A Harmonized Data Model for Noise Simulation in the EU

Mobile surveys and machine learning can improve urban noise mapping: Beyond A-weighted measurements of exposure

Urban noise pollution is a major environmental issue, second only to fine particulate matter in its impacts on physical and mental health. To identify who is affected and where to prioritise actions, noise maps derived from traffic flows and propagation algorithms are widely used. These may not reflect true levels of exposure because they fail to consider noise from all sources and may leave gaps where roads or traffic data are absent. We present an improved approach to overcome these limitations. Using walking surveys, we recorded 52,366 audio clips of 10 s each along 733 km of routes throughout the port city of Southampton. We extracted power levels in low (11 to 177 Hz), mid (177 Hz to 5.68 kHz), high (5.68 to 22.72 kHz) and A-weighted frequencies and then built machine-learning (ML) models to predict noise levels at 30 m resolution across the entire city, driven by urban form. Model performance (r²) ranged from 0.41 (low frequencies) to 0.61 (mid frequencies) with mean absolute errors of 4.05 to 4.75 dB. The main predictors of noise were related to modes of transport (road, air, rail and water) but for low frequencies, port activities were also important. When mapped to the city scale, A-weighted frequencies produced a similar spatial pattern to mid-frequencies, but did not capture the major sources of low frequency noise from the port or scattered hotspots of high frequencies. We question whether A-weighted noise mapping is adequate for health and wellbeing impact assessments. We conclude that mobile surveys combined with ML offer an alternative way to map noise from all sources and at fine resolution across entire cities that may more accurately reflect true exposures. Our approach is suitable for noise data gathered by citizen scientists, or from a network of sensors, as well as from structured surveys.

Traffic Noise Prediction Applying Multivariate Bi-Directional Recurrent Neural Network

With the drastically increasing traffic in the last decades, crucial environmental problems have been caused, such as greenhouse gas emission and traffic noise pollution. These problems have adversely affected our life quality and health conditions. In this paper, modelling of traffic noise employing deep learning is investigated. The goal is to identify the best machine-learning model for predicting traffic noise from real-life traffic data with multivariate traffic features as input. An extensive study on recurrent neural network (RNN) is performed in this work for modelling time series traffic data, which was collected through an experimental campaign at an inner city roundabout, including both video traffic data and audio data. The preprocessing of the data, namely how to generate the appropriate input and output for deep learning model, is detailed in this paper. A selection of different architectures of RNN, such as many-to-one, many-to-many, encoder–decoder architectures, was investigated. Moreover, gated recurrent unit (GRU) and long short-term memory (LSTM) were further discussed. The results revealed that a multivariate bi-directional GRU model with many-to-many architecture achieved the best performance with both high accuracy and computation efficiency. The trained model could be promising for a future smart city concept; with the proposed model, real-time traffic noise predictions can be potentially feasible using only traffic data collected by different sensors in the city, thanks to the generated big data by smart cities. The forecast of excessive noise exposure can help the regulation and policy makers to make early decisions, in order to mitigate the noise level.