From place to perception: the influence of geospatial context on soundscape assessment

Abstract

Research in auditory perception has sought to understand which factors, especially non-acoustic ones, influence the personal interpretation of acoustic environments. To that end, spatiotemporal information has been suggested to link to noise generation and propagation as well as to shape the listeners’ context. In this study, a campaign of soundwalks collected 3757 in-situ daytime soundscape evaluations with broad geographical coverage across Flanders (Belgium). This dataset was used to assess how spatiotemporal context explains listeners’ experience. Firstly, the physical context of the soundwalks was characterized by resourcing open-source geospatial data, using indicators of anthropogenic activity, nature/land use, and traffic infrastructure. These features were then used in random-forest models to predict the soundscape's dimensions—pleasantness and eventfulness—achieving RMSEs of 0.414 and 0.312 (2-point scale), and R2 of 13.4 % and 9.6 %, respectively. In parallel, pleasantness and eventfulness were also modelled using perceived sound sources as predictors, yielding substantially higher accuracy and explanatory power than the geofeatures models. Feature impact interpretation with SHAP values and partial dependence analysis identified traffic-related features as key drivers of lower pleasantness; greenery could mitigate this effect, but only in low traffic density areas. Eventfulness was increased by road density as well as linked to proxies of anthropogenic activity and the built environment. Although a large portion of the variance in the soundscape dimensions remains unexplained by geofeatures, these models’ coherence with perceptual models supports their value as a preliminary resource for identifying priority areas for soundscape interventions.