Prediction of sediment transport rates in gravel-bed rivers using Gaussian process regression

Estimating sediment transport rate in rivers has high importance due to the difficulties and costs associated with its measurement, which has drawn the attention of experts in water engineering. In this study, Gaussian process regression (GPR) is applied to predict the sediment transport rate for 19 gravel-bed rivers in the United States. To compare the performance of GPR, the support vector machine (SVM) as a common type of kernel-based models was developed. Model inputs of sediment transport were prepared based on two scenarios: the first scenario considers only hydraulic characteristics and the second scenario was formed using hydraulic and sediment properties. Obtained results revealed that the GPR models present better performance compared to the SVM models and other empirical sediment transport formulas. Also, it was found that incorporating the second scenario as input led to better predictions. In addition, performing sensitivity analysis showed that the ratio of average velocity to shear flow velocity is the most effective parameter in predicting the sediment transport rate of gravel-bed rivers.