Scientists buzzing with hopes for AI bee research

PhD student Alixandra Prybyla uses a parabolic microphone to record bees at Orkney's Ring of Brodgar

There are more than 20 species of bumblebee in the British Isles. But some have gone extinct since the Second World War - and others are at risk.

Now it is hoped teaching a computer to automatically identify threatened bees in the wild could help conserve them.

Researchers need to teach the system to recognise the different buzzes every species makes.

They do that by recording the noise and adding comprehensive details of the bee that produced it.

Alixandra Prybyla, who's studying for a PhD at the University of Edinburgh, and her undergraduate colleague Ava Nelson have been searching for bees at the rich wildflower meadows round the Ring of Brodgar in Orkney.

A Great Yellow Bumblebee foraging on clover at the Ring of Brodgar

The area is home to a number of species - including the great yellow bumblebee, which is one of the rarest bumblebees in the British Isles.

Alixandra Prybyla told BBC Radio Orkney that you have to think of the body of a bee as being "like a tiny musical instrument".

"In the same way that you would pluck a string on a guitar and a vibration would move through that string and into the body of the guitar, the same thing happens to a bumblebee", she says.

The vibrations it makes when it flies are shaped in its body and are emitted as a unique sound.

So human ears can hear the difference, for example, between the low buzz produced by the relatively slow wing beats of a buff-tailed bumblebee queen and the much higher - almost whiny - noise made by a small early bumblebee worker.

But more subtle or complex distinctions can only be detected using scientific instruments to analyse the frequencies the insects emit.

Undergraduate student Ava Nelson is also working on the bioacoustics project

It would take a human researcher "thousands of hours" to work through all the possibilities. So Alixandra Prybyla and her students are employing artificial intelligence.

"We're recording thousands and thousands of bee sounds and taking all sorts of environmental data, all sorts of morphological data, and we're giving it to this artificial intelligence algorithm that we're programming, and teaching it. Teaching it what to look for."

That means after they've recorded an individual they live trap it in a net, chill it to put it into torpor - a hibernation state so it can be handled safely. They then identify, weigh and measure it, tag it, warm it up again and release it.

The plan is to build up a database which links the sounds with all those details, along with environmental information like how warm it is, how much pollen the bees is carrying, and what time of year it is .

The white dot on this Garden Bumblebee shows it has previously been recorded, trapped, and measured by the researchers, then released back into the wild.

By measuring bees over the course of their full season, she says, they should get a full picture of all the variables that can affect what a bee sounds like.

Bees make buzzes for all sorts of different reasons. Sonication is a high-pitched buzz to get flowers to release pollen and they also have aggression buzzes - which are communication buzzes.

But in order to make sure that the system is comparing like with like, they focus on recording "foraging buzzes" - the noise a bee makes as it goes from flower to flower.

In future it is hoped that the system will be able to recognise and analyse recordings generated automatically by microphones left in bee habitats.

It had been thought Great Yellow Bumblebees were restricted to coastal machair but researchers now know they are also found in flower meadows like this one in Orkney

"It can difficult, and time consuming to have experts go to all the different places and identify the bees, do population examinations, and get the kind of data needed to do conservation interventions", Alixandra Prybyla says.

"So we're hoping that this technology can be deployed as something called a 'remote acoustic monitoring station'.

"You'd set up microphone arrays which would take in all sports of buzzes and automatically they could be identified. And if we notice concerning changes, or things that a researcher would flag if they were in the field. We could come out and see what's going on."