 The paper proposes a novel data-driven approach using long-short-term memory networks, LSTMs, for regional rainfall runoff modeling, which significantly improves performance compared to traditional hydrological models when calibrated for multiple basins together. The proposed approach also achieves better performance than individual basin calibration and learns catchment similarities as a feature layer in a deep learning model. This article was authored by F. Krotzert, D. Klotz, G. Shalev, and others.