Assessing Model Performance Using Weekday/Weekend Analysis

Comparison of photochemical air quality model predictions with observed ambient concentration data is necessary for assessing model performance. In modeling applications for the one-hour ozone standard, this comparison was often considered sufficient for determining whether the model performed well enough to predict future ozone concentrations and test proposed control strategies. Often, little or no consideration was given to the model’s performance in predicting the response to emission changes. With the advent of the new standards for eight-hour ozone and PM2.5, EPA devised a new attainment test based on the model’s relative response to emission changes (the one-hour test for ozone was based on predicted peak ozone concentrations). Thus, under the new standard, evaluating the model’s response to emission changes becomes at least as important as evaluation of its ability to reproduce historically observed events. In its recently finalized guidance for demonstrating attainment under the eight-hour standard, EPA (2007) recommended several possible means of assessing model response to emission changes, but these methods are either indirect (probing tools, alternative base cases, most observation-based models) or are difficult to employ in practice (retrospective analyses). However, weekday/weekend analysis (a type of observation-based model) is relatively straightforward and offers a direct comparison between modeled and observed responses to emission changes. Several researchers, including Blanchard and Tannenbaum (2003, 2005), Fujita, et.al. (2003), and Yarwood, et al (2003), have studied the socalled “Weekend Effect,” which often leads to an increase in measured ozone concentrations on weekends compared to weekdays. The “Weekend Effect” is somewhat counterintuitive, since emissions of NOX, an integral component in tropospheric ozone formation, are generally lower on weekend