fluids

Article
A Three-Dimensional Flow and Sediment Transport
Model for Free-Surface Open Channel Flows on
Unstructured Flexible Meshes
Yong G. Lai 1, * and Kuowei Wu 2,3
1 Technical Service Center, U.S. Bureau of Reclamation, Denver, CO 80225-007, USA
2 Water Resources Planning Institute, Water Resources Agency, Taichung 413, Taiwan; Kuowei@wrap.gov.tw
3 Department of Civil Engineering, National Chiao Tung University, Hsinchu 300, Taiwan
* Correspondence: ylai@usbr.gov; Tel.: +1-303-445-2560




Received: 7 December 2018; Accepted: 20 January 2019; Published: 24 January 2019


Abstract: Three-dimensional (3D) hydrostatic-pressure-assumption numerical models are widely

used for environmental flows with free surfaces and phase interfaces. In this study, a new flow
and sediment transport model is developed, aiming to be general and more flexible than existing
models. A general set of governing equations are used for the flow and suspended sediment
transport, an improved solution algorithm is proposed, and a new mesh type is developed based
on the unstructured polygonal mesh in the horizontal plane and a terrain-following sigma mesh in
the vertical direction. The new flow model is verified first with the experimental cases, to ensure
the validity of flow and free surface predictions. The model is then validated with cases having
the suspended sediment transport. In particular, turbidity current flows are simulated to examine
how the model predicts the interface between the fluid and sediments. The predicted results agree
well with the available experimental data for all test cases. The model is generally applicable to
all open-channel flows, such as rivers and reservoirs, with both flow and suspended sediment
transport issues.

Keywords: 3D hydraulic model; 3D sediment model; turbidity current; polygonal mesh;

hydrostatic assumption

1. Introduction
Numerical models are gaining popularity for solving a wide range of environmental fluid
mechanics problems. When dealing with free-surface flows and sediment transport processes in
open channel cases (e.g., rivers and reservoirs), one-dimensional (1D) and two-dimensional (2D)
numerical models have been widely used. 1D models are cross-section averaged, while 2D models are
mostly depth averaged. Some of the widely used 1D flow and sediment models include HEC-RAS [1],
MIKE11 [2], CCHE1D [3], and SRH-1D [4]. 2D flow and sediment transport models include the research
models of Wu [5], Hung et al. [6], and Huang et al. [7], and widely used public-domain models, such as
CCHE2D [8], TELEMAC-MASCARET [9], UnTRIM [10,11], Delft3D [12], and SRH-2D [13,14]. 1D and
2D models will remain useful, particularly for applications with a relatively large spatial extent or over
a long period of time.
Three-dimensional (3D) numerical models are less used in river engineering but start to gain
attention owing to the rapid advancement of computer hardware. There are many instances where
1D and 2D models are insufficient for environmental modeling. Flow examples include rivers with
sharp bends, impacts of in-stream structures, and lake stratification; sediment transport examples are
the local scour in rivers, sediment sluicing at reservoirs, and turbidity currents in lakes and estuary,

Fluids 2019, 4, 18; doi:10.3390/fluids4010018 www.mdpi.com/journal/fluids

Fluids 2019, 4, 18 2 of 19

among many others. As a result, many 3D environmental fluid dynamic models have been developed.
The most general 3D models are based on the solution of the Navier-Stokes equations (called NS
models in this paper); such models are the most accurate and subject primarily to the accuracy of
the turbulence models. NS models, however, are resource intensive and may require the use of
supercomputers for practical environmental problems [15]. Therefore, most current 3D models made
the hydrostatic-pressure assumption in the vertical direction [12]. Such models, called 3D hydrostatic
models in this paper, have been widely adopted in environmental flow simulations, where 3D effect
is important, but the hydrostatic-pressure assumption is adequate. 3D hydrostatic models are much
easier to use and faster in simulation speed than the NS models. The hydro-static assumption is valid
for rivers where water depth is much smaller than river width (shallow water assumption), and lakes,
reservoirs and estuaries, where the vertical velocity is much smaller than the horizontal current flow.
This study reports the research and development of a new 3D hydrostatic model for flow
and sediment transport modeling of environmental problems in waterbodies with free surface and
fluid-sediment interface. The focus is on the simulation of rivers and reservoirs. The objective of the
study is to overcome some shortcomings of the existing models so that the new model may be general
and flexible for engineering applications. New contributions of the present study include an improved
solution algorithm and the use of a flexible mesh system. To our knowledge, no 3D models have been
reported in the literature that have adopted the proposed algorithm or used the new mesh type.

2. Literature Review
A number of 3D hydrostatic models have been developed for river, lake and reservoir, estuary,
and costal simulation. A list of such models with sediment transport capability were reviewed in [16].
Some of the popular models are briefly discussed next.
Princeton Ocean Model (POM) is one of the early 3D models developed by [17]; it was primarily
for high-resolution modeling of estuary and coastal ocean processes. The model handled the free
surface of tides and the constant-changing bed terrain with the sigma mesh in the vertical direction.
This traditional sigma mesh means that the vertical mesh coordinate follows both free surface and
terrain changes in the process of an unsteady simulation, but the number of mesh point is fixed.
POM adopted the finite-difference discretization method and used an orthogonal curvilinear coordinate
system. Later, Estuarine and Coastal Ocean Model with Sediment (ECOMSED) was developed,
which built upon POM [18,19]. A key model development was the inclusion of the sediment transport
capability in ECOMSED. Based on a similar modeling philosophy, Environmental Fluid Dynamics
Code–3D (EFDC3D) was developed [20], which contained a hydrodynamic solver, a sediment transport
module, and a bed module. The flow solver was similar to POM in terms of the governing equations
and solution algorithms, except that the free surface was solved using a preconditioned conjugate
gradient solver rather than the Alternating Direction Implicit (ADI) method. EFDC3D also adopted
the finite difference method and the orthogonal curvilinear or rectilinear coordinate system.
Delft3D is a comprehensive numerical modeling suite consisting of a number of modules and
covering a range of engineering areas [21]. It may be applied to both river and ocean environments.
Delft3D was initially developed for commercial use, but a portion of the suite is becoming freely
available for public use [22,23]. Delft3D simulates the unsteady flows resulting from tidal or
meteorological forcing with the effect of density differences due to temperature and salinity included.
The model adopted the orthogonal curvilinear mesh in the horizontal plane and offered both the sigma
and Z mesh in the vertical direction. Z mesh refers to a mesh system where the horizontal mesh plane
is layered vertically so that each layer is orthogonal to the vertical direction.
CH3D is another widely used 3D hydrostatic model developed primarily for river engineering
applications. The model and its variants were documented by a number of researchers [24–26].
CH3D is a finite-difference model adopting the structured but non-orthogonal curvilinear mesh in the
horizontal plane and the sigma mesh in the vertical direction. The use of the non-orthogonality of the
mesh is more suitable to the river environment than the orthogonal mesh adopted by earlier models.
Fluids 2019, 4, 18 3 of 19

ROMS-CSTMS is a 3D hydrostatic model developed and used in Europe, and it has the circulation
and sediment transport modeling capability [27]. The model used also a curvilinear orthogonal mesh
in the horizontal and a stretched, terrain-following mesh in the vertical. The model has been applied
to many cases; it was applied to the Waipaoa River continental shelf offshore of the North Island in
New Zealand [28]. Satisfactory results were reported in comparison with a 13-month field data set.
Most of the 3D hydrostatic models adopted a restrictive mesh type in the horizontal plane and
the sigma or Z mesh in the vertical direction. The structured, curvilinear mesh is the most widely
used horizontal mesh type in the existing models. These meshes are inflexible in representing the
complex geometry of environmental flows. A structured mesh, for example, is difficult to generate
as there are requirements on the number of mesh points and the restricted shape and quality of the
mesh cells. The vertical sigma or Z mesh has its own issues. With the sigma mesh, the governing
equations are transformed from the physical space to the computational space. The benefit is that
the vertical coordinate ranges from −1 to 0, which does not change with the free surface and bed
movement. A drawback, however, is that the vertical mesh points are inflexible, which may lead to
inadequate resolution around a density interface [29]. Significant errors were reported in the areas
with steep bottom topography [30]. For flows with density interfaces, a significant number of vertical
points may be needed to reduce the numerical error. The Z mesh was developed to overcome the
problem facing the sigma mesh; it did not make the coordinate transformation and kept the physical
coordinates so that vertical points may be easily moved adaptively. Another benefit of the Z mesh
is that different numbers of vertical points may be used in different zones. The horizontal mesh
planes, however, needed to be orthogonal to the vertical coordinate; this restriction caused the mesh
layer to not align on the density interface, leading to large numerical errors [31,32]. Further, the Z
mesh is not terrain-following, so the bed is represented only by the staircase approximation. Such a
zig-zag representation was found to cause inaccuracy in the bed shear stress computation and near-bed
horizontal advection [31,32].
In this study, a new mesh type is proposed—an unstructured, physical-coordinate sigma mesh.
In the horizontal plane, the unstructured polygonal mesh is generated; in the vertical direction,
the physical-coordinate based sigma mesh is used. The polygonal mesh is the most general mesh
type; other types are merely special cases. For example, the hybrid mesh of mixed triangular and
quadrilateral cells may be used by the proposed new mesh. The hybrid mesh was shown to be a
flexible and accurate type with the 2D depth-averaged hydraulic models [14]. With the vertical physical
coordinate mesh, the mesh point distribution may be arbitrarily assigned. Points can freely follow
interfaces and remove the restrictions of both the traditional sigma mesh and Z mesh. Key features of
the proposed new mesh type, different from the traditional meshes, are listed below:

• The horizontal mesh uses the polygonal cells;

• The vertical mesh points may be distributed arbitrarily conforming to free surface, bed, or interface
changes; and
• The physical-Cartesian-coordinate based governing equations are solved directly without the
coordinate transformation.

3. Governing Equations

3.1. Flow Equations

The present model makes the following assumptions: (a) the vertical pressure distribution is
hydrostatic; (b) the density variation impacts only the buoyancy force (Boussinesque assumption);
and (c) the flow-sediment mixture is incompressible. We concern primarily with rivers and reservoirs
Fluids 2019, 4, 18 4 of 19

so that special costal processes, such as ocean wave-generated processes and Coriolis force, are not
included. The above assumptions lead to the following flow equations:

∂U ∂V ∂W
+ + =0 (1)
∂x ∂y ∂z
 
g
Z ζ
∂U ∂UU ∂VU ∂WU ∂ζ ∂ρ ∂Txx ∂Txy ∂ ∂U
+ + + = −g − dz0 + + + ( ν + υV ) (2)
∂t ∂x ∂y ∂z ∂x ρ Z ∂x ∂x ∂y ∂z ∂z
 
g
Z ζ
∂V ∂UV ∂VV ∂WV ∂ζ ∂ρ 0 ∂Tyx ∂Tyy ∂ ∂V
+ + + = −g − dz + + + ( ν + υV ) (3)
∂t ∂x ∂y ∂z ∂y ρ Z ∂y ∂x ∂y ∂z ∂z
Z ζ Z ζ
∂ζ ∂ ∂
+ Udz0 + Vdz0 = 0 (4)
∂t ∂x ZB ∂y ZB

In the above, t is time; (x, y, z) are the Cartesian coordinates (x and y are in the horizontal directions; z
is the vertical direction opposite to the gravity); ρ is the mixture density; U, V and W are the mean
velocity components along the Cartesian coordinates x, y, z, respectively; g is the acceleration due to
gravity; ζ is the free surface elevation; Txx , Txy , Tyx , and Tyy are the horizontal turbulence stresses;
νV is the vertical turbulent viscosity. Note that the horizontal and vertical turbulences are treated
separately due to the quite different turbulent characteristics caused by the different spatial scales
of typical open channel flows. It is a practice widely adopted by almost all hydro-static assumption
models [12].
The horizontal turbulence stresses in Equations (2) and (3) are computed by:

∂U 2k
Txx = 2(ν + v H ) − (5a)
∂x 3
∂V 2k
Tyy = 2(ν + v H ) − (5b)
∂y 3
 
∂U ∂V
Txy = Tyx = (ν + v H ) + (5c)
∂y ∂x
where ν is the kinematic fluid viscosity, νH is the horizontal turbulent viscosity, and k is turbulent
kinetic energy.
Various turbulence models may be used to compute the horizontal and vertical viscosities. In this
study, the large eddy simulation (LES) model is used for the horizontal turbulence and k-ε model is
used for the vertical turbulence. The horizontal LES is based on the Smagorinsky subgrid scale model,
following the formulation of [33]. The horizontal eddy viscosity is computed by:

νH = νSGS + νV (6)

"   #1/2
∂U 2 ∂V 2 ∂V 2
   
∂U
νSGS =α 2 + + +2 A2D (7)
∂x ∂y ∂x ∂y

In the above, A2D is the 2D horizontal cell area and α is a model constant (normally taken to be 0.11
but can be a user calibration parameter).
The k-ε model is used to compute the vertical viscosity; the formulation of [26] is adopted as:

k2
νV = Cµ (8)
ε
Fluids 2019, 4, 18 5 of 19

In the above, k is the turbulence kinetic energy and ε is the turbulence dissipation rate. The 3D
transport equations for k and ε are expressed as:
 
∂k ∂Uk ∂Vk ∂Wk ∂ νV ∂k
+ + + = +G+B−ε (9)
∂t ∂x ∂y ∂z ∂z σk ∂z

ε2
 
∂ε ∂Uε ∂Vε ∂Wε ∂ νV ∂ε ε
+ + + = + Cεq ( G + Cε3 B) − Cε2 (10)
∂t ∂x ∂y ∂z ∂z σε ∂z k k
where G is the turbulence generation due to vertical velocity gradient and B is turbulence generation
due to buoyancy force. The generation term G is computed by:
" 2  2  2  2  2 #
∂U ∂V ∂U ∂V ∂U ∂V
G = νV + +2 +2 + + (11)
∂z ∂z ∂x ∂y ∂y ∂x

The buoyance production represents energy conversion between turbulent kinetic energy and
potential energy in a stratified flow; it is computed by:

νV g ∂ρ
B= (12)
σρ ρ ∂z

In general, σρ = 1 is assumed for sediment induced stratification cases. Many turbulence model
constants have been proposed and the following standard values [34] are used: Cµ = 0.09 and Cε3 = 0
for stable stratification; Cε3 = 1 for unstable stratification.

3.2. Sediment Equations

A general set of sediment governing equations are adopted. The suspended sediments are
divided into a number of size classes so that the non-uniform processes are taken into account.
A non-equilibrium and fully unsteady sediment transport equation is utilized for each size class.
The sediment concentration is tightly coupled to the flow in that the flow dictates the sediment
concentration distribution, while the sediment alters the flow simultaneously through the baroclinic
term in the momentum equations. For size k, the following sediment transport equation is solved:

∂(W − ωk )Ck
     
∂Ck ∂UCk ∂VCk ∂ ∂C ∂ ∂C ∂ ∂C
+ + + = D Hk k + D Hk k + DVk k (13)
∂t ∂x ∂y ∂z ∂x ∂x ∂y ∂y ∂z ∂z

In the above, the subscript k designates the variable associated with the sediment size class k; Ck
is the volumetric concentration; ωk is the fall velocity; D Hk and DVk are the horizontal and vertical
diffusivities computed by:
ν ν
DVk = V ; D Hk = H (14)
σCk σCk
In the above, σCk is the Schmidt parameter that is assumed to be a constant from 0.5 to 1.0 [35].
The fall velocity is computed by the method of van Rijn [36].
The solution of Equation (13) needs a special boundary condition on the stream bed. The net
sediment flux on the bed reflects the net sediment exchange rate between sediments in water and those
on the bed; it is not zero unless the flow reaches equilibrium. The net sediment rate out of the water
column is computed by [35]:
∂C
ωk Ck + DVk k = Dk − Ek (15)
∂z
Fluids 2019, 4, 18 6 of 19

where Dk = ωk Ck and Ek are the sediment deposition and entrainment rate, respectively.
The entrainment rate is computed by:
(
ωk Ck∗ loose bed with unlimited supply
Ek = (16)
Min ωk Ck∗ , ωk Ck


fixed bed without supply

In the above, entrainment rate is proportional to the local equilibrium concentration near the bed.
The equilibrium concentration is determined by an empirical equation assuming unlimited sediment
supply from the bed. In the numerical implementation, (Dk − Ek ) is the net sediment rate out of the
water column and is added to bed as the depositional sediment.
The equilibrium concentration equation is often computed at a reference height δ above the bed,
but the interpretation of the bed location (z = 0) varies among researchers [37]. Theoretically, z = 0
should be the edge of the bed-load layer, but practically, such an edge is hard to measure as there is no
clear boundary between the bedload and the suspended load. In this study, the z = 0 bed is chosen
to be located at about ( 0.15 ∼ 0.35) d50 below the top of the roughness element, as recommended
by [38], where d50 is the medium size of the roughness elements on the bed.
Seven equilibrium equations were reviewed in [39]; later a more comprehensive list were provided
by [40]. In this study, the equilibrium concentration is determined by the formula of [41] as follows:

0.331(θ − 0.045)1.75
C∗ = (17a)
1+ 0.331
0.46 ( θ − 0.045)1.75

u2τ
θ= (17b)
(γ − 1) gd
In the above, uτ is the bed shear velocity, d is the representative sediment diameter such as d50 , and g
is the acceleration due to gravity.

4. Numerical Method
The above governing equations are first discretized on a suiTable 3D mesh; the discretized
equation set is then solved with a robust solution algorithm. The 3D mesh generation consists of two
steps in this study: a 2D horizontal mesh generation and an automatic vertical mesh point creation.
The 2D mesh may assume polygonal shapes and the 2D nodes are assigned with the bed elevation.
The vertical nodal distribution is carried out automatically by the model given: (a) the water surface
elevation predicted by the model, (b) the number of vertical nodes, and (c) the distribution criterion.
Vertical nodal distribution is performed every time step so that the free surface and bed changes are
followed. Each 3D mesh cell consists of a closed set of faces—a set of vertical faces coinciding with the
2D mesh and the top and bottom faces.
The discretization of the governing equations is performed using the finite-volume method.
The partial differential equations (PDE) are integrated over 3D mesh cells. The Gaussian integral is
used to transform the volume integral to flow fluxes on mesh faces. They are described next.

4.1. Flow Discretization

The continuity Equation (1) is discretized for a 3D cell k as:

∀k − ∀0k → →
 

∆t
+ ∑ Vf k · n f A f k + FTk + FBk = 0 (18a)
f aceV

→ → → →
   
FTk = VTk ·n Tk A Tk − FgTk ; FBk = VBk ·n Bk A Bk − FgBk (18b)
Fluids 2019, 4, 18 7 of 19

→ →
In the above, subscript k refers to a variable associated with the cell k; ∀ is cell volume; Vf k , VTk ,
→ → →
and VBk are the flow velocity vector at the centroids of the vertical, top, and bottom faces; n f , n Tk ,
→
and n Bk are the unit normal vectors of the vertical, top, and bottom faces; A f k , A Tk , and A Bk are the
areas of the vertical, top, and bottom faces; summation ∑ is over vertical faces only; FgTk and
f aceV
FgBk are volume fluxes at the top and bottom faces due to vertical mesh movement. In our solution
procedure, Equation (18a) is used to compute the vertical flow velocity once the horizontal velocities
are computed.
The 3D mesh nodes are allowed to move in the vertical direction to conform to the free surface
and bed elevation changes. The moving mesh fluxes, FgTk and FgBk , are computed with the method
of [42]. With this approach, a new mesh is first formed and the new cell volume is then computed.
The moving mesh fluxes, FgTk and FgBk , are finally computed to satisfy the volume conservation
∀ −∀0
constraint, i.e., k∆t k = FgTk − FgBk .
The depth-averaged continuity Equation (4) is used to compute the free surface elevation in
the flow solver. Instead of discretizing (4) directly, like most models did, we derive the discretized
depth-averaged equation by summing up Equation (18a) vertically. Doing so over all vertical 3D cells
at a 2D cell k leads to the following:

∑k ∀k − ∑k ∀0k → →
 

∆t
+ ∑ Ff ·n f =0 (19a)
f aceV

→ −→
 
Ff = ∑ A f k ·Φ f k =0 (19b)
k
→
In the above, Ff is the volume flow flux through the entire vertical face from bed to free surface;
−→  
Φ f k = U f k , Vf k is the 2D velocity vector comprised of only the two horizontal velocity components.
The momentum Equations (2) and (3) are rewritten in the following form:
→ → → →
     
∂Φ →→ ∂  ∂Φ  ∂Φ  ∂Φ  →∗∗
 
∂ ∂
+ ∇· V Φ = − g∇ζ + νH +  νH + νV + SΦ (20a)
∂t ∂x ∂x ∂y ∂y ∂z ∂z

    "Z ! #
∂ ∂U ∂ ∂V ∂ ζ ρ
∗∗
SU = νH + νH −g − 1 dz0 (20b)
∂x ∂x ∂y ∂x ∂x z ρre f
    "Z ! #
∂ ∂U ∂ ∂V ∂ ζ ρ
∗∗
SV = νH + νH −g − 1 dz0 (20c)
∂x ∂y ∂y ∂y ∂y z ρre f
→
In the above Φ = (U, V ) is the velocity vector in the horizontal plane; ∇ is the 3D Kroneker
Delta operator, ρre f is a constant reference density, e.g., the clear water or average mixture density.
Discretization of Equation (20a) is also carried out with the finite volume method. In the following,
the subscript k is dropped for clarity.
The discretization of the unsteady term is performed using the first-order Euler scheme.
The convection term is discretized as a second-order scheme:
y 
→→
 
→ →

−→

−→ −→

∇· V Φ d∀ = ∑ Vf ·n f A f Φ f + FT Φ T − FB Φ B (21)
f aceH
Fluids 2019, 4, 18 8 of 19

In the above, subscripts f, T, and B refer to the values at the centroids of the vertical, top,
−→
and bottom cell faces; Φ f is obtained with a second-order averaging equation involving known values
at the neighboring cells as [43]:
→ →
  → →

−→ −→
−→ δ2 Φ1 + δ1 Φ2 δ r edge × n f · δ1 r 2 − δ2 r 1 →
Φf = − ∑ Φedge (22)
δ1 + δ2 edge
(δ1 + δ2 ) A f

→
In the above, stands for summation over all edges of the cell face; Φedge is horizontal velocity at the
→ → →
center of the edge; δ r edge is the edge distance vector; r1 and r2 are left-to-right distance vectors between
→ → → → →
the neighboring cell center to the face centroid (n f is used to decide left or right); δ1 = r1 ·n f ; δ2 = r2 ·n f ;
and subscripts 1 and 2 denote values of the left and right cells. From now on, the above averaging of a
→
variable from cell center to cell face is denoted as hΦi with “< >” as the averaging operator.
Discretization of the diffusion term is more involved and follows the method of [43]. A detailed
derivation is omitted and the final second-order accuracy discretized equation for an arbitrary polygon
is expressed as:
→ → →
      
y
 ∂ νH ∂Φ  + ∂ νH ∂Φ  + ∂ νV ∂Φ d∀ =
−→
∂x ∂x ∂y ∂y ∂z ∂z ∑ Df (23a)
F

−→ −→ −→ edge →
 
Df = Dn Φ 2 − Φ 1 + ∑ Dc Φedge (23b)
edge
−→ −→ →  →
→ −→
ν f ·nf Af edge
r1 + r2 × ν f ·δ r edge → −→ −→ −→
Dn = ; Dc = ; ν = νH n x ex + νH ny ey + νV nz ez (23c)
(δ1 + δ2 ) (δ1 + δ2 )
In the above, ∑ is summation over all faces of the cell (vertical, top, and bottom faces).
F
The water elevation gradient term is the same for all vertical 3D cells that share the same horizontal
mesh (i.e., the 2D mesh). It is discretized as a second-order accuracy scheme by:
y g∀ −→
g∇ζd∀ = g∀∇ζ =
AH ∑ dF nF ζ F (24)
F2D

In the above, ∑ is summation over all edges of the 2D mesh cell; A H is horizontal area of the
F2D
cell, ζ F is free surface elevation at the center of the 2D cell edge, d F is edge distance of the 2D cell,
−→
and n F is unit normal vector on an edge of the 2D cell.
The final discretized horizontal momentum equations, say at a cell P, may be assembled as:
−→ −→ →∗
AP ΦP = ∑ Anb Φnb − g∀∇ζ + S Φ (25)
nb

where ∑ is summation over all neighboring cells which share the same faces with cell P.
nd

4.2. Flow Solution Procedure

A special procedure is used to obtain flow flux through vertical faces of a 3D mesh cell [43].
The horizontal velocity at a cell face is computed by:

−→ → g∀ g∀
Φf = hΦi − h i∇ζ f + h ∇ζ i (26)
AP AP
Fluids 2019, 4, 18 9 of 19

where Equation (26) is vertically summed to obtain the volumetric flow flux vector as follows:

−→ −→ −→ g∀ g∀
Ff = ∑ A f kΦ f k = ∑ A f k h Φk i − D f ∇ζ f + ∑ A f k h APkk ∇ζ i; Df = ∑ A f k h APkk i (27)
k k k k

insertion of Equation (27) into Equation (19a) leads to the following water elevation equation:

∑ ∀k − ∑ ∀0k "
−→
#
 −→
 g∀k −→
∑ ∑ ∑ ∑ A f k h APk ∇ζ i · n f
k k
− Df ∇ζ f · n f = − A f k h Φk i + (28)
∆t f aceV f aceV k k

and Equation (28) is used to solve the water surface elevation.

The iterative solution procedure proceeds as follows. First, the two momentum equations in
Equation (25) are solved. That is, given all variables at time level 0, the 3D horizontal velocities are
obtained (starred superscript denotes provisional values at the new time n):

−→∗ Anb −→∗ g ∀0 →0

ΦP =∑ Φ nb − ∇ζ 0 + S Φ (29)
nb
AP AP

the flow flux at vertical faces is updated using Equation (27) as:

→∗ −→∗ g∀0k
Ff = ∑ A f k h Φ k i − D f ∇ζ 0f + ∑ A f k h ∇ζ 0 i (30)
k k
A Pk

−→n →n
The corrector step is next performed so that the new values of Φ , ζ n , and F f are obtained
that satisfy the continuity Equation (28). Expressed in the incremental form, the following equations
are obtained:
−→0 A −→n −→∗ g ∀0
Φ P = ∑ nb ( Φ nb − Φ nb ) − ∇ζ 0 (31a)
nb
A P A P

→0 −→0
Ff = ∑ A f k h Φ k i − D f ∇ζ 0f (31b)
k

∑ ∀nk − ∑ ∀0k →∗ −→
−→
 
∑ ∑
k k
− D f ∇ζ 0f · n f = − (F f · nf ) (31c)
∆t f aceV f aceV

−→0 −→n −→∗ −→0 −→n −→∗ →0 →n →∗

In the above, Φ P = Φ P − Φ P , Φ k = Φ k − Φ k , ζ 0 = ζ n − ζ 0 and F f = F f − F f .
Equation (31c) is the elevation correction equation that is solved with the ILU preconditioned CGS
solver [14]. With the new water elevation obtained at time level n, the new velocity and flow fluxes are
updated and the vertical velocity at time level n is computed using Equation (18a). This completes
one cycle of the iterative solution of the governing equations. A number of iterations are usually
performed to achieve the convergence within a time step. Note that the sediment transport equation is
also included in the above iterative solution process if sediment transport is to be simulated.

4.3. Sediment Discretization

Discretization of the suspended sediment Equation (13) is carried out similarly to the momentum
equations. The concentration Equation (13) is rewritten in the following form:

→
       
∂C ∂ ∂C ∂ ∂C ∂ ∂C
+ ∇· VC = DH + DH + DV + SC (32a)
∂t ∂x ∂x ∂y ∂y ∂z ∂z

∂ωC
SC = (32b)
∂z
Fluids 2019, 4, 18 10 of 19

The discretization of the unsteady term adopts the first-order Euler scheme and the second-order
convection term is discretized as:
y 
→
 
−→ −→

∇· VC d∀ = ∑ Vf · n f A f C f + ( FT CT − FB CB ) (33)
f aceH

where C f , CT , and CB are the concentration at the centroids of vertical, top, and bottom cell faces.
Discretization of the diffusion term is second-order and is expressed as:
y ∂  ∂C

∂

∂C

∂

∂C

−→
∂x
DH
∂x
+
∂y
DH
∂y
+
∂z
DV
∂z
d∀ = ∑ Df (34a)
F

∑ Dc
edge
D f = Dn (C2 − C1 ) + Cedge (34b)
edge

edge
The definition of Dn and DC are discussed before for the momentum equations; ∑ is a sum over all
F
faces of the cell (vertical, top, and bottom faces).
The final discretized concentration equation, say at a cell P, is rearranged in the following form:

A P CP = ∑ Anb Cnb + SC∗ (35)

nb

where ∑ is summation over all neighboring cells which share the same faces with cell P.
nb

5. Model Verifications
Flow and suspended sediment transport cases are selected to test and verify the new numerical
model, and they are reported next. In the discussion, the proposed new 3D model is named SRH-3D,
consistent with the 2D model SRH-2D developed in [13,14].

5.1. Diversion Channel Flow

Channel bifurcation occurs frequently in natural rivers. Flow features are complex at the juncture.
A laboratory case studied in [44] is selected to verify the new flow solver. The case was also simulated
by other 2D and 3D models in the past [13].
The solution domain consists of a main channel, 6.0 m in length and 0.3 m in width, and a side
diversion channel normal to the main channel, 3.0 m in length and 0.3 m in width. Two horizontal
meshes are used: one is the same as the 2D mesh of [13], which used pure quadrilaterals, and the other
adopts the triangle-quadrilateral hybrid mesh. A total of 16 vertical cells are selected leading to 76,800
cells for the first mesh and 107,280 cells for the mixed mesh. A close-up view of the mixed mesh is
shown in Figure 1a.
The flow conditions are as follows. The main channel has a discharge of 0.00567 m3 /s (applied at
X = −3.0 m) and water surface elevation of 0.0555 m at the exit (X = 3.0 m); the water surface elevation is
0.0465 m at the exit of the diversion channel (Y = 3.3 m). The simulation with the two meshes produces
essentially the same results, so only results from the mixed mesh are reported herein. The predicted
flow pattern is shown in Figure 1b. Key flow features are predicted well, such as flow acceleration
near the junction and into the diversion channel and the flow recirculation in the diversion channel.
Simulated results are compared with the measured data of [44]; the water surface elevation is in
Figure 2 and the depth-averaged velocity is in Figures 3 and 4 in both the main and side channels.
Good agreements are obtained between the model results and the data. The SRH-3D and SRH-2D
results are similar, which points to the fact that 2D models are applicable for such flows, at least for the
case simulated.
 predicted flow
acceleration pattern
near is shown
the junction in into
and Figure
the1b. Key flow
diversion features
channel arethe
and predicted well, such as
flow recirculation in flow
the
acceleration near the junction and into the diversion channel and the flow recirculation
diversion channel. Simulated results are compared with the measured data of [44]; the water surface in the
diversion channel. Simulated results are compared with the measured data of [44];
elevation is in Figure 2 and the depth-averaged velocity is in Figures 3 and 4 in both the main and the water surface
elevation
side is in Good
channels. Figureagreements
2 and the depth-averaged velocity
are obtained between theismodel
in Figures 3 and
results and4thein data.
both the
Themain
SRH-3D and
sideSRH-2D
and channels. Goodare
results agreements are obtained
similar, which points tobetween the model
the fact that results
2D models areand the data.
applicable forThe SRH-3D
such flows,
Fluids
and2019, 4, 18 results are similar, which points to the fact that 2D models are applicable for such flows,
SRH-2D 11 of 19
at least for the case simulated.
at least for the case simulated.

(a) (b)
(a) (b)
Figure 1. A portion of the mixed mesh at the juncture (a) and the predicted surface flow velocity for
Figure
the
Figure flow 1.diversion
1. A A portion
portion of
case
of thethe
(b).mixedmesh
mixed meshatatthe
thejuncture
juncture (a)
(a) and
and the
the predicted
predictedsurface
surfaceflow
flowvelocity
velocityforfor
the flow diversion case
the flow diversion case (b). (b).

(a) (b)
Fluids 2018, 3, x FOR PEER REVIEW
(a) (b) 11 of 19
Figure
Figure 2. 2. Comparisonofofwater
Comparison waterelevation
elevationalong
along both walls of
of the
themain
mainchannel
channel(a)
(a)and
andthe side
the channel
side (b).(b).
channel
Figure 2. Comparison of water elevation along both walls of the main channel (a) and the side channel (b).

Figure3.3.Comparison
Figure Comparisonofof x-velocity
x-velocity profiles
profiles at selected
at selected x locations
x locations inmain
in the the main channel
channel (Solid(Solid Blue
Blue Line:
Line: SRH-3D; Red Dashed Line: SRH-2D; Symbol:
SRH-3D; Red Dashed Line: SRH-2D; Symbol: Experiment). Experiment).
 Figure 3. Comparison of x-velocity profiles at selected x locations in the main channel (Solid Blue
Fluids 2019, 4, 18 12 of 19
Line: SRH-3D; Red Dashed Line: SRH-2D; Symbol: Experiment).

Figure 4. Comparison of V velocity profiles at selected y locations in the side channel (Solid Blue Line:
Figure 4. Comparison of V velocity profiles at selected y locations in the side channel (Solid Blue Line:
SRH-3D; Red Dashed Line: SRH-2D; Symbol: Experiment).
SRH-3D; Red Dashed Line: SRH-2D; Symbol: Experiment).
5.2. Flow in a Sharply Curved Bend
5.2. Flow in a Sharply Curved Bend
An open channel flow with a sharp 180◦ rectangular bend was experimentally studied in [45].
A bendAnwith
openachannel flow with a sharp
mean radius-to-width ratio180°
of rectangular
3.0 and less bend was experimentally
is considered strongly curvedstudied
andinexhibits
[45]. A
bend with a mean radius-to-width ratio of 3.0 and less is considered strongly
highly 3D flow characteristics [46]. The Rozovskii bend has a mean radius-to-width ratio of 1.0, so it curved and exhibits
highly
belongs3D to flow
the verycharacteristics
sharp bend[46]. The Rozovskii
category. The same bendcase washas subject
a mean to radius-to-width ratio of 1.0,
a number of numerical so it
studies,
belongs
such as to the depth-averaged
a 2D very sharp bend category.modelingThe sameacase
in [46], 2D was subjectexplicitly
modeling to a number of numerical
taking studies,
into account the
such as a 2D depth-averaged modeling in
secondary flow in [47], and a 3D NS solver modeling in [48]. [46], a 2D modeling explicitly taking into account the
secondary
The modelflow in [47], and
domain anda bed
3D NS solver modeling
geometry are showninin[48]. Figure 5a. The approach and exit channels
The model
are straight, 1.6 m domain and bed
in length, and geometry are shown
0.8 m in width. The inbendFigure
has 5a. The approach
a radius andthe
of 0.4 m for exitinner
channels
wall.
are straight, 1.6 m in length, and 0.8 m in width. The bend has a radius
The numerical modeling parameters are as follows. The entire channel has a flat smooth bottom. of 0.4 m for the inner wall.
The numerical
The discharge at modeling parameters
the entrance is 0.0123aremas follows.
3 /s, resultingTheinentire channel
an average has a flat
velocity smooth
of 0.265 m/s,bottom.
ReynoldsThe
discharge at the entrance is 0.0123 m 3/s, resulting in an average velocity of 0.265 m/s, Reynolds
number of 15,600, and Froude number of 0.11. The water elevation at the model exit is extrapolated from
number of 15,600,data.
the experimental and TheFroude number
horizontal of 0.11.
mesh Thethe
adopts water
mixed elevation
trianglesat (two
the model
zones)exit
andisquadrilaterals
extrapolated
from the
(Figure 5a).experimental data. The
A total of 20 vertical meshhorizontal mesh adopts
cells are chosen, leading to thea total
mixed triangles
of 165,280 3D(two
meshzones)
cells. and
quadrilaterals (Figure 5a). A total of 20 vertical mesh cells are chosen, leading
Predicted water depth along the inner and outer sidewall is compared with the experimental data to a total of 165,280 3D
mesh
in cells.
Figure
Fluids 2018,5b. It is PEER
3, x FOR seen REVIEW
that the water surface elevation is higher at the outer bank than at the12inner of 19
Predicted
bank—the watersuper-elevation
expected depth along the inner
effect. and
The outerresults
model sidewallmatchis compared
well alongwith the experimental
the outer wall and the
data in Figure
prediction
elevation is than 5b.
theIt2Dis seen
under-predicted model,that the water
indicating
along the surface
that
inner vertical
wall. elevation
The variation
SRH-3D is of
higher at the outer
the velocity
model indeed and
has athebank than
secondary
better at flow
predictionthe
inner
than bank—the
produced
the 2Dby theexpected
model, sharp super-elevation
bend
indicating have effect.onThe
some effects
that vertical variation ofmodel
the flow.
the resultsand
velocity match well along flow
the secondary the outer wall
produced
and the elevation is under-predicted
by the sharp bend have some effects on the flow. along the inner wall. The SRH-3D model indeed has a better

(a) (b)
Figure5.5.Model
Figure Modeldomain
domainandand
thethe horizontal
horizontal mixed
mixed mesh
mesh (a) aand
(a) and a comparison
comparison of the of the depth
water water along
depth
along the inner and outer sidewall for the bend case
the inner and outer sidewall for the bend case (b). (b).

Predicted depth-averaged velocity is compared with the experimental data in Figure 6. The 3D
model results agree with the measured data well. It is noted that the 2D model results agree also well
with the data, contrary to the finding of [47]. The improvements of the 3D model over the 2D are
mainly in the shift of the maximum velocity from the inner to outer sidewalls when flow comes out
of the bend. This rapid shift can be explained by the falling transverse water elevation slope and
release of the additional momentum by the secondary flow when the radius of curvature at the bend
 (a) (b)
Fluids 2019, 4, 18 5. Model domain and the horizontal mixed mesh (a) and a comparison of the water depth13 of 19
Figure
along the inner and outer sidewall for the bend case (b).

Predicted
Predicteddepth-averaged
depth-averaged velocity
velocityisiscompared
comparedwith withthe the experimental
experimental datadata in
inFigure
Figure6.6.The
The3D3D
model results agree with the measured data well. It is noted that the 2D model
model results agree with the measured data well. It is noted that the 2D model results agree also well results agree also
wellwith
withthe
thedata,
data, contrary
contrary to to
thethe finding
finding of [47].
of [47]. TheThe improvements
improvements of 3D
of the themodel
3D model
over over
the 2Dtheare
2D
are mainly
mainlyin inthe
theshift
shiftofofthe
themaximum
maximum velocity
velocity from
from thetheinnerinner to outer
to outer sidewalls
sidewalls whenwhenflow flow
comes comes
out
out of
of the
the bend.
bend. This
This rapid
rapid shift
shift can
can be be explained
explainedby bythe the falling
fallingtransverse
transversewater
waterelevation
elevationslope
slopeandand
release of the
release additional
of the additional momentum
momentum bybythe
thesecondary
secondaryflow flowwhen
when thethe radius of
of curvature
curvatureatatthethebend
bend
exitexit abruptly
abruptly changes
changes to to infinity.
infinity. Finally,the
Finally, theability
abilityofofSRH-3D
SRH-3D model
model to predict
predict secondary
secondaryflows
flowsisis
examined,
examined, since
since primary
primary reasons
reasons totouse
usea a3D3Dmodel
modelare areto topredict
predict vertical
vertical velocity
velocity distribution
distributionand and
secondary flows. There is no secondary flow data with the present case, so a separatesimulation
secondary flows. There is no secondary flow data with the present case, so a separate simulation
using
using the the
NSNS solver
solver U2 U
2RANS [43] is carried out. Comparison of the predicted secondary flows at two
RANS [43] is carried out. Comparison of the predicted secondary flows at two
transects are compared in Figure
transects are compared in Figure 7 between 7 betweenSRH-3DSRH-3Dand andUU22RANS.
RANS. Very
Very similar
similar secondary
secondaryflows
flowsareare
predicted by the two models. This shows the hydrostatic models, such as SRH-3D, are adequateinin
predicted by the two models. This shows the hydrostatic models, such as SRH-3D, are adequate
predicting
predicting thethe secondary
secondary flows,
flows, eveneven inin sharpbends.
sharp bends.

(a) (b)

(c) (d)
Figure
Figure 6. Comparison
6. Comparison of of simulateddepth-averaged
simulated depth-averaged velocity
velocity (normalized
(normalized by by the
the average
averagevelocity)
velocity)
Fluids 2018, 3, x FOR PEER REVIEW
across
across the the channel
channel width
width at four
at four transects
transects forforthe
thebend
bendcase.
case.(a) 35°;
(a)35 ◦ ; (b) 100°;
(b)100 ◦ ; (c) 143°;
(c)143 ◦ ; (d)
(d) 186 ◦ .13 of 19
186°.

(a) (b)

(c) (d)
Figure
Figure 7. Comparison
7. Comparison of of predictedsecondary
predicted secondary flows:
flows:SRH-3D
SRH-3D onon
thethe
leftleft
andand
U2RANS
U2RANSon the
onright.
the right.
(a) SRH-3D at transect 90
◦ °; (b) U22RANS at transect 90°; (c) SRH-3D at transect 0.1 m after exit; (d)
◦
(a) SRH-3D at transect 90 ; (b) U RANS at transect 90 ; (c) SRH-3D at transect 0.1 m after exit;
U2RANS at transect 0.1 m after exit.
(d) U2 RANS at transect 0.1 m after exit.

5.3. Suspended Sediment in a Channel

3D modeling of suspended sediment transport is not trivial; correct implementation of even the
bed boundary condition is not well understood [37]. The flume results in [49] are used to test the
sediment module. The experiment was carried out in a flume with rough bottom created by placing
a single layer of pebble-sized stones (median diameter: 7 mm). The bed slope was 0.07252%. The
water elevation from the roughened bottom was 62 mm after the flow was fully developed. A log-fit
Fluids 2019, 4, 18 14 of 19

5.3. Suspended Sediment in a Channel

3D modeling of suspended sediment transport is not trivial; correct implementation of even the
bed boundary condition is not well understood [37]. The flume results in [49] are used to test the
sediment module. The experiment was carried out in a flume with rough bottom created by placing a
single layer of pebble-sized stones (median diameter: 7 mm). The bed slope was 0.07252%. The water
elevation from the roughened bottom was 62 mm after the flow was fully developed. A log-fit of the
measured velocity profile led to the shear velocity of uτ = 0.021 m. The mean velocity was 0.22 m/s.
The mesh has 62 cells along the 12.4 m flow direction and 31 vertical cells distributed non-uniformly.
The cell center of the first cell near the bed is δ1 = 2.4415 mm, so that the point is within the log
law (y+ = δ1νuτ = 53.4). Upstream discharge is 0.01364 m3 /s, downstream depth is 62 mm, and bed
roughness height is k S = 13 mm.
The computed velocity and vertical eddy viscosity distribution are compared with the
experimental data in Figure 8. Good agreement is obtained. Vertical eddy viscosity is important
in predicting the suspended sediment transport. For the present case, two mechanisms are important
for the sediment motion: settling velocity and turbulent resuspension. The well-cited experimental
data sets of [50,51] are used for comparison of the turbulent viscosity.
Two model runs are made to compute the suspended sediment corresponding to two sediment
sizes. The same runs were reported by [37]. Run 1 has a sediment size of 0.0236 mm, sediment fall
velocity of 0.0516 mm/s, and Rouse number of 0.06. Run 2 has the sediment diameter of 0.079 mm, fall
velocity of 5.79 mm/s, and the Rouse number of 0.672. The sediment sizes are chosen, such that the
Rouse parameters are significantly different. The simulation starts from clear water and stops once
the equilibrium sediment concentration profiles are obtained. Predicted concentration is compared
with the theoretical Rouse profile in Figure 9. It is seen that the Rouse profile is well predicted by the
model. Note that the Rouse profile is only approximate and was derived by assuming the parabolic
Fluids 2018, 3, xof
distribution FOR
thePEER REVIEW
vertical turbulence viscosity. 15 14 of 19

+
U
1.0
15
10
+ 0.8
U
z/H
10 5 0.6 Model Prediction
Data, Huhrman et al. (2010)
Model Prediction
0.4 U+=1/0.41 ln(30z/ks)
5 0 -1 Nezu and Rodi (1986)
Model Prediction 10 10
0
z/k 10
1
Ueda et al. (1977)s
Fuhrman et al. (2010) 0.2
U+=1/0.41 ln(30z/ks)
0 -1 0.0 -3
10 100 z/ks 101 10 10-2 t /(Hu) 10-1

(a) (b)
Figure8.8. Comparison
Figure Comparison ofof predicted
predicted results
resultswith
withthe
the measured
measureddata
datafor
forthe
the Fuhrman
Fuhrman case.
case. (a)
(a) velocity;
velocity;
1.0 6
(b)turbulent
(b) turbulentviscosity.
viscosity.
5
0.8
z/H
6
0.6
5

(a) (b)
 10 10
(a) z/ks 10 10 10
(b) t /(Hu) 10

Figure 8. Comparison of (a)predicted results with the measured data for the Fuhrman
(b) case. (a) velocity;
(b) turbulent
Figure viscosity.of predicted results with the measured data for the Fuhrman case. (a) velocity;
8. Comparison
1.0 6
(b) turbulent viscosity.
Fluids 2019, 4, 18 5
15 of 19
0.8
z/H
6
0.6
5

(a) (b)
(a) (b)
Figure 9. Comparison of suspended sediment concentration vertical distribution for the Fuhrman case
(Black 9.
Figure
Figure line:
9. the model;of
Comparison
Comparison ofRed line: Rouse
suspended
suspended equation).
sediment
sediment (a) Run 1;vertical
concentration
concentration (b) Rundistribution
vertical 2.
distribution for
for the
the Fuhrman
Fuhrman case
case
(Black line: the model; Red line: Rouse equation). (a) Run 1; (b) Run
(Black line: the model; Red line: Rouse equation). (a) Run 1; (b) Run 2. 2.
5.4. Lock Exchange: Intrusive Turbidity Current into a Two-layer Fluid
5.4. Lock Exchange: Intrusive Turbidity Current into a Two-layer Fluid
5.4. Lock Exchange:
Intrusion of aIntrusive
turbidityTurbidity
current Current into a Two-layer
into a two-layer fluid isFluid
simulated next in an attempt to verify
Intrusion
the ability of
of the a turbidity current into a two-layer fluid is simulated next in an
theattempt to verify the ability
Intrusion of amodel to simulate
turbidity currenttheintosediment
a two-layertransport
fluid is and capture
simulated next fluid-sediment
in an attempt interface.
to verify
of theexperimental
The model to simulate casesthe
of sediment transport
[52] are chosen and and capture
the setup the fluid-sediment
is illustrated inthe interface.
Figure The case
10. The experimental
was also
the ability of the model to simulate the sediment transport and capture fluid-sediment interface.
cases of [52] are chosen and the setup is flume
illustrated inglass
Figure 10.measuring
The case was also studied by [53] with
The experimental cases of [52] are chosen and the setup is illustrated in Figure 10. The case waswide,
studied by [53] with an NS solver. The had a tank 197.1 cm long, 19.9 cm also
an
andNS48.5
solver. TheThe flume had a glass tank measuring 197.1fixed
cm long, 19.9 cmandwide, and water
48.5 cm tall.
studied bycm[53]tall.
with anlock-length
NS solver. The(l) behind the gate
flume had was
a glass at 18.6
tank measuring cm
197.1 the
cm total
long, depth
19.9 cm wide,
The lock-length
(H) was (l) behind the gate was fixed at 18.6 cm and the total water depth (H) was 20 cm.
and 48.5 20
cmcm.tall. The lock-length (l) behind the gate was fixed at 18.6 cm and the total water depth
(H) was 20 cm.

Figure 10.Sketch
Figure10. Sketchand
andparameters
parametersfor
forthe
theintrusive
intrusivegravity
gravitycurrent
currentcases.
cases.
Figure 10. Sketch and parameters for the intrusive gravity current cases.
Two cases are simulated. The first is the symmetric flow turbidity current (SFTC), in which the
depth of the two layers in the ambient fluids is equal (h0 = h1 = 10 cm). The densities of the two
layers are: ρ0 = 1000 kg/m3 and ρ1 = 1020 kg/m3 . SFTC has the symmetrical flow type according
to [52], as the density of the lock fluid is equal to the depth-weighted average of the upper and lower
layers. The second case is an asymmetric flow turbidity current (AFTC), in which the depth of the two
layers in the ambient fluids are set as h0 = 17.5 cm and h1 = 2.5 cm, respectively. The densities of the
two layers are the same as in the first case.
In the numerical modeling, no variations are expected in the lateral direction (y), so only 3 lateral
cells are used with the symmetry boundary condition specified in the lateral back and front boundaries.
In the two other directions, 533 cells are in the longitudinal direction (x) and 50 cells are in the vertical
direction (z). Initially, fluid is stationary. For the SFTC case, the initial concentration is C = 0.006061 in
the lock, C = 0.0 and 0.012122 in the top and bottom of the ambient, respectively. For the AFTC case,
C = 0.009091 in the lock, and C = 0.0 and 0.012122 in the top and bottom of the ambient, respectively.
The SFTC case results are shown in Figure 11 to visualize the temporal evolution of the intrusive
gravity current. They are compared with the images taken from the laboratory experiments of [52].
After the lock gate was removed, the fluid contained behind the lock gate collapsed symmetrically and
 respectively.
The SFTC case results are shown in Figure 11 to visualize the temporal evolution of the intrusive
gravity current. They are compared with the images taken from the laboratory experiments of [52].
After the lock gate was removed, the fluid contained behind the lock gate collapsed symmetrically
and propagated along the interface. The head already started to form and was visible at 2 s. The initial
Fluids collapse
2019, 4, 18began with rapid acceleration. As it propagated to the right end of the wall, the gravity16 of 19
current brought strong mixing, resulting in mass entrainment and dilution. These processes are
predicted by the numerical model well. The results, however, confirm the finding of [54] that the
propagated along
hydrostatic the interface.
models, such as theThe head
current already
one, started
failed to predicttothe
form and was
formation visible
of the at 2 s. The initial
Kelvin-Helmholtz
billows.
collapse beganThe present
with rapidmodel otherwiseAs
acceleration. is capable of predicting
it propagated to the the overall
right end ofturbidity current
the wall, features,
the gravity current
as well as the propagation speed of the current.
brought strong mixing, resulting in mass entrainment and dilution. These processes are predicted
Simulatedmodel
by the numerical resultswell.
of theThe
AFTC case however,
results, are shownconfirm
in Figure the12;finding
the temporal
of [54] evolution of the
that the hydrostatic
intrusive gravity current is compared between the model and the experiment of [52]. Similar to the
models, such as the current one, failed to predict the formation of the Kelvin-Helmholtz billows.
findings of the SFTC case, the hydrostatic model predicts the initial turbidity current formation and
The present model otherwise is capable of predicting the overall turbidity current features, as well as
travel well. The model, however, is incapable of predicting the Kelvin-Helmholtz waves as well as
the propagation speed process
the wave reflection of the current.
after the front reaches the end wall.

Experiment

SRH-3D

0.0040 0.0050 0.0071 0.0081 0.0040 0.0050 0.0071 0.0081

(a) (b)

Experiment

SRH-3D
0.0040 0.0050 0.0071 0.0081 0.0040 0.0050 0.0071 0.0081

(c) (d)
Figure
Figure 11. Temporal
11. Temporal evolution
evolution of theofSFTC
the SFTC intrusive
intrusive gravity
gravity current—sediment
current—sediment concentration
concentration contours.
contours.isExperiment
Experiment is from
from [52] and [52] and visualized
visualized by addingbydye.
adding = 2Time
dye. (a)
(a) Time s; (b)= Time
2 s; (b)= Time = 14
14 s; (c) s; (c)= 26 s;
Time
Time= =3826s.s; (d) Time = 38 s.
(d) Time

Simulated results of the AFTC case are shown in Figure 12; the temporal evolution of the intrusive
Experiment
gravity current is compared between the model and the experiment of [52]. Similar to the findings
of the SFTC case, the hydrostatic model predicts the initial turbidity current formation and travel
well. Fluids 2018, 3, x FOR
The model, PEER REVIEW
however, 16 the
is incapable of predicting the Kelvin-Helmholtz waves as well as of 20wave

reflection process after the front reaches the end wall.

Experiment

SRH-3D

0.0045 0.0065 0.0086 0.0096 0.0106 0.0045 0.0065 0.0086 0.0096 0.0106

(a) (b)

Experiment

SRH-3D

0.0045 0.0065 0.0086 0.0096 0.0106 0.0045 0.0065 0.0086 0.0096 0.0106

(c) (d)

FigureFigure 12. Temporal

12. Temporal evolution
evolution of theof the AFTC
AFTC intrusive
intrusive gravitygravity current—sediment
current—sediment concentration
concentration contours.
contours.
Experiment is Experiment
from [52] andis from [52] andby
visualized visualized
adding by adding
dye. = 2Time
dye. (a)
(a) Time s; (b)=Time
2 s; (b) Time
= 14 = 14
s; (c) s; (c)
Time = 26 s;
Time = 26
(d) Time = 38 s. s; (d) Time = 38 s.

6. Conclusions
A new 3D hydrostatic numerical model is developed that may be used to simulate flow and
suspended sediment transport in rivers and reservoirs with free surface and fluid-sediment interface.
The new model is formulated in a general way so that it is applicable to a wide range of
environmental fluid flows. A new solution algorithm is developed that derives the water elevation
Fluids 2019, 4, 18 17 of 19

6. Conclusions
A new 3D hydrostatic numerical model is developed that may be used to simulate flow and
suspended sediment transport in rivers and reservoirs with free surface and fluid-sediment interface.
The new model is formulated in a general way so that it is applicable to a wide range of environmental
fluid flows. A new solution algorithm is developed that derives the water elevation equation from the
vertical summation of the discretized continuity equation. The resultant elevation correction equation
is consistent with the rest of the discretized equations, leading to a robust and stable iterative solver.
A major contribution of the present study is the adoption of a new mesh—the horizontal polygonal
mesh coupled with the arbitrarily distributed vertical mesh. The 3D mesh is terrain-following and
movable vertically to changes in water surface elevation and bed elevation.
The flow solver is verified with two 3D cases having experimental data. The model performs
well in stability and accuracy. The sharply curved bend flow modeling shows that the 3D hydrostatic
model works well for flows in meander bends, as the predicted secondary flows match that from
the non-hydrostatic NS solvers. In comparison with the 2D model, the 3D model improves upon the
flow through the sharp bend. A primary advantage of the 3D model over 2D ones is that the vertical
distribution and secondary flows are predicted.
The suspended sediment transport module is fully coupled to the flow solver, in that the flow
dictates the sediment movement and the changes of sediment concentration alter the flow velocity.
The coupling is achieved implicitly within the same time step. The sediment module is first tested for
its ability to predict the vertical distribution of the sediment concentration, and then used to predict
the turbidity current formation and travel. Despite the inability of the model to predict the waves
associated with the Kelvin-Helmholtz instability, the model is adequate in predicting the sediment
collapse, current formation and movement speed.
Future works include: (a) demonstration and application of SRH-3D to practical rivers;
(b) development of the adaptive mesh movement capability to track density interface; and (c) extension
to bedload sediment transport.

Author Contributions: Conceptualization, Y.L.; methodology, Y.L.; software, Y.L.; validation, Y.L. and K.W.;
investigation, Y.L. and K.W.; resources, Y.L. and K.W.; writing—original draft preparation, Y.L.; writing—review
and editing, K.W.
Funding: This research was funded by U.S. Bureau of Reclamation Science and Technology Program and Water
Resources Agency of Taiwan.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Gibson, S.; Brunner, G.; Piper, S.; Jensen, M. Sediment Transport Computations in HEC-RAS. In Proceedings
of the Eighth Federal Interagency Sedimentation Conference (8thFISC), Reno, NV, USA, 2–6 April 2006;
pp. 57–64.
2. MIKE 11. Users Manual; Danish Hydraulic Institute: The Hague, The Netherlands, 2005.
3. Wu, W.; Vieira, D.A. One-Dimensional Channel Network Model CCHE1D Version 3.0.—Technical Manual;
Technical Report No. NCCHE-TR-2002-1; National Center for Computational Hydroscience and Engineering,
The University of Mississippi: Oxford, MS, USA, 2002.
4. Huang, J.; Greimann, B.P. GSTAR-1D, General Sediment Transport for Alluvial Rivers—One Dimension;
Technical Service Center, U.S. Bureau of Reclamation: Denver, CO, USA, 2007.
5. Wu, W. Depth-averaged two-dimensional numerical modeling of unsteady flow and non-uniform sediment
transport in open channels. J. Hydraul. Eng. 2004, 130, 1013–1024. [CrossRef]
6. Hung, M.C.; Hsieh, T.Y.; Wu, C.H.; Yang, J.C. Two-dimensional nonequilibrium noncohesive and cohesive
sediment transport model. J. Hydraul. Eng. 2009, 135, 369–382. [CrossRef]
7. Huang, W.; Cao, Z.; Pender, G.; Liu, Q.; Carling, P. Coupled flood and sediment transport modelling with
adaptive mesh refinement. Sci. China Technol. Sci. 2015, 58, 1425–1438. [CrossRef]
Fluids 2019, 4, 18 18 of 19

8. Jia, Y.; Wang, S.S.Y. CCHE2D: Two-Dimensional Hydrodynamic and Sediment Transport Model for Unsteady
Open Channel Flows Over Loose Bed; NCCHE Technical Report, No. NCCHETR-2001-01; The University of
Mississippi: Oxford, MS, USA, 2001.
9. Hervouet, J.-M. Hydrodynamics of Free Surface Flows; John Wiley&Sons, Ltd.: Hoboken, NJ, USA, 2007;
pp. 1–360.
10. Casulli, V.; Walters, R.A. An unstructured grid, three dimensional model based on the shallow water
equations. Int. J. Numer. Meth. Fluids 2000, 32, 331–348. [CrossRef]
11. Bever, A.J.; MacWilliams, M.L. Simulating sediment transport processes in San Pablo Bay using coupled
hydrodynamic, wave, and sediment transport models. Mar. Geol. 2013, 345, 235–253. [CrossRef]
12. Deltares. Delft3D-FLOW. Simulation of multi-dimensional hydrodynamic flow and transport phenomena, including
sediments—User Manual; Version 3.04, rev. 11114; Deltares: Delft, The Netherlands, 2010.
13. Lai, Y.G. SRH-2D Version 2: Theory and User’s Manual; Technical Service Center, U.S. Bureau of Reclamation:
Denver, CO, USA, 2008.
14. Lai, Y.G. Two-Dimensional Depth-Averaged Flow Modeling with an Unstructured Hybrid Mesh.
J. Hydraul. Eng. 2010, 136, 12–23. [CrossRef]
15. Paik, J.; Eghbalzadeh, A.; Sotiropoulos, F. Three-Dimensional Unsteady RANS Modeling of Discontinuous
Gravity Currents in Rectangular Domains. J. Hydraul. Eng. 2009, 135, 505–521. [CrossRef]
16. Papanicolaou, A.N.T.; Elhakeem, M.; Krallis, G.; Prakash, S.; Edinger, J. Sediment Transport Modeling
Review—Current and Future Developments. J. Hydraul. Eng. 2008, 134, 1–14. [CrossRef]
17. Blumberg, A.F.; Mellor, G.L. Diagnostic and prognostic numerical circulation studies of the south Atlantic
bight. J. Geophys. Res. 1983, 88, 4579–4592. [CrossRef]
18. Blumberg, A.F.; Mellor, G.L. A description of a three-dimensional coastal ocean circulation
model. In Three-Dimensional Coastal Ocean Models, Coastal and Estuarine Science; Heaps, N.S., Ed.;
American Geophysical Union: Washington, DC, USA, 1987; Volume 4, pp. 1–19.
19. HydroQual, Inc. A primer for ECOMSED: Users Manual; HydroQual, Inc.: Mahwah, NJ, USA, 2002.
20. Hamrick, J.M. A Three-Dimensional Environmental Fluid Dynamics Computer Code: Theoretical and Computational
Aspects; Special Report 317; The College of William and Mary, Virginia Institute of Marine Science:
Gloucester Point, VA, USA, 1992; 63p.
21. DHI. Delft3D-FLOW: Simulation of Multi-Dimensional Hydrodynamic Flows and Transport Phenomena, Including
Sediments; Deltares: Delft, The Netherlands, 2011.
22. Roelvink, J.A.; van Banning, G.K.F.M. Design and development of DELFT3D and application to coastal
morphodynamics. In Hydroinformatics; Verwey, A., Minns, A.W., Babovic, V., Maksimovic, C., Eds.; Balkema:
Rotterdam, The Netherlands, 1994; pp. 451–456.
23. Lesser, G.R. Computation of Three-Dimensional Suspended Sediment Transport within the DELFT3D-FLOW Module;
WLjDelft Hydraulics Report Z2396; Delft Hydraulics: Delft, The Netherlands, 2000.
24. Johnson, B.H.; Kim, K.W.; Heath, R.E.; Hsieh, B.B.; Butler, H.L. Validation of three dimensional hydrodynamic
model of Chesapeake Bay. J. Hydraul. Eng. 1993, 119, 2–20. [CrossRef]
25. Spasojevic, M.; Holly, F.M. Three-Dimensional Numerical Simulation of Mobile-Bed Hydrodynamics; Contract Rep.
HL-94-2; U.S. Army Engineer Waterways Experiment Station: Vicksburg, MS, USA, 1994.
26. Gessler, D.; Hall, B.; Spasojevic, M.; Holly, F.M.; Pourtaheri, H.; Raphelt, N.X. Application of 3D mobile bed,
hydrodynamics model. J. Hydraul. Eng. 1999, 125, 737–749. [CrossRef]
27. Haidvogel, D.B.; Arango, H.; Budgell, W.P.; Cornuelle, B.D.; Curchitser, E.; Di Lorenzo, E.;
Fennel, K.; Geyer, W.R.; Hermann, A.J.; Lanerolle, L. Ocean forecasting in terrain-following coordinates:
Formulation and skill assessment of the Regional Ocean Modeling System. J. Comput. Phys. 2008, 227,
3595–3624. [CrossRef]
28. Moriarty, J.M.; Harris, C.K.; Hadfield, M.G. A Hydrodynamic and Sediment Transport Model for the Waipaoa
Shelf, New Zealand: Sensitivity of Fluxes to Spatially-Varying Erodibility and Model Nesting. J. Mar. Sci. Eng.
2014, 2, 336–369. [CrossRef]
29. Leendertse, J.J. Turbulence modelling of surface water flow and transport: Part IVa. J. Hydr. Eng. 1990, 114,
603–606. [CrossRef]
30. Stelling, G.S.; van Kester, J.A. On the approximation of horizontal gradients in sigma co-ordinates for
bathymetry with steep bottom slopes. Int. J. Numer. Meth. Fluids 1994, 18, 915–955. [CrossRef]
Fluids 2019, 4, 18 19 of 19

31. Bijvelds, M.D.J.P. Numerical Modelling of Estuarine Flow over Steep Topography; Delft University of Technology:
Delft, The Netherlands, 2001; 136p.
32. Cornelissen, S.C. Numerical Modelling of Stratified Flows Comparison of the Sigma and z Coordinate
Systems. Master’s Thesis, Delft University of Technology, Delft, The Netherlands, 2004.
33. Mandang, I.; Yanagi, T. Cohesive sediment transport in the 3D hydrodynamic baroclinic circulation model in
the Mahakam Estuary, East Kalimantan, Indonesia. Coast. Mar. Sci. 2009, 32, 1–13.
34. Rodi, W. Turbulence models and their application in Hydraulics, State-of-the-art paper article sur l’etat
de connaissance. In Proceedings of the IAHR Sectionon Fundamentals of Division II: Experimental and
Mathematical Fluid Dynamics, Delft, The Netherlands, 3–6 September 1984.
35. Celik, I.; Rodi, W. Modeling suspended sediment transport in nonequilibrium situations. J. Hydraul. Eng.
1988, 114, 1157–1191. [CrossRef]
36. Van Rijn, L.C. Principles of Sediment Transport in Rivers, Estuaries and Coastal Seas; Aqua Publications:
Amsterdam, The Netherlands, 1993.
37. Liu, X. New Near-Wall Treatment for Suspended Sediment Transport Simulations with High Reynolds
Number Turbulence Models. J. Hydraul. Eng. 2014, 140, 333–339. [CrossRef]
38. Fredsøe, J.; Andersen, K.H.; Sumer, B.M. Wave plus current over ripple-covered bed. Coast. Eng. 1999, 38,
117–221. [CrossRef]
39. Garcia, M.H.; Parker, G. Entrainment of Bed Sediment into Suspension. J. Hydraul. Eng. 1991, 117, 414–435.
[CrossRef]
40. Garcia, M.H. (Ed.) ASCE Manuals and Reports on Engineering Practice No. 110, Sedimentation Engineering:
Processes, Measurements, Modeling, and Practice; ASCE: Reston, VA, USA, 2008.
41. Zyserman, J.; Fredsøe, J. Data Analysis of Bed Concentration of Suspended Sediment. J. Hydraul. Eng. 1994,
120, 1021–1042. [CrossRef]
42. Lai, Y.G.; Przekwas, A.J. A Finite-Volume Method for Simulations of Fluid Flows with Moving Boundaries.
Int. J. Comp. Fluid Dyn. 1994, 2, 19–40. [CrossRef]
43. Lai, Y.G.; Weber, L.J.; Patel, V.C. Non-hydrostatic three dimensional method for hydraulic flow simulation. I:
Formulation and verification. J. Hydraul. Eng. 2003, 129, 196–205. [CrossRef]
44. Shettar, A.S.; Murthy, K.K. A numerical study of division of flow in open channels. J. Hydraul. Res. 1996, 34,
651–675. [CrossRef]
45. Rozovskii, I.L. Flow of Water in Bends of Open Channels; The Israel Program for Scientific Translations:
Jerusalem, Israel, 1961.
46. Molls, T.; Chaudhry, M.H. Depth-averaged open-channel flow model. J. Hydraul. Eng. 1995, 121, 453–465.
[CrossRef]
47. Lien, N.C.; Hsieh, T.Y.; Yang, J.C.; Yeh, K.C. Bed-flow simulation using 2D depth-averaged model.
J. Hydraul. Eng. 1999, 125, 1097–1108. [CrossRef]
48. Leschziner, M.A.; Rodi, W. Calculation of strongly curved open channel flow. J. Hydraul. Div. 1979, 105, 1297–1314.
49. Fuhrman, D.R.; Dixen, M.; Jacobsen, N.G. Physically-consistent wall boundary conditions for the k-omega
turbulence model. J. Hydraul. Res. 2010, 48, 793–800. [CrossRef]
50. Ueda, H.; Moller, R.; Komori, S.; Mizushina, T. Eddy diffusivity near the free surface of open channel flow.
Int. J. Heat Mass Transf. 1977, 20, 1127–1136. [CrossRef]
51. Nezu, I.; Rodi, W. Open-channel flow measurements with a laser doppler anemometer. J. Hydraul. Eng. 1986,
112, 335–355. [CrossRef]
52. Sutherland, B.R.; Kyba, P.J.; Flynn, M.R. Intrusive Gravity Currents in Two-layer Fluids. J. Fluid Mech. 2004,
514, 327–353. [CrossRef]
53. An, S.D. Interflow Dynamics and Three-Dimensional Modeling of Turbid Density Currents in Imha Reservoir,
South Korea. Ph.D. Thesis, Department of Civil and Environmental Engineering, Colorado State University,
Fort Collins, CO, USA, 2011.
54. Fringer, O.B.; Gerritsen, M.G.; Street, R.L. An Unstructured-grid, Finite-Volume, Nonhydrostatic, Parallel
Coastal Ocean Simulator. Ocean Model. 2006, 14, 139–173. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).