When honeybees are ready to establish a new colony, they initiate a coordinated procedure called swarming. The queen in residence departs with half the colony, leaving a set of virgin queens behind with the remaining workers. For beekeepers, swarming provides an opportunity to capture the departing bees and establish a new hive. To forecast a swarm, beekeepers regularly inspect their hives for the presence of larger honeycomb cells that host developing future queens. But those regular inspections are laborious. Now Martin Bencsik of Nottingham Trent University in the UK and his colleagues are automating the process by using machine learning.
Honeybees have a complex language encoded in their buzzing—for instance, newly emerged virgin queens vibrate their folded wings and send short pulses through the honeycomb to alert the hive. But the bees provide no obvious auditory cues that a colony and its old queen are preparing to swarm. Bencsik’s team hoped to find hidden patterns in the hive’s vibro-acoustic emissions. The researchers planted accelerometers in the heart of the hive to measure vibrations in the honeycomb caused by the colony’s buzzing. The bees, barely perturbed, encased the accelerometers in new honeycomb as the devices collected data.
Using the resulting vibrational spectra from swarming and nonswarming colonies, the researchers trained two machine-learning algorithms to search for buzzing patterns that might indicate a swarm. Each hour, the algorithms would evaluate whether the colony was preparing to swarm, and if they determined that it was, they would alert the researchers. The first algorithm based its decision on an hour of data at a time. During the swarming season of late spring to early summer, the algorithm correctly differentiated between swarming and nonswarming colonies 91% of the time, but it was ineffective at predicting off-month swarms. The second algorithm analyzed 10 days of successive buzzing to make its hourly predictions. It had an 80% success rate during the swarming season but performed better than the other algorithm year-round.
The researchers now are analyzing the short vibrational pulses of individual bees. Those pulses are an integral part of the honeybees’ complex language, so the encoded information could enhance researchers’ ability to predict swarming behaviors. The team aims to create a device that will monitor hives and warn of impending swarms so beekeepers can more efficiently care for their colonies. (M.-T. Ramsey et al., Sci. Rep. 10, 9798, 2020.)
