Variation of dominant discharge along the riverbed based on numerical and deep-learning models: A case study in the Middle Huaihe River, China.

• A bed-steadying discharge is proposed. • A flow-sediment model and an LSTM model are integrated. • The riverbed deformation is predicted. River morphology plays an important role in water environment and resources. The dominant discharge (Q D) is a crucial indicator for understanding river morphology and bed evolution under the impact of various interacting processes. At present, the identification of Q D mainly depends on the analysis of a large number of hydrological data derived from measuring stations, leading to difficulty in obtaining detailed Q D distributions along the study reach. In this paper, Q D is approximately expressed as the bed-steadying discharge (Q S) which is based on major factors of water level and sediment-carrying capacity. Subsequently, an integrated model combining a numerical fluid-flow and sediment model with a deep-learning algorithm is applied to analyze the changing process of the Q S. The flow and sediment transport processes are simulated by the calibrated mathematical model, which are then adopted as the input sequences for the long short-term memory (LSTM) module. The verification results of the established LSTM model show robustness and accuracy in predicting the flow and sediment transport processes under multi-stage incoming flow and sediment conditions. Furthermore, the proposed integrated model is applied to identify the detailed process of Q S in the Middle Huaihe River. Results show that the changing process of Q S along the study reach is characterized by a particular trend of "increase-decrease-rapid increase" due to natural changes and human activities. Additionally, the Q S agrees well with Q D at the hydrological station, showing that Q S can be applied as an approximation for Q D along the study reach. By analyzing longitudinal and transverse profiles, the rationality of using Q S as obtained by the newly presented model is demonstrated. Its temporal variation is also consistent with the cross-sectional changes for the specified stations. [ABSTRACT FROM AUTHOR]