Deep learning subgrid-scale parametrisations for short-term forecasting of sea-ice dynamics with a Maxwell elasto-brittle rheology
<p>We introduce a proof of concept to parametrise the unresolved subgrid scale of sea-ice dynamics with deep learning techniques. Instead of parametrising single processes, a single neural network is trained to correct all model variables at the same time. This data-driven approach is applied...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2023-07-01
|
| Series: | The Cryosphere |
| Online Access: | https://tc.copernicus.org/articles/17/2965/2023/tc-17-2965-2023.pdf |
| _version_ | 1797776021046427648 |
|---|---|
| author | T. S. Finn C. Durand A. Farchi M. Bocquet Y. Chen A. Carrassi A. Carrassi V. Dansereau |
| author_facet | T. S. Finn C. Durand A. Farchi M. Bocquet Y. Chen A. Carrassi A. Carrassi V. Dansereau |
| author_sort | T. S. Finn |
| collection | DOAJ |
| description | <p>We introduce a proof of concept to parametrise the unresolved subgrid scale of sea-ice dynamics with deep learning techniques.
Instead of parametrising single processes, a single neural network is trained to correct all model variables at the same time.
This data-driven approach is applied to a regional sea-ice model that accounts exclusively for dynamical processes with a Maxwell elasto-brittle rheology.
Driven by an external wind forcing in a <span class="inline-formula">40 km×200 km</span> domain, the model generates examples of sharp transitions between unfractured and fully fractured sea ice.
To correct such examples, we propose a convolutional U-Net architecture which extracts features at multiple scales.
We test this approach in twin experiments: the neural network learns to correct forecasts from low-resolution simulations towards high-resolution simulations for a lead time of about 10 min.
At this lead time, our approach reduces the forecast errors by more than <span class="inline-formula">75 <i>%</i></span>, averaged over all model variables.
As the most important predictors, we identify the dynamics of the model variables.
Furthermore, the neural network extracts localised and directional-dependent features, which point towards the shortcomings of the low-resolution simulations.
Applied to correct the forecasts every 10 min, the neural network is run together with the sea-ice model.
This improves the short-term forecasts up to an hour.
These results consequently show that neural networks can correct model errors from 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.</p> |
| first_indexed | 2024-03-12T22:44:58Z |
| format | Article |
| id | doaj.art-30a6b3d9132c45718bc62300d0899b29 |
| institution | Directory Open Access Journal |
| issn | 1994-0416 1994-0424 |
| language | English |
| last_indexed | 2024-03-12T22:44:58Z |
| publishDate | 2023-07-01 |
| publisher | Copernicus Publications |
| record_format | Article |
| series | The Cryosphere |
| spelling | doaj.art-30a6b3d9132c45718bc62300d0899b292023-07-21T07:57:11ZengCopernicus PublicationsThe Cryosphere1994-04161994-04242023-07-01172965299110.5194/tc-17-2965-2023Deep learning subgrid-scale parametrisations for short-term forecasting of sea-ice dynamics with a Maxwell elasto-brittle rheologyT. S. Finn0C. Durand1A. Farchi2M. Bocquet3Y. Chen4A. Carrassi5A. Carrassi6V. Dansereau7CEREA, École des Ponts and EDF R&D, Île-de-France, FranceCEREA, École des Ponts and EDF R&D, Île-de-France, FranceCEREA, École des Ponts and EDF R&D, Île-de-France, FranceCEREA, École des Ponts and EDF R&D, Île-de-France, FranceDept. of Meteorology and NCEO, University of Reading, Reading, United KingdomDept. of Meteorology and NCEO, University of Reading, Reading, United KingdomDept. of Physics and Astronomy “Augusto Righi”, University of Bologna, Bologna, ItalyLaboratoire 3SR, Grenoble INP, CNRS, Université Grenoble Alpes, Grenoble, France<p>We introduce a proof of concept to parametrise the unresolved subgrid scale of sea-ice dynamics with deep learning techniques. Instead of parametrising single processes, a single neural network is trained to correct all model variables at the same time. This data-driven approach is applied to a regional sea-ice model that accounts exclusively for dynamical processes with a Maxwell elasto-brittle rheology. Driven by an external wind forcing in a <span class="inline-formula">40 km×200 km</span> domain, the model generates examples of sharp transitions between unfractured and fully fractured sea ice. To correct such examples, we propose a convolutional U-Net architecture which extracts features at multiple scales. We test this approach in twin experiments: the neural network learns to correct forecasts from low-resolution simulations towards high-resolution simulations for a lead time of about 10 min. At this lead time, our approach reduces the forecast errors by more than <span class="inline-formula">75 <i>%</i></span>, averaged over all model variables. As the most important predictors, we identify the dynamics of the model variables. Furthermore, the neural network extracts localised and directional-dependent features, which point towards the shortcomings of the low-resolution simulations. Applied to correct the forecasts every 10 min, the neural network is run together with the sea-ice model. This improves the short-term forecasts up to an hour. These results consequently show that neural networks can correct model errors from 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.</p>https://tc.copernicus.org/articles/17/2965/2023/tc-17-2965-2023.pdf |
| spellingShingle | T. S. Finn C. Durand A. Farchi M. Bocquet Y. Chen A. Carrassi A. Carrassi V. Dansereau Deep learning subgrid-scale parametrisations for short-term forecasting of sea-ice dynamics with a Maxwell elasto-brittle rheology The Cryosphere |
| title | Deep learning subgrid-scale parametrisations for short-term forecasting of sea-ice dynamics with a Maxwell elasto-brittle rheology |
| title_full | Deep learning subgrid-scale parametrisations for short-term forecasting of sea-ice dynamics with a Maxwell elasto-brittle rheology |
| title_fullStr | Deep learning subgrid-scale parametrisations for short-term forecasting of sea-ice dynamics with a Maxwell elasto-brittle rheology |
| title_full_unstemmed | Deep learning subgrid-scale parametrisations for short-term forecasting of sea-ice dynamics with a Maxwell elasto-brittle rheology |
| title_short | Deep learning subgrid-scale parametrisations for short-term forecasting of sea-ice dynamics with a Maxwell elasto-brittle rheology |
| title_sort | deep learning subgrid scale parametrisations for short term forecasting of sea ice dynamics with a maxwell elasto brittle rheology |
| url | https://tc.copernicus.org/articles/17/2965/2023/tc-17-2965-2023.pdf |
| work_keys_str_mv | AT tsfinn deeplearningsubgridscaleparametrisationsforshorttermforecastingofseaicedynamicswithamaxwellelastobrittlerheology AT cdurand deeplearningsubgridscaleparametrisationsforshorttermforecastingofseaicedynamicswithamaxwellelastobrittlerheology AT afarchi deeplearningsubgridscaleparametrisationsforshorttermforecastingofseaicedynamicswithamaxwellelastobrittlerheology AT mbocquet deeplearningsubgridscaleparametrisationsforshorttermforecastingofseaicedynamicswithamaxwellelastobrittlerheology AT ychen deeplearningsubgridscaleparametrisationsforshorttermforecastingofseaicedynamicswithamaxwellelastobrittlerheology AT acarrassi deeplearningsubgridscaleparametrisationsforshorttermforecastingofseaicedynamicswithamaxwellelastobrittlerheology AT acarrassi deeplearningsubgridscaleparametrisationsforshorttermforecastingofseaicedynamicswithamaxwellelastobrittlerheology AT vdansereau deeplearningsubgridscaleparametrisationsforshorttermforecastingofseaicedynamicswithamaxwellelastobrittlerheology |