Statistical models for traffic noise at signalized intersections

A total of 14,235 noise levels measurements were utilized in developing statistical models that have the capability to predict different noise levels including: equivalent, maximum, or minimum noise level in terms of parameters affecting each level. Different parameters expected to have an effect on noise levels were collected. These parameters included traffic volume, composition of traffic, traffic speed, horn using effect, number of lanes, width of lanes, approach width, road slope, and pavement surface texture. The parameters affecting each noise level were selected based on simple correlation matrices, scatter plots, and statistical tt-test. Different forms of models were evaluated for each noise level. The best model describing the relationship between each noise level and parameters affecting it are presented in this paper. The reliability of the nonlinear developed models were judged based on coefficient of multiple determination (R2R2), the significance of each variable at αα-level of 0.05, and the standard error of the estimates. While the reliability of linear developed models were judged based on the general linear regression tests represented by FF-value and tt-value in addition to the coefficient of multiple determination (R2R2), the significance of each variable at αα-level of 0.05, and the standard error of the estimates.Based on the analysis of the collected data, three groups of models were developed. The first group of models predicts the equivalent noise level in terms of traffic volume, traffic speed, distance, heavy vehicles and British Pendulum Number (BPN). The second group presents models that describe the relationships between maximum noise levels, heavy vehicles and use of horn. While, the third group presents minimum noise level prediction models in terms of BPN and lane width. A verification of the developed models was performed by comparing the predictive noise levels with those measured at different sites. Results of this verification indicated that the developed models were found to have good prediction capability.