Statistical Environmental Noise Prediction Using Data from a Noise Monitoring Network

In this study, we developed a statistical method to predict environmental noise using machine learning models trained on measured data from a noise monitoring network in Seoul, Korea. Daytime and nighttime annual equivalent noise levels were used as dependent variables, and traffic, climate, topographic, landscape, and land-use characteristics were used as explanatory variables. Feature variables were aggregated within buffer distances of 20 to 80 m around monitoring sites to identify the optimal range of influence. Among several models that we evaluated, an extremely randomized trees (Extra-Trees) model showed the highest predictive performance with a coefficient of determination of 0.729 and a root mean square error of 3.4 dB(A) for daytime noise at a buffer radius of 30 m. We then applied Shapley additive explanations (SHAP) to analyze the contribution of each variable, and the results showed that factors related to traffic were the most influential, followed by land-use characteristics. The trained model was applied to a 10 m × 10 m grid to generate a statistical noise map. This study highlights the potential of explainable machine learning-based statistical noise mapping for urban noise management and land-use planning.