controls for O3 pollution long after science suggested that this strategy was not effective.
The ability to quantify uncertainty in the predictions of state-of-the-science chemical transport models remains inadequate, although improvements have been made.
In a number of areas of the country, there have been inadequate resources (financial and personnel) for the correct application of these tools to local and regional problems.
Beyond the specific concerns listed above, there are two broader concerns—model uncertainty and over-reliance on models for O3 SIPs—and one important opportunity—multipollutant models—that bear more detailed discussion.
An assessment of the uncertainties in model output has not been a routine component of the information used by regulators. Unfortunately, the uncertainty in model predictions depends on a variety of factors, including the type of prediction (such as variables and averaging time), the application, the uncertainties associated with the model’s input and internal parameters, and the model itself. As a result, a universal, comprehensive statement cannot be made about the uncertainties of any model simulation (see Oreskes et al. 1994). Literature estimates for individual components of an air quality model—emissions, chemistry, transport, vertical exchange, deposition—typically indicate uncertainties of 15–30%, but when the supporting data sets are weak, the uncertainties can be significantly higher. A consideration of all factors reveals that model uncertainty can be both significant and poorly characterized (NARSTO 2000). It follows, therefore, that relying solely on the output of an air quality model to resolve emission-control issues or to demonstrate attainment of an air quality standard or objective is problematic.
Despite these limitations, air quality models remain the only tools available for quantitatively simulating or estimating future outcomes. Although challenging to use, air quality models are essential to the current AQM system. Their use can be optimized by collecting and analyzing appropriate data sets, carefully assessing model sensitivity and uncertainty, and avoiding inappropriate applications.
As discussed in more detail later in this chapter, SIPs developed for O3 nonattainment areas have relied heavily on so-called attainment demon-