AN ACCURATE MODEL FOR SECONDARY FLOW EFFECT ON LATERAL DISTRIBUTION OF DEPTH AVERAGED VELOCITY IN COMPOUND CHANNELS

The analytical solution of the lateral distribution of depth averaged velocity in compound channels is known as the Shiono and Knight method (SKM). In comparison with 1-D methods (such as ``Divided Channel Method'', ``Coherence Method'' and ``Exchange Discharge Model''), SKM as a 2-D method enjoys the advantage of computing the distribution of local velocity across the channel. SKM uses three coefficients f, lambda and Gamma as the representatives of bed resistance, lateral viscosity and secondary currents, respectively. Obviously, accurate estimation of these parameters plays an important role on approaching to exact values. The estimation of f, lambda and Gamma has been the subject of numerous researches, but all of them have assumed constant values for Gamma in width of channel. This research is to shed more light on the secondary flow parameter and its accurate estimation effect on the results. As the first step, the exact experimental values for Gamma were calculated based on SERC-FCF data, Series No. 2. Then the distribution of Gamma across the channel was extracted for each sub-section. In opposition to current procedure of SKM calculation, the exact values of Gamma showed that this parameter is not really constant and follows a distribution of almost third order of the lateral channel coordinate. In the next step, the best function of Gamma for each experiment was calculated and the results of Shiono and Knight's analytical solution for velocity and shear stress, with and without using the suggested model for secondary flow effect, were compared with the observation data. The comparison certified a good agreement between the results of new model and the experimental data, especially in the border region between main channel and floodplain, where the current procedure on calculation of SKM does not work correctly. Comparison of relative errors between calculated and observed velocities showed that the suggested model was competent to reduce the average of errors from 9.1% to 2.3%.