Scientists counted elephants from space using satellites and AI

Landmarks that can be “seen from space”, from the pyramids to the Great Wall of China, are often talked about. What about something much smaller, like elephants or dolphins? They, too, can be detected using satellites and an algorithm trained to search.

Satellite imagery has long been used for counting wildlife. But an accurate result was obtained only if the image of the animal in the photo was in sharp contrast to the landscape and the environment. To date, no research has been able to locate animals in forests or open pastures.

African elephants are the first animals to be successfully counted from space as they move through a complex, diverse landscape, including grasslands and partially tree-covered savannahs. The new study used a highly accurate neural network model that was “trained” to automatically detect and count African elephants. The neural network used images that were taken in high resolution at an altitude of 600 kilometers above the Earth’s surface using the Maxar satellites (Worldview 3 and Worldview 4). Gray “drops” against the background of green spots of the forest, upon careful “examination” by the algorithm, turned out to be elephants wandering among the trees.

Counting and control of African elephants is usually done from low-flying aircraft. It is costly and takes many hours. The new survey technique allows you to scan large areas of up to 5000 km² in minutes, eliminating the risk of duplication. Then, computer machine learning algorithms analyzed these images and isolated the elephants from the complex background landscape. While this new study is still a private proof of concept, it is generally applicable at work.

Only 415,000 African elephants are left in the world due to poaching and extinction, according to the World Wildlife Fund. Their population has declined sharply over the past century. To preserve the species, it is necessary to conduct accurate monitoring, it is important to know where the animals are and how many there are.

The algorithm for counting and recognizing elephants from space was created by Dr. Olga Isupova, a scientist from the University of Bath (UK): “We just show the machine examples and say: This is an elephant, this is not an elephant.” She also learns to recognize small details that are elusive to the naked eye. Satellite imagery is increasing in resolution every two years and finer details can be seen with each increase. As satellite imagery improves, other smaller views may soon be captured from space.