Data Assimilation Project

Reconstruction based on sparse and noisy observations for Lorenz 63 and Lorenz 96 physical states with Data Assimilation and Neural Networks

Table of contents

• General info
• Technologies
• Setup
• Features
• Acknowledgements
• Contact

General info

This project was made during the Advanced Course on Deep Learning and Geophsyical Dynamics. This course was co-organized by AI Chairs OceaniX (https://cia-oceanix.github.io/) and AI4Child.

Technologies

Project is created with:

• Python : 3.7.7
• Pytorch Lightning : 1.5.2
  PyTorch Lightning is a PyTorch framework for building neural networks. Its documentation is here.

Setup

To run this project, install it locally :

Install PyTorch Lightning with pip install pytorch_lightning

Make sure the utils.py is in the same folder as your different Python Notebook.

Features

Four notebooks are ready to use :

• Baseline_L63.ipynb: 4DVar and CNN implemented in Pytorch Lightning for Lorenz 63.
• Baseline_L96.ipynb: 4DVar and CNN implemented in Pytorch Lightning for Lorenz 96.
• 4DVarNet_L63.ipynb: 4DVarNet implemented in Pytorch Lightning for Lorenz 63.
• 4DVarNet_L96.ipynb: 4DVarNet implemented in Pytorch Lightning for Lorenz 96.
• A gridsearch.py Python script can be run to find optimal parameters for the CNN.

Acknowledgements

• This project was based on this project.
• Many thanks to Ronan Fablet and the lecturers of his Advanced Course on Deep Learning and Geophsyical Dynamics course.

Contact

Created by Simon Driscoll , Charlotte Durand and Oscar Jacquot - feel free to contact us !