ATD: Stochastic algorithms for countering chemical and biological threats

The vision for the proposed work is to capitalize on experience in specific areas involving biological / chemical attacks and to develop as wide-ranging algorithms as possible for countermeasures to such threats. These will be based on stochastic modeling to describe (deterministic and random) features of contamination location and spread from which algorithms to detect threats may be developed. Standard parametric models will be examined for applicability, along with development of appropriate space-time random fields tailored to the specific situation. This is similar in approach to current use of Gaussian fields to model trends and fluctuations in pollution regulation studies, requiring few specific assumptions, but allowing application of general methods such as Central Limit and Extreme Value Theory. The starting point will be the PI's prior experience in assisting EPA scientists and engineers with the development of statistical designs for assuring confidence in complete decontamination of anthrax from buildings in the face of budgetary limitations which was based on very simple assumptions such as statistical independence of positions of anthrax deposits. Better and more cost effective statistical conclusions are likely to result from the proposed models - relaxing the independence assumptions.

Clearance of contaminated buildings is typically achieved by Chlorine Dioxide fumigation - a highly effective process since the gas under appropriate pressure and duration of application can be expected to reach all areas where contaminant may be lodged. The cost of and time required for such clearance can be quite staggering (e.g. estimates of $130 million for just the Brentwood postal facility over a 26 month period, and $1 billion in all for the total of the early post 9/11 incidents). Additionally, in spite of a high expectation that fumigation achieves total clearance, systematic sampling of surfaces for residual anthrax is done for confirmation. This cannot be exhaustive in view of the high cost of sampling and laboratory analysis and the sheer number of samples required for reasonably complete coverage. Hence, determinations are made (using simple statistical assumptions) of the extent of sampling required to ensure that if all samples turn out to be negative, there is specified very high confidence that the entire area of concern is entirely contaminant free. This seems a reasonable approach to give adequate added assurance of successful clearance since if even one sample tested positive for anthrax, entire fumigation would have to be repeated. Current routines for determining necessary sampling to achieve a given clearance confidence have been developed under simple assumptions. The future work on this topic will use the more realistic statistical assumptions of stochastic modeling which are expected to lead to algorithms, which are both more accurate and cost efficient. The above case of building decontamination is described as a starting point for much more complex investigations such as detection of the presence of contaminants being actively introduced in an HVAC system for circulation throughout a building - a topic which has received some previous attention in the literature. A related area of initial activity concerns the discrimination between toxic substances such as anthrax and harmless powders for which novel statistical (for example, wavelet) methods are already being developed in collaboration with EPA researchers. Further, the issues involved in such 'Homeland examples' arise in various forms in the detection of battlefield threats (such as hidden IED's) with attendant problems of signal detection - subjects for consideration under this grant. Finally, the methods are expected to have useful 'non-terrorism' applications such as to the spread of pandemics and risks of importation of anthrax bearing animal products such as meat and hides.