Comparison of Three Approaches to Atmospheric Source Localization

The paper describes three methods for finding the source of an airborne contaminant in a turbulent wind field. We assume that an unmanned aircraft samples the atmosphere at a constant altitude within the planetary boundary layer, measuring the contaminant concentration. The turbulent wind field is generated using the openly available software TurbSim. A concentration field, or plume, is created by advection and diffusion of the contaminant from its source. The plume is modeled using a filament-based method proposed in the literature. The unmanned aircraft is assumed to sample the concentration field quickly, relative to the plume dynamics; once the plume has matured, the concentration is fixed and the sampling strategy is executed. A recursive Bayesian estimation approach is compared with a gradient descent algorithm and an extended Kalman filter. By comparison, Bayesian estimation requires relatively weak modeling assumptions, and simulation results suggest this approach is less sensitive to error in the initial state.