SURFACE ROUGHNESS EFFECTS IN NEAR-BED TURBULENCE:

IMPLICATIONS TO SEDIMENT ENTRAINMENT

By A. N. Papanicolaou,1 P. Diplas,2 C. L. Dancey,3 and M. Balakrishnan4

ABSTRACT: In this study, the characteristics of near-bed turbulence were experimentally investigated for three
distinct roughness regimes, namely (1) isolated; (2) wake interference; and (3) skimming. Spherical particles of
the same size and density were placed upon a rough sediment bed to simulate the three regimes. Experimental
runs for the aforementioned regimes were performed in a tilting water-recirculating flume. Flow measurements
atop the spherical particles were performed by means of a 3D laser Doppler velocimeter. The aim of the tests
was to provide further evidence that the structure of turbulence is affected throughout the boundary layer by
the presence of roughness geometry. The measurements reported here include velocity profiles of the mean
streamwise and vertical velocity components and of the Reynolds shear stress distribution. To further quantify
the differences in turbulent structure under various surface roughnesses, a quadrant analysis was performed.

INTRODUCTION tics, culminating in the now familiar boundary layer theory.

Most of these investigations focused on turbulent character-
Turbulent boundary layers over roughness elements have istics over smooth surfaces, while fewer dealt with flows over
considerable engineering interest in several disciplines, e.g., rough elements (Nezu and Nakagawa 1993). It was probably
mechanical, aerospace, and hydraulic engineering. In sediment Schlichting (1936) who first studied the effects of various
transport knowledge of the effects of surface roughness on a roughness elements on flow characteristics using hemispheres
turbulent boundary is important for producing physically as roughness elements. Schlichting (1978) expanded his orig-
based models to predict sediment entrainment in natural wa- inal study by performing velocity measurements atop a matrix
terways. Such knowledge requires a good understanding of the of various immobile configurations of spheres with diameters
flow-bed interaction processes. of 4 mm. Several studies continued the work initiated by
Currently, empirical methods relating sediment flux with Schlichting. Andreopoulos and Bradshaw (1981) used sand-
bed shear stress are used to address the aforementioned prob- paper as roughness elements for their experiments, while Rau-
lem. Although these techniques often provide good predictions pach (1981) employed small vertical cylinders. Acharya and
for the range of flow and bed configurations for which they Escudier (1987) performed measurements over an expanded
were developed, they typically yield poor results when extrap- mesh roughness. Recently, Krogstad et al. (1992) used a wo-
olated to conditions other than those for which they were ini- ven mesh to generate their roughness. The findings reported
tially calibrated. Most of the empirical relations have been from the aforementioned studies clearly show that to charac-
developed for surface bed configurations that are densely terize the roughness simply by inspecting the effects it has on
packed (Fig. 1), which correspond to skimming flow regimes the mean velocity profile is inadequate. Various roughness ge-
(Montgomery and Buffington 1993). However, when these ometries may produce the same effect on the mean velocity
techniques are employed to bed geometries with less dense profile, but the turbulent characteristics (e.g., Reynolds stress,
surface roughness elements they may underpredict sediment microstructure of turbulence) may differ significantly (Krog-
entrainment. Low-density bed geometries are encountered in stad et al. 1992).
high-gradient streams with low relative submergences and/or The discovery of the so-called bursting phenomenon in tur-
pool-riffle sequences (Figs. 2 and 3). Streambeds with rela- bulent flows by Kline et al. (1967) [i.e., bursting phenomenon
tively low submergence correspond to isolated flow regimes, or interchangeably bursting process is defined in this paper as
while wake-interference regimes exist for pool-riffle sequences the metastable cycle of outward, ejections, inward, and sweeps
(Montgomery and Buffington 1993). To develop predictive (Robinson 1990 and Clifford et al. 1991)] generated a new
sediment transport models applicable to various bed surface
geometries, an understanding of the effect a rough surface has
on turbulent quantities is necessary.

BACKGROUND
The analysis of the turbulence properties of boundary layer
flows has been a vibrant area of research for almost a century.
Early pioneering work was directed toward the formulation of
time- and space-averaged representation of flow characteris-
1
Asst. Prof., Dept. of Civ. and Envir. Engrg., Albrook Hydr. Lab, Wash-
ington State Univ., Pullman, WA 99163-2910. E-mail: [email protected]
2
Prof., Dept. of Civ. and Envir. Engrg., Virginia Polytechnic Inst. and
State Univ., Blacksburg, VA 24061.
3
Assoc. Prof., Dept. of Mech. Engrg., Virginia Polytechnic Inst. and
State Univ., Blacksburg, VA.
4
Engr., Energy Applications, Columbia, MD 21045.
Note. Associate Editor: Alexander Cheng. Discussion open until Au-
gust 1, 2001. To extend the closing date one month, a written request
must be filed with the ASCE Manager of Journals. The manuscript for
this paper was submitted for review and possible publication on June 14,
1999; revised September 1, 2000. This paper is part of the Journal of
Engineering Mechanics, Vol. 127, No. 3, March, 2001. 䉷ASCE, ISSN
0733-9399/01/0003-0211–0218/$8.00 ⫹ $.50 per page. Paper No. 21199. FIG. 1. Isolated Roughness Regime

JOURNAL OF ENGINEERING MECHANICS / MARCH 2001 / 211

 Reynolds stresses in the longitudinal and vertical direction (or
interchangeably, the momentum fluxes in these directions)
␳u2 and ␳w 2, may be more important in sediment transport
than the shear stress component (-␳uw). This suggests that cal-
culations of sediment flux should not be based on shear stress
alone, as many of the equations predicting bed load, and/or
suspension rates generally do. Most of the preceding studies
have not focused on the effects that roughness has on the struc-
ture of near-bed turbulence since they used the same roughness
configuration throughout the tests.
The primary motivation of this study is to provide further
evidence that the structure of turbulence is affected throughout
the boundary layer by the presence of roughness geometry.
This is accomplished here by considering three roughness con-
figurations, corresponding to the isolated, wake interference,
and skimming flow regimes. Each of these configurations con-
sists of a matrix of immobile spheres equally spaced from each
other. Immobility is attained by using lead spheres, because
they are not entrainable by the flow conditions considered here
in addition to providing unimpeded acquisition of near-bed
flow measurements without any interference to the laser Dopp-
ler velocimeter (LDV) beams.
The emphasis is placed her on the description of the char-
acteristics of near-bed turbulence, for the three aforementioned
geometries, by recording detailed flow measurements very
close to the boundary. These measurements include velocity
profiles of the mean streamwise and vertical velocity compo-
nents and of the Reynolds shear stress distribution. To further
quantify the differences in turbulent structure under various
FIG. 2. Wake Interference Regime: (a) Wake Regime; (b) Pool- surface roughnesses, a quadrant analysis employing the
Riffle Sequence
method of Lu and Willmarth (1973) was performed.
Finally, the fixed-roughness tests reported here are coupled
with the entrainment tests, presented in Papanicolaou et al.
(1999) to provide some qualitative assessment of the turbulent
flow structures triggering sediment entrainment. Such coupling
is made possible because both tests were conducted under the
same flow conditions, initial-roughness configurations, and
roughness elements of identical geometry and surface rough-
ness. By coupling the results of the two studies, speculations
on how the structure of turbulence affects the entrainment of
sediment are provided.

EXPERIMENTAL ARRANGEMENT
The experiments were conducted in a tilting, water recir-
culating flume with a rectangular cross section and walls made
FIG. 3. Skimming Flow Regime of Plexiglas. The flume is 20.5 m long, 0.6 m wide, and 0.3
m deep. Its useful length is approximately 16 m. The test sec-
interest in further studying the structures of boundary layer tion, which is 3 m long and 0.4 m wide, was located 13 m
turbulence and then applying this new knowledge to the en- from the flume entrance, where fully developed turbulent flow
trainment problem. conditions were established during the experiments. Lead
In an attempt to link the characteristics of turbulent episodes spherical particles 8 mm in diameter, were placed atop a bed
with the entrainment of sediment, several researchers (e.g., of glass beads of identical size (8 mm in diameter) packed
Clever and Yates 1976; Sumer and Deigaard 1981; Grass four layers deep (with porosity of almost 26%) and distributed
1983; Dyer and Soulsby 1988; Rashidi et al. 1990; Lapointe uniformly along the flume bed. Detailed flow measurements
1992; Kirkbride 1994; Nino and Garcia 1996; Kaftori et al. were obtained by means of a 3D LDV. The LDV employed
1998) have suggested that the sweeps cause the initiation of here is a six-beam, nonorthogonal, color-separated, fringe-
bed-load motion in a stream bed, while the ejections are pri- mode, off-axis backscatter system. This particular nonintrusive
marily responsible for the particles’ suspended motion. Re- instrument uses three independent optical channels to measure
cently, a second school of thought (Nelson et al. 1995) has three nonorthogonal components of the velocity. Two of these
supported the opinion that the sweeps are not the only events components are approximately coplanar with a coupling angle
responsible for bed-load transport of gravel, but that the out- of approximately 30⬚. The third component is approximately
ward interactions are also responsible. Nelson et al. (1995) orthogonal to the other two. To improve optical access and
have clearly shown that when the magnitude of the outward facilitate near-bed measurements, the LDV system was tilted
interactions increases comparatively to the other events of a 4.8⬚ from the horizontal. The LDV measuring volume is
bursting phenomenon, the sediment flux increases too, al- roughly an ellipsoid about 0.08 mm in vertical and streamwise
though the magnitude of the Reynolds shear stress decreases. extent and 0.3 mm in cross-section extent. Average data rates
Along these lines, Sterk et al. (1998), Clifford et al. (1991), of about 20 measurements/s were obtained by seeding the flow
and Williams et al. (1989) have suggested that the normal with silicon carbide. An uncertainty analysis was performed to
212 / JOURNAL OF ENGINEERING MECHANICS / MARCH 2001
ensure the validity of the data (Balakrishnan 1998). This anal- two studies, conjectures of how the structure of turbulence
ysis suggests that only the spanwise velocity component is affects the entrainment of sediment are provided.
sensitive to errors, which in turn does not affect the results of For the 2 and 50% cases, detailed point velocity measure-
this study, since only the longitudinal and vertical components ments were carried out at a vertical distance of 0.13–0.19 mm
are used for the quadrant analysis. Similar observations have above the particle top surface and at a distance of 8–10 mm.
been reported in the literature by Cline and Deutsch (1993), For the 70% case, due to the presence of noise, measurements
Song et al. (1994), and Meyers (1985). Cline and Deutsch were obtained at a distance of 4.7 and 19 mm. Tables 1 and
suggested that the vibration of the laser probe volume is a 2 summarize the hydraulic conditions for the tests, namely, the
likely contributor to the elevated velocity fluctuation values in slope S of the flume; the depth H; the friction velocity u*; the
the spanwise direction. The mean velocity error in the stream- Reynolds number R = 4HV/␯, where V is the average depth
wise and vertical direction measurements is <0.8 and 2%, re- velocity, ␯ is the kinematic viscosity, the Reynolds shear stress
spectively. component ␳uw, and the local mean velocities and turbulent
Three packing configurations tests were performed: (1) The intensities in the streamwise and vertical directions, U , W, and
2% test, which represents the isolated flow regime; (2) the u⬘ and w⬘, respectively.
50%, which represents the wake interference regime; and (3)
the 70%, which corresponds to the skimming flow regime (Pa- RESULTS
panicolaou et al. 1999). For the 2% test, the spacing among
Methodology
the particles was almost 6 balls in diameter and a total of 530
particles were placed within the test section. In the 50% case, The backbone of the analysis performed here is based on the
the spacing was about 1 ball in diameter and 13,250 particles hypothesis that bursting events (i.e., defined as the cycle of
were employed throughout the run. Finally, for the 70% case, sweeps, ejections, inward, and outward interactions) are pri-
the particles were in contact with their neighboring particles marily responsible for the commencement of sediment motion
and about 18,600 particles were used. (Clifford et al. 1991). It is therefore, imperative to evaluate the
Each of these configurations consists of a matrix of im- effects that various roughness geometries have on the structure
mobile spheres. Immobility is attained by using lead spheres, of near-bed turbulence and use such knowledge in sediment
because they are not entrainable by the flow conditions con- transport. The methodology that is followed is unique, since it
sidered here. As clearly stated in the Introduction, the focus provides for the first time a combined analysis of the time-
of this investigation is to examine the near-bed flow charac- average flow characteristics (i.e., mean velocity profiles and
teristics for the three (immobile) roughness regimes. In order Reynolds shear stress component ␳uw) with the characteristics
to obtain near-bed flow measurements without any interference of a bursting cycle, i.e., the magnitude of the instantaneous
to the laser beams, lead particles were used. In short, use of stresses within a bursting cycle and the frequency of occurrence
the lead spheres as fixed roughness elements was dictated by of these events within that cycle. Approaches that do not ac-
the objectives of this study. count for frequency of occurrence and magnitude of turbulence
The lead spheres tests reported here are identical to other structures are rather incomplete (Jain 1992).
tests presented in Papanicolaou et al. (1999). These tests are To date, attention has been directed exclusively toward the
identical in terms of the flow conditions, initial roughness con- shear stress term ␳uw, although this term may not be the most
figurations, and the geometric characteristics of roughness el- relevant to sediment motion. The systematic exclusion of the
ements (the glass spheres have the same size and the same normal stress and other shear stress components from the sed-
surface roughness as the glass spheres). The focus of the two iment-flow interaction processes, implies that there is lack of
studies, however, is different. The focus of Papanicolaou et al. understanding of the effects that roughness has on turbulent
(1999) was on monitoring the entrainment of glass spheres via flow. In this study, the role of different instantaneous stress com-
an imaging software package. Reestablishment of these ex- ponents was evaluated by following a different approach. It was
perimental conditions is considered here in order to provide presumed that the particles do experience the action of the in-
some qualitative assessment on the turbulent flow structures stantaneous stress components (the particles do not distinguish
triggering sediment entrainment. By coupling the results of the the contributions corresponding to the mean and ‘‘apparent’’
stress components as we conveniently do for mathematical or
TABLE 1. Summary of Flow Parameters statistical expediency). This consideration is fully justified con-
sidering the fact that in sediment transport theory, the normal
Friction Reynolds and vertical drag force components, are defined as a function
Test case Slope S Depth H velocity u number
* of the instantaneous velocities U 2 and W 2, respectively (Naden
(percent) (percent) (m) (m/s) (R = 4HV/␯)
1987). Thus, the time series of the instantaneous normal stresses
(1) (2) (3) (4) (5)
U 2 and W 2 and of the shear stress UW (or equivalent momen-
2 0.2 0.05 0.0313 8 ⫻ 104 tum fluxes) were constructed for the three different packing con-
50 1 0.057 0.075 1.7 ⫻ 105
70 1.2 0.076 0.093 3.6 ⫻ 105
figuration tests while the quadrant analysis was employed to
determine the frequency of occurrence of each individual event,

TABLE 2. Flow Parameters per Measuring Point

Measuring
distance z above Streamwise mean Vertical mean Time average
Test case ball surface velocity U velocity W Intensity u⬘ Intensity w⬘ Reynolds stress
(percent) (mm) (m/s) (m/s) (m/s) (m/s) uw
(1) (2) (3) (4) (5) (6) (7)
2 0.19 0.2588 0.04914 0.07159 0.02235 0.00064
2 10.18 0.3369 0.02314 0.06357 0.03441 ⫺0.00019
50 0.13 0.2951 0.002454 0.1439 0.03271 ⫺0.00071
50 7.99 0.5407 ⫺0.01178 0.1278 0.06655 ⫺0.00271
70 4.7 0.3456 0.001528 0.1613 0.1048 ⫺0.00726
70 18.96 0.8812 0.005947 0.162 0.08291 ⫺0.00367

JOURNAL OF ENGINEERING MECHANICS / MARCH 2001 / 213

i.e., outward interactions, ejections, inward interactions, and
sweeps, within a bursting cycle. These results when coupled
with the visualization experiments of Papanicolaou et al. (1999)
can be used to identify the mechanisms associated to sediment
entrainment for the three roughness regimes.

Results
The steps that are involved in the postprocessing of the laser
measurements are summarized in Papanicolaou (1997). Figs.

FIG. 5. Reynolds Stress Distribution for: (a) 2%; (b) 50%; (c)
70% Roughness Configurations

4 and 5 present, respectively, the profiles of the mean velocity

components in the three directions and of the time-average
Reynolds stress. The mean velocities and the Reynolds shear
stress are normalized with the friction velocity u*. U ⫹ =
U/u* denotes the normalized mean velocity in the streamwise
direction, W ⫹ = W/u* is the normalized mean velocity in the
vertical direction, and V ⫹ = V/u* is the normalized velocity in
the spanwise direction. The normalized time-average Reynolds
FIG. 4. Velocity Profiles for: (a) 2%; (b) 50%; (c) 70% Rough- stress is defined as (uw)/u*. In Fig. 6(a)—the 2% case—U ⫹
ness Configurations varies within the range of 8–10, while W ⫹ and V ⫹ obtain
214 / JOURNAL OF ENGINEERING MECHANICS / MARCH 2001
 the ball there is a momentum flux exchange from the boundary
to the core of the flow. Term uw does not change sign, i.e., it
is always negative, for the rest of the packing cases implying
that the effects of secondary flow are significant for the iso-
lated flow regime only.
Time series plots in Figs. 6(a–c) provide the variation in
magnitude of the momentum (or instantaneous stress) terms
U 2, W 2, and UW. These measurements, which are on an av-
erage of 3,072 (or of duration of about 400 s)/measuring point,
are taken at close proximity to the top surface of a sediment
particle (at a distance of 0.2 mm above the top surface of a
sediment particle for the 2% and 50% cases and at a distance
of 5 mm for the 70% case). Figs. 6(a–c) reveal a striking
difference in magnitude among the three stress components.
Overall, U 2 obtains values that are at least 6 to 7 times higher
in magnitude than W 2 and UW. This is consistent with recent
experimental flow measurements related to coarse sediment
movement where it was shown that U 2 is a good predictor of
sediment (Clifford et al. 1991), while the ‘‘total momentum
flux’’ UW (as it is defined by Nelson et al. 1995) has a poor
correlation with sediment entrainment (Williams et al. 1989,
Nelson et al. 1995). The overwhelming importance of U 2 re-
instates the strong correlation that exists between the drag
force and the transport of coarse material (Sterk et al. 1998).
Moreover, Figs. 6(a–c) indicate differences in the turbulent
structure of flow, especially between the 2% and 70% cases.
In the 2% case, the time series plots demonstrate the presence
of relatively low-frequency events while higher-frequency
events are recorded for the 70% run. This is partially attributed
to the flow separation occurring in the 2% case. In this case,
the roughness elements (i.e., the 8 mm diameter spheres) pro-
vide sharp breaks in bed elevation, causing flow detachment
and slow return of the fluid parcel to the undisturbed boundary.
Similar flow boundary layer processes were discussed earlier
by Eaton and Johnston (1980).
The time series analysis performed here was complemented
with the construction of the joint frequency distributions of
u/u⬘ and w/w⬘ for the three roughness regimes. The quadrant
analysis (Lu and Willmarth 1973) was used to isolate the na-
ture of contributions to uw from the sweeps, ejections, inward,
and outward interactions. In the quadrant analysis four discrete
categories of momentum exchange are defined based on the
signs of u and w. To do so, a threshold criterion is usually
applied

冕
T
(uw)i (H ) 1
= lim uw (t)Si (t, H ) dt (i = 1, 2, 3, 4)
uw uw T→⬁ 0

where the subscript i refers to the ith quadrant and

Si (t, H ) = 1 if 兩uw(t)兩 > Huw
Si (t, H ) = 0 otherwise
By choosing different H values [H is defined as the hole size
(Lu and Willmarth 1973)], large contributions to Reynolds
stress from each quadrant can be extracted leaving the smaller
FIG. 6. Time Series Plots of U 2, W 2, and UW for: (a) 2%; (b) fluctuations in the ‘‘hole.’’ The selection of an appropriate
50%; (c) 70% Roughness Configurations threshold parameter H has been the source of considerable
ambiguity and inconsistency within published work. In this
study, H is chosen to be equal to 5 as suggested by Lu and
positive but small values, with W ⫹ being larger than V ⫹. The Willmarth (1973).
distributions of the nondimensional Reynolds stress are shown Figs. 7(a–c) were developed by plotting the normalized
in Figs. 5(a–c). The anticipated behavior for the Reynolds (u/u⬘)-(w/w⬘) pairs, contouring the density of the points, and
stress is that it will vary linearly with respect to z. Figs. 5(a– normalizing the results to peak values of 100%. The four quad-
c) reveal that the stress distribution very close to the bed is rants in the plots correspond to the four turbulent events (out-
complicated primarily due to the flow detachment around the ward interactions, ejections, inward interactions, and sweeps)
particle. Similar observations are shared by Steffler et al. that characterize the individual turbulent velocity measure-
(1985). It is also interesting to note that uw —in the 2% case ments. Figs. 7(a–c) reveal unique information about the tur-
—obtains positive values near the ball and negative values for bulence characteristics under various roughness configurations.
almost z /d > 1.4 [Fig. 5(a)]. This implies that very close to In the 2% case [Fig. 7(a)], the joint frequency distribution
JOURNAL OF ENGINEERING MECHANICS / MARCH 2001 / 215
 TABLE 3. Percentage of Time Occupied by Each Kind of Tur-
bulence Event

Test Distance z Inward Outward

case above ball Ejections inter- inter-
(percent) (mm) (bursts) Sweeps actions actions
(1) (2) (3) (4) (5) (6)
2 0.8 16.79 20.56 34.9 27.74
50 0.8 29.2 25.44 22.97 22.413
70 4 33.45 29.65 21.48 15.42

clearly demonstrates a positive correlation associated with the

bed shear stress at the measuring point and the tilting of the
joint frequency distribution into quadrants 1 and 3. Although
this is not the anticipated trend, it is not particularly surprising
if the nature of the flow for the 2% case is carefully considered
(such as fluid detachment occurring at the top of a particle
where a local acceleration zone occurs atop the particle) [Fig.
6(a)]. Moreover, Table 3 clearly illustrates that in the 2% case,
the inward and outward interactions, on the average, occupy
the highest percentage of time within a bursting cycle. The
foregoing finding seems to be in agreement with the recent
findings of Kaftori et al. (1998). They suggested that for the
flow regime in the presence of isolated roughness elements,
the joint frequency distribution among the quadrants in the
wall region changes dramatically. According to Kaftori et al.
(1998), the importance of the second and fourth quadrants di-
minishes as the roughness increases, while the contributions
of the first and third quadrants to the Reynolds stress become
more significant. Along the same lines Bennett and Best
(1995) illustrated, through measurements over bedforms, that
outward interactions events show an increased occurrence
within the separation zone both at and downstream from reat-
tachment. Visualization experiments conducted by the same
authors suggest that outward interaction events are restricted
mainly to near the bed. Instead, in the 50% case, the percent-
age of time that is occupied by each kind of turbulent event
is well balanced (Table 3). This is well demonstrated in Fig.
7(b), with the rather circular shape of the joint frequency dis-
tribution. This distribution does not have any pronounced
peaks or any preferential tilting toward one of the four quad-
rants and is rather symmetric with respect to the origin of the
u-w plane. Fig. 7(c)—70% case—depicts the anticipated neg-
ative correlation associated with the Reynolds stress and the
tilting of the distribution into quadrants 2 and 4. This trend is
typically encountered in flows over smooth boundaries (Nezu
and Nakagawa 1993) and it is fully justified here considering
the fact that fluid motion, for the 70% test occurs, over a well-
packed flat bed layer of identical spheres. Table 3 illustrates
that the percentage of time occupied from the ejections and
sweeps within a bursting cycle in the 70% case is higher than
that of the inward and outward interactions. Similar trends of
the results shown in Fig. 4(c) have been reported in the lit-
erature by Nezu and Nakagawa (1993) and Balakrishnan and
Dancey (1994) (for flows over smooth boundaries) and by
Nelson et al. (1995) for flows over roughness (sandy beds).

DISCUSSION

The information shown in Figs. 6(a–c) raise many questions

regarding the significance of the UW term in the entrainment
process. If the magnitude of the stress terms is indeed the
criterion that is used to decide which terms should be included
in the study of the entrainment problem, then the U 2 term
FIG. 7. Joint Frequency Distribution of Normalized u/uⴕ and
should be considered as the most relevant stress term (or the
w/wⴕ at First Measuring Point above Ball Surface for: (a) Isolated most relevant momentum flux term) to sediment motion (in-
Roughness Regime; (b) Wake Interference Regime; (c) Skimming stead of UW ). Significant peaks in the U 2 term provide cre-
Flow Regime (Events Corresponding to Quadrants Are Shown) dence to the idea that U 2 is probably the most important term
216 / JOURNAL OF ENGINEERING MECHANICS / MARCH 2001
 associated to incipient motion in gravel bed streams under the
presence of various roughness regimes.
Moreover, the joint frequency distributions constructed here
[Figs. 7(a–c)] clearly demonstrate that all events within a
bursting cycle (except the inward interactions) may contribute
to the sediment entrainment problem (depending on the exist-
ing roughness conditions). Therefore, in flows such as those
over the roughness elements examined here, the frequency of
the near-bed turbulence can vary significantly. Accordingly,
the percentages of the time during which sweeps, ejections,
outward, and inward interactions occur do not remain constant
but vary as the bed configuration changes. The conventional
thinking that ejections and sweeps are the most frequent events
appears to be true only for the skimming flow regime. In short,
Figs. 7(a–c) suggest that the ratio of the Reynolds stress to
standard deviation of the downstream velocity is not the same
for all roughness conditions. In the lower-packing density case,
this ratio is less comparatively to the other two cases implying
that empirical relations between stress and sediment transport
flux may underpredict transport in low-density packing cases.
Because Figs. 7(a–c) describe the joint frequency distribu-
tion of (u/u⬘)-(w/w⬘) at points found atop the particles (except
in the 70% case), further data analysis was considered. In this
case, flow data found at the most distant locations from the
top surface of a ball were examined (Table 1). The construc-
tion of the joint probability distributions of u/u⬘ and w/w⬘ for
the three roughness regimes, shown in Figs. 8(a–c), postulated
similar trends for the 70% case for the two points along the
vertical. Fig. 8(b) shows that the distribution is clearly tilted
toward the second and fourth quadrants (which is the antici-
pated trend). Finally, Fig. 8(a) illustrates that the distribution
of (u/u⬘)-(w/w⬘) for the isolated flow regime follows a rounder
shape than that of the wake and skimming flow regimes. The
rounder shape of the distribution in the lower-packing density
case is indicative of flow nonuniformity. By combining the
information shown in Figs. 6–8 one can claim that in higher-
packing densities the sweeps and ejections appear to be the
most dominant events while for flows over isolated roughness
elements nonuniformity is present with the outward interac-
tions becoming the most important.

CONCLUSIONS
The experimental evidence provided in this investigation in-
dicates that the various packing densities configuration en-
countered in natural gravel bed streams affect the turbulence
characteristics of the flow.
Based on the experimental findings presented in this paper
one can conclude:

1. The time-averaged characteristics of flow (i.e., Reynolds

stress and mean flow velocities) may be inadequate to
describe the effects of roughness on flow characteristics.
2. The Reynolds stress component is not the most relevant
component to sediment entrainment.
3. Flows over lower-packing densities may be envisaged as
extracting energy from turbulence to mean flow since
events described by quadrants 1 and 3 appear to be the
most dominant.
4. The quadrant analysis conducted here shows that the ra-
tio of the Reynolds stress to standard deviation of the
downstream velocity is smaller in the low-density cases
than the densely packed cases. Hence, empirical expres-
sions based solely on bed shear stress and sediment
transport relations may underpredict transport in low-
FIG. 8. Joint Frequency Distribution of Normalized u/uⴕ and
density packing cases.
w/wⴕ at Last Measuring Point above Ball Surface for: (a) Isolated 5. The joint frequency distributions of u/u⬘ and w/w⬘ (where
Roughness Regime, (b) Wake Interference Regime; (c) Skimming u⬘ and w⬘ are the turbulent intensities in the horizontal
Flow Regime (Events Corresponding to Quadrants Are Shown) and vertical directions, respectively) which were con-
JOURNAL OF ENGINEERING MECHANICS / MARCH 2001 / 217
 structed here, clearly demonstrate that the frequency of facing step.’’ Rep. MD-39, Dept. of Mech. Engrg., Stanford University,
occurrence and magnitude of turbulent events, under in- Stanford, Calif.
Grass, A. J. (1983). The influence of boundary layer turbulence on the
cipient flow conditions, varies significantly with bed mechanics of sediment transport, B. M. Sumer and A. Muller, eds.,
roughness. In particular, for the isolated flow regime the Balkema, Rotterdam, The Netherlands, 3–17.
outward and inward interactions are the most frequent Jain, S. C. (1992). ‘‘Note on lag in bed-load discharge.’’ J. Hydr. Engrg.,
events. For the skimming regime, the sweeps and ejec- ASCE, 118(6), 904–917.
tions are the most dominant, and for the weak interfer- Kaftori, D., Hestroni, G., and Banerjee, S. (1998). ‘‘The effect of particles
ence regime, all events appear to occupy the same per- on wall turbulence.’’ Int. J. Multiphase Flow, 24(3), 359–386.
Kirkbride, A. (1994). ‘‘Turbulence structure in straight gravel-bed chan-
centage of time. Additional analysis was provided by nels.’’ PhD dissertation, University of Sheffield.
developing the time series of the normal and shear stress Kline, S. J., Reynolds, W. C., Schraub, F. A., and Runstadler, P. W.
components U 2, W 2, and UW of the instantaneous stress (1967). ‘‘The structure of turbulent boundary layers.’’ J. Fluid Mech.,
tensor at a very close proximity atop a particle surface. Cambridge, U.K., 30, 741–773.
The results deduced here are consistent with the Sterk et Krogstad, P., Antonia, R., and Browne, L. (1992). ‘‘Comparison between
rough- and smooth-wall turbulent boundary layers.’’ J. Fluid Mech.,
al. (1998) and Nelson et al. (1995) measurements where Cambridge, U.K., 245, 599–617.
streamwise velocity was found to correlate well with sed- Lapointe, M. (1992). ‘‘Burst-like sediment suspension events in a sand
iment transport (for conditions well above critical). The bed river.’’ Earth Surf. Proc. Landforms, 17, 253–270.
findings, although quite academic at this stage of the in- Lu, S. S., and Willmarth, W. W. (1973). ‘‘Measurement of the structure
vestigation, should indeed be capable of extension to real of the Reynolds stress in a turbulent boundary layer.’’ J. Fluid Mech.,
cases to predict incipient motion of sediment in natural Cambridge, U.K., 60(Part III), 481–511.
Meyers, J. F. (1985). ‘‘The elusive third component.’’ Int. Symp. on Laser
streams under various hydrologic and bed roughness Anemometry, ASME, New York.
conditions. Montgomery, D., and Buffington, J. (1993). ‘‘Channel classification, pre-
diction of channel response, and assessment of channel condition.’’
ACKNOWLEDGMENTS Rep. THW-SH19-93-002.
Naden, P. (1987). ‘‘An erosion criterion for gravel-bed rivers.’’ Earth
This paper is based upon work supported by the U.S. Geological Sur- Surface Processes Landforms, 12, 83–93.
vey under Grant No. 14-08-0001-G2271. The help provided by Adam R. Nelson, J., Shreve, R. L., McLean, S. R., and Drake, T. G. (1995). ‘‘Role
Maxwell, Graduate Research Assistant at Washington State University, of near-bed turbulence structure in bed-load transport and bed-form
for the preparation of the plots included in this manuscript is gratefully mechanics.’’ Water Resour. Res., 31(8), 2071–2086.
acknowledged. Finally, the writers would like to thank the two anony- Nezu, I., and Nakagawa, H. (1993). Turbulence in open channel flow,
mous reviewers who provided many useful comments. Balkema, Rotterdam, The Netherlands.
Nino, Y., and Garcia, M. H. (1996). ‘‘Experiments on particle-turbulence
interactions in the near-wall region of an open channel flow: Implica-
APPENDIX. REFERENCES tions for sediment transport.’’ J. Fluid Mech., Cambridge, U.K., 326,
285–319.
Acharya, M., and Escudier, M. (1987). ‘‘Turbulent flow over mesh rough- Papanicolaou, A. (1997). ‘‘The role of turbulence on the initiation of
ness.’’ Turbulent shear flows, F. Durst, B. Launder, J. Lumley, F. sediment motion.’’ PhD dissertation, Virginia Polytechnic Institute and
Schmidt, and J. Whitelaw, eds., Vol. 5. State University, Blacksburg, Va.
Andreopoulos, J., and Bradshaw, P. (1981).‘‘Measurements of turbulence Papanicolaou, A. N., Diplas, P., Balakrishnan, M., and Dancey, C. L.
structure in the boundary layer on a rough surface.’’ Boundary-Layer (1999). ‘‘Computer vision technique for tracking bed-load movement.’’
Meteorol., 20:201–213. J. Comp. in Civ. Engrg., ASCE, 13(2), 71–79.
Balakrishnan, M. (1998). ‘‘The role of turbulence on the entrainment of Rashidi, M., Hestroni, G., and Banerjee, S. (1990). ‘‘Particle-turbulence
a single sphere and the effects of roughness on fluid-solid interaction.’’ interaction in a boundary layer.’’ Int. J. Multiphase Flow, 16, 935–949.
PhD dissertation, Virginia Polytechnic Institute and State University, Raupach, M. (1981). ‘‘Conditional statistics of Reynolds stress in rough-
Blacksburg, Va. wall turbulent boundary layers.’’ Appl. Mech. Rev., 44, 1.
Balakrishnan, M., and Dancey, C. L. (1994). ‘‘An investigation of tur- Robinson, S. K. (1990). ‘‘Coherent motions in the turbulent boundary
bulence in open channel flow via three-component laser Doppler ane- layer.’’ Annu. Rev. in Fluid Mech., 23, 601–639.
mometry.’’ Fundamentals and advancements in hydraulic measure- Schlichting, H. (1936). ‘‘Experimentelle untersuchung zum rauhigkeits
ments and experimentation, ASCE, New York, 159–175. problem.’’ Ingenieur-Archiv, 7(1), 1–34.
Bennett, S., and Best, J. (1995). ‘‘Mean flow and turbulence structure Schlichting, H. (1978). Boundary-layer theory, 7th Ed., McGraw-Hill,
over fixed, two-dimensional dunes: Implications for sediment transport New York, 656.
and bedform stability.’’ Sedimentology, 42, 491–513. Song, T., Graf, W. H., and Lemmin, U. (1994). ‘‘Uniform flow in open
Cleaver, J., and Yates, B. (1976). ‘‘The effect of reentrainment on particle channels with movable gravel bed.’’ J. Hydr. Res., Delft, The Nether-
deposition.’’ Chemical Engrg. Sci., 31, 983–992. lands, 32(6), 861–876.
Clifford, N. J., McClatchey, J., and French, J. R. (1991). ‘‘Measurements Steffler, P. M., Rajaratnam, N., and Peterson, A. W. (1985). ‘‘LDA mea-
of turbulence in the benthic boundary layer over a gravel bed and surements in open channel.’’ J. Hydr. Engrg., ASCE, 111(1), 119–130.
comparison between acoustic measurements and predictions of the bed- Sterk, G., Jacobs, A., and Van Boxel, J. (1998). ‘‘The effect of turbulent
load transport of marine gravels.’’ Sedimentology, 38, 161–171. flow structures on saltation sand transport in the atmospheric boundary
Cline, C., and Deutsch, S. (1993). ‘‘On elevated RMS level in wall- layer.’’ Earth Surface Processes and Landforms, 23, 877–887.
bounded turbulent flows when measured by laser Doppler velocimeter.’’ Sumer, B. M., and Deigaard, R. (1981). ‘‘Particle motions near the bottom
Experiments in Fluids, 15, 130–132. in turbulent flow in an open channel.’’ J. Fluid Mech., Cambridge,
Dyer, K., and Soulsby, R. L. (1988). ‘‘Sand transport on the continental U.K., 109, 311–338.
shelf.’’ A Rev. Fluid Mech., 210, 295–324. Williams, J., Thorne, P., and Heathershaw, A. (1989). ‘‘Comparisons be-
Eaton, J. K., and Johnston, J. P. (1980). ‘‘Turbulent flow reattachment: tween acoustic measurements and predictions of the bed-load transport
An experimental study of the flow and structure behind a backward- of marine gravels.’’ Sedimentology, 36, 973–979.

218 / JOURNAL OF ENGINEERING MECHANICS / MARCH 2001