Receptor analysis using neural networks

The apportionment of a thermal power plant, a large steel mill and other sources of suspended particulate matter to the deterioration of air quality is analyzed through receptor model analysis. Samples of suspended particulate matter (SPM) used for this evaluation were collected in the city of San Nicolas, Argentina, at four monitoring sites during a three months period. These sites were located in the zones where a relatively high contribution of the thermal power plant to the ambient air concentration of NOx, SO2 and SPM was expected. Multielemental composition of SPM for 1) a number of selected samples, 2) the average measured emission profile of the thermal power plant and 3) surrogate emission profiles for relevant local sources, served as the basis for Principal Component Analysis. This evaluation led us to conclude that the sum of the mass of the measured metals and the concentrations of iron, lead and potassium constituted the significant variables to discriminate the role of the different sources at the receptor samples. The apportionment of the relevant sources was calculated using an artificial neural network that was designed with the previously mentioned variables as inputs.