ARTICLE IN PRESS

Journal of Wind Engineering

and Industrial Aerodynamics 96 (2008) 1451–1471
www.elsevier.com/locate/jweia

Towards practical use of LES in wind engineering

Tetsuro Tamura
Tokyo Institute of Technology, G5-7, 4259 Nagatsuta, Midori-ku, Yokohama, Japan
Available online 9 April 2008

Abstract

This paper reviews large-scale computing results currently obtained by the large eddy simulation (LES)
technique for various wind engineering problems. The LES or direct numerical simulation (DNS)
techniques should be used to numerically simulate unsteady flow phenomena. LES is appropriate for
wind engineering applications because its computational power and memory requirements are
reasonable. The present study discusses the applicability of LES to several issues such as wind-resistant
design, prediction of wind velocity affected by terrain or ground surface conditions, and estimation of
turbulence structures and atmospheric diffusion in urban areas. Thus far, numerical validation has been
performed mainly by comparison with experimental data. However, when considering the complexity of
actual conditions and the natural wind environment, direct comparison with full-scale measurement data
is very important in all cases to verify the effectiveness of LES in wind engineering.
r 2008 Elsevier Ltd. All rights reserved.

Keywords: LES; Inflow turbulence; Full-scale measurement; Bluff body aerodynamics; Wind in cities; Ground
surface condition; Terrain effect; Atmospheric environment

1. Introduction

Large eddy simulation (LES) has become a powerful tool for turbulent flow analysis in
the computational wind engineering field as well as in computational fluid dynamics. This
has been enabled by the establishment of the following numerical techniques.

(1) Generation of inflow turbulence by numerical or statistical methods.

(2) Sophisticated sub-grid scale (SGS) turbulence modeling for unsteady separated flows
with vortex shedding or turbulent boundary layers over complicated undulations with
very rough surfaces.

E-mail address: tamura@depe.titech.ac.jp

0167-6105/$ - see front matter r 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jweia.2008.02.034
 ARTICLE IN PRESS
1452 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

(3) Numerical discretization with conservation of various physical quantities for modeling
complicated geometries.

Development of a numerical modeling has enabled LES to predict wind flows around
buildings and structures under conditions very close to the actual state. Therefore, the
practical use of LES in wind engineering is now possible. Here we discuss the applicability
of LES to wind-resistant design of very large buildings and structures, and determination
of wind velocity affected by hilly undulations, complex terrain or ground surface
roughness; turbulence structures inside or over urban canopies; and atmospheric
dispersion of mass and heat in urban areas. Based on previous examples reviewed in
following sessions, the current state and potential for use of LES in wind engineering
is examined. In particular, in order to independently estimate the accuracy or
reproducibility of numerical results obtained by LES for wind-related natural phenomena,
comparison with full-scale measurement data or characteristics presumed by full-
scale measurement, rather than wind tunnel test data, is tried for many cases and from
as many viewpoints as possible. This paper emphasizes the importance of such a direct
comparison.

2. Wind effects on buildings and structures

2.1. Wind loading estimation—AIJ activities in relation to LES for wind-resistant design

The Architectural Institute of Japan (AIJ) established a working group to investigate the
applicability of the CFD technique to the wind-resistant design of actual buildings and
structures (AIJ, 2005; Tamura et al., 2006a). This investigation has clarified the current
state of predictive accuracy of the numerical model and provided a guide for an
appropriate numerical model and method. This paper describes the activities of AIJ in
estimating the performance and limitations of various types of numerical modeling in
computational wind engineering. It deals with low-rise (1:1:0.5) and high-rise (1:1:4)
buildings. Both RANS simulation and LES were carried out to determine the
aerodynamics of these building models. The computed wind forces and pressures for
these simple configurations have been validated by comparison with experimental data.
AIJ used these computed data to determine wind loadings on these buildings as predicted
by RANS and LES. Fig. 1 displays wind loads on cladding of a high-rise (1:1:4) building
estimated by LES (Nozawa and Tamura, 2003) and experiments (Ohtake, 2002),
accompanied by AIJ recommendations (AIJ, 2004) determined by considering many field
measurement data as well as experimental data. Matching of these results indicates that
LES can accurately predict wind loadings on claddings of high-rise buildings of various
shapes.

2.2. Generation of inflow turbulence for LES

In many cases in which LES is applied to wind engineering problems, generation of

inflow turbulence, which provides time-sequential data with physically corrected spatial
structures, is a very important issue, because wind is essentially a turbulent and unsteady
flow at very high Reynolds numbers. Thus, the oncoming flow in the computational
domain should have these inherent characteristics of natural wind. One method for
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1453

LES (Nozawa,2003)
experiment (Ohtake,2002)
160 160 160
AIJ recommendations (2004)
140 140 140

120 120 120

100 100 100

height (m)

80 80 80

60 60 60

40 40 40

20 20 20

0 0 0
0 2 4 6 -6 -4 -2 0 -6 -4 -2 0 kN/m2
windward wall leeward wall side wall

Fig. 1. LES estimation of wind loads on cladding of a high-rise (1:1:4) building by AIJ.

obtaining this type of flow is to numerically simulate turbulent flows in the auxiliary
computational domain (often called a driver region where a fluid flow is driven). Lund
et al. (1998) proposed the technique of inflow generation, where various development
ratios of boundary-layer parameters such as depth and friction velocity are estimated
using the computed result just obtained sequentially. The velocity profile at the inflow
boundary is reset based on the law of the wall or defect law assuming the spatial
development of a boundary layer on a smooth surface. Kataoka and Mizuno (1999)
proposed a simplified version of Lund’s method, where the development of the boundary-
layer thickness is neglected and the mean velocity profile is fixed at inflow. Nozawa
and Tamura (2002, 2005) extended Lund’s method to a rough-wall boundary-layer flow,
using a roughness block arrangement based on roughness density and the experimental
formula for rough-wall friction to predict the development of a boundary layer over a
rough surface. All the above processes for deriving the inflow conditions allow the
simulation of a spatially developing neutral turbulent boundary layer with a small
computational domain.
Another method is to generate inflow turbulence statistically. This method is mainly
employed for homogeneous turbulence as a two-dimensional problem, such as flow around
a cylinder-type structure (Tamura and Ono, 2003). No driver region is required, so a very
small computational domain is possible. Artificial velocity fluctuations are generated from
the assumed turbulence characteristics. However, statistically generated inflow turbulence
might not satisfy the governing equations for incompressible flows, thus requiring velocity
correction. This usually requires a very large amount of storage for time-sequential data.
Recently, in view of the atmospheric boundary layer over the earth, a thermally stratified
effect can be considered for the above type of boundary-layer simulation (Tamura et al.,
 ARTICLE IN PRESS
1454 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

2003; Tamura and Mori, 2006). For an unstably stratified boundary layer, in the driver
region, velocity fluctuation is generated by using the quasi-periodic boundary condition for
a rough wall, while temperature is treated as a passive scalar. The generated inflow data for
temperature as well as velocity are introduced into the main computational domain, where
the solution of physical quantities takes into account buoyancy effects. For a stable
boundary layer, the generated inflow velocity data at the recycle station are introduced
into the main computational region. However, for the temperature data, the assumed
profile is imposed at the inflow boundary of the main region and the solution of
temperature starts to take into account buoyancy effects. This is because fluctuation of the
passive scalar is too large and not appropriate for the inflow condition of a stable turbulent
boundary layer.

2.3. Turbulence effects on bluff body aerodynamics

Thus far, several researchers have investigated the aeroelastic instability of prismatic
structures on the basis of the numerical results obtained from large eddy simulation for
turbulent flows around square and rectangular prisms. From the wind engineering point of
view, the turbulence effect on unstable oscillations should be clarified. Turbulence of
oncoming flows changes sensitively the behavior of separated shear layers and vortex
formation around a prism, so we need to investigate details of flow structures under
unstable oscillations, as well as the aeroelastic characteristics of a prism. By using inflow
homogeneous turbulence generated by the statistical method, we can carry out LES for the
aerodynamics of prisms in turbulent flows. The turbulence effects on the aerodynamic and
aeroelastic characteristics of a prism have been clarified (Tamura and Ono, 2003). Fig. 2
illustrates LES results of time-averaged streamlines around a square and a rectangular
(D/B ¼ 2.0, B, breadth of a rectangular prism; D, depth) prism in smooth and turbulent
(turbulent intensity: 6% and 13%) flows. For the square prism, the size of the wake
increases as the inflow turbulence increases. The center of the wake vortices is separately
placed behind the back surface, so the drag decreases. However, for a rectangular prism,
the separated shear layer comes close to the leeward corner and reattaches to it with higher

Smooth Smooth

u’/U = 0.06 u’/U = 0.06

u’/U = 0.13
u’/U = 0.13

Fig. 2. Time-averaged streamlines around a square and a rectangular prism in smooth and turbulent flows.
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1455

inflow turbulence. As a result, a weak vortex is formed behind the prism. For both square
and rectangular prisms, turbulence reduces drag on the prism, but the physical
mechanisms are different.
It is well known that a square or a streamwise-elongated rectangular cylinder is subject
to the possibility of the onset of unstable oscillation called galloping at reduced velocity
(Vr ¼ U/Bf0, U, mean velocity; f0, natural frequency) basically higher than the resonant
velocity. Fig. 3 displays the relation of the reduced velocity (Vr) and the response
displacement for aeroelastic oscillations of a square and a rectangular prism. For a
square prism, galloping occurs in both smooth and turbulent inflows. However, for a
rectangular prism, the oscillation caused by a motion-induced vortex is not affected
by inflow turbulence, but galloping disappears for turbulent inflow. Fig. 4 depicts
LES results of phase-averaged vorticity contours around a rectangular prism undergoing
aeroelastic oscillations. For the vortex-induced oscillation at Vr ¼ 5, the motion-induced
vortex formation around the prism is almost the same for both smooth and turbulent
inflow conditions. There are no clear turbulence effects under vortex-induced oscillation.
However, for galloping at Vr ¼ 25, inflow turbulence makes the separated shear
layer ambiguous and does not generate a negative damping force on the rectangular
prism.
For a short rectangular cylinder, a different type of unstable oscillation is observed.
High-speed galloping occurs due to the negative damping force generated by the after body
effect and its physical mechanism has previously been understood. However, for unstable
oscillation of a short rectangular cylinder, which is generally called low-speed galloping,
there remain many unsolved problems. We do not know what this instability stems from,
so its physical mechanism has not been clearly understood. In particular, there has been
little study on turbulent effects on low-speed galloping. Tamura et al. (2005a) describe
aerodynamic and aeroelastic characteristics of a rectangular cylinder with a small side ratio
for smooth and turbulent inflow conditions and examine their physical meanings based on
LES results.

Exp. by Miyazaki (Sc=20,Smooth) Exp. by Laneville (Smooth)

Exp. by Miyazaki (Sc=30,Smooth)
Exp. by Miyazaki (Sc=40,Smooth)
Exp. by Miyazaki (Sc=20,Turb,u'/U=0.11,Lx/B=0.97)
Exp. by Miyazaki (Sc=30,Turb,u'/U=0.11,Lx/B=0.97)
Exp. by Miyazaki (Sc=40,Turb,u'/U=0.11,Lx/B=0.97)
Present comp. (Sc=10,Smooth)
1.0 Present comp. (Sc=20,Smooth)
Present comp. (Sc=40,Smooth)
Present comp. (Sc=10,Turb,u'/U=0.1,Lx/B=1.5)
Present comp. (Sc=20,Turb,u'/U=0.1,Lx/B=1.5)
Present comp. (Sc=40,Turb,u'/U=0.1,Lx/B=1.5)
y/B
y/B
Exp. by Laneville (u'/U=0.067,Lx/B=1.58)
Exp. by Laneville (u'/U=0.09,Lx/B=1.58)
Exp. by Miyazaki (Sc=5,Smooth)
Exp. by Miyazaki (Sc=5,u'/U=0.11,Lx/B=1.94)
Present comp (Sc=5,Smooth)
Present comp (Sc=5,u'/U=0.06,Lx/B=1.5)
Present comp (Sc=5,u'/U=0.13,Lx/B=1.5)
0.5

0.5

0.0 0.0
5 10 15 20 25 30 0 5 10 15 20 25 30
Vr Vr

Fig. 3. Aeroelastic oscillations of a square and a rectangular prism.

 ARTICLE IN PRESS
1456 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

In turbulent flow
In smooth flow
u’/U = 0.13, Lx/B = 1.5

y = 0, dy/dt > 0, Vr = 5

In smooth flow In turbulent flow

u’/U = 0.13, Lx/B = 1.5

y = 0, dy/dt > 0, Vr = 25

Fig. 4. Phase-averaged vorticity contours around a rectangular prism in elastic oscillations (Lx, streamwise
turbulent integral scale). (a) Vortex-induced oscillations; (b) Unstable oscillations.

3. Wind flow velocity affected by terrain

3.1. Wind flow over a simply shaped hill with sinusoidal curve

LES was carried out to investigate turbulent boundary-layer flows over a hill-shaped
model with a steep or a relatively moderate slope at moderately high Reynolds numbers
(Tamura et al., 2005b). This study focused strongly on the hill’s surface condition, such as
roughness effects, as well as curvature effects. A rough surface was simulated by arraying
roughness blocks on the ground. For modeling of LES, both the dynamic Smagorinsky
model (DSM) and the dynamic mixed model (DMM) were used. It is well known that the
behavior of separated shear layers and the vortex motions are sensitive to oncoming
turbulence, such that the separated shear layer comes close to the ground surface, or the
separation region becomes small. Based on unsteady phenomena of wake flows over a
smooth and a rough two-dimensional hill-shaped obstacle, roughness effects on the flow
patterns and the turbulence structures distorted by the hill were clarified. The applicability
of DSM and DMM is also discussed, focusing on the recirculation region behind a steep
hill. The performance of the eddy viscosity and the scale similarity models is clarified from
the basic concept of their modeling process. Fig. 5 shows LES results for the vertical
profiles of mean velocities over a two-dimensional hill with a smooth surface in a turbulent
boundary layer. The mean velocity profiles predicted by DSM and DMM are similar, but
the locations of the reattachment points are clearly different. The reattachment point
behind a steep hill predicted by DMM is in good agreement with the experimental result.
Fig. 6 shows the surface roughness effects on mean flows over a two-dimensional hill
predicted by using DMM for LES. The reattachment points for both smooth and rough
surfaces agree well with experimental data. The smooth surface case shows the
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1457

x/H
-2.5 -1.25 0.0 1.25 2.5 3.75 5.0 6.25
3
LES(DSM)
2.5 LES(DMM)
Line at u=0(DSM)
2
Line at u=0(DMM)
y/H

1.5
1
0.5
0
-2 0 2 4 6
u/U8 5.4
0 1.0 (Exp., Smooth)

Fig. 5. Vertical profiles of mean velocities over a 2D hill with smooth surface in turbulent boundary layer.

x/H
-2.5 -1.25 0.0 1.25 2.5 3.75 5.0 6.25
3
Smooth
2.5 Rough
Line at u=0(Smooth)
2 Line at u=0(Rough)
y/H

1.5
1
0.5
0
6.5
u/U8 5.4 (Exp., Rough)
0 1.0 (Exp., Smooth)

Fig. 6. Roughness effects on mean flows over a 2D hill in turbulent boundary layer.

reattachment point closer to the hill and a smaller recirculation region than the rough
surface case. Totally, DMM can accurately predict the behavior of the separated shear
layer at an impinging region near the reattachment point because it is not limited to eddy
viscosity modeling.

3.2. Ground surface modeling for vegetation

This example of LES focuses on modeling of grasses and trees on the ground surface
based on the actual surface condition of a terrain (Tamura et al., 2007). In order to model
the vegetation effects for LES, we employ the feedback forcing method proposed by
Goldstein et al. (1993). In this model, the equation of motion for trees or grasses are
coupled with the Navier–Stokes equation, so turbulence in the vegetation canopy can be
expressed numerically. Fig. 7 shows the computational model of turbulent flow over a
three-dimensional hill with vegetation. Figs. 8 and 9 display the streamwise mean velocity
and turbulence intensity over a three-dimensional hill. The computed results over the hill
both with and without vegetation are in good agreement with the previous experimental
data (Meng and Hibi, 1998) for a rough and a smooth hill. Fig. 10 shows the streamwise
mean velocity in the inner region of vegetation over a three-dimensional hill. According to
 ARTICLE IN PRESS
1458 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

Vegetation model
Rescale

y
x Roughness
Inflow 3D hill
z blocks
blocks
Driver domain Main domain

Fig. 7. Computational model of turbulent flow over a 3D hill with vegetation.

5H Present comp.(smooth)
Meng's exp.(smooth)
Present comp.(rough)
4H Meng's exp.(rough)

3H

2H

0
-2.5H -1.25H 0 1.25H 2.5H 3.75H 5H 6.25H

u/U
8

0 1.0

Fig. 8. Streamwise mean velocity over a 3D hill in turbulent boundary layer.

the previous field measurement data, reasonable results were obtained, such as connection
to an exponential type of velocity profile. The effects of vegetation on turbulence statistics,
including the coherent structures above the vegetation, are also clarified. In particular, the
turbulence statistics inside the vegetation canopy are shown. According to comparison
with previous field measurement data, almost the same results are obtained qualitatively by
LES. The vertical profiles for each turbulence component show the same tendency. Fig. 11
depicts the instantaneous velocity deviation inside the vegetation canopy over a three-
dimensional hill. Inside the canopy, we find the isotropic turbulence structures caused by
the porous media of the vegetation.

3.3. Actual complicated terrain and comparison with full-scale measurement data

The purpose of this research was to establish flight safety around a heliport to be located
on a cliff top at the northern end of Aogashima Island (Sugio et al., 2005). Because the
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1459

5H Present comp.(smooth)
Meng's exp.(smooth)
Present comp.(rough)
4H Meng's exp.(rough)

3H

2H

0
-2.5H -1.25H 0 1.25H 2.5H 3.75H 5H 6.25H
u/U
8

0 0.2

Fig. 9. Streamwise turbulence intensity over a 3D hill in turbulent boundary layer.

Present comp.(smooth)
Meng's exp.(smooth)
Present comp.(rough)
Meng's exp.(rough)

1.5H

0.5H
h0
0
-2.5H -1.25H 0 1.25H 2.5H
u/U
8

0 1.0

Fig. 10. Streamwise mean velocity at the inner region of vegetation over a 3D hill.

heliport was surrounded by steep geographical features, there was a possibility that strong
wind gusts or large wind fluctuations would have an impact on flight safety. Fig. 12
illustrates the computational model for Aogashima Island, which was selected as an actual-
shaped hill. Fig. 13 depicts the instantaneous vorticity contours around Aogashima Island.
A very large recirculation region was formed inside the crater at the center of the island.
Previously, a flight test was conducted using a research helicopter to obtain observational
data of the local wind field around the heliport. The LES results are compared with the
measurement data obtained by the research helicopter to validate the numerical
simulation. Fig. 14 compares the LES results with actual flight measurement data for
instantaneous wind velocities at a helicopter flying over Aogashima Island. LES employs
two models: model 1 (low resolution) and model 2 (high resolution). A low velocity
location close to the cliff due to flow separation can be recognized in the high resolution
LES result. Thus, LES well reproduced the separation and reattachment of the unsteady
wind field over the complicated terrain of Aogashima Island. It is also shown that LES can
 ARTICLE IN PRESS
1460 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

Fig. 11. Instantaneous velocity deviation inside vegetation canopy over a 3D hill.

Fig. 12. Computational model of an actual-shaped hill—around Aogashima Island.

Fig. 13. Instantaneous vorticity contours around Aogashima Island.

provide an instantaneous flow field and can be used to identify the location and conditions
in which high turbulence is likely to occur. Details of the wind velocity and the turbulence
structures are clarified locally around the heliport as well as in the overall region of the
island wake.
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1461

1.4

1.2
Horizontal velocity ratio

0.8

0.6

0.4 LES by model 1

LES by model 2
Flight measurement
0.2

0
-600 -500 -400 -300 -200 -100 0 100 200
Horizontal location (east to west) [m]

Fig. 14. Comparison of LES and actual flight measurement data for instantaneous wind velocities at a flying
helicopter over Aogashima Island.

Table 1
Computational cases for actual urban areas and their roughness densities

Case l Domain width Site

Case 1 0.415 1000 m  1000 m Kanda

Case 2 0.397 500 m  1000 m Kanda
Case 3 0.298 500 m  1000 m Kanda
Case 4 0.484 500 m  1000 m Kanda
Case 5 0.412 500 m  2000 m Kanda
Case 6 0.511 1000 m  1000 m Shinjyuku
Case 7 0.220 1000 m  1000 m Meguro
Case 8 0.241 500 m  1000 m Meguro
Case 9 0.165 1000 m  1000 m Tsukuba
Case 10 0.125 500 m  1000 m Tsukuba
Case 11 0.190 500 m  1000 m Tsukuba

4. Wind in cities

4.1. Design wind velocity profile over various rough surfaces

This research example uses LES to estimate the vertical profiles of wind velocity in urban
areas (Okuno et al., 2005). The accuracy of GIS (Geographic Information System) data for
actual cities has recently been improved, so it is expected that we can utilize these data to
estimate wind profiles over complicated surface roughnesses in urban areas. Table 1 shows
computational cases for actual urban areas and their roughness densities. Fig. 15 illustrates
examples of geometric surfaces of actual urban areas. Fig. 16 shows LES results for mean
velocity profiles for various urban areas. Fig. 17 shows the relations between power-law
 ARTICLE IN PRESS
1462 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

Fig. 15. Geometric surfaces of actual urban areas. (a) Commercial area in Tokyo (Shinjuku); (b) residential area
in Tokyo (Meguro).

1
Case3
Case5
Case6
Case8
Case9
0.1 Case11
/Z

0.01

0.001
0.0001 0.001 0.01 0.1 1 10
U/U8

Fig. 16. Mean velocity profiles for various urban areas.

indices of mean velocity profiles and roughness densities for various urban areas. Power-
law indices obtained from geometric surface data of actual urban areas (refer to Fig. 18)
are also shown in this figure. Raupach et al. (1991), by using both experimental and field
measurement data, proposed a relation between roughness length normalized by
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1463

1
By Raupach and Counihan
0.8 By LES

0.6


0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6


Fig. 17. Relations between power-law indices of mean velocity profiles and roughness densities of various urban
areas.

Probability density of
Roughness density Building height

Raupach’scurve

Roughness length
Roughness height
Roughness height

Z0

Counihan (1975)’ s equation

α = 0.24 +0.096 (log10 Z0) + 0.016 (log10 Z0)2

Fig. 18. Estimation algorithm for power-law indices of mean velocity profiles by using geometric surface data of
actual urban area (Counihan, 1975).

roughness height and roughness density for a uniformly rough surface where the
boundary-layer thickness is estimated to be sufficiently larger than the roughness height.
However, actual urban areas with very rough surfaces do not have a sufficiently thick
boundary layer compared to building height, because they usually include several tall
buildings. This research aims at finding a universal relation between wind profile
parameters and surface roughness for cities. LES was carried out for wind flow over
models of various actual urban areas and the vertical wind velocity profile was obtained
from results computed by LES. The LES results show consistency with past accumulated
data obtained by experiment and field measurement for small roughness density. However,
for large roughness density, we can recognize a power-law index by LES larger than those
predicted by GIS data. The larger LES value was sometimes observed in field measurement
data of wind profiles in larger cities.
 ARTICLE IN PRESS
1464 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

4.2. LES versus RANS for turbulent flows in urban canopies

For wind in cities, it is still unclear which numerical model, RANS or LES, is the most
appropriate. Fig. 19 depicts the computational domain of LES for densely arrayed tall
buildings in Shinjuku. Fig. 20 shows the velocity profiles at the inflow boundary of the
driver for urban turbulent flows. Fig. 21 illustrates the horizontal distributions of time
averaged velocity contours computed by RANS and LES in the urban canopies (Kataoka
et al., 2006). LES depicts the instantaneous complicated structures in the flow field between
the building blocks completely inside the urban canopy close to the ground. Due to the
isotropic tendency of turbulence structures in the canopy, the time-averaged contours of
the LES results are in reasonably good agreement with the RANS results. However, at a

Fig. 19. Computational domain of LES for densely-arrayed tall buildings at Shinjuku.

Fig. 20. Velocity profiles at inflow boundary of the driver for urban turbulent flows.
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1465

Fig. 21. Horizontal distributions of time averaged velocity for urban canopies. (a) LES; (b) RANS.

Fig. 22. Computed and experimental velocity profiles (changes of profiles along leeward direction over urban
area). (a) LES vs. experiments; (b) RANS vs. experiments.

higher level between tall buildings, the wakes obtained by RANS and LES are not always
consistent. In the present result, the building wake of RANS becomes wider than that of
LES due to larger turbulence viscosity. LES still maintains applicability to wind flow
 ARTICLE IN PRESS
1466 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

among densely arrayed tall buildings where vortex shedding and interaction of separated
shear layers from tall buildings generate strong unsteady flow patterns. Fig. 22 displays
computed and experimental velocity profiles, which mean changes of profiles in the
leeward direction over urban areas. The LES results are in closer agreement with the
experimental data than the RANS results.

5. Atmospheric environments

5.1. Dispersion of hazardous gasses in urban areas

In the next example (Tamura et al., 2006b), LES is applied to the problem of
atmospheric dispersion of hazardous gases inside urban areas with oncoming turbulence,
where the estimation of concentration fluctuation is very important. This numerical model
has already been validated by comparison with previous field experimental results for
plume dispersion through regular arrays of roughness blocks. Here, the characteristics of
flow and plume dispersion are investigated in detail for actual urban areas. Fig. 23
illustrates instantaneous concentration contours of hazardous gases in two kinds of urban
areas. We can find high concentration fields along main streets and stagnation of gasses in
the wakes of buildings. The occurrence of peak-to-r.m.s. concentrations based on various
kinds of surface roughness is also examined for safety analysis. The obtained results show
that the patterns of spatial distribution of mean concentration in actual urban areas are

Fig. 23. Dispersion of hazardous gasses in urban area. (a) Kasumigaseki (government office quarter); (b) Kanda
(commercial area).
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1467

20 20

cU∞/δ2/Q

cU∞/δ2/Q
10 10

0 0
0 10 20 0 10 20
tU∞/δ tU∞/δ
Kasumigaseki In the wake of massive building Edge of concentration field
Smooth and continous dispersion Intermittent high concentration

20 20
cU∞/δ2/Q

cU∞/δ2/Q
10 10

0 0
0 10 20 0 10 20
tU∞/δ tU∞/δ
Street canyon Dense area of low-ride buildings
Kanada Frequency high concentration Smooth and continous dispersion

Fig. 24. Time variation of concentration at various location in urban area.

very different from those in regular arrays of roughness blocks. Fluctuating concentrations
in actual urban areas also have special features (Fig. 24), such as frequent high
concentrations along main streets or smooth and continuous dispersion among densely
spaced buildings, which cannot be explained by the physical mechanism of the dispersion
process obtained from regular arrays of roughness blocks.

5.2. Urban heat island and local-scaled atmospheric circulation

Heat island effects on coastal cities are sometimes expected to be mitigated by sea
breezes, so that cold air mingles with hotter air over and inside urban canopies. Downtown
Tokyo has a coastline along its southeast boundary. However, several very tall buildings
have recently been constructed in the Shiodome area near this coastline, between the center
of Tokyo and Tokyo Bay. We are concerned that these tall buildings will block the passage
of sea breeze into downtown Tokyo. Therefore, LES computation was carried out to
model wind flows over the actual roughened ground surface of Tokyo (Tamura et al.,
2006c; Nagayama et al., 2007). Fig. 25 illustrates the computational model of LES analysis
for urban heat island. For the horizontal inflow boundary condition of computational
domain 3 in the Shiodome area of 1 km  1.75 km, turbulent flow data including wind
velocity and temperature are inputted, taking into consideration the existence of the sea
upstream. The driver region of domain 1 generates a turbulent boundary layer by using the
re-scaling technique and domain 2 thermally stabilizes the boundary layer based on the sea
 ARTICLE IN PRESS
1468 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

Domain2 for stable TBL

(Driver region) Domain3
(Main region)

Domain1 for turbulence generation

800m

y
z x
1000m 1750m
Shiodomearea in Tokyo

Fig. 25. Computational model of LES analysis for urban heat island.

Fig. 26. Instantaneous velocity contours in urban canopy. (a) Instantaneous temperature contours; (b) time-
averaged temperature contours.

breeze characteristics. Radiation temperature measured by aircraft is utilized in setting up

of the boundary conditions at ground surface in cities. According to a previous study, field
measurements around the Shiodome area showed an extreme reduction in wind velocity
behind a group of tall buildings. LES results for turbulent flows in the roughness layer over
a city also show that the flow velocity greatly decreased behind the building blocks
(Fig. 26). This clearly demonstrates the occurrence of environmental degradation of the
thermal conditions due to densely arrayed tall buildings. Furthermore, in the urban
canopy, a very special aspect can be found: steady patterns of a path with strong wind in
the wake of a group of tall buildings. From temperature analysis of heat sources on the
surfaces of the building blocks, it can be seen that the heat has been convected and greatly
diffused far away at a higher location. However, near the bottoms of the buildings, there is
much less heat transported due to the effect of blocking of the sea breeze. Field
measurement data show non-uniformity in the horizontal profile of temperature due to
blocking of the sea breeze. Fig. 27 also shows that heat has stagnated in the building group
because a cavity region is formed among them.
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1469

Fig. 27. Temperature field in urban canopy.

Meso-scale
Meso-scale meteorological
meteorological model
model
Input: data Input: GPV data
MM5

Compute the wind pro file on the specified date

Method1 Method 2

Specific wind velocity and direction Vertical profiles of wind

to the whole region for LES velocity and direction

Set up of the reference wind data Generation of inflow wind data by

targeting the wind profiles

large-eddy simulation Driver computational region

Output:Wind data over and with in urban canopy

Main computational region for urban area

Fig. 28. Computational model for hybrid method of LES and meso-scale meteorological model.

6. Combined method with meso-scale meteorological model

The final example (Takemi et al., 2006) investigates the microscale structure and
variability of severe winds in an urban canopy by conducting coupled simulation of LES
and a meso-scale meteorological model (MM5). Fig. 28 illustrates a schematic of the
computational model for this hybrid method. Here, the analysis is focused on an actual
urban canopy in downtown Tokyo during a period of explosive cyclogenesis. Wind fields
of urban scale are simulated by MM5, while turbulent flow fields inside the urban canopy
are computed by LES that explicitly incorporates the effects of actual shapes of buildings
and structures. Generally, we cannot obtain the absolute values for wind flow by LES, but
only values relative to the reference wind can be estimated. Thus, combining the results by
MM5 (the two methods shown in the figure), we have estimated the wind velocity and
 ARTICLE IN PRESS
1470 T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471

compared it with observational data. The LES/MM5 combined analysis shows results
consistent with observational data, and also exhibits details of local gust wind and sudden
changes of canopy winds. This study suggests that the combined method is a powerful tool
for predicting a severe wind disaster in an urban area.

7. Concluding remarks

Many examples of LES in wind engineering are shown in order to understand the
current state and the future possibility of the practical use of the LES model. In the wind
engineering field, there remain many problems where peak-type values need to be
predicted, such as estimation of wind loadings on buildings and structures for wind-
resistant design, gust evaluation around tall buildings, inside urban canopies and over a
complicated terrain, and prediction of concentration fluctuation in the dispersion of
hazardous gases. LES is definitely capable of providing accurate predictive values that are
comparable to wind tunnel experimental data. The next step is to accumulate reliable
numerical data directly and compare them with data obtained by full-scale measurement in
order to establish a numerical model that can provide useful data for wind engineering
problems, as an independent and powerful tool. We would expect LES to be free from
wind tunnel experimental technique and to introduce a new concept of wind-related
structural and environmental design by CFD. Computation cost is not a serious problem.
Reduction in CPU price and innovative development of a parallel algorithm and its user-
friendly usage are expected to solve this problem. It is much more important to get a
consensus and to develop incentives for using a numerical model such as LES in the wind
engineering field.

References

Architectural Institute of Japan, 2004. Recommendations for loads on buildings.

Architectural Institute of Japan, 2005. Guide for numerical prediction of wind loads on buildings.
Counihan, J., 1975. Adiabatic atmospheric boundary layers. Atmos. Environ. 9, 871–905.
Goldstein, D., Handler, R., Sirovich, L., 1993. Modeling a no-slip flow boundary with an external force field. J.
Comput. Phys. 105, 354–356.
Kataoka, H., Mizuno, M., 1999. Numerical flow computation around 3D square cylinder using inflow turbulence.
J. Archit. Plann. Environ. Eng. Trans. AIJ 523, 71–77.
Kataoka, H., Tamura, T., Okuda, Y., Ohashi, M., 2006. Numerical evaluation of the wake field behind high-rise
buildings by RANS and LES—cross comparison among computed and wind—tunnel experimental results. In:
Proceedings of the 19th National Symposium on Wing Engineering, pp. 73–78.
Lund, T.S., Wu, X., Squires, K.D., 1998. Generation of turbulent inflow data for spatially-developing boundary
simulations. J. Comput. Phys. 140, 233–258.
Meng, Y., Hibi, K., 1998. An experimental study of turbulent boundary layer over steep hills. In: Proceedings of
the 15th National Symposium on Wing Engineering, pp. 61–66.
Nagayama, J., Tamura, T., Takemi, T., Okuda, Y., Mikami, T., Tanaka, H., Sato, M., 2007. LES analysis of
thermally stratified turbulent flow field over the actual urban area. Wind Eng. JAWE 32 (2), 133–134.
Nozawa, K., Tamura, T., 2002. Large eddy simulation of the flow around a low-rise building immersed in a
rough-wall turbulent boundary layer. J. Wind Eng. Ind. Aerodyn. 90, 1151–1162.
Nozawa, K., Tamura, T., 2003. Feasibility study of LES on predicting wind loads on a high-rise building. In:
Proceedings of 11th International Conference on Wind Engineering, pp. 2519–2526.
Nozawa, K., Tamura, T., 2005. Large eddy simulation of wind flows over large roughness elements. In:
Proceedings of EACWE4, pp. 1–6.
Ohtake, K., 2002. Takenaka time-sequential database for wind loading on buildings, Private communication.
 ARTICLE IN PRESS
T. Tamura / J. Wind Eng. Ind. Aerodyn. 96 (2008) 1451–1471 1471

Okuno, A., Tamura, T., Okuda, Y., Kikitsu, H., 2005. LES estimation of wind velocity profiles over various
roughened ground surfaces in cities. In: Proceedings of APCWE6, pp. 437–445.
Raupach, M.R., Antonia, R.A., Rajagopalan, S., 1991. Rough-wall turbulent boundary layers. Appl. Mech. Rev.
44, 1–25.
Sugio, Y., Matayoshi, N., Okuno, A., Tamura, T., 2005. Large eddy simulation of turbulent flow around
Aogashima Island—comparison with field measurement data obtained by the research helicopter. In:
Proceedings of APCWE6, pp. 1820–1829.
Takemi, T., Tamura, T., Takei, Y., Okuda, Y., 2006. Microscale analysis of severe winds within the urban canopy
during a period of explosive cyclogenesis by coupling large-eddy simulation and mesoscale meteorological
models. In: Proceedings of CWE2006, pp. 165–168.
Tamura T., Mori, K., 2006. LES of spatially-developing stably stratified turbulent boundary layers. Direct and
Large-Eddy Simulation, vol. VI, pp. 583–590.
Tamura, T., Ono, Y., 2003. LES analysis on aeroelastic instability of prisms in turbulent flow. J. Wind Eng. Ind.
Aerodyn. 91, 1827–1846.
Tamura, T., Tsubokura, M., Cao, S., Furusawa, T., 2003. LES of spatially-developing stable/unstable stratified
turbulent boundary layers. In: DLES5, pp. 65–66.
Tamura, T., Dias, P.P.N.L., Ono, Y., 2005a. Turbulence effects on aerodynamic unstable oscillations of a short
rectangular cylinder. In: 4th Conference on Bluff Body Wakes and Vortex-Induced Vibrations, pp. 99–102.
Tamura, T., Cao, S., Okuno, A., 2005b. LES study of turbulent boundary layer over a smooth and a rough 2D hill
model. In: ETMM6, pp. 257–266.
Tamura, T., Nozawa, K., Kondo, K., 2006a. AIJ guide for numerical prediction of wind loads on buildings. In:
Proceedings of the CWE2006, pp. 161–164.
Tamura, T., Nakayama, H., Okuda, Y., 2006b. LES analysis on atmospheric dispersion in spatially-developing
turbulent boundary layer in actual urban area. J. Environ. Eng. Trans. AIJ 604, 31–38.
Tamura, T., Nagayama, J., Ohta, K., Takemi, T., Okuda, Y., 2006c. LES estimation of environmental
degradation at the urban heat island due to densely-arrayed tall buildings. In: 17th Symposium on Boundary
Layers and Turbulence, AMS, pp. 22–25.
Tamura, T., Okuno, A., Sugio, Y., 2007. LES analysis of turbulent boundary layer over 3D steep hill covered with
vegetation. J. Wind Eng. Ind Aerodyn. 95, 1463–1475.