Modeling roadway traffic noise in a hot climate using artificial neural networks

Road traffic noise is the most prevalent form of environmental noise pollution, and it can either be measured in the field or predicted through verified mathematical models. While many road traffic noise models are available around the world, these models cannot be simply generalized because local conditions affecting such noise (e.g., vehicle type and weather) vary from one locality to another. In this paper, the Artificial Neural Network (ANN) technique was employed to model road traffic noise in a city with known hot climate, namely Sharjah City in United Arab Emirates. Toward this end, field data were collected from three different road sites which resulted in more than 420 hourly measurements of noise level, traffic volume and classification, average speed, and roadway temperature. Overall, a total of 16 feed-forward back-propagation ANN models with one and two hidden layers were undertaken in this research. The best-performing models were compared with two conventional models: The Basic Statistical Traffic Noise model (BSTN) and the Ontario Ministry of Transport Road Traffic Noise model (ORNAMENT). In general, results showed that ANN models outperformed the conventional models. Furthermore, the performance of both ANN and conventional road noise models was further improved by including roadway temperature. While the reported work emphasizes the importance of adapting models to different local conditions, it also encourages others to consider the roadway temperature when modeling traffic noise, especially in areas of hot climate. Given that ANN models are often designated as black-box, a few techniques were utilized to expand the capabilities of ANN models to offer explanatory insights in modeling the inter-relations between the variables considered in this study. Generally, it was found that the ANN model was capable of identifying these relations as expected. In-depth analysis showed that the distance from the edge of road was found to be the most significant contributing factor whereas heavy-vehicle volume was surprisingly found to be the least. (C) 2017 Elsevier Ltd. All rights reserved.