Preprints
https://doi.org/10.5194/egusphere-2023-3078
https://doi.org/10.5194/egusphere-2023-3078
20 Feb 2024
 | 20 Feb 2024
Status: this preprint is open for discussion.

HyPhAI v1.0: Hybrid Physics-AI architecture for cloud cover nowcasting

Rachid El Montassir, Olivier Pannekoucke, and Corentin Lapeyre

Abstract. This work proposes a hybrid approach that combines Physics and Artificial Intelligence (AI) for cloud cover nowcasting. It addresses the limitations of traditional deep learning methods in producing realistic and physically consistent results that can generalize to unseen data. The proposed approach enforces a physical behaviour. In the first model, denoted HyPhAI-1, a multi-level advection dynamics is considered as a hard constraint for a trained U-Net model. Our experiments show that the hybrid formulation outperforms not only traditional deep learning methods, but also the EUMETSAT Extrapolated Imagery model (EXIM) in terms of both qualitative and quantitative results. In particular, we illustrate that the hybrid model preserves more details and achieves higher scores based on similarity metrics in comparison to the U-Net. Remarkably, these improvements are achieved while using only one-third of the data required by the other models. Another model, denoted HyPhAI-2, adds a source term to the advection equation, it impaired the visual rendering but displayed the best performance in terms of Accuracy. These results suggest that the proposed hybrid Physics-AI architecture provides a promising solution to overcome the limitations of classical AI methods, and contributes to open up new possibilities for combining physical knowledge with deep learning models.

Rachid El Montassir, Olivier Pannekoucke, and Corentin Lapeyre

Status: open (until 09 May 2024)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
Rachid El Montassir, Olivier Pannekoucke, and Corentin Lapeyre

Data sets

A sample of the training data used in the paper EUMETSAT https://zenodo.org/doi/10.5281/zenodo.10642093

Model code and software

Github repository containing the source code of the implemented hybrid models. Rachid El Montassir, Olivier Pannekoucke, and Corentin Lapeyre https://doi.org/10.5281/zenodo.10676679

Interactive computing environment

Github repository containing forecasting examples using pre-trained models. Rachid El Montassir, Olivier Pannekoucke, and Corentin Lapeyre https://doi.org/10.5281/zenodo.10676679

Video supplement

HyPhAI-1 2-hour forecast on 01/01/2021 at 12:00 p.m. Rachid El Montassir, Olivier Pannekoucke, and Corentin Lapeyre https://zenodo.org/doi/10.5281/zenodo.10375283

Rachid El Montassir, Olivier Pannekoucke, and Corentin Lapeyre

Viewed

Total article views: 272 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
180 83 9 272 9 7
  • HTML: 180
  • PDF: 83
  • XML: 9
  • Total: 272
  • BibTeX: 9
  • EndNote: 7
Views and downloads (calculated since 20 Feb 2024)
Cumulative views and downloads (calculated since 20 Feb 2024)

Viewed (geographical distribution)

Total article views: 270 (including HTML, PDF, and XML) Thereof 270 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 15 Apr 2024
Download
Short summary
This study introduces a novel approach, combining Physics and Artificial Intelligence (AI) for improved cloud cover forecasting. This approach outperforms traditional Deep Learning (DL) methods in producing realistic and physically consistent results while requiring less training data. This architecture provides a promising solution to overcome the limitations of classical AI methods, and contributes to open up new possibilities for combining physical knowledge with deep learning models.