Increased safety for polar marine activities
Far from civilization with few icebreakers around to help, without being at risk of being enclosed by thick sea ice or unexpectedly meeting an iceberg – how can vessels safely navigate? What types of sea ice are out there in the Arctic Ocean? Where are they located, where are they drifting to, and how fast are they moving? How can we map sea ice in all its forms to help ships decide the best and safest course? How we do this efficiently for large and remote areas? There are many questions to address!
The traditional way of mapping sea ice relies on visual analysis and interpretation of radar satellite images by sea ice analysts. The results of this analysis are then provided to the ship captains in the form of sea ice type or sea ice concentration maps. However, manual production of these maps is a time consuming and arduous task.
Researchers at UiT and the partners in CIRFA are developing new ways to identify sea ice conditions. We use data from radar instruments carried by satellites such as the European SENTINEL-1 and Canadian RADARSAT-2. The satellites cross over the Arctic several times a day and scan the ocean surface in beams that can be up to a few hundred kilometres wide. In contrast to optical sensors, the radar can see the ocean surface through clouds and in darkness.
New ways to make sea ice maps
Instead of visually interpreting each satellite image, researchers at CIRFA and MET have developed an algorithm that automatically interprets many satellite images at a time. The algorithm built by CIRFA researchers Johannes Lohse, Wenkai Guo and Anthony Doulgeris can recognize different types of sea ice. Its ability to do this is rooted in the physical properties of the backscatter; the strength of the radar signal that is reflected from the Earth surface and recorded by the satellite.
To understand how backscatter works, imagine how the light of a flashlight reflects differently depending on what you shine the light on. Like smooth surfaces appearing bright and corners or edges making shadows, the reflected radar signal depends on whether the water or ice is rough and deformed, or if it’s smooth, almost like a mirror. The change in backscatter, if properly analysed, tells us about different sea ice types and open water areas.
To automatically process the satellite data and make ice type maps, the algorithm was optimized over months. During this optimisation, the researchers trained the algorithm by showing it a lot of data corresponding to different sea ice types recovered from manually analysed satellite images. This step is essential to gain confidence that the algorithm can distinguish and map smooth or deformed sea ice, wet surfaces, sea ice ridges, and open water areas.
The algorithm can make a map based on satellite imagery for a specific location, for example in the northern Atlantic Ocean, in as little as 15 minutes. For larger areas, or areas that are not as frequently overflown by satellites, it will take more time to acquire data and make maps.
Testing the algorithm in real conditions
PhD candidates Anna Telegina, Jozef Rusin and Laust Færch, and the Norwegian Meteorological Institute (MET) engineer Alistair Everett joined the UAK research cruise on KV Svalbard in June 2021 when CIRFA’s automatic sea ice classification was tested. Before and during the cruise, the algorithm was set up at MET, where the latest satellite images were analysed and the ice type maps were produced before they were sent to the ship. The team on the ship compared the sea ice maps generated by the algorithm with reality around the vessel to check and confirm the information from the algorithm.
In addition, Tom Rune Lauknes from NORCE used drones to map sea ice with ultra-high resolution optical images. This is important to correctly characterize the ice, to learn how well the algorithm works, and to know how to improve it.