Professional Documents
Culture Documents
2014,26(4):512-522
DOI: 10.1016/S1001-6058(14)60058-5
Abstract: The hydrodynamics of geophysical flows in oceanic shelves, estuaries, and rivers are often studied by solving shallow
water equations under either hydrostatic or non-hydrostatic assumptions. Although the hydrostatic models are quite accurate and
cost-efficient for many practical applications, there are situations when the fully hydrodynamic models are preferred despite a larger
cost for computations. The present numerical model is implemented by the finite volume method (FVM) based on unstructured grids.
The model can be efficiently switched between hydrostatic and non-hydrostatic modules. The case study shows that for waves pro-
pagating along the bar a criterion with respect to the shallowness alone, the ratio between the depth and the wave length, is insu-
fficient to warrant the performance of shallow flow equations with a hydrostatic approach and the nonlinearity in wave dynamics can
be better accounted with a hydrodynamic approach. Besides the prediction of the flows over complex bathymetries, for instance, over
asymmetrical dunes, by a hydrodynamic approach is shown to be superior in accuracy to the hydrostatic simulation.
In the above equations, g is the gravity acceleration, where ch is an arbitrary Smagorinsky constant, var-
f = 2ω sin φ is the Coriolis term, φ is the latitude, ying in the range of 0.01-0.5[12].
ω is the Earth’s angular velocity, and ζ is the free
1.3 Numerical scheme
surface elevation. When the non-hydrostatic pressure
pn is ignored, the vertical momentum equation should 1.3.1 Outline of the numerical scheme
also be neglected, and the system (1)-(4) then degene- In the numerical method, a semi-implicit scheme
rates to the conventional shallow water equations with is used, with a parameter θ . The importance of the
two horizontal momentum equations and the continui- value of θ was shown in previous papers[5,13]. The
ty equation. value of θ is taken as 0.5 in the following computa-
tions. The computational domain is meshed by the un-
1.2 Turbulence model structured grids and the flow variables are taken at the
To close the system the eddy viscosity concept is cell centers thus giving the method a name of the CC
employed in the model and the vertical eddy viscosity scheme. For the advection terms, the TVD (Total Va-
coefficient ν tV is determined by solving the one-equa- riation Diminishing) scheme is used to calculate the
tion Spalart-Allmaras (SA) model[11]. The transport cell face flux. An arbitrary variable ϕ
f
at the cell
equation for ν is
face is calculated with a second order TVD scheme
2
Dν ν 1 1
= cb1Sν − cw1 f w + {∇ ⋅ [(ν + ν )∇ν ] +
f
ϕ = ϕc + ψ (rf )(ϕ D − ϕC ) (9)
Dt d
σ 2
cb 2 (∇ν ) 2 } (6) where f denotes the cell face, ϕ D and ϕC are, re-
spectively, the cell-centered variables in the down-
where wind and upwind nodes around the face f . Using a
1/6 flux limiter function ψ (rf ) , which is simply a linear
ν 1 + c6
χ ≡ , f w = g 6 w63 , g = r + cw2 (r 6 − r ) , function of rf , one can obtain a higher order TVD sc-
ν g + cw3
heme. The ratio rf is calculated by the method exten-
ν ν χ 3 ded by Darwish[14].
r≡ , S = S + 2 2 f v 2 , f v1 = 3 3 , On the cell faces a non-orthogonal horizontal
Sκ d
2 2
κ d χ + cv1
local coordinate (ξ , η ) is used to replace the Cartisan
coordinate ( x, y ) for the derivative calculations. From
χ 1 ∂u ∂u j
fv 2 = 1 − , Si j = i + the chain rule of differentiations, one obtains
1 + χ f v1 2 ∂x j ∂xi
1 1
and d is the distance to the solid wall. The model φx = (φξ yη − φη yξ ) , φ y = (φη xξ − φξ xη ) (10)
J J
constants include
where J = xξ yη − xη yξ , and ξ is directed from a
2
cb1 = 0.1355 , σ = , cb 2 = 0.622 , κ = 0.41 , control cell center to a neighbor cell center across the
3
common face, and η is directed from one vertex to
cb1 1 + cb 2 another on the same face anticlockwisely around the
cw1 = + , cw2 = 0.3 , cw3 = 2.0 , cv1 = 7.1 control cell center (Fig.1).
κ 2
σ
ν tV = ν f v1 (7)
1.3.2 Predictor step with hydrostatic pressure ηi* − ηin KBM ∂ q*x
Firstly, the continuity equation is discretized as
∆t
+θ ∑ ∆σ k +
k =1 ∂ x i , k
ζ in +1 − ζ in KBM ∂ qxn +1
∆t
+θ ∑ ∆σ k +
k =1 ∂ x i , k
KBM
∂ qn KBM ∂ q*
(1 − θ ) ∑ x ∆σ k + θ ∑
y
∆σ k +
k =1 ∂ x i , k
k =1 ∂ y i , k
KBM KBM ∂ q n +1
∂ qn
(1 − θ ) ∑ x ∆σ k + θ ∑ KBM ∂ q n
y
∆σ k +
(1 − θ ) ∑
y
k =1 ∂ x ∂ y ∆σ k = 0 (15)
i , k k =1 i , k
k =1 ∂ y i , k
KBM ∂ q n and
(1 − θ ) ∑
y
∆σ k = 0 (11)
k =1 ∂ y i , k q*xi − qxin ∂ζ * ∂ζ n
= Fqxin − gDθ − gD (1 − θ ) +
∆t ∂ x i ∂ x i
and the discretized momentum equations are written
as
∂ ν tv ∂ q*x
(16)
qxin +1 − qxin ∂ζ n +1 D ∂σ D ∂σ i
= Fqxin − gDθ −
∆t ∂ x i
q*yi − q yin ∂ζ * ∂ζ n
= Fq yin − gDθ − gD (1 − θ ) +
∆t ∂ y i ∂ y i
∂ζ n D ∂ pnn +1
gD (1 − θ ) − +
∂ x i ρ 0 ∂ x i ∂ ν tv ∂ q*y
(17)
∂ ν tv ∂ q n +1 D ∂σ D ∂σ i
x
(12)
D ∂σ D ∂σ i q*zi − qzin ∂ ν tv ∂ q*z
= Fqzin + (18)
n +1 n
∆t D ∂σ D ∂σ i
q yi −q yi ∂ζ n +1
= Fq yin − gDθ −
∆t ∂ y i Furthermore, discretizing the above equations, one ob-
tains
∂ζ n D ∂ pnn +1 ∂ζ *
gD (1 − θ ) − + q*xi , k ∆σ k = Fqxin , k ∆σ k − gD ∆t ∆σ kθ
∂ y i ρ 0 ∂ y i +
∂ x i
∂ ν tv ∂ q yn +1 ν tv q*xi , k −1 − q*xi , k
(13) ∆t 2 −
D ∂σ D ∂σ i D i , k −1/ 2 ∆σ k −1/ 2
ν tv q*xi , k − q*xi , k +1
q −qn +1
1 ∂ p ∂ ν tv ∂ qzn +1
n n +1
∆t 2 (19)
= Fqzin −
zi zi
+
n
D i , k +1/ 2 ∆σ k +1/ 2
∆t ρ0 ∂σ i D ∂σ D ∂σ i
(14) ∂ζ *
q*yi , k ∆σ k = Fq yin , k ∆σ k − gD ∆t ∆σ kθ +
where F is an operator that includes the explicit dis- ∂ y i
cretization of the convective terms, the horizontal vis-
cosity terms and the Coriolis term. ν q*yi , k −1 − q*yi , k
∆t tv2 −
Setting qx∗ , q∗y , qz∗ , qσ∗ , η ∗ as the temporary D i , k −1/ 2 ∆σ k −1/ 2
flow variables controlled by the hydrostatic pressure
ph , the discretized continuity and momentum equa- ν q*yi , k − q*yi , k +1
∆t tv2 (20)
tions are written as: D i , k +1/ 2 ∆σ k +1/ 2
516
q*zi , k − q*zi , k +1
ν ti , k +1/ 2 (21) Z 2i = [∆ t (1 − θ ) ∆σ 1 , , ∆ t (1 − θ ) ∆σ k , ,
∆σ k −1/ 2
∆ t (1 − θ ) ∆σ KBM ]
Hence, the governing equations are summarized in
matrix notation as: Ai is a series of tri-diagonal matrixes.
For the discretizated Eqs.(23) and (24), the inter-
∂ Qx* ∂ Q*y ∂Qn
ζ + Z1i
i
*
+ Z1i = ζ i − Z 2i x −
n mediate variables Qxi*− , Q*yi− are introduced, so
∂ x i ∂y i ∂ x i
Ain Qxi*− = Gxin (26)
∂Q n
y
Z 2i (22)
∂y Ain Q*yi− = G yin (27)
i
n ∂ζ * n n
The intermediate variables Qxi*− , Q*yi− can be di-
A Q =G −B
ix
*
xi xi i (23)
∂ x i rectly calculated by solving the linear tri-diagonal
equations. Subtracting Eqs.(26) and (27) into Eqs.(23)
and (24), respectively, one obtains:
∂ζ *
Aiyn Q*yi = G yin − Bin (24)
∂ y i ∂ζ *
Ain (Qxi* − Qxi*− ) = − Bin (28)
∂ x i
Aizn Qzi* = Gzin (25)
∂ζ *
The details of the matrixes are Ain (Q*yi − Q*yi− ) = − Bin (29)
∂ y i
∆ tτ s Rewriting the governing Eq.(22)
Gxin = Fqxin ,1∆σ 1 + , Fqxin , 2 ∆σ 2 , , Fqxin , k ∆σ k ,
ρ
∂ (Qx* − Qx*− ) ∂ (Q*y − Q*y − )
ζ i* + Z1i + Z 2i = ζi −
n
Fq n
∆σ KBM ] , T
∂ x ∂ y
xi , KBM i i
Fq yin , KBM ∆σ KBM ]T , Notating Qxi* − Qxi*− = Qxi′ , Q*yi − Q*yi− = Qyi′ in Eqs.(28)
and (29), and then substituting Eqs.(28) and (29) into
G = [ Fq ∆σ 1 , Fq
n n n
∆σ 2 , , Fq n
∆σ k , (30) to obtain
zi zi ,1 zi , 2 zi , k
∂ n −1 n ∂ζ *
Fqzin , KBM ∆σ KBM ]T , ζ i* − Z1i A B −
∂x ∂x i
where
f
D ∆ lis
APi ,sk = ∆ t ∆σ k (cos α is yη − sin α is xη ) ,
ρ0 J is ∆ξis
∆ t ∆ si ∆ t ∆ si
APi ,Tk = , APi ,Bk = ,
D ρ0 ∆σ k −1/ 2 D ρ0 ∆σ k +1/ 2
Fig.2 Experiment setup
NS
APi , k = ∑ AP + AP + AP s
i, k
T
i, k
B
i, k
s =1 In the numerical experiments the wave is genera-
ted by adding source terms in the governing equation.
pnin +1,s is the pressure in the s th neighbor cell, and To absorb the incident waves in the simulations a
dampening layer is set on the beach side of the com-
BPi , k is the integration of the right hand side of putational domain. The quadrilateral horizontal mesh
Eq.(42). resolution is 0.025 m, and the vertical first grid point
519
Fig.3 Measured and modeled time series of free surface at different locations over the bar
nearest to the bottom is adjusted to be about 1.5y + shape of the waves becomes more sharp-crested
(Locations 3-7), the computations of the hydrostatic
( y + = yu∗ / ν ) , which is required by the non-slip wall model near the crests start to develop spurious oscilla-
condition. Then the vertical grid is stretched with a tions while the non-hydrostatic model is still capable
ratio 1.15 upward until about 0.025 m equal to the ho- of the accurate simulation of the wave’s pattern.
rizontal mesh size. The integration time step is One of the principal assumptions in the shallow
0.0001s, and the total simulation takes about 30T flow approach in the hydrostatic modeling is that the
(wave peroid). vertical acceleration can be ignored due to the insigni-
Measured and computed time series of the water ficant size of the flow depth compared to the horizon-
surface elevation are compared for seven characteris- tal extensions. In the present case, the ratio h / λ (the
tic locations, as shown in Fig.3. Both models simulate water depth to the wave length) is about 0.1 in front of
the approaching waves quite accurately (Location 1) the bar and the results of the hydrostatic modeling
even when the waves become slightly asymmetrical agree well with the measurements (Location 1). The
due to the interaction with the bar (Location 2). As the shoaling of the waves on the bar is concomitant with
520
the increase of the wave amplitude and the decrease of same flow parameters. The three-dimensional domain
the wave length, though the ratio h / λ is insignifican- covers the dune within an area of 0.4 m long and
tly enlarged when the wave approaches the crest of the 0.2 m wide. The horizontal grid is uniform with a
bar. Thus despite the fact that the shallowness crite- scale of 0.0025 m while the vertical grid is stretched
rion holds true, the behavior of the waves in this en- from the bottom with a ratio 1.15. The point nearest to
vironment is considerably nonlinear. The non-hydro- the bottom is set within the viscosity boundary layer
static pressure term introduced in the momentum of about 1.6z + . At the inlet, the discharge and the
equations enhances the model’s ability to capture the eddy viscosity are provided by another simulation in a
wave dispersion[9]. On the other hand, the hydrostatic long channel with a flat bottom to get a fully deve-
model follows the shallow water dispersion rule, i.e., loped turbulent flow. A zero water level boundary
c = gh . As one can guess the sharpness of the wave condition is set at the outlet. Simulations are carried
near the crest, one can then estimate the convective out with both hydrostatic and non-hydrostatic models.
acceleration of the fluid in the vertical direction and The results of modeling are compared for six charac-
compare its value with the gravity acceleration. When teristic locations in vertical profiles of the stream ve-
this convective acceleration is significant, the mode- locities, as shown in Fig.4.
ling will require a non-hydrostatic approach.
free-surface hydrodynamics[J]. Ocean modeling, 2005, [16] OHYAMA T., BEJI S. and BATTJES J. A. Experimen-
10(2): 137-151. tal verification of numerical model for nonlinear wave
[9] ZHANG Jing-xin, LIU Hua and XUE Lei-ping. A ver- evolutions[J]. Journal of Waterway, Port, Coastal,
tical 2-D mathematical model for hydrodynamic flows and Ocean Engineering, 1994, 20(6): 637-644.
with free surface in σ coordinatet[J]. Journal of Hy- [17] BALACHANDAR R., POLATEL C. and HYUN B.-S.
drodynamics Ser. B, 2006, 18(1): 82-90. et al. LDV, PIV, and LES investigation of flow over a
[10] PHILLIPS N. A. A coordinate system having some spe- fixed dune[C]. Sedientation and Sediment Transport:
cial advantages for numerical forecasting[J]. Journal of Proceedings of the Symposium Held in Monte Verta.
Meteorology, 1957, 14: 184-185. Monte Verita, Switzerland, 2002, 171-178.
[11] SPALART P. R. Strategies for turbulence modelling [18] YUE W., LIN C. L. and PATEL V. C. Large eddy si-
and simulations[J]. International Journal of Heat and mulation of turbulent open-channel flow with free sur-
Fluid Flow, 2000, 21(3): 252-263. face simulated by level set method[J]. Physics of fluids,
[12] ZHANG Q. Y., CHAN E. S. Sensitivity studies with the 2005, 17(2): 1-12.
three-dimensional multi-level model for tidal motion[J]. [19] YUE W., LIN C. L. and PATEL V. C. Large-eddy si-
Ocean Engineering, 2003, 30(12): 1489-1505. mulation of turbulent flow over a fixed two-dimensional
[13] CHEN C., LIU H. and BEARDSLEY R. C. An un- dune[J]. Journal of Hydraulic Engineering, ASCE,
structured grid, finite-volume, three-dimensional, primi- 2006, 132(7): 643-651.
tive equations ocean model: Application to coastal [20] BALEN W. V., UIJTTEWAAL W. S. J. and BLAN-
ocean and estuaries[J]. Journal of Atmospheric and CKAERT K. Large-eddy simulation of a curved open-
Oceanic Technology, 2003, 20(1): 159-186. channel flow over topography[J]. Physics of Fluids,
[14] DARWISH M. S., MOUKALLED F. TVD schemes for 2010, 22(7): 1-17.
unstructured grids[J]. International Journal of heat
and Mass Transfer, 2003, 46(4): 599-611.
[15] BEJI S., BATTJES J. A. Experimental investigation of
wave propagation over a bar[J]. Coastal Engineering,
1993, 19(2): 151-162.