The 6th Int. Conf.

on Hydroscience and Engineering (ICHE-2004), May 30-June 3, Brisbane, Australia 1

STRUCTURE OF THE TURBULENT FLOW OVER A ROUGH BED

J. J. R. Williams1 and Krishna M. Singh2

ABSTRACT

This paper presents the results of a direct numerical simulation of turbulent flow over a rough bed in
an open channel. We consider a hexagonal arrangement of spheres on the channel bed. The depth of
flow has been taken as four times the diameter of the spheres and the Reynolds number has been
chosen so that the roughness Reynolds number is greater than 70, thus ensuring a fully rough flow.
A parallel code based on finite difference, domain decomposition, and multigrid methods have been
used for the DNS. Computed results are compared with available experimental data. We report the
first and second order statistics, detailed description of the instantaneous flow field, detection and
analysis of coherent structures, variation of lift and exchange coefficients. Good agreement with
experimental results is seen for the mean velocity, turbulence intensities and Reynolds stress.
Further, the DNS results clearly bring out the discrepancies in the experimental statistics, and
provide an accurate quantitative statistics for the rough bed flow. Analysis of the instantaneous flow
field, correlations and coherent structures confirms the streaky nature of the flow near the effective
bed and the existence of a hierarchy of vortices aligned with the streamwise direction. The exchange
coefficients indicate a large degree of mixing between the fluid trapped below the mid-plane of the
roughness elements and that above it.

1. INTRODUCTION

A clear understanding of turbulent flow over rough surfaces is crucial in the accurate modelling of
sediment transport, dispersion of pollutants and heat transfer. Over the past few decades, significant
experimental work has been carried out on the measuring of flow structures that occur over rough
beds: Grass (1971), Nezu (1977), Grass, Stuart and Mansour-Tehrani (1991), Grass and Mansour-
Tehrani (1996), and Defina (1996). These investigations show that distinctive pattern of low-
velocity streaks observed for turbulent flow over smooth surfaces is also a characteristic of the near-
wall flow over rough boundaries. The gross features of the flow, e.g. variation of the mean
streamwise velocity, rms fluctuations, and Reynolds stresses, are also fairly similar. However, much
remains to be done to fully understand the flow behaviour and to quantify the sizes of coherent
structures and turbulent quantities. A direct numerical simulation (DNS) of such flows is able to
provide an accurate picture, and is the objective of the present work.
Direct numerical simulation provides an ideal approach for the solution of time dependent
turbulent flows. In DNS, the turbulent motions are resolved down to approximately the Kolmogorov

1
Professor, Department of Enginering, Queen Mary, University of London, Mile End Road, London E1 4NS, UK
(J.J.R.Williams@ qmul.ac.uk)
2
Assistant Professor, Department of Mechanical and Industrial Enginering, Indian Institute of Technology Roorkee,
Roorkee 247 667, India (singhfme@iitr.ernet.in, k_m_singh@yahoo.com)
 Williams and Singh. Structure of the turbulent flow over a rough bed 2

scales. There are no turbulence models involved as in RANS or large eddy simulation (LES)
modelling. The only downside is that computational demands become excessive at high Reynolds
numbers. LES alleviates this excessive demand by restricting the computational resolution to the
large eddies (or grid scales) and representing the subgrid scales by a subgrid model. In this work, we
employ this approach for an initial simulation to provide a nearly converged turbulent flow field to
use as the initial conditions for the DNS.
Beginning with a 323 simulation of isotropic turbulence by Orszag and Patterson (1972}, use
of DNS has become increasingly widespread in the turbulence research with the advances in
computing technology. Moin and Mahesh (1998) provide a review of DNS as a research tool
covering its various applications, and its impact on turbulence modelling and understanding the
structure of the turbulent flows. The computational requirements of DNS have restricted its
applications to low Reynolds number flows in relatively simple geometry. For example, let us
consider the simulation of channel flows. The first DNS of fully developed channel flow was
performed by Kim, Moin and Moser (1987). Since then, it has been applied to channel flows of
increasing physical and geometrical complexities, viz. free surface (Lam and Banerjee 1992),
transpiration (Sumitani and Kasagi, 1995), riblets (Choi et al., 1993) etc. However, all these
applications have been limited to smooth-wall or hydrodynamically smooth flows. DNS of the fully
rough flows have recently appeared in conferences (Bhaganagar and Kim 2002, Lee 2002, Leonardi
et al. 2002, Nozawa et al. 2002, Singh and Williams 2003), and the present study represents a DNS
of one of the most complicated configurations considered so far.
We consider the simulation of flow over a hexagonal arrangement of the spheres of diameter k
on the bed of an open channel. The choice of the geometry has been motivated by the availability of
the experimental data (Grass and Mansour-Tehrani, 1991,1996; Defina, 1996). The depth of the
flow, d, is taken to be four times the diameter of the spheres on the bed, and Reynolds number is
chosen to ensure a fully rough flow regime. Based on the data obtained from the DNS, we have
analysed the features of the flow and compared them with the published experimental data. Initial
results on statistical structure of the flow were presented in Singh and Williams (2003). In this
paper, we present further details on statistical structure (first and second order statistics) of the rough
bed flow, detection of the coherent structures and implications for the bed stability and sediment
transport.
We start with a brief description of the problem and numerical techniques in Section 2.
Statistical features are presented in Section 3, followed by the details of coherent structures in
Section 4. Some implications for the sediment transport are described in Section 5, followed by the
concluding remarks in Section 6.

2. PROBLEM DEFINITION AND NUMERICAL SOLUTION

We consider the flow in a open channel consisting of a packed hexagonal arrangement of spheres of
diameter k on its bed (Figure 1). We assume the turbulence to be homogeneous in horizontal planes,
and use a computational box of the dimensions 43d  4d  d (streamwise  spanwise  depth) with
periodic boundary conditions in the horizontal plane. The depth of the flow, d, is taken to be four
times the diameter of the spheres, k, on the bed, and Reynolds number is chosen to ensure a fully
rough flow regime. A no-slip boundary condition is assumed at the bed (and surfaces of spheres)
and a stress-free hard lid is used for the top surface.
The Navier-Stokes equations are solved using a second order finite difference method applied
to a cell-centred staggered grid for spatial discretization and a second order Adams-Bashforth
method for time integration. The Poisson equation for pressure at each time step is solved using a
fictitious domain multigrid preconditioner and conjugate gradient method. The surfaces of the
spheres have been approximated as a series of steps in the Cartesian grid and these numerical
 Williams and Singh. Structure of the turbulent flow over a rough bed 3

techniques have been implemented in a parallel complex geometry DNS-LES code CgLes. The
computations have been run on the Cray T3E at the University of Manchester.

Figure 1 Geometry of the computational domain.

We have used a Cartesian grid of resolution 1024  512  128 in the streamwise, spanwise and
vertical directions respectively. Flow Reynolds number Re+(= U*d/ν) is set to 533, and results in the
roughness Reynolds number ks+ = 102 indicating a completely rough flow regime. The Kolmogorov
length scale η+ is estimated to be 3.0 and the grid sizes (in wall units) Δx+, Δy+ and Δz+ are
calculated to be 3.6, 4.2 and 4.2 respectively. Thus, the gird resolution is sufficient to resolve all the
essential length scales, and hence, no subgrid model has been used in the computations.

3. STASTISTICAL STRUCTURE OF THE ROUGH BED FLOW

The statistical description of flow over a rough bed is normally characterized by two parameters: (a)
the effective location of the bed, yb, and (b) the equivalent sand roughness, ks. These are determined
from the flow statistics by solving a two parameter nonlinear curve-fitting problem using least
squares and Newton-Raphson techniques. Table 1 lists these values together with the other mean
flow properties. Note that the simulated flow corresponds to a fully rough flow regime with a
roughness Reynolds number ks+ = 102.

Table 2. Mean flow variables obtained from DNS

Effective location of bed, yb = 0.81k Roughness Reynolds number, ks+ = 102

Equivalent sand roughness, ks = 0.768k Re+ = U* d/ = 533.33
Effective flow depth, h = d – yb = 3.19 k Reh+ = U* h/ = 425.33
Bulk mean velocity, Um / U* = 7.318 Rem= Um h/ = 3112
*
Mean velocity of top surface, Uc / U = 11.649 Rec= Uc h/ = 4955

Based on the accumulated DNS data, we have obtained the distribution of the mean streamwise
velocity, turbulent intensities, Reynolds stress and its various correlations, and compared these with
 Williams and Singh. Structure of the turbulent flow over a rough bed 4

the available experimental data (Singh and Williams, 2003). We briefly describe the distribution of
the mean velocity and turbulence intensities in this section, followed by the quadrant analysis.

3.1 Mean Velocity Distribution

The mean streamwise velocity profile normalised with the bed shear velocity U* is presented in
Figure 3. We can clearly see the absence of the viscous sub-layer and the presence of a retarded
fluid layer near the bed (Y+ <40) due to the penetration of the roughness elements in the logarithmic
region. Comparison with the experimental results of Grass (1971), Grass et al. (1991) and Defina
(1996) show that the DNS profile follows the standard pattern of velocity defect increasing with
wall roughness. Let us note that the experimental data of Grass (1971) corresponds to the roughness
Reynolds number of 84.7 based on the roughness height k. To obtain the roughness Reynolds
number ks+, we have scaled it with ks/k = 0.88 (Grass and Mansour-Tehrani, 1996).

Figure 3 Mean velocity profiles. Thin Figure 4 Mean velocity profiles plotted
vertical line denotes the top of the against Y/ks. Thin vertical line denotes
roughness elements. the top of the roughness elements.

A better comparison of the DNS result with the experimental data can be made by plotting the
mean velocity with respect to the vertical length scaled by the equivalent sand roughness ks. Figure 4
presents the plot of U+ and Y/ks. The mean velocity profile obtained with DNS shows excellent
agreement with the recent experimental results (Defina, 1996 and Grass et al., 1991).
Let us have a look at the mean flow below of the top of the roughness elements (Figure 5).
Some interesting features can be observed in the flow around the roughness elements: the presence
of viscous sublayer near the bed (y/d < 0.06, y+ < 32), a fully turbulent layer on top of it, a mild back
flow near the mid-plane followed by sharp velocity gradient. Moreover, this slow moving (nearly
still) fluid flow around the mid-plane of the roughness elements is turbulent as indicated by the
variation of the effective eddy viscosity.
 Williams and Singh. Structure of the turbulent flow over a rough bed 5

Figure 5 Mean velocity and scaled effective viscosity near the bed.

3.2 Turbulence Intensities

The turbulence intensities normalised by the bed shear velocity are shown in Figure 6. We have also
included the experimental values from Grass (1971) and Nezu (1977) corresponding to roughness
Reynolds numbers of 84.7 and 136 respectively. Since, these roughness Reynolds numbers are
based on the roughness height k, these values actually correspond to ks+ ~ 74.5 for Grass (1971) and
ks+ ~ 100 for Nezu (1977). Thus, data from Grass (1971) correspond to a less rough bed, whereas
those of Nezu (1977) correspond to a roughness condition very similar to that of the present DNS.
The computed values are lower than the experimental values of Grass (1971), but show a good
agreement with the experimental data of Nezu (1977) everywhere except near the bed. This trend is
as expected. Let us note that there is considerable scatter in the experimental data, especially near
the bed. Hence, the DNS values are likely to provide a more accurate picture than the experimental
results of Nezu (1977). At the same time, the peak value of u ' / U * is 1.83 which is very close is the
experimentally observed value around 1.9 suggested by Nezu (1977). Two peaks can be seen in the
profile for the spanwise fluctuations: one below (y/d = 0.22, Y+ = 11) and one above (y/d = 0.29, Y+
= 48) the top of the roughness elements. This feature marks the difference in the near-bed profiles of
turbulent fluctuations between the smooth and rough bed flows.

Figure 6 Turbulent intensities

 Williams and Singh. Structure of the turbulent flow over a rough bed 6

3.3 Quadrant Analysis

Quadrant analysis provides detailed information on the contribution of various events in the flow to
the total turbulence production. The events in the flow are linked to the quadrants in the (u’, v’)-
plane (Lu and Willmarth,1973). The first quadrant contains outward motion of high speed fluid, also
referred to as an outward interaction event. The second quadrant is associated with the ejection of
low speed fluid away from the wall. The third quadrant is linked with inward interaction event (i.e.
motion of low speed fluid towards the bed); whereas the fourth quarter is associated with the sweep
event (i.e. the inrush of high-speed fluid). The even quadrants contribute to the positive production
and the odd quadrants to the negative production of the Reynolds shear stress. Figure 7 presents the
contribution to the Reynolds shear stress from each quadrant above the effective bed. We can clearly
see the dominance of ejection events (second quarter) away from top of the roughness elements (Y+
> 36). However, sweep events (fourth quarter) dominate near the top of the roughness elements (and
above the effective bed).

Figure 7 Contribution to Reynolds shear stress from each quadrant normalized by the local mean
Reynolds shear stress (RSj denotes the j-th quadrant). Thin vertical line denotes
the top of roughness elements.

4. COHERENT STRUCTURES IN THE ROUGHBED FLOW

4.1 Effect of roughness on flow structure and streak spacing

The turbulent flow around the roughness elements would certainly be influenced by the roughness
length scale. This effect is clearly demonstrated by the plot of the two-point correlations shown in
Figure 8 at different y-locations (y/d = 0.027 and y/d = 0.246). Numerous well-defined peaks in
these plots indicate the presence of small-scale disturbances around individual spheres. However,
the correlations just a small vertical distance above the top of the spheres at y/d = 0.308 show a
much smoother profile, and thus indicate the decoupling of the flow field from the influence of the
individual roughness elements. These results confirm the observation of Grass and Mansour-Tehrani
(1996) on the apparent ability of the rough wall flow to order itself very rapidly over a small vertical
distance above the roughness elements.
The distinctive pattern of low-velocity streaks have been observed in smooth as well as rough
wall flows. Plots of u’ < 0 at different vertical locations (not included here) confirm the presence of
elongated low-speed streaks in the rough bed flow. Further, we have observed an increase in average
separation between the streaks as we move away from the bed.
 Williams and Singh. Structure of the turbulent flow over a rough bed 7

Figure 8 Two-point correlations at different y-locations near the bed.

Left side: Streamwise separations; Right side: Spanwise separations.

We can estimate the mean transverse spacing of the streaks from the plot of two-point
correlations. Let z be the distance to the first minimum of the spanwise correlation of streamwise
velocity. The mean transverse spacing between the streaks, , is then roughly 2z. This mean
transverse spacing of low-speed streaks determined from spanwise correlations at different vertical
locations is plotted in Figure 9. For comparison with the experimental data of Defina (1996)
 Williams and Singh. Structure of the turbulent flow over a rough bed 8

(identified by : S2 and: S3 in this plot), we have scaled the spacing () as well as the distance
from the top of the roughness elements (yt) by the equivalent sand roughness ks. Near the top of the
spheres, ks  3.2 which is very close to ks  3.8 suggested by Grass and Mansour-Tehrani (1996).
We can observe that this value is nearly constant near the top of the roughness elements (ytks  0.5),
which indicates the presence of a single vortical hierarchy active near the top of the spheres. Further
up from the bed, the streak spacing grows, but the growth pattern is rather non-uniform suggesting
the presence of different sets of interlinked vortical hierarchies. A uniform linear growth is seen
only near the top surface.

Figure 9 Transverse spacing of streaks scaled by equivalent sand roughness ks.

(yt is the distance from the top of the roughness elements.)

4.2 Coherent Structures Identification from Instantaneous Flow Field

The velocity and pressure distribution on a vertical plane aligned with the streamwise direction is
shown in Figure 10. The velocity field is presented in a convected frame of reference moving with
the bulk velocity Ub. Cross-sections through tranverse vortices, ejections and sweep zones (two of
these marked by E and S for the sake of illustration). Occurrence of elongated low-pressure zones,
some of which extend close to the top of the channel, indicates the presence of a few large vortical
structures presumably formed by the coalescence of smaller vortices originating near the top of the
roughness elements.

Figure 10 Velocity vectors in advected frame of reference (u-Ub, v) and contours of

pressure on a streamwise-vertical plane z/d = 0.012
 Williams and Singh. Structure of the turbulent flow over a rough bed 9

The distribution of the velocity and pressure on a spanwise cross-section is presented in Figure
11. A few pairs of counter-rotating vortices can be seen near the top of the roughness elements in
almost all of the cross-sectional plots. However, single vortical cores are more common which
suggest a preponderance of single leg (or asymmetric) hairpin vortices in the flow. Further, we can
see a large number of small-scale vortical structures near the top of the spheres. The number of
vortex cross-sections decreases with increase in the distance from the bed, with far fewer vortex
cross-sections observed near the top surface of the channel. These observations further corroborate
the hypothesis of the presence of a hierarchy of vortical structures extending over the full depth of
the flow.

Figure 11 Velocity vectors (v,w) and contours of pressure on a spanwise cross-section (x/d = 0.105)

For further investigation of the near wall coherent structures, we have used the so-called 2
technique of Jeong and Hussain (1995). A plot (not included here) of mean and rms profiles of 2
(obtained from an instantaneous flow field) with depth shows two closely spaced peaks near the top
of the roughness elements, indicating intense vortical activity there. Figure 12 shows the iso-
surfaces of 2 = -240 (which 2% of the minimum value of 2) in the range 0.18 < y < 0.37 (which
represents the region of the most-dominant vortical activity). We can clearly observe a large
collection of slender structures, with the dominant vortices inclined at a small angle to the
streamwise direction. The streamwise extent of the dominant vortices varies from 100 to 200 wall
units which is very similar to the streamwise extent of near-wall vortices observed by Jeong et al.
(1997) in smooth channel flow.

Figure 12 Top-view of isosurfaces of 2 = -240 (ordinate is the z-axis, i.e. spanwise direction)
 Williams and Singh. Structure of the turbulent flow over a rough bed 10

5. IMPLICATIONS FOR SEDIMENT TRANSPORT

5.1 Distribution of drag and lift experienced by the roughness elements
The roughness elements experience drag and lift forces in the turbulent flow field. Of these, the
distribution of lift is very important in the context of sediment transport. This is because the
roughness elements subjected to lift in excess of their weight would be lifted up from the bed and
washed away. Drag plays an equally important role in uprooting the individual elements through
rolling action.
Let us note that a randomly varying distribution of velocity and pressure fields around the
roughness elements would give rise to a random distribution of lift and drag experienced by the
individual spheres. This is clearly seen in Figures 13 and 14. There are considerable variations in
the lift or drag experienced by individual roughness elements. Moreover, the lift is even negative for
many elements, which have a high-pressure zone near their top surface. Probability distributions of
the lift and drag in Figures 13 and 14 support these observations.
Another interesting feature can be observed from Figures 13 and 14. Most of the roughness
elements which experience very high lift are subjected to high drag as well, and their downstream
neighbours experience negative lift. This combination is highly favourable for lifting and rolling of
these elements over their downstream neighbours, and thereby disturbing the bed. Whether this
would actually happen would need detailed analysis of 3-D forces acting on each roughness
element. However, if we use the Shields diagram for the bed stability, the normalized lift L > 0.58
(corresponding to the Shields function Fs = 0.056 for fully rough flows) for a roughness element to
be uprooted from the bed. For the present case, Lmax = 0.14 which indicates that the flow Reynolds
number is too low to cause the bed instability by lifting and carrying away the roughness elements.
At the same time, the preceding analysis provides some insight into the process of possible bed
instability at a higher flow Reynolds number.

Figure 13 Distribution of lift Figure 14 Distribution of drag

Figure 15 PDF of lift on the elements Figure 16 PDF of drag on the elements
5.2 Exchange coefficients
 Williams and Singh. Structure of the turbulent flow over a rough bed 11

We have already seen that apparent quiescent fluid layer between the channel bed and mid-plane of
the spheres fully experiences the effects of the turbulent flow field in the fluid layer above it. In
many practical applications, it would be very useful to know the rate of mixing (or exchange) of this
retarded fluid below the mid-plane with the fast moving fluid above it. This would govern, for
example, the transport of fine grain sediments or pollutants on the floor of the channel. Based on the
DNS data, we can easily work out the rate of volume exchange and time scales involved. Let Vin and
Vout denote the influx and efflux of the fluid from the bottom layer. Let V be the volume of the fluid
contained below the mid-plane of the spheres. The volume and time exchange coefficients are given
by Vex  Vin / V and Tex  V / Vin respectively. Tex gives the time it would take for nearly complete
exchange of the volume V of the fluid in the bottom layer with the fast moving fluid above the
roughness elements. Table 3 gives the average values of these quantities. Let us note that the time
here is in the wall units (d/U*). A vigorous mixing is indicated by the values of the exchange
coefficients.

Table 3. Mean values of exchange coefficients (volume V = 1.36415916)

| Vin| Vout Vex Tex

Mean value 30.538 30.538 22.384 4.470e-02
Variance 0.0687 0.0687 0.0369 1.475e-06
Standard deviation 0.0829 0.0829 0.0607 1.214e-03
Maximum value 31.722 31.722 23.254 4.633e-03
Minimum value 29.447 29.447 21.586 4.300e-02

6. CONCLUDING REMARKS

We have presented a investigation of the structure of the fully rough turbulent flow over a rough bed
in an open channel consisting of a hexagonal arrangement of the spheres on its bed using DNS.
Comparison of the computed results, which correspond to ks+ = 102, shows a good agreement with
the available experimental data. Further, DNS results provide an accurate complement of the first
and second order statistics of the rough-bed flow specially near the roughness elements.
The DNS data has also provided us to detailed information on the flow within the roughness
elements (which is very difficult, if not impossible, to get from physical experiments). It is clearly
seen that even the highly retarded flow below the mid-plane of the roughness elements is fully
turbulent (barring a very thin layer close to the physical bed). Further, there is a large degree of
mixing between the fluid trapped below the mid-plane and that above it, as evidenced from the
estimates of the time and volume exchange coefficients. These coefficients would be of vital interest
in fine-grain sediment transport and dispersal of pollutants trapped on the bed.
We have also presented an investigation of the coherent structures. Streak spacing obtained
from the two-point correlation data compare favourably with experimental data of Defina (1996),
and points towards the existence of a hierarchy of vortices primarily aligned with the streamwise
direction. This observation is further strengthened by the instantaneous plots of the velocity and
pressure fields, plots of high and low-speed streaks and 2. Iso-surfaces of 2 clearly show the
presence of slender coherent structures having a streamwise extent of 100 to 200 wall units.
Investigation of the variation of drag and lift experienced by individual roughness elements provides
the pointers for the mode and mechanisms of the bed instability through coarse grain sediment
transport. It is seen that most of the elements which experience high lift are also subjected to high
 Williams and Singh. Structure of the turbulent flow over a rough bed 12

drag, whilst their downstream neighbours experience negative lift. This combination is highly
favourable for the lifting and rolling of these elements over their downstream neighbours, and
thereby disturbing the bed.

ACKNOWLEDGEMENTS
This work was carried out under the grant GR/M85241 from the UK Engineering and Physical
Research Council. Additional computing resources were provided by the UK Turbulence
Consortium from the HPC grant. Support of these organizations is gratefully acknowledged.

REFERENCES
Bhaganagar, K. and Kim J. (2002). “Physics of roughwall turbulent boundary layer”, Bull. Am.
Phys. Soc. Vol. 47, No. 10, page 55.
Choi, H., Moin, P. and Kim, J. (1991). “Direct Numerical Simulation of Turbulent Flow over a
Riblets”, Journal of Fluid Mechanics, Vol. 225, pp. 503-539.
Defina, A. (1996). “Transverse spacing of low-speed streaks in a channel flow over a rough bed”,
Coherent Flow Structures in Open Channels, ed. Ashworth, P.J., Bennett, S.J., Best, J.L. and
McLelland, S.J., John Wiley, London, pp. 87-99.
Grass, A.J. (1971). “Structural features of turbulent flow over smooth and rough boundaries”,
Journal of Fluid Mechanics, Vol. 50, pp. 233-255.
Grass, A.J. and Mansour-Tehrani, M. (1996). “Generalized scaling of coherent bursting structures
in the near-wall region of turbulent flow over smooth and rough boundaries”, Coherent Flow
Structures in Open Channels, ed. Ashworth, P.J., Bennett, S.J., Best, J.L. and McLelland, S.J.,
John Wiley, London, pp. 41-61.
Grass, A.J., Stuart, R.J. and Mansour-Tehrani, M. (1991). “Vortical structures and coherent motion
in turbulent flow over smooth and rough boundaries”, Phil. Trans. Roy. Soc. London, Vol.
A336, pp. 33-65.
Jeong, J. and Hussain, F. (1995). “On the identification of a vortex”, Journal of Fluid Mechanics,
Vol. 285, pp. 69-94.
Kim, J., Moin, P. and Moser, R. (1987). “Turbulence in a fully developed channel flow at low
Reynolds number”, Journal of Fluid Mechanics, Vol. 177, pp. 133-166.
Lam, K. and Banerjee, S. (1992). “On the condition of streaks formation in a bounded turbulenct
flow”, Physics of Fluids A, Vol. 4, pp. 306.
Lee, C. (2002). “Direct simulations of rough-wall channel flows”, Bull. Am. Phys. Soc., Vol. 47,
No. 10, page 111.
Leonardi, S., Orlandi, P., Djenidi, L. and Antonia, R.A. (2002). “DNS of a turbulent channel flow
with different rough walls”, Advances in Turbulence IX, ed. IP Castro, PE Hancock, TG
Thomas, pp. 387–90. Barcelona: CIMNE.
Lu, S.S. and Wilmarth, W.W. (1973). “Measurement of the structure of the Reynolds stress in a
turbulent boundary layer”, Journal of Fluid Mechanics, Vol. 60, pp. 481-511.
Moin, P. and Mahesh, K. (1998). “Direct numerical simulation: A tool in turbulence research”,
Annual Reviews of Fluid Mechanics, Vol. 30, pp. 539-578.
Nezu, I. (1977). “Turbulent structure in a open channel flow”, Ph. D. Thesis, Kyoto University,
Kyoto, Japan.
Orszag, S. and Patterson, G. S. (1972). “Numerical simulation of three-dimensional homogeneous
isotropic turbulence”, Physics Review Letters, Vol. 28, pp. 76-79.
Singh, K. M.. and Williams, J.J.R. (2003). “Direct Numerical Simulation of Flow Over a Rough
Bed”, Proc. International Symposium on Shallow Flows, Delft, Holland, Part III, pp. 45-50.
Sumitani, Y. and Kasagi, N. (1995). “Direct numerical simulation of turbulent transport with
uniform wall suction and injection”, AIAA Journal, Vol. 33, pp. 1220-1228.