Exploring the ability of LSTM-based hydrological models to simulate streamflow time series for flood frequency analysis

Abstract. An increasing number of studies have shown the prowess of Long Short-Term Memory (LSTM) networks for hydrological modelling and forecasting. One commonly cited drawback of these methods, however, is the requirement for large amounts of training data to properly reproduce streamflow events. For maximum annual streamflow, this can be problematic since they are by definition less common than mid- or low-flows, leading to under-representation in the model’s training set and, ultimately, parameterization. This study investigates six methods to improve peak streamflow simulation skill of LSTM models used to extend streamflow observation time series for flood frequency analysis (FFA). Methods include adding meteorological data variables, providing streamflow simulations from a distributed hydrological model, oversampling peak streamflow events, adding multihead attention mechanisms, adding data from a large set of “donor” catchments and combining some of these elements in a single model. Furthermore, results are compared to those obtained by the distributed hydrological model HYDROTEL. The study is performed on 88 catchments in the province of Quebec using a leave-one-out cross-validation implementation and an FFA is applied using observations as well as model simulations. Results show that LSTM-based models are able to simulate peak streamflow as well (for a simple LSTM model implementation) or better (with hybrid LSTM-hydrological model implementations) than the distributed hydrological model. Multiple pathways forward to further improve the LSTM-based model’s ability to predict peak streamflow are provided and discussed.