You are on page 1of 18

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

Vol. 49, No. 1 AMERICAN WATER RESOURCES ASSOCIATION February 2013

COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER1

Junqiang Xia, Zhengbing Wang, Yanping Wang, and Xin Yu2

ABSTRACT: Significant channel adjustments often occur during flood seasons in the Lower Yellow River (LYR),
and it is a challenging work to accurately simulate the morphodynamic processes in the LYR using numerical
models. A comparison of two morphodynamic models (Delft3D and 2DLLCDM) for the LYR is presented herein
to identify critical improvements for these models. The concepts of these models are first compared with each
other. The models were then used to simulate the processes of flood routing, sediment transport, and morpholog-
ical changes occurring in a braided reach of the LYR. The differences were investigated between the simulated
results from these models and corresponding field measurements, and the results indicate that: (1) the hydrody-
namic processes calculated by both models agree closely with the measurements if an appropriate Manning’s
roughness coefficient is used; (2) the concentrations of suspended load at the downstream boundary calculated
by the models agree reasonably with the observed data; and (3) the predicted cross-sectional profiles obtained
from these models do not correspond well with the measurements. Based on these findings, the weak aspects of
the models are clarified, and three critical improvements are recommended, including: (1) the development of
roughness predictor; (2) the refinement of graded sediment transport capacity formulation; and (3) the consider-
ation of bank erosion module. These improvements need to be implemented in the future.

(KEY TERMS: morphodynamic model; Manning’s roughness coefficient; sediment transport capacity; fluvial pro-
cesses; Lower Yellow River.)

Xia, Junqiang, Zhengbing Wang, Yanping Wang, and Xin Yu, 2012. Comparison of Morphodynamic Models
for the Lower Yellow River. Journal of the American Water Resources Association (JAWRA) 49(1): 114-131.
DOI: 10.1111 ⁄ jawr.12002

INTRODUCTION world’s rivers in terms of sediment load (Li, 2003; Wu


et al., 2008a). In general, the Lower Yellow River
(LYR) is defined as the reach between Mengjin in
The Yellow River (see Figure 1), the second largest Henan and Lijin in Shandong, with a total length of
river in China, is well known for its high concentra- about 786 km. The LYR is usually divided further
tions of suspended load. At Sanmenxia, the mean into three geomorphologically distinct reaches. The
annual natural runoff of the Yellow River is normally upper reach, from Mengjin to Gaocun, has a typical
58 billion m3, and the mean annual suspended sedi- braided channel pattern. The lower reach, from
ment load is 1.6 billion tons, ranking it first of all the Taochengpu to Lijin, is a well-restricted and stable

1
Paper No. JAWRA-11-0126-P of the Journal of the American Water Resources Association (JAWRA). Received October 12, 2011; accepted
August 23, 2012. ª 2012 American Water Resources Association. Discussions are open until six months from print publication.
2
Respectively, Professor (Xia), State Key Laboratory of Water Resources and Hydropower Engineering Science, Wuhan University, No. 8,
Donghu South Road, Wuhan 430072, China; Professor (Wang), Delft University of Technology, Faculty of Civil Engineering and Geosciences,
2600 GA Delft, The Netherlands; and Senior Engineer (Wang) and Professor (Yu), Yellow River Institute of Hydraulic Research, Yellow River
Conservancy Commission, Zhengzhou 450003, China (E-Mail ⁄ Xia: xiajq@whu.edu.cn).

JAWRA 114 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION


COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER

Sanhuhekou 200 km
Yellow River Basin
Bayangaole Toudaoguai
Main stream of YR
Wanjiazhai
Tributaries of YR

Bohai Bay
Shizuishan

Lijing
Liujiaxia Qingtongxia
Wubao
Longyangxia Luokou
Xiaheyan
Aishan
Taochengpu
III
Longmen Gaocun

Lanzhou
Sunkou
Tangnaihei
Xiaolangdi
Jiahetan II
Huayuankou
Maqu
Tongguan
I
Jiwan

Reservoirs Sanmenxia

Hydrological stations I=Braided reach; II=Transitional reach; III=Meandering reach


Tuchengzi Section(TCZ)

FIGURE 1. Sketch of the Yellow River Basin.

meandering river reach. The river reach lying 1994; Wu et al., 2004; Xia et al., 2004). Zhang (1994)
between Gaocun and Taochengpu has a transitional proposed a steady 2D model in a body-fitted coordinate
channel pattern from braided to meandering. Heavy system to simulate the fluvial processes in a braided
soil erosion on the Loess Plateau upstream has led to reach of the LYR. Xia et al. (2004) integrated a depth-
intensive sedimentation in the LYR. According to the averaged 2D flow and sediment transport submodel
observed data, the total deposition volume in the with a cohesive bank erosion submodel to simulate the
LYR reached about 5.52 billion m3 during the period lateral and longitudinal channel deformations, which
from 1950 to 1999, of which 60% was deposited in the accurately reproduced the processes of channel widen-
braided reach (Xia et al., 2010), and severe sediment ing and bed deposition in a braided reach of the LYR
deposition caused the bed in the braided reach to (Wang et al., 2008). Wu et al. (2004) used a depth-aver-
aggrade by 2-4 m over this period because of a reduc- aged 2D model to calculate the process of flow and sedi-
tion in the water volume entering the LYR (Li, 2003). ment transport in the reach of the LYR between
One effect of heavy sedimentation in the LYR was an Huayuankou and Jiahetan. The simulated time series
obvious shrinkage of the main channel accompanied of water level, discharge, and sediment concentration
by a sharp decrease of the flood discharging capacity, were in reasonably good agreement with the measure-
which severely influenced the management of flood ments. At present, depth-averaged 2D models are more
control and made the phenomenon of ‘‘secondary often adopted in practice when simulating the morpho-
perched river’’ more serious (Wu et al., 2008a; Xia logical changes in the LYR due to their relatively easy
et al., 2010). Therefore, river engineers and scientists implementation and applications, although the process
in China adopted various methods to study the pro- of bed evolution in a braided reach is very complex due
cesses of flood routing, sediment transport, and mor- to there being numerous bars and complicated
phological changes in the LYR, and these methods branches appearing at low water stages, frequent
usually comprised physical river modeling and math- shifting of the main channel, and significant channel
ematical river modeling (Zhang and Xie, 1993). adjustment during high discharges. Therefore, the
Physical river models were used widely in the 1980s application of 2D morphodynamic models to the LYR is
and the 1990s to investigate the fluvial processes, and a challenging task at the present stage. Cao and Car-
they played an important role in solving the sediment ling (2002) presented a general review on the current
problem of the LYR. With the rapid development of status of mathematical modeling of alluvial rivers, and
computer performance and numerical methods for pointed out that the current mathematical models for
nonlinear analysis during the last four decades, math- morphodynamic processes are far from being mature,
ematical river models have become more popular due and some key issues need to be investigated further,
to their low cost, flexibility in evaluating different including the formulae of sediment transport capacity
effects of river engineering works, and capability of and bed roughness. Therefore, it is important to iden-
delivering detailed predictions. Two-dimensional (2D) tify what kind of model improvement is most relevant
models capable of simulating the processes of flow and and most urgently required in the modes for the LYR,
sediment transport as well as morphological changes which can be further referred to other river morphody-
have been developed since the end of the 1990s (Zhang, namic models.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 115 JAWRA


XIA, WANG, WANG, AND YU

In this study, an intercomparison of two morphody- not only predicting the processes of flood routing and
namic models was conducted by considering both the longitudinal bed deformation in natural rivers such as
discrepancies between the model concepts and the dif- the LYR, but also simulating the process of lateral bed
ferences in the simulated results and corresponding deformation, especially the process of lateral erosion
observed data, with some experience being then and failure of cohesive, noncohesive, or composite river-
obtained from this intercomparison. The common banks (Xia et al., 2004; Wang et al., 2008).
shortcomings of both these models considered herein
were presented, including the inaccuracy of input data
concerned both the initial bathymetry and initial bed- Hydrodynamic Module
material composition. Furthermore, the most critical
model improvements for simulating the morphody- In this section, the governing equations for hydro-
namic processes in the LYR were identified in this dynamics in curvilinear coordinates are presented
study as our primary objective, covering the bed rough- first. In addition, treatments of key parameters
ness predictor, the formulation of graded sediment related to the equations are also given, covering the
transport capacity, and the module of bank erosion. effect of sediment on the hydrodynamics and the for-
mulation of bed roughness coefficient.

Governing Equations for Hydrodynam-


COMPARISON OF MODEL CONCEPTS ics. Both models in common use the 2D shallow
water equations (Lesser et al., 2004; Wang et al.,
2008). The depth-averaged continuity equation for
In this section, general information is first flow in curvilinear coordinates can be written as
described for these two morphodynamic models con-
sidered herein. Governing equations with the treat- @Z 1 @ 1 @
ments of key parameters and numerical solution þ ðhUCg Þ þ ðhVCn Þ ¼ 0 ð1Þ
@t Cn Cg @n Cn Cg @g
methods are also presented for each model.
The momentum equations for flow in the n and g
directions are given respectively by
General Information

Two morphodynamic models for the LYR are @U U @U V @U UV @Cn V 2 @Cg


þ þ þ 
included in the comparison: Delft3D (Delft Hydrau- @t Cn @n Cg @g Cn Cg @g Cn Cg @n
lics, 2003) and 2DLLCDM (2D longitudinal and lat- pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2
g @Z 2 U þV ð2Þ
eral channel deformation model) (Xia et al., 2004). ¼  gn U
Cn @n h4=3
Delft3D is a generic commercial software package,  2 
whereas another model is specially developed for the mt @ U @2U
þ þ 2
LYR. Each of the models can be considered as con- Cn Cg @n2 @g
sisting of four modules: (1) a hydrodynamic module;
(2) a sediment transport module; (3) a module for
updating the morphology; and (4) a module for updat- and
ing the bed-material composition.
Delft3D (Delft Hydraulics, 2003; Lesser et al., @V U @V V @V UV @Cg U 2 @Cn
þ þ þ 
2004) was developed by WL|Delft Hydraulics in the @t Cn @n Cg @g Cn Cg @Cn Cn Cg @g
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Netherlands. It is a model suite that consists of a 2 2
g @Z 2 U þV ð3Þ
number of integrated modules that together allow the ¼  gn 4=3
V
Cg @g h
simulation of 2D or 3D hydrodynamics, and the com-  2 
putation of sediment transport and morphological mt @ V @2V
þ þ ;
changes in various coordinate systems. During this Cn Cg @n2 @g2
study, only the 2D-FS mode of Delft3D was used; it
comprises a 2D flow and online-sediment module in where n and g are orthogonal curvilinear coordinates
orthogonal curvilinear coordinates. in the horizontal directions; Z is the water level; h is
2DLLCDM, developed by Xia et al. (2004), is a depth- the water depth; U and V are velocity components in
averaged 2D model for the full channel adjustment, the n and g directions, respectively; Cn and Cg are
which consists of a submodel of 2D flow and sediment Lamé coefficients; and mt (=/u*h) is the horizontal
transport in orthogonal curvilinear coordinates and a turbulent viscosity coefficient for flow, where / is the
submodel of riverbank erosion. The model is capable of empirical parameter ranging from 0.0 to 1.0, u* is the

JAWRA 116 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION


COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER

pffiffiffi
friction velocity calculated by u ¼ nu g=h1=6 , in
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi of a discharge, the coefficient gradually decreased
which u ¼ U 2 þ V 2 , g is the gravitational accelera- and approached a minimum value of 0.007 to 0.010.
tion and n is the Manning’s roughness coefficient, Chien and Wan (1999) pointed out that the variation
and t is the time. Equation (1) represents mass con- in Manning’s roughness coefficient is closely related
servation for the water volume. In Equation (2) or to the development of bed form in the LYR.
(3), the first term on the left-hand side represents the In Delft3D, the roughness coefficient is specified as
local variation in momentum with time, and the other a constant, or as a space-varying coefficient of Man-
terms represent the advective transport of momen- ning, White-Colebrook or Chezy, or computed with a
tum. The terms on the right-hand side in Equation roughness predictor proposed by Van Rijn (2007a).
(2) or (3) represent water-level gradient, friction loss, Van Rijn’s roughness predictor determines the
and diffusive turbulent momentum transfer associ- parameter of bed form from the hydraulic and sedi-
ated with the Reynolds’s stresses, respectively. mentary conditions such as flow velocity, water
Equations (1-3) only represent the classical clear- depth, and grain size, and the parameter of bed form
water governing equations, which usually need to be is then used to compute the roughness coefficient,
solved using a decoupled approach. Therefore, the which is used for computations of both sediment
influences of sediment transport and bed evolution on transport and flow motion. However, this roughness
the hydrodynamic processes are not directly taken predictor has not been extensively validated in rivers
into consideration in the governing equations in the with high sediment concentrations such as the LYR,
case of the decoupled solution (Cao and Carling, for which the morphology in this reach is poorly
2002; Wu, 2008). Furthermore, the parameter of understood. Therefore, only constant roughness val-
roughness coefficient in Equations (2-3) is a very ues are used in the present study for Delft3D.
important one, and the selected value of Manning’s 2DLLCDM uses the formula proposed by Zhang
roughness coefficient has a significant effect on the et al. (2001) to calculate Manning’s roughness coeffi-
variations in water level and flood propagation speed. cient, and this formula can take into consideration
Therefore, the selection of roughness coefficient needs the effect of hydraulic and sedimentary factors on the
to be treated specially. These two issues related to value of Manning’s roughness coefficient. It can be
Equations (1-3) are discussed separately as follows. expressed by

Effect of Sediment on the Hydrodynamic Pro-  1=6  


cesses. The sediment concentrations in the LYR are h d
n ¼cn pffiffiffi
sometimes extremely high, especially during flood g h
8 "   #5 91
seasons, so that the density of the water-sediment <  0:77  
mixture varies temporally and spatially, which in d d d 0:2 =
0:49 þ 1 3p=8 sin ;
turn influences the turbulent structure of the flow. In : h h h ;
the 3D mode of Delft3D, all these effects can be taken
into account, whereas only the density variation is ð4Þ
considered in the 2D-FS mode of Delft3D. 2DLLCDM
cannot directly account for the effect of sediment con- where cn = 0.375j and d ð¼ Db50 Þ is the roughness
centrations in the flow governing equations. Such a thickness related to the flow intensity and the grain
treatment is regarded to be reasonable without the size of bed material, in which b¼1
0:5 3
occurrence of hyperconcentrated sediment-laden flows þ108:113Fr ð1Fr Þ , Fr is the Froude number of flow;
or rapid bed evolution (Cao and Carling, 2002; Wu, and j is the Von Kármán coefficient
pffiffiffiffiffiffi for turbid water,
2008). Therefore, this treatment is applicable in the calculated by j ¼ 0:4  1:68 Sv ð0:365  Sv Þ, where Sv
current study because the sediment concentrations to is the sediment concentration by volume. This empiri-
be simulated were relatively low, with a maximum cal formula is valid for Fr <0.80. The validity of this
value of <20 kg ⁄ m3. However, the calculation of Man- formula was verified using ample laboratory and field
ning’s roughness coefficient in 2DLLCDM accounts data, and it has been used in other numerical models
for the effect of sediment concentrations, which indi- for the LYR (Zhang et al., 2001; Ni et al., 2004; Wang
rectly considers the effect of sediment concentrations et al., 2008).
on the hydrodynamic processes.

Formulation of Bed Roughness Coefficient. In Sediment Transport Module


the LYR, the magnitude of Manning’s roughness coef-
ficient was found to be quite large during low dis- In the LYR, the change in morphology is mainly
charges, with a highest measured value of more than caused by the nonequilibrium transport of suspended
0.045 (Chien and Wan, 1999). Following the increase load, which is closely related to the calculation

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 117 JAWRA


XIA, WANG, WANG, AND YU

method of the source term, and the formulation of niades, 1965). In 2DLLCDM, ask is determined by the
sediment transport capacity in the governing equa- method proposed by Zhang et al. (2001), and its
tions for sediment transport. detailed expression can also be found in Ni et al.
(2004) and Wang et al. (2008). Furthermore, all the
Governing Equations for Sediment Trans- fractions of suspended load are regarded as noncohe-
port. In the calculation of sediment transport, sive sediment in 2DLLCDM.
Delft3D always includes the transport of both bed
load and suspended load. 2DLLCDM specially devel- Formulation of Sediment Transport Capac-
oped for the LYR, does not include the simulation of ity. Delft3D uses formulas that calculate either the
bed load transport because the transport of sus- total sediment transport or the bed load and sus-
pended load in this river is fairly dominant. Accord- pended load separately. For the suspended load, a
ing to the analysis of observed data by Long and depth-averaged sediment concentration in equilib-
Zhang (2002), the average ratio of bed load to total rium (S*) can be calculated by the formula proposed
load of sediment in the LYR is only about 0.5%. by Van Rijn (1984a), which can be expressed as fol-
Therefore, the transport of bed load is often ignored lows:
and only the transport of suspended load is consid-
ered in this model. The advection-diffusion equation
for the fractional suspended load used in both models S ¼ 0:015qs FD50 T 1:5 =ðaD0:3
 Þ; ð7Þ
can be given as
where qs is the density of sediment, F the shape coef-
  ficient, D50 the median diameter of bed material, T
@ 1 @   @
ðhSk Þ þ Cg UhSk þ ðCn VhSk Þ the transport stage parameter, D* the dimensionless
@t Cn Cg @n @g
   
particle size, and a the reference height. In this
h @ Cg @ @ Cn @ ;
¼ es ðSk Þ þ es ðSk Þ study, the enhanced sediment transport formula of
Cn Cg @n Cn @n @g Cg @g Equation (7) is applied in Delft3D (Van Rijn,
þ ðEk  Dk Þ 2007a,b). This formula can calculate the sediment
ð5Þ transport capacity of the total load at the same time.
The bed load transport formula for steady flow, pro-
posed by Van Rijn (1984b), was modified by introduc-
where es is the horizontal turbulent diffusivity coeffi- ing the instantaneous bed-shear stress. The enhanced
cient for sediment; Sk is the sediment concentration sediment formula has been developed for sediments
for the kth fraction; and the source term is repre- with a grain size of 0.008 mm and coarser, for con-
sented by the rates of bed erosion (Ek) and deposition centrations up to 150 kg ⁄ m3, and water depths
(Dk), respectively. These models adopt different exceeding 1.0 m. However, the sediment transport
expressions for the terms of Ek and Dk. capacity computed by the enhanced formula is under-
estimated when the flow velocity is <0.6 m ⁄ s and the
Calculation of the Source Term. These two median diameter of suspended load is <0.1 mm (Wu
models apply a similar equation for the calculation of et al., 2008b).
the source term, given by: In 2DLLCDM, the sediment transport capacity for-
mula proposed by Zhang and Zhang (1992) is used,
Dk  Ek ¼ ask xsk ðSk  Sk Þ; ð6Þ which can be written as

"  #0:62
where xsk represents the effective settling velocity for ð0:0022 þ Sv Þu3 h
S ¼ 2:5 ln ; ð8Þ
the kth fraction; and S*k and ask are the sediment j cs ccm ghxm 6D50
m
transport capacity and coefficient of saturation recov-
ery or adjustment for the kth fraction.
In Delft3D, a distinction is made between noncohe- where S* is the sediment transport capacity; cs and
sive and cohesive fractions. For noncohesive fractions, cm denote the specific densities Pof sediment and tur-
Equation (6) is also used to calculate the source term, bid water, respectively; xm ¼ N k¼1 DPk xsk is the
but the coefficient ask is calculated according to the group settling velocity of nonuniform sediment,
formulation of Galappatti and Vreugdenhil (1985), or where DP*k is the percentage of the sediment trans-
Wang (1992). For cohesive fractions, the Krone- port capacity for the kth fraction; and D50 is the med-
Partheniades formulations are used to determine the ian diameter of bed material. SI units are used in
rates of erosion and deposition (Krone, 1962; Parthe- this formula. The accuracy of this formula was

JAWRA 118 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION


COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER

validated using a series of field and laboratory data, in Delft3D underpredicts the sediment transport
which confirmed its reasonable predictive accuracy. capacity if the concentration is <10 kg ⁄ m3, although
Therefore, this formula has been widely used in the it generally overestimates the capacity at high con-
computation of sediment transport in several numeri- centrations; and (2) Equation (8) in 2DLLCDM per-
cal models for the LYR (Zhang et al., 2001; Ni et al., forms well because it was calibrated by the data from
2004; Wang et al., 2008). the Yellow River. A statistical method of the mean
Figures 2a and 2b compare the results calculated normalized error (MNE) is usually used to evaluate
by the above sediment transport formulas with the the goodness-of-fit between the computed and mea-
observed data. These observed data were mainly from sured results. For a perfect fit, the value of MNE
the measurements at hydrometric stations in the approaches zero. The obtained values of MNE for
LYR (Wu et al., 2008b), with the flow velocities rang- Van Rijn’s formula and Equation (8) are 0.557 and
ing from 0.3 to 3.1 m ⁄ s, and with the water depths 0.478, respectively. Therefore, the predictive accuracy
ranging from 0.6 to 11.3 m. It can be seen from Fig- of these formulas is limited partly due to the compli-
ure 2 that: (1) Van Rijn’s formula (Van Rijn, 2007b) cated sediment transport in this reach.

3
10
HYK (Huayuankou) Module for Updating the Morphology
JHT (Jiahetan)
GC
(Gaocun)
10
2 SK
(Aishan)
Channel adjustments can occur in two spatial
Computed Concentration (kg/m )

AS
directions in the LYR. One is longitudinal channel
3

LK (Luokou)
LJ (Lijing) adjustment, characterized by morphological changes
1 TCZ (Tuchengzi) such as bed scour or sediment deposition on the bed,
10
and the other is lateral channel adjustment, distin-
guished by river width adjustments, usually involving
10
0
the process of bank erosion (Xia et al., 2004). In
updating the morphology in the LYR, the rate of
longitudinal channel adjustment can be determined
-1
10 using the bed deformation equation due to the non-
equilibrium transport of suspended load, whereas the
(a) lateral channel adjustment requires the simulation of
-2
10 bank erosion process.
-2 -1 0 1 2 3
10 10 10 10 10 10
3
Measured Concentration (kg/m ) Updating Bed Level. 2DLLCDM only accounts
for the transport process of suspended load. There-
1.0E+3 fore, the change in bed levels can be calculated by:

HYK XN   X N
JHT
@Zb
0 0 DZbk
q ¼q ¼ ðDk  Ek Þ; ð9Þ
1.0E+2 GC @t k¼1
Dt k¼1
Computed Concentration (kg/m3)

SK
AS
LK
LJ
where Zb is the bed level, q¢ the dry density of bed
1.0E+1 TCZ material, DZbk the thickness of bed evolution for the
kth fraction, N the total number of fractions used to
represent the gradation of nonuniform sediment, and
Dt the time step. Delft3D still includes bed load
1.0E+0 transport, and therefore the change in bed levels is
extended with the divergence of the bed load trans-
(b) port.
1.0E-1
1.0E-1 1.0E+0 1.0E+1 1.0E+2 1.0E+3
Simulating Bank Erosion. In the braided reach
3
Measured Concentration (kg/m ) of the LYR, bank erosion often occurs during the per-
iod of clear water scouring, due to water impound-
ment and sediment detention of the Xiaolangdi
FIGURE 2. Comparison Between the Calculated and
Observed Concentrations: (a) Delft3D (Van Rijn’s formula) and Reservoir. Analysis of the observed data during the
(b) 2DLLCDM (Equation 8). period from 1999 to 2005 showed that the amount of

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 119 JAWRA


XIA, WANG, WANG, AND YU

scouring in the braided reach was 0.594 billion m3, distribution of bed material in the surface layer can
which occupied 77% of the total scouring in the LYR be classified into two cases of bed scour and bed depo-
over the period. It also indicated that the sediment sition, with a detailed description of the procedure
quantity from bank erosion accounted for 30-50% of being given in Wang et al. (2008). Delft3D uses a sim-
the total sediment amount scoured in the braided ilar procedure as presented in 2DLLCDM. However,
reach (Xia et al., 2007). Delft3D and 2DLLCDM adopt the former usually accounts for the adjustment of
different methods to account for the process of bank bed-material composition in the mixing layer.
erosion.
Delft3D uses a factor of dry cell erosion to simulate
the process of bank erosion (Roelvink et al., 2003). In Numerical Solution Methods
the case of erosion near a dry bank, the standard
scheme will not allow erosion of the adjacent cells, In Delft3D, a model domain is usually covered by a
even when a steep scour hole would develop right set of curvilinear mesh, and the variables are
next to the bank. Therefore, a scheme has been arranged in a pattern called a staggered grid to dis-
implemented, where for each wet cell the erosion vol- cretize the 2D shallow water equations in space. In
ume is distributed over the wet cell and the adjacent this arrangement, a water-level point is defined in
dry cells if there are dry points adjacent to it. The the center of a cell and the velocity components are
distribution is governed by a user-specified parame- located on the grid cell faces normal to their orienta-
ter, which determines the fraction of the erosion to be tions. In this model, an alternating direction implicit
assigned to the adjacent dry cells. If this factor equals (ADI) method is used to solve the continuity and
0, the standard scheme is used; if this factor equals momentum equations (Leendertse, 1970). The advan-
1, all erosion that would occur in the wet cell is tage of the ADI method is that the implicitly inte-
assigned to the adjacent dry cells. grated water levels and velocities are coupled along
2DLLCDM simulates the process of bank erosion grid lines, leading to the discretized equations with a
using the methods of near-bank hydrodynamics and small band width. Stelling (1984) extended the ADI
soil mechanics (Darby, 1998; Mosselman, 1998; Xia method of Leendertse (1970) with a special approach
et al., 2004; Wang et al., 2008). For cohesive river- for the horizontal advection terms, and the corre-
banks, this simulation is based on the bank erosion sponding scheme is denoted as a cyclic method of
model proposed by Osman and Thorne (1988), and the ADI (Stelling and Leendertse, 1991). This leads to a
corresponding computational procedure includes two method that is computationally efficient, at least sec-
steps: (1) computation of lateral erosion distance; and ond-order accurate, and stable at Courant numbers of
(2) analysis of bank slope stability. The eroded mate- up to approximately 10. The sediment transport
rial from the bank is divided into two parts: suspended equation is formulated in a conservative form (finite
load and bed material. The former is the lateral input volume method) and is solved using the cyclic
term for the transport equation of suspended load, and method. In addition, a horizontal Forester filter (For-
the latter is used to determine the deposition thickness ester, 1979) based on the diffusion along the horizon-
in the near-bank zone. The detailed procedure for the tal direction is applied to remove any negative
enhanced bank erosion submodel can be found in Xia concentration values that may occur. The Forester fil-
et al. (2004) or Wang et al. (2008). ter is mass conserving and does not cause significant
amplitude losses in sharply peaked solutions.
2DLLCDM uses a three-step solution procedure
Module for Updating the Bed-Material Composition (Wang et al., 2008). First, the flow governing equations
in curvilinear coordinates are split into two sets of
In order to simulate the phenomenon of armoring equations in the longitudinal and lateral directions
or sorting of bed material during degradation or using the method of fractional steps (Yanenko, 1971).
aggradation, a procedure of updating the bed-mate- The ‘‘time marching’’ ADI scheme is employed to solve
rial composition is often applied. In 2DLLCDM, the the two sets of discretized equations on a staggered
bed material at each computational cell is divided grid (Leendertse, 1970). Secondly, a method of frac-
into two vertical layers: the upper one is called the tional steps in space and a hybrid scheme of explicit-
mixing or active layer and the lower one is called the implicit discretization are employed to solve the 2D
memory layer. The thickness of the mixing layer is transport equation of suspended load, and more details
denoted by Hb, with its gradation being represented of this method can be referred to Xia et al. (2004) or
by DPbk. The memory layer is further divided into m Wang et al. (2008). Thirdly, the bed level at each node
smaller sublayers, with the thickness and gradation by the end of the time level can be updated with the
of each sublayer being represented by DHm and DPmk, explicit scheme, and the calculation of updating the
respectively. The adjustment procedure of the size bed-material composition is then conducted.

JAWRA 120 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION


COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER

experiments of regulating water and sediment have


Summary of the Model Capabilities
been conducted annually since 2002. These experi-
ments usually adopt a scheme of a joint operation,
According to the aforesaid description, the main
with the water from other reservoirs along the Yellow
characteristics of the two models are summarized in
River being first transferred to the Xiaolangdi Reser-
Table 1, which clearly indicates the difference in the
voir. The water is then released from the Xiaolangdi
model capabilities, including calculations of bed
Reservoir with relatively large discharges (i.e., man-
roughness and sediment transport capacity, treat-
made flood peaks) to maintain a certain level of sedi-
ment of the source term in sediment transport, and
ment concentration. The man-made floods first route
calculations of updating the morphology and bed-
along the braided reach, and then propagate along
material composition. Furthermore, it should be
the transitional and meandering reaches of the LYR.
pointed out herein that the current governing equa-
The channel in the braided reach is scoured to a cer-
tions for flow and sediment transport are based on
tain extent due to low sediment concentrations and
the diffusion model using the assumption of a kind of
large discharges released from the Xiaolangdi Reser-
continuous medium (Wu, 2008), instead of the two-
voir, and the corresponding flood discharging capacity
phase model for sediment-laden flows.
in the main channel is estimated to increase signifi-
cantly in this reach. During these experiments,
detailed hydrological data are collected at many
MODEL SET-UP locations along the river. These measurements have
provided the necessary information required for
conducting the intercomparison of these models.
According to the flood defense strategy of the Chi- In the current study, the above models were set up to
nese government, the water levels in the reservoirs simulate a water and sediment regulation experiment
along the Yellow River need to be lowered to specified in a braided reach conducted before the 2004 flood sea-
pool levels before the flood season each year. A large son. An intercomparison of the model-predicted mor-
volume of storage water in the reservoirs needs to be phodynamic processes was undertaken. In addition,
released during this period. In order to optimize the these model predictions were also compared with the
use of the storage water, a series of large-scale exper- available field data. The study reach consisted of 15 con-
iments of a joint operation between several associated secutive river bends with different curvatures. There-
reservoirs have been undertaken in recent years. fore, it was also characterized as a meandering reach to
Xiaolangdi Reservoir has played a key role in this a certain extent, although it was part of a braided reach.
operation. This reservoir has a total storage capacity This geometry was caused by severe sedimentation and
of 12.65 billion m3 and is located at the exit of the shrinking of the main channel in the LYR since 1987.
last gorge in the middle reach of the Yellow River, Before the flood season in 2004, the mean width of the
130 km downstream of the Sanmenxia Dam (see Fig- main channel in the study reach was 780 m, with a min-
ure 1). The Xiaolangdi Reservoir was built in the late imum value of 300 m, and the mean longitudinal slope
1990s as a key management measure for flood and of the main channel was about 1.5 ⁄ 10,000. It should be
sediment control, irrigation and water supply, as well noted that all the hydrological data used in the models
as power generation. The reservoir started to were measured by the Bureau of Hydrology, Yellow
impound water in October 1999, and large-scale River Conservancy Commission (YRCC), including the

TABLE 1. Summary of the Model Capabilities.

Different Models

Model Capabilities Delft3D 2DLLCDM

Bed roughness Constant roughness Roughness predictor (Zhang et al., 2001)


Sediment transport capacity Sediment transport formula by Van Rijn Suspended load transport formula by Zhang
(2007a,b) and Zhang (1992)
Cohesive sediment and noncohesive sediment Only noncohesive sediment is considered
are considered
Source term in Equation (5) Galappatti and Vreugdenhil (1985) for Zhang et al. (2001)
noncohesive sediment
Partheniades (1965) for cohesive sediment
Morphological updating Factor of dry cell erosion (Roelvink et al., 2003) Bank erosion module based on soil mechanics
(Xia et al., 2004)
Bed-material composition updating Only updating in the mixing layer Updating in the mixing layer and memory layer

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 121 JAWRA


XIA, WANG, WANG, AND YU

measurements of water levels, discharges, concentra- at Jiahetan, and the observed hydrograph of water
tions, cross-sectional profiles, and so on. A list of all the level at Gaocun. Due to the operation of the Xiao-
hydrological data is shown in Table 2. langdi Reservoir, the released water consisted of two
phases with the flood peak discharges of 2,780 and
2,840 m3 ⁄ s, respectively, and with the corresponding
Computational Mesh and Initial Bathymetry peak sediment concentrations of 11.0 and 12.1 kg ⁄ m3.
It can also be found that the discharge hydrograph at
The study domain covered the braided reach Jiahetan and the water-level hydrograph at Gaocun
between Jiahetan and Gaocun in the LYR. It had a were well correlated.
length of about 87 km with 46 observed cross-sectional Figure 5 shows a mean size distribution of the sus-
profiles. The main channel and a part of the low flood- pended load at Jiahetan over the study period, and
plains were covered by the study domain. Because the mean median diameter at this section was
large-scale overbank flows did not occur during this 0.031 mm. The gradation of incoming suspended load
experiment, it was not necessary to cover the entire was used to determine the fractional sediment dis-
floodplains. The study domain was divided into charges at the upstream boundary. The study period
472 · 10 curvilinear cells with grid sizes varying was 600 h, from June 19 to July 14, 2004, and the
between 22 and 555 m. A part of computational mesh computational time step was 6 s. As compared with
in a local region is shown in Figure 3. The bed eleva- the water volume entering this reach, the volume of
tion of each cell was obtained from the observed cross- water diversion during the period was negligible in
sectional geometry by the technique of smooth interpo- this reach, and therefore the process of water diver-
lating in the separate regions of main channel and low sion was not accounted for.
floodplains (see Figure 3). The above computational
mesh was used by both Delft3D and 2DLLCDM.
Initial Bed-Material Composition

Boundary Conditions The data for the bed-material composition in this


reach observed in April 2004 were available at only
The upstream model boundary was located at the eight cross-sections, so the size distribution of bed
Jiahetan gauging station, 220 km downstream of the material at each node was obtained by linear interpo-
Xiaolangdi Dam. The downstream model boundary lation of these known bed-material gradations at the
was located at the Gaocun gauging station, 307 km sections. Figure 5 also indicates a mean size distribu-
downstream of the dam. Figure 4 shows the observed tion of the bed material in this reach at the initial
time series of discharge and sediment concentration time. It can be seen from Figure 5 that the size distri-
butions for both the suspended load and bed material
were nonuniform, with the particle sizes ranging
TABLE 2. List of All the Hydrological Data.
from 0.002 to 0.500 mm. In this study, five fractions
Figures Data Name Sources or References were used to represent the nonuniform sediment mix-
ture, with the corresponding representative diame-
2a Sediment transport data Wu et al. (2008b) ters of 8, 23.50, 46.50, 93.50, and 250.0 lm. At the
3 Initial bathymetry Interpolation from the
initial time, the mean median diameter of the bed
measured
cross-sectional profiles material in this reach was 0.056 mm, which was
4 Hydrographs of discharge YRCC coarser than that of the suspended load at Jiahetan.
and suspended load In Delft3D, the first fraction was treated as cohesive
concentration at sediment, whereas the other fractions were treated
Jiahetan
as noncohesive sediment. In 2DLLCDM, all the frac-
Water-level hydrograph
at Gaocun tions were regarded as noncohesive sediment.
5 Gradations of inflow
suspended load and bed
material in the reach
6-7 Water-level hydrographs
ANALYSIS OF MODEL RESULTS
at different sites
8 Discharge hydrograph at
Gaocun
9 Hydrograph of suspended Before application, numerical models are usually
load concentration at Gaocun first calibrated and subsequently verified with inde-
10-11 Measured cross-sectional
pendent datasets. The most common calibration
profiles
parameter for hydrodynamics is the bed roughness

JAWRA 122 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION


COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER

18
Jiahetan
Bed elevation (Unit:m)
15 Location of hydrometric station
59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 Location of water gauging station
12
Y(km) CS05 Local mesh Location of observed cross-section
CS30
9
5
CS15 CS3
6
CS40 ng

Gaocun
CS25 ua
3 CS10 CS20
g zh CS45
Qin
Dongbatou1

0
0 6 12 18 24 30 36 42 48 54 60 66 72
X(km)

FIGURE 3. Initial Bathymetry in the Study Region with the Location of Each Section.

3000 16 63.2

2700 63
14

2400 62.8
12

Concentration S (kg/m3)
Discharge Q (m3/s)

62.6
2100

Water level Z (m)


10
62.4
1800
8 62.2
1500
62
6
1200
61.8
Q 4
900 61.6
S 2
600 61.4
Z
300 0 61.2
0 60 120 180 240 300 360 420 480 540 600
June 19 July 13
Time (hours)

FIGURE 4. Boundary Conditions Used in the Models.

100
of the two models in the LYR can be referred to in
80 Wang et al. (2008) and Van Maren et al. (2011).
% by weight finer than

Therefore, the effect of different parameter settings


60
should be investigated more as a sensitivity analysis
40
rather than placed within a calibration ⁄ verifica-
Suspended load
tion ⁄ application sequence. The hydrodynamic pro-
20 Bed material cesses are also strongly coupled with the morpho-
dynamic processes in the LYR, and therefore, both
0 the hydrodynamic and morphologic results simulated
0.001 0.010 D(mm) 0.100 1.000
by both models are presented in detail.
A small bed roughness coefficient produces higher
FIGURE 5. Size Distributions for the Bed Material and flow velocities and lower water levels, and conse-
Suspended Load. quently increases the propagation speed of flood
waves (Julien, 2002). The sediment transport capac-
coefficient, whereas the calibration parameters for ity is modified indirectly by the roughness coefficient
morphology are numerous. In the present study, a through the water depth and flow velocity, but also
different approach was followed because numerous directly in Delft3D because the roughness value is an
datasets with high accuracy were not readily avail- input parameter for the sediment transport formula-
able for the LYR, but more importantly, the calibra- tion. In this section, the effects of different bed rough-
tion of the parameters themselves was a part of the ness coefficients were investigated on water levels at
analysis. A key step of this study was to preliminarily water gauge stations, discharges and concentrations
identify the effect of different model concepts on the at the downstream boundary, and bed evolution at
corresponding model results, based on different bed typical sections. Delft3D and 2DLLCDM usually had
roughness coefficients. Detailed validation processes to be run for a sufficiently long time or a spin-up

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 123 JAWRA


XIA, WANG, WANG, AND YU

period without bathymetric updating to achieve a 78.0


(a)
steady state under the given inflow discharge and sus- 77.5
pended load concentration at the upstream boundary,
and the given water level at the downstream bound- 77.0

Water level (m)


ary. These steady state results were then treated as 76.5
the initial conditions of flow and sediment transport.
76.0

75.5
Effect of Bed Roughness on Water Level 75.0

Figures 6 and 7 compare the water-level hydro- 74.5

graphs at the water gauge stations of Jiahetan, Don- 74.0


gbatou1, and Qingzhuang calculated using different 0 60 120 180 240 300 360 420 480 540 600
Manning’s roughness coefficients with the corre- Time (hours)
sponding measurements. 75.0
In the case of Delft3D (Figure 6), a Manning’s (b)
74.5
roughness coefficient of 0.012 provides the best agree-
ment between the simulated and measured water lev- 74.0
els at Jiahetan. However, at Dongbatou1, the

Water level (m)


73.5
calculated water levels are underestimated if Man-
73.0
ning’s roughness coefficients smaller than 0.015 are
used, whereas they agree closely with the observed 72.5
data for a value of n = 0.018. At Qingzhuang,
72.0
although the roughness coefficient does not greatly
influence the calculated water levels because the cali- 71.5
bration location is very close to the downstream 71.0
boundary, a value of n = 0.018 gives reasonable 0 60 120 180 240 300 360 420 480 540 600
agreement between the model predictions and mea- Time (hours)
surements. The sensitivity of water levels to the 65.0
selected roughness value varies at different sections. (c)
64.5
The mean water level at Jiahetan or Dongbatou1,
over the study period, was predicted to increase by 64.0
about 0.50 m as the value of Manning’s roughness
Water level(m)

63.5
coefficient varied from 0.012 to 0.018, whereas it was
predicted to rise by only 0.1 m at Qingzhuang for the 63.0

same change in the roughness value. Therefore, vari- 62.5


ous roughness values at different sites are required
62.0
in order to provide reasonable agreement between
Observed Manning's n = 0.012
the model predictions and corresponding measure- 61.5
Manning's n = 0.015 Manning's n = 0.018
ments. 61.0
Figure 7 shows that 2DLLCDM can satisfactorily 0 60 120 180 240 300 360 420 480 540 600
simulate water levels at the three stations when the Time (hours)
roughness predictor of Equation (4) is used. This is
expected because the roughness predictor in FIGURE 6. Comparison Between the Calculated Water Levels by
2DLLCDM was calibrated by the measurements in Delft3D Using Various Roughness Coefficients and Those Observed
this reach. At Jiahetan, the calculated water levels at (a) Jiahetan, (b) Dongbatou1, and (c) Qingzhuang.
are closer to the observed data if a constant rough-
ness coefficient of 0.010 is used. At the stations of The following conclusions can be drawn from these
Dongbatou1 and Qingzhuang, the calculated water results: (1) a calibrated Manning’s predicator pro-
levels are in agreement with the observed data if a vides better results than constant Manning’s values;
constant roughness coefficient of 0.015 is used. There- (2) the value of Manning’s roughness coefficient in
fore, this comparison between the calculated water the LYR usually changes with flow intensity, and
levels under different roughness values and corre- reasonable results can be obtained if it ranges from
sponding measured data can testify further that a 0.010 to 0.015; and (3) various roughness values at
reasonable roughness predictor is necessary for a different cells are required in order to increase the
morphodynamic model in the LYR. predictive accuracy of water levels.

JAWRA 124 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION


COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER

78.0 However, a decrease in the value of Manning’s rough-


(a)
77.5 ness coefficient from 0.015 to 0.012 results in an ear-
lier arrival of the first peak discharge by
77.0
approximately 6 h. In addition, the comparison shows
Water level (m)

76.5 that a lower roughness coefficient yields more accu-


76.0 rate results. In the case of n = 0.012, the correlation
coefficient squared (R2) between the observed and cal-
75.5
culated discharges is 0.980, with the corresponding
75.0 value of root mean square deviation (RMSD) of
74.5 116.5 m3 ⁄ s. In the case of n = 0.018, the value of R2
between them is 0.942, with the corresponding RMSD
74.0
value of 194.2 m3 ⁄ s. Therefore, the correlation degree
0 60 120 180 240 300 360 420 480 540 600
between them would reduce by about 4% as the value
Time (hours)
of Manning’s roughness coefficient increases from
75.0
(b) 0.012 to 0.018.
74.5 Figure 8b indicates that different bed roughness
74.0 coefficients have a marked influence on the dis-
charges predicted by 2DLLCDM. In the case of the
Water level (m)

73.5
roughness predictor adopted, the value of R2 between
73.0 the observed discharges and calculated values is
72.5 0.982, with the corresponding RMSD value of
135.1 m3 ⁄ s. With the increase of the roughness coeffi-
72.0
cient, the calculated peak discharge is underesti-
71.5 mated by about 300 m3 ⁄ s, and the calculated
71.0 minimum discharge is overestimated by about
0 60 120 180 240 300 360 420 480 540 600 200 m3 ⁄ s. During the period from t = 190 h and
Time (hours) t = 270 h, the calculated discharges using the rough-
65.0 ness predictor are greater than the observed data.
(c) This phenomenon can also be found in the results in
64.5
Figure 8a from Delft3D, which may have been caused
64.0 by the fact that the process of water diversion during
this period was not taken into consideration in the
Water level (m)

63.5
models.
63.0 Based on the above comparison, the following
62.5 Observed conclusions are drawn: (1) Manning’s roughness
Manning's n = 0.010 coefficient has a marked effect on the predicted
62.0 discharge hydrograph at the downstream section for
Manning's n = 0.015
61.5 2DLLCDM; and (2) a lower value of Manning’s
Manning's n = 0.020 Manning's n = Predictor
roughness coefficient (n = 0.012) improves the simu-
61.0
lated results in this case study.
0 60 120 180 240 300 360 420 480 540 600
Time (hours)

Effect of Bed Roughness on Sediment Concentration


FIGURE 7. Comparison Between the Calculated Water Levels by
2DLLCDM Using Various Roughness Coefficients and Those
Observed at (a) Jiahetan, (b) Dongbatou1, and (c) Qingzhuang. Figure 9 compares the calculated concentrations of
suspended load at the downstream boundary using
different values of Manning’s roughness coefficient
Effect of Bed Roughness on Discharge with the observed data. Temporal variation in the
suspended load concentrations calculated by Delft3D
Figure 8 compares the discharge hydrographs at is similar to that in the observed data, as shown in
the downstream boundary calculated by different Figure 9a. The sediment concentrations calculated
Manning’s roughness coefficients with the observed with high roughness coefficients of 0.015 and 0.018
data. The effect of the bed roughness coefficient on are lower than the observed concentrations, whereas
the discharges predicted by Delft3D is minor, as the sediment concentrations calculated with a Man-
shown in Figure 8a. The calculated two peak dis- ning’s roughness coefficient of 0.012 agree better with
charges are in good agreement with those observed. the observations. During the early period, the

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 125 JAWRA


XIA, WANG, WANG, AND YU

3500
(a)
3000

Discharge (m /s)
2500

3
2000 2000
Observed

1500
1600 Manning's n = 0.012
1200 Manning's n = 0.015
1000 Manning's n = 0.018
800
48 52 56 60 64 68 72
500
0 60 120 180 240 300 360 420 480 540 600
Time (hours)
3500
(b)

3000
Discharge (m /s)

2500
3

2000
Observed
1500 Manning's n = Predictor
Manning's n = 0.010
1000 Manning's n = 0.015
Manning's n = 0.020

500
0 60 120 180 240 300 360 420 480 540 600
Time (hours)

FIGURE 8. Comparison Between the Calculated Discharges at Gaocun Using Different Roughness Coefficients and the
Measured Data: (a) Delft3D and (b) 2DLLCDM.

20
(a)
18
Concentration at GC (kg/m )
3

16 Observed
14 Manning's n = 0.012

12 Manning's n = 0.015
Manning's n = 0.018
10
8
6
4
2
0
0 60 120 180 240 300 360 420 480 540 600
Time (hours)
20
(b)
18
Concentration at GC (kg/m )

Observed
3

16
Manning's n = Predictor
14 Manning's n = 0.010
12 Manning's n = 0.015
10 Manning's n = 0.020

8
6
4
2
0
0 60 120 180 240 300 360 420 480 540 600
Time (hours)

FIGURE 9. Comparison Between the Calculated Concentrations at Gaocun Using Different Roughness Coefficients and the
Measured Data: (a) Delft3D and (b) 2DLLCDM.

JAWRA 126 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION


COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER

calculated concentrations are underestimated by 2 to ment with the measurements during the early period,
4 kg ⁄ m3 if different roughness coefficients are used, while they are underestimated during the second
whereas they are in close agreement with the mea- flood peak, as shown in Figure 9b. In the case of the
surements during the second flood peak. In the case roughness predictor adopted, the value of R2 between
of n = 0.012, the value of R2 between the observed the observed and calculated concentrations is 0.68,
and calculated concentrations is 0.71, with the corre- with the corresponding RMSD value of 1.45 kg ⁄ m3.
sponding RMSD value of 1.47 kg ⁄ m3. In the case of Based on the above results, the following conclu-
n = 0.018, the value of R2 between them is 0.65, with sions are obtained: (1) the concentrations of sus-
the corresponding RMSD value of 2.73 kg ⁄ m3. There- pended load predicted by Delft3D and 2DLLCDM are
fore, the correlation degree between the observed and in general agreement with the measurements if an
calculated concentrations would decrease by about appropriate value of Manning’s roughness coefficient
8.5% as the value of Manning’s roughness coefficient is used; and (2) limited accuracy for predicting sedi-
increases from 0.012 to 0.018. In addition, Delft3D ment concentrations in these models indicates that
also predicted the magnitude of cohesive sediment for the simulation of graded sediment transport depends
different Manning’s roughness coefficients. In the on various formulations such as the roughness coeffi-
case of n = 0.012, the mean magnitude of cohesive cient, the sediment transport capacity, and the source
sediment concentration at the downstream boundary term in Equation (5).
was predicted to be about 2.17 kg ⁄ m3, whereas the
mean value of noncohesive sediment concentration at
the same section was estimated to be 5.24 kg ⁄ m3; the Effect of Bed Roughness on Changes in
mean rate of bed load transport at the downstream Cross-Sectional Geometry
boundary was predicted to be equal to 0.18 t ⁄ s, which
was much less than the mean rate of suspended load Figures 10 and 11 compare the predicted changes in
transport (15.72 t ⁄ s). cross-sectional geometry at Sections CS03, CS29, and
For 2DLLCDM, the sediment concentrations calcu- CS42 with different Manning’s roughness coefficients to
lated with the roughness predictor are in close agree- those observed. CS03 was located 3.4 km downstream

77 (a) CS03 71 (b) CS29 67 (c) CS42

70 66
76
69 65
Bed elevation (m)

Bed elevation (m)

Bed elevation (m)

75 68 64
67 63
74
66 62
65 61
73
64 60
72 63 59
2800 3000 3200 3400 3600 3800 4000 4200 4200 4400 4600 4800 5000 5200 5400 4200 4400 4600 4800 5000 5200 5400
Distance from the left bank (m) Distance from the left bank (m) Distance from the left bank (m)

Initial(April) Observed (July) Manning's n = 0.012 Manning's n = 0.015 Manning's n = 0.018

FIGURE 10. Comparison Between the Calculated Cross-Sectional Profiles by


Delft3D Using Different Roughness Coefficients and Those Observed.

77 71 67
(a) CS03 (b) CS29 (c) CS42
70 66
76
69
65
Bed elevation (m)

Bed elevation (m)


Bed elevation (m)

75 68
64
67
74 63
66
62
73 65
64 61
72
63 60

71 62 59
2800 3000 3200 3400 3600 3800 4000 4200 4200 4400 4600 4800 5000 5200 5400 4200 4400 4600 4800 5000 5200 5400
Distance from the left bank (m) Distance from the left bank (m) Distance from the left bank (m)

Initial(April) Observed (July) Manning's n = Predictor Manning's n = 0.010 Manning's n = 0.015 Manning's n = 0.020

FIGURE 11. Comparison Between the Calculated Cross-Sectional Profiles by


2DLLCDM Using Different Roughness Coefficients and Those Observed.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 127 JAWRA


XIA, WANG, WANG, AND YU

of the inlet section with a bankfull width of 915 m. DISCUSSION


CS29 was located about 48 km downstream of the inlet
section with an initial bankfull width of 460 m. CS42
was located about 73 km downstream of the inlet sec- The above analysis of the predicted results from
tion with a bankfull width of 760 m. these two models indicates that the determination of
Figure 10 shows that in Delft3D, different bed Manning’s roughness coefficient is a crucial parame-
roughness coefficients only have a little effect on the ter in the models. In general, the predicted water
changes of lateral profiles, with the exception of CS03. levels and discharges agree reasonably with the
At CS03, the simulated profile with a Manning’s observed data if an appropriate Manning’s roughness
roughness coefficient of 0.018 agrees generally with coefficient is used. Delft3D and 2DLLCDM can gener-
the measurement. At CS29, the simulated profiles ally simulate the sediment concentrations, whereas
with different roughness values are characterized by neither of the models is able to accurately predict the
about 390 m retreat of the right floodplain bank, inde- observed changes in cross-sectional geometry partly
pendent of Manning’s roughness coefficients used. due to the complex morphodynamic processes in the
However, this section did not experience any bank ero- LYR. It should be pointed out that some discrepan-
sion. At CS42, the left deep-channel was not formed cies between the simulated and observed data were
for different roughness values. In a word, the final caused by the inaccuracy of input data, and a part of
profiles with the roughness value of n = 0.015 from them were induced by the model limitations them-
Delft3D indicate that excessive channel incision was selves.
predicted to occur in the reach, and the mean values
of bed levels were predicted to be 74.51, 66.51, and
62.63 m at these sections, respectively, which were Inaccuracy of Input Data
lower than the corresponding measured values of
75.02, 67.06, and 62.84 m. The limited agreement The inaccuracy of input data in these models con-
between the measured and calculated profiles indi- cerned both the initial bathymetry and initial bed-
cates that the accuracy for predicting morphological material composition. The inaccuracy of initial
changes depends on various modules, such as the bathymetry resulted from three scenarios: (1) large
hydrodynamic module, the sediment transport mod- spacing existed between the two observed consecutive
ule, and especially the module of bank erosion. sections. The current initial bathymetry used in the
Figure 11 indicates that, in 2DLLCDM, different models was obtained by the special interpolation
Manning’s roughness coefficients only have a little method using the available geometry of 46 cross-sec-
effect on the changes of lateral profiles at CS29 and tional profiles. The average distance between the two
CS42. At CS03, the simulated lateral profile using consecutive sections was about 2 km. This spacing was
the roughness predictor agrees closely with the too large for the complex bathymetry of the LYR,
observed one. At CS29, all the simulated profiles which could not provide a detailed presentation of the
using different roughness values show bank erosion, complex topography for 2D models. (2) The latest
although no bank erosion was observed. At CS42, the bathymetry in this reach was not available. The initial
simulated profiles do not show the deepening of the cross-sectional profiles were measured on April 20,
left channel using different roughness coefficients. In whereas the simulation period started on June 19. (3)
general, the final profiles with the roughness predic- Coarser mesh dimensions could not provide accurate
tor from 2DLLCDM also indicate that significant bed bathymetry. In this study, only 11 computational
scour was predicted to occur, with the calculated nodes were allocated along the lateral direction, which
mean bed levels of 75.01, 66.04, and 62.35 m at these could not provide a detailed representation of the main
sections, respectively. This poor agreement between channel and floodplains in cross-sectional geometry.
the observed and simulated lateral profiles can be However, finer mesh dimensions would have greatly
partly attributed to the complexity of morphological increased the computer time in this study. An increase
changes in the LYR on the one hand, and it also pro- in the density of computational grids can slightly
poses a requirement to refine the current morphody- improve the calculated results, according to a previous
namic model on the other hand. mesh sensitivity test similar to the current case study
It can be seen from the simulated lateral profiles (Wang et al., 2008).
that these models cannot accurately predict the The local sediment transport capacity is closely
changes in cross-sectional geometry at different sec- related to the local composition of bed material.
tions, which may be caused by the models them- Therefore, the composition of bed material and its
selves, and inaccurate computational conditions spatial variation influence the pattern of erosion or
input. In the following section, a detailed discussion deposition. However, this spatial variation is not
is given to investigate these errors. known in sufficient detail. Vertical variation in the

JAWRA 128 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION


COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER

bed-material composition (the history of the initial 0.040

conditions) is also important, but is not known at 0.035


present. Therefore, the inaccuracy of bed-material

Manning roughness coefficient


Jiahetan
0.030
composition may result in the inaccuracy of sediment Gaocun

transport capacity and corresponding magnitude of 0.025

bed evolution in local zones. 0.020

0.015

0.010
Model Limitations
0.005

Model limitations for the LYR cover many aspects 0.000


due to the complex morphology of the LYR. In the 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

current case study, the effects of high sediment con- Froude number

centrations and rapid bed evolution on the hydrody-


namic processes are not considered, and therefore, a FIGURE 12. Relationship Between the Observed Manning’s
decoupled solution of the governing equations is used Roughness Coefficient and Froude Number of Flow at Sections of
Jiahetan and Gaocun.
in the models, which is appropriate for low sediment
concentrations and slow bed changes (Cao and Car-
ling, 2002; Wu, 2008). Methods of determining the through summing the thickness of bed deformation
bed roughness coefficient and sediment transport for each fraction. In fact, the current predictive accu-
capacity have an important effect on the predicted racy of the graded sediment transport capacity for-
results. Furthermore, inaccurate simulation of the mulae is relatively low in these two models.
bank erosion process also influences the predicted Figure 13 indicates a comparison between the graded
changes of cross-sectional profiles. sediment transport capacity computed by Equation
Based on the comparison between the simulated (8) and corresponding observed concentration for
and observed water levels, it can be concluded that three fractions. It can be seen from this figure that
the bed roughness coefficient should be variable in the points for the computed and observed concentra-
both space and time. Figure 12 indicates the values tions are extremely scattered for the fractions with
of Manning’s roughness coefficient at Jiahetan and sediment diameters of <0.025 and >0.050 mm. The
Gaocun observed in recent years, and it can be seen computed concentrations are greatly underpredicted
that: (1) the roughness value varies with the flow for the fraction of diameter <0.025 mm, whereas they
intensity (Froude number); and (2) this value at are significantly overestimated for the fraction of
Jiahetan is usually different from that at Gaocun diameter >0.050 mm. For the fraction of diameter
with the same flow intensity. Therefore, the first ranging from 0.025 to 0.050 mm, the computed con-
improvement in the models is to develop a roughness centrations generally agree with the observed data,
predictor with relatively high accuracy for the LYR, as shown in Figure 13b. In addition, it should be
which requires more detailed measured roughness noted that both models do not consider the hiding
data to be collected, including flow and sediment and exposure mechanisms in nonuniform bed mate-
characteristics at various hydrometric sections. Fur- rial, and the hiding and exposure effect is a very
thermore, these influencing factors can be directly important aspect of multiple-sized sediment trans-
considered in developing the new roughness predic- port, which also influences the predictive accuracy
tor, including the hydraulic factor such as the Froude (Wu, 2008). Therefore, the second improvement in
number and the factor of bed features such as dimen- the models is to enhance the accuracy of the current
sionless shear stress. formulas for the graded sediment transport capacity.
The accuracy of the sediment transport capacity The process of bank erosion plays an important
formulas used in the models directly influences the role in the channel adjustments of the LYR, which is
morphological changes, and has an indirect effect on related to the riverbank soil composition and mechan-
the flood routing and water levels along the reach. Of ical characteristics. A study using the field observa-
the formulas of sediment transport capacity used, tion and indoor experiments on the bank soil samples
Van Rijn’s formula is relatively inaccurate due to the in this reach conducted by Xia et al. (2008) proposed
lack of calibration with the observed data in the LYR. that the majority of the riverbanks are cohesive, and
In each of the morphodynamic models, the change of are characterized by an obvious structure of vertical
bed elevation at a node is calculated according to the layers. In addition, the variation in riverbank soil
difference between the suspended load concentration composition along the reach is very noticeable in the
and its equilibrium value for each fraction, and the braided reach. Two factors can cause serious bank
total magnitude of bed evolution is then obtained erosion in the braided reach. First, a local low clay

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 129 JAWRA


100 100
(a) d < 0.025 mm (b) d = 0.025 - 0.050 mm

Computed S *k (kg/m3)

Computed S *k (kg/m3)
10 10

1 1

0.1 0.1
0.1 1 10 100 0.1 1 10 100
Measured Sk (kg/m3) Measured Sk (kg/m3)

content in riverbank soil reduces the value of the crit- at the downstream section from Delft3D and
ical shear stress of bank soil (0.1-0.3 N ⁄ m2), which is 2DLLCDM agree reasonably with the measurements,
much smaller than the average near-bank flow shear whereas both models cannot predict accurately the
stress (2-3 N ⁄ m2). Second, frequent bank failure dur- observed changes of cross-sectional profiles.
ing flood seasons is mainly caused by a large reduc- Based on further consideration, two factors causing
tion in the apparent cohesion from 34 to 4 kN ⁄ m2, the inaccuracy of simulated results were presented,
due to the increase in water content of riverbank soil. including the inaccuracy of input data and model lim-
However, such influencing factors cannot be taken itations. Three urgent improvements for these models
into consideration in the models of Delft3D and have been identified in order to accurately simulate
2DLLCDM. Therefore, the third improvement in the the morphodynamic processes in the LYR, which are:
models is to develop a bank erosion module, which is (1) the development of bed roughness predictor; (2)
based on the near-bank hydrodynamics and soil the refinement of graded sediment transport capacity
mechanics, and can account for the temporal varia- formulation; and (3) the consideration of bank erosion
tions in shear strength and water content of river- module. These improvements need to be implemented
bank soil. in the future mathematical river models for the LYR.

CONCLUSIONS ACKNOWLEDGMENTS

This research was partly supported by the Natural Science


Foundation of China (Grant No. 51079103), and was partly sup-
In this study, two morphodynamic models (Delft3D ported by the Program for New Century Excellent Talents in Uni-
and 2DLLCDM) were intercompared as they were versity from the Chinese Ministry of Education (Grant No. NCET-
used to simulate the morphodynamic processes in the 10-0619) and by the Scientific Special Expenditure for Nonprofit
Public Industry from the Chinese Ministry of Water Resources
LYR. Both models applied the method of decoupled
(200901015).
solution due to low sediment concentrations and slow
bed changes in the current case study. The intercom-
parison included the comparison of the model con-
LITERATURE CITED
cepts, and the differences between the model
predictions and corresponding field measurements. Cao, Z.X. and P.A. Carling, 2002. Mathematical Modelling of Allu-
Moreover, the corresponding explanations were pre- vial Rivers: Reality and Myth. Part I: General Review. Proceed-
sented to investigate these differences. Therefore, ings of the Institution of Civil Engineers Water, Maritime and
Energy 154(3):207-219.
some experience from this intercomparison can also
Chien, N. and Z.H. Wan, 1999. Mechanics of Sediment Transport.
be further referred to the improvements for morpho- American Society of Civil Engineers, Reston, Virginia, 913 pp.
dynamic models in other alluvial rivers. Darby, S.E., 1998. Modelling Width Adjustment in Straight Allu-
The simulated results indicate that the predicted vial Channels. Hydrological Processes 12(8):1299-1321.
water levels along the reach and discharges at the Delft Hydraulics, 2003. User Manual of Delft3D-FLOW: Simulation
of Multi-Dimensional Hydrodynamic Flows and Transport Phe-
downstream boundary from these two models can
nomena, Including Sediments. Report of Delft Hydraulics, The
agree closely with the observed data with an appro- Netherlands, 614 pp.
priately selected value of Manning’s roughness coeffi- Forester, C.K., 1979. Higher Order Monotonic Convective Differ-
cient. The predicted concentrations of suspended load ence Schemes. Journal of Computational Physics 23(1):1-22.
COMPARISON OF MORPHODYNAMIC MODELS FOR THE LOWER YELLOW RIVER

Galappatti, R. and C.B. Vreugdenhil, 1985. A Depth-Integrated Wang, Z.B., 1992. Theoretical Analysis on Depth-Integrated Model-
Model for Suspended Sediment Transport. Journal of Hydraulic ling of Suspended Sediment Transport. Journal of Hydraulic
Research 23(4):359-377. Research 30(3):403-421.
Julien, P.Y., 2002. River Mechanics. Cambridge University Press, Wu, B.S., D.S. Van Maren, and L.Y. Li, 2008b. Predictability of
Cambridge, 433 pp. Sediment Transport in the Yellow River Using Selected Trans-
Krone, R.B., 1962. Flume Studies of the Transport of Sediment in port Formulas. International Journal of Sediment Research
Estuarial Shoaling Process. University of California, Berkeley, 23(4):283-298.
110 pp. Wu, B.S., G.Q. Wang, J.Q. Xia, X.D. Fu, and Y.F. Zhang, 2008a.
Leendertse, J.J., 1970. A Water Quality Simulation Model for Well- Response of Bankfull Discharge to Discharge and Sediment
Mixed Estuaries and Coastal Seas (Vol. 1), Principles of Compu- Load in the Lower Yellow River. Geomorphology 100(3-4):366-
tation. Rand Corporation, RM-6230-RC, New York, 71 pp. 376.
Lesser, G.R., J.A. Roelvink, V. Kester, and G.S. Stelling, 2004. Wu, W.M., 2008. Computational River Dynamics. Taylor & Francis,
Development and Validation of a Three-Dimensional Morpholog- New York, 494 pp.
ical Model. Coastal Engineering 51(8-9):883-915. Wu, W.M., E.H. Jiang, and S.S.Y. Wang, 2004. Depth-Averaged 2D
Li, G.Y., 2003. Ponderation and Practice of the Yellow River Con- Calculation of Flow and Sediment Transport in the Lower Yel-
trol. Yellow River Conservancy Press, Zhengzhou, China, 271 low River. International Journal of River Basin Management
pp. 2(1):51-59.
Long, Y.Q. and Y.F. Zhang, 2002. Study on the Yellow River Sedi- Xia, J.Q., B.S. Wu, G.Q. Wang, and Y.P. Wang, 2010. Estimation of
ment From the Viewpoint of Total Sediment Load. Journal of Bankfull Discharge in the Lower Yellow River Using Different
Yellow River 24(9):28-30 (in Chinese). Approaches. Geomorphology 117(1-2):66-77.
Mosselman, E., 1998. Morphological Modelling of Rivers With Erod- Xia, J.Q., B.S. Wu, and Y.P. Wang, 2007. Processes and Character-
ible Banks. Hydrological Processes 12(8):1357-1370. istics of Recent Channel Adjustment in the Lower Yellow River.
Ni, J.R., H.W. Zhang, A. Xue, S. Wieprecht, and A.G.L. Borthwick, In: Proceedings of the 10th International Symposium on River
2004. Modeling of Hyperconcentrated Sediment-Laden Floods in Sedimentation (Vol. V), M.V. Lomonosov (Editor). Moscow State
the LYR. ASCE Journal of Hydraulic Engineering 130(10):1025- University, Moscow, Russia, pp. 126-135.
1032. Xia, J.Q., B.S. Wu, Y.P. Wang, and S.G. Zhao, 2008. An Analysis of
Osman, A.M. and C.R. Thorne, 1988. Riverbank Stability Analysis Soil Composition and Mechanical Properties of Riverbanks in a
I: Theory. ASCE Journal of Hydraulic Engineering 114(2):134- Braided Reach of the Lower Yellow River. Chinese Science Bul-
150. letin 53(15):2400-2409.
Partheniades, E.A., 1965. Erosion and Deposition of Cohesive Soils. Xia, J.Q., G.Q., Wang., and B.S. Wu, 2004. Two-Dimensional
ASCE Journal of the Hydraulics Division 91(1):105-139. Numerical Modelling of the Longitudinal and Lateral Channel
Roelvink, J.A., T. Van Kessel, S. Alfageme, and R. Canizares, 2003. Deformations in Alluvial Rivers. Science in China (Series E)
Modelling of Barrier Island Response to Storms. In: Proceedings 47(Suppl. I):199-211.
of Coastal Sediments 2003, R.A. Davis, A. Sallenger, and P. Yanenko, N.N., 1971. The Method of Fractional Steps: The Solution
Howd (Editors). World Scientific Pub Co Inc., Florida, pp. 1-11. of Problems of Mathematical Physics in Several Variables (Eng-
Stelling, G.S., 1984. On the Construction of Computational Meth- lish translation edited by M. Holt). Springer-Verlag, Berlin.
ods for Shallow Water Flow Problem. Rijkswaterstaat Commu- Zhang, H.W., Y.D. Huang, and L.J. Zhao, 2001. A Mathematical
nications, Vol. 35. Government Printing Office, The Hague, The Model for Unsteady Sediment Transport in the LYR. Interna-
Netherlands. tional Journal of Sediment Research 16(2):150-158.
Stelling, G.S. and J.J. Leendertse, 1991. Approximation of Convec- Zhang, H.W. and Q. Zhang, 1992. Formula for the Sediment Trans-
tive Processes by Cyclic AOI Methods. In: Proceeding of the 2nd port Capacity of the Yellow River. Journal of Yellow River
ASCE Conference on Estuarine and Coastal Modelling, M.L. 14(11):7-9 (in Chinese).
Spaulding, K. Bedford, A. Blumberg, R. Cheng, and C. Swanson Zhang, R.J. and J.H. Xie, 1993. Sedimentation Research in China.
(Editors). The Waterway, Port, Coastal and Ocean Division of China Water and Power Press, Beijing, 260 pp.
ASCE, New York, pp. 771-782. Zhang, S.Q., 1994. Two-Dimensional Mathematical Model for
Van Maren, D.S., M. Yang, and Z.B. Wang, 2011. Predicting the Studying Erosion and Sedimentation of a Wandering Stretch in
Morphodynamic Response of Silt-Laden Rivers to Water and the Lower Yellow River. International Journal of Sediment
Sediment Release From Reservoirs: Lower Yellow River, China. Research 9(Special Issue):246-255.
ASCE Journal of Hydraulic Engineering 137(1):90-99.
Van Rijn, L.C., 1984a. Sediment Transport, Part II: Suspended
Load Transport. ASCE Journal of Hydraulic Engineering
110(11):1613-1641.
Van Rijn, L.C., 1984b. Sediment Transport, Part I: Bed Load
Transport. ASCE Journal of Hydraulic Engineering
110(10):1431-1456.
Van Rijn, L.C., 2007a. Unified View of Sediment Transport by Cur-
rents and Waves I: Initiation of Motion, Bed Roughness, and
Bed-Load Transport. ASCE Journal of Hydraulic Engineering
133(6):649-667.
Van Rijn, L.C., 2007b. Unified View of Sediment Transport by Cur-
rents and Waves II: Suspended Transport. ASCE Journal of
Hydraulic Engineering 133(6):668-689.
Wang, G.Q., J.Q. Xia, and B.S. Wu, 2008. Numerical Simulation of
Longitudinal and Lateral Channel Deformations in the Braided
Reach. ASCE Journal of Hydraulic Engineering 134(8):1064-1078.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 131 JAWRA

You might also like