Calibration and Validation of a Measurements-Independent Model for Road Traffic Noise Assessment

Featured Application: The road traffic noise predictive model presented in this paper can be calibrated without field measurements and can be applied in several traffic conditions, with an average error lower than 2 dBA. The assessment of road traffic noise is very important for the health of people living in urban areas. Noise is usually assessed by field measurements, and predictive models play an important role when experimental data are not available. Nevertheless, when they are based on regression techniques, predictive models suffer from the drawback of strong dependence on the calibration data. In this paper, the authors present a regressive model calibrated on computed noise levels without the need for field measurements. The independence from field measurements makes the model flexible and adjustable for any road traffic condition possible. A multilinear regression technique is applied to establish the correlation between the computed equivalent noise levels and several independent variables, including, among others, traffic flow and distance. The model is then validated on a large field measurement database to check its efficiency in terms of prediction accuracy. The validation is performed both via error distribution analysis and using different error metrics. The results are encouraging, showing that the model provides good results in terms of the average error (less than 2 dBA) and is not susceptible to the presence of outliers in the input data that correspond to unconventional conditions of the traffic flow.