Development of a regression model to forecast ground-level ozone concentration in Louisville, KY

To support ozone forecasting and episodic air pollution control initiatives in the Louisville metropolitan area, a multiple-linear regression model to predict daily domain-peak ground-level ozone concentration [O-3] has been developed and validated. Using only surface meteorological data from 1993-1996 and making extensive use of parametric transformations to improve accuracy, the ten parameter model has a standard error of prediction of 12.1 ppb and an explained variance of 0.70. Retrospective ozone forecasts were made for each day of the four ozone seasons (May-September) using archival meteorological data as input to the model. For the period 1993-1996 examined, 50% of days were forecast to within +/- 7.6 ppb, and on 80% of days the accuracy was within +/- 14.8 ppb. The model correctly predicted 74, 80, and 40% of occurrences of the daily "good" ([O-3] less than or equal to 60 ppb), "moderate" (60 < [O-3] less than or equal to 95), and "approaching unhealthful" (95 < [O-3] less than or equal to 120) air quality categories, respectively. The model did not predict any of the nine exceedances of the National Ambient Air Quality Standard ([O-3] > 120) which occurred over the four year period. Simple supplementary meteorological criteria were developed that correctly forecast 89% of NAAQS exceedances. Used in combination with forecaster experience, synoptic weather information, and supplementary meteorological criteria, the regression model can be a useful tool for improving the accuracy of local O-3 forecasts. (C) 1998 Elsevier Science Ltd. All rights reserved.