Urban Airborne and Blood Lead Analyses Using Neural Networks

Abstract

A feed-forward back-propagation–type neural network was applied to airborne lead data collected around a busy highway interchange in Birmingham, England. The data were taken during the early 1970s, when lead emissions from vehicles were at their highest. The neural computation was capable of identifying realistic isopleths of lead concentrations around the interchange for summer/winter and daytime/nighttime conditions. A second neural study on blood lead levels taken from residents all over Birmingham was also undertaken. This analysis augmented the findings of the original study and highlighted the importance of age, sex, location, and age of house on residents' blood levels. It was also found that multiple linear least-squares regression analysis gave very poor results and should not be used for interpreting this type of highly nonlinear data. The neural analyses provided a relationship between airborne and blood lead concentrations, and suggested that opening the interchange had caused the blood lead levels in schoolboys, living adjacent to the highways, to increase by about 9%.