Markerless tracking of bumblebee foraging allows for new metrics of bee behavior and demonstrations of increased foraging efficiency with experience

Abstract

Bumblebees have become model organisms for cognitive ecology and social learning. Quantifying the foraging behavior of free-flying bees, however, remains a methodological challenge. We describe and provide the code for a method of studying bee free flying foraging behavior using the open source neural-network based markerless tracking software DeepLabCut. From videos of bees foraging in an arena we trained a neural network to accurately track the position of each bee. We then used this approach to study foraging behavior and show that the ratio between flying time and flower visiting time decreases over repeated foraging bouts, indicating increasing efficiency of bee foraging behavior with experience. Visit durations, a laborious metric to measure by hand, were significantly shorter on flowers that had previously been visited. This experiment illustrates the usefulness of DeepLabCut for objective quantification of behavior, and in this case study shows that previous experience increases bee foraging efficiency.