Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

ARTICLE IN PRESS

Computers & Geosciences 31 (2005) 929–946


www.elsevier.com/locate/cageo

MOD_FreeSurf2D: A MATLAB surface fluid flow


model for rivers and streams$
Nick Martin, Steven M. Gorelick
Department of Geological and Environmental Sciences, Stanford University, Braun Hall, Bldg. 320, 450 Serra Mall,
Stanford, CA 94305-2115, USA
Received 6 June 2004; received in revised form 7 March 2005; accepted 7 March 2005

Abstract

MOD_FreeSurf2D is an open source MATLAB code that simulates fluid velocities and depths in rivers and streams.
Although this model was designed for a specific purpose, MOD_FreeSurf2D can be employed in general scenarios when
the depth-averaged, shallow water equations apply. The model approximates the depth-averaged, shallow water
equations with a finite volume, semi-implicit, semi-Lagrangian representation. This numerical solution method provides
accuracy and stability when using model time steps that exceed the Courant–Friedrichs–Lewy (CFL) restriction. An
additional benefit of the numerical representation is the ability to simulate moving land/water boundaries. Model
results were shown to be accurate when compared to published data from a dam-break experiment in a 21 m flume and
from velocity and depth measurements along a 400 m river reach in Idaho. Results from the dam-break experiment
simulations demonstrate the model’s ability to simulate wetting and drying. Additionally, sensitivity analyses conducted
on the two scenarios show model convergence and demonstrate that the model can employ time steps that exceed the
CFL restriction.
r 2005 Elsevier Ltd. All rights reserved.

Keywords: Numerical model; Semi-Lagrangian; Semi-implicit; Depth-averaged; Shallow water

1. Introduction Currently, his fluid flow module is the only existing


component of the suite of modules (but other modules
This paper describes the mathematical foundation, are in development).
numerical implementation, and testing of an open Although MOD_FreeSurf2D was designed to provide
source, MATLAB code that solves the depth-averaged, a flow velocity field to a series of coupled sediment
shallow-water equations. MOD_FreeSurf2D was devel- transport and deposition modules, this model provides a
oped to be one of a series of open source modules used stand-alone representation of fluid flow in rivers and
to simulate hydraulic and sedimentary processes in order streams. Our purpose in presenting this paper and
to model aquifer formation in fluvial environments. providing this MATLAB code is to provide an
accessible tool that other researchers can use and modify
$
Code on server at http://www.iamg.org/CGEditor/
for their particular applications. An important design
index.htm. criterion was to make MOD_FreeSurf2D as generally
Tel.: +1 5404212994; fax: +1 5404428863. applicable as possible. With this goal in mind, the model
E-mail address: nick.martin@stanfordalumni.org permits the specification of a variety of parameters that
(N. Martin). allow the model user to include or remove various

0098-3004/$ - see front matter r 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cageo.2005.03.004
ARTICLE IN PRESS
930 N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946

components of the governing equations according to the gravitational constant, t is time,  is the horizontal eddy
user’s requirements. These parameters and governing viscosity coefficient, H is the total water depth, and f is
equations are covered in more detail in the following the Coriolis parameter. Eq. (4) and Fig. 1 give the
sections and in the manual that is included with the relationship among total water depth; H, undisturbed
MATLAB code. water depth; h, free surface elevation, Z:
Additional important criteria in the design of
H ¼ Z þ h. (4)
MOD_FreeSurf2D were an open source, stable code
with the ability to simulate moving land/water bound- Assumptions of a well-mixed water column and of a
aries. With these considerations in mind, we examined small water depth-to-width ratio are required for
many different 2D model configurations (see Table 1) simplification to these equations from the 3D primitive
before selecting a semi-implicit, semi-Lagrangian, variable equations (Casulli and Cheng, 1992).
finite volume formulation. MOD_FreeSurf2D builds The vertical integration of the water column, that
on the semi-implicit, semi-Lagrangian finite-volume provides the depth-averaged representation, requires top
approximation to the depth-averaged shallow water and bottom boundary representations to replace the
equations of Casulli and Cheng (1992) and related vertical change in velocity in the viscous terms. A
work (Casulli, 1990; Casulli, 1997; Casulli, 1999; Casulli prescribed wind-stress coefficient, gT, and prescribed
and Cattani, 1994; Casulli and Cheng, 1992; Walters wind velocities, Va and Ua, in Eq. (5) provide the top
and Casulli, 1998). In addition, we include our own friction boundary:
semi-Lagrangian method in the model (Martin and
u qu
qz ¼ gT ðU a  UÞ u qv
qz ¼ gT ðV a  V Þ: (5)
Gorelick, 2005). The numerical representation employed
in the model has the advantages of stability and of A Manning–Chezy formula given in Eq. (6) provides a
the ability to represent moving land/water boundaries. bottom friction relationship (Casulli, 1999; Casulli and
Although many of the numerical methods combined Cheng, 1992):
in the model have been used in other models, pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi
U 2 þV 2 U 2 þV 2
MOD_FreeSurf2D is the only open source, depth- u qu
qz ¼ g Cz 2 U; u qv
qz ¼ g Cz2
v, (6)
averaged code that we know of that employs this
collection of methods. where Cz is the Chezy coefficient, n is the kinematic
velocity coefficient, and z represents the vertical direc-
tion.
Eqs. (1)–(3) provide three partial differential equa-
2. Governing equations
tions in three unknowns, Uðx; y; tÞ, V ðx; y; tÞ, and
Zðx; y; tÞ. The governing equations represent fluid flow
The governing equations for MOD_FreeSurf2D are
in streams, rivers, and shallow estuaries when the
the depth-averaged, shallow-water equations on a
hydrostatic pressure assumption, the well-mixed water
regular mesh:
column assumption, and the small depth-to-width ratio
qU qU qU assumption are valid. In spite of these limiting assump-
þU þV
qt qx qy tions, the 2D shallow water equations have been
 2  successfully employed to represent fluid flow in vertically
qZ q U q2 U g ðU a  UÞ
¼ g þ  þ þ T well-mixed rivers, streams, and shallow estuaries
qx qx2 qy2 H
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (Falconer and Chen, 1991; Garcia and Kahawita,
2 2
U þV 1986; Heniche et al., 2000; Leclerc et al., 1990; Zhao
g U þ fV , ð1Þ
Cz2 et al., 1994).

qV qV qV
þU þV
qt qx qy 3. Numerical approximations
 2 
qZ q V q2 V
¼ g þ  þ 2
qy qx2 qy Numerical approximations to the governing equations
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (Eqs. (1)–(3)) were selected to provide an accurate,
g ðV a  V Þ U2 þ V2
þ T g V  fU, ð2Þ stable representation of fluid flow and to simulate
H Cz2 moving land/water boundaries. In MOD_FreeSurf2D
semi-implicit and semi-Lagrangian numerical represen-
qZ qðHUÞ qðHV Þ
þ þ ¼ 0, (3) tations are combined with the finite volume method to
qt qx qy
solve the shallow water equations on a rectangular grid.
where U is the depth-averaged x-direction velocity Semi-implicit treatment requires that the gravitational
component, V is the depth averaged y-direction velocity terms in the momentum equations and the velocity
component, Z is the free surface elevation, g is the divergence in the continuity equation be treated
Table 1
Comparison of a selection of 2D models for rivers, streams, and estuaries presented in literature

Author Finite Volumeb Free Surfacec Lagrangian representation of advectiond Implicite Wetting and drying f

Martin and Gorelick (2005)a K K K K K


Akanbi and Katopodes (1988) K K
Alcrudo and Garcia-Navarro (1993) K
Anastasiou and Chan (1997) K K
Bates and Anderson (1993) K K
Bellos et al. (1991) K K

N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946


Benque et al. (1982) K K K K
Bermudez et al. (1998) K K K
Casulli and Cheng (1992) K K K K K
Falconer and Chen (1991) K K K
Fennema and Chaudhry (1990)
Fennema and Chaudhry (1989) K

ARTICLE IN PRESS
Fraccarollo and Toro (1995); Toro (1992) K
Fujihara and Borthwick (2000) K
Galland et al. (1991); Hervouet and Haren (1996) K K K K
Garcia and Kahawita (1986)
Guillou and Nguyen (1999) K K K
Heniche et al. (2000) K K
Hsu et al. (2000) K K K
Katopodes and Strelkoff (1978) K K
Kawahara and Umetsu (1986) K K
King (1977); King and Norton (1978) K K
Layton and Panne (2002) K K
Leclerc et al. (1990) K K
Molls and Chaudhry (1995) K
Petera and Nassehi (1996) K K K K
Tabuenca and Cardona (1992) K K
Tee (1976); Tee (1977) K K
Tseng and Chu (2000) K K
Tucciarelli and Termini (2000) K
Zhao et al. (1994) K K
Zhou and Goodwill (1997); Zhou and Stansby (1999) K
Zoppou and Roberts (2000) K
a
This article.
b
Finite volume designation requires integral representation of the conservation equations.
c
Models with a free surface are those that solve an equation for free surface elevation.
d
Lagrangian representation of advection cateogory includes purely Lagrangian methods, semi-Lagrangian methods, and characteristics-based methods.
e
Implicit model include those models that are semi-implicit and those that allow designation of the degree of implicitness with the y method.
f
Wetting and drying designation requires that a model represent moving land-water boundaries by solving the governing equations for both wet and dry areas or by using moving

931
or deforming calculationd grids.
ARTICLE IN PRESS
932 N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946

CFL restriction:
V
i,j+1/2 Dt
CFL ¼ w (7)
i-1/2,j i,j i+1/2,j Dxi
U traditionally governs the maximum time-step duration
i,j-1/2 in transient fluid flow simulations. In Eq. (7), w is the
velocity component in the xi-direction, Dt is the time
Top View step size, and Dxi is the cell dimension in the xi-direction
of flow. This restriction relates fluid velocity to time step
η size and to computational cell size and stipulates that the
CFL number should be less than 1.0.

3.1. Semi-implicit representation


h H

In the semi-implicit process, a system of equations is


generated where free-surface elevation, Z, is the only
Side View unknown variable at time N þ 1. This system of
equations with one unknown is obtained by substituting
Eqs. (9) and (10):
Fig. 1. Mod_FreeSurf2D variable location (top view) and
variable definition (side view). Top view displays location of Dt  N 
ZNþ1
i;j ¼ ZNi;j  y H iþ1=2;j U Nþ1 N Nþ1
iþ1=2;j  H i1=2;j U i1=2;j
velocity components, total depth, and free surface elevation on Dx
computational grid. Side view shows relationship between free Dt  N 
surface elevation, Z, total water depth, H, and undisturbed y H i;jþ1=2 V Nþ1 N Nþ1
i;jþ1=2  H i;j1=2 V i;j1=2
Dx
depth, h. Dt  N 
 ð1  yÞ H iþ1=2;j U N N N
iþ1=2;j  H i1=2;j U i1=2;j
Dx
Dt  N 
 ð1  yÞ H i;jþ1=2 V N N N
i;jþ1=2  H i;j1=2 V i;j1=2 ,
Dx
implicitly (Robert, 1981, 1982) while the remaining ð8Þ
terms are treated explicitly. In semi-Lagrangian repre-
sentations, advection is simulated by following the g Dt  N 
U Nþ1 N
iþ1=2;j ¼ FU iþ1=2;j  ð1  yÞ Ziþ1;j  ZN i;j
motion of the fluid. Dx
In the model, the mass- and momentum-conservative g Dt  Nþ1 Nþ1

y Ziþ1;j  Zi;j
finite volume method (Fletcher, 1991) was adopted for Dxrffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
stability and accuracy. The finite volume method is   2  2
derived from the integral form of the conservation UN
iþ1=2;j þ VN
iþ1=2;j
equations and is valid at both discontinuities and in  g Dt U Nþ1
iþ1=2;j
Cz2iþ1=2;j H N
iþ1=2;j
smooth portions of the flow field (Zhao et al., 1994).  
Nþ1
Although our implementation is on a regular rectan- gt U a  U iþ1=2;j
gular mesh, the finite volume method and not the finite þ Dt , ð9Þ
HNiþ1=2;j
difference method is employed.
The combination of semi-implicit and semi-Lagran-
g Dt  N 
gian methods provides a stable solution method that V Nþ1 N
i;jþ1=2 ¼ FV i;jþ1=2  ð1  yÞ Zi;jþ1  ZN i;j
permits relatively large model time steps without Dy
significant degradation of model accuracy. The advan- g Dt  Nþ1 Nþ1

y Zi;jþ1;  Zi;j
tage of relatively long model time steps is a smaller Dy
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   
number of time steps, and fewer total calculations, are 2 2
required to complete a simulation. A semi-implicit, semi- UN
i;jþ1=2 þ VN
i;jþ1=2
Lagrangian representation demonstrated a 6-fold in-  g Dt V Nþ1
i;jþ1=2
Cz2i;jþ1=2 H Ni;jþ1=2
crease in the maximum stable time step in applications  
of the atmospheric form of the primitive variable gt V a  V Nþ1
i;jþ1=2
equations (Robert, 1982). In free-surface flow modeling, þ Dt , ð10Þ
HNi;jþ1=2
a semi-implicit, semi-Lagrangian portrayal permitted
the relaxation of the Courant–Friedrichs–Lewy (CFL) where Dx is the distance across a computational volume
restriction on time-step duration (Casulli, 1990; Casulli in the x-direction; Dy is the distance across a volume in
and Cattani, 1994; Casulli and Cheng, 1992). The the y-direction, and Dt is the time step duration. Eq. (8)
ARTICLE IN PRESS
N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946 933

represents the discrete form of the depth-averaged 3.2. Semi-Lagrangian advection operator
continuity, Eq. (3). Eqs. (9) and (10) provide the discrete
forms of the depth-averaged momentum equations, Eqs. The advective operators, FU and FV in Eqs. (9) and
(1) and (2). The free-surface system of equations (10), contain the advective, viscous, and Coriolis
obtained after this substitution is symmetric, positive- components from the governing equations. Eqs. (12)
definite and can be solved with a pre-conditioned and (13) represent the semi-Lagrangian numerical
conjugate gradient solver (Casulli and Cheng, 1992). approximation of these operators. U nsLbicubic represents
After values of ZNþ1 are obtained by conjugate gradient the advective component; the  Dt terms are the
solution of this system of equations, the ZNþ1 values are viscous terms, and f Dt V nsLbilinear provides the Coriolis
used in Eqs. (9) and (10) to explicitly update the velocity component:
components U nþ1 nþ1
iþ1=2;j and V i;jþ1=2 .  n 
U isl þ1;j sl  2U nisl ;j sl þ U nisl 1;j sl
In Eqs. (8)–(10) subscripts i,j represent computational n n
FU iþ1=2;j ¼ U sLbicubic þ  Dt
volume centers, where i  1=2 or j  1=2 denote Dx2
 n 
computational volume faces. The computational U isl ;j sl þ1  2U isl ;j sl þ U nisl ;j sl 1
n

mesh is the Arakawa C-grid, displayed at the top þ  Dt


Dy2
of Fig. 1,which has directional velocity, U and V, and þ f Dt V nsLbilinear , ð12Þ
total water depth, H, defined at the centers of volume
faces, and free-surface elevation, Z, defined at  n 
volume centers. We adopt this rectangular computa- V isl þ1;jsl  2V nisl ;j sl þ V nisl 1;j sl
FV ni;jþ1=2 ¼ V nsLbicubic þ  Dt
tional grid in MOD_FreeSurf2D; however, these meth- Dx2
 n 
ods can be extended for use with unstructured grids V isl ;j sl þ1  2V nisl ;jsl þ V nisl ;j sl 1
(Casulli, 1997). þ  Dt
Dy2
The parameter y, in Eq. (8), determines the ‘‘degree of
þ f Dt V nsLbilinear . ð13Þ
implicitness’’ of the solution and y should range between
0.5 and 1.0 for a semi-implicit method. In MOD_Free- A semi-Lagrangian representation of advection is a
Surf2D, the model user specifies the y-value. When y two-step process (Staniforth and Cote, 1991), and the
equals 0.5, the approximation is centered in time. semi-Lagrangian method employed in MOD_Free-
When yequals 1.0, the approximation is fully implicit Surf2D is presented schematically in Fig. 2. In step
(Casulli, 1999). one, the Lagrangian step, the path of a fluid particle
The term gT, in Eqs. (9) and (10), is the wind-stress
coefficient. This parameter is user specified and setting
this parameter to zero effectively removes the wind-
surface stress terms from model calculations. Va and Ua destination point

in Eqs. (9) and (10) are user prescribed wind velocities. A 13/2
departure point
uniform, prescribed value for each wind velocity applies pathline
to the entire simulation domain. 11/2 advective term interpolation locations
In Eqs. (9) and (10) the value for the Chezy Coriolis term interpolation locations
9/2
coefficient, Cz, is obtained from the relationship with viscous term interpolation locations
Manning’s roughness coefficient, Mn:
7/2
 1=6
5/2
HN
iþ1=2;j
Cziþ1=2;j ¼ . (11)
Mniþ1=2;j 3/2
n
io
ct

1/2
re

Although it is convenient to consider Mn to be


di
j =

dimensionless (and we employ this convenience in the


flo

remainder of this paper), consistent dimensions for Mn i = 1/2 3/2 5/2 7/2 9/2 11/2 13/2 15/2 17/2
in Eq. (11) are TL1=3 ðs m1=3 Þ. The equivalent value of
Cz in US Customary units is obtained by multiplying H Fig. 2. Schematic of semi-Lagrangian advection representation
in Eq. (11) by 1.49 (Street et al., 1996). Because of this in MOD_FreeSurf2D. Semi-Lagrangian solution for FV n8;11=2
(destination point) is shown as a two-step process; similar
dimensional inconsistency and because Mn values are
calculations are preformed for all FV ni;jþ1=2 . In first step,
usually listed in reference sources as an acceptable range
pathline tracing is performed to locate departure point. In this
of values, we employ Mn as an estimated parameter. situation, departure point is located in computational volume
In MOD_FreeSurf2D, the model user supplies an Mn (4,2). Second step, interpolation, is performed at departure
value for each computational volume in the simulation point using different interpolation stencils for three different
domain. terms (advective, viscous, and Coriolis).
ARTICLE IN PRESS
934 N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946

located at each U iþ1=2;j and V ijþ1=2 location on the across this face is zero when total depth at a volume face
Eulerian computational grid (e.g. the destination point is zero. Because the free-surface representation of total
in Fig. 2) is traced back through Dt of travel time to the water depth (see Fig. 1 and Eqs. (15) and (16)) allows
particle’s departure point (isL,jsL). Pathline tracing, is volume faces to naturally wet or dry in a conservative
accomplished in MOD_FreeSurf2D with choice of manner as flow conditions warrant, the correct closed
either a semi-analytical particle path-line tracing method boundary conditions are enforced at every dry face. This
(Martin and Gorelick, 2005) or a classical four-step, treatment enables the model to determine automatically
explicit Runge–Kutta method for path line tracing the location of water/land boundaries:
(Zheng and Bennett, 2002).  
In step two, the Eulerian step, the velocity compo- H Nþ1
iþ1=2;j ¼ max 0; hiþ1=2;j þ ZNþ1
i;j ; hiþ1=2;j þ ZNþ1
iþ1;j , (15)
nents in the advective operator are interpolated from the
 
known U niþ1=2;j and V ni;jþ1=2 velocity components on the H Nþ1 max 0; h ZNþ1
; h ZNþ1
i;jþ1=2 ¼ i;jþ1=2 þ i;j i;jþ1=2 þ i;jþ1 . (16)
Eulerian grid around (isL,jsL). Because pathline tracing
follows the representative fluid particles back Dt in time,
the velocity values on the Eulerian grid are known from 3.4. Boundary conditions
the calculations for the previous time step (i.e. time step
n). Different interpolation stencils are employed for the Two general types of domain boundaries, open and
three components of advective operators as shown in closed, exist for any simulation domain involving a free
Fig. 2. The advective component is expected to be the surface. Closed boundaries are simply boundaries across
dominant term in our applications and a bicubic which no flow occurs, and generally represent a coast, a
Lagrange polynomial interpolation is employed in shoreline, or a riverbank. Open boundaries are virtual
MOD_FreeSurf2D to calculate U nsLbicubic . non-physical boundaries across which water may flow
The Coriolis and viscous terms in Eqs. (12) and (13) (Agoshkov et al., 1994), such as a river inflow boundary.
can be included or removed from calculations depending In general, open boundaries may be inflow only, outflow
on the requirements of the model user. The option to only, or both inflow and outflow. For both open and
retain these terms is included in the model to make the closed boundaries, three dependent variables, the total
model as generally applicable as possible. If the Coriolis water depth, the normal velocity component, and the
terms are retained, bilinear interpolation is employed to transverse velocity component, require specification or
calculate V nsLbilinear in the Coriolis term (see Fig. 2), and other treatment.
the Coriolis parameter, f, is calculated with the f-plane Open boundaries can be treated with either Dirichlet
model (Kundu, 1990) given by (fixed value) or radiation boundary conditions in
f ¼ 2O sin y0 , (14) MOD_FreeSurf2D. Depth-averaged velocity, total
water depth, and total water flux can be fixed at open
where y0 is the central latitude of the simulation region boundaries. Fixed total water depth and water flux
and O is the rotation rate of the earth. To remove the boundaries should only be applied to inflow boundaries.
Coriolis terms from Eqs. (12) and (13), the model user Dirichlet velocity conditions may be applied to both
would set O to zero. inflow and outflow boundaries (Agoshkov et al., 1994).
For the viscous terms, a constant horizontal eddy Two types of radiation boundary conditions are
viscosity coefficient, e, is specified by the user. Because available. Both radiation conditions should only be
the eddy viscosity terms are generally small relative to applied to outflow open boundaries in the model. The
the other terms in Eqs. (1) and (2),  is often set equal to first type of radiation boundary is a projection of
zero (Casulli and Cheng, 1992) effectively removing velocity normal to the domain boundary:
these terms from numerical representation. If the viscous
terms are included in model calculations, a centered qU qU
þ U upw ¼ 0, (17)
difference stencil is employed in the calculation of these qt qn
terms as shown in Fig. 2. where the drift velocity term, U upw , is simply the
upwinded, normal directional velocity component, and
3.3. Representation of wetting and drying n represents the direction normal to the domain
boundary (Tseng, 2003). This simple condition does an
An important feature of the model is that channel adequate job of allowing information to propagate out
boundaries within the simulation domain do not require of the simulation domain.
specification, as the model automatically determines The other radiation boundary condition is applied to
wetting or drying in response to any change in flow free-surface elevation, Z, Eq. (18). Orlanski (1976)
conditions. The Arakawa C-grid layout (see Fig. 1) and developed this boundary condition to limit, or absorb,
total depth variables, employed by Casulli and Cheng wave reflections at open boundaries. The core of the
(1992) and adopted here, ensure that the normal velocity method is the calculation of a propagation velocity, C n ,
ARTICLE IN PRESS
N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946 935

from grid points surrounding the boundary with the 1991; Hsu et al., 2000; Jovanovic’ and Djordjevic, 1995;
leap-frog finite-difference representation: Tseng and Chu, 2000). It involves a flume 21.2 m in
length with an open downstream end and a closed
qZ qZ
þ C n ¼ 0. (18) upstream end (Fig. 3). A dam is located 8.5 m from the
qt qn
upstream end (12.7 m from the downstream end). The
flume contains a curved constriction, beginning at 5.0 m
from the closed end and terminating at 4.7 m from the
4. Model applications open end. The narrowest part of the flume is 0.6 m and
occurs at the dam location. In the areas outside of the
MOD_FreeSurf2D was applied to two test applica- constriction, the flume is 1.4 m wide. In the experiment
tions for which there were published fluid-flow data sets. of interest, the slope of the flume is 0.002, and the water
The first application employed published data from a depth behind the dam is 0.15 m.
dam-break flume experiment to demonstrate the model’s Water depth during the experiment was measured at
simulation capability and to highlight the model’s eight different locations along the flume midline. Wave
success in treating rapidly wetting and drying condi- meters recorded measurements at areas of critical to sub-
tions. The second case used detailed river-flow data to critical flow regime. Specifically, wave meters were
check the model’s capability to reproduce velocities and located at two locations upstream of the dam
water depths on the reach scale. (x ¼ 8:5 and 4:5 m) and at locations directly adjacent
to the dam on each side (x ¼ 0:0 and þ0:0 m). Pressure
4.1. Application I: Flume experiment dam-break case transducers were employed to measure water depth,
assuming a hydrostatic pressure distribution, at four
A variety of 2D fluid flow models have been tested on locations downstream of that dam characterized by
dam-break style conceptual problems (Alcrudo and super-critical flow conditions (x ¼ þ2:5, þ5:0, þ7:5, and
Garcia-Navarro, 1993; Anastasiou and Chan, 1997; þ10:0 m). The removal of the dam, or the water release
Bermudez et al., 1998; Fennema and Chaudhry, 1989; time, is the starting point for measurements. The
Fennema and Chaudhry, 1990; Galland et al., 1991; experiment lasted for approximately 70 s with water
Garcia and Kahawita, 1986; Zhao et al., 1994; Zoppou depth measurements published for about 62 s after dam
and Roberts, 2000), have been used to model dam-break removal. Water depth data for the experiment are
style experimental setups (Bellos et al., 1991; Fraccarollo provided for each measurement location by approxi-
and Toro, 1995; Hsu et al., 2000; Jovanovic’ and mately 50 plotted points. These points were taken from
Djordjevic, 1995; Molls and Chaudhry, 1995; Tseng the published data with digitization software. Each
and Chu, 2000), and have been employed to model dam- point corresponds to the average signal value for a time
break induced inundation (Hervouet, 2000; Tucciarelli interval of 1.4 s (Bellos et al., 1992).
and Termini, 2000). Although the high gradient in the The model was used to simulate this experiment given
free surface across the dam produces substantial the flume topography, the initial water height behind the
deviations from the assumptions underlying the shallow dam, and a homogeneous value of Manning’s roughness
water equations, the dam break problem is often coefficient. A radiation velocity boundary condition,
considered as a 2D flow (Fraccarollo and Toro, 1995). Eq. (17), was specified for the outflow end of the flume.
A dam-break experiment, the initially dry bed case of Bellos et al. (1991) suggest employing a Manning’s
Bellos et al. (1992) with an initial water depth of 0.15 m roughness of 0.012 for the glass and smooth-steel
behind the dam, and a flume slope of 0.002, was chosen flume configuration. The experiment was simulated
for testing of MOD_FreeSurf2D. This particular case with Mn ¼ 0:012, but a slightly better match to the
has been employed to test other 2D models (Bellos et al., measured depth values downstream of the dam is

Fig. 3. Plan view flume layout for dam-break experiment of Bellos et al. (1992). Wave meter locations are shown with circles. Pressure
transducer locations are displayed with stars. At the beginning of the experiment, dam is removed and water behind dam travels from
left to right and out of the open end of flume. Flume slope is 0.002. Flume length is 21.2 m and width is 1.4 m. Location 1 is at
x ¼ 8:50 m; Location 2 is at x ¼ 4:00 m, Location 3 is at x ¼ 0:00 m, Location 4 is at x ¼ þ0:0 m, Location 5 is at x ¼ þ2:50 m,
Location 6 is at x ¼ þ5:00 m, Location 7 is at x ¼ þ7:50 m, and Location 8 is at x ¼ þ10:00 m.
ARTICLE IN PRESS
936 N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946

Hydrograph x = -8.50 m Hydrograph x = -4.00 m


0.20 0.20
0.18 Simulated Depth [m] 0.18 Simulated Depth [m]

Water Depth [m]


Water Depth [m]
0.16 Measured Depth [m] 0.16 Measured Depth [m]
0.14 0.14
0.12 0.12
0.10 0.10
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0.00 0.00 0 10 20 30 40 50 60 70
0 10 20 30 40 50 60 70
(A) Time [s] (B) Time [s]
Hydrograph x = -0.00 m Hydrograph x = +0.00 m
0.16 0.16
0.14 Simulated Depth [m] 0.14 Simulated Depth [m]
Water Depth [m]

Water Depth [m]


Measured Depth [m] Measured Depth [m]
0.12 0.12
0.10 0.10
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0.00 0.00 0
0 10 20 30 40 50 60 70 10 20 30 40 50 60 70
(C) Time [s] (D) Time [s]
Hydrograph x = +2.50 m Hydrograph x = +5.00 m
0.10 0.10
Simulated Depth [m] Simulated Depth [m]
Water Depth [m]

Water Depth [m]

0.08 Measured Depth [m] 0.08 Measured Depth [m]

0.06 0.06

0.04 0.04

0.02 0.02

0.00 0.00
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
(E) Time [s] (F) Time [s]
Hydrograph x = +7.50 m Hydrograph x = +10.00 m
0.08 0.08
Simulated Depth [m] Simulated Depth [m]
Water Depth [m]
Water Depth [m]

Measured Depth [m]


0.06 0.06

0.04 0.04

0.02 0.02

0.00 0.00
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
(G) Time [s] (H) Time [s]

Fig. 4. Dam-break flume data compared to simulated depth values. Simulated depths, for Mn ¼ 0:010, y ¼ 0:8, Dx ¼ 0:1250 m,
Dy ¼ 0:5000 m, and Dt ¼ 0:103 s, and measured depths are displayed for each experimental measurement location. No data are
provided for Fig. 4G. Crce ¼ 1:0 for this simulation.

obtained by a modest adjustment in Mn. Fig. 4 displays that the simulated versus measured depth comparison
simulated water depths at each measurement location deteriorated.
for a simulation with Dx ¼ 0:1250 m, Dy ¼ 0:0500 m, In Fig. 4, simulated depth values match the measured
Dt ¼ 0:103 sec, and Mn ¼ 0:010. A roughness coefficient values well for the three measurement locations above
of 0.010 is not unreasonable for a glass and smooth-steel the dam. Simulated values below the dam do not match
flume since a minimum value of 0.011 is given for a measured values as well as above the dam because the
smooth-steel, lined, open channel (Street et al. 1996) downstream locations are characterized by critical to
and for a plane-bedded sand, open channel (Shen and super-critical flow. The model does not simulate these
Julien, 1993). We decreased Mn below 0.010 and found downstream locations as well, because the important
ARTICLE IN PRESS
N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946 937

vertical velocity information in the transition from sub-


to super-critical flow is not reproduced by the model as a
result of depth-averaging. However, the model is stable
and robust enough to simulate this experiment moder-
ately well despite the super-critical flow regions.
In addition to the results displayed in Fig. 4, 40
simulations were completed to explore the effects of
temporal and spatial discretizations. The four discretiza-
tions employed were Dx ¼ 0:0625 and Dy ¼ 0:0250 m,
Dx ¼ 0:1250 and Dy ¼ 0:0500 m, Dx ¼ 0:2500 and
Dy ¼ 0:1000 m, and Dx ¼ 0:5000 and Dy ¼ 0:2000 m.
Ten different time steps, corresponding to Crce of
0.3–2.0 (see Eq. (19)), were used with each spatial
discretization:

pffiffiffiffiffiffiffi Dt pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Dt
Crce ¼ gH ¼ 9:81  0:15 . (19) Fig. 6. Convergence and accuracy for simulated depths at four
Dx Dx
sub-critical to critical locations. Sum of Normalized Errors,
Crce is employed to represent the CFL criterion in this SNE, is provided by Eq. (20A). For sub- to critical locations,
case because the celerity provided by the initial dam- Crce less than or equal to 1.4 are providing roughly the same
break is expected to be the dominant model speed. degree of accuracy.
Multiple spatial layouts and time steps provide further
evidence of model consistency and convergence, and
show that the model can provide accurate results when
the CFL criterion is exceeded.
Figs. 5–7 display results from a suite of simulations
employing the four different discretizations, different
time steps to provide the various Crce numbers, and an
Mn equal to 0.010. These figures demonstrate model
convergence, and display the relative accuracy of
simulations at relatively high values of Crce. Fig. 5
displays the difference in simulated water mass remain-
ing in the flume. Essentially the same amount of water is
left in the simulation domain after 70 s when the value of
Crce is less than or equal to 1.5. When the Crce value is

Fig. 7. Convergence and accuracy for simulated depths at three


super-critical locations. Sum of Normalized Errors, SNE, is
provided by Eq. (20A). For super-critical locations, Crce less
than or equal to 1.4 are providing roughly the same degree of
accuracy.

greater than 1.5, results show that water depth is no


longer accurately represented.
The Sum of Normalized root mean square Errors
(SNE) is employed as a measure of relative accuracy and
is displayed in Figs. 6 and 7. Here it provides a statistic
that is used to compare simulated water depths at each
location to the measured values from Bellos et al. (1992).
Fig. 5. Percentage of the initial water mass remaining in flume The SNE given in Eq. (20A), is obtained by computing
after 70 s. Simulations with Crce larger than 1.5 are not the root mean square error, RMSE, at a location
providing results of the same degree of accuracy as lower Crce (Eq. (20B)), dividing it by the maximum measured water
value simulations. depth at that location, and summing the values over all
ARTICLE IN PRESS
938 N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946

locations (S total locations) at which observations exist: 4.2. Application II: Flow in a 400 m reach of the Kootenai
River, Idaho
XS  
RMSE k
SNE ¼ , (20A) The second test case for the model comes from data
k¼1
max H k
provided by a US Geological Survey study of flow in the
Kootenai River, Idaho (Lipscomb et al., 1998). As part
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi of this study, velocity and water depth data were
u
u1 X M
RMSE ¼ t ðxf  xv Þ2 . (20B) collected along 41 transects approximately 10 m apart
M 1 with an Acoustic Doppler Current Profiler (ADCP) for
three reaches of the Kootenai River. The transects,
In Eqs. (20A) and (20B) xf is the simulated value; xv is taken together, provide a 3D profile of stream velocity in
the measured value, and M is the number of measure- each reach. To test the model’s predictive capability on
ments. the reach scale, simulated water depth and water
The SNE decreases as the spatial representation is velocities for Reach 1 of the Kootenai River were
refined, as shown in Figs. 6 and 7. Convergence occurs compared to the data collected by the USGS.
at both the sub-critical and critical locations. Accurate The ADCP data provide a set of 3D velocities.
simulated water depths are generated for the four sub- Because the model solves the depth-averaged shallow-
critical to critical measurement locations when Crce does water equations, the velocity values collected for each
not exceed 1.5. Similar results were obtained for the water column must be transformed into a single value
super-critical locations but at a lower Crce; accurate for each water column location. Depth averaging was
simulated depths are generated at super-critical loca- done using numerical integration and yielded 1591
tions when Crce does not exceed 1.4. depth-averaged values. Specifically, Simpson’s Rule
Wetting of the flume below the location of the dam is was used to integrate the velocity values in each water
accurately simulated as shown clearly in Fig. 4. In column. Then, the average value was obtained by
addition, Fig. 8, shows the simulation results when the dividing by the total water depth at each measurement
time frame in the dam-break case time was extended by location. The depth-averaged velocity and depth data
5 min. At the end of 6 min, all measurement locations are are presented in the contour plots of Figs. 9 and 10.
effectively dry. Taken together, these three figures demon- Topography, water depth, and a Manning’s rough-
strate the ability of the model to simulate wetting and ness coefficient were used in the model to simulate
drying in response to rapidly changing flow conditions. Reach 1. The topography data from the study are depths

Hydrograph x = -8.50 m Hydrograph x = +0.00 m


0.20 0.16
0.18 Simulated Depth [m] 0.14 Simulated Depth [m]
Water Depth [m]

Water Depth [m]

0.16
0.12
0.14
0.12 0.10
0.10 0.08
0.08 0.06
0.06
0.04
0.04
0.02 0.02
0.00 0.00
0 100 200 300 400 500 600 0 100 200 300 400 500 600
(A) Time [s] (B) Time [s]

Hydrograph x = +5.00 m Hydrograph x = +10.00 m


0.10 0.08
Simulated Depth [m] Simulated Depth [m]
Water Depth [m]

Water Depth [m]

0.08
0.06
0.06
0.04
0.04

0.02 0.02

0.00 0.00 0
0 100 200 300 400 500 600 100 200 300 400 500 600
(C) Time [s] (D) Time [s]

Fig. 8. Simulated dam-break flume experiment water depths at four measurement locations over 6 min. In this simulation,
Dx ¼ 0:1250 m, Dy ¼ 0:0500 m, Dt ¼ 0:103 s, y ¼ 0:8, and Mn ¼ 0:010. At end of 6 min, all measurement locations are effectively dry.
ARTICLE IN PRESS
N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946 939

0 0.7
0.6 0.8
5403900 0.8
0.2 0.4 0.7 0.6 0.9
Northing in Meters
0.9 1.0 0.9
0.8
0.7
0.8
0.6
5403800 0.7
0.4 0.2
0.7 0.4 0
0.6
0.2
0
5403700
543400 543500 543600 543700 543800
Easting in Meters

Fig. 9. Contours of velocity magnitudes obtained by depth-averaging velocity data for Reach 1 of the Kootenai River. Horizontal axis
is meters east-west. Vertical axis is meters north-south. Contours are velocity magnitudes in meters per second.

0
4 6 8 10 12
5403900 12 14
2 4 10 14
6 8
Northing in Meters

12
14 15 14
10
12
14 8
10
8 6
10 6 4
5403800 8
4 2
6
2 0
4
0
5403700
543400 543500 543600 543700 543800
Easting in Meters

Fig. 10. Contours of measured water depths in Reach 1 of the Kootenai River. Horizontal (east-west) and vertical (north-south) axes
in meters. Contours are total water depth in meters.

to the bottom measurements along each transect. These measurement, 1240 m3/s (Lipscomb et al., 1998), was
data were interpolated to square grids of 30, 15, 10, 6, 5, used for the inflow boundary condition and the free-
and 3 m using a quadratic radial basis function (Golden surface radiation condition from Eq. (18) provided the
Software, 1999). The grids are aligned in such a way that outflow boundary condition. Flow was distributed by
all the defined larger grid points are represented in every multiplying the vertically integrated velocities from the
higher resolution grid. A complementary initial water- ADCP by the local water depths. Given the Dirichlet
depth grid was generated in the same manner for each inflow boundary, steady state was achieved by marching
different grid size. A uniform value of Manning’s through time until all velocity and depth values were
roughness coefficient, Mn ¼ 0:055, was applied to every constant over time.
computational volume in the simulation domain. This A number of simulations were completed for Reach 1
roughness coefficient value is from the higher end of the using the initial and boundary conditions stipulated
expected range of roughness coefficients in rivers above to investigate the effects of time step size. The
(Barnes, 1967; Street et al., 1996) but provided a better simulated velocity and depth values were compared to
fit for simulated velocities than obtained with lower the depth-averaged measurements. Bilinear interpola-
coefficient values. The mean discharge on the day of tion of the simulated values provided velocity and depth
ARTICLE IN PRESS
940 N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946

0
0.4 0.6 0.7 0.8 0.9
5403900 0.2 0.4 0.6 0.7 0.8 0.9
1.0
Northing [m] 1.0
0.9
0.9 0.8
0.7
0.8
0.6
5403800 0.7 0.4
0.8 0.6 0.4 0.2
0
0.2
0.7 0.6
0

5403700
543400 543500 543600 543700 543800
Easting [m]

Fig. 11. Contours of simulated velocity values for Reach 1 of the Kootenai River. In this simulation, Dx ¼ Dy ¼ 5:0 m, Dt ¼ 2:4 s, and
y ¼ 0:8. Horizontal axis is meters east-west. Vertical axis is meters north-south. Contours are velocity magnitudes in meters per second.

0
4
6 8 14
5403900 10 12
2 4 14
6 8 12
10 12
14
Northing [m]

14 15
12 10
14 8
10
12 8 6
10 6 4
5403800 8
4 2
6
2 0

4
2 0

5403700
543400 543500 543600 543700 543800
Easting [m]
Fig. 12. Contours of simulated water depths for Reach 1 of the Kootenai River. In this simulation, Dx ¼ Dy ¼ 5:0 m, Dt ¼ 2:4 s, and
y ¼ 0:8. Horizontal axis is meters east-west. Vertical axis is meters north-south. Contours are total water depth in meters.

magnitudes at each water column measurement loca- 0.108 m/s. These values can be viewed in the context of
tion. Fig. 11 displays contours of the simulated velocity the depth-averaged maximum measured velocity for the
values at the measurement locations. Fig. 12 shows the reach, which is 1.175 m/s, and the median measured
contours of simulated water depth at the measurement depth-averaged velocity, which is 0.752 m/s. The median
locations. The simulation that generated Figs. 11 and 12 absolute residual of depth is 0.088 m and the RMSE of
employed Dx ¼ Dy ¼ 5:0 m, Dt ¼ 2:4 s, with y ¼ 0:8. depth is 0.337 m. These values can be viewed in the
Simulations with spatial discretizations of 10 m or context of the maximum measured water depth of
smaller provide similar results at steady state. 16.340 m, and the median measured depth of 7.580 m.
Visual comparison of velocities in Figs. 9 and 11 and Fig. 13 contains a contour map of velocity residuals,
water depths in Figs. 10 and 12 indicates the simulation ðxf 2xv Þ from Eq. (20B), and shows that the location of
ability of the model. The measured and simulated the larger magnitude velocity residuals is along the bank
velocity contour plots and depth contour plots or close to the outflow boundary. One reason that the
have similar contour shapes and magnitudes. Quantita- larger magnitude velocity residuals are located adjacent
tive comparison shows a median absolute residual of to the riverbank is that the rectangular finite-volume
0.065 m/s and a RMSE (Eq. (20B)) of velocity of discretization does not fully capture the irregular
ARTICLE IN PRESS
N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946 941

Fig. 13. Filled contour plot of velocity residuals calculated from the simulation displayed in Fig. 11. Large magnitude residuals are
primarily located near riverbank and near outflow boundary. Velocity residuals are in meters per second.

Fig. 14. Histogram of velocity residual values obtained from the simulation displayed in Fig. 11. These residual values were employed
to generate Fig. 13. Residual values are roughly centered on zero with a median residual value of 0.025, maximum residual value of
0.352, and minimum residual value of –0.546.

geometry of the riverbank. Although the model permits the velocity residual values are clustered about zero with
the use of varying roughness coefficients across the a slight positive bias of median residual of 0.025 m/s, or
simulation domain, a uniform Manning’s roughness 3.4% of the median velocity. Fig. 15 shows that
coefficient was used in this simulation. Improvement in the histogram of depth residuals is also centered
simulation results relative to data might be achieved about zero.
by varying the roughness coefficient across the channel We also investigated the accuracy of the simulations
or reach. when the CFLo1 condition was greatly exceeded.
In Figs. 14 and 15, histograms of velocity and depth Fig. 16 displays the change in the RMSE of velocity as
residual values are displayed. Fig. 14 demonstrates that the grid Courant number increases. The grid Courant
ARTICLE IN PRESS
942 N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946

Fig. 15. Histogram of water depth residuals obtained from the simulation displayed in Fig. 12. Residual values are roughly centered on
zero with a median residual value of 0.016, maximum residual value of 2.029, and minimum residual value of –2.278.

Fig. 16. RMSE of velocity, from Eq. (20B), for each Crmax value, from Eq. (21). All of the velocity RMSE values are close to 0.10
demonstrating that the program maintains accuracy for high values of Crmax . The decrease in RMSE with increase in Crmax is result of
smoothing effect shown in Fig. 18 and 19.

number for these simulations is given by where Umax is the maximum simulated directional
velocity component. As Crmax increases, the RMSE of
Dt velocity remains close to 0.10 and even decreases
Crmax ¼ U max , (21)
Dx slightly. At Crmax values above the largest value given
ARTICLE IN PRESS
N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946 943

Fig. 17. RMSE of simulated water depth values, from Eq. (20B), for each Crmax value, from Eq. (21). Consistent decrease in RMSE
with decrease in Dx demonstrates model convergence. RMSE decreases with Dx because resolution of irregular riverbank improves.

0
0.4 0.6 0.7 0.8 1.0
5403900
0.2 0.4 0.6 0.7 0.8 0.9
1.0
Northing [m]

0.9
0.9 0.8
0.9 0.8 0.7

0.7 0.6
5403800
0.8 0.6 0.4 0.2
0.4 0
0.2
0.7
0.6
0

5403700
543400 543500 543600 543700 543800
Easting [m]

Fig. 18. Contours of velocity magnitudes, simulated with Dx ¼ Dy ¼ 5:0 m, Dt ¼ 10:0 s and y ¼ 0:8, for Reach 1 of the Kootenai
River. Velocity contours are in meters per second. This simulation corresponds to Crmax ¼ 2:0. Size of 1.0 m/s contour has decreased
relative to Fig. 11.

for each Dx, the model did not converge to a solution. plots of velocity magnitude from two additional
Fig. 17 displays the change in the RMSE of simulated simulations, are presented. These two figures represent
depth values as Crmax increases. The RMSE depth simulations employing identical conditions and para-
values do not change with Crmax. However, the depth meter values to those that produced Fig. 11 except that
RMSE values become smaller as Dx decreases and better the model time step and Crmax are different for each
resolution is obtained of the irregular water/land simulation. In Fig. 11, (Dt ¼ 2:4; Dt ¼ 10:0 in Fig. 18,
boundary. Examination of Figs. 16 and 17 suggests that and (Dt ¼ 50:0 in Fig. 19). Using Eq. (21) and U max
the model provides a reasonable steady-state solution equal to the maximum simulated component velocity for
for Crmax exceeding 10. each time step, Crmax ¼ 0:5 when Dt ¼ 2:4; Crmax ¼ 2:0
To examine the slight decrease in the RMSE of when Dt ¼ 10:0, and Crmax ¼ 10:0 when Dt ¼ 50:0. In
velocity as Crmax increases, Figs. 18 and 19, contour these three figures, the size of the 1.0 m/s contour
ARTICLE IN PRESS
944 N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946

0
0.8 0.9
0.6 0.7
5403900
0.2 0.4 0.6 0.7 0.8

Northing [m] 0.9


0.9

0.9 0.9 0.8 0.7


0.8
0.7 0.6
5403800 0.8
0.6 0.4 0.2
0.7 0.4 0.2 0

0.6
0

5403700
543400 543500 543600 543700 543800
Easting [m]

Fig. 19. Contours of velocity magnitudes, simulated with Dx ¼ Dy ¼ 5:0 m, Dt ¼ 50:0 s and y ¼ 0.8, for Reach 1 of the Kootenai
River. Velocity contours are in meters per second. This simulation corresponds to Crmax ¼ 10:0. Size of 1.0-m/s contour has decreased
relative to Figs. 11 and 18.

decreases as Dt and Crmax increase while the other exceeds 10 in a steady-state solution. The dam-break
contours maintain shape and location in results from all flume simulations also demonstrate that the model’s
three simulations. This suggests that employing a larger ability to simulate wetting and drying of the domain.
time step smoothes the values with the largest magni-
tudes. The 1.0 m/s contours of the measured velocity
values in Fig. 9 are similar in size and location to the
equivalent contour in Fig. 19. The improvement in the Acknowledgements
fit of the 1.0 m/s contour as Crmax increases accounts for
the slight improvement in RMSE of velocity. This material is based upon work supported by the
National Science Foundation under Grant No. EAR-
0207177. Any opinions, findings, and conclusions or
5. Conclusions recommendations expressed in this material are those of
the authors and do not necessarily reflect the views of
MOD_FreeSurf2D solves the depth-averaged, shal- the National Science Foundation. This material is also
low-water equations using a semi-implicit, semi-Lagran- based upon work supported under a Stanford Graduate
gian numerical approximation. This open source code Fellowship. In addition, we would like to thank
primarily employs the numerical methods proposed by S. Lipscomb and C. Berenbrock for providing the
Casulli and Cheng (1992) and has the advantages of Kootenai River data. We would also like to thank R.L.
stability, representation of moving land/water bound- Street for his insight into and his suggestions on free
aries, and the ability to employ time step sizes that surface flow modeling. We want to acknowledge the
exceed the CFL criterion. Although the model boundary condition suggestions received from Y.-H.
was designed for a specific purpose, it can be employed Tseng. Finally, we would like to thank two anonymous
to simulate water depth and water velocity in reviewers for insightful suggestions that improved the
general situations where the governing shallow water quality of this paper.
equations apply.
Comparisons to data at the laboratory and field scales
show that the model accurately simulated both a
transient dam-break flume experiment and steady-state References
flow in the Kootenai River. Sensitivity analyses con-
Agoshkov, V.I., Quarteroni, A., Saleri, F., 1994. Recent
ducted on the two cases demonstrate that the model
developments in the numerical simulation of shallow water
yields accurate results when the CFL condition is equations I: boundary conditions. Applied Numerical
exceeded and demonstrate model convergence. Accu- Mathematics 15, 175–200.
racy was maintained when the CFL value based on Akanbi, A., Katopodes, N.D., 1988. Model for flood propaga-
celerity, Crce, was 1.4 in a transient solution and when tion on initially dry land. Journal of Hydraulic Engineering
the CFL value based on the maximum velocity, Crmax, 114, 689–719.
ARTICLE IN PRESS
N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946 945

Alcrudo, F., Garcia-Navarro, P., 1993. A high-resolution dam-break type problems. Journal of Hydraulic Research
Godunov-type scheme in finite volumes for the 2D 33 (6), 843–864.
shallow-water equations. International Journal for Numer- Fujihara, M., Borthwick, A.G.L., 2000. Godunov-type solution
ical Methods in Fluids 16, 489–505. of curvilinear shallow-water equations. Journal of Hydrau-
Anastasiou, K., Chan, C.T., 1997. Solution of the 2D shallow lic Engineering 126 (11), 827–836.
water equations using the finite volume method on Galland, J.-C., Goutal, N., Hervouet, J.-M., 1991. TELEMAC:
unstructured triangular meshes. International Journal for a new numerical model for solving shallow water equations.
Numerical Methods in Fluids 24, 1225–1245. Advances in Water Resources 14 (3), 138–148.
Barnes Jr., H.H., 1967. Roughness characteristics of natural Garcia, R., Kahawita, R.A., 1986. Numerical solution of the St.
channels. US Geological Survey Water-Supply Paper 1849, Venant equations with the MacCormack finite-difference
219pp. scheme. International Journal for Numerical Methods in
Bates, P.D., Anderson, M.G., 1993. A two-dimensional finite- Fluids 6, 259–274.
element model for river flow inundation. Proceedings of the Golden Software, I, 1999. Surfer 7 User’s Guide. Golden
Royal Society of London (A) 440, 481–491. Software, Inc., Golden, CO 619pp.
Bellos, C.V., Soulis, J.V., Sakkas, J.G., 1991. Computation of Guillou, S., Nguyen, K.D., 1999. An improved technique for
two-dimensional dam-break induced flows. Advances in solving two-dimensional shallow water problems. Interna-
Water Resources 14 (1), 31–41. tional Journal for Numerical Methods in Fluids 29,
Bellos, C.V., Soulis, J.V., Sakkas, J.G., 1992. Experimental 465–483.
investigation of two-dimensional dam-break induced flows. Heniche, M., Secretan, Y., Boudreau, P., Leclerc, M., 2000. A
Journal of Hydraulic Research 30, 47–62. two-dimensional finite element drying-wetting shallow
Benque, J.P., Cunge, J.A., Feuillet, J., Hauguel, A., Holly Jr., water model for rivers and estuaries. Advances in Water
F.M., 1982. New method for tidal current computation. Resources 23, 359–372.
Journal of the Waterway, Port, Coastal and Ocean Division, Hervouet, J.-M., 2000. A high resolution 2-D dam-break
ASCE 108 (WW3), 396–417. model using parallelization. Hydrological Processes 14,
Bermudez, A., Dervieux, A., Desideri, J.-A., Vazquez, M.E., 2211–2230.
1998. Upwind schemes for the two-dimensional shallow Hervouet, J.-M., Haren, L.V., 1996. Recent advances in
water equations with variable depth using unstructured numerical methods for fluid flows. In: Anderson, M.G.,
meshes. Computer Methods in Applied Mechanics and Walling, D.E., Bates, P.D. (Eds.), Floodplain Processes.
Engineering 155, 49–72. Wiley, Chichester, UK, pp. 183–214.
Casulli, V., 1990. Semi-implicit finite difference methods for the Hsu, C.-T., Yeh, K.-C., Yang, J.-C., 2000. Depth-averaged
two-dimensional shallow water equations. Journal of two-dimensional curvilinear explicit finite analytic model for
Computational Physics 86, 56–74. open-channel flows. International Journal for Numerical
Casulli, V., 1997. Numerical simulation of three-dimensional Methods in Fluids 33, 175–202.
free surface flow in isopycnal coordinates. Inter- Jovanovic’, M., Djordjevic, D., 1995. Experimental verification
national Journal for Numerical Methods in Fluids 25, of the MacCormack numerical scheme. Advances in
645–658. Engineering Software 23, 61–67.
Casulli, V., 1999. A semi-implicit finite difference method for Katopodes, N., Strelkoff, T., 1978. Computing two-dimen-
non-hydrostatic, free-surface flows. International Journal sional dam-break flood waves. Journal of the Hydraulics
for Numerical Methods in Fluids 30, 425–440. Division, ASCE 104 (HY9), 1269–1288.
Casulli, V., Cattani, E., 1994. Stability, accuracy and efficiency Kawahara, M., Umetsu, T., 1986. Finite element method for
of a semi-implicit method for three-dimensional shallow moving boundary problems in river flow. International
water flow. Computers and Mathematics with Applications Journal for Numerical Methods in Fluids 6, 365–386.
27 (4), 99–112. King, I.P., 1977. Finite element models for unsteady flow
Casulli, V., Cheng, R.T., 1992. Semi-implicit finite difference routing through irregular channels. In: Gray, W.G., Pinder,
methods for three-dimensional shallow water flow. Interna- G.F., Brebbia, C.A. (Eds.), Finite Elements in Water
tional Journal for Numerical Methods in Fluids 15, Resources. Proceedings of theFirst International Confer-
629–648. ence on Finite Elements in Water Resources, Princeton, NJ,
Falconer, R.A., Chen, Y., 1991. An improved representation of pp. 4.185–4.204.
flooding and drying and wind stress effects in a two- King, I.P., Norton, W.R., 1978. Recent application of RMA’s
dimensional tidal numerical model. Proceedings of the finite element models for two dimensional hydrodynamics
Institution of Civil Engineers 91(Part 2), 659–678. and water quality. In: Brebbia, C.A., Gray, W.G., Pinder,
Fennema, R.J., Chaudhry, M.H., 1989. Implicit methods for G.F. (Eds.), Finite Elements in Water Resources. Proceed-
two-dimensional unsteady free-surface flows. Journal of ings of the Second International Conference on Finite
Hydraulic Research 27 (3), 321–332. Elements in Water Resources, London, pp. 2.81–2.100.
Fennema, R.J., Chaudhry, M.H., 1990. Explicit methods for Kundu, P.K., 1990. Fluid Mechanics. Academic Press, San
two-dimensional transient free-surface flows. Journal of Diego, CA 638pp.
Hydraulic Engineering 116 (8), 1013–1034. Layton, A.T., Panne, M.v.d., 2002. A numerically efficient and
Fletcher, C.A.J., 1991. Computational Techniques for Fluid stable algorithm for animating water waves. Visual Com-
Dynamics, vol. I. Springer, New York 401pp. puter 18, 41–53.
Fraccarollo, L., Toro, E.F., 1995. Experimental and numerical Leclerc, M., Bellemare, J.-F., Dumas, G., Dhatt, G., 1990. A
assessment of the shallow water model for two-dimensional finite element model of estuarian and river flows with
ARTICLE IN PRESS
946 N. Martin, S.M. Gorelick / Computers & Geosciences 31 (2005) 929–946

moving boundaries. Advances in Water Resources 13 (4), Tee, K.T., 1977. Tide-induced residual current—verification of
158–168. a numerical model. Journal of Physical Oceanography 7,
Lipscomb, S.W., Berenbrock, C., Doyle, J.D., 1998. Spatial 396–402.
distribution of stream velocities for the Kootenai River near Toro, E.F., 1992. Riemann problems and the WAF method for
Bonners Ferry, Idaho, June 1997. US Geological Survey solving the two-dimensional shallow water equations.
Open File Report 97-830, 174pp. Philosophical Transactions of the Royal Society of London
Martin, N., Gorelick, S.M., 2005. Semi-analytical method for (A) 338, 43–68.
departure point determination. International Journal for Tseng, M.H., Chu, C.R., 2000. Two-dimensional shallow water
Numerical Methods in Fluids 47, 121–137. flows simulation using TVD-MacCormack scheme. Journal
Molls, T., Chaudhry, M.H., 1995. Depth-averaged open- of Hydraulic Research 38 (2), 123–131.
channel flow model. Journal of Hydraulic Engineering 121 Tseng, Y.-H., 2003. On the development of a ghost-cell
(6), 453–465. immersed boundary method and its application to
Orlanski, I., 1976. A simple boundary condition for unbounded large eddy simulation and geophysical fluid dynamics.
hyperbolic flows. Journal of Computational Physics 21, Ph.D. Dissertation, Department of Civil and Environ-
251–269. mental Engineering, Stanford University, Stanford, CA,
Petera, J., Nassehi, V., 1996. A new two-dimensional finite 207pp.
element model for the shallow water equations using a Tucciarelli, T., Termini, D., 2000. Finite-element modeling of
Lagrangian framework constructed along fluid particle floodplain flow. Journal of Hydraulic Engineering 126 (6),
trajectories. International Journal for Numerical Methods 416–424.
in Engineering 39, 4159–4182. Walters, R.A., Casulli, V., 1998. A robust, finite element model
Robert, A., 1981. A stable numerical integration scheme for the for hydrostratic surface water flows. Communications in
primitive meteorological equations. Atmosphere-Ocean 19, Numerical Methods in Engineering 14, 931–940.
35–46. Zhao, D.H., Shen, H.W., Tabios III, G.Q., Lai, J.S., Tan,
Robert, A., 1982. A semi-Lagrangian and semi-implicit W.Y., 1994. Finite-volume two-dimensional unsteady-flow
numerical integration scheme for the primitive meteorolo- model for river basins. Journal of Hydraulic Engineering
gical equations. Journal of the Meteorological Society of 120 (7), 863–883.
Japan 60, 319–325. Zheng, C., Bennett, G.D., 2002. Applied Contaminant
Shen, H.W., Julien, P.Y., 1993. Erosion and sediment trans- Transport Modeling. Wiley-Interscience, New York, NY
port. In: Maidment, D.R. (Ed.), Handbook of Hydrology. 621pp.
McGraw-Hill, Inc., New York, pp. 12.1–12.61. Zhou, J.G., Goodwill, I.M., 1997. A finite volume method
Staniforth, A., Cote, J., 1991. Semi-Lagrangian integration for steady state 2D shallow water flows. International
schemes for atmospheric models—a review. Monthly Journal of Numerical Methods for Heat & Fluid Flow 7 (1),
Weather Review 119, 2206–2223. 4–23.
Street, R.L., Watters, G.Z., Vennard, J.K., 1996. Elementary Zhou, J.G., Stansby, P.K., 1999. 2D shallow water flow model
Fluid Mechanics. Wiley, New York, NY 757pp. for the hydraulic jump. International Journal for Numerical
Tabuenca, P., Cardona, J., 1992. Numerical model for the study Methods in Fluids 29, 375–387.
of hydrodynamics on bays and estuaries. Applied Mathe- Zoppou, C., Roberts, S., 2000. Numerical solution of the two-
matical Modeling 16, 78–85. dimensional unsteady dam break. Applied Mathematical
Tee, K.T., 1976. Tide-induced residual current, a 2-D nonlinear Modeling 24, 457–475.
numerical tidal model. Journal of Marine Research 34 (4),
603–628.

You might also like