one or more localities of interest (for example, an urban center). Many of these models also have more advanced treatments of clouds and depositional processes, which can have a significant effect on O₃ and related gaseous pollutants.

The proposed fine particulate (PM_2.5)⁷ NAAQS and the need for models to carry out the related SIP attainment demonstrations will present an opportunity to build further upon the experience gained from modeling both O₃ and acid deposition. Fine particulate matter comes not only from direct emissions but also from atmospheric reactions that transform primary pollutants such as sulfur dioxide (SO₂), NO_x, ammonia, and VOCs into particulate sulfate, nitrate, carbonaceous material, and ammonium. Some of the processes that transform gases into PM take place in clouds and are therefore closely tied to the processes that lead to acidic precipitation. As a result, models for PM_2.5 require a conjoining of processes relevant to the formation of ground-level O₃ and to the formation of acid deposition. Because of this complexity, most models capable of simulating PM can also be used to address other air pollutant problems (O₃, acidity, and visibility reduction), and for that reason, they are also referred to as multipollutant or unified air quality models. A prime example of such a model is EPA’s new Models-3, which is available to both the scientific and regulatory communities in various forms. Seigneur (2001) has published a review of the strengths and weaknesses of these tools.

One significant problem with emissions-based air quality models for O₃ is their dependence on uncertain VOC and NO_x emission inventories (see Box 3-4). In recognition of that limitation, additional diagnostic tools have been developed to complement the air quality models. These tools, sometimes referred to as “observation-based methods,” use ambient air quality measurements rather than emission inventories to determine the relative effectiveness of VOC and NO_x emission reductions on O₃ pollution mitigation. The advantage of these methods is that they are not affected by uncertainties in emission inventories. On the other hand, they have two distinct disadvantages: (1) they require high-quality data on the ambient air concentrations of relevant chemical constituents, and much of those data are not gathered in routine monitoring networks; and (2) they are diagnostic and not prognostic (predictive). The latter limitation means that observation-based models can usually be used to estimate the types but not the amounts of emission reductions needed to meet a specific air quality