Photo: Sergey Torbik / Pexels

Using AI to study sea-ice thickness

The new ESA-funded project SuperIce, led by the Nansen Center, has high ambitions to improve the detailed information about the sea-ice thickness in the Arctic Ocean.

How thick sea ice in the Arctic is, is not just important for ships passing through locally, but also on a more global scale in a climate context. Thin ice melts faster than thick ice, and with a shrinking sea-ice cover we get less white area in the Arctic to throw sunlight back into space. This is one of the factors that leads to higher surface temperatures on our planet. Thin ice also cracks more easily than thick ice. And when cracks open in the ice cover, leads form and heat from the relatively warm Arctic Ocean below escapes into the atmosphere, representing a huge amount of energy. The heat escaping from the ocean water exposed in leads is more than hundred times as much as heat being released through thin ice, making information on lead size and location an important factor for the energy balance in the Arctic. Quantifying these heat fluxes also helps us to understand changes going on in the Arctic climate, which in turn affects the global climate through ocean and atmosphere teleconnections.

So, you see that sea-ice thickness is important to know as much as possible about. Today, many satellites orbit in space, giving us information on the sea ice, including how thick it is all over the Arctic. But the information from the satellite observations does not have a very high resolution, they are only at approximately 100 km of spatial resolution. And as with many things, the better the resolution, the better the results – in this case heat flux calculations, as well as seasonal forecasts of the sea ice. They can drastically improve with high-resolution information, because processes happening on small scales will be captured much better.

The SuperIce project (led by Julien Brajard at the Nansen Center) is currently working towards an increased resolution of sea-ice thickness information. The researchers in the project make use of the in-house developed sea-ice model neXtSIM, as well as an Artificial Intelligence-based model, which is trained to turn low-resolution sea-ice thickness data into high-resolution data (see fact box). They aim to achieve resolutions of down to 5 km, which is a huge jump from the current 100 km scales.

The SuperIce team is excited to contribute to improved seasonal forecasting and climate impact assessment through their work! The project fits right in with many of other sea-ice research activities at the Nansen Center, for example the development of our sea-ice model neXtSIM, and other projects developing artificial intelligence applications at the Nansen Center.

First SuperIce webinar, open to anyone!

On Wednesday, the 14h of February 2024, the SuperIce team is hosting the first webinar in a series of scientific presentations on sea ice and machine learning. Everyone is welcome to tune in and join!

Read more here or register directly here.

The SuperIce project

SuperIce is funded by the European Space Agency and led by the Nansen Center. It runs until September 2024. The project proposes a multi-step approach. First, a physically based sea-ice model, neXtSIM, is employed to generate high-resolution synthetic sea-ice thickness datasets. These synthetic datasets are then filtered to mimic the resolution of satellite products. An AI-based diffusion model is trained to super-resolve the low-resolution sea-ice thickness data. Finally, the AI-based model is applied to real Earth observation data, and its results are validated against high-resolution satellite data.

What is sea-ice modelling and neXtSIM?

You can find lots of information on our website, for example here.