Estimating urban noise levels from Multi-Scale and Multi-Spectral remote sensing imagery

Abstract

Establishing a high-quality urban sound environment is essential for the sustainable development of modern cities. Estimating the noise pollution levels in urban areas is integral to improving the overall well-being of city dwellers. However, current approaches to noise levels estimation present significant challenges. Existing approaches are highly data-dependent. They either rely on data from noise sampling networks or require urban geographical data related to noise. Moreover, the latter approach often involves relatively complex modeling processes. This reliance on data availability and granularity significantly constrains the applicability of these methods. In this study, we propose a novel framework for urban noise levels estimation, leveraging deep learning techniques and multi-scale, multi-spectral remote sensing imagery. Specifically, we utilize a noise recording device to sample sound pressure level (SPL) data through mobile measurements at various locations during the daytime, a Transformer-based model is then constructed to learn noise-related information embedded in the scale, spectral, and spatial contextual features of Sentinel-2 imagery. The extracted high-dimensional feature vectors are used to quantitatively estimate SPL, with the proposed Noise-Trans-Sentinel model achieving MAE, RMSE, and R² values of 3.48, 4.68, and 0.63, respectively. Finally, a SHAP method is employed to interpret the model, exploring the role of multi-scale and multi-spectral remote sensing information in urban noise levels estimation. Our proposed framework enables and validates low-cost, spatially continuous noise estimation in urban areas. It fills a critical gap by demonstrating, for the first time, that high-resolution urban noise mapping can be achieved solely from remote sensing imagery, without relying on dense sensor networks or GIS data. This research contributes to cross-modal studies in urban environmental science and informs the optimization of urban soundscapes.