Two types of statistical models for predicting high-ozone events in the Houston-Galveston region of the Texas Gulf Coast are developed for the two-fold purpose of determining both qualitatively and quantitatively the atmospheric conditions that favor such events. Because of its localized concentration of major point sources and its proximity to the Gulf of Mexico (and its attendant sea-breeze processes), the conditions under which high-ozone events occur are expected to be more complex than the conditions leading to high-ozone events in other regions.

The two types of models considered were a decision-tree model designed solely to predict EPA-defined high-ozone events and a MOS type model designed to specifically predict maximum 1- and 8-hour ozone concentrations. Analysis of the model residuals for the test period allows both models to make both deterministic and probabilistic forecasts (in the form of confidence intervals for the MOS type approach).

Because of its proximity to the critical latitude of 30N, Houston's diurnal wind evolves in the manner of an ellipse over a diurnal cycle. To quantify this variation and its effect on ozone, wind data from an offshore buoy just southeast of the Houston area was analyzed with the purpose of determining the magnitude and orientation of the ellipse. This is done by fitting a combined linear-sinusoidal four-parameter model to the hourly u and v wind components for each day of data. This model turns out to explain a large fraction of the daily variance in wind, especially during July and August.

The mean orientation and amplitude of this ellipse is a valuable predictor for high-ozone events in the Houston area because it reveals when, if ever, on a given day that the sea breeze will completely offset the large-scale background flow. If this occurs during a period with ample sunshine in a specific area with many sources of VOCs and NOx, or an already high concentration of ozone, a high-ozone event may be likely.

This sort of analysis enables us to use predictors in the statistical model that have a direct physical relationship to ozone formation. For example, rather than using forecasted wind speed and direction at one or two times during the day, we use the forecasted daily mean wind as a predictor and combine it with the expected characteristics of the sea breeze ellipse to develop additional predictors such as minimum likely wind speed and duration of stagnation. Such predictors would be difficult to forecast directly, but here they are derived from two easy-to-forecast quantities: mean wind and mean diurnal cycle.

The results of the statistical model development will be a model which can be used to predict the occurrence of high-ozone days in advance, allowing citizens and industries within this region to curtail their activities accordingly, without the cost burden of modifying activities on a permanent vasis. Qualitatively, this study also allows us to determine the synoptic and mesoscale conditions conducive to high-ozone events along the Texas Gulf Coast.