Near-Term-Climate-Prediction-BDL

Associated Papers

• Feature Importance in a Deep Learning Climate Emulator [ ArXiV ] [ ICLR 2021 Workshop on Modeling Oceans and Climate Change ]
• A Bayesian Deep Learning Approach to Near-Term Climate Prediction [ ArXiV ] [ Journal of Advances in Modeling Earth Systems ]

Xihaier Luo, Balasubramanya T Nadiga, Yihui Ren, Ji Hwan Park, Wei Xu, Shinjae Yoo

Dependencies

• python 3
• PyTorch
• matplotlib

Installation

• Install PyTorch and other dependencies

• Clone this repo:

git clone https://github.com/Xihaier/Near-Term-Climate-Prediction-BDL

Dataset

The dataset used have been uploaded to Google Drive and can be downloaded with corresponding links.

Link: https://zenodo.org/record/6822275#.YzMpJOyZMeY

Model

We ran extensive models and shared some of the best ones here, which can be divided into two categories: deterministic models and Bayesian models. For example, if a deterministic model, such as DenseNet, is chosen, one should run

cd src
python main_DL.py

If you are interested in ConvLSTM and its variants, we have included our implementation in the appendix folder.

cd appendix
python main.py

Citation

If you find this repo useful for your research, please consider to cite:

@article{luo2022bayesian,
  title={A Bayesian Deep Learning Approach to Near-Term Climate Prediction},
  author={Luo, Xihaier and Nadiga, Balasubramanya T and Ren, Yihui and Park, Ji Hwan and Xu, Wei and Yoo, Shinjae},
  journal={arXiv preprint arXiv:2202.11244},
  year={2022}
}

Questions

For any questions or comments regarding this paper, please contact Xihaier Luo via xluo@bnl.gov.