AI is helping scientists measure a forest’s recovery from animal sounds

Researchers working in Ecuador’s Chocó forest used AI to examine animal sounds across 43 plots. Their 2023 study compared agricultural land, naturally regenerating forest and old-growth sites to investigate how wildlife communities change as forest returns.[1]

Listening to a community

Automated recorders can collect sound over extended periods, but the resulting archive can quickly exceed what researchers can listen to manually. A recognition model turns selected calls into searchable observations. The task is complicated by overlapping species, rain and other background sounds: a recording rarely contains one perfectly isolated animal.[2]

A published pipeline developed by Jack LeBien and colleagues illustrates the approach. Sound is converted into spectrograms, images representing frequencies over time. Experts supply examples of target calls and review candidate detections. A convolutional neural network then learns to recognise several species in the same recording. False detections are useful training material too, helping it learn which similar sounds should be rejected.[2]

Why the species list matters as much as its length

Imagine two monitoring sites that each contain twenty detected bird species. One has a community similar to the nearby reference forest; the other is dominated by species associated with open agricultural land. The total is identical, but the ecological interpretation differs. This is an illustrative comparison: a count describes how many species were recorded, while composition describes which species they were.

The Society for Ecological Restoration’s standards frame recovery against an appropriate reference ecosystem. That reference considers the characteristic species, physical conditions and ecological functions the project aims to recover. It also allows for environmental change. A practical assessment therefore needs a clear destination, rather than treating any increase in a single measure as success.[4]

From an audio archive to a conservation question

Arbimon, the Rainforest Connection platform used in this research field, combines recording storage with several forms of acoustic analysis. Its documented workflow starts with defining the research question, then collecting recordings, inspecting spectrograms and choosing an analysis. Users can annotate sounds, create templates and prepare training examples before interpreting the results.[3]

That range makes the archive useful at more than one level. A researcher can inspect a particular call, compare sites or return to the recordings with a revised analysis. The original audio preserves evidence behind the summary.[3]

For a restoration project, the most useful report connects those observations with its ecological goals: which characteristic communities are returning, where progress differs and which parts of the ecosystem need closer investigation. Repeated measurement then supports decisions about the next stage of restoration work.[4]

Report an error or suggest a correction