|
|||||||
|
|
|
|||||
|
|
|||||||
October 15, 1998
Wesley E. Stimson
Haley & Aldrich, Inc.
465 Medford Street
Suite 2200
Boston, MA 02120-1400
Dear Mr. Stimson:
I am concerned about the W. R. Grace site proposed sampling plan as presented at the public meeting on October 1, 1998. While a uniform sampling plan (such as in this case, with sampling at regular intervals on a grid) may be useful in some cases, it is my understanding that it would not be optimal for identifying contamination of the type we might expect on this site. I believe that uniform sampling is useful if the entire area to be sampled is expected to have been exposed in similar fashion and/or if the contamination levels are smoothly distributed, either by diffusion after the contamination (as in the case of volatile substances), or by the original source of contamination for stationary contaminants (such as settling of pollutants around a smokestack), and the sampling interval is small enough to elucidate the gradual shifts in concentration across the region.
I thought that contamination is expected to be at a low level, in which case the sampling points may provide too little coverage to be useful. Furthermore, the uniform sampling is subject to systematic errors. My concern is that without randomization of the sampling points, the entire test may be of very little value if some of the assumptions are wrong. Additionally, the testing method will only give results of whether asbestos was detected, so we will not have the benefit of using concentration levels for statistical inferences.
From the initial sampling that has been performed, it appears that there are areas of asbestos contamination adjacent to areas with none detectable by TEM (AB-5 vs. AB-6 or AB-2 vs. AB-3). At the meeting, there was also mention that the concentration of asbestos found is indicative of "dumping," which could create regions of localized contamination (RoLC) with undetectable levels just outside the region of dumping.
The limitations of uniform sampling on a grid can be easily seen by noting that there could be such an RoLC running the length of a zone (e.g. 350 ft. in Zone 2) that could be 46 ft. wide (assuming 50 ft. grid spacing, and sampling with a 2 ft. radius) which would be entirely missed by this sampling plan. Although it would be unlikely (but possible) to miss such a large RoLC, there is a fairly high probability (67%) that the proposed sampling plan would miss a similar 15 ft. wide region of same length. I would think the probability of missing a randomly placed 15 ft. diameter pile of dumped asbestos to be fairly high under the current sampling plan.
A statistically based sampling plan would allow estimation of the probability of missing a region of given size. With the proposed plan, there is no way to estimate how much error can be expected. Since the level of contamination is, hopefully, small, I would think a statistically based, random sampling plan might give us better confidence with fewer samples than a reasonable uniform sampling plan.
Regardless of the method (random vs. uniform sampling) there needs to be a thoughtful estimation of the smallest size of an RoLC before the sampling interval can be determined. While the proposed plan may be reasonable for initial testing of an area where contamination has not yet been found (but would presumably be evenly distributed if present), we already have data indicating the presence of asbestos which, additionally, does not seem to be evenly distributed.
I would think that since we have identified sources of contamination, we should first characterize those areas to estimate the possible extent and distribution of contamination that we might expect to find in the area. Without taking the existing data into consideration in the design of the next testing plan, the assumptions in deciding the sampling plan become based on conjecture, and subsequently, we can only guess if the results will have any meaning. It surprises me that a plan is being undertaken where conclusive results could be made only if significant asbestos levels are found. If none is found, the only thing that can be said with some confidence is that no huge regions of asbestos contamination exist, but can be no inference of the existence of moderately sized RoLCs because the proposed plan is not statistically based.
I urge you to reconsider whether the proposed sampling plan is sufficiently robust before proceeding with the testing.
I'll close with a quote from the book "Sampling Design and Statistical Methods for Environmental Biologists" (1979):
It is critical that at some level the sampling be random or most statistical tests will be invalid. The assumption of independence of errors is the only one in most statistical methods for which violation is both serious and impossible to cure after the data have been collected.
Sincerely,
Michael Nakagawa
cc: David B. Struhs, Commissioner
Commonwealth of Massachusetts
Executive Office of Environmental Affairs
Department of Environmental Protection