Artificial intelligence can map the outline and area of ​​giant icebergs

Artificial intelligence can map the outline and area of ​​giant icebergs

Scientists have trained an artificial intelligence or artificial intelligence system to accurately map the surface area and outline of giant icebergs captured on satellite images in a hundredth of a second.

It’s a major advance over current automated systems that struggle to distinguish between icebergs and other features in an image.

Human or manual interpretation of satellite images is the other option, and although the results are accurate, it can take several minutes to outline a single iceberg. If this has to be repeated several times, the process quickly becomes time-consuming and tedious.

Icebergs have a major impact on the polar environment and monitoring them is crucial for both marine safety and scientific study. They can be very large – in some cases the size of small countries – and can pose a danger to passing ships. As they melt, icebergs release nutrients and fresh water into the seas, and this can have an impact on marine ecosystems.

“Icebergs exist in hard-to-reach parts of the world, and satellites are just a great way to explore them,” said Dr. Anne Brackman-Volgemann, who led the study while doing her doctoral research at the University of Leeds Polar Observation and Monitoring Centre. As a tool to monitor its whereabouts, it can help scientists understand the process of it melting and eventually beginning to break apart.

“Using the new AI system overcomes some of the problems with current automated methods, which can have difficulty distinguishing between icebergs and other ice floating on the sea or even a nearby coastline in the same image.”

Neural network

Dr. Brackmann-Volgemann and her colleagues used an algorithm called U-net — a type of neural network — to “train” a computer to accurately map the outline of icebergs from images taken by the Sentinel-1 satellites operated by the European Space Agency.

As part of the study, the effectiveness of the U-net algorithm was compared with two other state-of-the-art algorithms used to map icebergs. They are known as k-means and Otsu. The algorithms are programmed to identify the largest iceberg in a series of satellite images.

Image 1 shows the U-net algorithm correctly identifying the iceberg, which is highlighted in red. In comparison, the k-means algorithm incorrectly identified a group of small icebergs and ice fragments, shown in blue, as a single large iceberg. This is shown in picture 2.

Aerial photo of an iceberg shaded in pink.  The surrounding ocean is black and white floating icebergs surrounding the iconic iceberg.

Images 1. Image credit: Dr. Anne Brackmann-Volgemann and the European Space Agency

A block of small ice shards is highlighted in purple.

Images 2. Image credit: Dr. Anne Brackmann-Volgemann and the European Space Agency

Image 3 shows that the U grid correctly identifies the iceberg, but this time it is surrounded by sea ice. The iceberg is highlighted in red, and the sea ice is seen as a gray structure. However, the k-means algorithm identified the iceberg and sea ice as one iceberg. It is not possible to differentiate between the two, even though they are distinct bodies, as sea ice is flat ice in the sea and an iceberg that stands meters above it. Shown in picture 4

An iceberg shaded pink and surrounded by sea ice.

Images 3. Image credit: Dr. Anne Brackmann-Volgemann and the European Space Agency

The iceberg and the sea ice surrounding it as a single iceberg are highlighted in purple

Images 4. Image credit: Dr. Anne Brackmann-Volgemann and the European Space Agency.

Picture 4. Image credit: Dr. Anne Brackmann-Volgemann and the European Space Agency.

Dr Brackman-Volgman, who now works at the Norwegian Arctic University in Tromsø, said the technology could lead to new services that provide information about the shape and size of giant icebergs. Current mapping services only show the midpoint or central location and length of icebergs. Interpretation with this new approach means that its plan and area can be calculated.

“The ability to automatically map the extent of the iceberg while enhancing speed and accuracy paves the way for an operational service that provides iceberg outlines on a regular, automated basis,” she added.

“Combining it with measurements of iceberg thickness also enables scientists to monitor where giant icebergs are releasing huge amounts of fresh water into the oceans. There are services that provide data on the location of icebergs, but not their outline or area.

Accuracy of the mapping system

The system was tested on satellite images of seven icebergs, all of which were the size of the city of Bern – 54 square kilometers; And Hong Kong – 1052 km2. Up to 46 images were used for each of these icebergs, covering all seasons from 2014 to 2020.

Through a series of tests, U-net outperformed the other two algorithms and was more effective at identifying the outlines of an iceberg in images captured when environmental conditions were difficult, such as an image that captures a lot of ice structures.

In machine learning, an F1 score is an assessment of how well an algorithm is performing and ranges from 0 to 1, with values ​​closer to one being shown with greater accuracy. U-net achieved an F1 score of 0.84. The other two algorithms scored 0.62.

Andrew Shepherd, a professor at Northumbria University and one of the study’s co-authors, said: “This study shows that machine learning will enable scientists to monitor remote and inaccessible parts of the world in near real-time. With machine learning, the algorithm will become more accurate as it learns One of the errors in the way you interpret a satellite image.

More information

The paper – “Mapping the extent of giant Antarctic icebergs using deep learning” was published in the journal The Cryosphere.

Please use the email address of David Lewis at the University of Leeds Press Office

Top image: Adobe photostock – Iceberg in Disko Bay, Greenland.

You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *