Keywords: auto-regressive models with exogenous inputs, ozone forecasting, photochemical pollution, stationary and non-stationary grey-box models
Objective evaluation of statistical models for next day ozone forecasting
Different stochastic auto-regressive non-stationary models with exogenous inputs have been identified and validated in order to predict maximum ozone concentration expected for the next day. Selected inputs were one-day forecasted meteorological variables and average concentrations of ozone precursors (compounds involved in ozone formation chemical reactions, i.e. NO, NO2, NMVOC). The model performance was evaluated through objective statistical indices and by comparing the predictions with those of a simple persistency model. The results indicate a better performance for a station located outside of the urban area, where there is a reduced influence of ozone precursor emissions and large-scale meteorological conditions are more influent.