Particulate air pollution causes a wide range of effects on human health, including disorders of the respiratory and cardiovascular systems, asthma and can cause mortality. Hence, the development of an efficient air quality forecasting and early warning system is an obvious and imperative need. The objective of this work was to investigate this forecasting possibility using linear models (such as ARX, ARMAX, outputerror and BoxJenkins), and Neural Networks (NNs). The input data for the models were meteorological variables and the 24h average PM
concentration of the present day, while the output was the 24h average PM
concentration predicted for a 3day horizon. The results revealed that all the models yield fairly good estimates, but the BoxJenkins model showed the best fit and predictability.
Keywords: air quality forecasting, linear modelling, neural networks, particulate matter, public health, air pollution, Brazil, early warning systems