climate-super-resolution

This repo contains training and evaluation scripts of Neural Network for Climate Modelling Super Resolution

Local Env Installation

1. Run:

conda create -n climsr python=3.8
conda activate climsr
conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c nvidia
conda env update -f environment.yml

2. Done

Data pre-processing

1. Download World Clim and CRU-TS datasets
2. Ensure folder structure as in arguments in preprocessing.py
3. Run pre-processing script preprocessing.py

How to train

Run:

python train.py <args>

See train.py, datamodules.py and pl_sr_module.py for full list of arguments.

How to evaluate on the hidden test set

Testing should happen automatically when the test dataset was defined. See test.py for more details.

Inference

To run inference on CRU-TS, please see inference.py. The results can be visualized using provided notebooks.

Less interactive version of the notebook can be found in the inspect_results.py.