Comments to "Rainfall–Runoff Prediction at Multiple Timescales with a Single Long Short-Term Memory Network"

Multiple Timescales with a Single Long Short-Term Memory Network" (LSTM) by Martin Gauch et al. presents an extension of LSTM hydrological models to sub-daily time steps. In previous publications, LSTMs as hydrological models were used on a daily time step. The authors explore multiple approaches to achieve a 'multi-timescale' model, of which three (naive LSTM, sMTS-LSTM, MTS-LSTM) are evaluated in more detail and less promising experiments are briefly explained in an Annexe. Similar to previous applications of LSTMs, the models C1 HESSD Interactive comment Printer-friendly version Discussion paper are applied at the CAMELS dataset, encompassing 516 basins across the contiguous USA where hourly data is available. Results are compared to the NOAA National Water Model (NWM) and show that all LSTMs architectures outperform the NWM. The authors suggest that the MTS-LSTM provides most flexbility for future use.