In a significant development, researchers have innovatively used artificial intelligence (AI) to map the expanse of large Antarctic icebergs in satellite images, in just 0.01 seconds. This new approach is a major shift away from the time-consuming manual efforts that were required previously.
Large icebergs are a crucial part of the Antarctic environment. They affect ocean physics, chemistry, biology and maritime operations, explains Anne Braakmann-Folgmann, lead author of the study, which was published in The Cryosphere. “Hence it is crucial to locate icebergs and monitor their extent, to quantify how much meltwater they release into the ocean,” she added.
However, in the photos provided by satellites that carry camera-like instruments, icebergs, sea ice and clouds all appear white, making it difficult to identify actual icebergs. Researchers have struggled to separate icebergs from sea ice, which looks brighter in satellite images as it is rougher and older. Moreover, smaller iceberg pieces which are found near icebergs are often grouped with the main iceberg by mistake, the European Space Agency explains in a press statement.
The new neural-network approach, developed by researchers from the University of Leeds, is based on the U-net design and has been trained using the images provided by the Copernicus Sentinel-1 radar mission to chart icebergs, even in challenging circumstances. “Its power lies in the neural networks' ability to understand intricate non-linear relationships and take the whole image context into account,” the statement explains.
The neural network has been successful in identifying the largest iceberg in each image, unlike comparative methods, which often choose slightly smaller icebergs in proximity.
The network constantly refined its predictions during the training based on the difference between the manually derived outline and the predicted result, the statement elaborated. Its training ends when it reaches its optimum performance.
According to the European Space Agency, the algorithm has been tested on seven icebergs, ranging in size from 54 sq km to 1052 sq km, which is the size of the city of Bern in Switzerland and Hong Kong, respectively. The dataset used for training included 15 to 46 images for each iceberg, across various seasons and the years 2014 to 2020. Moreover, a single image by Sentinel-1 per month per iceberg was used to ensure variety. The findings showed that the AI’s accuracy was 99%.
“Being able to map iceberg extent automatically with enhanced speed and accuracy will enable us to observe changes in iceberg area for several giant icebergs more easily and paves the way for an operational application,” Braakmann-Folgmann said in the statement.