U Un ni iv ve er rs si it t d de eg gl li i S St tu ud di i d di i C Ca ag gl li ia ar ri i

F Fa ac co ol lt t d di i I In ng ge eg gn ne er ri ia a
D D. .I I. .M Me e. .C Ca a. .
D Do ot tt to or ra at to o d di i R Ri ic ce er rc ca a i in n I In ng ge eg gn ne er ri ia a I In nd du us st tr ri ia al le e
X XI IX X C Ci ic cl lo o



N Nu um me er ri ic ca al l S Si im mu ul la at ti io on ns s o of f t th he e
A At tm mo os sp ph he er ri ic c B Bo ou un nd da ar ry y L La ay ye er r



G Gi io or rg gi io o C Cr ra as st to o










T TU UT TO OR R
P Pr ro of f. . P Pi ie er rp pa ao ol lo o P Pu ud dd du u
C CO OO OR RD DI IN NA AT TO OR RE E D DE EL L C CI IC CL LO O
P Pr ro of f. .s ss sa a A Al le es ss sa an nd dr ra a F Fa an nn ni i

Cagliari, Febbraio 2007


I
TABLE OF CONTENTS
TABLE OF CONTENTS ............................................................................................... I
RINGRAZIAMENTI (ANKNOWLODGEMENTS)................................................ IV
NOMENCLATURE ......................................................................................................V
ABSTRACT ................................................................................................................... 1
CHAPTER 1 THE ATMOSPHERIC BOUNDARY LAYER ................................... 4
1.1 THE ATMOSPHERIC BOUNDARY LAYER (ABL)............................................... 4
1.2 VELOCITY VERTICAL PROFILE IN ABL ............................................................ 4
1.3 SIMILARITY THEORY ....................................................................................... 7
1.3.1 Introduction............................................................................................ 7
1.3.2 Vertical profiles of velocity .................................................................... 7
1.3.3 Vertical profiles of second moments of velocities (Reynolds stresses) ... 9
CHAPTER 2 CFD MODELLING OF THE ABL .................................................... 11
2.1 THE NAVIER-STOKES EQUATIONS................................................................. 11
2.2 TURBULENCE MODELLING............................................................................ 12
2.3 THE RANS EQUATIONS................................................................................. 14
2.4 EDDY VISCOSITY MODELS............................................................................. 17
2.4.1 A one equation turbulence model: the Spalart-Allmaras (SA) model... 17
2.4.2 Two equations turbulence models ........................................................ 19
2.5 EDDIES SOLVING METHODOLOGIES: LES AND DES ...................................... 23
2.5.1 Filtered Navier-Stokes equations ......................................................... 23
2.5.2 Sub Filter Scale (SFS) modelling ......................................................... 25
2.5.3 Hybrid RANS/LES models. The SA based DES model.......................... 27
2.6 WALL FUNCTIONS FOR ROUGH SURFACES ..................................................... 28
2.7 CONCLUSIONS............................................................................................... 30
CHAPTER 3 2D RANS SIMULATIONS.................................................................. 33
3.1 SETTINGS OF ROUGHNESS PARAMETERS........................................................ 33


II
3.2 GEOMETRY AND GRIDS.................................................................................. 34
3.3 BOUNDARY CONDITIONS ............................................................................... 34
3.4 RESULTS: VELOCITY PROFILES....................................................................... 36
3.5 COMPARISON OF TURBULENCE MODELS ........................................................ 39
3.5.1 Profiles observed at the outlet .............................................................. 41
3.6 2D SIMULATIONS ON AN ISOLATED HILL........................................................ 42
3.6.1 Choice of horizontal refinement............................................................ 47
3.6.2 Choice of discretization scheme............................................................ 52
CHAPTER 4 3D RANS SIMULATIONS OVER AN ISOLATED HILL............... 55
4.1 INTRODUCTION.............................................................................................. 55
4.2 SIMULATIONS ON ASKERVEIN HILL ............................................................... 57
4.2.1 General description .............................................................................. 57
4.2.2 Measurement reference data................................................................. 59
4.2.3 Simulations with Fluent ........................................................................ 65
4.2.4 Simulations with WindSim.................................................................... 74
CHAPTER 5 SIMULATIONS 3D DES/LES OVER FLAT TERRAINS* ............. 83
5.1 INTRODUCTION.............................................................................................. 83
5.2 STUDY OF THE ABL WITH LES TECHNIQUES................................................. 84
5.3 METHODOLOGY FOR THE DES/LES SIMULATIONS OF THE ABL ................... 85
5.3.1 Initial steady simulations ...................................................................... 85
5.3.2 Turbulence modeling ............................................................................ 90
5.3.3 The new LES-UDF implemented in Fluent ........................................... 91
5.3.4 DES settings tested................................................................................ 92
5.4 2D TESTS ....................................................................................................... 94
5.4.1 2D boundary layers on smooth walls.................................................... 95
5.4.2 2D boundary layers on rough walls.................................................... 102
5.5 SIMULATIONS DES 3D................................................................................ 104
5.5.1 Wall-shear stress ................................................................................ 107
5.5.2 Velocity profiles .................................................................................. 110
5.5.3 Eddy-viscosity profiles........................................................................ 112


III
5.5.4 Velocity components variances profiles ............................................. 118
5.5.5 Turbulent Kinetic Energy profiles ...................................................... 122
5.5.6 Comparison with literature ................................................................ 124
5.6 SIMULATIONS LES 3D OVER A ROUGH FLAT TERRAIN ................................ 132
5.6.1 A new LES-UDF to use with wall-functions ....................................... 132
5.6.2 Wall-shear stress ................................................................................ 133
5.6.3 Velocity profiles.................................................................................. 135
5.6.4 Eddy viscosity profiles........................................................................ 136
5.6.5 Velocity variances profiles ................................................................. 136
5.6.6 Turbulent Kinetic Energy vertical profiles ......................................... 136
5.7 CONCLUSIONS............................................................................................. 146
CHAPTER 6 FOREST MODELLING WITH WINDSIM.................................... 149
6.1 INTRODUCTION ........................................................................................... 149
6.2 POROSITY.................................................................................................... 151
6.3 MOMENTUM SINKS...................................................................................... 152
6.4 A CANOPY MODEL FOR WINDSIM................................................................ 153
6.5 WINDSIM SIMULATIONS.............................................................................. 155
CONCLUSIONS........................................................................................................ 165
REFERENCES .......................................................................................................... 168
APPENDIX A. MODIFICATION OF SKE MODELS CONSTANTS. .............. 173
APPENDIX B. DERIVATION OF FORMULA (2.33)........................................... 176
APPENDIX C. FURTHER 2D DES CASES ON ROUGH WALLS..................... 177
APPENDIX D. LES-UDF TEXT.............................................................................. 180
APPENDIX E. COMPUTATION OF DRAG COEFFCIENTS C1 OF TABLE 6.7
..................................................................................................................................... 182



IV
RINGRAZIAMENTI (ANKNOWLODGEMENTS)
Vorrei ringraziare il gruppo di lavoro di cui ho fatto parte nel dipartimento di meccanica
dellUniversit di Cagliari per il supporto datomi in questi anni. Prima di tutto Prof. Mandas che
mi ha introdotto nel campo della fluidodinamica sperimentale e dellenergia eolica, sin dai tempi
dellultimo anno del corso di laurea e come relatore del lavoro di tesi di laurea.
Sono particolarmente riconoscente nei confronti dellIng. Cambuli, che mi ha strettamente
seguito durante il lavoro svolto. I mei colleghi, in particolare Giorgio Melis, Carlo Enrico
Carcangiu, Roberto Fuliotto e Marco Pau, ovvero il CFD Group, con cui ho avuto la fortuna e il
piacere di lavorare ma anche di condividere momenti importanti di questi anni. Ringrazio inoltre
il mio tutor, il Prof. Puddu, sempre pronto ad ascoltarer e dare preziosi consigli in merito alle
attivit di dottorato.
Fase intensa delle attivit svolte stata durante il Diplome Course 2005-06 presso il von
Karman Institute. Vorrei ringraziare tutti i giovani ingegneri che anno preso parte al corso e
soprattutto quelli che mi hanno sostenuto in maniera fraterna e con cui ho convissuto questi
intensi mesi: Marco Pau (ancora lui!), Emmett Dempsey, Javier Gonzalez Castano, Julien Bodart,
Domenico Verrastro, Filippo Coletti, Flora Tomasoni, Anna Meller, Valentina Taviani, Mike
Bilka, Gkhan Balik e inoltre Alessandro Gambale e Raf Theunissen.
Il lavoro presso il VKI non sarebbe stato inoltre possibile senza la supervisione del Prof.
Carlo Benocci e di Patrick Rambaud.
I would like to thanks the cheerful and welcoming people I found in Norway at the Vector
A.S.. Arne Reidar Gravdahl, Tine Vlstad and her husband Asbjrn with the kids, and Andrea
Vignaroli, my housemate and colleague. All of them made the permanence in Norway pleasant
and unforgettable. I wish to thank also Arancha Pera Gilaberte that I met only for few days in
Norway but that showed to be a very good friend in the following months.
I wish to thank also Prof. Peter Taylor and collaborators who provided the experimental
data of the Askervein hill project with the orography maps.

Un particolare ringraziamento va alla mia famiglia: Carlo, Paola ed i miei genitori Antonio e
Giuseppina che continuano a sostenermi giornalmente e a credere in certe mie presunte
potenzialit.


V
NOMENCLATURE
Roman
B constant (~5,45) in the logarithmic law, equation (2.31)
C proportionality constant, equation (1.2)
C
w
, C
w2
, C
w3
, C
b1
, C
b2
, C
prod
parameters of the SA model
C
S
Smagorinsky constant; roughness constant in modeling of
boundary layer over rough surfaces
C

, C
1
, C
2
k- models constants
C
DES
DES models constant
C1 [s
-1
] drag coefficient of momentum source term proportional to
velocity
C2 [m
-1
] drag coefficient of momentum source term proportional to
the square of velocity
d

[m] modified wall distance in SA based DES model, equation

(2.29)
d [m] wall distance; displacement length in forest modeling
according to equation (6.2)
d
f
[m] displacement length of the florest floor according to
sketch of Figure 6.1
E [-] constant (= exp(B) 9,81)
f Coriolis parameter 2sin
f
i
, F
i
[m s
-2
] specific force in x
i
direction
v1 v2
f ,f , f
w
parameters of the SA model
k [m
2
s
-2
] turbulent kinetic energy (TKE)
g [-] parameter of the SA model
G filter function

G [kg m
-1
s
-2
] production term (SA model) in equation (2.8)
k b
G , G [kg m
-1
s
-3
] production of TKE due to the mean velocity gradients
and buoyancy


VI
G

[kg m
-3
s
-2
] production term of in the transport equation (k-
model), equation (2.13)
h [m] height of wall adjacent cell
h , h
C
[m] height of the canopy
H [m] height of an isolated hill
K
S
+
normalized
K
S
[m] roughness elements size (equivalent sand )
S
[m] length scale in the eddy viscosity models
L [m] half width of the hill measured at half height H/2
L
S
[m] mixing length in the SGS eddy viscosity definition,
equation (2.24)
L
x
, L
y
[m] length of the computational domain in x, y directions
N total number of cells in a computational grid
N
x
, N
y
number of subdivisions of a computational grid in the x
and y directions
p [Pa] space filtered pressure
r parameter of the SA model
S [s
-1
] module of the rate-of-strain tensor, equation (2.26)
S

[kg m
-3
s
-2
] source term of in the transport equation (k-
model), equation (2.13)

S [s
-1
] modified S (SA model)
ij
S [s
-1
] rate-of-strain tensor for resolved scale, equation (2.25)
t [s] time
u*, u

[m s
-1
] friction velocity
u
i
[m s
-1
] velocity along the Cartesian coordinate x
i

u,v,w [m s
-1
] velocities along the Cartesian axes x, y, z
u
+
, y
+
wall coordinates
U, V, W [m s
-1
] time averaged velocities along the Cartesian axes x, y, z
V [m
3
] volume of a general cell
S
V [m s
-1
] velocity scale for eddy-viscosity models
x
i
[m] Cartesian coordinates


VII
x, y, z [m] Cartesian coordinates
x [m] position vector
Y

[kg m
-3
s
-2
] dissipation term of in the transport equation (k-
model), equation (2.13)

Y [kg m
-1
s
-2
] destruction term of , (SA model) equation (2.8)
z
0
[m] roughness length
z
0f
[m] roughness length of the forest floor, according to sketch in
Figure 6.1
z
h
[m] gradients height (or height of the ABL)
Greek
expansion ratio in a computational grid
[m
2
] permeability
* k- models parameter
volume porosity

S
surface porosity

k
,

diffusive coefficient for k and in the k- model

ij
Kronecker delta
[m] length scale used in the modification of wall distance d

B shift of u
+
in the modified logarithmic law for rough
surfaces (2.32)
x, y, z [m] resolution of a mesh in the x, y, z directions
[m
2
s
-3
] dissipation rate of turbulent kinetic energy (TDR)
von Karman constant (~0,42)
[Pa s] dynamic viscosity

t
[Pa s] turbulent (or eddy) dynamic viscosity, SGS turbulent (or
eddy) dynamic viscosity
t sgs
[m
2
s
-1
] SGS dynamic eddy-viscosity
[m
2
s
-1
] kinematic viscosity
[m
2
s
-1
] modified eddy viscosity in the SA model
t sgs
[m
2
s
-1
] SGS kinematic eddy-viscosity


VIII
[m] position vector
[kg m
-3
] air density
[] generic variable, latitude

m
[-] normalized wind shear, defined in equation (1.4)
[] time averaged generic variable (x,t), spatial filtered
generic variable (x,t)

[-] constant diffusive term (SA model), equation (2.8)

k
, [-] turbulent Prandtl numbers for k and (k- model)

k
, [-] turbulent Prandtl numbers for k and (k- model)

w
[Pa] wall shear stresses

ij
[Pa] SGS shear stresses, equation (2.22)
[s
-1
] specific dissipation rate
[-] ratio modified eddy to laminar viscosity (SA model)
[s
-1
] computational domain, hearths rotational speed

ij
[s
-1
] rotation rate tensor
Special notations
(primes) fluctuating component in a Reynolds decomposition
(overbar) time averaged (or space filtered) component
space averaging over horizontal plane
Acronyms
ABL Atmospheric Boundary Layer
AEP Annual Energy Production
a.g.l. above ground level
CFD Computational Fluid Dynamics
CP Center Point (Askervein hill topography)
CWE Computational Wind Engineering
DES Detached Eddy Simulation
DNS Direct Numerical Simulation
HT Hill Top (Askervein hill topography)
KE k- model


IX
LES Large Eddy Simulation
NS Navier Stokes equations
PBL Planetary Boundary Layer
PDE Partial Differential Equation(s)
RANS Reynolds Averaged Navier Stokes
RC Roughness Constant (Fluents parameter for rough walls)
RH Roughness Height (Fluents parameter for rough walls)
RKE Realizable k- model
RNG Re-Normalization Group
RS Reference Site (Askervein hill topography)
RSM Reynolds Stress Model
SA Spalart Allmaras
SFS Sub Filter Scale
SGS Sub Grid Scale
SKE Standard k- model
SL Smagorinsky-Lilly
SST Shear Stress Transport
TDR Turbulence Dissipation Rate
TKE Turbulent Kinetic Energy
UDF User Defined Function
URANS Unsteady Reynolds Averaged Navier Stokes
WALE Wall-Adapting Local Eddy-viscosity (SGS model)
WMLES Wall Modeled LES








1
ABSTRACT
Nowadays there is a wide public interest for the exploitation of renewable
energies sources and energy saving policies, the European Union is clearly
aiming at increasing the production of energy from renewable sources and, at
the same time, promoting among the European citizens an attention to avoid the
waste of energy.
The development of renewable energies, particularly energy from wind,
water, solar power and biomass, is a central task of the European Union's
energy policy. There are several reasons for this orientation:
Renewable energy has an important role to play in reducing Carbon
Dioxide (CO
2
) emissions, a major European Community objective.
Increasing the share of renewable energy in the energy balance
enhances sustainability.
It also helps to improve the security of energy supply by reducing the
European Community's dependence on imported energy sources, mainly oil and
natural gas.
Renewable energy sources are expected to be economically competitive
with conventional energy sources in the medium to long term.
Several renewable energy technologies, especially wind energy, have
achieved a level of maturity which allows them to compete to the standard
energy sources even without the help of economic incentives as green
certificates.

The numerical simulations of the Atmospheric Boundary Layer (ABL)
afforded in the present PhD thesis can be seen as part of a more general
framework of the Computational Wind Engineering (CWE). The increase of
computational capacities for commonly used PCs and workstations and the
relatively high and stable price of wind tunnel facilities guide nowadays the
wind engineering to a wider usage of computational resources rather than
experimental facilities.


2
Although the increased importance of the CWE, the wind-tunnel and on-site
activities keep on being complementary and not alternative to computations as
can be stated for most of the engineering branches.
The numerical simulations of the ABL are applied to the evaluation of the
wind loads on structures, pollutants dispersion, the natural ventilation of
buildings, the assessment of the pedestrian comfort in a urban environment, the
sails aerodynamics of sails and many other topics.
For instance, CFD computations on given terrains coupled with the
meteorological data are the starting point for a business plan of a wind farm.
Some numerical tools able to predict the Annual Energy Production (AEP)
of a wind farm basing their analysis upon empirical relations are spread in the
wind engineering community. The empirical relations commonly used for the
computation of the speed-up due to the topography are prone to fail or loose
accuracy when dealing with complex terrains.
A terrain can be considered complex when its steepness exceeds a certain
threshold and above this level the empirical relations dont guarantee any
success. In fact, in the flow around steep terrains, hills separation bubbles are
very likely to occur and the level of turbulence is supposed to be quite high,
furthermore also unsteady phenomena like shedding of vortices can occur. This
kind of flows lead the commonly used empirical models to fail and the
numerical solution of the RANS equations remains therefore only analysis able
to reproduce the correct behaviour of the flow.
The present thesis focuses on the application of two commercial finite-
volume CFD codes for the numerical simulations of the ABL flow. The first
one is the more general finite volume solver Fluent and the second one,
WindSim, is devoted to wind energy assessment using the general RANS finite-
volume solver PHOENICS.
After a description of some topics regarding part of the physics of the ABL
discussed on the Chapter 1, the following Chapter 2 treats the modelling of the
ABL with finite volume solvers by integration of the Navier-Stokes equations
modified for turbulence modelling.


3
Numerical simulations of the ABL are presented in the following sections:
in Chapter 3 2D simulations while 3D simulations over an isolated hill
(Askervein hill) are described in the Chapter 4; the test case of Askervein has
been chosen since the wide range of anemometric data, gathered during two
experimental campaigns and published in the two reports by Taylor and
Teunissen (1985, 1986), [36] and [37].
A particular attention has been paid on the application of different
turbulence models. Numerical results have been compared to data from
Askervein Hill experimental campaign. The RANS equations have been solved
with the k- (standard, RNG and realizable), k- and RSM when the code
Fluent has been used, while only the standard k- model has been used with the
code WindSim when solving the Askervein Hill case.
More advanced techniques as DES and LES have been also considered
applied on flat terrains flow simulations, using the commercial code Fluent.
This part of the research project has been carried out in collaboration with von
Karman Institute for fluid dynamics during the 2005-06 Diploma Course,
therefore the Chapter 5 is a modification of the final report for the von Karman
Institute.
The last chapter of the PhD thesis is devoted to the description of a canopy
model implemented in WindSim in order to emulate the flow over a forested
area. A new canopy model has been proposed, based on the flow through
porous media and the simulations reported in the chapter are the results of the
three months training (april-june 2005) in the Vector A.S., Norway, and of the
following and still active collaboration between the author and the developers of
the code WindSim.


4
Chapter 1 THE ATMOSPHERIC BOUNDARY LAYER
1.1 THE ATMOSPHERIC BOUNDARY LAYER (ABL)
The Atmospheric Boundary Layer (ABL), or Planetary Boundary Layer
(PBL), can be described like the lowest 1-2 km of the atmosphere, the region
most directly influenced by the exchange of momentum, heat, and water vapour
at the earths surface, Kaimal and Finnigan (1994) [17].
Stull (1988) [33] instead defines the boundary layer as that part of the
atmosphere that is directly influenced by the presence of the earths surface, and
responds to surface forcings with a timescale of about an hour or less.
Moreover Stull (1988) [33] mentions several forcings perturbing the boundary
layer, i.e. frictional drag, evaporation, heat and pollutants transfer,
modifications of the flow due to topography. According to Stull (1988) [33] the
ABL depth ranges from one hundred metres to few kilometres, depending on
the forcings.
As regards the flows and phenomena inside an ABL a typical time scale is
one hour (as already stated) while typical space scales are few kilometres; these
scales are important to be also accounted for when numerical simulations have
to be performed.
1.2 VELOCITY VERTICAL PROFILE IN ABL
A neutral ABL, where heat transfer is negligible, can be subdivided into
several sub-layers, as sketched in Figure 1.1: a canopy layer attached to the
ground surface, where the obstacles constituting the roughness elements are
displaced; above the canopy layer there is the surface layer, where the effects of
Coriolis force are still negligible. It is normally considered extending up to one
tenth of the total height of the ABL; the outer layer of the ABL, or Ekman layer,
which is instead affected by the rotation of the Earth by means of the Coriolis
force. A rotation of the wind direction is therefore observed, gradually passing


5
from a crossing-isobars wind in the boundary layer to a parallel isobars wind
(Geostrophic wind) in the free atmosphere where frictional forces are
negligible.
The crossing isobars flow and the Geostrophic wind are due to the balance
of friction, pressure gradient and Coriolis forces, in the surface layer the only
important forces are friction and pressure gradient therefore its not observed a
significant variation of horizontal wind direction within it.
Geostrophic wind and wind in the Ekman layer are sketched in Figure 1.2.

Canopy
layer

Figure 1.1 Subdivision of the ABL (or PBL) into further sub-layers.

Figure 1.2 Geostrophic wind and wind in the Ekman layer.
ABL
Ekman
layer
Gradient height
Free atmosphere
H
e
i
g
h
t

z

[
m
]

Surface layer


6
Theoretical considerations (Tennekes, 1982 [38]) lead to assume the
boundary layer to have a depth proportional to u

/f, where u

is the friction
velocity defined in (1.1) and f is the Coriolis parameter 2sin, being the
hearths rotational speed and the latitude. The ABL height, in the case of
neutral stratification, can be expressed by the (1.2).
w

=
(1.1)

| |
=
|
\
h
u
z C
f
(1.2)

A value of C = 0.25 yields boundary layer heights close to observed
daytime heights.
If Coriolis, friction and pressure gradients are responsible for the wind flow
in the outer layer and free atmosphere, in the surface layer the Coriolis force
looses its importance while the roughness of the ground becomes a more
significant parameter, affecting both the velocity profile and the angle of
incidence of wind at the ground level and the isobars.
The roughness of the terrain influences the depth of the ABL as its
sketched in Figure 1.3, the rougher the terrain and the higher the ABL.

Figure 1.3 different velocity profiles for four kinds of terrain typologies.


7
In Figure 1.3 the roughness of the terrain is described by the parameter z
0

which is called aerodynamic roughness length, or just roughness length, whose
meaning will be introduced in the successive paragraph on similarity theory.
1.3 SIMILARITY THEORY
1.3.1 Introduction
Similarity theory provides a way to organize and group the variables of
interest in dimensionless groups. This procedure is helpful to a better
understanding of fluid-dynamics phenomena and provides guidelines for correct
scaling in the experimental facilities.
A proper choice of the dimensionless groups should allow to establish
universal relationships between the groups, valid in each condition, that can
be represented in graphs, numerical tables or regression curves.

Several kind of similarities approaches have been proposed in literature for
the ABL. In this paragraph only few relations are reported together with a
similarity theory known also as Monin-Obukhov theory or surface-layer
similarity so called since it is valid in the surface layer. For further information
on similarity theories on the ABL it is suggested to refer to Stull (1988) [33].
1.3.2 Vertical profiles of velocity
A similarity study can be carried out to describe vertical profiles of
turbulence statistics in the ABL when fully developed conditions (and therefore
horizontal homogeneity) are reached. In the surface layer the mean horizontal
velocity is commonly described by a log profile in neutral conditions (1.3). A
neutral condition stands when thermal effects are negligible.

| |
=
|
\ 0
U(z) 1 z
ln
u z
(1.3)
In the equation (1.3) the mean horizontal velocity U is normalised it by the
friction velocity u

, defined by (1.1), while the height z is compared to the

8
roughness length z
0
(the height where the mean velocity would go to zero
according always to (1.3)) and is the von Karman constant (~0.4).
The typologies of terrain can be classified by their roughness length z
0
,
according to the classification proposed in the European Wind Atlas [39].
Table 1.1 Classification of terrains by their roughness length, partially adapted from
the European Wind Atlas (1989) [39]
class terrains z
0
[m]
Muddy terrains, wetlands, icepack 10
-5
3 10
-5

I
Water areas * 3 10
-5
0,0002
II Sand 0,0002 0,001
III Airport runway areas, mown grass 0,001 0,01
IV
Farmland/Airports with very few
trees, buildings, etc.
0,01 0,04
V Many trees and/or bushes 0,04 0,1
VI Forests, suburbs 0,1 1
VII Cities 1 4
* Air and sea form a dynamically coupled system, the determination of the roughness
length of open sea and water surfaces is usually obtained by models taking into
account shapes and dimensions of the waves.

The first derivative of the mean horizontal velocity is referred as the wind
shear, which can be combined with , z and u

to build a dimensionless group

m
(1.4), which in neutral stability conditions equals one. In these conditions it
leads to the log-law (1.3) while in non-neutral conditions it is generally
expressed as a function of the dimensionless height z/L, where L is the Monin-
Obukhov length.


9

m
z U
u z
(1.4)
1.3.3 Vertical profiles of second moments of velocities (Reynolds stresses)
Different vertical profiles for higher moments of the velocities have been
also proposed, which pretend to be universal. According to Stull (1988) [33],
the turbulent momentum flux for a neutral boundary layer can be considered
decreasing linearly with the height z.
( )
=
h
S
u' w' z
1
z u' w'
(1.5)
For the variances of the velocities, Stull (1988) [33]:

| |
= +
|
\
2
2 2
top
2 2
h h
u' z z u'
6 1
u z z u
for stream direction (1.6)

| |
= +
|
\
2
2 2
top
2 2
h h
v' z z v'
3 1
u z z u
for span direction (1.7)

| |
=
|
\
1/ 2
2
2
h
w' z
1
u z
for vertical direction (1.8)
As can be noticed in these expressions for the momentum flux and variances
of velocity components (1.6)-(1.8) the altitude z is normalized against the depth
of the boundary layer z
i
rather than with the roughness length z
0
.
In the expressions for the variances of the horizontal velocities (1.6) and
(1.7) the values at the top of the boundary (assuming known a well-defined top
height z
i
), normalized by the surface stress, do appear. Although this ratio is
expected to vary, during the KONTUR experiment (Grant, 1986) [14] it was
found to equal 2.0 for both equations.
Taking into account the equations (1.6) to (1.8) and the definition of TKE
(1.9) it follows also a vertical profile for the normalised TKE (1.10).



10
( )
2 2 2
1
k u v w
2
= + + (1.9)
2 2 2
2 2 2 2
2 2 1/ 2
2 2
2 2
1
2
1 ' '
6 1 3 1 1
2


| |

= + + = |
|
\
| |
| | | | | |
| = + + + +
| | |
|
\ \ \
\
top top
h h h h h
k u v w
u u u u
z z u z z v z
z z u z z u z

2 1/ 2
2 2 2
2 2 2 2
1 1
9 1 4 1
2 2

| |
| |
| | | |
| = + + = + + |
| |
|
|
\ \
\
\
h h h
k u v w z z z
u u u u z z z
(1.10)


11
Chapter 2 CFD MODELLING OF THE ABL
2.1 THE NAVIER-STOKES EQUATIONS
The equations ruling the fluid mechanics are the mathematical expressions
of three physical principles:
the conservation of mass;
the conservation of momentum;
the conservation of energy.
Two of the mentioned principles (conservation of mass and energy) lead to
two scalar expressions while the conservation of momentum turns into a vector
one. The primitive equations of fluid mechanics that are always valid can be
furthermore simplified; if the fluid is Newtonian and the flow incompressible
only the continuity (mass conservation) and momentum equations can be
considered, also called Navier-Stokes equations from the scientists that firstly
derived them. The Navier-Stokes equations can be written in different forms
and notations; they can be expressed, using the summation notation, by:
conservation of mass (continuity equation),
( )
i
i
u 0
x

(2.1)
conservation of momentum (momentum equation),
i i
i k i
k i k k
u u p
u u f
t x x x x

| | | |
+ = + +
| |

\ \
(2.2)
The equations of motion can be written in four different forms: in
conservative (Eulerian reference frame) or non-conservative form (Lagrangian
reference frame), in integral or differential form. When a particular flow has to
be studied the equations of motion have to be integrated with given boundary
conditions. In most of the cases the equations of motion cant be solved
analytically and numerical solutions remain the only possible way.


12
From the purely theoretical point of view, the four forms are equivalent but
they are not from the numerical point of view since some numerical procedures
are referred to particular forms of the governing equations. For instance, in the
finite-volume technique the Navier-Stokes equations are considered in their
conservative integral form and applied to control volumes usually called cells
that cover the whole computational domain.
The present thesis deals with the numerical simulations of the neutral ABL
in micro-scale domains, which means dimensions of few kilometres; in such
kind of flows both compressibility and thermal effects are negligible.
What instead is not negligible and plays an important role in an ABL is the
turbulence. Turbulent flows can be solved directly in simulations named DNS
which are affordable nowadays only to study very low Reynolds number cases
and only by means of high performance computers. In a typical ABL the
Reynolds number ranges between 10
5
and 10
10
making these kind of flows
affordable by time-averaged Navier-Stokes equations (RANS) and, in smaller
domains, also with Wall Modelled Large Eddy Simulation (WMLES) and
hybrid methodologies RANS/LES as the Detached Eddy Simulation (DES). All
these methodologies will be described in the following sections dedicated to
turbulence modelling.
2.2 TURBULENCE MODELLING
A classical approach in the study of turbulence is carried out by means of
the Reynolds decomposition: the variables fluctuating in a turbulent flow can be
decomposed into a mean and a fluctuating term.
The time-averaged (or Reynolds averaged) Navier-Stokes equations, shortly
RANS, are obtained applying the Reynolds decomposition to the unknowns
appearing in the Navier-Stokes equations and time averaging. The RANS
contain further unknowns called Reynolds stresses which are due to all scales
turbulence and need to be modelled in order to close the set of equations. The
biggest advantage of using the RANS in simulating turbulence flows is that they
allow to treat the turbulence as a steady phenomena, i.e. if a flow is unsteady


13
only because of turbulence the RANS simulation will be a steady simulation,
with great saving of computational means.
The RANS based turbulence models are usually classified by the number of
additional differential equations needed to close the original set of PDE Some
of turbulence models used to close the RANS are:

algebraic or zero equation model: mixing length;
one equation model: Spalart Allmaras
two equations model: k- (e.g. standard, RNG, realizable)
two equations model: k- (e.g. standard, SST)
five (2D) and seven equations (3D): RSM (Reynolds Stress Model)

A second approach to the turbulence modelling which is gaining more and
more applications is the Large Eddy Simulation (LES), which is based on the
space filtered Navier-Stokes equations.
The finite volume technique is equivalent to apply a space filter to the
Navier-Stokes equations; in this fashion only the big eddies whose dimensions
are larger than the filter width, which is connected to the cells dimensions, are
computed directly. The non resolvable eddies or sub-filter eddies are modelled
with specific turbulence models named Sub Grid Scale (SGS) or Sub Filter
Scale (SFS) turbulence models.
Due to the fact that the biggest eddies have to be calculated directly, a LES
simulation can be only 3D and unsteady. Even if in some particular cases, for
instance for some external aerodynamics, good solutions can be obtained in 2D
but always with unsteady simulations.
Its well known that RANS turbulence models are characterised by few
constants that have to be tuned to best simulate the given problem. In some way
the RANS turbulence models are all problem dependent and this is one big
defect of the RANS models. Big eddies are strongly anisotropic and depend
upon boundary conditions and mean flow; considering instead smaller and
smaller eddies they loose information about mean flow and boundary


14
conditions, small vortices are also more homogeneously spread and more
isotropic. There is hope that small scale vortices are universal and that SGS
models can work for every kind of turbulence if the filter width is applied at the
proper size.
The LES still requires the usage of supercomputing for high Reynolds
numbers and the main reason of that is the fine grid needed to discretize the
near wall region in wall bounded flows. When the walls are smooth and pure
LES is desired than the constraints for the resolution of the grid for the wall
adjacent cells is about 100 wall units streamwise and 20 spanwise, while only
one wall unit normally to the wall. Among the methodologies born to treat high
Reynolds numbers the hybrid LES-RANS named Detached Eddy Simulation
(DES) have to be mentioned. With these techniques the near wall region is
modelled with a RANS approach and the zone distant from the wall the flow is
treated with a LES approach. The DES is constructed by modifying the usual
RANS model (SA, k-, k-) which acts in the standard way close to the wall
and in a modified one at a certain distance from the wall, as a SGS model. The
DES was born to deal with external aerodynamics issues, for instance to
compute the flow around an airfoil in stall conditions and the model is
conceived to treat the attached boundary layer with the URANS methodology
and the detached vortices with LES.
2.3 THE RANS EQUATIONS
The turbulent flows can be treat as steady if the flow statistics remain
constant when calculated over a period T large enough; in this case the flow is
called statistically steady.
The Navier-Stokes equation can be time averaged over a period T large
enough to mean the turbulent fluctuations in order to obtain the Reynolds
Averaged Navier-Stokes (RANS) equations.
A generic variable, fluctuating because of the turbulence, can be
decomposed in a mean part and a fluctuating part according to the Reynolds
decomposition.


15

(t)= +(t) Reynolds decomposition

In the present thesis both overbar and capital letter notation have been used
to identify a time averaged variable.
Where the mean value is obtained by:
T
0
1
(t)dt
T
=

(2.3)

And the integration time must be such that the average of the fluctuating
component is zero.
T
0
1
' '(t)dt 0
T
= =

(2.4)

The integration time T over which a general variable is observed in order
to smooth the fluctuations due to turbulence is comparable to several turn-over
periods of the vortices.
After applying the Reynolds decomposition and time averaging the NS
equations the RANS are obtained, for instance written for Cartesian coordinates
and uncompressible flow in (2.5), continuity, and (2.6) momentum equation.
Further unknowns are hence introduced,
i j
u' u' , named the Reynolds
stresses which need to be modelled to close the set of equations. Even if the
Reynolds stresses are originated by time averaging the convective term of the
Navier-Stokes they are usually grouped in the diffusive term of the RANS
momentum, like in first term of the right hand side of the (2.6). In fact they are
responsible for the turbulent diffusion of momentum which in a fully turbulent
flows is several order of magnitude higher than the molecular diffusion due to
viscosity.
j
j
U
0
x

(2.5)


16
( )
i
j ij i j i
j j i
U 1 1 P
U u u F
x x x

= +

(2.6)
A way to model the Reynolds stresses, which account for six variables in
3D and three in 2D, is by assuming true the Boussinesq hypothesis to relate the
Reynolds stresses to the mean velocity gradients introducing in this way the
eddy-viscosity concept.
j
i i
i j t t ij
j i i
U
U U 2
u u k
x x 3 x
| | | |
= + + |
|
|

\
\
(2.7)
The set of PDE could be closed by imposing a constant eddy-viscosity,
resulting in a very poor turbulence modelling. More often a transport equation is
used for the (modified) eddy viscosity like in the SA model, or the eddy
viscosity is computed with an algebraic equation taking into account a length
scale and a velocity scale of the turbulence, computed by other two variables
linked to turbulence and transported by two further PDE.
The turbulence models that close the set of equations are usually classified
by the number of additional transport equations: a one equation model is the
SA, while largely employed two equations models are the k-, the k- and their
modifications.
A way to close the RANS without the Boussinesq hypothesis is to apply a
transport equation for each Reynolds stress. This approach is followed in the
RSM model, which in 3D is a 7 equations model, 6 equations for Reynolds
stresses and one for dissipation.
Even if the RSM model is capable to simulate the anisotropy of the
turbulence it requires a large computational cost. This one reason why
nowadays is more common to perform unsteady simulations URANS, DES or
LES rather than steady RSM.
In the following sections some turbulence treatments will be described, in
particular eddy viscosity models in paragraph 2.4 and eddy solving
methodologies LES and DES in paragraph 0.


17
2.4 EDDY VISCOSITY MODELS
2.4.1 A one equation turbulence model: the Spalart-Allmaras (SA) model
The basic concepts of the Spalart-Allmaras model are now recalled in order
to make more comprehensible its modification leading to the SA based DES
model.
In the SA model a further transport equation (2.8) for a modified turbulent
viscosity is used to close the system of RANS equations.
( ) ( ) ( )

(
| |

(
+ = + + + |
`
|
(

) \

2
i b2
i j j j
1
u G C Y
t x x x x
(2.8)
where G

is the production of turbulent viscosity and Y

is the destruction of
turbulent viscosity that occurs in the near-wall region due to wall blocking and
viscous damping.

and C
b2
are constants and is the molecular kinematic
viscosity.
The turbulent viscosity is computed by:
=
t v1
f (2.9)
=
+
3
v1 3 3
v1
f
C

The production term G

is modelled by
=


b1
G C S

and,
+

v2 2 2
S S f
d

Where C
b1
is a constant of the model, the density of the fluid and

S given
by the previous equation, defined as the sum of S, depending on deformation


18
and rotation tensor, and a term accounting for the vicinity of the wall through
the distance to the nearest wall d; is the von Karman constant (~0,4187 in
Fluent 6.2) and the function f
v2
is evaluated by:
=
+
v2
v1
f 1
1 f

.
In the original form of the SA model the scalar S is computed by the
magnitude of the rotation-rate tensor
ij
.
( )
=
ij ij
S 2
with
| |

= |
|

\
j
i
ij
j i
U
U 1
2 x x

And this is also the default value for S in the Fluent Spalart-Allmaras
option. The second option available in Fluent for the choice of S is:
( )
+
ij prod ij ij
S C min 0, S (2.10)
Being S
ij
the element of strain-rate tensor.
| |

= + |
|

\
j
i
ij
j i
U
U 1
S
2 x x

With C
prod
= 2.0 as default value. The destruction term Y

is modelled by:


19

| |
=
|
\

2
w w
Y C f
d

( +
=
(
+

1/ 6
6
w3
w 6 6
w3
1 C
f g
g C

( )
= +
6
w2
g r C r r

2 2
r
S d

The default values for the SA model constants are given in table below:
Table 2.1 - default values in Fluent 6.2 for the Spalart-Allmaras model constants
=
b1
C 0.1355 =
b2
C 0.622

2
3

=
v1
C 7.1
( )

+
= +

b2 b1
w1 2
1 C C
C
=
w2
C 0.3
=
w3
C 2.0 = 0.4187
The standard Spalart-Allmaras model uses the distance to the closest wall as
the definition for the length scale d, which plays a major role in determining the
level of production and destruction of turbulent viscosity.
2.4.2 Two equations turbulence models
In the two equations turbulence models the eddy viscosity is generally
defined by means of a velocity scale V
S
and a length scale
S
using the (2.11).
A velocity scale which is generally accepted to describe turbulence is the square
root of the turbulent kinetic energy k, while for the length scale different
choices are available, velocity and length scale for the k- and k- model are
reported in Table 2.2.
S S t
V C

= (2.11)


20
Table 2.2 Velocity scale and length scale for k- and k- models
V
S
S

k- k
1/2
k
3/2
/
k- k
1/2
k
1/2
/
Model k- : :: : standard and RNG
The standard k- model is a two equations model based on the transport
equations for the turbulence kinetic energy (k) and its dissipation rate ():
( ) ( )
k M b k
j k
t
j
i
i
S Y G G
x
k
x
ku
x
k
t
+ + +
(
(

|
|

\
|
+

( ) ( )
( )

S
k
C G C G
k
C
x x
u
x t
b k
j
t
j
i
i
+ + +
(
(

|
|

\
|
+

=
=

2
2 3 1

In the additional transport equations two production terms do appear i.e. G
k

and G
b
, respectively the generation of TKE due to velocity gradients and
buoyancies. The term Y
M
takes into account compressibility effects while the
source terms are labelled with S; for a neutral ABL the generation due to
buoyancies is negligible, as well as the compressibility effects and there are no
sources, therefore the two transport equations for k and can be simplified:
( ) ( )

+
(
(

|
|

\
|
+

k
j k
t
j
i
i
G
x
k
x
ku
x
k
t


( ) ( )
k
C G
k
C
x x
u
x t
k
j
t
j
i
i
2
2 1

+
(
(

|
|

\
|
+

21

In the standard k- model the eddy viscosity is modelled by the equation
(2.11) with scales as in Table 2.2. The TKE generation term is modelled by
t

S
2
where S is the magnitude of rate-of-strain tensor.

Constants appear in the standard k- formulation: their values are far to
being universal and some values are generally accepted since they produce right
level of turbulence in common industrial flows.
In the case of neutral ABL flows other values of constants have been
proposed, Crasto et al. (2004) [8] and Mandas et al. (2004) [23], in order to
produce a proper level of turbulence in proximity of the ground. The default
values adopted in Fluent 6.2 for the model constants and the modified set of
constants as in Crasto et al. (2004) [8] and Mandas et al. (2004) [23] are shown
in Table 2.3.
Table 2.3 standard k- model constants
C

C
1
C
2

k



Default 0,09 1,44 1,92 1 1,3
Modified 0,03 1,44 2,223 1 1,3

In the commercial code Fluent 6.2 two k- models derived from the
standard one are available: the RNG k- and the realizable k-. Further
information about these two modifications of the standard k- are provided in
the Fluent Users Guide [10].

The RNG modification of the k- model provides the possibility to compute
the effective viscosity by the use of a further differential equation, in this
fashion the algebraic equation (2.11) is not any more employed. If the option is


22
activated the model acts fairly well in low Reynolds conditions hence near-wall
flows. An improvement is therefore expected also for ABL simulations.
The RNG k- does make also use of constants which have to be properly set
to achieve best results for each flow condition. The default values of constants
for the RNG k- model in Fluent 6.2 and the modification proposed are given in
the table below.
Table 2.4 RNG k- model constants
C

C
1
C
2

Default 0.0845 1.42 1.68
Modified 0.05 1.42 1.68
Model k- : :: : standard and SST (Shear-Stress Transport)
The standard k- model is characterized by the two transport equations:
( ) ( ) G Y S
| |
+ = + +
|

\
i k k k k
i i i
k
k ku
t x x x
(2.12)

( ) ( ) G Y S

| |
+ = + +
|

\
i
i i i
u
t x x x
(2.13)

Where G
k
e G

are respectively the production of k and due to velocity
gradients,
k
and

diffusive coefficients, Y are dissipation terms and with S
are labelled the sources.

The diffusive coefficients are modelled as in the standard k- by:

= +
t
k
k
(2.14)

= +
t
(2.15)



23
The molecular viscosity is incorporated in the diffusive terms while
k

and

are turbulent Prandtl numbers for k and . The turbulent viscosity is
computed in the k- turbulence model by the (2.16).
*
t
k

= (2.16)
Obviously the parameter * has the same meaning of C

employed in the
standard k- model but in the k- model * changes dynamically in order to
make the model fit to low Reynolds regions. Its for this reason that the k-
model is naturally more suitable than the SKE for wall bounding flows.
The k- model, as all the RANS models, is completed with a set of
parameters whose optimal value changes quite a lot depending on the flow
features: mean flow and level of turbulence.

In addition to the standard k- model described in the previous section the
standard k- model, FLUENT also provides a variation called the shear-stress
transport (SST) k- model, called in this way since the definition of the
turbulent viscosity is modified to account for the transport of the principal
turbulent shear stress. The transport equations for the SST k- model and its
features are reported in Fluent Users Guide [10].
2.5 EDDIES SOLVING METHODOLOGIES: LES AND DES
2.5.1 Filtered Navier-Stokes equations
Increasing computational capabilities and the possibility to build clusters of
computers allow nowadays to perform simulations of the ABL with eddies
solving methodologies like the LES (or more correctly with Wall Modelled
LES, WMLES) and hybrid methodologies RANS/LES as the DES.

In the LES model the governing equations are space filtered rather than time
averaged as in RANS. A general variable ( ,t) space filtered is:


24
( ) ( ) ( ) t , t G , d

, =

x x (2.17)
Where G is a filter function and the volume integral is extended on the
whole computational domain . f the filter function has a compact support the
integral is reduced to the support of the filter function.
For finite volume solvers like Fluent or PHOENICS the grid itself acts as a
Top-Hat filter, considering a domain subdivided in a mesh, in a inner cell of
volume V:
( )
1/ V, V
G ,
0, V

=

(2.18)

In this fashion the mesh itself acts as a spatial filter; other types of filters
can be used, for instance the Gaussian filter or the sharp Fourier cut-off. It
follows also that the filter width is directly connected to the cubic root of the
cell volume.
Application of the (2.17) and (2.18) brings to a filtered function given by:
( ) ( )
V
1
t , t d , V
V
, =

x (2.19)
Where now the overbar is used to indicate a space filtering procedure. The
filtered Navier-Stokes equations for the case of incompressible isothermal flows
are given by:
( )
0
i
i
u
t x

+ =

(2.20)

( ) ( )
ij
i
i i j
j j j i j
u p
u u u
t x x x x x

| |

+ =
|
|

\
(2.21)

With the filtering of the Navier-Stokes equations a closure problem is also
introduced since the
ij
, named SFS shear stresses, depend on further unknowns
i j
u u .


25
j i j i ij
u u u u = (2.22)
In order to close the set of equations the SFS stress terms have to be
modelled. The basic idea of the LES is that small scale vortices are universal,
do not depend on mean flow and boundary conditions. Therefore if the filter
width is set to a proper value, i.e. if the mesh is enough refined the SFS model
should be universal and acts correctly regardless the type of flow.
2.5.2 Sub Filter Scale (SFS) modelling
As introduced in the previous paragraph, the filtered NS equations introduce
a closure problem; the turbulence closure is assigned to the SFS model that is
devoted to simulate the filtered turbulence i.e. the shear stress term in the
momentum equation (2.22).

Pioneering work of Smagorinsky (1963) [29] on LES employed the concept
of SGS eddy viscosity (2.24); while equations (2.23)-(2.27) describe the
Smagorinsky-Lilly model how is implemented in Fluent 6.2.
Eddy viscosity models do not allow back-scatter of energy from small
eddies to bigger eddies which is not negligible when dealing with small scale
turbulence and for flows close to the wall. Attempts to cure the lack of back-
scatter in boundary layer flows have been proposed by Mason &Thomson
(1992) [24] by adding a stochastic forcing to the velocities in the wall adjacent
cells. The technique of the stochastic forcing allows to obtain a velocity profile
closer to the logarithmic profile.
1
2
3
ij kk ij t ij
S = (2.23)
2
t S
L S = (2.24)

1
2
j
i
ij
j i
u
u
S
x x
| |

= + |
|

\
(2.25)


26
ij ij
S S S 2 = (2.26)
( )
1/ 3
min , =
S S
L d C V (2.27)

The Smagorinsky constant C
S
employed in the Smagorinsky model more
realistically varies in a range of values depending on the local turbulence. A
way to improve the Smagorinsky model is to compute dynamically the
Smagorinsky constant as a function of the flow features, a dynamic
Smagorinsky model like the one proposed by Germano et al. (1991) [13] and
subsequently modified by Lilly (1992) [22] which is implemented in the code
Fluent 6.2 as described in Kim (2004) [19].

A different approach to the turbulent closure for the filtered Navier-Stokes
equations and further improvements to SGS modelling have been made by
Bardina et al. (1983) [2] with the first scale-similarity model. In the scale-
similarity models the unfiltered velocity is expressed as a function of the
filtered velocity ) (
i i
u f u ; for instance the unfiltered velocity can be written
as a truncated Taylor series of the filtered resolved velocity. The simplest scale-
similarity model is the one proposed by Bardina et al. (1983) [2] where the
authors proposed that the velocity could be approximated with the filtered
velocity
i i
u u . In this way the SFS stress term of (2.22) becomes:
j i j i ij
u u u u = (2.28)

Mixed models of scale similarity and dynamic eddy-viscosity have been
also proposed, for instance in Zang et al. (1993) [42]. Mixed models are able to
compute both back-scatter and forward-scatter of energy respectively from
small-scale vortices to bigger-scale and vice versa and viscous dissipation,
hence a behaviour closer to the reality compared to pure eddy-viscosity models
which dont allow back-scatter.
Although in the code employed Fluent 6.2 four types of eddy-viscosity SGS
models are available, these models resulted to be not applicable in the case of


27
ABL since the LES option in Fluent can be performed only for smooth surfaces.
In the Chapter 5 the results of simulations of the ABL over flat terrains are
illustrated. Some of them obtained by applying the DES module of Fluent and
other by using an UDF defining a Smagorinsky model; both the methodologies
were coupled with wall-functions for rough surfaces. Therefore a WMLES
approach rather than a proper LES.
2.5.3 Hybrid RANS/LES models. The SA based DES model
The Detached Eddy Simulation (DES) model belongs to the class of
LES/RANS hybrid models usually obtained by modifications of turbulence
models originally born to close the RANS equations; the modified models act in
their original form close to the wall while at a certain distance from the wall the
modification is activated and the model provides an SGS eddy-viscosity.
The main aim of hybrid RANS/LES modelling is to work in applications in
which pure LES is not affordable since the high number of cells required, i.e.
high Reynolds number flows. The DES models were originally born for
external aerodynamics applications where the attached boundary layers were
supposed to be solved completely by the URANS models while the LES zone
was aimed at solving the detached vortices. Also in flows over complex terrains
the DES models can find an area of application even this is not the case for
which they have been originally conceived. When DES is applied to the ABL
the URANS zone solves only part of the boundary layer while the grey zone
URANS/LES resides inside the logarithmic region and a typical shift of velocity
is generally observed in the interface region.
The DES model was originally proposed by Spalart et al. (1997) [31] by
modification of the Spalart-Allmaras RANS model.
In FLUENT 6.2 the only DES model available by default is based on the
one-equation Spalart-Allmaras model. Two-equations DES models have been
also proposed, based on transport equations of k and and also of k and ; two
equations DES models have been only recently introduced in the newest version
of Fluent, the 6.3, which has not been used for the present project.


28
The SA based DES model implemented in Fluent, as proposed by Shur et
al. (1999) [28] replaces the distance to the nearest wall d both in the production
and destruction terms of the SA model with a new length scale

d , defined as:
( ) =

DES
d min d, C (2.29)
When the equation (2.29) is substituted in the SA model, in a time marching
procedure the effect obtained is to have a URANS zone close to the wall where
(d < C
DES
) while the rest of the domain is treated in a LES way using the
modified SA as SFS model.
The filter size by default in Fluent 6.2 is based on the largest cell
dimension while the empirical constant C
DES
has a default value of 0,65.
The constant C
DES
affects both the position of the interface separating the
RANS and the LES zone but also the value of SGS turbulent viscosity in the
LES zone as it will be described in Chapter 5.
2.6 WALL FUNCTIONS FOR ROUGH SURFACES
According to the theory proposed by Cebeci and Bradshaw (1977) [6] a
turbulent boundary layer bounding a rough surface can be described with three
kinds of fluid dynamic regimes depending upon a Reynolds number
+
S
K based
on the dimension of roughness elements K
S
(equivalent sand diameter) and the
friction velocity u

u K
K
S
S
=
+
(2.30)
Cebeci and Bradshaw (1977) [6] proposed that the universal log-law for
smooth walls (2.31) could be modified for rough surfaces (2.32) by subtracting
a function B.
( ) B y u + =
+ +
ln
1

(2.31)


29

u
u
u =
+
;

u y
y =
+
Wall coordinates
Where u
+
and y
+
are the wall coordinates, is the von Karman constant and
B an experimental constant which is found to be around 5,45.
( ) B B y u + =
+ +
ln
1

(2.32)
The three hydro dynamic flow regimes previously mentioned are:
Hydro-dynamically smooth (
+
S
K < 2.25)
Transitional (2.25 <
+
S
K < 90)
Fully rough (
+
S
K > 90)
The new term B can be expressed as function of the dimensionless
roughness height
S
K
+
. In Fluent 6.2 the expressions of B for the three different
regimes are given by
in hydro-dynamically smooth regime:
B = 0
in transitional regime:
( ) [ ] 811 . 0 ln 4258 . 0 sin
75 . 87
25 . 2
ln
1

|
|

\
|
+

=
+ +
+
S S S
S
K K C
K
B

in fully-rough regime:
( )
+
+ =
S S
K C B 1 ln
1

Two parameters are used in Fluent to describe the roughness of a wall: the
constant C
S
and the roughness height K
S
. Fluent first computes the friction
velocity u

which is used to evaluate the dimensionless roughness height

+
S
K
that is used to calculate the B and finally the velocity at the wall adjacent cell
by means of the (1.32).


30
Experience and suggestions in the Fluent Users Guide [10] about the
proper usage of the wall functions for rough surfaces ensure that the height of
the wall adjacent cells has to be at least twice the roughness height K
S
, hence if
a more refined grid (normally to the wall) is desired the constant C
S
has to be
increased keeping constant the product C
S
K
S
.

By comparison of the equation (1.3) with the (2.32) in fully rough regime
after some algebra a simple relation is found between the product C
S
K
S
and the
aerodynamic roughness length z
0
.
0
z E K C
S S
(2.33)
Where E equals exp(B).

From a theoretical point of view the wall functions are correct only if the
horizontal pressure gradients are negligible (in equilibrium conditions) but in
the common usage of the code the wall functions are applied as boundary
conditions in every kind of wall bounding flow without penalising too much the
solution.

The mathematical procedure followed to derive the equation (2.33) is
reported in Appendix B.
2.7 CONCLUSIONS
In the present Chapter some basis of ABL modelling are presented. Firstly
an overview of the time averaged (or Reynolds Averaged) Navier-Stokes
equations is made; the RANS introduce a closure problem because of the
introduction of new unknowns, the Reynolds stresses. The closure of the
equations is obtained with turbulence models, which are generally classified
with the number of transport equations constituting a model.
Further approaches for solving of equations for turbulent flows like
WMLES and DES have been also described.


31

The described methodologies have been tested in CFD simulations. Results
of RANS simulations carried out with the code Fluent 6.2 on flat terrains aimed
at a proper setting of roughness parameters and turbulence modelling are
presented in Chapter 1. 3D RANS simulations on a isolated hill (Askervein
Hill) performed with the codes Fluent 6.2 and WindSim 4.5 and 4.6 are
presented in the Chapter 1.

In Chapter 5 a first investigation of the use of the DES and LES options of
the commercial code Fluent 6.2 for the simulation of a neutral ABL is
presented. The Chapter 5 reports with some modifications part of the Diploma
Course projects thesis Crasto G. (2006) [9] carried out by the author in 2005-
06 at the von Karman institute.



33
Chapter 3 2D RANS SIMULATIONS
3.1 SETTINGS OF ROUGHNESS PARAMETERS
A series of 2D RANS simulations have been performed with the code
Fluent 6.2 in order to set the proper values for the roughness parameters K
S
and
C
S
and the minimum height of the wall adjacent cell compatible with the wall
functions for rough surfaces. The dimension of the wall adjacent cells normally
to the wall should be at least twice the roughness parameters K
S
representing the
dimension of the roughness elements. Simulations have been performed for
each class of roughness defined in Table 1.1.
In the values of roughness parameters and cell height h in Table 3.1 C
S
has
been always fixed to one since it is the maximum allowed without importing in
the code files named profile files in the codes user manual [10], equation
(2.33) leads to a value for K
S
for a given value of roughness length z
0
.
If a more vertically refined grid is desired the constant parameter C
S
in
Fluent 6.2 has to be higher than one and therefore assigned by profile files. For
instance if C
S
equals E than K
S
equals z
0
and h at least twice z
0
.
Table 3.1 Classification in roughness classes, values of K
S
, C
S
and height of the
ground adjacent cell h for the given mean z
0
.
class
mean z
0

[m]
K
S

[m]
C
S

h
[m]
z
0

[m]
I 0,0001 0,000981 1 0.002 10
-5
0,0002
II 0,0005 0,004905 1 0.01 0,0002 0,001
III 0,005 0,04905 1 0.1 0,0010,01
IV 0,03 0,2943 1 0.6 0,01 0,04
V 0,07 0,6867 1 1.4 0,04 0,1
VI 0,55 5,3955 1 10.8 0,1 1
VII 2,5 24,525 1 50 1 4


34
3.2 GEOMETRY AND GRIDS
The 2D tests performed to evaluate the effects of the different turbulence
models have been performed on a simple domain 6 km long and 1 km high. A
uniform spacing has been used for the horizontal direction, while vertically a
fixed expansion ratio has been employed which is the ratio of the heights of
two adjacent cells; the height of the first cell h is in accordance with Table 3.1.
The geometric characteristics of the used grids are summarised in Table 3.2.
Table 3.2 geometric characteristics of the used grids.
height of the domain (y direction) Ly 1000 m
length of the domain (x direction) Lx 6000 m
length of the cell (stream-wise) x 20 m
number of horizontal intervals Nx 300

class
h
[m]
Ny N
I 0.002 1.100 114 34200
II 0.01 1.22 50 15000
III 0.1 1.126 60 18000
IV 0.6 1.086 60 18000
V 1.4 1.117 40 12000
VI 10.8 1.023 50 15000
VII 50 1.000 20 6000
3.3 BOUNDARY CONDITIONS
Physically an ABL has only one border, the terrain, but computational
domains have to be bounded in all directions and other unphysical borders have
to be introduced. Even if the terrain is the only physical border for the ABL it


35
provides some difficulties in modelling, difficulties connected to the roughness
of the ground, which force to the use of wall-functions.
As regards the other boundaries, the top of the domain is treated as
symmetry plane, the inlet as a velocity inlet, condition that needs the
assignment of velocity and turbulence, given as profiles of mean velocity, TKE
and TDR. The outlet is instead treated as outflow, i.e. all the normal gradients
but the pressure gradients are imposed zero at the outlet (Neumann condition).
The vertical profiles of velocity and turbulence assigned at the inlet were
supposed to regard a fully developed boundary layer for a given roughness, the
same roughness assigned to the ground; in this way the profiles of velocity and
turbulence should not change within the domain.
The difference of profiles observed between outlet and inlet in simulation of
fully developed flows are therefore due to errors in the expression of the inlet
profiles and usage of wall boundary condition.
The analytical expression of mean velocity, TKE and TDR vertical profiles
assigned at the inlet are:
Mean velocity

| |
=
|
\
m
0
u z
U(z) ln =1
z
(3.1)
TKE
2
2
h
u z
k(z) 1
z C

( | |
=
( |
( \
(3.2)
TDR
z
u
z
3
1
) (

= (3.3)
Where the (3.1) is the log-law introduced in the paragraph 1.3 on similarity
theory; the (3.2) and (3.3) match the boundary conditions for TKE and TDR
used in the code at the wall and provide also a proper reduction of TKE and
TDR from the wall to the free flow, they have been also used quite successfully
in Mandas et al. (2004) [23] using Fluent and by Leroy (1999) [21] using the
finite-volume CFD code PHOENICS.


36
Further velocity and turbulence profiles could be obtained for instance by
using transitional periodic conditions (streamwise) over flat terrains. In the
periodic streamwise condition the variables calculated at the outlet are assigned
to the inlet, with the possibility to point to a given mass flow, in this way a fully
developed flow can be obtained. Even if not reported some RANS simulation
have been performed using transitional periodic boundaries, especially as
starting point for the LES/DES simulations reported in Chapter 5. Periodic
conditions that are in this way useful to test furthermore the wall functions
modified for rough surfaces by verifying the settings of the roughness
parameters K
S
(RH) and C
S
(RC).
3.4 RESULTS: VELOCITY PROFILES
The aim of the 2D numerical simulations of flat terrains is to confirm that
the vertical profiles of velocity and turbulence given by equations (3.1)-(3.3) are
typical of a fully developed boundary layer. Moreover, simulations on flat
terrains allow to understand how to set the roughness parameters RH and RC in
order to reproduce the same roughness length of the profiles imposed at the
inlet. In this set of simulations the desired flow field is a fully developed, fully-
rough turbulent boundary layer, hence with the advective term that should drop
to zero in the whole domain.
In the present paragraph the numerical results u(z) and
m
observed at the
outlet of the domain are presented for a roughness length z
0
0,03 m (RH 0,2943
m, RC 1, h 0,6 m), in Figure 3.1 and Figure 3.3. The velocity profile observed at
the outlet of the domain is fairly logarithmic and the points are scattered with a
squared correlation factor R
2
of 0,9979 for an exponential regression computed
over the first seven nodes. The roughness length at the outlet computed with the
regression is 0,0439 m while the desired value is 0,03 m. There are two ways to
further improve the prediction of the roughness length: the first one is to slightly
reduce the RH factor (or K
S
) keeping constant RC (or C
S
) while the second one
is to increase RC and assign RH according to equation (2.33) (
0
z E K C
S S
)
and then adjust RH if necessary. Roughness constants RC higher than one in


37
Fluent 6.2 must be assigned with profile files, whose structure and procedure
for reading is explained in code users manual [10]. Its important to note that
reducing the roughness height RH allows to refine furthermore the grid in wall
normal direction since the height of the wall adjacent cell has to be at least
twice RH. For instance, with RH 0,25 m and RC 1 a better prediction of
roughness length is obtained for a target of z
0
0,03 m, see Figure 3.2.
The normalised wind-shear
m
, defined in the section on similarity theory in
Chapter 1, equals one in a perfectly logarithmic velocity profile. In Figure 3.3
the normalised wind shear obtained in the same conditions of Figure 3.1 is
equals to 1 0,1 which is clearly acceptable; the behaviour observed close to
the top of the domain in Figure 3.3 is forced by the symmetry boundary
condition and not disturbing the flow field since confined to the very upper
region.

y = 0,0439e
0,6149x
R
2
= 0,9979
0,1
1
10
100
1000
0 2 4 6 8 10 12 14 16 18
x-vel [m/s]
h
e
i
g
h
t

[
m
]

Figure 3.1 velocity profile at outlet for RH 0,2943 m as in Table 3.1 and according to
equation (2.33).


38
inlet whole domain
y = 0,0309e
0,6439x
R
2
= 0,9999
outlet first ten nodes
y = 0,0325e
0,6479x
R
2
= 0,9981
0,1
1
10
100
1000
0 2 4 6 8 10 12 14 16 18
x-vel [m/s]
h
e
i
g
h
t

[
m
]

Figure 3.2 velocity profiles at inlet and outlet for the same settings of Figure 3.1
except RH 0,25 m.
0,1
1
10
100
1000
0,6 0,7 0,8 0,9 1 1,1 1,2

m
h
e
i
g
h
t

[
m
]

Figure 3.3 normalised wind shear extracted at the same section and for the same
simulation of Figure 3.1.


39
For other two cases the roughness constant RH has been tuned to obtain a
better prediction of the roughness length z
0
; both first attempt and tuned values
are reported in Table 3.3 where the correlation factor is referred to the first five
nodes, excluded the one on the ground.

Table 3.3 Tuned values of RH parameter for better prediction of roughness length z
0
.
RH [m]
Class inlet z
0

First Tuned
RC
h
[m]
z
0 predicted
with tuned RH

R
2

IV 0,03 m 0,2943 0,25 1 0,6 0,0325 m 0,9981
VI 0,16 m 1,567 1,17 1 3,2 0,199 m 0,9981
VI 0,55 m 5,3955 2,61 1 10,8 0,5397 m 0,9985

Some tests have been performed also with RC = E, RH = z
0
and h slightly
higher that twice RH. More accurate predictions of roughness length are
obtained in RANS steady simulations but not so satisfactory when dealing with
LES/DES simulations since in this way several cells are used inside the canopy
layer while the logarithmic law is valid only in the surface layer.
3.5 COMPARISON OF TURBULENCE MODELS
The simulations for the comparison of the turbulence models have been
performed on a rectangular domain 1 km high and 6 km long for a roughness
length z
0
0.03 m (roughness class IV).

The models tested are:
Standard k- with default constants
Standard k- with modified constants
RNG k- with default constants
RNG k- with modified constants
Realizable k- with default constants



40
The models and relative modified constants have been already described in
the previous chapter about general settings of the numerical simulations. The
default and modified constant for the SKE and RNG k- models are reported in
Table 2.3 and Table 2.4 while the other models default constants are reported in
the Fluents guide [10]. For the RSM model, some differences exist between the
2D and the 3D versions and therefore the 2D RSM tests could not be
completely satisfactory in the preparation of 3D simulations. Some CFD results
obtained with the RSM will be presented directly in the next chapter for 3D
simulations over an isolated hill.
Several simulations have been carried out to evaluate the different
behaviours of the turbulence models on a flat terrain. Geometries and meshes
remain the same described in the previous paragraph. Further tests on different
turbulence models will be presented in the next Chapter 1 where the flow field
of the isolated hill of Askervein is investigated. The test case of Askervein is
very useful to understand the performances of the turbulence models, especially
in the prediction of the flow in the lee side of the hill, in an adverse pressure
situation where turbulence modelling is important.
Once the turbulence profiles (3.2) and (3.3) are substituted in the equation
(2.11) an eddy viscosity vertical profile is computed, shown in Figure 3.4. The
value of eddy viscosity in Figure 3.4 do not depend upon the constant C

while
depends on the friction velocity u

.



41
0,1
1
10
100
1000
0 5 10 15 20 25 30
h
e
i
g
h
t

[
m
]


t
[Pa s]
Figure 3.4 Vertical profile of eddy viscosity according to equation (2.11) and
turbulence profiles (3.2) and (3.3).
3.5.1 Profiles observed at the outlet
In the next graphs (Figure 3.5 and Figure 3.6) the TKE profile assigned at
the inlet of the domain and observed at the outlet are plotted respectively for the
SKE and RNG k- models. When an eddy viscosity model is applied it
computes at each iteration the eddy viscosity which is then plugged in the
momentum RANS equation. Its interesting therefore to see also if the eddy
viscosity assigned at the inlet by the turbulence parameters k and is kept
unperturbed through the whole domain to the outlet. The vertical profiles of
eddy viscosity are shown in the following Figure 3.7 and Figure 3.8 for the SKE
and RNG k-; it can be seen that in both models the TKE is slightly reduced in
the domain compared to the inlet condition, but this is not so dramatic, also the
eddy viscosity is maintained in all the cases but the RNG with modified
constants. Therefore the SKE and the RNG can be used in the reproduction of a
neutral ABL with the default constants. When there is the need to reproduce a
neutral ABL, and the ratio TKE/u

2
at the ground is known from experimental


42
data, the SKE model can be used with modified constants. The procedure for
the modification of the SKE models constants is reported in Appendix A.
The usage of the modified constants for the RNG, as in Table 2.4 is not
suggested since effects caused to the eddy viscosity profile of Figure 3.8 are not
satisfying. It would be necessary to derive a new set of constants for the RNG
model, for instance by following a procedure similar to the one described in
Appendix A for the SKE model.
3.6 2D SIMULATIONS ON AN ISOLATED HILL
Simulations on a given topography like an isolated hill rather than on a flat
terrain are more significant in order to set the effects of the horizontal
refinement and the discretization schemes employed. Therefore 2D tests have
been performed on a hill with a symmetric shape given by the equation (3.4).
2
( )
1
H
z x
x
L
=
| |
+
|
\
(3.4)
Where the height of the hill is named H and L is half width of the hill (x =
L) measured at its half height (z = H/2).
The domain analysed was 1 km high and 7 km long, the top of the hill was
positioned at 1.5 km from the inlet, the hill top H was 116 m and L = 250 m.
The roughness length chosen for this series of tests was 0,03 m therefore the
height of the first cell was set to 0,6m (RC = 1, RH = 0.2943 m).
Further characteristics of geometry, grids, and settings are reported in the
Table 3.4.








43
SKE (default constants) inlet outlet

0,1
1
10
100
1000
0 0,5 1 1,5 2 2,5 3
TKE [m
2
/s
2
]
h
e
i
g
h
t

[
m
]

SKE (modified constants) inlet outlet

0,1
1
10
100
1000
0 0,5 1 1,5 2 2,5 3
TKE [m
2
/s
2
]
h
e
i
g
h
t

[
m
]

Figure 3.5 vertical profiles of TKE for the SKE model with default and modified
constants.


44
RNG k- (default constants)
inlet outlet

0,1
1
10
100
1000
0 0,5 1 1,5 2 2,5 3
TKE [m
2
/s
2
]
h
e
i
g
h
t

[
m
]

RNG k- (modified constants)
inlet outlet

0,1
1
10
100
1000
0 0,5 1 1,5 2 2,5 3
TKE [m
2
/s
2
]
h
e
i
g
h
t

[
m
]

Figure 3.6 vertical profiles of TKE, for the model RNG k-, for default and modified
constants


45

SKE (default constants) inlet outlet

0,1
1
10
100
1000
0 5 10 15 20 25 30 35

eff
[Pa s]
h
e
i
g
h
t

[
m
]

SKE (modified constants) inlet outlet

0,1
1
10
100
1000
0 5 10 15 20 25 30 35

eff
[Pa s]
h
e
i
g
h
t

[
m
]

Figure 3.7 Vertical profiles of effective viscosity for the SKE model with default and
modified constants


46

RNG KE (default constants) inlet outlet

0,1
1
10
100
1000
0 5 10 15 20 25 30

eff
[Pa s]
h
e
i
g
h
t

[
m
]

RNG KE (modified constants) inlet outlet

0,1
1
10
100
1000
0 20 40 60 80

eff
[Pa s]
h
e
i
g
h
t

[
m
]

Figure 3.8 vertical profiles of effective viscosity for the RNG k- model with default
and modified constants


47
3.6.1 Choice of horizontal refinement
Five values of horizontal spacing have been chosen to study the error due to
spatial discretization, see Table 3.5 for further grids details, and if a possible
grid independence was reachable. For sake of simplicity the tests are carried out
with the first order up-wind scheme while the next paragraph is devoted to
spatial discretization schemes effects.
In Figure 3.9 the value of x-velocity is shown computed at an altitude of 10
m for the finest grid tested (5 m of resolution) and for the coarsest one (40 m of
resolution) in the whole length of the domain. In the next Figure 3.10 the error
due to horizontal refinement for all the cases is instead shown , in the vicinity of
the hill top (above) and its rear (down). Figure 3.11 shows the same trend of
Figure 3.10 but for the y-velocity.
The height of 10 m a.g.l. has been chosen to plot the velocity components
since typical for anemometric measurements. For example this was the
positioning height for most of he anemometers employed in the experimental
campaign of Askervein which is treated in the next chapter. Other typical
heights of displacement for the anemometers are 30 m and the hub height of
wind turbines (for wind energy applications) that can be even 100 m a.g.l..
While in the 2D simulations performed on domains with a horizontal
spacing of 5 m the time of computation resulted not acceptable, with a spacing
of 10 m a good compromise is reached between accuracy of results and time of
computation.
Nevertheless when the ABL is modelled over 3D domains of few kilometres
square a horizontal uniform grid of 10 m resolution still leads to a total number
of cells of few millions; horizontal resolution of 20 m or 30 m are therefore
used if the simulations have to be carried out on a single machine.
Its good practice to refine horizontally the grid in the area of interest, that
should be rather in the core of the domain, and expand the interval size outside
the area of interest towards the lateral borders. Nevertheless for the 2D tests
uniform grids have been preferred in order to keep the grid as simple as
possible.


48
Table 3.4 Geometry, mesh, boundary conditions and some further settings for the 2D
simulations on the isolated hill


Geometry
Lx 7000 m
Ly 1000 m
H 116 m
L 250 m
Hill x-position 1500 m from the inlet
Mesh
Ny 50
h 0.6 m
Nx dependent on refinement x
Boundary Conditions
Inlet Velocity Inlet
Outlet Outflow
Top Symmetry
Ground Wall functions for rough surfaces
Further Settings
Solver Segregated
Turbulence Model SKE (modified constants)
Pressure-Velocity Coupling SIMPLEC
Pressure Discretization PRESTO!
Table 3.5 Horizontal resolution and total number of cells for 2D tests on the hill
x [m] 5 10 20 30 40
N 70,000 35,000 17,500 11,650 8,750


49


Figure 3.9 x-velocity computed at 10 m a.g.l. for 5 m spacing (solid line) and 40 m
spacing (dashed line)
0 2 4 6 8
1
0
1
2
1
4
1
6
1
8
0
1
0
0
0
2
0
0
0
3
0
0
0
4
0
0
0
5
0
0
0
6
0
0
0
7
0
0
0
x

p
o
s
i
t
i
o
n

[
m
]
x - v e l [ m / s ]


50
16
16,2
16,4
16,6
16,8
17
17,2
1400 1420 1440 1460 1480 1500 1520 1540 1560 1580 1600
x position [m]
x
-
v
e
l

[
m
/
s
]
5
5,5
6
6,5
7
7,5
8
8,5
9
9,5
10
2000 2200 2400 2600 2800 3000 3200 3400 3600 3800 4000
x position [m]
x
-
v
e
l

[
m
/
s
]
5m
10 m
20 m
30 m
40 m
Figure 3.10 Spatial refinement effects, x- velocity at 10 m altitude


51
3
3,1
3,2
3,3
3,4
3,5
3,6
1300 1320 1340 1360 1380 1400 1420 1440 1460 1480 1500
x position [m]
y
-
v
e
l

[
m
/
s
]
5m
10 m
20 m
30 m
40 m
-3,5
-3
-2,5
-2
-1,5
-1
-0,5
0
0,5
1
1500 1520 1540 1560 1580 1600 1620 1640 1660 1680 1700
x position [m]
y
-
v
e
l

[
m
/
s
]
Figure 3.11 Spatial refinement effects, y-velocity at 10 m altitude




52
3.6.2 Choice of discretization scheme
Fluent 6.2 allows to model the convection terms of each governing equation
with different types of discretization schemes (viscous term is always modelled
with a central scheme with second order accuracy). In the settings employed,
summarised in Table 3.4, the possible discretization schemes for the convection
terms are the up-wind first and second order, the power-law scheme, the third
order MUSCL and the QUICK, which is a weighted average of a second order
up-wind and a second-order central scheme; they are all described in the Fluent
Users Guide [10] and related references.
Figure 3.12 (above) shows how the peak of velocity observed at 10 m a.g.l.
in proximity of the hill top is better predicted with second order and third order
schemes while first order up-wind and power-law introduce probably too much
numerical viscosity and therefore poor prediction; also in the rear of the hill, see
Figure 3.12 (down), approximately at three kilometres from the inlet the effects
of the discretization scheme can be appreciated; the choice of the discretization
scheme is therefore pushed to a second order up-wind scheme for all the
governing equations since the use of an higher order scheme doesnt look
motivated. The plot of the vertical component shows a behaviour similar to the
horizontal component, therefore it will not be shown.



53
16
16,2
16,4
16,6
16,8
17
17,2
1400 1420 1440 1460 1480 1500 1520 1540 1560 1580 1600
x position [m]
x
-
v
e
l

[
m
/
s
]
1st ord Up-Wind
2nd ord Up-Wind
3rd ord MUSCL
QUICK
power-law
5
5,5
6
6,5
7
7,5
8
8,5
9
9,5
10
2000 2200 2400 2600 2800 3000 3200 3400 3600 3800 4000
x position [m]
x
-
v
e
l

[
m
/
s
]
1st ord Up-Wind
power-law
2nd ord Up-Wind
3rd ord MUSCL
QUICK
Figure 3.12 Discretization schemes effects, x-velocity at 10 m a.g.l., on the hill top
(up) and rear (down)



55
Chapter 4 3D RANS SIMULATIONS OVER AN ISOLATED HILL
4.1 INTRODUCTION
The validity of a new CFD technique should be validated whenever possible
by means of comparisons with experimental data. One of the most employed
test-cases for the numerical analysis of the propagation of the ABL around
isolated hills is the wind field observed on the hill of Askervein during the two
experimental campaigns in September/October 82 and 83 carried out by a
team of researchers coming from different institutions in the world.
The research groups which took part to the Askervein project were the
Atmospheric Environment Service (AES), in Canada, the Ris national
laboratory, in Denmark, the University of Hannover, in Germany, the
University of Canterbury, in New Zealand, the Building Research
Establishment and the ERA Technology Ltd, both located in the United
Kingdom. Although the measurements gathered on the Askervein hill are
nowadays more than twenty years old, the big amount of data available is still
an important reference point for those researchers who want to investigate the
flow over hilly terrains.
The experimental results of the Askervein hill project are reported in Taylor
and Teunissen (1983) [34] for the 82 data and in Taylor and Teunissen (1985)
[36] for the 83 data. A comprehensive description of the experimental set ups
and instruments can be found in Taylor and Teunissen (1986) [37] together with
a selection of the experimental results.
After the publication of the gathered data by the involved institutions
several researchers tried to reproduce with numerical codes the flow field
observed on the Askervein hill.
Castro et al. (2003) [5] performed a series of numerical RANS simulations
reproducing some of the experimental runs, the TU03-B, MF03-D and TK03 as
named in Taylor and Teunissen (1985) [36], the considered incoming wind


56
direction was 210 measured clockwise from north while the stratification was
nearly neutral.
The turbulence model employed by Castro et al. (2003) [5] was the SKE
with the constants of the model (C
1
= 1.44, C
2
= 1.92 and
k
= 1.0) set as in
Launder and Spalding (1972), whereas the constants C

and

followed the
recommendations by Beljaars et al. (1987) [3] (C

= 0.033 and

= 1.85).
Castro et al. (2003) [5] proposed to pass from a steady to an unsteady state
analysis since in their RANS simulations in the lee side of the hill, close to the
hill top, the flow variables were oscillating around a mean value; in a unsteady
simulation an intermittent separation bubble is likely to occur as observed in
few of the experimental runs. If this is the case of the run TU03-B only an
unsteady simulation can reproduce correctly the flow field in the lee side of the
hill.
The simulations run by Stangroom (2004) [32] on the hill of Askervein dealt
with the incoming flow at 180 and 210. For the 180 direction the reproduced
run was the MF01-D e TU01-B while for the 210 the runs of reference are the
already mentioned TU03-B, MF03-D e TK03. The simulations presented in
Stangroom (2004) [32] were run with the commercial code CFX, the grids
employed were unstructured with few layers of prismatic cells adjacent to the
terrain. Stangroom (2004) [32] adopted the RNG k- and the RSM turbulence
models in the RANS simulations over Askervein hill. The RNG k- model
predicted the flow field in a manner comparable to the RSM. Depending on the
wind direction and area of the domain, the experimental results were sometime
closer to the RNG prediction and sometime closer to the RSM prediction.
Vertical profiles of turbulence (TKE or
u
) over the hill top obtained with RNG
k- and RSM models are also shown in Stangroom (2004) [32], that have been
compared to the experimental data collected with propeller and sonic
anemometers. Both the turbulence models underpredicted the measured level of
turbulence, with a higher error for the RNG k-.


57
4.2 SIMULATIONS ON ASKERVEIN HILL
4.2.1 General description
The hill of Askervein resembles a semi-ellipsoid with axes of few
kilometers with the hill top (HT) located few hundreds of meters far from its
centre point (CP). The topography of the hill is plotted in Figure 4.1, where the
measurements points CP, HT and RS are shown together with the points of
deployment of the propeller anemometers GILL UVW located at 10 m a.g.l.
along the line A passing through the hill top HT. Other two reference lines were
the line AA, parallel to the line A but passing through CP, and the line B,
passing through HT and CP, along the major axis.
Two digital maps of the area are available: the map-A that covers an area of
16 x 16 km
2
and the map-B that covers an area of 6 x 6 km
2
, both made by 256
x 256 points, the horizontal resolution of map-A being therefore 62.5 m while
the map-B has a resolution of 23.4 m. Moreover the map-A describes the hills
surrounding Askervein hill while the map-B shows only the hill isolated. Both
maps are shown in Figure 4.2. The whole topography described by map-A is
provided in Figure 4.3 together with the extension of map-B.
Since the presence of the mentioned group of neighbouring hills most of the
experimental runs performed by the research groups were relative to wind
blowing from the third and the fourth sector, directions for which the hills close
to Askervein are supposed to not affect the flow field around the hill.
The higher resolution of map-B provides a slightly higher altitude at the
summit of the hill; a less smoothed topography could lead to numerical results
much closer to the experimental.
Its not possible to state, without performing simulations, which one of the
two maps could lead to the best prediction: map-A, enclosing a wider area and
neighbouring topographies, or map-B, with an higher resolution.


58
Easting
N
o
r
t
h
i
n
g
0 1000 2000 3000 4000 5000 6000
0
1000
2000
3000
4000
5000
6000
RS
HT
CP
line AA
line A
line B

Figure 4.1 topography of Askervein hill, Reference Site (RS), Hill Top (HT) and
Centre Point (CP), obtained from map-B.
a) Map A b) Map B
10
10
10
10
10
10
10
20
20
20 20
20
20
30
30
30
30
30
40
40
40
40
40
50
60
50
50
50
60 60
60
70
70
70
80
80
80
90
90 90
100
100
100 110
120
130140
140
Ovest - Est
S
u
d
-
N
o
r
d
-2500 0 2500 5000
-4000
-2000
0
2000
4000

10
10
10
10
10
10
20
20
20
30
30
40
50
60
70 80
90
100
120
120
Ovest - Est
S
u
d
-
N
o
r
d
-3000 -2000 -1000 0 1000 2000 3000
-3000
-2000
-1000
0
1000
2000
3000

Figure 4.2 Maps A (a detail) and B (the whole map) of Askervein Hill


59
Easting
N
o
r
t
h
i
n
g
-5000 0 5000
-8000
-6000
-4000
-2000
0
2000
4000
6000
8000
Map-A extension
Map-B extension

Figure 4.3 Extensions of map-A and map-B over the topography described by map-A.
4.2.2 Measurement reference data
Among all the data gathered during the two campaigns, it has been chosen
to simulate the run TU03-B. During this run the wind was blowing almost
constantly for three hours from a direction of 210, a direction almost
perpendicular to the main axis of the hill, for which the hills at the east of
Askervein would produce no or minimal effects on the flow-field, moreover
during the run TU03-B the stratification was neutral, therefore stability effects
can be neglected simulating this run.



60
From the report of the experiments by Taylor and Teunissen (1985) [36] it
is extracted the Table 4.1 which shows part of the results obtained with the
GILL UVW anemometers displaced along line A and at the RS point.
The same horizontal velocities listed in the second column of Table 4.1 are
then plotted in Figure 4.4 along line A. Considering the definition of TKE as
given by equation (1.9) and the standard deviations of the velocity components
in Table 4.1 the TKE is calculated and plotted against the horizontal position
along the line A in Figure 4.5. The standard deviations of the velocity
components, also reported in Table 4.1, are plotted in Figure 4.6 against the
horizontal position along line A of the measurement points at 10 m ag.l..


v
e
l
o
c
i
t
y

[
m
/
s
]

0
2
4
6
8
10
12
14
16
18
-1000 -800 -600 -400 -200 0 200 400 600

line A horizontal position [m]
Figure 4.4 Horizontal velocity measured at 10 m a.g.l. along line A by GILL UVW
anemometers (as in Table 4.1).




61
Table 4.1 Averaged data from GILL UVW anemometers for turbulence run TU03-B
(extracted from table A1.3 of Taylor and Teunissen (1985) [36]).
Twr lcn Direct. Speed
u

Int U
v

Int V
w

Int W uv uw vw
RS 207.3 8.6 1.223 0.143 0.704 0.082 0.413 0.048 0.135 -0.247 -0.013
ASW85 201.6 7.8 1.200 0.153 0.762 0.097 0.463 0.059 0.007 -0.239 0.002
ASW50 192.9 6.7 1.350 0.202 0.683 0.102 0.475 0.071 0.099 -0.272 0.015
ASW35 196.0 7.2 1.243 0.174 1.038 0.145 0.580 0.081 0.042 -0.351 -0.033
ASW20 200.6 10.5 1.115 0.107 1.126 0.107 0.565 0.054 0.032 -0.287 -0.004
ASW10 207.9 13.2 1.059 0.080 1.232 0.093 0.577 0.044 0.098 -0.243 -0.049
HT 203.4 16.2 1.100 0.068 1.034 0.064 0.577 0.036 0.189 -0.153 0.031
ANE10 206.5 12.0 1.758 0.146 1.012 0.084 0.531 0.044 -0.004 -0.241 0.056
ANE20 195.9 5.6 2.560 0.458 1.502 0.269 0.881 0.158 -1.431 -0.560 0.206
ANE40 188.1 3.0 1.983 0.668 1.798 0.606 1.192 0.402 -0.275 0.694 0.459
Legend:
Twr lcn tower location (name)
Direct. direction, measured in degrees clockwise from north
Speed horizontal wind speed

u
estimator of standard deviation of streamwise component of velocity
Int U streamwise turbulent intensity (=
u
/Speed)

v
estimator of standard deviation of spanwise component of velocity
Int V spanwise turbulent intensity (=
v
/Speed)

w
estimator of standard deviation of vertical component of velocity
Int W vertical turbulent intensity (=
w
/Speed)
uv covariance streamwise-spanwise
uw covariance streamwise-vertical
vw covariance spanwise-vertical



62
T
K
E

[
m
2
/
s
2
]

0
1
2
3
4
5
6
-1000 -800 -600 -400 -200 0 200 400 600


line A horizontal position [m]
Figure 4.5 TKE computed with (1.9) and standard deviations of velocity components
measured at 10 m a.g.l. along line A by GILL UVW anemometers (as in Table 4.1).
s
t
a
n
d
a
r
d

d
e
v
i
a
t
i
o
n

[
m
/
s
]

0
0,5
1
1,5
2
2,5
3
-1000 -800 -600 -400 -200 0 200 400 600
sigma-u (streamwise)
sigma-v (spanwise)
sigma-w (vertical)


line A horizontal position [m]
Figure 4.6 Standard deviations of velocity components measured at 10 m a.g.l. along
line A by GILL UVW anemometers (as in Table 4.1).


63
Considering valid a log-law (1.3) at the reference site RS and assuming a
roughness length z
0
of 0.03 m, a friction velocity of 0.634 m/s is computed by
exponential interpolation (with 0.41); if the roughness length is not fixed the
regression provides a roughness length of 0.0486 m and a friction velocity of
0.685 m/s; experimental data and regression curves are reported in Figure 4.7.
h
e
i
g
h
t

[
m
]

y = 0,03e
0,6471x
R
2
= 0,9782
y = 0,0486e
0,5984x
R
2
= 0,9849
0
10
20
30
40
50
60
0 2 4 6 8 10 12 14

h
e
i
g
h
t

[
m
]

y = 0,03e
0,6471x
R
2
= 0,9782
y = 0,0486e
0,5984x
R
2
= 0,9849
1
10
100
6 7 8 9 10 11 12 13


velocity [m/s]
Figure 4.7 Vertical profile of mean horizontal velocity at RS during TU03-B run (from
Taylor e Teunissen (1985) [36]) and exponential interpolations.


64
The values of TKE measured at RS are instead plotted in Figure 4.8, they
can be used to evaluate the SKE C

constant: if the value of C

0.09 is kept
(default in Fluent) then the TKE values measured by the GILL UVW at the RS
are approximately matched by equation (3.2), see Figure 4.9. This observation
allows the usage of the default values for the SKE models constants and also
for the other models constants in the simulations of the TU03-B run, performed
in neutral conditions. Considering the velocitys standard deviation measured at
RS (table A1.6 and A1.7 of Taylor and Teunissen (1985) [36]), also a vertical
profile of TKE has been constructed, shown in Figure 4.8. The constructed
vertical profiles of both velocity and turbulence should resemble the profiles
assigned at the inlet of the computational domains.

h
e
i
g
h
t

[
m
]

0
10
20
30
40
50
60
0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4 1,6
cup anemometers
GILL UVW anemometers

TKE [m
2
/s
2
]
Figure 4.8 Vertical profile of TKE obtained from turbulence measurements (tables
A1.6 and A1.7 of Taylor and Teunissen (1985) [36]).


65
h
e
i
g
h
t

[
m
]

0
5
10
15
20
25
30
35
40
45
0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4 1,6
GILL UVW anemometers
equation (3.2)

TKE [m
2
/s
2
]
Figure 4.9 TKE values measured by the GILL UVW anemometers at RS during the run
TU03-B and profile given by equation (3.2) with C

0.09 and friction velocity 0.634 m/s

(solid line).
4.2.3 Simulations with Fluent
A series of 3D simulations have been carried out with the commercial code
Fluent on the isolated hill of Askervein. Both maps A and B are used in order to
study the effects of the original resolution and influences to the flow of the
group of hills at east of Askervein.
For both the maps a geometry of 3 km x 6 km enclosing the hill has been
drawn in order to have the sides of the computational domain parallel to the
incoming wind (210). The height of the computational domain has been set to
1 km while the hill of Askervein is placed approximately 1.6 km far from the
inlet border.
A view of part of the meshed borders for the computational domain
obtained from map-A is shown in Figure 4.10, above the whole domain
containing the hill of Askervein.


66
Further details of the geometry and grid are provided in the Table 4.2 while
the boundary conditions employed are summarised in Table 4.3.
0
500
1000
Z
0
1000
2000
3000
X
0
2000
4000
6000
Y
Y
X
Z


Figure 4.10 Grid obtained from map-A.


67
Table 4.2 Geometric characteristics of the meshes used for TU03-B runs comparisons
Height of the domain 1000 m
Width domain (dir. x, spanwise) 3000 m
Length domain (dir. y, streamwise) 6000 m
Height of wall adjacent cell (h)
0.6 m (z
0
0.03 m RC 1 RH 0,2943 m)
Width cells (x, y) 30 m
Number of vertical divisions 50
Number of horizontal divisions 100 x-axis, 200 y-axis
Total number of cells 1 000 000

Table 4.3 Boundary Conditions
Inlet
Velocity Inlet
Left side
Symmetry
Right side
Symmetry
Pressure Outlet (map-A), hydrostatic model
Outlet
Outflow (map-B)
Terrain
Wall with wall-functions
Top (geostrophic wind)
Symmetry

Several turbulence models have been tested in RANS simulations with the
described geometry, grid and boundary conditions; the turbulence models
employed are reported in Table 4.4.








68
Table 4.4 Turbulence models employed in the presented RANS simulations run with
the code Fluent.
Map-A Discretization scheme for advection term
SKE
default constants 2
nd
order upwind (momentum, k and )
RNG k- * default constants 2
nd
order upwind (momentum, k and )
RKE
default constants 2
nd
order upwind (momentum, k and )
RSM
default constants
2
nd
order upwind (momentum, k, and
Reynolds stresses)
Map-B
SKE
default constants 2
nd
order upwind (momentum, k and )
RNG k- * default constants 2
nd
order upwind (momentum, k and )
RKE
default constants 2
nd
order upwind (momentum, k and )
RSM
default constants
2
nd
order upwind (momentum, k, and
Reynolds stresses)
* with the Differential Viscosity model activated, see the Fluent Users Guide [10] for
the options description
4.2.3.1 Results: mean flow and turbulence
The horizontal velocity measured in the run TU03-B is plotted in Figure
4.11, where the vertical bars are proportional to the standard deviation of the
stream wise velocity and where the horizontal velocities obtained with the SKE
model in both maps are also shown. The two SKE simulations forecasted well
the flow up stream of the hill while the peak of velocity at the top resulted
slightly under predicted, a feature which is common in the reproduction of the
Askervein flow, see e.g. the plots presented by Castro et al. (2003) [5], an under
prediction of velocity that can be explained certainly by the too coarse meshes
originated from both maps. In fact, also in the 2D simulations of Chapter 1
about an isolated hill it was already shown that an horizontal resolution of 30 m
leads to a poor prediction of the wind peak at the top of the hill. Therefore
improvements could be expected refining the grids especially around the
summit of the hill.


69
The under prediction of the wind peak at the top of the hill has been also
explained in literature (Castro et al. (2003) [5]) by a roughness of the terrain
around the top of the hill smoother than the one considered.
The wind down stream of the hill is instead over predicted by both the
simulations based on the two maps, in the lee side of the hill, in an adverse
pressure-gradient flow, the turbulence model plays an important rule. The flow
downstream is not well captured because of the turbulence modelling.
Differences can be noticed between the wind predictions of the two simulations,
the map obtained from map-B looks behaving better than the one obtained from
map-A. It seems therefore that the neighbouring hills do not affect significantly
the flow at the lee side of the hill, while the higher resolution of map-B would
lead to a better prediction of the flow.
v
e
l
o
c
i
t
y

[
m
/
s
]

0
2
4
6
8
10
12
14
16
18
20
-1000 -800 -600 -400 -200 0 200 400
measurements
std k-e ASK A
std k-e ASK B


line A horizontal position [m]
Figure 4.11 Horizontal velocity along line A (10 m a.g.l.), SKE model map-A and
map-B.
The same experimental data of horizontal velocity and bars proportional to
streamwise velocity standard deviation are used in Figure 4.12 and Figure 4.13,
respectively for map-A and map-B, to make comparisons of four turbulence
models (SKE, RNG KE, RKE and RSM).


70
In Figure 4.12 and Figure 4.13 is shown how the mean flow along the line A
in the ups stream region and on the top of the hill is basically described in the
same manner by all the turbulence models employed. The flow in lee side, in
the adverse pressure gradient region, is the most sensitive to turbulence
modelling. The two equations turbulence model which was able to simulate in
the best way the mean flow in the lee side was the RNG, while the SKE was the
worst. The RSM produced the flow field closest to the experimental one in the
simulation on the grid built over map-A while in the simulation on the grid built
over the map-B the best forecasting was given by the RNG while the RSM
slightly underestimated the velocity in the lee side.
Depending on the topography the RNG or the RSM seem to behave better,
one big issue in the analysed run is that the flow was about to separate or an
intermittent bubble of separation was likely present. If unsteady phenomena
were present in the chosen run the only way to reproduce accurately the
experimental data would be by performing unsteady simulations.
Nevertheless, by observation of the mean flow the best behaving two
equation model would look the RNG KE. It would be interesting to analyse an
other run, always in neutral conditions, for instance from 180, possibly when
the flow was not prone to separate.
The turbulence results are instead presented in Figure 4.14 by the TKE
values collected along line A where the values of the measured TKE are the
same of Figure 4.5. The SKE model predicts well the level of turbulence
upstream and at the top of the hill while the level of turbulence is under
predicted in the lee side of the hill. Again, in the adverse pressure gradient flow,
poor matching is observed with experimental data with the grid obtained from
map-B producing an higher peak of turbulence, but still much smaller than the
measured.
When the same TKE data are compared with the numerical results obtained
with different turbulence models in Figure 4.15 and Figure 4.16 the two
equations models predicting better the TKE level are the SKE and the realizable
k- in a more or less equal way, while the RNG k- leads to a slightly level of


71
turbulence. The RSM model produces a slightly underestimated TKE at the top
of the hill, like the RNG k-e does, but its the turbulence model which predicts
the highest peak of turbulence at the lee side, still too low compared to the
experimental data. In the forecast of the TKE the grid dependency is even much
stronger than in the velocity prediction, therefore also for the turbulence plots a
much better result is expected if an higher horizontal discretization is
considered.
v
e
l
o
c
i
t
y

[
m
/
s
]

0
2
4
6
8
10
12
14
16
18
20
-1000 -800 -600 -400 -200 0 200 400 600
data
std k-e ASK A
real ke ASK A
RNG k-e ASK A
RSMASK A


line A horizontal position [m]
Figure 4.12 Horizontal velocity along line A (10 m a.g.l.), k- derived turbulence
models and RSM over map-A.


72
v
e
l
o
c
i
t
y

[
m
/
s
]

0
2
4
6
8
10
12
14
16
18
20
-1000 -800 -600 -400 -200 0 200 400 600
data
std k-e ASK B
realizable k-e ASK B
RNG k-e ASK B
RSM ASK B


line A horizontal position [m]
Figure 4.13 Horizontal velocity along line A (10 m a.g.l.), k- derived turbulence
models and RSM over map-B.
T
K
E

[
m
2
/
s
2
]

0
1
2
3
4
5
6
-1000 -800 -600 -400 -200 0 200 400
measurements
ske ASK A
ske ASK B


line A horizontal position [m]
Figure 4.14 TKE values along line A (10 m a.g.l.), SKE model for both maps


73
T
K
E

[
m
2
/
s
2
]

0
1
2
3
4
5
6
-1000 -800 -600 -400 -200 0 200 400 600
measurements
ske ASK A
realizable k-e ASK A
RNG k-e ASK A
RSM k-e ASK A

line A horizontal position [m]
Figure 4.15 TKE values along line A (10 m a.g.l.), k- derived turbulence models and
RSM over map-A.
T
K
E

[
m
2
/
s
2
]

0
1
2
3
4
5
6
-1000 -800 -600 -400 -200 0 200 400
measurements
ske ASK B
realizable ASK B
RNG ASK B
RSM ASK B

line A horizontal position [m]
Figure 4.16 TKE values along line A (10 m a.g.l.), k- derived turbulence models and
RSM over map-B.



74
In Figure 4.17 the standard deviations of the velocity vector are plotted
along line A, both measured and calculated by the RSM model, while the
standard deviations for the horizontal components are generally underestimated,
for the vertical component the standard deviation is overestimated, these
behaviours are then enhanced in the lee side of the hill at approximately at 200
m downstream of the hill top.

sigma-u (streamwise) sigma-v (spanwise) sigma-w (vertical)
sigma u RSMASK A sigma v RSMASK A sigma w RSMASK A
sigma u RSMASK B sigma v RSMASK B sigma w RSMASK B

s
t
a
n
d
a
r
d

d
e
v
i
a
t
i
o
n

[
m
/
s
]

0
0,5
1
1,5
2
2,5
3
-1000 -800 -600 -400 -200 0 200 400 600


line A horizontal position [m]
Figure 4.17 standard deviations of velocity components along line A (10 m a.g.l.),
measured and simulated with RSM over map-A and map-B.
4.2.4 Simulations with WindSim
The software WindSim is aimed at calculating the annual energy production
of a given wind farm; it firstly generates a structured grid made of hexahedral
cells, fitted to a given topography, and then solves the RANS equations by
means of the commercial CFD code PHOENICS, with a finite volume
approach, in steady state conditions for an uncompressible and isothermal flow.


75
The continuity and momentum equations are solved with a segregated solver
named SIMPLEST which is a modification of the widely used algorithm
SIMPLE.
The meshes generated with WindSim are built over a rectangular area with
the borders directed eastwards and northwards; horizontally they can be
uniformly spaced or with an inner refined region which is usually the area
enclosing the wind farm or the anemometric station. The grids can be stretched
also along the vertical direction; in this fashion the smallest cells result the
adjacent to the terrain while the others expand vertically with increasing
altitude.
The inlet conditions are the same used in the Fluent simulations presented in
the first part of this chapter: vertical profiles of velocity, TKE and TDR are
imposed at the inlet; the outlet is treated by providing a pressure profile while,
when the flow is directed in 0, 90, 180 and 270 the lateral borders are
treated as symmetry planes; a second approach available in the setting of the
boundary conditions is given by a nesting procedure, a meso-scale model is
performed, the flow field obtained is then used to provide boundary conditions
for an inner micro-scale model.
The turbulence model used by default by WindSim to close the RANS
equations is the standard k- model which can be selected with two different
sets of constants that are reported in Table 4.5.The sets of modified constants in
Table 4.5 are the same also used in Castro et al. (2003) [5] and proposed in
Beljaars et al. (1987) [3].

Table 4.5 Constants of the standard k- model in WindSim 4.5
C

C
1
C
2

k


Standard 0,09 1,44 1,92 1 1,3
Modified 0,034 1,44 1,92 1 1,85



76
The case study of Askervein have been also analysed with WindSim in
order to test the two sets of constants available and compare the numerical
results obtained with the corresponding data obtained with Fluent.
Grids with different resolution were analysed in order to study the grid
sensitivity of the results.
The simulations carried out with the code WindSim 4.5 were performed on
the digital map-B, covering an area of 6000 m x 6000 m centred on the hill of
Askervein.
Three grids has been constructed over the map-B and then employed in the
simulations performed with the code WindSim and some geometric features of
the computational domain and grids are reported in Table 4.6 and Table 4.7.
The horizontal resolution in the inner refined area of the coarsest grid is
comparable with the resolution of the meshes employed in the simulations
performed with the code Fluent which is 30 m, while grid 1 and 2 are
horizontally more refined in the inner region with a horizontal spacing around
20 m.
In the Figure 4.18 its shown the topography of the hill of Askervein while
in Figure 4.19 is visible a top view of the map-B discretized as in Grid 3,
measurements points RS, HT, CP and along line A are also shown.

Table 4.6 horizontal extension of domains an inner refined regions employed in the
simulations of the Askervein hill flow with WindSim
Height domain
1000m above highest terrain node
Easting (x) extension
X
min
072688m X
max
078688m X 6000m
Northing (y) extension
Y
min
820456m Y
max
820456m Y 6000m
Refined area
Easting (x) extension
X
min
074488m X
max
076688m X 2200m
Northing (y) extension
Y
min
822256m Y
max
824456m Y 2200m



77
Table 4.7 geometric features of the domains and meshes for the performed
simulations on Askervein hill with the code WindSim
Number of cells N
X
N
Y
N
Z
Total cells
Grid 1
168 168 34
959 616
Grid 2
168 168 40
1 128 960
Grid 3
119 119 40
566 440

Max resolutions
x
min
y
min
z
min
Grid 1
21,4m 21,4m 2,428m
Grid 2
21,1m 21,2m 0,375m
Grid 2
21,1m 21,2m 0,375m


Figure 4.18 Elevation map of Askervein hill obtained from map-B


78

Figure 4.19 Discretized map-B for the grid 3, points HT, CP, RS and 10 m masts
locations along line A. 210 wind direction as in run TU03-B of Taylor and Teunissen
(1985) [36]
The standard k- turbulence model has been employed to investigate the
flow on the hill of Askervein with both the sets of constants reported also in
Table 4.5. The friction velocity employed is about 0.634 m/s as calculated by
interpolating the experimental data at RS relative to the run TU03-B. In Table
4.8 the simulations whose results will be reported in following paragraphs are
summarised, classified by grid, turbulence model, number of iterations.
The chosen direction for the incoming wind is the 210 measured clockwise
from north as done for simulations with Fluent, in this way WindSim considers
two faces as inlets and two as outlets, this is also an important difference
existing between the simulations performed with Fluent and WindSim. The
computations performed with WindSim are meant, like the ones carried out
210
line A
line AA
line B


79
with Fluent, to reproduce the wind field observed during the experimental run
TU03-B, MF03-D and TK03 of the report Taylor and Teunissen (1985) [36].

Table 4.8 Set of simulations of Askervein (from map-B) run with WindSim 4.5
Turbulence models
SKE default SKE mod.
Grid 1 ( (( (u
12
) )) ) ( (( (u
1
) )) )
Grid 2 ( (( (u
2
) )) ) ( (( (u
2
) )) )
Grid 3 ( (( (u
12
) )) ) ( (( (u
12
) )) )

Number of iterations
SKE default SKE mod.
Grid 1
5500(1)/3000(2)
4000
Grid 2
3000
1100
Grid 3
3000(1)/3000(2)
4000(1)/3000(2)

Friction velocities
u
1
0,422 m/s u(500m)=10 m/s
u
2
0,6336 m/s u(500m)=15,02266 m/s (TU03-B)
4.2.4.1 Results: mean flow and turbulence
The calculated mean flow-field is compared to the experimental
measurements reported in table A1.3 and A1.4 of Taylor and Teunissen (1985)
[36]. The computed horizontal velocity along the line A is plotted in Figure 4.20
showing that also WindSim underestimates the velocity magnitude at the top of
the hill and overestimates at the lee-side, as already observed in the Fluent
simulations. The grids 2 and 3 provide basically the same results in terms of
velocity and the grid 1 differs from the other two meshes in the lee side of the
hill at about 0,5 km from the hill top which is very likely connected to the high
level of turbulence predicted. The grid dependency is shown also in Figure 4.21
by means of the TKE predicted along line A, the simulation carried out with the
grid 1 was probably close to diverge since the too high value of TKE obtained


80
at the lee side, the high value of TKE remained restricted in the lee area but the
peak of TKE about 10 m
2
/s
2
is not verisimilar. The grid 2 predicted a TKE peak
higher than the one computed with grid 3, showing the grid dependency of the
results; finally the flow field obtained with grid 3 is very similar to the one
calculated with Fluent in a similar horizontal resolution showing not many
differences between results obtained by the two numerical tools.
Also the effects of the usage of both sets of SKE models constants, see
Table 4.5, have been carried out, the numerical results are presented in the same
manner adopted for the grid dependency study, in Figure 4.22 is plotted the
horizontal velocity and in Figure 4.23 the TKE both calculated along line-A
passing through the hill top. In terms of velocity the usage of the two sets of
constants is not so significant, the two curves in Figure 4.22 do not differ that
much; the usage of the two sets of constants produce instead different values of
TKE which is generally higher for the modified set due to the smaller C

, while
the predicted peak of TKE is the same in the two cases resulting in a generally
smoother turbulence field when applying the modified set of constants.


81

0
2
4
6
8
10
12
14
16
18
20
-1000 -800 -600 -400 -200 0 200 400 600
line A horizontal position [m]
v
e
l
o
c
i
t
y

[
m
/
s
]
measurements
WindSim Grid 1 SKE (def)
WindSim Grid 2 SKE (def)
WindSim Grid 3 SKE (def)

Figure 4.20 Grid dependency: horizontal velocity along the line A, reproduction of
experimental data TU03-B.
0
2
4
6
8
10
12
-1000 -800 -600 -400 -200 0 200 400 600
line A horizontal position [m]
T
K
E

[
m
2
/
s
2
]
measurements
WindSim Grid 1 SKE (def)
WindSim Grid 2 SKE (def)
WindSim Grid 3 SKE(def)

Figure 4.21 Grid dependency: Turbulence Kinetic Energy (TKE) predicted with
WindSim 4.5 and measured (TU03-B) along line A at 10 m a.g.l.


82

0
2
4
6
8
10
12
14
16
18
20
-1000 -800 -600 -400 -200 0 200 400 600
line A horizontal position [m]
v
e
l
o
c
i
t
y

[
m
/
s
]
measurements
WindSim Grid 3 SKE (def)
WindSim Grid 3 SKE(mod)

Figure 4.22 Models constants dependency: horizontal velocity along the line A,
reproduction of experimental data TU03-B.
0
1
2
3
4
5
6
-1000 -800 -600 -400 -200 0 200 400 600
line A horizontal position [m]
T
K
E

[
m
2
/
s
2
]
measurements
WindSim Grid 3 SKE(def)
WindSim Grid 3 SKE(mod)

Figure 4.23 Models constants dependency: Turbulence Kinetic Energy (TKE)
predicted with WindSim 4.5 and measured (TU03-B) along line A at 10 m a.g.l.


83
Chapter 5 SIMULATIONS 3D DES/LES OVER FLAT
TERRAINS*
5.1 INTRODUCTION
Present literature and in-house experience pointed out that RANS approach
is insufficient to provide a satisfactory simulation of flows bounding complex
terrains features or architectural structures, leading to explore the application of
more precise turbulence solving methodologies such as Large Eddy Simulation
(LES) and hybrid methods URANS/LES such as the Detached Eddy Simulation
(DES). Further CFD techniques are obtained once LES or hybrid URANS/LES
are coupled with the use of wall functions to model the flow in the wall adjacent
cells, in this case the Wall Modelled LES (WMLES) is used rather than pure
LES. In this way turbulent flows with Reynolds numbers up to 10
9
are
affordable. Moreover, the turbulence plays the main rule to determine fluxes of
momentum, energy and moisture within the atmosphere. An accurate prediction
of the turbulence within the ABL is therefore necessary and an application of
eddies solving methodologies like LES and DES is nowadays simulating the
research since the improved capabilities of the modern computers and the lack,
up to now, to predict in a correct way the logarithmic wind profile in the surface
layer by LES methods.
In the present chapter will be described a series of DES and WMLES aimed
at evaluating the capabilities of the commercial code Fluent 6.2 in reproducing a
neutral ABL with eddy solving methodologies.
Since the scope of the presented simulations was to model a neutral
stratified flow, only the continuity and momentum equations have been solved;
a grids resolution study have been performed to determine the proper
discretization for all the different classes of terrain roughness.
* The present chapter is a rearranged version of the DC projects thesis
Investigation of the Atmospheric Boundary Layer applying Large Eddy Simulation
carried out at the von Karman Institute during the academic year 2005-06.


84
Boundary conditions for eddies solving methods should have some
characteristics, for instance, the vertical planes parallel to the flow cannot be
symmetry planes as done in RANS simulations, since symmetry planes kill the
turbulent fluctuations and the formation of the eddies themselves; periodic
conditions are therefore used in the span-wise direction. When the ABL to be
studied is a fully developed flow over a flat terrain periodic conditions stream-
wise are again successfully applied.
Inlet conditions with imposed velocity and turbulence are not taken into
account in the present PhD thesis but they deserve to be considered in
alternative to stream-wise periodic conditions especially when the ABL
considered flows not anymore over a flat terrain but over topographic features.
In this way the size of the domain itself can be kept not too wider than the
region of interest. For a neutral ABL such those studied and presented in this
chapter, the upper boundary can be imposed as a symmetry plane while wall-
functions are still used for the terrain, as done in the RANS simulations.
5.2 STUDY OF THE ABL WITH LES TECHNIQUES
In the last two decades several studies have dealt with the reproduction of
the ABL with eddies solving techniques, both LES and hybrid RANS/LES
methods (DES). LES of neutral stratified boundary layers have been presented
by Mason and Thomson (1987) [24] using a standard Smagorinsky SGS model
in the simulation of an Ekman layer. The same authors proposed in 1992 [25]
the use of the Smagorinsky model coupled with a back-scattering technique to
cure the failure of the simulations to well reproduce the logarithmic law,
equation (1.3). In fact, as also mentioned by Mason and Thompson [25], when
the neutral ABL is simulated with the original Smagorinsky-Lilly model, an
overshoot in the velocity is typically obtained with values of normalised wind
shear
m
reaching a maximum about 2.
In recent years two approaches have been followed in order to improve the
simulations: one is the refinement of the grids in order to set a proper filter


85
width, while the second one is the usage of improved and generally more
advanced SGS models compared to the standard Smagorinsky-Lilly.
Sullivan P.P. et al. (1994) [34] proposed a two-part eddy viscosity model
where the contribution of mean and fluctuating eddy-viscosities are introduced
and their effects are balanced by means of an isotropy factor.
Among most recent works dealing with LES simulations of ABL its worthy
to mention the Ph.D. thesis of F.T. Chow (2004) [7] where a dynamic
reconstruction model is applied to reconstruct the resolvable sub-filter-scale
(RSFS) stresses together with the dynamic eddy-viscosity model by Wong and
Lilly (1994) [42] for the sub-grid-scale stresses (SGS). Furthermore, an
example of DES applied to the planetary boundary layer is given by the multi-
planet EPIC General Circulation model proposed by Dowling T.E. and LeBeau
R.P..
5.3 METHODOLOGY FOR THE DES/LES SIMULATIONS OF THE
ABL
The flow field computed by both DES and LES methods need to be started
from a RANS steady solution properly perturbed in order to make develop the
turbulence more easily. Few pass-through times has to be waited for before
gather statistics, that should be done when statistically steady conditions are
reached.
5.3.1 Initial steady simulations
All the unsteady simulations, both DES and LES, 2D and 3D cases are
started from a converged solution of a RANS simulation of the ABL using the
SKE model for the turbulence closure.
The grids employed were all of structured type with uniform horizontal
spacing and ranging from 20 m to 60 m of size. For the vertical discretization of
the domain the prescription to assign the height of the wall adjacent cell h
higher than twice RH (Roughness Height) it has been followed, but the more
accurate results in term of predicted wall-shear stress are obtained with the


86
value of h only slightly higher than 2 RH, the rest of the vertical spacings are
assigned in order to obtain a successive ratio about 1,1; the meaning of the
parameter RH appearing in the wall function for rough surfaces is explained in
the Fluent Users Guide [10].
The initial RANS simulations are run with the SKE model, the air is treated
as uncompressible and dry; about the operating conditions the operating
pressure is set at 1 atm at the ground level, the gravitational acceleration is
enabled, while the operating density has been set to zero.
As regards the boundary conditions employed, the upper face of the domain
has been set as symmetry, while periodic boundaries have been used, only
streamwise in 2D and both spanwise and streamwise in 3D. The mass flow has
been assigned, usually calculated by assuming that a log-law is valid in the
whole domain. The terrain has been treated with a wall boundary condition,
using wall functions for rough surfaces, usually with RC (Roughness Constant)
one and RH calculated by the equation:
0
z E
RH
RC
, (5.1)
Which is obtained by the (2.33) with C
S
and K
S
respectively equal to RC and
RH. The equation (5.1) comes from the comparison of (1.3) and the wall-
function modified for rough surfaces as it is described in the Fluent users
manual [10].
Even if the maximum value accepted by Fluent for the RC parameter is one
(value accepted 01) thorough the graphical user interface, its also possible to
assign a value of RC higher than one, by reading a profile file for the lower
boundary where values of RC are specified for each cell of the terrain. For
instance RC can be set to E and RH to z
0
which also allows to draw a finer grid
close to the wall and to obtain a more precise wall shear stress forecast.
The described roughness settings for a finer grid with the parameter RC
higher than unity have been set successfully for RANS SKE simulations and


87
there is hope that their use will provide some improvements also in LES and
DES simulations.
As regards the discretization methods adopted, the default schemes have
been left active but the pressure discretization scheme is turned to PRESTO!
(PREssure STaggered Option) which simulates a staggered grid helping the
convergence process. The upper limit of the ratio eddy viscosity to laminar
viscosity is increased in such a way that the code wont prevent this ratio to
grow, in fact the eddy-viscosity for an ABL is naturally very high so there is no
reason to limit its value.
Finally the RANS simulation is run until reaching a complete convergence,
with the residual at minimum possible value, for a grid counting 125000 cells
this condition is generally reached in 6000 iterations.
Once the solution of the RANS simulation reaches the convergence, the
command solve>initialize>init-instantaneous-vel perturbs
the flow field calculated, the velocity field is in this way prepared to start up the
LES or the DES simulation. The initialization of the flow field superimposes
over the RANS final solution a synthetic turbulence. In this way its easier to
make develop a natural turbulence, i.e. ruled by the solved equations. In some
cases a turbulence will be never obtained without a initialization command
since turbulence needs instabilities and some perturbations in the flow field to
born and grow. The kind of instability present in a fully developed boundary
layer needs a perturbation of a certain amplitude in order to obtain turbulence;
this is not the case of other instabilities for instance of turbulent wakes, jets and
mixing layers where all scales perturbations, also very small and hence always
present, are amplified and finally lead to turbulence.
Unfortunately, in Fluent users guide [10] there is no further description
about the initialization of the instantaneous flow field even thought its very
likely that the same command is used in producing a synthetic turbulence for
the perturbed inlet conditions. In this case the perturbation of the flow field is
obtained by the free-divergence random-flow-generation function proposed by
Kraichnan (1970) [20] and modified by Smirnov et al. (2001) [30], this


88
command is based on the summation of a finite number of Fourier harmonics
(100 in the code Fluent) in order to construct a synthetic turbulence spectrum.
In the code users guide [10] its not mentioned if the initialization
command is available only for some turbulence models. Nevertheless it has
been noted during the performed tests that the initialization of the instantaneous
velocities is not valid when the SA turbulence model is activated, this is the
reason why the RANS simulations performed before the initialization of the
flow field are done with the standard k- model rather than the SA model,
which is the most natural choice when the scope is to switch to the SA based
DES.
In order to have a feeling of the effect of the initialization command
described on the flow field it can be seen from Figure 5.1 and Figure 5.2 how
the command produces a perturbation on the original velocity field (velocity
profiles of Figure 5.1) obtaining the velocity profiles like the ones shown on the
lower part of Figure 5.1; in Figure 5.2 can instead be seen how the turbulence
kinetic energy is not directly influenced by the perturbation command.



89

Figure 5.1 Example of vertical profiles stream-wise component of velocity for a
converged RANS SKE simulation (above) and the same value after initialization
command (below).


90

Figure 5.2 Examples of TKE vertical profiles for a converged RANS SKE (above) and
the same values after initialization (below).
5.3.2 Turbulence modeling
The study presented in this chapter dealt with the LES/DES simulations of a
neutral ABL over a flat terrain using the commercial code Fluent 6.2; hence
only the continuity and momentum equations have been employed, paying
attention to determine the sensitivity of the model to different parameters such


91
as grid resolution, roughness parameters and turbulence models parameters
(SGS).
Since the LES option in Fluent 6.2 can not operate with wall-functions
adapted for rough walls the choice of the turbulence treatment has been left in
the first part of the study to an hybrid URANS/LES methodology named DES
(Detached Eddy Simulation) based on a modification of the Spalart-Allmaras
model, while in a second moment of the study an home-made Smagorinsky-
Lilly model has been plugged in the code in order to use it coupled with the
wall-functions modified for the roughness.
In the eddies solving methods LES only the big-eddies are solved directly
(resolved scale) while the small-scale eddies effects are modelled with
turbulence models named sub-grid-scale (SGS) models, or more correctly the
sub-filter-scale (SFS).
A peculiarity of eddies solving methodologies is that the flows have to be
modelled always in unsteady simulations; also a 2D approach is theoretically
wrong since eddies are correctly developed only in tri-dimensional space.
Nevertheless 2D LES can be carried out and it has been successfully applied
especially in some cases of external aerodynamics of bluff bodies.
5.3.3 The new LES-UDF implemented in Fluent
The code Fluent 6.2 offers four methodologies to model the SGS eddy
viscosity
t-SGS
, four SGS models:

the Smagorinsky-Lilly (SL) model
the dynamic SL model
the Wall-Adapting Local Eddy-Viscosity (WALE) model
the dynamic kinetic energy model

All the listed models are not available coupled with the wall-functions for
rough walls while the DES option does. Therefore the need in the present


92
project for DES turbulence modeling or the implementation of a new SGS
turbulence model to be used coupled with the wall-functions.

In the standard Smagorinsky-Lilly SFS model, which has been implemented
in a new UDF and used for rough surfaces in the present project, the
t-sgs
is
modelled by:
( )
2
t sgs S
C S

= (5.2)
Being C
S
the Smagorinsky constant, the filter width (=
1/ 3
cell volume )
and S the magnitude of the resolved scale rate of strain tensor
( )
1/ 2
ij ij
S 2S S = (5.3)
5.3.4 DES settings tested
The SA based DES model is implemented in Fluent 6.2 as proposed by Shur
et al. (1999) [28], two further modifications of the model are available:

1. the possibility to substitute the length scale d only in the destruction term
2. and to compute the filter size from the volume of the cell in equation
(5.5) rather than on the maximum dimension (5.4).

( ) = max x, y, z (5.4)
=
3
cell volume (5.5)
The possibility to choose two values of filter width and the replacement of
the new length scale (2.29) in both the production and destruction term or only
in the destruction term lead to four possible settings of the DES model which
are named and summarized in the Table 5.1.
Its worthy to remember that the choices shown in Table 5.1 are available in
Fluent only through the text user interface.


93
The four possible DES models have been tested in 2D domains both for
smooth and rough walls, the model more similar to a LES with a Smagorinsky-
Lilly model has been finally chosen for the 3D simulations.

Table 5.1 classification of the four DES options in Fluent 6.2.

( ) = max x, y, z
=
3
cell volume
( ) =

DES
d min d, C
replaced in all the terms
(production + destruction)
DES1 DES2
( ) =

DES
d min d, C
replaced only in the
destruction term
DES3 DES4

Furthermore, if the variable S is computed with equation (2.8) and changing
the value of some constants (C
prod
and C
w3
)of the model further possible settings
are available; some of the possible settings have been tested on 2D and 3D
boundary layers, the settings tested are:

Case a: S based on vorticity;
Case b: S based on vorticity/strain ( equation (2.10)), C
prod
= 2, C
w3
= 2;
Case c: S based on vorticity/strain ( equation (2.10)), C
prod
= 1, C
w3
= 2;
Case d: S based on vorticity/strain ( equation (2.10)), C
prod
= 1, C
w3
= 0

Where all the other constants are left with the default values. The reason
why the C
prod
is set to 1 is to make S S when > S while the constant
C
w3
is set to 0 to make f
w
in the destruction term equal to 1.



94
Considering the transport equation for the modified eddy viscosity adopted
in the SA model, equation (2.8), in equilibrium conditions (production term =
destruction term) the following relation holds for the settings DES4d (which
means DES4 of Table 5.1 and case d of the previous list):
( ) =

2
b1
t des
w
C
C S
C
(5.6)
Comparing the equation (5.6) to the (5.2) and considering that far from the
wall
t t
and

S S it turns out that

2 2 b1
S des
w
C
C C
C
(5.7)
In order to make the Smagorinsky LES and the SA based DES models
predict the same order of magnitude of SGS turbulent viscosity the equation
(5.7) has to be respected.
5.4 2D TESTS
Before dealing with the description of the 3D simulations performed of a
neutral ABL, that will be presented in the paragraphs 5.5 and 5.6, the
preliminary tests 2D are going to be described in the present one. The DES and
LES tests 2D do not pretend to solve correctly the ABL since eddies are 3D
structures and only 3D simulations can solve them. The 2D simulations are
meant to test quickly some features and parameters of the models employed
even if a real response about the quality of a model can only come from the tri-
dimensional simulations. In 2D LES/DES, as has been observed , turbulence 2D
is hard to form especially when no important obstructions to the flow like bluff
bodies are present. For instance in the 2D studied cases of flat terrains that are
going to be presented in this chapter the perturbations superimposed with the
initialization command did not develop into turbulence.
All the 2D tests have been performed on a simple geometry, a rectangle 6
km long and 1 km high representing a flat terrain. The boundary conditions


95
employed are translational periodic for the inlet/outlet, wall functions for the
ground and symmetry for the upper boundary.
The grid employed is uniform horizontally while a stretching has been
prescribed for the vertical discretization, the minimum height of the cell is
named h and its the height of the wall adjacent cell; in all the grids drawn the
successive ratio has been kept around 1,1.
The methodology followed for the simulations is the one described in the
previous paragraph 5.3. Several tests have been performed over smooth walls
for DES, LES and LES with a user defined function implementing a
Smagorinsky model (from this point named simply LES-UDF), and, over rough
walls, for DES and LES-UDF.
While the 2D simulations over smooth terrains are meant more to compare
the LES option of Fluent and LES-UDF to the several tried settings of the DES
module, the tests over rough surfaces allowed comparisons of DES settings only
to LES-UDF.
5.4.1 2D boundary layers on smooth walls
2D simulations have been run over a smooth flat terrain always using a wall
function to model the ground. With this series of simulations it was possible to:
a) compare LES and DES
b) make a study of sensitivity to the parameter C
DES

c) make comparisons between different settings of the DES model

The 2D simulations are aimed at reproducing a fully developed boundary
layer by applying a translational boundary condition. The numerical results are
presented by vertical profiles of velocity, presented in wall units u
+
= u/u

and
y
+
= y u

/ since the flow is over a smooth surface and it is therefore possible to

compare the results with the universal logarithmic law of a turbulent boundary
layer.


96
5.4.1.1 2D boundary layers on smooth walls: comparing LES against DES
Two vertical profiles of normalized velocity are shown in Figure 5.3 in wall
units, one obtained with a DES and the other with LES with a Smagorinsky-
Lilly SGS model and the position of the switching interface is shown with a
dashed line; in the RANS region of the DES the universal logarithmic law is
predicted correctly while above the RANS region the profile deviates from the
log-law, this is due to the fact that in 2D simulations no turbulent structures at
all has been developed, so the only eddy-viscosity observed is the one produced
by the SGS model. From the same Figure 5.3 it can be seen that the LES model
of Fluent acts effectively as an other DES model in the sense that there is a
region unsteady RANS, where in fact a mixing length model is applied and at
an interface the model switches to pure LES according to equation (2.29).


Figure 5.3 Velocity profiles (u
+
, y
+
) for DES 2D (C
DES
0,065) and LES 2D (C
S
0,1) of
flat terrain. Dashed line is the position of the interface RANS/LES for the DES model.


97
5.4.1.2 2D DES boundary layers on smooth walls: C
DES
sensitivity
The constant C
DES
influences directly the position of the interface
URANS/LES, as its shown in Figure 5.4; in the DES model the interface is
located at the wall-distance C
DES
, moreover the constant C
DES
affects the
magnitude of the SGS turbulent viscosity exactly like the Smagorinsky constant
C
S
(see the procedure followed to obtain equation (5.7) and Gunot and Aupoix
(2003) [16]).

Figure 5.4 Velocity profiles (u
+
, y
+
) for DES 2D (C
DES
0,065 and C
DES
0,65). Dashed
lines are the positions of the interfaces RANS/LES.
Lilly derived a value of 0.17 for the Smagorinsky constant C
S
for
homogeneous isotropic turbulence in the inertial sub-range; according to the
equation (5.7) a value of C
DES
0.827 is obtained; these values of constants are
generally accepted to be too dissipative and smaller values of C
S
are therefore
used. Since the constant C
DES
is directly connected to the SGS eddy viscosity
and its value cannot change too much from the default value 0.65, the position
of the interface, which is placed at C
DES
from the wall, is correctly controlled
only by a proper refinement of the grid in the wall adjacent region since the


98
filter width is computed from the cells geometric characteristics by the (5.4)
or the (5.5).
5.4.1.3 2D boundary layers on smooth walls: DES settings
Four different settings have been tested on smooth surfaces, numbered from
DES1 to DES4 according to Table 5.1. All the cases produced normalised
velocity profiles departing from the log law from the interface URANS/LES
towards higher y
+
(LES region) as can be observed from Figure 5.5.
The filter width for cases 1 and 3 is computed by the max cell dimension
while for cases 2 and 4 is computed by the 1/3 power of the cell-volume; in
fact for the cases shown in Figure 5.5 the constant C
DES
has not been modified
and the position of the interface clearly changes with the filter width.
When comparing the cases shown in Figure 5.5 the higher wall shear is
obtained for cases 3 and 4 where the modification of the distance in the SA
model is applied only in the destruction term which leads also to make

S closer
to S, and therefore the model closer to a Smagorinsky model.


Figure 5.5 Normalized velocity profiles in wall-coordinates DES 2D (smooth), cases
1, 2, 3 and 4 of Table 5.1.


99
Since the case 4 is the more similar to a Smagorinsky model it has been
chosen to be used for the 3D simulations. There is a good experience in the
usage of the Smagorinsky model, even if its known that eddy-viscosity SGS
models like Smagorinsky fails for instance in predicting the log-law expected
from the similarity theory.
To make the case 4 even more similar to the Smagorinsky model other three
settings have been tried, named 4b, 4c and 4d already listed in the paragraph
5.3.4. With the configuration 4d S is computed only from the strain rate tensor
in the region where > S while in the other configurations 4a, 4b and 4c S is
computed from a linear combination of and S .
Velocity profiles for the further settings of the case 4 are shown in Figure
5.6 in wall-coordinates. It can be seen that the velocity profile for the case 4d is
quite different from the ones obtained with the other settings. The main task of
this series of tests was to find a set of parameters for the SA based DES which
make the hybrid model the closest possible to a Smagorinsky model, therefore
further comparisons of DES 4d with LES and LES-UDF will be presented.
1 E+00
1 E+01
1 E+02
1 E+03
1 E+04
1 E+05
1 E+06
1 E+07
1 E+08
0 20 40 60 80 100 120 140 160 180
u+
y
+
universal log-law
case 4a
case 4b
case 4c
case 4d
u 5,5 2,5ln(y )
+ +
= +
u y
+ +
=

Figure 5.6 Normalized velocity profiles for the four further settings tried for the case
DES 4.


100
In the following Figure 5.7, Figure 5.8 and Figure 5.9 its shown as with the
DES 4d, LES and LES-UDF similar results are obtained in terms of velocity
profiles and eddy-viscosity profiles.
Several tests have been performed on 2D domains and smooth walls, the
main conclusions coming from this series of settings is that the DES 4d, LES
and LES-UDF are almost equivalent.
Observing the eddy-viscosity profiles of Figure 5.9 it can be noticed how in
the LES and DES cases the eddy viscosity decreases to zero closer to the wall,
while in the LES-UDF case it approaches a value that is around 0,0002 Pa s, a
value that is roughly ten times the molecular viscosity.
This defect is indeed cured in the LES in Fluent by defining a mixing length
given by
( )
=
1/ 3
S S
L min d, C V (5.8)
and the SGS eddy-viscosity is computed by
2
t sgs S
L S

=
In this way also the LES model acts in hybrid way, with a mixing length
model between the wall and the interface and in LES model in the core of the
flow therefore, now the interface results placed at C
S
from the wall.
Depending on the aspect ratio of the wall-adjacent cells the model can act as
LES in the all cells or not even without the correction in (5.8), this could be the
case of rough walls where not an extreme vertical refinement is required at the
wall and more cubic wall-adjacent cells are used.


101

Figure 5.7 Vertical profiles of turbulent SGS viscosity, comparison of LES (default
Smagorinsky of Fluent) and DES with settings 4d.

1 E+00
1 E+01
1 E+02
1 E+03
1 E+04
1 E+05
1 E+06
1 E+07
1 E+08
0 50 100 150 200 250 300 350 400 450 500
u
+
y
+
universal log-law
DES 4d (Cdes 0,065)
LES (Cs 0,013)
LES-UDF (Cs 0,013)
u 5,5 2,5ln(y )
+ +
= +
u y
+ +
=

Figure 5.8 Normalized velocity profiles (in wall-coordinates), DES 4d, LES and LES-
UDF cases.


102

Figure 5.9 Vertical profiles of SGS eddy viscosity, comparisons of LES (default
Fluent), LES-UDF and DES (case 4d).
5.4.2 2D boundary layers on rough walls
The 2D simulations on smooth walls have been mainly done in order to
compare DES settings with LES and LES with LES-UDF. For the rough walls
the LES is not available and a confirm of the similarity of DES 4d and LES-
UDF is here shown.
In the present work three values of roughness for terrains have been
analyzed, they are summarized in Table 5.2 with indicated RH and RC, the
input parameters of the code for the roughness and h standing for the height of
the wall-adjacent cell.
Table 5.2 Meshes used for the rough walls 2D tests.
RH [m]
class z
0
[m]
used z
0

[m] First Tuned
RC
(01)
h
[m]
IV 0,010,04 0,03 0,2943 0,25 1 0,6
0,16 1,567 1,17 1 3,2
VI 0,1 1
0,5 4,896 3,39 1 10


103
The values of roughness chosen are not going to be presented for the 2D
simulations, while further results are presented in the following paragraphs
about 3D simulations. As regards the 2D simulations presented in the next
paragraph it will be shown the equivalence achieved between DES 4d and LES-
UDF when dealing with rough terrains.
5.4.2.1 LES-UDF and DES4d on rough walls
Comparison of velocity profiles and SGS viscosity profile are shown for the
DES 4d case and the LES-UDF with the constant Cs and C
DES
properly set,
Figure 5.10 and Figure 5.11. The two models give very similar results and its
therefore demonstrated that in equilibrium conditions the DES 4d behaves very
closely to the LES with Smagorinsky model. Due to the relative good
confidence in the Smagorinsky model it has been decided to run the DES 4d
and the LES-UDF for the 3D simulations.


Figure 5.10 Vertical profiles of velocity for DES 4d and LES-UDF.



104

Figure 5.11 Vertical profiles of SGS eddy viscosity for DES 4d and LES-UDF.
5.5 SIMULATIONS DES 3D
A series of DES 3D simulations over flat terrains have been carried out with
some of the settings tested in 2D tests, presented in previous paragraphs. The
simulations were performed over a domain 3 km wide in both horizontal
directions and 1 km high. The relative numerical results are presented in the this
and following paragraphs; a roughness sensitivity study has been carried out
considering three values of roughness length: 0.03 m, 0.16 m and 0.5 m.
The roughness length of 0.16 m was chosen in order to make comparisons
with the numerical results by Sullivan et al. (1994) [34] whose simulations were
performed for the same value of roughness length, same domain size but higher
resolution both horizontal and vertical, different boundary condition for the top
surface. Moreover, Sullivan et al. (1994) [34] adopted for their LES simulations
a different SGS model and different C
S
equivalent parameters, comparisons
with the presented results are therefore more qualitative than quantitative.
The DES model with settings 4d has been tested on a domain whose
discretization main features are listed in table below while the same geometric


105
features of the discretised domain used by Sullivan et al. (1994) [34] are given
in the Table 5.4.

Table 5.3 Grids features of the three simulations compared.
z
0

[m]
Lx, Ly
[m]
Lz
[m]
Nx, Ny Nz N x, y
[m]
h
[m]
0,03 3000 1000 50 50 125 000 60 0,6
0,16 3000 1000 50 50 125 000 60 3,2
0,50 3000 1000 50 30 75 000 60 10

Table 5.4 Grids features of the discretised computational domain used by Sullivan et
al. (1994) [34].
z
0

[m]
Lx, Ly
[m]
Lz
[m]
Nx, Ny Nz N x, y
[m]
h
[m]
0,16 3000 1000 96 96 884 736 ~ 30 -

The settings chosen for the DES model are the ones described in the
paragraph 5.3.4 and identified as 4d which resembles the closer possible the
Smagorinsky SGS model. The present simulations were aimed to assess the
sensitivity of the DES model to different values of roughness of the terrain. The
values of roughness length tested are 0,03 m, 0,16 m and 0,5 m, the first two
corresponding to areas of low vegetation, while the third one is typical of a
forested zone according to the classification of terrains, for instance the one
proposed by the European Wind Atlas [39] and reported in Table 5.5. The used
settings of the roughness parameters RH and RC for the three values tested have
been already presented in Table 5.2.
The constant C
DES
of the model has been set to 0,486 corresponding to a
Smagorinsky constant C
S
of 0,1 according to equation the equation (5.7) and
Spalart-Allmaras models constants as in Table 2.1.


106
Table 5.5 Classification of the terrains based on the roughness length z
0
according to
the Wind Atlas [39].
z
0
[m] Terrain typology Roughness Class
1-0,5 City Forest 3
0,5 0,2 Suburbs Shelter belts 3
0,2 Many trees and/or bushes 2-3
0,1 Farmland with closed appearance 2
0,03
Farmland with very few buildings,
trees, etc.
1
0,01 Airport runway areas mown grass -
3 10
-4
- 1 10
-4
Sand and water surfaces (smooth) 0

All the run DES 3D simulations were started from a perturbed RANS
solution. The DES simulations were first left run for ten flow-through times in
order to achieve a statistically steady turbulence, afterward the time statistics
were gathered over an equal time window of ten flow-through times. The flow-
trough time has been computed on the maximum velocity rather than on the
bulk velocity.
The statistics gathered by default in Fluent 6.2 are the mean and the RMS
values of the static pressure, velocity magnitude, x, y and z component of the
velocity.
A more complete study of the turbulence would be achieved by calculating
the co-variances of velocity components and higher order moments, while a
spectral analysis of turbulence can also help to understand if the turbulence has
been simulated correctly.
Nevertheless in the following paragraphs the most significant results are
presented, usually as vertical profiles of variables averaged on horizontal
planes.
The boundary conditions employed for both DES and LES simulations are
slightly different than those usually employed in RANS simulations, for a flat
terrain is convenient to set both the stream-wise and the span-wise borders as


107
translational periodic, in this fashion a fully developed situation is easily
reproduced allowing at the same time to maintain a rather correct shape of the
vortices in proximity of the lateral borders; the use of symmetry plane as lateral
boundaries would kill all the turbulent fluctuations. The top of the domain is
treated as a symmetry plane while the bottom as a wall with wall functions,
these last two settings remained therefore the same used for RANS simulations.
A mass-flow has been assigned to the translational periodic condition, Table
5.6, aiming to produce a wind velocity of 15 m/s assuming that a log-law
(equation (1.3)) would be valid over the whole domain.

Table 5.6 Mass flows assigned at the periodic boundaries and physical time step.
z
0

[m]
Mass-flow
[kg-s]
Time-step
[s]
Max CFL
cell number
0,03 4,983 10
+07
4 1
0,16 4,883 10
+07
4 1
0,50 4,787 10
+07
4 1
5.5.1 Wall-shear stress
The first type of numerical data that is going to be shown is the evaluation
of the wall-shear stress by the code Fluent.
The contour plots of wall shear stress on the ground (Figure 5.12 for the
smaller value of roughness tested, z
0
0.03 m) in the three cases studied show
that the wall shear stress is higher in elongated areas parallel to the flow which
are patterns of eddies organized in elongated structures whose length is not
estimable since the periodic conditions make them like infinite long. The same
behaviour has been described by Mason and Thompson (1992) [25] for
computations performed with the standard Smagorinsky model. The authors
claimed that the observed elongated structures are similar to the near-surface
streaks found in low Reynolds number simulations of turbulent boundary layers.
For this reason the observed elongated structures are named by some authors


108
super-streaks. Super streaks are structures not observed in actual turbulent
boundary layers, they appeared also in hybrid RANS/LES simulations of a
turbulent boundary layer, where the interface is placed in the logarithmic region
around 30 y
+
. In the mentioned hybrid RANS/LES simulations a shifting of the
velocity profile in the logarithmic region is observed, this shift is accompanied
by the presence of the super-streaks. In recent studies, for instance by Piomelli
et al. [26], the introduction of a stochastic forcing at the interface allowed to
reduce the shift in the velocity profile in the proximity of the interface while the
super-streaks structures disappeared. A rather similar phenomena is observed in
a WMLES and DES of the ABL over rough surfaces. A stochastic forcing of the
velocity field could likely improve the simulations performed.
Unfortunately, the shortness of the domain together with the stream wise
periodic conditions enhances the permanence of elongated structures; with the
hazard of construct structure infinitely long if the size of the structures is
comparable to the length of the domain.
By averaging the value of wall shear stress predicted by the DES
simulations values are obtained that are smaller than the wall shear stress
calculated by RANS SKE simulations and even smaller than the reference value
which is computed considering valid the log-law (1.3) for the whole domain and
for the given mass-flow (Table 5.6); the three values of wall-shear stress are
summarized in the table below, the corresponding friction velocities are linked
to the wall-shear stress by its definition, equation (1.1).

Table 5.7 Wall-shear stresses and friction velocities for log-law (1.3), SKE-RANS and
DES 4d simulations.
Logarithmic law RANS SKE DES 4d
z
0

[m]
x
[m]
w LOG LAW

[Pa]
ref
u

[m/s]
w SKE

[Pa]
ske
u

[m/s]
w
DES

[Pa]
u

[m/s]
0,03 60 0,445 0,603 0,387 0,562 0,167 0,370
0,16 60 0,633 0,719 0,514 0,648 0,245 0,447
0,50 60 0,836 0,826 0,643 0,724 0,374 0,553


109
There is a constant tendency for the DES 4d performed to underestimate the
friction velocity, the value calculated is generally 60% of the reference value.
The tendency to underestimate the wall-shear stress is directly connected to
the typical local deviation from a logarithmic profile (overshoot) in the
velocity profiles observed which occurs in the surface layer. Instead the velocity
gradient is higher than expected in the same zone where super-streaks are
observed.
According to Mason and Thompson (1992) [25] the presence of the
overshoot is connected to the lack of backscatter of turbulent kinetic energy
from smaller to bigger vortices; in the simulations by Mason and Thompson
(1992) [25] the lack of backscatter was due to the use of the Smagorinsky SGS
model which is a pure dissipative model as all the eddy viscosity models.
One of the questions to be answered in the present chapter is to understand
if the DES model predicts the overshoot typical of the Smagorinsky model,
and if this appears how much it differs from a log-law and from other cases.
It has to be pointed out that, when some variables will be presented
normalized by using the friction velocity, the underestimation of the friction
velocity can lead to an amplification or to a reduction of the normalized group.
Therefore both the non-dimensional groups based on computed friction velocity
and reference (based on log-law) friction velocity are presented in the following
sections of this report.



110

DES 4d C
DES
0,486 (z
0
0,03 m)

z
0
0,03 m z
0
0,16 m z
0
0,5 m
w
DES
= 0,167 Pa

w SKE
= 0,387 Pa

w LOG LAW
= 0,445 Pa

w
DES
w LOG LAW
0,375

w
DES
= 0,245 Pa

w SKE
= 0,514 Pa

w LOG LAW
= 0,633 Pa

w
DES
w LOG LAW
0,387

w
DES
= 0,374 Pa

w SKE
= 0,643 Pa

w LOG LAW
= 0,836 Pa

w
DES
w LOG LAW
0,447

Figure 5.12 Contours of wall shear stress [Pa] on the ground for the z
0
0,03 m
(above); averaged values for DES and RANS SKE are reported with their reference
value (log-law) for all the three cases.
5.5.2 Velocity profiles
In Figure 5.13 the vertical velocity profiles are shown, computed by
averaging on horizontal planes the already time averaged stream-wise velocity
component. Two trends are particularly clear, firstly, in the cells close to the


111
wall for which it results d < C
DES
, in the RANS region, a logarithmic
behaviour is fairly reproduced, in the interface RANS/LES the velocity profiles
deviate from the log-law, a behaviour typical also of simulations performed
with standard Smagorinsky-Lilly model, at a certain distance from the wall the
logarithmic profile is once again recovered.
The velocity profiles are then normalized with the friction velocity
computed and the one of reference (calculated assuming valid a log-law in the
whole flow field and with the given mass flow) and plotted against z/z
0

respectively in Figure 5.14 and Figure 5.15, its clear that the missing
overlapping of the curves obtained with DES 4d simulations and the log-law
curve of equation (1.3) come from the underestimation of the friction velocity;
from the collapsing of the curves in Figure 5.15 can be another time highlighted
how the expected friction velocity should be closer to the one of reference than
the one computed.
In Figure 5.16 a comparison is presented with the baseline model by
Sullivan et al. (1994) [34], in both cases the velocity profile deviates from the
logarithmic law, in the profile relative to the simulation DES 4d the low value
of predicted wall shear stress, i.e. of friction velocity, leads to amplify the value
of normalized velocity, moreover a value of the models constant C
DES
higher
than 0.486 should bring the profile computed to a shape closer to both the log-
law and the Sullivans baseline model.
In the Figure 5.17 the velocity profiles are shown in wall-coordinates u
+
y
+
,
this plot helps to understand if the boundary condition for the rough surfaces
has been applied correctly, expecting a value of B of equation (2.32)
corresponding to the given roughness length z
0
; in Figure 5.17 it can be seen as
the B seems to be better predicted for the lower values of roughness length.
In order to even better highlight and estimate the overshoot of velocity
given by the model the normalized wind-shear
m
defined by equation (1.4) is
shown in Figure 5.18, in the three cases studied the overshoot in
m
ranges
between 2,6 and 3,4 where the value of a log-profile is 1.


112
It can be pointed out that tests made with a grid with double horizontal
resolution (x = y = 30 m) and half horizontal dimensions (1500 m) of the
domain to keep the same number of cells, not shown here, gave a less dramatic
overshoot.
5.5.3 Eddy-viscosity profiles
In Figure 5.19 and Figure 5.20 vertical profiles of SGS turbulent viscosity
are shown, averaged on horizontal planes, the maximum of turbulent viscosity
occurs around 30 m of altitude regardless the roughness length. The poor
vertical resolution of the grid drawn for the most rough case doesnt allow to
make a good estimation of the SGS viscosity maximum. Nevertheless the
maximum value results higher for the higher roughness tested which is
influenced by the higher velocity gradients.

A way to normalize the turbulent viscosity is to multiply it by the vertical
gradient of the stream-wise velocity and divide by the wall-shear stress, the
non-dimensional group obtained, plotted in Figure 5.21, should assume a value
close to one at the surface if the SGS momentum flux is much higher than the
resolved scale one. The described non-dimensional group collapses to the same
value close to the wall for two out of three cases, for the most rough case the
group falls on a value slightly smaller than unity but the trend looks generally
acceptable.


1
1
3

z
0
0,03 m

z
0
0,16 m
z
0
0,50 m
Figure 5.13 Roughness sensitivity: velocity profiles, solid lines logarithmic laws, dashed lines horizontal plane averaged mean
velocities.


114


Figure 5.14 Velocity profiles normalized with u

and z
0
, solid line log-law (1.3),
dashed line z
0
0,03 m, dotted line z
0
0,16 m, dashed-dotted line z
0
0,50 m.


Figure 5.15 Velocity profiles normalized with u

and z
0
, solid line log-law (1.3),
dashed line z
0
0,03 m, dotted line z
0
0,16 m, dashed-dotted z
0
0,5 m.


115

Figure 5.16 Dashed line DES4d z
0
0,16 m, dotted line baseline model from Sullivan et
al. (1994) [34].

Figure 5.17 Velocity vertical profiles in wall-coordinates.


116



Figure 5.18 Normalized-wind shear profiles.



117

Figure 5.19 SGS turbulent viscosity profiles, solid line z
0
0,03 m, dashed line z
0
0,16
m, dashed-dotted line z
0
0,5 m.

Figure 5.20 Same values of Figure 5.19, semi-log coordinates.


118

Figure 5.21 Normalized SGS turbulent viscosity.
5.5.4 Velocity components variances profiles
The time RMS of the velocity components gathered by the code has been
space averaged on horizontal planes in order to collect a set of vertical profiles,
which are presented, both in dimensional form and in normalized form, in the
following, Figure 5.22 and Figure 5.23. The RMS of the velocity components
take into account only the resolved scale of the turbulence; the SGS part should
be added to the resolved scale in order to reflect the whole turbulence, its
influence is expected to be very important close to the terrain but negligible at a
certain distance from the wall, in the surface layer.
Both the collapse of the data and the values close to the wall expected from
similarity theory, equations (1.6), (1.7) and (1.8) lead again to evaluate as
underestimated the wall-shear stress by the code while a friction velocity closer
to the reference value looks more reasonable. Its certainly sure that for a more
precise judgement of the variances the filtered part should be evaluated and
summed.
t SGS
w
v
z

1
1
9


Figure 5.22 Vertical profiles of RMS of velocity components. Solid lines: z
0
0,03 m; dashed lines z
0
0,16 m, dotted
lines z
0
0,5 m.



1
2
0


Figure 5.23 Variances of velocity components normilized with predicted friction velocity. Solid lines: z
0
0,03 m; dashed
lines z
0
0,16 m, dotted lines z
0
0,5 m.



1
2
1


Figure 5.24 Variances of velocity components normilized with reference (log-law) friction velocity. Ssolid lines: z
0
0,03
m; dashed lines z
0
0,16 m, dotted lines z
0
0,5 m.



122
5.5.5 Turbulent Kinetic Energy profiles
Turbulent Kinetic Energy profiles of resolved turbulence scale are shown in
Figure 5.25, normalized values in Figure 5.26 and Figure 5.27. Since the TKE is
computed from the RMS of velocity components the behaviour shown is the
same observed for the RMS of the velocities plots, while an higher value of
turbulence is fairly reproduced for the higher roughness since the higher
velocity gradients, in the normalized profile a rather parabolic profile is shown,
which resembles the expected profiles.


Figure 5.25 TKE resolved scale vertical profiles for the DES4d simulations.


123

Figure 5.26 TKE resolved scale vertical profiles normalized with computed friction
velocity for the DES4d simulations.

Figure 5.27 TKE resolved scale vertical profiles normalized with reference friction
velocity for the DES4d simulations.


124
5.5.6 Comparison with literature
Among the cases presented in literature the simulations by Sullivan et al.
(1994) [34] have been chosen for comparison with the performed simulations,
even if the model used in [34] is not the same analysed in this section. In fact
Sullivan and collaborators compared two eddy-viscosity SGS models, both
considering an extra transport equation for the SGS energy e which in
equilibrium conditions (production of e equals dissipation ) resembles the
standard Smagorinsky model with C
S
0,18.
The main features for the meshes tested and the one in [34] are summarized
in Table 5.8. The horizontal resolution in the meshes used in the simulations by
Sullivan is approximately 30 m and, since its preferable compare results for
same resolutions, two DES4d simulations are here presented, one with x 60 m
and Lx 3000 m and one with x 30 m and Lx 1500 m, where for the latter the
resolution is fairly comparable with the resolution of the grids used in [34].
The filter width based on the cell-volume for the two grids employed is
plotted in Figure 5.28. In Figure 5.29 the normalized velocity profile and in
Figure 5.30 the normalized wind shear, both the figures show a better prediction
of the velocity field by the more refined grid, showing a grid dependency of the
results. Further improvements could be expected with a C
DES
higher,
corresponding to C
S
= 0,18.
Figure 5.31 and Figure 5.32 show vertical profiles of resolved scale TKE.
Figure 5.33, 5.34 and 5.35 show vertical profiles of RMS of velocity
components compared to Sullivans baseline model results and curves from
similarity theory. Not perfect agreement is found between the compared cases.
It has to be stressed that the vertical profiles presented are referred only to the
resolved scale while Sullivan added to the resolved scale RMS of velocity
components a SGS contribution calculated on the SGS eddy viscosity
What it can be seen from the vertical profiles Figure 5.33-5.35 of the RMS
on the velocity components is that there is a main tendency in under predict the
not-stream-wise, especially for the span-wise direction. Moreover the profiles


125
result comparable only in the case of similar resolution (x 30 m for DES4d
case).
Table 5.8 Grids features of the three simulations compared.
z
0

[m]
Lx, Ly
[m]
Lz
[m]
Nx, Ny Nz N x, y
[m]
h
[m]
DES 4d 0,16 m 1500 1000 50 50 125 000 30 3,2
DES 4d 0,16 m 3000 1000 50 50 125 000 60 3,2
Sullivan et al.
[34]
0,16 m 3000 1000 96 96 884 736 ~ 30


Figure 5.28 Filter width based on cell-volume: solid line domain 1,5 km x 1,5 km (x,
y 30 m), dashed line domain 3 km x 3 km (x, y 60 m).



126

Figure 5.29 Velocity profiles in non-dimensional units, solid line for domain 3 km x 3
km, dashed line domain 1,5 km x 1,5 km, dotted line Sullivans baseline model.

Figure 5.30 Normalized wind shear, solid line domain 3 km x 3 km, dashed line
domain 1,5 km x 1,5 km, dotted line Sullivans baseline model.


127

Figure 5.31 Vertical profile of TKE resolved scale.


128


Figure 5.32 Vertical profiles of normalized TKE.





1
2
9


<RMS vel> [m/s]
Figure 5.33 RMS of the velocities: solid lines stream-wise component, dashed lines span-wise component, dotted
lines vertical component; a) DES 4d with x 60 m (resolved scale), b) DES 4d with x 30 m (resolved scale), c)
baseline model [34] (total = resolved scale + SGS).



130
a)
0
100
200
300
400
500
600
700
800
900
1000
0 5 10
h
e
i
g
h
t

[
m
]



c)
0
100
200
300
400
500
600
700
800
900
1000
0 2 4 6
h
e
i
g
h
t

[
m
]


d)
0 2 4 6


Figure 5.34 Variances of the velocities normalized with computed friction velocity:
solid lines stream-wise component, dashed lines span-wise component, dotted lines
vertical component; a) DES 4d with x 60 m (resolved scale), b) DES 4d with x 30 m
(resolved scale), c) baseline model [34] (total = resolved scale + SGS), d) similarity
theory.


131
a)
0
100
200
300
400
500
600
700
800
900
1000
0 1 2 3 4
h
e
i
g
h
t

[
m
]


c)
0
100
200
300
400
500
600
700
800
900
1000
0 2 4 6
h
e
i
g
h
t

[
m
]

d)
0 2 4 6


Figure 5.35 Variances of the velocities normalized with reference friction velocity:
solid lines stream-wise component, dashed lines span-wise component, dotted lines
vertical component; a) DES 4d with x 60 m (resolved scale), b) DES 4d with x 30 m
(resolved scale), c) baseline model [34] (total = resolved scale + SGS) (normalized
with computed u

), d) similarity theory.


132
5.6 SIMULATIONS LES 3D OVER A ROUGH FLAT TERRAIN
5.6.1 A new LES-UDF to use with wall-functions
The only eddies-solving methodology which by default in Fluent allows to
deal with rough surfaces is the DES module (based on SA model in Fluent 6.2),
which have been tested and whose potentialities in reproducing an ABL have
been investigated and presented in the previous section 5.5. Even if the DES
model have been tested over a wide range of roughness length, not so
satisfactory results have been obtained up to now which led to the idea to
explore the possibility of implementation of a user defined function (UDF) to
prescribe a SGS eddy viscosity.
Since Fluent allows to implement extra functions to model the SGS eddy-
viscosity, an attempt has been done to build a classical Smagorinsky model. The
aim of the function was to recalculate the eddy viscosity as wanted and to
substitute the one given by the DES model.
Therefore, the first feature required by the testing of the function was to
understand if the boundary conditions for rough walls were still enabled when a
function is overwriting the calculated eddy-viscosity of the DES model.
The preliminary tests of the new SGS model gave positive results as
concerns the application of the wall boundary conditions. Further tests have
been carried out in order to quantify the sensitivity of the new model to some
features like resolution of the grid both horizontal and vertical, the Smagorinsky
constant C
S
and roughness parameters.
The typical errors of the standard Smagorinsky model for the
reproduction the surface layer of an ABL are well documented, for instance in
Mason and Thompson (1987) [24] and (1992) [25]. The simulations presented
in this section therefore dont pretend to model correctly the ABL but are more
aimed to reproduce the same results obtained by other authors with the
Smagorinsky model in order to understand if the whole numerical procedure is
properly set.


133
The Smagorinsky model has been implemented in the new function in its
standard form, already described in Chapter 2 and the new LES-UDF tested on
a set of structured grids whose main features are listed in Table 5.9; overall the
cases performed with the new function were the four listed in Table 5.10.
In the domain 2 the number of vertical divisions has been increased to 60 to
keep the expansion ratio under 1,1; the height of the first cell of 0,5 m is meant
to be used with a roughness constant RC = E and a roughness height RH = z
0
.
In the following paragraphs some features of the obtained flow fields will be
described, comparison with the baseline model by Sullivan [34] are also done to
assess the correctness of the procedure.
Table 5.9 Grids features of the three simulations compared.
z
0

[m]
Lx, Ly
[m]
Lz
[m]
Nx, Ny Nz N x, y
[m]
h
[m]
Domain 1 0,16 1500 1000 50 50 125 000 30 3,2
Domain 2 0,16 1500 1000 50 60 150 000 30 0,5
Domain 3 0,16 3000 1000 50 50 125 000 60 3,2
Table 5.10 Cases analyzed for the LES UDF.
Domain RH
[m]
RC C
S
Case 1 1 1,17 1 0,1
Case 2 1 1,17 1 0,18
Case 3 3 1,17 1 0,1
Case 4 2 0,16 E 0,18
5.6.2 Wall-shear stress
As for the cases presented in the previous chapter on the DES simulations
traces of elongated structures appear in the contour plots of wall-shear stress on
the ground, see for instance Figure 5.36 for case 1, the same behaviour has been
observed in all the tested cases. Moreover, these traces seem to be infinite
long because of the periodic conditions and the shortness of the domain in the


134
stream-wise direction y, anyway passing from the domain of 1.5 km long to the
one 3 km the traces of the super-streaks seem to cover still the whole domain.
As commented in the previous section 5.5 for DES simulations, the
presence of super-streaks is very likely associated with an underestimation of
wall-shear stress and the presence of an overshoot in the velocity profile,
features that were therefore expected to be observed also in the LES
simulations, as occurred.

Figure 5.36 LES-UDF, Contour plot of Wall-Shear Stress [Pa] (Case 1).
By averaging the wall shear stress on the ground its obtained a value which
is generally smaller than both the wall shear stress obtained with RANS SKE
simulations and the theoretical value computed taking into account a log-law
valid all over the domain. All the computed shear stresses are listed in the table
below. From the case 4 it turns out that with an higher vertical resolution and a
constant RC higher than one the friction velocity is much better predicted with
both the SKE and the LES-UDF method.





135
Table 5.11 Wall-shear stress and friction velocities for the cases analyzed with the
new LES UDF.

Mass-flow
[kg/s]
w LOG LAW

[Pa ]
ref
u

[m/s]
w SKE

ske
u

[m/s]

w
LES

[m/s]

Case 1 2,5 10
+7
0,664 0,736 0,539 0,663 0,446 0,603
Case 2 2,5 10
+7
0,664 0,736 0,539 0,663 0,522 0,653
Case 3 4,883 10
+7
0,633 0,719 0,514 0,648 0,347 0,532
Case 4 2,5 10
+7
0,664 0,736 0,581 0,689 0,539 0,663
5.6.3 Velocity profiles
Vertical velocity profiles are presented in Figure 5.37, Figure 5.38 and
Figure 5.39, normalized forms in Figure 5.40. In Figure 5.37 is shown how the
Smagorinsky constant of 0,18 leads to a better velocity profile compared to the
default value of Fluent 0,1. A C
S
equal to 0,18 is by the way the theoretical
value obtained by Lilly in equilibrium conditions and considering an inertial
range with -5/3 law.
Figure 5.38 shows how the horizontal resolution influences the stream-wise
velocity profile, since the slightly different boundary conditions (assigned mass
flow), comparisons with normalized velocities are therefore more significant.
In Figure 5.39 the effects of a higher vertical resolution, especially close to
the ground is shown, not satisfactory velocities are obtained within the canopy
region, probably a mixing length defined from the wall distance in the cells
close to the wall would help to have a velocity profile closer to the log-law
above the canopy (inside the canopy a log law its not expected).
The normalized wind shear, Figure 5.41, clearly shows an overshoot in all
the cases studied, the smaller value of overshot is obtained for the case 2 with a
maximum around 2,3, only slightly higher than the value obtained with the
baseline model by Sullivan et al. (1994) [34].


136
5.6.4 Eddy viscosity profiles
In the SGS eddy-viscosity profiles, Figure 5.42 and Figure 5.43, are clear
the effects of the Smagorinsky constant C
S
, the eddy viscosity is proportional to
the square of C
S
and to the square of the filter width . Moreover, unphysical
and too high values of eddy-viscosity are predicted close to the wall, therefore
the need of a alternative computation close to the wall which could be, for
instance, the choice of the minimum value between z and C
S
as mixing
length as it's implemented in Fluent in its LES module. It would be interesting
to see the results of the latter approach even if probably also a stochastic forcing
would be necessary to achieve satisfactory results.
The domain vertical refinement doesnt give any improvement as in steady
simulations; but likely, coupled with the corrections explained, stochastic
forcing and a new mixing length close to the wall, it could produce important
improvements.
5.6.5 Velocity variances profiles
The variances of the velocity components, resolved scale, are shown in
Figure 5.44-Figure 5.47, the general behaviour is in the range of a neutral ABL,
as already observed the friction velocity computed seems too low and
consequently amplifying the normalized TKE, values of normalized groups
obtained with the reference friction velocity look more realistic.
5.6.6 Turbulent Kinetic Energy vertical profiles
TKE profiles are presented in Figure 5.48, Figure 5.49 and Figure 5.50.



137


Figure 5.37 vertical profiles of velocity (Smagorinsky constant effect): case 1 (Cs 0,1)
and case 2 (Cs 0,18).


Figure 5.38 vertical profiles of velocity (resolution effect): case 1 (domain 1, x 30 m)
and case 3 (domain 3, x 60 m).


138


Figure 5.39 vertical profiles of velocity (vertical resolution effect): case 2 (domain 1,
h 3,2 m) and case 4 (domain 2, h 0,5 m).


139



Figure 5.40 Normalized profiles and comparison with Sullivans [34] baseline model.


Figure 5.41 - Normalized wind-shear
m


140


Figure 5.42 SGS eddy viscosity profiles.



Figure 5.43 SGS eddy viscosity profiles (lin-log coordinates).


141
d)
0 2 4 6

Figure 5.44 Variances of the velocities normalized with computed friction velocity:
solid lines stream-wise component, dashed lines span-wise component, dotted lines
vertical component; a) case 1, b) case 2, c) Sullivans baseline model, d) equations
(1.6)-(1.8).


142
d)
0 2 4 6

Figure 5.45 Variances of the velocities normalized with computed friction velocity:
solid lines stream-wise component, dashed lines span-wise component, dotted lines
vertical component; a) case 3, b) case 4, c) Sullivans baseline model, d) equations
(1.6)-(1.8).


143
d)
0 2 4 6

Figure 5.46 Variances of the velocities normalized with reference friction velocity:
solid lines stream-wise component, dashed lines span-wise component, dotted lines
vertical component; a) case 1, b) case 2, c) Sullivans baseline model, d) equations
(1.6)-(1.8).


144
d)
0 2 4 6

Figure 5.47 Variances of the velocities normalized with reference friction velocity:
solid lines stream-wise component, dashed lines span-wise component, dotted lines
vertical component; a) case 3, b) case 4, c) Sullivans baseline model, d) equations
(1.6)-(1.8).


145

Figure 5.48 TKE resolved scale vertical profiles.



Figure 5.49 Normalized TKE resolved scale, comparison with baseline model from
[34] and similarity theory from definition of TKE (1.9) and equations (1.6)-(1.8).


146
0
100
200
300
400
500
600
0 2 4 6
/u
ref
2
h
e
i
g
h
t

[
m
]


0 2 4 6
/u
ref
2



Figure 5.50 TKE normalized with reference friction velocity (resolved scale),
comparison with baseline model from [34] and similarity theory from definition of TKE
(1.9) and equations (1.6)-(1.8).
5.7 CONCLUSIONS
The possibility to investigate the ABL flow in neutral stratification with the
DES option of the code Fluent 6.2 has been investigated. A setting of
parameters which allow a rather acceptable ABL reproduction have been found
but further testing is probably required and increasing of the C
DES
constant, use
of discretization schemes different than the first order up-wind for the modified
turbulent viscosity have to be tested. Probably, the SA-based DES approach
could have some limits in the reproduction of the ABL if no stochastic forcing
is implemented. It has to be pointed out that in the newer version of Fluent, the
6.3, two further DES models are available, based on modifications of the k-
and k-, its worthy also to see how they behave in the reproduction of a neutral
ABL.


147
A new UDF has been implemented which allows the use of the
Smagorinsky model coupled with wall-functions, in particular for rough
surfaces. The preliminary 3D tests with the new UDF gave results more
promising than the ones obtained with the DES approach. In general all the
flow-field obtained show an overshoot in the velocity profiles, an
underestimation of the wall-shear stress and the presence of super-streaks. All
these features bring to think at a stochastic forcing as a possible valid treatment
to improve the results of the computations. Moreover, the unnaturally high level
of SGS eddy viscosity close to the wall, when the LES-UDF is applied suggests
to limit the eddy-viscosity close to the wall, perhaps by choosing the mixing
length of the model as the minimum value given by the Smagorinsky model and
a value proportional to the wall-distance, i.e. equation (2.27).
A series of future steps have been individuated for the DES/LES analysis of
the ABL:
Use of zonal embedded unstructured grids
Use of higher order discretization schemes for
sgs
in DES4d
Implement Ls=min(d;Cs) in the LES-UDF function
SGS models more advanced than Smagorinsky (WALE, Wong-Lilly,
dynamic Smagorinsky, etc.)
Bigger domains, maybe higher resolution
Stochastic forcing (synthesized or DNS data), both for DES and WMLES
(Wall Modelled LES)
Simulate the flow (neutral stratification) on a given topography (e.g.
Askervein Hill) with DES and WMLES
Stratification effects, both convective and stratified ABL
Passive scalar transport, punctual source of scalar (local pollutant
source), both for neutral and non-neutral conditions
Simulate the flow around very complex topographies, (e.g. presence of
buildings, steep hills, etc.)
Introduction of Coriolis force (nesting with meso-scale models)


149
Chapter 6 FOREST MODELLING WITH WINDSIM
6.1 INTRODUCTION
The roughness of the terrain has be taken into account in CFD simulations
by the usage of wall functions, as presented up to this point of the thesis. To
apply correctly the wall functions for rough surfaces the centroid of the wall
adjacent cells should reside in the surface layer where the logarithmic law is
valid, as already introduced in the previous chapters. Hence, when modelling
very rough surfaces with wall functions, for instance for values of z
0
higher than
0,1, quite coarse grids have to be drawn. It turns out that for very rough surfaces
the flow inside and close to the canopy layer, see the subdivision of the ABL in
Figure 1.1, has to be described more accurately and the effects of roughness on
the near-wall region cannot only be provided by the boundary conditions. The
velocity profile inside a forest canopy is sketched in Figure 6.1.

Figure 6.1 Flow inside and above a forest canopy, wind profiles in the forest sub-
regions, graph reconstructed from Gardiner B. (2003) [11].

| |
|
\ 0
u z - d
u(z) = ln
z
( ) ( )
| |
|
\
z
-1
h
u z = u h e
| |
|
\
f f
0f
u z - d
u(z) = ln
z


150
While the wind profile in the forest floor can be still described with a
logarithmic law; in the canopy layer, which is a partially occupied by the fluid,
the wind profile diverges from a logarithmic law and re-assumes a logarithmic
behaviour only at 2-3 canopy heights, in the surface layer. Trunks, branches and
foliage establish a complex system of flexible bodies constituting the forest
canopy partially obstructing the flow and extracting momentum from it. Hence
a possible modelling could account for the two features: porosity and extraction
of momentum from the flow by the canopy itself.
The present chapter deals with the implementation of a canopy model in the
CFD code WindSim, a software already introduced in the Chapter 4 which
solves the RANS equations by a finite-volume solver named PHOENICS. The
studies presented have been partially carried out during the internship of the
author at Vector A.S., Tnsberg, Norway, during the period April-June 2005.
The development of the canopy model in collaboration with Vector A.S. has
continued during the course of the PhD. Part of the results on the canopy model
has been already presented in occasion of the WindSim users meeting in June
2005, while further works have been also presented during the WindSim
workshop held in Kassel on the 13
th
February 2006 and in the poster session at
the EWEC 2006 held in Athens, see Gravdahl et al. [15].
The versions of the code used during the internship were the 4.4 and the 4.5
and the solver available was of segregated type. Starting from the version 4.6.1
a coupled solver is also available in WindSim, generally less subject to diverge
than the segregated one and suggested to be employed when the canopy model
is activated. The turbulence is treated with the standard k- model with the
models constants that can assume two different sets of values already reported
in Table 4.5. All the simulations run with the canopy model were performed
with the standard constants of the turbulence model.
The main reason of modelling a forest canopy with numerical tools is
motivated by the need of a more precise description of the flow field around
wind farms located in the vicinity of areas with high vegetation. When flow
simulations When flow simulations over a topography containing woodlands


151
are needed, the usual approach with wall-functions like the log-law (6.1) for
modelling the ABL bears to a poor prediction of wind energy production.
0
u z
u(z) ln
z

| |
=
|
\
(6.1)
Therefore the need to model the ABL inside and in the proximity of forest
in a more detailed manner. The simplest canopy model could come from the
introduction of a displacement length d in the logarithmic law.
0
u z d
u(z) ln
z

| |
=
|
\
(6.2)
The equation (6.2) can be used to describe the ABL over a forested area in
the surface layer, above 2-3 canopy heights of altitude, as also sketched in
Figure 6.1. A canopy model based on the equation (6.2) is considered in
Stangroom (2004) [32] by shifting the wall of a value d and then employing the
(6.1) in a CFD model constructed with the finite-volume RANS solver CFX.
The concept of the displacement length d is used also in the forest model of the
code WAsP and tested for instance by Raftery et al. [27] for three validation
cases and by using five methodologies to set the parameters d and z
0
; their main
conclusions were that usually WAsP still describes poorly the flow field around
heavy forestry and shows a general tendency to under predict canopy effects,
which means an over estimation of the mean terrain velocity.
A more detailed approach in the modelling of a canopy flow can be
provided discretizing the canopy with few layers of cells and use sink terms in
the momentum RANS equation coupled with a volume porosity in the cells
devoted to the canopy flow modelling.
6.2 POROSITY
Two kind of porosities can be set and applied in PHOENICS, and
consequently in WindSim, to each cell of a mesh: one concerning the volume of


152
the cell () and a second one concerning the faces (
S
). Both porosities are non-
dimensional groups since they describe the fluid fraction of the cell volume or
the fluid fraction of the faces bounding the cell.

volume occupied by the fluid/volume of the cell (VPOR)

S
free area/frontal area of the cell (EPOR, NPOR, HPOR)

For each hexahedral cell of a structured grid four porosities variables can be
assigned, one concerning the volume (VPOR) and three concerning areas
(EPOR, NPOR, HPOR) bearing in mind that opposite faces have equal porosity

S
.
6.3 MOMENTUM SINKS
The main effect of the porosity variables on the flow is to speed-up the flow
passing from a fluid cell to a porous one, in order to satisfy the mass
conservation. Therefore, the concept of porosity by itself is not sufficient to
describe completely the effects of forestry on wind; consequently the forces
exerted by the vegetation on the fluid have to be modelled too, for instance by
adding sink terms in the momentum RANS equations (6.3). The sink of
momentum in a forest canopy can be modelled by the sum of two terms, one
proportional to the velocity and a second one proportional to the square of the
velocity as in (6.4). Both equations (6.3) and (6.4) are for unitary porosity.

(
= + +
(


j j
i t j j
i i i j
U U
P
U f S
x x x x
[N/m
3
] (6.3)
j j i i j
S =-C1U -C2 U U U [N/m
3
] (6.4)
In the case of laminar flow, steady state, negligible advection and diffusion,
the source term can be considered only due to molecular viscosity, leading to
the Darcys law which is generally used to model the flow through media with
low porosity as the percolation of a fluid through the terrain (a description of


153
the law is provided in Appendix E). Since the high porosity and the high
Reynolds number in the flow inside a forest canopy, the drag forces should be
more realistically proportional to the square of the velocity.
6.4 A CANOPY MODEL FOR WINDSIM
A canopy model has been therefore introduced in WindSim. The basic idea
of the canopy model is to treat the vegetation (or urban canopy) as a porous
media, while sink terms are added in the momentum RANS equation. The
canopy forest is completely characterised by the following parameters:

a ground cover roughness z
of

a volume porosity
the height of the canopy h
c
( h in Figure 6.1)
two resistive coefficients C1 (1/s) and C2 (m
2
/m
3
)
The wall functions are therefore still used, and a roughness length still
provided, but the z
of
is now used to describe effect of the forest floor. Hence a
logarithmic law as sketched in Figure 6.1 for the forest floor without a
displacement length.
f
0f
u z
u(z) ln
z

| |
=
|
\
(6.5)
Several layers of cells are meant to model the canopy while the rest of the
domain is left to model the flow outside the canopy.
The roughness of the terrain was previously treated in WindSim 4.4 only
with the wall functions based on the logarithmic law (6.1), which remains the
way to model the roughness of the terrain in those regions where the canopy
model is not activated.
The porosity of a forest canopy changes depending on the species of plants,
their shape and scatter on the ground, hence the estimation of the porosity of a
vegetation canopy is quite difficult and furthermore complicated by the fact that


154
it changes during seasons and years. An estimation of the porosity of a
windbreak can be achieved by optical means, taking black-white pictures of the
windbreaks against a bright background; an evaluation of the porosity is
therefore obtained by counting the proportion of the white part in the picture.
A list of optical porosities for different species of plants are provided in the
Table 6.1 below; they have been estimated by Wei Li and collaborators and
published in internet at [41].
In WindSim 4.4 a first approach to solve the flow through a porous media
has been achieved by using the both porosities and
S
and sink terms.
The model is completed with the use of sink terms (drag forces). The total
drag force acting on each tree or element of the vegetation can be thought as the
sum of two terms, one proportional to the velocity (viscous forces) and a second
one proportional to the square of the velocity (pressure forces). As stated in
Kaimal and Finnigan (1994) [17] the Reynolds numbers inside a canopy are
usually enough high to make dominant the pressure forces and, therefore, a drag
force proportional to the square of the velocity can probably describe
completely the sink of momentum.
Table 6.1 Estimations of porosity for several species of trees in winter, from [41]
Specie of tree %
Beech 84
Birch 55
Acacia 40
Mature maple 47
Scots pine 38
Black cherry 77
Lime 54
Ash 48
Spruce 29
Firs 30
Cypresses 15


155
The canopy model has been firstly tested in 3D simulations for step changes
in roughness, from a low roughness area (treated with wall functions) to high
roughness (treated with the canopy model) that will be described in the
following paragraph.
6.5 WINDSIM SIMULATIONS
The main purpose of the simulations carried out with the code WindSim 4.4
and 4.5 was to test the new canopy model in a sudden change in roughness
condition where the inlet zone was set with wall-functions and the rougher with
the canopy model. Different approaches were taken into account to solve the
canopy layer and compare to experimental and numerical data in literature [17].
A mesh was built for different canopy parameters, a top view of discretised
terrain is given in Figure 6.2, the roughness length at the entrance of the domain
is 0,03 m, after 2 km from the inlet the flow reaches a vegetation canopy of
height as specified in Table 6.3. Hence an internal boundary layer develops
starting from the forest edge. A top view of the discretised terrain is shown in
Figure 6.2 while dimensions and characteristics of the mesh employed are
reported in Table 6.2.


Figure 6.2 View of the discretised terrain
x = 0 m
2km
Canopy (Forest):
z
0
as in
z
0
=0,03m
x = 2000 m x = 9000 m


156
Table 6.2 number of cells and dimensions in x (streamwise), y (spanwise) and z
(vertical) direction.
Mesh 1
Nx Ny Nz
117 58 40
Lx Ly Lz
9000 m 2000 m 1000 m

A first series of simulations have been performed with WindSim 4.4. For
this version of the code a canopy model was not already available and the whole
domain has been meshed vertically with a fixed height distribution factor
(HDF), which is the ratio between the height of the wall adjacent cell to the one
of the highest cell. The first ten layers of cells starting from the ground and
from 2 km from the inlet to the outlet have been used to describe the vegetation
canopy.

The simulations run have been named in the following manner:

Canopy 00: z
0
0,03 m in the whole domain
Canopy 01: step change in roughness (z
0
0,03 m to 0,6 m) with the
logarithmic law
Canopy 02: canopy model with only porosities
Canopy 03: canopy model with porosities and drag forces








157
Table 6.3 Characteristics of the grids and of the used canopy models with WindSim
4.4.
Cells in
the
canopy
z
0f
height
canopyh
c

porosities
and
S
DRAG
Canopy 00 - 0,03m - NO NO
Canopy 01 - 0,6m - NO NO
Canopy 02 10 0,6m 61,01m = 0,5

S
= 0,5
NO
Canopy 03b 10 0,6m 61,01m = 0,5

S
= 0,5
C1 =5.965E-5 1/s
Canopy 03c 10 0,6m 61,01m = 0,5

S
= 0,5
C1 = 1.0 1/s
Canopy 03e 10 0,6m 61,01m = 0,5

S
= 0,5
C1 = 1.0E-02 1/s

For the several used canopy models the most important settings are reported
in Table 6.3 and Table 6.4.

Table 6.4 Same grids characteristics as in for two further cases (WindSim 4.4).
Cells in the
canopy
z
0f
h
c
and
S
DRAG
Canopy 03f 10 0,6 m 61,01 m = 0,5

S
= 0,37
C1 = 0,01 1/s
C2 = 0.01 1/m
Canopy 03g 10 0,03 m 61,01 m = 0,5

S
= 1,0
C1 = 0,01 1/s




158
The normalised velocity profile u/u
C
is used to compare the numerical
results to data. The normalised velocity profile for some experiments are shown
in Figure 6.3 together with a normalised momentum flux. The curves are
obtained from Kaimal and Finnigan (1994) [17] and from Garratt (1994) [12],
and are relative to data from both wind tunnel and real vegetation canopies.
Walshe (2004) [40] proposed a canopy model based on resistive forces and the
velocity shear and momentum flux profiles obtained by him have been also
compared to data extracted from Kaimal and Finnigan (1994) [17] providing
good description of the canopy flow, much more precise compared to the pure
roughness length model (log-law).
Vertical profiles of velocity are extracted at the point of coordinates x =
7500 m and y = 1000 m where the flow in the canopy is supposed to be fully
developed. The normalized velocity profiles obtained are shown in Figure 6.4
for some of the cases run. From the set of simulations run the settings of the
Canopy 03f ( = 0,5;
S
= 0,37; C1 = 0,01 s
-1
; C2 = 0,01 m
-1
) seems to fit better
the experimental curves from Bordeaux and Uriarra (pine forest); nevertheless
further investigation was needed and this has been done with the newer version
of the code, the 4.6.1.


159

z
/
h
C

0
0,5
1
1,5
2
2,5
0 0,5 1 1,5
Bordeaux
eucalyptus forest
Uriarra (pine forest)


c
u/u
z
/
h
C

0
0,5
1
1,5
2
0 0,5 1 1,5

2
-u w /u
Figure 6.3 Vertical profiles of normalised wind shear (above) and turbulent
momentum flux (below), within and immediately above the canopy for several
experiments [17] and [12].



160
h
/
h
C

0
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
1,8
2
0 0,5 1 1,5 2
Bordeaux
Uriarra
canopy 01 (log-law)
canopy 02 (log-law + porosities)
canopy 03c
canopy 03e


velocity/uC
h
/
h
C

0
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
1,8
2
0 0,5 1 1,5 2
Bordeaux
Uriarra
Canopy 03e
Canopy 03f
Canopy 03g

velocity/uC
Figure 6.4 normalised velocity profiles obtained with WindSim 4.4.
The simulations performed with the version 4.6.1 were over a 3D domain,
meant to resemble a 2D condition that is not treatable with WindSim. The
dimensions of the whole control volume were in fact 0,1 km x 9 km x 1 km, the
grid constructed was uniform horizontally and a grading has been set vertically


161
as also summarised in Table 6.5. The flow condition was still a step change in
roughness, the roughness before the forest was set 0,03 m while the forest edge
has been placed at 2 km from the inlet.
The settings of the canopy model for some of the simulations performed
with the version 4.6 are reported in Table 6.6, whilst in the new version of the
code the faces porosities
S
are always left to one. With the new canopy model
a layer of porous cells is set at a certain height, for instance 10 m, and the
canopy region is discretised vertically in a uniform way while above the canopy
the vertical discretization is characterised by a given grading.
The vertical profiles of wind obtained with the most significant simulations
of the new series are shown in Figure 6.5 and Figure 6.6; the volume porosity
has been set equal to 0,38 in all cases but one since the experimental results
named Uriarra are referred to a pine forest (with should have an optical porosity
of 38 %). All the reference data have been reported in either Kaimal and
Finnigan (1994) [17] and Garratt (1994) [12].

Table 6.5 number of cells and dimensions in x (streamwise), y (spanwise) and z
(vertical) direction.
Mesh 2
Nx Ny Nz
360 4 35
Lx Ly Lz
9000 m 100 m 1000 m
x y z
25 m 25 m Variable







162
Table 6.6 Grid characteristics of simulations carried out with WindSim 4.6.1.
Cells in the
canopy
z
0f
h
c


DRAG
10 0,05 m 10 m 0,50 C1 0,1 1/s
10 0,05 m 10 m 0,38 C1 1,0 1/s
10 0,05 m 10 m 0,38 C1 0,5 1/s
10 0,05 m 10 m 0,38 C1 0,2 1/s
10 0,05 m 10 m 0,38 C1 0,1 1/s
10 0,05 m 10 m 0,38 C1 0,05 1/s
10 0,05 m 10 m 0,38 C2 0,5 1/m
10 0,05 m 10 m 0,38 C2 0,2 1/m
10 0,05 m 10 m 0,38 C2 0,1 1/m
10 0,05 m 10 m 0,38 C2 0,05 1/m



163
h
/
h
C

0
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
1,8
2
0 0,5 1 1,5 2 2,5 3 3,5 4
Bordeaux
Uriarra
porosity 0,38 C1 0,05
porosity 0,38 C1 0,1
porosity 0,38 C1 0,2

velocity/uC
Figure 6.5 Vertical profiles of wind shear for three cases run with mesh 2 of Table
6.5. (porosity and C1).
h
/
h
C

0
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
1,8
2
0 0,5 1 1,5 2 2,5 3 3,5 4
Bordeaux
Uriarra
porosity 0,38 C2 0,05
porosity 0,38 C2 0,1
porosity 0,38 C2 0,2


velocity/uC
Figure 6.6 Vertical profiles of wind shear for three cases run with mesh 2 of Table
6.5. (porosity and C2).


164
If the simulations with porosity 0,38, C1 0,05 1/s and C2 zero is considered
acceptable for simulating a pine forest then a set of values of C1 and are
calculated (the followed procedure is presented in Appendix E) and proposed in
Table 6.7 for the same species of trees in Table 6.1. The values of C1 and
needs of course to be validated but they provide a good starting point over
which refine the settings of the canopy model.
If the constant C2 is whished to be used instead of the C1 it should be noted
that a constant C2 almost providing the same effect of a C1 is the one which
gives the same drag force; therefore a method to estimate the equivalent C2
constant is to compute the sink term S = C1 U using for U the velocity at h
C
/2
and imposing the equivalence to S = C2 U
2
.

Table 6.7 Drag coefficient C1estimated for the values of porosity given in Table 6.1
starting from the value for porosity 0,38 used in the computations
Species of tree
optical porosity
(%)
C1
(1/s)
Beech 84 0,008
Black cherry 77 0,012
Birch 55 0,028
Lime 54 0,029
50 0,033
Ash 48 0,036
Mature maple 47 0,037
Acacia 40 0,047
Scots pine 38 0,050
Firs 30 0,067
Spruce 29 0,070
Cypresses 15 0,145


165
CONCLUSIONS
In the course of the present Ph.D. an analysis of the Atmospheric Boundary
Layer (ABL) has been carried out with CFD commercial codes. Two different
codes have been employed, i.e. Fluent and WindSim, both based on a finite
volume solver. The ABL has been always considered with negligible thermal
effects, in neutral stability conditions.
The main interest in investigating the ABL with CFD codes resides in the
assessment of wind energy by more accurate numerical tools than commonly
used. It is worthy to note that different applications may benefit from CFD
studies on the ABL, as the assessment of pollutants dispersion, wind loads on
structures and pedestrian comfort, just to mention few of them.
The first part of the work has been devoted to the applications of RANS
approach to the ABL simulations. Firstly 2D simulations over flat terrains and
an isolated 2D bump provided a full set of parameters to be used in the 3D
simulations for every kind of terrain roughness. The correct horizontal and
vertical resolution and the best discretization schemes were chosen. Secondly,
steady RANS 3D simulations over an isolated hill (test case of Askervein) were
carried out, allowing a comparison of the result with experimental data. The
flow around the hill of Askervein is quite challenging since the difficulties in
reproducing the lee side region, in adverse pressure gradient conditions, with
flow separation likely to occur.
The object of this part of the work was mainly to test different turbulence
models and different set of models constants. The results showed that
turbulence models behave differently in reproducing various aspects of the
flow. The k- turbulence model, in its RNG version, provided the more correct
estimation of the velocity profiles compared to standard and realizable version
results. On the contrary, The standard and realizable k- models predict better
the turbulence level, even if the peak of turbulence that is present in the lee side
of the hill is always under predicted.


166
The RSM model gave generally better predictions of the flow field in most
of the simulations, both in terms of velocity and turbulence, with the drawback
of a need for bigger computational resources. The differences with k- models
are enhanced especially in the lee zone of the hill.
A grid refinement study has been conducted showing the strong dependence
of the results on the correct grid spacing both in the vertical direction and in the
flow direction. It has been shown that finer grid determine better results with a
computational effort to be afforded only by parallel processing.
The not very positive results obtained by using the RANS equation are
probably due to the possibility that important unsteady phenomena are present
in the flow over the Askervein hill, like intermittent separation in the lee side.
This could result in a unsteady turbulence that can be only approximately
predicted by steady RANS simulations.
In order to overcome the difficulties found with RANS simulations, more
sophisticated and challenging methodologies of turbulence treatment have been
analysed. The DES and WMLES techniques have been tested on rough flat
terrains with the code Fluent 6.2. In particular, since the WMLES was not
available in the code, a User-Defined-Function (UDF) has been implemented,
where the Smagorinsky-Lilly model is coupled with wall functions for rough
surfaces. This LES-UDF (WMLES) model shows better results then those
obtained with the SA based DES for a neutral ABL over flat terrains. Still an
overshoot in the wind velocity profile is present, which diverges from the
logarithmic law. Some trial for grid sensitivity and results from the literature,
show that this defect can be cured by further refinement of the grid, i.e.
reducing the filter width, and/or with a SGS modelling more advanced than the
standard Smagorinsky.
In the framework of assess the capabilities of CFD codes to evaluate the
wind energy resources on real sites, a new canopy model has been proposed and
tested for the code WindSim. A canopy model was needed to simulate the
effects of forestry on the ABL flow. This new approach is based on the usage of
porous cells and sink terms in the RANS equations and has been tested in a step


167
change in roughness over a flat terrain. Results show a good agreement with
experimental data, allowing to propose a series of setting parameters for
different types of woodlands and providing guidelines for forest modelling.



168
REFERENCES
[1] Anderson J.D., Degrez G., Dick E., Grundmann R. (1995).
Computational Fluid Dynamics, John Wendt F. (Editor). Springer
[2] Bardina, J., Ferziger, J. H. & Reynolds, W. C. (1983). Improved
turbulence models based on large eddy simulation of homogeneous,
incompressible, turbulent flows. Technical Report TF-19. Department
of Mechanical Engineering, Stanford University, Stanford, California
[3] Beljaars A.C.M., Walmsley J.L., Taylor P.A. (1987). A mixed
spectral finite-difference model for neutrally stratified boundary-layer
flow over roughness changes and topography. Boundary Layer
Meteorology, vol. 38, pp. 273-303
[4] BWEA. Workshop on the influence of trees on wind farm energy yield.
Technical report, BWEA-British Wind Energy Association, 2004.
Workshop held at The Arthouse Hotel (Glasgow, Scotland), 17th March
2004. http://www.bwea.com/planning/trees.html
[5] Castro F.A., Palma J.M.L.M., Lopes Silvia A. (2003). Simulation of
the Askervein flow. Part 1: Reynolds averaged Navier-Stokes equations
(k- turbulence model). Boundary-Layer Meteorology, vol 107, pp.
501-530
[6] Cebeci T. and Bradshaw. P. (1977). Momentum Transfer in Boundary
Layers. Hemisphere. Publishing Corporation, New York.
[7] Chow F. K. (2004). Subfilter-scale turbulence modelling for large-eddy
simulation of atmospheric boundary layer over complex terrain. Ph.D.
Thesis, Stanford University.
[8] Crasto G., Cambuli F., Mandas N., Cau G. (2004). Simulazione
numerica degli effetti fluidodinamica di una collina artificiale per la
protezione di un parco carbone. Proceedings of the 59
th
ATI Congress.
(in Italian)
[9] Crasto G. (2006). Investigation of the atmospheric boundary layer
applying large eddy simulation. Diploma Course 2005-06 final
projects thesis, von Karman institute for fluid dynamics, Belgium.


169
[10] Fluent 6.2 Users Guide
[11] Gardiner B. (2003). Airflow Over Forests and Forest Gaps.
Presentation at the workshop BWEA [4]
[12] Garratt J.R. (1994). The atmospheric boundary layer. Cambridge
atmospheric and space science series.
[13] Germano, M., Piomelli, U., Moin, P. & Cabot, W. H. (1991). A
dynamic subgrid-scale eddy viscosity model. Physics of Fluids 3 (7), pp.
1760-1765
[14] Grant, A.L.M.. Observations of boundary layer structure made during
the 1981 KONTUR experiment. Quart. J. Roy. Meteor. Soc., Vol. 112,
pp 825-841.
[15] Gravdahl A.R., Crasto G., Quinn S., Mandas N., Cambuli F. (2006).
Canopy modelling with a CFD code. EWEC, Athens, 2006.
pdf document at:
http://www.windsim.com/documentation/papers_presentations/main.html
[16] Gunot D., Aupoix B. (2003). Etude de l'approche DES (Detached
Eddy Simulation) pour prvoir les coulements instationnaires
grande chelle. 16
eme
Congrs Franais de Mcanique, Nice, 1-5
September 2003.
[17] Kaimal J.C., Finnigan J.J. (1994). Atmospheric boundary layer flows:
Their structure and measurement. Oxford University Press, New York,
NY.
[18] Kim H.G., Lee C.M., Lim H.C., Kyong N.H. (1997). An experimental
and numerical study on the flow over two-dimensional hills. Journal of
Wind Engineering and Industrial Aerodynamics, vol. 66, pp. 17-33
[19] Kim S.E. (2004). Large eddy simulation using unstructured meshes
and dynamic subgridscale turbulence models. Technical Report AIAA-
2004-2548, American Institute of Aeronautics and Astronautics, 34th
Fluid Dynamics Conference and Exhibit.
[20] Kraichnan R. (1970). Diffusion by a Random Velocity Field. Physics
of Fluids, vol. 11, pp. 21-31.


170
[21] Leroy J. (1999). Wind field simulations at Askervein Hill. Technical
Report, Vector CFD Consulting.
[22] Lilly (1992). A proposed modification of the Germano subgrid-scale
closure method. Physics of Fluids 4 (3), pp. 633-635.
[23] Mandas N., Cambuli F., Crasto G., Cau G. (2004). Numerical
simulation of the Atmospheric Boundary Layer (ABL) over complex
terrains. EWEC 2004, London, November 2004.
[24] Mason P.J. and Thompson D.J. (1987). Large-eddy simulations of the
neutral-static-stability planetary boundary layer. Quart. J. Roy. Meteor.
Soc., Vol. 113, pp 413-443.
[25] Mason P.J., Thomson D.J. (1992). Stochastic backscatter in large-
eddy simulations of boundary layers. Journal of Fluid Mechanics, vol.
242, pp 51-78.
[26] Piomelli U., Balaras E., Pasinato H., Squires K.D. and Spalart P.R.
(2003). The inner-outer layer interface in large-eddy simulations with
wall-layer models. Int. J. Heat Fluid Flow, vol. 24, pp. 538-550.
[27] Raftery P., LeBlanc M., Manning J. (2004). WAsP Validation Study
of Forestry Effects. Presented at the Workshop [1].
[28] Shur M., Spalart P. R., Strelets M. and A. Travin (1999). Detached-
Eddy Simulation of an Airfoil at High Angle of Attack. In 4th Int.
Symposium on Eng. Turb. Modeling and Experiments, Corsica, France.
[29] Smagorinsky J. (1963). General circulation experiments with the
primitive equations. Monthly Weather Review, vol. 91, pp. 99-152.
[30] Smirnov R., Shi S., and Celik I. (2001). Random Flow Generation
Technique for Large Eddy Simulations and Particle-Dynamics
Modeling. Journal of Fluids Engineering, vol. 123, pp. 359-371.
[31] Spalart P.R., Jou W-H, Strelets M., Allmaras S.R. (1997).
Comments on the feasibility of LES for wings, and on hybrid RANS/LES
approach. 1
st
AFSOR International Conference on DNS/LES, Aug. 4-8
1997, Ruston.


171
[32] Stangroom P. (2004). CFD Modelling of Wind Flow Over Terrain.
Civil Engineering. Nottingham, University of Nottingham
[33] Stull R.B. (1988). An Introduction to Boundary Layer Meteorology.
Kluwer Accademic Publisher
[34] Sullivan P.P., Mc Williams J.C., Moeng Chin-Hoh (1994). A
subgrid-scale model for Large-Eddy simulation of planetary boundary-
layer flows. Boundary-Layer Meteorology, Vol. 71, 1994, pp 247-276.
[35] Taylor P.A., Teunissen H.W. (1983). ASKERVEIN 82: Report on the
September/October 1982 Experiment to Study Boundary-Layer Flow
Over Askervein, South Uist. Research Report MSRB-83-8, Atmospheric
Environment Service, Downsview, Ontario, Canada
[36] Taylor P.A., Teunissen H.W. (1985). The Askervein Hill Project:
Report on September/October 1983 Main Field Experiment. Internal
Rep. MSRB-84-6, Atmospheric Environment Service, Downsview,
Ontario, Canada
[37] Taylor P.A., Teunissen H.W. (1986). The Askervein Project:
overview and background data. Boundary Layer Meteorology, vol. 39,
pp. 15-39
[38] Tennekes, H. (1982): Similarity relations, scaling laws and spectral
dynamics. Atmospheric Turbulence and Air Pollution Modeling
(F.T.M. Nieuwstadt and H. van Dop, Eds.).Reidel, Hingham, MA, pp.
37-68
[39] Troen I., Lundtang Petersen E. (1989). European Wind Atlas. Ris
National Laboratory, Denmark.
[40] Walshe J. (2004). Effects of Forestry A CFD resistive volume model.
POWERGEN Power Technology Center.
[41] Wei Li, Fan Wang, Simon Bell.
http://www.sbe.hw.ac.uk/research/buildingeng/wind_modelling/gui_wi
ndbk.html


172
[42] Wong V. C., Lilly D. K. (1994). A comparison of two dynamic subgrid
closure methods for turbulent thermal-convection. Physics of Fluids 6
(2), pp. 1016-1023.
[43] Zang, Y., Street, R. L., Koseff, J. R. (1993). A dynamic mixed
subgrid-scale model and its application to turbulent recirculating flows.
Physics of Fluids 5 (12), pp. 3186-3196.


173
APPENDIX A. MODIFICATION OF SKE MODELS CONSTANTS.
The constant C

is computed taking into account measures of turbulent

kinetic energy at a given height, in conditions of neutral stratification and fully
developed boundary layer the vertical profile of TKE can be described by
equation (3.2) considering the height of the ABL equals to the eight of the
domain, for instance 1000 m.

If k
10
is the TKE measured at 10 m height than is C

is given by:
4
4
2
10
10
1
ABL
u m
C
k

( | |
=
( |
\

Given the transport equation for k:
( ) ( ) G
t
i k
i i k i
k
k ku
t x x x

| | | |
+ = + +
| |
|

\ \

For a steady flow:
( ) G
t
i k
i i k i
k
ku
x x x

| | | |
= + +
| |
|

\ \

And fully developed flow (2D):
0 G
t
k
k
k
y y

| | | |
= + +
| |
|

\ \

0 G
t
k
k
k k
y y y y

| | | |
= + +
| |

\ \

2
2
1
0 G
t k
k
k k
y y y

| |
= + +
|

\

t
k

>> implies
1
0 G
t k
k
k
y y

| |
= +
|

\

And since the first term of the RHS of the equation is negligible compared
to both the destruction and production terms it follows G
k
= .



174

Transport equation for :
( ) ( )
2
1 2
G
t
i k
i i i
u C C
t x x x k k

| | | |
+ = + +
| |
|

\ \

For steady flow:
( )
2
1 2
G
t
i k
i i i
u C C
x x x k k

| | | |
= + +
| |
|

\ \

Fully developed boundary layer:
2
1 2
0 G
t
k
C C
y y k k

| | | |
= + +
| |
|

\ \

t

>>
2
1 2
0 G
t
k
C C
y y k k

| |
= +
|

\

The eddy viscosity with the algebraic equation
2
t
k
C

= that can be
substituted in the transport equation for the TDR:
2
2
1 2
0 G
k
C k
C C
y y k k

| |

= +
|
|

\

2
2
1 2
1
0 G
k
C k
C C
y y k k

| |
= +
|

\

And considering the previous equilibrium hypothesis G
k
= :
2
2 2
1 2
1
0
C k
C C
y y k k

| |
= +
|

\

( )
2
2
2 1
1
C k
C C
k y y

| |
=
|

\

( )
2
2
2 1
1
C k
C C
k y y

| |
=
|

\


2
2
2 2
1 1 1
y y y y


| | | |
= +
| |

\ \




175
Considering the TDR inversely proportional to the wall distance,
3
*
( )
u
y
y

=
3
2
( ) * 1 y u
y y y

| |
= =
|

\

2 3
2 3 2
( ) * 2 2 y u
y y y

| |
= =
|

\

hence:
2 2
2
2 2 2 2 2 2 2
1 1 1 1 1 2 1 2 1
y y y y y y y y y


| | | | | |
= + = + = + =
| | |

\ \ \


That can be substituted in the transport equation for TDR,
( )
( )
2
2
2 1
2
3
3
2 1 2
1
* 1
C k
C C
k y y
C k
u
C C
y y

| |
=
|

\

| |
=
|
\

( )
3
6
2 1 2
*
C k
u
C C

=
In proximity of walls the TKE is given by
2
* u
k
C

= which leads to:

( )
6 6
2 1 2 3/ 2
* *
C
u u
C C
C

=
( )
2 1 2
1 1 1
C C
C



=
And finally a relation is found linking the constant Cm the difference C
2

C
1
and the turbulent Prandtl number for TDR.

2
2 1
C C
C

176
APPENDIX B. DERIVATION OF FORMULA (2.33).
Lets consider equations (1.3) and (2.32) reported below,

| |
=
|
\ 0
U(z) 1 z
ln
u z
(1.3)
( )
1
u ln y B B

+ +
= + (2.32)
Now, using the definition of wall coordinates and the relation linking E to B,

u
u
u =
+
;

u y
y =
+
Wall coordinates, ( ) E B ln
1

= the (2.32) becomes:

( ) B E
u y
u
u
+
|
|

\
|
= ln
1
ln
1

And in fully-rough regime (
+
S
K > 90)
( ) ( )
+
+ +
|
|

\
|
=
S S
K C E
u y
u
u
1 ln
1
ln
1
ln
1

Since the product
+
S S
K C is expected to be much greater than 1 it follows:
( ) ( )
+
+
|
|

\
|

S S
K C E
u y
u
u
ln
1
ln
1
ln
1

( )
|
|

\
|
+
|
|

\
|

u K
C E
u y
u
u
S
S
ln
1
ln
1
ln
1

|
|

\
|

S S
K C
y E
u
u
ln
1

Comparing the last expression to the (1.3) and considering that y and z are the
wall normal coordinate in both cases results the equation (2.33).

S S
K C
E
z

0
1
or
0
z E K C
S S



177
APPENDIX C. FURTHER 2D DES CASES ON ROUGH WALLS
Both the sensitivity to the model constant C
DES
and the four settings listed in
Table 5.1 have been tested on rough walls. Velocity and SGS eddy-viscosity
profiles are shown in Figure 6.7 and Figure 6.8. The constant C
DES
influences
clearly the position of the interface URANS/LES and the value of the SGS
eddy-viscosity.
In Figure 6.9 and Figure 6.10 are plotted velocity profiles obtained with the four
settings of Table 5.1.

Figure 6.7 vertical profiles of velocity, sensitivity to the C
DES
constant.


178

Figure 6.8 vertical profiles of SGS eddy viscosity, sensitivity to the C
DES
constant.


Figure 6.9 velocity profiles for DES case 1 and 3.



179

Figure 6.10 velocity profiles for DES case 2 and 4.


180
APPENDIX D. LES-UDF TEXT
/***************************************************************
UDF that specifies a standard Smagorinsky-Lilly SGS model

by Giorgio Crasto (DC 2005-2006) - 27th April 2006
***************************************************************/

#include "udf.h"

DEFINE_TURBULENT_VISCOSITY(user_mu_t,c,t)
{
real S;
real Cs = 0.1; /* Smagorinsky constant */
real mu_t;
real rho;
rho = C_R(c,t);

/* S = sqrt (2 ( Sij * Sij ) ) */

#if RP_3D
S = sqrt((
(C_U_G(c,t)[1]+C_V_G(c,t)[0])*(C_U_G(c,t)[1]+C_V_G(c,t)[0])+
(C_U_G(c,t)[2]+C_W_G(c,t)[0])*(C_U_G(c,t)[2]+C_W_G(c,t)[0])+
(C_V_G(c,t)[2]+C_W_G(c,t)[1])*(C_V_G(c,t)[2]+C_W_G(c,t)[1])
)+
2*(
(C_U_G(c,t)[0])*(C_U_G(c,t)[0])+
(C_V_G(c,t)[1])*(C_V_G(c,t)[1])+
(C_W_G(c,t)[2])*(C_W_G(c,t)[2])
)) ;
#else


181
S = sqrt((
(C_U_G(c,t)[1]+C_V_G(c,t)[0])*(C_U_G(c,t)[1]+C_V_G(c,t)[0])
)+
2*(
(C_U_G(c,t)[0])*(C_U_G(c,t)[0])+
(C_V_G(c,t)[1])*(C_V_G(c,t)[1])
)) ;
#endif

#if RP_3D
mu_t = rho*pow(Cs,2.0)*pow(C_VOLUME(c,t),2.0/3.0)*S;
#else
mu_t = rho*pow(Cs,2.0)*C_VOLUME(c,t)*S;
#endif

/* real L_s = min; mixing length */

return mu_t;
}


182
APPENDIX E. COMPUTATION OF DRAG COEFFCIENTS C1 OF
TABLE 6.7
First of all the optical porosity of the porous medium (Pine forest) has been
considered equal to the volume porosity . The value of C1 0,05 1/s is
considered acceptable for the pine forest.

Now taking into account the definition of permeability by the Darcys law
itself:
q L
A P
=

Where, given a tube of flow of length L and cross sectional area A in a
porous region, as sketched in Figure 11:

[m
2
] is the permeability;
q [m
3
/s] is the flow rate;
[Pa s ] is the dynamic viscosity;
P [Pa] is the pressure drop along the length L

The ratio of the flow rate q to the cross sectional area A yields the Darcys
velocity or superficial velocity, while considering the porosity of the medium
the interstitial or actual velocity, always space averaged, is given by the ratio of
the superficial velocity to the porosity .

Figure 11 sketch of a tube of flow in a porous media for the definition of the Darcys
law.


183
DARCY
q
U
A
= Darcys (or superficial) velocity
DARCY
U q
U
A
= = actual (or interstitial) velocity
A relation is proposed in order to link the permeability to the porosity of
the medium :
2
2

C
1
=

. (AE.1)
Where the following relation defining the drag coefficient C1 holds:

C1

=
Being:

[m
2
] the permeability;
[-] the porosity;
[Pa s] the dynamic viscosity;
[kg/m
3
] the density;

According to the proposed equation when the porosity equals one (fluid
cell) the permeability is infinite and drag coefficient zero, when the porosity is
zero (solid cell) also the permeability equals zero and the drag coefficient is
infinite.

Since for a pine forest ( = 0,38) the constant C1 is found to be around 0,05
1/s and assuming valid the equation (AE.1) the constant C is then calculated
equal to 6,9529E-04 m
2
, hence the values of C1 proposed in Table 6.7.