Assessment of Regression Trees and Multivariate Adaptive Regression Splines for Prediction of Scour Depth Below the Ski-Jump Bucket Spillway

Spillways are constructed in dams in order to discharge the excess water in the reservoir. In the skijump
bucket spillways, water jet impacts diagonally to downstream erodible bed and causes scour
hole downstream of the dam. The scour hole development may threaten the stability of the dam.
Hence, an accurate and correct estimation of scour depth is one of the most important issues in
hydraulic engineering. In recent years, soft computing tools have been widely used to model complex
and nonlinear phenomena. Therefore, in this study, using data mining algorithms such as
classification and regression trees and multivariate adaptive regression splines have been used for
estimation of maximum scour depth at the downstream of the ski-jump bucket spillway. For this
purpose, these models were developed using 95 experimental data and dimensionless parameters. The
results showed 3 q gH as the most important parameter in prediction of scour depth. In addition,
statistical indicators and scatter diagrams showed that multivariate adaptive regression splines have
the highest value of correlation coefficient CC=0.966 and minimum error measures RMSE=0.075 and
MAE=0.057 and were more accurate than regression trees in prediction of scour depth below a skijump
bucket spillway.