The influence of a vegetated bar on channel-bend flow dynamics

. Abstract. Point bars influence hydraulics, morphodynamics, and channel geometry in alluvial rivers. Woody riparian vegetation often establishes on point bars and may cause changes in channel-bend hydraulics as a function of vegetation density, morphology, and flow conditions. We used a two-dimensional hydraulic model that accounts for vegetation drag to 10 predict how channel-bend hydraulics are affected by vegetation recruitment on a point bar in a gravel-bed river (Bitterroot River, Montana, United States). The calibrated model shows steep changes in flow hydraulics for vegetated flows compared to bare-bar conditions for flows greater than bankfull up to a 10-year flow (Q 10 ), with limited additional changes thereafter. Vegetation-morphology effects on hydraulics were more pronounced for sparse vegetation compared to dense vegetation. The main effects were 1) reduced flow velocities upstream of the bar; 2) flow steered away from the vegetation patch with up to a 15 30 % increase in thalweg velocity; and 3) a shift of the high-velocity core of flow toward the cutbank, creating a large cross-stream gradient in streamwise velocity. These modeled results are consistent with a feedback in channels whereby vegetation on point bars steers flow towards the opposite bank, potentially increasing bank erosion at the mid-and downstream end of the bend while simultaneously increasing rates of bar accretion.


Introduction 20
Channel-bend morphodynamics along meandering rivers influence channel morphology, river migration rates, channelfloodplain connectivity, and aquatic habitat. River point bars, fundamental to channel-bend morphology (Blondeaux and Seminara, 1985;Ikeda et al., 1981), steer flow and induce convective accelerations (Dietrich and Smith, 1983) that influence boundary shear stress (Dietrich and Whiting, 1989) and sediment transport fields (Dietrich and Smith, 1983;Legleiter et al., 2011;Nelson and Smith, 1989). Channel migration rates are furthermore controlled by the collective processes of bar accretion 25 and bank erosion. Bars along the inner bends of river meanders, although typically broadly described as point bars, also comprise chute bars, tail bars, chute bars, and scroll bars that reflect distinct formative conditions (e.g., obstructions and/or stream power variations) and produce distinct morphodynamic feedbacks (Kleinhans and van den Berg, 2011).
2 Channel dynamics can be tightly coupled with the recruitment and succession of riparian vegetation on river bars (Amlin and Rood, 2002;Eke et al., 2014;Karrenberg et al., 2002;Nicholas et al., 2013;Rood et al., 1998). Plants change local hydraulics (Nepf, 2012;Rominger et al., 2010) and sediment transport conditions (Curran and Hession, 2013;Manners et al., 2015;Yager and Schmeeckle, 2013), resulting in strong feedbacks between the recruitment and growth of woody riparian vegetation and bar building (Bendix and Hupp, 2000;Dean and Schmidt, 2011) that can influence the morphology of rivers at 5 multiple scales (Bywater-Reyes et al., 2017;Osterkamp et al., 2012). Pioneer vegetation can occur on all bar types but is most likely to survive on nonmigrating bars, such as forced alternating point bars (Wintenberger et al., 2015). Plant traits including height, frontal area, and stem flexibility vary with elevation above the baseflow channel, influencing both the susceptibility of plants to uprooting during floods and their impact on morphodynamics (Bywater-Reyes et al., 2015Diehl et al., 2017a;Kui et al., 2014). Vegetation effects on hydraulics, bank erosion, and channel pattern also depend on the uniformity of 10 vegetation distribution on bars, which can vary depending on wind versus water-based dispersal mechanisms (Van Dijk et al., 2013), and on whether plants occur individually or in patches (Manners et al., 2015). Experimental work in flumes has shown that vegetation is vital to sustaining meandering in coarse-bedded rivers (Braudrick et al., 2009). Vegetation's effect on stabilizing banks, steering flow, and impacting morphodynamics furthermore depends on seed density and stand age. Uniform vegetation on bars has been shown, experimentally, to decrease bank erosion 15 rates, stabilize banks, and increase sinuosity of meander bends (Van Dijk et al., 2013). Gran and Paola (2001) showed that vegetation, by increasing bank strength, generates secondary currents associated with oblique bank impingement that may be more important than helical flows generated by channel curvature. Other experiments have generally suggested vegetated bars decrease velocities over the bar and push flow toward the outer bank. For example, tests in a constructed, meandering laboratory stream with two reed species planted on a sandy point bar showed that vegetation reduced velocities over the 20 vegetated bar, increased them in the thalweg, strengthened secondary circulation, and directed secondary flow toward the outer bank (Rominger et al., 2010). Another study in the same experimental facility, but using woody seedlings planted on the point bar, also found reduced velocities in the vegetated area of the bar, with the greatest reductions at the upstream end, and the effect varying with vegetation architecture and density (Lightbody et al., 2012). In a flume study where meandering effects were simulated in a straight channel by placing dowels representing vegetation patches in alternating locations along the edges 25 of the flume, vegetation reduced velocity within and at the edges of the vegetation patch and increased velocities near the opposite bank (Bennett et al., 2002). Experiments in a high-curvature meandering flume, in contrast, showed that vegetation inhibited high shear-stress values from reaching the outer bank (Termini, 2016), inconsistent with studies simulating moderate sinuosity channels.
Vegetation's effect on river morphodynamics have also been simulated with computational models. Reduced-30 complexity models that approximate the physics of flow have successfully reproduced many of the features observed in channels influenced by vegetation, such as the development of a single-thread channel (e.g., Murray and Paola, 2003). Twodimensional models that use shallow-water equations and, in some cases, sediment transport relations, provide an alternative that may be less dependent on initial conditions and more capable of representing the physics of vegetation-flow interactions 3 (Boothroyd et al., 2016(Boothroyd et al., , 2017Marjoribanks et al., 2017;Nelson et al., 2016;Nicholas et al., 2013;Pasternack, 2011;Tonina and Jorde, 2013). Investigations of channel-bend dynamics influenced by vegetation using two-dimensional models often represent vegetation by increasing bed roughness (see Green, 2005 andCamporeale et al., 2013 for comprehensive reviews). Nicholas et al. (2013) simulated bar and island evolution in large anabranching rivers using a morphodynamic model of sediment transport, bank erosion, and floodplain development on a multi-century timescale where vegetation was modeled 5 using a Chezy roughness coefficient. Asahi et al. (2013) and Eke et al. (2014) modeled river bend erosional and depositional processes that included a bank-stability model and deposition dictated by an assumed vegetation encroachment rule. Bertoldi and Siviglia (2014) used a morphodynamic model coupled with a vegetation biomass model, which accounted for species variations in nutrient and water needs to simulate the coevolution of vegetation and bars in gravel-bed rivers. Vegetation was modeled as increased bed roughness via the Strickler-Manning relation that varied linearly with biomass. Their model showed 10 two scenarios: one where flooding completely removed vegetation, and one where vegetation survived floods, resulting in vegetated bars. These two alternative stable states (bare versus vegetated bars) have been found experimentally as well (Wang et al., 2016).
Although the aforementioned models produce many of the features of river morphodynamic evolution, when vegetation drag is dominant over bed friction, using conventional resistance equations (e.g., Manning's n; roughness) to model 15 vegetation's effect on the flow introduces error. Increasing the roughness within vegetated zones increases the modeled shear stress and therefore artificially inflates the sediment transport capacity at the local scale (e.g., vegetation patch or bar), although reach-scale results may be appropriate (Baptist et al., 2005;James et al., 2004). Vegetation drag can also be treated in computational models by representing plants explicitly as cylinders (e.g., Baptist et al., 2007;Vargas-Luna et al., 2015a), comparable to the approach of many flume studies, or by accounting for drag from foliage, stems, and streamlined vegetation, 20 but such an approach is currently not widely adopted because of limited ability to specify all parameters (e.g., Boothroyd et al., 2015Boothroyd et al., , 2017Jalonen et al., 2013;Västilä and Järvelä, 2014). Vargas-Luna et al. (2015a) showed through coupling of numerical modeling and experimental work that representing vegetation as cylinders is most appropriate for dense vegetation. Iwasaki et al. (2015) used a two-dimensional model that accounted for vegetation drag to explain morphological change of the Otofuke River, Japan, caused by a large flood event in 2011 that produced substantial channel widening and vegetation-25 influenced bar building. They found that vegetation allowed bar-induced meandering to maintain moderate sinuosity, whereas in the absence of vegetation, river planform would switch from single-thread to braided. Marjoribanks et al. (2017) modeled the effects of vegetation on channel hydraulics for a small (~5 m wide by 16 m long), straight river reach and found velocity reduced broadly throughout the channel.
As the above review suggests, there have been considerable advances in laboratory and computational modeling of 30 vegetation effects on hydraulics that complement understanding of bar and bend morphodynamics and reciprocal interactions between riparian vegetation and river processes (Corenblit et al., 2007;Gurnell, 2014;Osterkamp and Hupp, 2010;Schnauder and Moggridge, 2009). Challenges persist, however, in representing field-scale complexities in a modeling framework that allows for testing field-scale interactions between plants, flow, and channel morphology on vegetated point bars. Here we 4 tackle key elements of this problem by investigating how the distribution of woody vegetation on a point bar influences bend hydraulics and flow steering across a range of flood magnitudes using a two-dimensional modeling approach informed by high-resolution topography and vegetation morphology data that spatially defines vegetation drag. We model a range of vegetation densities and plant morphologies representing different stages of pioneer woody vegetation growth on a point bar.
We vary discharge in the model to represent the stage-dependent effects of vegetation on hydraulics, as well as different flood 5 stages that may be important for the recruitment of plants and the erosion or deposition of sediment within the channel bend.
We predict that the presence of woody vegetation affects bar and meander dynamics by steering flow, thereby influencing the morphodynamic evolution of vegetated channels. Our objectives are to 1) Determine which vegetation morphology and flow conditions result in the greatest changes to channel-bend hydraulics; and 2) Infer how these changes in hydraulics would impact channel-bend morphodynamics and evolution. The insights derived from our analysis are relevant for understanding 10 ecogeomorphic feedbacks in meandering rivers, understanding how such feedbacks are mediated by plant traits and flow conditions, and for riparian plant species management along river corridors.

Study area
To meet our objectives, we model a point bar-bend sequence on the Bitterroot River, southwest Montana, United States ( Fig.  15 1). Our field site has a pool-riffle morphology and a wandering pattern, with channel bends, point bars, and woody vegetation on bars and floodplains. The study reach is located on a private reserve (MPG Ranch) with minimal disturbance to the channel and floodplain, and flow and sediment supply are relatively unaltered by flow regulation, because the only significant dam in the contributing watershed is ~120 km upstream of the study reach, on a tributary. Annual mean discharge is 68 m 3 s -1 , bankfull Shields number is 0.03, and median grain size is 23 mm, and drainage area of ~6,200 km 2 . Woody bar vegetation is composed 20 of sand bar willow (Salix exigua), and cottonwood (Populus trichocarpa) seedlings, saplings, and young trees (

Flow model
To characterize the influence of a vegetated bar on channel-bend hydraulics, we used an edited version of FaSTMECH (version 2.3.2), a hydrostatic, quasi-steady flow model contained within iRIC (Nelson et al., 2016; http://i-ric.org/en/index.html).
FaSTMECH solves the depth-and Reynolds-averaged momentum equations in the streamwise and cross-stream directions, in a channel-fitted curvilinear coordinate system, using a finite-difference solution (Nelson et al., , 2016. By convention, 5 values of u and v are positive downstream and toward the left bank, respectively. Bed stress closure is achieved through a drag coefficient (C d ) scheme. Details of the modeling process, beyond those provided in the text here, can be found in the Supplement.
We created the flow model domain in FaSTMECH by characterizing the topography and flow boundary conditions (discharge and water surface elevation at the downstream boundary) of a study reach on the Bitterroot River, Montana ( Fig.  10 1). We surveyed channel topography with a combination of airborne LiDAR, echosounder and RTK GPS surveys (see Supplement). The resulting curvilinear orthogonal grid we created had an average cell size of 2.5 by 2.5 m for calibration runs (described below), and 5 by 5 m for the remaining runs. We linked transducer stage measurements at the downstream end of the study reach to discharge derived from USGS 12344000 Bitterroot River near Darby MT, corrected by contributing area for our field site. Water surface elevations at the downstream boundary for modeled discharges were extracted from the stage-15 discharge relationship. Discharge was measured at the field site and compared to the adjusted USGS 12344000 value and found to agree within 10 % (Table 1).    FaSTMECH uses relaxation coefficients to control changes in a parameter between iterations (Nelson, 2013).
Convergence was found after 5000 iterations (mean error discharge < 2 %), considered indicative of adequate model performance for FaSTMECH (Nelson, 2013). We calibrated channel characteristics (bed roughness specified as C d and lateral eddy viscosity, LEV) and considered them fixed after calibration (Table 1). We used a constant C d , an approach that has been 5 shown elsewhere to perform comparably to variable roughness in FaSTMECH (e.g., Segura and Pitlick, 2015). We set C d to minimize the root mean square error (RMSE) of modeled water surface elevation (WSE) versus WSE measured in the field from 2011-2015, over a range of calibration flows (see Supplement).. In this calibration process we manually varied C d values from 0.01 to 0.001, resulting in a C d of 0.003 and lowest RMSE's for WSE from 0.11 to 0.16 m for the lowest and highest calibration flows, respectively (Table 1). Similarly, we manually varied LEV from 0.01 to 0.001 during model calibration, 10 resulting in a LEV value of 0.04, which minimized RMSE of depth-averaged velocity (Ū=0.24 m s -1 ; Table 1) between modeled values and those measured at four cross sections ( Fig. 1) (see Supplement for details). The RMSE ranges obtained through calibration are consistent with values reported in other studies that have used FaSTMECH (e.g., Legleiter et al., 2011;Mueller and Pitlick, 2014;Segura and Pitlick, 2015), providing confidence in model performance.

Modeling vegetation's impact on channel-bend hydraulics 5
We edited FaSTMECH to account for vegetation form drag (F D ) using the following drag equation for rigid vegetation: (Nepf, 1999;Vargas-Luna et al., 2015, 2016, and U c is the approach velocity. Drag (F D ) is calculated per bed area (distributed over vegetation polygons). 10 For U c we used node velocity (after Jalonen et al., 2013). The vegetation drag coefficient (C d,v ) was assigned a value of one, a first-order approximation also used by others (Boothroyd et al., 2016;Nepf et al., 2013;Vargas-Luna et al., 2016). We modeled vegetation as cylinders by assuming the cylindrical stem frontal area is equal to A c , specifying vegetation parameters by polygon with an associated stem density (#stems m -2 ) and height (m; allows for partitioning of A c by flow depth). The model assumes a logarithmic velocity profile, although we recognize this is an over-simplification of how factors such as vegetation submergence alter velocity profiles (e.g., Manners et al., 2015).
We focused our analyses on a point bar (Fig. 1) that supports woody riparian vegetation (Populus seedlings, saplings, and young trees) most likely recruited mainly by flood dispersal. In our model simulations, we varied vegetation density (#stems m -2 ) and A c (m 2 per plant) on the vegetated bar for each of the four flows, and we compared model output to a no-5 vegetation (no veg) scenario. We considered two vegetation density cases: sparse (sps) and dense (dns). Our sparse case was based on the average density (0.02 stems m -2 ) obtained from the airborne LiDAR (see Supplement for more detail). Our dense case (20 stems m -2 ) was based on the average from random vegetation density plots measured on the bar, which ranged from <1 stem m -2 to 227 stems m -2 and is consistent with other dense field-measured values (Boyd et al., 2015;van Oorschot et al., 2016;Wilcox and Shafroth, 2013). For A c , we used ground-based LiDAR to capture vegetation structure (Antonarakis et al., 10 2010;Bywater-Reyes et al., 2017;Manners et al., 2013;Straatsma et al., 2008). We scanned Populus patches representing different stages of pioneer woody vegetation growth: seedlings (sdl), saplings (sap), and young trees (tr). From these scans We considered hydraulic (u, v) solutions for three cross sections at locations across the bar and cutbank of the channel bend, representing the upstream, midstream, and downstream portion of the bar (Fig. 3). We additionally considered the hydraulics and potential for bed mobility spatially, where the Shields number, * , was used as an indicator of bed mobility: 25 where is boundary shear stress, is sediment density, is acceleration due to gravity and is grain diameter. We used the median grain diameter from pebble counts collected on the study bar and along cross sections (Fig. 1). We compared the solutions for vegetation runs for each flow to no-vegetation scenarios to evaluate which configurations had the greatest influence on hydraulics. 30

Results
The effects of point bar vegetation on modeled hydraulics across our study reach are presented here in several ways. First we compare vegetation results, for different density and growth stages, to the no-vegetation case; and second, we compare results spatially at different cross sections across the bar at different discharges. For the no-vegetation case, velocity and shear stress were generally highest in the thalweg and lower over the bar (Fig. 4). Downstream velocity (u) was generally greater than 5 cross-stream velocity (v). The greatest v magnitudes were for the downstream cross section (XS1; Fig. 5c,d). With increasing flow magnitude, both u (Fig. 5b) and v (Fig. 5d) decreased within the thalweg region, but stayed relatively constant over the bar. A similar trend was seen at the mid-bar cross section (XS2) with u decreasing within the thalweg region as flow magnitude increased, but remaining relatively constant over the bar (Fig. 6). In contrast, u increased within the thalweg region and over the bar with increasing flow (Fig. 7a,b) at the upstream cross section (XS3), whereas v stayed relatively constant (Fig. 7c,d). 10 (Fig. 3 shown). Velocity and Shields number are reduced on the bar with increasing size or density of plants, and flow paths within the thalweg and adjacent to the vegetation patch become more concentrated.  Q2 (a,c) and Q10 (b,d) flows. In the thalweg (j = 90) and at the head of the bar, u is decreased with increasing seedling size and density. For Q ≥ Q10, v became more negative adjacent to the vegetation patch.

Figure 4. Planview comparison of channel-bend hydraulics (velocity; a-c, and Shields number; c-f) for the Q10 no-vegetation (a,d), sparse young trees b,e), and dense seedlings (c,f) runs. Location of cross sections
The manner in which different vegetation densities and growth stages influenced hydraulics varied spatially around 5 the bend. In general, adding vegetation increased velocity within the thalweg and at the edge of the vegetation patch compared to the no-vegetation case, creating concentrated flow paths adjacent to the patch while reducing velocity and shear stress at the head of the bar and within the vegetation patch. The effect of the vegetated bar on channel-bend hydraulics became more pronounced with discharges increasing from the Q 2 to Q 10 . Furthermore, sparse vegetation behaved similarly to the novegetation scenario for low flows, but had an increasing effect on hydraulics at the Q 10 . Vegetation effects increased steeply 10 from Q 2 to Q 10 with modest changes thereafter. In general, hydraulics were more sensitive to plant morphology differences (A c ) for sparse conditions compared to dense conditions (Fig. 5, 6, 7).
At the downstream end of the bar (XS1; Fig. 5), vegetation increased the magnitude of downstream (u) and crossstream (v more negative) velocity within the thalweg region, and reduced velocities over the bar. For flows ≥ Q 10, the highvelocity core became more concentrated and shifted away from the bar. Results for the Q 20 and Q 100 flow were similar to that 15 of the Q 10 (Fig. S2). This thalweg effect became more pronounced with increasing plant density and plant size, except in the case of dense young trees, which behaved more similarly to the bare bar scenario for the Q 10 flow. Amplification of thalweg velocities at XS1 was greatest for the dense sapling scenario, with 17 % and 12 % increases in u and v, respectively, for the Q 10, and increases in velocity magnitude for flows > Q 10 . On the vegetated bar, u and v decreased within the vegetated patch, with u values reduced up to 56 % for the sparse young tree scenario, and up to 95 % for the dense scenariosthese magnitudes 20 are well above uncertainty in velocities. With increasing plant size and density, the values of u and v at the right edge of the vegetation patch were greater than or nearly equal to that in the thalweg, with a particularly large increase for dense scenarios.
Thus, flow velocities were decreased within the patch, increased adjacent to the patch, and were deflected toward the left bank.
At the midstream position (XS2), downstream velocities (u) in the thalweg region were greater than at XS1. The impact of the vegetation patch on u for XS2 was pronounced, with u increased up to 30 % within the thalweg and the maximum 25 value of u shifted toward the left bank with increasing plant size, density, and discharge (Fig. 6). Like XS1, the thalweg effect reached a maximum for dense saplings at the Q 10 . As flow increased (Q 20 and Q 100 ), dense trees had the greatest effect on increasing thalweg u. On the bar, the effect on u for XS2 was similar to XS1. Values of u decreased with increasing size and density of plants, and u increased at the right outer edge of the vegetation patch. Over the bar, u was reduced up to 99 % for the dense scenarios compared to the no-vegetation scenario, and increased at the edge of the patch up to 3300 %. At XS2, v 30 values were small compared to XS1 and XS3 and were relatively insensitive to the presence of the vegetation patch (Fig. S3).
At the upstream end of the bar (XS3; Fig. 7), an opposite trend in changes in u within the thalweg was observed. With increasing seedling size and density, u was decreased within the thalweg and at the head of the bar, with a maximum reduction in u of 29 % for dense scenarios. Results for the Q 20 and Q 100 flow were similar to that of the Q 10 (Fig. S4). For Q ≥ Q 10 , v was more positive to the left (70 %) of the vegetation patch and more negative to the right of the vegetation patch (180 % reduced).
Within the vegetation patch, u and v were reduced (96 % and 100 %, respectively). Thus, flow was steered away from the vegetation patch.

Impact of vegetation on channel-bend hydraulics 5
Our results illustrate that vegetation enhances flow steering on bars, complementing previous work on bend dynamics in the absence of vegetation. Dietrich and Smith (1983) showed that bars steer flow in a manner that forces the high-velocity core toward the concave bank. They additionally found that flow over the heads of bars resulted in cross-stream components of velocity (v) and boundary shear stress ( ) directed toward the concave bank. Laboratory studies by Blanckaert (2010), representing sharp meander bends, identified relationships between zones of inward versus outward mass transport, transverse 10 bed profiles, and curvature variations, as well as illustrating that curvature-induced secondary flow associated with topographic steering concentrates most discharge over the deepest, outer parts of a bend. Whiting (1997) hypothesized that convective accelerations arising from flow steering would be most important at low flows, whereas Legleiter et al. (2011) showed that steering from bars continued to be important with increasing discharge. We found that vegetation began to have a detectable impact on channel-bend hydraulics for flows greater than the Q 2 , when plants were inundated, and that vegetation-induced 15 alteration of hydraulics was initially steep from Q 2 to Q 10 , with modest changes thereafter. Vegetation effects on flow did not decrease with increasing discharge, consistent with Abu-Aly et al. (2014). Whereas flow steering from bars may be most important at low flows (Whiting, 1997), our results suggest that flow steering will continue to be important over a range of flows for vegetated bars.
In general, we found the impact of the vegetated bar on channel-bend hydraulics to vary with both plant density and 20 morphology, and our modeling illustrated nuances in these relationships. Plant morphology differences affected hydraulics preferentially for sparse cases, whereas dense cases were similar. Dense young trees did not always result in the maximum alteration to channel-bend hydraulics-particularly for u during the Q 10 flow. At the downstream end of the bar, the highvelocity core became more concentrated and shifted away from the bar with increasing plant density and plant size, except in the case of dense young trees. Dense young trees behaved more similarly to the bare bar scenario for the Q 10 . This indicates 25 there may be thresholds whereby increasing density and size of vegetation no longer results in an additional hydraulic effect in some cases. Together, these results suggest altered bend hydraulics caused by bar vegetation may be most pronounced for vegetation-inundating flows up to Q 10 under sparse-vegetation conditions. We may expect vegetation-morphodynamic interactions to be strongest as recruited sparse woody riparian vegetation matures under moderate flow conditions (> Q 2 to Q 10 ), or conversely, if a bare bar establishes dense vegetation. This is consistent with the biogeomorphic phase concept 30 (Corenblit et al., 2007(Corenblit et al., , 2015a(Corenblit et al., , 2015b, whereby established vegetation has strong feedbacks with geomorphic processes, but with the additional constraint of enhanced interactions under a specific range of flow magnitudes.

20
At the mid-and downstream sections of the channel bend investigated, the presence of dense vegetation increased downstream velocity (u) within the thalweg up to 30 % and shifted the high-velocity core toward the cutbank. Vegetation increased the magnitude of cross-stream velocity (v) at both the up-and downstream end of the channel bend by increasing cross-stream flow toward the cutbank at the head of the bar and around the toe of the bar. Positive v values within the thalweg region at the upstream cross section (XS3) indicate, indeed, flow is steered toward the concave bank. By extension, cross-5 stream shear stress,is directed toward the concave bank. At the head of the bar, flow was additionally slowed within the channel (u decreased), and steered away from the vegetation patch, increasing flow within a side channel adjacent to the bar head and creating concentrated flow paths adjacent to the patch.
We acknowledge that some of our findings may be influenced by limitations of our modeling approach, which reflect persistent challenges in characterizing the complexities of vegetation architecture and flow in a modeling framework. Our 10 analysis simplified vegetation drag by assuming rigid cylinders, which has been found to most accurate for dense vegetation (Vargas-Luna et al., 2016). We used a constant vegetation drag coefficient, consistent with other studies. This approach likely overestimates vegetation drag at higher discharges, when the canopy is inundated and plants are more streamlined, reducing A c and C d,v (James et al., 2004). Future research directions include: 1) refining how vegetation drag is represented, especially for sparse vegetation; 2) quantifying changes in drag that result from streamlining and reconfiguration during inundation; 3) 15 including variations in drag coefficient for vegetation to represent depth-dependence and complexities of vegetation architecture; and 4) evaluating effects of non-logarithmic vertical velocity structures Boothroyd et al., 2017;Nepf, 2012;Västilä et al., 2013;Västilä and Järvelä, 2014;Whittaker et al., 2013).

Implications for channel morphology and evolution 20
The reduction of velocity and shear stress within the thalweg at the bar head caused by the presence of the vegetated bar would be expected to decrease sediment transport in this region. Van Dijk et al. (2013), in an experimental channel, found bar vegetation to increase fine-sediment deposition upstream of the vegetation patch, analogous to the bar head of our work. This may contribute to bar-head maintenance, such that the head of the bar is not eroded. Maintenance of the bar head would be countered by the potential for chute cutoff  or channel switching that may result because of concentrated 25 flow paths. Along the inside (river right) edge of the vegetated bar, a lower-elevation, chute-channel-like region is present, in which flow was concentrated and velocities increased as vegetation size and density increased. Seedling establishment was not successful in the lower-elevation region during the study period, possibly because higher shear stresses in this region limited fine sediment deposition conducive to recruitment and/or exceeded uprooting thresholds (Bywater-Reyes et al., 2015).
The concentrated flow paths adjacent to the vegetation patch, on the inside of the bar, may be characteristic of conditions on 30 vegetated bars along channel bends more generally, where both ridge and swale topography and chute bars may be present (Kleinhans and van den Berg, 2011), and where chute cutoffs and vegetation roughness and cohesion interact to influence morphodynamics (e.g., Braudrick et al., 2009). Seedlings often recruit along floodlines (Schnauder and Moggridge, 2009), forming rows of trees. Low-velocity areas within the rows induce fine-sediment deposition, steering flow away from the rows, and increasing velocity and shear stress adjacent to the rows such that sediment is transported in these regions. This process has been invoked to explain how vegetation creates vegetated islands (Gurnell et al., 2001), alternating patterns of vegetated ridges and adjacent channels (Tooth and Nanson, 2000), and the evolution of anabranching channels (Tooth et al., 2008). Van Dijk et al. (2013) found flood-dispersed vegetation recruited on bars resulted in island braiding, whereas vegetation distributed 5 uniformly across the floodplain maintained a single-thread meandering channel with increased sinuosity and decreased bend wavelength. Our analysis, more comparable to the flood-dispersed case, shows the potential for development of vegetated islands, but also for prevention of chute cutoff through bar-head maintenance; chute cutoff may be more likely in the absence of vegetation (Constantine et al., 2010).
The production of a low-velocity region over the vegetated bar could increase fine-sediment deposition on the bar, 10 consistent with flume and field observations. Elevated sediment deposition within patches of woody seedlings, with variations depending on plant characteristics, has been documented in meandering (Kui et al., 2014) and straight (Diehl et al., 2017b) flumes. Gorrick and Rodríguez (2012), working in a flume in which vegetation patches were simulated with dowels, documented elevated fine-sediment deposition within the patches (Gorrick and Rodríguez, 2012). Zones of fine sediment deposition on bars associated with roughness from vegetation or instream wood can in turn create sites for plant germination 15 and seedling growth (e.g., Gurnell and Petts, 2006). If reduced velocities result in increased deposition of sediment on the bar, bar accretion would induce additional topographic steering. This feedback would be expected to accelerate channel migration rates.
The increase in velocity and shift of the high velocity core toward the cutbank combined with low velocities within the vegetation patch would create a large velocity gradient across the channel. A larger velocity gradient within the thalweg 20 compared to over the bar would be expected to alter the dynamics of bank erosion. As a simple rule, bank erosion rate, ṅ, according to Parker et al.'s (2011) HIPS model is proportional to an erosion coefficient, k, and half the streamwise velocity difference between the two banks, Δu: The parameter, k, represents the material cohesion and vegetation root properties that control bank erosion and varies between 25 10 -8 and 10 -7 (dimensionless). For an assumed k, vegetation-induced velocity gradients across the channel are expected to alter bank erosion rates. Thus, although we show the potential for concentrated flow paths to cause a morphology with vegetated islands, our analysis does not preclude bar vegetation from increasing bank erosion rates that would tend to increase sinuosity, in contrast to previous experimental work suggesting flood-established vegetation should result in island development (Van Dijk et al., 2013). 30 Vegetation "pushing" flow toward the outer bank is analogous to "bar push" (Allmendinger et al., 2005;Parker et al., 2011), whereby a rapidly accreting point bar may cause erosion at the outer bank (Eke et al., 2014;van de Lageweg et al., 2014). This increase in bank erosion would be countered by deposition of fine sediment on the bar resulting from the vegetation-induced reduction in velocity in this region, that may in turn induce addition "push" through bar building (e.g., Eke 22 et al., 2014). Coarse bank roughness counters this effect, pushing the high velocity core back toward the center of the channel (Gorrick and Rodríguez, 2012; Thorne and Furbish, 1995). The balance between erosion of the bank and deposition on the bar would thus dictate whether net erosion or net deposition within the active channel occurs, inducing changes in channel width (Eke et al., 2014), and altering channel morphology.

Conclusion
The presence of a vegetated bar in a gravel-bed river altered both streamwise and cross-stream components of velocity vectors for overbank flows, with an increasing effect with discharge and both plant density and size. Vegetation steered flow away from the vegetated bar, creating concentrated flow paths in surrounding low-elevation side channels and a low-velocity region over the vegetated patch. Flow was slowed at the apex of the bar, and increased within the thalweg around the bend. These 10 changes in hydraulics could increase fine sediment deposition on the bar, potentially creating hospitable sites for vegetation recruitment, and increasing bank erosion that is dependent on cross-stream velocity gradients. This pattern would tend to reduce cross-stream sediment transport at the bar head, but increase it around the remainder of the bend.
Following the pattern of hydraulics, we would expect vegetation to change the morphodynamic evolution of channels with vegetation pushing flow in a manner typically attributed to bars, and increasing bank erosion rates. Bank retreat may 15 induce bar building, which could be accelerated by fine-sediment deposition within the vegetation patch. This feedback would induce additional topographic steering from the presence of the bar. With a numerical model, we have characterized a mechanisms by which channels with vegetated bars may evolve different morphologies and rates compared to those without, thereby contributing to understanding of ecogeomorphic feedbacks in river-floodplain systems (Gurnell, 2014) and of how life influences landscapes (Dietrich and Perron, 2006). 20