Temporal and Spatial Reconstruction of Atmospheric Puff Releases using Bayesian Inference

An atmospheric dispersion release event involving the transport of contaminants may be categorized as either continuous or instantaneous. An instantaneous, or puff, release of hazardous contaminants can be accidentally or intentionally released into the atmosphere where its transport by wind is subject to various atmospheric conditions. Using a network of sensors, one can estimate the parameters of the release source using an appropriate forward model and the measurements obtained from the sensors. In this study, the Stochastic Event Reconstruction Tool (SERT), which was developed to work for continuous releases where time of release was not a factor, is extended to reconstruct instantaneous releases. Using a data-driven Gaussian puff model, we estimate the source location, quantity of release, and time of release, along with other diffusion parameters controlling the spread of the contaminant within the atmosphere. We use a Bayesian inference method with Markov Chain Monte Carlo (MCMC) sampling to obtain parameter estimates probabilistically. Having the contaminant release time as a new parameter in the inverse problem introduces new challenges to the Bayesian inference method. To address this new complication faithfully, we make modifications to the original Bayesian inference method in SERT to ensure the convergence of MCMC chains for puff releases. To ensure a quick and accurate source term reconstruction, we pursue a parallel implementation in our method for MCMC chains, and determine the convergence of the simulation using statistical post-processing. Thanks to our parallel implementation, the entire simulation takes a matter of minutes. We validate our method using real trial cases from the Fusion Field Trial 2007 (FFT07) study, and demonstrate successful temporal and spatial reconstruction with puff releases of passive contaminants.