Development of methods to predict the discharge capacity in model and prototype meandering compound channels
The author developed two methods for predicting the discharge capacity of uniform meandering compound channels. The first method utilised an Artificial Neural Network (ANN) functional approximator which was taught to replicate the relationship between 9 of the key parameters and the magnitude of F* which was exhibited by the flow data gathered during the Series B extension (1993-1996) programme. The ANN approximator was constructed using a 'Matlab' software environment and was supported on a PC. The second method consisted of a semi-physical / semi-empirical method which was named the Enhanced zonal method. This method comprised formulations which explicitly determined the discharge capacity of the 3 individual flow zones in meandering compound channels whilst accounting for difference induced by the characteristic 4 flow region behaviour. The author demonstrated that both of these methods produce more accurate discharge capacity predictions than the James and Wark   method for the majority of available flow data sets. The James and Wark   method was the optimal method prior to the Series B extension (1993-1996) programme. The ANN approximator gave the most accurate predictions when the parameters of the compound channels to which it was applied fell within the range of the parameters incorporated in the ANN training data set. However, the author demonstrated that the Enhanced zonal method is the most reliable discharge capacity prediction method over the full range of uniform meandering compound channel configurations. The author developed two refined one-dimensional (1D) numerical models (for application to both steady and unsteady flow conditions) which incorporated the Ackers  and James and Wark   methods to determine the conveyance characteristics at representative uniform cross-sections in natural meandering compound channels. The author demonstrated, using a case study of the River Dane, that these refined 1D models were able to predict the water surface profiles in natural channels to a high degree of accuracy.