Assessing soundscape disturbance through hierarchical models and acoustic indices: A case study on a shelterwood logged northern Michigan forest

Abstract

Abstract Assessing the effects of anthropogenic disturbances on wildlife and natural resources is a necessary conservation task. The soundscape is a critical habitat component for acoustically communicating organisms, but the use of the soundscape as a tool for assessing disturbance impacts has been relatively unexplored until recently. Here we present a broad modeling framework for assessing disturbance impacts on soundscapes, which we apply to quantify the influence of a shelterwood logging on soundscapes in northern Michigan. Our modeling approach can be broadly applied to assess anthropogenic disturbance impacts on soundscapes. The approach accommodates inherent differences in control and treatment sites to improve inference about treatment effects, while also accounting for extraneous variables (e.g., rain) that influence acoustic indices. Recordings were obtained at 13 sites before and after a shelterwood logging. Four sites were in the logging region and nine sites served as control recordings outside the logging region. We quantify the soundscapes using common acoustic indices (Normalized Difference Soundscape Index (NDSI), Acoustic Entropy (H), Acoustic Complexity Index (ACI), Acoustic Evenness Index (AEI)) and Welch Power Spectral Density (PSD) values. We build two hierarchical Bayesian models to quantify the changes in the soundscape over the study period. Our analysis reveals no long-lasting effects of the shelterwood logging on the soundscape as measured by the NDSI, but analysis of H, AEI, and PSD suggest changes in the evenness of sounds across the frequency spectrum, indicating a potential shift in the avian species communicating in the soundscapes as a result of the logging. Further, our analysis confirms previous findings that the ACI does not accurately reflect changes in landscape configuration. Multiple model validation techniques (i.e., comparison of parameter estimates and the widely applicable information criterion (WAIC)) reveal our proposed hierarchical Bayesian models outperform more simple models used for hypothesis testing. Acoustic recordings, in conjunction with this modeling framework, can deliver cost efficient assessment of disturbance impacts on the landscape and underlying biodiversity as represented through the soundscape.