Comparative Study for Daily Streamflow Simulation with Different Machine Learning Methods

Rainfall–runoff modeling has been of great importance for flood control and water resource management. However, the selection hydrological models is challenging to obtain superior simulation performance especially with rapid development machine learning techniques. Three under different categories methods, including support vector regression (SVR), extreme gradient boosting (XGBoost), long-short term memory neural network (LSTM), were assessed simulating daily runoff over a mountainous river catchment. The performances input scenarios compared. Additionally, joint multifractal spectra (JMS) method was implemented evaluate during wet dry seasons. results show that: (1) LSTM always obtained higher accuracy than XGBoost SVR; (2) impacts variables such as antecedent streamflow rainfall LSTM; (3) showed relatively high seasons, classification seasons improved performance, seasons; (4) JMS analysis indicated advantages hybrid model combined trained wet-season data dry-season data.