Optimal prediction of lateral velocity distribution in compound channels

Many rivers have deep main channels in the centre with one or two adjoining floodplains. Prediction of the lateral velocity distributions over the entire river cross-section is necessary for solving many river-related and hydro-engineering problems. Using Genetic Algorithm (GA), this paper proposes a simple model with two separate equations for predicting the lateral velocity distribution in the main channel and floodplains of straight compound channels. The proposed model is based on two key parameters of compound channels, i.e. depth ratio and the coherence parameter. The constants and exponents of the model are obtained by using a GA based on both experimental data of several compound flumes and measurements of the river Severn in the UK. Using several statistical measures, it is shown that the predictions of lateral velocity distribution and stage-discharge by the model agree well with the observed laboratory data and natural river measurements used for calibration and validation. Moreover, the model is shown to outperform the conventional vertical divided channel method with 11.2% less error on average in predicting the velocity distribution.