C.3.2.2 DU 2

This DU was a 52.5 × 210 foot rectangle that composed one-quarter of an acre. This DU was investigated with four different sampling approaches:

  • A grid was placed on the site with each grid cell being 17.5 × 21 feet such that there were 30 cells covering the site. A systematic random sampling approach was used to collect samples composed of 30 increments. Three such ISM samples were collected. 

  • Thirty discrete samples were collected from immediately adjacent to the 30 systematic random sample locations used in the first of the 30-increment ISM samples collected.

  • A grid was placed on the site with each grid cell being 10.5 × 10.5 feet such that there were 100 cells covering the site. A systematic random sampling approach was used to collect samples composed of 100 increments. Three such ISM samples were collected.

  • The 100-cell grid was divided into four equal-sized quadrants. These quadrants were set by putting a big cross through the middle of the rectangular shape of the DU and allocating one corner to each quadrant. Five ISM samples with 25 increments each were collected from each quadrant. Then the quadrants were redrawn based on the CSM and prior information regarding expected arsenic concentrations. Specifically, one quadrant covered only the area of the former green (expected to have higher arsenic than other quadrants that included portions of the fairway), the next quadrant included a small portion of the green, and the third and fourth quadrants were composed solely of fairway.

Data from each sampling approach were analyzed, and the mean, SD, standard error (SE) of the mean, and 95% UCL were calculated for each. The 95% UCLs were compared to the FDEP cleanup level of 2.1 mg/kg arsenic in soil. In all cases, the 95% UCLs for DU 3 exceeded the threshold value of 2.1 mg/kg; however, the 95% UCLs for some of the quadrants did not exceed this threshold.

Figures C.3-1 and C.3-2 represent the data from DU 2 in box plots. The black circles on these box plots show the actual data points. The thick black line across the middle of each box is the median result. The thinner black lines above and below that are the 75th and 25th quantiles, respectively. That is, they represent the range in which the middle 50% of the data fall. The middle red line is the mean of the data, and the red lines above and below it represent the upper and lower confidence limits on that estimate of the mean.

Figure C.3-1. Discrete and ISM results for DU 2.

Figure C.3-1. Discrete and ISM results for DU 2.



Figure C.3-1 shows side-by-side box plots of the results of the first three (full-DU) sampling approaches. The first box represents the discrete data. The second and third boxes show the “lab-prepped" sample results, that is, the results from the samples that were carefully dried, sieved, homogenized and subsampled in a laboratory in accordance with the methodologies presented in EPA Method 8330B. The “field prepped” results shown in the final two boxes come from the same ISM samples, but subsampling of the material was performed in the field prior to shipping the main ISM sample to the laboratory for processing. The purpose of doing the field subsampling and then careful lab subsampling was to determine whether the samples that were processed in the lab would have less variability than those subsampled in the field. In Figure C.3-1 it is easy to see that there is very little difference between the results for these two types of subsampling protocols; however, it is important to note that the material being sampled here was easily mixed and that due to the nature of the contamination and the soil type, it is not surprising that field-homogenizing was nearly as effective as the more stringent USEPA SW-846 Method 8330B–type preparation. Any extrapolation of this particular finding beyond such a simple matrix and contaminant should be considered with great caution. In other situations, it is very likely that in other situations there may be a pronounced decrease in the variability between results when a thorough homogenization protocol is used.

It is also interesting to note, and apparent on Figure C.3-1, that there was not a notable improvement in the results between samples that contained 30 increments and those that contained 100 increments. The simulations performed by the ITRC ISM Team and presented in Section 4 of this document support this finding, and in fact show that only in cases with strongly skewed or variable data is there much value in collecting more than 30 increments per sample.

Finally, in Figure C.3-1 shows that the discrete data behave exactly as expected in comparison to the ISM data. Due to the smaller sample support for the discrete data, they are expected to be much more variable than ISM data. ISM physically averages over 30 or 100 samples, thus making each result essentially an average of many single discrete samples. While the means from the different sampling approaches shown in this figure do not significantly disagree, it is very clear that the discrete samples span a much wider concentration range and are more variable than the ISM results. This is a finding that matches the theory behind ISM, is borne out in the simulation studies, and can generally be expected to be true for most types of environmental investigations. Accordingly, one would anticipate that the magnitude of the UCL generated with discrete sampling with typical samples sizes (e.g., n = 10 to 30) would be greater than UCLs generated with ISM sampling.

Figure C.3-2 shows the results (in box plot form) of the quadrant sampling. Ideally, a DU would be composed of largely homogeneous media (at least in regards to the parameters of interest). If the CSM is not convincing, or if there is some reason to believe that the DU may have gross spatial heterogeneity (i.e., different concentrations of the chemical of interest in different areas of the DU), then partitioning the DU and taking separate ISM samples in each partition might be a useful strategy. For this former golf course, there was reason to believe that the greens and tees would have different concentrations of arsenic than the fairways, so partitioning into quadrants was employed.

ISM results for DU 2 by quadrant.

Figure C.3-2: ISM results for DU 2 by quadrant.



In Figure C.3-2 the boxes representing samples A-1 through D-1 show the results from the samples based on the original quadrants selected purely by breaking the 100-cell grid into four conveniently shaped sections without any recourse to prior knowledge or expectations for the site. The final four boxes representing samples A-2 through D-2 show the results for the samples from the quadrant configuration based on the CSM and our prior knowledge of the site. It is evident that, indeed, beginning with the quadrant placed on the green (1) and moving out to the fairway (4), there is a clear and significant difference in concentrations of arsenic trending down with distance from the green. It is interesting that data presented in the box plots for quadrants A-1 through D-1 would certainly have provided some interesting conversation among the project team if we did not actually already have a reason to suspect there were spatial differences across this DU.

Based on these results, it is likely that remediation would be considered necessary throughout DU-2. However, the project team might also decide to revisit the CSM and conduct additional sampling to better define the areas that would require remediation.