Stochastic event reconstruction of atmospheric contaminant dispersion

Environmental sensors have been deployed in various cities for early detection of contaminant releases into the atmosphere. Due to the sparse distribution of these sensors, event reconstruction and high-fidelity dispersion modeling capabilities are needed to implement effective strategies for environmental management and emergency response. We present a stochastic event reconstruction method that is designed to process information from a network of environmental sensors. The method is based on Bayesian inference with Markov Chain Monte Carlo (MCMC) sampling. The Bayesian framework provides a convenient way to incorporate prior knowledge and to propagate uncertainty of measurements into the final results. To address the fast-response operational needs, fast-running Gaussian plume dispersion models are adopted as the forward model in the inverse problem. A probability model is proposed to take into account both zero and non-zero concentration observations that can be available from a sensor network. It is found that stochastic treatment of the empirical parameters in the Gaussian plume dispersion model improves the event reconstruction results significantly. Computational tests on dispersion problems have shown that stochastic event reconstruction method works effectively in finding the most probable release. Additionally, posterior distributions are used to calculate probabilistic projections of the contaminant release with specified confidence levels.