Sequential Monte Carlo algorithm for source release estimation

A key problem is that of source localization of a chemical release. A distributed set of sensors samples the surrounding air and, based on the concentration of particulates, may signal a detection event. We are developing a source localization algorithm, based on sequential Monte Carlo (SMC) methods, that estimates the location of the release, the size of the release, and the time of the release. The algorithm employs a transport and dispersion model to predict concentrations in the surveillance region for many different hypothesized release parameters, which, combined with a sensor model, predicts which sensors will signal detections and which will not. The algorithm considers both sensor detection events and, importantly, sensor non-detection events to select which of the source release hypotheses are most consistent with the observed sensor data and, ultimately, to estimate the source release parameters. Simulation results are given for different scenarios and conditions.