Transitions in river character and behavior are a key focal point of enquiry in fields such as geomorphology, hydrology, and sedimentology. Such concerns have significant management applications, especially relating to issues such as management of flood risk and sedimentation hazards. These issues are likely to become even more pronounced in the future, as rivers adjust in response to climate and land use changes, and management actions. Putting aside concerns for terminological issues associated with differentiation of river types and their morphological attributes (see Lewin and Ashworth, 2014; Carling et al., 2014; Tadaki et al., 2014), process-based understandings of morphodynamic adjustments are required to address concerns for prospective future river changes (Beechie et al., 2010). Here we evaluate the influence of riparian vegetation upon process interactions along relatively understudied reaches of the Upper Yellow River atop the Qinghai–Tibet Plateau in western China. Qualitative description and analysis of this complex influence on bedload transport capacity remains unclear to our knowledge.

Channel bars are products of instream deposition of bedload materials, whether at the channel margin (bank-attached forms) or mid-channel bars (Brierley and Fryirs, 2005). Typically, bars mutually adjust with channel geometry, such that they scale to the size of the channel in which they form (Task Force on Bed Forms in Alluvial Channels, 1966; Nicholas et al., 2013). If these features become vegetated and stabilized, they are referred to as islands (Fryirs and Brierley, 2012). Unit bars (migrating lobate bed forms with heights and lengths that scale with channel depth and width) are differentiated from larger, more complex compound bars (e.g., Bridge, 1993; Brierley, 1989, 1991; Brierley and Hickin, 1991; Smith, 1974). Compound bars are products of multiple phases of accretion and reworking, with stacked sequences of unit bar, dune, and smaller bed form deposits that are often trimmed at their margins by bank erosion processes or dissected by chute channels (Ashworth et al., 2011; Best et al., 2003; Bridge, 2003; Brierley and Fryirs, 2005; McGowen and Garner, 1970; Reesink et al., 2014; Sambrook et al., 2009). Various studies have characterized the main morphological elements of large bars and islands, while other studies have developed conceptual models of bar evolution (e.g., Ashworth et al., 2000; Gurnell et al., 2001; Latrubesse and Franzinelli, 2002; Mertes et al., 1996).

There is notable variability in the presence, form, and hydraulic/sedimentologic (morphodynamic) role of bars along the continuum of channel planform (Bridge, 1993; Brierley, 1996). By definition, as suspended-load rivers have limited bedload-calibre materials, they have very few, if any, bars. The prominence of fine-grained (silt-clay) deposits under low energy conditions (often very low channel gradient) promotes passive channel behavior, typically with a low sinuosity, passive meandering, or anabranching (anastomosing) planform (Eaton et al., 2010; Fryirs and Brierley, 2012; Makaske, 2001; Wang et al., 2005). Patterns of bar formation in mixed- and bedload-dominated rivers reflect the flow–sediment balance along any given reach, with a spectrum of planform types ranging from active meandering and wandering variants through to fully braided rivers (see Ashworth, 1996; Ashworth and Lewin, 2012; Burge, 2006; Church and Rice, 2009). Braiding results from the inability of flow to transport all sediments that are made available to the channel, such that mid-channel sedimentation occurs (i.e., competence and/or capacity limits are exceeded). Recurrent reworking of bedload materials via thalweg shift during flood events alters the number, shape, and location of bars. Bar dissection and avulsion create multi-thread channel systems with a disorderly river planform, extremely unstable bars, and inconstant flow paths (Ashmore, 1991; Ashworth et al., 2000; Jerolmack and Mohrig, 2007).

If channel boundary conditions induce sufficient bank strength, and flows are able to transport available bedload sediments, the river adopts a configuration with better-defined, less mobile channels with a much lower width–depth ratio, whether within a single-channel configuration (typically passive meandering) or a multi-channel anabranching configuration (Eaton et al., 2010; Song and Bai, 2015). Controversy abounds in our theoretical understanding of process controls upon anabranching river behavior (see Carling et al., 2014; Nicholas et al., 2013). While Huang and Nanson (2007) and Jansen and Nanson (2004, 2010) attributed an anabranching configuration to the least action principle, wherein channels adjust their form to transport available sediment in the most hydraulically efficient manner, Eaton and co-workers postulated an alternative theoretical framing in which anabranching channels adjust to minimize their capacity to transport materials (Eaton and Church, 2007; Eaton et al., 2004, 2010). It is not our concern here to address this issue directly. Rather, our focus lies with analysis of relationships between bedload transport capacity and channel morphodynamics along a continuum of channel planform types that is coincident with a gradation in riparian vegetation cover along the Upper Yellow River (Yu et al., 2014).

Bar stability is the key distinguishing attribute of braided and anabranching rivers. In some instances, vegetation may support the long-term development of stable sandbars within a stable multi-channel system – a variant of an anabranching river (Latrubesse, 2008; Nanson and Knighton, 1996; Murray and Paola, 2003; Tal and Paola, 2010). Vegetation increases flow resistance and stabilizes the channel bed and bank along anabranching rivers, thereby altering channel geometry, bedload transport rates, and the resulting rates and patterns of bed deposition or erosion. Once a particular morphology has been formed, the configuration of channels and associated distribution of bars and roughness elements fashion process responses to subsequent flood events (Hooke, 1986, 2015; Hooke and Yorke, 2011; Luchi et al., 2010). If critical threshold conditions are exceeded, alterations to the balance and patterns of erosion and deposition processes may bring about transitions to different planform types.

Mutual adjustments between patterns of vegetation types (size, spacing, and density) and flow–sediment dynamics (patterns and rates of erosion and deposition) vary at different positions on the valley floor. Vegetation encroachment by pioneer species and successional processes induce abiotic and biotic transitions in geomorphic processes from the unvegetated channel bed and bar surfaces to grassland, shrubs, and treed areas at the margins of bars/islands and on floodplains (Corenblit et al., 2007, 2011; Gurnell, 2014; Hickin, 1984; Hupp and Osterkamp, 1996; Millar, 2000; Tooth and Nanson, 2000). Vegetation attributes influence the pattern of roughness elements and the associated distribution of flow energy, thereby affecting the distribution of erosional and depositional processes, and resulting morphological attributes (including the grain size distribution of bed/bar materials). Hence, vegetative controls influence the stability and behavior of the alluvial bed and bars, and the influence of vegetation upon flow–sediment interactions vary for differing planform types (Gran and Paola, 2001; Gradzinski et al., 2003; Jang and Shimizu, 2007; McBride et al., 2007).

Although the prominence of seasonal low–flow stages and nutrient-rich fine sands may support the growth of annual or perennial herbs and shrubs on mid-channel and transverse bars in braided rivers, this sparse vegetation cover has a negligible impact upon sediment deposition and erosion patterns, and is removed easily at flood stage (Coulthard, 2005). This mutual interaction between vegetation and erosion-deposition can be viewed as a threshold condition: if sufficient vegetation establishment occurs, resistance may exceed the erosion-deposition capability of a normal flood such that stabilization ensues, prospectively altering sedimentation patterns, increasing bank strength, and reducing channel width–depth ratio (Gran and Paola, 2001; Coulthard, 2005; Eaton et al., 2010). In anabranching channels the vegetation cover on mid-channel bars inhibits lateral migration, inducing a stable branching channel condition. During lower frequency floods, when bars are partially or completely submerged by flow, vegetation increases flow resistance, traps sediment, and inhibits erosion.

This study builds upon previously reported exploratory analyses of river diversity in the source zone of the Yellow River (Blue et al., 2013; Brierley et al., 2016; Li et al., 2013; Yu et al., 2014). In this region, herbs and sparse shrubs that grow on the sand/gravel bars of braided rivers have a trivial influence upon channel morphodynamics, while establishment of dense shrubs and sparse trees on sand/gravel bars promotes the emergence of anabranching channel configurations. Building on these previous observations, here we appraise process interactions along a vegetative gradient of river morphologic adjustments for four reaches: the Dari and Maqu reaches of the Yellow River main stream, and the Daheba and Lanmucuo river tributaries of the Upper Yellow River (Table 1). The Dari reach has a semi-stable braided channel, where sandbars are covered by herbaceous vegetation and sparse shrubs. The Maqu reach has a very stable anabranching channel with dense willows (Salix atopantha) on sandbars. The study reach along the Lanmucuo River has a stable gravel meandering river with herb coverage. The study reach along the Daheba River has a very unstable gravel braided channel without vegetation cover. We develop and apply a simplified model to explain the interaction between sediment transport capacity and river bed deposition in these reaches, examining the effect of vegetation resistance and adjustment of fluvial hydraulic geometry. From this, we quantitatively analyze the stability and evolution of braided, anabranching, and meandering reaches during flood events.

Upstream of Tangnaihai hydrological station the source zone of the Yellow River drains an area of 132 000 km2 (see Fig. 1a). In the 1950s, the Yellow River Conservancy Commission established four hydrological stations along the Yellow River in this area, namely (from upstream to downstream), Huangheyan station in Maduo County, Jimai station in Dari County, Maqu station in Maqu County, and Tangnaihai station in Xinhai County. The reach from Huangheyan to Jimai station is 325 km long and drains an area of 24 089 km2. In this reach the valley is quite wide, with semi-braided and semi-anabranching planform morphologies characterized by disordered channels with many bars (Brierley et al., 2016). The reach from Jimai to Maqu is 585 km long and drains an area of 41 029 km2. The upper section of this reach has a deeply incised (confined), sinuous valley between the Anyemaqen and Bayan Har Mountains. Flowing into the Ruoergai alluvial basin, there is a diverse array of planform types (Blue et al., 2013; Brierley et al., 2016; Li et al., 2013). The reach from Maqu to Tangnaihai station is 373 km long and drains an area of 35 924 km2. Most of this reach comprises a steep and incised canyon, with many deeply carved meander bends.

For this study, field investigations of vegetative influences upon bed/bar geomorphic processes were conducted 4 times in the summers of 2011–2014. Particle size distributions of bed and bank materials size were analyzed using a laser particle size analyzer (Mastersizer 2000) and field sieves were used to test 10 samples of river bed and bank materials in each reach. As a supplement, photographs of gravel and cobbles on the bed/bar surface were taken to visually estimate bed particle size. To estimate bedload transport capacity, channel geometry was assessed in the field and from remote sensing images of the branching channel network (Google Earth images from 2005–2014, with a resolution of about 0.24 m).

The best available hydrological data that could be accessed for this study were daily stage–discharge data from Jimai (1964–1985), monthly stage–discharge data from the Maqu (1959–1970), monthly cross section elevation change data from Shangcun station along the Daheba River (1.8 km upstream from its confluence with the Yellow River, 2009–2011), and 2011–2014 field data for the Lanmucuo River (a tributary of the Yellow River in the Maqu–Tangnaihai section, at an elevation of 3400–4200 m a.s.l., for which upstream and mid-catchment reaches have a typically meandering channel, while the downstream reach has a confined bedrock channel). There are no intensive human activities in this area of the Yellow River source zone.

Figure 1b shows the planform morphology of the four channel reaches. Figure 2 shows the channel morphology, pattern of bar types, and bed sediment. Basic channel characteristics of the study reaches are summarized in Table 2.

The Dari reach has a semi-braided and semi-anabranching channel in a wide valley (Fig. 1 – R1 and Fig. 2 – R1). This anabranching–braided transition zone is considered to be semi-stable, with an active channel zone that is around 1 km wide. The braided part of the channel is made up of many small longitudinal and transverse bars, with multiple connected branching channels. In the anabranching part, the large bars/islands have a dense grassland vegetation. Given the extensive width of the active channel zone, annual floods during June–September exert negligible impacts upon these relatively stable surfaces.

The Maqu reach is located in a wide alluvial valley (1.5–3.0 km wide). The dense tree cover of the vegetated islands is indicative of a stable channel configuration (see Fig. 1 – R2 and Fig. 2 – R2). During the flood season, tree trunks are partly submerged into water, but the trees are sufficiently dense to limit erosion. As a result, the anabranching system as a whole is quite stable with high bedload transport capacity.

The Lanmucuo River is a meadow meandering river with nearly 100 % vegetation cover (see Fig. 1 – R3 and Fig. 2 – R3). The root system of riparian grasses induces considerable protection from near-bank erosion. Field investigations from 2011–2014 indicate that the lateral migration induced by cantilever bank failure occurred at a rate of less than 0.2 m yr-1. The gravel-bed channel has a low bedload transport rate in the flood season. In some local sections, mid-channel bars with dense grass coverage have developed at the apex of bends. The whole channel is quite stable, in spite of short-term outer bank failures and long-term meander neck cutoffs.

The Daheba River has incised into the Gonghe–Xinhai sedimentary basin. Severe gully erosion has incised river–lacustrine sediments to a depth of 50–100 m, supplying large volumes of gravel/cobble to the middle and lower Daheba channel, inducing significant bed aggradation and the formation of a braided planform. Alluvial fans at gully outlets not only supply additional sediment, but also push the channel to the opposite side of the valley floor (a big fan is shown near D point in Fig. 1 – R4 and Fig. 2 – R4). As a result, the main branching channels are subjected to frequent and recurrent avulsion. Flows erode new small branching channels during the flood season, but a main channel coexists with several branching channels in the non-flood season. Unstable mid-channel bars are unvegetated other than sparse vegetation coverage (grass and shrubs) on riparian banks. The gravel–cobble bed and high bedload transport rate restrict vegetation establishment and growth, resulting in a typically unstable braided river.

Bank strength induced by sediment material mix and vegetation root networks exerts a critical influence upon the stability of alluvial channels (Eaton and Giles, 2009). Reinforcement of bank strength reinforced by grass, shrub, and tree roots is related to the density, depth, and spatial structure of the root network (Abernethy and Rutherfurd, 2001). Figure 3 shows representative photographs of river banks in the four study reaches. The diverse bank material composition and vegetation cover affect the relative strength of banks and their capacity to resist near-bank flow scour. The river bank in the Dari reach has a two-layer structure, with a 20–30 cm deep soil-root layer (d50 is median size, d50 = 0.02 mm) lying atop a gravel-sand layer (d50 = 6.0 mm) (Fig. 3a). The river bank in the Maqu reach has a dense grass, shrub, and tree cover (Fig. 3b), with no indication of flow scour in the flood season. The study reach along the Lanmucuo River has a typical composite bank sedimentology of a mixed load river (Fig. 3c). An upward-fining sequence is characterized by a basal gravel unit (d50 = 5.5 mm) extending to a 10–30 cm thick silt/sand layer (d50 = 0.03 mm) that is capped by a 10–50 cm thick fine-grained soil-root complex (d50 = 0.02 mm). Conversely, the bank of the middle Daheba River has characteristic deposits of a bedload-dominated river, with non-cohesive gravels and a sparse grass cover (Fig. 3d). Adjacent terraces that are more than 10 m high limit the capacity for channel widening, while actively supplying gravels. Mobile gravel banks influence the braided characteristics of the Daheba River. In summary, bank strength of the four study reaches varies from high to low as follows: Maqu reach, Lanmucuo River, Dari reach, and Daheba River.

Given the lack of observed data of bedload transport rate, bedload transport capacity has been estimated for a rectangular cross section using the theoretical bedload formulae outlined below. Channel flow follows the laws of flow continuity, flow resistance, and sediment transport with flow continuity law taking the form: Q=AV, where Q, A, and V are flow discharge (m3 s-1), channel cross-sectional area (m2), and average flow velocity (m s-1), respectively, A = WH, W is channel width (m), H is water depth (m).

This study adopts the Manning formula to embody the law of flow resistance for uniform alluvial channel flow: V=1nR2/3S1/2, where R is hydraulic radius (m), R = WH/(2H + W), S is flow energy slope, n is Manning's roughness coefficient. In this study, following Chow (1959), n = 0.050 if no vegetation in gravel-bed channels at high stages (i.e., Daheba reach), n = 0.030 in floodplain with short grass (i.e., Lanmucuo River), n = 0.050 in floodplain with scattered brush and heavy weeds (i.e., Dari River), and n = 0.150 in floodplain with dense willows at flood stage (i.e., Maqu reach).

Among numerous bedload formulae, the Meyer-Peter and Müller equation has been extensively and successfully applied (Meyer-Peter and Müller, 1948). The modification developed by Wong and Parker (2006) has been used in this study: Φ=3.97(Ψ-0.0495)3/2, where Φ and Ψ are the dimensionless bedload transport rate per unit channel width and the dimensionless flow shear stress, respectively, that are defined as Φ=qbρs/ρ-1gd503,Ψ=RSρs/ρ-1d50, where qb is the dimensional bedload transport rate per unit channel width (kg s-1 m-1), ρs is the density of sediments transported (kg m-3), ρ is the density of water (kg m-3), g is the acceleration of gravity (m s-2), and d50 is the median sediment size (mm).

Cross section and water depth were measured based on field survey and remote sensing images (see Table 2). Estimated hydraulic parameters and bedload transport capacity for the four reaches, derived using Eqs. (1)–(5), are summarized in Table 3. Note that channel width is effective bankfull width in the flood season, not valley width. The adopted mean grain size is lower than bed sediment size. Results shown here are considered to be approximations, and are analyzed solely in relational rather than absolute terms. Results show that the bedload transport capacity of the four reaches from high to low is as follows: Maqu, Lanmucuo, Daheba and Dari reaches.

This study has outlined the complex interplay between flow and sediment supply in the flood season, and the geomorphic/hydrodynamic role of vegetation cover on the valley floor, as determinants of channel morphodynamics/stability and bedload transport capacity for four alluvial reaches of the Yellow River source zone. Although the elevation of four reaches is different (Dari is 3960 m, Maqu is 3465 m, Lanmucuo River is 3604 m, and Daheba is 2832 m), the precipitation, temperature, and bed sediment size are basically similar (Yu et al., 2014). Nevertheless, vegetation coverage in the four reaches is quite different. The Dari reach (anabranching–braided) has a herb and shrub cover, the Maqu (anabranching) reach has trees, the Lanmucuo River (meandering) has meadow, and the Daheba River (braided) has no vegetation cover. As shown elsewhere, bar morphodynamics vary markedly for differing planform types, with key differences outlined here for braided, anabranching, and meandering channels (cf., Hooke, 1986; Kleinhans, 2010; Kleinhans and van den Berg, 2010; Church and Ferguson, 2015). Bar development and stability reflect the ability of vegetation to trap sediments and stabilize banks, which in turn is directly influenced by flow energy relationships (i.e., these are mutual adjustments; Corenblit et al., 2007; Gurnell et al., 2012; Gurnell, 2014; Osterkamp and Hupp, 2010; Pietsch and Nanson, 2011). In this study, riparian vegetation and its root network are considered to restrict channel width and increase hydraulic efficiency, inducing greater bedload transport capacity in multi-thread channels (Allmendinger et al., 2005; Huang and Nanson, 2007). Islands and floodplains are able to trap more fine-grained sediment in the flood season, enhancing the longer-term (decadal) stability of anabranching channels, as shown by the stable islands of the Maqu reach.

Relative to the passive (resisting) role of vegetation, bedload transport actively affects short-term patterns and rates of bed erosion and deposition. This, in turn, is affected by relationships between the flow regime (especially flood events and formative flows) and the influence of sediment supply upon bedload transport for differing river types (Church and Ferguson, 2015; Dunne et al., 2010). The supply of bed material sediment to an alluvial channel accelerates the growth of longitudinal, transverse, and point bars, thereby enhancing thalweg development and locally increasing flow velocity. Non-equilibrium between sediment supply and transport induces local channel instability, accentuating either bed erosion or deposition (Jansen and Nanson, 2010; Nanson and Huang, 2008). In this study, a channel stability gradient accords with both sediment movement and vegetation cover, wherein bedload transport capacity (a function of bed slope, hydraulic geometry, and sediment particle size) is related to the influence of riparian vegetation upon channel geometry/planform. We contend that the differing vegetation cover and planform response reflect the delicate balance between erosion and deposition on the channel bed and bank as influenced by bedload sediment supply in the flood season. Only when the bedload transport capacity is equivalent or greater than sediment supply does vegetation act as a key determinant of channel stability.