|
|
||
   College of Science CAMP Home People └ Alumni ├ J. Lindeman, PhD ├ J. Bakosi, PhD ├ E. Novakovskaia, PhD ├ N. Ahmad, PhD ├ J. C. Chang, PhD └ F. Camelli, PhD Research Simulation gallery Publications Resources for students Data archive Annual conference Computing resources Related links About our webpage Contact us |
 |
Joseph C. Chang, PhDYear of graduation: January 2003 Degree: Doctor of Philosophy Research field: Evaluation and uncertainty analysis for atmospheric dispersion models Thesis title: Methodologies for Evaluating Performance and Assessing Uncertainty of Atmospheric Dispersion Models Thesis abstract: This thesis describes methodologies to evaluate the performance and to assess the uncertainty of atmospheric dispersion models, tools that predict the fate of gases and aerosols upon their release into the atmosphere. Because of the large economic and public-health impacts often associated with the use of the dispersion model results, these models should be properly evaluated, and their uncertainty should be properly accounted for and understood. Any evaluation exercise should start with clear definitions of the evaluation objectives. A set of model evaluation methodologies (including the BOOT evaluation software, Taylor's nomogram, the figure of merit in space, the measure of effectiveness, the ASTM procedure and the CDF approach) was reviewed. Some methodologies (i.e., the ASTM procedure and the CDF approach) were found to also consider model uncertainty. It is important to adopt a balanced approach, because there is not a single best performance measure or best evaluation methodology. Various issues concerning model uncertainty were also further reviewed in this study, including descriptions of the components of the overall model uncertainty, definitions of the commonly-used terms and discussions of the methods to account for these uncertainty components. The CALPUFF, HPAC and VLSTRACK dispersion modeling systems were applied to the Dipole Pride (DP26) field data (~20 km in scale), in order to demonstrate the evaluation and uncertainty assessment methodologies. In terms of the maximum dosage on a given sampling line, the overall mean biases for the three models were found to be not much different from zero and within a factor of two of each other and the random scatters were about a factor of three. Dispersion model performance was found to be strongly dependent on the wind models used to generate gridded wind fields from observed station data. This is because, despite the fact that the test site was a flat area, the observed surface wind fields still showed considerable spatial variability, partly because of the surrounding mountains. The data-withholding technique was used to estimate the uncertainty due to input data errors and HPAC's second-order turbulence closure scheme was used to estimate the variability due to random turbulence. It was found that the two components were comparable for the DP26 field data, with variability more important than uncertainty closer to the source and less important farther away from the source. Therefore, reducing data errors for input meteorology may not necessarily increase model accuracy due to random turbulence. DP26 was a research-grade field experiment, where the source, meteorological and concentration data were all well-measured. Another typical application of dispersion modeling is a forensic study where the data are usually quite scarce. An example would the modeling of the alleged releases of chemical warfare agents during the 1991 Persian Gulf War, where the source data had to rely on intelligence reports and where Iraq had stopped reporting weather data to the World Meteorological Organization since the 1981 Iran-Iraq war. Therefore, the meteorological fields inside Iraq must be estimated by models such as prognostic mesoscale meteorological models, based on observational data from areas outside of Iraq and using the global fields simulated by the global meteorological models as the initial and boundary conditions for the mesoscale models. It was found that while comparing model predictions to observations in areas outside of Iraq, the predicted surface wind directions had errors between 30 to 90 deg, but the inter-model differences (or uncertainties) in the predicted surface wind directions inside Iraq, where there were no onsite data, were fairly constant at about 70 deg. This large uncertainty in the predicted surface wind directions was partly due to the differences already existing in the global fields that were used as the boundary conditions for the mesoscale models. As a result, various mesoscale meteorological models might give rise to distinctly different dispersion patterns for an agent cloud. This seemingly discouraging disparity in the dispersion model results is in fact quite plausible in a data-void situation, even with the use of state-of-the-art mesoscale meteorological models. Any decision making process must take this into account. Current employment: Northrop Grumman Mission Systems Current activities: Transport and dispersion modeling; consequence assessment for chemical, biological, and radiological weapons; dense gas modeling; meteorological modeling; and technical analysis. Research affiliation: Comprehensive Atmospheric Modeling Program, School of Computational Sciences, George Mason University See the CAMP
contact page for detailed
information on contacting us. |
