The field of movement ecology has benefited hugely from tags that allow animals to be tracked as they use the landscape, and such information has been vital to many conservation efforts. However, tags for tracking bees are either too heavy for the study of most species, too expensive, or are unable to function in complex environments, precluding study of bees in many natural habitats.
We have developed and deployed a prototype landscape-scale bee tracking method using rotating Bluetooth transmitters placed across a landscape and <40mg tags attached to foraging bees. Power constraints cause uncertain and noisy data, so a Gaussian Process prior is placed on the flight path, incorporating our assumptions around possible flight paths made by insect foragers. Doubly stochastic variational inference is used, which results in ‘probabilistic triangulation’ of the probable flight path the bee took.
The system has successfully tracked and inferred the movement path of foraging Bombus terrestris workers through a complex outdoor landscape. Preliminary work has begun on integrating sensors including photodiodes and accelerometers with the tags to infer behaviour alongside position, enabling biologging of flying insects for use in fields such as energetics.

Bio-sketch:
Mike Smith is a senior lecturer in Machine Learning in the School of Computer Science, at the University of Sheffield. He currently works on developing new methods for tracking small animals at a variety of scales, from lab to landscape. His focus is how Bayesian machine learning can be employed to extract as much information as possible from situations where there are substantial limitations on energy- and compute-. He is also leading a Leverhulme grant that includes the development of novel hardware, including microbatteries and on-board active learning.