2.
ENVIRONMENTAL SAMPLING AND ANALYSIS
Sampling Techniques
This chapter describes and evaluates the LLNL environmental sampling program. The assessment is based on a 1.5d visit in October 1993 by two committee members to review the environmental laboratory directed by W. L. Robison at LLNL, on evaluation of data provided directly to the committee by W. L. Robison, and on evaluation of the procedures described by W. L. Robison in his publications (see, for example, Robison et al, 1993).
Description of the LLNL environmental sampling program
The LLNL environmental program has sampled soil, vegetation, water, and foodstuffs on several islands of Rongelap Atoll and numerous other islands and atolls in the Marshall Islands. These measurements provide information on the radionuclide content of components of the projected Rongelap diet and on material, such as soil, that might be inadvertently ingested. LLNL has also estimated wholebody exposure from external sources of radiation. Our review focuses on soil sampling because soils are the major source of exposure to plutonium and on the sampling of three foodstuffs—coconut, breadfruit, and pandanus fruits—because these are the primary sources of cesium137 exposure and are major components of both the importedfoodsavailable and localfoodsonly diets (W. Robison et al. 1993). Inhalation and ingestion via water do not appear as significant sources of internal radiation exposure.
The nature ofthe sampling program has changed over time. The initial 1978 sampling program, which complemented the EG&G Corporation's aerial gammaray spectroscopy survey, was undertaken to characterize the distribution and amounts of radionuclides on many of the northern Marshall Islands, not just those of Rongelap Atoll (Robison et al., 1993). That characterization sampling was not designed to estimate the statistical distributions of radionuclides on the island; instead, the focus was on estimating the radionuclide concentrations in selected soils and vegetation and the depth profiles of radionuclides in the soil. On Rongelap Island, soil and foodstuff samples were collected at arbitrary locations along the length of the island. Later trips used a variety of sampling designs to characterize different aspects of radionuclide distribution on Rongelap. Some samples were collected by walking along transects; others were collected at arbitrarily chosen locations. Finescale spatial variation within a 1 x 1m area was sampled in one region of the island with relatively high activity. The region around the village on Rongelap Island was extensively sampled. Such characterization sampling provides information about the locations actually sampled, but there is no statistically reliable method to draw conclusions about unsampled areas from such sampling.
In the last few years, LLNL has conducted systematic sampling from a 100m sampling grid, illustrated in Fig. 4, that covers the vegetated area of the island. The grid was oriented
parallel to the rows of trees in the coconut plantation, and the grid origin was defined as the northernmost coconut tree. For each point in the systematic sampling grid at which there was a coconut, breadfruit, or pandanus tree, a soildepth profile was determined by digging a soil pit and collecting soil from depths of 05 cm, 510 cm, 1015 cm, 1525 cm, 2540 cm, and 4060 cm. Deeper soil fractions were not sampled, because observed activities in the samples from 4060 cm were low. In the case of a few grid locations, no foodstuff was nearby and no soil profile was collected at these points. If there was a coconut, breadfruit, or pandanus tree close to the sampling point, nuts and fruits were also collected. Because of the widespread planting of coconuts across the island, coconuts were usually found close to the designated sampling points, but breadfruit and pandanus trees are less frequent and more scattered across the island. The staff that performed the field sampling appear to have taken considerable effort to follow the systematic sampling plan. On the committee's visit to Rongelap, a number of sites were observed where access paths were cut through the thick scrub to reach sampling points. In Robison's opinion (W. L. Robison, personal communication, 1993), all breadfruit and pandanus trees in the vicinity of the deserted village on Rongelap Island have been sampled.
Evaluation of the LLNL environmental sampling program
In general, the goal of any statistical sampling program is to draw conclusions about some general population from a set of observed values. A statistical sampling program can be evaluated by considering four questions:

What population is being described and is that choice appropriate?

Are sample locations chosen with an appropriate technique?

Is the number of samples sufficient to obtain the desired precision?

Are appropriate techniques used to draw conclusions from the results of sampling?
As noted above, there have been two sampling programs on Rongelap: the initial characterization sampling at arbitrarily chosen positions and a more recent systematic sampling based on a 100m grid. The characterization sampling cannot be used to draw conclusions about unsampled locations, because the probability that an individual location or source of vegetation was sampled is unknown and arbitrary. We focus on the systematic sampling because it can be used to draw conclusions about unsampled locations. We consider soil sampling and foodstuff sampling separately because their goals are slightly different.
Soil sampling
The goal of the soilsampling program that was conducted by LLNL was to draw conclusions about soil radionuclide concentrations at any location on Rongelap. It should be noted, however, that the only soils sampled were from locations in the vicinity of foodstuff trees. Such a protocol is conservative and is likely to overestimate soil transuranic concentrations, inasmuch as there is some evidence that the unsampled areas (between trees) are likely to have lower concentrations of transuranics than the sampled areas (W. L. Robison,
personal communication, 1993).
The preferred method to draw a systematic sample is to choose a starting point randomly. The LLNL grid was started at the northernmost coconut tree. Because the island has an irregular shape, that can be considered as an acceptable choice of starting point and equivalent to a random start.
There are two separate issues concerning contributions of soil radionuclides to the total dose: the exposure from ''average" conditions and the probability that there are local regions of high concentrations of radionuclides, commonly referred to as "hot spots." The estimated contributions of soil ingestion and inhalation to the total dose are considered small—1.6% for all isotopes (Tables 6 and 12 in Robison et al., 1993). The concern with hot spots is that a chance encounter with an unusually high concentration of radionuclides might contribute a substantial dose; this issue is of concern mostly for plutonium and americium isotopes because of their longevity, although their contribution to the total average dose is normally very small. In the context of this report and for the form of radionuclide contamination on Rongelap, a hot spot refers to the possibility of finding a small area with substantially higher radionuclide concentration than the measured average, not to a hot particle as might occur from fallout from a reactor accident. In its deliberations the committee has arbitrarily defined a hot spot to be a small region with radionuclide concentration of more than 10 times the measured average.
The probability that a hot spot is sampled on a systematic sampling grid depends on the size and shape of the hot spot and the spacing of the sampling grid (Gilbert, 1987). Given a single circular hot spot of radius r and a square grid of spacing s, the probability that a grid location falls within the hot spot can be derived to be
for s > 2r. Hotspot sampling probabilities for a selection of grid spacings and three radii are given in Table 21.
The 100m grid used on Rongelap Island has a moderate probability of sampling a large (28m) hot spot, but a very low probability of sampling smaller hot spots. However, an extremely fine grid would be needed to sample small (10m or 5m) hot spots reliably. For a given area to be sampled, the number of samples required increases as grid spacing is reduced. For example, a 20m grid requires 25 times as many samples as a 100m grid. The probabilities in Table 21 were derived for circular hot spots; the detection probabilities are different for square or rectangular hot spots, but the shape of hot spot does not affect the basic conclusions of the sampling program.
Any reasonable sampling grid is unlikely to detect a very small, very hot area that might be randomly placed on Rongelap Island. However, two considerations suggest that the probability of there being such a hot spot is small: the BRAVO fallout precipitated from a relatively high elevation and was finely dispersed. One statistical consequence is a smallscale spatial correlation between radionuclide concentrations in nearby samples (this is discussed in greater detail later).
Table 21. Probability That a Single Circular Hot Spot of Selected Radius Is Sampled by a Grid of Selected Spacing
The previous analysis of the probability of sampling hot spots did not incorporate any information from the observed amounts of radionuclides in Rongelap soil. A second approach to evaluating the probability of there being hot spots is to specify a radionuclide concentration that defines a hot spot, and then ask, what is the probability that there might be such a concentration in a small randomly chosen area based on observed distributions. The committee agrees that a concentration 10times the observed mean level is a reasonable definition of a hot spot. For Rongelap Island, this level of concentration for plutonium would be about 35 pCi/g. The observed data for soil concentrations can then be used to estimate the probability that the soil plutonium concentration in a small area might exceed 35 pCi/g.
If soil plutonium concentrations were known to be normally distributed, the desired probability of a given concentration can be calculated from the cumulative distribution function of the normal distribution. However, probability plots of the soil plutonium data (e.g., Robison et al., 1993, Fig. 10) suggest that neither a normal nor a lognormal distribution is appropriate to estimate probabilities in the extreme upper tail of the distribution. One solution to this problem is to find a transformation of the data that optimally fits a normal distribution, then estimate probabilities using the transformed distribution. One broad class of transformations that can be used is the class of power transformations:
A log transformation corresponds to λ = 0, the squareroot transformation corresponds to λ = 0.5, and no transformation corresponds to λ = 1. The optimal transformation to normality can be defined as 1) the λ that maximizes the correlation between the transformed data and normal order scores (Stoline, 1991), or 2) the λ that produces transformed data with a skewness closest to 0. These approaches are used with two samples of soil plutonium data: the 22 observations from the systematic sampling grid and 110 observations from mapped locations across Rongelap.
Because the mean and standard deviation are estimated from the data, there is some sampling variation in the estimated probabilities. This sampling variation is smaller in larger samples. An upper confidence bound on the hot spot probability can be calculated using tables based on noncentral t distributions (Owen and Hua, 1977).
The optimal transformations to normality are different for the two groups of data. Both the correlation and skewness criteria indicate the same choice of transformation: λ = 0.8 is the optimal transformation for the grid data, while λ = 0.4 is the optimal transformation for the larger data set (Table 22). Both sets of data indicate that the probability of a hot spot is extremely small, 1 in 100,000 or less. Upper 99% confidence bounds on the hot spot probabilities are also small, 40 in 100,000 or 13 in 100,000, depending on the choice of data set (Table 22). The exact values of the hot spot probabilities and upper confidence bounds do depend slightly on the choice of transformation, but they are all small. If the observed values of soil plutonium concentrations are taken as representative of soil plutonium concentrations in unobserved areas of Rongelap Island, then the probability that a small area contains a high concentration of plutonium is extremely small. The numerical values of hot spot probabilities depend on the extrapolation from the distribution of observed data to the distribution of high concentrations. Because the assumed critical value (35 pCi/g) is much greater than the largest observed value, some form of extrapolation is necessary to make the calculation, but it is very difficult to verify the validity of the extrapolation.
Foodstuffs
The sampling of foodstuff on Rongelap is more difficult to evaluate because no quantitative goal specified the precision required for the sampling program. For example, a survey to estimate the mean concentration of plutonium in soil might specify that the standard error of the mean is to be less than 1 pCi/m^{2}, and this required precision could then be used to calculate the required number of samples. The qualityassurance plan for the LLNL program (W. L. Robison, personal communication, 1993) specified acceptable counting errors for each sample, but these provided no specification of the desired accuracy for estimates of population parameters. The population parameters used in the dose assessment are the mean, variance, and type of distribution of radionuclide concentrations in foodstuffs. Sample numbers of foodstuff and soils are shown in Table 23.
Attention is focused on sample sizes to estimate the cesium concentration in foodstuffs because, according to the dose assessment, cesium in foodstuffs contributes a large fraction of the total dose (Robison et al., 1993). The committee estimated mean concentrations because the dose to the maximally exposed resident depends more on the mean cesium concentration
Table 22. Probability of hot spots exceeding 35 pCi/g of Pu on Rongelap.
Data from 100 m grid on Rongelap (n = 22). 


λ* 
r** 
Estimated probability 
upper 99% confidence bound 
optimal transformation to normality: 


0.8 
0.986 
0.00000 
0.0004 
other candidate transformations to normality: 


0.6 
0.981 
0.00000 
0.0020 

0.7 
0.985 
0.00000 
0.0009 

0.9 
0.984 
0.00000 
0.0002 

1.0 
0.981 
0.00000 
0.0001 
Data from all locations on Rongelap (n = 110) 


λ* 
r** 
Estimated probability 
upper 99% confidence bound 
optimal transformation to normality: 


0.4 
0.996 
0.00001 
0.00013 
other candidate transformations to normality: 


0.2 
0.989 
0.00039 
0.00292 

0.3 
0.994 
0.00006 
0.00093 

0.5 
0.995 
0.00001 
0.00002 

0.6 
0.992 
0.00000 
0.00000 
* λ specifices the transformation to normality. ** r is the correlation between the transformed data and normal order scores; values closer to 1 indicate better fit to a normal distribution. 
and on differences in diet and individual physiology among the Rongelap natives than on the variance in cesium concentration in foodstuffs (Robison et al. 1993).
The expected width of a 95% confidence interval for the mean is one of several approaches to determining an adequate sample size. One criterion might be that the sample be large enough for the expected width to be smaller than the population standard deviation. A stricter criterion might be that the expected width be smaller than half the population standard deviation.
Table 23. Environmental Samples and Interviews Contributing to Rongelap Diet Information
Environmental samples analyzed by LLNL as of October 1993* 


Total 
On sampling grid 
Cs in drinkingcoconut meat 
293 trees 
134 trees 
Cs in pandanus 
77 trees 
22 trees 
Cs in soils (05 cm) 
659 samples 
186 samples 
Pu in soils (05 cm) 
208 samples 
22 samples 
Samples used to estimate variability among individuals: 

Diet 
34 Ujelang women 

Biological halflife (cesium) 
23 Marshallese men 

* Numbers may vary from Robison et al., 1993, because some trees have been sampled more than once. 
The expected width of a confidence interval depends on the variance of the sample mean, which depends on the sample size and properties of the population. The variance of the mean from a systematic sample can be written as
where n is the sample size, N is the population size, is the population variance, and ρ_{w} S the correlation among values in a systematic sample. The correlation ρ_{w} is usually unknown, but if the population is in random order, ρ_{w} = 0. The number of coconut trees on Rongelap (N) is about 10,000, much larger than the sample size. Hence, the sampling variance simplifies to that for randomsampling:
If the data can be transformed to a normal distribution, then the width of a 95% confidence interval for the transformed mean is given by
where t_{0.975, n1} is the 97.5 percentile of a student t distribution with n1 degrees of freedom.
The sample standard deviation is a slightly biased estimator of σ, but an unbiased estimate
can be obtained by multiplying s by a correction factor (Gurland and Tripathi 1971) to get the expected widths shown in Table 24. The expected widths in Table 24 are calculated for simple random samples of various sample sizes from an infinite population and are expressed relative to the population standard deviation. Expected widths are slightly smaller for finite populations. At least 19 samples are needed for the expected width to be less than the population standard deviation. At least 65 samples are needed to meet the stricter criterion of a width less than σ/2. With 150 samples, the expected width is about σ/3 (Table 24).
The number of samples of drinking coconuts from the 100m systematic grid is large enough to provide a reasonably precise estimate of mean cesium. The expected width of the 95% confidence interval is σ/3. The number of pandanus and breadfruit on Rongelap is smaller; however, in the vicinity of the village all or most of these trees have been sampled. In this case, the sample estimates are also precise estimates of the population mean, although the sample sizes are smaller. It is difficult to evaluate the precision of the estimated cesium concentrations in the other foodstuffs used in the dose assessment because there was no probability sampling and the population sizes are generally unknown.
The mean cesium137 concentration in many foodstuffs is likely to be poorly estimated because of the small sample sizes. For example, there are 5 samples of arrowroot, 7 samples of pork muscle, 6 samples of pork liver, and 4 samples of pork heart (Robison et al., 1993; Tables 20 and B1). Because standard deviations are not computed by Robison et al. (1993; Table 20), it appears that there are even smaller numbers of samples, perhaps only one each, of papaya, chicken muscle, chicken liver and chicken gizzard. Chapter 6 on uncertainty and variability in dose projection explores some of the consequences of the small sample sizes.
An alternative criterion to assess a sampling plan is that it provide sufficiently accurate estimates of extreme values in a population. This criterion is especially relevant to Rongelap resettlement, where the MOU is concerned with the dose to the maximally exposed resident. One statistical evaluation of whether extreme values are adequately sampled is based on confidence bounds on the proportion of a population that exceeds the largest sample value. These can be calculated without assuming that values in a population have a specific distributional form (e.g., normal) by using nonparametric tolerance bounds (Hahn and Meeker, 1991). The tolerance bound that is relevant here is the estimate, with given confidence, of the proportion of the population that exceeds the largest observed value in a sample.
For an infinite population, a 95% upper confidence bound on the proportion of individuals that exceeds the largest sample value is given by
where n is the sample size and F is the 95th percentile of an F distribution with 2, 2n degrees of freedom (Hahn and Meeker 1991). For a finite population of size N, the upper 95% confidence bound can be expressed as a number of persons. It is given by the smallest
Table 24. Expected Widths of 95% Confidence Intervals for the Mean and Nonparametric 95% Confidence Bounds on the Number (in the Population) that Exceed the Largest Observation in the Sample
number, k, that satisfies the following expression:
The maximum observed concentration of cesium137 in the meat of drinking coconuts—those harvested at the proper time for drinking the liquid in the coconut—was 14.7 pCi/g (decay corrected to 1995). It is reasonable to treat the populations of soil and coconut trees as having an infinite size because there are many more locations and coconut trees than were sampled. According to Table 24 and the sample sizes given in Table 23, it is unlikely that the meat from any drinking coconuts will be as large as the maximum observed, 14.7 pCi/g, but it is reasonably certain that no more than 2% of the coconut trees will have values larger than that. If one assumes that there are about 100 pandanus trees on Rongelap, it is reasonably certain that the fruits of very few trees (no more than three) have cesium concentrations larger than 33.3 pCi/g, the highest concentration found in the 77 measured trees.
In contrast, measurements of the diet and biological characteristics of the Rongelap people provide much less precise estimates of mean or extreme values. Diet estimates are based on 34 women; biological halflife estimates of cesium are based on 23 men. The expected widths of confidence intervals for the means are 7585% of the population standard deviation (Table
24). If the population size is taken to be 200, there are likely to be about 22 people (about 10% of the population) with diets or biological halflives more extreme than the largest values found in the sample. Samples of about 25 persons of a population of this size do not provide good estimates of either the mean or the maximum.
The results from a systematic sample are commonly treated as though the sample was a simple random sample. This is true of the LLNL Rongelap dose assessment, the work plan of the Rongelap Atoll Resettlement Project (S. Simon, personal communication, 1994), and our analysis illustrated in Table 31. Analyzing data from a systematic sample as though they came from a simple random sample is statistically justifiable if three conditions are met (Gilbert, 1987):

There is no trend in the population.

There are no strata in the population.

Values are uncorrelated across space.
Each of those issues can be examined with the available data. We concentrate on plutonium in the soil, because of its interest to the people of Rongelap, and cesium in the soil, pandanus, and drinking coconuts, because it contributes the major fraction of the calculated dose. Pandanus and drinking coconuts were chosen because they are large components of the diet and have relatively high cesium content, according to the LLNL diet model and dose assessment.
Three possible trends were considered: variations in concentrations of radionuclides across the island in a roughly eastwest direction, across the island in a roughly northsouth direction, and from the beachline to the center of the island, measured by the distance from a sampling point to the nearest beach. None of the four variables (plutonium in soil or cesium in soil, pandanus, or drinking coconuts) shows any trends in mean concentration in the eastwest northsouth directions (Figs. 5, 6, 7, and 8). However, the variance in cesium in coconuts is higher at the eastern and western ends of the island than in the center of the island. Except for cesium in pandanus, the concentrations increase with distance from the beach.
The structure of the soil in the village area is quite different from that in the coconut groves. In the village, the soil has much less organic matter, has been disturbed by construction, and often is covered with layers of crushed coral. Soil surface concentrations of cesium and plutonium are significantly lower in the village area than in the coconut groves (Figs. 5 and 6). A small number of pandanus and other trees grow in the vicinity of the village. The concentrations of cesium in these trees are similar to those found in other areas of the island.
The only set of data that can be used to draw conclusions about unsampled places on Rongelap is the systematic sample, because there are notable trends and strata across the island. Mean cesium and plutonium concentrations calculated from all tree and soil samples differ from those calculated from the sampling grid alone. Sites and trees on the sampling grid cover all areas of the island with about equal probability. The characterization samples cover some areas of the island excessively, such as the village area (for soils) and the center of the island (for trees). Oversampling in this manner biases the mean. For example, oversampling of village areas, as was the case for cesium and plutonium measurements, can reduce the overall
mean by as much as 25%. The LLNL dose assessment should use either estimates calculated from just the sampling grid to describe radionuclide concentrations in the soil and coconut trees or use areaweighted averages. For pandanus, breadfruit, and other foodstuffs, for which most or all of the population has been sampled, it is appropriate to use all the available data.
Spatial correlation was examined on islandwide, 10m, and 10cm scales by estimating isotropic and anisotropic semivariograms. Semivariograms are commonly used to describe spatial variation in soil properties (for descriptions and uses of semivariograms, see, for example, Burrough, 1991; Cressie, 1991; and Warrick et al., 1986). In such plots, departure of the curve from the horizonal asymptote indicates a positive correlation. Smallscale cesiumconcentration data are available from pit 22, a 1 x 1m area in a region of the island with relatively high concentrations of cesium in the soil, and two 100 x 100m sites. Within pit 22, soil samples were taken at 10cm intervals in a square grid. In the other regions, soil samples were taken at 10m intervals. One site was entirely in the coconut grove; the other crossed from the village to the coconut grove. Because of the differences in cesium concentrations in village and coconutgrove soil, only the first site can be used to estimate the spatial correlations. There was no detectable spatial correlation on the islandwide scale or the 10m scale, but cesium concentrations showed substantial positive correlations in pit 22 (Fig. 9). Sites within 40 to 50 cm of one another have similar cesium concentrations. If the spatial correlation found in pit 22 is characteristic of the rest of the island, this smallscale positive correlation decreases the probability that a very small hot spot exists.
Conclusions

The systematic sampling grid used by LLNL is a widely used, commonly accepted technique to sample environmental characteristics. Only the data from the sampling grid should be used, however, to estimate the mean and variance, unless most or all of the items on Rongelap are sampled, e.g., breadfruit and pandanus trees.

The samples derived from the sampling grid are sufficient to provide reasonably precise estimates of the mean concentrations of radionuclides in the major foodstuffs. Estimates of the variance in the distributions of radionuclides in the environment are less accurate, but they have a much smaller impact on the conclusions of the dose assessment.

There are trends in soil cesium and plutonium and coconut cesium concentrations from the beach to inland. Some care is needed to analyze these data because of the trends. The cesium concentration of pandanus can be analyzed as though the data were collected from an islandwide simple random sample.
Analytical techniques
This section briefly evaluates the analytical methods used by LLNL to determine the concentration of radioactive isotopes of concern in soil, sediment, seawater, and biota samples from Rongelap Atoll. The procedures used are described in great detail in an LLNL report (Wong et al., 1994), and the method used for preconcentration of plutonium from large
volumes of natural waters has been published (Wong et al., 1978). Much additional information was obtained from extensive discussions with W. L. Robison of the Health and Ecological Assessment Division of LLNL.
In general, standard wellproven procedures were used in analyses for the isotopes of concern. Internal standards and carriers were used as appropriate. Samples to be analyzed for plutonium were concentrated by manganese dioxide precipitation if they were aqueous (Wong et al., 1978) or ashed and digested in concentrated nitric acid if they were soil, sediment, or biota (Wong et al., 1994). After dissolution, the plutonium was separated by anion exchange (Dowex1), electroplated onto a metal disc, and analyzed by alpha spectrometry. Sample preparation and analysis for americium were similar to those for plutonium, except that a chelating ionexchange resin, octyl(phenyl)N,Ndiisobutylcarbamoylmethylphosphine oxide (TRUSpec), was used instead of a strongbase anionexchange resin. Again, the sample was electroplated and counted by alpha spectrometry.
The other two elements of primary concern are in these studies were cesium and strontium. Cesium137 in aqueous samples was measured by extraction onto a microcrystalline ammonium phosphomolybdate cationexchanger followed by purification from potassium and rubidium by sorption onto a strongacid cation exchange resin (BIOREX 40) and the precipitation of the cesium with chloroplatinic acid for beta counting. The cesium137 content of most soil and biota samples can be determined directly by gammaray spectrometry without any preparation.
Determination of strontium90 content of aqueous samples required preconcentration by coprecipitation with ammonium oxalate, dissolution in nitric acid and precipitation as strontium nitrate, dissolution in water, and precipitation of strontium carbonate. Pretreatment of soil and biota samples was the same as that of plutonium samples. The strontium should be present as a strontium nitrate precipitate in the nitric acid solution; it was separated, dissolved in water, and precipitated as strontium carbonate. After dissolution in hydrochloric acid, the strontium solutions were allowed to stand at least 18 d for the strontium90yttrium90 equilibrium to be established and then counted on a lowbackground beta counter. The yttrium90 content must also be considered in the determination of strontium90 concentration.
Conclusions
The methods used in these analyses are well established, proven procedures, many of which were developed by LLNL and used there for a number of years. Where there are alterations in these procedures, they generally involve the use of new, more efficient ionexchange resins and extraction solvents. The written procedures are very thorough, first describing the chemical basis for the method and then giving a careful, stepbystep set of instructions that are easy to understand and follow. Also, possible sources of error are recognized and cautioned against at appropriate points in the procedure, for example, the need for care in pipetting a tracer solution, to colorcode wash bottles to avoid mistakes about what they contain, and to use a Tefloncoated stirring rod in transferring solutions. These might seem to be inconsequential points, but many analyses have been aborted by failure to observe such seemingly trivial points.
The analytical procedures were carried out under an extensive qualityassurancequality
control program (Mount et al., undated). In addition, a splitsample program was under way with the Nationwide Radiological Study of the Republic of the Marshall Islands (S. L. Simon, personal communication, 1994). Limited results (S. Duffy, personal communication, 1993) from the analysis of coconut milk and meat in this splitsample program indicate acceptable agreement between the two laboratories. Because of the sound chemistry and thoroughness of the analytical procedures, the effectiveness of the qualitycontrol program, and the early data from the splitsample program, there is little reason to question the accuracy and precision of the LLNL analytical data.
Moreover, an earlier concern of the committee appears to have been eliminated. For a time, there was a considerable backlog of unanalyzed samples that were collected in a grid sampling program; lack of data on those samples would have impaired proper dose assessment on Rongelap. Fortunately, analysis of the samples is now essentially complete, and all the planned analytical results are available for use in dose assessment.