4DVarNet

Prerequisite

• git
• conda

Install

Install project dependencies

git clone https://github.com/CIA-Oceanix/4dvarnet-starter.git
cd 4dvarnet-starter
conda install -c conda-forge mamba
conda create -n 4dvarnet-starter
conda activate 4dvarnet-starter
mamba env update -f environment.yaml

Download example data

From the directory

wget https://s3.eu-central-1.wasabisys.com/sla-data-registry/6d/206c6be2dfe0edf1a53c29029ed239 -O data/natl_gf_w_5nadirs.nc

Run

The model uses hydra see [#useful-links]

python main.py xp=base 

Saved weights:

Gulfstream training

A bigger model has been trained using the command

python main.py xp=base +params=bigger_model 

You can find pre-trained weights here

The test metrics of this model are ([see here for the details])(https://github.com/ocean-data-challenges/2020a_SSH_mapping_NATL60):

	osse_metrics
RMSE (m)	0.0211406
λx	0.716
λt	4.681
μ	0.96362
σ	0.00544

Animation:

Useful links:

• Hydra documentation
• Pytorch lightning documentation
• 4DVarNet papers:
  • Fablet, R.; Amar, M. M.; Febvre, Q.; Beauchamp, M.; Chapron, B. END-TO-END PHYSICS-INFORMED REPRESENTATION LEARNING FOR SA℡LITE OCEAN REMOTE SENSING DATA: APPLICATIONS TO SA℡LITE ALTIMETRY AND SEA SURFACE CURRENTS. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences 2021, V-3–2021, 295–302. https://doi.org/10.5194/isprs-annals-v-3-2021-295-2021.
  • Fablet, R.; Chapron, B.; Drumetz, L.; Mmin, E.; Pannekoucke, O.; Rousseau, F. Learning Variational Data Assimilation Models and Solvers. Journal of Advances in Modeling Earth Systems n/a (n/a), e2021MS002572. https://doi.org/10.1029/2021MS002572.
  • Fablet, R.; Beauchamp, M.; Drumetz, L.; Rousseau, F. Joint Interpolation and Representation Learning for Irregularly Sampled Satellite-Derived Geophysical Fields. Frontiers in Applied Mathematics and Statistics 2021, 7. https://doi.org/10.3389/fams.2021.655224.