As a world first, my colleague Simon Driscoll from The University of Reading has investigated to possibility to build a machine learning emulator of sea-ice melt ponds. Their parametrisation in sea-ice models are critical for a forecasting due to the change in albedo of the melt sea-ice over the ice shelf. He have first carried out a sensitivity study based on Sobol sensitivity analysis to determine the key parameters of such parametrisation. He then successfully learns and test such neural network-based paremetrisation from the physical Icepack model. This model is faster and restricted to the most influencial variables.

The paper, a contribution to the Schmidt Sciences SASIP project, is entitled Parameter sensitivity analysis of a sea ice melt pond parametrisation and its emulation using neural networks, and is published in the Journal of Computational Science.