cnn-downscaling

Downscaling global climate models with convolutional neural networks

Setup

• Download processed* historical data from Google Drive (click "Download All")
• Move the downloaded data to a new "./data/" subfolder in this repository in order to use scripts with default configurations. The final path should be "/path/to/repo/data/processed/..."
• Install the corresponding conda environment via environment.yml via conda env create --name {your env name here} --file=environment.yml

* Raw data can be downloaded by following the instructions on the WeatherBench repo. We downloaded 2m_temperature and total_precipitation at 1.4 and 5.625 degrees of resolution

Training

All experiments ran for this project can be replicated with src/training/experiments.sh