Ecoacoustic recordings capture animal population and community changes over time

Abstract:

Distributed monitoring technologies are needed to fill the taxonomic, geographic, and temporal data gaps that beset our understanding of global biodiversity. Ecoacoustics is an emerging low-cost avenue for direct observations of the animal community using networks of autonomous sound sensors. Continuous acoustic surveys are more taxonomically inclusive than traditional field inventory methods, capturing lesser-known organisms (e.g. insects, bats, anurans) that are not routinely identified during time-restricted field surveys for more common taxa, such as birds. Here, we analyzed ecoacoustic data using two novel complementary approaches, one based on network theory to assess animal community assembly following Amazon forest degradation, and another using deep learning to detect the acoustic signatures of individual species in real-time. At the community-level, animal communication networks provided clear evidence of a sustained shift in faunal community composition following recurrent fires in Amazon forests. Animal communication networks in forests affected by repeated fire events were more homogenous and quieter than those in once-burned or selectively-logged forests. Importantly, biomass was not a robust proxy for biodiversity in degraded forests, highlighting the importance of 24-hour ecoacoustic data to track changes in animal communities. At the species-level, convolutional neural networks (CNN) enabled multi-label recognition of calls in complex, noisy, and biodiverse environments. In Puerto Rico, we developed a regional CNN that accurately identified a set of 24 bird and frog species (0.975 total-average-precision). These cutting-edge and generalizable deep learning methods open a pathway for continuous, long-term monitoring of populations based on in-situ sensor networks that feed data into the cloud. Together, these two methodological advances extend our understanding of ecosystems beyond common proxies and capture important differentiation of community composition and disturbance through the use of diurnal measurements. Ecoacoustics therefore offers a promising cost-effective solution for routine biodiversity surveys and long-term monitoring of animal community composition and populations.