Bugs in the system: Passive acoustic monitoring of tropical rainforest insect communities

Insects play a crucial role in the terrestrial ecosystems, but are often labor intensive and difficult to census, particularly in habitats like the rainforest canopy. Despite urgent questions about the magnitude of change in insect populations and potential drivers of change, information on insect populations is often drawn from a relatively small number of locations and taxonomic groups. Passive acoustic monitoring (PAM) can provide detailed information on the spatial and temporal distribution of soniferous insects. When combined with machine learning approaches for automating event detection at multiple sites and multiple years, PAM can provide exceptionally detailed information about the dynamics of the calling insect community.

We present a case study of multi-year insect acoustic monitoring on Barro Colorado Island in Panamá, with a particular focus on katydids and other soniferous insects. This monitoring integrates deployments in the forest canopy, as well as intensive deployment in well-characterized sites in the forest understory. We implemented a bespoke neural network model, supported by custom-engineered data-processing, to cope with the wide range of durations and bandwidths of the sounds of many of the focal species and conducted detailed evaluation of performance. We used this model to evaluate the spatial, temporal, and seasonal detection patterns of rainforest insects, test for impacts of moonlight, evaluate effects of habitat structure, and test for associations between the species diversity and acoustic detection rates of insects and birds. In changing forests, consistent approaches for insect sampling will be key to understanding insect ecology and generating interpretable and actionable data.