Effective 1-day ahead prediction of hourly surface ozone concentrations in eastern Spain using linear models and neural networks

The aim of this research was to develop pure predictive models in order to provide 24 h advance forecasts of the hourly ozone concentration for the rural site of Carcagente (Valencia, Spain) and the urban sites of Paterna (Valencia, Spain) and Alcoy (Alicante, Spain) over 4 years from 1996 to 1999. The peculiarity of the model presented here is that it uses past and previously predicted information of inputs exclusively, thus being this is the first genuine 24 h advance O3 predictive model with neural networks. We used autoregressive-moving average with exogenous inputs (ARMAX), multilayer perceptrons and FIR neural networks. Five performance measures yield reasonably good results in the th…