Ecologists have a new way to observe bee activity without collecting the insects: record their sounds and use a classifier to identify the signals. A University of Massachusetts Amherst-led team developed two models, one for overall bee activity and another that separates honeybee flight, bumblebee flight and the buzzing used by bumblebees to release pollen from flowers.[2]
The models were trained and tested on 594 recordings from 43 sites, spanning five recording years and varied environments in the northeastern United States. They are available through Raven Pro, an acoustic-analysis platform used by researchers. The paper was published on 12 October, followed by a university account on 21 October.[2][1]
Reliable detections come with a trade-off
At a confidence threshold of 0.9, the models achieved 93%–97% precision: most events they labelled were correct. Recall was lower, at 13%–44%, so they missed many target events at that setting. The two measures describe different parts of performance and are both needed to understand the result.[2]
What the acoustic models measure
| Feature | Reported finding | Practical meaning |
|---|---|---|
| Precision at 0.9 threshold | 93%–97% | Most labelled events were correct |
| Recall at 0.9 threshold | 13%–44% | Many target events remained undetected |
| Detection range | Primarily within 50 cm | Microphone placement strongly shapes the sampled area |
| Field validation | Detections correlated with observer counts | Supports use as a measure of local bee activity |
The short effective range is particularly consequential. A microphone samples a small acoustic neighbourhood, and repeated sounds may come from the same insect. A survey can compare activity across places or times, but it needs a consistent recording setup and an interpretation matched to what the instrument actually detects.[2]
Listening fills gaps between field visits
Traditional sampling can produce specimens that specialists identify under a microscope. That remains important for many species. Lethal traps, however, also capture nontarget insects and require considerable processing after collection. Acoustic monitoring adds repeatable observation while leaving the subjects in place.[1][2]
The researchers see a role in places and conditions that are hard to sample repeatedly, including tree canopies and periods of extreme heat. Recordings can reveal when activity occurs during the day, adding a time dimension that an occasional visit or accumulated trap sample may not capture.[1][2]
For a restoration project, the method could make regular observation easier to maintain. Its strength is a dependable additional signal, provided that threshold, range and recording conditions remain part of the interpretation. More of the season can be heard without every observation requiring a captured bee.
Sources & further reading
- AI listens to bee buzzes to monitor activity without lethal traps University of Massachusetts Amherst via Phys.org · 21 Oct 2026 · Accessed 25 Oct 2026
- What’s all the buzz about? Passive acoustic monitoring of bees Approaches in Entomology: Methods, Philosophies, and Ethics · 12 Oct 2026 · Accessed 25 Oct 2026