Counting elephants… from space!
Performing wildlife counts from space might sound like something from science fiction, but a team, led by Isla Duporge from Wildcru, have now developed an exciting new technology.
With populations of African elephants (Loxodonta africana) declining due to habitat loss and poaching, and whilst communities lose crops before being able to harvest them, it is essential for conservationists to have up to date data on elephant populations – on population sizes, to monitor trends and to understand their movements across large areas. Accurate information is essential so that conservations can make decisions on how to protect our elephants and our farmers.
Traditional monitoring methods are prone to errors. The most common survey techniques in savannah habitats are done by performing aerial counts from small planes, on the ground transect surveys and dung counts. Due to human error and other factors such as poor visibility, these methods can be inaccurate. They are also costly to undertake, so only take place sporadically. Inaccurate data can lead to the wrong conservation decisions being made about elephant populations, which is also a waste of funds and resources.
This is why a team from the University of Oxford (WildCRU and the Machine Learning Research Group) along with Dr Olga Isupova from the University of Bath and Dr Tiejun Wang from the University of Twente, collaborated in order to resolve these monitoring problems. They came up with a new method to survey elephants by remote sensing using satellite imagery along with automation detection using deep learning.
“Satellites can collect upward of 5,000 km² imagery in one pass captured in a matter of minutes, eliminating the risk of double counting. Repeat surveys are also possible at short intervals.”
This unobtrusive technique requires no ground presence, so does not disturb wildlife and there is no element of risk for on the ground researchers whilst they are collecting data. Areas which are inaccessible by vehicle or small planes can be easily surveyed, along with cross-border areas which normally need the requirement of permits in order that surveys can be undertaken.
There are challenges in using satellite monitoring. These include the processing of huge quantities of images. But the team has come up with a process of automated detection, so checking images, that would have taken months, can now be done in just a few hours. This technology also is less prone to errors.
Using photos from an Earth-observation satellite, orbiting 600km above the planet’s surface, the researchers created a customised training dataset They used Addo Elephant National Park, South Africa as their study site, then tested it with images from known elephant populations in the Maasai Mara, Kenya. Addo has a high concentration of elephants who move between open savannah and woodland habitats, so this technology had to take into consideration that elephants constantly change shape and colour. Elephants adopt a range of postures when playing and foraging, with mud and dust bathing meaning that they are not always a standard colour.
To develop this new monitoring method, the team created a customised training dataset, feeding data into a Convolutional Neural Network. They then compared results with human surveys. It was found that elephants can be detected in satellite imagery, with accuracy as high as human detection capabilities. Once their model was trained to find adults, it was then also able to identify calves. This new method is able to survey up to 5,000 sq km of elephant habitat on a cloud-free day.
The researchers believe that by using this new technology it will go a long way to assist in elephant conservation:
“Satellite remote sensing and deep learning technologies offer promise to the conservation of these majestic mammals. Conservation technologies open a new world of possibilities, to be embraced with the urgency necessitated by the sixth mass extinction and the global plight of biodiversity”.
You can read more about their methods, analysis and findings in their research paper Using very high-resolution satellite imagery and deep learning to detect and count African elephants in heterogeneous landscapes online.