Beyond Species-Based Assessments: Automatic Acoustic Recognition of Pollinator Function

Alef Iury Siqueira Ferreira, José Mesquita-Neto, NÁDIA FÉLIX FELIPE DA SILVA, Thierson Couto Rosa

Keywords: bioacoustics, ecological monitoring, pollinator function, deep learning, convolutional neural networks

TL;DR: We show that a deep learning model can determine a bee's ecological function as an effective pollinator from its buzzing sound, a scalable assessment for agriculture and ecology that moves beyond simple species identification.

Abstract: Although effective crop pollination depends on pollinator function rather than species identity, most monitoring tools rely on labour-intensive taxonomic assessments. Here, we evaluated whether acoustic signals could directly classify the functional roles of flower-visiting bees. Using convolutional neural networks (CNNs) trained on the buzzing sounds of bees visiting blueberry (Vaccinium corymbosum) flowers in southern Chile, we achieved a Macro F1-score of 85.5% in distinguishing true pollinators from non-pollinators—exceeding baselines and taxonomic classification. Our findings demonstrate that bee buzzing can reveal ecological function for the first time, enabling the automated recognition of effective pollinators in crop systems. This approach provides scalable tools for pollination management and opens a new direction for ecological monitoring that extends beyond taxonomy to functional roles.

Submission Number: 10