A Statistical Model for Forecasting Hourly Ozone Levels During Fire Season

Concerns about smoke from large high-intensity and managed lowintensity fires have been increasing during the past decade. Because smoke from large high-intensity fires are known to contain and generate secondary fine particles (PM2.5) and ozone precursors, the effect of fires on air quality in the southern Sierra Nevada is a serious management issue. Various process-based models have been developed for forecasting PM and ozone levels in the presence and absence of fires. Although these models provide deterministic predictions, few of them give measures of uncertainties associated with these predictions. Estimates of uncertainties are essential for model evaluation and forecasting with known precision levels. In this chapter we present a statistical procedure for forecasting next-day ozone levels at given sites. The statistical model takes into account some of the known sources of ozone fluctuations, including changes in temperature, humidity, wind speed, wind direction and, during fire season, effects of smoke from fires. Other sources of variation not directly accounted for in the model—e.g., variability in daily amount of ozone produced by sources other than fire—are included in the uncertainty measure as random effect variables. The advantage of a model that is capable of estimating mean effects and uncertainties simultaneously is that evaluation of model performance is immediate and predictions are available with specific precision levels. The ability of the model in making accurate forecasting with specified precisions is demonstrated by applying it to real data set of observed ambient ozone and weather values at two sites in the Sierra Nevada for the Corresponding author: E-mail: [email protected] Author's personal copy