You are on page 1of 13

Advances in Water Resources 33 (2010) 171–183

Contents lists available at ScienceDirect

Advances in Water Resources


journal homepage: www.elsevier.com/locate/advwatres

Modelling dam-break flows over mobile beds using a 2D coupled approach


Junqiang Xia a,b,*, Binliang Lin a, Roger A. Falconer a, Guangqian Wang b
a
Hydro-environmental Research Centre, School of Engineering, Cardiff University, CF24 3AA, UK
b
State Key Laboratory of Hydroscience and Engineering, Tsinghua University, Beijing 100084, China

a r t i c l e i n f o a b s t r a c t

Article history: Dam-break flows usually propagate along rivers and floodplains, where the processes of fluid flow, sed-
Received 2 June 2009 iment transport and bed evolution are closely linked. However, the majority of existing two-dimensional
Received in revised form 3 November 2009 (2D) models used to simulate dam-break flows are only applicable to fixed beds. Details are given in this
Accepted 10 November 2009
paper of the development of a 2D morphodynamic model for predicting dam-break flows over mobile
Available online 15 November 2009
beds. In this model, the common 2D shallow water equations are modified, so that the effects of sediment
concentrations and bed evolution on the flood wave propagation can be considered. These equations are
Keywords:
used together with the non-equilibrium transport equations for graded sediments and the equation of
Morphodynamic model
Dam-break flows
bed evolution. The governing equations are solved using a matrix method, thus the hydrodynamic, sed-
Mobile bed iment transport and morphological processes can be jointly solved. The model employs an unstructured
Coupled solution finite volume algorithm, with an approximate Riemann solver, based on the Roe-MUSCL scheme. A pre-
Finite volume method dictor–corrector scheme is used in time stepping, leading to a second-order accurate solution in both
time and space. In addition, the model considers the adjustment process of bed material composition
during the morphological evolution process. The model was first verified against results from existing
numerical models and laboratory experiments. It was then used to simulate dam-break flows over a fixed
bed and a mobile bed to examine the differences in the predicted flood wave speed and depth. The effects
of bed material size distributions on the flood flow and bed evolution were also investigated. The results
indicate that there is a great difference between the dam-break flow predictions made over a fixed bed
and a mobile bed. At the initial stage of a dam-break flow, the rate of bed evolution could be comparable
to that of water depth change. Therefore, it is often necessary to employ the turbid water governing equa-
tions using a coupled approach for simulating dam-break flows.
Ó 2009 Elsevier Ltd. All rights reserved.

1. Introduction flows [3–7]. With the advancement of computer technology and


numerical solution methods of the shallow water equations,
Dam-break flows could lead to severe flooding with cata- hydrodynamic models based on one-dimensional (1D) and two-
strophic consequences, such as damage to properties and loss of dimensional (2D) approaches are increasingly being used for pre-
human life. Therefore, dam-break flows have been the subject of dicting dam-break flows. Currently, numerical solutions of the
scientific and technical research for many hydraulic scientists shallow water equations type are one of the most active topics in
and engineers. Earlier studies were primarily based on analytical the field of hydraulics research. Several numerical models pertain-
solutions for idealised conditions. For example, Stoker [1] devel- ing to dam-break flows can be found in the literature, and they
oped an analytical solution to predict dam-break flows in an idea- have been successfully used to predict flood inundation extent
lised channel, in which the bed slope was assumed to be zero and and velocity distributions. However, the majority of these models
the friction term was ignored. Chanson [2] proposed an analytical are only applicable to dam-break flows over fixed beds [3,8–11].
solution of dam-break waves with flow resistance, and then it was In some catastrophic flood events, particularly those caused by
applied to simulate tsunami surges on dry coastal plains. During dam or dike failures, flood flows have induced severe sediment
the past two decades, numerical models and laboratory experi- movements in various forms: debris flows, mud flows, floating
ments have become very popular for investigating dam-break debris and sediment-laden currents [12]. Capart et al. [13] pointed
out that in some extreme cases, the volume of entrained material
could reach the same order of magnitude as the volume of water
initially released from the failed dam. For example, the Chandora
* Corresponding author. Address: Hydro-environmental Research Centre, School
of Engineering, Cardiff University, CF24 3AA, UK. river dam-break flow occurred in India in 1991 scoured a 2 m thick
E-mail address: xiaj1@cf.ac.uk (J. Xia). layer of bed material from the reach immediately downstream of

0309-1708/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advwatres.2009.11.004
172 J. Xia et al. / Advances in Water Resources 33 (2010) 171–183

the dam [14]. It is often necessary to account for the process of 2. Governing equations
morphological changes when simulating such severe dam-break
flows. Currently, two approaches are often used to model the mor- Due to the interaction between the sediment-laden flow and
phodynamic processes: uncoupled and coupled solutions [15]. In river bed, a river will undergo continuous morphological changes.
order to model the morphodynamic processes caused by dam- For dam-break flows, the processes of flood wave propagation and
break flows, the second method may be more acceptable. This is river channel evolution will usually be very significant. In order to
due to the rate of bed evolution often being more comparable to accurately model these processes, the following equations are used
the rate of water depth variation. Early numerical models for sim- in the current model:
ulating dam-break flows over mobile beds often adopted uncou-
pled solutions that did not account for the effects of sediment (i) a set of modified shallow water equations, which are able to
transport and bed deformation on the movement of flow take into account the influences of sediment transport and
[4,16,17]. Fraccarollo and Capart [18], Spinewine and Zech [19] bed deformation on the flood flows;
used two-layer 1D models to simulate dam-break flows over mo- (ii) the non-equilibrium transport equations for both suspended
bile beds. These models were applicable to morphological changes and bed loads; and
caused predominately by the non-equilibrium transport of bed (iii) the equation of bed evolution.
load. The applicability of the models was considered to be limited
because of the assumption of a constant sediment concentration in In this study, graded sediments are considered, with a compu-
the lower layer [6]. More recently, several models for simulating tational procedure being employed to allow bed material composi-
dam-break flows over mobile beds based on the coupled solution tion adjustment to be made.
have been developed. Cao et al. [20] presented a 1D numerical
model to simulate the hydraulics of dam-break flows over mobile
beds and the induced sediment transport and morphological evo- 2.1. Hydrodynamic equations
lution, and provided a detailed description of dam-break fluvial
The hydrodynamic governing equations used are based on the
processes. Recently they extended the analysis of the multiple time
two-dimensional shallow water equations, but with additional
scales of subaerial (near-bed) sediment-laden flows over erodible
terms being included to account for the sediment effects on the
bed to subaqueous turbidity currents, and proposed a fully coupled
fluid density and bed level change [15,27]. The shallow water gov-
modelling study [21,22]. Wu and Wang [6] proposed a similar 1D
erning equations of the 2D hydrodynamic model comprise the
model to simulate dam-break flows over mobile beds using the
mass and momentum conservation equations for the water–sedi-
coupled solution, and applied the model to investigate the mecha-
ment mixture flow. Coupled equations have been presented in a
nisms of morphodynamic processes caused by dam-break flows. In
one-dimensional form by Fagherazzi and Sun [4] and Cao et al.
Wu and Wang’s model, a more complex method was used to calcu-
[20], and in a two-dimensional form by Simpson and Castelltort
late the rates of sediment deposition and entrainment, which could
[25], Yue and Cao et al. [28]. The modified continuity (Eq. (1))
account simultaneously for the process of bed evolution caused by
and momentum equations in the x and y-directions (Eqs. (2) and
the suspended and bed loads.
(3)) can be expressed in detail as follows:
Although many 2D dam-break flow models over non-mobile or
fixed beds have been developed over the past decade [3,5,8–
@ @ @ @Z b
11,23,24], 2D models for dam-break flows over mobile beds using ðhÞ þ ðhuÞ þ ðhv Þ ¼  ð1Þ
@t @x @y @t
the coupled solution are not often reported due to the complexity
of flow-sediment transport and bed evolution. Simpson and Cas-  
telltort [25] extended an existing 1D coupled model of Cao et al. @ @ 2 1 2 @
ðhuÞ þ hu þ gh þ ðhuv Þ
[20] to a 2D model for the free surface flow, sediment transport @t @x 2 @y
!
and morphological evolution. This model used a Godunov-type @2u @2u
2
Dqgh @S q0  qm u@Z b
method with a first-order approximate Riemann solver, and was ¼ ghðSbx  Sfx Þ þ hv t þ  þ
@x 2 @y 2 2qs qm @x qm @t
verified by comparing the computed results with the documented
solutions. As commented by Cao [26], the first-order numerical ð2Þ
scheme in solving the governing equations may have limitations
in modelling water levels and sediment concentrations with gradi-
 
@ @ @ 2 1 2
ent discontinuities. The model was applied to test cases with some ðhv Þ þ ðhuv Þ þ hv þ gh
@t @x @y 2
idealised flat bed channels, without the need to consider the com- !
2
plicated wetting and drying fronts. Therefore, it is necessary to de- @2v @2v Dqgh @S q0  qm v @Z b
¼ ghðSby  Sfy Þ þ hv t þ  þ
velop a morphodynamic model for simulating dam-break flows @x2 @y2 2qm qs @y qm @t
over mobile beds with more advanced solution schemes and wider ð3Þ
applicability.
In the current study, a 2D morphodynamic model has been in which t = time; h = water depth; u and v = velocity components in
developed to simulate dam-break flows over mobile beds. In the the x and y-directions, respectively; g = gravitational acceleration;
model, a 2D hydrodynamic module is coupled with the sediment mt = turbulent viscosity coefficient, mt ¼ hu h, where u = friction
transport and bed evolution modules to predict simultaneously velocity and h = empirical coefficient, 0:0 < h < 1:0; Dq ¼ qs  qf ,
the hydrodynamics, sediment concentrations and morphological in which qf = clear water density and qs = sediment density;
changes. The model was first verified against results from existing
qm = density of water–sediment mixture, qm ¼ ð1  Sm Þqf þ Sm qs ,
numerical models and experimental data from laboratory tests
where Sm is the volumetric sediment concentration, and Sm ¼ S=qs ,
documented in the literature, and was then used to investigate
in which S = total concentration of graded sediments; q0 = density
the influences of different bed material compositions on the flood
of saturated bed material, q0 ¼ ð1  q0 =qs Þqf þ q0 , in which q0 =
levels and the morphology of channel bed. Finally, this study sim-
ulated dam-break flows over a fixed bed and a mobile bed to dry density of bed material. The bed slope terms ðSbx ; Sby Þ and fric-
examine the differences in the predicted flood wave speed and tion slope terms ðSfx ; Sfy Þ are written as Sbx ¼ @Z b =@x; Sby ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4=3 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4=3
depth. @Z b =@y and Sfx ¼ n2 u u2 þ m2 =h ; Sfy ¼ n2 v u2 þ m2 =h in the
J. Xia et al. / Advances in Water Resources 33 (2010) 171–183 173

x, y-directions, respectively, where Z b = bed elevation; n = The equation used to represent the bed load induced during bed
Manning’s roughness coefficient. evolution is written as:
The last two terms on the right-hand side of Eqs. (2) and (3) rep-
DZ bk
resent respectively, spatial variations in sediment concentration, q0 ¼ abk xbk ðqbk  qbk Þ ð5:2Þ
Dt
and momentum transfer due to sediment exchange between the
flow and the erodible bed. These two terms are only significant Therefore, the thickness of the channel bed due to the total sedi-
during the transport of hyper-concentrated flows or during the ra- ment load is given as:
pid bed evolution, and they seldom appear in the classical clear- X
Ns X
N
water governing equations. DZ t ¼ DZ sk þ DZ bk ð5:3Þ
k¼1 k¼Ns þ1
2.2. Transport equations of suspended and bed loads
in which N = total number of fractions of non-uniform sediments;
Ns = number of fractions of non-uniform suspended sediments;
In the current model, the non-equilibrium transport processes
DZ sk and DZ bk = thicknesses of bed deformation caused by sus-
of graded suspended and bed loads are included in the equation
pended load and bed load, respectively, in one time step; and
of sediment transport.
DZ t = total thickness of bed evolution in one time step. Therefore,
For the suspended load, the 2D non-equilibrium transport equa-
the bed elevation at a cell after one time step can be obtained by
tion is given as:
ðZ b Þlþ1 ¼ ðZ b Þl þ DZ t , in which the superscript l represents the time
@ @ @ level.
ðhSk Þ þ ðhuSk Þ þ ðhv Sk Þ
@t @x @y
    2.4. Sediment transport capacity
@ @Sk @ @Sk
¼ hes þ hes  ask xsk ðSk  Sk Þ ð4:1Þ
@x @x @y @y
A formula proposed by Wu and Long [33] is used to compute
in which es = turbulent diffusion coefficient of sediment; subscript k the sediment transport capacity of the suspended load, which is gi-
represents the kth sediment fraction; Sk ; Sk and xsk represent, ven as:
respectively, the sediment concentration, sediment transport " #M
capacity and effective settling velocity for the kth fraction; cm U3
ask = non-equilibrium adaptation coefficient of suspended load, S ¼ K ð6Þ
cs  cm ghxm
which is an empirical coefficient associated with the rate of bed
evolution. where K and M are two empirical parameters, K = 0.452 kg/m3 and
Several empirical methods have been proposed in the literature pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
M = 0.762; U = depth-average speed, U ¼ u2 þ v 2 ; cs and cm = spe-
for determining the value of ask , with significant differences exist- cific weights of sediment and turbid water; xm = group settling
ing among these methods [20,29]. It has been found that if the P 1=M
Ns M
same ask value is used for all of the sediment fractions, then during velocity of graded suspended sediments, xm ¼ k¼1 DP k xsk ;
the process of bed degradation the amount of coarse sediment DP k = percentage of sediment transport capacity for the kth size
eroded from the bed is greater than that of the finer fraction. This PNs
fraction, which is determined using DP k ¼ ðDPbk =xsk Þ/ = k¼1
causes the composition of bed material to become finer, rather
ðDPbk =xsk Þ/ , in which / is an empirical coefficient, and DPbk = per-
than getting coarser, which is physically incorrect. In order to avoid
centage of bed material for the kth grain size fraction.
this, different values of ask are usually set for different fractions. In
A formula proposed by Dou et al. [31] is used to determine the
the present study, the following empirical formulae were used to
bed-load transport capacity, which is given as:
determine ask [30]:
K b qs qm U3
ask ¼ b=x0:3
sk ; when Sk > Sk ; and qb0k ¼ 2 q q
ðU  U ck Þ  DPbk ð7Þ
C0 s m g xbk
ask ¼ b=x0:7
sk ; when Sk 6 Sk ;

in which qb0k = bed-load transport capacity per unit width for the
in which is b a case-dependent coefficient.
kth grain size fraction in kg/ms; K b is an empirical coefficient;
For the bed load, the 2D non-equilibrium transport equation
U ck = incipient velocity of the kth bed-load fraction; C 0 = dimension-
proposed by Dou et al. [31] is used, given as: pffiffiffi 1=6
less Chézy coefficient with C 0 ¼ C= g , in which C ¼ h =n. The va-
@ @ @ lue of bed load in a unit volume is obtained according to
ðhqbk Þ þ ðhuqbk Þ þ ðhv qbk Þ ¼ abk xbk ðqbk  qbk Þ ð4:2Þ
@t @x @y qbk ¼ qbok =hU. This formula can be used to compute the transport
capacity of sand or gravel bed-load whose diameter ranges from
in which qbk = amount of bed load in a unit volume of water, in kg/ 0.05 to 200 mm. The value of parameter K b in Eq. (7) was calibrated
m3; xbk = setting velocity of bed load; qbk = transport capacity of by the experimental data and it was about 0.1.
bed load in a unit volume of water, in kg/m3; and abk = non-equilib-
rium adaptation coefficient of bed load. Several studies have been 2.5. Bed material composition adjustment during bed evolution
conducted to determine this coefficient [31,32]. In the current
study, the value of abk suggested by Dou et al. [31] was used, and In order to simulate the phenomenon of armouring, i.e. sorting
it ranged from 0.1 to 1.0. of bed material, caused by degradation or aggradation, the bed
material at each computational cell is divided into two vertical lay-
2.3. Bed evolution equations ers: the upper one is called the mixing or active layer and the lower
one is called the memory layer. The thickness of the mixing layer is
The equation used to represent the suspended load induced denoted by Hb , with its gradation being represented by DP bk . The
during bed evolution is written as: memory layer is further divided into m smaller sub-layers, with
the thickness and gradation of each sub-layer being represented
DZ sk
q0 ¼ ask xsk ðSk  Sk Þ ð5:1Þ by DHm and DP mk , respectively. The adjustment procedure of the
Dt
size distribution of surface bed material can be classified into
174 J. Xia et al. / Advances in Water Resources 33 (2010) 171–183

two cases of bed scour and bed deposition, with a detailed descrip- y
tion of the procedure being given in Wang et al. [34]. cell node
cell centre lij Uk
3. Numerical solution procedure of the governing equations k
In the present study, a numerical model has been developed to j=1
simulate the morphodynamic behavior of dam-break flows over
Uj
UR n y Aj
mobile beds. The finite volume method is used to solve the govern- n
UL nx
ing equations discussed above, while using an unstructured trian- b
j=2
gular mesh. A cell-centered finite volume method is used in this SR
model, in which the average values of conserved variables are SL
Ui i Γ
stored at the centre of each cell with the edges of a cell defining a
the interface between this cell and the neighboring cells (see
Fig. 1).
In the current model, the numerical methods used include the m
Roe-MUSCL scheme for computing flow fluxes, treatment of the Um Ai
j=3
source term, method of time integration, treatment of wetting
and drying fronts and treatment of boundary conditions. Some of
o x
these methods are similar to those published in the literature
[10,35–37]. A detailed description of the methods for solving the
Fig. 1. Sketch of a control volume.
hydrodynamic governing equations for flows over a fixed bed can
be found in Xia et al. [38]. In the coupled mobile-bed model, an up-
wind scheme is used to compute sediment fluxes, and an explicit Integrating Eq. (8) over a control volume Ai yields:
discretization is employed to treat the additional terms associated Z Z Z Z
@U
with sediment transport and bed deformation in the morphody- dA þ r ~
FdA ¼ r ~
TdA þ SdA ð10Þ
namic model. In addition, a procedure is used to compute the Ai @t Ai Ai Ai

bed material composition adjustment during bed evolution, which in which ~F ¼ ðE; GÞ and ~ E; ~
T ¼ ð~ GÞ. Assuming U to be the average va-
is necessary for modelling the transport of non-uniform sediments. lue of the variables stored in a cell, the area integral can be used to
calculate approximately the unsteady terms, i.e. the local time var-
3.1. Discretization of flow and sediment governing equations iation and source terms in Eq. (9). Thus Eq. (10) becomes
I I
In order to solve the hydrodynamic governing equations using
@U
DA i þ F n ðUÞdC ¼ T n ðUÞdC þ SðUÞDAi ð11Þ
the finite volume method, Eqs. (1)–(4) are rewritten in a conserva- @t C C
tive form as: in which C denotes the boundary of Ai ; F n ðUÞ ¼ ~ F  n ¼ Enx þ
Gny ; Tn ðUÞ ¼ ~
T  n ¼ En e y ; n = outward unit vector normal to
e x þ Gn
e @G
@U @E @G @ E e
þ þ ¼ þ þS ð8Þ the boundary C; nx and ny = components of unit normal vector in
@t @x @y @x @y the x and y-directions, respectively. If the line integral can be
in which approximated by summing the flux vector over each edge of a trian-
2 3 2 3 2 3 gular control volume, the terms of line integral in Eq. (11) can be
h hu hv discretized in the following way:
6 7 6 2 1 27 6 7
6 hu 7 6 hu þ 2 gh 7 6 huv 7 I X
3 I X
3
U¼6
6
7;
7 E¼6
6
7;
7 G¼6
6 2 1 2 7;
7
F n ðUÞdC ¼ F ij Dlij and T n ðUÞdC ¼ T ij Dlij ð12Þ
4 hv 5 4 huv 5 4 hv þ 2 gh 5 C C
j¼1 j¼1
hSk huSk h v Sk
2 3 2 3 in which, Dlij = length of the edge between cells i and j, i.e. lij ; F ij and
0 0 T ij numerical convective and diffusive fluxes across lij . T ij is usually
6 7 6 7 e y . Thus Eq. (11) can now be rewritten
e x þ Gn
6 m hð@u=@xÞ 7 6 m hð@u=@yÞ 7 estimated using T ij ¼ En
e¼6 t
E 7; e ¼6 t
G 7 and
6 7 6 7 as:
4 mt hð@ v =@xÞ 5 4 mt hð@ v =@yÞ 5
es hð@Sk =@xÞ es hð@Sk =@yÞ @Ui 1 X 3
1 X 3
e y ÞDlij þ SðUi Þ
e x þ Gn
¼ F ij Dlij þ ð En ð13Þ
2 3 @t DAi j¼1 DAi j¼1
2 3 @Z b =@t
0
6 7
6 7 6  Dqgh2 @S
þ q0qqm u @Z@tb 7 Therefore, the key task in the above discretization procedure is to
6 þghðS bx  Sfx Þ 7 6 2qm qs @x 7
S¼6 7þ6 m
7 ð9Þ evaluate the convective fluxes F ij for the flow and sediment trans-
6
4 þghðSby  Sfy Þ
7 6 Dqgh2
5 6 q0 qm @Z b 7 7
4 2qm qs
@S
@y
þ qm v @t 5
port. In the current model, a similar method used by Liang et al.
ask xsk ðSk  Sk Þ [39] is applied here. The flow fluxes in F ij are solved by the Roe’s
0
approximate Riemann solver, with the sediment fluxes in F ij being
in Eq. (9) U is a vector of the conserved variables; E and G are the solved directly.
convective flux vectors of the flow in the x and y-directions, respec-
e and G
tively; E e are the diffusive vectors in the x and y-directions, 3.1.1. Roe-MUSCL scheme for the flow flux
respectively; and S is the source term including the bed friction, At the interface between cells Ai and Aj , the problem could be
bed slope and additional factors associated with sediment transport taken as a locally one-dimensional Riemann problem in the direc-
and bed deformation. Eq. (9) is a set of equations used to predict the tion normal to the interface [40], so the normal fluxes can be ob-
turbid water flow and the transport of the suspended load. The tained by an approximate Riemann solver, i.e.
transport equation for the bed load can be written in the same
F ij ¼ F  ½ðUL Þij ; ðUR Þij  ð14Þ
way as Eq. (9), but with the diffusive terms being neglected.
J. Xia et al. / Advances in Water Resources 33 (2010) 171–183 175

lþ12
where F  is an approximate Riemann solver for flow, and ðUL Þij and where U ¼ ðUl þ U Þ=2 and U ¼ Ul þ DtLðUl Þ. Since this method is
ðUR Þij are the reconstructions of U on the right and left sides of the an explicit scheme, the time step is restricted by the Courant–
interface C, respectively. Using different Riemann solvers and dif- Friendrichs–Lewy (CFL) condition. In a practical simulation, a con-
ferent reconstructions of ðUL Þij and ðUR Þij , various numerical stant time step is usually used and the CFL condition is used to
schemes have been be obtained, e.g. the HLL or HLLC Riemann’ solv- determine approximately the time step before the simulation starts.
ers [35,41,42]. The Roe’s approximate Riemann solver with the
MUSCL scheme (it is called later as Roe-MUSCL scheme) has been 3.1.5. Treatment of wetting and drying fronts
considered as one of the most accurate schemes [36,43], in conjunc- For predicting the flooding in a natural river with complicated
tion with an effective means of dealing with wetting and drying bed topography, it is necessary to employ an approach to simulat-
fronts. In the practical application of Roe’s approximate Riemann ing the evolution of wetting and drying fronts. The presence of ex-
solver, the cell will be temporarily out of the computational domain treme bed slopes and strong changes in the irregular geometry
if a cell is dry, and a more detailed validation of this method can be often results in great difficulties for numerical solutions, since an
found in Xia et al. [38]. Therefore, this scheme is employed in the inaccurate treatment of the wetting and drying fronts may lead
current model to evaluate the normal flow flux across an interface. to a significant prediction error. In using FVM for solving the shal-
Eq. (14) is expressed by: low water equations, many improvements have been made to the
h i  _
  prediction of wetting and drying fronts. In the methods by Zhao
F  ½ðUL Þij ; ðUR Þij  ¼ 0:5 ~ FðUL Þij  n  A½ðUR Þij  ðUL Þij 
FðUR Þij þ ~
et al. [48] and Sleigh et al. [43], the computational cells are divided
ð15Þ into three types, including wet, dry and partially dry cells. For a
partially dry cell, the momentum flux across an interface is set to
in which, ~FðUL Þij and ~
FðUR Þij = normal_fluxes on the left and right
zero with only the mass flux being considered. Brufau et al. [49]
sides of the edge C, respectively and B = flux Jacobian matrix eval-
presented a method of zero mass error using unsteady wetting
uated by Roe’s average [44]. The MUSCL scheme proposed by Van
and drying conditions in shallow water flows, with this method
Leer [45] is used in the present model to reconstruct the state vari-
being valid for a first-order accurate FVM. However, in practice a
ables (UR and UL ). In addition, the minmod limiter proposed by Roe
small value of minimum water depth hmin is usually introduced
and Baines [46] is used to ensure solution to remain positive, with
to simulate the evolution of wetting and drying fronts [50].
this method being extensively used in TVD-type numerical
In the current study, a wetting and drying method developed
schemes. It can be shown that the Roe-MUSCL scheme is spatially
for a regular grid finite difference model [51] has been refined
second-order accurate.
for the unstructured triangular grids employed herein. In the re-
fined algorithm, computational cells are divided into three types:
3.1.2. Upwind scheme for the sediment flux
active wet cells, active dry cells and inactive dry cells. The first
The sediment flux in the convective fluxes is obtained by mul-
two cell types are substantially included in the computational do-
tiplying the sediment concentration and the flow flux through
main, while the last cell type is removed temporarily out of the do-
the cell interface. In Fig. 1, SR and SL are the sediment concentra-
main in order to reduce the computer time. The detailed treatment
tions on the left and right sides of the edge lij , respectively. P and
of wetting and drying fronts in this model can be referred to this
Q are discharges per unit width across the edge lij in the x and y-
document [38].
directions, respectively. Therefore, the normal flow flux across
the edge, F f , equals Pnx þ Qny and the sediment fluxes on the left
3.1.6. Boundary conditions
and right sides of the edge can be expressed as F f  SL and F f  SR ,
The boundary conditions used in this model include two types,
respectively. An upwind scheme is used herein to determine the
i.e. closed and open boundaries. At a closed boundary, a free-slip
sediment flux across the edge, which is written as:
condition is used, i.e. hn ¼ 0; un ¼ 0; Skn ¼ 0 and qbkn ¼ 0, where
F s ¼ ½F f  SL þ F f  SR  jF f jðSR  SL Þ=2 ð16Þ the subscript n denotes the direction normal to the land or solid
wall boundary. For an open boundary, the values of h; u; v are eval-
uated using the Riemann invariants problem [24,37,43], Sk and qb k
3.1.3. Treatment of the source term
or needs to be specified by a user. A free outfall condition is applied
The source term in Eq. (9) includes the bed slope terms, friction
at the outflow boundary, which allows waves to travel across the
slope terms and additional terms associated with sediment trans-
boundary without reflection [5,8].
port and bed evolution. The treatment of these terms has a great
influence on the accuracy of the numerical solution. In a triangular
grid, the bed slopes in different directions can be readily computed 3.2. Discretization of bed evolution
since the three nodes of a triangle lie on the same plane, unlike the
vertices of a quadrangular grid. For the friction slope terms, an ex- After each time step, the bed change is calculated explicitly for
plicit discretization may cause numerical instability when the each sediment fraction. Eqs. (18.1) and (18.2) are used to calculate
water depth is very small [35,37]. In order to reduce the numerical the bed level changes due to the suspended and bed loads,
instability related to the friction slope terms, a semi-implicit dis- respectively.
cretization is adopted in the current model [38]. Dt n o
DZ sk ¼ ½ask xsk ðSk  Sk Þl þ ½ask xsk ðSk  Sk Þlþ1 ð18:1Þ
2q0
3.1.4. Method of time integration Dt n o
The Runge–Kutta method is usually used to increase the time- DZ bk ¼ ½abk xbk ðqbk  qbk Þl þ ½abk xbk ðqbk  qbk Þlþ1 ð18:2Þ
2q0
wise accuracy of a numerical scheme. In the current model, an
optimal TVD Runge–Kutta time stepping method is used to main- Thus the total thickness of bed deformation is obtained by combin-
tain stability of solution and obtain the second-order accuracy in ing the effects of suspended and bed loads.
time [47]. Denoting all the right-hand side terms of Eq. (13) as
LðUÞ, the method of time stepping is undertaken using the follow- 4. Model tests
ing equation:
 1
A series of model tests were undertaken to verify the numer-
Ulþ1 ¼ Ul þ DtL Ulþ2 ð17Þ
ical model outlined above, with the model predictions being
176 J. Xia et al. / Advances in Water Resources 33 (2010) 171–183

Fig. 2. Simulated results at t = 7.2 s: (a) water level and (b) velocity field.

compared with alternative numerical solutions and laboratory 4.2. Simulation of a dam-break flow in a converging-diverging channel
experimental data published in the literature. Four test cases
of dam-break flows were undertaken, including: (i) a partial Bellos et al. [53] conducted a series of dam-break flow experi-
dam-breach flow over a wet and fixed bed, (ii) a dam-break flow ments for various flow conditions. One of the experiments was
in a converging-diverging channel over a fixed bed, (iii) a dam- used herein as a test case. This experiment was undertaken in a
break flow over a mobile bed in a channel with a sudden fixed-bed flume, with a length of 21.2 m and a longitudinal slope
enlargement, and (iv) a partial dam-breach flow over a mobile of 0.6%. Within the flume, a channel of a converging–diverging plan
bed. shape was made to create two-dimensional effects, see Fig. 3. The
dam was positioned in the narrowest part ðx ¼ 0 mÞ of the flume,
where the width was 0.60 m. The dam separated initially upstream
4.1. A partial dam-breach flow over an initially wet and fixed bed and downstream regions and was then removed abruptly. Water
depths were measured by four probes located along the centerline
This test was first introduced by Fennema and Chaudhry [52] in of the channel. The initial water level in the upstream region was
a numerical method study which was considered as a classical 2D 0.3 m above the bottom level at the dam location and the initial
dam-break flow case. This test case is widely reported in the liter- water depth in the downstream region was zero (dry).
ature [24,35,36]. However, there is no analytical solution for this In this test case, a slip boundary condition was applied at the
case. The model domain was composed of a 200 m by 200 m basin upstream boundary and two sidewalls. At the downstream outlet,
over a flat, fixed and frictionless bed, with a thin-walled dam being a free outflow boundary condition was applied. The study domain
used to partition the basin into two equal-sized regions (see Fig. 2). was represented using 3213 unstructured cells, with the grid being
The water depths were 10 m and 5 m on the left and right sides of refined locally near the dam location, using an area constraint of
the dam wall, respectively. In the current study, the model domain about 0:005 m2 . The Manning n value was set to 0:012 m1=3 s,
was divided into 21,752 unstructured triangular cells, and the with a time step equal to 0.05 s. Fig. 4a–d shows the observed
computational mesh was refined around a breach site using an and calculated flow depths at the four measurement points. It
area constraint equal to about 0:5 m2 . A free outflow boundary can be seen that the simulated water depths agree generally well
condition was specified at the outlet boundary at x ¼ 200 m, with the measurements at the two upstream points (P1 and P2),
whereas at the remaining three sides a slip boundary condition with the minimum correlation coefficient reaching 0.998. The pre-
was assumed. At t ¼ 0 s, a 75 m wide breach centered at dicted mean depths at the two downstream points (P3 and P4) are
y ¼ 125 m was assumed to form instantaneously. A shock wave about 10% higher than the measurements, which is considered
was found to form and propagate downstream while a depression acceptable.
wave spread upstream. The predicted water level and velocity dis-
tributions at t = 7.2 s are shown in Fig. 2a and b, respectively, 4.3. Simulation of a dam-break flow in a mobile channel with a sudden
which agree closely with those obtained from the existing models enlargement
(see [24,35,36]).
A series of laboratory experiments of dam-break flow over mo-
Observation Points bile beds were carried out at the Civil Engineering Laboratory of
1.6 P1 (x=-8.50 m)
1.4 P2 (x=-4.00 m)
the Université catholique de Louvain, Belgium [7,19,54]. The test
1.2 case considered in this paper was conducted in a 6 m long flume
BD

P3 (x=+2.50 m)
1.0 P4 (x=+10.0 m)
with a non-symmetrical sudden enlargement from 0.25 m to
Free outflow
Y (m)

Open

0.8 P1 Wall
P2 P4
0.6 0.5 m width, located 1.0 m downstream of the gate (Fig. 5). The
0.4
Dam

P3 breaking of the dam was simulated by the rapid downward move-


0.2
0.0 ment of a thin gate at the middle of the flume. The sediment used
-0.2 was uniform coarse sand with a median diameter of 1.82 mm and a
-10 -8 -6 -4 -2 0 2 4 6 8 10 12 14
X (m)
density of 2680 kg=m3 , deposited with a bulk concentration of 53%.
The initial conditions consisted of a 0.1 m high horizontal layer of
Fig. 3. Sketch of a dam-break flow experiment in a converging-diverging channel. fully saturated sand over the whole flume and an initial layer of
J. Xia et al. / Advances in Water Resources 33 (2010) 171–183 177

0.30 0.30
(a) Location: P1 (b) Location: P2
0.25 Obs. 0.25 Obs.
Cal. Cal.

Water depth (m)


0.20 0.20
Water depth (m)

0.15 0.15

0.10 0.10

0.05 0.05

0.00 0.00
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Time(s) Time(s)

0.20 0.20
(c) Location: P3 (d) Location: P4

0.16 Obs. 0.16 Obs.


Cal. Cal.
Water depth (m)

Water level (m)


0.12 0.12

0.08 0.08

0.04 0.04

0.00 0.00
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Time(s) Time(s)

Fig. 4. Comparisons between the observed and calculated water depths.

X (m) model predictions were compared with the experimentally mea-


0.0 1.0 2.0 3.0 4.0 5.0 6.0 sured water levels at six points and bed levels at two cross-
0.6 sections.
CS1
CS2

Fig. 6 shows the model predicted and laboratory observed water


0.5 Observation Points 1-6
level time series at the measurement locations. For this case, the
non-equilibrium adaptation coefficient abk was set to 0.9. It can
Observation CS 1-2
0.4 be seen that with the mobile bed the predicted water level hydro-
Free outflow BD

graphs agree quite well with the observed data, and the correlation
Y (m)

5 6
0.3 coefficient reaches its minimum of about 0.6 at Point3, with this
value reaching more than 0.85 at any other points. The relatively
0.2 lower correlation coefficient at Point3 could be caused by the
Dam

over-scouring of the channel bed at this site.


Initial depth = 25cm
0.1 1 2 3 4 However, the predicted final bed levels agree less closely with
the experimental data. Fig. 7a and b shows the comparisons be-
0.0 tween the predicted and observed bed level profiles at two mea-
sured cross-sections. At CS1 (Fig. 7a), the predicted maximum
Fig. 5. Sketch of a dam-break flow experiment over a mobile bed.
bed scouring depth agrees well the observed value, but the deposi-
tion process is not accurately predicted near the right side wall,
and thereby the predicted mean bed level is 3.7% lower than the
0.25 m clear-water upstream of the gate. The model domain was observed one. At CS2 (Fig. 7b), the lateral bed profile is predicted
divided into 8156 unstructured triangular cells, with the mesh correctly, but the maximum scouring depth is over-estimated by
being refined just downstream of the dam, using an area constraint about 1 cm. Besides, the deposition depth near the sidewall is
of about 2 cm2 . A free-slip boundary condition was applied at all underestimated by about 5%. This may be due to a recirculation
side walls, with a free outflow boundary condition being used at zone, which was observed visually near the glass wall of the chan-
the downstream outlet. A Manning n value equal to 0:025 m1=3 s nel [54], with the current depth-averaged model being unable to
was set, and this value refers to that used in the publication of a predict vertical velocities near the wall. At CS2, the predicted mean
similar study [7]. In a 1D modelling study of dam-break flows by bed level is 5.8% lower than the observed one.
Zech et al. [7], a Manning roughness of 0:026 m1=3 s was used in The low prediction accuracy in bed level changes could be due
an experiment using the same flume and with the same sediment. to the inaccuracy in estimating the value of qbk , which was mainly
Due to the relatively coarse bed material and relatively small used in natural alluvial rivers with large water depths and subcrit-
velocity of dam-break flow, in this case study the bed evolution ical flows. In this test case, the water depths downstream of the
was considered to be caused only by the bed-load transport. The dam were less than 20 cm and changed rapidly. The effect of differ-
178 J. Xia et al. / Advances in Water Resources 33 (2010) 171–183

0.22 0.22
Exp.at P1 Exp.at P2
0.20 Cal.at P1 0.20 Cal.at P2
0.18 0.18
Z (m)

Z (m)
0.16 0.16
0.14 0.14
0.12 0.12
0.10 0.10
0.08 0.08
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time (s) Time (s)

0.22 0.22 Exp.at P4


Exp.at P3
0.20 0.20 Cal.at P4
Cal.at P3
0.18 0.18
Z (m)

Z (m)
0.16 0.16
0.14 0.14
0.12 0.12
0.10 0.10
0.08 0.08
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time (s) Time (s)

0.22 0.22
Exp.at P5 Exp.at P6
0.20 0.20 Cal.at P6
Cal.at P5
0.18 0.18
Z (m)
Z (m)

0.16 0.16
0.14 0.14
0.12 0.12
0.10 0.10
0.08 0.08
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time (s) Time (s)

Fig. 6. Comparisons between the observed and calculated water levels.

ent values of bed-load transport adaption coefficient, abk , on the Fig. 9a and b shows a comparison between the observed and
bed level changes was also investigated. It was found that the pre- calculated cross-sectional profiles after 20 s. These figures show
dicted bed level was moderately sensitive to abk ; as the value of abk that the model predictions agree well with the observed profiles.
varied between 0.8 and 1.0, the maximum difference in the pre- In this case when the parameter b was set to 0.5 and the corre-
dicted lowest bed levels was about 0.3 cm. sponding values of ask for the six fractions were 6.19, 3.60, 2.45,
1.73, 1.28 and 1.05 in case of deposition, respectively. At CS1
ðx ¼ 2:5 mÞ, the predicted cross-sectional bed profile is similar to
4.4. Simulation of a partial dam-breach flow in a mobile channel the measured one, with the maximum scour depth being underes-
timated by about 1.7 cm, or 13%, and the predicted mean bed level
The model was also applied to a study of partial dam-breach is about 35% higher than the observed one. At CS2 ðx ¼ 3:5 mÞ, the
flow experiments over a mobile bed, carried out at the Hydraulics model predicted the erosion pattern in the central channel gener-
Laboratory of Tsinghua University, China. These experiments were ally well, with the maximum scour depth being over-estimated by
conducted in a 18.5 m long flume with a 1.6 m width (See Fig. 8). A about 2 cm, and with the mean bed level being about 0.6 cm or 60%
thin-walled dam was located about 2.0 m downstream of the inlet, lower than the observed one. This is thought to be caused by the
and the channel bed was mobile only in a 4.5 m long reach, starting simple turbulence model used, which may not be able to generate
from the dam site. Initially, the water depth was set to 40 cm in the the rapid formation of horizontal circulating flow downstream of
reservoir and 12 cm downstream of the dam. At t ¼ 0 s, a 20 cm the dam.
wide breach centered at y ¼ 0:80 m forms instantaneously, and a
shock wave propagated downstream along the mobile channel.
4.5. Summary of parameters and error analysis results
The mobile-bed material was composed of non-uniform coal ash,
with a median diameter of about 0.135 mm. The natural density
A summary of the key parameters used for all of the test and
and dry density were about 2248 and 720 kg=m3 , respectively.
application cases is shown in Table 1, and the results from error
The measured non-uniform bed material was represented by six
analysis are shown in Table 2.
fractions of 0.022, 0.054, 0.103, 0.193, 0.375 and 0.750 mm, and
percentages of these fractions were 9.0%, 18.7%, 22.8%, 14.4%,
27.5% and 7.7%, respectively. In this experiment, the bed levels at 5. Further applications
sections after 20s were measured by the ultra-acoustic topographic
surveying meter. The measurements of water depth and velocity The objective of this case study was to investigate the influence
variations were not conducted due to the use of coal ash. of different bed material compositions on dam-breach flow
J. Xia et al. / Advances in Water Resources 33 (2010) 171–183 179

0.15 Initial (a) CS1 at X=4.1m 0.02 (a) CS1 (x=2.5m)

0.14 Exp.
Cal. -0.01
0.13

0.12 -0.04
Zb (m)

Zb (m)
0.11 -0.07
z
0.10
-0.10
0.09 Ini.
-0.13 Exp.
0.08
Cal.
0.07 -0.16
0 0.1 0.2 0.3 0.4 0.5 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Y (m) Y (m)

0.02 (b) CS2 (x=3.5m)


0.15 (b) CS2 at X=4.4m
Initial
0.14 Exp. -0.01
Cal.
0.13
-0.04

Zb (m)
0.12
Zb (m)

0.11 -0.07

0.10 -0.10
Ini.
0.09 Exp.
-0.13
0.08 Cal.

0.07 -0.16
0 0.1 0.2 0.3 0.4 0.5 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Y (m) Y (m)

Fig. 7. Comparisons between the observed and calculated cross-sectional profiles. Fig. 9. Comparisons between the observed and calculated cross-sectional profiles.

induced morphodynamic processes. The model domain was com- water depth was set to 5 m, and on the right side of the wall it is
posed a 2000 m by 2000 m basin of a flat bed, with a wall being in- assumed to be dry. A free outflow boundary was assumed at
serted in the centre to partition the basin into two regions. A 40 m x ¼ 2000 m, whereas the other boundaries were assumed to be
wide dam was assumed to be located in the centre of the wall. wall boundaries. This domain was divided into 20,654 triangular
Fig. 10 shows a sketch of the model domain, in which two observa- cells, and the computational grid resolution was refined around
tion points are also marked. Initially on the left side of the wall the the breach site, with the minimum cell area being about 81:8 m2 .

1.8 2m 4.5m 12m


1.6
1.4
Free outflow BD

1.2
Coal ash

1.0
y (m)

Water

0.8 Non-erodible bed


0.6
0.4
0.2
0.0
CS1 CS2
-0.2
-2 0 2 4 6 8 10 12 14 16 18 20
x (m)

Fig. 8. Sketch of a dam-breach flow experiment over a mobile bed.

Table 1
Parameters used in test and application cases.

No. Parameters Values used


Test Case 1 Test Case 2 Test Case 3 Test Case 4 Application case
1 h in mt (turbulent viscosity coefficient) 0.5 0.5 0.5 0.5 0.5
2 n (Manning n) 0.00 0.012 0.025 0.015 0.015
3 b in ask (non-equilibrium adaptation coefficient of suspended load) 0.50 0.50
4 abk = (non-equilibrium adaptation coefficient of bed load) 0.90
5 / in DP k (percentage of sediment transport capacity) 0.90 0.90
6 kb in Eq. (7) (transport capacity of bed load) 0.10
180 J. Xia et al. / Advances in Water Resources 33 (2010) 171–183

Table 2
Error analysis results.

Tests Sources Descriptions of error analysis for each test case


Test 1 Fennema and Chaudhry [52] No analytical solution, the predicted water level and velocity distributions agree closely with those obtained from the
existing models
Test 2 Bellos et al. [53] The correlation coefficients reach 0.998 at the two upstream measurement points. The predicted mean depths at the
two downstream measurement points are about 10% higher than the measurements
Test 3 Palumbo et al. [54] The correlation coefficient between the predicted and observed water level hydrographs ranges from 0.6 to 0.9. The
predicted mean bed level is 3.7% lower than the observed one at CS1, and is 5.8% lower than the observed one at CS2
Test 4 Tsinghua experiment The predicted mean bed level is about 1.7 cm or 35% higher than the observed one at CS1, and is about 0.6 cm or 60%
lower than the observed one at CS2

At t ¼ 0 s, it was assumed that a 200 m wide breach at the centre 25.2%, respectively. A constant Manning n value equal to
forms instantaneously. The model was used to simulate the mor- 0:015 m1=3 s was assumed for these cases.
phodynamic processes in this basin to investigate the impact of
bed material compositions on the flood flow characteristics. Three 5.1. Water level and bed level variations downstream of the dam
scenarios were considered, including: a fixed bed (Case 0), a bed
made up of uniform sediment (Case 1) and a bed made up of Fig. 11a and b shows the variations of water and bed levels at
non-uniform sediments (Case 2). A constant mobile layer thickness sites P1 and P2. It can be seen from these two figures that the pre-
of 3.0 m was used in both Cases 1 and 2. In Case 1, the size of uni- dicted water and bed levels for Case 1 are different from those for
form sediment grain was set to 0.05 mm. In Case 2, the bed mate- Case 2. After the occurrence of dam-breach, the greatest flow
rial in the mobile layer was divided into four fractions, including: velocity occurred near the dam-breach site, which produced a very
0.005–0.025 mm, 0.025–0.050 mm, 0.050–0.100 mm and 0.100– large sediment transport capacity. Therefore, the thickness of bed
0.250 mm, with the median diameter also being 0.05 mm. The scouring at P1 for Case 1 was greater than that for Case 2. At P2,
mass percentages of these fractions were 7.0%, 43.0%, 24.6% and the scouring thickness for Case 1 was less than that for Case 2 be-
cause the bed material could be entrained as the suspended load in
Case 1 more quickly than in Case 2. Such a phenomenon can be ex-
plained by the predicted water levels at P1 and P2. At P1, the water
2200
level at t ¼ 1200 s for Case 1 was 1.45 m higher than that for Case
2000 2, while the water level at the same time for Case 2 was 0.85 m
1800 lower than that for the fixed bed case. At the site of P2, the water
levels for Case 1 and for Case 2 were very close before 1200 s, and
1600
the averaged water level for Case 1 was about 10 cm higher than
1400 that for Case 2 after 1200 s. The maximum scour depths for Case
initial depth = 5m

free outflow BD

1200 1 and Case 2 were 1.47 m and 1.18 m at P1 after 3600 s, respec-
P1 tively, while they were 0.26 m and 0.38 m at P2, respectively. In
dry bed

P2
y (m)

200m

1000
addition, there existed a significant difference in the speed of flood
800 wave propagation between the two mobile bed cases and the fixed
600 bed case. At the location of P1, the water levels peaked at t ¼ 240 s
for the fixed bed, at t ¼ 1260 s for the uniform bed material and at
400
t ¼ 1680 s for the non-uniform bed material, respectively. This dif-
200 ference in the speed of flood wave propagation between different
0 bed material compositions was caused by the effects of sediment
transport and bed evolution on the movement of flow, which led
-200
-200 0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200 to different water depths and corresponding speeds of flood wave
x (m) propagation.
The rates of bed evolution and water depth change can be ob-
Fig. 10. Sketch of a partial dam-breach flow. tained from Fig. 11a and b. For example, at P1 for Case 2, the mean

a 2.0 b 1.5
1.6
1.2 1.1
Water / Bed level (m)

Water / Bed level (m)

0.8 Case 0
Water

Case 0
Water

0.7 Case 1
0.4 Case 1
Case 2
Case 2
0.0 Case 0
Bed

Case 0 0.3 Case 1


Bed

-0.4 Case 1
Case 2
Case 2
-0.8
-0.1
-1.2
-1.6 -0.5
0 600 1200 1800 2400 3000 3600 0 600 1200 1800 2400 3000 3600
Time (s) Time (s)

Fig. 11. Water level and bed level variations downstream of the dam for (a) P1 and (b) P2.
J. Xia et al. / Advances in Water Resources 33 (2010) 171–183 181

rate of bed scouring during the period between t ¼ 240 s and of the maximum scour depth (1.18 m) was at 200 m downstream
t ¼ 1320 s was about 0:65  103 m=s, while the mean rate of of the dam, with the scour depth close to the downstream bound-
water depth decrease during the same period was about ary being about 15 cm. The scour rate after 1200 s was relatively
0:45  103 m=s. Since the rate of bed evolution was comparable slow due to the decrease of flow intensity and the armouring of
to the rate of water depth variation, it was necessary to use the tur- bed material. It was also found that the upstream slope was usually
bid water equations to simulate the morphodynamic processes in greater than the downstream one in a scour hole for both cases. In
such flows. order to test grid convergence, two additional sets of computa-
tional meshes were used. The coarser mesh comprised 5854 trian-
5.2. Sediment concentration distributions gular cells, while the finer mesh comprised 39,149 triangular cells.
For Case 1, the maximum scour depths after 1 h were 1.67 m for
During the process of bed scouring, the variation of sediment the coarser mesh and 1.74 m for the finer mesh, respectively. For
concentration is usually characterized by an increasing tendency Case 2, the maximum scour depths were 1.25 m for the coarser
toward the outlet boundary. Fig. 12a and b shows the distributions mesh and 1.21 m for the finer mesh, respectively. Therefore, differ-
of sediment concentrations along the channel centerline for Cases ent mesh sizes had a slight effect on the maximum scour depth for
1 and 2. For Case 1, the concentration at the downstream boundary each case.
was 59:2 kg=m3 at t ¼ 600 s, while it decreased to 19:6 kg=m3 at The planar shape of the scour hole for Case 1 is different from
t ¼ 3600 s. For Case 2, the distributions of sediment concentrations that for Case 2. The planar shape of scour hole was approximately
along the channel centerline were more complicated due to the ef- elliptical for the former, while it was almost circular for the latter.
fect of non-uniform bed material composition. The concentrations Fig. 14a and b shows, respectively, the contours of bed levels after
were generally higher, with the values at the downstream bound- 3600 s for Cases 1 and 2. The erosion extent for Case 1 was less
ary at t ¼ 600 and 3600 s being 93.9 and 26:4 kg=m3 , respectively. than that for Case 2, while the maximum erosion depth for Case
1 was greater than that for Case 2. In addition, the lateral range
5.3. Evolution processes of scour holes was 720 m for the elliptical erosion, and 1600 m for the circular
erosion. The scour depth was 0.68 m at the centre of the ellipse,
For different bed material compositions, there exists a signifi- while it was 0.36 m at the centre of the circle. During the routing
cant difference in the evolution of scour hole downstream of the of the dam-break wave, the flow from the dam-breach diffused to-
dam, as shown in Fig. 13a and b. For Case 1, the site of the maxi- wards the circumference, with the largest velocity occurring at a
mum scour depth (1.72 m) was located at 120 m downstream of point along the channel centerline. In Case 1, the bed was made
the dam, and the scour depth close to the outlet boundary was very up of uniform fine sediment, the value of sediment transport
small, with a mean scour depth of 0.001 m. For Case 2, the location capacity was proportional to the velocity. This led to the largest

a 100 b 100

80 80 t=0s
Concentration (kg/m3)

Concentration (kg/m3)

t=0s t=600s
t=600s
t=1200s t=1200s
60 t=1800s 60 t=1800s
t=2400s t=2400s
t=3000s t=3000s
t=3600s t=3600s
40 40

20 20

0 0
600 800 1000 1200 1400 1600 1800 2000 600 800 1000 1200 1400 1600 1800 2000
Distance along the channel (m) Distance along the channel (m)

Fig. 12. Variations of sediment concentrations along the channel centerline for (a) Case 1 and (b) Case 2.

a 0.2 b 0.2
0.0 0.0
-0.2 -0.2
-0.4 -0.4
Bed Level (m)
Bed Level (m)

-0.6 -0.6
-0.8 -0.8
-1.0 t=0s -1.0 t=0s
t=600s t=600s
-1.2 t=1200s -1.2 t=1200s
t=1800s t=1800s
-1.4 t=2400s -1.4 t=2400s
-1.6 t=3000s -1.6 t=3000s
t=3600s t=3600s
-1.8 -1.8
600 800 1000 1200 1400 1600 1800 2000 600 800 1000 1200 1400 1600 1800 2000
Distance along the channel (m) Distance along the channel (m)

Fig. 13. Variations of longitudinal bed profiles along the channel centerline for (a) Case 1 and (b) Case 2.
182 J. Xia et al. / Advances in Water Resources 33 (2010) 171–183

a 1800 Scour depth(m) b 1800 Scour depth(m)


1700 -0.1
-0.2 1700 -0.1
1600 -0.4 -0.2
-0.6 1600 -0.4
1500 -0.8 -0.6
-1 1500 -0.8
1400 -1.2 -1
1400 -1.2 -0.1
-1.4
1300 -1.6 -0.1 -1.4
1300 -1.6 2
-0.

.1
1200

-0
-0.4
2
-0. -0.6 1200
1100 -1
Y(m)

-1

-1.4 1100

Y(m)

.2
-0.2

-0
-0.2

1000

-0.6
-1
.4

1000
-1
-0.1

-1.2
900 -0.8 900 -0.8 -0.4
-0.4 -0.2 -0.6
800 800
-0.1
700 700 -0.2
600 600
-0.1
500 500
400 400
300 300
200 200
800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000
X (m) X (m)

Fig. 14. Contours of bed levels after 3600 s for (a) Case 1 and (b) Case 2.

bed scour depth occurring close to the centerline and its value de- (i) for a dam-break induced flow at the initial stage, the rate of
creased away from the centerline. In Case 2, the bed was composed bed evolution is comparable to the rate of water depth var-
of graded sediments. All of the sediment fractions close to the iation near the dam site;
channel centerline were scoured due to the relatively large flow (ii) for mobile beds, the erosion extent over the uniform sedi-
velocity, while the flow velocity reduced away from the centerline, ment bed is less than that over the non-uniform sediment
the finer sediment fractions were still scoured, which led to a bed, while the maximum erosion depth obtained over the
wider planar shape of the scour hole. For Case 2, the composition former is greater than that over the latter; and
of bed material had a tendency of armouring due to bed scouring, (iii) the planar shape of the scour hole is approximately elliptical
and the corresponding value of sediment transport capacity de- over the uniform sediment bed and it is almost circular over
creased gradually. However, for Case 1, the bed material composi- the non-uniform sediment bed, which indicates increased
tion did not change during the process of bed evolution, since it erosion in the lateral direction.
had no effect on the value of sediment transport capacity. There-
fore, the maximum erosion depth in Case 1 was greater than that Acknowledgements
in Case 2.
The research reported in this paper was conducted as part of the
6. Concluding remarks Flood Risk Management Research Consortium (Phase II), supported
by the UK Engineering and Physical Sciences Research Council (GR/
In this paper, details are given regarding a 2D morphodynamic S76304). The National Basic Research Program of China
model for predicting dam-break flows over mobile beds. The model (2007CB714102) ‘‘Mechanisms of dam and levee breaking under
employs the shallow water equations for turbid waters, which take complex uncertainties and theories of risk mitigation and control”
into account the effects of sediment density and bed level changes is also gratefully acknowledged. Another National Basic Research
on the flood flows. In addition, the model is capable of simulating Program of China (2006CB403303) is appreciated for offering the
the transport of graded suspended load and bed load, and the experimental data of the forth test case in the model verification.
adjustment of bed material composition. A finite volume method Lastly, the authors thank very much the reviewers for their con-
is used to solve the governing equations. The model adopts an structive comments that improved the paper.
unstructured triangular mesh and is second-order accurate in both
time and space. A coupled approach has been used to solve simul-
References
taneously the flow and sediment transport processes induced by
dam-breaks, in which flood wave fronts move rapidly and lead to [1] Stoker JJ. Water waves. Pure and applied mathematics, vol. 4. New
significant bed level changes. York: Interscience Publishers; 1957.
The model was verified against the predictions from existing [2] Chanson H. Analytical solution of dam break wave with flow resistance:
application to tsunami surges. In: Jun BH, Lee SI, Seo IW, Choi GW, editors.
numerical models and laboratory experimental data published in Proceedings of the 31 IAHR biennial congress, Seoul, Korea; 2005. p. 3341–53.
the literature, including two test cases of flow over fixed beds [3] Lin GF, Lai JS, Guo WD. Finite-volume component-wise TVD schemes for 2D
and two cases of flow over mobile beds. It was then used to inves- shallow water equations. Adv Water Res 2003;26:861–73.
[4] Fagherazzi S, Sun T. Numerical simulations of transportational cyclic steps.
tigate the influences of different bed material compositions on the Comput Geosci 2003;29:1143–54.
hydrodynamics and morphology of a channel. It is demonstrated [5] Liao CB, Wu MS, Liang SJ. Numerical simulation of a dam break for an actual
that the model is capable of simulating the interactive processes river terrain environment. Hydrol Process 2007;21:447–60.
[6] Wu WM, Wang SSY. One-dimensional modeling of dam-break flow over
between the water flow, sediment transport and morphological
movable beds. ASCE J Hydraulic Eng 2007;133(1):48–58.
changes caused by dam-break flows for different bed material [7] Zech Y, Soares-Frazão S, Spinewine B. Dam-break induced sediment
compositions. From the model predictions it has been found that movement: experimental approaches and numerical modeling. J Hydraulic
the behavior of dam-break flows over a mobile bed is significantly Res 2008;46(2):176–90.
[8] Zhou JG, Causon DM, Mingham CG, Ingram DM. Numerical prediction of dam-
different from that over a fixed bed. The main findings from these break flows in general geometries with complex bed topography. ASCE J
tests are that: Hydraulic Eng 2004;130(4):332–40.
J. Xia et al. / Advances in Water Resources 33 (2010) 171–183 183

[9] Liang DF, Lin BL, Falconer RA. A boundary-fitted numerical model for flood [32] Wu WM. Depth-averaged 2D numerical modeling of unsteady flow and
routing with shock-capturing capability. J Hydrol 2007;332:477–86. nonuniform sediment transport in open channels. ASCE J Hydraulic Eng
[10] Begnudelli L, Sanders BF. Conservative wetting and drying methodology for 2004;130(10):1013–24.
quadrilateral grid finite-volume models. ASCE J Hydraulic Eng [33] Wu BS, Long YQ. Corrections for the formula of sediment transport capacity in
2007;133(3):312–22. the Yellow River. Yellow River 1993(7):1–4 [in Chinese].
[11] Gallegos HA, Schubert JE, Sanders BF. Two-dimensional, high-resolution [34] Wang GQ, Xia JQ, Wu BS. Numerical simulation of longitudinal and lateral
modeling of urban dam-break flooding: a case study of Baldwin Hills, channel deformations in the braided reach. ASCE J Hydraulic Eng
California. Adv Water Res 2009;32:1323–35. 2008;134(8):1064–78.
[12] Costa JE, Schuster RL. The formation and failure of natural dams. Geol Soc Am [35] Caleffi V, Valiani A, Zanni A. Finite volume method for simulating extreme
Bull 1988;100(7):1054–68. flood events in natural channels. IAHR J Hydraulic Res 2003;41(2):
[13] Capart H, Young DL, Zech Y. Dam-break Induced debris flow and particulate 167–77.
gravity currents. In: Kneller B, McCaffrey B, Peakall J, Druitt T, editors. Special [36] Wang JW, Liu RX. A comparative study of finite volume methods on
publication of the international association of sedimentologists, vol. 31; 2001. unstructured meshes for simulation of 2D shallow water wave problems.
p. 149–56. Math Comput Simul 2004;53:171–84.
[14] Kale VS, Ely LL, Enzel Y, Baker VR. Geomorphic and hydrologic aspects of mon- [37] Yoon TH, Kang SK. Finite volume model for two-dimensional shallow water
soon floods on the Narmada and Tapi Rivers in central India. Geomorphology flows on unstructured grids. ASCE J Hydraulic Eng 2004;130(7):678–88.
1994;10:157–68. [38] Xia JQ, Falconer RA, Lin BL, Wang GQ. Modelling floods routing on initially dry
[15] Zhang RJ, Xie JH. Sedimentation research in China. Beijing: China Water and beds with the refined treatment of wetting and drying. Environ Modell Softw,
Power Press; 1993. submitted for publication.
[16] Ferreira R, Leal J. 1D mathematical modeling of the instantaneous dam-break [39] Liang L, Ni JR, Borthwick AGL, Rogers BD. Simulation of dike-break processes in
flood wave over mobile bed: application of TVD and flux-splitting schemes. In: the Yellow River. Sci Chin (Ser E) 2002;45(6):606–19.
Proceedings of the european concerted action on dam-break modeling, [40] Godunov SK. A difference method for the numerical calculation of
Munich; 1998. p. 175–222. discontinuous solutions of hydrodynamic equations. Matemsticheskly
[17] Fraccarollo L, Armanini A. A semi-analytical solution for the dam-break Sboraik 47, (US joint publications research service); 1959.
problem over a movable bed. In: Proceedings of the european concerted action [41] Fraccarollo L, Toro EF. Experimental and numerical assessment of the shallow
on dam-break modeling, Munich; 1998. p. 145–52. water model for two-dimensional dam-break type problems. IAHR J Hydraulic
[18] Fraccarollo L, Capart H. Riemann wave description of erosional dam-break Res 1995;33(6):843–64.
flows. J Fluid Mech 2002;461:183–228. [42] Toro EF. Shock-capturing methods for free-surface shallow flows. John Wiley
[19] Spinewine B, Zech Y. Small-scale laboratory dam-break waves on movable and Sons; 2001. 326p.
beds. IAHR J Hydraulic Res 2007;45(Extra Issue):73–86. [43] Sleigh PA, Gaskell PH, Berzins M, Wright NG. An unstructured finite-volume
[20] Cao Z, Pender G, Wallis S, Carling P. Computational dam-break hydraulics over algorithm for predicting flow in rivers and estuaries. Comput Fluids
mobile sediment bed. ASCE J Hydraulic Eng 2004;130(7):689–703. 1998;27(4):479–508.
[21] Cao Z, Li Y, Yue Z. Multiple time scales of alluvial rivers carrying suspended [44] Roe PL. Approximate Riemann solvers, parameter vectors and difference
sediment and their implications for mathematical modeling. Adv Water Res schemes. J Comput Phys 1981;43:357–72.
2007;30(4):715–29. [45] Van Leer B. Towards the ultimate conservative difference scheme. V. A.
[22] Hu P, Cao Z. Fully coupled mathematical modeling of turbidity currents over second-order sequel to Godunov’s method. J Comput Phys 1979;32:101–36.
erodible bed. Adv Water Res 2009;32(1):1–15. [46] Roe PL, Baines MJ. Algorithms for advection and shock problems. In: Viviand H,
[23] Fraccarollo L, Toro EF. Experimental and numerical assessment of the shallow editor. Proceedings of the fourth GAMM conference on numerical methods in
water model for two-dimensional dam-break type problems. IAHR J Hydraulic fluid mechanics. Braunschweig: Vieweg; 1981. p. 281–90.
Res 1995;33(6):843–64. [47] Tan WY. Shallow water hydrodynamics. New York: Elsevier; 1992.
[24] Zhao DH, Shen HW, Lai JS, Tabios III GQ. Approximate Riemann solvers in FVM [48] Zhao DH, Shen HW, Tabios III GQ, Lai JS, Tan WY. Finite-volume two-
for 2D hydraulic shock wave modelling. ASCE J Hydraulic Eng dimensional unsteady flow model for river basins. ASCE J Hydraulic Eng
1996;122(12):692–702. 1994;120(7):863–83.
[25] Simpson G, Castelltort S. Coupled model of surface water flow, sediment [49] Brufau P, Garcia-Navarro P, Vazquez-Cendon ME. Zero mass error using
transport and morphological evolution. Comput Geosci 2006;32:1600–14. unsteady wetting-drying conditions in shallow flows over dry irregular
[26] Cao Z. Comments on the paper by Guy Simpson and Sebastien Castelltort topography. Int J Numer Methods Fluids 2004;45:1047–82.
coupled model of surface water flow, sediment transport and morphological [50] Bradford SF, Sanders BF. Finite-volume model for shallow water flooding of
evolution. Comput Geosci 2007;33:976–8. arbitrary topography. ASCE J Hydraulic Eng 2002;128(3):289–98.
[27] Xie JH. River modelling. Beijing: China Water and Power Press; 1990 [in [51] Falconer RA, Chen Y. An improved representation of flooding and drying and
Chinese]. wind stress effects in a 2D tidal numerical model. Proc Inst Civil Eng
[28] Yue ZY, Cao ZX, Li X, Che T. Two-dimensional coupled mathematical modeling 1991;2(2):659–72.
of fluvial processes with intense sediment transport and rapid bed evolution. [52] Fennema RJ, Chaudhry MH. Explicit methods for 2D transient free-surface
Sci Chin (Ser G) 2008;51(9):1427–38. flows. ASCE J Hydraulic Eng 1990;116(8):1013–34.
[29] Zhou JJ, Lin BN. One-dimensional mathematical model for suspended sediment [53] Bellos V, Soulis JV, Sakkas JG. Experimental investigations of two-dimensional
by lateral integration. ASCE J Hydraulic Eng 1998;124(7):712–7. dam-break-induced flows. IAHR J Hydraulic Res 1992;30(1):47–63.
[30] Wang GQ, Xia JQ. Channel widening during the degradation of alluvial rivers. [54] Palumbo A, Soares-Frazão S, Goutiere L, Pianese D, Zech Y. Dam-break flow on
Int J Sediment Res 2001;16(2):139–49. mobile bed in a channel with a sudden enlargement. In: Altinakar MS,
[31] Dou XP, Li TL, Dou GR. Numerical model of total sediment transport in the Kokpinar MA, Aydin I, et al., editors. Proceedings of river flow, Turkey; 2008, p.
Yangtze Estuary. Chin Ocean Eng 1999;13(3):277–86. 645–54.

You might also like