Visual Neural Mapping Technique for Locating Fine Airborne Particles Sources

Abstract

A combination of two pattern recognition methods has been developed that allows the generation of geographical emission maps from multivariate environmental data. During such a projection into a visually interpretable subspace by a Kohonen self-organizing feature map, the topology of the higher dimensional variables space can be preserved, but parts of the information about the correct neighborhood among the sample vectors are lost. This loss can partly be compensated for by the additional projection of Prim’s minimal spanning tree onto the trained neural network. This new environmental receptor site modeling technique is theoretically discussed for measurements from single sampling sites. In order to obtain a further quantitative evaluation of such a combined mapping of minimal spanning tree and Kohonen neural network, the concept of a geographic unit circle (GUC) is introduced as well. The GUC around the single sampling site in Granite City, IL, yielded estimates of the emission levels, the trace element profiles, and the geographic directions for a number of airborne particle sources.