Source term estimation uncertainty analysis using a genetic algorithm coupled with dispersion models

The stochastic nature of atmospheric turbulence makes it impossible to ensure that source term estimation algorithms will work in real-world conditions until they are tested against atmospheric transport and dispersion (AT&D) field datasets. The FUsing Sensor Information from Observing Networks (FUSION) Field Trial (FFT07) was conducted to create an abundant reliable dataset. The genetic algorithm (GA) coupled AT&D model method used here has been successful at estimating source characteristics and meteorological parameters necessary to predict the AT&D of a contaminant. As part of the FFT mission we have submitted predictions of Phase I FFT07. These included the GA coupled with both the Gaussian and SCIPUFF AT&D models. In this case, sensor thresholding is applied and a new statistical approach to determine the best estimate of the unknown parameters is incorporated. The GA couples AT&D models with observations using a dual cost function approach, rather than the single cost function method used previously. Additionally, a method was developed to determine sensor noise thresholds for concentration data and the statistical analysis uses bootstrap sampling to quantify the uncertainty in the estimates of the individual cases. This current study quantifies the uncertainty of the GA coupled AT&D method using a subset of the FFT07 dataset, that is, those trials that include a single source only, for both continuous and instantaneous releases.