外文翻译---CCHE1D渠道网络模型的灵敏度分析.docx

附 录附录一：外文原文Sensitivity Analysis of the CCHE1D Channel Network Model Weiming Wu (1), Dalmo A. Vieira (2), Abdul Khan (3) and Sam S. Y. Wang (4) (1), (2), (3) and (4), National Center for Computational Hydroscience and Engineering, School of Engineering, The University of Mississippi, MS 38677; PH (662) 915-5673 / (662) 915-7788; FAX (662) 915-7796; E-mail: AbstractThe CCHE1D model was designed to simulate long-term flow and sediment transport in channel networks to support the DEC project. It uses either the dynamic wave or the diffusive wave model to compute unsteady flows in channel networks with compound cross sections, taking into account the effects of in-stream hydraulic structures, such as culverts, weirs, drop structures, and bridge crossings. It simulates non-uniform sediment transport using a non-equilibrium approach, and calculates bank toe erosion and mass failure due to channel incision. The CCHE1D model decouples the flow and sediment transport calculations but couples the calculations of non-uniform sediment transport, bed changes and bed material sorting in order to enhance the numerical stability of the model. In this paper, the sensitivity of CCHE1D to parameters such as the non-equilibrium adaptation length of sediment transport and the mixing layer thickness is evaluated in cases of channel aggradation and degradation in laboratory flumes as well as in a natural channel network. In the case of channel degradation, the simulated scour process is not sensitive to variation in values of the non-equilibrium adaptation length, but the determination of the mixing layer thickness is important to the computations of the equilibrium scour depth and of the bed-material size distribution at the armoring layer. The simulated bed profiles in the case of channel aggradation and the calculated sediment yield in the case of natural channel network are insensitive to the prescription of both the non-equilibrium adaptation length and the mixing layer thickness. The CCHE1D model can provide reliable results even when these two parameters are given a wide range of values. Introduction The CCHE1D model was designed to simulate long-term flow and sediment transport in channel networks to support the Demonstration Erosion Control (DEC) project, which is an interagency cooperative effort among the US Army Corps of Engineers (COE), the Natural Resources Conservation Service (NRCS) and the Agricultural Research Service (ARS) of the US Department of Agriculture. The CCHE1D version 2.0 was based on the unsteady flow model DWAVNET (Diffusion WAVe model for channel NETworks, Langendeon, 1996) and the sediment transport model BEAMS (Bed and Bank Erosion Analysis Model for Streams, Li et al., 1996). It was significantly improved by implementing the dynamic wave model and the non-equilibrium sediment transport model (Wu, Vieira and Wang 2000). The CCHE1D was integrated with the landscape analysis tool TOPAZ (Garbrecht and Martz, 1995) and with the watershed models AGNPS (Bosch et al., 1998) and SWAT (Arnold et al., 1993), through an ArcView GIS-based graphical user interface (Vieira and Wu, 2000). The CCHE1D has been successfully tested in various experimental and field cases. Because several parameters in CCHE1D must be prescribed empirically, it is very important to know the response of the model to the uncertainty of these parameters. In this study, the sensitivity of CCHE1D to model parameters such as the non-equilibrium adaptation length of sediment transport and the mixing layer thickness is analyzed in cases of channel aggradation and degradation in laboratory flumes as well as in a natural channel network. Description of the CCHE1D Channel Network Model Hydrodynamic Model. The CCHE1D flow model simulates unsteady flow in channel networks with compound cross-sections using either the diffusive wave model or the dynamic wave model. The dynamic wave model solves the full St. Venant equations. The Preissmann implicit,four-point, finite difference scheme is used to discretize the governing equations. Linearized iteration schemes for the discretized governing equations are established and solved using a double sweep algorithm. The influence of hydraulic structures such as culverts, measuring flumes, bridge crossings and drop structures has been considered in the CCHE1D model. Stage-discharge relations for hydraulic structures are derived so that the hydraulic structures become an intrinsic part of the numerical algorithm. Sediment Transport Model. The CCHE1D model calculates non-uniform sediment transport in rivers using a non-equilibrium approach. The governing equation for the non-equilibrium transport of non-uniform total load is (1)where A is the flow area; Ctk is the depth-averaged total-load concentration of size class k; Qtk is the actual total-load transport rate; Qt*k is the total-load transport capacity; Ls is the adaptation length of non-equilibrium sediment transport; and qlk is the side sediment discharge from banks or tributaries per unit channel length, with the contribution from banks being simulated by CCHE1D bank erosion and bank failure module, and the contribution from upland erosion being simulated by SWAT or AGNPS. The sediment transport capacity can be written as a general form (2)where pbk is the bed material gradation; Q*tk is the potential sediment transport rate, which is determined with SEDTRA module (Garbrecht et al., 1995), Wu, Wang and Jias formula (2000), the modified Ackers and Whites 1973 formula (Proffitt and Sutherland, 1983), or the modified Engelund and Hansens 1967 formula (with Wu, Wang and Jias correction factor, 2000). The bed deformation due to size class k is determined with (3)where p is the bed material porosity, which is calculated with the methods of Komura and Simmons (1967), Han et al (1981), or is specified by the user according to available measurement data; Abk / t is the bed deformation rate of size class k. The bed material is divided into several layers. The variation of bed material gradation pbk at the mixing layer (surface layer) is determined by the following equation (Wu and Li, 1992) (4)where Am is the cross-sectional area of the mixing layer; Ab / t is the total bed deformation rate, defined as Ab/ t =k=1N Ab/ t; N is the total number of size classes; p*bk is pbk of the mixing layer when Am/ t Ab/t0 , and p*bk is the percentage of the kth size class of bed material in subsurface layer (under mixing layer) when Am/ t Ab/t 0. Eq. (1) is discretized using the Preissmann implicit scheme, with its source term being discretized by the same formulation as that for the right-hand term of Eq. (3) in order to satisfy the sediment continuity. Eq. (4) is discretized by a difference scheme that satisfies mass conservation. A coupled method for the calculations of sediment transport, bed change and bed material sorting is established by implicitly treating the pbk in Eq. (2) as pbkn+1and simultaneously solving the set of algebraic equations corresponding to Eqs. (1)-(4) by using the direct method proposed by Wu and Li (1992). This coupled method is more stable and can more easily eliminate the occurrence of the computed negative bed material gradation, when compared to the decoupled method, in which the pbk in Eq. (2) is treated explicitly. However, the aforementioned coupling procedure for sediment transport, bed change and bed material sorting computations is still decoupled from the flow calculation. Model Parameters to be Analyzed The parameters in numerical models of flow and sediment transport in rivers can be classified into two groups: numerical parameters and physical parameters. The numerical parameters result from the discretization and solution procedures, while the physical parameters represent the physical properties of flow and sediment, or the quantities derived from the modeling of flow and sediment transport. In the CCHE1D channel network model, the numerical parameters include computation time step and grid length, and the physical parameters are the Mannings roughness coefficient, non-equilibrium adaptation length of sediment transport, mixing layer thickness, bed material porosity, etc. Usually, the numerical parameters can be more easily handled than the physical parameters. Some of these physical parameters, such as the Mannings roughness coefficient and bed material porosity, have been studied by many investigators and may be determined by measurement. However, the non-equilibrium adaptation length and the mixing layer thickness are less understood and must be prescribed empirically. Therefore, the main concern in this paper is to analyze the influence of these two physical parameters on the simulation results. The non-equilibrium adaptation length Ls characterizes the distance for sediment to adjust from a non-equilibrium state to an equilibrium state. Wu, Rodi and Wenka (2000) and Wu and Vieira (2000) reviewed in detail those empirical and semi-empirical methods for determining Ls published in the literature, such as Bell and Sutherlands (1981), Armanini and di Silvios (1988), etc. It was found that those methods provide significantly different estimations of Ls. In CCHE1D, the adaptation length for wash load transport is set as infinitely large because the net exchange between wash load and channel bed is usually negligible. The adaptation length for suspended load transport is calculated with Ls=uh/s, in which u is the section-averaged velocity, h is the flow depth, s is the settling velocity of sediment particles, and is the adaptation coeficient which can be calculated with Armanini and di Silvios (1988) method, or specified as a constant value by the user. The adaptation length for bed load transport is suggested to set as the length of the dominant bed forms, such as 7.3h, the length of sand dunes (van Rijn, 1984), or 6.3B, the length of alternate sand bars in the channel (Yalin, 1972). Here, B is the average channel width. The mixing layer thickness is a key parameter in the determination of bed material gradation, which in turn influences the whole simulation. However, the evaluation methods for this parameter found in the literature are highly empirical. Physically, it is related to bed-form movements. Therefore, in CCHE1D, the mixing layer thickness is set as half the sand dune height, which is calculated with van Rijns (1984)formula.Case Studies Case A: Channel Degradation. The experiment of bed degradation and armoring processes performed by Ashida and Michiue (1971) was used to test the CCHE1D model in a previous study (Wu, Vieira and Wang, 2000), and it is here adopted to conduct the sensitivity analysis of the model. The experimental flume was 0.8m wide and 20m long. The flume bed was filled with non-uniform sediment with a median size of 1.5mm and a standard deviation of 3.47. In experimental run 6, the inlet flow discharge was 0.0314 m3/s, and the initial bed slope was 0.01. In this sensitivity analysis, only one parameters value is changed at a time, while all other parameters are kept the same as those used in the previous test. In order to examine the influence of Ls, several functions, such as Ls=7.3h, Ls=t and Ls=1+0.5t, have been used. Here, t is the time in hours. Figure 1 shows the comparison of the measured and calculated bed scour depths at 7m, 10m and 13m upstream from the weir at the end of the flume. The function Ls=7.3h provides the best result for the bed scour process, especially the time to reach equilibrium state. The results from Ls=t and Ls=1+0.5t are also very close to the measured data. It appears that the calculated scour depth is insensitive to Ls. It is also found that the calculated equilibrium bed material gradation at the armoring layer is insensitive to Ls. Figure 1. Sensitivity of the Calculated Bed Scour Depth to Ls in Ashida and Michues (1971) Run 6 In addition, the influence of mixing layer thickness on the calculated scour depth and bed material gradation is examined by changing the value of mixing layer thickness from the median size of parent mixture to twice that value. Figure 2 shows that the thicker the mixing layer, the larger the equilibrium scour depth. The time to reach the equilibrium scour depth and the equilibrium bed material gradation increases as the thickness of the mixing layer increases. The mixing layer thickness is important in the case of bed scour. Figure 2. Sensitivity of the Calculated Bed Scour Depth to Mixing Layer Thickness in Ashida and Michues (1971) Run 6Case B: Channel Aggradation. The channel aggradation experiments performed at the St. Anthony Falls Hydraulic Laboratory (SAFHL; see Seal et al., 1995) were used to test CCHE1D model (Wu, Vieira and Wang, 2000). Here, the experimental run 2 is used to conduct the sensitivity study. The experimental reach of the flume was 45m long and 0.305m wide, with an initial bed slope of 0.002. The tailgate was kept at a constant height that was high enough to produce an undular hydraulic jump at the downstream end of the main gravel deposit. The sediment fed at the flume entrance was a weakly bimodal mixture comprising a wide range of sizes, from 0.125mm to 64mm, which was transported mainly as bed load. Due to sediment overloading, an aggradational wedge developed. Its front gradually moved downstream while the upstream bed elevation continued to rise. In run 2 the water discharge was 0.049m3/s, the sediment feed rate was 5.65kg/min, and the tailgate water elevation was 0.45m. The influence of the adaptation length Ls on the calculated bed profile is analyzed by setting Ls as 0.5m, 2m and 7.3h. Here, h is set to the average flow depth over the wedge from the inlet to the gravel deposit front, and 7.3h equals to about 1m. As shown in Figure 3, Ls has little influence on the location, height and celerity of the gravel deposit front. It seems that Ls does not affect the top slope of wedge. The only noticeable influence of Ls is on the slope of the deposit front. The longer the adaptation length, the milder the slope of the deposit front. However this occurs over a limited distance, and the influence of L s on the calculated bed profile is limited. Figure 3. Sensitivity of the Calculated Bed Profile to Ls in SAFHLs (1995) Run 2 Figure 4 shows the calculated bed profiles with the mixing layer thickness being given values of d50, 6d50 and 0.5. The difference among the calculated bed profiles is very small. As the mixing layer thickness increases six times, the deposit front just moves downstream about 1.3%. The influence of the mixing layer thickness on the deposition case is much less than on the previous scouring case.Figure 4. Sensitivity of the Calculated Bed Profile to Mixing Layer Thicknessin SAFHLs (1995) Run 2 Case C: Goodwin Creek Watershed: Goodwin Creek in Panola County, Mississippi, is an experimental watershed for the DEC project. The drainage area above the watershed outlet is 21.3km2, and the average channel slope is about 0.004. Most of the channels in the watershed are ephemeral, with perennial flows occurring only in the lower reaches of the watershed. The runoff produced by storm events swiftly exits the watershed, and the discharge returns to base flow levels within one to three days. The sediments are transported in the channels as bed load and suspended load, and range from silt (0.062mm) to sand to gravel (65mm). Fourteen fully-instrumented flumes were constructed in the channels to control degradation of the channel bed and to monitor runoff and sediment yield. Figure 5 shows the channel network of Goodwin Creek extracted from a Digital Elevation Model using TOPAZ. Ten in-stream measuring flumes and four culverts, located in the channels of Strahler order two or higher, are considered. The simulation duration is 18 years, from January 1978 to December 1995, with a total of 1192 storm events. The time step used in the calculation is 15 minutes. The runoff and sediment yield from the upland fields generated with SWAT (Bingner et al., 1997) are used as the inflow conditions for the simulation of flow and sediment transport in the channel network. The model calibration done by Wu, Vieira and Wang (2000) showed that the CCHE1D model provided good predictions of the channel evolution and sediment yield in this watershed. Here, the sensitivity analysis of this model to the non-equilibrium adaptation length and the mixing layer thickness is conducted. Figure 5. Channel Network with Hydraulic Structures in Goodwin Creek At first, the non-equilibrium adaptation coefficient , which is used to calculate the non-equilibrium adaptation length for suspended load, is given values of 0.001, 0.01, 0.1 and 1.0, while other parameters are kept invariable. Here, the sediment transport capacity is calculated by Wu, Wang and Jias (2000) formula. Figure 6 shows the comparison of the calculated silt, sand and gravel yields using various . Table 1 provides the quantitative information about this comparison. As increases from 0.001 to 1.0, the gravel yield increases 27%, and the silt yield decreases 9%; the sand yield decreases first and then increases with a net increase of 8%; the total sediment yield slightly decreases 5% as a result. Figure 6. Sensitivity of the Calculated Sediment Yield to Table 1 Calculated Sediment Yield at Watershed Outlet Using Various Table 2 shows the calculated sediment yield at the watershed outlet using various non-equilibrium adaptation lengths for