The shape and distribution of icebergs and their trajectory through the oceans from source, to where they break up and melt, can be diagnostic of ice-sheet dynamics, and ocean and atmospheric conditions. Existing methods to detect icebergs in satellite imagery are optimised for open ocean and work less well for icebergs within the sea ice pack. This project seeks to use machine learning techniques to identify, track and follow the disintegration of icebergs in the Amundsen Sea near Antarctica, using radar satellite images. The study may reveal temporal trends in icebergs populations and their trajectories through the coastal seas.

Explaining the science

The Antarctic ice sheet loses ice in roughly equal parts through direct melting and the formation of icebergs. Icebergs range from the massive tabular icebergs, that cover many hundreds or square kilometres, to small fragments the size of a car. In the Amundsen Sea - an arm of the Southern Ocean off Marie Byrd Land in western Antarctica - there is an expectation the current retreat of the ice sheet, which is now making a significant contribution to sea-level rise, is causing a transition in the dominant processes of iceberg formation, from the larger tabular icebergs to small narrow icebergs which topple over. This project is using machine learning techniques to identify, track and follow the disintegration of icebergs using all-weather radar satellite images from ESA satellites.

Project aims

There are established methods for identification of icebergs in radar satellite imagery mainly based on CFAR (Constant False Alarm Rate) detection methods. However there are a number of limitations to these approaches, e.g. they work well in open ocean, but are less effective for icebergs within the sea ice pack or tabular icebergs in close proximity. New approaches will aim to identify icebergs in both the open ocean and in sea ice. Further improvements will consider information about iceberg size and shape, and tracking icebergs trajectories.


Ben Evans

Researcher in Machine Learning, British Antarctic Survey

Researchers and collaborators