Sediment Rating Curve Estimation Using Robust Regression

A sediment rating curve is the most known method in the hydrological approach for suspended sediment load estimation that is a power equation (or linear equation based on logarithmic data transformation) to relate suspended sediment load to the river discharge. The conventional method to determine the sediment rating curve is ordinary least square (OLS) that outliers can significantly influence. The robust estimations are statistical methods developed to overcome the limitations of OLS. In this study, two robust estimation methods, including MM and least trimmed square (LTS) methods, were investigated for sediment rating curve estimation, and their results were compared with OLS. To compare these three estimators, recorded data sets of four hydrometry stations, Baghou, Alang-Darreh, Anjirab, and Jafakandeh, located in Golestan province with 33-279 pairs of data were used. Coefficient of determination, root mean square error, and mean absolute error as numerical criteria, and graphical criterion was used to compare the results. Assessment of numerical precision criteria showed suitable efficiency for equations estimated based on robust estimator application and revealed their ability to improve OLS estimation. Evaluation of graphical criterion revealed that robust regression estimations could be similar or different to the OLS.