Evaluation of different weighting methods to predict scour depth on grade control structures

The scour depth estimation is of great importance. Researchers have provided many empirical relationships based on laboratory and field work, but so far no relationship has been found to be satisfactory in different situations. In this study, the accuracy of different experimental (individual) relationships was evaluated in two stages before and after the correctional correction. The results show that before and after the bias correction, the individual relations of the National Institute and Mason with the mean square error of 0.87 and 0.23 meters are the most accurate. The combination of individual relationships with combined methods models shows that the GRA and EWA methods with a mean square error of 0.25 and 0.23 meters were the most accurate among the direct methods in before and after bias correction stage. The error of the AICA and BICA indirect methods in before and after bias correction is similar to the best single relation and could not improve the results of the individual relationships. The results of local method (KNN) and artificial intelligence (LS-SVM) method before and after bias correction are equal and estimate scour depth more accurately than individual relations. Comparison of different combinational methods in before and after bias correction shows that individual relationships with maximum scour depth using different combinational methods can improve predictive accuracy.