Answers are grounded in this paper and cite their sources.
Functional classifications of organisms by their ecological roles provide a clearer understanding of ecosystem processes and services. However, functional roles are often inferred from taxonomic relatedness rather than being directly measured, despite evidence showing that closely related taxa can differ significantly in their ecological contributions. Therefore, automating the recognition of ecological function could offer a more scalable and ecologically meaningful alternative to traditional taxonomy-based approaches. Here, we evaluated the potential of bioacoustic signals to directly classify the functional roles of flower-visiting bees in blueberry crops in southern Chile. We tested the model’s ability to distinguish effective and ineffective pollinators using Pre-trained Audio Neural Networks (PANNs), a family of pre-trained models based on convolutional neural networks (CNNs), and fine-tuned on the buzzing sounds of bees visiting blueberry flowers in southern Chile. The best-performing CNN model achieved an 79.72% Macro F1-score, far exceeding the 45.44% achieved by the random and majority-class baselines. Performance varied among functional groups (69–92%), reflecting the moderate effects of class imbalance. Models trained exclusively on floral sonication sounds (distinctive vibrations produced during pollen extraction) outperformed models based on flight sounds or a combination of both sounds, indicating that sonication encodes stronger functional cues. T-SNE visualizations revealed partial disentanglement of functional features in the model’s embeddings. Our results demonstrate, for the first time, that bee buzzing can be used to infer pollination function automatically. Although we acknowledge that the dataset originates from a single crop system (highbush blueberry) and geographic region, this bioacoustic approach offers a scalable, noninvasive tool for identifying effective pollinators and advancing pollination management. It also opens up new avenues for ecological monitoring, shifting the focus from taxonomic grouping to ecological function, although functional assignments remain based on species-level evidence.
No takes yet. Share an insight, caveat, or question.
Ferreira et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: