Rainfall-runoff Simulation in the Small and Medium-sized Catchments Using the Long Short-Term Memory Network
Flood forecasting is one of the important non-engineering measures for flood prevention and mitigation in river basins.Presently,heavy rainfall and flood disasters occur frequently in the small and medium-sized catchments,and the capability of flood forecasting remains limited when facing short-term rainstorm.Taking a small and medium-sized catchment in the eastern mountainous area of Anhui Province as the study site,this paper built a rainfall-runoff model u-sing the Long Short-Term Memory(LSTM)network,and the modelling performance of the LSTM model in the selected study site was explored.The results show that the selecting the stational measured rainfall as model inputs can improve its performance on rainfall-runoff modelling and prediction,and the LSTM model achieves higher precision than tradition-al artificial neural networks.Besides,the LSTM model effectively captures the complicated rainfall-runoff relationship in the selected river basin,and can achieve satisfactory performance on the flood peak simulation.For the training and tes-ting flood events,the calculated qualified rates of flood peak simulation are both over 90%.Meanwhile,the internal structural characteristics of the LSTM model have strong similarity with the hydrological process over the river basin,which contributes to a good fitting performance of the nonlinear rainfall-runoff relationship in the selected catchment.
mountainous regionlong short-term memory networksmall and medium-sized riversrainfall-runoff simulation
NETL
NSTL
万方数据
