The Feasibility of Integrative Radial Basis M5Tree Predictive Model for River Suspended Sediment Load Simulation

Accurate suspended sediment transport prediction is highly significant for multiple river engineering sustainability. Conceptually evidenced, sediment load transport is highly stochastic, spatial distributed and redundant pattern due to the incorporation of various hydrological and morphological variables such as river flow discharge and sediment physical properties. The motivation of this study is to explore the feasibility of newly intelligent model called Radial basis M5 model tree (RM5Tree) for suspended sediment load (S_t) prediction for daily scale information at Trenton hydrological station, Delaware River. Numerous input combination attributes are formulated based on the preceding information of sediment and river flow discharge. The prediction accuracy “based statistical and graphical visualizations” of the proposed model validated against numerous well-established predictive models including response surface method (RSM), artificial neural network (ANN) and classical M5Tree based models. The investigated input combinations behaved differently from one case to another. The optimum input combination attributes are included two months lead times of sediment and discharge information to predict one step ahead S_t. The attained results of the proposed RM5Tree model exhibited a remarkable prediction accuracy with minimal values of root mean square error (RMSE≈2091 ton/day) and coefficient of determination (R²≈0.86). This presenting a percentage of enhancement in the prediction accuracies by (51.6, 53.1 and 26.3) over (RSM, ANN and M5Tree) optimal models over the testing phase.