See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/229924843

River channel and bar patterns explained and predicted by an empirical and
physics-based method

Article  in  Earth Surface Processes and Landforms · May 2011

DOI: 10.1002/esp.2090

CITATIONS READS

231 6,409

2 authors:

Maarten G Kleinhans Jan H Van den Berg

Utrecht University Utrecht University
395 PUBLICATIONS   7,752 CITATIONS    77 PUBLICATIONS   2,626 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Bar and channel patterns in estuaries View project

Morphodynamics of channel networks in tidal deltas View project

All content following this page was uploaded by Jan H Van den Berg on 29 November 2017.

The user has requested enhancement of the downloaded file.

EARTH SURFACE PROCESSES AND LANDFORMS
Earth Surf. Process. Landforms (2010)
Copyright © 2010 John Wiley & Sons, Ltd.
Published online in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/esp.2090

River channel and bar patterns explained

and predicted by an empirical and a
physics-based method
Maarten G. Kleinhans* and Jan H. van den Berg
Department of Physical Geography, Faculty of Geosciences, Utrecht University, Utrecht, The Netherlands

Received 7 January 2010; Revised 10 August 2010; Accepted 19 August 2010

*Correspondence to: Maarten G. Kleinhans, Department of Physical Geography, Faculty of Geosciences, Utrecht University, PO Box 80115, 3508 TC Utrecht, The
Netherlands. E-mail: m.kleinhans@geo.uu.nl

ABSTRACT: Our objective is to understand general causes of different river channel patterns. In this paper we compare an
empirical stream power-based classification and a physics-based bar pattern predictor. We present a careful selection of data from
the literature that contains rivers with discharge and median bed particle size ranging over several orders of magnitude with various
channel patterns and bar types, but no obvious eroding or aggrading tendency. Empirically a continuum is found for increasing
specific stream power, here calculated with pattern-independent variables: mean annual flood, valley gradient and channel width
predicted with a hydraulic geometry relation. ‘Thresholds’, above which certain patterns emerge, were identified as a function of
bed sediment size. Bar theory predicts nature and presence of bars and bar mode, here converted to active braiding index (Bi). The
most important variables are actual width–depth ratio and nonlinearity of bed sediment transport. Results agree reasonably well
with data. Empirical predictions are somewhat better than bar theory predictions, because the bank strength is indirectly included
in the empirical prediction. In combination, empirical and theoretical prediction provide partial explanations for bar and channel
patterns. Increasing potential-specific stream power implies more energy to erode banks and indeed correlates to channels with
high width–depth ratio. Bar theory predicts that such rivers develop more bars across the width (higher Bi). At the transition from
meandering to braiding, weakly braided rivers and meandering rivers with chutes are found. Rivers with extremely low stream
power and width–depth ratios hardly develop bars or dynamic meandering and may be straight or sinuous or, in case of
disequilibrium sediment feed, anastomosing. Copyright © 2010 John Wiley & Sons, Ltd.
KEYWORDS: braided river; meandering river; anabranching; scroll bar; chute bar

Introduction
Alluvial channels classically are categorised as braided, mean-
dering and straight (Leopold and Wolman, 1957; Ferguson,
1987; Nanson and Knighton, 1996) (Figure 1). These patterns
form members of a continuum without physical thresholds
(Carson, 1984; Ferguson, 1987). In this paper we consider
individual channels of an anabranching system on the
continuum from straight to braided (Nanson and Knighton,
1996).
Channel pattern and the style of meandering or braiding are
closely related to the nature of the bars (see Kleinhans, 2010,
for review). Rivers self-organise their planform pattern through
feedbacks between bars, channels, floodplain and vegetation,
which emerge as a result of the basic spatial sorting process of
washload sediment and bed sediment. While bars and flood-
plain form, the banks are eroded on the other side of the
channel. Thus the balance between floodplain formation and
destruction determines the width of channels. Bar pattern, in
turn, is determined mostly by width–depth ratio. Relatively
narrow rivers may have alternate scroll bars and point bars and
are single-thread. Wide rivers with multiple bars in their cross-
section have multiple channels and are therefore braided.
Transitional states include weak braiding, chute bars and chute
channels.
Bar patterns provide the template of bank erosion and for-
mation as well as the dynamics of channel networks in
anabranching and braiding rivers. In meander bends, bars are
forced to their positions by the bends, in contrast to free bars
that may migrate. It is therefore tempting to infer that alternat-
ing bars lead to alternating bank erosion so that meandering
rivers emerge. But this requires that new floodplain forms on
the opposite side at the same rate as it is eroded, otherwise the
channel widens and changes to a more braided state. Thus
floodplain formation by washload deposition, levee formation
and vegetation enters the explanation of channel and bar
pattern (Kleinhans, 2010).
Simple parameters related to the channel and unrelated
to the floodplain have been used with some success for inde-
pendent prediction of the pattern despite the complexity of
bar–channel–floodplain interaction. One general approach
was semi-empirical, wherein certain parameters for channel
2 M. G. KLEINHANS AND J. H. VAN DEN BERG
Figure 1. Classification of alluvial river patterns including single
channel and anabranching forms. Laterally inactive channels consist
of straight and irregular sinuous forms, whereas active channels
consist of more regular sinuous (=meandering) and braided forms
(modified after Nanson and Knighton, 1996).
flow strength such as gradient, stream power or sediment
mobility were plotted for a range of rivers (see Ferguson,
1987, for review). The other approach was more physics-
based, whereby the physics of flow and sediment transport
were simplified to such an extent that analytical solutions
emerged for incipient bar or channel patterns (Parker, 1976;
Fredsøe, 1978; Struiksma et al., 1985; Seminara and Tubino,
1989).
The relevance of such simple predictors is twofold. Firstly,
reasonable success of simple predictors tells us that channel
pattern is a relatively simple emergent property of the full
underlying complexity. Secondly, simple predictors and clas-
sifications remain useful assessment tools for river renaturali-
sation purposes, geological reconstruction and efficient
characterisation of many rivers. Many small rivers all over the
world are being renaturalised and allowed to re-meander.
However, some of these rivers never meandered, which could
have been predicted with a simple predictor. Moreover, as
hydrological and sediment transport boundary conditions have
changed since the historical time that those rivers were
Copyright © 2010 John Wiley & Sons, Ltd.
natural, their new pattern may differ from that on historical
maps. The effect of such change can again be assessed by
simple predictors. Furthermore, many large rivers in the world
are changing from one to another channel pattern and the
simple predictors provide hypotheses for the causes. Our work
on the simple predictors is not intended to replace more
sophisticated modelling and experimental work, but merely to
feed it with new hypotheses, to compare the historically sepa-
rated geomorphological and engineering approaches, and to
outline the limitations of the simple predictors. Indeed, our
own work in progress includes such experimental and numeri-
cal recreation of these patterns (see Kleinhans, 2010, for
outline).
The first objective of this paper is to extend the van den Berg
(1995) semi-empirical predictor of channel pattern to more
detailed prediction of bar and channel pattern, while stressing
that channel patterns form a bar pattern-related continuum
without hard thresholds (Ferguson, 1987). The second objec-
tive is to compare the empirical predictions to a simplified
physical predictor (Struiksma et al., 1985; Crosato and Mos-
selman, 2009) for bar pattern, in order to explore the explana-
tory and predictive power of the semi-empirical and physical
approaches.
The setup of the paper is as follows. First, the semi-empirical
prediction method for channel pattern of van den Berg (1995)
is outlined, followed by our extension to bar pattern based on
literature and observation. Next, the physics-based prediction
methods of bar stability by Struiksma et al. (1985) and number
of braids by Crosato and Mosselman (2009) (based on
Struiksma et al., 1985) are described. We then describe how
data on natural rivers and their patterns were collected (pro-
vided as online supporting information). The results section
describes how the data support the theory and empirical pre-
dictors and how the latter two compare. In the discussion we
counter earlier criticism on the empirical discrimination,
address the effect of using different predictors for channel
width, and discuss the anastomosing and anabranching pat-
terns and the relation between stream power, floodplain for-
mation and resulting bank strength.
Theory
Empirical discrimination
The van den Berg (1995) discriminator between meandering
and braiding
Stream power is a useful parameter in river pattern prediction
simply because it represents the energy to move sediment
(Ferguson, 1987). Van den Berg (1995) developed a method
of channel pattern prediction that incorporates valley slope,
mean annual flood discharge and a predicted channel width
by a hydraulic geometry relation (given below), so that no
information of actual, pattern-dependent channel character-
istics is required for the prediction. The valley slope is inde-
pendent of the channel sinuosity; the mean annual flood
discharge is a hydrological measure much less dependent
on pattern than bankfull discharge, and predicted width is
independent of measured width. In other words, this method
only uses independent variables, contrary to other empirical
methods.
The method allows a discrimination of braided and mean-
dering river patterns in unconfined alluvium from general
pattern-independent boundary conditions of median grain size
of the river bed D50 mm and a potential specific stream power
(W/m2), wpv, defined as
Earth Surf. Process. Landforms (2010)
 RIVER CHANNEL AND BAR PATTERNS EXPLAINED AND PREDICTED
ρ gQSv
ω pv = (1)
Wr
-3
where r = water density (kg m ), g = acceleration due to
gravity (m s-2), Sv = valley slope (–), related to channel slope as
S = PSv, Wr = reference channel width (m) and Q = channel-
forming discharge (m3 s-1) (mean annual flood or bankfull
discharge, discussed later). Essentially, wpv is a parameter for
the potential maximum of the available flow energy corre-
sponding to a minimum sinuosity P = 1.
The reference width Wr is predicted to remain independent
of the actual, pattern-dependent width Wa as (see van den
Berg, 1995, for discussion of this choice)
Wr = α Q (2)
with α = 4⋅7 s m−1 for sand defined as D50 < 2 mm and
a = 3·0 for gravel.
A discriminant analysis based on a large number of river
data indicated that for stream power versus median particle
size of the channel bed (in metres) the discriminator between
dominantly braiding and meandering was found at
ω bm = 900D50
0⋅42
(3)
where subscript bm indicates braided meandering (van den
Berg, 1995).
Bledsoe and Watson (2001) confirmed this empirical finding
of van den Berg using logistic regression analysis. They showed
that it represented the 50% probability of braiding, which
again emphasises that these discriminants do not discri-
minate but indicate a gradual transition. Rather than stream
power Bledsoe and Watson (2001) used a mobility parameter
defined as
M = QS (4)
( m3 s−1 ) and found the discriminator at
− β0 + q
β1 − β 2 β1 (5)
Mbm = 10 D50
with coefficients (their model 72) b0 = 12·78, b1 = 8·08 and
b2 = -2·40. The probability distribution around this discrimi-
nator is described by the parameter q:
p charges in accretionary inner meander bends and emerge at
q = ln (6)
1− p
where P = probability (for instance, 0·1 or 0·9 for the 10% and
90% percentiles). Substitution of Equation (2) into Equation (1)
shows that the mobility parameter of Bledsoe and Watson
(2001) can be reinterpreted as specific stream power wherein
prediction of width is implicitly included. It comes as no
surprise, then, that stream power and mobility parameter gave
nearly exactly the same discriminator between braiding and
meandering (Bledsoe and Watson, 2001). Henceforth we refer
to both as the empirical method and use the stream power
method.
Makaske et al. (2009) showed that the empirical method
also allowed the prediction of low-energy stable channels by a
discriminant at a tenfold lower stream power:
Copyright © 2010 John Wiley & Sons, Ltd.
3
ω ia = 90D50
0⋅42
(7)
where subscript ia refers to the discrimination between inac-
tive channels and active channels with scroll bars. Inactive or
stable channels are here defined as being immobile, i.e.
having no lateral migration because their flow energy is below
the threshold to erode the bank material. This discriminator
was hinted at in Ferguson (1987, p. 131).
We prefer to call these stable or laterally immobile channels
rather than straight as in Leopold and Wolman (1957), as in
reality many of these channels have an irregular course, some
with a high sinuosity, which erroneously suggests a process of
active bank erosion, lateral migration and scroll bar formation
– in other words, meandering. The immobile anastomosing
channels of the Columbia River are indeed straight (Makaske
et al., 2002), but the nearly immobile channels on intertidal
mud flats only erode their banks in the sharpest bends due to
flow separation (Leeder and Bridges, 1975; Kleinhans et al.,
2009). As we will argue in this paper, it is the presence and
nature of the bars combined with nature and rates of flood-
plain formation and bank retreat that define a more meaning-
ful river pattern.
Finally, we have considered non-dimensionalising the
stream power but decided against it. Stream power could be
made non-dimensional in a similar way to the Shields number,
but this would require estimating the flow velocity or the
sediment-settling velocity, which introduces inaccuracies due
to friction parameters. Furthermore, it would suggest a gener-
ality that is not valid; the present analysis is limited to water,
siliclastic sediment and terrestrial gravity. Alternatively, stream
power could be made non-dimensional by defining a non-
dimensional discharge similar to that of Parker et al. (2007),
using particle size and gravitational acceleration. Again, it is
the question whether this is physically correct and it does not
improve the predictive result.
Discrimination between bar patterns by stream power?
Based on a number of observations we will hypothesise how
and why bar pattern can be predicted from stream power. This
will lead to an extension of the original diagram of van den
Berg (1995), which can more directly be compared to bar
theory.
In the inner bank of a meander three types of bars may be
found: scroll bars, chute bars and tail bars (Figure 2). Tail bars
are generated in the wake of obstructions, such as large woody
debris as described by Edwards et al. (1999). Thus tail bars can
be formed in all river systems, meandering or braided, irre-
spective of flow energy conditions, so we will further disregard
this type. Scroll bars and chute bars form during peak dis-
low stage. Based on the literature we will argue below that
chute bars are associated with higher stream power whereas
scroll bars may occur in low and high stream power.
A scroll bar is a curvilinear ridge to the side of the channel
and more or less parallel to the channel. In planform it often
points down-channel and is therefore sometimes named a
point bar (Smith, 1974; Church and Jones, 1982). Yet this is
confusing as the frequently used term point bar in sedimen-
tology denotes the accretional part of the inner meander bend.
A point bar extends vertically from the deepest part of the main
channel to the starting point of accumulation of top-stratum
fines associated with floodplain formation. As such they com-
monly preserve transverse sorting in the channel bend as
upward fining of the entire point bar deposit.
Scroll bars can result from several processes (Nanson and
Croke, 1992; Bridge, 2003, pp. 142–145). The best-known and
Earth Surf. Process. Landforms (2010)
4 M. G. KLEINHANS AND J. H. VAN DEN BERG
Figure 2. Definition sketch of three bar types in meandering rivers
that may emerge at low discharge in meander bends. The tail bar is
caused by accidental presence of obstructions such as large woody
debris. The chute and scroll bars are related to the bend and hydro-
dynamics. In the event of a chute cutoff the chute bar is eroded.
possibly most important process is their generation by
landward migration and coagulation of transverse bars, down-
stream of the bend apex in the decelerating, slightly upslope-
directed near-bed helical flow. The flow over a scroll bar
curves towards the inner bend, because of difference in head
between the main flow and the more or less stagnant water in
the slough behind the bar (Lewin, 1970; Ackers, 1982). As a
result, transverse bars migrate upslope and are remoulded into
a dynamic scroll bar with a steep downstream side almost
perpendicular to the main channel flow direction (Jackson,
1976). In contrast to these dynamic scrolls, a more stable type
of scroll bar may form from deposition of fines from suspen-
sion (Nanson, 1980).
Active scroll bars form when accumulation space is created
in the inner meander by outer bank erosion. This explains their
absence in rivers that are stable due to low flow energy or
human interference. Scroll bars develop a wider spacing and
form more frequently as the rate of erosion of the opposed
outer channel bank increases (Hickin and Nanson, 1975).
Accretion of the inner bend – or point bar – and formation of
successive scroll bars often but not always results in a charac-
teristic ridge and swale topography. The curvilinear ridges of
this topography mark time lines of the accretion by scrolls of
the upper point bar (Tooth et al., 2008).
Chute bars are common in braiding rivers and also occur in
meandering rivers. Chute bars are horse-hoe-shaped lobes
formed at the downstream end of a chute or chute channel that
crosses a braid bar at peak discharge. The flow first converges
into the chute, and diverges past the chute lobe on both sides,
which therefore aggrades (Ferguson et al., 1992). Chute chan-
nels and chute bars are also common on inner bends of
Copyright © 2010 John Wiley & Sons, Ltd.
meandering rivers with relatively high specific stream power
values at peak discharge (McGowen and Garner, 1970;
Gustavson, 1978; Blacknell, 1982; Bridge et al., 1988; Field-
ing and Alexander, 1996; Bartholdy and Billi, 2002; Kemp,
2004). In addition, they are known from several less energetic
river systems, as relict structures of former braided patterns
(Marston et al., 1995).
When chutes end in chute bars they do not count as braids
under the definition of braiding and braiding intensity. Only
when the growth of a chute results in a chute-cut off is a new
braid formed. Indeed, experiments have shown that chute
cutoffs indicate the transition from meandering to braiding
(Friedkin, 1945). Chute cutoff is also an important process in
the initiation of active braiding (Ashmore, 1991).
Most studies that focus on chute bars do not mention
dynamic scroll bars, implicitly suggesting that they are not
present and that chute bars would exclude their formation
(McGowen and Garner, 1970; Nanson, 1980; Blacknell, 1982;
Brierley and Hickin, 1991; Kemp, 2004). However, several
investigations report the occurrence of both types of bars on
one and the same point bar (Lewin, 1970; Bridge et al., 1988;
Fielding and Alexander, 1996). For instance, 18th-century
maps of the Rhine near the Dutch–German border show that
this river had scroll bars as well as chute bars in association
with eroding banks in the outer bends. Shortly after that it was
engineered into a stable navigation channel devoid of emer-
gent bars at low flow (Figure 3).
The configuration of scroll and chute bars depends on the
curvature of the main channel in the Allier River. In wide
bends chute bars migrate downstream to about halfway the
point bar. In more tightly curved bends dynamic scroll bars
generally become wider and chutes tend to crosscut most of
the point bar. Chute bars sometimes merge with scroll bars
(Figure 4), and chute bars appear to develop and reactivate
relatively infrequently, depending on the recurrence of high
flood events (van den Berg and Middelkoop, 2007).
To summarise, the available evidence suggests that chute
bars and chute cutoffs are related to the transition between
meandering and braiding at relatively high stream power just
below the discriminator of braided rivers (Figure 5). Chutes are
therefore expected to overlap with weakly braided rivers.
Scroll bars, on the other hand, are associated with actively
meandering rivers at all stream powers below the discrimina-
tor of braided rivers and above that of stable rivers. Thus the
energy-based classification of river patterns can be detailed
further with discriminators between immobile rivers without
bars (<wia; Equation (7)), meandering rivers with scroll bars
(<wsc), meandering rivers with chute bars (<wab; Equation (3))
and rivers with braid bars (>wbm). Stressing again that the
discriminators in this paper are not hard thresholds but indi-
cators of transitions (‘transitionators’), we suggest a new dis-
criminator for meandering rivers with scroll bars and
meandering rivers with chute bars (and scroll bars) in between
Equations (3) and (7):
900 0⋅42
ω sc = 0⋅42
D50 ≈ 285D50 (8)
10
where the subscript sc indicates the discrimination between
scrolls and chutes.
Theoretical prediction of bar pattern
Bar theory (Struiksma et al., 1985; see also Seminara and
Tubino, 1989) predicts the existence of bars and bar regime,
Earth Surf. Process. Landforms (2010)
 RIVER CHANNEL AND BAR PATTERNS EXPLAINED AND PREDICTED 5

Figure 3. The River Rhine at the German–Dutch border in 1751 had several well-developed scroll bars and chutes (flow from left to right, location
127 in online supplementary material). Red C indicates a chute and chute bar; red S indicates scroll bars. In the middle and right part of the map
notes dating 1754 were drawn, demonstrating that the river at that time was still able to erode its banks. This figure is available in colour online at
wileyonlinelibrary.com

1 km
Channel bank
escarpment

Scroll bar front

Chute bar front

Transitional bar front

Low stage flow
direction
High stage flow
direction on pointbar

Figure 4. Example of occurrence of chute and dynamic scroll bars in gentle and tight bends of a high-energy meandering river with a mixed
sand–gravel bed. In tight bends transitional chute/scroll bars may form (after van den Berg and Middelkoop, 2007). Allier River in 1997 near
Bressoles, France (location 117 in online supporting information). Aerial photograph source (infrared false colour): Inventaire Forestier National,
Lyon, France. This figure is available in colour online at wileyonlinelibrary.com

Copyright © 2010 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms (2010)
6 M. G. KLEINHANS AND J. H. VAN DEN BERG

transverse and longitudinal slopes and because of spiral

A flow in bends. The steady bed topography in river bends can
be understood as a combination of a transversely sloped
bed depending on the local channel curvature and a pattern
of steady alternate bars induced by upstream variations (or
perturbations) in channel curvature. Struiksma et al. identified
four characteristic length scales in the linearised equations
for the steady alternate bars: namely the adaptation length
of flow lw, the adaptation length of a bed disturbance ls,
the wavelength of the bar Lp and the damping length of the
bar LD.
The adaptation length of flow lw (m) is given as

B C 2ha
λw =
2g (9)

where ha = actual mean (measured) water depth (m) and

C = the Chézy friction coefficient ( m s−1 ), derived from the
data as

u
C=
RS (10)

C
where u = cross-sectionally and depth-averaged flow velocity
(m s-1), S = channel gradient and R = hydraulic radius (m), here
calculated as

Waha
R=
Wa + 2ha (11)

where Wa = actual (measured) channel width (m). With speci-

fication of water depth or velocity, the missing parameter
Figure 5. Energy-related continuum of bar types in single-thread
follows from continuity Q = uhaWa, where Q = flow discharge
channel patterns. From high to low stream power. (A) Meandering with (m3 s-1, choice discussed later).
chute bars. Sheep River at Black Ranch, Alberta, Canada (Kellerhals The adaptation length of a bed disturbance ls (m) is
et al., 1972), location 51 in online supporting information). (B) Mean- calculated as
dering with scroll bars. Assiniboine near Portage La Prairie, Manitoba,
Canada (Rannie, 1990); location 96 in online supporting information).
(C) Immobile sinuous channels. Nqoga south of Omdop airstrip, 2
ha ⎛ Wa ⎞
Botswana (Tooth and McCarthy, 2004); location 120 in online sup- λs = f (θ )
π 2 ⎝ ha ⎠ (12)
porting information. This figure is available in colour online at
wileyonlinelibrary.com

where the magnitude of the transverse slope effect is calcu-

lated from an empirical function (Koch and Flokstra, 1981;
given flow conditions. The theory of Struiksma et al. (1985) is Talmon et al., 1995):
valid for forced alternate bars, where ‘forced’ means that bars
are fixed in location to bends or other perturbations in the
D 0⋅3
f (θ ) = 9 ⎛ 50 ⎞
channel boundary, in contrast to free bars that may migrate.
θ
The theory of Struiksma et al. (1985) can also easily be ⎝ ha ⎠ (13)
adapted to predict whether multiple bars can exist across the
channel; that is, whether it braids. Crosato and Mosselman
(2009) derived a predictor for the number of bars from the where q is the non-dimensional shear stress (Shields number)
theory of Struiksma et al. (1985), which is sketched below. Free defined as
bar theory gives similar predictions of bar regime (Marra,
2008; Crosato and Mosselman, 2009) (though not of bar
dimensions, but that is not our objective here), so that we can τ
θ=
apply the theory given below to all alluvial rivers. ( ρs − ρ ) gD50 (14)
The Struiksma et al. (1985) theory is based on interaction
between flow and a deformable sediment bed. The direction
of sediment transport may differ from the direction of in which r and rs are the density (kg m-3) of water and sedi-
depth-averaged flow because of gravitational effects on ment, respectively. The shear stress t (Pa) is calculated as

Copyright © 2010 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms (2010)
 RIVER CHANNEL AND BAR PATTERNS EXPLAINED AND PREDICTED 7

τ = ρ gRS (15) mode m = 3, braiding index Bi = 2

Often f (θ ) = αθ θ is used to describe or numerically

model bar regime, dimensions and dynamics, where aq is used
for calibration. This indicates that the transverse slope-related
part of the theory is rather uncertain, so that the predictions of
the theory are uncertain as well.
Intuitive understanding of these theoretical length scales is
offered as follows. Consider a long straight channel with per-
fectly uniform steady flow which suddenly enters a bend. This
bend acts as a perturbation to the flow: as momentum is
conserved, the flow is directed onto the outer bend, forcing the mode m = 2, braiding index Bi = 1.5
water surface to rise at the outer bank. This additional pressure
causes a spiral flow to set up that, near the bed, is directed
towards the inner bend. This flow pattern does not appear
instantaneously but develops asymptotically towards equilib-
rium, which is characterised by an adaptation length lw at
which about 63% of the adaptation has been accomplished. In
response to the flow pattern the bed deforms through sediment
transport. As the near-bed flow is directed slightly towards the
inner bend, sediment transport is directed slightly inwards as unstable, growing, m = 1, Bi = 1
well and a bar is built up in the inner bend. On the resulting
transverse bed slope, gravity opposes the inward movement of
sediment to some extent. The bed cannot adapt immediately
downstream of the bend (or there would be a ridiculous dis-
continuity in the bed surface) but adapts asymptotically, char-
acterised by the adaptation length ls.
Non-dimensional bar period (or wavelength) Lp (m) is
calculated by underdamped

2πλw 1 λ λ 2 n−3
( )
2
= ( n + 1) w − ⎛ w ⎞ − (16)
Lp 2 λs ⎝ λs ⎠ 2

where n = the degree of nonlinearity of sediment transport
versus depth-averaged flow velocity (qb = f(un)). For a classical
bed load transport predictor such as Meyer-Peter and Müller
(1948), n = 3 for high Shields numbers and increases to infinity
towards the critical Shields number for sediment motion. We
choose n = 4 for sand-bed rivers and for gravel-bed rivers
n = 10 (following Crosato and Mosselman, 2009) as gravel is
closer to the threshold of motion, so that the nonlinearity is
stronger. The effect of this choice will be assessed later.
Non-dimensional damping length Ld (m) of the bars is
calculated by
λw 1 ⎛ λw n − 3 ⎞
= − (17)
Ld 2 ⎝ λs 2 ⎠
The theory predicts whether forced bars dampen out in less
than one bar length (overdamped regime) or over longer dis-
tance, so that multiple bars along the river may exist (under-
damped regime) or excite (excitation regime, Ld < 0) (Figures 6
and 7). As shown above, this characteristic of bars is a function
of the non-dimensional interaction parameter (IP):
λ < IP <
IP = s (18)
λw
which depends strongly on width–depth ratio, and weakly on
hydraulic roughness and sediment mobility. For narrow and
deep channels the bars are overdamped (see Figure 6 bottom
and Figure 7). For a local perturbation this results in the dis-
appearance of the bar within a short distance downstream of
overdamped
Figure 6. Definition of bar regime and bar mode (after Parker,
1976; Struiksma et al., 1985; Mosselman et al., 2006; Crosato and
Mosselman, 2009).
the perturbation. For a bend the transverse slope adapts to the
equilibrium transverse slope within a short distance. Over-
damping occurs for
2
IP ≤ (19)
n + 1+ 2 2n − 2
as can be derived by equating lw/Ld = 0 in Equation (17) and
solving the resulting quadratic equation for the case of IP < 1.
For channels of intermediate width–depth ratio the bars are
underdamped:
2 2
(20)
n + 1+ 2 2n − 2 n−3
where the right-hand side was derived by equating lw/Lp = 0
and solving Equation (16) for IP. Underdamping leads to over-
deepening of the outer-bend pool and associated enhance-
ment of the bar in the inner bend just downstream of the
entrance to the bend or other perturbations (such as sudden
widening, narrowing or bank irregularities). For very wide and

Copyright © 2010 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms (2010)
8 M. G. KLEINHANS AND J. H. VAN DEN BERG

width−depth ratio further adaptation of the deformable bed downstream, and

so on. In this condition the bed deformation grows in the
10 25 50 75 100
3 downstream direction. The resulting bed topography ranges
A. Sand λw/Ld from pronounced finite-amplitude alternating bars to braided
2.5 channel patterns. For lower IP the bed deformation dampens
2πλ /L out downstream of the perturbation (Mosselman et al., 2006).
w p
2 λ = 111 m Note that the linear stability analysis of Struiksma et al. (1985)
w
h = 3.0 m is valid for bars of small amplitude. For larger amplitudes
p

approaching the water depth, nonlinear effects become

λ /L , 2πλ /L

1.5 C = 28
w

D = 0.34 mm important.
50 The theory presented by Struiksma et al. for mode 1 bars can
1
θ = 1.9 easily be adapted to higher-mode bars (see also Parker, 1976)
d

0.5 n=4 through

w

underdamped
overdamped

( )fθ
0 h 1 Wa 2
excitation λs = ( ) (22)
π2 m h
−0.5

−1 where m = mode as defined in Figure 6. For any mode, the

−1 0 1 threshold between underdamped and excited is found at
10 10 10
IP = λs/λw
2m2
width−depth ratio IPue = (23)
n−3
10 25 50 75 100
3
B. Gravel λ /L A river is predicted to braid if the actual IP for higher modes is
w d
2.5 above this threshold.
2πλ /L The range of width–depth ratios where underdamped bars
w p
2 λ = 59 m form is mostly determined by the nonlinearity of sediment
w
h = 1.6 m transport n. Figure 7 shows that this range is much narrower for
p

gravel-bed rivers than for sand-bed rivers. For width–depth

λ /L , 2πλ /L

1.5 C = 28
w

D = 33 mm ratios of 50 and higher the bar regime in gravel-bed rivers is

50 excitation, and underdamped for mode 5 (Bi = 3). We can now
1
θ = 0.1 hypothesise that braided rivers are more numerous in gravel
d

0.5 n = 10 than in sand if we assume that width–depth ratio is indepen-

w

excitation B =3

dent of bed particle size.

underdamped
overdamped

i
excitation

0
Bar mode theory of Crosato and Mosselman (2009)
−0.5 Crosato and Mosselman (2009) derived a mode predictor from
the theory of Struiksma et al. (1985), which compared favour-
−1 ably with observed modes in a number of braided rivers in
−1 0 1
10 10 10 sand and gravel:
IP = λ /λ
s w
0⋅17 g ( n − 3) Wa3S
Figure 7. Regimes of mode 1 bars generated at a perturbation for m2 =
ρs − ρ CQ (24)
sand-bed rivers (A) and for gravel-bed rivers (B). Most of the differ- D50
ence between sand-bed and gravel-bed rivers is caused by the non- ρ
linearity of sediment transport (n). Width–depth ratio is indicated
at the top of the figures. Dashed vertical lines indicate theoretical where the relation between mode m and braiding index Bi is
thresholds of bar regime. This figure is available in colour online at
defined as
wileyonlinelibrary.com

m −1
Bi ≡ +1 (25)
shallow channels the bars become unstable and theoretically 2
grow in height downstream of the perturbation. This excitation
occurs for Herein, a river is considered single-thread for Bi ⱕ 1·2,
moderately braided for 1·2 < Bi < 3 or braided for Bi ⱖ 3,
where Bi = number of active channels across the river width Wa
2
IP ≥ (21) during channel-forming discharge (Figure 6) (Egozi and
n−3 Ashmore, 2008).
The threshold from underdamped to excited bars (for a given
Intuitively this can be understood as a positive feedback mode) is here interpreted as the transition to dominant pres-
between the flow and the deforming bed. Both flow and bed ence of these bars in reality. These thresholds are mathemati-
adapt to the perturbation, but over a different length. If the flow cally identified as hard thresholds, but in practice merely give
has already adapted to the upstream perturbation while the reference values in the continuum from overdamped to under-
bed has not yet adapted to the same perturbation, then the damped to excited bars (Figure 7). The theory is strictly valid
flow will again adapt to the changing bed, which will cause for forced bars, so that underdamped forced bars of mode 1 are

Copyright © 2010 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms (2010)
 RIVER CHANNEL AND BAR PATTERNS EXPLAINED AND PREDICTED
equivalent to alternate bars or scroll bars in meander bends
where the bend provides the forcing. If theory predicts
1·2 < Bi < 3 then this is interpreted as chute cutoffs or weakly
braided.
Furthermore, the theory is valid for bars of limited amplitude
where interaction with the water surface is limited because of
the linearisation of the equations (as also clearly stated by
Crosato and Mosselman, 2009). Our preliminary results of
numerical modelling show that the initial bars and channels of
small amplitude are all of similar size (see Figure 7 in Klein-
hans, 2010). Thus the theory predicts correctly that the chan-
nels in higher-mode bar patterns are of equal importance, even
though this is not the case in real rivers where the instability of
the bifurcations leads to one or a few dominant channels
(Bolla Pittaluga et al., 2003; Kleinhans et al., 2008; Egozi and
Ashmore, 2009). Consequently, the theory predicts too many
braids for strongly braided rivers, and does not make any
prediction on the importance of individual channels.
It is important to realise that the physics-based method is not
really predictive because it requires knowledge of the actual
channel width, depth and channel gradient. These variables
are strongly coupled to the river pattern itself through the
feedback between bar pattern, bank erosion and style of flood-
plain formation. However, it provides a partial explanation,
based on physics, for the empirical method, as bar pattern is an
important aspect of channel pattern.
Data Collection and Analysis
Selection of equilibrium rivers
The present analysis refers to natural rivers in unconfined
alluvium that are in a dynamic equilibrium condition over a
length scale of a few tens of meander lengths (hundreds of
widths). The selection criteria of rivers were the same as in van
den Berg (1995):
• Q > 10 m3 s-1;
• perennial flow regime;
• no valley constraints;
• no dams, jetties, groynes, sills, bank protection;
• no artificial cutoffs;
• no roads or rural areas bordering on the channel;
• no clear bed degradation (e.g. no clear terraces);
• no sign of strong modification of the hydrological regime
(e.g. bars overgrown with forest or clearly underfit patterns
such as presence of small-wavelength meanders in long
bends).
These are all related to unwanted effects on the width of the
channels as this would affect the bar and channel pattern and
their natural dynamics. We stress that the criteria for river
selection and classification outlined above were strictly fol-
lowed irrespective of their potential specific stream power,
which was calculated only after selection.
Based on Google Earth images (accessed February–July
2009) we classified observed rivers by active braiding index,
presence of scroll bars, chute bars and scrolled point bars.
Single-thread river patterns were subdivided into meandering
with chute bars, meandering with scroll bars, and stable
straight or sinuous rivers (Figure 5). We characterised braiding
intensity by the braiding index, defined as the average cross-
sectional number of active, unvegetated or barely vegetated
braids (Bi), which gives the best combination of rapid measure-
ment and precision for Google Earth (Egozi and Ashmore,
2008). We counted active braids visually using indications
such as trees falling over into the channel in outer bends,
unvegetated higher parts of point bars and administrative
Copyright © 2010 John Wiley & Sons, Ltd.
9
borders that no longer follow the river. In multi-thread systems
highly braided and moderately braided were distinguished
above and below Bi = 3. To minimise interpretation differences
this visual classification was done by only one operator
(JHvdB). The data and geographical coordinates are provided
in the online supplementary information.
The van den Berg data were revisited and 74 out of 143
rivers were removed. In most cases the reason for removal was
an obviously underfit river pattern. Information gathered from
various sources for most cases revealed that this resulted from
various man-induced causes, such as water extraction or dam
building. Most removed cases were straight rivers with a sinu-
osity P < 1·3. The cause of the straight channel pattern usually
was a narrow valley, either by being confined between rocks
or by incision of the river. Van den Berg (1995) found that
straight rivers were not classified well but occurred in both
the meandering and braided fields of the diagram. The reanaly-
sis now explains this: those straight rivers did not meet the
criteria. Their stable occurrence is typical for moderately
entrenched or valley-confined rivers of moderate flow energy
as described by Nanson and Croke (1992) and Rosgen (1994).
In an unconfined setting these rivers would have developed
meandering or braiding patterns. Our focus on alluvial rivers
without bed degradation also led to exclusion of the incising
rivers used by Bledsoe and Watson (2001). Many of the rivers
selected by Crosato and Mosselman (2009) may have been
suitable to test bar pattern but were unsuitable for our purpose
of river pattern analysis as they did not meet our criteria.
Collection of new data and analysis
A literature survey resulted in an extension of the dataset by 58
rivers to a total of 127 rivers. Required variables were mean
annual flood discharge (Qaf) and bankfull discharge (Qbf),
median particle size (D50) of the channel bed, valley slope (Sv)
or sinuosity (P) and channel slope (S, where by definition
P = S/Sv). Data of actual width (Wa), depth (h) and velocity (u)
during conditions with bankfull discharge Qbf or mean annual
flood discharge Qaf were collected as far as available. Data
were converted to SI where necessary.
The data are presented as online supporting information. A
summary of the data is shown as probability density distribu-
tions for several variables in Figure 8. The data are log-
normally distributed for most variables, except particle size
and the related Shields number, which is explained by the
well-known sparsity of rivers with pea gravel (2–8 mm) as bed
sediment. This demonstrates that the data are well representa-
tive of rivers from fine sand to coarse gravel, from small stream
to major river and from bedload to suspended bed material
load-dominated river.
Mean annual flood discharge was used as effective channel-
forming discharge. Reported bankfull discharge was used
instead if mean annual flood discharge was not found in the
literature. This choice was made for several reasons. Mean
annual flood is less dependent on channel pattern than bank-
full discharge. Furthermore, bankfull discharge reported in the
literature is less reliable, as there are a large number of defi-
nitions (Williams, 1978) and in the data sources generally it is
not clear what definition was used. Formative discharge can
also be defined as the discharge at which on average the same
annual bed sediment load is transported as by the full prob-
ability distribution of discharge. In fact, this also is not a
completely pattern-independent measure of discharge, as it is
influenced by the width and depth of the channel. The
maximum of average annual bed material load commonly
coincides with the water level at which the flow resistance per
Earth Surf. Process. Landforms (2010)
10 M. G. KLEINHANS AND J. H. VAN DEN BERG

0.99
0.99

0.95 0.95
0.9 0.9

0.75 0.75

0.5 0.5

0.25 0.25
0.1 0.1
0.05 0.05

0.2 2 20 10 30 100
bed particle size D50 friction C

0.99
0.99

0.95 0.95
0.9 0.9

0.75 0.75

0.5 0.5

0.25 0.25
0.1 0.1
0.05 0.05

10 100 1000 10000 0.1 1 10

discharge Qma mobility θ

0.99 0.99

0.95 0.95
0.9 0.9
0.75 0.75

0.5 0.5

0.25 0.25
0.1 0.1
0.05 0.05
Figure 8. Cumulative distributions of
variables representing the dataset. Full
0.0001 0.001 0.01 1 10 100 dataset provided in online supporting
channel gradient S stream power ω information. This figure is available in
pv colour online at wileyonlinelibrary.com

unit of discharge is minimal, which is the bankfull level. As a
consequence, the formative discharge changes if the width or
depth of the channel is changed. The dimensions of the
channel also determine how a flood wave is transformed in a
downstream direction, so that downstream discharge is not
independent from channel characteristics. Yet the least pattern-
dependent formative discharge measure is a frequency-based
hydrological variable: the mean annual flood discharge.
The discriminators Bi = 1·2 and Bi = 3 as well as Equation
(23) for these values of Bi were compared to the data by
calculating Equations (18) and (25) for all rivers where a depth
or velocity was available. The predictive capacity of the bar
pattern from theory was then compared to the empirical dis-
criminators for specific potential valley stream power.
To assess the nonlinearity of sediment transport, we also plot
the criteria for bedload and suspension (Soulsby, 1997). These
must be converted from non-dimensional shear stress (Shields
number) to stream power. From their definitions and the law of
Chézy, it can be derived that
τ 15⋅ C [(ρs − ρ ) gD50 ]15⋅
ω= = θ 15⋅ C (26)
ρg ρg
For our dataset the average Chézy number is 28 and the data
confirm no correlation between Chézy number and water
depth or bed sediment size. Typically, rs = 2650 kg/m3,
r = 1000 kg/m3 and g = 9·81 m/s2, so that w ª 0·283t1·5.
Inapplicability of the prediction methods to
anabranching, anastomosing and wandering rivers
The anabranching river pattern is often added as a separate
category or higher level to channel pattern classifications

Copyright © 2010 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms (2010)
 RIVER CHANNEL AND BAR PATTERNS EXPLAINED AND PREDICTED
(Schumm, 1968; Rust, 1978; Alabyan and Chalov, 1998;
Latrubesse, 2008) (Figure 1). It is defined as a system of mul-
tiple channels characterised by vegetated or otherwise stable
alluvial islands (floodplain) that divide flows at discharges up
to nearly bankfull (Nanson and Knighton, 1996). Here we
argue that the higher-level multi-thread property is unrelated
to stream power.
Anastomosing rivers are anabranching systems with low
stream power insufficient for significant channel mobility, so
that individual channels may be straight or sinuous but not
meandering (Makaske et al., 2002). A possible explanation of
anastomosis is that it is caused by avulsion driven by upstream
sediment overloading (Abbado et al., 2005). More precisely,
the aggradation of the floodplain and levees is exceeded by the
aggradation of the channel bed. The bed sediment cannot be
accommodated in extending point bars because the river is not
strong enough to erode banks. Hence the river is forced to
avulse by channel aggradation (Makaske et al., 2009; Klein-
hans et al., work in progress). This strongly suggests that the
anastomosing river pattern is a disequilibrium pattern, which is
outside the scope of this paper.
Some 40 years ago wandering rivers were added to the
classical meandering-braided continuum (Neill, 1973; Church
and Rood, 1983). The wandering river pattern, sometimes
called island-braided (e.g. Ward et al., 2002), has a compli-
cated channel planform with long sections exhibiting
multiple-channel anabranches surrounding semi-permanent
islands supporting mature forests, separated by single-channel
sections Church and Jones (1982). Wandering emphasises the
activity of the channels and originally the term was used for a
transitional pattern between meandering and braiding (Neill,
1973; Church and Rood, 1983). In Figure 1 the wandering
pattern is classified accordingly. Unfortunately, since then the
term has been used in more than one sense in the literature. In
the past decade or so the emphasis in the definition of wan-
dering seems to shift from channel mobility to channel pattern;
and several completely different causes may result in a similar
pattern of anabraching reaches connected by single-thread
reaches. For instance, Burge and Lapointe (2005) demon-
strated that avulsions that create new anabranches in many
cases are forced by beaver dams or ice jams. Ward et al. (1999)
described wandering systems that evolved from coagulating
tail bars generated in the wake of large woody debris. Such
‘wandering’ patterns can form under much lower energy con-
ditions than the transitional meandering-braided systems for
which the term was reserved in the early days. A wandering
pattern may also be created when the effective discharge
reduces, for instance due to the construction of storage reser-
voirs. As a result, braid bars may become stable, covered by
dense vegetation, and coalesce by the siltation of smaller braid
channels. Examples of such systems misclassified as (in equi-
librium) ‘wandering’ are indicative of a future transition to
meandering (as probably occurred in the Allier River, as indi-
cated on historical maps in the local archives). Thus wandering
is the result of either an unusual non-fluvial forcing mecha-
nism or disequilibrium. Such mechanisms are not included in
the river pattern prediction method presented here. Also, the
prediction method refers to single-channel systems. Therefore
we do not present wandering rivers as a separate river pattern.
However, it has been made clear that without such external
forcing mechanisms wandering rivers would be braided or
meandering near the meandering–highly braided discriminant
(Nanson and Knighton, 1996; Burge and Lapointe, 2005).
Anabranching is common in the largest rivers of the world
with Qbf > 17 000 m3 s-1, such as the lower Brahmaputra, the
Congo and the Amazon. Latrubesse (2008) suggested that these
systems plot in a very restricted area of Figure 9, with stream
Copyright © 2010 John Wiley & Sons, Ltd.
11
power values 3 < wpv < 25 W m-2 and grain sizes 0·1 < D50 <
0·5 mm. There are exceptions to this rule, however, such as
the Brahmaputra at Dibrugarh, Assam, India, with Qbf =
34 000 m3 s-1 (Chitale, 1973; location 19 in online supporting
information), which has a highly braided pattern and plots
accordingly in the stability diagram. Another hypothesis for the
anabranching of the largest rivers is superposition of a bankfull
discharge-related pattern on a large-scale flood-related pattern
(Alabyan and Chalov, 1998). For instance, the Brahmaputra/
Jamuna River shows intricate anabranching or braiding during
low flow, but a mode 2 (Bi ª 1·5) river during a high monsoon
flood (Thorne et al., 1993). We can speculate that large floods in
these large rivers are strong enough to overcome floodplain
sediment and vegetation strength, so that the floods deform
floodplains as if they were channels, but this is beyond the
scope of this paper. We conclude that the existence of rivers
with multiple channels, separated by floodplain, is not clearly
empirically or theoretically related to stream power.
Results
Empirical prediction of bar and channel pattern
Four stability fields can be distinguished in the new empirical
diagram: rivers with stable channels, meandering rivers with
scroll bars, meandering rivers with scroll and chute bars as
well as moderately braided rivers, and finally braided rivers
(Figure 9). Moderately braided rivers and meandering rivers
with chute bars plot within the same field (Figure 9). This is not
a surprise, as a small increase of channel-gradient-related
stream power may increase the activity of chutes in meander-
ing rivers, causing chute cutoffs which change the river pattern
to moderately braided (see Discussion). Note that if actual
stream power based on actual channel slope and measured
water depth were used then the channel patterns would not be
separated at all.
The discriminators indicate which pattern occurs above the
line, but do not indicate that a lower-energy pattern does no
longer occur above this line. For example, scroll bars occur at
wpv ⱖ wia (Equation (7)), while rivers without bars may also
occur in this field. Likewise, braided rivers occur for wpv ⱖ wbm
(Equation (3)) but hardly below this discriminator, while some
chutes and moderately braided rivers plot both above and
below this same discriminator. The present discriminators are
here reinterpreted as lower thresholds rather than discrimina-
tors, but it is again stressed that these lines represent an esti-
mate of probability of existence – not a hard threshold. Table I
indicates that most rivers correctly plot above the applicable
threshold.
The original discriminator between meandering and braid-
ing (Equation (3)) must be reinterpreted. This line was thought
to represent the 50% probability line separating stability fields
of dominantly braiding or meandering (Bledsoe and Watson,
2001). The present distinction between moderately braided
(Bi < 3) and braided (Bi ⱖ 3) rivers revealed that only braided
rivers plot above the discriminator. Furthermore, only very few
meandering rivers of the new dataset are located above the
discriminator (Figure 9), which invalidates the 50% interpreta-
tion. Therefore, Equation (3) is reinterpreted as a lower ‘thresh-
old’ for highly braided systems.
Inactive channels are not necessarily straight. For example,
the highly sinuous Barwon River (near Walgett, Australia, loca-
tion 113 in online supporting information) has no significant
bank erosion or lateral migration. Classifying this river as mean-
dering based on sinuosity would contradict the low stream
power value previously associated with straight rivers (Leopold
Earth Surf. Process. Landforms (2010)
12 M. G. KLEINHANS AND J. H. VAN DEN BERG

3
10 A

Eq.3
potential specific stream power ωpv (W/m2)

ed
braid Eq.8
highly
2
10
ande ring,
ed a nd me
braid utes
mod. nd ch Eq.7
o l l s a
scr
olls
g w i th scr
1
derin
10 mean
bars
m o b ile, no
im
la terally

0
10
no bars
scroll bars
chute bars
moderately braided
braided
scrolled point bars
−1
10
−4 −3 −2 −1
10 10 10 10
median grain size D50 (m)

3
10 B

Eq.3
potential specific stream power ωpv (W/m2)

2 Eq.8
10

Eq.7

1
10
d
an
rt

0
d

10
po
de
ns
en

P≤1.1
tran nded

tra
sp

1.1<P≤1.3
rt

su
spo

n 1.3<P≤1.5
loa

io
pe

ot
d
xe

1.5<P≤1.8
sus

m
d

no
be
mi

P>1.8
−1
10
−4 −3 −2 −1
10 10 10 10
median grain size D50 (m)

Figure 9. Patterns of equilibrium alluvial rivers plotted with the potential specific stream power related to valley gradient and predicted width. (A)
Data subdivided by bar pattern. (B) Data subdivided by sinuosity. Equation (3) (top) is the upper limit of meandering and lower limit of highly
braided. Equation (8) (middle) is the lower limit of meandering with chute bars or moderately braided. Equation (7) (bottom) is the lower limit of
meandering with scroll bars. Dotted lines are approximate criteria for beginning of motion and beginning of suspension. This figure is available in
colour online at wileyonlinelibrary.com

Copyright © 2010 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms (2010)
 RIVER CHANNEL AND BAR PATTERNS EXPLAINED AND PREDICTED 13

Table I. Classification of channel pattern by potential specific stream 10

power above the lower thresholds (but not below the higher threshold) single mod. braided A. All data

mb
thread braid.

/ω
Correctly Incorrectly

pv
Class classified classified

relative stream power ω

braided
1
No bars wpv ⱖ wia 11 11 mod. braided
Scroll bars wpv ⱖ wia 50 1 & meandering,
scrolls&chutes
chutes wpv ⱖ wsc 25 2
Moderately braided wpv ⱖ wsc 10 0 meandering,
Braided wpv ⱖ wbm 20 2 scroll bars
0.1
no bars,
laterally stable
and Wolman, 1957), whereas classifying this river as a single-
thread river devoid of any scroll bars and without significant
bank erosion agrees with the low stream power value associ- 0.01
ated here with immobility (Makaske et al., 2002). Thus immo- 1 10 100
bile sinuous channels may seem to meander at first sight, but predicted braiding index Bi
this is merely a relic (see Ferguson, 1987, p. 153) or is caused by
10 no bars
non-fluvial antecedent relief (as in Kleinhans et al., 2009). B. Gravel−bed rivers

mb
scroll bars
Vegetation significantly affects channel mobility in a similar

/ω
chute bars
way to engineering structures, hard rock and strong clay layers

pv
mod. braided
on the point bar. Several meandering rivers with scroll bars,

relative stream power ω

braided braided
but without chutes and chute bars, plot near or even slightly 1 mod. braided
above the meandering stability field. These examples refer to & meandering,
tortuously meandering systems in the Brazilian tropical forest, scrolls&chutes
such as the Jurua (locations 58 and 59 in online dataset). The meandering,
high sinuosity of these rivers is caused by the dense vegetation scroll bars
on the point bar that resists cutoff by chuting (Baker, 1978). 0.1
no bars,
This suggests that a clearing of the forest for some of these laterally stable
cases might result in a change of channel pattern into a less
tortuous meandering system with chute bars, or – in view of under− excitation, B =3
i
damped excitation, B =1
the high potential energy values – even into a moderately i
0.01
braided system. Immobile channels may occur at relatively 0.1 1 10 100 100010000
high stream power values in the case that roots of the bank interaction parameter λ / λ
vegetation reach below the channels floor. Channels with s w
Q < 15 m3 s-1 in a forested area were also immobile for this 10
C. Sand−bed rivers
mb

reason (Beechie et al., 2006).

/ω

Lines indicating the threshold of motion and of suspension over− under−

pv

show that gravel-bed rivers generally occur close to the thresh- damped damp.
relative stream power ω

old of motion, whereas sand-bed rivers are dominated by braided

1 mod. braided
suspension. This confirms that the chosen degree of nonlinear-
ity n of sediment transport versus depth-averaged flow velocity & meandering,
scrolls&chutes
should indeed be higher for the gravel-bed rivers (n = 10) than
for the sand-bed rivers (n = 4), in agreement with Crosato and meandering,
scroll bars
Mosselman (2009). Note that this remains a simplification as 0.1
the sediment of gravel-bed rivers is commonly much more no bars,
poorly sorted than that of sand-bed rivers, so that the begin- laterally stable
ning of motion will be more gradual and the mode of sediment excitation, Bi=3
transport mixed bedload and suspended load. excitation, B =1
i
As a test of whether the original discriminant (Equation (3) is 0.01
still valid for the present dataset we recalculated the power by 0.1 1 10 100 100010000
ordinary least square fitting of a power function through poten- interaction parameter λs / λw
tial specific stream power versus particle size for the rivers
with scroll bars, chute bars and moderate braiding. The power Figure 10. Comparison of empirical and theoretical predictions for
was 0·41, close to the original 0·42, which we take as cor- bar pattern, and, by implication, for channel pattern. Stream power is
roboration. This is further supported by Bledsoe and Watson normalised with the braided–meandering discriminator (Equation (3)).
(2001), who found the same power. We compared the stream Same legend as Figure 9a. (A) Comparison of stream power-based
power and mobility number approaches of van den Berg prediction and theoretically predicted bar mode (Equation (24)), here
expressed as active braiding index. (B) Comparison of empirical pre-
(1995) and Bledsoe and Watson (2001) but no difference in
dictions and theoretical bar regime (Equation (23)) for gravel-bed
classification was found as expected.
rivers. (C) Same as B, for sand-bed rivers. This figure is available in
colour online at wileyonlinelibrary.com
Theoretical prediction of bar pattern compared to
empirical prediction the empirical discriminators on one axis (Figure 10), the
trend of increasing stream power with particle size is
Here we compare theoretical and empirical predictions of removed by normalisation: the unit potential valley-related
bar pattern. To represent the position of rivers relative to stream power of all individual rivers is divided by the stream

Copyright © 2010 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms (2010)
14 M. G. KLEINHANS AND J. H. VAN DEN BERG
Table II. Classification of channel pattern by theoretical braiding
index Bi
Correctly Incorrectly
Class classified classified
Single-thread (no bars, scroll bars) Bi ⱕ 1·2 21 25
Chutes and mod. braided 1·2 < Bi ⱕ 3 20 3 predicted. If stream power were that large in reality because
Braided Bi > 3 9 0
power at the braiding threshold for the same particle size as
wpv ⱖ wbm.
The mode is straightforwardly interpreted as active braiding
index. Braided rivers are reasonably separated at Bi = 3 from
moderately braided rivers or meandering rivers with chute
cutoffs (Figure 10A). The latter predict many meandering
rivers to be braided or moderately braided or to have chute
bars that they do not have. The overlap between the various
river patterns indicates that the result would not improve for
different choices of threshold braiding index or sediment
transport nonlinearity. Table II indicates that most multi-
thread rivers are correctly classified, but that single-thread
rivers occur up to Bi ª 4. This agrees with the finding that
rivers with scroll bars and chute bars overlap in the empirical
stream power-based diagram. Note that the braiding index is
very much overpredicted (up to 100), in agreement with
Crosato and Mosselman (2009), because the theory is based
on linearised equations, whereas in reality the number of bars
decreases as they grow in height up to a significant fraction of
the water depth.
If the interaction parameter is considered, four bar regimes
are distinguished by three theoretical thresholds (Figures 10B
and 10C). The first is the transition from overdamped to under-
damped bars, indicative of the emergence of bars in very
narrow and deep channels. The second is the transition from
underdamped bars to excited bars of Bi = 1, indicative of pro-
nounced scroll bar formation. The third is the transition to
excited bars of Bi = 3, indicative of braiding. For gravel-bed
rivers, many meandering rivers with chute bars are theoreti-
cally predicted to braid weakly. We consider this correct as the
number of parallel active channels is larger than unity. For
sand-bed rivers, many channels without bars are indeed pre-
dicted to be overdamped, but many rivers with chute bars are
predicted in the underdamped regime, which we do not con-
sider correct.
Observations are in agreement with the prediction from
theory that braided rivers are more numerous in gravel than in
sand because the range of width–depth ratios where alternate
or low mode bars occur is narrower for higher nonlinearity of
sediment transport (n) (see Figures 9A, 10B and 10C).
In short, the theory shows the same trends as the empirical
predictors. The empirical predictors are more accurate, which
is attributed to the fact that they were based on the dataset.
However, the theoretical predictions have the advantage over
the empirical predictions that they use actual measured
channel widths rather than deliberately mispredicted widths as
in the empirical method. Despite this disadvantage for the
empirical method, it is, surprisingly, still better.
Discussion
Channel width prediction and channel pattern
The empirical method predicts the patterns well precisely
because the width is deliberately predicted incorrectly. This
Copyright © 2010 John Wiley & Sons, Ltd.
can be understood as follows. If actual width were used then
the stream power decreases for rivers where width was over-
predicted, and increases for rivers where width was underpre-
dicted, so that the discriminative power is entirely lost. In other
words, by using the pattern-independent width predictor the
stream power for rivers with high width–depth ratios is over-
the river were too narrow and deep, then the excess energy
would likely be spent on channel enlargement, in which case
the width–depth ratio would increase to the actual value. The
reverse would be the case for rivers where stream power is
underpredicted.
Bar theory, on the other hand, requires the actual observed
channel widths and channel gradient for good prediction. If
these are unknown an accurate prediction of width is required,
but this is not trivial. Only width predictors with a power larger
than 0·5 and information on the strength on the banks for
the multiplication coefficient would have predicted the
widths correctly, but then the prediction would no longer be
independent. More sophisticated channel width prediction
that includes bank strength (Eaton and Church, 2007) would
require information on bank or floodplain material and
vegetation, which depends on channel pattern, and also
requires an optimality assumption, which is controversial.
Such detailed prediction is beyond the scope of this paper.
The empirical method has been criticised by Lewin and
Brewer (2001, 2003) for its use of a regime-based estimate of
channel width rather than actual width, because this results in
incorrect values of the potential maximum of available energy.
Although the criticism is valid it is beside the mark, as the
computation of channel width aims at a pattern-independent
reference value for channel pattern, and precisely for this
reason intends not to predict actual width, which is pattern-
dependent (van den Berg, 1995).
To demonstrate the effect of the used predictor for channel
width we plot predicted against measured width (Figure 11).
Channel widths were predicted with Equation (2) and with the
channel width predictor of Parker et al. (2007). Both were
found to underpredict the width of many braided rivers in our
dataset and overpredict that of many meandering rivers. Parker
et al. (2007) also predicts channel depth. Although much scat-
tered, it overpredicts depth of many braided rivers and under-
predicts depth in meandering rivers. Thus the Parker et al.
predictors yield underpredicted width–depth ratios for rivers
that braid and overpredicted width–depth ratios for rivers that
meander.
Finally, we note that the effect of using different width
predictors for sand-bed and gravel-bed rivers is small. The
empirical channel pattern discriminators (Figure 9) were recal-
culated (not shown) using the predictor for reference width
(Equation (2)) with an average coefficient a = 4·0 rather than
two different ones for sand-bed rivers and gravel-bed rivers.
The effect was limited to a slight rotation of the data on the
centre of the graph, in agreement with the findings by Bledsoe
and Watson (2001). The power on diameter would change
from 0·42 to 0·33 approximately.
Sinuosity and stream power
Meandering patterns have been subdivided in the past by
sinuosity P (Schumm, 1963; Rust, 1978). It is a common
perception that sinuosity tends to increase in response to a
steepened valley gradient and vice versa (e.g. Ackers and
Charlton, 1970), because energy gradient is adapting to an
equilibrium between flow transporting capacity and sediment
supply. Evidence for this was found in a number of case studies
Earth Surf. Process. Landforms (2010)
 RIVER CHANNEL AND BAR PATTERNS EXPLAINED AND PREDICTED 15

A A
predicted channel width W (m)
no bars
−2
scroll bars 10
1000
r

chute bars
mod. braided

v
valley gradient S
braided
y=x
−3
10
100

−4
10
10

10 100 1000 1 1.5 2 2.5 3

observed channel width Wa (m) sinuosity P

no bars
B B
predicted channel width W (m)

1 scroll bars
10
bm
chute bars
1000 /ω moderately braided
braided
pv
relative stream power ω

0
10

100
−1
10

10
−2
10
10 100 1000 1 1.5 2 2.5 3
observed channel width W (m) sinuosity P
a
10 Figure 12. (A) Lack of correlation between valley slope Sv and sinu-
C osity P. (B) Complex correlation between relative potential stream
predicted channel depth h (m)

power wpv /wab and sinuosity P. This figure is available in colour online
at www.interscience.wiley.com/journal/espl

of alluvial rivers in plains subject to active tectonic movements

(Schumm, 1969; Burnett and Schumm, 1983; Schumm et al.,
2000). At first sight this might suggest that tortuous meander-
ing rivers in general would represent relatively high-energy
conditions.
1 Yet, other factors influence sinuosity. Vegetation (Baker,
1978), the percentage silt–clay in the bank material (Schumm,
1963) and channel bed grain size (Rosgen, 1994; Dade, 2000)
may all obscure or even counter valley gradient-related trends
found elsewhere. For instance, in river channels of the Oka-
vango wetlands, Botswana, which is also an area with differ-
ential tectonic movements, sinuosity decreases with increasing
1 10 valley gradient (Tooth and McCarthy, 2004). Also on intertidal
observed channel depth h (m) mud flats the sinuosity decreases with increasing gradient
(Kleinhans et al., 2009). As the size distribution and quantity of
Figure 11. Effect of hydraulic geometry on channel pattern. All data
sediment supplied to a river are different for each case, the
are shown. (A) Width prediction according to Equation (2) as in van
den Berg (1995). (B) Width prediction with the Parker et al. (2007)
mentioned equilibrium energy slope is different for each river.
predictor for gravel bed rivers. (C) Depth prediction using Parker et al. Therefore, specific downstream relations between sinuosity
(2007). This figure is available in colour online at www.interscience. and valley slope found in a number of river plains cannot be
wiley.com/journal/espl generalised. Our dataset supports this statement (Figure 12A).
Rosgen (1994) and Dade (2000) showed a clear inverse rela-

Copyright © 2010 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms (2010)
16 M. G. KLEINHANS AND J. H. VAN DEN BERG
1960 1980
Bressoles
N •
P = 1.53 P = 1.13
tion between channel slope and sinuosity, which might suggest
a similar relation between stream power and sinuosity. Our
data show, in contrast, that no such simple relation exists
(Figure 12B). Sinuosity is high for the range of stream power
where meandering prevails, and low for both low and high
stream powers where rivers are laterally stable or braided. This
corroborates that the inverse relation of channel slope with
sinuosity is a consequence of sinuosity, not a cause.
Moreover, P for a given reach of a river does not take a
characteristic, stable value but typically shows important fluc-
tuations in time and space. Very large changes in P may be
caused by series of meander cutoffs that often cluster in time
due to events such as large floods, vegetation mortality in
extreme weather, and human interference. For instance, P
fluctuated between 1·40 and 2·92 in a reach of the River Bolin,
UK, over a time period of 162 years (Hooke, 2004). In the
Allier, France, a cascade of chute cutoffs triggered by the
construction of a bridge and a large flood resulted in a tem-
porary reduction of the sinuosity from 1·53 to 1·13 and an
increase in braiding index (Figure 13). The rapid recovery to
P = 1·4 after this event illustrates that rate of change of sinu-
osity is strongly related to the velocity of channel migration
and thus to flow energy; but a flow energy relation with
sinuosity is not obvious from the data. This invalidates the
use of sinuosity-based classifications for meandering pattern
prediction.
By implication this also invalidates the use of channel
gradient-based classifications for prediction. Although there is
consensus that channel pattern is ultimately related to dis-
charge, gradient and the nature of the self-formed banks which
Copyright © 2010 John Wiley & Sons, Ltd.
1998
1 km
Figure 13. Aerial photographs of the
Allier River near Bressoles, France
(location 117 in online supporting
information). In 1980 several bends
were cut off by chutes, temporarily
reducing sinuosity and transforming
the river to weakly braided. Flow is to
the north. Rectangle indicates position
of Figure 4. Source of 1960 and 1980
photographs: Institut Géographique
P = 1.40 National, Paris, France.
determines channel width–depth ratio, Leopold and Wolman
(1957) and Eaton et al. (2010) use channel gradient; but
channel gradient depends on sinuosity and thus on pattern as
well (Figure 12). Therefore we use valley gradient as an inde-
pendent parameter, which leads to equal or better empirical
success (Figure 9).
Conclusion: A Bird’s Eye View of River
Channel Patterns
River channel patterns can be defined by bar pattern, number
of parallel channels, and whether the channels are separated
by channel sediment or floodplain. Bar and channel patterns
can well be discriminated and predicted with channel par-
ticle size and specific potential stream power calculated from
mean annual flood discharge, valley gradient and width pre-
dicted with a hydraulic geometry relation. Empirical power
functions of channel particle size are given as lower thresh-
olds for meandering rivers with scroll bars, for meandering
rivers with chutes or weakly braided rivers, and for braided
rivers.
Bar pattern is reasonably well predicted by forced-bar
theory, particularly the transition from single-thread (straight
and meandering) to multi-thread (chute cutoffs and braided).
Bar theory demonstrates the importance of actual channel
width–depth ratio and channel gradient. These are, however,
strongly pattern-dependent, so that bar theory is incomplete as
an independent predictor of channel and bar pattern. Rather,
bar theory explains why the deliberate misprediction of
Earth Surf. Process. Landforms (2010)
 RIVER CHANNEL AND BAR PATTERNS EXPLAINED AND PREDICTED
channel width for the empirical pattern prediction is required
for the predictive success of the latter.
Similar channel and bar patterns emerge along the entire
particle size range from fine sand-bed to cobble rivers. For
meandering, the range of specific potential stream power is
narrower in gravel than in sand. The difference arises from the
higher nonlinearity of sediment transport in the bed load-
dominated regime, for which bar theory predicts a narrower
range of width–depth ratios where meandering may occur.
Anabranching of rivers is unrelated to stream power but is
possibly related to additional factors such as river confine-
ment, bank and floodplain strength, degradation or aggrada-
tion and avulsion. The empirical prediction of pattern is
applicable to the individual branches.
Supporting Information
Supporting information may be found in the online version of
this article.
Acknowledgements—MGK is supported by the Netherlands Organi-
sation for Scientific Research (NWO) (grant ALW-Vidi-864·08·007).
We are grateful to Rob Ferguson for constructive and detailed com-
ments and to an anonymous reviewer for showing where clarification
was required. Discussion on channel patterns in general and on bar
mode specifically with Erik Mosselman and Alessandra Crosato is
cordially acknowledged. Comments on an earlier draft and discussion
by Wout van Dijk, Wietse van de Lageweg and Filip Schuurman
greatly helped to improve the paper. Margot Stoete and Fred Trappen-
burg of Geomedia are thanked for drawing Figures 1 and 2. The
authors contributed in the following proportions to conception and
design, data collection, analysis and conclusions, and manuscript
preparation: M.G.K. (50%, 10%, 60%, 80%) and J.H.B. (50%, 90%,
40%, 20%). Online supporting information contains the dataset
including coordinates, e.g. for use in Google Earth.
References
Abbado D, Slingerland R, Smith N. 2005. Origin of anastomosis in the
upper Columbia River, British Columbia, Canada. Special Publica-
tion of the International Association of Sedimentologists No. 35, pp.
3–15.
Ackers P. 1982. Meandering channels and the influence of bed mate-
rial. In Gravel-Bed Rivers, Hey RD, Bathurst JC, Thorne CR (eds).
Wiley: Chichester; 389–393.
Ackers P, Charlton F. 1970. Meandering geometry arising from varying
flows. Journal of Hydrology 11: 230–255.
Alabyan A, Chalov R. 1998. Types of river channel patterns and their
natural controls. Earth Surface Processes and Landforms 23: 467–
474.
Ashmore P. 1991. How do gravel-bed rivers braid? Canadian Journal
of Earth Sciences 28: 326–341.
Baker V. 1978. Adjustment of fluvial systems to climate and source
terrain in tropical and subtropical environments. In Fluvial Sedimen-
tology. Canadian Society of Petroleum Geologists Memoir; 211–
230.
Bartholdy J, Billi P. 2002. Morphodynamics of a pseudomeandering
gravel bar reach. Geomorphology 42: 293–310.
Beechie T, Lierman M, Pollock M, Baker S, Davies J. 2006. Channel
pattern and river-floodplain dynamics in forested mountain river
systems. Geomorphology 78: 124–141.
Blacknell C. 1982. Morphology and surface sedimentary features of
pointbars in Welsh gravel-bed rivers. Geological Magazine 119:
181–192.
Bledsoe B, Watson C. 2001. Logistic analysis of channel pattern
thresholds: meandering, braiding, and incising. Geomorphology 38:
281–300.
Copyright © 2010 John Wiley & Sons, Ltd.
17
Bolla Pittaluga M, Repetto R, Tubino M. 2003. Channel bifurcation in
braided rivers: equilibrium configurations and stability. Water
Resources Research 39: 1046.
Bridge J. 2003. Rivers and Floodplains. Blackwell: Oxford.
Bridge J, Smith N, Trent F, Gavel S, Bernstein P. 1988. Sedimentology
and morphology of a low-sinuosity river: Calamus River, Nebraska
Sand Hills. Sedimentology 33: 851–870.
Brierley G, Hickin E. 1991. Floodplain development based on selec-
tive preservation of sediments, Squamish River, British Columbia.
Geomorphology 4: 381–391.
Burge L, Lapointe M. 2005. Understanding the temporal dynamics of
the wandering Renous River, New Brunswick, Canada. Earth
Surface Processes and Landforms 30: 1227–1250.
Burnett A, Schumm S. 1983. Alluvial-river response to neotectonic
deformation in Louisiana and Mississippi. Science 222: 49–50.
Carson M. 1984. The meandering-braided threshold. Journal of
Hydrology 73: 315–334.
Chitale SV. 1973. Theories and relationships of river channel patterns.
Journal of Hydrology 19: 85–308.
Church M, Jones D. 1982. Channel bars in gravel-bed rivers. In Gravel-
Bed Rivers, Hey RD, Bathurst JC, Thorne CR (eds). Wiley: Chiches-
ter; pp. 291–338.
Church M, Rood K. 1983. Catalogue of alluvial river channel regime
data. Technical report, Department of Geography, University of
British Columbia, Vancouver.
Crosato A, Mosselman E. 2009. Simple physics-based predictor for the
number of river bars and the transition between meandering and
braiding. Water Resources Research 45: W03424.
Dade W. 2000. Grain size, sediment transport and alluvial channel
pattern. Geomorphology 35: 119–126.
Eaton B, Church M. 2007. Predicting downstream hydraulic geometry:
a test of rational regime theory. Journal of Geophysical Research
112: F03025.
Eaton, B, Millar R, Davidson S. 2010. Channel patterns: braided,
anabranching, and single-thread. Geomorphology 120: 353–
364.
Edwards P, Kollman J, Gurnell A, Tockner K, Ward J. 1999. A concep-
tual model of vegetation dynamics on gravel bars of a large Alpine
river. Wetlands Ecology and Management 7: 141–153.
Egozi R, Ashmore P. 2008. Defining and measuring braiding intensity.
Earth Surfaces Processes and Landforms 33: 2121–2138.
Egozi R, Ashmore P. 2009. Experimental analysis of braided channel
pattern response to increased discharge. Journal of Geophysical
Research 114: F02012.
Ferguson R. 1987. Hydraulic and sedimentary controls of channel
pattern. In River Channels: Environment and Process, Richards K
(ed.). Institute of British Geographers Special Publication 18. Black-
well: Oxford; 129–158.
Ferguson R, Ashmore PE, Ashworth PJ, Paola C, Prestegaard KL. 1992.
Measurements in a braided river chute and lobe 1. Flow pattern,
sediment transport, and channel change. Water Resources Research
28: 1877–1886.
Fielding C, Alexander J. 1996. Sedimentology of the upper Burdekin
river of North Queensland, Australia: an example of a tropical,
variable discharge river. Terra Nova 8: 447–457.
Fredsøe J. 1978. Meandering and braiding of rivers. Journal of Fluid
Mechanics 84: 609–624.
Friedkin J. 1945. A laboratory study of the meandering of alluvial
rivers. US Army Corps of Engineers, US Waterways Experiment
Station, Vicksburg, MS.
Gustavson T. 1978. Bed forms and stratification types, Nueces River,
Texas. Sedimentology 25: 401–426.
Hickin E, Nanson G. 1975. The character of channel migration on the
Beatton River, north-east British Columbia, Canada. Bulletin of the
Geological Society of America 86: 487–494.
Hooke J. 2004. Cutoffs galore!: occurrence and causes of multiple
cutoffs on a meandering river. Geomorphology 61: 225–238.
Jackson R II. 1976. Depositional model of point bars in the Wabash
River. Journal of Sedimentary Petrology 46: 579–594.
Kellerhals R, Neill C, Bray D. 1972. Hydraulic and geomorphic
characteristics of rivers in Alberta. Technical report 72-1,
Alberta Cooperative Research Program in Highway and River
Engineering.
Earth Surf. Process. Landforms (2010)
18 M. G. KLEINHANS AND J. H. VAN DEN BERG
Kemp J. 2004. Flood channel morphology of a quiet river, the Lachlan
downstream from Cowra, southeastern Australia. Geomorphology
60, 171–190.
Kleinhans MG. 2010. Sorting out river channel patterns. Progress in
Physical Geography 34: 287–326.
Kleinhans MG, Jagers H, Mosselman E, Sloff C. 2008. Bifurcation
dynamics and avulsion duration in meandering rivers by one-
dimensional and three-dimensional models. Water Resources
Research 44: W08454.
Kleinhans MG, Schuurman F, Bakx W, Markies H. 2009. Meandering
channel dynamics in highly cohesive sediment on an intertidal mud
flat in the Westerschelde estuary, the Netherlands. Geomorphology
105: 261–276.
Koch F, Flokstra C. 1981. Bed level computations for curved alluvial
channels. In Proceedings of the XIX Congress of the International
Association for Hydraulic Research, Vol. 2, New Delhi, India; 357.
Latrubesse E. 2008. Patterns of anabranching channels: the ultimate
end-member adjustment of mega rivers. Geomorphology 101: 130–
145.
Leeder M, Bridges P. 1975. Flow separation in meander bends. Nature
253: 338–339.
Leopold L, Wolman M. 1957. River channel patterns: braided, mean-
dering and straight. Geological Survey professional paper 282-B. US
Government Printing Office, Washington, DC.
Lewin J. 1970. Initiation of bed forms and meanders in coarse-grained
sediment. Bulletin of the Geological Society of America 87: 281–
285.
Lewin J, Brewer P. 2001. Predicting channel pattern. Geomorphology
40: 329–339.
Lewin J, Brewer P. 2003. Reply toVan den Berg and Bledsoe’s comment
on Lewin and Brewer (2001) ‘Predicting Channel Patterns’. Geomor-
phology 40, 329–339. Geomorphology 53: 339–342.
Makaske B, Smith D, Berendsen H. 2002. Avulsions, channel evolu-
tion and floodplain sedimentation rates of the anastomosing upper
Columbia River, British Columbia, Canada. Sedimentology 49:
1049–1071.
Makaske B, Smith D, Berendsen H, de Boer A, van Nielen-Kiezebrink
M, Locking T. 2009. Hydraulic and sedimentary processes causing
anastomosing morphology of the upper Columbia River, British
Columbia, Canada. Geomorphology 111: 194–205.
Marra W. 2008. Dynamics and interactions of bars in rivers and the
relation between bars and a braided river pattern. BSc thesis, Depart-
ment of Physical Geography, Utrecht University, Netherlands.
Marston R, Girel J, Pautou G, Piegay H, Bravard J-P, Arneson C. 1995.
Channel metamorphosis, floodplain disturbance and vegetation
development: Ain river, France. Geomorphology 13: 121–131.
McGowen J, Garner L. 1970. Physiographic features and stratification
types of coarse-grained point bars: modern and ancient examples.
Sedimentology 14: 77–111.
Meyer-Peter E, Müller R. 1948. Formulas for bed-load transport. In
Proceedings of 2nd Meeting: International Association for Hydraulic
Structures Research, Stockholm, Sweden; 39–64.
Mosselman E, Tubino M, Zolezzi G. 2006. The overdeepening theory
in river morphodynamics: two decades of shifting interpretations. In
River Flow 2006: International Conference on Fluvial Hydraulics,
Ferreira R, Alves E, Leal J, Cardoso A (eds). Lisbon, Portugal. Taylor
& Francis/Balkema: London; 1175–1181.
Nanson G. 1980. Point bar and floodplain formation of the meander-
ing Beatton River, northeastern British Columbia, Canada. Sedimen-
tology 27: 3–29.
Nanson G, Croke J. 1992. A genetic classification of floodplains.
Geomorphology 4: 459–486.
Nanson G, Knighton A. 1996. Anabranching rivers: their cause, char-
acter and classification. Earth Surface Processes and Landforms 21:
217–239.
Neill C. 1973. Hydraulic and morphologic characteristics of Atha-
basca River near Forth Assiniboine: the anatomy of a wandering
Copyright © 2010 John Wiley & Sons, Ltd.
View publication stats
gravel bed river. Technical report REH/73/8. Alberta Research
Council, Highways and River Engineering Division, Edmonton.
Parker G. 1976. On the cause and characteristic scales of meandering
and braiding in rivers. Journal of Fluid Mechanics 76: 457–479.
Parker G, Wilcock P, Paola C, Dietrich WE, Pitlick J. 2007. Physical
basis for quasi-universal relations describing bankfull hydraulic
geometry of single-thread gravel bed rivers. Journal of Geophysical
Research 112: F04005.
Rannie, W. 1990. The Portage La Prairie ‘flood plain fan’. In Alluvial
Fans: A Field Approach, Rachocki AH, Church M (eds). Wiley:
Chichester; 179–193.
Rosgen D. 1994. A classification of natural rivers. Catena 22: 169–199.
Rust B. 1978. A classification of alluvial channel systems. In Fluvial
Sedimentology. Canadian Society of Petroleum Geologists: Calgary,
Canada; 187–198.
Schumm S. 1963. Sinuosity of alluvial rivers in the Great Plains.
Geological Society of America Bulletin 74: 1089–1100.
Schumm S. 1968. River adjustment to altered hydrologic regime:
Murrumbidgee River paleochannels. Technical report 598, US Geo-
logical Survey professional paper.
Schumm S. 1969. River metamorphosis. Proceedings of the American
Society of Civil Engineers, Journal of the Hydraulics Division 95:
255–273.
Schumm S, Dumont J, Holbrook J. 2000. Active Tectonics and Alluvial
Rivers. Cambridge University Press: Cambridge, UK.
Seminara G, Tubino M. 1989. Alternate bars and meandering: free,
forced and mixed interactions. In River Meandering, No. 12, Ikeda
S, Parker G (eds). American Geophysical Union: Washington, DC;
153–180.
Smith N. 1974. Sedimentology and bar formation in the Upper Kicking
Horse River, a braided outwash stream. Journal of Geology 82:
205–223.
Soulsby R. 1997. Dynamics of Marine Sands. Thomas Telford Publi-
cations: London.
Struiksma N, Olesen K, Flokstra C, De Vriend H. 1985. Bed deforma-
tion in curved alluvial channels. Journal of Hydraulic Research 23:
57–79.
Talmon A, Struiksma N, van Mierlo M. 1995. Laboratory measure-
ments of the direction of sediment transport on transverse alluvial-
bed slopes. Journal of Hydraulic Research 33: 495–517.
Thorne C, Russell A, Alam M. 1993. Planform pattern and channel
evolution of the Brahmaputra River, Bangladesh. Special Publication
Vol. 75. Geological Society, London; 257–276.
Tooth S, McCarthy S. 2004. Controls on the transition from meander-
ing to straight channels in the wetlands of the Okavango delta,
Botswana. Earth Surface Processes and Landforms 29: 1627–
1649.
Tooth S, Jansen J, Nanson G, Coulthard T, Pietsch T. 2008. Riparian
vegetation and the late Holocene development of an anabranching
river: Magela Creek, northern Australia. Geological Society of
America Bulletin 120: 1212–1224.
van den Berg J. 1995. Prediction of alluvial channel pattern of peren-
nial rivers. Geomorphology 12: 259–279.
van den Berg J, Middelkoop H. 2007. Scroll bar and chute bar devel-
opment in meandering rivers. Two contrasting examples: the lower
Volga (Russia) and the Allier (France). In Proceedings of the 10th
International Symposium on River Sedimentation, Vol. 5, Moscow,
Russia, 1–4 August; 282–300.
Ward J, Tocker K, Edwards P, Kollman J, Bretschko G, Gurnell A, Petts
G, Rossaro B. 1999. A reference river system for the Alps: the ‘Fiume
Tagliamento’. Regulated rivers: Research and Management 15:
63–75.
Ward JV, Malard F, Tockner K. 2002. Landscape ecology: a framework
for integrating pattern and process in river corridors. Landscape
Ecology 17 (Suppl. 1): 35–45.
Williams G. 1978. Bankfull discharge of rivers. Water Resources
Research 14: 1141–1154.
Earth Surf. Process. Landforms (2010)