 This paper proposes a novel approach to regional rainfall runoff modeling by utilizing long short term memory networks, LSTMs. This approach was tested against other hydrological models that had been calibrated for individual basins or for multiple basins simultaneously. The result showed that the LSTM model outperformed all other models, even those that had been calibrated for individual basins. Additionally, the authors proposed an adaptation to the standard LSTM architecture called Entity Aware LSTM ELSTM that allowed for learning catchment similarities as a feature layer in a deep learning model. These learned catchment similarities corresponded well with prior hydrological understanding. This article was authored by F. Krotzert, T. Klotz, G. Shalev, and others.