The unintentional or intentional release of a harmful contaminant is a serious threat to civilians and to our nation's military forces. Responding quickly and with the best information possible is critical to mitigating the threat and minimizing the impact on population and property. Characterizing the source accurately is vital to driving atmospheric transport and dispersion (AT&D) models and thus predicting the future state of the puff. Unfortunately, a field sensor network is unlikely to provide adequate spatial or temporal resolution in either the meteorological or chemical data.

In this study we use a Genetic Algorithm (GA) coupled with the Second-order Closure Integrated Puff model (SCIPUFF) to back-calculate several parameters describing a contaminant release. A set of trial solutions, each representing a possible source term, is randomly initialized. The GA evolves the population through mating and mutation operators and for each new trial solution a new forecast is created via SCIPUFF. This resulting forecast concentration field is compared to the observed concentration field.

Initially the model is validated by using identical twin data, that is observation data generated by SCIPUFF itself. Second, more realistic observation data generated by a Computational Fluid Dynamics (CFD) model is used. The final stage of model testing includes using observation data from a field trial. We demonstrate that the genetic algorithm back-calculation model coupled with SCIPUFF is successful at identifying the basic source information.