Assessment and prediction of benzene concentrations in a street canyon using artificial neural networks and deterministic models

• Published in: Ecological modelling Vol. 193; no. 3; pp. 253 - 270
• Main Authors: Karakitsios, Spyros P., Papaloukas, Costas L., Kassomenos, Pavlos A., Pilidis, Georgios A.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2006
• Subjects: Artificial neural networks; Benzene; Modeling; Traffic flow patterns; Modeling; Benzene; Traffic flow patterns; Artificial neural networks
• ISSN: 0304-3800; 1872-7026
• DOI: 10.1016/j.ecolmodel.2005.07.024

The work deals with the comparison of two models: (i) an artificial neural network (ANN) and (ii) a semi empirical deterministic model (DET), used to simulate benzene concentrations in a street canyon. Furthermore, the response of models to ‘what if scenarios’ was also examined. The ANN was based on a training procedure using measurements collected in a specific street canyon (benzene concentrations, traffic density, vehicle's type distribution). The DET model was based on road traffic emission rate, wind speed and direction, and the geometrical characteristics of the road. Although both models produced very good results, given the limited amount of data available, the ANN succeeded slightly better than DET in predicting benzene concentrations. On the other hand, the ANN is less able to reproduce the effect of significant changes in traffic flow patterns on benzene concentrations. The results from the simulations indicate that the ANN is a promising technique for benzene modeling in an urban environment and in can be used for environmental management purposes.