J. Plwid Mech. (1971), vol. 48, part 4, pp.

721-761 72 1
Printed in Great Britain

The response of a turbulent boundary

layer to a step change in surface roughness
Part 1. Smooth to rough
By R. A. ANTONIA AND R. E. LUXTON
Department of Mechanical Engineering, University of Sydney

(Received 5 January 1970 and in revised form 30 November 1970)

The structure andgrowth of the internal boundary layer which forms downstream
of a sudden change from a smooth to a rough surface under zero pressure gradient
conditions has \been studied experimentally. To keep pressure disturbances
due t o the roughness change small, the level of the rough surface was depressed,
so that the crest of the roughness was aligned with the level of the smooth surface.
It has been found that, in the region near the change, the structure of the internal
layer is largely independent of that in the almost undisturbed outer layer, whilst
both the zero time delay and the moving axis integral length scales in the internal
layer are significantly reduced below those on the smooth wall. The growth-rate
of -:he internal layer is similar to that of the zero pressure gradient boundary
layer, whilst the level of turbulence inside the internal layer is high because of the
large turbulent energy production near the rough wall. From the mixing length
results, and an analysis of the turbulent energy equation, it is deduced that the
internal layer flow near the wall is not in energy equilibrium, and hence the
concept of inner layer similarity breaks lown. From an initially self-preserving
state on the smooth wall, the turbulent boundary layer approaches a second
self-preserving state on the rough wall well downstream of the roughness step.

1. Introduction
The changes which occur in a turbulent boundary layer following a sudden
perturbation from a self-preserving state have been the subject of several experi-
mental and theoretical investigations. The majority of this work has been re-
viewed by Tani (1968). Both perturbations applied at the wall (as discontinuities
in surface conditions) and perturbations applied to the free stream as step changes
in pressure gradient have been studied. Of the investigations of the f i s t type of
perturbation, most have been concerned with the response of a turbulent bound-
ary layer to a step change in surface roughness, perhaps because of the import-
ance of this problem in micrometeorology.
The existing theories and calculation methods for the flow downstream of a
change in roughness (e.g. Elliott 1958; Panofsky & Townsend 1964; Townsend
1965, 1966; Bradshaw, Ferriss & Atwell 1967; Taylor 1969) effectively require
that the thickness of the perturbed flow region, or internal layer, is small com-
pared with an overall length scale of the boundary layer. These theories are,
46 FLU 48
722 R.A . Antonia and R.E . Luxton
therefore, reasonably applicable in the atmospheric boundary layer for a limited
distance downstream from the step change. Measurements by Bradley (1965)
are typical of those in this region of the atmospheric boundary layer. There
have also been some wind-tunnel investigations of a turbulent shear flow follow-
ing a smooth-to-rough change (Jacobs 1939; Clauser 1956; Logan & Jones 1963;
Plate & Hidy 1967;Makita 1968; Antonia & Luxton 1971 a)and following arough-
to-smooth change in surface (Jacobs 1939; Taylor 1962; Makita 1968), but it is
unlikely that the above prediction methods would be applicable to these experi-
ments, as in all cases the internal layer represents a significant fraction of the total
boundary-layer thickness.
I n his review of the information provided by the above experiments, Tani
(1968) has concluded that the flow near the wall readjusts rapidly to the new
surface condition, but that the wall shear stress downstream from the stept
overshoots before slowly returning to its equilibrium value. Although this latter
conclusion appears to be qualitatively supported by the available measurements,
there is a need for accurate quantitative measurements of wall shear stress in
the region immediately downstream from the step. I n the experiments referenced
above, the wall shear stress was inferred either from extrapolation of measure-
ments of the Reynolds shear stress - UV, or from the slope of an assumed semi-
logarithmic velocity distribution. The accuracy of - UV measurements close to
a wall is usually not good, and as extrapolation involves effectively a differentia-
tion of the data, one cannot expect that an accurate skin-friction coefficient cf
could result from this technique. For the rough-to-smooth change, the Clauser
chart method for determining cf appears at first sight to be attractive, but an
independent measurement of wall shear stress is needed to establish the univer-
sality of the constants involved. Estimation of cr on a rough wall is a much more
difficult matter. Even for a self-preserving boundary layer (Perry, Schofield &
Joubert 1969), it is necessary to know the effective position of the wall and the
value of a roughness function, or ‘slip’ velocity AUjU, before the skin-friction
coefficient can be determined. I n the non-equilibrium flow close to a step change
in roughness, the basis of this technique becomes suspect, and in any case both
the effective origin and the roughness function are likely to be functions of stream-
wise position.
Many workers have advanced the existence of a logarithmic mean velocity
distribution in the region near the wall as evidence that the layer immediately
downstream of the surface change is an equilibrium layer (in the sense described
by Townsend 1961).
All the theories mentioned above effectively make this assumption, but no
direct experimental evidence of equilibrium is yet available.
Of the experimental studies referred to above, only those by Logan & Jones
(1963) in a pipe, and Makita (1968) in a channel, report measurements of the
turbulence intensities and the turbulence shear stress. These experiments may be
expected t o provide a reasonably complete picture of the readjustment of the
flow following the surface change, but this picture is complicated by the sudden
t For convenience, ‘step’ is used to mean ‘step change in roughness’ here and in other
places throughout the text.
 Turbutent boundary layer over a change in roughness 723
change in pressure gradient which accompanies the roughness change in fully
developed internal flows. To date, there has been no satisfactory experimental
study reported of the response of a zero pressure gradient turbulent boundary
layer to a change in surface roughness.
Antonia & Luxton ( 1 9 7 1 ~investigated
) the flow field downstream of an up-
standingt change in surface roughness. From a limited number of turbulence
measurements, it was deduced that in the region near the change the structure
of the internal layer was largely independent of that in the undisturbed outer
layer, whilst the integral length scales in the internal layer were significantly
reduced below those on the smooth wall. It was found, however, that the dis-
turbance introduced by the first roughness element had a significant effect on the
flow immediately downstream. I n order to minimize this disturbance, the level
of the rough surface has for the present experiments been depressed below that
of the smooth wall, so that the crest of the roughness is now aligned with the
level of the smooth surface. One of the aims of the present study is to verify the
above deductions for the new surface configuration. Particular attention is given
to the structure and growth of the internal layer downstream of the step and to
the mean flow velocity distributions ($6).
The variation of the wall shear stress in the region near the step is deduced
from the measurements of profile drag obtained by pressure tapping the roughness
elements. This variation is compared in $ 3 with that of the skin-friction coeffi-
cient inferred from the slope of the logarithmic mean velocity profiles. The con-
clusion drawn by Tani (1968), that the surface shear stress at first overshoots
and then gradually falls t o its final value, the whole readjustment t o the new
surface condition being quite rapid, is supported by the present wall shear stress
measurements.
The rate of growth of the internal layer is discussed in $ 6, whilst the distribu-
tions of turbulence intensity are presented in $7. These results show that the
edge of the internal layer grows a t a rate similar to that of the zero pressure
gradient boundary layer, whilst the level of the turbulence inside the internal
layer is high because of the large turbulent energy production near the rough
wall. From the mixing length results ($8) and the analyses of the turbulent energy
equation ($9),it is deduced that the internal layer flow near the wall is not in
energy equilibrium, and that inner layer similarity arguments are inapplicable.
It is seen ($ 10) that both the zero time delay and moving axis length scales are
significantly reduced as the flow enters the internal layer.
The approach of the turbulent boundary layer to a second self-preserving
state is verified by the results of $ 4 and $7. These turbulence measufements in the
self-preservingrough wall boundary layer are an essential starting point for the
study of the response of a boundary layer to a rough-to-smooth change in surface
(Antonia & Luxton 1969).
In this investigation, the roughness elements were upstanding from the level of the
upstream smooth surface, the faoe of the first element being totally exposed to the on-
ooming flow.

46-2
724 R.A . Antonia and R.E.Luxton

2. Experimental arrangement
The experimental boundary layer is formed on the floor of the working section
of a moderately low-turbulence wind tunnel. The tunnel, which is of the open
type, has a 9: 1contraction feeding a 15in. wide by 9 in. high working section. The
working section is 16ft long, and has a variable roof geometry to permit adjust-
ment of the pressure gradient. The experimental configurationbeing investigated
consists of a smooth floor 8ft long followed by a rough floor of similar length.
The roughness is of the ‘k-type’,? and is the same geometry as that used by
Moore (1951) and by Perry & Joubert (1963). It consists of rectangular slats of
iin. square cross-section and a pitch of & in. The first roughness element is de-
pressed below the smooth surface, the crest of the roughness being aligned with
the smooth wall surface (figure 1).The slats span the entire width of the tunnel

FIGURE
1. Geometry of surface and co-ordinate system.

and are faired into the corner fillets of the working section. The rough floor is
made up of four timber panels and a 1ft. long steel section which may be moved
to almost any streamwise position. This section provides an accurately defined
example of the roughness with elements located to within & 0.001 in. It can also
accommodate pressure-tapped elements for form drag measurements. The height
of the elements on the timber sections is accurate to about & 0-005in.
Measurements have been made at values of U,,the free-stream velocity, of
approximately 18ft-l see and 33ft-l see, with a zero streamwise pressure
gradient. With a $in. diameter tripping rod spanning the working section about
one foot downstream from the contraction, the boundary-layer thickness 8,
at the smooth-rough surface change is about 1.9 in. at the lower Reynolds number
U,&,/v 1: 1-9x lo4and 1-gin. at U,&,/v21 3.1 x lo4.
The majority of the mean velocity and turbulence intensity traverses were
carried out normal to the wall and in the centre plane of the tunnel. The geometric
centre of the measuring probe was usually located at the mid-point on the crest of
t A ‘h-type’ roughness is defined as one for which the roughness function AUlU,
(defined in $ 6 of the paper) scales on k, the physical height of the roughness.
 Turbulent boundary layer over a change in roughness 725

a roughness element but a few traverses were made in between roughness ele-
ments. The mean velocity profiles for both tunnel speeds were obtained with a
single hot wire (0.00012in. dia. tungsten, 0.040in. long) and a constant-tempera-
ture anemometer system designed by Fraser (1969). This was also used for
measuring the u-component turbulence intensity. The mean velocity profiles
at the higher tunnel speed were also obtained with a Pitot-static probe, used in
conjunction with a Texas Model 145 Precision Pressure Gauge. The v-component
turbulence intensity was obtained with a miniature DISA X-probe with 0.0002 in,
dia. platinum coated tungsten wires, 0-OQOin.long. The shear stress was measured
with this X-probe and with a single rotating inclined hot wire.
All signals were recorded in digital form after passing through sharp 1kHz
low-pass filters. The digital records were then processed on the English Electric
KDF9 computer in the Basser Computing Department of the University of
Sydney. Details of the digital data system may be found in Luxton, Swenson &
Chadwick (1967).
The form drag results presented in $ 3 were obtained with a pressure tapped
roughness element, which could be located at any desired position on the steel
roughness section. The vertical faces of the steel element were drilled at various
distances up tcr the height of the element. No pressure taps were inserted on the
roughness crest. The pressure holes were approximately 0.016 in. in diameter and
were spaced at regular intervals of 0.15in. in the transverse direction. The holes
were situated approximately on a straight line inclined at about 4" to the hori-
zontal, and spanned a distance of about 1.20 in. across the tunnel floor. These
sensing holes were connected to hypodermic tubes set into the base of the ele-
ment.
The pressures were measured with the Texas Precision Pressure Gauge Model
145, used in the null mode, and sensitive to 0.01 mm of water. Frequent checks
of these pressures were also made with a null reading Combist micromanometer
of rather greater sensitivity (about 0.003 mm of water). The agreement between
the two instruments was good but the Texas pressure transducer was found to be
easier to read.
The space-time correlation measurements presented in $ 10 were obtained with
the use of two single wires. The wires were made of 0-00012in. dia. tungsten
and were 0.040in. long. The probes carrying these wires were mounted on sepa-
rate traverse gears. The downstream traverse gear allowed accurate movements
in the x,y and x directions. The transverse gear used for the upstream probe allow-
ed accurate traversing in the y and z directions only. Most of the measurements
presented here are for nominally zero separation of the wires in both the y and z
directions. The wires used usually had fairly closely matched cold resistance
values as they were chosen from a batch of wires which had been plated during the
same process. The stings to which the wire ends were soldered were shaped to
minimize the interference t o the downstream stings. Also, the sin. dia. stem of
the downstream probe was a t a distance of about 60diameters from the stem
(also of sin. dia.) of the upstream probe at the position of zero 3 separation. This
position was in general determined by eye, and an accuracy of no better than
0.010 in. can be claimed for the x direction. The zero y separation was found to be
726 R.A . Antonia and R.E. Luxton
more critical as a small relative y displacement had a significant effect on the cor-
relation for small 1:separations. The zero y displacement setting was usually deter-
mined by comparing the oscilloscope traces of the wire signals. Although the
results presented in 5 10 were normalized with respect to the intensities at the two
wire positions, the wires were usually calibrated for each series of experiments,
and turbulence intensities could therefore be computed at a few points in the
boundary layer to check that the wires were operating satisfactorily.

b: +

FIGURE
2. Control volume used in determination of effective wall shear stress 7,.

and are likely to be in a state of energy equilibrium. In the present experiments,

close to the change from the smooth to the rough surface, the rough wall layer
is not fully developed and is unlikely to be in energy equilibrium. Accordingly,
the indirect ‘error in origin’ method, which presumes the existence of a logarith-
mic region and effectivelyforces the measured profile into a logarithmic form by
trial-and-error shifting of the origin, is unlikely to give reliable results. Two
direct methods are available: the floating element skin-friction balance (which
has been used successfully on smooth walls with small pressure gradients, and on
a large scale rough surface, in the micrometeorological studies by Bradley (1965)),
and the measurement of the form drag of individual roughness elements (as
used recently by Perry et al. (1969), and by Antonia & Luxton (1971 a ) ) .
I n the present work the form drag method has been used. As the theory of this
method has been given in Perry et al. (1969) only points of question will be raised
here. Referring t o figure 2, the method assumes (i)the combined contribution to
the momentum balance in the 5 direction due to the viscous shear stresses on
 Turbulent boundary layer ouer a change in roughness 727
AB, CD and EP is negligible; (ii)the momentum efflux out of FG is equal to the
momentum influx through HA; (iii) the integrals of the pressure distributions
over PG and HA are equal. The most questionable part of the first assumption is
that the shear stress on the crest of the roughness element is small.
Surface flow visualization studies by Schofield (1969) on a comparable rough
surface under fully developed conditions suggested the presence of a separation
bubble affecting the early part of the flow on the roughness crest. Further, it is
likely that in the cavities between elements reversed flow exists over the whole of
EP and part of AB. A s the shear stresses on these surfaces are also neglected it is
possible that the combined neglect of stresses on CD, EP and AB results in small
error. Neverthless, reliable, independent measurements of the shear stress on
the roughness crest would be useful. Assumptions (ii) and (iii) are difficult to
test experimentally with roughness of the present scale as the introduction of a
probe into the cavities between the elements is likely to disturb the flow in the
cavities. However, the assumptions appear to be reasonable at least for elements
not too close to the start of the roughness. A small streamwise pressure gradient
could be detectednear the crests of the first two or three roughnesselements and it
is likely that this also existed between the elements.
With the above assumptions the 2-momentum balance reduces to

THCA = (P1-P.) dy', (1)

Jok
where h is the pitch of the roughness, Pland P2are the pressures on the front and
rear faces of the roughness element, k is the height of the element and y' is mea-
sured from the base of the element. It must be noted that i-I;rG is a measure of the
momentum transport across the surface HG, and as such i t has contributions
from (--Z),(vaU/ay) and (- U V ) , where U and V are the values of the time
mean streamwise and vertical velocities, and ( ) denotes the average over one
pitch length A. Thus, if there is a periodic mean flow field induced by the periodic
roughness, and there is a phase difference between U and V , there will be a con-
tribution to THG due to mean flow momentum transport. The importance of
this term cannot be assessed for the present but measurements of UV presented
in Q 7 suggest that it may be significant if HC is taken close to the crest of the rough-
ness.
At the lower of the two Reynolds numbers the pressure difference (Pl-P2)
across a roughness element was small and consequently the accuracy of measure-
ment was poor. For this reason, most of the measurements presented in § 3 were
made at UISs/v21 3.1 x lo4. The reproducibility of these profiles was good, but
even at the higher Reynolds number the minimum difference recorded (approxi-
mately half way up the element) was only about 0.1 mm of water, and a resolution
error of up t o 13 % is possible for this value. The pressures Pland P2were measured
relative to the pressure at a hole near the crest of the trailing face of the element.
Very little variation was found for the distribution of Pz, and for this reason only
the pressure difference (Pl-P2) is shown in figure 3 plotted against y'/k. It is
seen in figure 3 that the pressure distribution rises near the crest and near the
base of the roughness, and has a broad minimum in the region of y'lk = 0.5.
728 R. A . Antonia and R. E . Luxton
The distribution is not symmetrical about the mid-height, as the pressure rise
near the crest is greater than that near the base. This asymmetry persisted to the
last measuring station at x = 46.5 in. (figure 4), and was evident at both Reynolds
numbers. It must also be remarked that the spatial resolution of the pressures
plotted in figures 3 and 4 is not good, owing to the unavoidably large diameter
of the measuring holes in comparison with the element height. I n particular, the
shape of the distribution near the crest and near the base of the element is poorly
defined. It may be noted that no evidence of a stagnation line on the leading face
of an element may be seen, nor was there any evidence of stagnation in the flow-
visualization experiments of Liu, Kline & Johnston (1966) and of Schofield
(1969), both for comparable roughness geometries.

1.o

0.8

0.2

(P,- P*)/+U;x 102

3. Pressure difference profiles in the region near the step. U,S,/v
FIGURE 2: 3.1 x lo4.
Note the shift in origin.

Effective wall shear stress values derived from the pressure distributions of
figures 3 and 4 through ( l ) , assuming the effective T~ to be equal to T ~ are~ ~ ,
plotted in figure 5. For comparison, skin friction deduced from the ' error in origin'
method is also shown (assuming the error in origin E = 0.08in. or elk = 0.64 at
all stations, a value which seems t o give an acceptable, though by no means
unique, logarithmic region even at x = 2in., as shown in figure 6), but, as has
already been stated, the basis of this method is questionable at least near the
start of the roughness. Close to the start of the roughness, the form drag results
may also be in error due to the doubt expressed concerning assumptions (ii) and
(iii) used in the derivation of equation (1).GeneralIy there may also be an error
due to uncertainty in the shape of the pressure distributions near the crest and
near the base of the elements, and to the errors involved in measurement of small
pressure differences. The wall shear stress, derived from the momentum thick-
ness distribution shown later in figure 8, is, however, in better agreement with
 Turbulent boundary layer over a change in roughness 729
the trend and magnitude of the form drag results than with those obtained from
the error in origin method.
All methods used give a distribution of cf which is in qualitative agreement with

r
a
0.8 I I Dsensing
i a . of hole 0 0
00

*
0.6 1 o n
o n
2 00
0.4
aa
0.2
0 0
0

0
0 1.o 2.0 3.0 4.0 5.0 6.0
( P I - P a ) / t U ;x 1 0 2
FIGURE
4. Pressure differenceprofiles at x = 46.5in.from step.
0, U,SJV N 3-1 x 104; n, 1.9 x 104.

't

Smooth wall
(x= - 4 in.)
I I I I I I
2 4 6 8 10 12
x (in.)
FIGURE 5. Variation of effective wall shear streas near the step. U,S,/v N 3.1 x lo4.
0,determined from form drag of roughnesselements; + ,determined from ' error in origin'
method assuming B = 0.08 in. at all stations.
730 R.A . Antonia and R.E.Luxton
the prediction methods referred to in $ 1 , even though these methods are probably
not strictly applicable to the present case. There is a sudden rise in c, at the start
of the roughness, followed by a fairly rapid fall towards the value applicable to
the fully rough layer. Thus, the skin friction appears to adjust rapidly, within
3 or 4 boundary-layer thicknesses, to the new rough wall boundary condition.

1.0 -

0.8 -

0.6 -
s"
5
0.4 -

0.2 -
01 I 1 1.
0.01 0.1 1.0 5.0
y + E (in.)
6. Example of ' error in origin' method of obtaining a logarithmic profile
FIGURE
a t z = sin. U18& 2: 1.9 x 104.

4. Self-preservation of mean velocity field

Mean velocity profiles at four stations on the rough wall are plotted in figure 7
in the form (Ul- U ) / U ,as a function of (y +€)/A, where A = S*Ul/U, and e, the
error in origin, is assumed to be constant and equal to 0.08in. The distributions
at x = 35in. and x = 45in. fall, within experimental scatter, on the same curve
over almost the complete range of (y + €)/A. Since the mean velocity defect at
x = 20in. exhibits only a slight departure from this curve for the larger values
of (y+e)/A, it is reasonable to assume that self-preservation of the mean flow
field is attained by x = 35 in. The profile defect parameter,

is found to be equal to about 6.8 for the mean velocity distribution at x = 45 in.
Values of Greported in the literature for the universal velocity profile on a smooth
or rough wall in a zero pressure gradient usually lie between 6.0 and 7.0.
The shape parameter H = P / O ,plotted in figure 8, increases from a value of
1.4 on the smooth wall to a value of about 1.8 at x 2: 25 in., and it maintains this
value up to x 2: 60 in. At larger values of x a slow decrease in H is observed.
 Turbulent boundary layer over a change in roughness 731
The displacement thickness 6" and the momentum thickness 8 increase
almost linearly with x, except very close to the surface change (figure 8). For
the range of x over which H is approximately constant, the streamwise variation

n $
n o
nv+
Ova

+ g : c1
n
:v 0

0 I I p X I n V " p 0

0 10 20 30 40
(y +€)/A x 10*
FIGURE 7. Mean velocity defect profles on the rough wall assuming 6 = 0.08in.
for each profile. U,6,/v N 1.9 x 1
0. 0 ,
' ~ v, 20; +, 35; 0,
= loin.; 45.

4 0.9

- 0.8
'ca
- 0.7

- 0.6
- 0.5
- 0.4

20
I I
30 40
I I
50
I
60
I
70 sbo3
x (in.)
FIGURE
8. Distributions of a* (0,0 ) ;6(13, m);H = 8*/6(A, A).
0,0,A,-, u,s,jv- i ~ ~ x 1 0 4 . ~ , ~ , ~ , - - - , ~ , ~ , / ~ ~ 3 ~ 1 x 1 0 4 .
732 R.A . Antonia and R. E . Luxton
of 0 implies a constant value of cf of about 0.0084 for both Reynolds numbers
investigated. From the above definition of G , a relation between H and G can
be written H = (1- G(cf/2)*)-l. (3)
For cf = 0.0084 and G = 6.8,H equals 1.79 which is in good agreement with the
experimental values of figure 8.

5. Method of plotting mean profiles

I n the study of the response of a turbulent boundary layer to an upstanding
step-change in surface roughness (Antonia & Luxton 1971 a),it was observed that
the mean velocity profiles inside the internal layer in the region near the step
exhibited a linear trend when plotted in the form U vs.y4. It was also found that
the mean velocity distribution outside the internal layer could conveniently be
plotted in the same co-ordinates, exhibiting a different linear trend except very
near the outer edge of the layer. The intersection of the two straight lines was
shown to be closely related to the edge of the internal layer. A weak dimensional
argument was given to support the half-power plotting scheme, but this required
that the streamwise gradient of wall shear stress d~,,/dx immediately downstream
of the step be included in the argument in place of the wall shear stress T,,.
Difficulties in making adequate measurements in that particular flow, occasioned
mainly by the presence of a separation bubble with unhappy stability charac-
teristics behind the upstanding first element, and the problem of measuring
7w2,left the dimensional argument with neither direct support nor denial.
I n the present experimental configuration, no first element separation bubble
was present, and hence quite detailed measurements could be made, though
some doubt must remain about the accuracy of the effective wall shear stress.
Nevertheless, the variation of cf shown in figure 5 hardly suggests that d7,,/dz
is a dominant parameter close to the start of the roughness. The local shear
stress gradient la~/ayl= a is, however, likely to be large in the region near the
step, as the magnitude of T,, is certainly large compared to that of 7wlon the
smooth wall immediately upstream of the roughness, and the shear stress in the
internal layer must vary between T,, at the surface and a value of the same order
as 7,1 near the edge of the internal layer. It is probable, then, that ~?7/ayhas a
strong influence on the velocity profile in the internal layer. This profile will also
be influencedby the distance from the surface y, thewall stress 7wzand somelength
scale associated with the roughness height k. Viscosity is not likely to be signifi-
cant, as the flow is fully turbulent at least down to the crests of the roughness
elements. Thus, we write u = ~ ; r (a,~7wz,
, k). (4)
If it is now assumed that the relative motions inside the internal layer are
mainly determined by a, i.e.
a u / a y = W Y ,a)/ay,
dimensional considerations then yield
 Turbulent boundary layer over a change in roughness 733

1.o in.

0.8

0 45
0.6

0.4
d
1
R
0.2

x= - 4 4 in.0
1 0
2 0 0
4 0
0
6 0
8.5 0
I I 1 1 1 1 1 1 1
45 0
0.2 0.4 0.6 0.8 I *o 1a 2 1.4 1.6 1a 8

y4 (in.)*
(a)

1.0 - x=l in.

2.5
4
0.8 - e 6
a 0 85
11

06 -

.-
5
0.4

0.2 -

0-
0-

:
0
0 0 0.4 0.8 1.2 1.6

yQ (in.)*
(a)
9. Mean velocity profiles plotted as a function of y*. ( a )U18,/v N 1.9 x lo4,
FIGURE
(b)U18,/v N 3-1 x 10'. Note shift in origin.
734 R.A . Antonia and R.E . Luxton
If it is further assumed that a is independent of y, then integration of ( 5 a )
results in a half-power velocity profile, with the integration constant being
presumably a function of ak/U$ Examples of mean profiles plotted in this form
are shown in figures 9(a) and (a). This last assumption, that a = la~/ay(is
independent of y, is supported by the distributions of the shear stress -uV in
the region near the step, presented in figure 12 (b) and discussed in $ 7.For present
purposes, it is sufficient to note that, within experimental accuracy, reasonable
straight line fits are obtained for the majority of the UV profiles in the internal
layer. As x increases, there is a tendency for -G to decrease as the wall is ap-
proached. From the slopes of the straight lines as shown in figure 12 (b), it is seen
that l8~/8yldecreases as x increases. From the half-power mean velocity profile
plots, figures 9 (a)and (b),it is evident that aU/ay!i also decreases as x increases
lending a t least qualitative support for the form of (5a).The main argument
in favour of the half-power method of plotting mean profiles must remain, how-
ever, the convenience of being able to estimate the position of the edge of the
internal layer from the 'knee ' in the velocity profile plots.

6. Growth of the internal layer

The approximate position of the edge of the internal layer has been estimated
in two ways. As mentioned in 9 5, it was shown in Antonia & Luxton (1971a) that
the 'knee' point, the intersection of the straight lines on the half-power plot
U us. yg, could conveniently be used to define the edge of the internal layer. The
position of the knee point deduced from the half-power plots presented in $5, is
shown in figure 10 for various values of x. An adequate fit to the results is given
by 8, cc x O " ~corresponding
, to a faster rate of growth of theinternal layer than for
the case of an upstanding roughness change?.
A physically more realistic determination of Si may be obtained by inferring
the ' merge ' point (or the position of merging) between consecutive mean velocity
profiles. The results presented in $2 show that this point also closely coincides
with the merging of the turbulence intensity profiles. I n the present study, the
position of the merge point was inferred by superposing mean velocity profiles
(plotted on either a linear or a half-power scale)obtained at successive streamwise
stations. The values of 8, obtained (figure 10) are slightly higher than those de-
termined by the knee point, whilst the rate of growth of 8,is now represented by
8, cc X O ' ' ~ , which is reminiscent of the growth rate of a turbulent boundary layer
developing in a uniform free stream. Here, the effect of a small streamline dis-
placement is included in the determination of 8i.
The experimental values of St, estimated from the merge point, are compared in
figure 11with the theoretical predictions of Elliott (1958)and of Townsend (1965)
in the region near the step. Although the theory of Elliott, and in particular that
of Townsend, is not strictly valid for small values of x/zo2,it is seen that the experi-
t Note, however, that because of the initial thickening of the internal layer produced
by the separation bubble from the fist element in the upstanding roughness case, the
physical thickness WES greater than in the present case even though the growth rate was
less.
 Turbulent boundary iayer over a change in roughness 735
mental rate of growth of Si is in reasonable agreement with the theories; but the
experimental values of Si/zo2are significantly smaller than the predicted values.
To compute these values, a constant value ofthe equivalent roughness length
scale zo2was assumed on the rough wall. This value is obtained by extrapolating

0.1 I I J
I .5 5 10
x (in.)
FIGURE A,estimates from profile
10. Growth-rate estimates for the internal layer. 0,
'merge'points. A,estimatesfromthe'knee'points. O,U,&,/vz 3 . 1 ~ 1 0A,
~ ; A,1 . 9 ~ 1 0 ~ .

the (assumed)logarithmic distribution to zero velocity. The assumed distribution

for a fully rough wall layer is

where K is the K&rm&nconstant, y" is the effective distance from the wall, A is
a constant taken here to be 4.9, and AU/U, is the roughness function (Clauser
1956) given by

For the present experiments, the best fit line gives D = 1.8. Using these expres-
sions we obtain
zo2 = kexp [ K ( D - A ) ] , (8)
736 R.A . Antonia and R.E . Luxton
which gives, for the stated values of A and D, xo2 = k13.56. The roughness step
M is defined by M = In zo2/xo1,where zol is the equivalent roughness length scale
on the smooth wall; and, for the present case, M is about - 4-6.
I n considering the above comparison, we must anticipate a conclusion drawn
in 5 8, that the assumption L, = qy breaks down in the inner layer similarity is not
applicable. As both the theoretical methods require such similarity, it is not
surprising that the predictions are poor for the present case.

Townsend (1965)- /

I I
I 100 500

xlzoz

FIGURE
11. Comparison of experimental growth of internal layer with theories of Town-
send (1966) and Elliott (1958).U1S$ II 1.9 x lo4, zo2 2: k/3.56, M = lnz,,/z,, N -4.6.

7. Discussion of turbulent intensity data

Figure 12 (a) shows distributions of J."/Ul and JUTS,the normalized root-
mean-square values of the turbulence intensities in the longitudinal and vertical
directions respectively. Figure 12 ( b ) shows the distribution of - SUVlUf, the
normalized Reynolds shear stress, immediately downstream of the roughness
step at the Reynolds number of UIS,/v2: 1.9 x lo4. As all the quantities shown in
figure 12 were obtained with an X-probe, measurements very near the wall were
not possible. For this reason, a single wire was used to obtain the shapes of the
u-component distributions in the region near the wall. These distributions,
 Turbulent boundary layer over a change in roughness 737
which are plotted on a semi-logbasis in figure 13,are in reasonable agreement with
those of figure 12 ( a )for y greater than 0.1 in. They show that near the wall (y less
than0.1 in.), ?is first increasedrelative to its smoothwall distribution (x= - 4in.)
in accord with the sudden increase in the average velocity gradient aU/ay and
the accompanying increase in the turbulent energy production raU/ay. As x
increases, however, aU/ay and r aUjay both decrease in this region (this appears
to be consistent with the decreasing trend of the skin-friction results, shown
in figure 5), and 2 is correspondingly reduced. Away from the wall (y greater
than 0.1 in.), 2 increases with x in accord with the streamwise increase in aU/ay
over the outer part of the internal layer, and the internal layer extends progres-
sively further into the outer region of the original smooth wall boundary
layer.
The distributions of 3 and - UV show the same trend? as that of t h e 2 profiles.
It is observed that, within the experimental accuracy, all of the turbulence in-
tensity profiles over the initial part of the rough wall merge into the corresponding
distributions on the smooth wall at values of y which are in close agreement with
the edge of the internal layer as inferred from the ‘merge ’ point of the mean velo-
-
city profiles (see 5 6). I n figure 12(b) the - uv distribution on the smooth wall
(z = - 4.4 in.), which was obtained with a DISA X-probe, extrapolates to a wall
value which corresponds to a skin-friction coefficient of about 0.0026. This value
is significantly lower than the Clauser chart value of 0.0034$ deduced from a mean
velocity profile at this station. As a check, a complete set of -zLv distributions
were obtained with a rotating wire. At x = - 2-5in., a wall value of about 0.0032 is
suggested from three measurements in better agreement with the Clauser chart
value. The shape of the smooth wall UV distribution obtained with the X-wire
is compatible with a constant shear stress in the region near the wall, but its
magnitude in this region is lower than that obtained with the rotating probe $. I n
the outer part of the smooth wall boundary layer, the two distributions are in
t This trend is in agreement with that indicated by the turbulence measurements down-
stream of an upstanding roughness change (Antonia & Luxton 1971a).
$ It should be noted that independent checks of this value were obtained with a Preston
tube, and by application of the momentum integral method.
$ It was found that, although 2 and 7 distributions on the smooth wall agreed well
with those obtained in self-preserving smooth wall boundary layers by other workers, the
-
uv distribution was generally lower than expected. Careful testing of a range of probe
designs, including commercial probes, indicated that in all but one case the measured
value of UV was low at the Reynolds number of the tests reported in this paper. The shapes
of the distributions were not distorted. At higher Reynolds numbers, the distributions
all agreed with those of Klebanoff (1955). The probe that gave the ‘correct’ distribution
at the lower Reynolds number (i.e. a distribution which extrapolated to the wall shear
value as determined from a Clauser chart) was of the single inclined rotating-wire type,
and this probe was also ‘correct’ at the higher Reynolds number. It was concluded that
the low values of UV a t the lower Reynolds number did not indicate a lack of full boundary-
layer development, but did indicate that X-wire configurations may be Reynolds number
dependent a t low Reynolds numbers. The results presented here are those obtained with an
X -wire configuration, and were all obtained a t the lower Reynolds number. Accordingly,
the UV distributions lie below those of Klebanoff, but, as the shapes of the distributions
are not affected, and the results are internally consistent, the conclusions drawn from them
are not seriously in question.
47 F L M 48
738 R.A . Antonia and R.E . Luxton
reasonable agreement with each other. Over the roughness, the magnitude of the
X-wire distribution of - UV in the internal layer is lower than that of the rotating
wire distribution a t the same station, but, apart from the scatter in the experimen-
tal points, these distributions are similar in shape. The fact that the shapes of

14 -

co
12 - %
* *+
2 0AD
X
4 0
d
1% 10 - 0

OA
0

VO
t

0
6 A
5
X 0

.
5
‘3
A

4 s B 0
k
A
Y

2 6 +
%
+b
I I I I I
Q

the distributions are similar will be seen to be particularly important in discus-

sion of the - UV distributions near the rough wall. It is difficult to reconcile the
absolute values of G in the region near the wall with the skin-friction results pre-
sented in $3. The straight-line approximations t o the UV distributions obtained
with the rotating wire extrapolate to wall values which are smaller than those
deduced from the form drag measurements, but there does not appear to be any
irrefutable physical reason to expect such an extrapolation to reach the effective
wall shear stress value.
 Turbulent boundary layer over a change i n roughness 739
The shear stress distributions at x = Gin. andx = loin. (figure12 ( b ) )show that
-uV tends to decrease as the wall is approached. This trend becomes more
pronounced at larger values of x (figure 14(b)). Figures 14 show the distributions
6

02

E:
X

F3
is
c.l
I

Q
1

FIGURE 12. Turbulence intensities and shear stress at stations near to the step change in
roughness, UIS,/uN 1.9 x 10'. (a)u- and u-component intensities. Unflagged symbols refer
to @/U1 and flagged symbols to J'/Ul. ( b )Shear stress - h i i / U : obtained with an X-wire.
The straight-line fits to distributions near the wall support the half-power data correlation
scheme for mean velocity profiles.
V o a o + e
x = -44.4in. 1 2 4 6 10

of Ju"/U,, J7/V, and - 2UVlUf plotted against y/6*, where 6" is the displacement
thickness of the boundary layer for larger values of x.At x = 35 in., the distribu-
tions of u'i andT2 become approximately similar in shape over a large fraction
of the boundary-layer thickness tending to suggest that the self-preservation of
the normal stresses is attained reasonably rapidly. The distributions of 2 and 3
on the smoothwall are also plotted in figure 14 (a)for comparison with those on the
rough wall. Fromfigures 13and 14 (a),afterrescalingwith 6, it may be seen that the
magnitudes of 3 and vTin the outer part of the rough wall boundary layer are
significantly higher than those on the smooth wall. It is reasonable to infer that
47-2
740 R. A . Antonia and R. E. Luxton
- - -
the integrated turbulent energy 4 I? dy where = u2+ v2 + w2,is higher on the
rough wall, and, as a large proportion of this energy occurs well away from
the wall, one might anticipate that it is associated with the larger length scales.
Near the roughness, and well downstream of the change, 2 and 3 increase with
distance from the wall, the increase in 3 being slower; but over most of the inner
layer (y/S < 0.15))2 and zv may be assumed to remain approximately constant.

Y (in.)
FIGURE
13. Streamwise component of turbulence intensity obtained with a single wire,
U,S,/v N 1-9x lo4. A,z= -4411.; 0,1; v, 2; 0,4;0 , 10; +,
45; B, 52.

The shapes of the - Gi distributions in the outer part of the rough wall boundary
layer also appear to be reasonably similar, tending t o confirm the above claim of
approximate self-preservation. I n the region correspondingto the inner layer, the
shear stress decreases as the wall is approached. At x = 46-5in., for example, the
decrease starts from g/S 21 0.25, where - 2 G / U ? 21 0.0052, this value being con-
siderably lower than the skin-friction coefficient value of 0.0084 obtained by
-
applying the momentum-integral method at this station. A similar trend of - uv
near the wall is also evident in the limited experimental data of Chanda (1958)
and Makita (1968). Chanda reports two distributions of UV measured at the same
station on a rough surface, which consisted of crushed stone ranging from 0.25 in.
to 0.315in. in size. The boundary-layer thickness was about 6.5in. These dis-
tributions exhibit a peak at y/S -N 0.25, the value of - 2 Z / U ? at this peak being
 Turbulent boundary layer ouer a change in roughness 741

A
0
+
00
A

6-
0
0
0 00
5 -
0
A
0' A A 00

4 - A
01
0 A
2 = A
X 0

.
5
4
3 - 0
0
A
A

E.l
0
I
2 O A
-0
0

0
0

0' I I I I
0 1 2 3 4
Y/S*
FIGURE14. Turbulence intensities and shear stress at stations well downstream from the
step change in roughness, U,S,/v N 1.9 x 104. ( a )u- and v-componentintensities. Unflagged
symbols refer to @/U1 and flagged symbolsto JvZ/Ul. Smooth wall results are included for
comparison. ( b ) Shear stress - 2uV/U: obtained with an X-wire. Values of c, obtained from
the momentum integral equation are 5 = 20in., of = 0.0086;46.6,0.0084;60,0.0084.
z = -4*4in., 6* = 0.31, 8 = 1.9
0 20 0.64 2.4
+ 36 0.67 2.7
0 46.5 0.77 2.9
A 50 0.81 3.0
742 R. A. Antonia and R. E. Luxton
approximately 27% lower than the value of cf given by 2 dO/dx.Makita’s investiga-
tion was mainly concerned with the experimental study of the flow field down-
stream of arough-to-smooth step change in surfaceroughnessin a two-dimensional
channel. The roughness geometry used was somewhat similar to that being
studied here except that the ratio Alk was equal to about 6.7. The distribution
of -UV on the roughness element just ahead of the smooth surface reveals a peak
at yld N 0.20 (d is the half-height of the channel), the value of - 2uVlU; at this
peak being approximately 40% lower than the wall value obtained from the
static pressure drop.
Chanda tentatively suggested that this apparent discrepancy in his results
may be due to the neglect of the normal stress terms in the momentum integral
equation,

At 2 = 46.5 in. in the present experiments, the contribution due to the second
term on the right-hand side of equation (9) represents only about 5 % of the
magnitude of 2d0/dx. Chanda has also stated that the total sesistance in the
layer near the Iough surface is due to the drag on the roughness, as well as to
the Reynolds shear stress, and therefore the measured values of UV will not give
the total shear stress near the surface. It is likely that the periodic nature of the
present rough wall causes the mean streamlines near the roughness crest to be
wavy, and a term such as ( U V ) /UZ,(see5 3) may be significant.The possibilitythat
the distortion of the mean streamlines in the vicinity of the roughness elements
may affect the measurements of UV cannot, however, be entirely discarded. For
the present investigation, it was assumed that the mean velocity vector near
the wall was parallel to the base of the rough surface. The mean velocity profiles
obtained with a single wire showed that the mean value of V (as obtained with
the equation of continuity) was small compared with the local value of U. The
actual variation of Vover a distance of one roughness pitch could not be extracted
from these results as the single wire responds to the total velocity vector. Also,
an X-wire was found t o be unreliable for the determination of V . It should also
be noted that several UV profiles, taken at intermediate positions between two
consecutive roughness elements, reveal no significant departure from the shape
of the profiles observed above the crest of the roughness.
Another indirect way of determining the shear stress distribution through a
boundary layer is by integration of the momentum equation. This method has
been used in a rough wall zero-pressure-gradient boundary layer by various
workers. The calculated shear stress distributions have in general been similar
to those obtained on the smooth wall, with the trend in the vicinity of the wall
consistent with the assumption of a constant shear stress (see e.g. Liu et al.
1966; Doenecke 1964).
It has been found that the mean velocity profile in the outer part of a rough
wall boundary layer follows the same similarity law as that for the outer region
of a smooth wall layer. Here we try to ascertain whether the same similarity form
also exists for the turbulence intensities measured on the smooth and rough walls.
The u- and w-componentturbulence intensity profiles a t x = 46.5 in. are replotted
 Turbulent boundary layer over a change in roughness 743
in figures 15 ( a )and (b),together with data taken from various sources,in the form
J.”/U7 vs. ylA and J7/qvs. ylA. Because of the uncertainty in the deter-
mination of error in origin 8,the value of y has been left uncorrected for the data
presented in figures 15t. The addition of e to y should not greatly affect the shapes
of the curves, particularly at the larger values of y. It should be noted that the
distributions of 3 and 3 at x = 46.5 in. may be regarded as being representative
of the majority of the profiles of figure 14 (a)for the larger values of x,where the
variation of the skin-friction coefficient is small. The data reported by Moore
(1951),Corrsin & Kistler (1954) andLiu et al. (1966),wereobtained at onestation
only in the flow, and therefore no definite statement can be made about the self-
preserving nature of these distributions. I n particular, the two profiles of Liu
et al. (figure 15 (a)),which correspond to two different Reynolds numbers, tend to
indicate a lack of self-preservation, at least for the lower Reynolds number.
Moore’s distribution of JG/q,although somewhat similar in shape to that for
the present rough wall data, lies significantly above the other data. The present
distribution on the rough wall is in reasonable agreement with that on the smooth
wall, except perhaps near the surface where the trend of the smooth wall data is
towards the higher values of ,@/I&. The distribution of Corrsin & Kistler
appears to be high for y/A less than about 0.15, but is in reasonable agreement
with the present data in the outer region of the boundary layer. Figure 15(b)
shows that the present distribution of JvT/U7 on the rough wall follows the shape
of the smooth wall distribution fairly closely, but its magnitude is slightly low.
Also, the decrease in J v ~ / U in
7 the wall region is more pronounced in the rough
wall boundary layer. Bearing in mind the uncertainty in the origin for y, and the
inaccuracy
- in the determination of cr,$ the assertion that the distributions of
u.2 and 2 assume the same similarity shape in the outer part of smooth and rough
wall boundary layers is not disproved by the data. The same can probably be
said about the shear stress profiles, although the results of figure 16 are even less
conclusive, and considerably more evidence is required. The present values of
-- uv are slightly lower in the outer part of the rough wall boundary layer than
those on the smooth wall, but the trends followed arereasonably similar. Theshear
stress values of Corrsin & Kistler appear t o decrease near the wall, but their
magnitude is significantly higher than the wall shear stress value obtained from
the momentum integral equation.
I n order to get some idea of the distribution of the turbulence intensities in
the region very close to the rough wall, a few measurements were made inside
the cavities between roughness elements. These measurements showed that both
the mean velocity and the u-component turbulence intensity rise fairly sharply
as the top of the cavity is neared. The mean velocity gradient and the gradient
o f 2 are large near the crest of the roughness, a n d z 2reaches a maximum value
t The data reported in Corrsin & Kistler (1954) included a correction for an error in
the origin, the ‘y = 0’ reference being chosen by extrapolation of the mean velocity profiles
in the region outside the corrugation peak8 to zero.
$ The values of c, indicated in figures 21 and 22 were obtained from the momentum-
integral method. One exception is the value of c, for the results of Liu et aZ. (1966) a t
U,6/v II 1.8 x 104, which was inferred from the slope of the logarithmic mean velocity
profile.
744 R.A . Antonia and R.E.Luxton

(4
3 ’ 0 ~

I I I
0 0.1 0.2 0.3
YIA
FIGURE 15. Comparison of the distributions of turbulence velocity components on smooth
and rough walls. ( a ) u component, ( b ) w component.
0 Present data U,S/v= 2.7 x lo‘, 6*/8 = 0 . 2 7 5 , ~ =
, 0.0084
(at x = 46.5in.)
+ Moore(1951)
0 LiuetaL(1966)
Liu et al. (1966)
A Corrsin & Kistler (1954)
x Present data
7.8 x lo4
1.8 x 104
1.0 x 104
6.5 x lo4

1.9 x 104
0.232
0.264
0.262
0.204

0.172
0.010

0.0009 i
Square section
roughness
0.0106 hlk = 4

0.0048 Corrugated
roughness
0.0034 Smooth wall
 Turbulent boundary layer over a change in roughness 745
a short distance outside the cavity. The distribution of the Reynolds shear
stress was measured with a specially designed X-wire, and, while the numerical
values obtained must be regarded with suspicion, the shape of the distribution
is probably correct. This distribution appears to have a maximum just outside
the top of the cavity (at roughly the position at which the maximum of 2,
occurs), but a second (and more convincinglydefined)peak occurs at about 10 yo
of the boundary-layer thickness. As mentioned earlier in 0 2, this latter peak is
observed for all downstream values of x. It is interesting to note that measure-
ments made between elements at x = 10.75 in. showed that the maximum value

1‘2

1.0

0.8

.3
t-
1s 0.6
I

0.4

0.2

FIUTJRE 16. Comparison of the distributions of turbulent shear stress on smooth and rough
walls. Symbolsand data are as for figure 15.----, the smooth wall distribution of Klebanoff
(1955)for 77., S/v = 7.5 x lo4 and C, = 0.0028.

of - UV aUlay, the turbulent energy production, occurs immediately above

the cavity where the mean velocity gradient is largest. As x increases, however,
the peak of - UV, which appears at the larger value of y, gives rise to a second and
much smaller peak in the distribution of - G a U / a y (see Antonia & Luxton
1971 b ) . The distributions of 2 and - UV in the outer part of the cavity are in
qualitative agreement with those obtained by Tani, Iuchi & Komoda (1961)
and Haugen & Dhanak (1966)in the free shear layer associated with a single two-
dimensional cavity under a thin oncoming turbulent boundary layer.
746 R.A . Antonia and R.E . Luxton

8. The mixing length and eddy viscosity distributions

The distribution of mixing length I ( = d / a U / a y ) is shown in figure 17 for
various values of x. On the smooth wall (x = - 4.4 in. and x = - 4 in.) the values
of 1 obtained in the wall region lie fairly close to the straight line 1 = 0.41y, as
expected from local similarity requirements. I n the outer region of the boundary
layer, 1 is approximately constant and equal to about 0.086, in agreement with
the value reported by Bradshaw ( 1 9 6 7 ~ )Very
. near the edge of the boundary
layer, it is expected that 1 will increase fairly sharply because aU/ay decreases

30 - U

25 -

n
. $ * + + +
A

V V A
V

10 O A

I I I I I I I
0 0.4 0.8 1.2 1.6 2.0 24 2.8
Y (in.)
FIGURE17. Mixing-length distributions. U18& N 3.1 x lo4: v, x = -4h.
U,S,/vII 1.9 x 10’: +, z = - 4.4 in.; 0,2; A,4; 0 , 6; U, 46.5.

much more rapidly than T in this region. As the surface roughness changes, aU/ay
increases near the wall, and, although the shear stress T is also increased, 1appears
to be significantlyreduced relative to I = 0.4 1y. I n the outer part of the boundary
layer, where T and aU/ay essentially retain their smooth wall values, I should
remain unaffected. The scatter in the experimental results in this region is partly
due to the small changes in the boundary-layer thickness, and also partly due to
errors in the reduction of the data. At x = 46.5 in., it is found that the values of I
near the wall again follow the line 1 = 0-41y quite closely, whilst in the outer
region 1 is approximately constant and equal to about 0.076.
 Turbulent boundary layer over a change in roughness 747
The distribution of eddy viscosity vT( = r/aU/ay)may be deduced from figures
17 and l a @ ) .It shows that, in the region very near the wall at small distances
downstream fkom the step, no significant reduction of vT relative to its distribu-
tion on the smooth wall can be detected. This tends to suggest that, at least in
this region, the change in the shear stress r is closely allied with that in the mean
velocity gradient aU/ay. The observed reduction in the mixing length 1 would be
explained by the dependence of 1 on d.
The distributions of the ratio a,( = -@/?), where is assumed to be equal
to (#) (G+F),are shown in figure 18 for the region near the step. The main
feature of these results is that very little change from the distribution of a, on the
smooth wall can be detected in the region near the wall for small values of x.

** 0

00
A

ry
f 0

O.I0l
0.08
0

?#

0.04

0
r
A.

T
0
.o
0.2 0.4
YlS
0.6 0.8
1 1.2

18. Distribution of al = -Z/nz in region near the step change, U,S,lv 1: 1.9 x lop.
FIGURE
0 + A U v 0 *
G= -2.5in. 1 2 4 6 10 15
d = 1.9in. 2.0 2.0 2.1 2-1 2.2 2.3

The implication is that a close relationship appears to be maintained between 7

and 5, even though the flow near the wall has been strongly perturbed, and, as
will be seen, is far from equilibrium.A comparison of the distribution of a, on the
smooth wall with that in the self-preservingroughwall boundary layer well down-
stream shows that a, is nearly constant and equal in both layers over a large
fraction of the outer layer, with a value of 0.12, which is smaller than the value of
0.14 deduced by Bradshaw ( 1 9 6 7 ~from
) the measurements of Klebanoff (1955).
In the inner region of the boundary layer, the values of a, on the rough wall fall
progressively below those on the smooth wall as the wall is approached. It is
748 R. A. Antonia and R. E . Luxton
possible that this reduction is due to the reported decrease in the values of - UV
near the rough wall. It is unlikely to be a consequence of the ‘inactive’ motion,
as one suspects that this is increased in strength in the rough wall layer owing to
the considerable increase in the energy in the outer region. This contention is
partly supported by the integral length-scale measurements presented in 0 10.

9. The breakdown of inner layer similarity

A discussion of the various terms in the turbulent energy balance at a station
immediately downstream from the step is given below. From these results, and
from the observations on mixing length made in 8 8, it is deduced that the flow in
the internal layer is not in a state of energy equilibrium.
A simplified form of the turbulent energy equation for a two-dimensional in-
compressible mean flow is (Townsend 1956)
au a - -
(u-+
a
ax
advection
”)
Y - gq2-T-+-(pv+gq2w)+E
a Y -. ay ay
production dimion
= 0,
dissipation
(10)

- - - - - - -
where q 2 = u2 + v2 + w2 and T = - UV. We now assume that w2 = &(u2 + w2) so that
the total turbulence intensity is represented by 42 = $(s+
wT). It
- is-further
assumed that j Z is negligible, so that the diffusion term becomes Qa(u2v+v3)/ay.
The dissipation e is obtained by difference.? The first three terms in the above
equation have been evaluated at x = 4in. for U,S,/v= 1.9 x lo4 and the results,
normalized with respect to 6 and U,, are plotted in figure 19. The thickness of the
internal layer at 2: = 4 in. is approximately 0.4in., correspondingto a value of yl6
of about 0.20.
The contribution to the production of turbulent energy by the interaction of
the normal stresses with the mean velocity gradient in the streamwise direction
(G-vz)aU/ax, neglected in (lo), is largest near the wall (y/6 < 0.1) as both
(2-3)and aU/ax are large in this region. The smooth wall distribution of UV
aU/ay shown in figure 19is that of Klebanoff (1955),obtained at a Reynolds num-
ber V,S/v 2~ 7.8 x lo4 for a zero pressure gradient. Outside the internal layer the
two distributions of UVaU/ay are in good agreement. The large values of UV
8U/ay inside the internal layer do not arise from a transfer of energy flux from
the unaffected flow outside the internal layer through the working of the mean
flow against the Reynolds shear stress, a(zLvU)/ay. They are most probably caused
by the extraction of the energy from the retarded mean flow in the region near
the wall.$
The relatively large advection in the internal layer arises mostly from
Ua(@)/ax, for, although @/ayislargeandnegative over most of theinternallayer,
V , which is positive, is found to be less than 1yo of the local mean velocity U.
The plot of ?/IT: ws. x (figure 20) shows that, at x = 4in., @/ax increases with
distance from the wall, attains a maximum value near y = 0.3 in., then decreases
-f The curves labelled ‘dissipation’ in figure 19 thus realIy include diffusion.
3 It is not useful to think in terms of Townsend’s (1956) two-layer concept which is
applicable to a nearly self-preserving situation. It is more realistic to consider the internal
layer as a separate and almost independent boundary layer.
 Turbulent boandary layer over a change in roughness 749
fairly sharply outside the internal layer. The maximum value of the advection
occurs near y/S -N 0.14, and represents almost 50 % of the value of UV aU/ay
there. I n the central portion of the boundary layer, the advection is negligibly

70

60

5 40
X
rn
I4

20

-20

r*

2
X
.g -40
c5

-60
of internal layer
-70
tI

.,
I I I
1
0 0.2 0.4 0.6 0.8
YP
FIGURE19. Turbulence energy balance at x = 4in. U,S,/v N 1.9 x lo4. production
UV aU/ay; A , diffusion by (SU;)a(-
by (6U;) &n”v)/ay
- only ; 0, Bdvection (6U:)[U a( +?)/
ax+ Va(@)/ay]; +, production by (SU;)(uz-wz) aV/ax; vy dissipation by difference,
0 ,Klebanoff (1955)smooth wall production at U,S/v = 7.8 x 104.

small. Although no measurements were made near the edge of the boundary
layer, it is expected that the advection will rise in this region andremain approxi-
mately equal to the diffusion (see e.g. Klebanoff 1955).The details of the energy
balance in this region are well documented in the literature, and were not of
prime concern in this investigation.
750 R. A . Antonia and R.E . Luxton
The curves in figure 20 show two main features:
- (i) I n the range of y from about 0.20in. to 0-50in., for a constant value of y,
q2 rises fairly sharply for small values of x, with its gradient @/ax reaching a
maximum at the inflexion point.
(ii) For the smaller values of y, appears to overshoot the value fhally ob-
tained at the larger values of 2.

3.0 -

2.0 -
2
X
h

fY
1g
1-0 -
1.20

0
0 4 8 12 16
x (in.)
FIGURE
20. Variation of turbulence energy N 3 ( 2+2 ) / 2 U :
in region near step change for constant values of y.

Before discussing the diffusion results, some of the main short-comings in the
measurement of &&)lay should be noted.
(i) Because of the non-linearity of the anemometer, the value of q q , partic-
ularly in the region near the wall, will be in error. It is unlikely that this error will
be large, as the third-order quantity that is most affected by the non-linearity,
for a constant temperature anemometer, is 2, which does not appear in q x
(ii) There will be a small error due to phase shifts in the anemometer and in
the 1 kHz band limiting filters.
(iii) Although the assumption 2= $(G+ 2) appears reasonable, no clear
assessment can be made of the importance of w22). The measurements of Brad-
shaw (1967 b ) in a strong adverse pressure gradient show that the absolute value
of &is larger than that of uxover the whole thickness of the boundary layer. The
measurements of Johnson (1959) in a zero pressure gradient indicate that % is
not much different from 3 over most of the boundary-layer thickness.
(iv) Finally, the data for & must be differentiated graphically to obtain the
diffusion term a(p22.’)/ay.
 Turbulent boundary layer over a change in roughness 751
The main feature of the diffusion results in figure 19 is that the gain of energy
by the diffusion represents a significant proportion of the total energy gain in the
region near the edge of the internal layer. The peak in the diffusion curve a t
x = 4in. occurs at the value of y a t which the & distribution (figure 21) exhibits
an inflexion. This value: of y corresponds fairly closely to the
- edge
- of the internal
layer. It is interesting to note that the distributions of v3/(v2)#,the skewness
of the v-component turbulence intensity, also show a maximum near the edge
of the internal layer. The maximum may be interpreted as a maximum in the
transport of turbulent energy by the z1 fluctuations, though it could also be
interpreted as a result of a random
- - switching between two different turbulence
signals. The distributions of v3/(v2)8on the smooth wall closely resemble those
obtained by Comte-Bellot (1965) for a developing boundary layer on the smooth
wall of a two-dimensional channel, and by Johnson (1959) for a self-preserving
turbulent boundary layer. The magnitude of the skewness rises fairly sharply
as the edge of the boundary layer is approached. This rise is probably caused en-
tirely by the highly intermittent nature of the turbulence in this region. Beyond
the edge of the boundary layer, it is expected that the skewnesswill again decrease
fairly rapidly to reach its Gaussian value of zero in the tunnel free stream.
As evidenced by the results of figure 21, there is also an energy gain by dif-
fusion near the edge of the boundarylayer. Although this latter gain will be small
compared to the gain near the outer edge of the internal layer, it follows that a
very large loss of energy by diffusion must occur in the region very close to the
wall, so that the area under the diffusion curve is zero. The majority of the &
distributions over the early part of the roughness indicate that the gradient
of %
q near the wall is positive and large.?
However, as the extent over which the positive a(&&)/ay occurs is small, it is
unlikely that this contribution will be of sufficient magnitude to offset the energy
gain in the region y / 8 greater than about 0-1.The present distributions of T
v
on the smooth wall indicate a substantial gain of energy in the outer part of the
constant stress region, but, again, the energy loss in the region very close to the
wall is by no means large enough to close the diffusion curve. Since the distribu-
tion of6 outside the internal layer at x = 4 in. is the same as that on the smooth
wall, a large loss of energy by diffusion very near the rough wall appears to be the
only solution that will bring about a closure of the diffusion curve. It is likely that
there will be an excess of production over dissipation very near the rough wall,
which will eventually maintain the high level of turbulence observed in the outer
part of the rough wall boundary layer.
Townsend (1961) showed that the major requirements for the existence of an
equilibrium layer are that (i) the advection is small, (ii)the dissipation length L,,
defined as L, = 78/e, is directly proportional to y, therelevant length scale8 in the
inner part of the boundary layer.
The advection results of figure 19 clearly indicate that condition (i) is not

t At y = 0, a(&&)/ay = 0.
Bradshaw (19676) points out that y is, strictly speaking,the length scale of the active
motion only.
752 R.A . Antonia and R.E . Luxton
satisfied in the outer part of the internal layer. Further, the results of figure 22
show that, in the region occupied by the internal layer, the values of L, inferred
from the present dissipation results are clearly below the line L, = ~ y tThis
. agrees
with the trend of the mixing length distribution presented in $8. The agreement
is not surprising, as the production is very nearly equal to the dissipation over

FIGURE
0.2 0.4 0.6
YP
0.8

21. Distributions of diffusion energy flux gq%/U!. -,

x = -4-4in.; 0.4; v, 6; 0 . 46.6. ----, U,&,/UII3.1~10~:
A,5; +, 8.
1.o

U,S,/vN 1.9 x lo4:

V, x = - 4 h . ; 0,3;
1.2

.,
most of the internal layer, making L, very nearly identical with 1. It is not un-
reasonable to expect that the rapidly growing internal layer imposes its own
length scale on the flow near the rough wall. If it is assumed that this length scaJe
is given by Si,the thickness of the internal layer, then a possible modification
to the L, = KY relation could be L, = tcyf(y/S,). The failure to satisfy the above
t Peterson (1969) points out that there is no theoretical justification for assuming that
I = KY is valid in non-equilibrium conditions. He uses some unpublished mixing length
results of Plate obtained downstream of a smooth-to-rough step in a wind tunnel, and those
of Busch & Panofsky (1968) downstream of a rough-to-smooth step in the atmosphere, t o
demonstrate his point. The results of Plate indicate ti reduction of the mixing length rela-
tive to I = ~ y in, agreement with the present finding. The results of Busch & Panofsky
indicate an increase in 1 with respect t o KY, in agreement with the results of Antonia &
Luxton (1969}for the boundary-layer fiow downstream of a rough-to-smooth step.
 Turbulent boundary layer over a change in roughness 753
conditions (i) and (ii)indicates that the concept of inner layer similarity is not
applicable for the internal layer flow, and helps to explain the inadequacy of
the various calculation methods (see $ 1 and $ 6) in predicting this flow.

YP
FIUVRE22. Dissipation length scale L,a t z = 4in. derived from
turbulent energy balance shown in figure 19.

10. Length scales on the rough wall

I n this section the streamwise variation of the longitudinal length scale in the
region near the step is discussed. Some measurements have also been made as
several stations in the self-preserving rough wall boundary layer and are com-
pared with those obtained on the smooth wall. The length scale variation near the
step further suggests a breakdown of inner layer similarity as presentedin $9.The
measurements well downstream on the roughness indicate that the structure of
the inner layer on the rough wall is different from that on the smooth wall.

10.1. Region near step

The distribution of the longitudinal length scale L in the region near the step it
first obtained from the autocorrelation curves at a fixed point. Use is made of
Taylor’s hypothesis as is defined by
tmax ~

L = Ruudt,
48 PLH 48
754 R.A . Antonia and R.E. Luxton
*
where R,, W(X, t))/~?
= (u(x) (12)
The averaging time for these results was of the order of 30 sec, and themaximum
time delay t,, was equal to 30 msec. It was found that, when this value of t,,
was reached, the autocorrelation was very nearly zero, or was sometimes slightly
negative.
Figure 23 shows the variation of L with x for three constant values of y in the
region near the step. Also shown in this figure is the approximate edge of the inter-
nal layer. For y = 0.10 in., the length scale is reduced by a factor of almost two
relative to its value on the smooth wall, and the position at which this reduction
first takes place roughIy coincides with the edge of the internal layer. As y
increases, the reduction becomes less significant. At y = 0*40in.,which is just

0
0

-1 0 1 2 3 4 5 6
x (in.)
FIUTJRE 23. Variation of the streamwise integral length scale L with x in the region near
the step change. ---, the point a t which each distribution enters the internal layer.

outside the logarithmic region of the smooth wall boundary layer, it is expected
that L would tend to scale more on the boundary-layer thickness 8.Yet the results
show that L is a slowly decreasing function of x, whereas S increases slightly,
if anything, in this range of values of x.This may be due to an ‘effectat a distance ’
by the internal layer, analogous with the concept of inactive motion. The reduc-
tion in L near the wall seems to support the reported variation of the mixing
length 1 in the region near the step.
A limited number of two-point space-time correlations have also been measured
to verify the trend for L indicated by the fixed point autocorrelations. The nor-
malized longitudinal space correlation coefficient for a zero time delay is defined

where the normalization is made, for convenience, with respect to the turbulence
intensities at the two points that are separated by a distance T ~measured
, in the
x direction. The longitudinal length scale L, is then given by
 Turbulent boundary layer over a change in roughness 755
If the time delay T is non-zero, the normalized space correlation coefficient then
becomes
Bl., = ( 4 0 , O )w - 1 , 7 ) ) / ( 4 2 ( 04&)1)). (15)
An optimum correlation envelope in space can be obtained by plotting the
maximum value of &, denoted here by B, against rl. The time delay (for a
given r,), at which this maximum value occurs, can be used to define a convection
velocity V, for the energy containing eddies, U, being given by the ratio of rl

I .o

0.8

0.6

0.4

0.2

0
0 1.o 2.0 3.0 4.0
r1 (in.)
FIUURE 24. Distributions of A, for three positions of fixed upstream probe at y = 0.1 in.
0, fixed probe at z = -ix6in.; 0,+ -&; A , + 2. Inset shows positions of fixed probe
relative to internal layer.

to this time delay. The moving-axis integral length scale L,,,, which defines the
rate at which the turbulence pattern changes in space, can then be written as

The distributions of &, correspondingt o three streamwise positions of the fixed

upstream probe for the same value of y in the vicinity of the step are shown in
figure 24. It is clear that the extent of the correlation rapidly decreases as the
flow responds to the rough wall boundary condition. The length scales L, derived
from these curves, together with the distribution on the smooth wall at
x = - 3.6 in., are in good agreement with the fixed point autocorrelation results
as seen in figure 25. This seems to support the validity of Taylor’s hypothesis
even in a region where the mean velocity gradient and the turbulence intensities
are high.
Calculations of the convection velocity U,, defined as above, suggest that for
small values of the separation distance r,, U, is close to the local velocity U .
48-2
756 R. A . Antonia and R.E. Luxton
As r l increases, U, becomes smaller than U , implying that the large eddies are
moving at a slower velocity than the local mean. If the value of the time delay,
at which the R,,curve (for a particular value of r l ) is tangent to the optimum
correlation envelope in time, had been chosen instead of the present value, the
departure of the newly defined convection velocity from the local mean would

0
-4 -2 0 2 4 6 8 10
x (in.)
FIGURE 25. Comparison between estimates of streamwise integral length scale from fixed-
point autocorrelation (L,0)
and two-point space-time correlation (L,,,n),in region near the
step chango.

have been somewhat larger. The shortcomings of the above definitions for the
convection velocity have been pointed out by Wills (1964)’who proposes a more
rigorous definition based on the wave-number/phase velocity spectrum, which
may be obtained by Fourier transforming the space-time correlations.
The changes in the shape of the optimum correlation envelope in space,
along a horizontal plane 0.01 in. distant from the wall are shown in figure 26 for
the region near the step. It is seen that the distributions of &,,,ax are lowered as
the fixed upstream probe moves closer to the edge of the internal layer. It is
reasonable to interpret the decrease of, &
, (with distance r l ) relative to a fixed
value of unity, as being due to the rate of change of the turbulence pattern in
time or space. The implication is that the eddies which are convected downstream
by the mean flow at a velocity nearly equal to the convection velocity are being
slowly distorted by the changingrate of strain imposed by the mean flow, or, more
specifically, by the rate of generation of ‘new’ turbulence. The rapid decrease
of BmaX over a short streamwise distance in the region near the step can therefore
be attributed to the increased production of turbulent energy inside the internal
layer, and consequently to the new and different turbulence structure of the
internal layer.
 Turbulent boundary layer over a change in roughness 757

The values of L, shown in figure 26, were not obtained from (16), as f,
was still relatively large at the largest value of rl used. Instead, they were inferred
by assuming an exponential fit to the Am, curve of the form
Rrnm = ~ X (-ri/Lrnax).
P

c
4

O’? t I
0.2
I
0.4
I
0.6
I
04
I
1.0
I
1.2
I
1.4
I
1.6
I
1.8
J
2.0
rl (in.)
B~IGURE26. Variation of the streamwise space-time correlation a t optimum time delay,
RmBX, in the region near the step change a t a distance y = 0.1 in. from the crests of the rough-
ness. Fixed probe positions are :() ,z = - 3.6 in. ;0, -&; +, +iec The corresponding values
of the streamwise moving axis length scale obtained from fitting a curve of the form
R,,, = exp ( -r1/Lw,) to these points are Lmx= 1*66in.,1.28, and 1.03, respectively.
+
0 0

I
0 0.2 0.4 0.6 0.8 1.0

FIGURE 27. Comparison between distributions of longitudinal length scales and rough walls,
U,S,/v = 1 . 9 104.Smoothwall:
~ +, z = -1in. Roughwall: LL46-5; 0,45.2; 0 , 6 7 6 . The
two points a t z = 67.5in. are from two-point correlations (LJ. All other points are from
autocorrelations (L).
758 R. A . Antonia and R. E . Luxton
I n figure 27, distributions of the longitudinal length scales across the smooth
wall boundary layer are compared with those well downstream on the rough
wall. The two values of L,indicated for the rough wall were obtained from the
zero-time-delay space-correlation curves of figure 28. The curves of a,,,
1.0

0.8

0.6
rl

5.
f
<c;
0.4

0.2

0
0 1 2 3 4 5 6
r l (in.)
FIGURE
28. Distributions o f & and A, on rough wall at x = 67.5in.
Fixed probe posititions are: +, y = 0.10 in.; 0,0.60.
also shown in figure 28 tend t o indicate a significant increase in L
,, across the
inner layer on the rough wall. More extensive space-time correlation measure-
ments in the vicinity of a rough wall in a self-preserving boundary layer are
needed to allow further comparison of the turbulence structure in the inner region
of smooth and rough wall layers. The wave-number phase velocity presentation
of these measurements should also provide valuable information about the spread
and convection velocity of turbulent energy over a range of wave-numbers.

10.2. Length scales in the self-preserving boundary layer

In figure 27, the distribution of the length scale L across the self-preserving
boundary layer on the rough wall is compared with that for an approximately
self-preserving smooth wall boundary layer (x= - 1 in.). The two distributions
are in reasonable agreement (taking into account the inevitable experimental
scatter) for values of y/6greater than about 0.3, where it can be assumed that L/6
is approximately constant and equal to about 0.36. This value is in close agree-
ment with the value of 0.37 inferred from spectral measurements by Klebanoff
 Turbulent boundary layer over a change in roughness 759
& Diehl(l951) in a zero-pressure-gradient turbulent boundary layer. Bradshaw
& Ferriss (1965)also reported a value of 0.37 in the outer part of a self-preserving
boundary layer in an adverse pressure gradient (U,cc X - O . ~ ~ ~ )I.n the region near
the wall (y/6 < 0.3))the values of L/6 over the rough wall are significantly lower
than those on the smooth wall. In the latter case, the values of L/6 are progessively
reduced, relative to the constant outer value of 0.36, from y/6 of about 0.1 in. to
the wall. The decrease in L/Gis much more sudden on the smooth wall than on the
rough wall. It was suggested previously that the reduction in L,in the region near
the step was probably caused by the diffusion of turbulence generated directly
by the shear layers associated with the individual roughness elements. It is not
unreasonable to assume that, at even larger values of z, the dominant length
scale for the inner layer turbulence on the rough wall will be a function of the
roughness geometry.? It should be pointed out that the observed decrease in
length scalenear the rough wall is not necessarily incompatible with the previously
reported result, that the mixing length 1 in the region close to the rough wall
varies as KY, since 1 and L need not be directly related near the wall. $:
Some supporting evidence for the observed reduction of L near the rough wal
is provided by the space correlation measurements at zero time delay of Chowd-
hury (1966).These measurements were made at only one point in the inner region
of a zero-pressure-gradient boundary layer at a value of y/6 of about 0-023, for
both a smooth and a sand-roughened surface. Chowdhury found that the integral
length scales in the z and y directions are significantly lower on the rough wall
than on the smooth wall, but the length scale in the z direction was higher on the
rough wall. Robertson et al. (1968) have obtained values of the length scale L
from spectral measurements at a few points across the section of a 3 in. diameter
sand roughened pipe and of an Sin. diameter pipe of ‘natural’ roughness. For a
given pipe, reasonably large increases in L are obtained by increasing the Rey-
nolds number, but as the authors themselves state the dependence of L on pipe
size and roughness is in need of’clarification. No positive statement can be made
about the magnitude of Lf6 near the rough wall of the pipe as very few measure-
ments are available in this region. Also, since the values of AUlU, obtained by
Robertson et al. clearly show that the roughness behaviour was in the transitional
rhgime, the results would probably be of little relevance to the present study. The
spectral measurements of Liu et al. (1966))made for the same roughness geometry
as used here, indicate values of L/6 which are smaller than those on the smooth
wall almost throughout the boundary layer, the average value of L/6 in the outer
rough wall layer being equal to about 0.30.

11. Summary of conclusions

One of the main points to emerge from the present experimental investigation
is that the study of the response of a turbulent boundary layer to a smooth-to-
rough change in surface condition is in effect the study of the development of the
t The distribution of q& at z = 46.5 in. (figure 21) indicates a significant loss of energy
by diffusion in the region near the wall.
$ In the outer region of the self-preserving boundary layer i t is found that both I and
L are proportional t o the boundary-layer thickness 6.
760 R. A . Antonia and R. E . Luxton
internal layer. The structure of the flow outside the internal layer does not seem
to be affectedby the new surface condition apart from a small streamline displace-
ment. The readjustment of the boundary layer to the new surface condition takes
place fairly rapidly, a distance of less than twenty boundary-layer thicknesses
from the start of the roughness being required for the mean flow integral para-
meters, such as cf and H , t o assume values that are appropriate to the self-
preserving boundary layer on the rough wall. The distributions of the turbulence
intensities also become approximately self-preserving within a similar distance
from the smooth-rough junction.
For the internal layer flow in the region near the step, the following observa-
tions are made.
(i) Because of the large production of turbulent energy that occurs near the
wall, the level of turbulence intensities is high. In the outer region of the internal
layer, the turbulence intensity gradients are also found to be large.
(ii) A significant reduction in the mixing length I relative to the smooth wall
distribution KY is observed in the region near the wall. The results of a turbulent
energy balance show that the dissipation length scale L, is nearly equal to 1 and is,
therefore, also reduced. It is further observed that both the zero time delay and
the moving-axis integral length scales are considerably decreased near the rough
wall. These observations clearly demonstrate the inapplicability of the equili-
brium layer concept to the internal layer. The assumption that L, = ~3needs
to be revised before a calculation method can successfully predict the perturbed
flow downstream of the step.
(iii) The diffusion of turbulent energy away from the wall plays a dominant
part in controlling the growth rate of the internal layer. The energy gain by
diffusion has a maximum very near the edge of the internal layer.
(iv) The overshoot of the wall shear stress, inferred from form drag measure-
ments immediately downstream of the step, is not as pronounced as that indicated
by the slope of the logarithmic mean velocity profiles.
(v) The mean velocity profiles exhibit a linear trend when plotted in the form
U us. yP. The edge of the internal layer obtained from these plots appears t o grow
in much the same way as does the zero-pressure-gradient boundary layer.
The measurements in the self-preserving boundary layer on the rough wall
show that the turbulence intensities in the outer layer appear to follow the same
self-preservingform as exhibited by the corresponding distributions on a smooth
wall. It is suggested that the turbulence structure in the inner layer is strongly
influenced by the roughness geometry. The integral length scales are lower than
those on the smooth wall. It has also been found that the Reynolds shear stress is
not constant, but decreases in the region near the wall. It is tentatively suggested
that the missing momentum transport term may be provided by the wavy nature
of the streamlines.

The work described in this paper represents part of a programme of research

supported by The Australian Research Grants Committee, The Australian
Institute of Nuclear Science and Engineering, and The Commonwealth Scientific
and Industrial Research Organization.
 TurbuJent boundary h y e r over a change in roughneSs 761

REFERENCES
ANTONIA,R. A. & LUXTON, R. E. 1969 Department of Mechanical Engineering, University
of Sydney, Charles Kolling Research Laboratory, T.N.F. 8.
ANTONIA, R. A. & LUXTON, R. E. 1971a Trans. A S M E , J . Basic Engng. 93, 22.
ANTONIA,R. A. & LUXTON, R. E. 1971b Phys. Fluids, 14, no. 5.
BRADLEY, E. F. 1965 Ph.D. Thesis, Australian National University.
BRADSHAW, P. 1967 a J . Fluid Mech. 29, 625.
BRADSHAW, P. 1967b J. Fluid J!lech. 30, 241.
BRADSHAW, P. & FERRISS, D. H. 1965 N.P.L., Aero Rept. 1145.
BRADSHAW, P., FERRISS, D. H. & ATWELL, N. P. 1967 J. Pluid Mech. 28, 593.
BUSCH, N. E. & PANOFSKY, H. A. 1968 Quart. J . Roy. Meteor. SOC.94, 132.
CHANDA, B. 1958 Colorado State University, Sci. Rept. 1 CER58BC21.
CHOWDHURY, S. 1966 Colorado State University, Tech. Rep. CER 65SC-EJP57.
CLAUSER, F. H. 1956 Advan. Appl. Mech. 4, 1.
COMTE-BELLOT, G. 1965 Publ. Sci. Tech. M i n . Air. 419.
CORRSIN, S. & KISTLER, A. L. 1954 N A C A T . N . 3133.
DOENECKE, J. 1964 Int. J . Heat Mass Transfer, 7, 133.
ELLIOTT, W. P. 1958 Trans. Am. Geophys. Union, 39, 1048.
FRASER, D. 1969 University of Sydney, Department of Mechanical Engineering, Chrlea
Kolling Research Laboratory T.N.F. 9.
HAUGEN, R. L. & DHANAK, A. M. 1966 Trans. A S M E , J . Appl. Mech 641.
JACOBS, W. 1939 2. angew Math. Mech. 19, 87. Translation 1940, N A C A T.M. 951.
JOHNSON, D. S. 1959 Trans. A S M E , J . Appl. Mech. 26, 235.
KLEBANOFF, P. S. 1955 N A C A Rept. 1247.
KLEBANOFF, P. S. & DIEHL,P. W. 1951 N A C A T.N. 2475.
LIU, C. K., KLINE,S. J. & JOHNSTON, J. P . 1966 Stanford University, Departmnt of
Mechanical Engineering, Rept. M D -15.
LUXTON, R. E., SWENSON, G. G. & CHADWICK,B. S. 1967 The Collection and Processing
of Field Data (ed. E. F. Bradley and 0. T. Denmead), p. 497. Interscience.
LOUAN, E. & JONES,J. B. 1963 Trans. A S M E , J . Bmic Engng 85,36.
MAKITA, H. 1968 M. Eng. Thesis, University of Tokyo.
MOORE,W. F. 1951 Ph.D. Thesis, State University of Iowa.
PANOFSKY, H. A. & TOWNSEND, A. A. 1964 Quart. J . Roy. Meteor. SOC.90, 147.
PERRY,A. E . & JOUBERT, P. N. 1963 J. Fluid Mech. 17, 193.
PERRY,A. E., SCHOFIELD, W. H. & JOUBERT, P. N. 1969 J. Fluid Mech. 37, 383.
PETERSON, E. W. 1969 J. Atmos. Sci. 26, 773.
PLATE,E. J. & HIDY,G. M. 1967 J . Geophys. Res. 72, 4627.
ROBERTSON, J. M., MARTIN,J. D. & BURKHART, T. H. 1968 I. & E.C. Fundamentals, 7,253.
SCHOFIELD, W. H. 1969 Ph.D. Thesis, University of Melbourne.
TANI,I. 1968 AFOSR-IFP-Stanford Conference on Computation of Turbulent Boundary
Layers, Stanford University.
TANI,I., IUCHI,M. & KOMODA, H. 1961 University of Tokyo, Rept. 364.
TAYLOR, P. A. 1969 Quart. J . Roy. Meteor. Soc. 95, 77.
TAYLOR, R. J. 1962 J . Fluid Mech. 13, 529
TOWNSEND, A. A. 1956 The Btructureof Turbulent Shear Flow.CambridgeUniversityPress.
TOWNSEND, A. A. 1961 J . Pluid Mech. 11, 97.
TOWNSEND, A. A. 1965 J . Fluid Mech. 22, 773 and 799.
TOWNSEND, A. A. 1966 J . Fluid Mech. 26, 255.
WILLS,J. A. B. 1964 J. Fluid Mech. 20, 417.