Corresponding author: R. N. Szupiany, Faculty of Engineering and Water Sciences, Littoral National University, Paraje El Pozo, RN 168, km 472, Santa Fe, CP 3000, Argentina. (email@example.com)
 Channel bifurcations associated with bars and islands are important nodes in braided rivers and may control flow partitioning and thus affect downstream confluences, as well as the formation and dynamics of bars. However, the morphodynamic processes associated with bar formation are poorly understood, and previous studies have largely concerned laboratory experiments, small natural streams, or numerical analyses with large Froude numbers, high slopes, and low Shields stresses. In these cases, the morphologic changes at bifurcations are relatively rapid, with predominant bed load transport and the suspended load playing a minor role. In this paper, the evolution of the flow structure and suspended bed sediment transport along four expansion-diffluence units in the Rio Paraná, Argentina, are described. The Rio Paraná is a large multichannel river with a bed composed of medium and fine sands and possesses low Froude numbers and high suspended bed material transport. Primary and secondary flow velocity components were measured with an acoustic Doppler current profiler (ADCP) along the expansion-diffluence units, and the backscatter signal of the ADCP was calibrated to allow simultaneous measurements of suspended bed sediment concentrations. The interactions between these variables show that the cores of primary flow velocity and suspended bed sediment concentration do not necessarily follow the thalweg at the bifurcation and that inertial effects on the suspended bed sediment may influence the morphodynamics of bar formation. It is suggested that changes in flow stage, as well as the presence of vegetation, may further increase the deposition of suspended bed sediment at the bar head. This study suggests that the ratio of suspended bed material to bed load is an important factor controlling the morphodynamics of bifurcations in large sand bed braided rivers.
where τ is the bed shear stress (N m−2), is the sediment density (kg m−3), ρ is the water density (kg m−3), g is acceleration due to gravity (m s−2), and is the median bed grain size (m). These experiments with low Shields numbers ( ranging between 0.04 and 0.29) effectively replicate conditions in higher-gradient, coarser-grained, gravel bed rivers, where morphological changes are relatively rapid, bed load predominates and the suspended bed sediment load is small or absent [Bolla Pittaluga et al., 2003; Federici and Paola, 2003; Bertoldi and Tubino, 2005; Kleinhans et al., 2008; Zolezzi et al., 2006]. Using a one-dimensional numerical analysis,Wang  developed predictive relationships to determine the division of sediment within the bifurcation as a direct function of the discharge ratio and bifurcation angle. These relationships governing sediment partitioning were later extended by Bolla Pittaluga et al. , Federici and Paola , and Bertoldi and Tubino  to include more complex nodal configurations. In their study, Bolla Pittaluga et al.  concluded that when bed load is the dominant mode of sediment transport, both the upstream channel slope and the bifurcation itself tend to control flow dynamics and sedimentation at the bifurcation. However, when suspended sediment dominates under high Shields numbers, Bolla Pittaluga et al.  postulated the existence of a close relationship between the partitioning of flow and the suspended sediment dynamics. Federici and Paola  and Bolla Pittaluga et al.  also studied the process of bar formation and its impacts on stability of the bifurcation. Their findings show that when the Shields number in the upstream channel ( ) is larger than 0.15, and the upstream channel aspect ratio is low, a stable central bar tends to form downstream. Conversely, for low values of (<0.15), and with increasing values of channel aspect ratio, unstable bars tend to form, prompting successive bifurcation closures and openings. Tarekul Islam et al. applied a similar analysis to results from laboratory experiments while exploring the applicability of a general nodal point relation to describe the distribution of sediment at channel bifurcations. Their study shows that the routing of sediment is strongly dependent on the bifurcation angle, yet is independent of variations in flow discharge. Despite the usefulness and insightfulness of these past studies, the majority of nodal point relationships that seek to predict bifurcation stability have a strong experimental, one-dimensional basis. In addition, most of these past studies have been derived for gravel bed channels with low Shields numbers and predominant bed load sediment transport. The validity of these experiments is thus restricted, particularly when considering the dynamics of large sand bed rivers [Miori et al., 2006].
 Large multithread sand bed rivers typically have very low Froude numbers (Fr≪ 1), gentle slopes, and suspended bed material to bed load ratios greater than unity. Such channels also show complex planform geometries, within-channel morphologies, and large bed forms. Knowledge of the morphodynamics of large multithread sand bed channels is still lacking, mainly due to the lack of appropriate technology and suitable measurement techniques. As a result, our ability to understand and predict their dynamics and evolution is also limited. Studies such as those ofAshworth et al. , Ferguson et al. , and Dargahi suggest that secondary flows upstream of a braid bar play an important role in the morphodynamic behavior of channel bifurcations. These flows are the result of a pair of surface-convergent secondary cells in which the flow to the right and left of the bifurcation (looking downstream) rotate in a counterclockwise and clockwise direction respectively. This secondary flow pattern is responsible for sediment sorting at the bar head, with finer sediments being driven toward the channel margins and coarser sediment being deposited at the bar head.Richardson et al. , Richardson , Richardson and Thorne , and McLelland et al. found no evidence for the existence of such secondary flow patterns around large braid bars in the Jamuna River, Bangladesh. On the contrary, they observed a pair of surface-divergent secondary cells just upstream of the bifurcation, and near-bed secondary flow convergence at the zone of flow division. Evidence of flow upwelling at the interface between these cells was also found at the bifurcation, with near-bed, sediment-laden fluid being drawn upward. The consequent decrease in velocity gradients resulted in the reduction of the bed shear stresses, and therefore in the deposition of sediment along the line of flow division. Outside of the cells, a zone of downwelling was identified, where relatively sediment-free, near-surface, water was advected toward the channel bed, increasing the near-bed shear stresses and promoting erosion [Richardson et. al., 1996; Richardson, 1997]. Measurements of three-dimensional flow structure in large rivers byMcLelland et al. , Parsons et al. , and Szupiany et al. have shown that the magnitude of secondary flows may be damped, or even absent, in large, low-sinuosity channels with high width to depth (W:D) ratios, and where the form roughness generated by bed forms is significant.
 Although Ferguson et al.  identified the importance of differences between the flow and bed sediment vectors at bifurcations under low Shields numbers, the relationship between flow division and the trajectory of the suspended sediment has only recently attracted attention [Tarekul Islam et al., 2006]. Tarekul Islam et al. highlight the need to consider inertial effects in the distribution of sediment at bifurcations, especially when the suspended bed sediment load is predominant over the bed load transport. However, the relative influence of inertial effects over other variables such as channel geometry, bifurcation angle, and discharge partitioning, is still unknown, especially when considering a three-dimensional approach.
 In order to address the need for well-constrained field data on flow and sediment partitioning at bifurcations, the present paper presents results from four large expansion-bifurcation units in the Rio Paraná, Argentina. The Rio Paraná is a large river with characteristics that set it apart from other past studies conducted in higher Froude number channels. Notably, the river banks and bed are composed mostly of medium and fine sands; it possesses Froude numbers of 0.1, slopes of O(10−5), and a suspended bed material to bed load ratio of approximately eight. By analyzing the relationships between the flow structures, trajectory of suspended sediment, channel geometry and bed morphology, the present study aims to investigate their possible influence on braid bar formation and thus channel evolution.
2. Study Sites, Field Procedures, and Methodology
2.1. Study Sites
 The Rio Paraná is the sixth largest river in the world by mean annual discharge [Schumm and Winkley, 1994], with a drainage basin of 2.3 × 106 km2 that includes parts of Brazil, Bolivia, Paraguay and Argentina. Downstream of the major confluence with the Rio Paraguay (Figure 1), the mean annual discharge of the Rio Paraná is 19,500 m3 s−1, and the water surface slope is O(10−5). The channel bed is composed largely of fine and medium sand [Drago and Amsler, 1998] and the planform pattern is multithread with a meandering thalweg. The Rio Paraná is also characterized by a series of bifurcations and confluences around large midchannel bars, which results in a succession of nodal sections of varying width. Mean channel widths and depths range between 600 to 3000 m and from 5 to 16 m, respectively.
 The study areas are located near the cities of Paso de la Patria (PP, upper reach), Paraná (PA, middle reach), and San Martín (SM) and Rosario (R) in the lower reach (Figure 1). These sites correspond to large, asymmetrical, bar bifurcations. The Paso de la Patria, Paraná, San Martín and Rosario sites were surveyed on April 2005, March 2009 and June 2006 (SM and R), respectively. At the time of the surveys, the total flow discharges were 11,560 m3 s−1 at PP, 12,126 m3 s−1 at PA, 16,740 m3 s−1 at SM, and 14,670 m3 s−1 at R. Table 1 shows additional geometric and hydraulic parameters. Note that the discharge values presented above only reflect the main channel discharge, which is 14,100 m3 s−1 at PP upstream of the river confluence (mean value, period 1990–2011), 15,200 m3 s−1 at PA (period 1970–2011), and 17,900 at SM and R (period 1971–1997 [Giacosa et al., 2000]). These measured flow stages at PP, PA, SM, and R were, respectively, 18%, 20%, 7%, and 18% smaller than the mean values for these locations.
Table 1. Geometric and Hydraulic Parameters of Bifurcations PP, PA, SM, and R Upstream of the Midchannel Barsa
 In the middle reach, the bed material is composed almost completely of sand (>90%), with small amounts of silt and clay (<4% according to Drago and Amsler ). Drago and Amsler  also report between 28% and 63% of very coarse and medium sand (mean grain size of 412 μm) in the upper reach between Yaciretá Dam and Corrientes City, whereas in the middle and lower reaches these percentages decrease markedly to 11% and 51% for mean grain sizes of 320 μm and 233 μm, respectively. Table 2 shows the mean grain size distributions of the bed material at the study sites. Alarcón et al.  demonstrated, through measurements and careful calibration of transport formulae, that the average suspended sand to bed load ratio is approximately 10 in the middle reach. This ratio reduces to approximately six in the upper reach when the same formulae are applied and mean values for sediment size and hydraulic variables are used. The reduction in this ratio is associated with the presence of coarser bed sediment in the riverbed in the upper reach.
Table 2. Bed Sediment: Mean Size and Grain Size Distribution at PP, PA, and SM/R
Percentage of Sand
2.2. Bathymetric Surveys, Three-Dimensional Flow Mapping, and Measurement of Bed Shear Stresses
 At three of the bar bifurcations (PA, SM and R), the river bed morphology was surveyed using a Raytheon single-beam echo sounder (SBES), coupled to a differential global positioning system (DGPS) deployed on a small vessel. A real-time kinematic differential global positioning system (RTK DGPS) provided horizontal positions with an accuracy of ±0.02 m at approximately 1 Hz. Depth measurements were obtained along cross sections every 100 m. In order to create bathymetric maps of the three bar bifurcation sites, the resulting spatially distributed point cloud information in thex, y, zcoordinates was interpolated using standard kriging methods onto a regular grid with a grid spacing of 50 m. At site PP, a RESON 8125 multibeam echo sounder (MBES) was used to map the bathymetry of the braid bar head. The MBES employed the same RTK DGPS technology as the SBES surveys for spatial and temporal information, and a TSS MAHRS gyrocompass was used to record the three-dimensional motion and heading of the vessel. Soundings from the MBES survey were processed with CARIS-HIPS-SIPS software to merge and filter the point cloud information. A base surface grid was obtained with a horizontal resolution of one meter, and a digital elevation model (DEM) was constructed with an accuracy of approximately ±0.06 m. In order to assess the geometry and spatial variation of bed forms through the bifurcations, a series of longitudinal profiles were extracted from the SBES DEMs (Figure 2). In PA, longitudinal profiles A-A′, B-B′ and C-C′ (Figure 2) were obtained following surface flow lines previously determined with floats deployed in the field. ADCP measurements were also made along profile A-A′. Three longitudinal transects (E-E′, D-D′ and F-F′) were also extracted from the PP MBES data (Figure 2). Furthermore, in order to account for the longitudinal variation in bed form geometry along the profiles at PA, these measurements were divided into three segments: (i) upstream, (ii) middle, and (iii) downstream. In this way, the different hydraulic conditions responsible for changes in bed form geometry could be identified.
 Once a bathymetric survey had been conducted at each study site, the three-dimensional flow velocity field was measured using a 1000 kHz Sontek ADCP at PA, SM, and R, and a 600 kHz Teledyne RDI ADCP at PP. Flow measurements were made at 5, 3, 4 and 4 predetermined cross sections at PP, PA, and SM and R, respectively, with the goal of characterizing the flow field structure along the bifurcations (Figures 1 and 2). Since the ADCP was deployed on a moving vessel, it was linked to the RTK DGPS to determine the position and velocity of the boat. The bottom track mode of the ADCP was not used during the surveys so as to avoid measurement errors introduced by the presence of bed load transport [e.g., Kostaschuk et al., 2005; Rennie et al., 2007]. In order to obtain accurate cross-sectional data, the boat velocity (∼1.5 m s−1) and its trajectory were constantly monitored by the helmsman during the survey (see Szupiany et al. [2007a] for details). The Sontek ADCP was set to a cell size (CS) and an averaging interval (AI), or ensemble, of 0.75 m and 10 s, respectively. The Teledyne RD Instruments ADCP employed Water Mode 1 [Teledyne RD Instruments, 2006] with a bin size of 0.25 m and a sampling frequency of one ensemble every 0.64 s. Primary and secondary flow structures around the studied bars were analyzed according to the methodologies discussed by Szupiany et al. [2007a, 2009]and D. R. Parsons et al. (Velocity mapping toolbox (VMT): A new post-processing suite for acoustic Doppler current profiler data, submitted toEarth Surface Processes and Landforms, 2012). This methodology, based on a spatial-averaging procedure, is sufficiently accurate to identify large coherent flow structures with intensities higher than the instrument noise and the effects of small-scale turbulence. Although more sophisticated methodologies can be applied in order to filter the influence of small-scale turbulence and noise from the velocity magnitude in boat-mounted ADCP's [Hoitink et al., 2009; Sassi et al., 2011; Tsubaki et al., 2012], these were not required or applied herein. This reasoning is due to the fact that the magnitude of the eventual secondary currents resulting from applying such a filter would be as small, and would not affect substantially the results related to the distribution of suspended sediment. For example, Szupiany et al.  recorded vertical velocities as small as 0.01 to 0.02 m s−1 in the Rio Paraná, and intensities of secondary currents of this order or lower will have a minimal influence on suspended sediment.
 Average velocity profiles obtained from the ADCP data were used to estimate the total bed shear velocity, , and bed shear stresses, τ, at sites PA and SM using the law of the wall, as suggested by Kostaschuk et al. . Szupiany et al. also presented more details concerning the applicability of this method in the Rio Paraná. It is important to note that the velocity profiles and depth measurements obtained using the echo sounder were averaged over a flow column 15 m in width (∼1% of the total cross-section width) and approximately 20 m in downstream length. A more detailed account of this averaging technique is given bySzupiany et al. [2007a]. Using this procedure, bottom irregularities are thus averaged out, and a mean bottom level is obtained, which serves as a reference elevation for the averaged point velocities. Uncertainties derived from an incorrect elevation of the point velocities, commonly encountered when applying the law of the wall in natural rough beds, are thus reasonably reduced [Sime et al., 2007]. This relatively simple approach allows for both a qualitative and quantitative analysis of the distribution of bed shear stresses through the bifurcation zones of a large river. It also allows an analysis of the interrelationships between the hydraulic and sedimentological variables, something that is very difficult to achieve using traditional techniques.
2.3. Suspended Bed Sediment Mapping
 Based on the analysis of Szupiany et al. [2007b, 2009], a good correlation can be obtained between the corrected backscatter signal (Sv) of the Sontek ADCP and the directly measured suspended sand concentrations (CS [see Szupiany et al., 2009, Figure 2]). Although results indicate that wash load concentrations are usually higher than suspended sand concentrations, wash load particle sizes are relatively small, and therefore their influence on the ADCP backscatter signal is negligible compared to that of suspended sand particles. The correlation between suspended sand concentration and the corrected backscatter intensity yields a coefficient of determination (R2) of 0.83, with a relationship of [Szupiany et al., 2007b, 2009]:
 Since the backscatter signal is a function of both the suspended sediment mass concentration and the grain size distribution, it is thus site and flow stage dependent. In order to ensure that equation (2) could be applied to all sites and flow stages in the present study (PA, SM and R), eight suspended bed sediment samples (four from PA and four from R and SM) were analyzed with a scanning electron microscope (SEM) to characterize and examine the homogeneity of size distribution across samples. At present, no such calibration exists for the Teledyne RDI ADCP for the PP region of the Rio Paraná. In this case, the uncalibrated backscatter intensity was used to qualitatively assess the distribution of the suspended bed sediment. For the data acquired with the Sontek ADCP, the product of the derived point concentrations and the corresponding flow velocities was used to determine the suspended sand transport across each cell, CS, and thus the suspended bed sediment transport through the entire cross section, Gss. The suspended sand transport through portions of the cross sections not measured by the ADCP (i.e., the top blanking distance (0.8 m) and approximately 10% of the cells closest to the bed), were estimated using simple linear interpolations to the boundaries.
3.1. Morphologic Features
 The bathymetric maps shown in Figure 2reveal complex morphologic characteristics and distinct thalweg behaviors at each of the four sites. Bifurcations SM and PA show considerable thalweg curvature in the expansion-bifurcation area, whereas the shift in thalweg position at R and PP is more gradual and uniformly divided. As a result, the bifurcate channels tend to have similar depths. In addition, in all bifurcations it is possible to observe a general decrease in flow depth along the thalweg as it approaches the bifurcation. Once the flow is divided, channel depth increases downstream, especially in the bifurcate channel that carries the largest flow discharge (similar findings were observed byZolezzi et al. ). Note that in all bifurcations, the longitudinal and transverse bed slopes are small, with maximum values of 0.012 (0.68°) and 0.0099 (0.57°), respectively. The small slopes and high width:depth ratios produce homogenous flow conditions that explain the logarithmic behavior of the vertical profiles of downstream velocity. As a result, accurate values of bed shear stress can be obtained using the full flow depth in the law of the wall equation [Szupiany et al., 2007a].
 In all four cases, significant differences were observed between the thalweg depths of the bifurcate channels downstream of the bifurcation. This difference is defined as the “inlet step” of a bifurcation [Zolezzi et al., 2006]. Observed values of 10 m, 8 m, 6 and 4 m at SM, PA, R, and PP, respectively, help explain the flow discharge division among the bifurcate channels, as well as other topographic forcing effects [Thomas et al., 2011; Miori et al., 2012]. Note that this depth difference, or inlet step, can be directly related to the observed flow ratios (QR) of 3.7, 2.8, 2, and 1.21 (at SM, PA, R, and PP, respectively) between the bifurcate channels (Table 1).
 A submerged bar was present on the right margin of SM, and an island was observed downstream of PA (Figure 2) where the thalweg changes direction. In both cases, a second shift in thalweg position was also identified. At SM, this second thalweg shift occurs close to the bar head, resulting in higher flow depths and erosion (as also observed during field work) on the left margin of the central bar downstream of the bifurcation. This prevents deposition upstream of the bar. In PA, the second thalweg shift occurs mainly within the left bifurcate channel (left margin), thus increasing the flow depths in this area.
 In terms of channel evolution, historical records from the past 70 years indicate that bifurcations SM, R and PA have each experienced important changes in thalweg position and therefore in flow division. Indeed, past thalweg locations and discharge values were very different from the present observations. During the period from 1941 to 1970, the thalwegs at SM and R were located along the left bifurcate channel [Parody and Estruco, 1975]. Facultad de Ingeniería y Ciencias Hídricas (FICH)  reported, through analysis of historical bathymetric maps from 1970 to the present that the thalweg has gradually shifted to the right channel, where it still remains. Ramonell et al.  investigated the temporal evolution of the thalweg at PA and observed that this suddenly shifted during the course of the 1998 flood, the fourth largest flood on record. Additionally, Ramonell et al.  indicated that the shift in 1998 was preceded by instances of thalweg migration that began around three decades previously. In spite of important dredging activities in the Rio Paraná (for example, see the right margin at the SM bifurcation, Figure 2), FICH  demonstrated that these activities have had only a local effect in the redistribution of flow, and that they do not affect the general morphodynamic behavior of the river.
Figure 3shows the morphologic variations observed at PA over the last decade (from 2000 to 2010), together with flow stage variations for the same period. A period of low- to medium-flow stage was present between 2000 and 2006, with an average flow stage of 2.82 m at the Puerto Paraná gauging station (note that the average flow stage for the period from 1971 to 2011 was 3.29 m, i.e., 0.47 m higher than the measured flow stage). After 2006, two ordinary floods were identified. The first flood occurred between 2006 and 2007 (12 months in duration considering the rising and falling limbs of the hydrograph), with the river reaching a maximum height of 5.46 m. The second flood took place between 2009 and 2011 (16 months in duration) with a maximum stage of 5.78 m.Figure 3 shows a significant increase in the curvature of the thalweg in the left bifurcate channel between 2000 and 2009, as well as an increase in depth along the right margin of the bifurcation. Figure 3 also reveals deposition on both sides of the island, as suggested by the increase in area between the 2 m and −2 m isobaths.
 The significant increase in flow depths at PA caused by the 2009–2010 flood resulted in considerable morphological changes. An example lies in the significant erosion observed in Figure 3around the island and upper portion of the right channel (as shown by the evolution of the −4 and 2 m isobaths). As a result of a hydrodynamic 2-D numerical model (SisBahia®) of the flow at bifurcation PA for high-flow stage,Fornari et al.  (Figure 4) showed that flow lines followed a straighter path in the expansion-bifurcation area, which suggests the presence of a small topographic forcing effect. This resulted in a decrease inQR, which may explain the observed erosion around the island. Flow measurements conducted before and after the flood peak (Figure 3) show a decrease in QR,(the discharge ratio between the left and right channels was 2.8 and 2.1 during 2009 and 2010, respectively). This indicates that, unlike low- and medium-flow stages, high-flow stages play an important role in the morphology and hydrodynamics of bifurcations. The magnitude of the peak discharge and duration of the 2006–2007 flood, however, were not large enough to produce significant changes in the general morphology of the bifurcation.
Table 3reports the average dimensions of dunes measured along the longitudinal profiles surveyed at PA and PP, and the observed changes in bed form geometry can be explained by changes in flow depth and flow velocity along each profile. Along profiles A-A′ (center) and B-B′ (right channel) at PA (Table 3 and Figure 5), dunes are larger in zone i (upstream of the bifurcation) and become progressively smaller in the downstream direction. In A-A′, this decrease in size is somewhat sudden through zone ii, with no dunes being identified in zone iii, 500 m upstream of the bar head. Along profile C-C′, however, dunes increase in size through zone iii, in correspondence with an increase in channel depth and flow velocities. Unlike PA, the maximum dune dimensions at PP occur upstream of the bifurcation along profile D-D′, with a decrease in dune size toward the downstream channels (Table 3 and Figure 5). Due to the limited spatial extent of the MBES survey (Figure 2), the minimum channel depth along profile D-D′ was 4 m, approximately 600 m upstream of the bar head. This makes direct comparison with profile A-A′ at PA difficult. However, inspection of aerial photographs and satellite images at PA, R and PP for similar flow stages indicates the existence of a wide range of dune sizes (seeFigures 5c, 5d, and 5e). At PP, bed forms of significant size were identified in proximity to the upper portion and margins of the bar head, whereas at PA and R these features were missing. Dunes are only observed at PA and R on the bar surface.
Table 3. Average Dune Dimensions Derived From Single and Multibeam Surveys Along Longitudinal Profilesa
Averages of height and wavelength were computed from individual dunes present in each subsection.
3.2. Flow Structure
 A clear flow deceleration was recorded in all four bifurcations, between cross sections 1 and 3 at PA and PP, and between cross sections 1 and 4 at SM and R. This flow deceleration shows maximum velocities decreasing from 1.4 to 1.2 m s−1 at SM and R, from 1.6 to 1.2 m s−1 at PA, and from 1.5 to 1.3 m s−1 at PP (see Figure 6). Another distinctive feature was the observed nonuniform distribution of the cross-sectional velocities upstream of the bifurcations (i.e., the cores of maximum velocity were located on the right margin at site R and PA, and toward the center at PP and SM). This behavior, which is typical of large rivers with large width:depth ratios, could have an important impact on the morphological stability of these channels [Bolla Pittaluga et al., 2003; Federici and Paola, 2003]. At R, the core of maximum velocities is located near the channel center, yet further downstream it shifts to the right bank following the thalweg. A similar behavior occurs at sites SM and PA. In these cases, the cores of maximum velocity, located at the channel center at SM1 and near the right bank in PA1, shift to the left/center of the channel following the thalweg curvature. In PP, the thalweg does not shift and the flow goes through a more gradual and uniform division (Table 1 shows that the discharge ratio at PP is close to unity). The flow is divided at PP3 and PP4, with a greater discharge entering the right channel (Table 1). At SM2, SM3 and PA3, the core of maximum velocity does not follow the strong curvature and maximum depth of the thalweg, which suggests the presence of inertial effects acting on the flow. This phenomenon is accentuated during high-flow stages [Fornari et al., 2012, Figure 4], and can result in significant morphological changes. Similar to the results found at PA, Zolezzi et al.  report a decrease in the discharge ratio (i.e., a more uniform division of discharges between the two bifurcate channels) as a result of an increase in the discharge upstream of the bifurcation. It is important to note that the increase in flow discharge during the 2009–2010 flood in the smaller bifurcate channel (left channel at PA) produced significant morphological changes in the riverbed (see section 3.1). The resulting topographic effect could help explain the decrease in flow discharge ratio observed after the flood.
 Clear evidence of flow division upstream of the bifurcation was not found at SM, R, and PA. At PP, flow division starts at PP3 near the bar head. Miori et al.  suggest that bed roughness influences the distance at which this division takes place, but that it does not affect the flow discharge ratio.
 At SM and PA (sections SM3, SM4 and PA3), consistent secondary flow structures can be identified (counterrotating cells C1, C2, and C3, respectively; Figure 6) in the areas of maximum thalweg curvature (Figures 2 and 6). However, secondary flow intensities are small, c. 0.05 m s−1, which is less than 5% of the primary flow velocity magnitude. These values are approximately 10% lower than those measured by Thomas et al.  in an experimental bifurcation with a much lower width: depth ratio (˜3). These findings suggest that secondary flow structures may be more influenced by topographic forcing (thalweg curvature) than the flow structure created by flow division and curvature of the bifurcation [Miori et al., 2012]. Note that in PP, the absence of thalweg shifting leads to the lack of any coherent flow structures.
3.3. Sediment Transport, Bed Shear Stresses, and Shields Parameter
 The grain size distributions of the suspended bed sediment samples obtained at PA, SM, and R are presented in Figure 7. Note that similar distributions were found at PA, R, and SM at different flow stages of 12,126 m3 s−1, 14,670 m3 s−1, and 16,740 m3 s−1, respectively. Mean particle diameters ranged from 84 μm to 129 μm. The observed homogeneity in grain size distributions of the suspended bed sediment for these three flow stages ensures the applicability of the calibration procedure reported by Szupiany et al. [2007b, 2009] for the middle reach of the Rio Paraná.
 The total mean transport of suspended sand, as estimated from the ADCP calibration (Table 4), is similar across all bifurcations (standard error lower than 45%). This calibration also follows closely the suspended sand rating curve for the Rio Paraná derived by Alarcón et al.  for similar flow conditions (433 kg s−1, 301 kg s−1 and 179 kg s−1 for bifurcations SM, R, and PA, respectively). Table 4 also reports small deviations in the values of suspended bed sediment transport (Gss) among individual cross sections. Deviations from the mean are smaller than ±14%, which provides confidence in the use of the ADCP to measure Gss and its size distribution in large rivers. In order to obtain the suspended bed sediment discharge ratio, the cross sections closest to the bifurcations (i.e., PA3, SM4, and R4) were divided into two segments corresponding to each bifurcate channel (with an origin marked by the bar head location). The suspended bed sediment discharge ratio for each bifurcation (QS) is presented in Table 4.
Table 4. Suspended Bed Sediment Transport Along the Studied Bifurcationsa
Here L and R denote the left and right branches, respectively.
Suspended bed sediment discharge ratio QS (GssL/GssR)b
Figure 6shows the spatial-temporal distributions of suspended sand concentrations for each cross section at bifurcations R, SM, PA and PP. Well-defined cores of maximum sediment concentration can be identified for each cross section, with the patterns and distributions of these cores through each of the bifurcations showing some distinctive characteristics. These cores are located at the channel center at R and SM (cross sections R1, R2, R3, SM1, and SM2) and near the right bank at PA (cross sections PA1 and PA2), and occupy nearly 30% of the channel width. In most cross sections, the cores of high suspended sand concentration correlate directly with the highest flow velocities. However, it is notable that in the expansion zones just upstream of the bifurcations, there is a distinct deviation from this trend, with the maximum concentrations not following the path of the cores of maximum flow velocities. This pattern is present in the region of the right bank (at R) and within the thalwegs (at SM and PA) where the maximum sediment concentrations remain along the channel center down to the very beginning of the bar, while the flow velocity maximum deviates more rapidly (in SM, R, and PA; seeFigure 6). If the cores of maximum velocity are mapped with respect to the thalweg, at R, SM and PA, the deviation angles have values of 17°, 23°, and 29°, respectively.
 In the case of PP, the correlation between flow and suspended bed sediment is even more evident than for the other bifurcations. At cross section PP2, the division and curvature of the thalweg is accompanied by a shift in the core of maximum sediment concentration, notably in the right bifurcate channel where the discharge is larger. Although the effects of sediment inertia can also be proposed here, i.e., higher sediment concentrations on the sides of the island head as opposed to lower concentrations on the left margin that are associated with higher flow velocities, the correlations between sediment concentration and flow are not as apparent as at R, SM and PA. The deviation angle at PP is approximately 17°.
 The irregular behavior shown by the cross-sectional distribution of flow velocity, suspended sand transport, and the bed shear stress is shown inFigure 8 for sites SM and PA. At the onset of the expansion (cross sections SM1 and PA1) these variables are in phase, but the correlation is lost farther downstream, especially when considering sediment transport. Due to fluctuations in the velocities measured by the ADCP, the values of bed shear stress possess some dispersion, but the general patterns across the cross sections are clear. Note that in shallow water zones, the ADCP point velocity data were insufficient (less than three points in the vertical) to apply the law of the wall method. As a result, gaps appear in section SM4 between 1900 and 2200 m, in PA1 between 0 and 500 m, and in PA3 between 1800 and 2200 (Figure 8). Figure 9shows the correlation coefficients obtained for the cross-sectional distribution of flow velocity, sediment transport and Shields stress at SM and PA. It is observed that better fits are obtained for the most upstream cross sections SM1 and PA1, but that the fit becomes poorer in the downstream direction. This loss of correlation due to the presence of inertial sediment effects may explain the differences between discharge and sediment transport ratios between the two bifurcate channels (Table 4).
 Mean cross-sectional Shields stress parameters possess values ranging from 0.13, 0.09, 0.08, and 0.06 at SM1, SM2, SM3, and SM4, respectively, and values of 0.11, 0.09, and 0.07 at PA1, PA2, and PA3, respectively. These values are lower than the threshold of 0.15 proposed byFederici and Paola . Bolla Pittaluga et al. , and Zolezzi et al.  also propose a threshold of 0.1 for , but they argue that this value can be higher in cases where the width:depth ratio is large.
Figure 10shows the distribution of suspended bed sediment load, flow velocity, and shear stress along longitudinal profile A-A′ at PA. The decrease in flow velocity and bed shear stress downstream of, and near, the bar head reaffirms the contention regarding inertial effects on the suspended bed sediment. In other words, the observed suspended bed sediment is not the result of a resuspension mechanism, as both flow velocity and bed shear stress decrease in the downstream direction approaching the bar head. In fact, bed sediment samples obtained at the center of each bifurcate channel and upstream of the bar head (Figure 11, samples M1, M2, and M3, respectively; see Figure 2 for locations) show a finer grain size distribution for M3, similar to the grain size distribution of the suspended bed sediment (Figure 7). This suggests that at the time of the survey, the finer material in suspension was being deposited on the bar head.
Figures 11 and 12show the grain size distribution and location of sediment samples taken from a 0.5 m deep pit at the bar head of PA. Visual inspection of the pit revealed the presence of ripples and dune-scale cross stratification. Both types of stratification showed comparable grain size distributions, with a mean grain size (d50) and grain size distribution also similar to that of the suspended bed sediment in the main channel (Figure 11). Little overlap was found, however, between the grain size distribution at this bar head location and those located in the thalweg (M1 and M2; Figure 11). The stratification observed in the bar head pit (Figure 12) shows: (1) bottom layer: climbing ripples, (2) middle layer: dunes with their stoss side preserved, and (3) top layer: draped laminae. These preserved sedimentary structures indicate that the suspended bed sediment deposition rate was relatively high in relation to the bed form migration rate [Allen, 1971a, 1971b; Ashley et al., 1982; Baas et al., 2000]. These findings reaffirm, from a sedimentological perspective, the important role played by the suspended bed sediment in this region of deposition at the bar head.
4. Discussion and Conclusions
 This study provides a first analysis on the interactions between bed morphology, flow structure, suspended bed sediment transport, and depositional processes at the bar head of bifurcations in a large sand bed river (width ∼1500 m and width: depth ratios of ∼150 in average).
 As previously shown by Parsons et al.  and Szupiany et al. , the results in this study further illustrate that expansion-bifurcation units in rivers like the Rio Paraná cannot be studied without first evaluating the bed morphology. In the Rio Paraná, it was observed that in reaches with a higher width:depth ratio, thalweg meandering is more prominent [Ramonell et al., 2002]. This behavior generates complex bed morphologies that depart from the typical morphologies observed in smaller channels or in laboratory studies [e.g., Zolezzi et al., 2006]. As a result, precise values of bifurcation angles and asymmetries, which have been proposed as key variables governing flow and sediment partitioning [Wang, 1995; Tarekul Islam et al., 2006], can be difficult to define. The presence of an inlet step, defined as the difference in thalweg depths between the bifurcate channels, was found to influence the discharge ratio of the bifurcation. This is because the topographic effect causes an increase in the water surface slope at the onset of the bifurcation, resulting in a higher water surface elevation in the smaller channel, and in more flow being diverted through the larger channel [Thomas et al., 2011; Miori et al., 2012]. A similar behavior was observed by Zolezzi et al.  and Edmonds and Slingerland  in small braided channels and in deltaic distributary networks during flow stages close to the “state of formative condition”. During flood stages, stream power increases, topographic effects are less significant, since the flow is more uniformly distributed, and the discharge ratio changes according the planform geometry of the bifurcation. As a consequence, the relatively greater discharge captured by one of the bifurcate channels has a larger impact upon its morphology. This, in turn, affects the hydrodynamic conditions of the bifurcation after the flood has passed, which is when topographic effects become predominant. It follows from this analysis that floods play a significant role in the dynamics and stability of bifurcations, especially when dealing with large fluvial channels in which morphological changes occur over longer time scales (of the order of decades for sites SM, R, and P).
 In relation to the flow structure, the present results differ from the findings of Ashmore  and Ashworth et al. obtained for channels with smaller width:depth ratios (i.e., surface-convergent secondary cells upstream the bar head with downward vertical velocities between these cells) andRichardson et al.  and Richardson (i.e., surface-divergent secondary cells with upward vertical velocities upstream of the bar head). The present results show the presence of only one cell measured in the cross sections nearest the bar head at PA and SM, which is where the flow curvature is close to its maximum. In spite of this, secondary velocities were found to be rather small. Although it could be argued that these thalweg-driven flow structures contribute to the process of thalweg meandering, in rivers with high width:depth ratios they are unlikely to generate a significant superelevation of the water surface. In bifurcations where no thalweg diversion was observed, the radius of flow curvature at the bifurcation tended to be high, which resulted in coherent flow structures that were weak or absent. These factors, coupled with the effect of form resistance [Parsons et al., 2007], could explain the weak secondary flows observed. Vertical gradients in horizontal velocity, however, can have the opposite effect, as they make centrifugal forces more important in this region, and, as a consequence, promote the formation of coherent secondary flow cells. These cells are directed toward the channel banks at the surface, and toward the central bar at the bed. Due to their weak intensity, however, their net effect on the redistribution of sediment toward the island is likely to be minimal, thus exerting a limited influence on the bar morphodynamics. Note that no redistribution of the suspended bed sediment transport produced by this secondary cell was observed (Figure 6). Additionally, the horizontal secondary velocities induce a vertical flow, but with an intensity that is smaller than the horizontal flow velocity, and which lies below the resolution of the ADCP. Consequently, the effect of this flow on the resuspension/deposition of sediment toward the bed or the surface can also be considered negligible. More measurements of bed load transport are needed, however, to determine whether these small secondary flows are strong enough, and long enough in duration, to transport sediment toward the bar and banks. During high-flow stages, the observed increase in flow uniformity, the change in the flow discharge ratio [Fornari et al., 2012], and the increase in turbulence due to the presence of bed forms, are factors that could contribute to the decrease in flow curvature at the bifurcation. As a result, secondary flow structures are not expected to form or become stronger downstream. Additional investigations are required, however, to further corroborate these findings.
 The lack of correlation between the rapid shift of the core of maximum velocity (and unit discharge) and the core of maximum suspended bed sediment concentration suggests that inertial effects on the suspended sand particles are important, and these inertial effects are the reason why sand particles tend to maintain their paths along the channel centerline toward the bar heads (Figure 6). This lack of correlation between flow and suspended bed sediment concentration is also responsible for significant changes in the flow and sediment discharge along the bifurcate channels (Tables 1 and 4). For example, the reduction in QS with respect to QR at SM and PA, and the increase in QS with respect to QR at R, imply that more Gss enters the smallest bifurcate channel at PA and R. Similarly, the shift in thalweg curvature at SM causes the maximum concentrations of suspended bed sediment to remain between the center of the channel and the right margin. This produces a reduction in QS with respect to QR, so that most of the sediment transport occurs within the right bifurcate channel. This behavior may affect the stability of the system, as it causes further morphological change in the channels downstream of the bifurcation. This is due to the fact that the transport capacities of the bifurcate flows are likely not in equilibrium with the amount of sediment they transport, thus causing erosion or deposition.
 During low- and medium-flow stages, unit discharge and flow transport capacity progressively decrease toward the bar head (Figure 10), promoting deposition of the suspended bed sediment. As a result, four factors favor the upstream accretion of bars and islands in the middle Rio Paraná: (i) the existence of inertia-driven sediment advection, (ii) the occurrence of low- and medium-water stages, (iii) the complex behavior of bed forms, which were observed to decrease in size with flow depth as they approach the bar head, and (iv) the grain size distributions of the bed material and suspended load sediment, which determine which sediment is directed in suspension onto the bar head. The morphological analysis conducted before and after a flood (Figure 3) suggests the opposite behavior during high flows, with erosion at the bar head and around the bar. For the upper Rio Paraná, the ratio of bed load: suspended bed sediment transport is much larger, and the bed load grain size is coarser. These factors could influence the morphodynamic evolution of bars and islands in this reach. At site PA (Figure 11), the grain size distributions of the bed material and the suspended bed sediment show little overlap. For instance, if a grain size of 0.2 mm is considered, then 86% of the sediment is larger than this diameter in bed samples M1 and M2, 48% in sample M3 just upstream of the bar, while only 2% and 10% exceed this value in the suspended bed sediment and bar head pit samples, respectively. These observations suggest the presence of two important processes: (a) a progressive decrease in the grain size of the bed sediment transport and/or (b) the progressive deposition of suspended bed sediment, as shown by the presence of climbing ripples in the preserved sediments (Figure 12). Once on the bar head region, this deposited sediment may be later reworked as bed load, especially during higher water levels. Figures 5a, 5c, and 5d, which are representative of medium- and low-flow stages, show that dunes can only be formed just upstream of the bar head. It is only during high-flow stages (highest velocities and bed shear stresses) that dunes are able to migrate to the top of the bar head (Figure 5a). This behavior shows that the deposition of bed load is dependent on flow stage, whereas the suspended bed sediment load is able to be deposited across a full range of stages. The presence of vegetation (Figure 12) just upstream and around the bar head may also decrease the flow velocity and promote further deposition of the suspended bed sediment. At lower river stages, the vegetation density increases due to new growth, creating additional roughness and favoring deposition of suspended bed material during subsequent high-flow stages. Data available at bifurcation PP is scarce in comparison with that at PA, especially in relation to longitudinal profiles, hydraulic and sedimentological data, and bed samples. In shallowing flows, however, dune height tends to diminish toward the bar head in profile D-D′ (Figure 5b), which is compatible with the small dunes visible on the bar surface in Figure 5e.
Edmonds and Slingerland  and Thomas et al. report that low Shields numbers cause a smaller superelevation of the water surface, which results in smaller cross-sectional pressure gradients, which in turn cause a more gradual flow partitioning and smaller bifurcation angles. The observed geometry of islands in the Rio Paraná could be explained by this phenomenon, since these bifurcations are characterized by low Shields numbers and small bifurcation angles. Contrary to the findings from laboratory experiments and numerical models, flow in large rivers such as the Rio Paraná possesses small curvatures, thus preventing the formation of strong secondary currents. The bifurcations in this study can be classified as unstable due to (i) the low values of Shields stress [Bolla Pittaluga et al., 2003; Federici and Paola, 2003; Zolezzi et al., 2006]. (ii) the nonuniform, cross-sectional distribution of flow velocities [Bolla Pittaluga et al., 2003; Federici and Paola, 2003] (iii) unbalanced QR and QS ratios in the bifurcate channels, which cause deposition or erosion, and (iv) complex planform geometries associated with bed form morphology and thalweg diversion. The present study has shown that even though significant changes, such as changes in channel hierarchy, usually take long periods of time (in the order of decades), an ordinary flood can modify, in a relatively short period of time, the morphology of a bifurcation as well as the hydrodynamic conditions downstream.
 The present study has also shown that the combined analysis of flow distribution, suspended bed sediment transport, bed shear stress (Figures 8 and 10) and bed material grain size can help explain the important role played by inertial effects on the fate of the suspended sediment load. The fact that these effects can impact thalweg morphology, as well as the formation and dynamics of bars and islands downstream of the bifurcation, is something that has not been considered in previous studies. Additional quantification and modeling of such effects and their relationship with the bed load sediment transport are still required, especially for different flow stages and in rivers showing different bed load grain sizes. Such future work will help further elucidate the importance of bed load to suspended bed load transport ratios and the effects of sediment inertia on the morphodynamics of channel bifurcations.
ADCP averaging interval, s.
ADCP cell size, m.
suspended bed sediment concentration, mg L−1.
gravitational acceleration, m s−2.
bed suspended sediment transport, kg s−1.
total discharge, m3 s−1.
left branch discharge, m3 s−1.
right branch discharge, m3 s−1.
discharge ratio (QL/QRt).
suspended bed sediment discharge ratio (GssL/GssR).
volume scattering strength, dB.
position coordinate, m.
position coordinate, m.
position coordinate, m.
median bed grain size, m.
water density, kg m−3.
sediment density, kg m−3.
bed shear stress, N m2.
 The authors would like to gratefully acknowledge the help provided by Roberto Viejo Mir, Santiago Cañete, and Roque Negro during the field measurements. This study was conducted within the framework of the project “Analysis of Construction Processes in the Floodplain of a Large River: The Rio Paraná in Its Middle Reach” granted by Universidad Nacional del Litoral (Santa Fe, Argentina). D.R.P. and J.L.B. thank the UK Natural Environment Research Council for funding through grants NER/A/S/2001/00445 and NER/B/S/2003/00243 and the UK Royal Society for a Joint International Project. D.R.P. also thanks NERC for his fellowship funding in grant NE/C002636/1. The authors would also like to acknowledge the constructive and detailed reviews of Jochen Aberle, Scott Wright, and anonymous reviewers, whose comments have considerably strengthened this paper.