characteristics of these potential agent sinks and their patterns of operation may produce more helpful information than their physical coordinates. Here, potential hot spots coincide with these physically identifiable entities, and one or a few point-wise measurements within each may be a sufficient characterization.
The distinguishing characteristic of this model is the disconnected way in which it characterizes the contamination at each location, even within a spatial domain; essentially nothing is being assumed about spatial structure.7 That is, the model provides little or no basis for making any claim about what might exist at any unsampled locations. More specifically, there is no statistical basis for predicting or estimating the contamination at an unsampled location, no matter how close it is to the locations actually sampled. With sufficient sampling, including replicate measurements at some points, the model parameters (such as the mean and standard deviations representing measurement errors and the spatial variability of the contamination) may be estimable, and through this it may be possible to predict, for example, the proportion of the region of interest that exhibits significant contamination. But there may be little or no basis for identifying the sub-regions that may be most problematic.
From a modeling standpoint, hot spot phenomena are much more difficult to characterize than more gradually varying spatial patterns because of the noted lack of spatial coherency. Specifically, when hot spots comprise a relatively small area of volume compared to the region that must be screened, their location cannot be inferred by spatial interpolation from locations where the measurand is low. For practical purposes, a hot spot can be detected only by a measurement reflecting the concentration at the unknown location of the hot spot. For this purpose, sequential sampling plans that utilize multiple sampling bases may be most effective, as described below.
The fixed and sequential sampling plans described above can sometimes be useful in detecting isolated hot spots of agent concentration in spatial domains. However, if the hot spots are small in volume or area and the background concentrations do not “ramp up” to these elevated levels at nearby locations, the likelihood of identifying a hot spot with a spatially distributed collection of near-point measurements will often not be great.
Locating isolated agent deposits may be more effectively accomplished by a sequential strategy in which early samples are made on a broad area or air volume basis (e.g., headspace determinations), followed by measurements of wipe samples taken from limited surface areas, and finally with spatially resolved measurements.
The efficiency of sampling to identify hot spots may be improved substantially through the use of reliable generator knowledge in the temporal ordering of measurements. Ordered sampling suggests that there is some prior reason for making measurements at some locations earlier than at others. For example, in occluded space
7The kinds of spatial structure referred to might include continuity of the quantity of interest as a function of location, or any other characteristic that suggests a systematic connection between location and that quantity. In contrast, here the committee is discussing a model for which all pairs of spatial locations have the same relationship; for example, distance has no relationship to how much difference might be expected in the quantity of interest.