02 Jan 2023
02 Jan 2023
Status: this preprint is open for discussion.

Deep learning of subgrid-scale parametrisations for short-term forecasting of sea-ice dynamics with a Maxwell-Elasto-Brittle rheology

Tobias Sebastian Finn1, Charlotte Durand1, Alban Farchi1, Marc Bocquet1, Yumeng Chen2, Alberto Carrassi2,3, and Véronique Dansereau4 Tobias Sebastian Finn et al.
  • 1CEREA, École des Ponts and EDF R&D, Île-de-France, France
  • 2Dept. of Meteorology and NCEO, University of Reading, Reading, United Kingdom
  • 3Dept of Physics and Astronomy "Augusto Righi", University of Bologna, Italy
  • 4Université Grenoble Alpes, CNRS, Grenoble INP, Laboratoire 3SR, Grenoble, France

Abstract. We introduce a scalable approach to parametrise the unresolved subgrid-scale of sea-ice dynamics with deep learning techniques. We apply this data-driven approach to a regional sea-ice model that accounts exclusively for dynamical processes with a Maxwell-Elasto-Brittle rheology. Our channel-like model setup is driven by a wave-like wind forcing, which generates examples of sharp transitions between unfractured and fully-fractured sea ice. Using a convolutional U-Net architecture, the parametrising neural network extracts multiscale and anisotropic features and, thus, includes important inductive biases needed for sea-ice dynamics. The neural network is trained to correct all nine model variables at the same time. With the initial and forecast state as input into the neural network, we cast the subgrid-scale parametrisation as model error correction, needed to correct unresolved model dynamics. We test the here-proposed approach in twin experiments, where forecasts of a low-resolution forecast model are corrected towards high-resolution truth states for a forecast lead time of about 10 min. At this lead time, our approach reduces the forecast errors by more than 75 %, averaged over all model variables. The neural network learns hereby a physically-explainable input-to-output relation. Furthermore, cycling the subgrid-scale parametrisation together with the geophysical model improves the short-term forecast up to one hour. We consequently show that neural networks can parametrise the subgrid-scale for sea-ice dynamics. We therefore see this study as an important first step towards hybrid modelling to forecast sea-ice dynamics on an hourly to daily timescale.

Tobias Sebastian Finn et al.

Status: open (until 27 Feb 2023)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse

Tobias Sebastian Finn et al.

Model code and software

Code for experiments and neural networks (Git repository that includes all code except the regional sea-ice model) Tobias Sebastian Finn

Tobias Sebastian Finn et al.


Total article views: 220 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
127 88 5 220 4 4
  • HTML: 127
  • PDF: 88
  • XML: 5
  • Total: 220
  • BibTeX: 4
  • EndNote: 4
Views and downloads (calculated since 02 Jan 2023)
Cumulative views and downloads (calculated since 02 Jan 2023)

Viewed (geographical distribution)

Total article views: 229 (including HTML, PDF, and XML) Thereof 229 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
Latest update: 03 Feb 2023
Short summary
We combine machine learning with a simplified regional sea-ice model to correct errors in sea-ice dynamics of low-resolution forecasts towards a high-resolution truth. The combined model improves the forecast by up to 75 % and surpasses thereby the performance of persistence. As the error correction can also represent physically-explainable relations, this study highlights the potential of combined modelling for short-term sea-ice forecasting.