Using hierarchical Bayesian models to analyze acoustic data

Abstract

Acoustic data provide numerous challenges for statistical modeling and analysis. Such challenges include data that are multivariate, spatially and temporally correlated, non-Gaussian, and non-stationary. Longterm acoustic monitoring studies result in high-dimensional time series that are both statistically and computationally difficult to analyze. The hierarchical Bayesian framework is a statistical modeling framework applicable to a wide range of ecological problems that can account for the complexities of acoustic data in a logical and computationally efficient manner. The Bayesian structure enables the incorporation of previous research into models, and allows for parameter estimates, fitted values, and predictions to be obtained within the same algorithm. The hierarchical framework enables thinking about modeling in an ecologically relevant manner. Here, we describe the general framework and it’s broad applicability to the fields of bioacoustics and ecoacoustics, and use it to analyze a soundscape data set from western New York to predict the soundscape over a sample region in western New York from public road data. We build soundscape maps over the sample region, along with estimated uncertainty maps, to display the power of such a modeling framework. Wide adoption of such a framework in the fields of bioacoustics and ecoacoustics could enhance collaboration between researchers, increase our understanding of the acoustic environment, and improve our ability to answer ecologically-relevant questions using acoustic data.