You are on page 1of 15

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/285824285

Discontinuous Galerkin Method for 1D Shallow Water Flows in Natural Rivers

Article  in  Engineering Applications of Computational Fluid Mechanics · March 2012


DOI: 10.1080/19942060.2012.11015404

CITATIONS READS

30 182

2 authors:

Wencong Lai Abdul A. Khan


Clemson University Clemson University
20 PUBLICATIONS   240 CITATIONS    111 PUBLICATIONS   1,309 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

parallel flood modeling View project

CI-WATER View project

All content following this page was uploaded by Abdul A. Khan on 05 February 2016.

The user has requested enhancement of the downloaded file.


This article was downloaded by: [177.234.0.174]
On: 26 August 2015, At: 22:53
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick
Place, London, SW1P 1WG

Engineering Applications of Computational Fluid


Mechanics
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tcfm20

Discontinuous Galerkin Method for 1D Shallow Water


Flows in Natural Rivers
a a
W. Lai & A. A. Khan
a
Department of Civil Engineering, Clemson University, Lowry Hall, Clemson, SC
29634-0911, USA
Published online: 19 Nov 2014.

To cite this article: W. Lai & A. A. Khan (2012) Discontinuous Galerkin Method for 1D Shallow Water Flows in Natural
Rivers, Engineering Applications of Computational Fluid Mechanics, 6:1, 74-86, DOI: 10.1080/19942060.2012.11015404

To link to this article: http://dx.doi.org/10.1080/19942060.2012.11015404

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of
the Content. Any opinions and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied
upon and should be independently verified with primary sources of information. Taylor and Francis shall
not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other
liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 1, pp. 74–86 (2012)

DISCONTINUOUS GALERKIN METHOD FOR 1D SHALLOW WATER


FLOWS IN NATURAL RIVERS
W. Lai and A. A. Khan *

Department of Civil Engineering, Clemson University, Lowry Hall, Clemson, SC 29634-0911, USA
* E-Mail: abdkhan@clemson.edu (Corresponding Author)

ABSTRACT: A numerical model is proposed for the solution of one-dimensional shallow water flow equations for
natural rivers. This model is based on the total variation diminishing Runge-Kutta discontinuous Galerkin finite
element method. In natural rivers, the cross-section shape and bed slope can be quite irregular, which requires a
compatible discretization scheme for the bed slope term and net pressure force term. Therefore, in this model, the
hydrostatic pressure force term and the wall pressure force term are combined and a new discretization for the
resulting term is introduced. This formulation is shown to prevent unphysical flow due to improper treatment of
bottom slope term. The mass and momentum flux term are calculated by HLL Riemann solver. A scheme is
presented to model flow over dry bed. To evaluate the numerical scheme, tests are conducted for idealized dambreak
problems in parabolic (with wet and dry beds), and rectangular channels, hydraulic jump in a rectangular channel,
dambreak in the Teton River (Idaho, USA) and the Toce River (Northern Alps, Italy), and flooding event in the East
Downloaded by [177.234.0.174] at 22:53 26 August 2015

Fork River (Wyoming, USA). The comparison of the computational results with analytical and laboratory results of
dam break flows shows that the model is capable of handling flow over dry areas. The simulation results for
hydraulic jump show the discharge conservation property and shock prediction capability of the model. The
dambreak and flood simulations in natural channels show that the model is capable of handling flows in highly
varying bed topography and channel geometry.
Keywords: Runge-Kutta discontinuous Galerkin, total variation diminishing, shallow water flows, Saint-Venant
equations, natural rivers

1. INTRODUCTION many numerical schemes have been developed to


solve the Saint-Venant equations (Capart et al.,
Numerical modeling of natural river flows is of 2003; Cattella et al., 2008). The difficulties in
paramount importance in river management, modeling these equations come from the
floodplain delineation, habitat assessment, water treatment of convective flux and source terms
supply, morphological changes, et cetera. Apart (Garcia-Navarro and Vázquez-Cendón, 2000).
from the one-dimensional schemes, two- More, careful treatment of the source terms of bed
dimensional models have been extensively elevation change and channel width change is
developed and recently three-dimensional models also required to preserve the conservative
have become a viable tool for river flow modeling property of the system (Nujić, 1995; Zhou et al.,
(Tritthart and Cutknecht, 2007; Ying et al., 2009; 2001).
Erpicum et al., 2010). However, the 2D and 3D Since the shallow water flow equations are
models are computationally expensive and hyperbolic, a discontinuous solution may be
detailed topographic information is required generated even if the initial and boundary
(Catella et al., 2008). For practical applications, conditions are smooth. To achieve accurate
the prediction of discharge and water level approximations for the flux terms in solving
profiles of unsteady flows along a river reach hyperbolic conservation laws, exact and various
would provide enough information for approximate Riemann solvers can be utilized.
engineering needs. Therefore, one-dimensional Godunov (1959) used the exact formulation of
models would remain a practical tool for Riemann problems to introduce a discontinuous
modeling natural river flows, especially for river solution for the flux terms. Other methods include
flow simulation over long reaches and for/during Roe flux (Roe, 1981), HLL flux (Harten et al.,
extended times. 1983), HLLE flux (Einfeldt, 1988), weighted
The governing equations for the one-dimensional average flux method (Toro, 1989), and HLLC
shallow water flows are represented well by the flux (Toro et al., 1994). Rhebergen et al. (2008)
Saint-Venant equations. Over recent decades, introduced a numerical flux (NCP-flux) for non-

Received: 12 Mar. 2011; Revised: 18 Jul. 2011; Accepted: 5 Sep. 2011

74
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

conservative terms in the hyperbolic system of positivity-preserving limiter to handle dry bed and
equations. Improper treatment of source terms in still water for shallow water flows.
numerical modeling of Saint-Venant equations In this paper, the total variation diminishing
may result in non-physical flow in the Runge-Kutta discontinuous Galerkin method is
computational domain. Several methods have applied to solve the Saint-Venant equations. In
been proposed to appropriately model the source the studies cited above, the RKDG method was
term, for example, Nujić (1995), Garcia-Navarro applied to rectangular channels. In this study, the
and Vázquez-Cendón (2000), Perthame et al. RKDG method is used to model flow in natural
(2001), Zhou et al. (2001), Ying et al. (2004), and channels, where the hydrostatic pressure force and
Catella et al. (2008). the wall pressure force discretization has to be
In this paper, the Runge-Kutta discontinuous compatible with that of the gravity force term,
Galerkin (RKDG) method is evaluated for otherwise bed slope generated unphysical flows
modeling flow in natural rivers. Reed and Hill may occur. The net pressure force term, arising
(1973) introduced the Runge-Kutta discontinuous from hydrostatic pressure force and wall pressure
Galerkin method for the solution of neutron force, is combined with the bed elevation slope
transport equation. The method was further term. A new formulation is introduced for
developed by Cockburn et al. (1989 and 1990), discretization of this combined term that
Cockburn and Shu (1989 and 1998) for solving guarantees elimination of non-physical flows due
non-linear, time-dependent, hyperbolic equations. to irregular bed for a horizontal water surface
Downloaded by [177.234.0.174] at 22:53 26 August 2015

The RKDG method combines the advantages of under no flow condition. The mass and modified
the finite volume method and the finite element momentum flux terms are evaluated by HLL flux
method and is ideally suited to handle convection- function. The wetting and drying is an essential
dominated flows and shocks. The discontinuous component of the natural river flow modeling. A
elements allow for easy implementation of inter- simple scheme is adapted to handle wetting and
element fluxes using Riemann solvers and slope drying. To test the model, it is first applied to
limiters. Each element is solved independently benchmark dambreak problems in rectangular and
and an element is connected to the surrounding parabolic channels with wet and dry bed
elements through boundary fluxes, thus the downstream of the original dam location. The
method yields to an easy implementation of TVD hydraulic jump is simulated with different
Runge-Kutta time integration scheme. For upstream Froude numbers and the results are
discontinuous Galerkin method, the nodes compared with the measured data. The scheme is
connecting the elements will have two values, then used to simulate dambreak flow resulting
thus discontinuity can be resolved easily. As with from Teton Dam failure, dambreak flow in the
others schemes that are applied to convection Toce River, and a flood event in the East Fork
dominated flows, spurious oscillations may River.
appear in numerical solutions near discontinuities
and steep gradients and extra effort is required to 2. SHALLOW WATER FLOW
achieve smooth solutions. Non-Oscillatory EQUATIONS
schemes, flux limiters, and slope limiters are the
most common methods for achieving smooth The mass and momentum conservation equations
solutions. Schwanenberg and Köngeter (2000) that form one-dimensional shallow water flow
and Kesserwani et al. (2008) applied the equations for natural rivers with irregular cross
discontinuous Galerkin method to solve the sections are given, respectively, below
shallow water equations using slope limiters.
A Q
Bokhove (2005) used HLLC flux with selective  0 (1)
slope limiting using the shock detection method t x
of Krivodonova et al. (2004) in the discontinuous
Q (Q 2 / A) Z
Galerkin method. Bokhove (2005) also presented    gA  gAS f (2)
an Eulerian-Lagrangian approach to handle wet- t x x
dry boundary, where the positivity of water depth
In the equations described above, A is the cross
was guaranteed albeit under restricted time step.
sectional flow area, Q is the flow rate, g is the
Ern et al. (2008) applied discontinuous Galerkin
gravitational acceleration, and Z is the water
method using flux modification to preserve steady
surface elevation. The water surface elevation is
state at rest and a slope modification to deal with
the sum of the distance from the datum to the
flooding and drying. Xing et al. (2010) used
lowest point in the channel cross section and the
discontinuous Galerkin method using the

75
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

flow depth. The flow depth is measured from the


Z 2 Z3
lowest point in the channel cross section to the
Z1 Z3
water surface and must be determined from the Z 4
flow area considering the cross section geometry. Z 2
For horizontal, prismatic channel Z can be easily
replaced by A and the equation can be written in
conservation form. However, for natural channels, x
where cross section geometry is varying in
1 2 3 4
addition to bed slope, Z and A have a complex
relationship through flow depth. Fig. 1 Spatial discretization for discontinuous
In the momentum equation, the wall pressure elements.
force and the net hydrostatic pressure force terms
are combined with the weight term to obtain the sections. Further details can be found in Cockburn
first term on the right hand side (Cunge et al., (2001) and Li (2006) for convection-dominated
1980). When appropriately discretized, the hyperbolic equations. A typical spatial
momentum equation in the above form helps discretization for discontinuous elements is
prevent the unphysical bed slope generated flow. illustrated in Fig. 1. In an element, the variation of
That is, if the water in the channel with irregular area, discharge, and any other function, f , are
bed is at rest initially it will remain at rest. Nujić
Downloaded by [177.234.0.174] at 22:53 26 August 2015

represented by nodal values and shape functions


(1995) reported that the accuracy of the numerical
scheme was dependent on the treatment of the or interpolating functions, N j , as given below
convective flux term. The treatment of the
convective flux and source terms will be given Aˆ  N j A j , Qˆ  N j Q j ,

 
later. In the momentum equation, the friction (6)
slope, S f , is given by and fˆ  A, Q   f Aˆ , Qˆ ,

n 2Q Q where  represents variation of A within an


Sf  , (3) element and A j is the value at node j.
R 4/3 A2
To apply the discontinuous Galerkin finite
where n is the Manning’s roughness coefficient element method to Eqs. (1) and (2), first the
and R is the hydraulic radius (ratio of flow area equations are multiplied by weight or test
to the wetted perimeter). functions Ni , taken to be the same as shape
The eigenvalues (  1 , 2 ) and eigenvectors ( k1 , functions N j in the Galerkin method. The
k 2 ) for the Saint-Venant equations are given, resulting mass and momentum equations are
respectively, by integrated over an element with boundary
[ xs , xe ] and after integration by parts the two
 1  Q A  gA b  u  c,
 equations, respectively, can be written as
 (4)
 2  Q A  gA b  u  c,  e Aˆ
x
 e Ni Qdx
x
 N dx   ˆ 
i
xs t xs x (7)
k  [1 , u  c]T ,
1
 (5) Ni ( xe ) P( xe )  Ni ( xs ) P( xs )  0
k 2  [1 , u  c]T ,

Qˆ  e Ni Qˆ 2
x x
where b is the width of the channel at the water  e
 Ni dx   dx 
t  xs x Aˆ
surface and c is the wave celerity.  xs
3. DISCONTINUOUS GALERKIN Ni ( xe )G ( xe )  Ni ( xs )G ( xs ) (8)
METHOD FORMULATION AND SOURCE x
TERM TREATMENT  e  ˆ gn 2Qˆ Qˆ 
 Z  dx,
  Ni  gA
ˆ 
A brief description of the RKDG method for the   x Rˆ 4 3 Aˆ 
shallow water equations is given in the following  xs  

76
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

wave system. The HLL solver provides an


where P( x)  Q and G( x)  Q2 A are the
economical way to find the wave structure and
numerical fluxes for mass and momentum, evaluate intercell flux in general Riemann
respectively, at an element boundaries. problem (Fraccarollo and Toro, 1995). The wave
The water surface elevation term ( gA Z x ) speeds are calculated using a method suggested
represents a non-conservative term. Rhebergen et by Fraccarollo and Toro (1995). For the
al. (2008) implemented a technique to find weak conserved variables vector U  [ A, Q]T and the
formulation and boundary fluxes for the non-
corresponding boundary flux vector
conservative terms. The non-conservative
2 T T , the approximate
boundary fluxes were approximated using NCP- F  [Q, Q A]  [ P( x), G( x)]
flux as discussed earlier. An alternative method is numerical flux is given as
presented below to handle the non-conservative
term related to water surface gradient that F  for 0  S
L

 
produces a well-balanced scheme. Since linear
 S F  S F   S S U  U

 
elements are used, there are two test functions N1
hll -   R L L R
F U ,U   SR  SL
and N 2 . The water surface elevation gradient

term is discretized using the following expression for S L  0  S R ,
(see Fig. 1)  
F for 0  S R
Downloaded by [177.234.0.174] at 22:53 26 August 2015

 ˆ ( A  Axe ) ( Z xe  Z xs )


 gN1 Aˆ Z   gN1 xs
(10)

 x 2 ( xe  xs )      
  S L  min  u  g  A / b  , u  c 
  ˆ ( Ax  Ax ) ( Z x  Z x )   
 gN 2 Aˆ
Z
  gN 2 s e e s  (11)
 x ( xe  xs )  S  max  u   g  A / b  , u  c 
 2
 R  
(9)  


where Ax and Ax are the values of A at the u 
1  
(u  u  )  g  A / b   g  A / b 

s s
2
boundary xs from the left and right elements and (12)
so on. A standard discontinuous Galerkin
discretization for the water surface elevation 1   1
gradient term results in an ill-balanced c  ( g  A / b   g  A / b  )  (u   u  )
2 4
formulation, where non-physical flows due to
(13)
irregular bed topography may be generated in
case of horizontal water surface with zero where F  F(U- ) and F  F(U ) . The wave
discharge (Kesserwani et al., 2010). The present
speeds to the left and right of an element
discretization scheme for the water surface
elevation gradient term will not produce non- boundary are given by S L and S R , respectively.
physical flows due to irregular bed topography. The velocities to the left and right of the element
 
That is, for a level water surface, the water boundary are given by u and u .
surface gradient term will vanish. With discharge In dry bed problems, the wave speeds for the dry
and water surface gradient being zero, all terms in bed to the right and left of the element under
Eqs. (7) and (8) except for the time derivatives consideration are given, respectively, by the
terms of area and discharge, respectively, will following equations
disappear. Hence, the still water level will be

maintained. The friction term in Eq. (8) is SL  u   g  A / b 
discretized using a standard finite element (14)
 
method. SR  u  2 g  A / b 


SL  u  2 g  A / b 
4. NUMERICAL FORMULATION
(15)
Some widely used approximate Riemann solvers  
SR  u  g  A / b 
have been discussed before. In this paper, the
HLL flux is adopted for the one dimensional two-

77
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

For an initially dry bed, a sufficiently small depth,  j 1   j


hd (10-8 m), and zero velocity are defined at the b (19)
( x j 1, m  x j , m )
dry nodes (Ying et al., 2004). After every time
step the water depth in the whole domain is
checked. If the water depth at a node falls below ab  j 1   j 1
 . (20)
hd , then the water depth is set to hd and the flow 2 ( x j 1, m  x j 1, m )
rate is set to zero at that node.
For the flux at the element boundary, if the water 6. RUNGE-KUTTA TVD TIME
depth on one side is greater than hd and the water INTEGRATION
depth on the other side is less than or equal to hd ,
Former studies have shown that the TVD Runge-
then the left and right wave speeds are calculated Kutta time integration scheme should be one
based on Eq. 14 or 15. If water depth on both order higher than the order of the basis function
sides of the boundary is less than or equal to hd , used to achieve variation of a variable over an
the numerical flux in Eq. (10) would be zero since element (Cockburn et al., 1989; Cockburn and
the flow rate is zero, so the mass and momentum Shu, 1989; Cockburn et al., 1990). Since linear
flux is conserved. In this way, a positive water elements are used as basis function, the second
depth is provided and the dry element can be order TVD Runge-Kutta scheme is implemented
Downloaded by [177.234.0.174] at 22:53 26 August 2015

evaluated with satisfactory accuracy. here. Eqs. (7) and (8) can be written in the
following general form
5. SLOPE LIMITER U
 L(U). (21)
Flux limiters and slope limiters are often applied t
in combination with TVD time integration To advance the solution from time step n to n+1,
scheme to achieve overall TVD property for the the second order TVD Runge-Kutta scheme, as
numerical scheme. Generally, the slope limiter is given by Gottlieb and Shu (1998), can be written
applied to the conserved variables, which in this as
case are the cross-section area, A , and the flow
rate, Q . The slope limiter in a general form for a
variable  in an element can be written as
U[1]  U n  tL U n ,
  
 (22)
 ( x)    ( x  xm ) for xs  x  xe , (16) 
1
2
1
2
1
U n 1  U n  U[1]  tL U[1] .
2
 
where  is the average value of a variable over For the explicit scheme adopted here, the
Courant-Friedrichs-Lewy condition must be
an element, xm is the midpoint of the element,
fulfilled and is given by
and  is a monotonized central slope limiter. The
variable  can be the cross section flow area or  t 
max   u  c  
1
(23)
the flow rate. The monotonized central slope  x  2 p 1
limiter (Li, 2006) for an element j can be written
as where p is the order of polynomial used for space
discretization (Cockburn, 2001).
j 
sign(a)  sign(b) min  a  b , 2 a , 2 b 
  7. NUMERICAL TESTS
2  2 
(17) Test 1: Idealized Dambreak in Parabolic Channel
The upwind slope a, the downwind slope b, and To examine the accuracy and shock capturing
the central slope (a+b)/2 are, respectively, given ability of the proposed model in a non-rectangular
by channel, the numerical scheme was first applied
to the idealized dam break problems in a
 j   j 1 parabolic channel. Dambreak tests with wet bed
a (18)
( x j , m  x j 1, m ) and dry bed downstream of the dam were
performed and computed results were compared
with analytic solutions (Henderson, 1966). The
test was conducted in a 1000 m long horizontal,

78
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

60
frictionless, parabolic channel. The width of the Computed
parabolic channel was given by b=Z0.5. The dam 50 Exact

was located at the middle of the channel. The


40
water depth upstream of the dam was 10 m,

Q (m3/s)
downstream water depth was 1 m and zero for 30
wet bed and dry bed tests, respectively. Element
20
size of 1 m and time step of 0.005 s were used in
both tests. 10

Numerical solutions and analytical solutions of 0


0 200 400 600 800 1000
water surface and flow rate at 20 s after the x (m)
removal of the dam in both cases are shown in
Fig. 5 Computed and exact flow rate for dry bed
Figs. 2-5. Numerical results are in excellent dambreak in parabolic channel.
agreement with analytic solutions for both tests.
Dambreak test with wet bed shows the scheme is 0.08
Measured at 3.75s
capable to capture the shock wave. While the dry 0.07 Measured at 9.40s
Simulated at 3.75s
bed test demonstrates suitability of the model to 0.06 Simulated at 9.40s
simulate flow over an initially dry bed. 0.05

Z (m)
12 0.04
Computed
Downloaded by [177.234.0.174] at 22:53 26 August 2015

10 0.03
Exact
0.02
8
0.01
Z (m)

6 0
0 5 10 15 20
x (m)
4
Fig. 6 Computed and measured water surface at 3.75
2
s and 9.40 s.
0
0 200 400 600 800 1000
x (m) Test 2: Dambreak in Rectangular Flume
Fig. 2 Computed and exact water surface for wet In this test, the model was applied to a dambreak
bed dambreak in parabolic channel. problem in a rectangular, horizontal flume for
60
which measured water surface profiles after the
Computed dambreak event were available (Schoklitsch,
50 Exact
1917). The flume was 0.096 m wide, 0.08 m high,
40 and 20 m long. The dam was located at the
Q (m3/s)

middle of the flume, with water ponded to a


30
height of 0.074 m upstream of the dam and a dry
20 bed downstream. The flume was made of smooth
wood, and the Manning’s roughness coefficient of
10
0.009 s/m1/3 was used in the simulation. The
0
0 200 400 600 800 1000 removal of the dam was assumed instantaneous,
x (m)
and the water flow after the dam removal was
Fig. 3 Computed and exact flow rate for wet bed simulated. Element size of 0.1 m and time step of
dambreak in parabolic channel. 0.0001 s were used in this test.
Fig. 6 shows the computed water surface profiles
12
with measured data at 3.75 seconds and 9.40
Computed
10
Exact seconds after the removal of the dam. The
8
computed water surface profiles are in good
agreement with the measured profiles. The results
Z (m)

6 show that the scheme is capable of modeling the


4 progressive wave after the dam removal over an
initially dry bed with friction.
2
Test 3: Hydraulic Jump in Rectangular Channel
0
0 200 400 600 800 1000
x (m) Gharangik and Chaudhry (1991) undertook
Fig. 4 Computed and exact water surface for dry bed physical model study to measure water surface
dambreak in parabolic channel. profiles of hydraulic jumps for various upstream

79
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

0.3
the jumps are observed in the discharge and may
0.25 be due to the breakdown of the hydrostatic
pressure assumption. Overall, the scheme is
0.2
capable to model hydraulic jump accurately in
Z (m)

0.15 terms of its location and water surface profile.


Measured
0.1
Test a Test 4: Teton Dam Failure
Measured
Test b
0.05
The one-dimensional numerical model was
applied to model the Teton dam failure that took
0 2 4 6
x (m)
8 10 12 14 place on June 5, 1976. The Teton dam was a
92.96 m (305 feet) high earth-fill dam with a
Fig. 7 Computed and measured water surface 914.4 m (3000 feet) long crest located on the
profiles for hydraulic jumps. Teton River in southeastern Idaho. The inundated
0.1 area after the dambreak is shown in Fig. 9. River
Test a cross sections, Manning’s roughness coefficient,
Test b
0.08 reservoir storage depletion, and the flow rate at
the dam site were documented by the U.S.
Q (m3/s)

0.06
Geological Survey (Ray and Kjelstrom, 1976).
The flooded area and measured cross sections,
Downloaded by [177.234.0.174] at 22:53 26 August 2015

0.04
shown as straight lines across the river, are shown
0.02
in Fig. 9. The cross sections used in the
computation were interpolated from the available
0
0 2 4 6 8 10 12 14
data. The discharge at the dam site is shown in
x (m) Fig. 10. The dam site discharge and water surface
Fig. 8 Computed flow rate variation for hydraulic level were used as inflow boundary conditions,
jumps. corresponding to the supercritical inflow
condition. Since the initial flow conditions before
and downstream boundary conditions. The study the dambreak were not specified, a downstream
was conducted in a 14 m long and 0.46 m wide dry bed was assumed as an initial condition for
rectangular, horizontal flume. Two cases (labeled simulation. In addition, inflows from Henry’s
as ‘a’ and ‘b’) were selected to validate the Fork and Snake River were ignored as these
current model. For the cases ‘a’ and ‘b’, the inflows were small compared to the large flood
upstream velocities were 3.831 m/s and 1.826 event. The simulation ended at 10 hours after the
m/s, the upstream depths were 0.031 m and 0.064 dambreak.
m, and downstream depths were 0.265 m and Teton Dam

0.168 m, respectively. These conditions


correspond to upstream Froude numbers of 6.947
and 2.305, and downstream Froude numbers of
Sugar City
0.278 and 0.542, respectively. The given
Menan Buttes
discharge and upstream water depth were used as Rexburg

upstream boundary conditions (compatible with


supercritical flow at the upstream end of the N
Snake River
channel). At the downstream end, the water depth
maintained in the physical model at the end of the
channel was used as a boundary condition. The
initial condition for the water surface was set by 0 5 10 km

linearly varying the water depth between the inlet


and outlet. The Manning’s roughness coefficient
was set to 0.008 s/m1/3.
The final computed water surface and flow rate
for the two cases are shown in Figs. 7 and 8. Idaho Falls

Element size of 0.1 m and time step of 0.002 s


were used in this test. Fig. 7 shows that the
locations and water surface profiles for the two
hydraulic jumps are predicted accurately by the
model. In Fig. 8, small oscillations at the toe of Fig. 9 Flood area downstream of Teton River.

80
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

8
10
Measured
Discharge 1600 Computed
Bottom
7
1550

Max Z (m)
10
Q (m3/s)

1500
6
10
1450

5
10 1400
Frame 001  26 Apr 2011  0
0 2 4 6 8 10 0 20 40 60 80 100
Time (hr) x (km)

Fig.10 Discharge at the dam site after dambreak. Fig. 13 Maximum water elevation during the flood
event.
1550
Water surface
Bottom

+P26
1500 +P1
+P5 +P21
+P18
Z (m)
Downloaded by [177.234.0.174] at 22:53 26 August 2015

1450
Fig. 14 Plan view of Toce River topography.

1400
Test 5: Toce River Case
0 20 40 60 80 100
x (km)
A physical model at 1:100 scale of a reach of the
Fig. 11 Water surface along the river at 10 hours after Toce River valley (Northern Alps, Italy) was
dambreak. developed at the ENEL-HYDRO laboratory in
Milan, Italy. The physical model tests were used
2.5
in the CADAM project (Frazão and Testa, 1999).
2
Modeling parameters, such as topographic data,
inflow hydrograph, and Manning’s coefficient,
1.5 were specified by Electricité de France (EDF).
Results of the physical model were also provided
Fr

1 so that modelers could make an objective


comparison of their numerical modeling results.
0.5
The topography of Toce River physical model is
0
shown in Fig. 14. The physical model covered an
0 20 40 60 80 100
x (km) area of approximately 50 m × 12 m. To measure
the water surface level, 32 water level gauges
Fig. 12 Froude number along the river at 10 hours
after dambreak.
were installed in the physical model. Selected
gauges along the main river axis (P1, P5, P18,
Fig. 11 and 12 show the computed water surface P21, and P26) were used for comparing the
and Froude number, respectively, at 10 hours simulated results with the measured data. For
after the dam break. Fig. 13 shows the maximum simulation, 62 cross-sections were used. These
water surface during the flood event along the cross-sections are shown in Fig.15. The elements
river. As shown in Fig. 11, the water surface used in the computation domain were of non-
exhibits large variation in the study reach. In Fig. uniform size and the length of the element varied
13, the computed maximum water surface from 0.25 m to 1.94 m.
elevation along the river reach compares well A rectangular tank was located at the upstream
with the measured data. The difference between (left) end of the physical model. Discharge from
the computed result and the measured data in the the rectangular tank was used as an inflow
middle of the river (30-50 km) is mainly due to boundary condition during numerical calculation.
neglect of the side flow from the Snake River. The inflow hydrograph at the rectangular tank is
The results demonstrate that the numerical shown in Fig. 16. The flow conditions at the inlet
scheme is capable of modeling dambreak and outlet ends of the physical model were forced
problems in natural rivers and can provide to be critical and the same conditions were
satisfactory results. applied in the computation. The channel
downstream of the tank was initially dry.

81
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

12 2

10
P26

8
P1
y (m)

Fr
6 P5 P21 1
P18

0 0
0 10 20 30 40 50 0 10 20 30 40 50
x (m) Distance (m)

Fig. 15 Locations of computational crosssections of Fig. 18 Computed Froude number along the river at
Toce River. time=100s.
8
Discharge

0.2
7.5

Max Z (m)
Q (m3/s)

7
0.1 Computed
Physical model
Downloaded by [177.234.0.174] at 22:53 26 August 2015

ISIS model
6.5
Bottom

0 6
0 60 120 180 0 10 20 30 40 50
Time (s) x (m)

Fig. 16 Inflow boundary conditions at the river inlet. Fig. 19 Computed maximum water level of Toce
River.
7.8
Water Surface
7.6 Bottom Coare mesh
7.8 Fine mesh
7.4 Measured
P1
Z (m)

7.5
7.2 P5
Z (m)

P18
7 7.2 P21

6.8
6.9
P26
6.6
0 10 20 30 40 50
6.6
Distance (m)
0 60 120 180
Time (s)
Fig. 17 Computed water surface profile at time=100s.
Fig. 20 Computed and measured stage-time
The value of Manning’s roughness coefficient hydrographs.
was taken to be 0.0162s/m1/3 based on the value
proposed in the physical model study. The good agreement with the physical model
numerical simulation ended at 180 seconds. The measurement. The root-mean-square error
computed water surface and Froude number at (RMSE) is 0.0473 m between ISIS model and
100 seconds are shown in Figs. 17 and 18, measured data and is 0.0347 m between present
respectively. The predicted water surface level, as model and measured data. The present model
displayed in Fig. 17, shows a lot of variation due provides better results than the ISIS model at the
to bed slope and channel width changes. Several downstream end of the river.
hydraulic jumps can be identified as confirmed by Fig. 20 shows a comparison of computed stage-
the Froude number variation along the channel time hydrographs with the measured data at five
shown in Fig. 18. gauge points. Besides the coarse mesh in Fig. 15,
The computed results of maximum water level are a refined mesh, with four elements added between
plotted along with measured data from the every cross section, was used to investigate effect
physical model study and results from ISIS model of mesh size. The difference in the water surface
in Fig. 19. The ISIS model was based on the finite level between the coarse and refined mesh is more
difference Preissmann implicit scheme (Rosu and pronounced at the downstream stations (P21 and
Ahmed, 1999). The computed results of P26), with the refined mesh providing better
maximum water level, as shown in Fig. 18, are in comparison with the measured data. For the

82
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

refined mesh, the arrival time of the surge and the The average water depth on June 1 was used as
variation of water surface elevation with time initial depth, while the initial flow rate was set to
show good agreement with the measured data. be 6.0 m3/s throughout the study reach. Since
value of roughness coefficient is unavailable, the
Test 6: East Fork River Case
Manning’s roughness coefficient was varied to
In this test, the numerical scheme is used to model find the appropriate value.
a flood event in the East Fork River, Wyoming. The computed and measured water surface at
The East Fork River flows in the Wind River sections 2505 and section 3295 are shown in Fig.
Range of Wyoming, west of the Continental 22. The computed and measured discharges at
Divide and east and southeast of Mt. Bonneville. section 0000 are shown in Fig. 23. In Fig. 24, the
The study reach configuration is shown in Fig.21. computed water surface at noon on June 12, 1979
The meandering study reach is approximately 3.3 is compared with the measured water surface for
km in length and terminates downstream at a different values of Manning’s roughness
bedload trap constructed across the river. The coefficient. Results show that Manning’s
number shown at each cross section in the figure roughness coefficient of 0.028 s/m1/3 provides
is the centerline distance (in meters) upstream results that are in agreement with measured data.
from the bedload trap (section 0000). Bed In general, increase in roughness will increase
elevation at 39 cross sections were measured water surface and vice versa. For the Manning’s
daily during a month-long period (Meade et al., roughness coefficient of 0.028 s/m1/3, the
Downloaded by [177.234.0.174] at 22:53 26 August 2015

1980).The influence of the sediment transport and difference in the measured and computed water
changes of channel bed (due to short simulation surface elevation and discharge values at the
period) are not considered in this test. beginning is mostly due to the uncertainty in the
The simulation period of the river flow was 12 initial conditions, the difference during the last
days from June 1 to June 12, 1979. The hourly four days may originate from the higher sediment
discharge measured at section 3295 was used as transport within the reach. In general, the
the inflow boundary condition (Emmett et al., numerical results with Manning’s roughness
1980), while the hourly gage height at section coefficient of 0.028 s/m1/3 are in good agreement
0000 was used as outflow boundary condition. with the measured data and the test demonstrates
Subcritical flow condition existed at the inflow that the model is capable of simulating flood flow
and outflow boundary and the boundary in natural rivers.
conditions used in the model reflected that fact. 9
n=0.028
Measured
Z (m)

7
0 2 4 6 8 10 12
Time (Day)

Fig. 22 Computed and measured water surface at


section 3295 and section 2505 (lower lines).
20
n=0.028
Measured
15
Q (m3/s)

10

0
0 2 4 6 8 10 12
Time (Day)

Fig. 21 Map of the 3.3 km study reach in the East Fig. 23 Computed and measured discharge at section
Fork River (Emmett et al., 1980). 0000.

83
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

9
n=0.035 computations. Journal of Hydraulic
n=0.028
8
n=0.020 Engineering 125(5):385-393.
7
Measured
Bed
3. Catella M, Paris E, Solari L (2008).
Conservative scheme for numerical modeling
Z (m)

6
of flow in natural geometry. Journal of
5 Hydraulic Engineering 134(6):736-748.
4. Cockburn B (2001). Discontinuous Galerkin
4
methods for convection dominated problems.
3
0 500 1000 1500 2000 2500 3000 Lecture Notes in Computational Science and
Distance (m) Engineering, Springer, Berlin, Vol. 9, 69-224.
Fig. 24 Computed and measured water surface around 5. Cockburn B, Hou S, Shu CW (1990). The
noon June 12, 1979. Runge-Kutta local projection discontinuous
Galerkin finite element method for
8. CONCLUSIONS conservation laws IV: The multidimensional
case. Mathematics of Computation
A numerical model based on discontinuous 54(190):545-581.
Galerkin method is developed to solve one- 6. Cockburn B, Lin SY, Shu CW (1989). TVB
dimensional shallow water equations for natural Runge-Kutta local projection discontinuous
rivers. The Saint-Venant equations are modified Galerkin finite element method for
Downloaded by [177.234.0.174] at 22:53 26 August 2015

by combining the pressure force and bed conservation laws III: One dimensional
elevation change terms. The treatment of the systems. Journal of Computational Physics
combined term to account for the momentum flux 84(1):90-113.
is given. The inter-element flux terms are 7. Cockburn B, Shu CW (1989). TVB Runge-
approximated using HLL flux. The total variation Kutta local projection discontinuous Galerkin
diminishing (TVD) property is achieved by TVD finite element method for conservation laws
time integration and TVD slope limiters. The II: General framework. Mathematics of
numerical scheme is evaluated by simulating dam Computation 52(186):411-435.
break problems in rectangular and non- 8. Cockburn B, Shu CW (1998). The Runge-
rectangular channels for dry bed and wet bed Kutta discontinuous Galerkin method for
downstream of the dam and hydraulic jumps in a conservation laws V: Multidimensional
rectangular channel. The numerical scheme is systems. Journal of Computational Physics
then applied to simulate flow in Teton River, 141(2):199-224.
Toce River, and East Fork River. These test cases 9. Cunge JA, Holly FM, Verwey A (1980).
show that the numerical scheme is capable of Practical Aspects of Computational River
modeling both sudden dambreak type of flows Hydraulics. Pitman, London.
and natural floods in natural channels. The results 10. Einfeldt B (1988). On Godunov-type methods
show that the discharge, water surface level, and for gas dynamics. SIAM Journal on
surge arrival time are accurately predicted using Numerical Analysis 25(2):294-318.
the developed scheme. In addition, the numerical 11. Emmett WM, Myrick RM, Meade RH (1980).
scheme is capable of simulating flow over an Field data describing the movement and
initial dry bed in natural channels. The highly storage of sediment in the East Fork River,
varying bed topography and channel cross section Wyoming, Part I. River hydraulics and
shapes, which are normal features of natural sediment transport, 1979. USGS Open-File
channels, can be handled with ease. Report 80-1189, Denver.
12. Ern A, Piperno S, Djadel K (2008). A well-
REFERENCES balanced Runge-Kutta discontinuous Galerkin
method for the shallow-water equations with
1. Bokhove O (2005). Flooding and drying in flooding and drying. International Journal for
discontinuous Galerkin finite-element Numerical Methods in Fluids 58(1):1-25.
discretizations of shallow-water equations. 13. Erpicum S, Dewals B, Archambeau P,
Part 1: One dimension. Journal of Scientific Detrembleur S, Pirotton M (2010). Detailed
Computing 22-23(1-3):47-82. inundation modeling using high resolution
2. Capart H, Eldho TI, Huang SY, Young DL, DEMs. Engineering Applications of
Zech Y (2003). Treatment of natural Computational Fluid Mechanics 4(2):196-
geometry in finite volume river flow 208.

84
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

14. Fraccarollo L, Toro EF (1995). Experimental 27. Meade RH, Myrick RM, Emmett WW
and numerical assessment of the shallow (1980). Field data describing the movement
water model for two-dimensional dam-break and storage of sediment in the East Fork
type problems. Journal of Hydraulic River, Wyoming, Part II. Bed elevations,
Research 33(6):843-846. 1979. US Geological Survey, Open-File
15. Frãzao SS, Testa G (1999). The Toce River Report 80-1190, Denver.
test case: Numerical results analysis. 28. Nujić M (1995). Efficient implementation of
Proceedings of the 3rd CADAM Workshop, non-oscillatory schemes for the computation
Milan, Italy. of free-surface flows. Journal of Hydraulic
16. Garcia-Navarro P, Vázquez-Cendón M Research 33(1):101-111.
(2000). On numerical treatment of the source 29. Perthame B, Simeoni C (2001). A kinetic
terms in the shallow water equations. scheme for the Saint-Venant system with a
Computers & Fluids 29(8):951-979. source term. Calcolo 38(4):201-231.
17. Gharangik AM, Chaudhry MH (1991). 30. Ray HA, Kjelstrom LC (1976). The flood in
Numerical simulation of hydraulic jump.” southeastern Idaho from the Teton Dam
Journal of Hydraulic Engineering failure of June 5, 1976. U.S. Geological
117(9):1195-1211. Survey, Open-File Report 77-765, Boise,
18. Godunov SK (1959). A finite difference Idaho, USA.
method for the numerical computation of 31. Reed WH, Hill T (1973). Triangular mesh
Downloaded by [177.234.0.174] at 22:53 26 August 2015

discontinuous solutions of the equations of method for the neutron transport equation.
fluid dynamics. Math Sbomik 47(3):271-306. Los Alamos Report LA-UR-73-479.
19. Gottlieb S, Shu CW (1998). Total variation 32. Rhebergen S, Bokhove O, Van Der Vegt JJW
diminishing Runge-Kutta schemes. (2008). Discontinuous Galerkin finite element
Mathematics of Computation 67(221):73-85. methods for hyperbolic nonconservative
20. Harten A, Lax PD, Van Leer B (1983). On partical differential equations. Journal of
upstream differencing and Godunov-type Computational physics 227(3):1887-1922.
schemes for hyperbolic conservation laws. 33. Roe P (1981). Approximate Riemann solvers,
SIAM Review 25(1):35-61. parameter vectors, and difference schemes.
21. Henderson FM (1996). Open Channel Flow. Journal of Computational Physics 43(2):357-
McGraw-Hill, New York. 372.
22. Kesserwani G, Ghostine R, Vazquez J, 34. Rosu C, Ahmed M (1999). Toce River dam-
Ghenaim A, Mosé R (2008). Application of a break test case – A comparison between the
second-order Runge-Kutta discontinuous ISIS numerical model and the physical model.
Galerkin scheme for the shallow water Proceedings of the 3rd CADAM Workshop,
equations with source terms. International Milan, Italy.
Journal for Numerical Methods in Fluids 35. Schwanenberg D, Köngeter J (2000). A
56(7):805-821. discontinuous Galerkin method for the
23. Kesserwani G, Liang Q, Vazquez J, Mosé R shallow water equations with source terms.
(2010). Well-balancing issues related to the Lecture Notes in Computational Science and
RKDG2 scheme for the shallow water Engineering, Springer, Berlin, Vol. 11, 419-
equations. International Journal for 424.
Numerical Methods in Fluids 62(4):428-448. 36. Schoklitsch A (1917). Ueber
24. Khalifa AM (1980). Theoretical and Dammbruchwellen. Sitzungsberichte der
Experimental Study of the Radial Hydraulic Kaiserlichen Akademie Wissenschaften,
Jump. Ph.D. Thesis, University of Windsor, Viennal, 126,1489-1514.
Windsor, Ontario, Canada. 37. Toro E (1989). A weighted average flux
25. Krivodonova L, Xin J, Remacle J-F, method for hyperbolic conservation laws.
Chevaugeon N, Flaherty JE (2004). Shock Proceedings of the Royal Society of London.
detection and limiting with discontinuous Series A, Mathematical and Physical Sciences
Galerkin methods for hyperbolic conservation 423(1865):401-418.
laws. Applied Numerical Mathematics 48(3- 38. Toro E, Spruce M, Speares W (1994).
4):323-338. Restoration of the contact surface in the HLL-
26. Li BQ (2006). Discontinuous Finite Element Riemann solver. Shock Waves 4(1):25-34.
in Fluid Dynamics and Heat Transfer. 39. Tritthart M, Gutknecht D (2007). Three-
Springer, London. dimensional simulation of free-surface flows
using polyhedral finite volumes. Engineering

85
Engineering Applications of Computational Fluid Mechanics Vol. 6, No.1 (2012)

Applications of Computational Fluid


Mechanics 1(1):1-14.
40. Xing Y, Zhang X, Shu C (2010). Positivity-
preserving high order well-balanced
discontinuous Galerkin methods for the
shallow water equations. Advances in Water
Resources 33(12):1476-1493.
41. Ying X, Jorgeson J, Wang SSY (2009),
Modeling dam-break flows using finite
volume method on unstructured grid.
Engineering Applications of Computational
Fluid Mechanics 3(2):184-194.
42. Ying X, Khan AA, Wang SSY (2004).
Upwind conservative scheme for the Saint
Venant equations. Journal of Hydraulic
Engineering 130(10):977-987.
43. Zhou JG, Causon DM, Mingham CG, Ingram
DM (2001). The surface gradient method for
the treatment of source terms in the shallow-
Downloaded by [177.234.0.174] at 22:53 26 August 2015

water equations. Journal of Computational


Physics 168(1):1-25.

86

View publication stats

You might also like