VOLUME 88, NUMBER 1 P H Y S I C A L R E V I E W L E T T E R S 7 JANUARY 2002

014501-1

Scaling and Similarity in Rough Channel Flows

G. Gioia1 and F. A. Bombardelli21Department of Theoretical & Applied Mechanics, University of Illinois, Urbana, Illinois 61801

2Department of Civil & Environmental Engineering, University of Illinois, Urbana, Illinois 61801(Received 21 June 2001; published 17 December 2001)

We show that Manning’s empirical formula for the mean velocity of turbulent flows in channels rep-resents the power-law asymptotic behavior of a flow of incomplete similarity in the relative roughness.We then derive the formula based on the phenomenological theory of turbulence. Our derivation yieldsthe correct similarity exponent; it justifies Manning’s use of a single parameter, the hydraulic radius, tocharacterize the geometry of the cross section; and it affords insight into the mechanism of momentumtransfer.

DOI: 10.1103/PhysRevLett.88.014501 PACS numbers: 47.27.Nz

Manning’s empirical formula for the mean velocityof gravity-driven, uniform, fully developed turbulentflows in rough open channels is among the better knownexpressions used by hydrologists, geomorphologists, andhydraulic engineers. The formula is customarily usedto determine the capacity of natural streams and floodplains, and to design artificial channels [1,2]. It has alsobeen used to quantify the vast flows which appear to haveoccurred on Mars in pre-Amazonian times [3]. Becauseit embodies a large corpus of experimental results [1],and it is known to work very well, Manning’s formulaaffords a singular opportunity for gaining insight into aproblem of considerable theoretical interest and numerousapplications. Yet, there exists no theory of Manning’s for-mula, and the following assertion, made in a classical texton geomorphology [4], remains valid after thirty-sevenyears: “It is truly surprising that engineering practice hasdepended to such an extent on a formula as empirical asthis one, derived nearly a century ago.”

Manning’s formula is usually written in the dimension-ally inconsistent form

V �1n

s1�2R2�3, (1)

where V is the mean velocity of the flow, s is the slope ofthe channel, R is the hydraulic radius of the cross sectionof the channel, and n is the roughness coefficient. Thehydraulic radius is defined as the area of the cross sectionof the channel divided by the length of the wetted perime-ter; for example, for a rectangular channel of width b anddepth h, R � bh��b 1 2h�, and limb!` R � h. (The useof the single parameter R to characterize the geometry ofthe cross section has been amply verified experimentally,at least for the case of rectangular channels.) Appropri-ate values of n have been measured for different types ofchannel walls, and tabulated [2,5].

Before deriving Manning’s formula, Eq. (1), we ascer-tain to what extent it can be predicated on dimensionalanalysis and suitable assumptions of similarity. Based on(1), we start by including V , R, and gs in our set of vari-ables, where gs is the active component of the gravitational

0031-9007�02�88(1)�014501(4)$15.00

acceleration. To characterize the roughness of the channelwalls we follow several authors (e.g., [2]) in using a vari-able r, the absolute roughness, which has units of length.The dimensional equations �gs� � �V 2�R� and �r� � �R�show that the dimensions of two of the variables (gs andr) can be expressed as products of powers of the dimen-sions of the other variables; it follows from Buckingham’sP theorem [6] that we can reduce the functional relation-ship among V , R, gs, and r to an equivalent functionalrelationship between two dimensionless variables. A sen-sible choice of dimensionless variables is F � V�

pgR

(the Froude number) and r�R (the relative roughness). Weexpress the functional relationship between F and the rela-tive roughness in the form F � F �r�R�, or, equivalently,

V � F

∑rR

∏ pRgs , (2)

where F is a dimensionless function of r�R. To make fur-ther progress, we note that in rivers and artificial channelsr�R ø 1, and seek to formulate an asymptotic similar-ity law for r�R ! 0. There are two possible similarities:complete and incomplete [6]. In the case of complete sim-ilarity in r�R, F �r�R� tends to a constant as r�R ! 0.This would make V independent of the roughness in riversand artificial channels (where r�R ø 1), which is con-trary to experimental observation. In the case of incom-plete similarity in r�R, (2) admits the following power-lawasymptotics [6],

V � K

µr

R

∂apRgs 1 o

∑µr

R

∂a∏, (3)

where K is a dimensionless constant, and a is a similar-ity exponent, which cannot be determined by dimensionalanalysis. A comparison of (3) with (1) shows that the lead-ing term of (3) is compatible with Manning’s formula, andthat a � 21�6. The value of a is the most important em-pirical result implicit in Manning’s formula. A comparisonof (3) with (1) also shows that n � K21r1�6g21�2. Twopieces of this expression for n have been proposed pre-viously. The scaling n � g21�2 was suggested by Chow[2], and later justified dimensionally by Yen [7]; it was first

© 2001 The American Physical Society 014501-1

VOLUME 88, NUMBER 1 P H Y S I C A L R E V I E W L E T T E R S 7 JANUARY 2002

used by Carr [3] to adapt the tabulated values of n to thegravitational field of Mars. The scaling n � r1�6 was pro-posed by Strickler [2] based on the analysis of extensiveexperimental data.

In deriving Manning’s formula we expect to verify(i) that an incomplete similarity in r�R prevails forr�R ø 1; (ii) that the similarity exponent is a � 21�6;and (iii) that for rectangular channels the hydraulic radiussuffices to characterize the geometry of the cross section.

We start by considering a rectangular channel of slope s.Then, the streamwise component of the gravitational forceper unit length of channel is Fg � rbhgs, where r is thedensity of the fluid. Let us call S a wetted surface tangentto the peaks of the roughness elements, Fig. 1. (For thetime being, we need only consider roughness elements ofuniform size r.) Under conditions of fully developed tur-bulence, the streamwise component of the force on S perunit length of channel is Ft � �b 1 2h�t. In this expres-sion, b 1 2h is the wetted perimeter and t � rjynyt jis a Reynolds shear stress, where yn and yt are the fluctu-ating velocities normal and tangent to S , respectively, andan overbar denotes time average. We study yn first, andstart by making a crucial observation: when the relativeroughness is small �r�R ø 1�, turbulent eddies of sizeslarger than, say, 2r, can provide only a negligible velocitynormal to S , Fig. 1. On the other hand, turbulent eddiessmaller than r fit in the space between successive rough-ness elements, and they can provide a velocity normal toS . However, when these eddies are smaller than, say, r�2,their characteristic velocities are negligible compared withthe characteristic velocity of the eddies of size r. Thus,yn is dominated by ur , which is the characteristic velocityassociated with the eddies of size r (a suitable mathemati-cal expression for ur is given in [8]). In other words,yn � ur , where the symbol “�” means “scales with.” Wenow turn to yt . Eddies of all sizes can provide a velocitytangent to S . It follows that yt is dominated by V , whichis the characteristic velocity associated with the largesteddies, and yt � V . We surmise that jynyt j � urV ,which together with the equation of balance of momentumtransfer, Fg � Ft, leads to

ur V �µ

bhb 1 2h

∂gs � Rgs . (4)

We now seek to relate ur and V . To that end we useKolmogórov’s scaling. This scaling can be easily derived

FIG. 1. Immediate vicinity of a channel wall with roughnesselements of characteristic size r . The dashed line is the trace of awetted surface S tangent to the peaks of the roughness elements.

014501-2

for isotropic turbulence. It has been proved, however, thatthe scaling applies as well to turbulence which is not onlyanisotropic, but also inhomogeneous [9] (the turbulence isinhomogeneous in the vicinity of the wall). If the eddies ofsize r are within the inertial range (i.e., if r ¿ h, whereh is the Kolmogórov length), then u3

r �r � ´, where ´

is the rate of dissipation of turbulent energy per unit mass.According to Kolmogórov’s theory of turbulence, ´ equalsthe rate of production of turbulent energy per unit mass,and is independent of the viscosity [8,10]. It follows that ascaling expression for ´ can be obtained in terms of V , b,and h. The largest eddies possess an energy per unit mass�V 2; of these, the ones with horizontal vorticity vectorare characterized by a turnover time h�V , whereas theones with vertical vorticity vector are characterized by aturnover time �b�2��V , Fig. 2. We conclude that

u3r

r� ´ �

V 2

h�V1

V2

b�2V�

µb 1 2h

bh

∂V3 �

V 3

R,

(5)

whereupon

ur �µ

rR

∂1�3

V . (6)

This equation indicates that ur is self-similar in r with ex-ponent 1�3, a well-known result of Kolmogórov’s theory[8]. More surprisingly, r appears normalized by the hy-draulic radius R. Substituting (6) into (4) yields

V �µ

rR

∂21�6pRgs , (7)

which is the leading term of (3) with a � 21�6, asexpected. This concludes our derivation.

We have derived Manning’s formula for the case of chan-nel walls with roughness elements of uniform size r. Wenow generalize our derivation to the case of channel wallswith roughness elements in a range of sizes. Consider

FIG. 2. Largest-length-scale eddies in a rectangular channel ofwidth b and depth h. The velocity of these eddies scales withthe mean velocity of the flow V .

014501-2

VOLUME 88, NUMBER 1 P H Y S I C A L R E V I E W L E T T E R S 7 JANUARY 2002

a channel wall W1 characterized by a probability distri-bution p�s�,

R`0 p�s� ds � 1, where p�s� ds measures

the probability of finding a roughness element of a sizebetween s and s 1 ds. Assume that for the channel wallW1 the average roughness element is of size 1, i.e., thatR`

0 sp�s� ds � 1. Then, we can use W1 to generate afamily of geometrically similar channel-wall surfaces�Wr. For a generic member Wr of this family ofchannel-wall surfaces the average roughness element isof size r, i.e.,

R`0 sp�s�r� d�s�r� � r. (The concept of

geometrically similar channel-wall surfaces dates back tothe early Twentieth Century; see, e.g., [11]). We nowrederive Manning’s formula for a generic member Wr ofthe family of geometrically similar channel-wall surfaces�Wr . The average Reynolds stress on the wetted surfaceS of Fig. 1 is t � rV

R`0 usp�s�r� d�s�r�, and

we can rewrite (4) in the form

VZ `

0usp�s�r� d�s�r � � Rgs . (8)

On the other hand,

us �µ

s

R

∂1�3

V �

µs

r

∂1�3µr

R

∂1�3

V . (9)

Substituting (9) into (8) leads to the leading term of (3)with

K � K0

µ Z `

0j1�3p�j� dj

∂21�2

, (10)

where K0 is a constant.We obtained (7) based on three assumptions. The first

one is that r�R ø 1. In keeping with this assumption, (7)corresponds to the leading term in the power-law asymp-totics of Eq. (3). The second assumption is that the tur-bulent eddies in the vicinity of the walls are governed byKolmogórov’s scaling (6). This is justified because Kol-mogórov’s scaling has been shown to apply to inhomoge-neous turbulence. The third assumption is that the spacesbetween roughness elements are occupied by eddies of sizer, in the form shown in Fig. 1. We now discuss this thirdassumption.

It is apparent that an eddy of size r could be found be-tween any two successive roughness elements. In deriving(7) we have assumed, however, that one such eddy does oc-cupy the space between each pair of consecutive roughnesselements. Our assumption could be justified by recallingthat in Kolmogórov’s theory eddies of any given size withinthe inertial range are space filling (this is required for ´ tobe scale invariant within the inertial range [8]). It is perhapsmore illuminating to think of the assumed set of eddies ofsize r as akin to the arrays of parallel vortices that havelong been documented in the vicinity of smooth channelwalls, and which constitute the most common form ofcoherent structures. (Note, however, that the eddy of size rin Fig. 1 need not have a vorticity vector oriented stream-wise.) We know from theoretical work on the etiologyof coherent structures that numerous instabilities are pos-

014501-3

sible leading to arrays of vortices of specific wavelengths[12,13]. Interestingly, it has been conjectured that thepresence of periodic forms of wall roughness (such as, forinstance, riblets) may excite instabilities of similarwavelength [12]. This conjecture affords a compellingexplanation for the incomplete similarity in the relativeroughness, r�R, displayed by Eq. (3). In fact, thissimilarity is quite puzzling: given that turbulence involvesa wide spectrum of wavelengths, spanning many orders ofmagnitude, why would r, which is just one wavelengthsomewhere within that spectrum, appear so conspicuouslyin (3)? The puzzle is explained if the wall roughnessinduces arrays of eddies of size r in the immediate vicinityof the wall and if, as suggested by our derivation, theseeddies effect most of the momentum transfer. Thus, if rdiminishes, the capacity for momentum transfer alsodiminishes; as a result, the fluid friction diminishes, and themean velocity increases, as indicated by Eq. (7). Giventhat the size of the eddies is bounded below by theKolmogórov length h, it is interesting to investigatewhat happens when r approaches h. To that end, westart by recalling that h � n3�4´21�4, where n is thekinematic viscosity. From (5) we have ´ � V3�R,and therefore h�R � �n�VR�3�4 � Re23�4, whereRe � VR�n is the Reynolds number. Therefore, as theroughness approaches the Kolmogórov length (i.e., asthe channel walls become hydraulically smooth), weexpect (7) to become

V � Re1�8p

Rgs . (11)

The appearance of the Reynolds number in (11) indicatesthat in the limit r ! h the momentum transfer is viscous.It is convenient to write (11) in terms of the resistancecoefficient, f � Rgs�V2; the result is f � Re21�4, whichwe recognize as Blasius’s classical empirical scalingfor hydraulically smooth channels [2]. This result un-veils the existence of a relationship among the three wellknown, and apparently unrelated, scalings due to Blassius,f � Re21�4, Kolmogórov, h � n3�4´21�4, and Manning,V � r21�6.

We have provided a derivation of Manning’s empiricalformula. Besides the final result, we have reached a num-ber of interesting conclusions. For example, in rectangularchannels the Reynolds stress in the immediate vicinity ofthe walls depends on (i) the mean velocity of the flow,(ii) the local wall roughness, and (iii) the depth and widthof the cross section through the hydraulic radius only. Thisconclusion suggests ways of formulating generalized Man-ning formulas for channels in which different portions ofthe channel walls are characterized by different familiesof geometrically similar channel-wall surfaces, of con-siderable interest in applications. It also has momentousgeomorphological implications, since it allows for the de-termination of absolutely stable aspect ratios, b�h, in natu-ral channels. We shall study these and other related issuesin a separate paper.

014501-3

VOLUME 88, NUMBER 1 P H Y S I C A L R E V I E W L E T T E R S 7 JANUARY 2002

G. G. is very grateful to Professor W. R. C. Phillips forencouragement, for many illuminating discussions, and forhis comments on several versions of the manuscript.

[1] J. C. I. Dooge, in Channel Wall Resistance: Centennial ofManning’s Formula, edited by B. C. Yen (Water ResourcesPublications, Littleton, Colorado, 1992).

[2] V. T. Chow, Open-Channel Hydraulics (McGraw-Hill, NewYork, 1988).

[3] M. H. Carr, J. Geophys. Res. 84, 2995 (1979).[4] L. B. Leopold, M. G. Wolman, and J. P. Miller, Fluvial Pro-

cesses in Geomorphology (W. H. Freeman and Co., SanFrancisco, 1964).

[5] For a relatively recent example, see G. J. Arcement andV. R Schneider, Guide for Selecting Manning’s RoughnessCoefficients for Natural Channels and Flood Plains,Water-Supply Paper No. 2339 (Department of the Interior,

014501-4

U.S. Geological Survey, Reston, VA, 1990). Also seehttp://wwwrcamnl.wr.usgs.gov/sws/fieldmethods/Indirects/nvalues/ for a table with beautiful illustrations.

[6] G. I. Barenblatt, Scaling, Self-Similarity, and IntermediateAsymptotics (University Press, Cambridge, 1986).

[7] B. C. Yen, J. Hydraul. Eng. 118, 1326 (1992).[8] U. Frisch, Turbulence (University Press, Cambridge, 1995).[9] B. Knight and L. Sirovich, Phys. Rev. Lett. 65, 1356

(1990); R. D. Moser, Phys. Fluids 6, 794 (1994).[10] D. Lohse, Phys. Rev. Lett. 73, 3223 (1994). The existence

of an upper bound on ´ that is independent of the viscosityhas been proved mathematically; see Ch. R. Doering andP. Constantin, Phys. Rev. Lett. 69, 1648 (1992).

[11] T. von Kármán, Turbulence, Aeronautical Reprints No. 89(The Royal Aeronautical Society, London, UK, 1937).

[12] W. R. C. Phillips, in Eddy Structure Identification in FreeTurbulent Shear Flows, edited by J. P. Bonnet and M. N.Glauser (Kluwer Academic Publishers, Dordrecht, 1993).

[13] L. N. Trefethen, A. E. Trefethen, S. C. Reddy, and T. A.Driscoll, Science 261, 578 (1993).

014501-4

RIVERS UNIFORM FLOW

if : 1) there are no appreciable variations in the channel geometry (width, slope , roughness/grain size), for a certain length of a river reach 2) flow discharge does not vary then, UNIFORM FLOW conditions are established

0

t

U

x

U

Given the cross section characteristics the location of the free surface / flow depth will result from the balance between gravity and the flow resistance/friction

gR

V

R

kVRf

V

CkRVg

r

geometryroughradiushydraulic

,,

),,,,,,,(

2

Dimensional analysis:

Reynolds relative roughness Froude

In an open channel flow the flow is certainly turbulent and the regime certainly rough ; thus at a scale of a river, we - neglect viscous effect and - lose the dependency on the Reynolds

number (Re high enough ! )

BLACKBOARD 1

Hydraulic radius R=A/P

because the flow is uniform: the bed topography // free surface water slope // energy grade line

Thus the energy losses hf in a L reach can be estimated by :

assuming:

factorfriction f with 8

2

f

R

kfunction

V

r

Note that L=4R is the key length scale to consider when extending the Moody diagram from pipe to open channel flow (Dpipe 4Rhydraulic radious )

22

2

1

82

1Vc

fV f

solving for the shear stress

we need an empirical closure providing f or cf , C, hf for a given cross sectional geometry and roughness (usual problem in turbulent flow)

slope of the energy grade line

cf = roughness coeff.

figure_09_12

Moody diagram constant Re f1/2 curves (given D, L, Ks find losses hf)

2

9.010Re

74.5

7.3log

25.0

D

kf

s

explicit formula

Re= 4R V/ν

ks / 4R

Empirical closure 1: friction factor in open channel flow

Empirical closure 2: roughness and friction factors (how can we define ks of a river to inform the Moody diagram?)

Manning proposed a dimensional coeff.:

2/13/2

2/13/2

1

SRn

KV

SRCV

n

n is the dimensionless Manning coefficient: while [Kn]= m1/3 /s is dimensional (Kn dimensional is the price we pay to have n a number) Kn=1 m1/3/s (SI m, m/s) Kn=1.49 ft1/3/s (English Units ft, ft/s) n large : low velocity, large roughness : e.g. n=0.05 for high grass or cobbles n small : high velocity, low roughness : e.g. n=0.01 clean straight concrete

Gauckler-Manning

original (old) Manning

The V dependency on R2/3 has been elegantly demonstrated only in 2002 Gioa et al. PRL

BLACKBOARD 2-3-4

8

TOUR OF BEDFORMS IN RIVERS: RIPPLES (L<0.6m, a few cm in height)

Ripples in the Rum River, Minnesota USA at very low flow; ~ 10 - 20 cm.

Ripples are characteristic of a) very low transport rates in b) rivers with sediment size D less than about 0.6 mm. Typical wavelengths are on the order of 10’s of cm (up to 60cm) and and wave heights are on the order of cm. Ripples migrate downstream and are asymmetric with a gentle stoss (upstream) side and a steep lee (downstream side). Ripples do not interact with the water surface.

flowmigration

View of the Rum River, Minnesota USA

from Gary Parker ebook , former UMN, SAFL, professor currently at UIUC

9

TOUR OF BEDFORMS IN RIVERS: DUNES (L>0.6m, scale with the depth, also in amplitude)

Dunes in the North Loup River, Nebraska USA. Two people are circled for scale. Image courtesy D. Mohrig.

Dunes are the most common bedforms in sand-bed rivers; they can also occur in gravel-bed rivers. Wavelength can range up to 100’s of m, and wave height can range up to 5 m or more in large rivers. Dunes are usually asymmetric, with a gentle stoss (upstream) side and a steep lee (downstream) side. They are characteristic of subcritical flow (Fr sufficiently below 1). Dunes migrate downstream. They interact weakly with the water surface, such that the flow accelerates over the crests, where water surface elevation is slightly reduced. (That is, the water surface is out of phase with the bed.)

flowmigration

from Gary Parker ebook , still affiliated SAFL professor, currently at UIUC

10

TOUR OF BEDFORMS IN RIVERS: ALTERNATE BARS

Alternate bars in the Naka River, an artificially straightened river in Japan. Image courtesy S. Ikeda.

Alternate bars occur in rivers with sufficiently large (> ~ 12), but not too large width-depth ratio B/H. Alternate bars migrate downstream, and often have relatively sharp fronts. They are often precursors to meandering. Alternate bars may coexist with dunes and/or antidunes.

11

BEDFORMS IN THE LABORATORY AND FIELD: DUNES

Dunes on an exposed point bar in the meandering Fly River, Papua New Guinea

Dunes in a flume in Tsukuba University, Japan: flow turned off. Image courtesy H. Ikeda.

12

Rhine River, Switzerland

BEDFORMS IN THE LABORATORY AND FIELD: ALTERNATE BARS

Alternate bars in a flume in Tsukuba University, Japan: flow turned low.

Image courtesy H. Ikeda.

Alternate bars in the Rhine River between Switzerland and Lichtenstein.

Image courtesy M. Jaeggi.

13

BEDFORMS IN THE LABORATORY AND FIELD: MULTIPLE-ROW (LINGUOID) BARS

Linguoid bars in a flume in Tsukuba University, Japan: flow turned off.

Image courtesy H. Ikeda.

Linguoid bars in the Fuefuki River, Japan. Image courtesy S. Ikeda.

14

Ohau River, New Zealand

WHEN THE FLOW IS INSUFFICIENT TO COVER THE BED, THE RIVER MAY DISPLAY A BRAIDED PLANFORM

Braiding in a flume in Tsukuba University, Japan: flow turned low.

Image courtesy H. Ikeda.

Braiding in the Ohau River, New Zealand. Image courtesy P. Mosley.

15

Alluvial rivers construct their own channels and floodplains. Channels and floodplains co-evolve over time.

Fly River, Papua New Guinea

16

Beni River, Bolivia Images courtesy R. Aalto

17

Fly River, Papua New Guinea

Okavango River, Botswana Image courtesy N. Smith

18

GRAVEL-BED

A river with a characteristic surface size in excess of 16 mm can be termed a gravel-bed river. Here the term “gravel” is used loosely to encompass cobble- and boulder-bed streams as well. Three such streams are shown below.

Genessee River, New York, USA.

Raging River, Washington, USA. Rakaia River, New Zealand

1D SEDIMENT TRANSPORT MORPHODYNAMICS

with applications to

RIVERS AND TURBIDITY CURRENTS © Gary Parker November, 2004

19

THRESHOLD OF MOTION

DgD,1006.022.0 p

)7.7(6.0

pc

6.0p

RReRe

Re

The Shields number * is defined as

RgD

u

RgD

2

*b

Shields (1936) determined experimentally that a minimum, or critical Shields

number is required to initiate motion of the grains of a bed composed of non-

cohesive particles.

Brownlie (1981) fitted a curve to the experimental line of Shields and obtained the

following fit:

c

Based on information contained in Neill (1968), Parker et al. (2003) amended the

above relation to

]1006.022.0[5.0)7.7(6.0

pc

6.0p

ReRe

In the limit of sufficiently large Rep (fully rough flow), then, becomes equal to

0.03.

c

1

sR

1D SEDIMENT TRANSPORT MORPHODYNAMICS

with applications to

RIVERS AND TURBIDITY CURRENTS © Gary Parker November, 2004

20

MODIFIED SHIELDS DIAGRAM

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.1

1 10 100 1000 10000 100000 1000000

Rep

c*

sandsilt gravel

The silt-sand and sand-gravel

borders correspond to the values

of Rep computed with R = 1.65, =

0.01 cm2/s and D = 0.0625 mm

and 2 mm, respectively.

1D SEDIMENT TRANSPORT MORPHODYNAMICS

with applications to

RIVERS AND TURBIDITY CURRENTS © Gary Parker November, 2004

21 21

0.01

0.1

1

10

1 10 100 1000 10000 100000

Rep

*

motion mod Brownlie

ripples

suspension

dunes C&C

ripples C&C

extrap C&C dunes

2

p

v

6.11orD

Re

2pfs )(orvu ReR

modified Brownlie

C&C ripples/no ripples

C&C no dunes/dunes

extrapolated C&C

no dunes/duneslower regime plane bed

dunes

ripples

no motion

suspension

SHIELDS DIAGRAM INCLUDING RESULTS OF CHABERT AND CHAUVIN (1963)