Weather forecasting based on methods of artificial intelligence (AI) is a rapidly evolving field. A recent study from KIT has shown for West Africa that a deterministic 24-hour rainfall forecast using a „convolutional neural network“ architecture, which is then converted into a probabilistic forecast using the new Easy Uncertainty Quantification (EasyUQ) method (Walz et al. 2024), has more skill than any of the conventional numerical weather prediction (NWP) based forecasts tested.  Earlier, work by Little et al. (2009) had shown that simple statistical (Markov) models (Markov-SMGLM) provide better precipitation forecasts than NWP ensembles + post-processing for a one-day lead time over England.  This leads us to assume that a one-day forecast of precipitation based on CNN + EasyUQ could challenge NWP forecasts for Germany, too.  In the Master thesis you will train an existing CNN + EasyUQ model architecture on radar-based precipitation and verify the forecasts against station data from the German Weather Service (DWD).  The thesis will be co-supervised by the group of Prof. T. Gneiting (Computational Statistics).