4.2
Application of a Genetic Algorithm-Coupled Receptor/Dispersion Model to the Dipole Pride 26 Experiments

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 31 January 2006: 2:00 PM
Application of a Genetic Algorithm-Coupled Receptor/Dispersion Model to the Dipole Pride 26 Experiments
A407 (Georgia World Congress Center)
Christopher T. Allen, Penn State Univ., University Park, PA; and S. E. Haupt and G. S. Young

Presentation PDF (410.4 kB)

A receptor model is coupled with a dispersion model, SCIPUFF, by using a genetic algorithm (GA). Previous work has validated this approach using a simpler dispersion model and shown that a GA is superior to least squares as a method by which to solve for the apportionment vector. The current coupled model uses SCIPUFF to calculate expected pollutant concentrations at a given receptor for each of a set of candidate sources. The GA then determines the source apportionment vector that best matches the expected concentrations with the actual receptor data. The model is validated using synthetic data produced by SCIPUFF for three distinct source emission setups along with previously defined meteorological conditions. Monte Carlo simulations are run for each setup, incorporating various amounts of noise to alter the receptor data to show that the GA coupled model is capable of producing the previously known source apportionment vector to a good approximation.

The GA coupled model incorporating SCIPUFF is then tested using real receptor data from the Dipole Pride 26 experiments at the Nevada Test Site. The model is first modified to make use of the Dipole Pride data, including the use of pollutant data from more than one receptor. The model is used to determine the locations and times of the emissions during the experiments using the provided receptor and meteorological data. In this way we show the real-world applicability of the model as well as its limitations. The GA is successful at pinpointing the exact location and time of each emission to a good approximation in light of the uncertainty. This uncertainty is associated with the accuracy of the pollutant data, meteorological characterizations, and the inherent difference in SCIPUFF-predicted ensemble mean concentrations and a specific realization. Future work will further develop the concept of using a GA to characterize other factors associated with pollution emission, including a specification of the meteorological conditions from the pollutant data provided.