09/11/2023
25 scenes
1 Like it
In a groundbreaking development, researchers at the University of Leeds have unveiled a neural network that can quickly and accurately chart the expansion of large Antarctic glaciers in satellite images, accomplishing the task in just 0.01 seconds. This novel approach is in stark contrast to the laborious and time-consuming manual efforts previously required.
Anne Brockman-Folkman, lead author Findings Published today in Cryosphere, conducted her research while she was a doctoral student at the University of Leeds in England. Now working at Norway’s Arctic University in Tromsø, he emphasized the importance of large glaciers in the Antarctic environment.
„Giant glaciers are important components of the Antarctic environment. They affect ocean physics, chemistry, biology and, of course, ocean activity. Therefore, it is important to identify glaciers, monitor their size, and quantify how much meltwater they release into the ocean.
The Copernicus Sentinel-1 radar mission plays a key role in an innovative approach to mapping rocks using artificial intelligence, providing images of glaciers regardless of cloud cover and lack of daylight.
In images from satellites carrying camera-like instruments, glaciers, sea ice and clouds all appear white, making it difficult to pick out real glaciers.
Radar view of Sentinel-1
Whereas in most radar images, as returned by Sentinel-1, icebergs appear as bright objects against a dark ocean and sea-ice background.
However, when the surroundings are complex, it can sometimes be difficult to distinguish icebergs from sea ice or even beaches.
Dr Brockman-Folkman explained, „Sometimes the sea ice around glaciers is harder and older, and therefore looks brighter in satellite images. The same is true for wind-roughened seas.
„Furthermore, smaller glacier fragments, which often occur near glaciers, continue to lose ice around their edges, which can easily be mistakenly grouped with the main glacier.
„Additionally, the Antarctic coastline can resemble glaciers in satellite images, so standard segmentation algorithms often select coastlines instead of actual glaciers.”
However, the new neural network approach excels at mapping glacier size even under these challenging conditions. It has the power to understand the complex nonlinear relationships of neural networks and take into account the entire image environment.
To effectively monitor changes in glacier area and thickness, which is essential to understanding how glaciers melt and release freshwater and nutrients into the ocean, pinpointing a specific giant glacier for continuous monitoring is critical.
Using AI to measure the size of glaciers
The neural network introduced in this study is highly efficient at identifying the largest glacier in each image, unlike comparative methods, which often select nearby slightly smaller glaciers.
The architecture of the neural network is based on the famous U-Net design. It was trained using Sentinel-1 images revealing giant glaciers in various settings, with manually derived outlines serving as targets.
Throughout the training process, the system continuously refines its predictions, adjusting its parameters based on the difference between the manually obtained outline and the predicted result. Training stops automatically when the system reaches its optimal performance, ensuring its adaptation and success in new examples.
The method was tested on seven glaciers ranging in size from 54 sq km to 1052 sq km, equivalent to the areas of Bern in Switzerland and Hong Kong, respectively.
A diverse dataset covering different seasons and years 2014–2020 was compiled from 15 to 46 images for each glacier.
One Sentinel-1 image per month per glacier was used to confirm the dataset type. With 99% accuracy, the results are impressive.
Dr. Brackman-Folkman added, „Automated glacier extent mapping with improved speed and accuracy will make it much easier to observe changes in glacier area for many large glaciers and pave the way for operational use.”
ESA’s Mark Drinkwater noted, “Of course, satellites are essential for monitoring changes and understanding processes that occur far from civilization. This new neural network automates what would otherwise be a manual and labor-intensive task of detecting and reporting glacier size.
„We congratulate the team for introducing this innovative machine learning approach to achieve a robust and accurate approach to monitoring changes in the vulnerable Antarctic region.”