WIT Press

PM10 Forecasting In Santiago, Chile


Free (open access)








485 kb

Paper DOI



WIT Press


P. Perez & J. Reyes


We show the results produced with a neural network model designed to forecast tomorrow’s maximum of the 24 h average of the PM10 concentrations in the city of Santiago, Chile. We were able to generate a daily report at 8 PM based on concentrations measured until 7 PM at five stations on the present day plus measured and forecasted values of meteorological variables. The report delivers maxima for the following day at the site of the same five stations. The dominant condition is assumed to be given by the greatest of the concentrations among the five forecasts. Since, according to the range where the concentrations fall, three classes of air quality are defined, good, bad and critical, we have focused our efforts on the correct prediction of this air quality class. We have trained the models using 2001 and 2002 data to test 2003 conditions and 2002 and 2003 data in order to forecast 2004 values. According to the results reported here, 87% of the 2003 days and 92% of the 2004 cases were correctly classified using the neural model. The forecasting accuracy is significantly better than that obtained with the model that is being used at present by environmental authorities. Keywords: air pollution, forecasting, PM10, neural networks. 1 Introduction Air quality in Chile is defined in terms of the 24 hour moving average (24MA) of concentrations of particulate matter with diameter smaller than 10 microns (PM10). These particles are small enough as to penetrate the respiratory tract of humans and for this reason they are potential disrupters of the normal functioning of the organism. According to the value of the observed maximum of the 24MA of PM10 (P24MAM) the day is classified as a day of class A (good) if P24MAM < 195 µg/m 3 , class B (bad) if 195 µg/m 3 ≤P24MAM < 240 µg/m 3 or class C (critical), if P24MAM ≥240 µg/m 3 . The condition for the whole city is


air pollution, forecasting, PM10, neural networks.