Machine learning methods for sea ice analysis, prediction and projection

Neven Fučkar (neven.fuckar[at]

The scientific questions that will be studied:  

–    Detection and classification (supervised and unsupervised) of anomalies in observed, reconstructed and modelled sea ice variables.
–    Comparison of machine learning, statistical and dynamical predictions.
–    Attribution of sea ice trends and extreme events to anthropogenic forcing factors.  
–    How does sea ice variability and predictability evolve with global climate change?

The processes that will be investigated:

–    Sea-ice-atmosphere-ocean interaction
–    Influence of the atmosphere and oceans on formation of sea ice modes of variability and extreme events
–    Storage and release of climate signals in sea ice variables with climate memory effects (thickness, volume and enthalpy)

The type of analyses that will be conducted:

–    Analysis of CMIP6 simulations, and related large ensembles, observations and reanalysis products.
–    Multi-method detection and attribution of trends and extreme events.
–    Tools: K-means clustering, SOM analysis, mixture models, kernel methods, …

Examples of active projects
•    How the structure and occurrence of Arctic SIT modes of variability evolves in CMIP6 projections
•    Attribution of Arctic SIE extremes in today’s climate and at the end of 21st century.


Reichstein, M., et al. 2019: Deep learning and process understanding for data-driven Earth system science. Nature, 566, 195–204.

Kim, Y. J., Kim, H.-C., Han, D., Lee, S., and Im, J.: Prediction of monthly Arctic sea ice concentration using satellite and reanalysis data based on convolutional neural networks, The Cryosphere Discuss.,, in review, 2019.

Fučkar, N.S., Guemas, V., Johnson, N.C. et al. 2019: Dynamical prediction of Arctic sea ice modes of variability, Clim Dyn 52: 3157.