Short-term prediction of urban NO2 pollution by means of artificial neuralnetworks

A neural network model for the short-term prediction of concentrations of u rban pollutants was developed and applied to the Turin (Northern Italy) air quality network. In particular, the study was focused on NO2 concentration s measured at five stations; t + 3 and t + 24 hour NO2 concentration foreca sting based on hourly meteorological and concentration data gave good agree ment with observed concentrations. This is particularly true for the mean c oncentration values and concentration distribution. The time of occurrence of peak values was correctly forecast but the amounts were generally undere stimated. To reduce this underestimation, an empirical step function was ap plied in the t + 24 case. This allowed an accurate estimate to be obtained of the few cases in which 50% of the air quality monitoring stations exceed ed the attention level (200 mug m(-3)) during the following day for at leas t one hour.