This is the code for this paper https://doi.org/10.1029/2022MS003132
This project can be built and trained on Ubuntu 18.04.3 LTS, with python3.7 and CUDA 10.0/cudnn 7.6.5.
conda create -n enso python=3.7
source activate enso
pip install [some torch-related libs](https://drive.google.com/drive/folders/1hHQC0Ku1Vm4pLd2F3wVb2f5wnxx9ZyH6?usp=sharing)
pip install netCDF4==1.5.3
pip install progress==1.5
pip install loguru==0.3.2
pip install cmaps
pip install pyproj
pip install h5py
conda install -c conda-forge cartopy
Met Office Hadley Centre observations datasets (HadISST) is used for this model. Download it and put it in ./file/.
The archieved dataset is also in 
Firstly, use the following commands to parse and parpare training data.
python -m data.prepare_data
The output training data files are also in ./file/
Then, train the model:
python -m train_multi_gpus
Firstly, download the latest HadISST from the above wetsites and replace the new data for data preprocessing.
Secondly, fine-tune the trained model:
python workflow.py
Finally, make forecasts for the future 18 months:
python forecast.py
The forecast results will be recorded in ./result-{year}-{month}.csv