Recurrent neural networks (RNNs) can effectively model river routing processes and outperform physics-based models when trained on global datasets of runoff and streamflow.
