River Sediment Amounts Prediction with Regression and Support Vector Machine Methods.

Abstract

Accurate estimation of the amount of sediment in rivers; determination of pollution, river transport, determination of dam life, etc. matters are very important. In this study, sediment estimation in the river was made using Interaction Regression (IR), Pure-Quadratic Regression (PQR) and Support Vector machine (SVM) methods. The observation station on the Patapsco River near Catonsville was chosen as the study area. Prediction model was developed by using daily flow and turbidity data between 2015- 2018 as input parameters. Models were compared to each other according to three statistical criteria, namely, root mean square errors (RMSE), mean absolute relative error (MAE) and determination coefficient (R2 ). These criteria were used to evaluate the performance of the models. When the model results were compared with each other, it was seen that the IR model gave results consistent with the actual measurement results.

Keywords

• estimation
• turbidity
• regression
• support vector machine
• streamflow