Cloud Sampling

Page 1

This is an informal report intended primarily for internal or limited external distribution. The opinions and conclusions stated are those of the author and may or may not be those of the Laboratory. Work performed under the auspices of the U.S. Department of Energy by the Lawrence Livermore National Laboratory under Contract W-7405-Eng-48.

Page 2

This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency Portions of this document may be illegible in electronic image products. Images are produced from the best available

Page 3

LIVERMORE: Lawrence Radiation Laboratory SUBJECT: Cloud Sampling Among the more formidible problems facing the Plowshare Program is the development of our ability to evaluate the radioactivity content of the clouds released by nuclear cratering operations, ledge of this content may be inferred from an analysis of device per formance and fallout pattern data, For purposes of predicting potential hazard or treaty violation, the fraction of radioactivity resulting in "driftout" (long-range fallout) The only practical method of doing this is by direct Appropriate cloud sampling, however, im plies the avoidance of a number of pitfalls, some of which have been Such an analysis, however, leaves must be determined, sampling of the cloud itself. 1) Sampling at late times involves difficulties in: a. finding the cloud; b. formulation of proper sampling techniques which allow for variation of concentration Within the cloud and diffusion of the cloud; and Sampling at early times has certain advantages in that the easy to find and the concentration of material is high, allowing

Page 4

for greater analytical sensitivity. However, we again face the question of whether the sample(s) taken are truly statistically representative of the total cloud or "universe" sampled. In addition, for both the early and late time sampling, we must consider whether the sampling technique itself is valid. In other words, does the sample taken actually contain all of the radioactivity present in that volume of air which has passed through the sampler? In stating the previous two problems, an assumption of "perfect sampling" has been made. This assumption, in many cases, may be weak, especially when one considers that some sampling techniques are size selective, due either to the collection efficiency of the instrument used or to non-isokinetic sampling. Improper volumetric calibration of sampling instruments can also lead to serious errors. The first stated problem has been dealt with by Todd Crawford in his UCRL-14983, titled "Predicting and Sampling of Nuclear Clouds From the Viewpoint of Diffusion Theory." The second, or statistical problem, is discussed in the attached draft. A future memo will deal with the problem of sampling techniques.

Page 5

ON THE EARLY TIME SAMPLING OF "LUMPY" CLOUDS (A Cursory Statistical View) In past nuclear cratering events, cloud sampling has been accomplished primarily by use of aircraft-mounted samplers. Early time sampling has typically been performed by sending three or four airplanes through the cloud at different times to obtain a line integrated sample, of the relative volumes of the cloud and the sample taken, and assuming a cloud of homogenous composition, the sample content is then multiplied by the ratio of volumes to obtain the total cloud radioactivity. With only one sample taken at any given time, and no idea of the manner in which the activity is distributed within the cloud, homogeneity is assumed, since no evidence to the contrary is available. Such an assumption can lead to gross errors in the assessment of cloud content, as will be demonstrated. Since little or no information is presently available on the nature of distribution of activity in actual nuclear cratering clouds, we must seek elsewhere for this knowledge. To gain some insight into these phenomena, we have chosen to rely upon events of a similar nature to supply the nec One such series of events was the "ROLLER COASTER" Operation, events were designed to evaluate the hazard resulting from accidental non nuclear HE detonation of plutonium-bearing device components, produced were sampled and evaluated in detail, cloud concentration of oC bearing particulates, a grid of samplers was placed in the path of the cloud by use of a barrage balloon. The samplers used for the purpose were "Sticky It was assumed that the relative quantities of <X contamination To determine variation in The sampler arrange ment is shown in Figure 1.

Page 6

Early Time Sampling... on the cylinders is indicative of the variation in cloud content at the various grid points, and that each cylinder represented a line integrated sample along the X axis (axis parallel to wind direction) through the cloud. Of the five events comprising the ROLLER COASTER series, the Clean Slate I event was apparently successful in capturing the entire cloud; the Double Tracks event captured almost all of it, and the remaining three events were unsuccess ful. We can, therefore, use Clean Slate I and Double Tracks for statistical analysis of Of activity (Pu) distribution. Pictures of these clouds at time of grid intercept are shown in Figures 2C and 2D. Results of analysis for of contamination are shown in Figures 3C and 3D. These results are expressed in terms of /i /i amperes, which, in this case, is directly proportional to Pu content. On the same charts, we have drawn in an approximate outline of the visible cloud (dotted lines) scaled from the cloud photographs and an outline of the "radioactive" cloud (solid line) based on the one /i a isometric line. A statistical analysis of cloud content, and its variation within the clouds was performed by tallying the results for all samples taken within the "radioactive"clouds, as shown in Table 1. Figures AC and AD present histograms of these tallies. Since the histograms indicate the sample results to be ex tremely skewed towards the low readings, one may conclude that the statistical variation is not normal. To obtain, a better spread in the results, a compilation of the data by log increments was then performed (Tables 2C and 2D). Logprobability plots of these data are given in Figures 5C and 5D. Chi-square analysis indicates that a log-normal distribution characterizes the data better than does the Normal, Binomial or Poisson distributions. The chi-square tab ulations for the log-probability distributions are given in Tables 3C and 3D. A summary of the analyses is given in Table A. For log-normal distributions, the standard error of the mean (SE“) is where 6”g » the geometric standard deviation of the log-normal population, and

Page 7

The 68% confidence range for any given set of data therefore extends of 0.68 that the true X of the universe sampled will fall within a range de fined by the calculated X times or divided by SE-. or, to state it another way, there is a probability Lower 68% confidence level “ SEX The range may be stated as X x/t As an example, suppose a setSE- . (8 x/t 2), then there are roughly two chances out of three that any other set of samples taken from the same universe would indicate an X within a range extending from 4 to 16. The 95% confidence range would be X x/t (SE^) , so that in the above example there would be a 0.95 probability that any group of samples would yield an X within a range extending from 2 to 32. Figure 6 shows the relationship between SE- and N for log-normal clouds of various (fg. This relationship enables us to determine the number of samples required to attain any desired confidence level. To apply this to nuclear cratering clouds, we must make the following 1) The distribution, of radioactivity within the cloud is log normal, as apparently was the case with the ROLLER COASTER 2) Samples are taken on a random basis within the cloud. Some knowledge of tf’g, the geometric standard deviation, Now, assuming the (fg of a nuclear cratering cloud is about the same as that of the ROLLER COASTER clouds ( <f g *’ 10), we see that, with only one sample taken, there is roughly two chances in three of being representative of the true mean concentration within a factor of 10. Ten samples would re duce this range to a factor of 2.2 and 100 samples would bring it down to a

Page 8

Based upon this very cursory analysis, two things seem to be clear. The first is that we need to take more samples in order to characterize cloud content to any reasonable degree of accuracy, the whole matter of statistical analysis is of sufficient importance that it should be pursued jointly with a statistician versed in sampling The second is that

Page 14

CLEAN SLATE I DOUBLE TRACKS Median activity per sample ^tyua) Mean activity per sample (ppa) Median concentration fyi a/ftJ) Visible Cloud Volume (ft"*) Number of Observations, N Based upon following assumptions: 1) Each sticky cylinder has a cross-sectional area ■ 1.375 in 3) Sticky cylinder has 50% sampling efficiency Therefore, each cylinder samples

Page 16

fcI Figure 2C. Two views of CLEAN SLATE No. 1 cloud at time of balloon

Page 17

Figure 2D. Two views of DOUBLE TRACKS cloud at time of first balloon intercept. (Sandia Corporation photos) f

Page 18

Lawrence Radiat ion Le:>. ra: FROM: Jerry J. Cohen In the above memo, the chi-square analyses shown in Tables 'C and 3D were performed by an invalid procedure and are therefore not correct. Please remove and destroy these tables and attach this correction. Subsequent analysis indicates that a log normal distribution for the given data is, at best, unlikely. Accordingly, we can only consider the statistical treatment as having "normalized" the data to the indicated distribution. This serves only as a first approximation until better information as to the true nature of the distribution is available. A further study of the data is presently being performed to identify this distribution more precisely. However, because of the high variance in this data, the conclusions reached in the memo still appear to be

Page 19

KEUFFEL 4 ESSER CO. MADE IN U.5.A» ) ssafeafefe Sfea as fefe *iM{iPjte pte feffe g|in jfejfel fenvux