Make sure to install all requirements

conda env create -f environment.yml

and activate the conda environment before using any code

conda activate diffscale

The data used in this project is publicly accessible from:

https://cds.climate.copernicus.eu/

https://apps.ecmwf.int/datasets/data/s2s-realtime-instantaneous-accum-ecmf/levtype=sfc/type=cf/

respective licenses apply.

All data and pretrained models can be downloaded from:

https://datacloud.hhi.fraunhofer.de/s/3qfnBewLjGmMGSZ

Place the training dataset dict_s2s.pt (with reforecasts) and the testing dataset dict_s2s_test.pt (forecasts only) into the folder:

datasets/

and the pretrained models lrws_sf.ckpt, lrws_sf.ckpt, and lrws_sf_df.ckpt into the folder:

weights/

of this root-directory.

You can reproduce trainings with the bash script reproduce_train_all.sh

bash reproduce_train_all.sh

and samplings with the bash script reproduce_sample_all.sh

bash reproduce_sample_all.sh