Uav Navigation by Expert System for Contaminant Mapping

The intentional or unintentional release of a harmful atmospheric contaminant is a potentially devastating threat to homeland and defense security. Accurate identification of the source location and intensity is essential to predicting subsequent transport and dispersion of the contaminant. Insufficient spatial and temporal resolution of the available contaminant and wind-field observations make source characterization extremely difficult (Allen et al. 2006) when using typical networks of fixed sensors. Given sufficient observational data, however, the problem is feasible as shown by Long et al. (2008). That study demonstrated the use of the Gaussian puff equation as the dispersion model in identical twin numerical experiments applying a Genetic Algorithm (GA) to back calculate the required source characteristics solely from the observations on grids ranging from 8×8 to 2×2 of fixed location concentration sensors. It is not practical, however, to cover all societally important regions with a dense enough fixed sensor network to make that method practical. Therefore we study an alternative, the use of a mobile sensor system. The coupling of concentration observations to dispersion model forecasts by a GA proved to be a fruitful way to determine both source characteristics and those of the transport and dispersion process. Using the Gaussian plume equation as the dispersion model, Allen et al. (2007a) applied a genetic algorithm to identify four parameters: source location (x,y), source strength, and wind direction. Even when noise was added to the concentration data to simulate the non-Gaussian nature of instantaneous turbulent dispersion, the results remain excellent for sensor grids of 8×8 and larger. That grid size presumes that the wind direction is unknown and requires receptors in all possible directions from a potential source. An earlier version of the model was validated with circular and spiral source array synthetic data configurations before being applied to field test data from Logan, Utah (Haupt 2005) then was validated in the context of superimposed noise (Haupt et al. 2006). Allen et al. (2007b) extended that analysis with a more sophisticated dispersion model, SCIPUFF and correctly identified the time of release, source location, and apportioned contaminant contributions from multiple sources, even for concentration observations contaminated with moderate amounts of white noise as well as testing the model on ____________________________________________