The 24th IAHR-APD Congress (24th IAHR-APD Congress)

534 / 2024-06-30 22:40:06

Flood routing network construction in mega-basin using data-driven artificial intelligence model

Flood routing; Muskingum model; Long short-term memory; Data-driven model; Mega-basin

Subtheme 2: Integrated Water Resources Management > Topic 1: Water Resources Management system, policy and regulation

Rapid and accurate simulation of flood routing network in mega-basin is key to the development of water resources management policies. However, there is no generalized method for dealing with multi-tributary flood routing problems. In this study, we propose a generalization method applicable to complex river systems to construct flood routing network. What’s more, we use a data-driven long short-term memory (LSTM) method to simulate flood routing in rivers and choose the linear Muskingum model (LMM) as a comparative model of LSTM. In this study, the Upper Yangtze River basin is divided into four sections, and the flood routing network is simulated using LMM and LSTM methods. Compare to LMM method, LSTM method can achieve higher Nash-Sutcliffe Efficiency (NSE) values in all scenarios. The NSE values of four hydrological stations are above 0.99 in training, validation and test periods, which indicates that LSTM can be applied to simulate the flood routing network in the Upper Yangtze River basin.