granite-wxc

This repository contains code and examples to apply the Prithvi WxC foundation model to downscaling tasks. In particular, the repository contains both code and instructions for generic fine-tuning tasks as well as fine-tuned models for MERRA2 2m temperature and the ECCC v10 and u10 wind components as reference.

Figure 1: 6x downscaling of MERRA-2 2m temperature

Figure 2: 8x downscaling of ECCC's u10 wind component

Getting started

1. Create a virtual environment
2. Clone this repository as well as that of the foundation model. Install both in the virtual environment.:

   git clone https://github.com/NASA-IMPACT/Prithvi-WxC
   git clone https://github.com/IBM/granite-wxc.git
   cd Prithvi-WxC
   pip install '.[examples]'
   cd ../granite-wxc
   pip install '.[examples]'
   

3. Run the notebooks in the examples directory
   • MERRA2 example: This notebook will download model weights as well as sample data for basic illustration from Hugging Face.
   • ECCC example: This directory contains notebooks for both fine-tuning and inference. It also includes instructions for downloading and setting up the data, model and the required files from Hugging Face.

Fine-tuned model

The fine-tuned model for MERRA-2 2m temperature data is available via Hugging Face. For an application to EURO-CORDEX data please refer to the paper.

The fine-tuned model for ECCC v10 and u10 wind component data is available via Hugging Face.