One of the main advantages of some AI models is that they can significantly speed up demanding tasks. Researchers from the University of Leeds have introduced a neural network that they say can map the outline of a large glacier in just 0.01 seconds. Scientists can determine the location of large glaciers by hand. However, it is not possible in this way to monitor changes in the area and thickness of glaciers, or, for example, how much water they release into the seas.
“Giant glaciers are an important part of the Antarctic environment,” said Anne Braakmann-Folgmann, lead author of the paper. “They affect ocean physics, chemistry, biology and, of course, maritime traffic. That is why it is very important locate glaciers and track their extentso that we can calculate how much meltwater they release into the ocean.“
So far, manual mapping has proven to be more accurate than automated approaches, but it has been time and labor intensive. Researchers trained an algorithm called U-net based on images taken by ESA’s Copernicus Sentinel-1 Earth monitoring satellite. The algorithm was tested on seven glaciers. The smallest of these was about the size of Bern, Switzerland, and the largest was about the same size as Hong Kong.
The new model is said to outperform previous automation attempts with 99 percent accuracy. In addition, existing methods also had a problem with distinguishing icebergs from sea ice and other elements, for example. A significant advantage of the neural network is that it is 10,000 times faster than a human when mapping glaciers.