Estimation of scour downstream of a ski-jump bucket using the multivariate adaptive regression splines

In this paper, modeling the scour downstream of a flip bucket of spillways was considered using empirical formulas, soft computing techniques such as multilayer perceptron (mlp) neural network, and multivariate adaptive regression splines (mars). for this purpose, 95 data sets were collected with regard to the most affective parameters on the scouring phenomena at downstream of spillways. during the mlp model development, it was found that the two transfer functions, such as log-sigmoid and radial basis, had very suitable performances for predicting the desired scouring phenomena. the results of mars model showed that this model with coefficient of determination 0.99 and 0.91 during the development and testing stages, respectively, had suitable performance for modeling the scouring depth at downstream of flip bucket structure. the results of gamma test and mars model indicated that q/(gd^3w%), r/dw, and h/dw were the most affective parameters on the scouring phenomena.