spateGAN: Spatio-Temporal Downscaling of Rainfall Fields using a cGAN Approach - Showcase downscaling and animations

Supplementary material "spateGAN: Spatio-Temporal Downscaling of Rainfall Fields using a cGAN Approach"<br> ========= spateGAN is a conditional Adversarial Generative Network for spatio-temporal precipitation downscaling. Corresponding to the downscaling of global climate model outputs within the climate sciences domain, it reconstructs the resolution of coarsened weather radar image sequences from 32x32 km to 2x2 km and 1 hour to 10 minutes. This example repository includes a deterministic and probabilistic spateGAN model, illustrates their applicability and shows animations of downscaled rain events for three different days. The animations are also directly accessible via: Downscaling animation <br> Running the code<br> -----<br> Install conda envirnoment. The Jupyter notebook provides a downscaling scenario for a time sequence of one day, including:<br> * RADKLIM-YW data coarsening * probabilistic and deterministic trained spateGAN models<br> * spatio-temporal downscaling example<br> * plotting of the results<br> * animation to illustrate the temporally consistent and advection of model predictions For efficient model prediction, the use of a GPU with approximately 32 GB of memory is recommended.