Validation of Delft3d
Validation of Delft3d
Validation of Delft3d
Report
December 2007
X0356, M3470
Report
X0356, M3470
WL | delft
hydraulics
Client:
Title:
Abstract:
This report is the validation document for the mathematical model Delft3D-FLOW. The document is organised
conforming to the Guidelines for documenting the validity of computational modelling software.
Chapter 1 contains a short overview of the computational model and introduces the main issues to be addressed
by the validation process. The model overview includes information about the purpose of the model, about preand post-processing options and other software features, and about reference versions of the software.
Validation priorities and approaches are briefly described, and a list of related documents is included.
Chapter 2 summarises the available information about the validity of the computational core of the model. In
this chapter, claims are made about the range of applicability of the model and about the accuracy of
computational results. Each claim is followed by a brief statement regarding its substantiation. This statement
indicates the extent to which the claim has in fact been substantiated and points to the available evidence.
Chapter 3 contains such evidence, in the form of brief descriptions of relevant validation studies. Each
description includes information about the purpose and approach of the study and a summary of main results
and implications. About thirty validation studies have been documented.
This Delft3D-FLOW validation document, the input and result files of the validation can be supplied to users of
the Delft3D-FLOW system, so that the validity and performance of Delft3D-FLOW can be verified.
References:
Ver
0.3
1.0
Author
H. Gerritsen,
E.D. de Goede
Date
16 March 2004
Remarks
Incomplete, draft
version
H. Gerritsen,
E.D. de Goede
30 December 2007
Final version
Project number:
X0356, M3470
Keywords:
Validation, Delft3D-FLOW
Number of pages:
266
Classification:
None
Status:
Review
B. Jagers
Approved by
A.E. Mynett
X0356, M3470
31 December 2007
Final version 1.0
Contents
WL | Delft Hydraulics
Introduction ......................................................................................................11
1.1
1.2
1.3
1.4
1.5
1.6
Physical system......................................................................................21
2.2
Model functionality................................................................................22
2.2.1 Applications ..............................................................................22
2.2.2 Processes ...................................................................................26
2.3
2.4
2.5
31 December 2007
Final version 1.0
X0356, M3470
3.2
3.3
3.4
References......................................................................................................... 41
Glossary ............................................................................................................ 51
ii
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
List of Figures
Chapter 1: Introduction
1.1 System overview of Delft3D
1.2 Horizontal curvilinear grid concept
1.3 Vertical grid concepts: -model and z-coordinate model of Delft3D-FLOW
WL | Delft Hydraulics
iii
31 December 2007
Final version 1.0
X0356, M3470
3.1.4.3 Water level for 10 m/s wind velocity (both analytical and numerical solutions)
3.1.4.4 Horizontal velocity profile for 10 m/s wind velocity (both analytical and numerical
solutions).
Lock exchange flow
3.1.5.1 Initial distribution of salinity
3.1.5.2 Distribution of salinity at t = 120 s; hydrostatic mode
3.1.5.3 Distribution of salinity at t = 120 s; non-hydrostatic mode
Wave force and a mass flux in a closed basin
3.1.6.1 Wave force and mass flux effect
3.1.6.2 Water level (in m) along the basin; computed (in red) and according to the analytical
solution (in black); on x-axis the x-coordinate (in m) of the basin
Flow over a weir
3.1.7.1 Cross sectional discharges for Cyclic scheme and a 30 m grid size
3.1.7.2 Cross sectional discharges for Flooding scheme and a 30 m grid size
3.1.7.3 Cross sectional discharges for Cyclic scheme and a 10 m grid size
3.1.7.4 Cross sectional discharges for Flooding scheme and a 10 m grid size
Coriolis testcase
3.1.8.1 The spherical-curvilinear 50x50 grid
3.1.8.2 Water level for the -plane solution. The analytical (initial) solution is depicted on the
left and the numerical solution on the right
3.1.8.3 V-velocity for the -plane solution. The analytical (initial) solution is depicted on the
left and the numerical solution on the
3.1.8.4 U-velocity for the -plane solution. The analytical (initial) solution is depicted on the
left and the numerical solution on the right
Equilibrium slope for a straight flume
3.1.9.1 Equilibrium bed slope
Tidal flume
3.2.1.1 Computed and observed salt intrusion; -model
3.2.1.2 Computed and observed salt intrusion; Z-model
Water elevation in a wave flume
3.2.2.1 Bathymetry of laboratory experiment of Beji and Battjes
3.2.2.2 Water level history at a station located at 13.5 m from the inflow boundary
Measurements in blue (with markers), Delft3D-FLOW results in red (no
markers)
3.2.2.3 Water level history at a station located at 15.7 m from the inflow boundary
Measurements in blue (with markers), Delft3D-FLOW results in red (no
markers)
iv
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
3.2.2.4 Water level history at a station located at 19.0 m from the inflow boundary
Measurements in blue (with markers), Delft3D-FLOW results in red (no
markers)
Vertical mixing layer (horizontal splitter plate)
3.2.3.1 Horizontal velocity profiles at 2 m, 5 m, 10 m and 40 m behind the splitter plate.
Red circles represent the measurements and the solid lines represent the
computed results
3.2.3.2 Density profiles at 2 m, 5 m, 10 m and 40 m behind the splitter plate. Red circles
represent the measurements and the solid lines represent the computed results
3.2.3.3 Turbulent kinetic energy profiles at 2 m, 5 m, 10 m and 40 m behind the splitter
plate. Red circles represent the measurements and the solid lines represent the
computed results
One-dimensional dam break
3.2.4.1 Water surface slope for a dam break scenario with initially dry bed
3.2.4.2 Water level for flooding of a dry bed, computed (in blue and black) and analytical
(in red)
3.2.4.3 Water level for flooding of a wet bed, computed (in blue and black) and analytical
(in red)
Horizontal mixing layer (vertical splitter plate)
3.2.5.1 Set-up of experiment (top view)
3.2.5.2 Computational grid at the tip of the splitter plate
3.2.5.3 Mixing layer thickness as a function of distance from the tip of the splitter plate
(which lies at x = 0)
3.2.5.4 Observed (triangles) and simulated (lines) properties of mean flow and turbulence at
a series of downstream cross sections; blue: offline computed flow
characteristics, red: online computed flow characteristics
Numerical scale model of an estuary
3.2.6.1 Detailed numerical scale model grid, Deurgangdock section, situation without dock
3.2.6.2 Depth of detailed numerical scale model, Deurgangdock section, situation without
dock
3.2.6.3 Computed (red) and measured (black) water levels at six stations along the flume
on15 April 2003
3.2.6.4 Computed (red & magenta) and measured (blue & green) salinity at 7.2 cm - TAW
in six stations along the flume on 15 April 2003
3.2.6.5 Computed (red) and measured (black) current magnitude and direction in four
stations along the flume on 15 April 2003
Numerical scale model of an estuary and a tidal dock
3.2.7.1 Numerical scale model grid for domain decomposition, situation with
Deurgangdock and positions of instruments
WL | Delft Hydraulics
31 December 2007
Final version 1.0
X0356, M3470
3.2.7.2 Depth of numerical scale model with domain decomposition, situation with
Deurgangdock
3.2.7.3 Computed (red) and measured (black) water levels at six stations along the flume on
26 May 2003
3.2.7.4 Computed (red & magenta) and measured (blue & green) salinity at six stations
along the flume on 26 May 2003
3.2.7.5 Computed (red & magenta) and measured (blue & green) salinity at two stations in
the Deurgangdock on 26 May 2003
3.2.7.6 Computed (red) and measured (black) current magnitude and direction in four
stations along the flume on 26 May 2003
vi
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
WL | Delft Hydraulics
vii
31 December 2007
Final version 1.0
X0356, M3470
Zegerplas
3.4.2.1 Schematic picture of the Zegerplas (left) and grid that was used (right)
3.4.2.2 Zegerplas, temperature distribution over layers of the -model for numerical
simulation of one year
3.4.2.3 Zegerplas, temperature distribution over layers of the Z-model for numerical
simulation of one year
3.4.2.4 Zegerplas, temperature profile at RO371 (measured at August 14 1998 (the "+"signs), measured at August 11 1999 (the crosses), and computed for
August 12 1996 (solid line)
Lake Grevelingen
3.4.3.1 Overview of the Lake Grevelingen measurement locations (red dots); cross-sections
will be plotted along the red, dashed line
3.4.3.2 Time series of salinity (blue line) at 1 m (upper plot) and 15 m depth (lower plot)
compared to measurements (red dots) at Station GTSO-08
3.4.3.3 Time series of temperature (blue line) at 1 m (upper plot) and 15 m depth (lower
plot) compared to measurements (red dots) at Station GTSO-08
3.4.3.4 Z,t-diagrams of salinity (upper plot) and temperature (lower plot) for Station GTSO08. The coloured dots represent measured values
3.4.3.5 Salinity contour plots at 1 m (upper plot) and 15 m depth (lower plot) compared to
measurements (coloured dots) at 14 June
3.4.3.6 Temperature contour plots at 1 m (upper plot) and 15 m depth (lower plot)
compared to measurements (coloured dots) at 14 June
3.4.3.7 Salinity (upper plot) and temperature (lower plot) along a cross section through
Lake Grevelingen. The coloured dots represent measurements
Sea of Marmara
3.4.4.1 Computed (dashed line) and measured (solid line) water level elevation in Pendik
(near southern Bosporus entrance) for December 2003
3.4.4.2 Computed (dashed line) and measured (solid line) water level elevation in Kavak
(near northern Bosporus entrance) for January 2003
3.4.4.3 Modelled upper (blue), lower (red) and net (black) flux variation through northern
cross section in the Bosporus for January 2003. The coloured dots represent
fluxes derived from ADCP measurements (units are in km3/yr)
3.4.4.4 Modelled upper (blue), lower (red) and net (black) flux variation through northern
cross section in the Bosporus for July 2003. The coloured dots represent fluxes
derived from ADCP measurements (units are in km3/yr)
3.4.4.5 Time series of temperature (blue line) at 1 m (upper plot) and 40 m depth (lower
plot) compared to measurements (red dots) at Station M23 in the north east of
the Sea of Marmara
3.4.4.6 Z,t-diagrams of salinity (upper plot) and temperature (lower plot) for Station M23 in
the north east of the Sea of Marmara. The coloured dots represent measured
values
3.4.4.7 Z,t-diagrams of salinity (upper plot) and temperature (lower plot) for Station K2 in
the south west of the Black Sea. The coloured dots represent measured values
viii
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
3.4.4.8 Salinity (upper plot) and temperature (lower plot) along a cross section through the
Dardanelles, the Sea of Marmara, the Bosporus and the south east of the Black
Sea for January 21, 2003. The coloured dots represent measurements
3.4.4.9 Salinity (upper plot) and temperature (lower plot) along a cross section through the
Dardanelles, the Sea of Marmara, the Bosporus and the south east of the Black
Sea for July 5, 2003. The coloured dots represent measurements
3.4.4.10Salinity (upper plot) and temperature (lower plot) along a cross section through the
Bosporus for January 21, 2003. The coloured dots represent measurements
South China Sea
3.5.1.1 Surface layer temperature during the Northeast (NE, January) and Southwest (SW,
August) monsoon highs. Model results compared with monthly-mean,
climatological Remotely Sensed Sea Surface Temperature (RS SST) data
obtained from (Vazquez 2004)
3.5.1.2 Profile data at selected model stations. Model results (lower panel) compared with
monthly-mean, climatological profile data from the World Ocean Atlas 2001
(upper panel) (Levitus 1982)
List of Abbreviations
2D
Two dimensional
3D
Three dimensional
BSS
IAHR
SWE
RANS
WL | Delft Hydraulics
ix
31 December 2007
Final version 1.0
X0356, M3470
Summary
This document is the validation document for the mathematical model Delft3D-FLOW. The
document is organised conforming to the Guidelines for documenting the validity of
computational modelling software (IAHR, 1994).
The subject of this document is the validation of a computational model. The term
computational model refers to software which primary function is to model a certain class of
physical systems, and may include pre- and post-processing components and other
necessary ancillary programmes. Validation applies primarily to the theoretical foundation
and to the computational techniques that form the basis for the numerical and graphical
results produced by the software. In the context of this document, validation of the model is
viewed as the formulation and substantiation of explicit claims about applicability and
accuracy of the computational results. This preface explains the approach that has been
adopted in organising and presenting the information contained in this document.
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
features, and about reference versions of the software. Validation priorities and approaches
are briefly described, and a list of related documents is included.
Chapter 2 summarises the available information about the validity of the computational core
of the model. In this chapter, claims are made about the range of applicability of the model
and about the accuracy of computational results. Each claim is followed by a brief statement
regarding its substantiation. This statement indicates the extent to which the claim has in
fact been substantiated and points to the available evidence.
Chapter 3 contains such evidence, in the form of brief descriptions of relevant validation
studies. Each description includes information about the purpose and approach of the study
and a summary of main results and implications.
A glossary and complete list of references are contained in this document too.
A word of caution
This document contains information about the quality of a complex modelling tool. Its
purpose is to assist the user in assessing the reliability and accuracy of computational
results, and to provide guidelines with respect to the applicability and proper use of the
modelling tool. This document does not, however, provide mathematical proof of the
correctness of results for a specific application. The reader is referred to the License
Agreement for pertinent legal terms and conditions associated with the use of the software.
The contents of this validation document attest to the fact that computational modelling of
complex physical systems requires great care and inherently involves a number of uncertain
factors. In order to obtain useful and accurate results for a particular application, the use of
high-quality modelling tools is necessary but not sufficient. Ultimately, the quality of the
computational results that can be achieved will depend upon the adequacy of available data
as well as a suitable choice of model and modelling parameters.
Acknowledgements
The guidelines [IAHR, 1994] provide a logical set up of the document. In the preparation of
the present document, we aimed to share the look and feel with other validation documents
in the market place, notably that of TELEMAC-2D [EDF-DRD, 2000] and of
UnTRIM [Bundesanstalt fr Wasserbau, 2002].
WL | Delft Hydraulics
xi
31 December 2007
Final version 1.0
X0356, M3470
List of Symbols
Symbol
Bk
Units
m2/s3
Be
m2/s4
C , C2 D
C3D
m1/2/s
Meaning
buoyancy flux term in transport equation for turbulent kinetic
energy
buoyancy flux term in transport equation for the dissipation of
kinetic energy
2D Chzy coefficient
m1/2/s
3D Chzy coefficient
Cd
c
c( )
kg/m3
kg/m3
mass concentration
cD
cp
c
J/kgC
calibration constant
c'
D
Dback
Dh , Dv
Dmol
d
ds
d50
d90
kg/m2s
m2/s
m2/s
m2/s
m
m
sediment diameter
E
E
Fx
m/s
kg/m2s
m/s2
evaporation
erosion rate cohesive sediment
radiation stress gradient in x-direction
Fy
m/s2
f
g
H
H rms
I
k
ks
L
1/s
m/s2
m
m
m/s
m2/s2
m
M xS
m
kg m/s
mixing length
index number of sediment fraction
depth-averaged mass flux due to Stokes drift in x-direction
M yS
kg m/s
xii
WL | Delft Hydraulics
Symbol
X0356, M3470
31 December 2007
Final version 1.0
Units
m/s2
Meaning
source or sink of momentum in x- and y-direction
m1/3/s
kg/ms2
m/s
m2/s3
Manning's coefficient
hydrostatic water pressure
precipitation
production term in transport equation for turbulent kinetic energy
P P
kg/m2s2
m2/s4
Q
qin
qout
R
Ri
S
Sb
s
m/s
1/s
1/s
m
ppt
-
T
T
T , T ,T
t
U
U
u
u*
M M
n
P
P
Pk
C
K
kg/ms2
s
m/s
m/s
time
depth-averaged velocity in x-direction
velocity of water discharged in y-direction
m/s
m/s
u*b
m/s
u*s
m/s
m/s
U10
U
U orb
m/s
m/s
m/s
uS
u
v
vb
V
V
vS
v
w
ws , 0
m/s
m/s
m/s
m/s
m/s
m/s
m/s
m/s
m/s
m/s
ws(
m/s
WL | Delft Hydraulics
xiii
31 December 2007
Final version 1.0
Symbol
X0356, M3470
Units
m
m
Meaning
Cartesian co-ordinates
bed roughness length
deg
s
m
m2/s4
m2/s
m2/s
m2/s
m2/s
deg.
J/sm2C
m
m
eo
deg
1/s
m2/s
m2/s
m2/s
m2/s
x, y, z ,
z0
c
t
x ( n ,m )
y ( n,m )
zb
zs
b
m2/s3
f
f,x,
f,y,
f,z
s,y,
s,z
s,x,
back
H
mol
V
2D
m /s
m2/s
3D
a
0
( )
s
m /s
kg/m3
kg/m3
density of water
density of air
kg/m3
kg/m
J/m s K
xiv
Stefan-Boltsmanns constant
WL | Delft Hydraulics
Symbol
b
b
b
c
w
m
b,c
b ,cr
b ,cw
b,w
cr
cr , d
cr , e
X0356, M3470
Units
-
Meaning
scaled vertical co-ordinate;
z
d
; (surface
0;
Prandtl-Schmidt number
kg/ms2
N/m2
shear stress
bed shear stress due to current and waves
kg/ms2
kg/ms
kg/ms
kg/ms2
kg/ms
1)
kg/ms2
kg/ms2
kg/ms
kg/ms
kg/ms2
kg/ms
kg/ms
kg/ms2
kg/ms2
kg/ms2
kg/ms2
kg/ms2
m/s
1/s
deg./hour
J/m2s
m
m
earth tide
eo
tidal loading
cw
max
mean
s
s
WL | Delft Hydraulics
31 December 2007
Final version 1.0
xv
X0356, M3470
31 December 2007
Final version 1.0
Introduction
This chapter refers to the model Delft3D-FLOW as a software product, and clarifies the
relation of that which is being validated to the rest of the software. It includes brief
descriptions of pre- and post-processing options, as well as an explanation of the modular
structure of the computational core of the model.
Delft3D-FLOW is the hydrodynamic module of Delft3D, which is Delft Hydraulics' fullyintegrated program for the modelling of water flows, waves, water quality, particle tracking,
ecology, sediment and chemical transports and morphology. In Figure 1.1 a system overview
of Delft3D is given.
Figure 1.1:
1.1
Model overview
1.1.1
Purpose
The primary purpose of the computational model Delft3D-FLOW is to solve various one-,
two- and three-dimensional, time-dependent, non-linear differential equations related to
hydrostatic and non-hydrostatic free-surface flow problems on a structured orthogonal grid
to cover problems with complicated geometry. The equations are formulated in orthogonal
curvilinear co-ordinates on a plane or in spherical co-ordinates on the globe. In Delft3DFLOW models with a rectangular or spherical grid (Cartesian frame of reference) are
considered as a special form of a curvilinear grid, see [Kernkamp et al., 2005] and
[Willemse et al., 1986].
WL | Delft Hydraulics
11
31 December 2007
Final version 1.0
X0356, M3470
The equations solved are mathematical descriptions of physical conservation laws for:
water volume (continuity equation),
linear momentum (Reynolds-averaged Navier-Stokes (RANS) equations), and
tracer mass (transport equation) , e. g. for salt, heat (temperature) and suspended
sediments or passive pollutants.
Furthermore, bed level changes are computed, which depend on the quantity of bottom
sediments.
The following physical quantities can be obtained in dependence on three-dimensional
space (x,y,z) and time t:
water surface elevation (x,y,t) with regard to a reference surface (e. g. mean sea level),
current velocity u(x,y,z,t), v(x,y,z,t), w(x,y,z,t),
non-hydrostatic pressure component q(x,y,z,t),
tracer concentration C(x,y,z,t), e. g. temperature, salinity, concentration of suspended
sediments or passive pollutants; and
bed level d(x,y,t),representing changes in bathymetry.
Delft3D-FLOW can be used in either hydrostatic or non-hydrostatic mode. In case of
hydrostatic modelling the so-called shallow water equations are solved, whereas in nonhydrostatic mode the Navier-Stokes equations are taken into account by adding nonhydrostatic terms to the shallow water equations. A fine horizontal grid is needed to resolve
non-hydostatic flow phenomena.
When the computational model Delft3D-FLOW is used in one- or two-dimensional mode
(with one z-layer in vertical direction) the results for u, v and C will be the respective depth
averaged values for current velocity and tracer concentration.
For the vertical grid system two options are available in Delft3D-FLOW, namely so-called
- or z-coordinates. For a detailed discussion of these two grid systems we refer to
Section 1.1.4. In the remainder of this document z will be used as vertical coordinate.
1.1.2
12
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
1.1.3
Horizontal grid
In Delft3D-FLOW the horizontal physical model domain (x,y) is covered with a curvilinear
orthogonal grid, designed and optimised for a given application through a grid generator.
This includes simple rectangular, spherical and curvilinear grids. The computations are
performed on a transformed, simple rectangular computational domain. For the horizontal
direction the grid concept is illustrated in Figure 1.2.
C e l l d r aw n by p l o t p ro g r a m s ,
R G F G R I D a nd Q U I C K I N
+
+
+
o
+
+
+
+
+
+
+
+
+
o
o
+
+
+ +
+
+
P o i n t s re q u i r e d b y
f o r b o u nd a r y d e f in it i o n/ s p e c .
+
+
+
o
+
+
+
+
+
o
De p t h p o i n t i
M o d e l b o u n d a r ie s
Wat e r l e ve l p o i n t i
o
o
o
o
+
+
+
+
+
+
+
+
o
o
o
o
o
o
o
o
o
o
+
+
+
+
+
+
+
+
+
+
+
+
+
+
o
o
o
o
o
o
o
o
+
+
+
+
+
+
Figure 1.2
1.1.4
I d e n t ic a l g r id i n d e x n u m b e r
Vertical grid
3D numerical modelling of the hydrodynamics and water quality in these areas requires
accurate treatment of the vertical exchange processes. The existence of vertical stratification
influences the turbulent exchange of momentum, heat, salinity and passive contaminants.
The accuracy of the discretisation of the vertical exchange processes is determined by the
vertical grid system. The vertical grid should:
WL | Delft Hydraulics
13
31 December 2007
Final version 1.0
X0356, M3470
resolve the boundary layer near the bottom and surface to allow an accurate evaluation
of the bed stress and surface stresses, respectively;
be fine around pycnoclines;
avoid large truncation errors in the approximation of strictly horizontal gradients.
The two commonly used vertical grid systems in 3D shallow-water models are the zcoordinate system (Z-model) and the so-called -coordinate system ( -model), see
Figure 1.3. Neither meets all the requirements. The Z-model has horizontal coordinate lines
which are (nearly) parallel with isopycnals, but the bottom is usually not a coordinate line
and is represented instead as a staircase (zig-zag boundary). This leads to inaccuracies in the
approximation of the bed stress and the horizontal advection near the bed. The sigma-model
has quasi-horizontal coordinate lines. The first and last grid line follow the free surface
( = 0) and the sea bed boundary ( = -1), respectively, with a user defined -distribution in
between. The grid lines follow the bottom topography and the surface but generally not the
isopycnals. Inaccuracies associated with these numerical artefacts have been addressed in
Delft3D, which has led to acceptable solutions for practical applications.
In Delft3D-FLOW both the options of fixed horizontal layers (Z-model) and the sigma grid
( -model) are operational. The two grid concepts are illustrated in Figure 1.3.
In practice, this means that depending on the application the user can choose the best option
for the representation of the processes in the vertical. In case of stratified flow problems in
coastal seas, estuaries and lakes where steep topography is a dominant feature, this is an
important issue. For lakes a Z-model is preferred, because the vertical exchange process
should not be dominated by truncation errors.
1.1.5
14
WL | Delft Hydraulics
X0356, M3470
RGFGRID
QUICKIN
Delft3D-FLOW GUI
TRIANA
TIDE
NESTHD
GPP
QUICKPLOT
Table 1.1
31 December 2007
Final version 1.0
For details on using these utility programs we refer to the respective User Manuals.
1.1.6
Version information
The content of this document is consistent with the (operational) version 3.55.04 of the
Delft3D-FLOW software, which has been released in November 2007 [WL | Delft
Hydraulics, 2007].
1.2
In this DELFT3D-FLOW validation document at first the model functionality of Delft3DFLOW is described by means of its applications (see Chapter 2.2.1) and its physical
processes (see Chapter 2.2.2). Next, the following three phases are distinguished:
Conceptual model (mathematical description of a physical system together with some
fundamental assumptions and/or simplifications), see Chapter 2.3.
Algorithmic implementation (conversion of the conceptual model into a set of
procedures for computation; e.g., discretisations, solution procedures), see Chapter 2.4.
Software implementation (conversion of algorithmic implementation into computer
code; coding of algorithms, data structures, etc.), see Chapter 2.5.
These three phases are according to the IAHR guidelines for validation, as described in
[IAHR, 1994], see also [Dee, 1993].
In Chapter 3 the claims and substantiations that have been formulated in Chapter 2 for the
model functionality, the conceptual model, the algorithmic implementation and the software
implementation of Delft3D-FLOW are validated for a large number of validation studies.
WL | Delft Hydraulics
15
31 December 2007
Final version 1.0
X0356, M3470
This Delft3D-FLOW validation document, the input and result files of the validation studies
(of Chapter 3) can be supplied to users of the Delft3D-FLOW system, so that they are able to
verify the validity and performance of Delft3D-FLOW.
1.3
This version hardly contains any quantification of model results. All validation studies have
been described and lot of figures are shown. By visual inspection and by reading the
conclusions for each validation study, the reader will have an impression of the quality of the
computed results. However, a quantitative assessment of the accuracy of model predictions is
lacking. In a next version of this validation document, a quantitative assessment will be
added.
For a quantitative assessment various options are available. Preferably, a uniform
performance indicator will be applied for all validation studies, possibly supported by
additional quantification parameters. An appropriate candidate might be the so-called Brier
Skill Score (BSS), which is an objective performance indicator. The BSS is defined as:
BSS =1
B X
2
2
(1.3.1)
1.4
Related documents
The validation studies, as described in Chapter 3 of this validation document, are also
available via the Wiki/Internet side of WL | Delft Hydraulics, see
http://wiki.wldelft.nl/display/DSC/Validation+document.
Further documents related to the current version of the computational model Delft3DFLOW can be found in the User Manual [WL | Delft Hydraulics, 2007].
1.5
Project team
This validation document has been prepared by Herman Gerritsen and Erik de Goede. The
figures and WL | Delft Hydraulics Wiki/Internet pages with model results of the validation
studies have been prepared by Frank Platzek, Menno Genseberger. Rob Uittenbogaard,
Firmijn Zijl, Daniel Twigt and Jan van Kester have provided valuable assistance.
16
WL | Delft Hydraulics
1.6
X0356, M3470
31 December 2007
Final version 1.0
WL | Delft Hydraulics
17
X0356, M3470
31 December 2007
Final version 1.0
Model validity
This chapter summarises all available information pertaining to the validation of the
computational core of the model. This includes the assumptions and approximations that
were introduced during the design and implementation of the model. It further includes
claims about the applicability and/or accuracy of (aspects of) the model, together with
statements about the substantiations of those claims.
The nature of a claim and its substantiation varies depending on the subject, as explained
below under the headings of the various subsections in which they appear. We have aimed to
make claims as explicit as possible and to provide useful information about model validity.
Substantiation aims to be thorough but brief, which can be achieved by using references.
Note that a substantiation may be incomplete, due to the nature of the claim, or because the
evidence is not (yet) available. In such cases we prefer to admit this rather than to invent a
substantiation that appears convincing.
These claims and substantiations together comprise the essential information in this
document. The remainder of the document serves either to provide context, necessary
background material or substantiating evidence.
2.1
Physical system
This section describes the physical system or systems being modelled. It describes what is
being modelled, rather than how it is being modelled. The hydrodynamic module Delft3DFLOW simulates two-dimensional (2D, depth-averaged) or three-dimensional (3D) unsteady
flow and transport phenomena resulting from tidal and meteorological forcing, including the
effect of density differences due to a non-uniform temperature and salinity distribution
(density-driven flow). The flow model can be used to predict the flow in shallow seas,
coastal areas, estuaries, lagoons, rivers and lakes. It aims to model flow phenomena of
which the horizontal length and time scales are significantly larger than the vertical scales,
which is the so-called shallow water assumption.
If the fluid is vertically homogeneous, a depth-averaged approach is appropriate. Delft3DFLOW is able to run in two-dimensional mode (one computational layer), which
corresponds to solving the depth-averaged equations. Examples in which the twodimensional, depth-averaged flow equations can be applied are tidal waves, storm surges,
tsunamis, harbour oscillations (seiches) and transport of pollutants in vertically well-mixed
flow regimes.
Three-dimensional modelling is of particular interest in transport problems where the
horizontal flow field shows significant variation in the vertical direction. This variation may
be generated by wind forcing, bed stress, Coriolis force, bed topography or density
differences. Examples are dispersion of passive materials or cooling water in lakes and
coastal areas, upwelling and downwelling of nutrients, salt intrusion in estuaries, fresh water
river discharges in bays and thermal stratification in lakes and seas.
WL | Delft Hydraulics
21
31 December 2007
Final version 1.0
2.2
X0356, M3470
Model functionality
This section describes the functionality of the model by referring to specific instances or
configurations of the physical system described in Section 2.1 above. It consists of claims
about what the model is actually able to represent, and (to the extent that this is possible)
how well it does so. For the purposes of this section the model can be regarded as a black
box, taking input information and producing computational results.
2.2.1
Applications
This section presents an overview of the domain of applicability of the model. This is done
by making claims about the types of practical and realistic situations in which the model can
be employed, and showing the nature and quality of the information that the model is
capable of generating in those situations.
The purpose of providing the reader with this inventory of application types is to allow the
reader to quickly recognise whether the model is indeed suitable for a particular application.
The computational model Delft3D-FLOW can be used in a wide range of applications,
which are listed below:
Tide and wind-driven flow resulting from space and time varying wind and atmospheric
pressure (See Section 2.2.1.1)
Density driven flow and salinity intrusion (See Section 2.2.1.2)
Wind driven flow (See Section 2.2.1.3)
Horizontal transport of matter on large and small scales (See Section 2.2.1.4).
Hydrodynamic impact of engineering works such as land reclamation, breakwaters,
dikes (See Section 2.2.1.5)
Hydrodynamic impact of hydraulic structures such as gates, weirs, barriers and floating
structures (See Section 2.2.1.6)
Spreading of waste water discharges from coastal outfalls (See Section 2.2.1.7)
Thermal recirculation of cooling water discharges from a power plant
(See Section 2.2.1.8)
Hydrostatic and non-hydrostatic flow (See Section 2.2.1.9).
Thermal stratification in seas, lakes and reservoirs (See Section 2.2.1.10).
Small scale current patterns near harbour entrances (See Section 2.2.1.11).
Flows resulting from dam breaks (See Section 2.2.1.12).
The model results have the form of distributions of the simulated quantities (water levels,
currents, salinity, temperature, pollutant concentrations) in all grid points at user specified
points in time, plus detailed time series of such parameters at user-selected locations. Each
application is described in more detail in subsequent sections.
22
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
WL | Delft Hydraulics
23
31 December 2007
Final version 1.0
X0356, M3470
24
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
WL | Delft Hydraulics
25
31 December 2007
Final version 1.0
X0356, M3470
2.2.2
Processes
This section further characterises the domain of applicability of the model. This is done by
making claims about the individual physical processes represented in Delft3D-FLOW. The
idea is to break down the physics into elements that are as simple as possible, yet still
meaningful.
The information contained in this section supplements that in the previous section. It is
intended to allow the reader to judge whether or not the model is suitable for his purpose, by
considering separately the individual processes that play a role in the application he has in
mind.
Delft3D-FLOW is able to represent a large number of processes, which are listed below:
Propagation of long waves (barotropic flow) (See Section 2.2.2.1)
Baroclinic flow, salinity, suspended sediment and temperature driven flow, including
prognostic or diagnostic modelling (See Section 2.2.2.2)
Transport of dissolved material and pollutants (See Section 2.2.2.3)
Transport of sediments, including erosion, sedimentation and bed load transport
(See Section 2.2.2.4)
Propagation of short waves (See Section 2.2.2.5)
Subcritical and supercritical flow (See Section 2.2.2.6)
Steady and unsteady (time varying) flow (See Section 2.2.2.7)
Drying and flooding of intertidal flats (See Section 2.2.2.8)
The effect of the Earth's rotation (Coriolis force) (See Section 2.2.2.9)
26
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
For long waves (in shallow water) the vertical acceleration can be
assumed to be negligible and the pressure to be hydrostatic. Under these
assumptions the celerity of the wave only depends on gravity and water
depth. It is also independent of the wave length. Relevant examples of
long waves are tidal waves. The free surface gradients represent so-called
barotropic flow. A special surface gradients generated by of long waves
are the surges along coasts generated by wind (storm surge modelling).
Claim 2.2.2.1:
WL | Delft Hydraulics
27
31 December 2007
Final version 1.0
X0356, M3470
Claim 2.2.2.2:
Substantiation: Validation Study 3.1.5 (Lock exchange flow with hydrostatic pressure or
non-hydrostatic pressure).
Validation Study 3.2.1 (Salt intrusion in laboratory flume)
Validation Study 3.2.6 (Salt intrusion in 3D Numerical Scale Model with
-coordinates of an estuary)
Validation Study 3.2.7 (3D Numerical Scale Model with -coordinates
and HLES for the complex exchange flow between a tidal dock and the
estuary)
Validation Study 3.4.1 (3D -model of the North Sea)
Validation Study 3.4.3 (3D Z-model of Lake Grevelingen)
Validation Study 3.4.4 (3D Z-model of Sea of Marmara)
Validation Study 3.4.5 (3D -model of the South China Sea)
Claim 2.2.2.3:
28
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
Claim 2.2.2.4:
For short waves the vertical acceleration of the fluid can no longer be
neglected and the pressure is non-hydrostatic. The celerity of the wave
then depends on gravity, water depth as well as wave length.
Claim 2.2.2.5:
Claim 2.2.2.6:
WL | Delft Hydraulics
29
31 December 2007
Final version 1.0
Reference:
X0356, M3470
Claim 2.2.2.7:
In estuaries and coastal seas with significant tidal range quite often vast
areas of land (tidal fiats) are subsequently covered and uncovered with
water during each tidal cycle.
Claim 2.2.2.8:
Claim 2.2.2.9:
Delft3D-FLOW can take into account the impact of the Coriolis force
arising from the rotation of the earth.
210
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
Claim 2.2.2.11: Delft3D-FLOW can take into account time varying sources and sinks for
e.g. river flows and discharges from outfalls.
Substantiation: Validation Study 3.4.1 (3D North Sea)
Example(s) of application studies:
1. Spreading of the Rhine plume along the Dutch Coast [De Kok et al., 2001]
Impact of space and time varying wind shear stress at the water
surface
WL | Delft Hydraulics
211
31 December 2007
Final version 1.0
Description:
X0356, M3470
Wind causes a wind stress at the free surface, which can yield a
considerable set-up in the water elevation. Storm surge modelling is also
an example of space and time varying wind stresses.
Claim 2.2.2.12: Delft3D-FLOW can accurately simulate the effects of space and time
varying wind stresses at the free water surface.
Substantiation: Validation Study 3.1.4 (Wind driven flow);
Validation Study 3.4.3 (Sea of Marmara).
Example(s) of application studies:
1. Generation and dissipation of storm surges in the North Sea [Bijlsma, 1988]
2. Typhoon driven storm surges in the South China Sea [Kernkamp e.a., 2004]
Claim 2.2.2.13: Delft3D-FLOW can take into account the impact of the space varying
shear stress at the bottom. Several formulations are possible (Chezy,
Manning, White-Colebrook or roughness height Z0). The input
coefficients for bottom friction can vary in space.
Substantiation: Validation Study 3.1.6 (Wave force and a mass flux in a closed basin).
Claim 2.2.2.14: Based on an accurate space and time varying wind and pressure fields as
input, Delft3D-FLOW can compute accurate storm surges.
Substantiation: Validation Study 3.4.3 (Sea of Marmara)
212
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
The heat radiation emitted by the sun reaches the earth in the form of
electromagnetic waves with wavelengths in the range of 0.15 to 4 m. In
the atmosphere the radiation undergoes scattering, reflection and
absorption by air, cloud, dust and particles. On average neither the
atmosphere nor the earth accumulates heat, which implies that the
absorbed heat is emitted back again. The wavelengths of these emitted
radiations are longer (between 4 and 50 m) due to the lower prevailing
temperature in the atmosphere and on Earth.
Claim 2.2.2.15: In Delft3D-FLOW the heat exchange at the free surface is accurately
modelled by taking into account the separate effects of solar (short wave)
and atmospheric (long wave) radiation, and heat loss due to back
radiation, evaporation and convection.
Substantiation: Validation Study 3.4.1 (North Sea Model, 3D -model);
Validation Study 3.4.3 (Zegerplas, 3D Z-model and 3D -model);
Validation Study 3.4.3 (Sea of Marmara, 3D Z-model);
Validation Study 3.4.4 (Lake Grevelingen, 3D Z-model);
Validation Study 3.4.5 (South China Sea, 3D -model)
Reference(s):
Claim 2.2.2.16: Delft3D-FLOW can take into account wave and wave-induced effects by
coupling with wave models. In Delft3D-FLOW coupling with two wave
models are operational: HISWA, a second generation wave model
(Holthuijsen, 1998) and SWAN, a third generation wave model
(Ris, 1997).
WL | Delft Hydraulics
213
31 December 2007
Final version 1.0
X0356, M3470
Substantiation: Validation Study 3.1.6 (Wave force and a mass flux in a closed basin).
Reference:
Claim 2.2.2.17: Delft3D-FLOW can take into account the impact of secondary flow in a
depth-averaged model. This is done by adding a term in the momentum
equations.
Substantiation: Not in list of validation studies yet.
Claim 2.2.2.18: Delft3D-FLOW can take into account tide generating forces when
simulating very large scale water systems.
Substantiation: Not in list of validation studies yet.
Example(s) of application studies:
1. Tidal model validation of the seas of South-East Asia [Gerritsen et al., 2003]
214
WL | Delft Hydraulics
2.3
X0356, M3470
31 December 2007
Final version 1.0
Conceptual model
This section describes technical aspects of the conceptual model that are relevant to the
validation process. In particular, it addresses the simplifications and assumptions of the
conceptual model compared to the actual physical world.
2.3.1
Governing equations
u
u
u
+u
+v +
x
y d+
t
v
v
v
+u + v +
y d+
t
x
=-
1
fv = - Pu + Fu +
1
+ fu = - Pv +Fv +
in the
1
2
(d + )
1
(d + )2
H (qin
v
V
qout ) P E
(2.3.1.1)
(2.3.1.2)
(2.3.1.3)
[(d + )U]
[(d + )V]
+
=Q
x
y
(2.3.1.4)
WL | Delft Hydraulics
215
31 December 2007
Final version 1.0
X0356, M3470
H
+
+v
x
x
w= + u
H
+
+
y
y
H
+
t
t
In Eq. (2.3.1.4), Q represents the contributions per unit area due to the discharge or
withdrawal of water, precipitation and evaporation:
0
1
qin qout d
P E,
In Eqs. (2.3.1.1-4) u(x,y, ,t,), v(x,y, ,t) and (x,y, ,t) are the velocity components in the
horizontal x, y and vertical -directions, respectively; (x,y) is the water level above a
reference plane; d(x,y) is the depth below this plane; H(x,y)=d(x,y)+ (x,y) is the total water
depth, t is the time; f is the Coriolis parameter; g is the gravitational acceleration and v is
the vertical eddy viscosity coefficient. Furthermore qin and qout are the local sources and
sinks of water per unit of volume (1/s), respectively, P represents the non-local source term
of precipitation and E non-local sink term due to evaporation. We remark that the intake of,
for example, a power plant is a withdrawal of water and should be modelled as a sink. At the
free surface there may be a source due to precipitation or a sink due to evaporation.
From now, it is assumed that v can be prescribed as a non-negative function of space and
time (often a constant) and that no specific closure model is required. In later paragraphs the
turbulence closure models in Delft3D-FLOW are described, which determine v. The
horizontal pressure terms Pu and Pv are described in (2.3.1.8), the horizontal viscosity terms
Fu and Fv in (2.3.1.9). Equations (2.3.1.1) and (2.3.1.2) are the momentum equations,
whereas the (depth-integrated) continuity equation is described in (2.3.1.4). The depth
averaged velocities read
0
U= ud
-1
V= vd
-1
z+d
(2.3.1.5)
the so-called -coordinate is used, introduced in (Phillips, 1957). At the bed =-1 and at the
free surface =0. The -coordinate system is boundary fitted in the vertical plane.
By integration of the equations (2.3.1.1)-(2.3.1.2) from the bottom to the free surface we
arrive at the 2D (depth averaged) momentum equations, which are described in Eq. 2.3.1.25.
We remark that the continuity equation (2.3.1.4) already uses depth-averaged velocities.
Barotropic and baroclinic effects
The hydrostatic pressure is given by:
O
P = P atm + gH
216
(x, y, ,t)d
(2.3.1.6)
WL | Delft Hydraulics
X0356, M3470
Pu = g
Pv = g
31 December 2007
Final version 1.0
(2.3.1.7)
In case of a non-uniform density, Leibnitz' rule is used to obtain the following expressions
for the horizontal pressure gradients:
Pu = g
+ g
Pv = g
+ g
z FGH
z FGH
0
d+
d+
0
IJ d
K
IJ d
K
(2.3.1.8)
with 0 the ambient density of water. In (2.3.1.8) the first term in the right-hand side
represents the barotropic effect, whereas the second term describes the baroclinic influence.
In (2.3.1.8) the density variations are neglected except in the buoyancy term. This is known
as the Boussinesq approximation.
Horizontal viscosity
The forces Fu and Fv in the momentum equations (2.3.1.1-2) represent the horizontal
Reynolds stresses. In Delft3D-FLOW, however, for Fu and Fv a simplified version of the
above-described formulations is applied, see [Mellor and Blumberg, 1985]. The formulation
reads
Fu =
Fv =
FG
H
FG
H
x
2
IJ
K
vI
J
y K
u
2
y
(2.3.1.9)
in which the gradients are taken along -planes (planes of constant -value).
WL | Delft Hydraulics
217
31 December 2007
Final version 1.0
X0356, M3470
of the pressure has to be resolved in order to get physically realistic flow patterns. Delft3DFLOW incorporates such a non-hydrostatic model. Note that the non-hydrostatic
functionality is only available for the z-coordinate model and not for the grid system (see
Section 1.1.4 for a description of the vertical grid systems).
Non-hydrostatic equations
In case of a non-hydrostatic model, the momentum equations of the Reynolds-averaged
Navier-Stokes equations may be written as:
1
fv = - Pu + Fu
u
u
u
u
+u +v + w
x
y
z
t
v
v
v
v
+ u + v + w + fu =
y
t
x
z
w
w
w
w
+ u + v +w
y
t
x
z
Pv +Fv
Pw +Fw
q
1
+
x (d + )2 z
q
y
q
z
1
2
(d + ) z
1
2
(d + ) z
u
z
(2.3.1.10)
u
z
(2.3.1.11)
w
z
(2.3.1.12)
with
Fw =
FG
H
w
2
y
IJ
K
(2.3.1.13)
The pressure terms have been split into a hydrostatic part (see Eq. (2.3.1.6)) and a nonhydrostatic part q. The non-hydrostatic is computed via a so-called pressure correction step.
This requires the solution of a Poisson-type equation. The continuity equation is given by
Eq. (2.3.1.4), which is valid for the -model as well.
[(d + )C]
[(d + ) u C]
[(d + ) v C]
( C)
+
+
+
=
t
x
y
x
with
218
D h (d + )
d
C
C
+
D h (d + )
x
y
y
1
d+
Dv
(2.3.1.14)
C S
representing the first-order decay process. S represents the source and sink terms
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
per unit area due to the discharge qin or withdrawal qout of water and the exchange of heat
through the free surface Qtot .:
qinCin
qout C
Qtot .
(2.3.1.15)
and eddy diffusivity coefficient DH are much larger than the vertical
SGS
back
H
simulates the larger scale horizontal turbulent motions through a methodology called
Horizontal Large Eddy Simulation (HLES). The associated horizontal viscosity coefficient
SGS will then be computed by a dedicated SGS-turbulence model. For details of this
approach, see Section Turbulence modelling in horizontal direction.
The 3D part
3D
WL | Delft Hydraulics
SGS
3D
back
H
back
H
. Consequently, in Delft3D-
is defined by
(2.3.1.17)
219
31 December 2007
Final version 1.0
X0356, M3470
For turbulence closure models responding to shear production only, it may be convenient to
specify a background or ambient vertical mixing coefficient in order to account for all
other forms of unresolved mixing. Therefore, in addition to all turbulence closure models in
back
H
may be
specified by you, which is a background value for the vertical eddy viscosity in the
momentum equations (2.3.1.1) and (2.3.1.2). Consequently, the vertical eddy viscosity
coefficient V is defined by:
V
with
mol
max (
mol
back
V
3D
),
(2.3.1.18)
combinations of horizontal and vertical eddy viscosity are optional, Table 2.1 below
presents an overview.
model description
back
H
SGS
2D, no HLES
computed
by HLES
-
3D, no HLES
Table 2.1
computed
by HLES
back
V
3D
2D-turbulence +
dispersion coefficient
3D-turbulence +
dispersion coefficient
2D-turbulence
computed
by
vertical turbulence
model.
computed
by
vertical turbulence
model.
background
vertical viscosity
background
vertical viscosity
Remarks:
1. We note that the background horizontal eddy viscosity represents a series of
complicated hydrodynamic phenomena. Table 2.1 shows that this background horizontal
eddy viscosity
back
H
back
H
GUI in addition to the Elder formulation for 3D-turbulence and dispersion for 2D
simulations.
3. It is important to emphasize that the under limit of vertical eddy viscosity
back
V
is only
used in the momentum equations (see Eqs. 2.3.1.1 and 2.3.1.2) and is not used in the
vertical turbulence models (see for example Eqs. (2.3.20a-b).
220
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
3D
D3D :
1. Constant coefficient;
2. Algebraic eddy viscosity closure model;
3. k L turbulence closure model (1-equation model);
4. k
turbulence closure model (2-equation model).
The turbulence closure models differ in their prescription of the turbulent kinetic energy k,
the dissipation rate of turbulent kinetic energy , and/or the mixing length L.
The first turbulence closure model is the simplest closure based on a constant value which
has to be specified by the user. We remark that a constant eddy viscosity will lead to
parabolic vertical velocity profiles (as in laminar flow). The other three turbulence closure
models are based on the so-called eddy viscosity concept of Kolmogorov and Prandtl
[Kolmogorov, 1942; Prandtl, 1945]. The eddy viscosity is related to a characteristic length
scale and velocity scale. The eddy viscosity has the following form:
V
c' L k ,
(2.3.1.19)
in which c ' is a constant determined by calibration, derived from the empirical constant c
model; c '
in the k
c1/ 4 , c
k- turbulence model
In the k- model, which is a second-order turbulence closure model, the turbulent energy k and
dissipation are calculated by the following transport equations:
k
k
k
+u +v +
t
x
y d+
+u
+v
d+
1
(d + )2
1
(d + )2
+ Pk + Bk -
(2.3.1.20a)
+ P + B - c2
(2.3.1.20b)
For a detailed description of the turbulence models in Delft3D-FLOW, we refer to (WL | Delft
Hydraulics, 2007).
Turbulence modelling in horizontal direction
Two types of turbulence closure models have been implemented to determine the horizontal
viscosity coefficient
2D
1. A constant or space-varying coefficient, which has to be specified by the user via the
Delft3D-FLOW Graphical User Interface.
2. A subgrid-scale closure or Horizontal Large Eddy Simulation (HLES) model, which
computes the horizontal viscosity coefficient and the eddy diffusivity coefficient.
WL | Delft Hydraulics
221
31 December 2007
Final version 1.0
X0356, M3470
generally also needed to damp small-scale noise introduced by the advection terms. They
must be chosen dependent on the grid size. Usually, the horizontal coefficients are an order
of magnitude larger than the vertical coefficients determined by the turbulence closure
model in vertical direction.
Below we summarise the essential aspects of the sub-grid scale model for Horizontal Large
Eddy Simulation (HLES) of shallow water flows subjected to bed friction. The result of this
HLES model is a formulation of the horizontal component of the sub-grid eddy viscosity
and the sub-grid eddy diffusivity.
Horizontal Large Eddy Simulation (HLES) model
The HLES model is based on theoretical considerations presented by [Uittenbogaard, 1998]
and summarised in [Uittenbogaard and Van Vossen, 2003]. This model for the sub-grid eddy
viscosity xx includes the damping of sub-grid eddies by bed friction reads:
1
k s2
xx
S*
B2
(2.3.1.21)
with:
3g U
4H C 2
(2.3.1.22)
where C is the Chzy coefficient and H is the total water depth. Correspondingly, the sub-grid
eddy diffusivity xx for horizontal mixing of heat, salt, mud and dissolved constituents reads:
xx
xx
In Eq. (2.3.1.21) the sum of strain rates squared (i.e. double contraction of horizontal
component of the strain rate tensor) reads:
* 2
u*
x
v*
y
u*
y
v*
x
u* v *
.
y x
(2.3.1.23)
In Eq. (2.3.1.23) the superscript (..)* denotes fluctuating flow variables and these are defined
through the following recursive high-pass filter operator [Bendat and Piersol, 1971]:
t
*
n 1
n 1
(2.3.1.24)
with:
t
n 1
222
1 a
n 1
t
n
t
0
0 and a
exp
t/
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
in which t is the time step and is the relaxation time. In Eq. (2.3.1.24) represents any of
the velocity components (u,v). For a more detailed description of the HLES model we refer
to the User Manual of Delft3D-FLOW [WL | Delft Hydraulics, 2007].
U
U
U
1
+U
+V
+ fV = - P u +
t
x
y
- bx
+ Fu
(d + )
sx
- by
+ Fv
(d + )
V
V
V
1
+U
+V
- fU = - P v +
t
x
y
sy
(2.3.1.25)
gU U 2 + V 2
2
C 2D
by
cfU U 2 +V 2 ,
gV U 2 + V 2
cfV U 2 +V 2
2
C 2D
and the horizontal viscosity terms analoguous to Eq. (2.3.1.9). The depth-averaged continuity
equation has already been described by (2.3.1.4) and does not depend on the dimensionality
in the vertical direction, since it only contains the depth-averaged velocities U and V.
c( )
t
uc (
x
x
( )
s,x
vc (
y
c
w ws(
c(
( )
( )
s, y
( )
( )
s,z
(2.3.1.26)
( )
0,
The local flow velocities and eddy diffusivities are based on the results of the hydrodynamic
computations. The three-dimensional transport of sediment is computed in exactly the same
way as the transport of any other conservative constituent, such as salinity, heat, and
constituents. Important additional processes of sediment over other constituents are, for
example, the exchange of sediment between the bed and the flow, and the settling velocities
ws( ) of sediment under the action of gravity. The sediment transport formulation allows the
combined use of cohesive and non-cohesive sediment. Other processes such as the effect
that sediment has on the local mixture density, and hence on turbulence damping, can also
be taken into account. In addition, if a net flux of sediment from the bed to the flow, or vice
versa, occurs then the resulting change in the bathymetry should influence subsequent
hydrodynamic calculations. The formulation of several of these processes is sediment-type
specific. In particular, this applies for sand and mud. In this section we discuss some of the
differences, in general terms only.
WL | Delft Hydraulics
223
31 December 2007
Final version 1.0
X0356, M3470
The full 3D sediment transport is computed without resorting to the use of equilibrium
sediment concentration profiles. The variation of sediment transport in regions of
accelerating and decelerating flow is accurately accounted for. The effects of density
currents, stratification, and spiral flow are automatically taken into account without
requiring any special parametric formulations. Typical time scales of the intended
applications may range from those used in hydrodynamic simulations to some orders of
magnitude longer, i.e. from hours to many years. Long morphological simulations are
achieved by using a morphological time scale factor. Typical length scales can range from
those of near field morphology such as local scour near the head of breakwaters to those of
tidal inlets and estuaries and coastal areas.
For the sediment fractions one can select any mixture of sand and mud. Different
formulas are used to calculate the fall velocity, erosion and sedimentation of each type of
sediment. One can specify up to five sediment fractions in a single simulation. This means
one can execute a simulation with, for example, more than one sand fraction, or a number
of sand and mud sediment fractions present at any location. The simultaneous presence
of multiple sediment fractions has implications to the computation of the local hindered fall
velocity of any one sediment fraction as well as for the resulting mixture density.
In its standard form Delft3D-FLOW uses an empirical relation (Eckart, 1958) to adjust the
density of water in order to take into account varying temperature and salinity. For sediment
transport this relation is extended to include the density effect of sediment fractions in the
fluid mixture. This is achieved by adding (per unit volume) the mass of all sediment
fractions, and subtracting the mass of the displaced water. As a mathematical statement this
translates as:
lsed
mix
S , c(
)
w
c(
S
1
S
( )
s
(2.3.1.27)
Horizontal density gradients (now also due to differences in sediment concentrations) can
create density currents. Vertical density gradients can also have a significant effect on the
amount of vertical turbulent mixing present.
The settling velocities ws(
mixtures, the settling velocity of a single particle is reduced due to the presence of other
particles. In order to account for this hindered settling effect we follow Richardson and Zaki
(1954) and determine the settling velocity in a fluid-sediment mixture as a function of the
sediment concentration and the non-hindered settling velocity:
( )
s
cstot
1
CSOIL
ws( ,0) ,
(2.3.1.28)
in which CSOIL is the reference density (input parameter), ws ,0 is the 'basic' sediment
fraction specific settling velocity. The total mass concentration cmtot is the sum of the mass
concentrations of the sediment fractions and is equal to
224
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
lsed
c( ) .
cstot
(2.3.1.29)
2.3.2
The following assumptions and approximations that have been introduced into the
formulation of the conceptual model equations (2.3.1.1)-(2.3.1.29):
1. The continuous medium assumption.
2. The fluid (water) is assumed to be incompressible.
3. Reynolds averaging use of equations which have been averaged over the time and
length scales of turbulent fluctuations.
4. The shallow water assumption, which means that the depth is small compared to
horizontal length scales and that vertical accelerations are assumed to be negligibly small
compared to the gravitational acceleration. The vertical momentum equation is then
reduced to the hydrostatic pressure relation.
5. The effect of variable density is only taken into account in the horizontal pressure gradient
term (Boussinesq approximation).
6. The Eddy viscosity concept, in which the Reynolds-stresses are parameterised by the
product of a so called eddy-viscosity with the spatial gradient of the mean quantities.
7. In a Cartesian frame of reference (model domains smaller than ~ 500 by 500 km2) and
applications at moderate latitude (| |> 40 ) the Coriolis parameter may be assumed to be
uniform and the effect of the Earth's curvature is not taken into account.
8. The horizontal turbulent stresses at the surface and sea bed (wind and bottom friction)
are parameterised as quadratic friction terms, representing momentum transfer (drag) from
the atmosphere and momentum removal, respectively.
9. The gravitational acceleration is taken as uniform.
10. The immediate effect of buoyancy on vertical acceleration is not considered in the
hydrostatic version. In Delft3D-FLOW density differences are taken into account in the
horizontal pressure gradients and in the vertical exchange coefficients. In order to take
into account such effects, the non-hydrostatic option of Delft3D-FLOW should be
switched on. However, real-world applications with Delft3D-FLOW lead to a
computational effort that are not feasible yet, which is due the small horizontal mesh sizes
(namely in the order of the vertical mesh size) that are required in such applications..
11. The part of the horizontal eddy viscosity due to 2D turbulence is flow independent and
must be specified by the user, unless the HLES approach is used.
12. The curvature of the grid in both the horizontal and the vertical plane is neglected for the
horizontal turbulent stresses.
2.3.3
This section serves to account for the choices that were made in formulating the conceptual
model, that is, the assumptions and approximations listed in Section 2.3.2, and to explain the
implications of those choices for applicability and/or accuracy.
WL | Delft Hydraulics
225
31 December 2007
Final version 1.0
X0356, M3470
Substantiation: See [Malvern, 1969]. All validation studies of Chapter 3 are relevant to
this claim.
Incompressibility
If a fluid is treated as incompressible the density of the fluid will depend on temperature and
the concentration of dissolved substances but not on pressure. Under the assumption that a
fluid element (volume) does neither exchange heat nor dissolved substances with its
surroundings (isentropic deformation) the fluid can be regarded as incompressible
[Batchelor, 1967] if the:
1. fluid particle velocity is much smaller than the speed of sound, and the
2. phase speed of the disturbances (e. g. speed of the free surface waves) is also small
compared to the speed of sound, and the
3. vertical scale of motion (water depth) must be small compared to a mean value of
/(d /dz).
Claim 2.3.3.2:
Substantiation: See [Batchelor, 1967]. All validation studies of Chapter 3 are relevant to
this claim.
Reynolds averaging
For almost all practical applications it is impossible to solve for the detailed (turbulent)
velocities in rivers, estuaries, coastal seas and other similar water bodies. Around 1880
Reynolds suggested the first time, to split the variables u, v, w and p into a mean and a
fluctuating component. After insertion into the Navier-Stokes equations and subsequent
time- averaging we obtain the Reynolds-Averaged Navier-Stokes equations (RANS). They
contain new terms, the so-called Reynolds-stresses, which describe the mean momentum
transfer due to turbulence. The solutions to these equations can be obtained by making
226
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
special assumptions about the nature of these Reynolds stress terms (turbulence closure), see
also Claim 2.3.3.4 below.
Claim 2.3.3.3:
Boussinesq approximation
In natural waters variations of density (
reference density
WL | Delft Hydraulics
227
31 December 2007
Final version 1.0
Claim 2.3.3.5:
X0356, M3470
Substantiation: The variations in momentum changes due to mass variations are at most 3
- 4 % when dealing with natural water bodies. Therefore, it is a valid
approach to apply the Boussinesq approximation.
Eddy-viscosity concept
In the Boussinesq eddy-viscosity concept the Reynolds-stresses are parameterised by the
product of a so called eddy-viscosity with the spatial gradient of the mean quantities. For
simplicity the resulting eddy-viscosity tensor is assumed to be diagonal. The eddy-viscosity
tensor depends on the state of the turbulent motion and has therefore to be determined by a
separate turbulence model. Note that in Section 2.3.1 the turbulence models (both in
horizontal and vertical direction) of Delft3D-FLOW have been described.
Claim 2.3.3.6a: The conceptual description of the turbulence models in Delft3D-FLOW
(e.g., the eddy-viscosity concept, k- model, HLES model, etc.) is valid
for tidal flows in rivers, estuaries and coastal seas.
Substantiation: Validation Study 3.2.3 (Vertical mixing layer); See also [Abbott, 1997] for
a description of the ranges of common turbulence models.
Claim 2.3.3.6b: Delft3D-FLOW can also compute sub-grid scale or averaged 2Dturbulence. The computation is performed by the so-called Horizontal
Large Eddy Simulation (HLES) approach.
Substantiation: Validation Study 3.2.5 (Horizontal plus vertical mixing layer).
Coriolis
The Coriolis term in the equation represents the influence on the flow exerted by the earths
rotation. The Coriolis parameter f depends on the geographic latitude and the angular speed
of rotation of the earth, : f 2 sin .
228
WL | Delft Hydraulics
X0356, M3470
Claim 2.3.3.7:
31 December 2007
Final version 1.0
A space varying Coriolis parameter is a standard option for all Delft3DFLOW applications. At moderate latitudes (| |> 40 ) and for relative
small model dimensions (< ~500 by ~500 km) the Coriolis variation can
be assumed constant. For larger model areas, or model areas close to the
equator, full variation in f can be accounted for in Delft3D-FLOW.
cos
(2.3.3.4)
sin
(2.3.3.5)
where
is the angle between the wind stress vector and the local coordinate direction.
Without wind the free surface stress is zero. The magnitude of the wind shear-stress is
defined as:
s
(2.3.3.6)
The magnitude is often determined by the following widely used quadratic expression:
s
CdU102
(2.3.3.7)
in which U10 is the wind at 10 metre height. Delft3D-FLOW offers the possibility to
prescribe either uniform or space varying wind. The latter is generally applied in
combination with space and time varying atmospheric pressure.
At the sea bed, the boundary conditions for the momentum equations are:
V
H
with
=
=-1
bx
and
by
bx
by
(2.3.3.8)
=-1
the bed stresses in x- and y-direction, respectively. The bed stress may be the
combined effect of flow and waves. In this section we restrict ourselves to the resistance due
to flow only.
Claim 2.3.3.8:
The parameterisations as described above are valid for free surface flows.
WL | Delft Hydraulics
229
31 December 2007
Final version 1.0
X0356, M3470
Gravity
The gravitational acceleration g is not a constant but varies from place to place, which
depends on the geographic latitude . In Delft3D-FLOW the gravitational acceleration g
can be specified by the user (via the input file). The default value is 9.81.
Claim 2.3.3.9:
Substantiation: This assumption is valid for relatively small model domains (and e.g. not
necessarily true for ocean models). Furthermore, the user can specify the
gravitational acceleration g via the GUI of Delft3D-FLOW. This means
that values in the order of 9.78 near the equator to 9.83 near the poles can
be applied.
2.4
Algorithmic implementation
This section describes the technical aspects of the algorithmic implementation (or
discretised version) of the conceptual model that are relevant to the validation process. This
section contains claims about the validity of the algorithmic implementation, and statements
about the substantiation of those claims. Several items with respect to the algorithmic
implementation are discussed.
The Alternating Direction Implicit solution technique used in Delft3D-FLOW has as
consequence that the Algorithmic Implementation (the discretisation of the model equations)
and the Software Implementation (the time stepping solution technique) are strongly linked.
We have chosen to treat these issues together in this section, and discuss the claims and
substantiations accordingly, instead of describing all claims and substantiations in separate
sections.
2.4.1
General
Following [Stelling, 1984], a robust space and time discretisation scheme for the shallow
water equations has to satisfy the following demands:
Robustness (unconditionally stable).
Accuracy (at least second-order).
Suitable for both time-dependent and steady state problems.
Computationally efficient.
Claim 2.4.1:
230
WL | Delft Hydraulics
2.4.2
X0356, M3470
31 December 2007
Final version 1.0
Computational grid
Figure 2.1
Legend:
+
water elevation ( ) / density (
velocity point(u, v, or w)
The arrangement of the flow variables presented in Figure 2.1 is called a staggered grid.
This particular arrangement of the variables is called the Arakawa C-grid. The water level
points (pressure points) are defined in the centre of a (continuity) cell. The velocity
components are perpendicular to the grid cell faces where they are situated.
In the horizontal plane Delft3D-FLOW offers the opportunity to use
orthogonal curvilinear co-ordinates ( , ), with Cartesian rectangular co-ordinates
(x,y) as a special case;
orthogonal (rectangular) latitude-longitude co-ordinates ( , ) on the sphere.
orthogonal curvilinear latitude-longitude co-ordinates ( , ) on the sphere, see
[Kernkamp et al., 2005].
Curvilinear co-ordinates allow smooth alignment of the grid with land boundaries and allow
local grid refinement in areas with large horizontal gradients.
Claim 2.4.2:
In Delft3D-FLOW the user can choose a type of grid that is suitable for
the application involved. Furthermore, using a domain decomposition
technique (not described in this document, but in Appendix B.13 of the
user manual) additional high horizontal and vertical grid resolution can be
realised in regions where it is specifically needed.
WL | Delft Hydraulics
231
31 December 2007
Final version 1.0
2.4.3
X0356, M3470
The discretised equations describe the relations between the flow values in the points of the
grid and between one time step and the next. A time stepping scheme is applied to determine
the evolution of the grid values in time. In Delft3D-FLOW a stable combination of secondorder central and third order upwind spatial discretisation is used, plus a so-called ADI-type
time stepping scheme to solve the discretised equations in time. This offers a combination of
accuracy, stability and acceptable computation times. The two-step Alternating Direction
Implicit solution concept reads, still formulated using differential operators for spatial
discretisations:
1
2
1
2
U
t
1
2
U
t
1
2
with U
1
Ax U
2
u , v,
1
Ax U
2
1
Ay U
2
f
u
0
H
1
2
BU
1
1
2
BU
d,
1
(2.4.1)
d,
and
0
Ax
1
Ay U
2
1
2
x
0
g
0
y
u
x
,
Ay
f
0
0
g
0
H
0 0
B
0
0
0 ,
0 0
with the linearised bottom friction coefficient. The essence of implicit techniques is that
the equations contain more than one variable at the unknown new time step, and can
therefore not be solved separately or explicitly. The linking of the variables at the new time
step increases accuracy and stability, but results in a large matrix equation, which has to be
solved. To improve stability the bottom friction is integrated implicitly for each stage. d is
the right-hand side containing external forcings like wind and atmospheric pressure.
In the first stage the time level proceeds from
t to t
1
2
to
1
2
followed by the U-momentum equation, which is implicitly coupled with the continuity
equation by the free surface gradient. In the second stage the time level proceeds from
to
1
2
1 . In this stage first the U-momentum equation is solved explicitly, followed by the V-
232
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
momentum equation which is implicitly coupled with the continuity equation by the free
surface gradient. In the (x- or y-)direction in which the barotropic pressure term (i.e. water
level gradient) is integrated implicitly, the horizontal advection terms and viscosity terms
are integrated explicitly. Similarly in the direction in which the barotropic pressure term is
integrated explicitly, the advection terms and viscosity terms are integrated in an implicit
way.
The second stage in the ADI-method is almost similar to the first stage. The grid
coefficients, direction dependent roughness coefficients and the u-velocity and v-velocity
are interchanged. The only principal difference between the u- and v-momentum equations
is the sign of the Coriolis term.
The above-described ADI solution technique of the shallow water equations avoids the
strong numerical stability restrictions of an explicit time integration. The latter would be
subject to a time step condition based on the Courant number for wave propagation (on a
rectangular grid):
CFLwave
2 t gH
1
x2
1
y2
1,
(2.4.2)
For many practical applications this would require a time step of only a few seconds to
simulate tidal propagation. Exceeding the time step would generate an instability and from
the view of robustness this is not acceptable.
T
40
Cf
2 t gH
tu
x
Table 2.2
WL | Delft Hydraulics
1
y2
gH
1
x2
1
y2
1
x2
1
y2
2 t
Stability of the explicit algorithm
for flooding
1
x2
4 2
233
31 December 2007
Final version 1.0
X0356, M3470
From Table 2.2 it can be seen that time step limitations occur in Delft3D-FLOW, namely for
the free surface wave propagation, baroclinic flow, the advection scheme(s), drying and
flooding and the horizontal viscosity. Which of the limitations is most restrictive is
dependent on the kind of application: length scale, velocity scale, with or without densitycoupling, etc. Note that for the vertical terms there is no (additional) stability criterion. This
means that the Delft3D-FLOW when applied in 3D mode is similarly robust compared to
the application in 2D. We remark, however, that the horizontal velocities in 3D are (slightly)
larger than in 2D, which might result in a (slightly) smaller time step for 3D applications.
In general, one would not expect a time step limitation for the barotropic mode. However,
the ADI-method may lead to inaccurately predicted flow patterns when a relatively large
time step is applied, see e.g. [Stelling, 1984]. This inaccuracy is called the ADI-effect and is
introduced by the splitting of the spatial operator in two directions. The upper bound for the
barotropic mode of 4 2 occurs in the most critical ADI situation, namely in case of a
narrow channel (width of few grid sizes) that makes an angle of 45 degrees with the
computational grid [Stelling et al, 1986], which is illustrated in Figure 2.2.
Figure 2.2
234
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
Cadv
u t v t
,
x
y
max
1,
(2.4.3)
with x and y the grid spaces in the physical space. Explicit integration of the horizontal
diffusion term yields an upper limit of:
1
DH
1
x2
1
y2
(2.4.4)
For the 3D transport equation the scalar concentrations are coupled in the vertical direction
by the vertical advection and diffusion term. An explicit time integration of the vertical
exchange terms on the -co-ordinate grid would lead to very severe time step limitations:
H
2 DV
H
(2.4.5)
(2.4.6)
Therefore, in the vertical direction a fully implicit time integration method is applied in
Delft3D-FLOW, which is first-order in time and leads to tridiagonal systems of equations.
The vertical coupling of the discretized transport equations is solved by a double sweep
algorithm.
To ensure that the total mass is conserved the transport equation in Delft3D-FLOW is
discretized with a mass conserving Finite Volume approach (flux form).
The transport equation is coupled with the momentum equations via the baroclinic pressure
term, see Eqs. (2.3.1.8). The temporal variations in salinity are slow compared to the
variations in the flow and therefore the baroclinic term in the momentum equations is
treated explicitly, introducing a stability condition for internal gravity waves (baroclinic
WL | Delft Hydraulics
235
31 December 2007
Final version 1.0
X0356, M3470
mode). The coupling with the flow is weak and in Delft3D-FLOW the transport equation is
solved independently of the flow for each half time step.
For the time integration of the horizontal diffusion term along -planes, the CrankNicholson method is applied. If the spatial discretisation of the horizontal diffusion term is
based on a Cartesian grid using the Finite Volume approach of [Stelling and Van
Kester, 1994] the integration is explicit. Source terms are integrated explicitly. In order to
avoid negative concentrations and instabilities, sink terms are integrated fully implicit.
If we substitute into the discretised transport equation a constant concentration field (e.g.
C=1), then we arrive at the discretised continuity equation.
To keep the numerical diffusion as small as possible the horizontal advection terms in the
scalar transport equation are approximated by the sum of a third-order upwind scheme and a
second-order central scheme. A second-order central scheme is applied for the
approximation of the vertical advection term. This so-called Cyclic method is the default
option for the advection terms. In Section 2.4.4 this method and the other advective schemes
in Delft3D-FLOW are described.
For the Cyclic method the time integration follows the ADI-method for the continuity
equation. In the first stage all space derivatives in x-direction are integrated implicitly in
time and all derivatives in the x-direction are taken explicitly. In the second stage the
directions for explicit and implicit integration are interchanged. If the upwind discretisation
is used in the stage in which both the horizontal advection and vertical viscosity term are
integrated implicitly, the resulting linear system of equations has thirteen diagonals but the
matrix is diagonally dominant. Thus, the system can be solved effectively by a Red Black
Jacobi iterative scheme in the horizontal direction and a double sweep in the vertical
direction.
Claim 2.4.3b:
236
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
The time integration consists of two stages. At both stages the same time integration method is
applied for the k- turbulence model. We will therefore only give a description of the first
stage. An operator splitting method is applied. A stage consists of two steps. In the first step
the following equation is approximated:
k
k
k
+u +v =0
t
x
y
t
+u
+v
(2.4.7)
=0
In the first step a first-order upwind discretisation is used for the horizontal advection terms.
The discretisation of the horizontal advection terms leads to a system of equations which is
decoupled in the vertical direction, because the vertical terms are not present in this step.
In the second step we have the remaining terms
k
k
1
+
=
t H
H
t
1
H
DV k + + Pk Bk
H
DV
H
(2.4.8)
2
+ P + B - c2
For the time integration a fully implicit method (Backward Euler) is applied. This yields, in
combination with the first-order upwind discretisation for the advection terms, a positive
scheme. For turbulence modelling a first-order discretisation for the advection terms is
assumed to be accurately enough, because it is assumed that production, buoyancy and
dissipation are the dominant terms [Van Kester, 1994].
Claim 2.4.3c:
2.4.4
WL | Delft Hydraulics
237
31 December 2007
Final version 1.0
X0356, M3470
this artificial viscosity dominates the physical viscosity and the computed solution is much
smoother than the correct one. Higher order upwind discretisation is not free from numerical
oscillations and introduces fourth-order artificial viscosity. This higher order viscosity
suppresses the wiggles without smoothing the solution too much.
u
x m , n ,k
um, n , k
um
1, n , k
um
1,n ,k
2 x
(2.4.9)
u
x m, n, k
um , n ,k
um , n ,k
3um, n,k
3um, n,k
4um 1, n , k
2 x
um
2,n ,k
4um
um
2,n , k
1, n , k
2 x
um, n , k
0
(2.4.10)
um, n ,k <0
which are successively used in both stages of the ADI-scheme. This combination is applied
for both the normal advection term u u / x and the cross advection term v u / y . The
scheme is denoted as the Cyclic Method [Stelling and Leendertse, 1991]. Near the
boundaries the higher order discretisation stencils for the advection terms contain grid points
on or across the boundary. To avoid an artificial boundary layer or instabilities, the
discretisations are reduced to lower order discretisations with smaller stencils, see
[Stelling, 1984].
For sufficiently smooth solutions the above described cyclic method works well, because its
accuracy is of third order and the numerical viscosity is minimal. However, near local
discontinuities in the solution, e.g. due to sharp bottom gradients or hydraulic jumps, the order
of accuracy is a meaningless concept. Then, conservation properties are more important, such
as conservation of mass, conservation of momentum or conservation of energy head.
238
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
In the WAQUA scheme [Stelling, 1984] the normal advection term u u / x is discretised
with central differences and the cross advection term v u / y , based on the dissipative
reduced phase error scheme. For the cross advection term the spatial discretisation is given
by (at the first and second stage, respectively):
u
y m ,n ,k
u
y m ,n, k
xy
m, n , k
xy
m, n ,k
xy
m, n ,k
um,n+1,k um,n-1,k
,
2 y
3um, n , k
4um,n-1,k
2 y
3um, n ,k
4um,n+1,k
2 y
(2.4.11)
um,n-2,k
xy
m , n ,k
um,n+2,k
0
(2.4.12)
xy
m , n ,k
<0
Thus, the difference between the Cyclic and the WAQUA scheme is the discretisation of the
normal advection term u u / x .
For the third method (i.e. the flooding scheme), Delft3D-FLOW determines for each grid
which of the two methods (either momentum conserving or energy head conserving) is most
appropriate, depending on whether the local flow is contracting or expanding. This method,
which is based on [Stelling and Duinmeijer, 2003], can be applied to rapidly varying depthaveraged flows for instance the inundation of dry land or flow transitions due to large
gradients of the bathymetry (obstacles). The scheme is also accurate for obstacles,
represented by only one point on coarse grids. For the Flooding scheme the bottom is
assumed to be represented as a staircase (DPUOPT=MIN) of tiles, centred around water
level points. In combination with the local invalidity of the hydrostatic pressure assumption,
conservation properties become crucial. In flow expansions a numerical approximation is
applied that is consistent with conservation of momentum and in flow contractions a
numerical approximation is applied that is consistent with the Bernouilli equation. For
sufficiently smooth conditions, and a fine grid size, both approximations converge to the
same solution. The local order of consistency depends on the solution. The approximations
are second-order, but the accuracy reduces to first-order near extreme values by the use of
the so-called Minmod slope limiter [Stelling and Duinmeijer, 2003]. The limiter prevents the
generation of wiggles. We remark that the conservation of momentum has been derived only
for a Cartesian rectangular grid and depth averaged velocities.
We describe the discretisations for positive flow direction. For negative flow direction the
discretisations are defined accordingly. The momentum conservative approximation for the
normal advection term u u / x derived for a control volume around a velocity point is
given by:
WL | Delft Hydraulics
239
31 December 2007
Final version 1.0
u
x
X0356, M3470
q mx , n
q mx
u mx
1, n
q mx
q mx , n
1, n
1, n
m,n
um ,n
q mx
q mx ,n
1, n
1, n
(2.4.13)
u m ,n
ru
H mx , n
m ,n
x
m,n
2
H mx
um
1
2
1, n
ru
u m ,n
u m -1,n
u m -1,n
u m -2,n
u m -1,n
ru
q mx,n
u m - 2,n
max(0, min( ru , 0 ))
The momentum conservative approximation for the cross advection term v u / y is given
by:
u
v
y
q my , n
q my , n
u my ,n 1
ru
H m ,n
m ,n
u m , n -1
u m ,n
q my
q my , n u my ,n
2
Hm
2
1
2
u m , n-1
u m,n-1 u m , n- 2
q my , n-1u my ,n 1
y
1, n
q my , n
um ,n
y
q m,n
-1
y
1, n 1
(2.4.14)
ru
ru
u m , n-1 u m , n-2
q my ,n-1
q2
2
Hm 1,n
Hmx ,n
x
4Hmx ,n Hm+1,n
x
q 0 Hmx ,n Hm+1
,n
1
x
Hmx ,n Hm+1,n
2q2
Hmx ,n Hmx 1,n
1
x
m 1,n
1
Hmx ,n
Momentum _ Conserv
(2.4.15)
2
x
q 0 Hmx ,n Hm+1,n
The energy conservative discretisation is applied for contractions in both directions. For 2D
flow the direction of the grid lines do not always coincide with streamlines and this will
generate small head losses.
Near the boundaries the higher order discretisation stencils for the advection terms contain
grid points on or across the boundary. To avoid an artificial boundary layer or instabilities, the
discretisations are reduced to lower order discretisations with smaller stencils. Stelling [1984]
developed the numerical boundary treatment that is implemented in Delft3D-FLOW.
240
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
The fourth scheme, which is the only advection scheme in the Z-model and is not available
for a -model, is a so-called multidirectional upwind scheme. It is an extension to two
dimensions of the first-order upwind method. It is a positive and monotone scheme. For
positive u- and v-velocity, this method can be written as:
y m,n,k
um,n,k
um,n,k
um-1,n,k
x
u
u
um,n,k m+1,n-1,k m,n-1,k
x
xy
m,n,k
xy
m,n,k
um,n,k um,n-1,k
y
um,n+1,k um,n,k
,
y
um,n,k
vm,n,k
um,n,k
x
vm,n,k
y
(2.4.16)
WL | Delft Hydraulics
241
31 December 2007
Final version 1.0
X0356, M3470
accurate than the scheme of Stelling and Leendertse. It combines two numerical schemes,
namely a first-order upwind scheme and the second-order upwind scheme developed by
Fromm. In case of a local minimum or maximum the first-order upwind scheme is applied,
whereas the upwind scheme of Fromm is used in case of a smooth numerical solution. The
time integration of the Van Leer-2 scheme is explicit and therefore a CFL condition for
advection and diffusion must be fulfilled. Owing to the explicit time integration the Van
Leer-2 scheme requires per time step less computation time than the Cyclic method of
Stelling and Leendertse. However, the Van Leer-2 scheme produces a more diffusive
numerical solution, because of the fact that a first-order upwind discretisation is applied in
case of a local maximum or minimum. The transport scheme for the Z-model is described in
[Bijvelds, 2003].
In 3D, for both transport schemes, the central differences in the vertical may give rise to
non-physical spurious oscillations, so-called "wiggles" [Gresho and Lee, 1979] in the
solution. These wiggles arise in the vicinity of steep gradients of the quantity to be resolved.
In case of negative concentrations an iterative filter procedure based on local diffusion along
-lines followed by a vertical filter can be switched on in order to remove the negative
values. The filtering technique in this procedure is the so-called Forester filter
[Forester, 1979], a non-linear approach which removes the computational noise without
inflicting significant amplitude losses in sharply peaked solutions.
Cyclic method
For the Cyclic method the upwind discretisation of the horizontal advective fluxes in xdirection is described by:
huc
x
Fm ,n, k
Fm
1, n , k
m ,n ,k
(2.4.17)
For the scalar flux Fm, n, k at the U-velocity point the interpolation is given by:
10cm, n,k
Fm, n ,k
um , n ,k
hm ,n ,k
10cm
1, n , k
5cm
1,n ,k
6
5cm
6
2, n ,k
cm
2, n , k
cm
um ,n ,k
3, n , k
, um ,n ,k
(2.4.18)
hvc
y
Gm ,n ,k
Gm,n
1,k
m ,n, k
(2.4.19)
with the scalar flux Gm ,n, k at the V-velocity point determined by:
Gm ,n ,k
242
m , n ,k
hm, n ,k
cm, n ,k
cm, n
2
1, k
(2.4.20)
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
Near open and closed boundaries the approximations for the fluxes are reduced to lower
order.
cm,n,k
Fm,n,k
1 CFLadv
when um,n,k
um,n,k hm,n,k
cm 1,n,k
cm,n,k cm 1,n,k
1 CFLadv
cm 1,n,k cm,n,k
,
cm 1,n,k cm 1,n,k
0,
u
cm,n,k cm 1,n,k
cm 1,n,k cm
cm,n,k cm
(2.4.21)
2,n,k
2,n,k
when um,n,k 0,
with:
CFLadv
tu
x
and:
0,
1,
c m+1,n,k
2cm ,n ,k
cm
cm
1,n , k
cm
1, n ,k
1, n , k
c m+1,n,k 2cm ,n ,k cm
cm 1,n , k cm 1, n , k
(2.4.22)
1, n ,k
1, (monotone).
WL | Delft Hydraulics
243
31 December 2007
Final version 1.0
X0356, M3470
2.4.5
co-ordinates
In the boundary-fitted -grid the sea bed and free surface are represented by = -1 and = 0,
respectively. The water column is divided into the same number of -layers independent of the
water depth. Vertical resolution increases automatically in shallow areas. For steep bottom
slopes combined with vertical stratification, the use of -transformed grids may lead to
artificial vertical mixing and artificial flow (creep) due to truncation errors in the
approximation of horizontal gradients both in the baroclinic pressure term and in the
horizontal diffusion term [Leendertse, 1990] and [Stelling and Van Kester, 1994].
Claim 2.4.5:
2.4.6
In a numerical model the process of drying and flooding is represented by blocking flow
through a face of a grid cell when the water level locally falls below a certain threshold and
so removing grid points from the flow domain that become dry and by again allowing
flow and so adding grid points that become wet when the local water level rises above a
second threshold. Drying and flooding gives a discontinuous movement of the closed
boundaries and may locally generate small oscillations in water levels and velocities. The
oscillations introduced by the drying and flooding algorithm are small if the grid sizes are
small and the bottom has smooth gradients.
The crucial items in a wetting and drying algorithm are:
The way in which the bottom depth is defined at a water level point from the four
neighbouring grid values.
The way in which the water level is defined at velocity points from the two
neighbouring grid values.
The user-defined threshold criteria for setting velocity points and water level points
wet or dry.
The time step limitation for drying/flooding specified in Table 2.2 is mainly related to
flooding. It limits the propagation of area flooding to one grid cell per time step, ensuring
244
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
that the decision based on local flow values leads to a local adjustment only.
Mathematically, this corresponds to satisfying the criterion set by the advective
characteristic direction.
Claim 2.4.6:
2.4.7
Hydraulic structures
In a Delft3D-FLOW model, so-called hydraulic structures can be defined to model the effect
of obstructions in the flow. The implementation is based on the assumption that the scales of
the structures are smaller than the local grid size - the algorithmic approach is based on subgrid scale representations. Examples of such physical structures are: 3D gates, porous plates,
local weirs, floating structures and 2D weirs.
A hydraulic structure generates a loss of energy apart from the loss by bottom friction. At
hydraulic structure points, an additional force term is added to the momentum equation, to
parameterise the extra loss of energy. The term has the form of a friction term with a
contraction or discharge coefficient.
The hydraulic structures that are available in Delft3D-FLOW, are divided into four basic
types:
closed (e.g., a gate)
quadratic friction
linear friction
floating structure
The special points are defined in a moving -co-ordinate, so the varying vertical position in
time is accounted for when needed. The flow condition at hydraulic structures may be
supercritical. For supercritical flow, the downstream water level has no influence on the
flow rate. The energy loss formulations presently available in Delft3D-FLOW assume
subcritical flow. Only for the hydraulic structure of the type 2D weir also the supercritical
flow rate is computed accurately.
WL | Delft Hydraulics
245
31 December 2007
Final version 1.0
Claim 2.4.7:
X0356, M3470
2.4.8
246
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
particular situation being modelled, and will remain a matter of judgement, tests have shown
that the computations remain stable in moderately morphologically active situations.
The interpretation of the morphological factor differs for coastal and river applications. For
coastal applications with tidal motion, the morphological variations during a tidal cycle are
often small and the hydrodynamics is not significantly affected by the bed level changes. By
increasing the morphological factor to for instance 10, the morphological changes during
one simulated tidal cycle are increased by this factor. From a hydrodynamic point of view
this increase in morphological development rate is allowed if the hydrodynamics is not
significantly influenced. In that case the morphological development after one tidal cycle
can be assumed to represent the morphological development that would in real life only
have occurred after 10 tidal cycles. In this example the number of hydrodynamic time steps
required to simulate a certain period is reduced by a factor of 10 compared to a full 1:1
simulation. This leads to a significant reduction in simulation time.
Bathymetry updating including bed-load transport
The change in the quantity of bottom sediments caused by the bed-load transport is
computed by:
Sb( ,nuu,m
t f MORFAC
A( n ,m )
( n ,m )
SED
y ( n ,m
1)
Sb( ,nvv1,m ) x ( n
1)
1, m )
Sb( n,vv,m ) x ( n ,m )
(2.4.23)
(n+1)
Key
S
( n , m)
b ,vv
(n)
Sb( n,uu, m
Sb( ,nuu, m )
1)
Velocity point
Depth point
Control volume
Sb( ,nvv1, m )
(n-1)
(Positive) bed-load
transport component
y
x
Figure 2.3
(m-1)
(m)
(m+1)
where:
( n ,m )
SED
WL | Delft Hydraulics
247
31 December 2007
Final version 1.0
f MORFAC
A( n ,m )
Sb( n,uu,m )
x ( n ,m )
y ( n ,m )
X0356, M3470
This computation is repeated for all sand and bed load sediment fractions, if more than
one is present, and the resulting change in the bottom sediment mass is added to the change
due to the suspended sediment sources and sinks and included in the bed composition and
bed level updating scheme.
For further details about the interaction of hydrodynamic and morphodynamic flow we refer
to Chapter 11 of the User Manual Delft3D-FLOW (WL | Delft Hydraulics, 2007).
Claim 2.4.8:
2.5
Software implementation
This section describes technical aspects of the software implementation that are relevant to
the validation process that have not been addressed in the previous section. In particular, it
addresses the implications of software implementation choices and techniques for the
technical quality of the computational model as a whole.
2.5.1
Implementation techniques
For Delft3D-FLOW the following choices have been made to convert the algorithmic
implementation into software:
Transition to the FORTRAN90 programming language;
Application of guide lines for programming in FORTRAN90, see [WL | Delft
Hydraulics, 2001];
Dynamic memory management (via a coupling with the C programming language).
(Delft3D-FLOW was originally coded in FORTRAN77, in which dynamic memory
management was not possible. That problem was circumvented by using C subroutines
for memory management. In near future it is foreseen that memory management will be
implemented in FORTRAN90);
Application of efficient iterative solvers for the momentum equations (a Red-Black
Jacobi solver), for the continuity equation (a (direct) tridiagonal solver), for the
248
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
2.5.2
A key aspect of complex modelling software that is used by a large community is the
integrity of the whole system. The following practices and procedures have been
implemented to ensure system integrity:
Extensive testing of all new implementations for backward compatibility using a
standard and continually updated testbank of analytical, laboratory and real life
applications; For each update a testbank is run, which consists of more than 100
testcases;
Full internal documentation of all modules before release;
Rigorous version control;
Adherence to formal module exchange and acceptance procedures;
Explicit release schemes for version updates and upgrades;
Extensive technical and user documentation;
Documentation also exists for ancillary modules such as preprocessing, simulation,
postprocessing.
Delft3D-FLOW uses ASCII input files (only in case of a restart a binary file might be used).
The output files are (mainly) in NEFIS format, which is a Neutral File System that has been
developed by WL | Delft Hydraulics. This format is platform independent.
Claim 2.5.3:
WL | Delft Hydraulics
249
31 December 2007
Final version 1.0
Claim 2.5.3:
X0356, M3470
Substantiation: The present Validation document and the Delft3D-FLOW User Manual
[WL | Delft Hydraulics, 2007].
2.5.3
Computational efficiency
Substantiation: See validation studies of Chapter 3.4 (real-world applications), for which
computation times are given.
250
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
Validation Studies
This chapter summarises validation studies and contributes to the substantiating evidence
for the claims made in the previous chapter. Each section of this chapter corresponds to a
validation study which purpose can be clearly identified in the context of the material
presented in the previous chapter. Such a study may involve case studies, theoretical
analysis, comparison with measurements, comparisons with other models, etc., as long as it
is relevant to the purpose of the study.
WL | Delft Hydraulics
31
Page 2
Page 3
Purpose
The purpose of this validation study is to show that for a schematised homogeneous channel, Delft3DFLOW computes accurate water elevations, logarithmic velocity profiles in the vertical direction, and
parabolic vertical viscosity profiles. This validation study is performed for the -model. In addition, both
2D and 3D models are investigated, with both the algebraic and the
turbulence model. For the 3D
models, different vertical layer distributions are investigated. A tracer is also added to check conservation
of mass.
Linked claims
Claim 2.2.2.7: Accurate simulation of steady and unsteady flow.
Claim 2.4.1: Accurate, robust and computationally efficient algorithmic implementation.
Approach
Flow in a simple channel with sloping bathymetry is investigated. A steady solution is reached, in
which the vertical viscosity term balances the barotropic pressure gradient. For this steady situation
an analytical solution is available from the 2D shallow water equations. Results from Delft3D-FLOW
simulations are compared to the analytical solution. Both 2D and 3D simulations are performed. For 2D
simulations water depth and depth-averaged velocity are compared to the analytical solution and for
3D simulations also vertical profiles for velocity and eddy viscosity are investigated. In the latter case,
solutions are obtained with both the algebraic and the
turbulence model. It is validated whether
both models produce logarithmic velocity profiles and parabolic eddy viscosity profiles. Also, a comparison
has been made between the -layer and with the Z-layer model results.
Model description
The properties of the model used in this validation case, are summarised in Table 3.1.1.1.
Length L (m)
Constant slope ib (-)
10000
0.0001
5
Discharge q (m /s)
1/2
/s)
65
500
60
Table 3.1.1.1: Properties for the simple channel flow validation case.
Using these parameters the equilibrium depth can be determined to be:
The discharge specified in table 3.1.1.1 is used as the upstream (inflow) boundary condition. For a cell
3
width of 500 m, the specified discharge becomes Q = 2500 m /s. At the downstream (outflow) boundary
a water level is prescribed:
The correction of 0.025 m is due to the staggered grid numbering. The water level point is located half
a grid cell (250 m) further than the depth point at which the domain boundary is located. Therefore,
the water level at the boundary is prescribed 1e-4*250 = 0.025 m lower than the water level at the
domain boundary. Using the conditions and parameters specified above, a Delft3D-FLOW simulation can
be performed. Both 2D and 3D models have been investigated, using many different settings concerning
boundary conditions and turbulence modelling.
Page 4
Results
The results obtained for the different test cases within this validation study are summarised in Table
3.1.1.2. The validity is checked by considering the (relative) difference in equilibrium depth between the
simulated solution and the analytical solution.
Dimensionality No. of layers
Turbulence
model
2D
3D
(equidistant)
N.A.
Algebraic
1
10
"
20
"
3D
10(*1)
(non-equidistant) "
10(*2)
"
10(*3)
"
Layering
(*1)
(*2)
(*3)
Algebraic
Algebraic
Algebraic
Algebraic
Depth at
inflow
[m]
3.89688
3.90756
3.91653
3.91159
3.93502
3.90322
3.93293
3.89296
3.94340
3.88902
3.93620
Layer thickness
(bottom to top)
[%]
5; 6; 7; 8; 9;
10; 12; 13; 15;
15
2.5; 3.5; 5; 6;
8; 10; 12; 15;
18; 20
4; 5.9; 8.7;
12.7; 18.7;
18.7; 12.7; 8.7;
5.9; 4
Table 3.1.1.2: Water depths at inflow boundary and comparison to analytical equilibrium depth for the
different test cases of the simple channel flow validation study.
In figure 3.1.1.1 the water depth at the simulation end time is given for several situations. It can be
seen that on average, simulations with the
turbulence model produce somewhat higher equilibrium
depths than with the algebraic model. For the
turbulence model the relative difference is in the
order of 1%, whereas for the algebraic turbulence model the relative difference is less than 0.5%. This
shows that the water levels are computed in an accurately.
Figure 3.1.1.2 shows four plots, two for the vertical velocity profiles and two for the vertical eddy
viscosity profiles for different layer distributions and turbulence models. Results are shown at the end
time of the simulation at location x = 9500 m. The top figures show results for equidistant vertical layers
and the bottom figures for a non-equidistant layering. The velocities are on average somewhat over
predicted compared to the analytical logarithmic profile. Small differences in profile shape can be noticed
between results for the algebraic and the
turbulence model. The algebraic turbulence model require
less layers to achieve an accurate logarithmic velocity profile compared to the
model. For the
algebraic turbulence model, the difference with the analytical solution is more or less constant across the
velocity profile (approx. 1%). The
model shows larger variation from surface level to bottom. In the
bottom layers, the largest differences with the analytical solution can be observed (approx. 4%).
Using more layers and/or different layering profiles has no significant effect on the results. Results with
the algebraic model show a slightly larger velocity in the bottom layers and a smaller velocity in the top
layers. The same holds for the
model, except for the bottom layers, where the difference with the
analytical solution increases rapidly. For the algebraic turbulence model, results improve when increasing
the number of layers. For the
model, using more layers has a small negative effect on the accuracy
Page 5
of the velocity profile. Using a logarithmic layering in vertical direction has a positive effect on the velocity
profile results with the algebraic model and a negative effect for the
model. Summarizing, a 10layer equidistant distribution already computes accurate vertical velocity and viscosity profiles and
accurate water levels.
Although results are shown for the -model, we remark that also with the Z-model accurate model results
are computed in Delft3D-FLOW. Despite the fact that we have a staircase bottom in a Z-model, accurate
water levels and vertical velocity and viscosity profiles are computed.
Figure 3.1.1.1: Water depths at end time of the simulation for the different test cases of the simple
channel flow validation study.
Figure 3.1.1.2: Vertical velocity (left) and eddy viscosity profiles (right) for a selection of the performed
test cases of the simple channel flow. Results shown are at the end time of the simulation at location x =
9500 m.
Many more cases have been investigated, where also domain length, slope, discharge, grid size and
time step have been varied. Results obtained with these variations show the same order of accuracy and
stability as the results presented in this validation study.
Note:
All simulations are conducted with the same boundary and initial conditions. For the outflow boundary
condition the 2D equilibrium water depth was specified, i.e. 3.89677 m. From the water depth profiles
(figure 3.1.1.1) it is apparent that the equilibrium depths in 3D are not equal to the analytical equilibrium
depth in 2D. The result of the small difference in equilibrium depth is that the water level does not have
the exact same slope as the bottom. This also gives rise to small acceleration terms in channel direction,
i.e. the flow velocity is not completely homogeneous along the channel length. This discrepancy (although
small) can be reduced by taking other boundary conditions. A possible remedy is to specify two water
level boundary conditions, resulting in a perfect water level slope, but with a slightly different discharge
2
than specified (5 m /s). Another option is to specify only a different outflow water level, which should
Page 6
be more in line with the specific 3D equilibrium depth. This results in a better water level profile, with
a slightly different equilibrium depth than the analytical 2D depth, but with the discharge as specified
above.
Results for a different outflow boundary condition is shown in Figure 3.1.1.1 as '3D 10 layers (nonequidistant no. 2) k-e adapted BCs'. An equilibrium depth is found of he = 3.9773 m, which is 2% higher
than the 2D equilibrium depth. In this case, the analytical velocity profile has been calculated using
this equilibrium depth. Differences in velocity profile and water depth compared to the solution with the
normal boundary conditions occur, but are marginal for simulations with these adaptations in boundary
conditions.
A similar analysis has been done for the eddy viscosity profile. The analytical solution is a parabolic
profile. The results show maximum relative differences with the analytical solution (of all considered
cases) of 10-12% (for the
models). For the eddy viscosity profile, a logarithmic layering has a small
positive effect for both the algebraic and the
turbulence model.
Conclusions
From the simple channel flow validation study, it can be concluded that Delft3D-FLOW accurately
reproduces steady flow (which is partly Claim 2.2.2.1) along a sloping bathymetry and the propagating
waves that proceed the steady state solution (Claim 2.2.2.7). Results show differences with a maximum
of 2% with respect to the analytical 2D solution for water depths, with a maximum of 4% for the vertical
velocity profiles and 10-12% for the vertical eddy viscosity profiles. The computational algorithm shows
robust and accurate results for different grid sizes, time steps and model properties (Claim 2.4.1). The
use of more layers or local refinement in the vertical direction has a positive effect on results with an
algebraic turbulence model and a negative effect on results with the
model. Application of different
types of (consistent) boundary conditions has a marginal effect on the results.
Page 7
Purpose
The purpose of this validation study is to investigate the ability of Delft3D-FLOW to accurately compute
the propagation of long waves, in which the treatment of boundary conditions plays a role as well. This is
achieved by considering a standing wave in a closed basin. Delft3D-FLOW uses an Alternating Direction
Implicit (ADI) time integration in the solution of the flow equations. This method switches between an
explicit and implicit time integration method for the x- and y-directions. We examine the behaviour of a
standing wave at a 45 angle with respect to the grid lines. In this way, we check whether the accuracy of
the ADI solver depends on the orientation of the grid. In the absence of bottom friction, all waves should
reflect on the closed boundaries without energy loss and the standing wave should not damp or spread.
Therefore, the effect of numerical damping is investigated in this validation study.
Linked claims
Claim 2.2.2.1: Propagation of long waves.
Claim 2.4.1: General (robustness, accuracy, efficiency).
Claim 2.4.3a: Delft3D-FLOW uses a robust ADI solver.
Approach
The propagation of long waves is investigated for propagation in x-direction, in y-direction and under an
angle of 45 with respect to the coordinate axes. The effect of the Alternating Direction Implicit (ADI)
time integration solver and the added numerical damping then becomes important. In addition, the
influence of the ratio of wave amplitude over water depth on the results is investigated.
Model description
A closed square basin of dimensions 20 x 20 m is used as domain. Waves with different direction, length
and amplitude have been used as initial water elevation. Here, the results are presented for 3 cases:
1. a wave in x-direction
2. a wave in y-direction
3. a wave with a 45 angle with respect to both axes.
For the first two cases a wave length of 20 m is taken and for the third case a wave length of
x 20 =
28.28 m. For all three cases an amplitude of 0.01 m is used. This small amplitude is chosen to ensure
purely linear wave behaviour. The model uses the following parameters:
/s (frictionless bed).
Results
The effect of the angle of the wave with respect to the grid is investigated. In figures 3.1.2.1-3.1.2.4 the
wave celerity c and period T are plotted as a function of the non-dimensional time step dt/T. The results
for the wave in x- and y-direction are exactly equal, which illustrates that in Delft3D-FLOW results for
models in either x- or y-direction are identical. For non-dimensional time steps larger than approximately
0.05-0.03 the celerity and period start to differ significantly from the analytical solution. For the 45angle wave the effect is slightly larger. This is most likely an ADI-effect. However, the absolute (and
relative) errors for the 45-angle wave are in the same order of magnitude, which shows that the
accuracy in wave propagation for Delft3D-FLOW is more or less independent of the directions of the
waves.
Page 8
More importantly, for values of dt/T that are sufficiently small (say < 0.025) the figures show that the
absolute and relative errors are small. The value of 0.025 corresponds to 40 points per wave period,
which time step limitations was given in Table 2.2. In practice, the values for dt/T are even much smaller.
For example, in a North Sea model, in which the semi-diurnal tide is dominant, and a time step of 5
minutes, dt/T reads 0.007.
Figure 3.1.2.1: Wave celerity c versus scaled time step dt/T, for the different wave directions.
Figure 3.1.2.2: Wave period T versus scaled time step dt/T, for the different wave directions.
Figure 3.1.2.3: Relative error (in %) in wave celerity c with respect to the analytical solution, versus
scaled time step dt/T, for the different wave directions.
Page 9
Figure 3.1.2.4: Relative error (in %) in wave period T with respect to the analytical solution, versus
scaled time step dt/T, for the different wave directions.
Conclusions
The propagation, reflection against boundaries and interaction of long waves (small amplitude-water
depth ratio) is simulated correctly with Delft3D-FLOW. A standing wave in a closed basin is simulated
to show nice symmetric behaviour and little numerical damping. This is not only the case for wave
propagation that is aligned to the computational grid, but also for waves that make an angle of 45 with
the grid.
Page 10
Purpose
The purpose of this validation study is to investigate the effect of a misaligned grid, i.e. a grid that makes
an angle with the flow direction, on the accuracy of the model results.
Linked claims
Claim 2.4.2: Computational grid.
Claim 2.4.3a: Robust ADI solver.
Claim 2.4.4a: Hydrodynamic advection.
Approach
A uniform flow in a simple straight channel with a sloping bottom is prescribed (see also validation study
3.1.1 Simple channel flow). Then, the slope of the water level should be equal to the slope of the bottom.
Also the barotropic pressure force balances the vertical viscosity force and the flow is in equilibrium. In
theory, the orientation of the grid should not have any effect on the accuracy of the model results. In this
validation study it is investigated to which extent this is valid for a sinusoidal grid, see Figure 3.1.3.1.
Model description
A straight channel with a sinusoidal grid is used. A bathymetry with a linear slope is applied in xdirection. The grid and bathymetry are shown in Figure 3.1.3.1. The main model dimensions and
parameters are:
2DH model
domain length L = 500 m, width 240 m.
bottom slope 0.001: Bottom height range: -4 to -4.5 m.
1/2
Figure 3.1.3.1 The computational grid and accompanying depth profile. Bottom height ranging from -4 m
(left) to -4.5 m (right).
Results
Figure 3.1.3.2 shows the water depth at the end of the simulation time. One can see that small deviations
from the analytical equilibrium depth are present, but that a large part of the interior domain has a depth
approximately equal to the analytical equilibrium depth of 2.037 m. As one may expect, deviations are
most prominent in the areas where the grid misalignment is largest (also largest grid size and aspect
ratio). The maximum relative error is (2.052 - 2.037)/2.037 x 100 % = 0.74 %.
Page 11
In addition, errors occur at inflow and outflow. This is due to the fact that the boundary conditions are
constant along the width of the channel, while the grid orientation is not. Along the boundaries the flow
is assumed to be perpendicular to the open boundary in Delft3D-FLOW, because so-called tangential
velocity can not be specified. This is a local (open boundary) effect. This can partly be circumvented by
adding one or a few columns of rectangular cells at both boundaries, as shown in Figure 3.1.3.3. The
maximum relative error then reduces to (2.047 - 2.037)/2.037 x 100 % = 0.49 %. Grid refinement, for
which no results are shown, even further reduces the overall error. After a mesh refinement with a factor
of two the maximum relative error to approximately 0.4 % for both the original mesh and the extended
mesh. This shows that the error made at the boundary dominates the error made by the grid distortion in
this case.
Figure 3.1.3.2 Water depth at the end of the simulation time (steady state solution).
Figure 3.1.3.3 Water depth at the end of the simulation time (steady state solution) with the extended
grid.
Conclusions
A misaligned mesh has a marginal distortion effect on the accuracy of the model results. Maximum errors
are in the order of 1 % or less. Delft3D-FLOW can therefore deal with flows that are not aligned with the
grid. As expected, grid refinement results into even smaller errors.
Page 12
Purpose
The purpose of this validation study is to show that Delft3D-FLOW can accurately simulate the flow
induced by a constant wind forcing in a straight channel.
Linked claims
Claim 2.2.1.3: Wind driven flow and storm surges.
Claim 2.2.3.8: Wind and bottom friction.
Approach
For the wind induced flow in a straight channel analytical solutions are available [Kocyigit and Falconer,
2004]. A comparison is made between the analytical solutions in these publications and results obtained
with Delft3D-FLOW. This is done for the water levels and vertical velocity profiles.
The analytical solution presented by Kocyigit and Falconer reads:
= 1026 kg/m .
= 0.03 m .
= 5e-3 m/s.
In Delft3D-FLOW one can not specify the wind shear stress, but only the friction coefficient. The latter is
determined using
where
m/s and
values of 0.1 and 0.235 N/m respectively were specified by Kocyigit and Falconer. Using the
above-described, friction coefficients
should be 0.004 and 0.00325, respectively, in order to arrive
at the wind stresses of Kocyigit and Falconer. In this way, the simulations of Kocyigit and Falconer are
reproduced with Delft3D-FLOW.
A same approach is followed for the linear bottom friction coefficient k1. In Delft3D-FLOW the bottom
roughness is prescribed in a different way. The standard approach is the (2D) Chzy coefficient. In
Delft3D-FLOW this value is converted in a 3D friction coefficient in case of 3D modelling. A 2D Chzy
roughness of 30 m
1/2
0.005 m/s. In case of a wind velocity of 10 m/s, a Chzy value of 35 m /s is used. We remark that the
two simulations (wind speed of 5 or 10 m/s) yields different bottom currents. In order to meet the linear
bottom friction coefficient k1, different Chzy values have to be applied.
Page 13
Model description
For this test case, the closed straight channel of [Kocyigit and Falconer, 2004] is used. The model
parameters are:
Length of the channel L = 12000 m.
Width of the channel W = 1000 m.
Depth of the channel h = 40 m.
Grid size x = y = 1000 m.
Simulation time T = 1 day (14400 minutes), with a time step t = 1 minute.
a double-logarithmic vertical -layering is used with 14 layers.
All boundaries are closed.
Initial condition: flow at rest with uniform depth 40 m.
A constant external wind forcing was applied, in the direction of the length of the channel (xdirection, 270 ):
Test 1: wind forcing of 5 m/s.
Test 2: wind forcing of 10 m/s.
Two different Chzy values were used for the two tests, derived in accordance with roughness values
specified by Kocyigit and Falconer:
Test 1: Chzy = 30 m
1/2
Test 2: Chzy = 35 m
1/2
/s.
/s.
The difference in Chzy values for the two test cases can be subscribed to the conversion from a 2D to
a 3D Chzy value in Delft3D-FLOW. The resulting bottom friction stresses were compared to make sure
identical bottom boundary conditions were applied. To this end the second test case requires a Chzy
value of 35 m
1/2
/s.
Results
Results from Delft3D-FLOW simulations are compared to the analytical solution described above. This is
done for water levels and horizontal velocity profiles. Figure 3.1.4.1 and 3.1.4.2 show the water level and
velocity profiles for the 5 m/s wind velocity case (test 1). Both the analytical and numerical results are
shown. The same is done for test 2 with 10 m/s wind velocity in figures 3.1.4.3 and 3.1.4.4.
Figure 3.1.4.1 Water level for 5 m/s wind velocity. Both analytical and numerical solutions are shown.
Page 14
Figure 3.1.4.2 Horizontal velocity profile for 5 m/s wind velocity. Both analytical and numerical solutions
are shown.
Figure 3.1.4.3 Water level for 10 m/s wind velocity. Both analytical and numerical solutions are shown.
Page 15
Figure 3.1.4.4 Horizontal velocity profile for 10 m/s wind velocity. Both analytical and numerical solutions
are shown.
A different number of layers and also different sorts of layering were used. For less than 10 layers the
results become inaccurate. For more than 20 layers, no improvement is seen in the results.
Conclusions
It can be concluded that Delft3D-FLOW can accurately reproduce the water level and horizontal velocity
profiles induced by wind. Numerical results agree very well with the analytical solution provided by
Kocyigit and Falconer.
Page 16
Purpose
The purpose of this validation study is to investigate the propagation of baroclinic flow and transport of
matter due for density driven flow. Both the hydrostatic and the non-hydrostatic mode of Delft3D-FLOW
are tested.
Linked claims
Claim
Claim
Claim
Claim
2.2.1.2:
2.2.1.4:
2.2.1.9:
2.2.2.2:
Approach
In this so-called lock exchange test case, a closed basin with two water masses with a different density
is initially separated by a vertical wall. Next, this vertical wall is removed. Due to the existence of an
internal gravity gradient, the fluid layers adjust and form a stably stratified, two-layer system with the
heavier water at the bottom and the lighter water at the surface. Sharp density fronts divide the two
layers both horizontally and vertically. Density fronts of this type are often observed in estuaries.
The length L of the basin is 112.5 m and the depth H is 10 m. The initial salinity concentrations of the two
water bodies are:
According to the above-described equation the celerity of the front is equal to 0.30 m/s for the present
case. The shear that is introduced due to the gravity current gives rise to instabilities that are allowed
to grow if the Richardson number falls typically below 0.25. Then so-called Kelvin-Helmholtz instabilities
may be observed. According to theoretical analysis, the slope of the fronts at the stagnation point (water
surface and bottom) is 60 to the horizontal [Turner, 1973].
Model description
A 2DV model is applied with 150 grid cells in the horizontal ( x = 0.75 m ) and 20 layers in the vertical
( z = 0.5 m ). In the vertical, the Z-model schematisation is used. The time step used is 0.3 sec. In this
way, the stability criterion for the propagation of internal waves is satisfied. For turbulence closure, the
model is used, without background values for viscosity and diffusivity.
The model is initialised at t = 0 s with on the left side of the basin a salinity of 5 ppt and at the right side
of the basin a salinity of 10 ppt (cf. Figure 3.1.5.1). A run has been made in both the non-hydrostatic and
the hydrostatic mode.
Page 17
Results
After the removal (at t = 0 s) of the vertical wall separating the two fluids, two discontinuities are
moving in opposite direction. Figures 3.1.5.2 and 3.1.5.3 show the salinity distribution after 120 s for the
hydrostatic and the non-hydrostatic mode, respectively. In the hydrostatic mode, the shape of the front is
not smooth. In the non-hydrostatic mode, the front is more smoothly curved. For both computations, the
bottom and surface front speeds have been derived. In the hydrostatic mode, the computed front speeds
are 0.22 m/s, whereas in the non-hydrostatic mode, the computed front speeds are 0.27 m/s.
Conclusions
The computational results are consistent with theory, as the heavier water intrudes underneath the less
dense fluid resulting in two fronts moving in opposite directions. The computed front speed of 0.27 m/s
in the non-hydrostatic mode is close to the analytical front speed of 0.3 m/s. In the hydrostatic mode the
computed front speed is considerably less.
This test case shows that Delft3D-FLOW is capable of correctly reproducing density driven flow (Claim
2.2.1.2 and Claim 2.2.2.2) as well as predicting the horizontal transport of matter (in this case salinity)
in an accurate way (Claim 2.2.1.3). Delft3D-FLOW offers the possibility of either a hydrostatic or a nonhydrostatic mode (Claim 2.2.1.9). This experiment shows that significant differences can occur between
the hydrostatic and non-hydrostatic results. Depending on the application, the user can choose the most
suitable approach. For example, the front speeds and shapes in general differ considerably between both
modes, whereas final steady state results (a stable, stratified, two density layer system) will be more or
less similar.
Page 18
Purpose
To investigate the combined effect of wave force and mass flux in a closed basin and comparison with a
semi analytical solution.
Linked claims
2.2.2.16: Accurate flow for wave induced forces and mass fluxes.
Approach
In order to get an impression of the effect of the combination of a wave force and a mass flux on the
behaviour of a water system, an analytical solution was determined for a simple case. By comparing with
the analytical solution the accuracy of the Delft3D-FLOW model results is quantified.
In a stationary situation, the forces generated by the pressure gradient, by waves and those generated
by the bottom shear stress are in equilibrium. Figure 3.1.6.1 shows a sketch of all horizontal forces that
act on a control volume.
with h the local water depth. The corresponding Chzy expression for the bottom stress reads
If we assume the wave force to be a positive constant, the sum of all forces on the control volume can in
the stationary situation be written as:
Multiplying by h and substituting the expression found for M yields a differential equation for the water
depth:
Page 19
Next, we arrive at
with given values for A, B and the depth h0 at a fixed position x0, the value h can be solved numerically
from the last equation at any given value x.
The initial water level H, to be used in the FLOW computation, can be determined by first solving h(xL)
and h(xR) at the left and right boundaries of the basin xL and xR respectively. Secondly, the following
expression is determined numerically:
If we choose:
3
1/2
1000.0 m and B = (0.4/45.0) m . The left boundary of the model is chosen at xL = 0.0 m, the right
boundary at xR = 500.0 m.
By choosing x0 equal to xL, we can iterate the last two equations above to find a value for h0 = h(xL) such
that H equals 1 m. We find h(xL) = 0.688453379 m. By inverting the equation for x - x0, the solution for
the free surface can be found.
Requirement: The results of a computation should not deviate more than 10
solution for the water level.
-2
Model description
The values for xL and xR were chosen as 0 and 500 m, respectively. A 2D homogeneous model (closed
basin) is applied with a uniform depth of 1 m. The grid size in x-direction is 10 m. In y-direction the
length of the computational domain is 60 m with grid sizes of 10 m.
Results
The above described procedure is verified for the left boundary only. Delft3D-FLOW computes at the
left boundary a total water depth of 0.6968 m. The difference with the analytical solution h(xL) =
0.688453379 m is 0.008 m, which is within the acceptance criterion. In Figure 3.1.6.2 computed water
levels are shown (in red) along the basin. The analytical solution is represented by the black line. From
this figure it can be seen that for the whole basin the computed and analytical solution are in good
agreement with each other. The two lines are almost on top of each other. On the vertical axis the water
elevation is shown. We remark that the water elevation above the reference level (of 0 m) is shown. In
order to determine the total water depth H, these values should be increased by 1 m.
Page 20
Figure 3.1.6.2: Water level (in m) along the basin; computed (in red) and according to the analytical
solution (in black); on x-axis the x-coordinate (in m) of the basin.
Conclusions
Delft3D-FLOW is capable of accurately computing a stationary free surface that is driven by a constant
wave force and a constant mass flux.
Page 21
Purpose
Verification of (super)critical flow over a weir. Moreover, the advection schemes in Delft3D-FLOW are
tested.
Linked claims
Claim 2.2.1.6: Delft3D-FLOW can be used to investigate the impact of hydraulic structures, such as
gates, weir and barriers.
Claim 2.2.2.6: Subcritical and supercritical flow.
Claim 2.2.3.4b: Abruptly changing bathymetry, orbital movements in short wave motions, or intensive
vertical circulations such as buoyant jet plumes: non-hydrostatic (Navier-Stokes) equations.
Claim 2.4.4a: Algorithm for hydrodynamic advection.
Claim 2.4.7: Algorithm for hydraulic structures is robust, accurate and efficient.
Approach
The flow condition over a weir may be sub- or supercritical. For supercritical flow the discharge at the
weir is completely determined by the energy head upstream. In such a case, the discharge is limited by:
which is described in detail in Section 10.9.2.4 of the Delft3D-FLOW manual. The purpose of this
validation study is to verify whether Delft3D-FLOW is able to accurately compute this theoretical
maximum. In Delft3D-FLOW multiple advection schemes have been implemented. One of these schemes,
namely the so-called Flooding scheme is suited for flow over obstacles like weirs. By comparing with
another advection scheme of Delft3D-FLOW (i.e. the cyclic scheme, which is the default option in Delft3DFLOW), we will show that the flooding scheme yields accurate results for such applications, while the
cyclic scheme will appear to be less accurate.
Model description
The upstream water level boundary is 2.0 m and the downstream boundary equals 1.7 m. The sill height
is 1.0 m above the bottom of the channel. The discharge is critical over the top of the sill. The energy
height upstream E1 is about 1.0 m. The width B of the channel is 90 m. The Chzy coefficient is 100
1/2
m /s (no water level gradient due to bottom friction). We simulate a channel with a grid size of 10 m
and 30 m, respectively. On the coarse grid the sill is represented by one grid cell and on the fine grid by
3
three grid cells. For our test case Qcritical yields a maximum discharge of 153 m /s.
Results
We simulate the flow with two advection schemes: cyclic, which is the default option in Delft3D-FLOW and
flooding. In table 3.1.7.1 below the discharge over the weir is presented, of which the theoretical value is
3
153 m /s. For this table it is evident that the discharge computed by the flooding scheme is close to the
theoretical value, which is much less the case for the cyclic scheme.
Cyclic
Flooding
161 m /s
157 m /s
= 10 m
170 m /s
= 30 m
207 m /s
3
3
Table 3.1.7.1: Discharge over the weir as simulated with two different grid sizes, using both the Cyclic
and the Flooding scheme.
Below four figures are presented that show the discharge per cross section of one grid cell, for these four
simulations. For the 30 m test model we have three grid cells in cross sectional direction, which means
Page 22
that we have to multiply the results plotted in Figures 3.1.7.1 and 3.1.7.2 with a factor of three to arrive
at the numbers in the table above. For the 30 m test model we have to multiply with a factor of nine.
The figures are in the following order:
Figure
Figure
Figure
Figure
3.1.7.1:
3.1.7.2:
3.1.7.3:
3.1.7.4:
Cyclic scheme - 30 m
Flooding scheme - 30
Cyclic scheme - 10 m
Flooding scheme - 10
grid size
m grid size
grid size
m grid size
Figure 3.1.7.1: Cross sectional discharges for Cyclic scheme and a 30 m grid size.
Figure 3.1.7.2: Cross sectional discharges for Flooding scheme and a 30 m grid size.
Page 23
Figure 3.1.7.3: Cross sectional discharges for Cyclic scheme and a 10 m grid size.
Figure 3.1.7.4: Cross sectional discharges for Flooding scheme and a 10 m grid size.
For the Flooding scheme the relative error in discharges is a few percent (2-5% for the 10 m and 30 m
models, respectively). This shows that the Flooding scheme is able to accurately compute flows over
a weir when the flow becomes supercritical. For other (i.e. subcritical) flow conditions Delft3D-FLOW
has been tested as well. However, that is not part of this test case. For subcritical flows the discharges
over a weir and the corresponding energy loss is based on experimental data ("Tabellenboek van
Rijkswaterstaat", Vermaas 1987) and/or the formula of Carnot. For more details we refer to the Delft3DFLOW manual.
Conclusions
This validation study that Delft3D-FLOW is able to accurately compute the flow over a weir. One should
use the Flooding advection scheme for this application. The Delft3D-FLOW default Cyclic advection
scheme is less suited for such situations. For the Flooding scheme, the relative difference with the
analytical discharge for a supercritical flow over a weir is 2-5%.
Page 24
Purpose
This validation study investigates the influence of the Earth's rotation, i.e. Coriolis forcing. For models
that are of large geometrical scale, the Coriolis forcing plays an important role.
Linked claims
Claim 2.2.2.9: The effect of the Earth's rotation (Coriolis force).
Claim 2.3.3.7: Coriolis force.
Approach
A prismatic channel is considered, under the influence of wind, Coriolis forcing and bed friction. Under
such conditions, the equations can strongly be simplified, so that an analytic solution can be derived.
Results obtained with Delft3D-FLOW are compared to this analytical solution. Applying the -plane
approximation following Wyrtki (1961) results into a solution for the water level variation. The solution
represents an equilibrium between wind forcing, bed friction and the Coriolis force.
Model description
For this large scale model spherical coordinates are used. A prismatic grid is then specified from 115 to
118.2
West-East by -6 to 5
grid resolution of 0.066667 x 0.22917 is used, resulting into a 50x50 spherical-curvilinear grid, see
Figure 3.1.8.1. This model is a schematic representation of Makassar Strait [Hulsen et al. (1998)].
The most important model parameters are:
Page 25
Results
Starting from the initial
-plane solution the flow situation should not change when propagating in time.
grid and due to discretisation errors, disturbances occur and the flow slightly deviates from the -plane
solution. Results computed by Delft3D-FLOW are compared to the analytical (initial) solution for the water
level (Figure 3.1.8.2), v-velocity (Figure 3.1.8.3) and u-velocity (Figure 3.1.8.4).
Page 26
Figure 3.1.8.2: Water level for the -plane solution. The analytical (initial) solution is depicted on the left
and the numerical solution on the right.
Figure 3.1.8.3: V-velocity for the -plane solution. The analytical (initial) solution is depicted on the left
and the numerical solution on the right.
Page 27
Figure 3.1.8.4: U-velocity for the -plane solution. The analytical (initial) solution is depicted on the left
and the numerical solution on the right.
Figures 3.1.8.2 and 3.1.8.3 show that for the water levels and v-velocities no differences can be
observed. The u-velocity figure 3.1.8.4 might suggest that significant differences occur compared to
the analytic solution. This is due to the fact that analytic solution is a zero-velocity field. However, the
absolute differences in u-velocity are also very small (i.e. smaller than 0.01 m/s).
Conclusions
From this validation study it can be concluded that Delft3D-FLOW can accurately simulate flows for large
geometric areas, in which the Coriolis force is important. Delft3D-FLOW can accurately reproduce the
plane solution from Wyrtki (1961).
Page 28
Purpose
Test coupling of hydrodynamic and morphology processes.
Linked claims
Claim 2.4.8: Robust and accurate implementation of the coupling of hydrodynamic and morphology
processes.
Approach
In this validation study case a relatively short straight flume with a movable bed is simulated. Initially
the bathymetry is horizontal. Due to the flow through the flume and to the sediment transport, erosion of
the bed starts to occur. This process continues till an equilibrium is reached (the boundary conditions are
steady state). Then, the bed should match the slope of the water surface. Moreover, a two-domain model
(i.e. domain decomposition) is used. In this way, it is verified whether the use of multiple domains leads
to non-physical disturbances near the coupling interface of the domains.
the upstream boundary a constant discharge boundary condition is applied and the flow enters the flume
carrying the local equilibrium suspended sediment concentration profile. At the downstream end of the
flume a constant water level is specified. As the bed of the flume is initially horizontal, an accelerating
flow is created. This in turn causes an increasing sediment transport rate along the length of the flume
and erosion of the bed. This process continues until the bed of the flume matches the slope of the water
surface and the process becomes stationary; equilibrium conditions have been achieved. In Delft3DFLOW at each time step the hydrodynamic quantities and the morphodynamic quantities are updated.
Substitution of this discharge gives the following expression for the bottom slope i:
Model description
The main characteristics are:
Results
Figure 3.1.9.1 shows the profile of the bed of the flume at steady state for the 2-domain model. A stable
solution is reached after approximately 30 hours and that, after a small adaptation near the upstream
boundary, the equilibrium bottom profile forms a straight line at a constant slope. By using the equation
for i given above, it can be verified that the slope of the bed is very close to the theoretical slope of the
2
water surface, given the specified discharge and bed roughness. Since H = 0.4072 m, Q = 0.198 m /s
1/2
and C = 40 m /s, we obtain a theoretical slope of 0.0193. For the model results we have a bottom slope
of 0.0183 (roughly 0.04 m over a distance of 12 m), which is close to the theoretical value.
Page 29
Conclusions
The equilibrium solution in a straight flume has been computed with Delft3D-FLOW and compared to the
analytical solution. The results show a close resemblance. Therefore, it can be concluded that Delft3DFLOW can correctly simulate coupled hydrodynamic and morphology processes for this validation study.
Page 30
Tidal flume
Water elevation in a wave flume
Vertical mixing layer (horizontal splitter plate)
One-dimensional dam break
Horizontal mixing layer (vertical splitter plate)
Numerical scale model of an estuary
Numerical scale model of an estuary and a tidal dock
Two-dimensional dam break
Page 31
Purpose
This validation study simulates the intrusion of salinity in a tidal flume. The goal is to validate the
numerical code of Delft3D-FLOW with respect to salinity stratification by comparison with laboratory
measurements in a scale model of a tidal flume at WL | Delft Hydraulics (Van Kester et al., 1993).
Another goal is to examine the accuracy of the turbulent transport computed by Delft3D-FLOW.
Linked claims
Claim 2.2.1.2 Density driven flow and salinity intrusion.
Claim 2.2.1.4 Horizontal transport of matter on large and small scales.
Claim 2.2.2.10 Turbulent mixing including Internal Wave Model.
Approach
2
The scale model of the tidal flume has a basin with a surface area of 120 m , representing a sea, and
a flume with a width of 1 m and a length of 130 m, representing a river. In the numerical simulation,
2
we use a schematisation of the river (130 x 1 grid cells of 1 x 1 m each). At the sea side a water level
boundary is applied. The model is used to simulate salt intrusion in a river for both the - and the Zmodel of Delft3D-FLOW.
Model description
Some characteristics of this model are:
-model and Z-model.
2DV model (20 uniformly distributed layers for both the - and Z-model).
Inflow of 12.5 ppt at the sea boundary and a fresh water inflow at the downstream river boundary.
turbulence model.
The set up of the experiment with the tidal flume was chosen such that a partly stratified tidal flow occurs
for a smooth bottom with:
a minimal salt intrusion of order 20 m,
a maximal salt intrusion less than 75 m and
a vertical stratification characterised by a gradual transition from salt water to fresh water.
For that purpose numerical values were chosen for the quantities of interest, of which the most important
ones are:
for a smooth bottom: a 2-dimensional Chzy coefficient between 65 and 70 m1/2/s,
a water depth in the river of 0.2 m,
a density difference of 10 kg/m3 and
for horizontal tidal forcing at the end of the river: a tidal period of 600 seconds such that there is a
reasonable displacement due to the tide.
In this validation study the salt intrusion is roughly 45 m.
Results
Results are shown in Table 3.2.1.1 and Figure 3.2.1.1 ( -model) and Table 3.2.1.2 and Figure 3.2.1.2 (Zmodel).
Cross
section
at (from
river)
Mean of
difference
Standard
deviation
of
difference
RMS of
difference
Maximum
of
difference
Minimum
of
difference
Range of
Range of
observation simulation
Page 32
3m
12 m
24 m
36 m
48 m
-0.05313
0.56552
0.106474
1.27404
1.205782
2.158168
2.26669
0.701648
2.29299
2.278201
2.131682
2.308516
0.700956
2.597987
2.552324
3.753557
7.594468
1.695141
7.669337
8.640538
-9.01805
-3.97777
-1.3517
-2.39202
0.163501
13.79644
11.91159
12.3959
12.11799
8.47773
13.13299
13.09592
13.01761
12.44464
0.472296
Table 3.2.1.1: Comparison of observed (measurement laboratory experiment) with simulated salinity in
case of a -model.
Cross
section
at (from
river)
3m
12 m
24 m
36 m
48 m
Mean of
difference
-0.4863
-1.15774
-0.56057
-0.28123
-0.07713
Standard
deviation
of
difference
2.173942
1.70966
1.445507
1.096491
0.890576
RMS of
difference
Maximum
of
difference
Minimum
of
difference
Range of
Range of
observation simulation
2.200991
2.047007
1.533457
1.118628
0.882749
4.817328
1.163412
1.730078
1.646807
0.901117
-6.47257
-6.01981
-5.41341
-3.63715
-3.58255
13.79644
11.91159
12.3959
12.11799
8.47773
12.51534
12.52684
12.5
12.43679
9.198372
Table 3.2.1.2: Comparison of observed (measurement laboratory experiment) with simulated salinity in
case of a Z-model.
Figure 3.2.1.1: Comparison of observed (measurement laboratory experiment) with simulated salinity in
case of -model.
Page 33
Figure 3.2.1.2: Comparison of observed (measurement laboratory experiment) with simulated salinity in
case of Z-layers.
The two figures show the evolution of the salt intrusion for one tidal cycle (of 10 minutes) for a -model
and a Z-model. The measurements are shown in open circles. For both models the computed results are
in good agreement with the measurements. During inflow and outflow the salinity profiles are accurately
computed. A salt wedge is entering this schematised river and the heavier salt water remains in the lower
(vertical) part of the water column, with fresh water on top of it. However, from Figure 3.2.1.1 one can
observe that the salt intrusion simulated with the -model does not reach that far the sea side (right
side) as measured in the laboratory experiment (compare for instance the bullets at the 48 meter crosssection for t = 1 hour and 34 minutes). With respect to this, Figure 3.2.1.2 shows that the simulation
with Z-layers performs better. This is also indicated by the last row of both tables: for the Z-model
the mean, standard deviation and RMS (root mean square) of the difference between measured and
simulated salinity at the 48 meter cross-section is much smaller than the corresponding values in case of
-layers.
Conclusions
This validation study shows that Delft3D-FLOW is able to accurately model the time evolution of a salt
plume that enters and leaves a river. Both the Z-model and the -model yield accurate results. The Zmodel performs slightly better than the -model when considering the maximum distance of intrusion.
Page 34
Purpose
The goal of this validation study is to investigate the propagation and dispersion of waves over a
submerged bar. For this purpose a non-hydrostatic model is used. A comparison is conducted with
measurements from the so-called Beji & Battjes experiment (Beji and Battjes, 1994).
Linked claims
Claim 2.2.1.9: Hydrostatic and non-hydrostatic flow.
Claim 2.2.2.5: Propagation of short waves.
Claim 2.2.3.4b: Abruptly changing bathymetry, orbital movements in short wave motions, or intensive
vertical circulations such as buoyant jet plumes: non-hydrostatic (Navier-Stokes) equations.
Approach
Beji and Battjes performed several experiments concerning the propagation of waves over a submerged
bar. Due to dispersion effects, high and low frequency waves are generated. Due to the abrupt changes
in bathymetry over the bar, vertical accelerations are non-negligible and the flow is no longer hydrostatic.
Using the non-hydrostatic module of Delft3D-FLOW, this experiment is simulated and the results are
compared with the measurements of Beji and Battjes.
Model description
The laboratory experiment performed by Beji and Battjes comprises the geometry shown in Figure
3.2.2.1. At the left boundary of the domain a wave generator introduces sinusoidal waves that propagate
into the domain and over the submerged bar. At the right boundary a sloping beach is present, absorbing
the waves. At a number of measuring stations (see also Figure 3.2.2.1) the computed Delft3D-FLOW
results are compared with measurements.
x = 2.5 cm.
-5
-3
Results
Using the non-hydrostatic module of Delft3D-FLOW the Beji and Battjes experiment has been conducted.
Figures 3.2.2.2-3.2.2.4 show time series of water levels at three locations on top and behind the bar: at
13.5 m, 15.7 m and 19 m from the inflow boundary. From the figures it can be seen that the amplitudes
and frequencies of the longest waves agree to a large extent with the measurements. In particular, this is
Page 35
the case near the bar (Figure 3.2.2.2). The shorter waves are represented less accurately. Further away
from the bar, the differences between the computed and measured results become apparent.
A sensitivity analysis has shown that a relatively small time step has to be applied in Delft3D-FLOW.
This is due to the fact that the time integration of the Z-model in Delft3D-FLOW is first-order accurate
in case of non-hydrostatic modelling. We remark that in case of hydrostatic modelling Delft3D-FLOW is
second-order accurate in time. A research version of Delft3D-FLOW is available with a higher accuracy
with respect to the time integration, allowing a larger time step. However, all validation studies in this
document are done with the operational version of Delft3D-FLOW.
This validation study focuses on computing water levels in a non-hydrostatic model. In most of the
practical applications of non-hydrostatic modelling, however, plume dispersion in horizontal and vertical
direction is the main issue, for which a first-order time integration method of the non-hydrostatic part is
more than sufficient.
Performing simulations with the hydrostatic model of Delft3D-FLOW yield very inaccurate water levels
compared to the measurements, due to the large dispersion effects that occur when the waves pass
the bar. This shows that modelling the physical phenomenon of flow over a submerged bar can not be
modelled in detail with hydrostatic models.
Figure 3.2.2.2: Water level history at a station located at 13.5 m from the inflow boundary.
Measurements in blue, Delft3D-FLOW simulations in red.
Figure 3.2.2.3: Water level history at a station located at 15.7 m from the inflow boundary.
Measurements in blue, Delft3D-FLOW simulations in red.
Figure 3.2.2.4: Water level history at a station located at 19.0 m from the inflow boundary.
Measurements in blue, Delft3D-FLOW simulations in red.
Conclusions
From this validation study it can be concluded that accurate propagation and dispersion of short waves
over a bar can only be simulated using the non-hydrostatic module of Delft3D-FLOW. The results are in
reasonable agreement with the measurements. Improvements are possible (and have been realised in a
Page 36
research version of Delft3D-FLOW). However, accurately computing the propagation of such small waves
is not our main area of interest for non-hydrostatic modelling. The main goal is simulating dispersion of
buoyant plumes (see e.g. Validation Study 3.3.4).
Page 37
Purpose
Test of vertical mixing and vertical exchange of momentum in stably stratified (salinity) flow. We remark
that this validation study has been reported earlier, see [Van Kester, 1994], [Uittenbogaard, 1989, 1992].
Linked claims
Claim 2.2.1.2: Density driven flow and salinity intrusion.
Claim 2.2.1.11: Small scale current patterns near harbour entrances.
Claim 2.2.2.10: Turbulent mixing including Internal Wave Model.
Claim 2.4.3c: Delft3D-FLOW uses an accurate and robust algorithmic implementation for turbulence
modelling.
Approach
For this validation study experimental data from a tidal flume is compared with numerical results obtained
with Delft3D-FLOW.
Model description
The main characteristics are:
Length of the test basin L = 40 m.
Width of the test basin W = 1 m.
2DV Flow, with different currents and salinity above and below the (horizontal) splitter plate. We
assume that there is no influence from the side walls.
Space varying horizontal grid size x in the first meter from the left boundary. From the splitter
plate to 1 meter inside the domain, x increases from 5 cm to 20 cm. In the remaining part there is
a constant grid size of 20 cm.
Uniform vertical grid size z with 40 layers.
Time step t = 0.36 s.
Simulation time: 6.0 minutes.
Initial conditions:
Upper layer (thickness 0.324 m): salinity 22.6 ppt (water density approximately 1016 kg/m ).
Lower layer (thickness 0.250 m): salinity 43.3 ppt (water density approximately 1031 kg/m ).
Boundary conditions:
Inflow (left): measured vertical velocity, turbulent kinetic energy and energy dissipation
profiles are prescribed.
Outflow (right): water level is 0.570 m. Boundary is weakly reflective:
= 10.0 s.
Bottom friction: A Chzy coefficient of 65 m1/2/s.
A constant horizontal background eddy viscosity and diffusivity of 1.0e-6 m2/s.
Results
At four locations along the centreline of the flume the Delft3D-FLOW results are compared with
measurements. This is done at 2 m, 5 m, 10 m and 40 m behind the splitter plate. The comparison is
conducted for the horizontal velocity profiles (Figure 3.2.3.1), the relative density (Figure 3.2.3.2) and
the turbulent kinetic energy (Figure 3.2.3.3).
Page 38
Figure 3.2.3.1: Horizontal velocity profiles at 2 m, 5 m, 10 m and 40 m behind the splitter plate. Red
circles represent the measurements and the solid lines represent the profiles as simulated by Delft3DFLOW.
Figure 3.2.3.2: Density profiles at 2 m, 5 m, 10 m and 40 m behind the splitter plate. Red circles
represent the measurements and the solid lines represent the profiles as simulated by Delft3D-FLOW.
Figure 3.2.3.3: Turbulent kinetic energy profiles at 2 m, 5 m, 10 m and 40 m behind the splitter plate.
Red circles represent the measurements and the solid lines represent profiles as simulated by Delft3DFLOW.
Page 39
Conclusions
From the results we concluded that Delft3D-FLOW can accurately simulate the flow around a horizontal
splitter plate, resulting in a vertical mixing layer. Flow velocities, density profiles and turbulent kinetic
energy profiles computed with Delft3D-FLOW agree very well with the measured quantities.
Page 40
Purpose
The purpose of this validation study is to test the Delft3D-FLOW modelling accuracy for flooding over
both a dry bed and a wet bed situation, resulting from a dam break. The Delft3D-FLOW model results are
compared with the analytical solution, see (Stoker, 1957).
Linked claims
Claim
Claim
Claim
Claim
Claim
Figure 3.2.4.1: Water surface slope for a dam break scenario with initially dry bed.
In this validation study the flooding scheme is examined for a dam break situation. In Figure 3.2.4.1
the surface profile after a dam break is shown. In (Stoker, 1957) it is shown by using the theory of
characteristics that the velocity u and wave celerity c can be computed according to:
For our test case this solution has been determined numerically. This analytical solution will be compared
with the solution computed by Delft3D-FLOW.
Model description
The dam break is simulated by a discontinuity in the water elevation at t = 0 (initial condition). The water
level before the dam is 2.0 m and behind the dam 0 m (dry bed) or 0.1 m (wet bed).
Some model characteristics:
Page 41
Results
Figure 3.2.4.1 contains the results for a dam break over a dry bed. The Delft3D-FLOW results for the
Cyclic scheme (black line) and the Flooding scheme (blue line) are presented, as well as the analytical
solution (in red). In this figure the front velocity is plotted. At about x = 15000 m the dam break occurs,
which starts with a front velocity of 2 m/s. At about x = 60000 the front velocity is zero. In Figure 3.2.4.2
it can be seen that the Flooding scheme accurately predicts the front velocity, which means the flooding
of the dry area occurs with a correct speed. The Cyclic scheme is clearly less accurate.
Figure 3.2.4.2: Water level for flooding of a dry bed, computed (in blue and black) and analytical (in red).
For the dam break in case of a wet bed (in our example 10 cm) the same conclusions can be drawn,
see Figure 3.2.4.3. The Flooding scheme accurately predicts the flooding velocity of this initially wet
(shallow) area, while the Cyclic scheme does not. This also illustrates that it is important to select for
each application the most suitable advection scheme. Delft3D-FLOW offers several advection schemes,
see Section 10.5.1. of the Delft3D-FLOW manual for further details.
Figure 3.2.4.3: Water level for flooding of a wet bed, computed (in blue and black) and analytical (in
red).
Conclusions
This validation shows proofs that Delft3D-FLOW is able to accurately predict the flow resulting from dam
breaks. This is valid for dam breaks over either initially dry or initially wet beds. The Flooding advection
scheme is more suitable for this application than the Cyclic scheme. Simulations with the Flooding
scheme yield accurate flooding velocities and wave heights, which is illustrated by comparison with the
analytical solution.
Page 42
Purpose
The purpose of this validation study is to show that Delft3D-FLOW can accurately compute properties
of quasi-2D turbulence as observed in the free-surface mixing-layer experiments of Uijttewaal & Booij
(2000). This means that the Delft3D-FLOW in combination with the HLES turbulence modelling technique
is tested on its capability to reproduce the mean streamwise velocity U, its RMS |u'| and the lateral
momentum flux
Linked claims
Claim 2.2.1.11: Delft3D-FLOW can be used for an accurate prediction of small scale current patterns near
harbour entrances. For example, a so-called Horizontal Large Eddy Simulation (HLES) can be applied to
resolve small scale turbulent behaviour.
Claim 2.2.2.10: In Delft3D-FLOW turbulence quantities in the horizontal direction can be modelled with
increasing complexity: a constant viscosity, or space varying (and time constant) viscosity or a space and
time varying viscosity can be applied (Horizontal Large Eddy Simulation (HLES) model).
Claim 2.2.3.6b: HLES model.
Claim 2.4.1: Delft3D-FLOW uses an accurate, robust and computationally efficient algorithmic
implementation for the shallow water equations (hydrostatic model) and for the incompressible NavierStokes equations (non-hydrostatic model). This is the case for both time-dependent and steady state
problems.
Claim 2.4.3c: Delft3D-FLOW uses an accurate and robust algorithmic implementation for turbulence
modelling.
Approach
A Delft3D-FLOW model has been set-up to simulate mixing layer characteristics in the laboratory
experiment by Uijttewaal & Booij (2000). Figure 3.2.5.1 shows the laboratory experiment.
Model description
The grid is such that its closed boundaries follow the flume walls and the splitter plate, see Figure
3.2.5.2. The dimensions of the numerical model are 20 m x 3 m. The mesh size of the grid varies from 4
mm x 16 mm near the tip of the splitter plate to 40 mm x 40 mm at the boundaries. To avoid influences
of spin-up effects in the analysis of flow characteristics, a relatively long simulation period of 50 minutes
Page 43
with a time step of 0.006 s is used. For computational reasons the simulations are split in a spin-up
simulations and a subsequent production simulation. The model runs in depth-averaged (2DH) mode to
keep computation times reasonable.
65 m
/s.
In the HLES simulation, effects of turbulent momentum exchange at scales that cannot be resolved
on the computational grid (close to and below the grid resolution) are taken into account by the HLES
method. HLES decomposes the flow field in a slow-varying or steady part and a fluctuating part by
means of a temporal filter technique. Based the fluctuating flow field HLES computes the effect of the
unknown small-scale turbulence on the flow by means of a space - and time-varying eddy viscosity. In
this simulation the filter time parameter is set to 1 min, which means that roughly all flow variations
with periods smaller than 1 minute are considered as fluctuation, and the remaining part as mean flow
behaviour. Furthermore, default HLES settings are used, see Table 3.2.5.1. Effects of three-dimensional
turbulence are accounted for by applying the Elder formulation, which is a 3D turbulence closure for
depth-averaged models. No background eddy viscosity is applied.
HLES Parameter name
Htural
Hturnd
Hturst
Hturlp
Hturrt
Hturdm
Hturel
Meaning
Slope in log-log spectrum
Dimensional number (=2 for
quasi-2D turbulence)
Prandtl-Schmidt number
Numerical low-pass coefficient
Filter relaxation time
Molecular diffusivity
Application of so-called Elder's
term to account
for 3D turbulence contributions to
2D eddy viscosity
Value
3
2
0.7
0.3
1 min
2
0 m /s
Yes
Results
The computed flow fields are stored every three seconds over a period of 25 minutes, hence yielding 501
stored flow fields. From these data the time-averaged flow field is computed. By subtracting the timeaveraged flow field from the instantaneous flow fields a series of fluctuating flow fields is obtained. The
fluctuating flow fields are also determined "online" by the HLES method. Characteristics of the online
computed fluctuating flow fields are compared with the "offline" computed fluctuations, and with the
experimental data.
Below, the results of the simulation with both processing methods are presented in Figures 3.2.5.3 and
3.2.5.4, and compared with measurement data from Uijttewaal en Booij (2000). The definitions of the
parameters that are visualised in the figures, are given in Table 3.2.5.2.
Page 44
Name
Mixing layer
Symbol
u
|u'|
Definition
, where
u1 = velocity of undisturbed flow
above plate
u2 = velocity of undisturbed flow
below plate
RMS value of u
Lateral momentum flux
Figure 3.2.5.3: Mixing layer thickness as a function of distance from the tip of the splitter plate (which
lies at x = 0).
Figure 3.2.5.4: Observed (triangles) and simulated (lines) properties of mean flow and turbulence at a
series of downstream cross sections; blue: offline computed flow characteristics, red: online computed
flow characteristics.
Page 45
Figure 3.2.5.3 shows that the growth of the width of the mixing layer is very accurately computed in
the first 2 m downstream of the splitter plate tip. Further downstream, between 2 m and 11 m, the
computed mixing layer width becomes increasingly larger than the measured width. The development of
the mean mixing layer width is slightly better represented by the results obtained by offline processing.
Figure 3.2.5.4 shows that the computed mean and turbulent characteristics at a series of downstream
locations in general have comparable characteristics compared to the measurements. The computed
profiles of the mean flow agree well with the measurements. This indicates that the momentum exchange
across the mixing layer is accurately simulated. The computed RMS values |u'| have comparable crosssection profiles and magnitudes. The computed lateral momentum exchange
agrees well with the
measurements at 0.5 m and 5.8 m downstream of the splitter plate tip, but deviates significantly at
2 m and 11 m . At 2 m the deviation is expected to be due to incorrect measurements, because no
explanation is known for this dip in the lateral momentum flux. At 11 m the computed magnitudes are
significantly lower than observed. The cause of this deviation is unclear. Furthermore, Figure 3.2.5.4
shows that the online computed turbulent fluctuations have similar characteristics as the fluctuations that
were computed by postprocessing. This means that large scale turbulent characteristics are accurately
computed with the online time-filter method.
Conclusions
From this validation study we conclude that:
In general, the turbulent characteristics of the mixing layer computed with the HLES turbulence
modelling technique are in reasonable agreement with the observations.
The computed steady (mean) and unsteady (fluctuating) flow behaviour agrees well with the
measurements.
Turbulent fluctuations determined with the time-filtering method by Delft3D-FLOW have similar
characteristics as turbulent fluctuations, which can be determined by Reynolds decomposition of the
resolved flow fields.
Page 46
Purpose
Validation of a 3D Numerical Scale Model application of Delft3D-FLOW with -coordinates and HLES for
vertical and horizontal mixing in the Physical Scale Model of a section of the Scheldt estuary without
dock.
Linked claims
Claim
Claim
Claim
Claim
Claim
Claim
Approach
The Physical Scale Model was constructed in Delft Hydraulics' Tidal Flume. The central part of the
130 m long flume was replaced by a scaled (and mirrored) section of the Scheldt estuary near the
Deurganckdock. The model was slightly distorted. For more information on the numerical scale model,
see [Winterwerp et al, XXXX].
Model description
The numerical flow model is characterized by:
3D Model.
Curvilinear grid in the horizontal plane and sigma coordinates in the vertical.
Tidal discharge with net fresh water inflow at the up-estuary boundary.
Water levels and salt water at the down-estuary boundary.
k- Turbulence model for vertical mixing.
HLES sub-grid scale model for horizontal mixing.
The physical scale model aims at a (scaled) reproduction of prototype conditions in a bend of the Scheldt
estuary:
Tidal levels and currents extremes and time evolution (including distortion), range, current in
prototype.
Salt intrusion little stratification during flood, larger during ebb.
Some characteristic numerical parameters:
River bend with scaled 3D bathymetry, typical depth 10 - 15 cm.
Density difference of kg/m3 (equivalent to prototype).
Tidal period of 1840 s.
Results
Figure 3.2.6.1 contains an overview of the model domain (small top plot) and zooms in on the
Deurganckdock section (see detailed plot with model grid). For the Deurganckdock section the
bathymetry is plotted in Figure 3.2.6.2. Water levels at six different location are presented in Figure
3.2.6.3, while Table 3.2.6.1 contains absolute error values for water levels. This shows that the water
levels are in good agreement with measurements, because on average the RMS errors are (much)
smaller than 1 mm for a tidal range of about 40 mm.
Page 47
Figure 3.2.6.1: Detailed numerical scale model grid, Deurganckdock section, situation without dock
(Bijlsma & Van Vossen, 2004).
Figure 3.2.6.2: Depth of detailed numerical scale model, Deurganckdock section, situation without dock
(Bijlsma & Van Vossen, 2004).
Page 48
Figure 3.2.6.3: Computed (red) and measured (black) water levels at six stations along the flume, run
s12 and test of 15 April 2003, cycle 23 (Bijlsma & Van Vossen, 2004).
Obs.
SimulationMean
H-regel-C
Peilnaald
0.5
Wavo
32
Wavo
44-1
WV51/
BEZO/
T/EP98-2
Wavo
60
Wavo
68
Peilnaald/
Wavo-80-1
Wavo
96m
Wavo
120m
Wavo
31m
Wavo
44m
Wavo
52m
Wavo
60m
Wavo
68m
Wavo
81m
Wavo
96m
Wavo
120m
RMS
Max
Min
-0.02
Stand.
dev.
0.15
-0.41
Obs.
range
39.30
Simulation
range
39.15
0.15
0.35
0.01
0.68
0.68
1.11
-1.47
37.99
39.32
-0.02
0.61
0.61
1.17
-1.38
38.22
39.55
-0.23
0.64
0.68
1.09
-1.60
38.52
39.88
-0.16
0.70
0.72
1.15
-1.68
39.06
40.43
-0.06
0.72
0.73
1.35
-1.72
39.77
40.89
-0.17
0.68
0.70
1.29
-1.75
40.50
41.70
-0.18
0.71
0.73
1.27
-1.86
41.66
43.04
-0.10
0.76
0.76
1.64
-1.83
43.63
44.78
Table 3.2.6.1: Absolute error of water levels in mm (observed - simulated, run s12).
Time series of salinity are shown in Figure 3.2.6.4 and Table 3.2.6.2 contains RMS errors in salinity. From
these results it can be concluded that salinity is reproduced in an accurate way as well, because the
absolute errors are roughly 0.25 ppt, while the variation in salinity is approximately 5 ppt.
Page 49
Figure 3.2.6.4: Computed (red & magenta) and measured (blue & green) salinity at 7.2 cm - TAW in six
stations along the flume, run s12 and test of 15 April 2003, cycle 23 (Bijlsma & Van Vossen, 2004).
Obs.
VAZO1
10m
VAZO2
52m
SimulationMean
Vazo/
0.13
Temp-10
WV51/
0.12
BEZO/
T/EP98-1
VAZO3
BEZO/
0.09
60m
Temp/
EMS-60-1
VAZO4
BEZO
0.02
68m
68
VAZO
Vazo/
-0.03
96m
Temp-95
Beckmann Temp/
-0.02
cel 114m Bechu?-114
Stand.
dev.
0.28
RMS
Max
Min
-0.22
Obs.
range
1.68
Simulation
range
2.50
0.31
1.01
0.48
0.49
1.36
-0.83
5.45
5.91
0.35
0.36
0.74
-0.76
5.83
5.62
0.34
0.34
0.82
-0.64
4.38
4.34
0.11
0.12
0.17
-0.28
2.58
2.45
0.09
0.10
0.15
-0.31
0.79
1.06
Page 50
Figure 3.2.6.5: Computed (red) and measured (black) current magnitude and direction in four stations
along the flume, run s12 and test of 15 April 2003, cycle 23 (Bijlsma & Van Vossen, 2004).
Conclusions
This validation study shows that the numerical scale model that was set-up with Delft3D-FLOW is able to
reproduce the complex flow that is measured in the physical scale model for the Scheldt estuary.
Page 51
Purpose
Validation of a 3D Numerical Scale Model application of Delft3D-FLOW with -coordinates and HLES for
the complex exchange flow between a tidal dock and the estuary. We remark that the previous validation
study the hydrodynamics in the Scheldt river itself was investigated, while now the focus is on flow in the
dock as well.
Linked claims
Claim
Claim
Claim
Claim
Claim
Claim
Claim
Approach
The Physical Scale Model was constructed in Delft Hydraulic's Tidal Flume. The central part of the
130 m long flume was replaced by a scaled (and mirrored) section of the Scheldt estuary near the
Deurganckdock. The model was slightly distorted. For more information on the numerical scale model,
see [Winterwerp et al., XXXX].
Model description
The numerical flow model is characterized by:
3D Model.
Curvilinear grid in the horizontal plane and sigma coordinates in the vertical.
Tidal discharge with net fresh water inflow at the up-estuary boundary.
Water levels and salt water at the down-estuary boundary.
k-# Turbulence model for vertical mixing.
HLES sub-grid scale model for horizontal mixing.
Domain decomposition (see Figure 3.2.7.1).
We remark that the horizontal resolution bear the Deurganckdock is higher compared to the previous
validation study.
The set-up of this physical scale model aims at a (scaled) reproduction of prototype conditions in a bend
and in a dock of the Scheldt estuary:
Tidal levels and currents extremes and time evolution (including distortion), range, current in
prototype.
Salt intrusion little stratification during flood, larger during ebb.
Characteristic numerical parameters:
River bend with scaled 3D bathymetry, typical depth 10 - 15 cm.
Density difference of kg/m3 (equivalent to prototype).
Tidal period of 1840 s.
Results
Similarly to the previous validation study, Figure 3.2.7.1 contains the model grid and Figure 3.2.7.2 the
bathymetry. In Figure 3.2.7.3 time series of computed and measured water levels are shown, which are
in good agreement with each other. This also follows from Table 3.2.7.1 with RMS errors in water levels.
The RMS errors are about twice as large compared to the situation without a dock, but are still small
(roughly 1.5 mm) compared to its tidal range of 40 mm.
Page 52
Figure 3.2.7.1: Numerical scale model grid for domain decomposition, situation with Deurganckdock and
positions of instruments.
Figure 3.2.7.2: Depth of numerical scale model with domain decomposition, situation with
Deurganckdock.
Page 53
Figure 3.2.7.3: Computed (red) and measured (black) water levels at six stations along the flume, run
dd14 and test of 26 May 2003, cycle 20 (Bijlsma & Van Vossen, 2004).
Obs.
Simulation Mean
H-regel-C
PEILNAALD -0.07
0.5
WAVO
-0.39
32
WAVO
-0.44
44
WAVO
-0.49
51
WAVO
-0.45
60
WAVO
-0.37
HARB
WAVO
-0.21
68
WAVO
-0.34
81
WAVO
-0.33
96
Wavo
-0.34
120
Wavo
31m
Wavo
44m
Wavo
51m
Wavo
60m
Wavo
Harbour
Wavo
68m
Wavo
81m
Wavo
96m
Wavo
120m
Stand.
dev.
0.40
RMS
Max
Min
-0.95
Obs.
range
38.79
Simulation
range
39.13
0.40
0.87
1.13
1.19
1.29
-2.61
37.21
39.41
1.23
1.31
1.57
-2.90
37.38
39.87
1.35
1.43
1.71
-3.20
37.87
40.34
1.42
1.49
1.91
-3.28
38.40
40.89
1.38
1.43
1.99
-3.17
38.85
41.23
1.44
1.45
2.29
-3.21
39.15
41.35
1.43
1.47
2.33
-3.33
40.46
42.23
1.54
1.57
2.55
-3.50
41.54
43.51
1.57
1.61
2.81
-3.79
43.69
45.25
Table 3.2.7.1: Absolute errors of water levels in mm (observed - simulated, run dd14).
In Figures 3.2.7.4-5 computed and measured salinities are shown, while Table 3.2.7.2 contains RMS
errors in salinity. The RMS errors in salinity are in the order of 10-35% of its tidal range, which is
reasonable.
Page 54
Figure 3.2.7.4: Computed (red & magenta) and measured (blue & green) salinity at six stations along the
flume, run dd14 and test of 26 May 2003, cycle 20 (Bijlsma & Van Vossen, 2004).
Figure 3.2.7.5: Computed (red & magenta) and measured (blue & green) salinity at two stations in the
Deurganckdock, run dd14 and test of 26 May 2003, cycle 20 (Bijlsma & Van Vossen, 2004).
Obs.
Simulation Mean
RMS
Max
Min
-0.09
Stand.
dev.
0.14
-0.52
Obs.
range
2.03
Simulation
range
2.20
Vazo
10m
Bezo
52m
Vazo
60-p1
Vazo
H
Bezo
H
Vazo
68m
Beckmann
cel-114m
VAZO/
TEMP-10
BEZO/
TEMP-52
VAZO
60-p1
VAZO
HARB
BEZO
HARB-p1b
BEZO
68
TEMP-113/
BE-CEL
0.16
0.17
-0.01
0.29
0.29
1.25
-0.78
5.06
4.36
-0.42
0.80
0.91
1.23
-1.86
4.98
3.88
0.35
0.51
0.62
1.47
-0.38
2.45
3.76
-0.28
0.46
0.54
1.04
-1.64
4.31
4.16
0.06
0.46
0.47
2.60
*) -1.27
*) 5.32
3.82
0.16
0.22
0.28
0.81
-0.29
2.42
2.27
Table 3.2.7.2: Absolute errors of salinity in ppt (observed - simulated, run dd14). *) due to observed S =
0 at LW.
Page 55
Figure 3.2.7.6 shows that computed and measured current magnitude are in good agreement with each
other.
Figure 3.2.7.6: Computed (red) and measured (black) current magnitude and direction in four stations
along the flume, run dd14 and test of 26 May 2003, cycle 20 (Bijlsma & Van Vossen, 2004).
Conclusions
This validation study shows that the Numerical Scale Model is able to reproduce the complex exchange
flow between the tidal dock and the estuary to an acceptable level, despite limitations e.g. due to the
local laminar conditions in the physical scale model.
Page 56
Purpose
This validation study investigates the flow resulting from a dam break including the interaction of
reflecting bores in a 2D model. We remark that in validation study 3.2.4 a one-dimensional dam break
was examined.
Linked claims
Claim
Claim
Claim
Claim
Approach
For this validation study a model is applied that is described in Stelling & Duinmeijer (2003). It consists of
two experiments:
a dam break with an initially wet region; and
a dam break with an initially dry region.
The experiments are performed in a closed domain. So, reflecting bores interact with each other, resulting
into a more "challenging" validation case compared to the one-dimensional dam break (see Section
3.2.4). Results of both experiments are compared to numerical results obtained using Delft3D-FLOW.
Model description
The experimental set-up consists of two reservoirs, A and B, separated by a wall with a gate of width
0.4m that can be lifted (Figure 3.2.8.1). Reservoir B is initially filled with water of height 0.6 m. Two
experiments are performed. In the first experiment, reservoir A is initially dry and in the second
experiment it contains a thin layer of water with a depth of 0.05 m. The gate is then lifted with a speed of
0.16 m/s and the subsequent flooding in reservoir A is studied. The simulation period is 21 s for the wetbed experiment and 24 s for the dry-bed experiment.
Some model parameters are:
Domain length L = 31 m.
Domain width D = 8.3 m.
Grid size x = y = 0.1 m.
Time step t = 0.006 s (0.0001 minutes).
-1/3
s.
Results
The results obtained using Delft3D-FLOW are compared to the experimental results obtained with
Delft3D-FLS and in the measurement done by Stelling and Duinmeijer at the Delft Technical University.
Page 57
Delft3D-FLS is a software system that is no longer operational. However, the numerical discretisation of
the advective terms in Delft3D-FLS is similar to the Flooding scheme of Delft3D-FLOW. In the left panel
of the figures, the Delft3D-FLOW results are shown, while in the right panel contains the Delft3D-FLS
and the Stelling and Duinmeijer measurements. Figures 3.2.8.2-3.2.8.7 show time histories at different
locations along the centre line of the domain, for both the initially wet bed case and the initially dry bed
case. Figures 3.2.8.8 and 3.2.8.9 show the 3D surface plots of the two experiments at the end time of
the simulation, respectively. From all these figures it can be concluded that both codes yield accurate
results for the simulation of a two-dimensional dam break when compared to the measurements.
Although differences can be observed, the peak in water levels and the time evolution of the peaks are in
general in agreement with each other.
Additional simulations have been performed using the Cyclic scheme of Delft3D-FLOW. These simulations,
for which no results are presented in this section, show that the Cyclic scheme leads to less accurate
results when modelling a bore and produces spurious oscillations around the bore. The Flooding scheme
does not suffer from this discrepancy. This clearly illustrates that for simulation of bore phenomena the
advective terms are important and that a well-suited advection discretisation method is required.
Figure 3.2.8.2 Time history of the water level in the reservoir, for the initially wet bed simulation.
Delft3D-FLOW simulation results are shown in the left plot and the Stelling and Duinmeijer measurements
and Delft-FLS results are shown in the right plot.
Figure 3.2.8.3 Time history of the water level 6 m behind the aperture, for the initially wet bed
simulation. Delft3D-FLOW simulation results are shown in the left plot and the Stelling and Duinmeijer
measurements and Delft-FLS results are shown in the right plot.
Figure 3.2.8.4 Time history of the water level 17 m behind the aperture, for the initially wet bed
simulation. Delft3D-FLOW simulation results are shown in the left plot and the Stelling and Duinmeijer
measurements and Delft-FLS results are shown in the right plot.
Page 58
Figure 3.2.8.5 Time history of the water level in the reservoir, for the initially dry bed simulation.
Delft3D-FLOW simulation results are shown in the left plot and the Stelling and Duinmeijer measurements
and Delft-FLS results are shown in the right plot.
Figure 3.2.8.6 Time history of the water level 6 m behind the aperture, for the initially dry bed
simulation. Delft3D-FLOW simulation results are shown in the left plot and the Stelling and Duinmeijer
measurements and Delft-FLS results are shown in the right plot.
Figure 3.2.8.7 Time history of the water level 17 m behind the aperture, for the initially dry bed
simulation. Delft3D-FLOW simulation results are shown in the left plot and the Stelling and Duinmeijer
measurements and Delft-FLS results are shown in the right plot.
Figure 3.2.8.8 Three-dimensional surface level plot at the end of the wet bed simulation.
Page 59
Figure 3.2.8.9 Three-dimensional surface level plot at the end of the dry bed simulation.
Conclusions
It can be concluded that Delft3D-FLOW can accurately simulate dam breaks and the interaction of
reflected bores, provided that the Flooding scheme of Delft3D-FLOW is applied. The scheme is well-suited
for this application and does not produce spurious oscillations around the bores. The numerical results
obtained closely resemble those obtained in the scale experiments.
Page 60
Page 61
Purpose
The aim of this validation study is to compare discretisation errors for rectangular and curvilinear grids in
a channel bend. Furthermore, the use of the cut-cells functionality is tested. We remark that cut-cells are
an approach to suppress inaccuracies near staircase boundaries.
Linked claims
Claim 2.4.2: In Delft3D-FLOW the user can choose a type of grid that is suitable for the application
involved.
Approach
For this validation study a channel bend with a U-shape is investigated. The flow through such a bend is
characterised by the centrifugal force in the bend, which causes a recirculation pattern in the bend. Three
different model grids are examined in this study, a curvilinear grid, a rectangular grid and a rectangular
grid adapted using the cut-cells functionality. Due to (mis)alignment of the grid with respect to the
flow, the solutions on the three grids will be different. The discretisation errors will be the smallest on
the curvilinear grid, since the grid is fully flow aligned with the closed boundaries. The solution on the
curvilinear grid is therefore used as the reference solution.
Model description
The three grids are shown in Figure 3.3.1.1. At the south-west (left) end of the channel, a constant inflow
velocity u = 0.5 m/s is specified. At the south-east (right) end of the channel, a water level of 10 m is
prescribed. The initial water level is set at 10 m as well. The bed level is zero in the whole channel. The
different model parameters for the three grids (three simulations) are given in Table 3.3.1.1.
Curvilinear grid
Rectangular grid
Cut-cells grid
60
60
5
5
0.1
200
5 (average)
5 (average)
0.02
200
Parameter
1/2
Chzy coefficient (m
x (m)
y (m)
t (minutes)
Simulation time T
(minutes)
/s) 60
10 (average)
5
0.1
200
Figure 3.3.1.1: The three investigated grids: curvilinear, rectangular and rectangular with cut-cells.
Page 62
Results
In Figure 3.3.1.2, the water level in the channel bend is shown at the end of the simulations for all
three grids. The left panel shows the water level on the curvilinear grid, which is used as the reference
solution. The mid panel shows the water level on the rectangular grid. One clearly sees the influence of
the staircase boundary on the water levels. The are significant differences in water levels compared to the
curvilinear grid solution. The right panel shows the water levels on the rectangular grid with the cut-cells
approach switched on. Application of cut-cells significantly suppresses the stair-case problems.
In this validation study the applied grid is relatively fine. As a result, the stair-problems are relatively
small as well. In the mid panel of Figure 3.3.1.2 a water level difference of 1-2 cm is observed at the
inflow boundary, while for the right panel with the cut-cells correction it is in the order of a few mm. In
(Vatvani, 2004) a similar test case was conducted, which shows much larger discretisation errors for a
curved bend model.
Figure 3.3.1.2: Water level in the channel bend on the curvilinear, rectangular and the cut-cells grid.
In Figure 3.3.1.3, the velocity profile in the bend is shown, for the three grids. In this figure one can
observe a circulation pattern that results from the centrifugal force in the bend. The water is 'pushed'
to the outer bend near the surface level and returns to the inner bend near the bed. Except for local
disturbances in the inner bend, the circulation pattern for both the rectangular grid and the cut-cells grid
agree well with the currents on the curvilinear grid.
Figure 3.3.1.3: Velocity profile in the channel bend on the curvilinear, rectangular and the cut-cells grid.
It should also be noted that the simulation with the curvilinear grid converges in approximately 1 hour
and 40 minutes, while the simulations on the rectangular grid - with and without cut-cells - takes
more than three hours. This also confirms the introduction of disturbances in the flow due to the flow
misalignment and the staircase boundaries, leading to more iterations for the underlying numerical
solution methods. For consistency all simulations are performed using the same simulation period.
For the grid with the cut-cells, the time step is reduced by a factor five, to ensure convergence of the
solution. This time step reduction is to be expected, since the spatial resolution has become much finer
at some locations along the cut-cell boundary. The smallest grid cells in general determine the maximum
applicable time step.
Conclusions
This validation study shows that the flow through a bend can be simulated accurately using a grid that
is aligned with the closed boundaries. Both the difference in water level between inner and outer bend
and the resulting recirculation pattern are reproduced by Delft3D-FLOW. When using other grids, e.g.
Page 63
rectangular grids, the cut-cells functionality is able to suppress the occurrence of disturbances in the flow
due to the staircase boundaries.
Page 64
Purpose
Test for robustness of drying and flooding algorithms in a square basin with varying depth. A 3D model is
tested, using both a - and a Z-model.
Linked claims
Claim 2.2.2.8: Delft3D-FLOW can accurately simulate drying and flooding of tidal areas.
Claim 2.4.1: General (robustness, accuracy, efficiency).
Claim 2.4.6: Moving boundaries - representation of drying and flooding.
Approach
This validation study is conducted to examine the accuracy and the robustness of the drying and flooding
2
algorithm in Delft3D-FLOW. This validation study consists of a square domain of 900 x 900 m with
an irregular bathymetry (see Figure 3.3.2.1). In southwest to northeast direction the bottom values
are constant. The bathymetry varies from 5 m on the southwest to northeast diagonal to -5 m at the
northwest and southeast corner.
Drying and flooding is simulated by withdrawing water at the northwest and southeast corner points, at
which the bottom is deepest. Initially the water level is 4.5 m. The discharges are distributed uniformly
over all layers. The simulation period is 720 minutes. In the beginning of the simulation, water is
discharged into the test basin, yielding in increase in water levels. After about four minutes the discharge
start to withdraw water from the basin and the water level drops. At the end of the simulation the whole
model is set dry. A tracer is added to check conservation of mass in combination with drying.
Figure 3.3.2.1: Bathymetry of the test case for drying and flooding.
Model description
Some characteristics of this model are:
3D Model.
2 Subdomains, almost of equal size (subdivision in x-direction).
-model and Z-model.
Page 65
Closed basin.
In the horizontal 9 x 9 grid cells of 100 x 100 m2 each.
In the vertical direction 10 uniformly distributed layers for both for the
Zbot = -5.21 and Ztop = 5.5).
Time step of 1 minute.
Initial tracer concentration of 10 ppm.
Results
In Figures 3.3.2.2 and 3.3.2.3, the time history of water elevation at two (arbitrary) locations are shown
for both the - and the Z-model, respectively. From the water levels one clearly sees that flooding (from
T = 0.5 min to about T = 5 min) and drying (roughly from T = 8 min) occurs. Moreover, this occurs in
a smooth way, because there are hardly any oscillations. In addition, these figures illustrate that also
in case of local grid refinement the implementation of drying and flooding in Delft3D-FLOW is simulated
in an accurate and robust way. In case of the -model only some small oscillations occur when the area
becomes flooded. Near drying no non-physical oscillations are visible. We remark that drying and flooding
yields a discontinuous movement of the closed boundaries and may therefore generate small oscillations
in water levels and velocities. The oscillations introduced by the drying and flooding algorithm are in
general small.
Page 66
Figure 3.3.2.3: Time history of water levels (in m) for Z-model at two locations; on x-axis the time in
minutes.
Conclusions
Delft3D-FLOW can be used to accurately predict drying and flooding phenomena. Simulations with both
- and Z-models yield accurate results. In flooding situations, the -model gives small (numerical)
oscillations near the flooding boundary due to discontinuous effects.
Page 67
Purpose
This validation study is used to investigate the vertical mixing behaviour of Delft3D-FLOW. The Lake
Veere model is a weakly dynamic system with salinity stratification. A simplified 2DV model of Lake Veere
is applied and tested for both a - and Z-model. [Bijvelds & Nolte, 2000].
Linked claims
Claim 2.2.1.10: Thermal stratification in seas, lakes and reservoirs.
Claim 2.4.5: Suppression of artificial mixing due to - co-ordinates.
Approach
This validation study is a two-dimensional vertical (2DV) schematic model for Lake Veere in the
Netherlands. It was developed for further investigation by WL | Delft Hydraulics of the effects on the
water quality of Lake Veere after completion of the Delta works. In this simplified test case, effects of
wind and temperature are neglected. The main goal is to compare the mixing behaviour of the - and Zmodel.
Model description
The length of the model is 7000 m. Its bathymetry (see Figure 3.3.3.1) is a composition of two parabolic
shaped pits (depth 20 m) with bottom slopes and depths similar to those in the actual bathymetry of
Lake Veere. A third (non-physical) deep pit (depth 500 m) is added in order to prevent reflections of the
salt wedge at the left boundary (Figure 3.3.3.1).
Figure 3.3.3.1: Bathymetry for the two-dimensional model for Lake Veere.
Initially, the water level of the lake is set to 5 m and there is no salinity in the lake. At the start of the
h
numerical simulation, from 10-02-1998 00 :00 :00 up to 13-02-1998 11 :20 :00 , salt water is
3
discharged (a discharge of 7.5 m /s with a salinity of 12.5 ppt) at the other boundary (right boundary in
Figure 3.3.3.1). The discharge is only during a part of the simulation in order to test the - and Z-model
under both forced and unforced conditions.
2
In the horizontal directions we discretise the domain with 70 x 1 grid cells of 100 x 150 m each. For the
-model we use 20 uniformly distributed layers in the vertical direction. For the Z-model the number of
layers is 30 (with Zbot = -500 m and Ztop = 10 m). Because of the deep non-physical pit we distribute
these layers in the vertical direction as is depicted in Table 3.3.3.1:
Page 68
Layer
1 (bottom)
2
3 - 10
11 - 15
19 - 20
21 - 29
30 (top)
Thickness
94.20 %
0.30 %
0.25 %
0.20 %
0.15 %
0.10 %
0.85 %
-model a
discharge of 0.375 m /s per grid cell at layers 11 - 20, for the Z-model 0.75 m /s per grid cell at layers
19 - 23).
h
The total simulation period starts at 10-02-1998 00 :00 :00 . The time step reads 15 seconds. For both
models the
turbulence closure model is used.
Results
We focus on the development of a stratified flow during the simulation. For that purpose, Figure 3.3.3.2
shows the change in salinity profile in pit 1 at six stages in time for both Z- and -model. From the figure
it can be seen that for the Z-model strong stratification occurs. For the -model this is not the case.
To explain this, we plotted in Figure 3.3.3.3 the horizontal velocity in pit 1 at a stage in the beginning
(left plot) and at the end (right plot) for both models (similar to [Bijvelds, 2001]). From this figure it
can be observed that at the end of the simulation the horizontal velocity for the Z-model is almost zero,
which is in line with the strong stratification. For the -model the horizontal velocity oscillates strongly,
causing extra vertical mixing, which prevents the built-up of stratification. This mixing is illustrated by
Figure 3.3.3.4, which shows the vertical velocity profile at the same state in the beginning (left plot)
and at the end (right plot) of the numerical simulation. Note that the horizontal scales of these plots are
much smaller than those of the plots in Figure 3.3.3.3. From the right plot in Figure 3.3.3.4 we see that,
compared to the -model there is almost no mixing in the Z-model at the end of the simulation.
Figure 3.3.3.2: Time evolution of salinity profile at pit 1 during the numerical simulation with the Z-model
(solid line) and the -model (dash-dotted line).
Page 69
Figure 3.3.3.3 Horizontal velocity at pit 1 after 1/6 (left plot) and at the end of the numerical simulation
with the Z- (solid line) and the -model (dash-dotted line).
Figure 3.3.3.4 Vertical velocity at pit 1 after 1/6 (left plot) and at the end of the numerical simulation
with the Z- (solid line) and the -model (dash-dotted line).
Conclusions
From this validation study it can be concluded that for simulating lakes or seas with deep pits and
stratification, the Delft3D-FLOW Z-model shows better results than the -model. When the -model is
used, artificial mixing occurs and no stratification. This is due to the oscillation of horizontal and vertical
velocities near deep pits. We remark that in Delft3D-FLOW the artificial vertical mixing in -models can
be significantly suppressed by switching on the so-called anti-creep option.
Page 70
Purpose
This validation study is a test for near-field modelling of a buoyant jet. The performance of the
hydrostatic and non-hydrostatic module of Delft3D-FLOW is investigated, see also [Bijvelds, 2003]. We
remark that the non-hydrostatic module can only be used in combination with the Z-model.
Linked claims
Claim 2.2.1.4: Hydrostatic and non-hydrostatic flow.
Claim 2.2.2.2: Baroclinic flow - salinity and temperature driven flow.
Claim 2.3.3.4b: Abruptly changing bathymetry, orbital movements in short wave motions, or intensive
vertical circulations such as buoyant jet plumes: non-hydrostatic Navier-Stokes equations.
Approach
The mixing behaviour of discharges is determined by the conditions in the receiving water body and by
the discharge characteristics. In the near-field region of a discharge the hydrostatic pressure assumption
is invalid and a accurate prediction of the jet trajectory requires a non-hydrostatic module. In this
validation study, a buoyant effluent is discharged in a stagnant water body.
Model description
In order to study the behaviour of such a buoyant jet, a 2DV model is set-up. The computational domain
has a length L of 60 m and a depth H of 10 m and consists of 120 by 20 grid cells (i.e. x = 0.5 m and
z = 0.5 m). The validation study is performed using either the standard hydrostatic simulation or the
non-hydrostatic module. For consistency, both simulations have been done using the Z-model.
The discharge is schematised by means of a 3D boundary condition in the Z-model. This means that the
jet is modelled by an inflow velocity near the bottom. At all open boundary layers the velocity is set equal
to zero, except at the second layer above the bed (k=2), where a velocity of 1.0 m/s is imposed. The
discharge that enters the water body has a temperature which differs from the receiving water body by
Tdis - Tambient = 20-15 C = 5 C. At x=L a water level boundary is imposed, keeping the average free
surface level constant in time. The time step in the simulations is equal to 0.6 s. The eddy viscosity and
diffusivity are computed by means of the
turbulence model. No additional background viscosity or
diffusivity have been imposed in the current model.
No measurements or analytical solutions are available for this validation study. However, in Figure 3.3.4.1
the expected trajectory of the buoyant plume is shown. From a qualitative point of view, we will compare
this trajectory with the computed results.
Page 71
Results
Two simulations have been carried out. One with the standard hydrostatic option and one with the nonhydrostatic option of Delft3D-FLOW. In Figures 3.3.4.2 (hydrostatic) and 3.3.4.3 (non-hydrostatic)
the temperature evolution in time is shown for both cases. The figures clearly show that for the nonhydrostatic simulation, the trajectory of the plume is much more in line with the expected trajectory, as
presented in Figure 3.3.4.1. In case of hydrostatic modelling, the hydrostatic pressure distribution yields
that the plume rises much too quickly to the surface. For such applications, a non-hydrostatic module is
therefore compulsory.
Figure 3.3.4.2: Temperature of the jet after 30 minutes, 1.5 hours, 2.5 hours, 3.5 hours and 4.5 hours,
for the hydrostatic simulation.
Figure 3.3.4.3: Temperature of the jet after 30 minutes, 1.5 hours, 2.5 hours, 3.5 hours and 4.5 hours,
for the non-hydrostatic simulation.
Conclusions
For the simulation of dynamics of a rising buoyant plume in the near vicinity of its discharge, a
hydrostatic model fails to represent physics in an accurate way. In these situations, the dynamic pressure
is non-negligible compared to the hydrostatic pressure and a non-hydrostatic module should be used.
Page 72
Purpose
Test for coupling of hydrodynamic and morphology processes.
Linked claims
Claim 2.2.2.4: Transport of sediments, including erosion, sedimentation and bed load transport.
Claim 2.4.8: Robust and accurate implementation of the coupling of hydrodynamic and morphology
processes.
Approach
In this validation study, water flows across a steep-sided trench cut in the sand bed of a flume. The water
reaches the upstream edge of the trench with an equilibrium suspended sediment concentration profile.
As the flow decelerates over the deeper trench some sediment is deposited. Sediment is then picked back
up by the accelerating flow at the downstream edge of the trench. Due to the spatial difference between
the areas of deposition and erosion the trench appears to migrate downstream. Figure 3.3.5.1 shows the
initial situation, before the trench has started to deform. Both the results of measurements carried out by
Van Rijn (1984) and the computed results of Delft3D-FLOW are presented.
Model description
Initially the model has a constant depth with a trench with vertical walls in the mid of the model area.
Due to the flow field (velocity ~ 0.5 m/s), the trench starts to move and changes its shape. The goal
is to verify whether the trench development due to morphological impact is in agreement with the
measurements.
Some model characteristics:
Dimension of model: length of 30 m and a width = 0.5 m.
x = 0.3 m and y = 0.1 m.
10 Non-equidistant layers.
Time step of 0.24 seconds.
Simulation period of 10 minutes; morphological changes start after T = 5 min.
Morphological scale factor of 180, which means that at the end of the simulation the position of the
trench after 15 hours is computed (5 min. of morphological changes multiplied by 180).
Algebraic turbulence model.
Horizontal viscosity coefficient of 5e-4 m2/s.
Results
Despite significant variation in both velocity and sediment concentration profiles over the width of the
trench, the Delft3D-FLOW solution generally accurately reproduces the measured results, which can
be seen in Figures 3.3.5.1 and 3.3.5.2. In Figure 3.3.5.1 it is shown that the computed and measured
velocity and the computed and measured sediment concentrations are in good agreement with each other
at the start of the morphological changes (at T = 7.5 hours). Next, the trench starts to move. Figure
3.3.5.2 shows the measured and computed position of the trench after 15 hours. It can be verified that
the trench has been reduced to approximately one half of its initial depth, and has migrated about 3 m
downstream. The computed results are in good agreement with the measurements.
Page 73
Figure 3.3.5.1: Measured and computed velocity and sediment concentration at the beginning of the
morphological changes (T = 7.5 hours).
Figure 3.3.5.2: Measured and computed trench after 15 hours; initial trench (in green), computed trench
after 15 hours (in black) and measured trench after 15 hours (in red).
Conclusions
This validation study shows that Delft3D-FLOW can be used for accurately predicting morphological
processes in one-dimensional channels. The morphological changes that result from hydrodynamic
processes in the channel agree very well with measurements.
Page 74
Purpose
Test for wind forcing. Additionally, domain decomposition is tested in this validation case.
Linked claims
Claim 2.2.1.3: Wind driven flow and storm surges.
Approach
The goal is to compute a velocity field that is consistent with the wind forcing. In the shallow parts of the
model area the velocity field should be in the direction of the wind. Furthermore, this is a test case for
domain decomposition, which means that artificial flow may not occur near coupling boundaries.
Model description
This validation study represents a schematised lake. It is a closed basin in which a (constant) northwestern wind forcing is applied (under an angle of 45 with the grid orientation). Thus, the wind direction
is not aligned along the grid lines. The left part of the schematised lake has a constant grid size of 2000
m and the right part of 1000 m, yielding a grid refinement of a factor of two. A space varying depth
is applied. The schematised lake is deepest in the centre of the basin, see Figure 3.3.6.1. Due to the
constant wind forcing a steady state solution is reached.
Results
Figure 3.3.6.2 contains the horizontal velocities at the steady state. The green velocity arrows correspond
to the left (coarse) domain and the red arrows belong to the right (fine) domain. In the shallow areas
(i.e. all areas except the mid of the basin) the flow direction is in line with the wind direction (towards
the southeast corner). Consequently, in the deeper part (i.e. the mid of the lake) a return flow occurs.
Furthermore, the figure shows that there is a smooth horizontal velocity field near the subdomain
interface. Thus, Delft3D-FLOW computes velocity patterns that are consistent with the wind direction,
also over subdomain boundaries. It is noted that the right domain has twice as many grid points as the
left domain (1-to-2 refinement). However, to obtain a clear figure the velocity arrows have not been
plotted at all grid points in the refined domain.
Page 75
Conclusions
This validation study shows that Delft3D-FLOW can reproduce the flow pattern in a lake that is known
to occur when the flow is driven by wind only. In shallow parts of the lake, the flow velocity is in the
direction of the wind and consequently, in the deeper parts, the return flow is directed opposite to the
wind direction.
Page 76
3.3.7 Tsunami
This page last changed on 19-08-2008 by platzek.
Purpose
The aim of this validation study is to examine the different advection schemes in Delft3D-FLOW (i.e. the
standard Cyclic scheme and the Flooding scheme) for the simulation of a Tsunami wave height and runup.
Linked claims
Claim
Claim
Claim
Claim
Approach
The 1993 Okushiri tsunami caused many unexpected phenomena. One of them was the extreme runup height of 32 m that was measured near the village of Monai in Okushiri Island (see Figure 3.3.7.1).
This tsunami run-up mark was discovered at the tip of a very narrow gulley within a small cove. This
benchmark problem is a 1/400 scale laboratory experiment of the Monai run-up, using a large-scale tank
(205 m long, 6 m deep and 3.4 m wide) at Central Research Institute for Electric Power Industry (CRIEPI)
in Abiko, Japan. It is emphasized that the focus is not to simulate the run-up of the real event, but to
reproduce the laboratory measurements.
Figure 3.3.7.1: Approximate coverage of the model area in Okushiri Island (Japan).
The definition of tsunami wave height (or water level), inundation and run-up that is applied is conform
the definitions in Figure 3.3.7.2:
Figure 3.3.7.2: Definitions of tsunami wave height (water level), inundation and run-up (adapted from
"Tsunami Glossary" - IOC/UNESCO - International Tsunami Information Centre, 2006 leaflet).
Page 77
Model description
The bathymetry data file is provided on a 14 cm rectangular grid. Based on this data the Delft3DFLOW model bathymetry is constructed (see Figure 3.3.7.3). The incident wave from offshore, at the
water depth d = 13.5 cm, is specified. Note that there are reflective vertical side walls at Northern and
Southern boundary of the model. The primary goal of this validation study is the temporal and spatial
variations of the shoreline location, as well as the temporal variations of the water-surface variations at
three specified near-shore locations.
Figure 3.3.7.3: Overall (left) and detailed (right) bathymetry of the model.
The Delft3D-FLOW rectangular model for this test contains 199 by 125 points with a resolution of 28 cm.
Originally, a finer resolution model with grid size of 14 cm was constructed. After some experiments it
was found that the solutions of the fine model can be fully replicated by a coarser model. Therefore, it
was decided to carry out the final runs with the coarse model with a mesh size of 28 m.
The simulation runs for 24 seconds as the boundary condition that was downloaded only contains data for
22.5 seconds. During the last 1.5 seconds the boundary condition is gradually decreased from 1 mm to
zero. The integration time step applied is 0.025 seconds. The maximum Courant number corresponding
to this time step at the start of the simulation equals 2.9. The time step has been determined after some
experiments to ensure that the simulated results have converged.
-4
which is assumed to be the case. An eddy viscosity value equal to 1x10 m /s has been applied and at
the start of the simulation a zero water level initial condition is prescribed.
Results
The table below shows the benchmark parameter values for the two simulations using different solvers
is presented below followed by figures showing the computed and measured wave height at stations 5, 7
and 9. Data are only available for these three stations.
Quantity: Water level [m]
Station
name
Run Q09
5
7
9
Average
Run Q9C
5
7
9
Average
RMS err.
obs.
range
simulat.
range
RMS/Obs.
range
0.0010
0.0015
0.0009
0.0011
0.0042
0.0033
0.0049
0.0042
0.0043
0.0036
0.0050
0.0043
0.0093
0.0077
0.0112
0.0094
0.0158
0.0125
0.0093
0.0125
0.0459
0.0462
0.0479
0.0467
0.0452
0.0437
0.0515
0.0468
0.0947
0.0790
0.1038
0.0925
0.0014
0.0017
0.0001
0.0010
0.0051
0.0046
0.0056
0.0051
0.0052
0.0049
0.0056
0.0052
0.0098
0.0080
0.0110
0.0096
0.0200
0.0191
0.0161
0.0184
0.0459
0.0462
0.0479
0.0467
0.0467
0.0408
0.0478
0.0451
0.1139
0.1051
0.1167
0.1119
Table 3.3.7.1: "Okushiri" benchmark results of the Delft3D model for 3 stations using the flood solver
(top, run q09) and standard solver (bottom; run q9c).
Page 78
Figure 3.3.7.4: Simulated tsunami wave heights (blue line) compared to measured data (red crosses) at
selected stations using the Flooding scheme (left; run q09) and Cyclic method scheme (right; run q9c).
In general, the model is able to reproduce the main characteristics of the tsunami wave quite correctly,
especially the results from the Flooding method. Table 3.3.7.1 and Figure 3.3.7.4 show that the
performance of the Flooding scheme is 20 to 25% better than the performance of the standard advection
scheme in Delft3D-FLOW (i.e. the Cyclic method). The RMS error is for the results obtained with the
Flooding scheme equals 8 to 10% of the observed wave range. The mean error is approximately 1 mm.
The wave heights computed by the Cyclic method are slightly higher. Especially the first wave peak
overshoots the actual wave height. For this application, it is evident that the Flooding scheme has an
advantage compared to the standard advection scheme in Delft3D-FLOW.
From the figures we also observe that there is a slight mismatch between the bathymetry and initial
condition that was applied in the model and the measured data. The reason for this mismatch is yet
unclear. When the bathymetry is adjusted downwards uniformly with an amount of 5 mm, the location
of the wave peaks (e.g. at t = approximately 20 seconds) can be shifted to arrive at earlier moment as
shown in Figure 3.3.7.5. Consequently, further improvements on the performance are expected to be
possible when this bathymetry discrepancy can be corrected.
Figure 3.3.7.5: Sensitivity of the computed tsunami wave at station 7 on bathymetry; original
bathymetry (straight line) and depth lowered uniformly by 5 mm (dashed line).
Page 79
Conclusions
For this Tsunami validation study it can be concluded that the Flooding advection scheme is more
suitable than the Cyclic method advection scheme, which is the default option in Delft3D-FLOW. The
Flooding scheme computes tsunami wave heights and wave speeds that are in good agreement with the
measurements.
Page 80
Page 81
Purpose
This validation study involves the onset, duration and areal extent of thermal stratification for the
Southern North Sea. This study illustrates the ability of Delft3D-FLOW to accurately model the horizontal
and vertical temperature distribution, including the depth of a thermocline.
Linked claims
Claim
Claim
Claim
Claim
Approach
A one-year simulation is conducted for a model representing the Southern North Sea. The period of 1
November 1988 till 1 November 1989 is chosen because of the availability of extensive observations. Insitu data for the North Sea for 1989 was measured by the NERC in the North Sea project (NSP) (Lowry
et al. 1992). The NSP data set of 1989 is a very comprehensive one, having six locations on the North
Sea with a series of continuous three-dimensional temperature measurements. Figure 3.4.1.1 shows the
trajectory of the ship cruise and the locations of these six stations (A-F) on the North Sea.
Geographical location
550'N; 054'E
5530'N; 531'E
5420'N; 024'E
5330'N; 30'E
Depth (m)
73.5
49.5
55.5
29.5
Table 3.4.1.1: Overview of locations and depths of measuring stations in the NSP project.
Model description
134 by 65 horizontal grid points.
10 non-equidistant -layers.
Curvilinear grid with mesh sizes varying from 1 km to 30 km.
Page 82
turbulence model.
Time step of 300 sec.
Results
For five periods the ship track measurements have been gathered, namely for:
The measurements for each period are now compared with model results at a time that is roughly in the
mid of the period. Moreover, for each period both surface and bottom temperatures are shown. In the
figures the measurements are in open circles.
Figures 3.4.1.12-15 contain the computed vertical profiles of temperature and the NSP measurements (in
open circles) of the four stations, respectively. Day 0 corresponds to 1 November 1988 and day 365 to 1
November 1989. The figures show that the measurements from the NSP project are not available for the
whole simulation period. For these four stations the vertical temperature profiles in the measurements
are as follows:
For station A data is available till 29 September '89 with a thermocline at approximately 55 m depth
from 8 to 15 degrees at 29 September '89.
For station B data is available till 28 September '89 with a thermocline at approximately 35 m depth
from 11 to 15 degrees at 28 September '89.
For station C data is available till 30 September '89 with weak stratification at 30 September '89.
For station D data is available till 24 September '89 without any thermal stratification during the
whole period (Dogger Bank location; well mixed).
Page 83
Figures 3.4.1.2-3.4.1.5
Page 84
Figures 3.4.1.6-3.4.1.9
Figures 3.4.1.10-3.4.1.11
Page 85
Figures 3.4.1.12-3.4.1.15
Conclusions
From the model results in Figures 3.4.1.2-15 it can be concluded that horizontal and vertical temperature
distribution throughout the whole year in good agreement with the measurements. With respect to the
vertical profiles, at the deepest stations (namely A and B) the onset of the thermocline is in accordance
with measurements (around day 200). The model shows a more diffuse thermocline in comparison
to the measurements (around day 200). The model shows a more diffuse thermocline in comparison
to the measurements. Note that only ten vertical layers are applied in the vertical. At station C the
measurements show a weak stratification from day 200 to day 275, which to some extent is reproduced
by the model. For station D the model does not compute any stratification, which is in agreement with the
measurements.
This validation study clearly demonstrates the potential of Delft3D-FLOW to accurately model thermal
stratification in seas. It is noted that to accomplish this, accurate model forcing is required, in particular
the meteorological forcing.
Page 86
3.4.2 Zegerplas
This page last changed on 25-08-2008 by platzek.
Purpose
Test for temperature stratification using a 3D model. Both the - and Z-model have been tested. The
simulation deals with a one-year simulation. For the simulated period measurements are available.
Detailed documentation can be found in [Genseberger et al., 2003].
Linked claims
Claim 2.2.1.10: Thermal stratification in seas, lakes and reservoirs.
Claim 2.4.2: Computational grid.
Claim 2.5.4: Computationally efficient.
Approach
The present validation study originates from a project in which water quality in deep pits was studied.
For training purposes a relatively coarse model was set-up for lake Zegerplas in The Netherlands. The
Zegerplas is located at the east of Alphen aan de Rijn and has been used for producing salt. For this
lake field measurements are available. Deep pits have their own special physical, chemical and biological
properties. They are characterised by the presence of a steep gradient during part of the year. In warmer
periods stratification is very strong. Due to the steep gradient of the bottom of deep pits, numerical
simulations with the -model approach seems to be less suitable.
Model description
Two physical processes are involved, namely temperature and wind. For the wind forcing we use field
measurements from 1996-1997. Initially the water level is 0 m and the temperature is 5 C. In the
horizontal direction a grid with 10 x 14 grid cells of 1000 m each is applied. A schematic view of the
Zegerplas and the model grid are shown in Figure 3.4.2.1.
Figure 3.4.2.1 Schematic picture of the Zegerplas (left) and grid that was used (right).
For the discretisation in the vertical direction both for the - and Z-model (with Zbot = -32.2 m and Ztop =
0.0 m) the non-uniform distribution of layers is as is listed in Table 3.4.2.1. The simulation period is from
h
01-01-1996 00 :00 :00 to 31-12-1996 00 :00 :00 with a computational time step of 1 minute. The
is used for turbulence modelling.
Layer
Thickness
Page 87
1
2
3
4
5
6
7
8
9
10
11
12
13-50
51
9.34
6.85
6.23
5.45
5.14
4.36
3.42
2.80
2.49
2.18
1.86
1.55
1.24
0.93
%
%
%
%
%
%
%
%
%
%
%
%
%
%
Results
For this validation study, the bathymetry is non-uniform and has a steep bottom gradient. At first, we
examine at the temperature distribution in the vertical for both models during the simulation of one year.
Figure 3.4.2.2 shows the temperature results for the -model, while Figure 3.4.2.3 contains the model
results for the Z-model. From these two figures we observe that the Z-model reproduces the stratification
in the warmer period in an realistic way, while the -model does not. Additional simulations with a
uniform bathymetry did not show such differences between the two models. We therefore conclude that
the difference between the results with the - and the Z-model are due to the non-uniform bathymetry
with the steep gradients.
To verify that the results computed with the Z-model yield a good representation of the real life
stratification, we compare the results of this simulation with field measurements. Figure 3.4.2.4 shows
the temperature profile measured at August 14 1998 at the deep pit RO37 (see "+" signs), the measured
profile at August 11 1999 (see crosses), and the computed results for August 12 1996 (solid line). The
computed and measured results are in good agreement with each other, although some differences can
be observed. Over a distance of 10 m in the vertical there is a salinity difference of roughly 15C. The
thermocline depth is computed accurately by the Z-model. Only near the surface differences with the
measurements are observed, which is probably due to the heat flux input. We remark that for other
periods in August 2008 we have a similar agreement between measurements and computed results.
Page 88
Figure 3.4.2.3: Zegerplas temperature distribution over layers of the Z-model for numerical simulation of
one year.
Figure 3.4.2.3: Zegerplas temperature profile at RO371; measured at August 14 1998 (the "+"-signs),
measured at August 11 1999 (the crosses), and computed for August 12 1996 with the Z-model (solid
line).
Conclusions
From this validation study we conclude that deep pits with relatively steep bottom gradients should be
modelled with the Z-model instead of with the -model. Then, salinity stratification can be computed in
an accurate way.
Page 89
Purpose
The purpose of this validation study is to show that Delft3D-FLOW is able to reproduce both thermal and
salinity stratification in terms of its duration as well as the depth of the pycnocline. The stratification is
caused in this case by heat exchange through the surface as well as by baroclinic pressure gradients.
Since Lake Grevelingen can be characterised as a weakly dynamic water system with steep bottom
gradients, a Z-model schematisation of Delft3D-FLOW has been applied.
Linked claims
Claim
Claim
Claim
Claim
Claim
Claim
Approach
This validation study involves a horizontal and vertical spatial temperature and salinity distribution as well
as the variation of salinity and temperature over the year. Stratification in Lake Grevelingen is caused by
heat exchange through the surface and by the inflow of water with a different salinity or temperature.
This process is influenced by wind and wind induced currents. In Lake Grevelingen, the balance between
these processes is very important from a physical point of view. The 3D Lake Grevelingen model has been
validated against data for the year 2000, using 1999 as spin-up year. Extensive salinity and temperature
profile measurements (GTSO) at 20 different locations at intervals of about 2-3 weeks have been used.
The measurement locations are presented in Figure 3.4.3.1.
Figure 3.4.3.1: Overview of the Lake Grevelingen measurement locations (red dots); cross-sections will
be plotted along the red, dashed line.
Model description
Lake Grevelingen is a closed water system. This implies that the water level is determined by the
differences between the incoming and outgoing discharges. The main discharges included in the model
are caused by flow through the Brouwerssluis, by precipitation, by evaporation and by river run-off. The
computational grid has about 4200 active cells in the horizontal plane, with cell sizes varying between
250 m and 400 m. In the vertical, the Z-model schematisation is used with a maximum of 28 layers at
Page 90
the deepest locations. The layer thickness varies from 2.5 m at the surface to 5.0 m near the bottom at
the deepest locations. The bathymetry of Lake Grevelingen can be described as a system of gullies with a
few deep pits (up to about 45 m), surrounded by a shallow area.
Uniform wind forcing with daily varying magnitudes and directions are applied at the surface. For
computing the heat flux through the surface, the so-called Ocean Heat Flux model is applied, using daily
uniform values for air temperature, relative humidity and cloud cover. A Secchi depth of 2.0 m is used,
-3
whereas the Stanton and Dalton number for the convective and evaporative heat flux are set at 3.0x10
-3
viscosity and diffusivity of 0.5 m /s is applied. The computation is started with a uniform salinity of 27
ppt and a uniform temperature of 5 C. The year 1999 is used for spin-up. Given the basin averaged
residence time of about 2 months this is certainly long enough. With the applied time step of 2 minutes
computing one year takes about 28 hours on a 3.6 MHz single CPU.
Results
The computational results are compared with the measurements in various ways. In Figure 3.4.3.2 and
3.4.3.3 time series of salinity and temperature are compared to measured values at both 1 m and 15 m.
In Figure 3.4.3.4 the salinity and temperature results are presented in z,t-diagrams, whereas in Figure
3.4.3.6 contour plots at both 1 m and 15 m depth are shown. In Figure 3.4.3.7 cross sectional contour
plots along Lake Grevelingen are compared with measurements.
Figure 3.4.3.2: Time series of salinity (blue line) at 1 m (upper plot) and 15 m depth (lower plot)
compared to measurements (red dots) at Station GTSO-08.
Page 91
Figure 3.4.3.3: Time series of temperature (blue line) at 1 m (upper plot) and 15 m depth (lower plot)
compared to measurements (red dots) at Station GTSO-08.
Figure 3.4.3.4: Z,t-diagrams of salinity (upper plot) and temperature (lower plot) for Station GTSO-08.
The coloured dots represent measured values.
Page 92
Figure 3.4.3.5: Salinity contour plots at 1 m (upper plot) and 15 m depth (lower plot) compared to
measurements (coloured dots) at 14 June.
Figure 3.4.3.6: Temperature contour plots at 1 m (upper plot) and 15 m depth (lower plot) compared to
measurements (coloured dots) at 14 June.
Page 93
Figure 3.4.3.7: Salinity (upper plot) and temperature (lower plot) along a cross-section through Lake
Grevelingen. The coloured dots represent measurements.
Conclusions
Figure 3.4.3 shows that there is a very good agreement between the computed temperature time series
and the corresponding measurements. This holds both for the surface and the bottom layer, which implies
that the temperature stratification in Lake Grevelingen, and especially the seasonal variation thereof, is
modelled accurately. As the time series shown here represent only one location, the station averaged
RMS temperature errors are computed at 1 m and 15 m. These average RMS errors are 0.61 C and 0.53
C, respectively. Compared to a seasonal variation in temperature of about 20 C at the water surface,
this is a very good result. Therefore, it can be concluded that Delft3D-FLOW can be used to accurately
predict thermal stratification in lakes (Claim 2.2.1.10). Furthermore, it can be concluded that the heat
exchange through the surface can be computed using Delft3D-FLOW, yielding very good results with
respect to seasonal variations in thermal stratification (Claim 2.2.2.15). For salinity the agreement of
computed time series with measurements is also good (cf. Figure 3.4.3.2). The station averaged RMS
errors are 0.55 ppt and 0.60 ppt at 1 m and 15 m respectively, which means a good agreement with
measurements is obtained. This, together with the very good results in terms of temperature, shows that
Delft3D-FLOW is capable of accurately simulating baroclinic (density driven) flows (Claim 2.2.2.2 and
Claim 2.2.1.2).
The same claims also hold when looking at the results in the z,t-diagrams (cf. Figure 3.4). In particular,
this is the for the accurate reproduction of the measured pycnocline depths and the variations therein.
The station averaged RMS errors for the results in the z,t-diagrams are 0.61 ppt and 0.74 C for salinity
and temperature, respectively. Compared to the variations both in time and in the vertical, these are
very good results. The contour plots at 1 m and 15 m depth (cf. Figure 3.4.3.5 and 3.4.3.6) show a
good agreement with measured salinity and temperature data in a spatial way. Both in the upper and
the lower layer the horizontal distribution of temperature and salinity is reproduced well. This supports
Claims 2.2.1.2 and 2.2.2.2. The accurate prediction of thermal stratification using Delft3D-FLOW (Claim
2.2.1.10) is shown in Figure 3.4.3.7, in which computed results are presented in terms of cross-sectional
contour plots.
Page 94
Purpose
The purpose of validation study is to show that Delft3D-FLOW is able to reproduce both thermal and
salinity stratification in terms of its duration as well as the depth of the pycnocline. The stratification is
caused by heat exchange through the surface as well as by baroclinic pressure gradients caused by a
permanent density difference between the Black Sea and the Aegean Sea.
Linked claims
Claim
Claim
Claim
Claim
Claim
Claim
Claim
Claim
Claim
Approach
The Sea of Marmara is a permanently stratified water body. Differences in salinity in the two large
adjacent basins, with the lower salinity in the Black Sea and higher salinity in the Northern Aegean, result
in an overall upper layer of lower salinity flowing from Black Sea through the Sea of Marmara to the
Northern Aegean, and a lower layer return flow of higher salinity. This basic density driven circulation
varies under the influence of seasonally varying fresh water river discharges of major rivers on the Black
Sea, precipitation and evaporation. The waters are also thermally stratified. The variation in temperature
stratification is mainly determined by the seasonal variation in the surface heat flux. Varying winds and
pressure variations further affect the hydrodynamic situation.
The 3D Sea of Marmara Model is validated against data for the year 2003, using 2002 as a spinup year. We use extensive salinity and temperature profile measurements from the regular monthly
ISKI monitoring campaign. These data are available at 19 different stations in the Bosporus and its
Black Sea and Sea of Marmara approaches. The validation involves horizontal and vertical spatial
temperature and salinity distribution as well as the variation of salinity and temperature over the year.
Furthermore, hourly water level series for 2003 are available for stations representative of the northern
and southern approaches of the Bosporus. The water level response due to wind and pressure forcing
is validated against these data. Apart from comparison with measurements, validation will be based on
the representation of observed characteristic features and natural phenomena such as the existence
in the Sea of Marmara of a Cold Intermediate Layer (CIL) during summer, the occurrence of inverse
stratification (where the upper layer is colder than the lower layer) and a seasonally changing pycnocline
depth.
Model description
The area of interest for the model is the Sea of Marmara and the two straits (Bosporus and Dardanelles)
connecting it to adjacent water bodies (Black Sea and Aegean Sea). For reasons of consistency (of
forcing), flexibility and simplicity, both the Black Sea and the northern Aegean Sea are included in the
model as well. A 3D Z-model is applied with about 10,000 grid cells in the horizontal and a maximum
of 30 layers in the vertical direction. Horizontal grid sizes vary from about 250 m in the Bosporus to
about 25 km in the northern and eastern part of the Black Sea to match local spatial scales, whereas
the vertical resolution varies between 2.5 m at the surface to about 450 m in the deepest part of the
Black Sea. The model consists of three domains (grids) and are online coupled by means of domain
decomposition.
The model has one open boundary, running across the Aegean Sea. A relatively weak tidal water level
elevation is imposed here. More important is the wind and pressure surface forcing, provided on a 0.5 by
0.5 data grid in a space and time varying (6 hourly) manner. The same holds for the input parameters
of the heat flux model (Ocean Heat Flux Model). This model uses relative humidity, cloud cover, air
Page 95
temperature, wind and pressure as well as the latitude of each grid cell to compute the net heat flux
through the water surface.
For vertical turbulence closure the
diffusivity. The horizontal viscosity and diffusivity is constant and varies spatially between 10 m /s and
2
50 m /s. With an applied time step of 2.5 minutes, a simulation of one year takes about two weeks on a
single CPU (3.6 GHz).
Results
The computational results of the 3D Sea of Marmara model are compared with the available
measurements. water levels, fluxes through cross sections, salinity and temperature (both in horizontal
and vertical direction) are considered. In Figures 3.4.4.1 and 3.4.4.2, time series of measured water level
elevation are compared to the model results. In Figures 3.4.4.3 and 3.4.4.4, the computed fluxes through
the upper and lower layer of the Bosporus are presented for different seasons, whereas in Figure 3.4.4.5
time series of temperature are compared to measured values at both 1 m and 40 m (representing the
upper and lower layer). In Figures 3.4.4.6 and 3.4.4.7, salinity and temperature results are presented
in z,t-diagrams for two different locations. The last type of presentation is in the form of cross sectional
contour plots. In Figures 3.4.4.8 and 3.4.4.9 this is done for two different times, showing a cross section
along the Dardanelles, the Sea of Marmara, the Bosporus and a small part of the Black Sea, whereas
Figure 3.4.4.10 has been zoomed in on the Bosporus only in order to make comparison with measured
values easier.
Figure 3.4.4.1: Computed (dashed line) and measured (solid line) water level elevation in Pendik (near
southern Bosporus entrance) for December 2003.
Figure 3.4.4.2: Computed (dashed line) and measured (solid line) water level elevation in Kavak (near
northern Bosporus entrance) for January 2003.
Page 96
Figure 3.4.4.3: Modelled upper (blue), lower (red) and net (black) flux variation through northern
cross-section in the Bosporus for January 2003. The coloured dots represent fluxes derived from ADCP
3
Figure 3.4.4.4: Modelled upper (blue), lower (red) and net (black) flux variation through northern
cross-section in the Bosporus for July 2003. The coloured dots represent fluxes derived from ADCP
3
Figure 3.4.4.5: Time series of temperature (blue line) at 1 m (upper plot) and 40 m depth (lower plot)
compared to measurements (red dots) at Station M23 in the north east of the Sea of Marmara.
Page 97
Figure 3.4.4.6: Z,t-diagrams of salinity (upper plot) and temperature (lower plot) for Station M23 in the
north east of the Sea of Marmara. The coloured dots represent measured values.
Figure 3.4.4.7: Z,t-diagrams of salinity (upper plot) and temperature (lower plot) for Station K2 in the
south west of the Black Sea. The coloured dots represent measured values.
Figure 3.4.4.8: Salinity (upper plot) and temperature (lower plot) along a cross section through the
Dardanelles, the Sea of Marmara, the Bosporus and the south east of the Black Sea for January 21, 2003.
The coloured dots represent measurements.
Page 98
Figure 3.4.4.9: Salinity (upper plot) and temperature (lower plot) along a cross section through the
Dardanelles, the Sea of Marmara, the Bosporus and the south east of the Black Sea for July 5, 2003. The
coloured dots represent measurements.
Figure 3.4.4.10: Salinity (upper plot) and temperature (lower plot) along a cross section through the
Bosporus for January 21, 2003. The coloured dots represent measurements.
Conclusions
A first indicator of the model representation is the verification of the barotropic response to space- and
time-varying wind. The water level variation in station Kavak in the northern part of the Bosporus, which
is mainly determined by wind in the Black Sea, is well represented in Figure 3.4.2. The modelled water
levels at Pendik, near the southern entrance of the Bosporus, depend much more on the representation
of the exchange flows through the Bosporus and the water level variation at the open boundary due
to inverse barometer effects. The results presented in Figure 3.4.1, show a good reproduction of these
water level variations. These results show that Delft3D-FLOW is clearly capable of modelling the impact of
space and time varying wind shear stress and atmospheric pressure at the water surface (Claims 2.2.2.12
and 2.2.2.14).
The variation in time of the modelled upper and lower layer fluxes through the Bosporus are presented
in Figures 3.4.3 and 3.4.4 for January 2003 and July 2003, respectively. These fluxes and variations
thereof are induced by variations in water level at both entrances of the Bosporus (barotropic effects) as
well as by a permanent density difference between the two adjacent water bodies for which it forms a
connection (baroclinic effects). The results show that both the upper layer and lower layer fluxes show
a larger variability in January compared to July. This was to be expected given the fact that wind speeds
are generally higher during winter as this is the storm season. From measurements it is known that
blocking events occur in the Bosporus, where one of the layers ceases to exist. These blocking events
are represented in the model. This can be seen in the plots when one of the coloured lines coincides with
the black line representing the net flux, with the other line being at zero level. For January, the model
simulates seven blocking events (five events of blocking of the upper layer and two events of the lower
layer). In July one blocking event (of the upper layer) occurs. Again, the seasonally varying frequency of
blocking shown by the model is to be expected, due to seasonal variations in wind climate. Furthermore,
Page 99
the results show good agreement with the ADCP derived upper and lower fluxes of the January and July
measurements, indicated by dots. From all this, it can be concluded that Delft3D-FLOW is capable of
reproducing wind and density driven (baroclinic) flow (Claims 2.2.1.3, 2.2.1.2 and 2.2.2.2).
Figure 3.4.5 presents the simulated temperature annual cycle against measurement data in Station
23 at a depth of 1 m (upper layer) and 40 m (lower layer). The winter results are very good, as is the
surface temperature increase in spring and early summer. Only, the modelled cooling of the surface layer
lags behind the measured cooling. As in the measurements, the lower layer has a more or less constant
temperature over the year. Comparing the modelled temperatures in the upper and lower layer shows
that in winter, the upper layer is colder than the lower layer (inverse temperature stratification). Only
due to strong salinity stratification does this yields a stable temperature profile. This inverse stratification
is a known natural phenomenon in the Sea of Marmara. These results confirm that Delft3D-FLOW is
capable of correctly modelling heat exchange through the free surface as well as the resulting thermal
stratification (Claims 2.2.1.10 and 2.2.2.15).
In Figure 3.4.6, results in station M23, located in the Sea of Marmara near the southern entrance
of the Bosporus, are presented as z,t-plots. The modelled results presented in this figure show very
good agreement with observations, in terms of both the values of upper and lower layer salinities and
temperatures. Apart from a good agreement with measurements, known natural phenomena like the
forming of the Cold Intermediate Layer during summer, the inverse temperature stratification during
spring and the seasonal vertical variation of the halocline are also represented correctly in the model.
To give an idea of the extent of the spatial variation in vertical temperature and salinity distributions,
the z,t-plot of measurement location K2, located in the Black Sea, near the northern entrance of the
Bosporus, is shown in Figure 3.4.7. The model results at K2 accurately capture the blocking event which
occurs around February/March 2003. In general, the lower layer salinity near the bed in Stations K2
is somewhat too low, however. Also here, a Cold Intermediate Layer exists in both the model results
and the measurements. These results further corroborate the Claim that Delft3D-FLOW is capable of
reproducing thermal stratification and density driven flow (Claims 2.2.1.10 and 2.2.1.2), as well as the
heat exchange through the free surface (Claim 2.2.2.15).
The vertical cross section in Figure 3.4.8 shows again the inverse temperature stratification in January,
whereas in Figure 3.4.9 the modelled Cold Intermediate Layer is clearly visible in July. Figure 3.4.10 is a
zoom-in at the Bosporus in January. In this figure the good agreement between modelled and measured
temperature and salinity profiles along the Bosporus is clearly shown. This holds for both the agreement
with measurements and for the reproduction of the inverse temperature stratification representing the
Cold Intermediate Layer (Claims 2.2.1.10 and 2.2.1.2).
Page 100
Purpose
This validation study involves the modelling of the seasonal, basin-scale temperature cycle of the South
China Sea, including large-scale baroclinic flow, heat exchange through the free surface, lateral exchange
with external systems and thermal stratification. This is examined by using a non-tidal model.
Linked claims
Claim
Claim
Claim
Claim
Claim
Claim
Approach
Model simulations for a climatological year are performed using a non-tidal three-dimensional baroclinic
model representing the South China Sea and Indonesian Archipelago. Focus is on the large-scale,
seasonal temperature cycle due to the availability of synoptic forcing and validation data at this scale.
The model applies a reduced depth approach, in which the depth is truncated at 300 meters depth, which
is below the maximum annual thermocline depth. This reduced depth approach is justified because the
energy exchange to deeper levels is assumed to be negligible, see [Qu, 2001].
The model has 20 equidistant layers in the vertical and includes both temperature and salinity as
transport parameters. Heat exchange at the free surface is determined by the Ocean heat flux model
based on space and time varying atmospheric state input. Exchange with external systems like the Pacific
Ocean is achieved by lateral temperature and salinity forcing and by water level forcing at the open model
boundaries [Twigt et al., 2006].
Model description
124 by 133 horizontal grid points, equidistant with a spacing of 1/4 [Gerritsen et al. 2003].
Reduced depth, truncated at 300 meters with 20 equidistant layers.
Water level boundaries driven by altimetric sea surface anomalies [Gerritsen et al. 2004].
Space and time varying lateral temperature and salinity open boundary forcing.
Space and time varying wind and atmospheric pressure forcing.
Space and time varying heat flux forcing (relative humidity, air temperature, cloudiness).
turbulence model.
Time step of 2 hours.
Results
The heat flux model coefficients and model forcing applied are optimized based on a series of sensitivity
analysis as described in [Twigt et al. 2006]. Figure 3.4.5.1 shows the models surface layer temperature
during the Northeast (NE, January) and Southwest (SW, August) monsoon highs. Model results are
compared with climatological, monthly-mean sea surface temperature data obtained from AVHRR remote
sensing satellites [Vazquez 2004]. Figure 3.4.5.2 shows model temperature profiles compared with profile
data from the World Ocean Atlas 2001 [Levitus 1982] at two stations.
Page 101
Figure 3.4.5.1: Surface layer temperature during the Northeast (NE, January) and Southwest (SW,
August) monsoon highs. Model results compared with monthly-mean, climatological Remotely Sensed
Sea Surface Temperature (RS SST) data obtained from [Vazquez 2004].
Figure 3.4.5.2: Profile data at selected model stations. Model results (lower panel) compared with
monthly-mean, climatological profile data from the World Ocean Atlas 2001 (upper panel) [Levitus 1982].
Conclusions
From the model results in Figure 3.4.5.1, we conclude that the characteristic large-scale surface layer
temperature features as observed from the remotely sensed validation data are resolved by the model
to a reasonable degree. These features include large-scale cold water intrusions through the open
model boundaries, wind driven mixing and baroclinic transport during the NE monsoon. During the
SW monsoon a strong increase in net surface heat flux amounts to higher and more uniform surface
layer temperatures. The results imply that the model resolves the large-scale baroclinic flow, external
lateral forcing and heat exchange through the free surface to a good degree. Model discrepancies, like
higher surface layer temperatures in the Southern South China Sea during the NE monsoon period, are
attributed mainly to the quality of the model forcing data [Twigt et al., 2006].
From model results in Figure 3.4.5.2 it is concluded that the model captures the characteristic cycle of
seasonal stratification to a good degree. This cycle varies for the different regions of the South China
Sea as a result of different forcing processes involved. Station 1 shows results for a station situated
in the Northern South China Sea, near Luzon Strait and Taiwan Strait. The sudden breakdown of the
thermocline around November as observed at this station is explained by cold water influx through the
open model boundaries and large-scale (horizontal) transport of this colder water by baroclinic flow. The
model resolves this to a good degree, implying that both lateral exchange through the open boundaries
and baroclinic transport are resolved. Station 2 shows model results in the central South China Sea,
where the surface heat flux is the main driver in the seasonal stratification cycle. The model captures
this cycle to a good degree, implying that the heat exchange at the free surface is resolved. Results at
Stations 1 and 2 indicate that the model resolves the seasonal thermal stratification to a good degree.
This validation study clearly demonstrates the ability of Delft3D-FLOW to model baroclinic circulation
and thermal stratification in a deep-water region like the South China Sea. To achieve this, free surface
heat flux forcing and external lateral forcing play an essential role. Both features are resolved to a good
degree by the model.
Page 102
X0356, M3470
31 December 2007
Final version 1.0
References
Abbott, M.B., 1997: Range of tidal flow modelling. Special Issue of J. Hydr. Eng., April 1997, ASCE.
Balzano, A., 1998: Evaluation of methods for numerical simulation of wetting and drying in shallow water
models. Coastal Engineering, Vol. 34, 83-107.
Batchelor, G., 1967: An Introduction to Fluid Dynamics. Cambridge University Press, London.
Beji, S. and Battjes, J.A., 1994: Numerical Simulation of Nonlinear Waves Propagation Over a Bank. Coastal
Engineering, 23:1-16. 3.2.1.
Bendat, J.S. and A.G. Piersol, 1971: Random data: analysis and measurement procedures. John Wiley-& Sons,
Inc. New York.
Bijvelds, M.D.J.P., 2003: A non-hydrostatic module for Delft3D-FLOW; technical documentation, verification
and practical aspects. Internal documentation for CDW project.
Bijvelds, M.D.J.P. and A.J. Nolte, 2000: 3D modelling of Lake Veere. Delft Hydraulics report Z2800 (for RIKZ,
in Dutch)
Bijlsma, A.C., 1988: Generation and dissipation of storm surges in the North Sea, with application to the major
storm surges of 1953 and 1983. Delft Hydraulics Report Z96-01.
Bijlsma, A.C., S.A.H. van Schijndel and H.R.A. Jagers, 2003: Computations with Delft3D, in combination with
HLES. Shallow Flows Symposium 2003, Vol III, 213-219.
Bijlsma, A.C., R.E. Uittenbogaard and T. Blokland, 2003: Horizontal large eddy simulation applied to stratified
flows. Proc. Shallow Flows. Delft, 16-18 June, Balkema, Rotterdam.
Bijlsma, A.C., 2007. Numeriek modelonderzoek naar de reductie van de neer in de monding van de voorhaven
van IJmuiden. Delft Hydraulics Report H4926.Bundesanstalt fr Wasserbau, 2002: Mathematical
Model UnTRIM. Validation Document Version 0.3, 61 pp. (see
http://www.hamburg.baw.de/hnm/untrim/hnm_untrim-en.htm).
Dee, D., 1993: A framework for the validation of generic computational models. Delft Hydraulics Report X109.
De Goede, E.D., J.N. Roozekrans, J.M. de Kok and R.J. Vos, 2000: REST3D Remote Sensing sea surface
temperature for 3D North Sea modelling, Netherlands Remote Sensing Board Report NRSP-2-00-16,
BCRS, Delft.
De Goede, E.D. and B. van Maren, 2005: Impacts of Maasvlakte 2 on the Wadden Sea and North Sea coastal
zone. Track 1: Detailed modelling research. Part I: Hydrodynamic modelling. WL | Delft Hydraulics
report Z3945.20. Royal Haskoning report 9R2847.A0.
De Kok, J. M., C. de Valk, J.A.Th.M. van Kester, E.D. de Goede and R.E. Uitenbogaard, 2001: Salinity and
temperature stratification in the Rhine plume. Estuarine, Coastal and Shelf Science, Vol. 53, 467-475.
WL | Delft Hydraulics
41
31 December 2007
Final version 1.0
X0356, M3470
Eckart, C., 1958: Properties of water, Part II. The equation of state of water and sea water at low temperatures and
pressures. American Journal of Science, Vol. 256, pp. 225-240.
EDF-DER, 2000: TELEMAC-2D Validation document version 5.0., Electicit de France, 124 pp.
Forester, C.K., 1979: Higher Order Monotonic Convective Difference Schemes. Journal of Computational Physics,
Vol. 23, pp. 1-22.
Genseberger, M. and J.A.Th.M. van Kester, 2002: Delft3D-FLOW test cases for Z model. WL | Delft Hydraulics
Research Report M3421,.
Gerritsen, H. and A.C. Bijlsma, 1988: Modelling of tidal and wind-driven flow; the Dutch Continental Shelf
Model. In: B.A. Schrefler, ed. Proc. Int. Conf. on Computer Modelling in Ocean Engineering, pp. 331338, A.A. Balkema, Rotterdam.
Gerritsen, H. and G.K. Verboom, 1994: The Coastal Modelling System. pp. 118-128 in: M.B. Abbott, ed. J. of
Hydraulic Research, 1994, Vol. 32.
Gerritsen, H., J.W. de Vries and M.E. Philippart, 1995: The Dutch Continental Shelf Model, in: D.R.Lynch and
A.M.Davies, eds. AGU Coastal and Estuarine Studies, Volume 47, pp. 425-467.
Gerritsen, H., R.J. Vos, Th. van der Kaaij, A. Lane and J.G. Boon, 2000: Suspended sediment modelling in a shelf
sea (North Sea), Coastal Engineering Vol. 41,1-3, 317-352.
Gerritsen, H., E.J.O. Schrama and H.F.P. van den Boogaard, 2003: Tidal model validation of the seas of South
East Asia using altimeter data and adjoint modelling. Proc. XXX IAHR Congress, Thessaloniki, 24-28
August 2003, Vol. D, 239-246.
Gresho, P.M. and R.L. Lee, 1979: Dont suppress the wiggles, theyre telling you something. Finite Element
Methods in Convection Dominated Flows (Ed. T.J.R. Hughes), ASME Winter Annual Meeting.
Hesselink, A. W., G.S. Stelling, J.C.J. Kwadijk and H. Middelkoop, 2003: Inundation of a Dutch river polder,
sensitivity analysis of a physically based inundation model using historic data. Water Resources
Research, Vol. 39, No. 9.
Holthuisen, L.H., R.C. Ris and N. Booij, 1998. A verification of the third-generation wave model SWAN, 5th
International Conference on Wave Hindcasting and Forecasting, Melbourne, Florida, 223-230.
Hulsen, L.J.M., 1989: Region Model: TRISULA-2DH hydrodynamic model WL | Delft Hydraulics report Z358.
IAHR, 1994: Guidelines for Documenting the Validity of Computational Modelling Software. International
Association for Hydraulic Research, P.O. Box 177, 2600 MH Delft, The Netherlands.
Karelse, M., 1996: Validation of Delft3D-FLOW with tidal flume measurements. WL | Delft Hydraulics report
Z810.
Karelse, M. and L.J.M. Hulsen, 1995: Pembroke Power station, marine discharge study, Milford Haven-phase 11,
3D hydrodynamic and water quality model. WL | Delft Hydraulics report Z755.
42
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
Kernkamp, H.W.J. and R.E. Uittenbogaard, 2001: 2D-LES of a free-surface mixing layer. In B.J. Geurts, R.
Friedrich & O. Metais (ed.) Direct and Large-Eddy simulation Workshop, Kluwer Academic
Publishers, pp. 409-418.
Kernkamp, H.W.J., H. Gerritsen and L.J.M. Hulsen, 2004: A spherical curvilinear approach in coastal ocean
modeling typhoon driven storm surges in the South China Sea. Proceedings 6 th International
Conference on Hydrodynamics, Singapore 2004, Vol.1, pp. 586-593.
Kernkamp, H.W.J., H.A.H. Petit, H. Gerritsen and E.D. de Goede, 2005: A unified formulation for the threedimensional shallow water equations using orthogonal co-ordinates: theory and application. Ocean
Dynamics. Vol. 55. pp. 351-369.
Kolmogorov, A.N., 1942: Equations of turbulent motion in incompressible fluid, Izv. Akad. Nauk. SSR, Seria
fizicheska Vi., No.1-2, pp. 56-58 (English translation: 1968 Imperial College, Mech. Eng. Dept. Rept.
ON/6).
Leendertse, J.J., 1990: Turbulence modelling of surface water flow and transport: part IVa. J. Hydr. Eng., Vol.
114, No. 4, 603-606.
Leer, van, B., 1974: Towards the ultimate conservative difference scheme II. Monotonicity and conservation
combined in a second-order scheme. J. Comput. Phys., Vol. 32, 101-136.
Lesser, G., J. van Kester and J.A. Roelvink, 2000. On-line sediment transport within Delft3D-FLOW, Report
Z2899, WL | Delft Hydraulics.
Lesser, G., J.A. Roelvink, J. van Kester and G.S. Stelling, 2004. Development and validation of a threedimensional morphological model. Coastal Engineering, Vol. 51, 883915.
Malvern, L.E. 1969: Introduction to the Mechanics of a Continuous Medium. Prentice-Hall, Englewood Cliffs,
N. J.
Mellor, G.L. and A.F. Blumberg, 1985: Modelling vertical and horizontal diffusities and the sigma coordinate
system, Monthly Weather Review, Vol. 113, 1379-1383.
Murphy, A.H. and E.S. Epstein, 1989: Skill scores and correlation coefficients in model verification. Monthly
Weather Review, Vol. 117, pp. 572581.
Pedlosky, J., 1979: Geophysical Fluid Dynamics. Springer Verlag, New York.
Phillips, N.A., 1957: A coordinate system having some special advantages for numerical forecasting, J. of
Meteorology, vol. 14.
Postma, L., G.S. Stelling and J. Boon, 1999: Three-dimensional water quality and hydrodynamic modelling in Hong
Kong. Stratification and water quality. Proceedings of the 2nd Int. Symp. on Environmental Hydraulics,
Hong Kong, Balkema, Rotterdam. December 1998, 43-49.
Prandtl, L., 1945: ber ein neues Formelsystem fr die ausgebildete Turbulenz, Nachr. Akad. Wiss., Gttingen,
Math.-Phys. Klasse, p. 6.
Ruessink, G. and J.A. Roelvink, 2000. Validation of On-line Mud Transport within Delft3D-FLOW. Report, WL |
Delft Hydraulics.
WL | Delft Hydraulics
43
31 December 2007
Final version 1.0
X0356, M3470
Richardson, J.F. and W.N. Zaki, 1954: Sedimentation and fluidization: part I. Transactions of the Institute of Civil
Engineers, Vol. 32, pp. 35-53.
Ris, R.C., 1997. Spectral Modelling of Wind Waves in Coastal Areas. Ph.D. Dissertation Delft University of
Technology), also Communications on Hydraulic and Geotechnical Engineering, report No. 97-4,
Delft, the Netherlands
Roelvink, J.A., R.E. Uittenbogaard and G.A. Liek, 1999: Morphological modelling of the impact of coastal
structures. Proceedings Coastal Structures 99, Vol.2, pp. 865-871.
Roelvink, J.A., T. van der Kaaij and B.G. Ruessink, 2001: Calibration and verification of large-scale 2D/3D flow
models. Phase 1, Parcel 2, Subproduct 2, ONL Flyland project.
Rodi, W., 1984: Turbulence models and their application in Hydraulics, State-of-the-art paper article sur l'etat de
connaissance. Paper presented by the IAHR-Section on Fundamentals of Division II: Experimental and
Mathematical Fluid Dynamics, The Netherlands.
Schrama, E.J.O., 2002: Availability and quality of global tide models, Delft Hydraulics, report Z3341.
Smits, J., 2000: Modelling framework for water quality management in Lake Victoria (LVMF), rapid assessment
for phase 2 of the development of LVMF, Delft Hydraulics, report Q2652.
Stelling, G.S., 1984: On the construction of computational methods for shallow water flow problems.
Rijkswaterstaat communications, No. 35, The Hague, Rijkswaterstaat.
Stelling, G.S., A.K. Wiersma, J.B.T.M. Willemse, 1986: Practical aspects of accurate tidal computations, J. of
ASCE Hydraulics Engineering, Vol. 9, pp. 802-812.
Stelling, G.S. and J.J. Leendertse, 1991: Approximation of Convective Processes by Cyclic AOI methods,
Proceeding 2nd ASCE Conference on Estuarine and Coastal Modelling, Tampa, pp. 771-782.
Stelling, G.S. and J.A.Th.M. van Kester, 1994: On the approximation of horizontal gradients in sigma co-ordinates
for bathymetry with steep bottom slopes, Int. J. Num. Meth. Fluids, Vol. 18, 915-955.
Stelling, G.S., Duinmeijer, S.P.A., 2003: A numerical method for every Froude number in shallow water flows,
including large scale inundations, Int. J. Num. Meth. in Fluids. Vol. 43, 1329-1354.
Tartinville, B., E. Deleersnijder, P. Lazure, R. Proctor, K.G. Ruddick and R.E. Uittenbogaard, 1998. A coastal
ocean model intercomparison study for a three-dimensional idealised test case. Appl. Math. Modelling
vol. 22, pp. 165-182.
Twigt, D.J., E.D. de Goede, H. Gerritsen and E.J.O. Schrama, 2007: Analysis and modeling of the seasonal
South China Sea temperature cycle using remote sensing, Int. Journal for Ocean Dynamics, Vol. 57,
467-484.
Uijttewaal, W.S.J. and R. Booij, 2000: Effects of shallowness on the development of free-surface mixing layers.
Physics of Fluids, Vol. 12, No. 2, pp. 392-402.
Uittenbogaard, R.E., 1988: Measurement of turbulence fluxes in a stably stratified mixing layer. Proceedings 3rd
Int. Symp. Refined Flow Modelling and Turbulence Measurements, Tokyo, 26-28 July, paper T11-3.
44
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
Uittenbogaard, R.E., 1998: Model for eddy diffusivity and viscosity related to sub-grid velocity and bed
topography. Note, WL | Delft Hydraulics.
Uittenbogaard, R.E., J.A.Th.M. van Kester and G.S. Stelling, 1992: Implementation of three turbulence models in
3D-TRISULA for rectangular grids. Delft Hydraulics, report Z81.
Uittenbogaard, R.E. and J. Imberger, 1993: The importance of internal waves for mixing in a stratified estuarine
tidal flow. Delft Hydraulics, report Z694.
Uittenbogaard, R.E. and B. van Vossen, 2003: Subgrid-scale model for quasi-2D turbulence in shallow water. Proc.
Shallow Flows. Delft, 16-18 June, Balkema, Rotterdam.
Van Banning, G.K.F.M., 1995: Stroomonderzoek sluizen IJmuiden, verslag numeriek modelonderzoek. Delft
Hydraulics Report H2251.
Van der Kaaij, T., 2007: Near- and far-field modelling of bouyant discharges. Delft Hydraulics Report
Z4147.
Van Kester, J.A.Th.M., 1994: Validatie Delft3D voor menglaagproef. Fase 1: Verbeterde implementatie k- model.
Delft Hydraulics Report Z810 (in Dutch).
Van Kester, J.A.Th.M. and G.S. Stelling, 1992: Versnellen van TRISULA-3D. Delft Hydraulics Report Z81.
Van Kester, J.A.Th.M., R.E. Uittenbogaard and G.S. Stelling, 1993: Some verifications of TRISULA-3D with
three turbulence models. Delft Hydraulics report Z81.
Van Rijn, 1993: Principles of sediment transport in rivers, estuaries and coastal seas. Aqua publications,
Amsterdam.
Van Schijndel, S.A.H. and H.R.A. Jagers, 2003: Complex flow around groynes. Proc. Shallow Flows. Delft, 1618 June, Balkema, Rotterdam.
Vatvani, D.K., 2004: Delft3D-FLOW met CDW functionaliteit; eindrapportage. Delft Hydraulics Report M3521.
Vatvani, D.K., H. Gerritsen, G.S. Stelling and A.V.R. Krishna Rao, 2002. Cyclone-induced storm surge and
flood forecasting system for India. In: Proc. Int. Conf. Solutions to coastal disasters, San Diego,
February 2002, pp. 473-487. Reston, VA: ASCE.
Verboom, G.K., J.G. de Ronde and R.P. van Dijk, 1992: A fine grid tidal flow and storm surge model of the
North Sea. Continental Shelf Research, Vol. 12, No. 2/3, pp. 213-233.
Vreugdenhil, C.B., 1994: Numerical methods for shallow-water flow. Water Science and technology library.
Kluwer Academic Publishers.
Wallingford, 200X: HYDRALAB-III JRA1; Validation of numerical models. Technical Note CBS0327/01.
Walstra, D.J.R., J.A. Roelvink and J. Groeneweg, 2000: Calculation of wave-driven currents in a 3D mean flow
model. Proc. of 27th Int. Conf on Coastal Engng., Sydney, pp. 1050-1063.
WL | Delft Hydraulics
45
31 December 2007
Final version 1.0
X0356, M3470
Wang, Z.B., C. Jeuken, H. Gerritsen, H.J. de Vriend and B.A. Kornman, 2002: Morphology and asymmetry of
the vertical tide in the Westerschelde estuary, Continental Shelf Research, Vol/Issue 22/17, p25992609.
Willemse, J.B.T.M., G.S. Stelling and G.K. Verboom, 1986: Solving the shallow water equations with an
orthogonal coordinate transformation. DELFT HYDRAULICS Communication No. 356.
WL | Delft Hydraulics, 1994: Test document Delft3D-MOR. WL | Delft Hydraulics report Z630.
WL | Delft Hydraulics and Rijkswaterstaat, 2001: Standard FORTRAN90 programming rules. OMS report number
2001-01.
WL | Delft Hydraulics, 2007: Delft3D-FLOW. Simulation of multi-dimensional hydrodynamic flows and transport
phenomena, including sediments, Version 3.13. User Manual. Delft Hydraulics. (November 2007).
46
WL | Delft Hydraulics
X0356, M3470
31 December 2007
Final version 1.0
Glossary
Algorithmic implementation: the conversion of the conceptual model into a finite set of
rules suitable for computation. This may involve spatial discretisation schemes, time
integration methods, solution procedures for algebraic equations, decision algorithms, etc.
Conceptual model: a mathematical/logical/verbal representation of a physical system or
process. This representation may involve differential equations, discrete algebraic equations,
decision graphs, or other types of conceptual descriptions.
Computational model: software whose primary function is to model a certain class of
physical systems. The computational model may include pre- and post-processing features, a
user interface, and other ancillary programmes necessary in order to use the model in
applications. However, this validation document primarily concerns the core of the
computational model, consisting of the underlying conceptual model, its algorithmic
implementation and software implementation.
Software implementation: the conversion of the algorithmic implementation into a
computer code. This includes coding of algorithms, use of standard mathematical software,
design and implementation of data structures, etc. The term software implementation, for the
purposes of this document, is limited to the computational core of the model. It does not
include pre- and post-processing software, user interfaces, or other ancillary programmes
associated with the computational model.
WL | Delft Hydraulics
51