Source Term Characterization of FFT07 Data using a Genetic Algorithm

The national strategy for Homeland Security states that the nation should focus its efforts on preventing and disrupting terrorist attacks, as well as responding to and recovering from incidents that do occur. Such incidents include an intentional release of hazardous chemical, biological, radiological, or nuclear (CBRN) material into the atmosphere. With this threat it becomes critical to military and civilian health and safety to predict the atmospheric transport and dispersion (AT&D) of the CBRN contaminant. Since accurate data regarding the source term are often unavailable it becomes necessary to estimate the source term from remote concentration measurements. However, due to the stochastic nature of atmospheric turbulence, it is impossible to ensure that source term characterization algorithms will work in real-world conditions until they are tested against atmospheric dispersion datasets. The FUsing Sensor Information from Observing Networks (FUSION) Field Trial (FFT07) was conducted to evaluate sensor data fusion algorithms under known environmental conditions, thus creating an abundant reliable dataset. Our method, which couples a genetic algorithm with a dispersion model has been previously successful at estimating not only source characteristics (location and strength) but also the meteorological parameters necessary to predict the AT&D of a contaminant with simulated data. The method's performance has been assessed for Phase I of the FFT07 data for both the Gaussian and SCIPUFF dispersion models. We have compared and contrasted results with and without the meteorological data reconstruction due to the fact that real-time wind direction and speed are sparse. A quantitative sensitivity study of concentration time averaging periods, selected and random sensor removal, and data censoring is done in preparation of Phase II of FFT07. In addition, we develop methods to quantify the uncertainty in our estimates.