A Comparative Analysis of Multiple Machine Learning Methods for Flood Routing in the Yangtze River

Obtaining more accurate flood information downstream of a reservoir is crucial for guiding regulation and reducing the occurrence disasters. In this paper, six popular ML models, including support vector regression (SVR), Gaussian process (GPR), random forest (RFR), multilayer perceptron (MLP), long short-term memory (LSTM) gated recurrent unit (GRU) were selected compared their effectiveness in routing two complicated reaches located at upper middle main stream Yangtze River. The results suggested that performance MLP, LSTM GRU models all gradually improved then slightly decreased as time lag increased. Furthermore, outperformed SVR, GPR RFR model demonstrated superior across range efficiency criteria, mean absolute percentage error (MAPE), root square (RMSE), Nash–Sutcliffe coefficient (NSE), Taylor skill score (TSS) Kling–Gupta (KGE). Specifically, achieved reductions MAPE RMSE least 7.66% 3.80% first case study 19.51% 11.76% second study. paper indicated was most appropriate choice