ASSESSING DISPERSION MODEL PERFORMANCE TO SIMULATE AVERAGE CENTERLINE CONCENTRATION VALUES

摘要

Most operational atmospheric simulation models are deterministic. They provide estimates of the average time- and space-variations in the conditions (e.g., mesoscale meteorological models), or they provide estimates of the average time- and space-variations of effects (e.g., air quality models). The observations used to test the performance of these models are individual realizations (which can be envisioned as coming from some ideal ensemble), and are affected by stochastic variations unaccounted for within the simulation model. If we believe this, then it makes sense to ask the models to replicate average patterns seen in the observations, but it does not make sense to ask the models to replicate the effects of stochastic variations unaccounted for within the simulation model (e.g., observed maxima, or total variance). The American Society for Testing and Materials (ASTM), published in December 2000 D 6589, entitled, "Standard Guide for Statistical Evaluation of Atmospheric Dispersion Model Performance", to provide a framework for developing techniques that are useful for comparison of modeled and observed concentrations, that addresses the concern that models provide deterministic estimates of ensemble averages, and observations are individual realizations from imperfectly defined ensembles. The Guide suggest a two step process: Step 1) analyze the observations to provide average patterns for comparison with modeled patterns, and Step 2) employ bootstrap resampling when comparing these patterns, which accounts for uncertainties in performing Step 1, and provides a means for objectively testing whether differences seen are statistically significant. An example procedure is provided in the Annex of D 6589 for evaluating performance of models to estimate the average maximum ground-level centerline concentration. In the example procedure, observations having similar meteorological conditions are grouped together at each downwind distance. In the following discussion, we summarize the results obtained in applying the example procedure to test the performance of four plume dispersion models: ADMS 3.1( Carruthers et al., 1994), AERMOD (version 01247, Cimorelli et al., 1996), HPDM (version 4.3, level 920605, Hanna and Paine, 1989), and ISCST3 (version 00101, U.S. Environmental Protection Agency, 1995), with tracer field data from three studies: Prairie Grass (Barad, 1958; Haugen, 1959), Kincaid (Bowne et al., 1983), and Indianapolis (Murray and Bowne, 1988).