Special session of California 2000 field study - Air Pollution Modeling and Its Application - page 547

Temporal evaluation . For long-term simulations, a model's ability to represent
processes that occur on a wide range of time scales must be evaluated. This ability
is obscured in plots of hourly values over an annual period (e.g., Fig. 4). Spectral
decomposition methods can often separate contributions with different characteristic
frequencies. Such methods enable key time scales to be identified so that perfor-
mance evaluations can focus on those scales. The long-term component of hourly
PM 2.5 measurements is shown with predictions in Fig. 5.
Fig. 5. Modeled and simulated long-term component (>about 21 days) of PM 2.5 at FSF site
Subsets and spatial evaluation
For a given simulation, a variety of data subsets can be defined based on, for
example, surface elevation, meteorological conditions, land use, etc. Performance
evaluations should demonstrate that a model adequately simulates conditions for
subsets associated with poor air quality. The ability of a model to predict the
spatial distribution of pollutants should also be evaluated. In Fig. 6, average nitrate
predictions are compared with observations along transect (A)-(A) of Fig. 1 fo r a
subset of months in 1999 and 2000 with high PM 2.5 concentration. Model perfor-
mance is not consistent along this transect.
Fig. 6. Modeled and simulated average PM 2.5 nitrate and nitrate ratio along Fig. 1 t ransect