Rainfall–runoff prediction at multiple timescales with a single Long Short-Term Memory network

Long Short-Term Memory Networks (LSTMs) have been applied to daily discharge prediction with remarkable success. Many practical scenarios, however, require predictions at more granular timescales. For instance, accurate of short but extreme flood peaks can make a life-saving difference, yet such may escape the coarse temporal resolution predictions. Naively training an LSTM on hourly data, entails very long input sequences that learning hard and computationally expensive. In this study, we propose two Multi-Timescale (MTS-LSTM) architectures jointly predict multiple timescales within one model, as they process long-past inputs single branch out into each individual timescale for recent steps. We test these models 516 basins across continental United States benchmark against US National Water Model. Compared naive distinct per timescale, multi-timescale are efficient no loss in accuracy. Beyond quality, different variables timescales, which is especially relevant operational applications where lead time meteorological forcings depends their resolution.