CREDIT arXiv Paper Code Repository

About

This repository hosts the main results of the Community Research Earth Digital Intelligence Twin (CREDIT).

• Preprint: https://arxiv.org/abs/2411.07814

Python Environment

The Python environment is the same as in the main CREDIT repository.

Navigation

• data_preprocessing: This folder contains scripts and Jupyter notebooks for the computation of z-score and residual normalization. Follow these steps:

  • Copy data_config_6h_template.yml to data_config_6h.yml (similarly for 1h) and modify it based on your file directories.
  • Run through the Jupyter notebooks from STEP00 to STEP03.
  • Access the scripts folder for large-scale verification runs.

• verification: This folder contains verification steps for CREDIT model runs. Follow these steps:

  • Copy verif_config_6h_template.yml to verif_config_6h.yml (similarly for 1h) and modify it based on your file directories.
  • Run through the Jupyter notebooks from STEP00 to STEP05.
  • Access the scripts folder for large-scale verification runs.

• visualization: This folder hosts the results of the CREDIT arXiv paper. It is runnable once all verification outputs are prepared.

• libs: A collection of functions/scripts used by this repository.

  • preprocess_utils.py: Functions for z-score and residual normalization calculations.
  • verif_utils.py: Functions for forecast post-processing and verification.
  • score_utils.py: Functions for bootstrapping and computing the zonal energy spectrum.
  • graph_utils.py: Functions for data visualization.

Production File Storage

If you have access to NCAR/UCAR HPCs, you can access production output files at the following locations:

/glade/campaign/cisl/aiml/ksha/CREDIT_arXiv/GATHER/fuxi_6h_20241029/
/glade/campaign/cisl/aiml/ksha/CREDIT_arXiv/GATHER/wxformer_6h_20241029/
/glade/campaign/cisl/aiml/ksha/CREDIT_arXiv/GATHER/wxformer_1h_20241029/
/glade/campaign/cisl/aiml/ksha/CREDIT_arXiv/GATHER/IFS/

Verification results are available at:

/glade/campaign/cisl/aiml/ksha/CREDIT_arXiv/VERIF/

The pre-processed ERA5 and solar irradiance data (i.e., training and verification target) are available at:

/glade/derecho/scratch/ksha/CREDIT_data/ERA5_mlevel_arXiv/SixHourly_y*
/glade/campaign/cisl/aiml/wchapman/MLWPS/STAGING/y_TOTAL*

Model weights

Model weights are not stored in this repository. They can be found at the following locations:

/glade/campaign/cisl/aiml/ksha/CREDIT_runs/arXiv_project/fuxi_6h/model_checkpoint.pt
/glade/campaign/cisl/aiml/ksha/CREDIT_runs/arXiv_project/wxformer_6h/model_checkpoint.pt
/glade/campaign/cisl/aiml/ksha/CREDIT_runs/arXiv_project/wxformer_1h/model_checkpoint.pt

We will upload the model weights online soon. Stay tuned.

Contact

• John Schreck schreck@ucar.edu
• Yingkai Sha ksha@ucar.edu
• William Chapman wchapman@ucar.edu
• David John Gagne II dgagne@ucar.edu