Abstract

Box-Jenkins Time Series (ARIMA) and the multivariate linear models (MLM) have been important and popular linear tools in air quality forecasting during the past decade for urban areas. On the other hand, artificial neural networks (ANN) recently have been used successfully as a nonlinear tool in several research studies of pollution forecasting. A hybrid model that combines both ARIMA and ANN tools was proposed to improve the unique capabilities of ARIMA and ANN tools in linear and non linear modeling on particulate matter forecasting. To examine the effectiveness of the proposed hybrid model over real particulate matter data, the time series of PM10 and meteorological data observed in ambient air during 2000-2006 at a site in Temuco, Chile, was used In 2005, this city was declared a non-attainment area for PM10, whose pollution is the result of a great economic growth, a fast urban expansion, woodstoves, industrial sources, and a strong diesel vehicles growth. Experimental results with meteorological and PM10 data sets indicated that the hybrid model can be an effective tool to improve the forecasting accuracy obtained by either of the models used separately, and compared with a statistical multivariate linear regression. This is an abstract of a paper presented at the 100th Annual Conference and Exhibition of the Air and Waste Management Association 2007 (Pittsburgh, PA, 6/26-29/2007).