Water surface profile prediction in compound channels with vegetated floodplains

Floods have become one of the most dangerous and frequent natural disasters. Most rivers are characterised by compound cross-sections that are usually contain vegetation. The ability to simulate water surface profiles (WSPs) in vegetated rivers quickly and accurately is crucial in flood forecasting operations. The aim of this study was to develop a low-cost and practical tool for predicting the WSP in compound channels with vegetated floodplains. In particular, artificial neural network (ANN) and support vector machine (SVM) techniques were used to devise a model for the prediction of the WSP in an experimental channel. Two approaches were employed: the first was based on the use of non-dimensional data and the second used dimensional data. The performance of the prediction methods was determined using a ten-fold cross-validation approach. Comparative results revealed that the SVM algorithm outperformed the ANN and regression models. The performance of the SVM model using dimensional data (correlation coefficient of 0.99 ± 0.005 and mean absolute error of 0.0019 ± 0.0002) was shown to be marginally better than the model using dimensionless data. Sensitivity analysis also indicated that the relative discharge and relative depth played the most important roles in estimating the WSP.