Estimation of Bedload Sediment Using Decision Trees Method and Comparison with Empirical Methods

Estimation of the amount of sediment material (carried by a specific stream) is one of the main topics of sediment research. Sediment research is important in many engineering projects such as planning and design of water storage resources, morphology and changes in river bed, annual sediment estimation for reservoirs, sustainable irrigation, coastal protection, channel dredging, and etc.The evaluation of the sediment occurrences, and estimation of the sediment carried by rivers has a special importance and it is necessary to develop new methods with easy application for estimating the sediment transport. This study seeks to develop a method to be able to predict the amount of bedload with classification and regression trees and compare with some well-known empirical formulas such as Meyer Petter Muller, Shields, Einstein Brown, and Shoklitch. Measured data from Yazdakan station, located in Qotour River in Azarbayjan Gharbi, northwest of Iran, has been used in this research. It is a mountainous river with gravel bed that carries a high volume of sedimentation. A set of 76 field measurements over 8 years was used in modeling after verification and refinement. The model was trained on 80 percent of data and tested on the remained 20 percent. The best input contained whole four variables; flow discharge, suspended sediment discharge, flow depth and velocity, and the output was bedload discharge. The quantitative results of the best Decision Tree include MAE = 2134, RMSE = 2668 and % R = 89. The results of Regression Trees demonstrate more accuracy in the prediction of bedload compared with empirical formulas.