ML20099J522

Applicant Exhibit A-151,consisting of Pages 3-2 Through 3-5 of Undated Rept Re Weather Sequence Sampling Method

ML20099J522
Person / Time
Site:	Limerick
Issue date:	05/22/1984
From:	AFFILIATION NOT ASSIGNED
To:
References
OL-A-151, NUDOCS 8411290101
Download: ML20099J522 (6)

Text

.

O !ltw v, O 3.1 ' Neather Sequence Sampling Method 2

n V

/MsV e -

The atmospheric dispersion of radioactive material froup l

4 u

p d[7 e,

g D]jj

+

, postulated 1ccident depends on the weather from the start o d g b; the accident through a period of tens to hundreds of hours M g Q following the accident. The character of the accident toget g 4/

with the weather accident determines coincident with and the transport anddispersion immediately following process that the % M ~i "

follows, and thus, the magnitude of the consequences that will result. Since the weather that could occur coincident with the accident is diverse, representative weather data sequences are selected as input to the dispersion model to reflect the depen-

-dence of the transport and dispersion process on the site wea-ther. The selection process is done by means of sampling tech-niques from a full year of hourly weather data characteristic of the plant site. The CRAC2 model allows a choice between four sampling techniques: (1) random sampling: (2) stratified ran-dos sampling: (3) stratified sampling; and (4) importance sampling.* Whatever sampling technique is chosen, the goal is to realistically represent the distribution of dispersion model results as a function of the site characteristic weather.

The first three of these sampling methods were included in the original CRAC model. A description of these three methods can be found in Section 13 of Appendix VI of the Reactor Safety Study [1]. The sampling method recommended for use in CRAC, and OusedinmostCRACapplications,isthestratifiedsampling i method. The stratified sampling method ensures a complete j coverage of diurnal, seasonal, and four day cycles without the i

statistical noise-of methods that utilize random sampling (1].

Sensitivity studies performed using CRAC indicate considerable l variability in predicted results attributable to sampling by l this method, however (2]. The importance sampling method available in CRAC2 greatly reduces the variability due to sampling observed with any of these three other techniques.

The basis of the importance sampling method is an initial assessment of the full set of hourly weather data. This initial assessment provides information about the types of weather sequences contained in the data and the frequency of these wea-ther types. With this information, weather sequences can be l' sampled to reflect the full year's weather data. This ensures representation of each type of weather sequence, those important to realistic representation of the weather data set, and those important to the occurrence of the most serious accident conse-quences.

^

I The sampling methods described in this section are implemented in subroutines BINMET and RANBIN and parts of MAIN and DAMAGE.

O 84}ih 0 '

g 4-2 L

O O

1

' $g

_\ q.(, f,

\

s s. ' ' N s k \

e 'g &s

, 's

\ 4 ',, S, L **g 6

- % vs%v *%

i 1 , The weather data assessment is done by sorting it into t'

-weather categories, categories that provide a realistic

. representation of the year's weather without overlooking those kinds of weather that are instrumental in producing major consequence impacts. A set of 29 weather categories has been

, selected for the CRAC2 model to reflect these requirements.

Heuristic judgment played a significant role in the choice of the 29. categories into which the data is sorted. Experience with the CRAC model revealed the impact.of weather events on L the consequence magnitudes resulting from the accident. Given a postulated large accident. large numbers of early deaths and injuries are normally associated with relatively low probability weather events such as rainfall or wind speed slowdowns within 30 miles of the plant site or with stable weather and moderate

wind speeds at the start of the release. In CRAC2 these weather data types have been selected to be among the 29 categories utilized in the assessment process.

The 29 categories are described in Table 3.1-1. An example of weather data sorted into these categories is shown in Table

3.1-2. The weather data for this example represent one year of meteorological data for the City of New York. The entire year of data, 8760 hourly recordings, are sorted into the 29 weather

- categories. Each sequence is examined to determine (1) the i first occurrence of rain within 30 miles of the site, or (2) the first occurrence of a wind speed slowdown within 30 miles of the accident site, or (3) the stability category and wind' speed at the start of the sequence. The first of these condi-tions that is satisfied by the sequence determines the weather i category to which it is assigned. Following the assessment process, the start hour of'each weather sequence will have been assigned to one and only one weather category. Each of the weather categories then includes a set of weather sequences representing the corresponding weather type. The probability '

i of occurrence of that weather type is the ratio of the total 1

number of weather sequences in the category to the total number of sequences in the year's weather data set.

l l- The sampling procedure now has two key items of information l . available to it: (1) the category of each weather sequence, and (2) the probability of occurrence of each category of weather.

A sample consists od a set of weather sequences selected from each of the categories. Normally, four sequences are selected i from each category by the " Latin hypercube" sampling scheme l- (3]. .Nith this sampling method, random samples are drawn from i

sets evenly spaced within the weather category. This assures that the model uses an even representation of the weather data over the full year. Assume that a weather category contains N g

weather sequences and that K g of the sequences are to be selected as samples. 0<Kg1Ng. The Ng weather sequences are then grouped into Kg evenly spaced sets Sg. .... S g.

3-3 -

i l

I

.d.,,-r- .w~~,, .n,.._my,.- _,_ ,

. l Set S) contains j

\ 'f *

• K l -

(3-1)

Qi) . .

("i) .

weather sequences **. One weather sequence is'then randomly colected from each set. Since the total number of weather i sequences selected from category i would be K g , the total

{

number of sequences selected from all 29 of the categories would be 29 Ei i=1 The assigned-probability for a meteorological sequence sampled from category i would be

"i/ El 29 E "i i=1 i

A i

• The notation . -

j

~

l fy) means the largest integ'er contained in the number jl g .

\

Y l **Since the Ngweather sequences of category i have a natural order i

determined by the initial time of each of the weather sequences, the evenly s Sg are ordered; i.e., Si S t./K ....

j consists of' paced tLe first sets.:N .] ele nts of category i. 5 consists g 3 2 of the next 2(Ng /K g) - N g/Kg elements of category 1, and so on. - -

O 3-4

• Nww -tyw-- -mw9--+.+y-se----e9%9ye i% . e w &S +w m ,w _ -

_99,us'h w

• w es--t w w wen-emm es -. -.

Consider a simple example. Let category i contain 10 weather sequences from which four are;to be sampled. Then Ng = 10, R, K g = 4, and S y contains two sequences, S 2 contains three sequences S contains two sequences, and S 4 contains three 3

sequences. One sequence is randomly drawn from each set S),

j = 1, ... 4, as in,the figure below.

I  % @l  %%@l@ % l%  % Yl 34 S1 S2 33 The' assigned probability for a sequence chosen from this

' 19.41, category would be , since CRAC2 requires the year's 8760 weather data to contain 8760 sequences.

+

The technique of importance sampling described here selects l-weather sequences that accurately represent the range of weather sequences in the weather data and their probability of occur-rence, and assures selection of sequences that yield severe consequences. The inclusion of these severe accident conse-quences and of weather sequence probabilities representing each i

O category is key in the realistic representation of the proba-bility distribution function of consequences. The technique is simple and does not require significant additional computation time compared to other sampling methods.

References for Section 3.1

1. WASH-1400 (1975), Reactor Safety Study. Aooendix VI:

Calculation of Reactor Accident Consecuences, NUREG75/014, i

US. Nuclear Regulatory Commission.

l- 2. Ritchie, L. T. . Aldrich, D. C. and Blond, R. M. (1981),

" Weather Sequence Sampling for Risk Calculations "

Transactions of the American Nuclear Society, Vol. 38.

l 3. Iman, R. L. and Conover, W. J. (1982), Short Course on l Sensitivity Analysis Techniaues, NUREG/CR-2350, SAND 81-1978.

I l

l l

e

1. .

O

'3-5 ,

i r

-