Sat. Dec 2nd, 2023
Artificial Intelligence System Trained to Map Icebergs in Seconds

Scientists have developed an artificial intelligence (AI) system that can accurately map the surface area and outline of icebergs captured on satellite images in just one-hundredth of a second. This breakthrough is a significant improvement on existing automated systems, which struggle to differentiate icebergs from other features in satellite images.
Icebergs have a significant impact on the polar environment, making their monitoring crucial for maritime safety and scientific study. They can be extremely large, posing a risk to passing ships and releasing nutrients and freshwater into the seas.
Led by Dr. Anne Braakmann-Folgmann, researchers at the Centre for Polar Observation and Modelling trained an AI algorithm called U-net to accurately map iceberg outlines using images taken by Sentinel-1 satellites operated by the European Space Agency. Manual interpretation of satellite images is more accurate but time-consuming. The U-net algorithm analyzes the pixels in the image to determine the boundary or outline of the iceberg.
The team compared the U-net algorithm to two other existing algorithms, k-means and Otsu, to identify the largest iceberg in a series of satellite images. Over a six-year period, they tested the algorithms on seven large icebergs, ranging in size from 54km2 to 1052km2. The U-net algorithm outperformed both k-means and Otsu, clearly delineating the outline of the iceberg in test images.
The U-net algorithm showed a 5% lower estimate of iceberg area compared to the k-means and Otsu algorithms, which returned figures that were 150% to 170% too large. The U-net algorithm avoids including sea ice or nearby coastline in its calculations.
This AI technology can provide new services that offer information about the shape and size of giant icebergs. Current mapping services only show the midpoint or central location and length of icebergs. Automated mapping with the U-net algorithm allows for the calculation of iceberg outlines and areas.
According to Professor Andrew Shepherd, Director of the Centre for Polar Observation and Modelling at Northumbria University, this study demonstrates the potential of machine learning to monitor remote parts of the world in real-time. Dr. Braakmann-Folgmann suggests that combining the iceberg mapping technology with measurements of iceberg thickness can help scientists further understand the release of freshwater into the oceans.
This research is published in The Cryosphere journal. Northumbria University is a leader in studying ice-sheet interactions with oceans and has recently received funding to become a Centre for Doctoral Training in Artificial Intelligence, focusing on citizen-centered AI projects.