Weather forecasts from numerical weather prediction models suffer from systematic errors. Statistical post-processing corrects these errors by removing forecast biases and improving the uncertainty quantification. A major challenge is to retain spatial and temporal dependencies in the post-processed forecasts. The aim of this project is to develop new machine learning based post-processing methods that yield spatially consistent predictions. In the first part of this project we target sub-seasonal weather forecasts that have a forecast horizon of three to six weeks. We adapt deep learning methods for image data to post-process global weather forecasts, thereby improving forecast calibration and skill. This project is part of the Young Investigator Group “Artificial Intelligence for Probabilistic Weather Forecasting”  of Sebastian Lerch funded by the Vector Stiftung.