Uncertainties in air quality model predictions

Abstract

As a result of several air quality model evaluation exercises involving a large number of source scenarios and types of models, it is becoming clear that the magnitudes of the uncertainties in model predictions are similar from one application to another. When considering continuous point sources and receptors at distances of about 0.1 km to 1 km downwind, the uncertainties in ground-level concentration predictions lead to typical mean biases of about ±20 to 40% and typical relative root-mean-square errors of about 60 to 80%. In fact, in two otherwise identical model applications at two independent sites, it is not unusual for the same model to overpredict by 50% at one site and underpredict by 50% at the second site. It is concluded that this fundamental level of model uncertainty is likely to exist due to data input errors and stochastic fluctuations, no matter how sophisticated a model becomes. The tracer studies that lead to these conclusions and have been considered in this study include: (1) tests of the Offshore and Coastal Dispersion (OCD) model at four coastal sites; (2) tests of the Hybrid Plume Dispersion Model (HPDM) at five power plants; (3) tests of a similarity model for near-surface point source releases at four sites; and (4) tests of 14 hazardous gas models at eight sites including six sets of experiments where dense gases were released.