COMPARISON OF FINITE DIFFERENCE AND FINITE ELEMENT SOLUTIONS OF UNSTEADY FLOW IN COMPOUND CHANNELS SAYYED ALI AYYOUBZADEH Faculty of Irrigation Hydraulic Structures, Tarbiat Modares University, Tehran, Iran ABDOLREZA ZAHIRI PhD Student of Irrigation Hydraulic structures, Tarbiat Modares University, Tehran, Iran Unsteady flow in compound open channels is complicated by large differences in hydraulic properties (flow depth, resistance) and cross sectional geometry of the main channel and the floodplains. In this paper, the flood routing in compound open channels has been conducted using a depth-averaged flow model in collaboration with diffusion wave model by finite difference numerical scheme. Implicit scheme has been used in this study. Comparison of these results with the finite element ones of Abril [1] for a hypothetical river with a symmetric compound channel show good agreement between the two approaches. This model has extended to heterogeneous compound channels. INTRODUCTION Flood routing in rivers with wide floodplains has much concerns by the researchers during the last few decades. In most of these studies, the floodplains have been considered as ineffective sections with very low flow velocity. In addition, the onedimensional Saint-Venant equations have been used for flood routing. Due to large differences in hydraulic properties (flow depth, resistance) and cross sectional geometry of the main channel and the floodplains, flow hydraulics in these sections is so complex. Momentum transfer between main channel and flood plains are so important and must be considered in flood routing computations. One of the most practical models for analysis of flow hydraulics in compound channels is the two-dimensional depth-averaged Shiono and Knight model [14]. This model is based on the Navier-Stocks equations and solves the lateral distribution of velocity and boundary shear stress. Recently, Abril [1] used the finite element solution of this model jointed with the convection-diffusion equation for flood routing computations in a symmetric hypothetical compound channel. In this paper, the finite difference solution of Shiono-Knight model has been used for unsteady flow simulation in compound river channels. Although the finite element solution has many advantages related to finite difference solution, but the later is simple and has wide applications in hydraulic engineering problems. The present model has extended to symmetric compound channels with heterogeneous roughness. 1 2 GOVERNING EQUATIONS Two-dimensional Shiono and Knight model Flow hydraulics in compound channels can be described by two-dimensional depthaveraged mathematical model (Shiono and Knight [14]). This model is based on the Navier-Stokes momentum equation. For streamwise motion: g H S0 f 1 ud 2 1 2 8 y s 1/ 2 ud 2 f H u d 8 y H ( U V )d y (1) where ρ=fluid density; g = acceleration of gravity; H=local water depth; S0= channel streamwise slope; f=Darcy-Wiscbach friction factor, ud=depth-averaged velocity; s=channel side slope of the banks (1:s, vertical:horizental); =dimensionless eddy viscosity; {UV}=velocity components in the {xy} directions, x=streamwise parallel to the bed; and y=lateral direction. The right hand side of the above equation is the secondary flow term, which is significant in some cases, e.g. meandering compound channels (Shiono and knight [15], Ervine et. al. [7]). Calibration processes carried out by Abril showed that the secondary flow term in the main channel and floodplains may be estimated from equation (2): H ( UV )d s gS0 H y (2) where s is calibration coefficient. The value of this coefficient has been suggested 0.05 in main channel for inbank flow. In the overbank flow, these values are 0.15 and – 0.25 for main channel and floodplains respectively (Shiono and Knight [15], Knight and Abril [10], Abril [1]). Shiono and Knight [15] have solved Eq. 1 analytically for trapezoidal compound cross sections. Although the analytical solution of this equation has also been used in irregular compound river channels (Ayyoubzadeh and Zahiri [5]), but in this paper, the numerical solution is used (Ayyoubzadeh and Zahiri [4]). Eq. 1 is a nonlinear partial differential equation. Using the transform of u 1 u 2 and assumption of X u 2 , a linear equation will be appeared: u d y 2 y g H S0 1/ 2 X f 1 1 2 f X 1 2 H S gHS 0 8 2y 8 y s (3) The above equation can be solved numerically by implicit finite difference approach. For calculation of Darcy-Wiscbach and dimensionless turbulent eddy viscosity coefficients in main channel and floodplains, the Abril’s calibration relationships [1] are used in this paper. These relationships for homogeneous compound channels for each water depth are: 3 f f fc R f & f c R R f 0.669 0.331 Dr R 0.20 1.2 Dr 0.719 1.44 (4) (5) (6) where R refers to the ratio of floodplain to main channel’s value (friction factor and dimensionless eddy viscosity) and Dr is relative depth (ratio of flow depth in floodplain to the main channel). The c and f subscripts refer to the main channel and floodplain respectively. Friction factor in main channel for inbank flow is calculated based on the Manning’s roughness coefficient. Abril [1] assumes a constant value of friction factor in main channel for overbank flow and for all flow depths, equal to bankfull flow value. The most appropriate calibration value for main channel’s eddy viscosity was 0.07 for all flow depths (Abril [1]). With these information, Eqs. 4-6 are utilized for calculation of friction factor and eddy viscosity of floodplains. THE CONVECTION-DIFFUSION EQUATION The Saint-Venant equations are the basic governing equations for flood routing in rivers. Different forms of these equations have been used by various authors for flood routing computations in compound channels (Abril [1], Ayyoubzadeh and Zahiri [6], Fread [8], Mizanur and Chaudhry [13], Moussa and Bocquillon [14], Tuitoek and Hicks [17],). The full dynamic wave, diffusion analogy and characteristic wave are some kinds of hydraulic flood routing. In this paper, the diffusion analogy has been utilized. The convectiondiffusion equation is: 1 Q Q D Q 0 c t x x c x (7) where Q is flow discharge, t is time, c is flood wave celerity and D is the flood propagation coefficient. The c and D hydrodynamic coefficients are functions of flood flow discharge: 1 dQ T dh Q D 2TS 0 c (8) where T is top water surface width and h is flow depth. These coefficients are calculated from stage-discharge rating curve of the compound channel. Eq. 7 is a non-linear second-order differential equation and can be solved numerically by implicit finite difference scheme to determine the unknown Q as function of x and t. Of course, the solution is subject to boundary condition at the upstream end of 4 the subcritical unsteady flow reach and initial conditions of Q along the reach at start of the time. These boundary and initial conditions are specified in the next sections. APPLICATION OF MODEL IN A HYPOTHETICAL RIEVR Ackers [3] defined a hypothetical symmetric compound cross section similar to natural rivers with trapezoidal sections in main channel and floodplains. This hypothetical river is used in this paper for flood routing computations. This river has a main channel of 15 m length and 1.5 m depth. The two floodplains are 20 m wide. Both side slopes of main channel and floodplains are 1:1. The selected reach is 20 km long with longitudinal bed slope of 0.003. The Manning roughness coefficients of 0.03 are assumed for both main channel and floodplains. Using the friction factors, turbulent eddy viscosity and secondary flow coefficients specified in the previous section, the lateral distribution of depth-averaged velocity across the compound channel can be solved by numerical finite difference solution of Eq. 1 for any arbitrary flow depth. An example of this lateral distribution is shown in figure 1 for flow depth of 2.0 m. Because of symmetry of the section, the results are illustrated only for a half-cross section. Velocity, Ud ( m /s ) . . . . . . Lateral Distance (m ) Figure 1. Lateral distribution of depth-averaged velocity for flow depth of 2.0 m Depth of Flow ( m ) . . . . . Finite Difference . Finite Element . Flow Discharge (m /s) Figure 2. Stage-discharge curves for hypothetical river simulated by two numerical methods 5 With lateral integration of this velocity distribution, the total flow discharge of compound channel for flow depth of 2.0 m can be determined. Repeating this procedure gives the stage-discharge curve of the section. This rating curve is shown in figure 2. In this figure the rating curve simulated by finite element approach (Abril [1]) is seen, too. As can be seen, the results of two numerical methods have close agreements. The bankfull discharge of the section has estimated nearly to 53.4 cms by both two methods. For flood routing computations in hypothetical river, firstly the variations of convection and diffusion coefficients as function of flow discharge are calculated from stage-discharge rating curve. Then the solution of Eq. 7 is obtained by implicit finite difference scheme according to the initial and upstream boundary condition. The convection and diffusion curves versus flow discharge are shown in figure. 3 for hypothetical river. These curves are exactly similar to the Abril’s results [1]. As can be seen, an abrupt reduction has occurred in these curves at bankfull flow discharge of 53.4 cms. This happens due to sudden increase of top water surface width once the flow initially fills the main channel and enters to the wide floodplains. These curves are essential for flood routing analysis since they express the speed and attenuation behaviour of a flood wave routed in the river (Abril [1]). C (m /s) & DS (m /s) Diffusion Convection Discharge (m /s) Figure 3. Variation of convection and diffusion coefficients vs. discharge in the hypothetical river INITIAL AND UPSTREAM BOUNDARY CONDITIONS Abril has used a mathematical inflow hydrograph for flood routing in hypothetical river by the form of: t t Q (t ) Qbase Q peak Qbase exp 1 t p t p (9) where Qbase is the initial or base flow of the river before the flood occurring, Qpeak is the peak flow discharge, tp is the time to peak discharge and is the curvature parameter which defines the width of the inflow hydrograph. The following values have been 6 utilized by Abril [1]: Qbase=10m3/s ; tp=15hr and 6 . The Qbase equal to 10m3/s was assumed along the river at the start time as initial uniform flow condition. Two values have been considered for peak flow discharges in this paper, one for inbank flow (40 m3/s) and another one for overbank flow (80 m3/s). The implicit finite difference solution of flood routing for the hypothetical river is shown in figure 4. In this figure both inbank and overbank outflows are illustrated. Fixed space and time steps of 500m and 100s are adapted in this numerical solution. For comparison, the finite element solutions of Abril [1] are observed, too. Close agreements between two numerical solutions can be seen from this Figure for both inbank and overbank flows. The peak discharge of outflow hydrographs and time to peaks are the same in two methods. As Abril [1] has stated, no attenuation of the peak discharge is observed in the outflow hydrographs 20 km downstream. Discharge (m /s) INFLOW Outflow - Finite Difference Outflow - Finite Elem ent Time (hr) Figure 4. Outflow hydrographs for hypothetical river simulated by two numerical methods . Depth (m ) . INFLOW Outflow - Finite Difference . . . . . Time (hr) Figure 5. Depth hydrographs at downstream end of the reach simulated by finite difference scheme for inbank and overbank flow 7 The numerical solution of convection-diffusion equation (7), gives only the outflow hydrographs for any arbitrary location along the river. However, using this outflow hydrographs and the stage-discharge rating curve of the river, both simulated by the finite difference scheme, the routed depth hydrographs can be calculated. Such calculations for downstream end of the reach are illustrated in figure 5 for example. These results haven’t presented by Abril [1] therefore only the finite difference solutions are shown here. The peak flow depth of 1.81 m has obtained for inflow and outflow depth hydrographs. EXTENDING THE MODEL FOR HETEROGENEOUS ROUGHNESS The governing equations of (1) and (7) are solved here numerically for homogeneous symmetric hypothetical river with compound cross section. Also, Abril [1] has solved these equations for compound channel with homogeneous roughness. In the natural river, however, the floodplains have roughness coefficients much greater than the main channels due to high growth of vegetations on floodplains during the low flow in the main channel. This heterogeneity makes the flow hydraulics in compound channels more complex and should be considered in design of flood alleviation schemes (Lai et. al. [12]). In this paper, the modified relationships for friction factors and secondary flow contributions obtained by Abril and Knight [2] are used for flood routing computations in heterogeneous compound channels. Abril and Knight [2] developed these relationships for heterogeneous compound channels: f c Rn2 f scHet scHom 1 I c (10) (11) where fc is the corrected friction factor of floodplains for heterogeneity, f is the floodplain’s friction factor obtained from Eq. 4 for homogeneous roughness, Rn is the ratio of Manning’s roughness coefficients between floodplain and main channel, s c Hom and s c Het refers to the main channel’s secondary flow coefficients in homogeneous and heterogeneous compound channel and Ic is the percentage increase in main channel’s secondary flow coefficient due to heterogeneity of compound channel with respect to the corresponding value found in homogeneous conditions. It has been observed from experimental data in compound channels with rough floodplains that the secondary currents in the main channel are much increased by the increase in roughness of the floodplains (Tominaga and Nezu [16]). The value of Ic is only a function (third order polynomial) of the roughness ratio Rn (Abril and Knight [2]): I c 0.0347 Rn3 0.485Rn2 3.03Rn 2.57 (12) 8 In the above equation, the variations of Ic with respect to the relative depth are ignored. For heterogeneous compound channels, the same values of turbulent eddy viscosity coefficients found in homogeneous conditions are used for main channel and floodplains, because the results of the two-dimensional Shiono-Knight model have very low sensitivity of these coefficients (Knight et. al. [11], Knight and Abril [10]). For flood routing simulations, the same hypothetical river is considered except that the Manning’s coefficient of 0.06 is assumed for floodplains, e.g. twice as rough as the main channel (Rn=2). The coefficients fc=0.065 and c 0.07 are considered for all overbank flow depths like the homogeneous conditions. The corrected values of main channel’s secondary flow and floodplains’ friction factor are obtained from Eqs. 10-12. Using this information, the flood routing computation carried out for the hypothetical river with the pre-specified initial and upstream boundary conditions. In figure 6, the finite difference solutions of the outflow hydrographs are illustrated for the reach of 20 km length. As can be seen, there is no attenuation of the peak flow discharge for heterogeneous conditions. INFLOW Discharge (m /s) OUTFLOW Time (hr) Figure 4. Outflow hydrographs for hypothetical river simulated by finite difference method for heterogeneous roughness CONCLUSIONS In this paper, using the implicit finite difference solution of two-dimensional depthaveraged Shiono-Knight model combined with the convection-diffusion equation, the flood routing for hypothetical compound channel with homogeneous roughness has been simulated. For both inbank and overbank flow conditions, the outflow hydrographs are obtained and compared against the finite element solutions of Abril [1]. The close agreement between two numerical methods, showed the suitable applicability of the finite difference scheme for flow hydraulic analysis of compound channels for both inbank and 9 overbank flows. Also, using the modified relationships of Abril and Knight [2], the present model has extended to the heterogeneous compound channels. ACKNOWLEDGMENTS The authors greatly thank the Prof. Abril, for his very useful comments. Also thanks are due to the committee of water engineering standards of the Khozestan Water and Power Authority for financial supports. REFFERENCES [1] Abril, J. B., “Overbank flood routing analysis applying jointly variable parameter diffusion and depth-averaged flow finite element models”, Proceedings of the International Conference on Fluvial Hydraulics, Belgium, (2002), pp 161-167. [2] Abril, J. B. and Knight, D. W., “Stage-discharge prediction for rivers in flood applying a depth-averaged model”, Journal of Hydraulic Research, IAHR, (2004), (To be published). [3] Ackers, P., “Stage-Discharge functions for two-stage channels”, Water and Environmental Management, Vol. 7, (1993), pp 52-61. [4] Ayyoubzadeh, S. A. and Zahiri, A., “Numerical solution of depth-averaged lateral distribution of velocity and bed shear stress in simple and compound channels”, 6th River Engineering International conference, Ahwaz, Iran, Vol. 1, (2003), pp 285-293 (in Persian). [5] Ayyoubzadeh, S. A. and Zahiri, A., “New Envelope Sections Method to Study Hydraulics of Compound Varying River Channels Using a Depth-Averaged 2D Model”, Journal of Engineering Sciences, Vol. 14, No. 2, (2003), pp 103-116 (in Persian). [6] Ayyoubzadeh, S. A. and Zahiri, A., “Numerical study of flood routing in compound channels”, Int. Conference on Hydraulics of Dams and River Structures, Tehran, Iran, (2004), pp 353-358. [7] Ervine, D. A., Babaeyan-Koopaei, K. and Sellin, R. H. J., “Two-dimensional solution for straight and meandering overbank flows”, Journal of Hydraulic Engineering, ASCE, Vol. 126, (2000), pp 653-669. [8] Fread, D. L., “Flood routing in meandering rivers with floodplains”, Proceedings Rivers ’76, Third Annual Symposium of Waterways, Harbors and Coastal Engineering Division, American Society of Civil Engineers, Vol. 1, (1976), pp 16-35. [9] Graf, W. H., “Fluvial Hydraulics”, John Wiley & Sons Ltd, England, (1998). [10] Knight, D. W. and Abril, J. B., “Refined calibration of a depth-averaged model for turbulent flow in a compound channel”, Journal of Water, Maritime and Energy, Vol. 118, (1996), pp 151-159. [11] Knight, D. W., Shiono, K. and Pirt, J., “Prediction of depth mean velocity and discharge in natural rivers with overbank flow”, International Conference on 10 [12] [13] [14] [15] [16] [17] Hydraulic and Environmental Modelling of Coastal, Estuarine and River Waters, Bradford, England, (1989), pp 419-428. Lai, C. J., Liu, C. L., and Lin, Y. Z., "Experiments on flood-wave propagation in compound channel", Journal of Hydraulic Engineering, ASCE, Vol. 126, No. 7, (2000), pp 492-501. Mizanur, R. and Chaudhry, M. H., “Flood routing in channels with flood plains”, Journal of Hydrology, Vol. 171, (1995), pp 75-91. Moussa, R. and Bocquillon, C., “Approximation zones of the Saint-Venant equations for flood routing with overbank flow”, Journal of Hydrology and Earth Sciences, Vol. 4, No. 2, (2000), pp 251-261. Shiono, K. and Knight, D.W., “Turbulent open-channel flows with variable depth across the channel”, Journal of Fluid Mechanics, Vol. 222, (1991), pp 617-646. Tominaga, A. and Nezu, I., “Turbulent structure in compound open channel flows”, Journal of Hydraulic Engineering, ASCE, Vol. 117, No. 1, (1991), pp 21-41. Tuitoek, D. K. and Hicks, F. E., “Modelling of unsteady flow in compound channels”, African Jouranl of Civil Engineering, JKUAT, Vol. 6, (2001), pp 45-54.