Sediment transport forecasting with machine learning methods and comparison of their performances

The management of water resources, flood control, and the construction and maintenance of water structures like dams all depend heavily on the precise computation of the suspended sediment load. Stream flow, water temperature, and daily suspended sediment concentration data from USGS station number 11447650 in the Sacramento River for 1966–1973 were utilized in this work to estimate the suspended sediment load, which is dependent on numerous factors. Fifteen distinct models were provided in this study, which used lag values from various intervals to examine various combinations of suspended particle concentration, stream flow, and water temperature parameters. Four machine learning techniques that are often used in the literature—ANFIS, Support Vector Machine, Random Forest, and Gaussian Process Regression—as well as the wavelet transform to improve estimate precision were used to assess the built models. Model performances were compared both before and after wavelet transform. In the estimation of daily suspended sediment concentration, ANFIS, Support Vector Machine, Random Forest and Gaussian Process Regression methods were found to perform quite well both before and after wavelet transformation. The model that gave the best result was the W-ANFIS M02 model after wavelet transformation (Correlation: 0.98, NS: 0.97, KGE: 0.97, PI: 0.07, RMSE: 0.17).