NSF NCAR MILES Community Runnable Earth Digital Intelligence Twin (CREDIT) - Experimental & Research version
This repository is my personal version of CREDIT, diverged from the main CREDIT repository under the NSF NCAR GitHub Organization. It aims to provide a fast and light-weight computational environment for AI weather prediciton models. New features experimented within this repository may be forwarded to the main CREDIT. This repository will also updated with my own AI-weather-forecast-related research works.
This repository is primarily hosted on NSF NCAR HPCs (casper.ucar.edu and derecho.hpc.ucar.edu). Clone this repository, make sure you have Pytorch with GPU access, and go through the following steps:
mamba env create -f environment.yml
conda activate credit_mini
pip install .
- Switch from the current ERA5 training set to the model resolution / level ERA5 from Google.
- Add more forcing and static inputs.
- Base class and unit testing.
Yingkai Sha [email protected]