Combination of time series forecasting models with a microscopic and stochastic approach to predict road traffic noise

Road traffic noise represents a major source of environmental pollution, and its prediction remains a critical task. This challenge particularly emerges when traffic data are not available, such as during the design phases of new infrastructures, where it becomes necessary to predict the noise exposure affecting nearby residents, even in the absence of measurement data. To address this issue, this work augments a previously developed microscopic and stochastic-core traffic noise model, integrating it with forecasting time series models for traffic flows and average vehicle speeds. This integration produces a hybrid model that enables the estimation of hourly traffic noise levels based solely on historical traffic patterns, even in the absence of direct traffic observations for the period under investigation. The methodology has been evaluated through a statistical analysis of simulated noise levels, with a focus on error distribution and conventional error metrics. The mean error of 0.43 dBA and the mean absolute error of 1.30 dBA confirm the accuracy of the proposed approach for estimating road traffic noise in data-scarce scenarios. A comparison with the CNOSSOS-EU model's performance highlights the possibility of using such methodology in early-stage infrastructure design and planning.