Evaluation and comparison of statistical forecast models for daily maximum ozone concentrations

Three statistical models that estimate daily maximum ozone (O₃) concentrations in the lower Fraser Valley of British Columbia (BC) are specified using measured concentrations from two monitoring stations during the time period 1978–1985. The three models are (1) a univariate deterministic/stochastic model, (2) a univariate autoregressive integrated moving average (ARIMA) model, and (3) a bivariate temperature and persistence based regression model.The three models as well as a persistence forecast are tested by comparison with O₃ concentrations observed during 1986; it is concluded that the bivariate model is superior to both unvariate models and persistence. The ARIMA model has nearly the same predictive capability as persistance while the mixed deterministic/stochastic model performs the worst. This suggests that the traditional time series technique of decomposing a series into a trend, a cycle and a stochastic component may not be appropriate for O₃ air quality forecasting.