One product of the simulation modeling of fugitive dust emissions on Owens dry lake is a series of maps indicating which areas of the lake bed need to be controlled in future dust mitigation efforts.These"dust control area"(DCA)maps vary considerably depending on the modeling inputs and assumptions that are used.For example,differences in data screening criteria,and in the summary statistic chosen,affect the modeling results.Statistical techniques that evaluate the performance of alternative modeling approaches may be useful for selecting the approach that provides the most accurate modeling results and therefore the"best"DCA map.Three types of model performance evaluation methods are discussed:(1)an"exceedance comparison",in which the numbers of predicted-and observed exceedances of the air quality standard are compared,(2)unpaired data comparisons,including Quantile-Quantile plots and the Robust Highest Concentration,and(3)paired data comparisons,including linear regression and error analysis techniques.The latter provide a more refined model performance evaluation than the unpaired comparisons.The characteristics of each of these methods are discussed.All of these techniques were applied to different subsets of the 2000 to 2002 monitoring data from the Owens dry lake,based on screens for PM_(10)concentration ranges and other factors.While the goal of model performance testing is to provide an objective approach to choose the best model,this study shows that many subjective choices have to be made.For example,the definition of the"best"performing model scenario is dependent on the evaluation method(s)chosen,and the interpretation of the results.These definitions will differ based on the perspective of the regulator,the regulated,or from a purely scientific perspective.
展开▼
