Urban airborne pollutant analysis by a neural network

Abstract

To be more useful, the measurements have often been combined with some atmospheric physics and chemistry observations. The sources like traffic and power plants may be separable by the measurements. However, the overall follow up of the profiles of the pollution characteristics by the measurements requires so many components in measurement vector that the correlations and clear view of the profile is difficult to discover. This chapter discusses the phenomena associated with atmospheric pollution and in particular, with aerosol particles. It applies a neural network calculation procedure to airborne pollutants and prevailing meteorological conditions to develop new receptor modeling procedure. It uses the self organizing feature map (SOM) by T. Kohonen. In that neural network analysis, the data set was clustered for finding reasonable groups of measured vectra to explain the sources of the pollutants. In general, the results show that a neural network procedure can be applied to the data set based on measured concentrations of airborne particles and gaseous pollutants, and it serves a useful tool to analyze large data sets.