Deep learning for eastimating dissolved oxygen in global ocean
Details for Oxyformer will be publicly available upon publication!
View Demo Β· Documentation Β· Report Bug Β· Request Feature
π Table of Contents
- About the Project
- Getting Started
- Data preparation
- Develop the Oxyformer
- Data and Results
- Contributing
- License
- Contact
- Acknowledgements
π About the Project
π₯ Update
- [2023/07/11] We have released the annual Dissolved Oxygen Products derived by Oxyformer, details are shown in here.
π§° Getting Started
βΌοΈ Prerequisites
- Python 3.8
- Pytorch 1.10.1
- CUDA 11.3 or higher
βοΈ Installation
First, install dependencies
# clone project
git clone https://github.com/Vipermdl/Oxyformer
# install project
cd Oxyformer
pip install -r requirements.txt
π§ Data preparation
- Dissolved Oxygen measurements
- Driven factors
π¨ Dissolved Oxygen measurements
Name | Date accessed |
---|---|
World Ocean Database | 02-2023 |
CLIVAR and Carbon Hydrographic Database | 02-2023 |
Pangaea Database | 02-2023 |
Global Ocean Data Analysis | 02-2023 |
- After download the measurements of oxygen data, run the following command to data compilation and quality control
# interpolate depth mapping using pchip function and data integrated.
python scripts/interpolate.py
# quality control for observational oxygen data
python scripts/quality_control.py
π Driven factors
- More details will be updated soon...
- After download the driven factors, run the following command to obtain the dataset dicatating Oxyformer's input and output data and train/val/test splits.
python scripts/label_utils.py --merge --splits
π Develop the Oxyformer
π· Model architecture
π§ͺ Training Oxyformer
To train Oxyformer, run the following command
python main.py experiments/Oxyformer.toml
π Run inference
To save predictions between July 2002 and December 2020 as NetCDFs for Oxyformer run
python inference.py experiments/inference.toml -o outputdir
π© Post process
The post-processing algorithm is then applied
python post-processing/mask_result_by_bathymetric.py
π Data and Results
- See DATA.md for instructions on how to download the data of our Oxyformer.
- Please read the DRAW.md to generate the paper figures.
π Contributing
Contributions are always welcome!
β οΈ License
Distributed under the no License. See LICENSE.txt for more information.
π€ Contact
Dongliang Ma - @dongliangma1 - mdl.viper@gmail.com
Project Link: https://github.com/Vipermdl/Oxyformer
π Acknowledgements
Use this section to mention useful resources and libraries that you have used in your projects.