Flower visitation through the lens: Exploring the foraging behaviour of Bombus terrestris with a computer vision-based application

Abstract

To understand the processes behind pollinator declines, and thus to maintain pollination efficiency, we also have to understand fundamental drivers influencing pollinator behaviour. In this study, we aim to explore the foraging behaviour of wild bumblebees, recognizing its importance from economic and conservation perspectives. We recorded Bombus terrestris on Lotus creticus, Persicaria capitata, and Trifolium pratense patches in five-minute-long slots in urban areas of Terceira (Azores, Portugal). For the automated bumblebee detection, we created computer vision models based on a deep learning algorithm, with custom datasets. We achieved high F1 scores of 0.88 for Lotus and Persicaria, and 0.95 for Trifolium, indicating accurate bumblebee detection. We found that flower cover per cent, but not plant species, influenced the attractiveness of flower patches, with a significant positive effect. There were no differences between plant species in the attractiveness of the flower heads. The handling time was longer on the large-headed Trifolium than those on the smaller-headed Lotus and Persicaria. However, our result did not indicate significant differences in the time bumblebees spent on flowers among the three plant species. Here, we also justify computer vision-based analysis as a reliable tool for studying pollinator behavioural ecology.