The accurate prediction of rainfall, and in particular of the heaviest rainfall events, remains challenging for numerical weather prediction (NWP) models. This may be due to subgrid-scale parameterizations of processes that play a crucial role in the multi-scale dynamics generating rainfall, as well as the strongly intermittent nature and the highly skewed, non-Gaussian distribution of rainfall. Here we show that a U-Net-based deep neural network can learn heavy rainfall events from a NWP ensemble. A frequency-based weighting of the loss function is proposed to enable the learning of heavy rainfall events in the distributions' tails. We apply our framework in a post-processing step to correct for errors in the model-predicted rainfall. Our method yields a much more accurate representation of relative rainfall frequencies and improves the forecast skill of heavy rainfall events by factors ranging from two to above six, depending on the event magnitude.
