A pioneering method has been developed at the Nansen Center that uses machine learning to improve both the accuracy of sea-ice models. The previously demanding manual adjustment processes of model parameters are automated and thereby sped up, and it is possible to produce maps of how the sea ice deforms that are so detailed that they resemble what we see in satellite images. The method can also be applied to other models and marks a breakthrough for climate research and forecasting of sea-ice conditions.
Climate change greatly affects the extent and movement of sea ice in the Arctic, which in turn has a major impact on local ecosystems, ocean currents, and global weather patterns. The sea ice is therefore considered a key component of the climate system.
Advanced modelling is required in order to calculate ice conditions. One of the biggest challenges in creating such models is to accurately describe the rheology of the sea ice, i.e. how the ice responds to forces such as wind, ocean currents, and waves.
A specialised rheology has already been developed at the Nansen Center, known as brittle Bingham-Maxwell (BBM) rheology, which is integrated into our sea-ice model neXtSIM. Using machine learning, we have now trained the model to simulate data with very high spatial accuracy, which can then be compared with real measurements from orbiting satellites. Through this process, the machine learning model calculates the optimal parameter values for the sea-ice model, quickly and accurately.
The method drastically improves the accuarcy of the models. This automated process not only saves time but also increases the reliability of the model by eliminating sources of error inherent in the manual calculation methods.
The new method is not limited to use in specific sea-ice models such as neXtSIM but can also be adapted to other sea-ice models. This expands the possibility for more accurate analyses of climate change, and impacts on and from sea ice, and could provide better warnings of sea ice conditions in the future.
The method represents a major advance in climate research, with the potential for a better understanding of polar ecosystems, more accurate sea ice forecasts, and a better decision-making basis for stakeholders working in the High North or coastal communities that are directly affected by sea-ice dynamics.
The use of machine learning in sea-ice modelling shows how technological advances can address previously unsolved problems and contribute to a deeper understanding of complex climate processes.
Key researchers: Anton Korosov, Michael Ying, Einar Ólason
