Prediction of River Pipeline Scour Depth Using Multivariate Adaptive Regression Splines

In this study, the multivariate adaptive regression splines (MARS) technique was applied to estimate scour depth around pipelines. To this purpose, 90 data sets related to effective dimensionless parameters on pipeline scouring phenomena were gathered from literature. A gamma test (GT) was used to define the most-effective parameters on scouring phenomena below pipelines. Performance of MARS model was compared with multilayer perceptron (MLP) neural network and empirical formulas. Results of the GT showed that e/D, τ*, and y/D are the most important parameters for scour depth. Results of MARS model with coefficient of determination (0.91) and root-mean square error (0.05) indicated that this model has suitable performance for predicting scour depth under pipelines and results of this model are more accurate compared to empirical formulas. Comparing results of MARS model and MLP showed that accuracy of MARS model is slightly lower than that of the MLP.