Saving red squirrels with AI and cloud computing

A red squirrel eating a nut
(Image credit: Huawei Technologies)

Red squirrels are close to becoming extinct in the UK. Over the last 13 years, the species has lost roughly 60% of its habitat in England and Wales, and it’s estimated that fewer than 290,000 red squirrels remain.

Over the last 100 years, or so, red squirrel populations have been declining largely due to the introduction of grey squirrels from North America, Dr Stephanie Wray, chair of the Mammal Society, tells IT Pro. Not only are their trans-Atlantic cousins much larger, but they also carry a disease called the parapox virus, which is harmless to them, but fatal to red squirrels.

Monitoring these animals has been a critical part of conservation efforts, as it enables scientists to better understand the habitats they live in, their behaviours and how they relate with other species. Monitoring, however, has traditionally been very labour intensive, and involves sending people into the wild to look for drays – the characteristic nests that red squirrels make – which can often be made in difficult-to-find places, such as high up on trees.

To that end, the Mammal Society has teamed up with the University of Bristol, the Rainforest Connection (RFCx), and networking giant Huawei, to preserve red squirrels. This landmark project combines bioacoustics with cloud computing and artificial intelligence (AI) to help conservationists assess and monitor squirrel populations, with a view to preventing their extinction and restoring their numbers.

Listening in

The project uses custom-built Guardian and Audiomoth monitoring devices in combination with Huawei software to analyse the natural noise of the environment; it’s the first time this equipment has been used this way in the UK.

“What this technology does is it allows us to put up recording devices that are picking up the whole soundscape in real-time,” Wray explains. “Then, by using the AI to recognise the cause of the different species of squirrels, we can just sort of filter out the bits that were interested in.”

A Guardian is essentially a minicomputer with solar panels, an antenna, and a microphone. It continuously records audio and connects through a mobile network to upload these sound files to a cloud repository. These devices use edge computer processing to perform AI-based processes themselves, then send out alerts through satellite or mobile signals. An Audiomoth, meanwhile, is an offline low-cost, open-source acoustic monitoring device used for monitoring wildlife.

Chrissy Durkin, RFCx director of international expansions, says the University of Bristol, in collaboration with the Mammal Society, is deploying these offline devices to collect some valuable initial data on red squirrels.

The Guardian bioacoustic device in a woodland

(Image credit: Huawei Technologies)

Guardian devices will be installed once initial data from Audiomoth devices are analysed

“The first step is we really need to have a lot of examples of the calls red squirrels make to start to understand, in detail, where they're present, and to be able to recognise them and understand their distribution,” she says.

By understanding the calls of the red squirrel, the project can produce an initial distribution map of where red squirrels are really based. This helps to determine where the long-term Guardian monitoring devices can be positioned. These devices can then give the project year-round data as well as an understanding of which species are present at any given time.

The data is then fed into RFCx’s platform, Arbimon, a biodiversity monitoring system that, according to Durkin, is useful for understanding and analysing the species present. It eliminates traditional sound analysis, which has involved office-based workers listening to audio files one by one, trying to interpret the species they're interested in by ear. Arbimon, instead, serves as a central hub for sound data funnelled into it from various recording devices. It can also perform several types of analysis, such as pattern matching, soundscape analysis, and random forest model analysis – a machine learning technique that’s used to solve regression and classification problems (it doesn’t have anything to do with actual forests).

No time to die


How to manage AI risk

Recommendations from the Cyber Resilience Think Tank


Perhaps the most important tool is pattern matching, which drastically reduces the time taken for scientists to label species calls, from up to two years to just a few hours. The system works by running red squirrels’ calls from an audio file, for example, as a template against all the data that’s been collected. “It’s essentially visual analysis within the spectrogram,” Durkin explains.

It then finds all the matches within a certain threshold for the species call. “Using that tool, alone, you can figure out where the red squirrel has been present across all of your data insights,” she adds. A scientist can then examine all matches and cross-reference, by ear, which calls correctly match with a red squirrel’s, and which don’t.

Engineers installing bioacoustic devices in the wild

(Image credit: Huawei Technologies)

Engineers installing bioacoustic devices in the wild

One possible expansion of this technology is to label a dataset in order to build out an AI model for the red squirrel. In the meantime, the data collected can easily produce distribution maps to reveal where red squirrels live. With these maps, scientists can create conservation plans to preserve, and bolster, the population of red squirrels and make sure their species is sustainable.

The red squirrel conservation project is still in its early stages, so the plans haven’t been finalised. Durkin says that in other projects she’s been involved with, however, data gathered has allowed conservationists to run a breeding programme, such as in the campaign to preserve Puerto Rican parrots. In this project, conservationists used distribution maps to find the best places to release birds bred in captivity back into the wild for the best chances of survival.

The initial analysis of audio recordings, in addition to the generation of population distribution maps, will help determine where the University of Bristol and the Mammal Society will position their permanent monitoring stations. Once this is accomplished, RFCx will physically travel to these locations to install the more advanced Guardian devices.

Looking ahead, Huawei’s technology could also be used in a host of further conservation projects. Although the initial layout has been made in the interests of protecting red squirrel populations, Wray suggests these same systems can be used to recognise the sounds of any species in order to help conservationists understand the extent of biodiversity in the UK, and the threats that British wildlife faces.

Rene Millman

Rene Millman is a freelance writer and broadcaster who covers cybersecurity, AI, IoT, and the cloud. He also works as a contributing analyst at GigaOm and has previously worked as an analyst for Gartner covering the infrastructure market. He has made numerous television appearances to give his views and expertise on technology trends and companies that affect and shape our lives. You can follow Rene Millman on Twitter.