Forecasting of River Sediment Amount using Machine Model

Abstract

Accurate estimation of sediments is important in river structures. The amount of suspended sediments is mostly determined by measurements from observation stations, sediment key curve, artificial intelligence modeling methods. In this study, the estimation of the sediments content was performed by using hydro-meteorological parameters such as river flow, air temperature and precipitation measured between 2011-2017 at Omaha Station in Nebraska. For the estimation of sediments amount, Support Vector Machines (SVM) and Generalized Regression Neural Network (GRNN) methods were used. These models were compared by using correlation coefficient (R), mean absolute error (MAE) and root of mean square errors (RMSE). When the measurement and model results were compared, SVM and GRNN models gave consistent results in the estimation of sediments content in rivers. Nevertheless, the SVM method showed slightly better correlation and lower error performance than the GRNN method.