Publications

Air pollution is usually driven by a complex combination of factors in which meteorology, physical obstacles, and interactions between pollutants play significant roles. Considering the characteristics of urban atmospheric pollution and its consequent impacts on human health and quality of life, forecasting models have emerged as an effective tool to identify and forecast air pollution episodes. The overall objective of the present work is to produce forecasts of pollutant concentrations with high spatio-temporal resolution and to quantify the uncertainty in those forecasts. Therefore, a new approach was developed based on a two-step methodology. Firstly, neural network models were used to generate short-term temporal forecasts based on air pollution and meteorology data. The accuracy of those forecasts was then evaluated against an independent set of historical data. Secondly, local conditional distributions of the observed values with respect to the predicted values were used to perform spatial stochastic simulations for the entire geographic area of interest. With this approach the spatio-temporal dispersion of a pollutant can be predicted, while accounting for both the temporal uncertainty in the forecast (reflecting the neural networks efficiency at each monitoring station) and the spatial uncertainty as revealed by the spatial variograms. Based on an analysis of the results, our proposed method offers a highly promising alternative for the characterization of urban air quality.