A major challenge in accidental or unregulated releases is the ability to identify the pollutant source, especially if the location is in a large industrial area. Usually in such cases, only a few sensors provide non-zero signal. A crucial issue is therefore the ability to use a small number of sensors in order to identify the source location and rate of emission. The general problem of characterizing source parameters based on real-time sensors is known to be a difficult task. As with many inverse problems, one of the main obstacles for an accurate estimation is the non-uniqueness of the solution, induced by the lack of sufficient information. In this study, an efficient method is proposed that aims to provide a quantitative estimation of the source of hazardous gases or breathable aerosols. The proposed solution is composed of two parts. First, the physics of the atmospheric dispersion is utilized by a well-established Lagrangian stochastic model propagated backward in time. Then, a new algorithm is formulated for the prediction of the spacial expected uncertainty reduction gained by the optimal placement of an additional sensor. These two parts together are used to construct an adaptive decision support system for the dynamical deployment of detectors, allowing for an efficient characterization of the emitting source. This method has been tested for several scenarios and is shown to significantly reduce the uncertainty that stems from the insufficient information.
