Estimation of discharge capacity in meandering compound channels using artificial neural networks

Abstract

Flow in compound (or two-stage) channels is very complex and different energy loss mechanisms operate under different geometric and flow conditions. Neither theoretical analyses nor current empirical approaches are sufficiently developed for practical calculation of conveyance for all conditions experienced in practice. An alternative approach, using artificial neural network modelling, has been successfully applied to predict conveyance under a wide range of conditions. The model proposed uses a feed-forward system with one hidden layer and an error back-propagation learning procedure. It predicts a dimensionless discharge using input describing the main channel and floodplain flow depths, vegetation density over the cross section, channel sinuosity, transverse floodplain slope, and floodplain bend tightness. The discharge is dimensionalized by multiplication with the composite discharge calculated assuming frictional resistance only. The model was trained using 45 data sets representing a range of main channel and floodplain characteristics and tested using 15 additional data sets. The discharge prediction error for all the data used in development and testing the model was -0.19% on average and exceeded 15% for one condition only.Key words: compound channels, channel conveyance, flow resistance, neural networks.