You are on page 1of 9

Applied Mathematical Modelling 31 (2007) 490498

www.elsevier.com/locate/apm

Computations of transport of pollutant in shallow water


a,*
Li Cai , Wen-Xian Xie b, Jian-Hu Feng c, Jun Zhou a

a
College of Astronautics, Northwestern Polytechnical University Xian, Shaanxi 710072, PR China
b
School of Science, Northwestern Polytechnical University Xian, Shaanxi 710072, PR China
c
College of Science, Changan University Xian, Shaanxi 710064, PR China

Received 1 May 2005; received in revised form 1 July 2005; accepted 1 November 2005
Available online 27 December 2005

Abstract

The focus of this paper is to simulate the transport of a passive pollutant by a ow modelled by the two-dimensional
shallow water equations. Considering the friction terms, new model for simulating the steady and unsteady transport of
pollutant is established. Then the adaptive semi-discrete central-upwind scheme based on central weighted essentially non-
oscillatory reconstruction is utilized for simulating the two-dimensional steady and unsteady transport of pollutant. The
non-oscillatory behavior and accuracy of the scheme are demonstrated by the numerical result.
 2005 Elsevier Inc. All rights reserved.

Keywords: Transport of pollutant; 2-D shallow water equations; Adaptive semi-discrete central-upwind schemes; Central weighted ess-
entially non-oscillatory (CWENO) scheme

1. Introduction

In past literature, the shallow water equations have wide applications in ocean and hydraulic engineering,
e.g. tidal ows in estuary and coastal water regions; bore wave propagation; the stationary hydraulic jump;
and river, reservoir and open channel ows. Recently, prediction of a pollutant transport in ows is a newness
and important subject in many industrial and environmental projects.
Here, we consider the transport of a passive pollutant by a ow modelled by the two-dimensional shallow
water equations. Our new two-dimensional pollutant model based on shallow water equations is given as
follows:
oU
rFS 1:1
ot
where U is the vector of the conserved variables, F is the ux vector function, S is the vector of the source
terms, and r i oxo j oyo is the gradient operator. In practice, U = (h, hu, hv)T, F hiu jv; huiu jv;
i 12 gh2 ; hviu jv j 12 gh2 T and S = Sb + Sf + SS with the bed slops Sb = (0, ghBx, ghBy)T, the bed

*
Corresponding author.
E-mail addresses: licai@mail.nwpu.edu.cn, eign@eyou.com (L. Cai).

0307-904X/$ - see front matter  2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2005.11.011
L. Cai et al. / Applied Mathematical Modelling 31 (2007) 490498 491
 p pT
ug2 u2 v2 vg2 u2 v2
frictions Sf 0;  h4=3
; h4=3
, the pollutant source terms SS = (S, 0, 0)T, in which B(x) denotes
the bottom elevation, h describes the uid depth above the bottom, u and v represent the ow velocity in
the x and y-directions respectively, the acceleration due to gravity is donated by g and g is the Manning resis-
tance coecient. If g = 0 and S = 0, the system (1.1) is the so-called SaintVenant system (see [14] and ref-
erences therein).
The transport equation
hT t uhT x vhT y T S S 1:2

is utilized to describe the propagation of the pollutant, where T is the pollutant concentration and TS is a given
concentration of the pollutant at the source. The system (1.1) and the equation (1.2) are coupled through the
source terms.
In this paper, we focus on the transport of pollutant considering the friction terms (i.e. g 5 0). This new
pollutant model was rstly proposed in our previous Ref. [5]. Likewise, the new hybrid system (1.1) and
(1.2) reduces to the SaintVenant-type model of transport of pollutant in [68] if g = 0. Furthermore, if
S  0, our models should also have steady states in which the ux gradients are non-zero but exactly balanced
by source terms. Obviously, the unsteady states will occur if S f 0 (see [9,10]) or g 5 0. We will discuss some
dierences between above two models in Section 4.
In recent years, an amount of research was done in developing and implementing modern high-resolution
nite dierence methods for approximating solutions of the system (1.1) and the equation (1.2). The upwind
schemes have to solve Riemann problems on the boundaries of each cell, which is interpreted as an upwinding
procedure. However, a general scheme for the (exact or approximate) solution of the Riemann problems is not
known, and the upwind approach may be rather complicated and costly, especially in multidimensional cases.
On the other hand, by a straight-forward centered computation of the quantities involved, the central schemes
avoid the Riemann solvers and characteristic decomposition of the Jacobians. Considering the local speeds of
non-linear wave propagation at the cells boundaries, Kurganov et al. presented central-upwind schemes in
[1113]. Since the non-uniform width of the overall local Riemann fans is calculated more accurately, the cen-
tral-upwind schemes enjoy a much smaller numerical viscosity as well as the staggering between two neighbor-
ing sets of grids is avoided.
In this paper, we combine the fourth-order central weighted essentially non-oscillatory (CWENO) recon-
struction with the adaptive semi-discrete central-upwind scheme in [14]. To the best of the authors knowledge,
no one has ever applied this particular combination of the adaptive semi-discrete central-upwind scheme and
the CWENO reconstruction for this particular model. The central-upwind scheme in [14] is derived in a gen-
eral form which is independent of the reconstruction step, as long as the reconstructed interpolant is su-
ciently accurate and non-oscillatory. That is, if the reconstructions we used have rth-order, r 6 4, accuracy,
we can easily get the rth-order schemes to approximate conservation laws without source terms. At the same
time, the mixed rth-order schemes seem to provide a much sharper resolution than the purely rth-order
schemes. These are the key mechanism to our method (similar to [15,5,16]).

2. The 2-D CWENO reconstruction

Consider the 2-D system of conservation laws

ut fux guy 0; u 2 Rd ; d P 1. 2:3

We discretize both space and time assuming uniform mesh spacings of D:Dx = Dy and Dt respectively. The
fourth-order CWENO-type reconstruction, which is based on a fundamental biquadratic polynomial, is ap-
plied to our 2-D scheme. Analogous to [17], we can achieve the fourth-order CWENO-type reconstruction
of components of u on cell Ij,k = [xjD/2, xj + D/2] [ykD/2, yk + D/2] at time tn:
X
1
Rj;k x; y xl;m
j;k P jl;km x; y; 2:4
l;m1
492 L. Cai et al. / Applied Mathematical Modelling 31 (2007) 490498

with
P j;k x; y b0 b1 x  xj b2 y  y k b3 x  xj y  y k
2 2 2
b4 x  xj b5 y  y k b6 x  xj y  y k
2 2 2
b7 x  xj y  y k b8 x  xj y  y k ; 2:5
where the nine coecients b0, . . . , b8 are uniquely determined by the interpolation conditions
Z xj D=2lD Z y k D=2mD
1
P j;k x; ydy dx 
ujl;km ; l; m 1; 0; 1. 2:6
D2 xj D=2lD y k D=2mD
The weights xl;m
j;k in (2.4) are written as

al;m
j;k
xl;m
j;k P1 l;m
; 2:7
l;m1 aj;k

where
C l;m
al;m
j;k 2
; e < 106 . 2:8
e ISl;m
j;k

The smoothness indicator, ISl;m j;k , is done by evaluating a suitable function on the norms of the polynomials
derivatives on the cell Ij,k, namely
R 2 2 2 2
ISl;m 2 2
j;k I j;k jox P jl;km x; yj joy P jl;km x; yj D joxx P jl;km x; yj D joyy P jl;km x; yj dx dy.

2:9
l,m
All that is left is to compute the constants C in (2.8). In order to make the reconstruction more ecient but
less robust, we select C1,1 = 1/16, C1,0 = 1/16, C1,1 = 1/16, C0,1 = 1/16, C0,0 = 1/2, C0,1 = 1/16,
C1,1 = 1/16, C1,0 = 1/16, C1,1 = 1/16.

3. Central-upwind scheme for 2-D model of transport of pollutant

Our new 2-D model of transport of pollutant (1.1) and (1.2) is rewritten as
wt hux hvy S; 3:10
" #   p!
2
hu g 2 hu  hv ug2 u2 v2
hut w  B gw  B Bx 4=3
; 3:11
wB 2 wB y w  B
x
  " # p!
2
hu  hv hv g 2 vg2 u2 v2
hvt w  B gw  B By 4=3
; 3:12
wB x wB 2 w  B
y
   
hu  hT hv  hT
hT t T SS 3:13
wB x wB y
We use the notation from [12,13,4,14]
u
j1=2;k Rj1;k xj1=2 ; y k ; u
j1=2;k Rj;k xj1=2 ; y k ;

u
j;k1=2 Rj;k1 xj ; y k1=2 ; u
j;k1=2 Rj;k xj ; y k1=2 ;

uj;k Rj;k xj ; y k ; uNj;k Rj;k xj ; y k1=2 ; uSj;k Rj;k xj ; y k1=2 ; 3:14


uEj;k Rj;k xj1=2 ; y k ; uWj;k Rj;k xj1=2 ; y k ; uNE
j;k Rj;k xj1=2 ; y k1=2 ;
uNW
j;k Rj;k xj1=2 ; y k1=2 ; uSE
j;k Rj;k xj1=2 ; y k1=2 ; uSW
j;k Rj;k xj1=2 ; y k1=2 .

where u = (w, hu, hv, hT)T.


L. Cai et al. / Applied Mathematical Modelling 31 (2007) 490498 493

The piecewise polynomial interpolant Rj,k(x,y), (2.4), may have discontinuities along the lines {xj1/2} and
{yk1/2}, which propagate with dierent right- and left-sided local speeds. To estimate them is a non-trivial
problem, but in practice we may use
8 9
<  of   
of  =
axj1=2;k max q  ;q  ; 3:15
: ou u  
ou u ;
j1=2;k j1=2;k
( 9
    =
y og  og 
aj;k1=2 max q  ;q  ; 3:16
ou  ou u ;
u
j;k1=2
j;k1=2

(  )
p  p 
a
j1=2;k max u gh ; u gh E ; 0 ; 3:17
W uj;k
uj1;k
( )
p  p 
a
j1=2;k min u  gh W ; u  gh E ; 0 ; 3:18
uj1;k uj;k
( )
p  p 
b
j;k1=2 max v gh ; v gh N ; 0 ; 3:19
uSj;k1 uj;k
( )
p  p 
b
j;k1=2 min v  gh ; v  gh N ; 0 ; 3:20
uSj;k1 uj;k

where q(A) denotes the spectral radius of a matrix A.


The 2-D semi-discrete central-upwind scheme can be written in the following form:
x x y y
d Hj1=2;k  Hj1=2;k Hj;k1=2  Hj;k1=2
uj;k t 
  Sj;k t; 3:21
dt D D
Now we have two approaches to calculate the fourth-order numerical uxes:

(a) in smooth regions, we use


a h i
j1=2;k
Hxj1=2;k fuNE 4fu E
fu SE

6a 
j1=2;k  aj1=2;k
j;k j;k j;k

a h i
j1=2;k
 
fuNW W SW
j1;k 4fuj1;k fuj1;k 3:22
6aj1=2;k  aj1=2;k
a  h i
j1=2;k aj1=2;k

uNW NE W E SW SE
j1;k  uj;k 4uj1;k  uj;k uj1;k  uj;k
6aj1=2;k  aj1=2;k

and

b h i
j;k1=2
Hyj;k1=2  guNW N NE
j;k 4guj;k guj;k
6bj;k1=2  bj;k1=2
b h i
j;k1=2 S
  guSW SE
j;k1 4guj;k1 guj;k1 3:23
6bj;k1=2  bj;k1=2
b  h i
j;k1=2 bj;k1=2 S
uSW NW N
j;k1  uj;k 4uj;k1  uj;k uj;k1
SE
 uNE
j;k ;
6bj;k1=2  b 
j;k1=2

(b) in rough regions, we use


fu 
j1=2;k fuj1=2;k axj1=2;k h i
Hxj1=2;k  u 
j1=2;k  uj1=2;k ; 3:24
2 2
494 L. Cai et al. / Applied Mathematical Modelling 31 (2007) 490498

and
gu 
ayj;k1=2 h
j;k1=2 guj;k1=2
i
Hyj;k1=2  uj;k1=2  u j;k1=2 ; 3:25
2 2
 2
T  2
T
2 huhv huhT 2 hvhT
where f hu; hu
wB
g
2
w  B ; ;
wB wB
, g hv; huhv hv
;
wB wB
g
2
w  B ; wB
.

Paralleling the procedure presented in Ref. [14], we impose 2-D weak local truncation errors,
fEnj;k : EuD ; /nj;k g (for the sake of brevity, the expression of fEnj;k g will not be addressed here). For our
scheme, fEnj;k g has a magnitude of O(D6) in smooth regions and of O(D2) in discontinue neighborhoods. In
this case, a criterion for adaption may be
(
if Enj;k =D4 > K; rough regions;
otherwise; smooth regions;
where K determines the tolerance for adaption. we notice that the method is not too sensitive to the choice of
the smoothness indicator constant K. So we select K = 1 in our numerical example.
Suppose that there is no adaption, (i) the resolution of the numerical solutions becomes lower if only imple-
menting the numerical uxes (3.24) and (3.25) or (ii) the sharper resolution can be obtained at the cost of
increasing the complexity in computation if only using the scheme numerical uxes (3.22) and (3.23).
The integration of source terms, Sj;k t, are
Z xj1=2 Z y k1=2
1 1
S j;k t 2 Sdx dy; 3:26
D xj1=2 y k1=2
Z xj1=2 Z y k1=2 p!
2 1 ug2 u2 v2
S j;k t 2 gw  B Bx dx dy 3:27
D xj1=2 y k1=2 w  B4=3
Z xj1=2 Z y k1=2 p!
3 1 vg2 u2 v2
S j;k t 2 gw  B By 4=3
dx dy 3:28
D xj1=2 y k1=2 w  B
Z xj1=2 Z y k1=2
4 1
S j;k t 2 T S Sdx dy; 3:29
D xj1=2 y k1=2
To preserve the stationary steady-states, the integration of source terms, Sj;k t, can be estimated as (see [4])
1
S j;k t 0; 3:30
"
2 g Bxj1=2 ; y k1=2  Bxj1=2 ; y k1=2 wNE NW
j;k  Bxj1=2 ; y k1=2 wj;k  Bxj1=2 ; y k1=2
S j;k t  
6 D 2
Bxj1=2 ; y k  Bxj1=2 ; y k wEj;k  Bxj1=2 ; y k wWj;k  Bxj1=2 ; y k
4 
D 2 #
Bxj1=2 ; y k1=2  Bxj1=2 ; y k1=2 wSEj;k  Bx SW
j1=2 y k1=2 wj;k  Bxj1=2 ; y k1=2
;
 ; 3:31
D 2
"
3 g Bxj1=2 ; y k1=2  Bxj1=2 ; y k1=2 wNE SE
j;k  Bxj1=2 ; y k1=2 wj;k  Bxj1=2 ; y k1=2
S j;k t  
6 D 2
Bxj ; y k1=2  Bxj ; y k1=2 wNj;k  Bxj ; y k1=2 wSj;k  Bxj ; y k1=2
4 
D 2 #
Bxj1=2 ; y k1=2  Bxj1=2 ; y k1=2 wNW
j;k  Bx SW
j1=2 y k1=2 wj;k  Bxj1=2 ; y k1=2
;
 ; 3:32
D 2
4
S j;k t 0. 3:33
L. Cai et al. / Applied Mathematical Modelling 31 (2007) 490498 495

If we intend to simulate the unsteady transport of pollutant with bed friction terms, we have to construct
1 4
unsteady discretizations of the integration of source terms: S j;k t and S j;k t depend on the type of the pollutant
1 4
source terms S, that is, S j;k t and S j;k t can be estimated by Simpsons rule or other special quadrature; while
20
2 g 4@Bxj1=2 ; y k1=2  Bxj1=2 ; y k1=2
S j;k t 
6 D
q 1
2 NW 2 NW 2
uNE NW
j;k uj;k gj;k1=2 uNEj;k uj;k vj;k vj;k
NE

2=3 A
4=3
2 wNE j;k  Bx j1=2 ; y k1=2 w NW
j;k  Bx j1=2 ; y k1=2
wNE NW
j;k  Bxj1=2 ; y k1=2 wj;k  Bxj1=2 ; y k1=2

2 q 1
0
E W 2 2 2
Bxj1=2 ; y k  Bxj1=2 ; y k u j;k u g
j;k j;k uEj;k uWj;k vEj;k vWj;k
4@ 2=3 4=3
A
D 2 wEj;k  Bxj1=2 ; y k wWj;k  Bxj1=2 ; y k
wEj;k  Bxj1=2 ; y k wWj;k  Bxj1=2 ; y k

0 2
Bxj1=2 ; y k1=2  Bxj1=2 ; y k1=2
@
D
q 1
2 SW 2 SW 2
uSE
j;k u SW
j;k g j;k1=2 uSEj;k uj;k vj;k vj;k
SE

A
4=3
22=3 wSE SW
j;k  Bxj1=2 ; y k1=2 wj;k  Bxj1=2 ; y k1=2
3
wSE  Bx j1=2 ; y k1=2 w SW
 Bx j1=2 ; y k1=2

j;k j;k 5 3:34
2

and 20
3 g 4@Bxj1=2 ; y k1=2  Bxj1=2 ; y k1=2
S j;k t 
6 D
q 1
SE 2 SE 2 SE 2
uNE
j;k u g
j;k j1=2;k uNEj;k uj;k vj;k vj;k
NE

2=3 A
4=3
2 wNE SE
j;k  Bxj1=2 ; y k1=2 wj;k  Bxj1=2 ; y k1=2

wNE SE
j;k  Bxj1=2 ; y k1=2 wj;k  Bxj1=2 ; y k1=2

2 q 1
0
2 2
Bxj ; y k1=2  Bxj ; y k1=2 uNj;k uSj;k g2j;k uNj;k uSj;k vNj;k vSj;k
4 @ 2=3 A
D 4=3
2 wNj;k  Bxj ; y k1=2 wSj;k  Bxj ; y k1=2
wNj;k  Bxj ; y k1=2 wSj;k  Bxj ; y k1=2

0 2
Bxj1=2 ; y k1=2  Bxj1=2 ; y k1=2
@
D
q 1
2 SW 2 SW 2
uNW SW
j;k uj;k gj1=2;k uNW
j;k uj;k vj;k vj;k
NW

A
4=3
22=3 wNW SW
j;k  Bxj1=2 ; y k1=2 wj;k  Bxj1=2 ; y k1=2
#
wNW SW
j;k  Bxj1=2 ; y k1=2 wj;k  Bxj1=2 ; y k1=2
 . 3:35
2
496 L. Cai et al. / Applied Mathematical Modelling 31 (2007) 490498

The semi-discrete central-upwind scheme (3.21) is a time-dependent ODEs, which can be solved by any sta-
ble ODE solver which retains the spatial accuracy of the scheme. In the numerical example below, we have
used the fourth-order Natural Continuous Extensions (NCE) of RungeKutta (RK) method, proposed in [18].

4. Numerical results

We solve the system (3.10)(3.13) in the square domain [300, 1100] [0, 1400] with the approach proposed
in the paper. The shape of the dam is given by
( h 2
i
min 200 y700
400
; 600 ; if 0 6 y < 700;
Cy
200; if 700 6 y 6 1400.
On the left of the dam, the initial conditions are w(x, y, 0) = 11, h(x, y, 0)u(x, y,0) = 100 and h(x, y, 0)
v(x, y, 0) = 0. On the right of the dam, the initial conditions are w(x, y, 0) = 10, h(x, y, 0)u(x, y, 0) = 0 and
h(x, y, 0)v(x, y, 0) = 0. Outow boundary conditions are imposed on all domain boundaries. The gravitational

Fig. 1. The solutions at time t = 30.


L. Cai et al. / Applied Mathematical Modelling 31 (2007) 490498 497

Fig. 2. Top view on T: (1) the pollutant concentration T of SaintVenant-type model and (2) the pollutant concentration T of new model.

constant g = 9.8, the mesh ratio is 0.2 and all computations are made on an uniform mesh 100 100 cells. The
non-at bottom consists of three elliptic-shape exponential humps, that is
h 2 2 2 2 2 2
i
Bx; y 4:5 ej1 x500 j2 y700 ej2 x300 j1 y600 ej2 x700 j1 y700 ;

with j1 = 104 and j2 = 103. The concentration of pollutant TS = 20 and the source term is
2
0:00001xy10002 0:0005xy2002
Sx; y; t 0:5e0:5t8 .
The water surface w and the pollutant concentration T of this initial-boundary value problem are shown in
Fig. 1, where Fig. 1(1) and (3) are solutions which come from the SaintVenant-type model (g = 0),
Fig. 1(2) and (4) are unsteady-state solution with g = 0.2. Due to the bed frictions, the pollutant concentration
T alters smoothly and large distortion should not appear, see Fig. 1(3)(4) and Fig. 2(1)(2).

5. Conclusions

We present the second-order adaptive CWENO-type central-upwind scheme for 2-D new model of pollu-
tant: the fourth-order adaptive semi-discrete central-upwind scheme based on CWENO reconstructions for
LHS of 2-D system (3.10)(3.13) and the second-order discretization of the RHS source terms. Numerical
results indicate that our method has high-resolution, non-oscillatory behavior and robustness.

Acknowledgements

The authors would like to thank the supports by the Center for High Performance Computing of North-
western Polytechnical University. All calculations were performed at the Center.

References

[1] A. Bermudez, M.E. Vasquez, Upwind methods for hyperbolic conservation laws with source terms, Comput. Fluids 23 (1994) 1049.
[2] P.G. Navarro, M.E. Hubbard, A. Priestley, Genuinely multidimensional upwinding for 2D shallow water equations, J. Comput. Phys.
121 (1995) 79.
[3] R.J. LeVeque, Balancing source terms and ux gradients in high-resolution godunov methods: the quasi-steady wave-propagation
algorithm, J. Comput. Phys. 146 (1998) 346.
[4] A. Kurganov, D. Levy, Central-upwind schemes for the SaintVenant system, Math. Model. Numer. Anal. 36 (2002) 397.
498 L. Cai et al. / Applied Mathematical Modelling 31 (2007) 490498

[5] L. Cai, W.X. Xie, J.H. Feng, J. Zhou, Computations of steady and unsteady transport of pollutant in shallow water, Math. Comput.
Simul., in press.
[6] E. Audusse, M. Bristeau, Transport of Pollutant in Shallow Water. A Two Time Steps Kinetic Method, INRIA Report, RR-4529.
[7] A. Chertock, A. Kurganov, Finite-volume-particle methods for models of transport of pollutant in shallow water, J. Sci. Comput., in
press.
[8] A. Chertock, A. Kurganov, On a hybrid nite-volume-particle method, Math. Model. Numer. Anal. 38 (2004) 1071.
[9] C. Zoppou, S. Roberts, Numerical solution of the two-dimensional unsteady dam break, Mathematical Sciences Institute, Research
Report MRR98-039.
[10] J.G. Zhou, D.M. Causon, C.G. Mingham, D.M. Ingram, The surface gradient method for the treatment of source terms in the
shallow-water equations, J. Comput. Phys. 168 (2001) 1.
[11] A. Kurganov, E. Tadmor, New high-resolution central schemes for non-linear conservation laws and convectiondiusion equations,
J. Comput. Phys. 160 (2000) 241.
[12] A. Kurganov, D. Levy, A third-order semidiscrete central schemes for conservation laws and convectiondiusion equations, SIAM
J. Sci. Comput. 22 (2000) 1461.
[13] A. Kurganov, S. Noelle, G. Petrova, Semi-discrete central-upwind schemes for hyperbolic conservation laws and HamiltonJacobi
equations, SIAM J. Sci. Comput. 23 (2001) 707.
[14] S. Karni, A. Kurganov, G. Petrova, A smoothness indicator for adaptive algorithms for hyperbolic systems, J. Comput. Phys. 178 (2)
(2002) 323.
[15] L. Cai, J.H. Feng, W.X. Xie, A CWENO-type central-upwind scheme for ideal MHD equations, Appl. Math. Comput. 168 (2005)
600.
[16] W.X. Xie, L. Cai, J.H. Feng, W. Xu, Computations of shallow water equations with high-order central-upwind schemes on triangular
meshes, Appl. Math. Comput. 170 (2005) 296.
[17] D. Levy, G. Puppo, G. Russo, A fourth-order central WENO schemes for multidimensional hyperbolic systems of conservation laws,
SIAM J. Sci. Comput. 24 (2002) 480.
[18] M. Zennaro, Natural continuous extensions of RungeKutta methods, Math. Comp. 46 (1986) 119.

You might also like