Flow, Turbulence and Combustion 72: 463–492, 2004.

C 2004 Kluwer Academic Publishers. Printed in the Netherlands. 463

Effect of Roughness on Wall-Bounded Turbulence

KIRAN BHAGANAGAR1 , JOHN KIM1 and GARY COLEMAN2

1
Department of Mechanical and Aerospace Engineering, University of California, Los Angeles, Los
Angeles, CA, U.S.A., E-mail: [email protected]
2
School of Engineering Sciences, University of Southampton, Southampton, U.K.

Received 3 October 2003; accepted in revised form 11 January 2004

Abstract. Direct numerical simulation of turbulent incompressible plane-channel flow between a
smooth wall and one covered with regular three-dimensional roughness elements is performed. While
the impact of roughness on the mean-velocity profile of turbulent wall layers is well understood, at
least qualitatively, the manner in which other features are affected, especially in the outer layer, has
been more controversial. We compare results from the smooth- and rough-wall sides of the channel
for three different roughness heights of h + = 5.4, 10.8, and 21.6 for Reτ of 400, to isolate the effects
of the roughness on turbulent statistics and the instantaneous turbulence structure at large and small
scales. We focus on the interaction between the near-wall and outer-layer regions, in particular the
extent to which the near-wall behavior influences the flow further away from the surface. Roughness
tends to increase the intensity of the velocity and vorticity fluctuations in the inner layer. In the outer
layer, although the roughness alters the velocity fluctuations, the vorticity fluctuations are relatively
unaffected. The higher-order moments and the energy budgets demonstrate significant differences
between the smooth-wall and rough-wall sides in the processes associated with the wall-normal
fluxes of the Reynolds shear stresses and turbulence kinetic energy. The length scales and flow
dynamics in the roughness sublayer, the spatially inhomogeneous layer within which the flow is
directly influenced by the individual roughness elements, are also examined. Alternative mechanisms
involved in producing and maintaining near-wall turbulence in rough-wall boundary layers are also
considered. We find that the strength of the inner/outer-layer interactions are greatly affected by the
size of the roughness elements.

1. Introduction
Surface roughness is a defining feature of many of the high Reynolds-numbers
flows found in engineering. In fact, the higher the Reynolds number, the more likely
the effects of roughness are significant, since the size of the roughness elements
becomes increasingly large compared to the near-surface viscous length appropriate
for smooth-wall flows. As a result, turbulent boundary layers over the hulls of ships
and submarines, within turbo-machinery, and over the surface of the earth are
all cases to which the smooth-wall idealization rarely applies. Unfortunately, the
impact of surface roughness is not entirely understood, and a number of important
fundamental questions have not yet received a satisfactory answer.
The turbulent boundary layer over a rough surface contains a roughness sublayer,
within which the flow is directly influenced by the individual roughness elements
464 K. BHAGANAGAR ET AL.

and is therefore not spatially homogeneous (i.e. time-averaged statistics are not in-
dependent of location, at the same mean wall-normal distance). The height of this
sublayer presumably depends upon the height of the roughness elements, as well as
their shape and density distribution. Plane averaging over a repeating unit of a uni-
form array will lead to a representative profile within the roughness sublayer, which
could then be regarded as spatially homogeneous on scales larger than the unit, if it is
much larger than the lateral size of individual elements. This type of unit averaging
has been done by Raupach et al. [1], Wood and Mason [2], and Cheng and Castro [3].
The questions that arise, which we attempt to answer below for one type of represen-
tative roughness, are (1) What is the height of the roughness sublayer and how does
it compare to the other length scales of the roughness? (2) What is the significance
of the roughness sublayer with respect to the dynamics of turbulence in this region?
(3) Are the turbulent statistics in the logarithmic region independent of the flow in
the roughness sublayer? and if not, (4) How do the turbulence structures generated
within the sublayer interact with and determine the eddy structure in the log region?
Perry et al. [4] performed experiments using both three-dimensional (3D)
diamond-shaped mesh roughness with height 29 mm (0.3δ, where δ is the boundary
layer thickness for the smooth-wall turbulent boundary layer) and streamwise and
spanwise mesh dimensions respectively of 10.5 mm (0.1δ) and 1.5 mm (0.01δ).
They also considered a two-dimensional (2D) wavy surface with peak-to-valley
normalized height of 17 mm (0.1δ) and streamwise wavelength of 76 mm (0.8δ). In
both cases they observed that smooth- and rough-wall boundary layers have quite
different structures and are controlled by different length scales in the inner layer,
but that low-order statistics are similar well away from the surface, implying that
the outer layer is unaffected by the details of the surface, be it smooth or rough.
Raupach et al. [1] present other data, laboratory and atmospheric, that reinforce the
outer-layer similarity hypothesis. However, there is other evidence, as described
below, that in some cases the turbulence over rough surfaces can be very different
from that over smooth surfaces throughout the boundary layer.
The experiments of Krogstad et al. [5] used a rough surface consisting of a square
mesh of wire diameter 0.69 mm (0.01δ), wire centerline spacing 3.18 mm (0.04δ)
and screen thickness of 1.55 mm (0.02δ). Antonia and Krogstad [6] used 2D rods
in spanwise direction with diameter 1.6 mm (0.02δ) and height 1.6 mm (0.02δ).
Tachie et al. [7] used three different surface roughness: the first one a perforated
plate with holes of diameter 2.2 mm (0.04δ) and thickness of 1.4 mm (0.03δ); the
next one sand grain roughness of 1.2 mm (0.02δ); and the third a square mesh
of wire diameter 0.6 mm (0.01δ) with wire centerline spacing of 7 mm (0.15δ).
Keirsbulk et al. [8] used roughness elements consisting of 2D square bars with
dimension of 3 mm (0.05δ) with the axes of the bars placed along the flow direction.
George and Simpson [9] performed experiments for 2D turbulent boundary layers
with sparsely and uniformly distributed 3D roughness elements; the roughness
elements were circular cylinder roughness elements with diameter 0.6 mm (0.01δ),
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 465

spacing in streamwise- and spanwise- directions 1.8 mm (0.03δ), and for three
different roughness heights 0.3 mm (0.0075δ), 0.6 mm (0.01δ) and 1.2 mm (0.02δ).
More recently, 2D roughness has been investigated by Ashrfian and Andersson [10]
using direct numerical simulations (DNS) at Reynolds number based on the mean
pressure-gradient of 400. Two-dimensional square rods, periodically arranged in
the streamwise direction, with a roughness height of 0.034δ, where δ is the half-
channel width, were introduced on both walls of the channel. They observed the
values of all components of the Reynolds stress tensor were modified in the inner-
layer as well as outer-layers due to the presence of roughness. Leonardi et al. [11]
investigated the effect of square bars on the bottom wall in a turbulent channel flow
using DNS for four different values of longitudinal separation (w) to the height (k)
ratios of (1, 3, 7, 19). They observed increased coherence in the spanwise direction
and decreased coherence in the streamwise direction with increasing w/k ratio.
All these experiments demonstrate significant differences in the shear stress,
uv, and root-mean-square (rms) of fluctuating velocity components throughout the
inner- and outer-layer. Shafi and Antonia [12] measured the rms vorticity fluctu-
ations normalized by the friction velocity u τ and boundary-layer thickness δ, and
found a moderate increase of the wall-normal and spanwise components in the
outer layer. They concluded that the effect of roughness on the vorticity is less pro-
nounced than on the Reynolds stresses. These results conflict with the traditional
picture of wall similarity, which assumes that outside the roughness sublayer the
turbulent motions are independent of the details of the (smooth or rough) surface
at sufficiently large Reynolds numbers.
To get a clearer picture of the impact of roughness in turbulent wall layers,
we investigate the effects of 3D roughness arranged in an “egg-carton” shape (see
Section 2) on turbulent boundary layers, and differences between the rough- and
smooth-wall flows. It is important to identify the physical mechanisms responsible
for the production and maintenance of turbulence adjacent to a rough wall, and
to ascertain if they are similar to those in the smooth-wall boundary layer, or if
instead alternative mechanisms are involved (this has profound implications, for
example, for control strategies applied to rough-wall flows). We investigate the
physical mechanisms represented by the various terms in the transport equations
for the Reynolds stresses, and consider whether alternatives (such as wakes induced
by the roughness elements [13]) are involved in the turbulence production process
in the inner region.
In light of the full range of experimental results, it now appears that rough-wall
boundary layers can be categorized according to whether or not the surface rough-
ness affects the outer layer (although to some extent this classification will depend
on which statistic is examined). Some of the above researchers have hypothesized
that the exact nature of the roughness surface is the differentiating parameter—
i.e. that some surfaces “communicate” with the outer layer while others do not,
despite the fact that they may produce similar first-order statistics (and thereby
466 K. BHAGANAGAR ET AL.

equivalent sand-grain roughness) or even share some geometric features (such as

mean height or spacing). Hence, there is a need to fully understand the influence of
roughness geometry. To do this, it is necessary to identify the relevant parameters
and study their influence on the turbulence dynamics. Possible parameters include
the size and shape of the roughness elements (and whether or not there is a het-
erogeneous range of each, or simply a repeating pattern of a single “unit cell” ),
and their streamwise and spanwise distribution (spacing, alignment and density).
A number of studies of the effect of roughness height of various element shapes
have already been done. One of the aims of this paper is to categorize the influence
of roughness for a single generic 3D element shape of fixed height, in terms of the
other geometric parameters (streamwise and spanwise spacing), and to quantify
how these parameters do or do not affect the outer layer. So far, little has been
attempted in this direction. The results are expected to provide a useful guide as to
when outer-layer similarity holds, and for what statistics, for rough-wall boundary
layers.
The paper is organized as follows. Section 2 deals with the numerical approach.
In Section 3 we discuss the effect of 3D roughness on the large-scale and small-
scale features of the roughness-sublayer (RSL), the inner (containing the RSL)
and outer layers, for one class of turbulent rough-wall boundary layer. Next we
discuss the physical mechanisms involved in the production and maintenance of
turbulence adjacent to and well away from the rough-wall side of the channel.
(It should be noted that for small roughness heights, as in the present case, the
roughness sublayer is part of the inner layer. In cases such as meteorological flows
with large roughness heights (e.g. buildings), the roughness sublayer can extend
into the outer layer, and the analysis will be more complicated than that presented
here.) In Section 4 we show the results for various numerical experiments conducted
with different roughness surfaces. This is followed by a summary in Section 5.
In this paper, u, v, w denote the velocity fluctuations in the streamwise (x), wall-
normal (y) and spanwise (z) directions, respectively. The superscript + denotes a
quantity normalized by the wall-shear velocity, u τ , and the kinematic viscosity, ν.

2. Numerical Approach
We have developed a numerical tool to simulate turbulent flow over a complex
boundary while retaining the simplicity and efficiency of computation in a Cartesian
system. This was done using an immersed boundary method (IBM) [14]. The
concept and details of this approach can be found in [15–17]. The equations of
fluid motion are calculated on the regular geometry of a periodic channel. The
virtual roughness boundary σ is prescribed within the channel as a function of the
streamwise (x) and spanwise (z) variables, such that the active flow domain D is
given by D = (x, y, z)|y = σ (x, z). To enforce the no-slip condition at this virtual
boundary, a linear profile is assumed for the streamwise and spanwise components
of velocity between zero at y = σ (x, z) and the velocity at a grid point above the
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 467

virtual boundary. The wall-normal component of velocity is prescribed from u and

w, mass conservation and the no-slip condition. The immersed no-slip boundary
is prescribed via a body force term. For purposes of defining the body force, we
employ a first-order temporal discretization of the Navier-Stokes equations:
un+1 − un
+ un · ∇un = −∇ p n + ν∇ 2 un + f n , (1)
t
where f n = ( f x , f y , f z )n is the body force, u = (u, v, w) the velocity vector, p the
pressure, ν the kinematic viscosity, t the time-step increment, and the superscripts
n and n + 1 respectively indicate the current and next time level. On the immersed
boundary σ (x, z), the velocity is zero, such that
un+1 = (0, 0, 0) (2)
and we approximate the body force as:
V − un
fn = + un · ∇un + ∇ p n − ν∇ 2 un , (3)
t
where V = (0, 0, 0). The time-dependent body force is applied at a set of two
points, the one just below the immersed boundary and the one just above. (When
the boundary coincides with the grid, the body force is applied at the boundary and
at a point below.) This method gives flexibility in choosing the immersed boundary
not found in some other methods, since there is no need to line up the boundary with
a grid. We use this method here to simulate turbulent channel flow between a smooth
wall and one covered with regular 3D roughness elements. Apart from the additional
forcing term, and the wall-normal discretization, the numerical method is similar
to the one presented by Kim et al. [18]. We use Fourier series in the streamwise and
spanwise directions. In the wall-normal direction, however, we employ a fourth-
order compact finite-difference scheme instead of the Chebyshev polynomials used
by Kim et al. [18]. The code has been parallelized using message-passing-interface
(MPI). The parallelization strategy is based on the domain decomposition technique,
since the IBM approach requires communication between the processors, which
contain the interpolating velocity grid points, and those with the grid points where
the body force needs to be applied. This required development of an efficient
bookkeeping strategy and MPI broadcasting technique.
DNS have been performed for Reτ = 400 (Reynolds number based on wall-
shear velocity u τ and channel half-height δ) in a periodic channel of streamwise
and spanwise size L x /δ = 2π and L z /δ = π , where 2δ is the distance between the
plane walls (one of which now lies outside the active flow domain, below the virtual
rough surface). The spatial discretization used 256 streamwise Fourier modes, 257
wall-normal compact finite-difference grid points of fourth-order accuracy and 256
spanwise Fourier modes. In the wall-normal direction, nonuniform mesh was used.
The grid spacing varied from 0.94 wall units (based on u τ at the rough wall) adjacent
to the virtual no-slip surface to 6.5 at the centerline. In the horizontal directions
x + was approximately 15 and z + was approximately 8.
468 K. BHAGANAGAR ET AL.

Figure 1. The roughness surface used in the simulations for h + = 21.6. The channel is shown
from the lower-wall (y = −1) to the center of the channel (y = 0). The virtual no-slip roughness
surface is at σ = −0.96.

The virtual no-slip surface consists of smooth 3D “egg carton”-shaped surface

σ (x, z) such that

  
h 2π x 2π z
σ (x, z) = σ0 + −1 + 1 + sin +
4 lx lz
  
2π x 2π z
× 1 + sin − , (4)
lx lz
where σ (x, z) is measured with respect to the channel coordinates, in units of
δ, h is the (peak to valley) roughness height, σ0 defines the mean offset of the
immersed boundary, and l x and l z are the streamwise and spanwise wavelengths
(peak-to-peak distance) of the roughness elements. We chose σ = −0.96 for all
simulations presented here, which prescribes the virtual no-slip roughness surface
at the bottom of the DNS domain, just above the lower wall. For this surface, the
roughness “bumps” extend 3h/4 above σ0 , while the valleys lie h/4 below it. The
roughness surface corresponding to h + = 21.6 is shown in Figure 1.

3. Turbulent Flow over a Rough-Wall with 3D Roughness Elements

3.1. TURBULENCE STATISTICS

We analyze a turbulent flow over a rough-wall with 3D roughness elements using

DNS. The shape of the roughness elements is expressed by Equation (4). We spec-
ify roughness elements with peak-to-peak spacing in the streamwise and spanwise
directions of 100 wall units in terms of smooth-wall u τ and three different heights
of approximately h + = 5.4, 10.8 and 21.6. Unless otherwise stated, results shown
here are for the case of h + = 21.6, with the other two cases included where com-
parison is appropriate. Once the velocity field reached a statistically steady state,
the computations were continued in time for about 10 nondimensionalized units (in
terms of the smooth-wall u τ and δ) to obtain mean statistics, which were gathered
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 469

by averaging over x and z directions as well as time. Convergence was verified by

examining the Reynolds shear stress, which must converge to a straight line.
The classical framework established by Nikuradse [19] predicts that the effect
of roughness on the mean-velocity distribution is confined to a thin wall layer. In
the log-region, assuming a logarithmic velocity distribution for flow over a smooth
wall given by
 
U 1 yw u τ
= ln + C0 , (5)
uτ κ ν
the rough-wall modification is
 
U 1 yw u τ U
= ln + C0 − , (6)
uτ κ ν uτ
where U/u τ = f (ks+ ) is the roughness function, ks+ the equivalent sand grain
roughness, and we assume κ = 0.41; C0 = 5.5 is the additive constant for both
the rough- and smooth- wall, u τ is the local wall-shear velocity (u τ of each wall
is referred to as local u τ , hereinafter). For the smooth-wall case u τ is obtained
using the shear at the upper smooth wall, and for the rough wall u τ is obtained
from the mean momentum balance. Comparing the rough- and the smooth-wall
distributions, the roughness results in a downward shift of the logarithmic profile
with no discernible change in slope.
To represent the mean velocity profile over a rough surface, the first task is to
determine the virtual offset α for the rough-wall side. This is done by expressing
the mean velocity in the log-region in the following form
U 1 (yw − α)
= ln (7)
uτ κ y0
Here, α is the virtual offset, y0 is the roughness length, and y0 and α are determined
by fitting the mean velocity profile in the inertial sublayer to the above equation.
All the results shown consider the virtual offset at the rough-wall side. Figure 2
shows the mean velocity profile normalized by the local wall-shear velocity, plotted
in wall units for the smooth-wall side and rough-wall side, for roughness heights of
h + = 5.4, 10.8 and 21.6. In this figure, y + represents the distance from the wall in
wall units. All results shown below compute y + taking the virtual offset into account
and using the local u τ unless stated otherwise. The roughness produces the expected
downward shift in U + , which increases with increasing h + . This results also serves
as a validation for our numerics. Raupach et al. [1] present a relationship between
U + and h + for wire screen roughness, bar roughness, sand-grain roughness and
natural vegetation. For our h + of 21.6, the corresponding U + is around 6.0, which
is close to the value of 6.4 observed here. Ashrafian and Andersson [10] obtained
U + of 7.0 for a 2D spanwise roughness of height 0.034δ, which is higher than
that given by our simulations, although not appreciably. The differences could be
attributed to the difference in streamwise spacing of the roughness elements.
470 K. BHAGANAGAR ET AL.

Figure 2. Mean velocity normalized by local u τ shown for the smooth-wall side and the rough-
wall side for h + of 5.4, 10.8, 21.6.

The three rough-wall cases correspond respectively to equivalent sand-grain

roughness of ks+ = 10, 20, and 48; the ratio of ks to the physical peak-to-valley
height h for the three cases is thus 1.85, 1.85 and 2.2. These ks / h values are
analogous to Case 11 of Schlichting’s regular roughness patterns [20]. This implies
the “egg carton” roughness used here is comparable in terms of its effect on the
mean velocity to uniformly packed spheres, with distance to diameter ratio of 0.46.
Figure 3 shows the mean-velocity defect for the smooth-wall side and the rough-
wall side of the channel. Here the mean velocity is normalized by local u τ , and the
distance from the wall yw is normalized by δt , where δt is defined as the distance
from the wall to the y location corresponding to minimum rms velocity fluctuations.
It is worth mentioning that the results in the outer-layer for the rough-wall and
smooth-wall collapse better when scaled by δt compared to δ, suggesting that δt is
a better scaling measure for the outer layer. The mean-velocity profile in the outer
layer is independent of the roughness elements. This is consistent with the classical
notion that roughness affects the mean velocity only in the inner layer and the mean
velocity in the outer layer is unaffected by the roughness elements.
We next examine higher-order statistics. We are particularly interested in the
turbulent intensities for velocity and vorticity in both the inner and outer layers, since
behavior of the former allows us to infer how roughness affects the largest scales
of motion, while that of the latter indicates how it alters the small-scale features.
Figure 4a shows the rms velocity fluctuations normalized by u τ at the smooth
wall. They are plotted against wall-normal distance y normalized by the channel
half height δ. For the rough-wall side we present results only above the virtual
origin. All three components are larger above the rough-wall than the smooth-wall,
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 471

Figure 3. Mean velocity defect normalized by local u τ . The wall-normal distance is normalized
by δt (defined as the length from the wall to the y-location corresponding to the minimum rms
velocity fluctuations from Figure 4(a)), which is the effective boundary layer thickness.

(a) (b)
Figure 4. (a) Turbulent intensity of velocity components normalized by u τ at the smooth
wall. The wall-normal distance is normalized by the channel half-height. (b) The wall-normal
distance is normalized by δt .

showing that roughness results in an increase of turbulence intensity in agreement

with existing experimental and numerical observations. The trend (not shown) is
for the fluctuations to become more intense with increasing roughness height, all
other parameters being the same. Moreover, the peak location moves further away
from the wall, with increasing roughness. As the rough-wall layers and smooth-
wall layers have different thicknesses, an appropriate scaling is required. From
Figure 4a, we select δt corresponding to the location of the minimum velocity
fluctuations from the respective wall. In Figure 4b the results with the new scaling
472 K. BHAGANAGAR ET AL.

are shown. With this scaling all the velocity components are larger for the rough-
wall compared to the smooth-wall, indicating clearly that the outer-layer is affected
by roughness.
Figure 5a presents the velocity fluctuations normalized by their respective local
u τ , from the appropriate smooth or rough wall, with the distance from the wall ex-
pressed in wall units. With this normalization, the maximum streamwise fluctuation
and spanwise fluctuation components near the wall are smaller for the rough-wall
case than for smooth-wall, while the wall-normal component is larger for the rough-
wall case. Further away from the wall, the three components are smaller for the
rough-wall side. Note that although the streamwise fluctuation is larger than the
smooth wall in absolute terms, it represents a smaller fraction of the local u τ (which
now includes both pressure and viscous drag). Figure 5b shows the same results
now plotted in the outer units, with the distance from the wall normalized by δt .
Away from the wall all three velocity components for the rough-wall are smaller
compared to the smooth-wall. These results illustrate that for this flow the surface
roughness does indeed directly affect the outer layer, and that this effect involves
the large scales of motion.
Krogstad and Antonia [21] investigated two different rough surfaces, one mesh
roughness and the other rod roughness, but both having a similar roughness func-
tion. For mesh roughness they observed that streamwise fluctuation decreased and
the wall-normal fluctuation increased in the wall-region, similar to our results.
However, for their rod roughness, both streamwise and wall-normal fluctuations
decreased in the wall region. In the outer region, for both roughnesses, they ob-
served a prominent increase for the wall-normal fluctuations as observed in our
simulations. Other investigators, George and Simpson [9] among others, also re-
ported that the streamwise fluctuations decreased while the wall-normal fluctuations
increased in the inner region of rough wall.

(a) (b)
Figure 5. (a) Turbulent intensity of velocity components normalized by local u τ and plotted
in y + . (b) The distance from the wall (yw ) is normalized by δt .
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 473

(a) (b)
Figure 6. (a) The rms vorticity fluctuations normalized by the smooth-wall u τ . The distance
from the wall is normalized by δt . (b) Normalized by local u τ .

The behavior of the small scales can be revealed by the rms vorticity fluctuations.
In Figure 6a the rms vorticity fluctuations are normalized by u τ at the smooth
wall and they are plotted in terms of yw /δt . Roughness results in an increase of
each component of vorticity in the inner layer, in absolute terms. In Figure 6b
we scale them with local u τ . In contrast to the velocity fluctuations, the vorticity
fluctuations in the outer layer are not affected by the rough wall. Further, the
roughness-induced trend toward isotropy is apparent. This could be interpreted
as follows: the structure of small-scale turbulence is about the same for the smooth
and rough wall, in that in both cases the outer-layer vorticity is nearly isotropic and
of the same magnitude. Therefore, no change in small-scale statistics is observed in
the presence of roughness. The tendency toward isotropy of large-scale turbulence
has been reported before. For example, Antonia and Krogstad [6] reported that
rough wall resulted in an increased tendency toward isotropy from their analysis of
the Reynolds stress anisotropy tensor. Using a similar approach, Leonardi et al. [11]
also reported that roughness results in the tendency to isotropy.
To analyze the turbulent transport process we investigated further higher-order
statistics and the turbulence energy budget. The skewness of the u velocity is defined
as Su = u 3 /u 3rms , with similar definitions for the wall-normal v and spanwise w
components, Sv and Sw , respectively. Figure 7 shows the skewness of the three
velocity components Su , Sv and Sw . The streamwise skewness changes from a
positive value to a negative value for the smooth-wall side at around y + of 10,
while for the rough-wall case the change occurs at y + ≈ 20. Beyond y + of 30, Sv
for the smooth-wall case is distinctly different compared to the rough-wall side,
as Sv rough is mostly negative for the rough wall, whereas for the smooth wall
it is mostly positive and changes sign at the end of the outer layer. This effect of
roughness on Sv reflects a significant change in the large-scale structures in the outer
region. It also suggests that the turbulence transport in the horizontal directions is
not significantly affected, whereas that in the wall-normal direction is affected due
474 K. BHAGANAGAR ET AL.

Figure 7. Skewness of velocity fluctuations: Su , Sv and Sw respectively denote the skewness

of u, v and w velocity.

to the presence of roughness. This is in agreement with an experimental observation

by Cheng and Castro (Castro, private communication). Keirsbulck et al. [22] used
2D bars of square cross-section placed perpendicular to the flow direction. Their
second-order statistics indicated that the outer layer was unaffected by the type of
the roughness used by them. Their skewness results indicated that Su and Sv follow
the same trend for both the smooth and rough wall throughout the boundary layer,
suggesting that skewness is a good indicator in determining the changes in the large
scale structures in the outer layer.
The Kurtosis, or fourth-order moment, defined as K u = u 4 /u 4rms is quite different
near the smooth and rough-walls, but away from the wall it approaches a Gaussian
value for all three components for both rough- and smooth-wall cases (not shown
here). This indicates that the fourth-order statistics are affected only in the inner
layer due to the presence of roughness.
The transport equation for turbulence kinetic energy E = 12 u i u i is given by
DE
= P+T + +D− , (8)
Dt
where
∂U ∂ Ev ∂ pv
P = −uv , T =− , =− ,
∂y ∂y ∂y
1 ∂2 E 1 ∂u i ∂u i
D= , = . (9)
Reτ ∂ y 2 Reτ ∂ x j ∂ x j
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 475

Figure 8. Terms in the turbulent kinetic energy budget normalized by u τ at the smooth-wall.

Here, P, T , , D and respectively denote the rate of production, turbulent trans-

port, pressure transport, viscous diffusion, and dissipation.
Figure 8 shows the turbulent energy budget normalized by u τ at the smooth
wall and the wall-normal distance by yw /δt . Normalized in this way, the maximum
production rate as well as the maximum dissipation rate at the rough-wall side
is larger than those at the smooth side, as expected. The maximum value of the
turbulent transport is also larger than that of the smooth-wall case. The viscous
diffusion terms are different in the inner region. The pressure transport term is not
of significant value and the change is limited to very close to the rough wall in
spite of increased form drag due to roughness. In Figure 9, the same terms are now
normalized by the local u τ (i.e. the respective wall-shear velocity) and plotted in
y + . The production rate for the rough-wall case is less than that of the smooth-
wall case. The maximum value of the turbulent transport term remains about the
same. There are regions in the inner layer where the turbulent transport is toward
the wall for the smooth-wall case, whereas it is always away from the wall for
the rough-wall case. A similar trend is observed for the viscous diffusion term.
The direction of the transport of the turbulent kinetic energy due to the turbulent
velocity fluctuations as well as due to viscous stresses is different in the inner layer
due to the presence of roughness; no significant difference is discernible in the outer
layer. The main difference between the smooth- and rough-wall turbulent kinetic
energy budget observed by Keirsbulck et al. [22] was that in the inner region, the
viscous diffusion, advection and turbulent transport terms showed different trends
for the rough-wall case. Krogstad and Antonia [21] reported that the major effect
476 K. BHAGANAGAR ET AL.

Figure 9. Terms in the turbulent kinetic energy budget normalized by local u τ and plotted in
y+.

of roughness was in the turbulent transport term, and the direction of the turbulent
transport process depended on the nature of the roughness. Our results indicate that
both the turbulent transport and viscous diffusion terms are significantly modified
due to roughness.
At this point, as we are interested in the turbulent transport of the turbulent
structures, we concentrate on the transport term for the u 2 and uv transport equa-
tion, u 2 v and uv 2 , respectively (the transport terms for v 2 , w 2 are not significant
compared to these terms). It is evident from Figure 10 that the transport term for
u 2 is significantly modified by the surface roughness, whereas that for uv is not af-
fected significantly, indicating that the effect of surface roughness on u 2 compared
to other components extends much beyond the inner layer. Both the energy budget
and the transport terms of the Reynolds stresses indicate a modified wall-normal
transport process in the outer layer induced by the roughness. Further, the modifi-
cation of the transport term due to roughness arises mainly due to the contribution
from u 2 . This information should be valuable for turbulence modeling of rough
walls.

3.2. TURBULENCE STRUCTURES

To investigate the effect of surface roughness on turbulence structures, two-point

correlations of the velocity fluctuations are examined, from which the length
scales associated with the near-wall streaky structures and streamwise vortices
can be estimated. Figure 11 shows the two-point correlations of the streamwise and
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 477

Figure 10. Reynolds stress transport terms, u 2 v and uv 2 (transport terms in u 2 - and uv-budget,
respectively).

Figure 11. Two-point spanwise correlations with the separation distance normalized by the
channel half-height.

wall-normal velocity components separated in the spanwise direction. The correla-

tions are shown at wall-normal locations of y + = 30 and y + = 80. The separation
distance is normalized by the channel half height. Normalized in this way, no dis-
cernible difference in the location corresponding to the negative peaks in Ruu and
Rvv (respectively, corresponding to one half of the average streak spacing in the
spanwise direction and the mean radius of the near-wall streamwise vortices) be-
tween the smooth- and rough-wall cases is seen, suggesting that the average streak
spacing and the diameter of the streamwise vortices are not affected by the surface
478 K. BHAGANAGAR ET AL.

Figure 12. Two-point spanwise correlations with the separation distance normalized by local
wall variables.

roughness. In Figure 12, on the other hand, the separation distance is normalized
by the wall variables (local wall-shear velocity and kinematic viscosity). The two-
point correlation for the streamwise velocity, Ruu , indicates that the streak spacing
is increased to 140 wall units for the rough-wall case from the 100 wall units for
the smooth-wall case at y + = 30. A similar trend of increased streak spacing
is observed at y + = 80. Likewise, the two-point correlations of the wall-normal
velocity, Rvv , indicates that the average diameter of the near-wall streamwise vor-
tices is increased from 30 to 45 wall units at y + = 30, and a similar increase is
observed at y + = 80. The streamwise extent of these structures can be estimated,
similarly, from two-point correlations separated in the streamwise directions. Fig-
ure 13 shows Ruu and Rvv separated in the streamwise direction at y + = 30 and
y + = 80. The separation distance is again normalized by the channel half-height.
The length of the streaks decreases for the rough-wall case. Figure 14 shows the
separation distance normalized by the wall variables. No significant difference is
observed.
Figures 15 and 16 illustrate the comparison between streaks in the x-z plane
for the rough-wall and the smooth-wall at two different y + locations, y + = 5
and 80. At y + = 5 the streaks are elongated on the smooth side of the channel,
whereas at the same y + location the streaks look significantly different on the
rough-wall side. At y + = 80, on the smooth-wall side of the channel the organized
structures are not very apparent, whereas at the same y + location on the rough-
wall side of the channel they appear more organized. This suggests that roughness
results in organized structures for a larger y + compared to the no roughness case.
Summarizing, the effects of roughness on streaks are as follows: in wall units,
the streaks spacing increases in the spanwise direction, and the diameter of the
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 479

Figure 13. Two-point streamwise correlations with the separation distance normalized by
channel half-height.

Figure 14. Two-point streamwise correlations with the separation distance normalized by local
wall variables.

streamwise vortices (from Rvv ) also increases. In physical units, the streaks extend
to a larger y extent compared to the smooth-wall case.
Contour plots of u, v and w velocity components in an x-y plane were examined
(not shown here) to investigate the effect of roughness on turbulence structures. The
structure of the u component at the rough-wall side of the channel was distinctly
different from the smooth-wall side in that increased activity were present near
the rough-wall side. Contours of v indicated that the roughness resulted in more
480 K. BHAGANAGAR ET AL.

Figure 15. Contours of u at y + = 5 above the smooth-wall side (top) and the rough-wall side
(bottom).

elongated structures along the wall-normal direction. This is consistent with the con-
clusions drawn from the statistical results, that roughness modifies the cross-shear
transport process. The structure of the w component of velocity also indicated an
increased activity on the rough-wall side of the channel, and a change in the angle
of inclination of the structures. Figure 17 shows contour plots of the three vorticity
components. The spanwise vorticity (ωz ) contours reveal that an irregular pattern
of back flow is present both at the peak and valley locations of the roughness el-
ements. The turbulence near the elements affects the vorticity on the bumps. The
wall-normal vorticity (ω y ) contours show an organized pattern of dominant vorticity
close to the peaks of the roughness bumps. From contours of velocity fluctuations
(not shown) and Figure 17, it is apparent that the structures in the outer layer are
significantly modified due to the presence of roughness. This further confirms the
presence of communication between the inner and the outer layers of turbulence.
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 481

Figure 16. Contours of u at y + = 80 above the smooth-wall side (top) and the rough-wall side
(bottom).

Contours of streamwise velocity and v-w velocity vectors in a y-z plane (not
shown here) were also examined. The near-wall streamwise vortices were much
stronger at the rough-wall side compared to those at the smooth-wall side. Increased
turbulence activities at the rough-wall side were apparent. Different turbulence
structures observed in the rough-wall region, especially the stronger streamwise
vortices close to the roughness elements, led to the question of whether there
is a fundamentally different self-sustaining mechanism by which turbulence is
maintained in rough-wall-bounded turbulent flows. Our next step is to consider this
possibility.
Kim and Lim [23] reported that the virtual flow without the linear coupling term
between the wall-normal velocity and wall-normal vorticity contained no turbulence
482 K. BHAGANAGAR ET AL.

Figure 17. Contours of ωx (top), ω y (middle) and ωz (bottom) in an x-y plane. The lower wall
is the rough side of the channel and the upper wall is the smooth side of the channel.

structures (near-wall streamwise vortices and streaks in particular), resulting in

complete laminarization of an initially turbulent channel flow. They concluded that
the near-wall turbulence structures are maintained through the linear coupling term,
without which turbulence cannot be sustained. We performed similar numerical
experiments to investigate the role of the linear coupling term in the presence of
surface roughness. We want to address the question of whether the linear coupling
term continues to play an important role in maintaining near-wall turbulence in the
presence of surface roughness.
A numerical experiment involved artificially removing the linear coupling term
on the rough-wall side, while it was kept on the smooth-wall side. This numerical
experiment was performed at Reτ = 180, since we wanted to compare the results
with those in Kim and Lim [23]. Figure 18 shows the rms velocity fluctuations
from this calculation. Substantial reductions are discernible in the rough-wall side,
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 483

Figure 18. The rms velocity fluctuations with no linear coupling term on the rough-wall side
(left half in the figure) of the channel.

where the coupling term is absent. However, the reduction—the streamwise com-
ponent in particular—is much less compared to that observed in Kim and Lim’s
virtual flow. Apparently the surface roughness contributes directly to the mainte-
nance of near-wall turbulence, thus preventing the complete laminarization found in
the smooth-wall case. The rms vorticity fluctuations shown in Figure 19 indicate
more substantial reductions in the vorticity fluctuations.
We also examine turbulence structures in the wall region to investigate the effects
of surface roughness in the absence of the linear coupling term. Contour plots of

Figure 19. The rms vorticity fluctuations with no linear coupling term on the rough-wall side
(left half in the figure) of the channel.
484 K. BHAGANAGAR ET AL.

Figure 20. Contours of ω y at y + = 18 above the rough-wall with no linear coupling term on
the rough-wall side.

the wall-normal vorticity at y + = 18 on the rough-wall side of the channel (not

shown here) revealed the usual high- and low-speed streaks. In contrast, the same
contours from the rough wall without the linear coupling term showed no apparent
turbulence structures, as shown in Figure 20. The same trends were observed from
contour plots of other turbulence quantities in the x-z plane as well as those in x-y
planes (not shown here), indicating that the usual dynamical activity of near-wall
turbulence is absent without the linear coupling mechanism. It can be concluded
from the observations—the suppression of the linear mechanism associated with
the linear coupling term results in reduction of velocity and vorticity fluctuations
and complete disappearance of the near-wall turbulence structures—that the self-
sustaining mechanism of near-wall turbulence in the rough-wall channel is similar
to that of the smooth-wall channel. The only additional effect of surface rough-
ness is “kinematic effects” due to the presence of surface roughness. This has an
important implication for turbulent boundary layer control: one can continue aim-
ing at suppressing or reducing the linear mechanism due to the coupling term in
designing a robust controller for drag reduction in rough-wall turbulent boundary
layers, as was shown for the smooth-wall boundary layer by Kim [24].

3.3. R OUGHNESS SUBLAYER

The depth of the roughness sublayer is a subject of ongoing debate, and there is
little information available on the characterization of the flow within this region. In
this section we determine the depth of the sublayer for the large- and small-scale
features of the flow. As the roughness elements in the x-z plane are represented by
the double sine function, the averaging is performed at two different inhomogeneous
locations—the “peak” and “valley” regions of corrugated surface. The underlying
assumption is that the statistics at these two locations will yield a good estimate of
the extent of inhomogeneity present in the layer. The rms velocity fluctuations are
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 485

Figure 21. The rms velocity fluctuations normalized by u τ at the smooth wall, shown above
the peak and valley locations of roughness elements.

shown in Figure 21 at peak and valley locations of the corrugated surface. Spatial
homogeneity is not achieved until y/δ = −0.82. We performed this for three
different roughness heights and obtained the depth of the roughness layer i.e ζlarge /δ
to be about 1.5h, where subscript large represents the length scale for the large scale
features of the flow, since it was revealed by a large-scale dominated statistics. A
similar examination of the rms vorticity fluctuations revealed a somewhat smaller
roughness sublayer, as shown in Figure 22. Spatial homogeneity of the small-scale
structures is achieved at y/δ of −0.86, resulting in the depth of the roughness
sublayer for small-scale roughness sublayer i.e ζsmall /δ to be about 1.1h; here the
subscript small represents the length scale for the small scale features of the flow. It
is worth noting that the horizontal components of the vorticity, ωx and ωz , exhibit
significant spatial inhomogeneity due to the roughness elements, but that the wall-
normal component of the vorticity ω y does not.

Figure 22. The rms vorticity fluctuations normalized by u τ at the smooth wall, shown above
the peak and valley locations of roughness elements.
486 K. BHAGANAGAR ET AL.

Table I. Case study for varying

l z /δ with fixed l x /δ = 0.6 and
ρ = 0.21.
Cases l z /δ
A1 0.5
A2 0.25
A3 0.12
A4 0.08
A5 0.02
A6 0.01

Table II. Case study for

varying l x /δ with fixed
l z /δ = 0.25 and ρ = 0.21.
Cases l x /δ
B1 1.2
B2 0.8
B3 0.24
B4 0.04

4. Parametrization of the Roughness Surface

The roughness parameters that we will consider are the streamwise l x and spanwise
l z size (i.e. wavelength) of the roughness elements (see Equation (4)) for a given
roughness density (ρ), which is defined as the ratio of the area of the roughness
elements in the horizontal plane at y = σ0 to the total horizontal area. We chose
the roughness elements to be ellipsoidal in shape, and so that we had the flexibility
to vary the dimensions and the roughness density of these roughness elements.
Considering that a large number of simulations had to be performed for this para-
metric study, all simulations were performed at Reτ ≈ 180, based on u τ at the
smooth wall, using 192 × 257 × 192 grid points in the streamwise (L x /δ = 4 π),
wall-normal, and spanwise (L z /δ = 4 π/3) directions. Computed statistics for this
Reynolds number revealed the same trend as those for Reτ = 400 presented in
previous sections.
We began our study by varying the size of the elements in the spanwise direc-
tion, and with all the other parameters kept constant. We performed a case study
for six different cases shown in Table I, where l z /δ is varied from 0.5 to 0.01, with
fixed l x /δ = 0.6 and ρ = 0.21. Figure 23 shows the mean velocity normalized by
the local u τ for all six cases plotted in wall units. For all the cases, the roughness
function (U + ) shows a negligible variation, indicating that that l z /δ does not play
a crucial role in determining the shift in the mean velocity profile. The roughness
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 487

25 25

20 rough 20 rough
smooth smooth

15 15

U+
U+

10 10

5 5

0 0 1 2 0 1 2
10 10 10 10 10 10
+
(a) Case A1 y+ (b) Case A2 y

25 25

rough rough
20 smooth 20
smooth

15 15

+
U+

U
10 10

5 5

0 1 2
100 101 102 10 10
+
10

(c) Case A3 y+ (d) Case A4 y

25 25

rough 20 rough
20
smooth smooth

15 15
U+
U+

10 10

5 5

0 1 2 0 1 2
10 10 10 10 10 10
+
(e) Case A5 y+ (f) Case A6 y

Figure 23. Mean velocity normalized by local u τ for cases with varying l z /δ and all other
parameters fixed.

length scale y0 (defined in Section 3), which is a surface property determined by

roughness geometry, also indicates little variation with the size of the roughness
elements in the spanwise direction. Figure 24 shows velocity fluctuations normal-
ized by the local u τ and δt . All six cases are shown. For Cases A1–A3, the u and w
fluctuations are smaller on the rough-wall side compared to the smooth-wall side,
while v fluctuations are larger on the rough-wall side close to the wall (i.e. inner
layer). Away from the wall (outer layer), all three velocity components are smaller
on the rough-wall side, which signifies that for these cases the outer layer of the
turbulent boundary layer are also affected by the presence of roughness. For Cases
A4–A6, all three velocity components in the inner layer show the same trend as the
previous cases. In the outer layer, however, u and w fluctuations of the rough-wall
side and smooth-wall side almost collapse, and the differences in the v fluctuations
are not very large. The outer layer is not significantly affected by the presence of
roughness for these cases. This leads to an interesting conclusion that the spanwise
size of the roughness elements plays an important role in determining whether the
outer layer is altered by roughness. The size of the roughness in the spanwise direc-
tion (normalized by the boundary layer thickness) does not play a significant role in
mean velocity statistics, but it is significant to determine the nature of the velocity
fluctuations, and hence the dynamics of the outer layers. This perhaps explains why
488 K. BHAGANAGAR ET AL.

Figure 24. The rms velocity fluctuations normalized by local u τ for cases with varying l z /δ
and all other parameters fixed.

many investigators with different roughness geometries obtained more or less sim-
ilar roughness shifts of the mean velocity profile, but reported different results for
the higher-orders statistics, such as the rms velocity fluctuations in the outer layer.
In the next set, we varied the size of the roughness elements in the streamwise
direction but kept all the other parameters constant. (Table II). We fixed l z /δ at
two specific values, at l z /δ = 0.25 (where the outer layer statistics depended on
the presence of roughness), and l z /δ = 0.02 (where the outer layer statistics did
not). The mean velocity normalized by the local u τ is shown in Figure 25. A slight
variation in the roughness function (from 6.14 to 7.12) was observed. This is in
contrast to the previous varying l z case, but the trend is not very pronounced and
more study is needed to determine the exact nature of the dependence of U + with
l x . Figure 26 presents the rms velocity fluctuations. For all the cases, in the inner
layer the u and w fluctuations for the rough-wall side are lower than the smooth-
wall side, while the v fluctuations are higher. In the outer layer, all three velocity
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 489

Figure 25. Mean velocity normalized by local u τ for various l x /δ with other parameters fixed.

components are lower for the rough-wall side compared to the smooth-wall side.
The same test has been performed at a different l z /δ = 0.02. Results similar to
Case A5 (Figure 24) have been observed, indicating that the physics in the outer
layer is dominated by the size of the roughness element in the spanwise direction,
rather than the size in the streamwise direction.

5. Summary and Concluding Remarks

Direct numerical simulations of a turbulent channel flow between a smooth and
rough wall have been performed to investigate the effects of surface roughness
on wall-bounded turbulence. To get a clearer picture of the impact of roughness
in turbulent boundary layers, we have investigated the effects of 3D roughness
arranged in an “egg carton” pattern. We performed a statistical analysis of the
large- and small-scale features of the flow. When normalized by u τ at the smooth-
wall side, the rms velocity fluctuations at the rough-wall side are higher than the
smooth-wall side. A similar effect is seen for the vorticity fluctuations. But when
normalized by the local u τ , the u and w fluctuations are smaller and the v fluctuation
is higher for the rough-wall side in the inner layer, indicating a more isotropic state.
In the outer layer all three velocity fluctuations are a smaller fraction of u τ on the
rough-wall side. The velocity fluctuations are thus altered throughout the boundary
490 K. BHAGANAGAR ET AL.

Figure 26. The rms velocity fluctuations normalized by local u τ for various l x /δ and all other
parameters fixed.

layer due to the presence of roughness. The vorticity fluctuations, on the other
hand, are not significantly altered in the outer layer, in that for both surfaces they
exhibit the same dependence on u τ and yw /δ. Hence there is interaction between
the inner and outer layers of the turbulent boundary layer at the large scales but not
at the small scales. The results from the skewness and the transport terms of the
Reynolds-stress budgets indicate modified wall-normal transport in the outer layer
due to the roughness. Since the structures in the outer layer depend on whether the
surface is smooth or rough, the present results can be added to the body of evidence
that contradicts classical outer-layer similarity theory. The two-point correlations
for the streamwise velocity reveal that the streak spacing in wall units increases
for the rough-wall case. Likewise, the two-point correlations of the wall-normal
velocity imply that the average diameter of the near-wall quasistreamwise vortices
increases in terms of wall units. The correlation results also indicate that the mean
streamwise length of the streaks decreases, measured with respect to local u τ /ν,
on the rough-wall side.
The thickness of the roughness sublayer depends on which statistic is used
to characterize it. The roughness-induced inhomogeneity influences the velocity
fluctuations (i.e. the larger scales) over a depth of 1.5 times the bump height h, while
EFFECT OF ROUGHNESS ON WALL-BOUNDED TURBULENCE 491

the roughness sublayer defined by the small-scale-dominated vorticity fluctuations

is 1.1h.
One of the important conclusions that can be drawn from this study is that the
streamwise and spanwise dimensions of roughness elements of fixed height play
a crucial role in determining whether the roughness affects the outer layer. The
spanwise size l z /δ of the roughness does not influence the mean velocity statistics,
but does have a large impact on the velocity fluctuations in the outer layer. This may
explain why investigators with different roughness geometries can observe similar
log-law shifts U + , but offer different interpretations of the outer-layer physics
based on their observations of higher-order statistics in the outer layer.
Much of the present results summarized above are in agreement with previ-
ous experimental/numerical results or verify conjectures that have been proposed
before. But some of the present results are at odds with some recent experimental re-
sults, which seem to indicate that the outer layer is unaffected by surface roughness.
We have presented one plausible scenario how such difference can arise. There are
other different factors (e.g. Reynolds number, different shapes and configurations
of roughness elements) that can lead to different results. Much more work is needed
for better understanding of the effect of roughness on wall-bounded turbulence.

Acknowledgments
This paper is a written version of the presentation by John Kim at an international
conference on fluid mechanics, which was held in honor of Professor Robert An-
tonia on the occasion of his 60th birthday. John Kim acknowledges the fruitful
collaborations he has had over the years with Professor Antonia. We are grateful
to Professor Ian Castro for his comments on a draft of this manuscript, and to Dr.
John Mansfied for his contribution to the immersed-boundary approach used in the
present work. This work has been supported by Office of Naval Research (N00014-
01-1-0811, Dr. Ronald Joslin). The computer time provided by NSF NPACI Centers
(Blue Horizon at SDSC and Copper at NCSA) is also gratefully acknowledged.

References
1. Raupach, M.R., Antonia, R.A. and Rajagopalan, S., Rough-wall turbulent boundary layers.
Appl. Mech. Rev. 44 (1) (1991) 1–20.
2. Wood, N. and Mason, P., The pressure force induced by neutral turbulent flow over hills. Quart.
J. Roy. Meteorol. Soc. 119 (1993) 1233–1267.
3. Cheng, H. and Castro, I.P., Near wall flow over urban-like roughness. Boundary-layer Meteo-
rology (2002) 229–257.
4. Perry, A.E., Lim, K.L. and Henbest, S.M., An experimental study of the turbulence structure in
smooth- and rough-wall boundary layers. J. Fluid Mech. 177 (1987) 437–466.
5. Krogstad, P., Antonia, R.A. and Browne, W.B., Comparison between rough-and smooth-wall
turbulent boundary layers. J. Fluid Mech. 245 (1992) 599–610.
6. Antonia, R.A. and Krogstad, P.A., Turbulence structure in boundary layers over different types
of surface roughness. Fluid Dyn. Res. 28 (2001) 139–157.
492 K. BHAGANAGAR ET AL.

7. Tachie, M.F., Bergstrom, D.J. and Balachandar, R., Rough wall turbulent boundary layers in
shallow open channel flow. J. Fluids Eng. 122 (2000) 533–540.
8. Keirsbulck, L., Mazouz, A., Labraga, L. and Tournier, C., Influence of the surface roughness on
the third-order moments of velocity fluctuations. Exp. in Fluids 30 (2001) 592–594.
9. George, J. and Simpson, R.L., Some effects of sparsely distributed three-dimensional roughness
elements on two-dimensional turbulent boundary layers. AIAA Paper 2000–0915, (2000).
10. Ashrfian, A. and Andersson, H., DNS of turbulent flow in a rod-roughened channel. In: Third
International Symposium on Turbulence and Shear Flow Phenomena, Sendai, Japan, June 25–27
(2003) Vol. 1, pp. 117–122.
11. Leonardi, S., Orlandi, P., Djenidi, L. and Antonia, R., Structure of turbulent channel flow with
square bars on one wall, In: Third International Symposium on Turbulence and Shear Flow
Phenomena, Sendai, Japan, June 25–27 (2003) pp. 123–128.
12. Shafi, H.S. and Antonia, R.A., Small-scale characteristics of a turbulent boundary layer over a
rough wall. J. Fluid Mech. 342 (1997) 263–293.
13. Bandyopadhyay, R. and Watson, R.D., Structure of rough-wall turbulent boundary layers. Phys.
Fluids 31 (1988) 1877–1880.
14. Mohd.-Yusof, J., Combined immersed-boundary/B-spline methods for simulations of flow in
complex geometries. CTR-Annual Research Briefs, Stanford University/NASA, American
Samoa (1997).
15. Fadlun, E.A., Verzicco, R., Orlandi, P. and Mohd-Yusof, J., Combined immersed-boundary/
finite-difference methods for three-dimensional complex flow simulations. J. Comput. Phys.
161 (2000) 35–60.
16. Goldstein, D., Handler, R. and Sirovich, L., Modeling a no-slip flow boundary with an external
force field. J. Comp. Phys. 105 (1993) 354–366.
17. Verzicco. R. and Orlandi, P., A finite-difference scheme for three-dimensional incompressible
flows in cylindrical coordinates. J. Comp. Phys. 123 (1996) pp. 402.
18. Kim, J., Moin, P. and Moser, R.D., Turbulence statistics in fully developed channel flow at low
Reynolds number J. Fluid Mech. 177 (1987) 133–166.
19. Nikuradse, K., Stromungsgesetze in Rauhen Rohren. [English Trans. NACA Tech. Memo No.
(1292).
20. Schlichting, H., Boundary Layer Theory 7th edn. (1979).
21. Krogstad, P. and Antonia, R.A., Surface roughness effects in turbulent boundary layers. Expts.
Fluids 27 (2001) 450–460.
22. Keirsbulck, L., Labraga, L., Mazouz, A. and Tournier, C., Surface roughness effects on turbulent
boundary layer structures., J. Fluid Engrs. 124 (2003) 127–135.
23. Kim, J. and Lim, J., A linear process in wall-bounded turbulent shear flows. Phys. Fluids 12(8)
(2000) 1885–1888.
24. Kim, J., Control of turbulent boundary layers. Phys. Fluids 15(5) (2003) 1093–1105.
25. Krogstad, P. and Antonia, R.A., Structure of turbulent boundary layers on smooth and rough
walls. J. Fluid Mech. 277 (1994) 1–20.