Photographic Visualization of Weather Forecasts with Generative Adversarial Networks

This is the companion upload for our paper Photographic Visualizations of Weather Forecasts with Generative Adversarial Networks. 'photocast.zip' contains the Tensorflow implementations of the generator and discriminator networks, as well as the code to reproduce the experiments presented in Sections 1.2 and 4 of the paper. The data used in the expert evaluations and their detailed results (summarized in Tables 2, 3 and 5 of the paper) are also available. Checkpoints of the trained models for the three camera locations are available in the respective .zip files. Simply uncompress the contents in the corresponding checkpoint folder, e.g. 'experiment/cevio/run__/20210716_080451/checkpoint'. The complete image archive and COSMO-1 forecast fields used for the training and evaluation of the networks can not be published online due to licensing restrictions. But it can be obtained free of charge for academic research purposes by contacting the MeteoSwiss customer service desk. Once obtained, add the images to 'data/raw__/image' and the COSMO-1 .csv files to 'data/raw__/cosmo_1-2019-2020' (see the examples therein). Then regenerate the training and testing data sets using the scripts in 'data/cosmo/surface' and 'data/nowcasting'. If you want to make use of the code, we suggest to fork our GitHub repository.Added additional results in the evaluation of matching weather conditions and the reproduction of the current imag