Evaluation of two pollutant dispersion models over continental scales

Two long-range, emergency response models—one based on the particle-in-cell method of pollutant representation (ADPIC/U.S.) the other based on the superposition of Gaussian puffs released periodically in time (EXPRESS/Russia)—are evaluated using perfluorocarbon tracer data from the Across North America Tracer Experiment (ANATEX). The purpose of the study is to assess our current capabilities for simulating continental-scale dispersion processes and to use these assessments as a means to improve our modeling tools. The criteria for judging model performance are based on protocols devised by the Environmental Protection Agency and on other complementary tests. Most of these measures require the formation and analysis of surface concentration footprints (the surface manifestations of tracer clouds, which are sampled over 24-h intervals), whose dimensions, center-of-mass coordinates and integral characteristics provide a basis for comparing observed and calculated concentration distributions. Generally speaking, the plumes associated with the 20 releases of perfluorocarbon (10 each from sources at Glasgow, MT and St. Cloud, MN) in January 1987, are poorly resolved by the sampling network when the source-to-receptor distances are less than about 1000 km. Within this undersampled region, both models chronically overpredict the sampler concentrations. Given this tendency, the computed areas of the surface footprints and their integral concentrations are likewise excessive. When the actual plumes spread out sufficiently for reasonable resolution, the observed (O) and calculated (C) footprint areas are usually within a factor of two of one another, thereby suggesting that the models possess some skill in the prediction of long-range diffusion. Deviations in the O and C plume trajectories, as measured by the distances of separation between the plume centroids, are on the other of 125 km d−1 for both models. It appears that the inability of the models to simulate large-scale vertical motions and, more generally, to adequately characterize the meso- and macro-scale flows (mainly as the result of constraints imposed by the temporal and spatial resolution of the rawinsonde network) are major impediments to improved accuracy.