Braided rivers are relatively simple to produce in the laboratory, whereas dynamic meandering rivers have not been sustained beyond initial bend formation. Meandering is theoretically explained by bend instability growing from planimetric perturbation, which convects downstream. In this study, we experimentally tested the importance of upstream perturbation and chute cutoff development in the evolution and dynamics of a meandering channel pattern. The initial straight channel had a transversely moving upstream inlet point and silt-sized silica flour was added to the sediment feed to allow floodplain formation. We obtained a dynamic meandering river with scroll bars. Bend growth was alternated by chute cutoffs that formed across the point bars. Meandering was maintained as one channel was disconnected by a plug bar. The curvature at the chute bifurcation transported sediment and build a new floodplain, while the other channel widens. At the end of the experiment, the fluvial plain exhibited a meandering channel, point bars, chutes and abandoned and partially filled channels with a slightly cohesive floodplain surface similar to natural meandering gravel bed rivers. We conclude that the necessary and sufficient conditions for dynamic meandering gravel bed river are a sustained dynamic upstream perturbation and floodplain formation.
 Many rivers on Earth, from the smallest to the largest, have a meandering planform. Meandering rivers are generally single-thread and are dynamic in the sense that their bends migrate through bank erosion. Erosion is balanced by floodplain formation on the other side of the channel, so that the channel migrates while approximately maintaining a constant width (Figure 1). Despite numerous field and model studies, the necessary and formative conditions for meandering remain unclear [Kleinhans, 2010]. Numerical models suggest that bend growth is balanced by neck cutoffs which leads to a statistical steady state of meander properties [Camporeale et al., 2005; Frascati and Lanzoni, 2010].
 Producing a dynamic meandering channel in laboratory experiments has proven to be curiously difficult. It is the question to what extent this is due to scaling problems and to what extent it reflects our limited understanding of the controls and process of meandering. Flume experiments that added small amounts of vegetation produced characteristics of meandering [Tal and Paola, 2010] or sustained a meandering river [Braudrick et al., 2009]. These experiments showed that braiding initiated by chute cutoffs [Friedkin, 1945] was prevented by strengthening the floodplain. The general aim of this study is to understand what the necessary and sufficient conditions are for the formation of a dynamic meandering gravel bed river and how a single-thread river reforms after chute cutoff.
 Theoretical understanding of the formation of meanders follows from linear stability analyses [Struiksma et al., 1985; Parker and Johannesson, 1989]. According to the linear stability analysis, a straight channel configuration is unstable both planimetrically, i.e., bend instability, and altimetrically, i.e., bar instability, for turbulent flow on a cohesionless sediment bed [Ikeda et al., 1981; Lanzoni and Seminara, 2006; Seminara, 2006]. Consequently, a perturbation of an initially straight channel alignment grows to form meandering patterns. The bar instability response of the channel can result in a planimetric perturbation due to bank erosion, which then leads to bend instability [Blondeaux and Seminara, 1985; Crosato et al., 2011]. The typical wavelengths of the bars and bends resonate at a specific width-depth ratio, which case non-migrating alternate bars form in the bends.
 The nature of bend instability determines meander planform development. Theoretically, instability could be absolute or convective (Figure 2). The convective instability requires a persistent perturbation located at some initial cross-section of the channel and affects the planform only in a single direction. The absolute instability develops even if perturbations are triggered only at some initial time and then cease. In addition, the absolute instability influences planform in both directions [Lanzoni and Seminara, 2006]. The direction of the convection is determined by the resonance condition: for relatively low width-depth ratios sub-resonant instability convects in the downstream direction, while for high width-depth ratios super-resonant instability convects in the upstream direction.Lanzoni and Seminara found that bend instability is mostly convective in nature, which for sub-resonant rivers imply that an upstream boundary must continuously be perturbed to maintain dynamic meandering. A sub-resonant instability is mostly found in meandering rivers, which are in an underdamped regime [Struiksma et al., 1985; Lanzoni and Seminara, 2006]. Underdamping leads to overdeepening of the outer-bend pool and associated enhancement of the bar in the inner bend just downstream of the entrance to the bend or other perturbations [Kleinhans and Van den Berg, 2011].
 In nature, chute cutoffs in meandering rivers can form in three different ways or combinations thereof. First, swales within the floodplain may capture overbank flow and incise until cutoff occurs [Fisk, 1947; Hickin and Nanson, 1975]. Second, as a result of localized bed aggradation along a meander, the conveyance capacity of the channel is reduced, and flood flow is forced overbank, resulting in headcut extension from the downstream rim in upstream direction across the point bar and chute cutoff [Gay et al., 1998; Zinger et al., 2011]. Thirdly, an embayment upstream on the point bar forms by localized bank erosion during floods and downstream extension of the embayment leads to a cutoff [McGowen and Garner, 1970; Constantine et al., 2010].
 Here we report on an experimental river representing a meandering gravel bed river that exhibits lateral and downstream migration, downstream convection and chute cutoffs. The objective of this paper is to assess the importance of sustained upstream perturbation on chute cutoff development in the evolution of meandering channel patterns. The inlet of water and sediment was moved continuously in a transversal direction to simulate a meander bend that migrates into the flume.
 This paper is set up as follows. We first give a detailed description of the layout and boundary conditions of the flume and experiments in the laboratory and the used measurement techniques. Then, we present detailed morphology in the form of maps and time series. Furthermore, we compare our experimental river to empirical, analytical and physical descriptions of natural meandering rivers, including bend migration and a physically based predictor of the transverse bed slope in bends. Finally, we elaborate on the set of processes that cause chute cutoff yet maintain a single-thread channel.
2. Experimental Setup, Methods and Materials
 The experiment was designed to represent a meandering gravel bed river which is 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 [Kleinhans et al., 2010]. We designed experimental conditions that compromise between the most important scaling issues; in particular, low sediment mobility, prevention of scour holes and cohesion of the floodplain sediment. To obtain meandering in a relatively short flume, we designed an inflow duct perpendicular to the upstream edge of the tank which moves transversely as a sustained perturbation. We compared the results obtained using the shifting inflow to those obtained for a control experiment without transverse inflow movement but with otherwise identical conditions. To assess whether the resulting experimental meandering channel was representative for natural meandering rivers, we quantitatively compared the most important properties and dynamics of meandering to empirical and analytical predictors. We studied the formation of chute cutoffs and the processes of channel closure.
2.1. Flume Setup and Experimental Procedure
 The experiment represented a meandering gravel bed river in underdamped regime and was accomplished by scale rules of some non-dimensional variables for hydraulic conditions, sediment transport conditions and morphological features, which had to remain between specific ranges (Table 1). The flow had to be subcritical (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 (Re > 2000). In contrast, braided rivers could be represented by laminar flow where floodplain interaction was not important [Métivier and Meunier, 2003; Malverti et al., 2008; Lajeunesse et al., 2010]. For sediment transport conditions bedload sediment should be mobile θ > θcr [Kleinhans and Van den Berg, 2011]. Because a laminar sublayer condition was known to be conducive to formation of ripples or scour holes [Kleinhans et al., 2010], the channel should 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 was represented by bar wavelength and interaction parameter (Table 1 ). This required that the channels had enough bank strength so they did not became too wide and shallow which would led to braiding.
Table 1. Boundary, Initial, and Designed Conditions
We used Keulegan with ks = D90.
Struiksma et al. [1985, equations 26 and 28] The scale rule for IP is to be within an underdamped regime (see text).
 The experimental setup was kept as simple as possible with constant influxes and a straight initial slope and channel. The experiments were carried out in the Eurotank at Utrecht University, a flume which is 6 m wide and 11 m long (Figure 3) [Postma et al., 2008]. The flume was filled with a 10 cm thick layer of poorly sorted sand (Figure 4) and the initial bed surface was set at a gradient of 5.5 · 10−3m/m based on pilots. We carved a 30 cm wide by 1.5 cm deep straight channel in the sediment, corresponding approximately to the predicted hydraulic geometry in a non-cohesive gravel bed river [Parker et al., 2007] and with self-formed channels in the pilot experiments. The downstream boundary was formed by a deep basin with constant water level. The base level was controlled by a fixed overflow level at downstream floodplain level, which led to sedimentation and formation of a fan delta. The constant input discharge of 1.0 l/s was controlled by a rotameter. The upstream inflow point was initially offset transversely at half a channel width as a static planimetric perturbation to start meandering.
 The sediment feed consisted of coarse (sand) and fine (silica flour) sediment, which were mixed in a ratio of 4:1. The sediment feed rate was 1.00 kg/hr of bedload sediment (sand) and 0.25 kg/hr of suspended sediment (silica flour). The sand had a specific weight of 2650 kg/m3, a median grain size D50 = 0.51 mm, and was poorly sorted (Figure 4). The silica flour was necessary for having sufficient suspended sediment transport, that could settle on the floodplain and point bar resulting in slightly more cohesive banks [Peakall et al., 2007] and in the filling of lows in the floodplain to limit the potential for chute cutoffs [Braudrick et al., 2009]. The silica flour had a specific weight of 2600 kg/m3, and a median grain size of D50 = 0.03 mm.
 The experiment ran for 260 hours under a constant formative (bankfull) discharge and transverse movements of the inflow point for water and sediment over time. First, from 0–30 hrs the inlet point was at a fixed position with an offset of half the channel width to study the evolution with a static upstream perturbation. Secondly, from 30–80 hrs the inflow point was transversely moved to the left with a constant rate of 1 cm/hr. Finally, from 100–260 hrs the inflow point was moved to the right with a constant rate of 1 cm/hr. The migration rate was determined from measurements of the average bend migration rate from earlier pilot experiments and about equal to a movement of W/30 per hour. Tests (not reported here) without migration resulted in a decrease in channel dynamics, while a faster movement of 5 and 10 cm/hr of the inlet point resulted in channel abandonment upstream, as lateral and longitudinal channel migration could not keep up with the migration speed of the inlet point. These results suggested that a minimum rate of perturbation is needed and that channel development is bounded to a maximum perturbation rate which depended on the erodibility of the banks.
2.2. Data Collection
 Several measurements were done during the experiment. Overhead photographs were taken at 10-min intervals to record channel migration by low-resolution images. Further, the flume was equipped with an automated gantry with a line-laser scanner (0.7 mm vertical resolution) and a high-resolution (0.25 mm/pixel) Digital Single Reflex camera. The dry bed topography was scanned on average every 7 hours. The high-resolution camera was remotely controlled and captured 55 images per full coverage of the flume at on average a 7 hour interval during wet and subsequent dry bed, so that both could be matched to the laser scan. The flow was shut off for scans and photographs of the dry bed.
2.3. Data Processing
 The collected data were processed to obtain digital elevation maps and silica flour distribution maps. The images from the low and high-resolution cameras were first rectified for lens distortion [Heikkila and Silven, 1997]. The high-resolution images from wet conditions were used to identify the channel, while high-resolution images from dry conditions (Figure 5a) produced a silica flour distribution map (Figure 5c). Silica flour was detected by changes in luminosity of the photographs, as silica flour was white compared to the bed material, but is rather inaccurate as it was affected by ambient lighting. The water was colored with a red dye, Rhodamine B, for identification of channels on the images from wet conditions.
 We used digital elevation models (DEMs) to identify channel pattern changes. The bed topography was measured by projecting the line-laser onto the bed normal to the mean downstream direction of the channels and photographing the line with a digital camera (0.7 mm resolution) mounted at an oblique angle with a 2 mm interval in longitudinal direction (Figure 5b). The point cloud with x, y and zcoordinates were gridded by a median filter on a 2, 4 or 10 mm grid depending on the size and details of the region of interest. These high-resolution DEMs as well as the time interval between scans were important for the quality of morphologically inferred process rates [Lane et al., 2003].
 The initial bed surface slope of the DEMs was subtracted to detrend the DEMs. For areas where no changes occurred between current DEM and original DEM, values were indicated with zero. DEMs of difference (DoD) were constructed by subtracting DEM pairs. The DoD showed that points known to be constant changed slightly in height over time. Differences in height of these stable points could be the result of interpolation of the point cloud on a grid and sub-millimeter movement of the gantry suspended from the ceiling. We therefore corrected the DoDs by setting no changes for values which were smaller than 0.7 mm which was a conservative estimate of the vertical scan resolution.
 The sediment balance was calculated by summation of the corrected DoDs for the fluvial part (first 10 m, equation (1)). The sediment balance volumes between timesteps t and t + 1 were calculated for all grid cells m based on the inversed Exner equation:
where V is volume (l), z is bed level (in dm), dxis grid size in x-direction (in dm) anddyis grid size in y-direction (in dm),i is grid cell index and m is total number of grid cells.
2.4. Data Reduction
 The experimental river was quantified by the total braiding index (TBI), the active braiding index (ABI) [Bertoldi et al., 2009], bend migration rate and transverse bed slope. The total braiding index, defined as the number of wetted channels per cross-section, was taken as the average number of channels (selected at six fixed cross-sections 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 active braiding index, defined as the number of channels that transport sediments in a cross-section, was the average number of channels which also had net morphological change (e.g., lateral migration) observed on the DoD maps at six identical cross-sections.
 Bend migration rate was determined by the displacement of the bend apex and related to bend curvature (R/W). Three phases (initiation, growth and termination) of bend development could be distinguished depending on non-dimensional bend curvature (R/W) and normalized migration rates (M/W), where R = radius of curvature (m), W = channel width (m) and M = migration rate (m/hr) [Hickin, 1974; Furbish, 1988; Hooke, 2003]. Therefore, the centerline of the channel was used to calculate the curvature and bend radius along the bend (equations (2)–(3)) [Fagherazzi et al., 2004; Crosato and Mosselman, 2009]. The apex of the bend was determined as those parts where the radius of curvature was smallest, measured as largest curvature. The curvature was calculated as [Fagherazzi et al., 2004]:
where s is the curvilinear streamwise coordinate (per mm), x- andy- coordinates of the grid and Λ is the curvature, which is positive (negative) when the bend is turning right (left) for increasing values ofs. Here we calculated the radius of curvature streamline (R) as:
At the bend apex, we measured the maximum bend migration rate [Hickin and Nanson, 1975, 1984; Hooke, 2003] automatically from DEMs and manually from overhead photographs. For the bend migration rate, the absolute distance between bend apex positions was calculated, so that the migration indicates lateral as well as longitudinal directions. The data was filtered for points which were related to a direction shift of the outer bank, for example due to chute cutoffs.
 We measured transverse bed slope in the bends for comparison to a transverse bed slope predictor, to test helical flow and the transverse bed slope effect in experimental bends for scale problems. Transverse bed slope was measured on the DEMs in profiles perpendicular to the channel centerline. Figure 5d illustrates several profiles along a meander bend and shows the channel width (W), maximum channel depth (hmax), mean channel depth (h), bank height (hb) and the transverse bed slope in a bend ( , where n is the transverse coordinate in a curvilinear channel). The transverse bed slope was determined between 2 cm from the deepest point until 2 cm from the waterline following Struiksma et al. . We used the theory of Struiksma et al.  adapted by Talmon et al.  to predict the transverse bed slope for damped conditions in an infinitely long bend:
where κ = 0.4 Von Karman's constant, g = acceleration of gravity (9.81 m/s2), and C = Chezy coefficient (here calculated as ). depends on and the helical flow strength [Struiksma et al., 1985], which could be calibrated for a better representation of the transverse bed slope of our experimental river. The water depth (h) was kept constant, while the bend radius (R) changed in this prediction. Data was filtered by excluding points that mismatched between the photographs during wet conditions and the DEMs.
 In our experiment a low sinuosity meandering river developed, where outer bend erosion was balanced by floodplain formation of scroll ridges and swales. Bend growth was alternated by chute cutoffs, and former channels were closed by plug bars (Figure 6). The plug bar is defined as a bar that hinders flow into a channel [Fisk, 1947; Toonen et al., 2012]. Below we report in detail on channel pattern evolution and chute cutoff mechanisms. We quantitatively compared the important properties and dynamics of a meandering river to empirical and analytical predictors, i.e., sinuosity, braiding index, bank erosion related to bend curvature and transverse bed slope.
3.1. Channel Pattern Evolution
 The experiment showed different stages of channel development which could be related to the upstream boundary conditions. The experiment was therefore subdivided into three stages: 1. (0–38 hrs) incipient meandering of a low-sinuosity channel with alternating bars developed without movement of the inflow point; 2. (38–100 hrs) meander growth which resulted in a higher-sinuosity meandering channel and the development of a mature point bar with successions of scroll bars that was later cut off as the inflow point moved to the left side; and 3. (100–260 hrs) increasing floodplain complexity as a major chute cutoff occurred and a single-thread meandering channel redeveloped as the inflow point moved to the right side (Figure 7 and Animations S1 and S2 in the auxiliary material).
3.1.1. Incipient Meandering
 Initially, a weakly sinuous quasi-meandering river with narrow alternating point bars developed (Figures 7 and 8a, 0–30 hrs). The initial offset of the inflow caused bank erosion and overbank flow. Bank erosion and formation of forced alternating bars produced an incipient meandering river with three bars and bends, a sinuosity of 1.12 (Figure 8a) and a bar wavelength corresponding to the predicted wavelength of 3.0 m (Table 1). After formation of these three bends, the initial perturbation migrated downstream and straightened the channel (Figures 8a and 8b, 14–30 hrs). The bar and bend stretched out to a length longer than the predicted bar wavelength and two bends remained in the domain of the flume (Figure 7, 30 hrs).
 Channel adjustment to the base level and formation of bars and bends led to sediment loss in the first 30 hrs (Figure 8c). Sediment which deposited downstream of the fluvial plain produced a fan delta, which after 30 hrs was already half the volume of the total fan delta volume measured at the end of the experiment (260 hrs). At the end of this first stage, the channel straightened and lateral migration ceased. Channel mobility was insignificant, i.e., neither lateral channel movement nor the development of a new floodplain was observed (Figure 9a).
3.1.2. Meander Growth
 In the second stage, movement of the inflow point in the transverse direction resulted in formation and growth of bends and increased sinuosity (Figure 7, 30–58 hrs). Development of bars and bends were initiated when the inlet point moved transversely over a distance of 2/3 of the channel width. Then, in the middle section of the flume, a bar and bend formed that later promoted the development of another bend downstream (64 hrs). Lateral migration of the channel by bank erosion and formation of bars was clearly visible in the DoD (Figure 9b). The transversal movement of the inlet resulted in bend expansion and the bend amplitude tripled (Figure 7, 58 hrs). Furthermore, the channel width changed and bend wavelength became shorter, so that three bends instead of two bends formed. Erosion and sedimentation resulted in width-depth variation of the channel along the experimental river (Table 2).
Range indicates variation along the entire channel including straight and bended parts.
3 · 103
4 · 103
 In this stage, net sediment loss occurred due to channel widening, channel extension by increase in sinuosity, and channel bank shaving while producing lower floodplains than the pristine floodplain (Figure 8c), but not by net channel deepening (W. I. van de Lageweg et al., Channel belt architecture formed by an experimental meandering river, submitted to Sedimentology, 2012). While the floodplain evolved by continuous migration of the meandering river, sediment loss was reduced to zero and the fan delta hardly grew. The channel sinuosity increased to a maximum of 1.3 (Figure 8a), until the middle bend was cut off. After the cutoff, the total braiding index increased (Figure 8b), because water flowed in the residual channel. The active braiding index remained about unity (Figure 8b), because the residual channel had no morphological change and the system remained single-threaded.
3.1.3. Floodplain Complexity
 In the third stage (Figure 7, 128–246 hrs), the inflow point was moved twice as far from the flume center as in the former opposite sweep. Movement of the inlet point in the reverse direction produced a multiple bend cutoff, so that the channel straightened (Figure 8a). In the straight channel new meander bends formed at opposite side of the earlier bends. The formation of bends resulted in an increase in sinuosity and sediment loss again (Figures 8a and 8c). Again a bend cutoff occurred, so that the residual bends were approximately mirrored after 191 hrs (Figure 6).
 As the experiment progressed, floodplain complexity increased. Former channels were transformed into floodplain lakes that only received a minor amount of silica flour by sheet flow over the plug bar. Depressions of the residual channels could easily be re-activated by channel migration at a later time to transform into chute channels. As a result bend translation onto depressions, i.e., former channels, resulted in twice the amount of cutoffs later in the third stage (between 191–260 hrs,Figure 8). Further, in the third stage the main channel did not produce a constant maximum sinuosity of 1.3, but the sinuosity varied between values of 1.1 and 1.2 due to repetitive cutoffs and redevelopment of bends (Figure 8a). Floodplain complexity was quantified by the total braiding index, which increased as disconnected residual channels remained unfilled with sediment. The active braiding index, on the other hand, ranged between 1 and 1.5 (Figure 8b), demonstrating that the channel remained largely single-threaded (seeFigure 10 and Animation S2).
3.2. Transverse Bed Slopes in Bends
 We examined if the transverse bed slope of the experimental channel bends behaved according to well verified theory (equation (4)). The transverse bed slope increased with decreasing bend radii (R/h). Transverse bed slope in the experiment was often steeper than the predicted transverse bed slope because of overdeepening at the beginning of the bend (Figure 11). This was expected as the channel was in an underdamped regime (Table 2). The transverse bed slope at overdeepening locations were generally three times larger than the transverse bed slope according to the predictors (Figure 11, blue line). We found no relation between discrepancies of measured and predicted transverse bed slope and bend migration rate or sediment sorting in the bend.
3.3. Bend Migration Styles
 Styles of bend migration are here described in the framework of the normalized bend migration rate (channel width/ hour; W/hr) and non-dimensional bend curvature (R/W) (Figure 12). To test the detailed variation in time, the position of some outer banks were also manually traced on the overhead photographs, which had a shorter time interval than the scans (Figure 12a). Furthermore, the normalized migration rate versus channel curvature was compared to migration rates of the upstream boundary (Figure 12b).
 Eroding banks in the experiment did not show any slump failure; erosion occurred gradually by sediment entrainment. Individual bend migration was characterized by an initiation, a growth and a termination phase (Figure 12a). Initially, bend migration was slow for high bend curvatures. In the pristine floodplain, the migration rate peaked at a R/W-ratio of 12 (Figure 12a). The bend migration rate in the reworked floodplain peaked at higher R/W-ratio of 18 and the maximum migration rate increased from 0.5 W/hr to 1.5 W/hr (Figure 12a). The peak occurred when the bend migrated through the depression of the former channel, so that less sediment had to be removed. The total volume of sediment that was removed for the formation of the bend in the pristine floodplain was twice the volume compared to bend formation with the same amplitude in the reworked floodplain, 21.5 l (in 28 hrs) versus 11.2 l (in 14 hrs) of sediment from initiation until cutoff. Similarity, in the development of both bends was that bend migration was terminated at the same bend curvature. For a R/W between 6–8, no lateral migration was observed and bends were abandoned (e.g., chute cutoffs).
 The lateral migration of the first bend was slower than the second bend (Figure 12b). The migration rate of the second bend exceeded the migration rate of the transversally moving upstream inlet point for about 50% of the time, while migration rate of the first bend only exceeded migration rates of the transversal moving inlet point for 25% of the time (Figure 12b). Apparently, the migration rate of the first bend was more controlled by the transversal movement of the inlet point while the second bend developed freely.
 To assess if the bend migration rates were mostly forced by the transversal moving inlet point as a standing or advecting wave, we compared lateral positions and migration rates of several points along the river with the position and migration rate of the inlet (Figures 13a and 13b). This demonstrated that the rate and direction of the bends differed non-systematically from that of the transverse movement of the inflow point in three ways. First, bend amplitude increased in the downstream direction (Figure 13a). Secondly, the maximum bend amplitude exceeded the maximum amplitude of the inlet point (Figure 13a). Thirdly, the migration rate increased downstream and initiation of channel migration was retarded. This indicated that the perturbation had to convect downstream and that the downstream bends were initiated but not forced by the transversal moving inlet point.
3.4. Chute Cutoffs
 In the experiment, channel shifting and new bend formation were generally caused by single chute cutoffs and once by multiple chute cutoffs. The occurrence of chute cutoffs prevented the development of high sinuosity meanders while the rapid closure of the channel abandoned by chute cutoff prevented development of a braided river.
3.4.1. Single Bend Cutoffs
 Single bend cutoffs occurred as a bend migrated downstream and occupied depressions, i.e., topographic lows, in the self-formed floodplain. Depressions formed on the inner side of the bend as a side-bar did not fully attach to the floodplain (Figure 14a). We define a side bar as a bar attached to one side of the channel, but not as part of a train of alternate bars. Furthermore, depressions formed when the residual channel was not filled with sediment (Figure 6). The self-formed floodplain was lower as the bend migrated and shaved the floodplain by eroding high banks and forming lower point bars (Figures 14a and 14b). Below we describe the chute cutoff development in five phases starting from incipient meandering.
 In the first phase (Figure 14a, 44–50 hrs), lateral migration induced by upstream perturbation instantly initiated an ‘alternate’ side bar. The upstream part of the side bar attached to the earlier formed floodplain. The downstream part of the side bar did not attach to the floodplain, so that a large swale depression formed. In a later stage, the depressions were only filled by suspended material that was transported by overbank flow over the bar into the swale. Due to a limited amount of suspended material and the absence of floods, depressions were not entirely filled by sediment.
 In the second phase (Figure 14b, 50–78 hrs), rapid meander growth proceeded as the perturbation by the upstreamflow orientation continued. Bank erosion and lateral movement of the bend produced a higher amplitude bend in the middle section of the flume. We visually observed that sediment eroded from the outer bank was transported and at least partly deposited on the inner bank downstream. As lateral successive scrolls were attached to the inner bank, a point bar formed which was characterized by scroll ridges and swales. On the upstream side of the point bar, accretion by sheet deposition of bedload material disconnected the swale depressions from the active channel.
 In the third phase (Figure 14c, 78–108 hrs), overbank flow became more important as water level fluctuated, even with constant discharge. Overbank flow occurred in response to in-channel sedimentation and channel geometry changed, which caused flow diversion at the outer bank. As the outer bank migrated laterally and longitudinally, outer bank erosion and sediment input in the channel continued. Downstream sediment transport decreased as sinuosity increased and lowered the local stream gradient. The decrease in stream gradient caused a decrease in flow velocity, so that water level rose, starting overbank flow at the outer bank. Diverted flow led to further reduction of sediment transport, and more in-channel deposition. Changes in the channel width-depth ratio and the occurrence of overbank flow continued for a long period.
 In the fourth phase (Figure 14d, 108–114 hrs), channel shallowing caused a further increase in overbank flow. In-channel sedimentation and bar accretion resulted in disconnection between the channel upstream and downstream of the point bar. Overbank flow became more focused as it rapidly incised a chute channel in the outer bank and formed a chute splay downstream of the chute channel. Overbank flow converged from the floodplain into the large swale depressions downstream and produced headcuts that propagated in upstream direction (Figure 14d). Silica flour deposition in the depression of the point bar and on the outer bank did not form large cohesive drapes (Figure 15), so that bank strength and the critical shear stress for sediment entrainment probably did not increase much.
 In the last phase (Figure 14e, 122 hrs), the chute channel rapidly incised the chute splay and connected with a headcut. Following the chute cutoff, the channel widened and a new side bar formed. Then the channel started to migrate laterally and formed a new bend just downstream of the former bend (Figure 7, 128 hrs). Former channels were observed in the DEM as a depression, but were disconnected from the main flow by a plug bar as shown in the overhead photographs (Figure 7 and Animation S2).
3.4.2. Multiple Bend Cutoff
 Multiple bend cutoffs developed in the third stage of channel development when the inflow point moved in the reverse direction. In the beginning of the reverse movement, a fan built at the upstream boundary and closed off the upstream part of the channel (Figure 7, 128 hrs). As the fan built out in the downstream direction, a chute channel incised in the fan from the inlet point position to the main channel at the first point bar (Figure 7, 150 hrs). Within 8 hours, the second bend was cut off due to the upstream change in flow orientation and an impulse of sediment from the upstream chute channel led to in-channel sedimentation. This processes continued until all three bends were cut off and the channel straightened (after 158 hrs). Meanwhile, floodplain formation by plug bars disconnected former active channels, so that the system remained single-threaded.
 Straightening of the active channel by several cutoffs was associated to the onset of braiding, but as the inflow point was continuously perturbed, alternate bars and meander bends were initiated in the opposite direction. In pilot experiments without transverse movement of the inflow point, multiple bend cutoff also occurred which resulted in a straight channel [van Dijk et al., 2010]. Without the continuous upstream disturbance, no subsequent lateral migration occurred after the multiple bend cutoff.
 Our experimental results show that the combination of a dynamic inflow point, poorly sorted sand and silt promotes the development of a dynamic meandering river exhibiting a richer morphology than hitherto produced in a flume, including cycles of meander growth and chute cutoff.
 Linear stability analysis suggests that meandering can form in a straight channel when non-migrating alternate bars grow and lead to bend instability [Blondeaux and Seminara, 1985]. However, the bend amplitude and the number of bends in the experiment were limited (Figure 9a). This may be due to the limited flume length. Although we observed that bend amplitude increased in downstream direction (Figure 13a).
 Bend migration in the experiment had similar characteristics as that in natural rivers. Bend growth could be characterized by channel extension, expansion and translation [Hooke, 1984]. Bend evolution followed phases of initiation, growth and termination as like in nature [Hickin, 1974; Furbish, 1988; Hooke, 2003]. The bend curvature related to the phases of bend evolution differs in our experiment compared to nature. The minimum radius of curvature in the experiment was around 4, and bend growth terminated between values of 4–8. In natural fluvial systems, termination occurred in bends ranging at bend curvature between 1–6, but with a maximum frequency in bend curvature of 1–2 [Hooke, 1984; Furbish, 1988]. Termination of bend growth was coupled to the occurrence of chute cutoffs. After reformation of bends the sinuosity did not reach the same high value of 1.3 before the first chute cutoff, which was also observed in numerical simulations with cutoffs [Camporeale et al., 2005; Frascati and Lanzoni, 2010].
4.2. Cutoff Processes
 Several shortcuts over the point bar developed and caused single bend or multiple bend cutoffs in the experiment. Multiple bend cutoffs in meandering rivers have been reported by several studies [e.g., Stölum, 1996; Hooke, 2004; Camporeale et al., 2005; Kleinhans and Van den Berg, 2011]. These cutoffs were caused by the propagation of a change in flow orientation and triggered by floods, and temporarily reduced sinuosity. In the experiment, a chain of events promoted the multiple bend cutoff. The reverse direction of the inlet point caused a shift in flow direction and triggered a chute cutoff upstream that was followed by multiple cutoffs.
 Chute cutoff development in the experiment was a result of overbank flow, which either occurred when bends became sharper and sinuosity increased or when sedimentation occurred in the channel (Figure 14). Eventually, a chute cutoff occurred when the unconfined overbank flow was captured by the topographic depressions in the floodplain, i.e., former channels, forming upstream propagating headcuts, while upstream bank erosion formed a chute channel. These processes were also observed in natural meandering rivers, e.g., bank incision [McGowen and Garner, 1970; Constantine et al., 2010], headcut propagation on the point bar [Gay et al., 1998; Zinger et al., 2011] or located specifically in the lower swales [Fisk, 1947; Hickin and Nanson, 1975].
 We summarize with a conceptual model (Figure 16) how the experiment elucidated how chute cutoffs develop in a dynamic meandering river with bend growth, floodplain formation and overbank flow. In general, a sinuous channel will rapidly develop until a critical bend amplitude (Figure 16a), which is also observed in numerical simulations [e.g., Howard, 1984; Sun et al., 1996]. Then, bend translation and expansion increases the bend curvature and alters the channel dimensions. A high-momentum flow advects onto and over the point bar in the curved channel [Dietrich and Smith, 1983]. At the sharper bend overbank flow occurs and flow is diverted, similar to bifurcations [Miori et al., 2006]. Water flows overbank and incises in the outer bank (flow strength > bank strength [Ferguson, 1987; Constantine et al., 2010]) and deposits a chute bar, whereas sediment is transported through the inner bend, i.e., the active channel. Downstream of the point bar overbank flow converged in depressions forming headcuts. Later, the actual chute cutoff is triggered by for example a flood (reported by Ghinassi  and Zinger et al. ) or when a bar blocks the flow in the active channel (Figure 16c). A single-thread channel is maintained as the curvature upstream [Kleinhans et al., 2008] leads to closing one of the two channels with a sediment plug. We argue that in braided rivers, even with one dominate channel [Egozi and Ashmore, 2009], channels are straight, i.e., less curvature, hence former active channels are not as much disconnected with a plug bar.
 Disconnection of a former channel by a plug bar has also been found in natural meandering gravel bed rivers, for example in the river Rhine in the past millennia before human interference (Figure 17) [Toonen et al., 2012]. In the Rhine, the plug bar consists of bedload sediment (sand-sized sediment), while the downstream residual channel is filled with finer material during low flow conditions [e.g.,McGowen and Garner, 1970; Toonen et al., 2012]. Fine material hardly filled the residual channels in the experiment, so that when the floodplain became more complex residual channels were easily reoccupied. Despite the addition of slightly cohesive silica flour to this experiment, chute cutoffs formed. In future experiments, we will add more silica flour to create a more cohesive floodplain and reduce the numbers of chute cutoffs. Further, we will vary the discharge, so that the overbank and floodplain lows can receive more silica flour.
4.3. Scaling of Flow and Sediment Transport
 The overall channel pattern of the experiment strongly resembles a meandering pattern dominated by chutes and scrolls, which is representative of a low-sinuosity meandering gravel bed river [Leopold and Wolman, 1957; Kleinhans and Van den Berg, 2011]. Scroll bars with typical ridges and swales formed at a constant discharge. The predicted interaction parameter indicated an underdamped river, meaning that in the outer bends overdeepening occurred [Struiksma et al., 1985; Johannesson and Parker, 1989]. The transverse bed slope observed in our experiment corresponds with a transverse bed slope predictor (equation (4)) [Struiksma et al., 1985; Talmon et al., 1995]. Upstream of the bend apex, the transverse bed slope was three times steeper as expected in underdamped regime with overdeepening. Such agreement between theory and the experiment strongly suggests that there are no severe scale problems.
 Our experimental meandering river was designed to minimize scaling issues. Indeed, flow was mostly subcritical, sand was dominantly transported as bedload while silica flour was transported in suspension, and scour holes were rare and usually the result of flow confluencing. As with natural gravel bed rivers [Lewin, 1976], we observed that changes in channel width-depth ratio resulted in localized critical flow conditions. However, we observed that overbank flow was laminar, so that the critical shear stress for sediment entrainment on the floodplain will be higher [Zanke, 2003], which may have reduced the number of chute cutoffs relative to a prototype situation with turbulent flow over the floodplain. On the other hand, the cohesionless sand allowed rapid bank erosion [Kleinhans et al., 2010], such that channel bend growth resulted in channel extension, floodplain shaving [Lauer and Parker, 2008] and net sediment loss. Despite the weak banks and the scale problem of laminar flow on the floodplains, the experimental river remained meandering.
4.4. Scaling of Meander Migration
 Comparing the meander evolution of the experimental river to that of real rivers requires a timescale. This is not trivial because that timescale is partly determined by the inherited strength of the banks and floodplain. For the purpose of scaling experimental time, we compare the time required to build a channel belt from bend reworking of the floodplain on both sides of the river. The number of bends that our experimental meandering built in 260 hr with a single cycle of transverse movement (Figure 10) can correspond to approximately a timescale of a century for the meandering gravel bed river Allier in France, where lateral mobility of the channel is very high (Figure 1) [Kleinhans and Van den Berg, 2011]. For the river Rhine at the Dutch-German border our experimental river represents a millennium of bends reworking the floodplain (Figure 17) [Kleinhans and Van den Berg, 2011; Kleinhans et al., 2011]. Both rivers have low sinuosity migrating meanders with scroll bars and chute cutoffs similar to our experiments. Meandering rivers with lower lateral channel mobility and fewer chutes, such as sinuous rivers in tropical forests where vegetation grows rapidly on the point bar [Kleinhans and Van den Berg, 2011], are perhaps better represented by meandering experiments with dense vegetation [Braudrick et al., 2009] or very high bank cohesion [Smith, 1998]. Our experimental meandering river with modest bank strength is highly dynamic, suggesting that meanders in nature without strong bank protection (e.g., vegetated banks) can also form very fast.
 This experiment shows the influence of a continuous planimetric instability, which can be in natural meandering gravel bed rivers meander growth, confluences of rivers, or channel width variations [Luchi et al., 2010] on the dynamics of a meandering river. Our findings may have implications for the downstream impacts of bank stabilization and other structures (e.g., bridges and groins) on the meander development, i.e., amplitude and migration rate. For example, directly downstream of the bridge and the fixed banks the amplitude and lateral migration of the bends decreases in the river Allier (Figure 1). Further, our results suggest that restored streams, where morphodynamics are required for rehabilitation of riparian processes [Geerling et al., 2006], could be made more dynamic by perturbing the upstream reach. For example, the downstream trapped sediment can be supplied at one side of the channel as a transverse perturbation.
 The dynamic meandering river which formed in our experiment represents a meandering gravel bed river with scroll and chute bars. Typical characteristics of this river were high lateral channel mobility, alternate bar initiation, meander growth, overdeepening in the bend and chute cutoffs. Channel shifting, reactivation of former channels and meander growth resulted in a mature and similar-to-natural floodplain complexity. Bend migration showed different phases of development, comprising initiation, growth and termination by cutoff. The time to form a channel belt by bend migration indicates that the timescale for our experimental meandering river corresponds approximately to a century for the river Allier in France and a millennium for the river Rhine at the Dutch-German border.
 We conclude that a dynamic upstream perturbation is needed to sustain an instability and which maintains meandering. An initial perturbation upstream results in initiation of bars and bends; the perturbation subsequently migrates downstream and results in a low-sinuosity quasi-meandering channel. This experiment demonstrates that sustained upstream inflow perturbation causes dynamic meandering even if banks are hardly strengthened by cohesive sediment and vegetation is absent. In nature, the inflow perturbation corresponds to perturbations caused by for example bend curvature.
 Chute cutoffs occur when overbank flow excavate depressions by incision and headcut propagation on the self-formed floodplain, i.e., point bar. Cutoff processes limit bend growth, but meandering is sustained. The sediment balance between bend dynamics, chute cutoffs and floodplain formation is important. The curvature at the chute bifurcation between the former active channel and the chute channel leads to such a division of water and sediment that the former channel is closed with a plug bar.
 This project is supported by the Netherlands Organisation for Scientific Research (NWO)(grant ALW-Vidi-864 08 007 to MGK) and funding by ExxonMobil Upstream Research (grant EM01734 to MGK and George Postma). We are grateful to the technicians of the physical geography lab (Henk Markies, Marcel van Maarsseveen and Chris Roosendaal) and of the Eurotank lab (Thony van der Gon and Henk van der Meer), for their technical input and support during the experiments. Comments on an earlier draft and discussion by Filip Schuurman, Hans Middelkoop and John Lewin greatly helped to improve the paper. We thank Kim Cohen for producingFigure 17. We acknowledge Christian Braudrick and two anonymous reviewers for comments that significantly improved the manuscript, and editor Alex Densmore and associate editor John Buffington for their helpful guidance. The authors contributed in the following proportions to conception and design, data collection, analysis and conclusions, and manuscript preparation: W.M.vD. (25, 40, 60, 80%), W.I.vdL. (25, 40, 20, 5%) and M.G.K. (50, 20, 20, 15%). Online supporting information contains two animations of the experiment.