Suspended Sediment Estimation Using Machine Learning Methods

Abstract

Suspended sediment in rivers is important for efficiently using water resources and hydraulic structures. In this study, the suspended sediment load of rivers was estimated using traditional multi-linear regression (MLR), machine learning methods such as the support vector machines (SVM) and M5 decision tree (M5T). Data on daily stream flow, daily maximum and minimum water temperature and suspended sediment concentration in the river were used as input data in all models to predict daily suspended sediment discharge. The performance of all methods is evaluated based on a statistical approach. Determination coefficient (R2), root mean square error (RMSE) and mean absolute error (MAE) are used as comparison criteria. Overall, the machine learning approaches better predict suspended sediment discharge.

Keywords

• sediment discharge
• prediction
• linear regression
• support vector machines
• m5 tree.