The Arctic covers an area of about 16.5 million square kilometers and consists mostly of the Arctic Ocean. To track animals here and gather information about their population sizes, scientists actually have to rely on satellite imagery. And the task of spotting, identifying and counting individual animals in the images from high altitude is difficult and time-consuming. Artificial intelligences would be ideally suited for this task. A Dutch research group working on a project with seals in the Arctic has taken a big step in this direction and discovered something astonishing.
Using drones, researchers from the Royal Netherlands Institute of Marine Research NIOZ, Wageningen University WMR and the company Aeria tracked seals around Svalbard and took images from high altitude. They then compared these images to high-resolution satellite imagery and were able to process the drone footage so that they could incorporate it into a computer algorithm that would then independently search the satellite imagery for the seals. This will allow seals to be counted more accurately over a larger area and thus generate more information on population sizes.
The project “Seals – An agent-based modeling tool”, led by Geert Aarts of NIOZ and WMR, aims to use satellite imagery and drone footage to show how the habitat of Arctic seals is changing due to climate change and what impact this on these marine mammals. The researchers in the group are taking advantage of the fact that the quality of the imagery has improved massively in recent years. For example, the satellites used in the work can provide image resolutions of 30×30 centimeters from an altitude of over 600 kilometers, which is enough to identify even the small ringed seals, about 120 centimeters in size, with great accuracy. But to automate the analysis of such images, the researchers first had to teach the computer to recognize the different seal species. To do this, they used drone footage taken on Svalbard over the course of this summer.
The resolution of the drone images was tremendously high and helped to identify the seals and even assign the individual breathing holes to individual animals. Because the work was conducted in an area where fast ice, ice that is connected to the land, lies, it was easy for the team to map the location of the holes. On pack ice, which shifts with the currents and changes its appearance constantly, especially in spring and summer, such fixed points are admittedly more difficult. But the team, which had studied ringed seals as well as walruses on Svalbard, is convinced that other pack-ice-loving seal species, such as hooded and harp seals, will also benefit from their research and that the algorithm can be adapted accordingly. The main thing is that the images, both satellite and drone, are high enough resolution. “These images can then be fed into a machine learning algorithm and used to train a neural network,” explains study co-author and PhD student Jeroen Hoekendijk. He is currently working on this development with the Swiss university EPFL.
The fact that the images are of such high quality that the individual seal species are identifiable was already a highlight for the research team. But Aarts, Hoekendijk and the rest of the team made another surprising discovery while analyzing the footage: lines leading to the seals’ breathing holes. Upon closer inspection, the scientists were able to identify these lines as polar bear tracks. Sometimes even the individual paw prints were visible on the images. “Although we were aware of the potential of such satellite imagery for observing mammals from space, we were amazed when we noticed the white lines on the ice connecting the ringed seals’ breathing holes,” Geert Aarts explains.
This allowed the group to demonstrate the potential of these high-resolution images for the Arctic as well. In Antarctica, a similar technique is already being used to count penguins, and there they are also counting on the help of private helpers as part of citizen science projects.
For the seals of the Arctic, it would definitely be an important contribution to their protection, because if nobody knows how many seals are left, they could suddenly disappear, despite the view from the sky.
Dr Michael Wenger, PolarJournal
More on the topic