Scientists have developed a new Artificial Intelligence, or AI, technique using radar images from Europe’s Copernicus Sentinel-1 satellite mission, to reveal how the Thwaites Glacier Ice Tongue in West Antarctica is being damaged by squeezing and stretching as it flows from the middle of the continent to the coast. Being able to track fractures and crevasses in the ice beneath the overlying snow is key to better predicting the fate of floating ice tongues under climate change.
A paper published today in Nature Geoscience describes how ESA’s development of the Copernicus Sentinel-1 satellites and routine synthetic aperture radar imaging of coastal Antarctica coupled with a novel artificial intelligence algorithm, has provided a glaciology tool with forensic science capability.
The paper’s authors – a team of scientists from the University of Leeds and the University of Bristol in the UK – developed an AI algorithm, originally used to identify cells in microscope images, to spot crevasses forming in the ice in Sentinel-1’s radar images of Thwaites Glacier Ice Tongue, as shown in the animation below.
Crevasses, or cracks opening in the moving ice mass, are indicators of stresses building up in the glacier.
Thwaites is a particularly important part of the Antarctic Ice Sheet because it holds enough ice to raise global sea levels by around 60 cm and is considered by many to be at risk of rapid retreat, threatening coastal communities around the world.
This new use of artificial intelligence will allow scientists to more accurately monitor and model changes to this important glacier.
The research focused on a part of the glacier system where the ice flows into the sea and begins to float – a point known as the grounding line. It forms the start of the Thwaites Eastern Ice Shelf and the western Thwaites Glacier Ice Tongue, which is also an ice shelf.
Despite being small in comparison to the size of the entire glacier, changes to the ice shelves could have wide-ranging implications for the whole glacier system and future sea-level rise.
The scientists wanted to know if crevassing or fracture formation in the glacier was more likely to occur with changes to the speed of the ice flow.
Using machine learning, the researchers taught a computer to look at radar images from the Copernicus Sentinel-1 mission and identify changes over the last decade.
Sentinel-1’s all-weather radar allows scientists to track ice movement and peer through snow cover to unveil Thwaites Glacier Tongue’s crocodile skin-like appearance, which is normally hidden from sight.
The analysis revealed that over the last six years, the Thwaites Glacier Ice Tongue has sped up and slowed down twice, by around 40% each time – from 4 km a year to 6 km a year before slowing. This is a substantial increase in the magnitude and frequency of speed change compared with past records.
The study found a complex interplay between crevasse formation and speed of the ice flow. When the ice flow quickens or slows, more crevasses are likely to form. In turn, the increase in crevasses causes the ice to change speed as the friction between the ice and underlying rock alters.
Dr Anna Hogg, a glaciologist at the University of Leeds, said, “Dynamic changes on ice shelves are traditionally thought to occur on timescales of decades to centuries, so it was surprising to see this huge glacier speed up and slow down so quickly.
“The study also demonstrates the key role that fractures play in un-corking the flow of ice, a process known as unbuttressing.
“Ice-sheet models must be evolved to account for the fact that ice can fracture, which will allow us to measure future sea-level contributions more accurately.”
Trystan Surawy-Stepney, lead author of the paper and a doctoral researcher at the University of Leeds, added, “The nice thing about this study is the precision with which the crevasses were mapped.
“It has been known for a while that crevassing is an important component of ice-shelf dynamics and this study demonstrates that this link can be studied on a large scale with beautiful resolution, using computer vision techniques applied to the deluge of satellite images acquired each week.”
Satellites orbiting Earth provide scientists with new data over the most remote and inaccessible regions of Antarctica. The radar on Sentinel-1 allows places like Thwaites Glacier to be imaged day or night, every week, all year round.
ESA’s Dr Mark Drinkwater commented, “Studies like this would not be possible without the large volume of high-resolution Antarctic ice-sheet image data provided by Sentinel-1.
“By continuing to plan and prepare future polar-orbiting missions, we can carry on supporting work like this and broaden the scope of scientific research on vital areas of Earth’s climate system.”
As for the Thwaites Glacier Ice Tongue, it remains to be seen whether such short-term changes have any impact on the long-term dynamics of the glacier, or whether they are simply isolated symptoms of an ice shelf close to its end.