You are on page 1of 19

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

net/publication/41733339

Defining and Measuring Braiding Intensity

Article in Earth Surface Processes and Landforms · December 2008


DOI: 10.1002/esp.1658 · Source: OAI

CITATIONS READS

199 7,649

2 authors, including:

Roey Egozi
Ministry of Agriculture, Israel
30 PUBLICATIONS 690 CITATIONS

SEE PROFILE

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

Water resources management in agricultural landscapes View project

The role of estuaries in controlling pollution of the Mediterranean Sea: The Alexander River as a case study View project

All content following this page was uploaded by Roey Egozi on 16 September 2016.

The user has requested enhancement of the downloaded file.


Earth Surface Processes and Landforms
Defining
Earth Surf.and measuring
Process. braiding
Landforms intensity (2008)
33, 2121–2138 2121
Published online 24 April 2008 in Wiley InterScience
(www.interscience.wiley.com) DOI: 10.1002/esp.1658

Defining and measuring braiding intensity


Roey Egozi† and Peter Ashmore*
Department of Geography, University of Western Ontario, London, Ontario, Canada N6A5C2

*Correspondence to: P. Ashmore, Abstract


Department of Geography,
University of Western Ontario, Geomorphological studies of braided rivers still lack a consistent measurement of the com-
London, Ontario, Canada plexity of the braided pattern. Several simple indices have been proposed and two (channel
N6A5C2. E-mail: count and total sinuosity) are the most commonly applied. For none of these indices has
pashmore@uwo.ca there been an assessment of the sampling requirements and there has been no systematic

Current address: R. Egozi, study of the equivalence of the indices to each other and their sensitivity to river stage.
Division of Environmental, Resolution of these issues is essential for progress in studies of braided morphology and
Water and Agricultural dynamics at the scale of the channel network.
Engineering, Faculty of Civil A series of experiments was run using small-scale physical models of braided rivers in a
and Environmental Engineering,
3 m × 20 m flume. Sampling criteria for braid indices and their comparability were assessed
Technion, Haifa 32000, Israel.
using constant-discharge experiments. Sample hydrographs were run to assess the effect of
Tel: +972-4-829-2625. E-mail:
regozi@techunix.technion.ac.il flow variability.
Reach lengths of at least 10 times the average wetted width are needed to measure braid
indices with precision of the order of 20% of the mean. Inherent variability in channel
pattern makes it difficult to achieve greater precision. Channel count indices need a mini-
mum of 10 cross-sections spaced no further apart than the average wetted width of the river.
Several of the braid indices, including total sinuosity, give very similar numerical values but
they differ substantially from channel-count index values. Consequently, functional relation-
ships between channel pattern and, for example, discharge, are sensitive to the choice of
braid index. Braid indices are sensitive to river stage and the highest values typically occur
below peak flows of a diurnal (melt-water) hydrograph in pro-glacial rivers. There is
no general relationship with stage that would allow data from rivers at different relative
stage to be compared. At present, channel count indices give the best combination of rapid
measurement, precision, and range of sources from which measurements can be reliably
Received 19 June 2007; made. They can also be related directly to bar theory for braided pattern development.
Revised 20 November 2007; Copyright © 2008 John Wiley & Sons, Ltd.
Accepted 10 December 2007
Keywords: Braiding index; gravel-bed river; braiding; channel pattern; physical model

Introduction
Braided rivers are recognized by the multiplicity of laterally mobile, intersecting channels (Leopold and Wolman,
1957). While the occurrence of braiding has been extensively analysed theoretically and empirically, there has been
little discussion or quantitative analysis of the degree of braiding and its response to various controls. The degree of
braiding is a basic morphological property of the river analogous to elements of meander geometry in single thread
channels. Understanding of the variation in the degree of braiding in response to external controls (e.g. Howard et al.,
1970; Robertson-Rintoul and Richards, 1993; Surian, 1999), and of braided river hydraulics and morpho-dynamics,
depend on the definition and measurement of the braided pattern morphology. The most basic element of this is the
complexity of the channel pattern–braiding intensity.
Several indices of braiding intensity have been proposed and used. However, there has been almost no analysis of
the comparability of these indices or of the reliability and sampling requirements needed to establish the precision
of the measurements. This is a significant hindrance to functional understanding of variations in braiding intensity in
response to external controls such as discharge regime, sediment supply and valley gradient. Here we address the issue
of the measurement of braiding intensity based mainly on experimental physical models of braided rivers along with
comparable field measurements.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
2122 R. Egozi and P. Ashmore

Braiding intensity indices


In general, braiding intensity indices have been based on one of three characteristics: bar dimensions and frequency
(e.g. Brice, 1960, 1964; Rust, 1978; Germanoski and Schumm, 1993); the number of channels in the network (e.g.
Howard et al., 1970); and the total channel length in a given river length (e.g. Hong and Davies, 1979; Mosley, 1981;
Friend and Sinha, 1993).

Bar indices
Brice (1960, 1964) originally devised a braid index based on the length of bars and islands in sand-bed braided rivers.
This index, denoted here as BIB, is defined as the sum of twice the length of all islands and (or) bars (Lb) in a reach
divided by the length of the reach (Lr) measured along the centre line of the river (Figure 1A). When bars alone were
used Brice referred to this as the ‘transient’ index because it changes with water stage. Germanoski and Schumm
(1993) adopted Brice’s index but modified the index by adding to it the total number of bars (Nb) per reach length (Lr)
(Figure 1B), denoted BIB* in this paper. The purpose of this modification was to reduce the possibility that a reach with
one large bar could have an equivalent, or even greater, braiding intensity than a reach with several smaller bars
(Germanoski and Schumm, 1993). Rust (1978) adopted a similar approach to Brice (1964), but measured braiding
intensity as the number of braids (NL) per mean channel wavelength where wavelength (λ) was defined as ~1·25 times
the distance between successive confluence and bifurcation (BIλ, Figure 1C). Rust (1978) measured braids and channel
lengths using the mid-line of the channel surrounding each bar in order to minimize sensitivity to stage variations.

Channel count index


Howard et al. (1970) proposed two indices of braided network complexity. The first is the mean number of links
intersected by cross-sections of the river in successive river lengths (BIT1, Figure 1D). River lengths are defined by
cross-sections sufficiently far apart that the cross-sections do not cross the same links (or segments) of the network
more than once. The number of channels (segments) is then averaged for the river lengths of interest. The second
index is the total number of links (or segments, <NL>; Figure 1E) in the measured reach (BIT2).
Other studies (e.g. Hong and Davies, 1979; Mosley, 1982; Ashmore, 1991; Chew and Ashmore, 2001) have also
used a channel count index. However, these studies did not divide the measurement reach into lengths in the way that
Howard et al. (1970) defined them. Instead, measurements were made at a number of cross-sections within river
reaches or flume length of interest such that cross-sections could cross a given channel segment more than once (BIT3,
Figure 1D). Bridge (1993) has argued that a channel count index is theoretically justified because it can be related
directly to the mode of the bar array from which braiding is theorized and shown to originate (Kinoshita, 1957; Parker,
1976; Fujita, 1989).

Channel length (‘sinuosity’) indices


Hong and Davies (1979) proposed an index (PT, Figure 1E) of the total length of channels (ΣLL) per unit length
of river (Lr); now often referred to as ‘total sinuosity’ (Richards, 1982 p. 182; Robertson-Rintoul and Richards, 1993;
Smith et al., 1996). Mosley (1981) proposed a variant of total sinuosity that substitutes the (straight line) reach length
with the length of the main channel (PT*, Figure 1F). This modification may reduce Bridge’s (1993) concern that the
sinuosity index combines two different aspects of the channel pattern – the number of separate channels and the
sinuosity of those channels – that makes it problematic as an index intended to measure channel multiplicity only.

Application of braid indices


The various braiding indices have been used to differing extents. Some (e.g. Rust’s wavelength-based index) appear to
have seen very limited use, perhaps for practical reasons or uncertainty about their validity (Bridge, 1993). Others
have clear limitations such as Howard et al. (1970) index (BIT2), which is not normalized for reach length. Channel
count (BIT3) and sinuosity (PT) indices are those most commonly used in correlating braided channel pattern with flow,
stream power, sediment transport, morphology and vegetation parameters, and in observing variations over time under
experimental conditions (Mosley, 1982, 1983; Ashmore, 1991; Warburton and Davies, 1994; Chew and Ashmore,
2001; Gran and Paola, 2001).
Although these braiding indices have been used for analysis of braided patterns, there are some unresolved issues
about the definition and measurement of braiding intensity.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
Defining and measuring braiding intensity 2123

Figure 1. Types of braiding intensity indices and the parameters measured for each index (see text for more details). Note that
in diagram C the thickness of the lines refers to channel order, e.g. the thickest line denotes first-order channel, as suggested by
Williams and Rust (1969).

(1) There has been almost no assessment of the inherent variation in braided pattern complexity under given, fixed,
external controls. Ashmore (1991) looked at variation over time during constant-discharge experiments and
assessed the standard deviation from the mean over the length of each experiment to estimate sample size (run
length) needed to estimate the mean value within a given error. Apart from this one example, there has been no
assessment of the statistical properties and sampling considerations (reach length, sample size, cross-section

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
2124 R. Egozi and P. Ashmore

spacing) for each braiding index. In general, there appears to have been almost no discussion of the basic
sampling issues that would allow statistical confidence limits to be placed on a given data set or analysis. Only
Brice (1964) appears to have considered the necessary channel length for a representative single measurement of
braiding intensity.
(2) The various braiding indices have never been systematically compared with each other in order to establish
whether or not they are equivalent and whether it is possible to convert between indices, especially between the
commonly used ‘channel count’ (BIT3) and ‘total sinuosity’ (PT). This is a fundamental issue in analysis of braided
channel pattern characteristics and in retrospective pooling and comparison of results. Many of the indices are
similar to one another and therefore may yield similar trends although not (numerically) identical values. Robertson-
Rintoul and Richards (1993) have shown that total sinuosity (PT) and a channel count index (BIT3) correlate very
closely. However, Bridge (1993) has argued that total sinuosity confounds two distinct pattern properties and
would not necessarily yield similar values as a channel count index for a given channel pattern. The issue is
unresolved.
(3) Many researchers have mentioned possible bias due to stage-related variation in braiding intensity at the time of
measurement (e.g. Surian, 1999; van der Nat et al., 2002) and have proposed standardizing the stage at which
measurements are made (Kellerhals et al., 1976; Rust, 1978; Begin, 1981; Mosley, 1982; Carson, 1984; Bridge,
1993). This assumes that data are available for a range of stages for any given river and that the stage variation is
a significant concern. However, none have actually quantitatively examined this stage effect either at a given
location, or under controlled, experimental conditions in order to assess the sensitivity of braiding intensity to
stage differences and rising/falling stage.

These three issues are the focus of this paper.

Methods
The analysis in this paper is based mainly on results from physical model experiments, with some comparison with
field observations. The model and experimental approach is advantageous because the overall channel controls can be
held constant and the conditions under which the channels form are known. This restricts variability in channel pattern
to that inherent in braided river morpho-dynamics and removes the effect of hydrological and geomorphological
history. In addition, channel patterns in the model evolve quickly and provide a long record of variation that can be
sampled at high frequency. This provides a large data set from which to derive statistical precision estimates. How-
ever, the experiments have the disadvantage of covering only a limited range of channel scale and morphology.
The study used generic Froude-scaled physical models of braided rivers of the kind that have often been used in
geomorphological studies of gravel braided rivers (e.g. Davies, 1987; Ashmore, 1991; Hoey and Sutherland, 1991;
Shvidchenko and Kopaliani, 1998; Warburton and Davies, 1998). The model is not designed to replicate the actual
morphology of a particular prototype river at any given time but does use the overall gradient, discharge and grain-size
characteristics of Sunwapta River, Alberta as a basis for modelling calculations. Therefore the model does not produce
exact plan-form similarity for a particular river but yields sampling results applicable to full scale rivers in general.
The experiments were run in a tilting (maximum slope of 2·5%) flume 3 m wide, 20 m long, with a sand layer
0·15 m thick (Figure 2A). Discharge was measured by a calibrated trapezoidal, sharp-edged, weir at the upstream end
of the flume (Figure 2B). The sediment gradation was similar to that of natural gravel-bed rivers but truncated at the
equivalent of coarse sand size (2·8 mm). Overall the grain sizes range from 0·1 mm to 8 mm, with D50 = 1·17 mm
(Figure 3). The transported sediments are circulated from the tail tank and fed back into the flume (after separating
excess water) via a vibrating feed tray (Figure 2C). In this way the sediment feed rate varied ‘naturally’ but in the long
term is expected to equal the sediment delivery rate by the river to the tail tank.
There were two sets of runs in which both the discharge and slope were the independent variables. The first set was
a sequence of three successive runs all with a slope of 1·5%. These runs had constant discharges of 1·4 L s−1 (experi-
ment 7); 2·1 L s−1 (experiment 8), and 2·8 L s−1 (experiment 14) (Egozi, 2006). Experiment 7 began from a straight
channel with trapezoidal cross-section cut into a flat bed with dimension (0·5 m wide, 0·015 m deep) just large enough
to accommodate the discharge. The other two runs followed directly from the channel configuration at the end of the
previous run, without re-flattening of the bed. Each experiment was run for about 70 hours. In each run, reach-
averaged braiding intensity initially increased from the beginning of the run following the increase in discharge, and
then fluctuated around a constant mean value (Figure 4).
The second series of experiments consisted of three sequential runs each simulating a diurnal hydrograph typical of
a pro-glacial braided river. The hydrograph simulation is based approximately on the daily melt-water flow cycle of

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
Defining and measuring braiding intensity 2125

Figure 2. University of Western Ontario (UWO) experimental flume: (A) plan and side views; (B) a close up of the weir; (C)
sediment feeder; (D) location of fixed cameras; (E) high-resolution mobile camera.

the pro-glacial Sunwapta River such as that measured in 2003 (Figure 5A) using Froude scaling of the discharge and
duration in each time-step (assuming time scale is square root of the length scale). Each hydrograph was run through
an established braided channel using nine step increases/decreases in discharge (Figure 5B). Because of the sediment
recirculation system, sediment supply rate lagged slightly behind the changes in discharge but the sediment circulation
time (from tail tank to upstream feed) is only about 30 seconds so sediment feed response time to changed discharge
was short compared with the length of each hydrograph step.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
2126 R. Egozi and P. Ashmore

Figure 3. Grain-size distribution for the flume and the Sunwapta River (Varkaris, 2002).

Figure 4. (A) First series of experiments with constant channel-forming discharge. (B) Stable mean BIT3 values at the second part
of each experiment indicate that the braid plain has developed to equilibrium state with Qchf.

In all of the runs, channel pattern geometry was measured in a 10-m-long section of the flume between 7 and 17 m
from the flume entrance in order to avoid entrance effects on the channel pattern. The measurement reach length is
approximately equal to 2·2–3·1 times the braid wavelength (Ashmore, 2001) and between 10 and 17 times the average
wetted width of the channel (total wetted area/measurement length). Therefore the measurements cover pattern varia-
tion at a length scale equivalent to multiple confluence-bar units.
Channel pattern was measured from imagery from three digital cameras mounted vertically above the flume. Two
(Olympus C5060) cameras were in fixed positions 3 m above the flume (Figure 2D) each covering the full width of
the flume and a length of approximately 5·4 m, for a total coverage of about 10 m (due to ~0·5 m overlapping length
between images). These cameras automatically collected images every 15 or 10 min during the first and second series
of experiments, respectively (using the software PTC camera controller from Pine Tree Computing, 2003). The third
camera (Canon 10D SLR) was attached to a moving carriage 2·9 m above the flume (Figure 2E) and used to acquire
high resolution images of smaller areas of the flume bed (3·3 m wide and 2·2 m long) manually. This camera collected
images every 4 hours during the first series of experiments and for each flow stage during the second series of experiments.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
Defining and measuring braiding intensity 2127

Figure 5. (A) Sunwapta River diurnal snow-melt flood hydrograph. The dashed lines indicate the range of discharge stages
simulated. (B) The three simulated hydrographs based on downscaling of an average diurnal cycle.

Time-lapse images captured with the Olympus cameras were ortho-rectified (using ERDAS Imagine Pro 8.5.1) to
eliminate image distortion, which was significant because of the wide angle lenses used on the cameras. The ortho-
rectified images were the base layer of channel pattern maps from which the channel pattern parameters were
extracted, measured and compared. Ortho-rectification used 46 photograph-control targets distributed along the flume
and surveyed using a total station. Target coordinates were calculated using three-dimensional intersection software
developed by Dr Jim Chandler, Loughborough University. After ortho-rectification, the images from each camera were
stitched together using ERDAS Imagine to obtain an image map of the whole reach (Figure 6).
The image maps were exported to ArcGIS 9.1 and water surface area was digitized manually (Figure 6) aided by
the higher resolution images from the Canon 10D. Channel network maps were developed from which the channel
parameters were measured.

Field data
For comparison with the experimental results from the hydrograph experiments, data were also collected from Sunwapta
River, Alberta during summer melt-water flows in 2003 (Figure 7). The field site is a braided reach 180 m long
(180 m × 140 m) about 3 km from the glacial source of the river at Athabasca Glacier. Valley slope is approximately
1·4% (Chew and Ashmore, 2001) and median grain size is about 40 mm. River stage was measured continuously
using an ultrasonic depth gauging (UDG) at a confined section about 2 km upstream of the study site. Stage was then
calibrated to discharge using velocity–area gauging in the study reach. Braiding intensity was measured from a series
of images from two 35 mm cameras positioned on a cliff overlooking the field site that were programmed to take pictures
automatically at 4 hours intervals through daylight hours each day. The oblique photographs were ortho-rectified and
used to measure BIT3 along 10 cross-sections (10 m apart) using the same procedures as the flume images and
measurements. Details of the flow gauging and image ortho-rectification procedures are in Ashmore and Sauks (2006).

Results

Effect of reach length on precision of total sinuosity index


Both the bar length and channel sinuosity indices are based on a defined reach length but the effect of reach length on
the measured index has not been assessed. Here the focus is on total sinuosity but the results are expected to apply to
other indices. Note that channel count indices do not explicitly involve a defined reach length, but the measured braid
index will still be sensitive to the length of reach used.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
2128 R. Egozi and P. Ashmore

Figure 6. Image processing and interpretation. (A) pairs of images are enhanced, cropped and ortho-rectified. (B) A mosaic is
produced from the images and exported to ArcGIS. (C) Braided channel pattern maps are then generated and braiding indices are
calculated. ‘MACh’ stands for main active channel.

Figure 7. Hydrograph for Sunwapta River during the period of measurements of braiding intensity. The grey solid symbols mark
the times when images where taken and braiding intensity values were computed.

To examine this effect of reach length we first define a length unit equal to the average wetted width (AWW),
termed a sub-reach length. The AWW has been shown to scale with total river discharge (Ashmore and Sauks, 2006)
and therefore is a useful scaling unit. The number of sub-reaches within a fixed flume length is 9–17 within the runs.
Braid index (PT) was measured along a fixed length of the flume using a range of multiples of the sub-reach length. At

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
Defining and measuring braiding intensity 2129

Figure 8. Mean and standard deviations of total sinuosity (PT) at varying reach lengths (Lr) for different channel-forming discharges:
(A) 1·4 L s−1, (B) 2·1 L s−1 and (C) 2·8 L s−1. The dashed line denotes the mean total sinuosity.

the smallest multiples there are several samples along the flume length but sample size decreases progressively at
larger multiples of the sub-reach length. This design was applied at five different times in each of the three runs of the
first series after the braided pattern had fully developed and adjusted to the new experimental controlling conditions
(Figure 4), and therefore assesses the variation among sample times, as well as spatial effects, in the braided pattern.
Figure 8 shows that longer sample reaches (larger multiples of sub-reach length) reduce the variability in the total
sinuosity braid index (PT). The coefficient of variation decreases from over 20% to less than 10% as the sampled reach
length increases from two times AWW to over 15 times AWW. Mean braiding index (PT) is not affected by the sample
sub-reach length because the same channel patterns were sampled in each case. Substantial improvement in precision
occurs up to, and possibly beyond, sample reach lengths on the order of 10 times the AWW.

Minimum number of cross-sections for channel-count indices


In the case of channel counts the effect of the number of cross-sections on measured braid index is also a significant
sampling issue. This effect was examined using the channel pattern at 13 different times during the same run over the
same sample length of 10 m. Braiding index was measured for 37 cross-sections with a spacing interval of 0·25 m
between neighbouring cross-sections. Then, in successive samples, a pair of cross-sections was removed and the
braiding index was recalculated. This procedure continued down to a minimum of three cross-sections. At each step
the pair of cross sections that was removed consisted of one cross-section from the upstream half of the reach and one
cross-section from the downstream half of the reach, and both cross-sections were at an equivalent position relative to
the middle of the reach. In this way, the whole reach length was sampled systematically to avoid bias towards any one
part of the reach. However, following this thinning procedure, the spacing between cross-sections is not exactly the
same for all samples. Therefore in addition to a sample size of 37 cross-sections with a spacing interval of 0·25 m, we
focused on sample sizes of 19 and 9 cross-sections where spacing intervals are constant (0·5 m and 1 m respectively).
The results show that mean BIT3 is independent of sample size at n > ~10 cross-sections, provided the cross-sections
are spaced evenly over a minimum representative reach length (Figure 9A). In addition, the standard deviation
decreases with an increased number of cross-sections, but stabilizes at n > ~10 cross-sections. In this case, n < 10
coincides with cross-section spacing greater than the average wetted width of the channel, therefore, it is suggested
that the spacing interval between neighbouring cross-sections should be no larger than the average wetted width at
channel-forming flow. Even so, the 95% confidence bands give a standard deviation of ±20% of the mean. This is
consistent with results of Ashmore (1991). Adding more cross-sections within a given reach length may not give
substantial improvement in precision. The large deviation reflects the magnitude of inherent variability in local
channel pattern characteristics. Further significant improvement in precision of the estimate of the mean would require
sample sizes of >100, which is not practical (Figure 9B). A similar coefficient of variation was obtained for a 1·14 km
braided reach of Sunwapta River (length equal to about 32 times AWW) with spacing between the cross-sections
approximately equal to the average wetted width at high flow. At this sample size the estimated standard error is
~15%. However, the actual coefficient of variation is larger than 15% and that is probably due to additional factors,

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
2130 R. Egozi and P. Ashmore

Figure 9. (A) Mean braiding intensity (for experiment 8) calculated for the number of cross-sections added to cover the same
reach length uniformly. Note the dashed line marks n = 10 cross-sections. (B) Increased standard deviation for decreasing sample
size for different channel-forming discharges. Note that for the same error magnitude there is a need for a higher density of cross-
sections at higher channel-forming discharges.

such as longitudinal grain-size sorting and channel adjustment (e.g. Chew and Ashmore, 2001), indicating that the
model-based sample criteria may be an absolute minimum requirement.

Comparison of braiding intensity indices


Previously, the issue of similarity between braiding indices has been examined for the limited case of BIT3 and PT
(Robertson-Rintoul and Richards, 1993) and it has been shown that the two indices are essentially identical. However,
Bridge (1993) demonstrated that it is possible to have two braided reaches with the same total sinuosity index but with
different channel count index. The PT index would be equivalent to BIT3 only if the individual anabranches had very
low sinuosity. This issue requires resolution and raises the question of the equivalency of other braiding indices.
Braiding indices were compared using simple bivariate least-squares (LS) linear regression. This regression struc-
ture gives the relationship for predicting the value of one index from the other, rather than the ‘true’ relationship
between them (Clarke, 1980). Figure 10A shows the relationship between BIT3 and PT based on 33 means of braiding
index, nine each at 1·4 L s−1, 13 at 2·1 L s−1, and 11 at 2·8 L s−1. While there is a significant correlation between these
two indices they do not follow the 1:1 line proposed by Robertson-Rintoul and Richards (1993). The slope of 1·57 for
the regression line (in Figure 10A) supports the idea that PT values should, in general, be higher than BIT3 values.
Many of the other braiding intensity indices are significantly intercorrelated. Two of these (PT* and BIb, Figure 10B
and C) have a trend close to a 1:1 relationship with PT. The similarity of PT and PT* is not surprising given the
similarity in their definition, but indicates that sinuosity of the main active channel is not high in this case. The
similarity of the total sinuosity and bar index also reflects their similar basis and definition. This similarity also
explains why comparison of BIT3 and BIb (Figure 10D) is very much like the linear regression line of BIT3 and PT
(Figure 10A). The two bar indices BIb and BIb* are correlated but do not follow a 1:1 line (Figure 10E) because BIb*
has an additional term (number of bars) in its definition.
If braiding indices are to be used as measures of channel pattern that are functionally related to external controls
such as discharge and stream power, it is useful to see what effect the choice of braiding index might have on these
derived relationships. This can be illustrated using simple best-fit power functions of braiding index and channel-
forming discharge for the three constant-discharge runs:

BIT3 = 2·39Q0·52 (1)


PT = 2·91Q0·73 (2)
PT* = 2·74Q0·75 (3)

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
Defining and measuring braiding intensity 2131

Figure 10. Pair-wise comparison of braiding intensity indices. Data points denoted by circle symbol in (A) are extracted from
Robertson-Rintoul and Richards (1993, p. 115). Diamond symbol denotes data from this study.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
2132 R. Egozi and P. Ashmore

Figure 11. (A) Conversion relations between BIT3 and other braiding intensity indices. The conversion fits the range of flume
discharges. (B) Braiding intensity calculated based on conversion equations in (A) versus measured values.

BIb = 2·26Q0·88 (4)


BIb* = 3·34Q 0·78
(5)

This is a very small data set and therefore there are large errors in the regression functions, but there is an apparent
influence of choice of braiding index on the functional relationship with discharge and in particular the relationship for
BIT3 (Equation 1) differs from the others. In this case the difference between the exponents is statistically significant
( p < 0·05). This is important given that BIT3 is the index most commonly used as a measure of braiding intensity.
Figure 11A gives a graphical conversion between channel count index (BIT3) and the four other indices. If these
conversions are performed then most of the points, regardless of which braided index is used, fall around the 1:1 trend
line (Figure 11B). In addition, a new set of equations similar to Equations 2–5, can be derived based on BIT3 values
substituted into the conversion equations of Figure 11A:

PT = 3·02Q0·68 (6)
PT* = 2·85Q 0·70
(7)
BIb = 2·43Q0·78 (8)
BIb* = 3·53Q0·72 (9)

The exponents in Equations 6–9 are not significantly different ( p > 0·05) from the ones in Equations 2–5, thus
conversion between BIT3 and other braiding intensity indices is possible and consistent functional relationships with
control variables may be developed using these conversions. This outcome may be specific to these experiments but
with additional data it may prove possible to generalize these conversion relationships.

Sensitivity of braiding intensity indices to flow stage


Flow-stage variation affects braided channel pattern parameters more than single thread channels because braided
rivers are characterized by higher width/depth ratios and are strongly dependent on exposure of bar-scale topography
(Bridge, 1993). Although stage effects on braiding intensity have often been raised as a concern, the only previous
study of this effect (Mosley, 1983) discerned no clear relationship between braiding intensity and discharge. However,
Mosley’s (1983) data were collected at discharges below the mean daily discharge and may not cover the full range of
variation.
In this study we examine the effect of flow stage on braiding indices averaged through three identical hydrographs
(Figure 5). Figure 12 shows that mean braiding intensity values of various braid indices followed the changes in flow

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
Defining and measuring braiding intensity 2133

Figure 12. Variation in mean braiding index with hydrograph stage.

stage throughout the hydrograph. However, some of the highest braiding intensity values occur below peak discharge,
reflecting the merging of channels (submerging bars) at peak flow, and also changes in channel pattern during the
hydrograph peak, that cause differences in braiding intensity between hydrograph limbs.
There were no differences in braid index between the hydrograph limbs for BIT3, PT and PT* (Figure 12). The
indices BIb and BIb* are less consistent; the measured values were lower during the rising limb than during the falling
limb. However, these differences decrease with increasing flow stage. In general, coefficients of variation in braid
indices for the hydrograph simulations are higher at any given discharge than in the constant discharge experiments
(Figure 13). In addition, the coefficient of variation of indices BIb and BIb* are higher than those of indices PT, PT* and
BIT3 at different flow stages for both falling and rising limbs. Common to all indices is that the highest coefficient of
variation is not associated with the peak discharge.

Figure 13. Coefficient of variation of different indices at varying flow stages (Q) during rising (r) and falling (f ) limbs. Subscript ‘p’
denotes peak discharge. Inset: coefficient of variation of different indices for constant flows.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
2134 R. Egozi and P. Ashmore

Figure 14. (A–C) Ortho-rectified images showing an example of changes in braiding intensity with increasing stage on Sunwapta
River. Note that highest braiding intensity is in B, below the peak daily discharge. For scale car length at the lower right corner of
the images is 3 m. The cross-section on the left-hand side of image C is 90 m. (D) Mean braiding intensity versus flow stage over
a series of rising limbs of diurnal hydrograph on the Sunwapta River. The arrows indicate the general trend and were fitted by eye.
The solid point indicates BIT3 peak value. The outlier marked by the triangle symbol was measured during the flood peak discharge
when a change in channel morphology occurred.

The field data (using only the BIT3 index) also indicate that braiding intensity increases with flow stage but peaks at
a discharge value of ~11 m3 s−1 (equivalent to stage 3 in the simulated hydrograph), which is lower than the highest
discharge (13·7 m3 s−1) for which braiding intensity was measured (Figure 14). Figure 14 shows that over a range of
9 m3 s−1 (4·7–13·7 m3 s−1), which is equivalent to the five stages of the simulated hydrograph, the change in BIT3 is
between a minimum of 5·8 and a maximum of 8·8, which is a statistically significant difference (based on a t-test with
p < 0·05). Although both laboratory and field cases show similar response of braid index to flow stage, the Sunwapta
River has higher braiding intensity than the physical model for equivalent channel-forming conditions (slope, grain
size and discharge).

Discussion
In principle, as with other morphological properties of a river, it is necessary to make measurements of braid index
along a representative reach length that should be scaled with the size of the river and the length scale of the braided
pattern (Ashmore, 2001) so as to sample the bar-scale variability in channel pattern. In fluvial morphology studies it is
common to normalize reach length in terms of channel-width units because it is the lower limiting scale unit of a river

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
Defining and measuring braiding intensity 2135

channel pattern (Nikora, 1991; Church, 2007). Previous studies have sometimes prescribed a reach length but have
provided very limited rationale for the choice of reach length. Rust (1978) suggested a reach length of several
‘wavelengths’ but recognized some practical problems with his definition. Howard et al. (1970) suggested a reach of
at least twice the braid-plain width, while Brice (1964) suggested a reach length equal to ten times the bank-full width.
Our analysis shows that a minimum sample length of at least 10 times the channel width (measured as the average
wetted width at channel-forming flow) is needed to measure braiding indices with a coefficient of variation ≥20%. The
inherent local variability in braid index prevents any greater precision. From these experiments, ten times the channel
width is equivalent to approximately two to three times the braid wavelength (as defined by Ashmore, 2001) and this
wavelength gives an alternative estimate of the required sample length if wetted width at channel-forming discharge is
not known.
Total sinuosity index, PT is ~1·6 times higher than BIT3 (Figure 10A) when applied to the same braided pattern. This
confirms Bridge’s (1993) concern that PT and BIT3 are not the same and contradicts Robertson-Rintoul and Richards’
(1993) result that they are equivalent. However, the cause of the difference between the two indices is not the
influence of sinuosity that Bridge (1993) proposed. The channels all have low sinuosity and also PT and PT* (which
incorporates some of the sinuosity effect) give very similar values. An alternative explanation for the difference
between BIT3 and PT is the orientation of the anabranches relative to the river orientation. Based on reported values of
junction and bifurcation angles, anabranch orientation could average 30°, and locally exceed 45°, relative to the river
axis (Mosley, 1976; Ashmore and Parker, 1983; Federici and Paola, 2003; Bertoldi and Tubino, 2005; Zolezzi et al.,
2006; Bertoldi et al., 2006). If anabranch orientation is a significant effect then it could be hypothesized that

BIT3 ≈ PT × cos α (10)

where α denotes median angular deviation of anabranches from the long axis of the river. Figure 15 shows the
consequence of making this adjustment for α of 40°, which brings the two indices much closer to a 1:1 relationship.
A similar analysis is performed on a 1·14 km braided reach of the Sunwapta River. Both PT and BIT3 were measured
from the same aerial photograph. The PT value was 7 but when adjusted using Equation 10 gives a value of 5·4, which
is almost the same as a BIT3 of 5·3. This shows that adjustment is possible between these two indices but more work
is needed on the validity of this effect and on average anabranch orientation angles.
It is clear that, as has often been realized in the past, there is a systematic effect of stage on the braided index and
a single measurement of braiding intensity at one value of discharge cannot adequately characterize a braided reach. A
reference discharge for making measurements (Surian, 1999), such as the ‘bankfull’ discharge, is difficult to define
(Carson, 1984), and furthermore, at sustained high flow, the channel morphology can change during the flood
event (Mosley, 1982). Kellerhals et al. (1976) argued that mean daily discharge is an appropriate flow to make
measurements at because high and low elevation bedforms are exposed at this common stage. Others have suggested

Figure 15. BIT3 versus corrected PT (using Equation 10) for experiments 7, 8 and 14.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
2136 R. Egozi and P. Ashmore

variations on this including half bank-full discharge (Rust, 1978), mean annual discharge (Begin, 1981), or discharges
based on frequency analysis (e.g. two year flood), which do not depend on a reference to channel geometry (Carson,
1984; Bridge, 1993). An alternative would be a use of a reference discharge related to the ‘effective discharge’ for
bedload transport (Davies, 1988), but this may be more complicated to apply than an index discharge. Van der Nat
et al. (2002) show that the full range of water-stage variation is important in order to understand braiding dynamics,
and recent observations (Kiss and Sipos, 2007) point out the geomorphological significance of low-flow stages to
braided bars. Here we have begun to quantify the magnitude of the stage effect but more work is needed before it is
possible to conclude if there is a general relationship that might be used to define a suitable sampling strategy or adjust
braiding index measurements made at differing relative discharges in a river or rivers. In any case it is clear that
analysis of braiding index as a morphological response variable requires measurement at a range of stages up to that
at which peak braiding index occurs.

Conclusions
Systematic experimental assessment of various braiding indices indicates that a stable estimate of reach-averaged
braiding intensity, using any index, needs a sample length equivalent to a minimum of 10 times the average wetted
width at channel-forming discharge. Even so, braiding intensity can be estimated to a precision of only about 20% due
to natural bar-scale variability in braiding intensity over time and space. In addition, channel count indices need a
minimum number of cross-sections (10 in our experiments) spaced no further apart than the average wetted width of
the river.
The two most commonly used indices, total sinuosity index (PT) and channel count index (BIT3), are not directly
equivalent, in contrast to the findings of a previous analysis (Robertson-Rintoul and Richards, 1993). Anabranch
orientation relative to the main flow direction, rather than channel sinuosity (Bridge, 1993), is proposed as the main
reason for dissimilarity between the two indices.
Many of the existing indices are very similar and give equivalent functional relations with discharge. However, the
channel count index is distinctly different from the others and conversion factors are needed to relate the other indices
to the channel count index.
A channel count index is preferred because it is not sensitive to variations in channel sinuosity and orientation, has
the smallest coefficients of variation and can be measured very quickly and reliably even from oblique photographs of
a reach. It is also the least sensitive to river-stage effects.
Braiding intensity is sensitive to flow stage, although this varies between indices. In all cases braid indices are
significantly different between maximum and minimum flow stage of a daily melt-water hydrograph. Consequently
the conversion between indices may also depend on the relative stage at which measurements are made. As previously
suggested, measurement of braid index should take account of the effects of river stage and therefore braiding
intensity should be measured at several flow stages or be referred to an index discharge, on which further research is
needed.

Acknowledgments
This research was funded by a NSERC grant awarded to P. Ashmore and supported by a Canada Foundation for Innovation grant
used in constructing the flume. R. Egozi was partially supported by an Ontario Graduate Scholarship. We thank Tim Davies and
Nicola Surian for their comments which improved the manuscript, K. VanKerkoerle for help with the figures, Dr. J. Chandler for
assisting in the orthorectification process, and several people who assisted with the experiments, especially Tobi Gardner. We also
thank Parks Canada for allowing us to take measurements in Sunwapta River, Jasper National Park, AB and several fieldwork
collaborators who assisted in the field measurements, notably Thomas Buffin-Belanger, Walter Bertoldi and Emilie Sauks.

References
Ashmore PE. 1991. Channel morphology and bed load pulses in braided gravel-bed streams. Geografiska Annaler 73A: 37–52.
Ashmore PE. 2001. Braiding phenomena: statics and kinemetics. In Gravel-Bed Rivers V, Mosley PM (ed.). New Zealand Hydrological
Society: Christchurch; 95–120.
Ashmore PE, Parker G. 1983. Confluence scour in coarse braided stream. Water Resources Research 19: 392– 402.
Ashmore PE, Sauks E. 2006. Prediction of discharge from water surface width in braided river with implication for at-a-station hydraulic
geometry. Water Resources Research 42: W03406. DOI: 10.1029/2005WR003993
Begin ZB. 1981. The relationship between flow-shear stress and stream pattern. Journal of Hydrology 52: 307–319.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
Defining and measuring braiding intensity 2137

Bertoldi W, Tubino M. 2005. Bed and bank evolution of bifurcating channels. Water Resources Research 41: W07001. DOI: 10.1029/
2004WR003333
Bertoldi W, Miori S, Salvaro M, Zanoni L, Tubino M. 2006. Morphological description of river bifurcations in gravel bed braided networks.
In Proceedings of the International Conference on Fluvial Hydraulics, River Flow 2006, 6–8 September, Lisbon, Portugal.
Brice JC. 1960. Index for description of channel braiding. Geological Society of America Bulletin 71: 1833.
Brice JC. 1964. Channel patterns and terraces of the Loup Rivers in Nebraska. Geological Survey Professional Paper 422-D.
Bridge JS. 1993. The interaction between channel geometry, water flow, sediment transport and deposition in braided rivers. In Braided
Rivers, Best JL, Bristow CS (eds). The Geological Society: London; 13–72.
Carson MA. 1984. The meandering–braided river threshold: a reappraisal. Journal of Hydrology 73: 315–334.
Chew LC, Ashmore PE. 2001. Channel adjustment and a test of rational regime theory in a proglacial braided stream. Geomorphology 37:
43–63.
Church M. 2007. Multiple scales in rivers. In Gravel-bed Rivers VI: from Process Understanding to River Restoration, Habersack H,
Piegay H, Rinaldi M (eds). Elsevier: Amsterdam; 3–32.
Clarke MRB. 1980. The reduced major axis of bivariate sample. Biometrika 67: 441– 446.
Davies TRH. 1987. Problems of bed load transport in braided gravel-bed rivers. In Sediment Transport in Gravel-bed Rivers, Thorne CR,
Bathurst JC, Hey RD (eds). Wiley: Chichester; 793–828.
Davies T. 1988. Modification of bedload transport capacity in braided rivers. Journal of Hydrology (N.Z.) 27: 69–72.
Egozi R. 2006. Channel pattern variation in gravel bed braided rivers. PhD Thesis, The University of Western Ontario: London.
Federici B, Paola C. 2003. Dynamics of channel bifurcations in noncohesive sediments. Water Resources Research 39(6): 1162. DOI:
10.1029/2002WR001434
Friend PF, Sinha R. 1993. Braiding and meandering parameters. In Braided Rivers, Best JL, Bristow CS (eds). The Geological Society,
London: 105–112.
Fujita I. 1989. Bar and channel formation in braided stream. In River Meandering, Ikeda S, Parker G (eds). Water Resources Monograph 12,
American Geophysical Union: Washington, DC; 417– 462.
Germanoski D, Schumm SA. 1993. Changes in braided river morphology resulting from aggradation and degradation. Journal of Geology
101: 451–466.
Gran K, Paola C. 2001. Riparian vegetation controls on braided stream dynamics. Water Resources Research 37: 3275–3283.
Hoey T, Sutherland AJ. 1991. Channel morphology and bedload pulses in braided rivers: a laboratory study. Earth Surface Processes and
Landforms 16: 447– 462.
Hong LB, Davies TRH. 1979. A study of stream braiding. Geological Society of America Bulletin 90(Part II): 1839 –1859.
Howard AD, Keetch ME, Vincent CL. 1970. Topological and geometrical properties of braided streams. Water Resources Research 6: 1674 –
1688.
Kellerhals R, Church M, Bray DI. 1976. Classification and analysis of river processes. Journal of Hydraulic Division, ASCE 102: 813–829.
Kinoshita R. 1957. Formation of dunes on river beds – an observation on the condition of river meandering. Proceedings of the Japanese
Society of Civil Engineers 42: 1–21.
Kiss T, Sipos G. 2007. Braid–scale channel geometry changes in sand-bedded river: significance of flow stages. Geomorphology 84: 209–
221.
Leopold LB, Wolman MG. 1957. River channel patterns: braided, meandering and straight. Geological Survey Professional Paper 282-B:
85 pp.
Mosley PM. 1976. An experimental study of channel confluences. Journal of Geology 84: 535–562.
Mosley PM. 1981. Semi-determinate hydraulic geometry of river channels, South Island, New Zealand, Earth surface Processe and
Landforms 6: 127–137.
Mosley PM. 1982. Analysis of the effect of changing discharge on channel morphology in stream uses in a braided River, Ohau River, New-
Zealand. Water Resources Research 18: 800–812.
Mosley PM. 1983. Response of braided rivers to changing discharge. Journal of Hydrology (N.Z.) 22: 18–67.
Nikora VI. 1991. Fractal structures of river plan forms. Water Resources Research 27: 1327–1333.
Parker G. 1976. On the cause and characteristic scales of meandering and braiding in rivers. Journal of Fluid Mechanics 76: 457–480.
Pine Tree Computing (2004). https://secure3.ntwebb.com/pinetreecomputing-com/default.asp (accessed 12 July 2004).
Richards KS. 1982. Rivers: Form and Processes in Alluvial Channels. Methuen, 358 pp.
Robertson-Rintoul MSE, Richards KS. 1993. Braided channel pattern and paleohydrology using an index of total sinuosity. In Braided
Rivers, Best JL, Bristow CS (eds). The Geological Society: London; 113–118.
Rust BR. 1978. A classification of alluvial channel systems. In Fluvial Sedimentology, Miall AD (ed.). Canadian Society of Petroleum
Geologist: Alberta; 187–198.
Shvidchenko AB, Kopaliani ZD. 1998. Hydraulic modeling bed load transport in gravel bed Laba River. Journal of Hydraulic Engineering
124: 778–785.
Smith LC, Isacks BL, Bloom AL, Murray BA. 1996. Estimation of discharge from three braided rivers using synthetic aperture radar
satellite imagery: potential application to ungauged basins. Water Resources Research 32(7): 2021–2034.
Surian N. 1999. Channel changes due to river regulation: the case of the Piave River, Italy. Earth Surface Processes and Landforms 24:
1135–1151.
Van der Nat D, Schmidt AP, Tockner K, Edwards PJ, Ward JV. 2002. Inudation dynamics in braided floodplains: Tagliamento River,
Northeast Italy. Ecosystems 5: 636–647.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp
2138 R. Egozi and P. Ashmore

Varkaris FC. 2002. Variability in Bedload Sediment Transport Estimates Sunwapta River, Alberta, Canada. MSc thesis, The University of
Western Ontario: London.
Warburton J, Davies TRH. 1994. Variability of bedload transport and channel morphology in braided river hydraulic model. Earth Surface
Processes and Landforms 19: 403–421.
Warburton J, Davies TRH. 1998. Use of hydraulic models in management of braided gravel-bed rivers. In Gravel-bed Rivers in the
Environment, Klingeman PC, Beschta RL, Komar PD, Bradley JB (eds). Water Resources Publications: Highlands Ranch, CO; 512–542.
Williams PF, Rust BR. 1969. The sedimentology of a braided river. Journal of Sedimentary Petrology 39: 649–679.
Zolezzi G, Bertoldi W, Tubino M. 2006. Morphological analysis and prediction of river bifurcations. In Braided Rivers: Process, Deposits,
Ecology and Management, Sambrook-Smith G, Best JL, Bristow CS, Petts GE (eds). Special Publication 36, International Association of
Sedimentologists. Blackwell Publishers: Oxford; 233–256.

Copyright © 2008 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 33, 2121–2138 (2008)
DOI: 10.1002/esp

View publication stats

You might also like