the Southern California Edison Report of February 1972, since there are many differences in the analysis and assumptions. The Woodward-Clyde Consultants do indicate, how ever, a definitely higher relative probability of exceedance associated with the same general period range of the Housner Reanalysis Spectrum that also has 08/13/82 22 SAN ONOFRE 1 SEISMIC INPUT

been determined as relatively deficient from non-probabilistic studies discussed above. It is our position that this indication of relative deficiency is an important result of the probabilistic studies.

Conclusions We have evaluated the theoretical modelling and empirical studies of ground motion for the SONGS site.

Each study has associated with it both strengths and weaknesses. The site-specific nature of the theoretical studies and their ability to make direct use of of physics indicate that the mean estimates of ground motion from these studies may be the best that have been presented. The difficulty in defining the uncertainty associated with the theoretical modelling studies is however greater than that determined for the empirical studies. We find that the final results, that is estimates of 84th percentile ground motion, are such that no relative ranking of significance is justified. We agree with one of our consultants who suggested that at the present time equal weight should be given to both modelling and empirical estimates.

Figure 1 and 2 show the ratios of the most relevant empirical and theoretical estimates of horizontal ground motion to the 0.67g Housner Reanalysis Spectrum at 2% and 10% damping. Over most of the period range the Housner spectrum exceeds the various estimates. The single highest exceedance is 19% at 0.12 seconds for 2% damping from the TERA estimate in'the Southern California Edison Report of June 1982. The average of theoretical and empirical studies indicate an exceedence of 10% or less between the period range of.07 and 0.25 seconds (see shaded area on Figures). The relative deficiency in this period range has also been indicated in the probabilistic studies. We estimate for example, that a 10% increase in the Housner Response Spectra would reduce the probability of exceedence by about a factor of two in this range. Only theoretical estimates of vertical ground motion were made. As discussed above one of these estimates showed exceedance of the vertical Housner Reanalysis Spectrum anchored at 0.44 g and this reached a maximum of 15% at 0.08 seconds for 2% d.amping.

Based upon the above comparison of theoretical, empirical and probabilistic estimates it is our position that while the Housner spectra are in general 08/12/82 23 SAN ONOFRE 1 SEISMIC INPUT

appropriate, our best estimates of the 84th percentile spectra from the con trolling earthquake would exceed the horizontal Housner spectra by up to 10%

in the 0.07 to 0.25 second period range and the vertical Housner spectra by up to 10% in the 0.05 to 0.15 second period range.

The licensee has argued that estimates of the 84th free field spectra from the controlling earthquake should be compared to instrumental versions of the Housner design reanalysis spectra. The instrumental versions are increased from the design version taking into account ductility and soil-structure interaction.

It is our position, taken in concurrence with the SEP Branch, that such factors, to the extent that they are applicable, should be considered in the engineering analysis phase of the review rather than in the seismological-input stage.

08/12/82 24 SAN ONOFRE 1 SEISMIC INPUT

References Day, S.M., 1979, Three Dimensional Finite Difference Simulation of Fault Dynamics:

Systems, Science, and Software, SSS-R-80-4295, Final Report Sponsored by:

National Aeronautics and Space Administration, Ames Research Center, December 1979.

Del Mar Technical Associates, 1978, Simulation of Earthquake Ground Motions for the San Onofre Nuclear Generating Station -- Unit 1, May 1978, Final Report.

Del Mar Technical Associates, 1979a, Simulation of Earthquake Ground Motions for San Onofre Nuclear Generating Station -- Unit 1, Supplement 1, July 1979.

Del Mar Technical Associates, 1979b, Earthquake Ground Motion Simulations for San Onofre -- Unit 1 Response to Proposed Task 4, September 1979.

Del Mar Technical Associates, 1980a, Simulation of Earthquake Ground Motion for San Onofre Nuclear Generating Station Unit 1, Supplement II, August 1980.

Del Mar Technical Associates, 1980b, Simulation of Earthquake Ground Motions for San Onofre Unit 1 Supplement III, August 1980.

Kanamori, H., and Jennings, P. C., 1978, Determination of Local Magnitude ML, from Strong-Motion Accelerograms. Bulletin of the Seismological Society of America, V. 68, pp. 471-485.

Seed, H. B., and Idriss, I. M., 1982, Soil Behavior During Earthquakes, Monograph, Earthquake Engineering Research Institute, Fall 1982.

Swanger, H. J.,

Murphy, J. R., Bennett, T. J.,

and Guzman, R., 1980, State-of the-Art Concerning Near-Field Earthquake Ground Motion; Annual Report September 1978 -

September 1979, NUREG/CR-1340.

08/13/82 25 SAN ONOFRE 1 SEISMIC INPUT

1.4 Empfrical Studies Basedobn June 1982 WCC 1.2 Based on April 1982 WCC 1sed on February 1982 VCC 1.0\\

0.8 4.

1Supp. I x 1.5 0.6 Theoretical Mode1ing'

's Su I

Supp.

IIIT

.xT 1 S tudies 0.4

-Based on June 1982 TERA 0.2 Darnping = 0.02 0

0.02 0.05 0.1 0.2 0.5 Period (sec)

Figure 1. Comparison of SONGS 84th percentile empirical.and theoretical spectra.to.Housner reanalysis spectrum.

Curves show ratios of individual study to Housner reanalysis spectrum.

Shaded portion represents.staff's best estimate of exceedance of Housner reanalysis spectrum by 84th percentile ground motion spectrum.

[-Empirical Studies Based on June.1982 1.2WC WCC Based on April 1982 WCC Based on Feb -1982 WCC 0~~1

0.

Supp. Ix1.5 Theoretical Modeling 0.6 -

Supp.

I Supp.

III x 1.5 Studies 0.4 as Based on June 1982 TERA 0.2 Darnping =0. 10 0.02 0.05 0.1 0.2 0.5

-Period (sec)

Figure 2.. Comparison of SONGS 84th percentile empirical.and theoretical spectra.to Housner reanalysis spectrum. Curves show ratios of individual study to Housner reanalysis spectrum.

Shaded portion represents.staff's best estimate of exceedance of Housner reanalysis spectrum by 84th percentile ground motion.

spectrum.