Development of expert systems for the prediction of scour depth under live-bed conditions at river confluences: Application of different types of ANNs and the M5P model tree

The three-dimensional structure of water flow at river confluences makes these zones of particular importance in the fields of river engineering, fluvial geomorphology, sedimentology and navigation. While previous research has concentrated on the effects of hydraulic and geometric parameters on the scour patterns at river confluences, there remains a lack of expert systems designed to predict the maximum scour depth (d_sm). In the present study, several soft computing models, namely multi-layer perceptron (MLP), radial basis function (RBF) and M5P model tree, were used to predict the d_sm at river confluences under live-bed conditions. Model performance, assessed through a number of statistical indices (RMSE, MAE, MARE and R²), showed that while all three models could provide acceptable predictions of d_sm under live-bed conditions, the MLP model was the most accurate. By testing the models at three different ranges of scour depths, we determined that while the MLP model was the most accurate model in the low scour depth range, the RBF model was more accurate in the higher range of scour depths.