Author(s): Azadeh Gholami; Ozgur Kisi; Andrew Binns; Saba Shaghaghi; Bahram Gharabaghi; Hossein Bonakdari

Linked Author(s):

Keywords: River equilibrium; Regime rivers; Stable channels; Geometry; Sensitivity analysis; Uncertainty analysis

Abstract: The complex dynamic equilibrium state of rivers, in which the amount of deposition and erosion are in balance, has been a fundamental research topic in river engineering. In this research, three advanced machine learning approaches, including M5 Model Tree (M5Tree), Multivariate Adaptive Regression Splines (MARS) and Least Square Support Vector Regression (LSSVR), are employed to gain new insights and develop more accurate methods for assessment of the longitudinal slope ( S ), water-surface width ( W ) and mean water depth ( D ) of rivers in regime state. Geometric and hydraulic characteristic of 85 cross-sections of Gamasiab River (located in western of Iran), Kaaj River (located in southwestern of Iran) and Behesht-Abad River (located in southwestern of Iran) are used to train and evaluate the employed methods (M5Tree, MARS, and LSSVR). Seven different models comprising various combinations of effective parameters influencing regime river geometry (the flow discharge ( Q ), median bed grain size ( d 50 ) and Shields parameter ( τ ∗ )), are developed to evaluate the effect of each of these variables on the prediction of the geometry of regime rivers ( S , W and D ). The M5Tree method outperformed the other approaches with respect to correlation coefficient ( R ) values of 0.872, 0.951, and 0.770 and Mean Absolute Relative Error (MARE) values of 0.484, 0.102, and 0.126 for slope, width, and depth prediction, respectively. Furthermore, the flow discharge Q was the key variable governing regime channel width and depth while the regime channel slope was found to be mainly controlled by the Shields parameter.

DOI: https://doi.org/10.1080/15715124.2018.1546731

Year: 2019