Formation of a cohesive floodplain in a dynamic experimental meandering river

Field studies suggest that a cohesive floodplain is a necessary condition for meandering in contrast to braided rivers. However, it is only partly understood how the balance between floodplain construction by overbank deposition and removal by bank erosion and chutes leads to meandering. This is needed because only then does a dynamic equilibrium exist and channels maintain meandering with low width–depth ratios. Our objective is to understand how different styles of floodplain formation such as overbank deposition and lateral accretion cause narrower channels and prevent chute cutoffs that lead to meandering. In this study we present two experiments with a self‐forming channel in identical conditions, but to one we added cohesive silt at the upstream boundary. The effect of cohesive silt on bank stability was tested in auxiliary bank erosion experiments and showed that an increase in silt reduced erosion rates by a factor of 2. The experiment without silt developed to a braided river by continuous and extensive shifting of multiple channels. In contrast, in the meandering river silt deposits increased bank stability of the cohesive floodplain and resulted in a reduction of chute cutoffs and increased sinuosity by continuous lateral migration of a single channel. Overbank flow led to deposition of the silt and two styles of cohesive floodplain were observed: first, overbank vertical‐accretion of silt, e.g. levee, overbank sedimentation or splays; and second, lateral point bar accretion with silt on the scrolls and in the swales. The first style led to a reduction in bank erosion, while the second style reduced excavation of chutes. We conclude that sedimentation of fine cohesive material on the floodplain by discharge exceeding bankfull is a necessary condition for meandering. Copyright © 2013 John Wiley & Sons, Ltd.


Introduction
Rivers can have various channel patterns, such as braided and meandering (e.g. Leopold and Wolman, 1957;Schumm and Khan, 1972). It has long been hypothesized that cohesive floodplain material or vegetation adds strength to river banks, and that this is a necessary condition for meandering (e.g. Ferguson, 1987). Rivers with a cohesive floodplain develop into a meandering river, while non-cohesive floodplains lead to channel widening which eventually results in braiding (Parker, 1979;Ferguson,1987;Kleinhans, 2010). Still, to experimentally reproduce a sustained dynamic meandering channel pattern in the laboratory has proven difficult, so that the exact conditions for meandering remained unclear. In the experimental work of Braudrick et al. (2009), meandering was sustained by the addition of vegetation to the floodplain. Geomorphological evidence of meandering rivers has been found on Mars, where vegetation cannot have played a role (Howard, 2009). Here we report on experiments which resulted in a braided and a meandering river, where the only difference was the addition of cohesive fines in the sediment feed.
Meandering rivers are characterized by a high sinuosity as meander bends can migrate laterally and increase the bend amplitude, while remaining single-threaded. Braided rivers have a low sinuosity and are characterized by multiple threads.
In meandering rivers, channel migration of the bends is slow (Hickin and Nanson, 1984) and bends develop in phases of creation, growth and abandonment (Camporeale et al., 2005). An important aspect of bend migration is bank erosion (Kleinhans, 2010). First, banks are undercut by fluvial erosion at the base and lower portion of the banks; second, bank retreat occurs by mass failure of the bank (Darby et al., , 2007Simon et al., 2000;Simon and Collinson, 2002). Then, the bank sediment settles at the bank toe and armoring protects the bank against fast erosion (Thorne, 1982;Parker et al., 2011) and shifts the locus of the high flow velocity, which reduces the shear stress acting on the bank Smith, 2006a, 2006b;Darby et al., 2010). Several studies attempted to predict bank erosion rates by calculating the bank erosion processes (e.g. Ikeda et al., 1981;Langendoen and Simon, 2008;Rinaldi and Darby, 2008;Parker et al., 2011). The strength of the bank depends on the floodplain style and floodplain cohesion. Other studies empirically linked bank erosion rates in bends to flow processes of channel depth, bend curvature, friction with the bank (Hickin and Nanson, 1984;Furbish, 1988) and the adaptation length of momentum redistribution of the flow across the curved channel (Struiksma et al., 1985). Bend migration rate generally increases when bends become sharper. On the other hand, in braided rivers migration rates are high even without sharp bends (Hooke, 2003). High bank erosion rates in experiments with cohesionless sediment led to channel widening and the development of a braided river (Ashmore,1991). Therefore, we hypothesize that cohesive floodplains are required to have stronger banks to sustain low width-depth ratios.
Chute cutoffs, which shorten the flow path, are a limiting process in the development of highly sinuous bends. The development of chute cutoffs is described in several field studies (e.g. Constantine et al., 2010;Micheli and Larsen, 2011;Zinger et al., 2011). Furthermore, chutes have limited the development of high sinuous meandering rivers in earlier experiments (Friedkin, 1945;Peakall et al., 2007;Braudrick et al., 2009;Tal and Paola, 2010;Van Dijk et al., 2012). To sustain meandering, chute cutoffs have to be limited. Cutoffs can be prevented by vegetation growth to stabilize banks, but meandering rivers form also in areas where vegetation does not play a role, e.g. intertidal muds (Kleinhans et al., 2009), Martian rivers (Howard, 2009), glaciers (e.g. Gorner Glacier, 45˚58' 11' N, 7˚48' 6" E; observation by WMvD) and deserts (Matsubara et al., 2011). Therefore, we hypothesize that cohesive sediment deposition on the point bars is a sufficient condition to prevent chute cutoffs.
Sediment erosion by bend migration and cutoff is balanced by deposition of sediment forming new floodplains. Lateral migration of the channel leads to erosion of the higher outer bank, while lateral accretion and floodplain construction on the inner side of the bend is lower (also known as floodplain shaving; Lauer and Parker, 2008). The process of floodplain shaving and channel extension results in local differences between erosion and deposition. This difference is balanced by overbank deposition or by filling depressions (e.g. abandoned channels; Lauer and Parker, 2008). Floodplains of silt and clay are constructed during floods (e.g. Middelkoop and Asselman, 1998), with more deposition near the channel and a general decrease of fine deposition with increasing distance from the channel (e.g. Walling and He, 1997;Törnqvist and Bridge, 2002). This shows that to build a cohesive floodplain flow discharges are required that at least temporarily exceed bankfull height, so that fines deposit on the high non-cohesive banks and in the disconnected lows to balance the local differences in elevation.
Construction of a new floodplain can occur in different styles. First, an important process in this context is lateral point bar accretion (e.g. Walling and He, 1997;Törnqvist and Bridge, 2002) on the inner side of the bend, which forms scroll bars that, with overbank deposition, becomes a floodplain (Jackson, 1976;Nanson, 1980). This floodplain consists of varying grain sizes and is mostly occupied during high flow stages (Nanson and Croke, 1992). Second, overbank flow on the outer bank and at the edge of the channel forms vertical accretion (Nanson and Croke, 1992), e.g. levees and splays (Brierley et al., 1997;Cazanacli and Smith, 1998). These splays could build out forming crevasses, but could also lead to avulsion (Pérez-Arlucea and Smith, 1999). Overbank sedimentation produces a floodplain that consists of a non-cohesive bed with a cohesive layer on top. Abandoned channels are filled by deposition of relatively coarse sediments that build plug bars (Toonen et al., 2012). After disconnection, finer sediments fill the remaining depressions (e.g. Lewis and Lewin, 1983). An experimental test of the construction of a cohesive floodplain and how bank stabilization leads to a meandering river has not been done to date.
Earlier studies have shown that the addition of bank cohesion decreases channel migration and changes channel width-depth ratio when bank stability increases (Friedkin, 1945;Schumm and Khan, 1972;Smith, 1998;Gran and Paola, 2001;Peakall et al., 2007;Braudrick et al., 2009;Tal and Paola, 2010). However, in prior experiments bank strength was provided by adding cohesive sediments or vegetation seeds manually on top of the floodplain (Friedkin, 1945;Schumm and Khan, 1972;Gran and Paola, 2001;Braudrick et al., 2009;Tal and Paola, 2010). In an earlier experiment (Van Dijk et al., 2012) we showed the development of a meandering river with a weakly cohesive point bar cover that nevertheless developed several chute cutoffs. Therefore, in this study we added more fines and used a simple hydrograph for overbank sedimentation compared to the experimental meandering river in Van Dijk et al. (2012) with constant discharge. In this paper we test how the river sustains meandering when the bank is stabilized by a self-formed cohesive floodplain, while a floodplain without cohesion results in a braided river. We refer to self-formed cohesive floodplain as the floodplain formed by sediment deposition distributed by the water flow after the initial conditions. The unchanged initial banks are referred to as pristine plain and are non-cohesive. The area where the river changes the bed/bank elevation is referred to reworked floodplain.
Here we show how the cohesive floodplain forms over the duration of the experiment and how this maintained a meandering channel, as well as demonstrating how the lack of cohesive floodplain led to braiding under otherwise equal conditions. To systematically evaluate bank erodibility in the experiment, tens of small-scale bank erosion tests were conducted. The objective of this study was to assess the effect of cohesive floodplain fines on (1) the floodplain formation (2) bank erosion and cutoff processes and (3) the channel pattern. This paper is structured as follows. First, we describe the setup and boundary conditions of the experiments, the measurement techniques, and the setup for the bank erosion tests. Second, we present results describing the bank erosion test, the detailed morphology, the water depth and the construction of the cohesive floodplain. Finally, we discuss floodplain formation and bank erosion processes based on the results of the bank erosion tests and the sequence of bend development observed in the experiments.

Experimental Setup, Methods and Materials
The experiments were set up to represent a gravel-bed river dominated by bedload transport (Kleinhans and Van den Berg, 2011). The designed conditions were not based on direct scaling from a particular natural river, but on an optimal reduction of scaling issues derived from a large number of pilot experiments ( Van de Lageweg et al., 2013;Van Dijk et al., 2012). We designed experimental conditions that compromise between the most important scaling issuesin particular, low sediment mobility, prevention of scour holes and cohesion of the floodplain sediment. The experiments were scaled down in discharge compared to our earlier experiment (Van Dijk et al., 2012;Table I), so that in the same length of the flume more bends could develop. Additionally, small-scale bank erosion tests were conducted to quantify the influence of silt on bank erosion rates.

Bank erodibility
Earlier experiments Most experiments resulted in a braided planform by reoccupation of depressions, which form when erosion exceeds deposition (Ashmore, 1991). To obtain self-formed meanders in the lab, earlier experiments reduced the bank erosion rate by having stronger banks. A decrease in erosion rate should lead to a longer time period for sediment deposition on the inner side of the channel, so that the local differences between erosion and deposition were balanced. The earlier experiments could be divided into two different types of bank stabilization. First, several prior studies added a cohesive mixture of clay in the bed, so that inner bend floodplain formation should keep up with the outer bank erosion. These experiments led to the formation of static meanders as lateral migration ceased when the bank cohesion was too high (Friedkin, 1945;Schumm and Khan, 1972;Smith, 1998). The addition of a less cohesive silt increased bank strength of the non-cohesive bed and formed a single bend (Peakall et al., 2007). Second, others have used vegetation to add bank strength. Alfalfa (Medicago sativa) sprouts seeded on a braided experimental river led to local bend migration but the pattern that formed in these experiments is best characterized as wandering (Gran and Paola, 2001;Tal and Paola, 2010) rather than truly meandering. Furthermore, alfalfa sprouts increased floodplain deposition of cohesionless fines, resulting in a sustained meandering system with a moderate sinuosity (Braudrick et al., 2009).
Bank erosion tests Too much bank stability decreases the dynamics of the river (Friedkin, 1945;Schumm and Khan, 1972;Smith, 1998;Gran and Paola, 2001). Therefore, we tested systematically the effect of different amounts of silt concentration on bank erosion rates. These experiments were inspired by the work of Friedkin (1945). Tens of small-scale experiments of bank retreat were conducted ( Figure 1). These tests were carried out to quantitatively assess the effect of different sediment mixtures on bank erosion rates and processes. Experiments were conducted in a flume with a duct of 50 mm wide and 1 m long on a slope of 0.01 m/m and a discharge of 400 L/hr. At the end of the entrance an experimental sediment block was placed. Here the water flow attacked the bank with a sharp angle of 45å nd an initial 50 mm channel width (see also Van de Lageweg et al., 2010;Kleinhans et al., 2010).
Two series of bank erosion experiments were performed to test erosion rates for two different styles of floodplains. (1) The effect of sediment mixtures with different silt concentration represented continuous entrainment of sediment from the banks with lateral silt accretion and without undercutting of the bank. These banks were observed in the point bar, where overbank flow caused chute excavation. (2) The effect of a vertical accretion of stacked silt layers on top of a non-cohesive bed was tested, which was observed in the experiment when levees, overbank sedimentation or remnants of the crevasse were undercut by the flow at the outer bend. For the first, a floodplain consisting of different ratios between sand and silt was tested for erodibility (e.g. inner bend floodplain), where the experimental sediment block was 20 mm thick. For the other one, a floodplain was tested which had a cohesive silt drape on top of the non-cohesive sand (e.g. outer bank floodplain).
Here the experimental sediment block consisted of an 8 mm, i.e. bankfull level, thick poorly sorted sand, which was draped with different thickness of silt layers.
Bank erosion rates were measured from time-lapse photography of the experimental sediment block. The progress retreat of the bankline of the experimental sediment block was obtained by image processing. The bankline was used to calculate sediment area and, as thickness was known, the volume of the experimental sediment block. Data were then reduced to half-life times to characterize bank erosion rates, which is defined as the time taken to reduce the volume of the experimental sediment block to half the initial volume.

Flume setup and experimental procedure
The experiments represented a gravel-bed river and were scaled by similarity of dimensionless variables for hydraulic conditions, sediment transport conditions and morphological  features, which had to remain within specific ranges ( Table I).
The flow had to be subcritical (Froude number, Fr < 1) as in most rivers. Turbulent flow was necessary to rework the sediment and to transport sediment in suspension in the channel and on the floodplain (Reynolds number, Re > 2000). For sediment transport conditions bedload sediment should be mobile θ > θ cr (Shields mobility number). A small ratio of particle size to laminar sublayer thickness is known to be conducive, so that the near-bed flow conditions affect bed scouring and ripple formation . The channel should therefore have a hydraulically rough bed, for which large particles were needed to disrupt the laminar sublayer (grain Reynolds number, Re * > 11.6). For morphological features the channel width-depth ratio determined the bar mode and bar formation, which is determined by bar wavelength and interaction parameter (Kleinhans and Van den Berg, 2011; Table I). This required that the channels had enough bank strength so that they did not became too wide and shallow, which ultimately leads to braiding. There are no rules for scaling bank strength, except that t/s > 1 (t is the shear strength and s is the normal stress). Therefore, we conducted small-scale bank erosion tests to estimate sufficient conditions for erosion processes to continue, yet maintain bank stability at values higher than for cohesionless sediment.
The experiments were carried out in a flume of 6 m wide and 11 m long, which was divided into two separate plains of 3 m wide and effectively 10 m long. The flume was filled with a 100 mm thick layer of poorly sorted sand ( Figure 2). The initial bed was set at a gradient of 0.01 m/m. We carved a 150 mm wide by 10 mm deep straight channel in the sediment, corresponding approximately to the predicted hydraulic geometry of a non-cohesive gravel-bed river (Parker et al., 2007) and self-formed channels in pilot experiments. The downstream boundary had a fixed weir, so that the base level was kept at a constant level. Upstream, we varied the inlet position of the sediment and water feeder with a lateral migration rate of 10 mm/hr, to mimic a bend that translates into the flume (see also Van Dijk et al., 2012).
We carried out two identical experiments, which only differ in the availability of cohesive fines in the sediment feed. The addition of fines in the sediment feed represents the addition of cohesive fines to the sediment load, which led to the transition between braiding and meandering in the Rhine-Meuse delta. To one experiment we added silt-sized silica flour (D 10 , D 50 and D 90 ; 3.7, 32 and 97 mm, respectively) in a ratio of 1:4 in the sediment feed ( Figure 2). Furthermore, an extra amount of 0.5 L silt was separately supplied during each high discharge to build a cohesive floodplain in the experiment. The Rouse number (Equation 1 in Rouse, 1937) for the silt-sized fraction was smaller than 1.2, which indicates that the fines will transport in suspension: where P is the non-dimensional Rouse number, w s is the sediment fall velocity (in m/s), k is the Von Kármán constant (0.4) and u * is the shear velocity (in m/s).
Deposition of the fine silt-sized fraction will probably not affect sediment entrainment due to changes in critical shear stress despite the reduction of median grain size. First, the addition of silt to the sand made the mixture bimodal, so that mobility differed between the two sediments and not increased mobility of the total mixture (Wilcock and Southard, 1988;Kleinhans and Van Rijn, 2002). Second, although the silt is not cohesive like clay, Lick et al. (2004) show that the critical shear stress increases for particles smaller than 50 mm. Third, fine particles will percolate into the bed (Frings et al., 2008) and calculations of the cutoff size of the sediment mixture shows that particles smaller than 20 mm (40% of the silt-sized silica flour) will affect neither bed level nor bed roughness ( Figure 2; Frings et al., 2008Frings et al., , 2011.
A simple schematic hydrograph was used with a Q high = 1800 L/hr (0.5 L/s) for 30 min and Q low = 900L/hr (0.25 L/s) for 2.5 h. We ignore hysteresis of wash load supply that is often observed in natural river floods (Asselman, 1999). In the hydrograph, low flow represents approximately bankfull discharge based on the predicted hydraulic geometry in a non-cohesive gravel-bed river (W = 200 mm, h = 9 mm, according to Parker et al., 2007), whereas high flow exceeds bankfull and distributed the fine sediment on the floodplain. The flood flow had an intermittency of 1:5, where 20% is flood stage and 80% of the time is bankfull stage. With constraints on the maximum discharge in the flume, the sediment had to be on average the right mobility. Furthermore, it was the duration and magnitude of low and high flow that were determined together. We designed the hydrograph to the same average discharge as in an experiment with a constant flow discharge of Q c = 1080 L/hr (0.3 L/s), so that a volume of about 3200 L in 3 h flowed through both experiments. The sediment feed was kept constant at 0.2 L/hr of bedload sediment ( . Hydrograph of one low and high discharge stage with sediment feed rates. In both experiments 0.2 L/lhr sand was continuously fed. In the meandering river 0.05 L/lhr of silt was continuously fed and during a high stage an extra 0.5 L of silt was added. The star indicates the moment that the bed topography was scanned and photographed. This figure is available in colour online at wileyonlinelibrary.com/journal/espl rotameter and the sediment feed was controlled by a sediment feeder. Each experiment ran for 120 h with one full cycle of the upstream moving boundary, with an amplitude of 300 mm in both directions.

Measurements and calibration
Several measurement techniques were used to record the morphodynamics of the experimental rivers. Overhead photos were taken at 5 min intervals to create a time-lapse video of the experiment. Further, the flume was equipped with an automatic gantry, on which we mounted a high-resolution camera (0.25 mm ground resolution) and a laser line scanner (0.2 mm vertical resolution). We measured and photographed the bed after each high discharge by pausing the experiment ( Figure 3). Two LED floodlights were mounted on the automatic gantry to suppress ambient lighting. The point cloud from the line-laser was gridded on a 4 mm grid by median filtering to produce digital elevation models (DEMs). The initial bed surface slope of the DEMs was subtracted to detrend the DEMs. The detrended elevation was expressed relative to the surface that remained unchanged. DEMs of difference (DoD) were calculated by subtracting DEM pairs. DoDs were thresholded by the vertical resolution (0.2 mm) of the laser line scanner. A sediment balance was calculated by summation of the thresholded DoDs (Equation (2)). The sediment balance volumes between time steps t and t + 1 were calculated for all grid cells m based on the inversed Exner equation: where V is volume (in dm 3 ), z is bed level (in dm), dx is grid size in the x-direction (in dm) and dy is grid size in the y-direction (dm), i is grid cell index and m is total number of grid cells.
To describe the evolution of experiments in the flume, the total and active braiding index (TBI and ABI; Egozi and Ashmore, 2009;Bertoldi et al., 2009), the sinuosity and distribution of the surface elevation were calculated for every time step.The water was dyed with a red color dye (Rhodamine B) to determine the channel position and water depth. The TBI, defined as the number of wetted channels per cross-section, was taken as the average number of channels (from six crosssections at distance of 2, 3.5, 5, 6.5, 8 and 9.5 m along the flume) identified on the DEM and high-resolution photographs where the red color band of the images corresponded to the red dyed water. The ABI, defined as the number of channels that transport sediment in a cross-section, was the average number of channels which also had net morphological change (e.g. erosion or deposition) observed on the DoD maps at the six identical cross-sections as the TBI. The frequency distribution of the detrended surface elevation (characterized by percentiles Z 5 , Z 50 and Z 95 ) was used to check whether the experiments did not aggrade or degrade, and to test whether the experiment with cohesive silt developed deeper channels and higher floodplains, as compared to the experiment without cohesive silt.
The high-resolution images were used to derive the concentrations of silt on the floodplain, to segment channels and to deduce water depth (based on Carbonneau et al., 2006;Tal and Paola, 2010). The high-resolution camera with RGB band gives values for green, red and blue, which can be transformed to a L*a*b* color space (CIELAB). Herein, L* represents the luminosity (low = black and high = white), a* is the position between red/magenta (high values) and green (low values), and b* is the position between yellow (high values) and blue (low values). The luminosity was used to make distribution maps of the highly reflective silt. Therefore, 18 samples of silt were related to the luminosity difference between the current image and theinitial image of the bed (Figure 4a). The intensity of the redness (a*) was related to water depth for each time step, as the redness reduced during the experiment. The relation between water depth and redness intensity was found by relating the bed elevation on a cross-section to the redness intensity at that cross-section. For this relation, only points were included of the active channel and the overbank flow on the outer bank (Figure 4b, c).

Results
In the experiment without fines a braided river formed, while the addition of cohesive fines produced a single-thread meandering river. In this section, we describe the effect of silt on bank erosion, morphodynamics of the braided and meandering river, cohesive floodplain formation and the effect of cohesive floodplain fines on bank stabilization and chute excavation.  erosion rates. Initially, erosion was rapid and declined with the decrease of the experimental sediment block volume and increased channel width. Erosion of the experimental sediment blocks without a silt layer on top occurred continuously by sediment entrainment. Coarse grains of the poorly sorted sand were detached from the experimental sediment blocks. The coarse grains were not transported downstream immediately and caused local bank toe protection. A cohesive silt layer on top stuck together and was undercut, resulting in the detachment of failure blocks. Silt was not directly transported and deposited at the bank toe. We suspect that silt depositing on the bank toe led to hydraulic smoother conditions and increased critical shear stresses, which caused a decrease in bank erosion. The increase of silt on top of the bank made the bank line more irregular as erosion rates differed locally. Nevertheless, the sediment mixtures eroded continuously and no failure blocks were observed. Silt fraction within the non-cohesive sands decreased the entrainment of sediment and bank erosion rates. The addition of 20% silt decreased the bank erosion rate by a factor of 2 ( Figure 5). Bank strength increased with an increase of silt concentration, until the pores were filled by silt (around 20%, Figure 5a). When the pores of the sand were filled, silt became part of the bed structure (Frings et al., 2008). The addition of water led then to fluidazation of the silt, so that bank strength did not increase. A further increase in silt resulted in more variation of the half-life time caused by variation in compaction and a variance of the cutoff size between mixtures, so that shear stresses differ (Frings et al., 2008). The mean halflife time was the same for concentrations higher than 20% silt in the mixture (Figure 5a), while lower concentrations showed no significant effect on bank stability.

Auxiliary bank erosion experiments
The bank strength increased when a silt layer was on top of the non-cohesive poorly sorted sand. A 1 mm thick layer of silt about 13% of the total volume of the experimental sediment blockdid not reduce the half-life time (Figure 5b). An experimental sediment block with a 1.5 mm (19%) thick silt layer had a much higher bank strength and reduced bank erosion by almost a factor of 3. A poorly sorted experimental sediment block with a 2 mm (25%) silt layer decreased the bank erosion to a rate four times slower than without a silt layer.

Channel pattern
Our experiments resulted in a braided river as well as a meandering river (Figure 6 and supporting information Movie 1).  Figure 6. Channel evolution of a braided river (a) and a meandering river (b); shaded elevation maps (DEMs) detrended with the initial slope. Unchanged surface is masked by gray. Cross-profiles are indicated for Figure 14.

FORMATION OF A COHESIVE FLOODPLAIN IN AN EXPERIMENTAL MEANDERING RIVER
Lateral channel migration was the dominant process in both experiments. Ultimately, meandering was sustained because the channel experienced fewer chute cutoffs. This was caused by the addition of cohesive fine material that deposited on the floodplain and stabilized the banks.
Initial alternate bars The initial conditions for both setups were the same, so that in both experiments alternate bars formed. In this first phase, the alternate bars grew to an incipient meandering river. Thereafter, the bend wavelength increased and bends migrated in the downstream direction. The most upstream bend reached the downstream bar and a chute cutoff shortened the channel in both experiments ( Figure 6, 27 h (a) and 39 h (b), respectively). The incipient meandering rivers straightened and new bends formed as the upstream perturbation was maintained. The development after the cutoff of the incipient meandering river produced a braided and a meandering river. In both setups, bend formation was initiated by the upstream boundary, which continuously moved in the transverse direction.

Braided river
The river without cohesive fines produced a braided planform as the channel repeatedly cross-cut the self-formed floodplain. The braided river was characterized by multiple channels (Figure 7a), as indicated by the TBI, which was around 2 (Figure 8a). Nevertheless, sediment transport occurred mostly in one main channel (ABI just above 1.7). Lateral channel migration formed a point bar with typical scroll ridges and swales. Further, a subsidiary channel formed that later developed into a chute channel. After each cutoff, perturbation of the channel by the moving upstream boundary caused lateral migration, so that new bends were initiated again.
Later, continuous migration of the channel resulted in reoccupation of older channel depressions and more channels became active in the braided river (Figure 8a, b). Overbank flow was observed when the bend developed and later followed by a chute cutoff. The chute cutoff straightened the channel and lowered the water level, so that the total wetted area of the floodplain in the experiment decreased (Figure 7c). The wetted area is the area where water flows during low flow. We distinguish the total wetted area and the area that has been reworked, and the difference between both is the overbank flow.  (Figure 9b). Later, sediment loss continued as the reworked area increased (Figure 9a) and the floodplain shaved off further. As a result the overbank flow in the braided declined ( Figure 7c). Eventually, the total sediment balance shows a loss of 73 L, which is about 4.5 mm in height per unit area that had been reworked by the channel. Nevertheless, the channel did not incise in the floodplain as the Z 5 , Z 50 and Z 95 of the reworked area did not change over time (Figure 9c).

Meandering river
The meandering river was characterized by sustained lateral migration of the channel in the middle section of the flume. The bend in the middle section of the experiment translated and expanded, so that a large bend formed ( Figure 6, 117 h). The curvature, expansion and translation of the bend in the middle section controlled the bend downstream (e.g. migration and cutoffs). The ABI of the braided river was just above 1.5, while the ABI for the meandering river was smaller than 1.5 ( Figure 8b). Eventually, sustained lateral channel migration produced a single-thread meandering river where the sinuosity increased up to 1.4 (Figure 8b, c).
The meandering river developed after the cutoff of the incipient meandering river. In the initial channel alternate bars developed, which increased the sinuosity and roughness of the channel. When the bend became sharper, water level rose and overbank (floodplain) flow occurred even during low flow, so that for example the total wetted area was larger than the wetted area of reworked floodplain, while in the braided river they became equal (Figure 7c). Concentrated overbank flow resulted in a chute cutoffof the non-cohesive incipient meandering river and decreased the wetted area as the channel straightened ( Figure 6, 39 h). The upstream moving boundary triggered lateral migration of the bend and the eroded sediment deposited upstream of the chute channel and formed a plug bar ( Figure 6, 60 h). Later, the bend extended and translated, which resulted in a continuous increase of the bend amplitude and bend length ( Figure 6, 99 h). Eventually, a large bend developed in the middle section of the flume with a typical scroll ridge-swale topography ( Figure 6, 117 h). The bend was not cross-cut even when the upstream moving boundary moved in the reverse direction seen in Van Dijk et al. (2012).
In the meandering river fines settled on the floodplain when water level exceeded bankfull elevation. Overbank flow on the pristine plain occurred when the bends became sharper in the meandering river, but also for the incipient meandering river in both experiments (Figure 7c). The wetted area without overbank flow on the pristine plain was equal for the braided and the meandering river (Figure 7c). Overbank flow in the meandering river resulted in a large area of shallow water depth at the end of the experiment (Figure 7d). The water depth distribution without pristine overbank flow and the surface elevation at the end (Figure 9d) showed that the channels were slightly deeper and the floodplain slightly higher for the meandering river. Channels in the braided river were less deep as water was divided over multiple branches ( Figure 6, 117 h).
As in the braided river, channel adjustments and rapid migration of the bends resulted in sediment loss over the first 20 h (Figure 9b). The meandering river reworked a smaller area and the absolute sediment loss was also less (Figure 9a, b). Overall, the sediment loss was 4.9 mm in height over the reworked area compared to 4.5 mm in height for the braided river. One reason for this difference could be that the channels in the braided river were less deep and the floodplain shaving effect was smaller. Although the sediment loss was higher, the surface elevation did not show more degradation (Figure 9c). The Z 5 , related to the deepest part of the reworked area, did not became lower, i.e. deeper. At the end of the experiment, the surface elevation illustrated that the meandering river, as compared to the braided river, had deeper (Z 5 ) and shallower points (Z 95 ) in the reworked area, i.e. the meandering river had both a deeper channel and a higher floodplain (Figure 9d).

Floodplain deposition and styles
The floodplain was formed by deposition and erosion of sand and silt and by enrichment of silt during overbank flow. Two characteristic floodplain styles were observed by the deposition of cohesive silt. The first style was formed by deposition of silt on the outer bank (vertical accretion), such as crevasse splays,  Figure 10a).

Silt distribution
Silt deposited in different styles on the meandering river and was not uniformly distributed along the meandering river. Continued silt addition resulted in an increase in the surface area fraction containing silt and an increase in silt concentrations over time (Figure 10b). Low concentrations of silt settled along the main channel during the formation of the incipient meandering river (Figure 10c). Later, the highest silt concentrations were observed near the flume inlet as overbank flow occurred on a small area, while silt concentrations in the water were relatively high, causing deposition. In the middle section, a large bend developed and captured most of the silt. Continued deposition of silt increased the silt concentration in the middle section (Figure 10c, d). Deposition of the silt led to depletion of the silt concentration, so that the surface fraction area of silt and concentration decreased in the downstream direction. After the incipient meandering river was cut off, silt was removed by channel migration without depositing a new layer of silt in the downstream section.
Inner bend deposits Point bars developed as the channel migrated laterally and sediment deposited on the inner side of the bend forming scroll Silt was spread on various elevations of the point bar: in the low swales, in a subsidiary channel and on the higher scrolls. Most of the silt deposited between the detrended elevation of À2.9 and À9.9 mm and the highest concentrations were clustered at detrended elevations of À7.4, -5.4 and À3.9 mm (Figure 11b). The lowest elevation of silt on the surface (À7.4 mm) was located in the most downstream part of the swales and in the remnants of the chute (subsidiary) channel (Figure 11c, blue). The chute channel was abandoned due to a plug bar upstream, while downstream silt deposited without filling the channel remnants. Silt on the surface (À5.4 mm) deposited also on the upstream part of the swales and on the scrolls along the channel (Figure 11c, yellow). The highest elevation of silt on the surface (À3.9 mm) was located on the chute bar formed during the incipient meandering river and on the active scrolls (Figure 11c, red). The lowest areas were not entirely filled with silt, so that depressions such as chutes and swales remained visible.

Outer bank deposits
The sharper bends in the meandering river promoted overbank flow on the outer bank, which caused formation of crevasse splays and levees. Downstream of the sharpest part of the bend, a high-momentum flow advected on to and over the point bar in the curved channel, so that overbank flow occurred and diverted on to the pristine plain. The interaction between flow strength and bank strength resulted in crevasse splays or levees, respectively. Crevasse splays or levees did not form in the braided river as flow curvature was generally low, so that water level did not rise above the pristine plain (Figure 7c).
The crevasse channel with splay formed during the incipient meandering phase when the banks were not yet cohesive. Downstream of the crevasse channel, a sandy splay formed with silt deposits at the lee side ( Figure 12a). Later, the upstream bend migrated in a lateral and downstream direction and the flow direction in the crevasse became more perpendicular to the channel. The former crevasse channel was abandoned and filled with fines, while in the new crevasse channel silt deposited again at the lee side of the crevasse splay (Figure 12b). Eventually, a chute cutoff in the main channel shifted the flow direction, so that the crevasse was abandoned. The remnant of the crevasse splay was partly filled with silt by overbank flow occurring on the crevasse splay (Figure 12c).
In the meandering river, continuous overbank flow with silt on the pristine plain led to formation of a levee. The banks were stronger, so that sediment deposited on the outer bank instead of channel incision and the formation of a crevasse. The thickness and location of the levee depended on the occurrence and direction of the overbank flow. Overbank flow followed the initial slope and spread outwards from the sharpest point of the bend (Figure 13a). In planform view, the overbank deposit formed a splay shape that was interrupted due to the low ridges on the initial pristine plain (Figure 13a). On these ridges, the silt fraction decreased as water depth  decreased and apparently flow velocity increased (Figure 13b). Deposition of sand and silt raised the floodplain elevation. Near the channel, sand deposition was dominant compared to silt deposition. The deposited sand fraction raised the floodplain elevation by 2 mm (Figure 13b). Spreading of the overbank flow resulted in deposition of more silt and less sand in a downward direction. The concentration of silt deposits varied in a longitudinal and lateral direction. First, in the longitudinal direction the percentage of the area covered by silt increased, but further downstream the percentage of silt decreased as most silt was already deposited (Figure 13b). Second, the percentage of silt was non-uniformly distributed in a lateral direction (Figure 13c). High concentrations of silt raised the floodplain elevation by 0.5-1.5 mm (Figure13b, c).

Bank stabilization
In this section, the relation between bank erosion and floodplain style is explored. First, the floodplain area that stabilized the banks is described. Second, bank erosion is related to meander migration in several bends.

Bank stabilization by floodplain construction
In the meandering river, several floodplain styles formed and stabilized the banks. These floodplain styles were eroded by different processes, i.e. bank undercutting or sediment entrainment (Table II). The self-formed floodplain reduced bank erosion upstream, chute incision and headcut formation downstream of the point bar, so that the meander bend was not cut off (as in Van Dijk et al., 2012). At the meander bend in the middle section, a total area of 7.9 m 2 (see Figure 10), several cohesive floodplain styles developed and covered about 40% of the area. On the pristine plain (total area of 4.3 m 2 ) silt covered an area of 1.75 m 2 , which was 41% of the total pristine plain area and was mostly concentrated on one side of the channel (Table II). On the inner bend (total area of 3.6 m 2 ) 43% of the area was covered by silt. Here, silt deposited during lateral accretion and during overbank flow on the higher scrolls (1.0 m 2 ) and in the lower swales (0.5 m 2 ). The bank erosion test demonstrated that addition of silt in the floodplain reduced erosion rates.

Bank erosion in channel experiments
Distinct floodplain styles and bank erosion processes resulted in differences in channel bend displacement in our experimental setup. Here we describe the migration of four bends (for location of cross-profiles see Figure 6). Three bends formed in the meandering river, where silt concentration around the bends decreased in the downstream direction ( Figure 10). The first bend was characterized by silt deposition on the inner side of the bend and on the outer bank. The second bend had silt deposition dominantly on the inner side of the bend. At the third bend some silt deposition occurred on the outer and inner bend. Also, one bend was analyzed from the braided river, representing a bend without any effect of a cohesive floodplain.
In the upstream bend, deposition of silt on the outer bank resulted in the formation of a levee and an increase in bank strength. In the experiment, channel migration rate decreased despite the sharper bend curvature over time (Table III). Because deposition of silt continued, chute excavation decreased on the inner side of the bend (Figure 14a). The migration rate of the bend in the middle section was faster despite the gentle bend (Table III). Here, bank strength was not increased by silt deposits on the outer bank (Figure 14b). Consequently, there was less time for silt deposition on the inner side and the concentration was therefore lower. Cross-profile B shows that silt deposition occurred mainly in the lows of the point bar, which halted chute incision.
Reoccupation or excavation of channel depression remnants occurred more often where silt had hardly deposited. The downstream bend of the meandering river had some small chute cutoffs on the active point bar. The first chute cutoff occurred after 57 h due to a flow direction shift caused by a bar upstream of cross-profile C. Later, upstream meander growth caused a local avulsion of the main channel as remnants of the former channel were reoccupied. Outer bank resistance by silt deepened the channel (Figure 14c). The second chute cutoff occurred on the active point bar, which was caused by the upstream bend that translated downstream and shortened the flow path. Remnants of the crevasse splay were not reoccupied as silt deposits stabilized the floodplain surface, preventing incision of chute channels (left side of the cross-profile C). Due to chute cutoffs, the average migration rate of the bend was lower, despite low silt concentrations on the outer bank and a bend curvature comparable to that in cross-profile B (Table III).
Local avulsions and chute cutoffs were more common in the experiment without fines. The main channel was located for most of the time at the outer bank, but local avulsions and chute cutoffs limited the bend growth (Figure 14d). Flow direction shifted continuously reoccupied channel remnants as the floodplain consisted only of erodible non-cohesive sediments ( Figure 5). Here, the subsidiary channel was excavated by overbank flow on the point bar, but it did not become the main channel ( Figure 14d). The cohesive floodplains on the outer bank reduced outer bank erosion and thus lateral migration, while the cohesive floodplain in the inner bend decreased channel incision and headcut formation.

Channel pattern and scaled conditions
Our experimental results demonstrate the development of a braided and meandering river. Cohesive fines deposited in the floodplain stabilized banks and led to a sustained meandering river. In the braided river, one or two channels remained active Table II. The surface area with floodplain formation styles occurring in relation to floodplain removal by bank undercutting and chute incision in the middle section of the meandering river (see Figure 10a  most of the time (ABI above 1.7). As observed in other studies on braided rivers in the flume (e.g. Ashmore, 2001;Egozi and Ashmore, 2009), continuous cross-cutting of the channel on the floodplain resulted in extensive shifting of the channel and bars. Field studies (e.g. Reinfeld and Nanson, 1993) have described braided river evolution by lateral migration of a braid-train, but this kind of braiding did not develop in this experiment. Lateral channel migration is the dominant process in the braided and meandering river as the channel belt increases (Van de Lageweg et al., 2013). Later, cutoffs in the braided river decreased lateral channel migration. In this experiment a cohesive silt was added for bank stabilization. In earlier experimental work, bank stabilization was accomplished by manual addition of vegetation on the  (Tal and Paola, 2010). Braudrick et al. (2009) added low-density material behaving as fine sediment, which was captured by the vegetation and filled potential chutes, so that single-thread meandering was sustained. Other studies (Friedkin, 1945;Schumm and Khan, 1972;Smith, 1998) tested the effect of initial cohesive banks on meandering river development. Bank stabilization should decrease lateral migration, so that sedimentation on the inner bend increases to original floodplain level. In an earlier experiment (Van Dijk et al., 2012) a large chute cutoff reset the development of the meander bend after the upstream perturbation moved in reverse direction and restarted new bend development. Here we show that the addition of even more silt prevented cut-off even after reversed movement of the upstream boundary. Our results indicate sustained meandering forms when there is a self-formed cohesive floodplain without manual interference as required in the case of seeding vegetation.
Our results showed the difference in channel development between experiments in which the only condition that was varied was the availability of cohesive fine material in the sediment feed. In an earlier experiment with less silt and a constant discharge, floodplain formation was limited and chute cutoffs could occur (Figure 15a; Van Dijk et al., 2012). A downscaled experiment with constant discharge (0.3 L/s) shows that overbank flow was limited and bends became shorter ( Figure 15b). The lack of overbank flow supports the idea of the importance of a varying discharge for the formation of a cohesive floodplain (Figure 15c, d). Another effect of the varying discharge is that the wavelength of the bend increases due to the effectiveness of sediment transport during high discharges (Figure 15b versus Figure 15c). We observed that for the low discharge and shallow channels average sediment mobility decreases, so that when flow dispersed over the floodplain the morphological changes reduced.

Bank erosion and chute cutoffs
The bank erosion experiments show that the addition of slightly cohesive silt decreases bank erosion and increases bank stability. The erosion rate can be quantified using an excess shear stress formula in which bank erosion is related to flow shear stress, critical shear stress and an erodibility coefficient (e.g. Rinaldi and Darby, 2008;Darby et al., 2010). With the bank erosion experiments we tested the erodibility coefficient of the sediment mixtures. However, prediction of the critical shear stress for cohesive material is complex as for cohesive sediments the fluvial entrainment threshold increases (Zanke, 2003;Lick et al., 2004;Rinaldi and Darby, 2008). We suspect that an increase of silt in the sediment mixture at the bank toe causes hydraulically smooth conditions and increases the critical shear stress for sediment entrainment as the sand is protected against turbulence by a viscous sublayer (e.g. Zanke, 2003;Vollmer and Kleinhans, 2007), both of which effects cause a decrease in bank erosion.
Silt on the bank toe originating from the bank top reduced bank erosion in the experiment. Deposition of silt on the outer banks forms a layer of cohesive silt on top of a non-cohesive bank. The erodibility of the non-cohesive sand is higher, so that flow undercuts the cohesive silt and failure processes with bank toe protection determine bank erosion rates, as in natural rivers with cohesive banks (e.g. Simon and Collinson, 2002;Darby et al., 2007;Langendoen and Simon, 2008;Parker et al., 2011).
Floods enhanced the occurrence of chute cutoffs by bank incision and excavating floodplain depressions. Channel migration causes a local imbalance of more erosion compared to deposition and forms depressions (Lauer and Parker, 2008). Without bank stability, these depressions are captured by chute cutoffs causing the channel to braid. Chute cutoffs formed by upstream bank incision and downstream headcut formation in  = 114 h). c) Experiment with varying discharge and without silt. The bar wavelength increased compared to constant discharge and a larger area was reworked at t = 114 h. d) Experiment with varying discharge and with silt addition. A large bend developed, while a cohesive floodplain formed and stabilized the banks (t = 114 h). This figure is available in colour online at wileyonlinelibrary.com/journal/espl the point bar lows (Constantine et al., 2010;Zinger et al., 2011;Van Dijk et al., 2012). In the meandering river, the erosion processes were balanced by the cohesiveness of the fines reducing bank erosion and chute excavation. In the experiment without fines, the local imbalance between erosion and deposition was not compensated by stronger banks so that chute cutoffs cause the channel to braid. Silt depletion in the meandering experiment decreased cohesive floodplain formation and increased chute incisions and cutoffs in the downstream section.

Floodplain sedimentation
Overbank flow is required for floodplain formation (e.g. Lewin, 1978;Nanson and Croke, 1992;Zwolinski, 1992). Overbank sedimentation is enhanced by several factors, e.g. river slope, lateral channel movement, base level and the occurrence and magnitude of floods (Zwolinski, 1992). In our experimental setup we used a simple hydrograph, with a long period of low flow representing bankfull discharge and a short period of high flow representing a flood. During the high-flow period, sedimentation of the fine material on the banks decreased the local sediment imbalance between erosion and deposition (Lauer and Parker, 2008). Furthermore, the floodplain with silt became more cohesive. Overbank sedimentation of fines causes two styles of floodplain formation: in the outer bank forming levees and crevasse splays and in the inner bend in chutes and between and on scroll ridges and swales.
The floodplain in the experimental meandering river was similar to floodplains found in natural river systems. The deposition of fines on the outer bank forms a levee with coarse grains near the channel and finer, but also thinner, deposits further away from the channel (e.g. Brierley et al., 1997;Walling and He, 1997;Ferguson and Brierley, 1999). Natural levees form along the channel when water level exceeds bankfull levels. Here, the levee formed only at the outer banks downstream of the bend apex when overbank flow spread outwards and contains fine material. The formation of a crevasse on the outer bank formed when the flow strength was stronger than the bank strength, as described for natural systems (O'Brien and Wells, 1986;Bristow et al., 1999). Fines settled on the distal side of the crevasse splay as observed in the Brahmaputra and the Cumberland Marshes (e.g. Coleman, 1969;Pérez-Arlucea and Smith, 1999). Later, the crevasse splay was abandoned by the main flow and fine silt particles buried the crevasse channel, as formed in aggrading rivers with suspended fine material (e.g. the Colombia River; Makaske, 2001). An important process in the construction of the distal levee and crevasse splay was advective deposition of the suspended material (Cazanacli and Smith, 1998) rather than diffusive mechanisms (Törnqvist and Bridge, 2002). The advective process was more important in the experiment as the floodplain flow was not turbulent, while in natural systems the effect of turbulence results in a stronger decrease in thickness and grain size in the levee (Törnqvist and Bridge, 2002).
Silt deposition affected meander formation mostly in the middle section of the meandering river. In the upstream reach extensive silt deposition decreased bank erosion and the bend amplitude, which caused a lower bend amplitude, whereas downstream silt deposition was supply-limited, which allowed the development of chute cutoffs on the point bar. We hypothesize that depletion of silt deposition in the downstream section would become less when the initial bed of a sediment mixture consists of at most 10% silt, which would not affect bank stability as shown in the bank erosion experiments.

Relevance for natural rivers
The experimental results support earlier ideas that bank strength is a necessary condition and a key parameter for river meandering (Ferguson, 1987;Eaton, 2006;Kleinhans, 2010;Kleinhans and Van den Berg, 2011). Cohesive outer banks will reduce lateral migration and decrease channel dynamics. More important is that the self-formed floodplain prevented chute cutoffs. To form a cohesive floodplain discharge variation is important because it allows fines to settle on higher banks (Middelkoop and Asselman, 1998). In nature, levees and cohesive banks form when water level fluctuates (Brierley et al., 1997).
This experiment suggests that the initial plain does not need to be cohesive. When the initial bed is non-cohesive, alternate bar formation and bend migration occur rapidly. The supply of cohesive material entering from the hinterland, our upstream boundary, could stabilize the banks. For example, the addition of cohesive fines to the sediment load of rivers at the transition of a glacial to interglacial climate could stabilize banks, decrease cutoffs and could cause the transition between braiding and meandering, as observed in the Rhine-Meuse delta (e.g. Vandenberghe, 2003;Erkens et al., 2011). Meandering patterns as present in the middle and late Holocene may have formed rapidly in the early Holocene, but were fixated when the amount of cohesive material on the floodplain increased while gradient reduced due to base level rise. Rivers with less cohesive material in the sediment load are more dynamic and have more cutoffs, such as the River Allier in France and the River Rhine at the apex of the delta around the border of the Netherlands and Germany.
Several studies ascribe the transition between braided and meandering to the vegetation cover in the river plain, which is largely controlled by climate conditions (e.g. Millar, 2000;Gibling and Davies, 2012). Vegetation will add strength to the bank (Millar et al., 1993;Eaton, 2006;Braudrick et al., 2009;Tal and Paola, 2010) and hydraulic resistance causes more deposition of fines on the banks (Darby, 1999), so that the influence of the cohesive sediment on the banks can be important. These experimental results illustrate that bank strength by cohesive materials can be sufficient to sustain meander development even without the growth of vegetation, as also observed on Mars (Howard, 2009).

Conclusions
A braided and a meandering gravel-bed river was developed in our experimental flume study. For the first time, we conclusively linked the development of the different channel patterns to the formation of a cohesive floodplain and resulting bank stability. We conclude that the necessary conditions to form and sustain a meandering channel pattern are cohesive floodplain material and overbank flow in addition to a dynamic upstream perturbation. Results show the following: • Bank erosion rates decrease significantly for slightly cohesive floodplains, tested with systematic small-scale bank erosion rate tests. • The braiding experiment is characterized by alternate bar formation in the initial straight channel, lateral channel migration and chute cutoff occurrence. • Sustained lateral migration and cohesive floodplain formation result in a meandering river, without the occurrence of chute cutoffs. The cohesive floodplain stabilizes the banks, so that lateral bank migration and chute excavation decreases. Discharge variations are necessary to form a cohesive floodplain of overbank sedimentation on the higher outer banks and to fill potential chutes.
• The deposition of fines forms two styles of cohesive floodplains: first, a layer of cohesive material on top of a noncohesive bank in the outer bank, e.g. levees and crevasse splays; second, lateral accretion of different grain sizes in the inner bend. • The experiments show that a meandering river can develop without having an initial cohesive bank. This suggests that the formation of a cohesive floodplain can result in the transition from a braided to a meandering river.