There are three main parts to getting the cGAN up and running for regional post-processing of global Numerical Weather Prediction (NWP) forecasts. For simplicity these are modularised into sub-directories with individual instructions contained within. When training and running the cGAN, we recommend visiting and following instructions from the sub-directories in the following order:
- data: Loading data and creating tfrecords for training.
- model: Setting up the model architecture and training the model.
- scripts: Generating forecasts.
Additionally, sub-directories evaluation and config contain evaluation scripts and the necessary configuration files for setting data paths and model architecture.