Numerical Models of Regional Air Quality:
(Lining Up Model and Measurements to Probe and Address Uncertainty)

The basic paradigm of a numerical model for regional air quality will be briefly described and the main objectives of air quality models for EPA will be noted. Two key conceptual models of scientific processes being simulated (ozone and inorganic fine particles) underlying many of the predictions of interest will be briefly described. A general perspective on uncertainties, from an air quality modeling viewpoint, separated into transient (random) and persistent (systematic) uncertainties will be illustrated with examples. Then four areas for potential collaboration will be discussed: Persistent errors and model evaluation, reducing (hopefully) uncertainty in inputs through inverse modeling, combining models and measurements through spatial-temporal modeling, and probing process relationships through multivariate analysis. Issues of model performance uncovered in highly resolved data turn into persistent errors in more typical spatial data sets that have longer integration times. Some of the sources of persistent error will be noted and the fact they can change with season will be illustrated. The importance of input errors will be illustrated with the example of ammonia emissions and the inorganic fine particle system. The use of inverse modeling to try to address input errors is presented as a valuable tool to be further developed. The spatial exploration of input errors is an important need, particularly for the health community. This user community desire for space-time interpolation that combines model and measurements will be discussed.