Detecting Noise Pollution through a Spatio-temporal Deep Learning Model: A Case Study of Barcelona

Noise pollution prediction affects the quality of life. Traditional approaches based on mathematical and statistical models face limitations in capturing the complexity of the spatiotemporal relationships that characterize noise patterns in cities. Moreover, methods based solely on time series tend to ignore spatial dependencies, while graph-based architectures, such as GCNs, lack an adequate representation of temporal dynamics.This study proposes a combination of Graph Convolutional Networks (GCN) and LSTM networks to overcome these limitations, simultaneously capturing the spatial and temporal dependencies present in monitoring station data from Barcelona. The proposed model, trained with historical noise level data, demonstrates superior performance compared to other approaches, reducing the error compared other approaches, validating the effectiveness and suggest that the proposed methodology can serve as a robust tool for predicting urban noise.