Previous flume-based research on braided channels has revealed four classic mechanisms that produce braiding: central bar development, chute cutoff, lobe dissection, and transverse bar conversion. The importance of these braiding mechanisms relative to other morphodynamic mechanisms in shaping braided rivers has not yet been investigated in the field. Here we exploit repeat topographic surveys of the braided River Feshie (UK) to explore the morphodynamic signatures of different mechanisms of change in sediment storage. Our results indicate that, when combined, the four classic braiding mechanisms do indeed account for the majority of volumetric change in storage in the study reach (61% total). Chute cutoff, traditionally thought of as an erosional braiding mechanism, appears to be the most common braiding mechanism in the study river, but was more the result of deposition during the construction of diagonal bars than it was the erosion of the chute. Three of the four classic mechanisms appeared to be largely net aggradational in nature, whereas secondary mechanisms (including bank erosion, channel incision, and bar sculpting) were primarily net erosional. Although the role of readily erodible banks in facilitating braiding is often conceptualized, we show that bank erosion is as or more important a mechanism in changes in sediment storage than most of the braiding mechanisms, and is the most important “secondary” mechanism (17% of total change). The results of this study provide one of the first field tests of the relative importance of braiding mechanisms observed in flume settings.
 Of all the planforms and river styles alluvial rivers may exhibit, braided rivers are the most dynamic [Brierley and Fryirs, 2005]. They owe this dynamism to their abundant bedload and readily erodible banks, which results in a high frequency of avulsions and complex flow patterns converging and diverging around active central bars [Ashmore, 1982; Charlton, 2007; Chew and Ashmore, 2001; Jerolmack and Mohrig, 2007; Miall, 1977]. Indeed, the maintenance of braiding is partly dependent on this very dynamism, as high rates of channel turnover inhibit the growth of bar-top vegetation which may otherwise stabilize the bed and banks through root reinforcement and increased flow resistance [Hicks et al., 2007; Paola, 2001; Tal and Paola, 2007].
 The continuously shifting network of channels, splitting at diffluences, and converging at confluences, give rise to a distinctive set of three-dimensional morphologies. Characteristic forms include a range of active bars, including mid-channel, bank-attached, and compound bars, with locally deep scour holes formed by high rates of sediment transport at confluences [Ashmore, 1982; Bridge, 1993]. The formation of multiple mid-channel bars (i.e., braiding) requires large width-to-depth ratios, which can only be accommodated by readily erodible banks [Ferguson, 1987; Millar, 2000; Zubik and Fraley, 1988].
 Existing conceptual models of braiding emphasize that there is no single process that leads to the division of flow and the evolution of mid-channel bars [Bridge, 1993; Ferguson, 1993; Leddy et al., 1993]. Rather, there are a suite of depositional (e.g., bar building) and erosional (e.g., channel cutting and bar dissection) processes that operate over time to develop and maintain the multi-thread character of these systems [Bridge, 1993; Ferguson, 1993; Kleinhans, 2010]. Much of our understanding of braided river dynamics comes from flume [e.g., Ashmore, 1982, 1991; Ashworth, 1996; Germanoski and Schumm, 1993] and, to a lesser extent, numerical modeling studies [Hicks et al., 2009; Murray and Paola, 1994; Nicholas et al., 2006]. In Ashmore's  seminal work on “how do gravel-bed rivers braid?” and Ferguson's  review of “understanding braiding processes in gravel-bed rivers”, four principal mechanisms of channel change and evolution emerge: central bar development, transverse bar conversion, chute cutoff of point bars, and lobe dissection. In addition, there is a large body of research that has articulated conceptually and documented qualitatively, the mechanisms that underpin braided river morphodynamics [Best and Bristow, 1993; Bridge, 1993; Sambrook-Smith et al., 2006].
 Despite the collective significance of this body of literature, little has been done to quantify the relative importance of specific braiding mechanisms in terms of change in sediment storage and their contribution to the dynamics of the river. Recent advances in repeat topographic surveying techniques and geomorphic change detection [Hicks et al., 2007; Milan et al., 2011; Williams et al., 2011] now permit such a quantitative analysis of braided river morphodynamics, but to date, have not yet been adequately exploited for this purpose.
 The goals of this paper are to (a) classify the morphodynamic signature of observed channel changes in a braided river, and (b) test the perceived dominance of the “classic braiding mechanisms” (as defined by Ashmore  and Ferguson ) to the overall volumetric changes in sediment storage in an actively braided, gravel-bed river. To accomplish this, we use high-resolution digital elevation models (DEMs) derived from repeat topographic surveys of a braided system over a 5-year period. We take the “dynamism” of the channel to be expressed geomorphically by turnover due to erosion and deposition, which is manifested topographically by changes in sediment storage. Specifically, we seek to quantify, for potentially the first time in the field, the relative contribution of different braiding mechanisms to the storage dynamics of sediment within a braided gravel-bed river. Further, we explain ways in which the relative importance of different braiding mechanisms are controlled by variations in discharge. The significance of this contribution is partly conceptual and partly methodological. We provide empirical insight into how much work (i.e., net displacement of sediment in storage due to fluvial forces) is done by these key braiding mechanisms and how complete current conceptual models of braiding actually are. Methodologically, we lay out a pragmatic form of empirical analysis others can apply to the expanding collection of repeat topographic time series data on their own study rivers.
2 The Glen Feshie Study Site
2.1 Physical Setting
 The River Feshie is a braided gravel-bed river well known for its extensive gravel bar deposits [Ferguson and Werritty, 1983]. The Feshie has been the focus of numerous studies of bar development [Ferguson and Werritty, 1983], sediment transport [Ferguson and Ashworth, 1992], geomorphic change detection [Brasington et al., 2000, 2003], historical planform change [Werritty and Ferguson, 1980; Werritty and McEwen, 1993], and hydrology [Soulsby et al., 2006]. Situated in the heart of the Scottish Highlands, the Feshie has three actively braided reaches: one at its confluence with the River Spey and two reaches upstream, the most active of which is a 3 km long reach near Glenfeshie Lodge that encompasses the ~ 1 km study reach in this paper (Figure 1).
 Glen Feshie itself is a glacial trough, which was deglaciated roughly 13,000 years B.P. [Gilvear et al., 2000] and is now flanked by fluvio-glacial outwash terraces [Robertson-Rintoul, 1986]. The flow-regime is unregulated and flashy, reflecting the steep and rugged terrain of the catchment [Soulsby et al., 2006]. This catchment is underlain by Moinian schist, which dominates the coarse bedload, and a small proportion of granite of the Cairngorm batholith. Average valley gradient is ~0.01 and the median surface grain size declines downstream through this braided reach from 110 to 35 mm, with significant grain size variability between wetted channels and bar tops. In this paper, we focus on a ~700 m long sub-reach at Glen Feshie, which was topographically surveyed annually during summer low flows between 2000–2007 [Rumsby et al., 2008; Wheaton, 2008]. Data are presented for a 5 year period from 2003 to 2007, from which four analysis periods (epochs of change) are derived.
 Since at least the mid 1800s, this reach of the Feshie has maintained a persistent braided character [Figure 2f; see also Rumsby et al., 2001; Werritty and Ferguson, 1980; Werritty and Brazier, 1991]. As documented by aerial photos dating back to 1946 and UK Ordnance Survey maps dating back to 1869, the Feshie has maintained a braiding index [Robertson-Rintoul and Richards, 1993] between 2 and 4 and a braiding intensity between 2 and 5 [Egozi and Ashmore, 2009]. The braiding index was calculated by counting the number of active channels in each of five cross sections and averaging them, whereas the braiding intensity was calculated by dividing the length of all braid channels by the length of the reach as per Mosley .
2.2 Hydrologic Record
 A long-term flow record does not exist at the study reach, but Brasington and Cox (personal communication) and Ferguson and Werritty  have gauged the study reach for short periods. A more robust record of continuous, 15 min resolution discharge dating back to 1992 is available from the Scottish Environmental Protection Agency (SEPA) at Feshie Bridge, 11 km downstream of the Glen Feshie study site. The maximum recorded daily discharge at Feshie Bridge was measured on 24 December 1999 at 121.5 m3s−1 with an instantaneous peak of 222.5 m3s−1 from the same day. Ferguson and Werritty  reported that the same gauging station was also maintained by the Department of Agriculture and Fisheries for Scotland from 1951 to 1974. The summary statistics for peak flows and mean discharge [Ferguson and Werritty, 1983] are very similar to the 1992–present record (Figure 2d).
 In all epochs, some portion of the braidplain in the study reach was likely inundated, but both the frequency and severity of those floods is significantly less in the 2004–2003 and 2006–2005 epochs than during either the 2005–2004 or 2007–2006 epochs. Ferguson and Werritty  reported that “bankfull flows” in the study reach are somewhere in the 20–30 m3s−1 range, which corresponds to a discharge of 28–42 m3s−1 at Feshie Bridge using a downscaling coefficient we developed empirically of 0.71. For this short duration record, there are essentially two drier epochs (2004–2003 and 2007–2006) and two wetter epochs (2005–2004 and 2006–2005) with 2007–2006 significantly wetter. Note, that we report all epochs as “new” minus (−) “old” (e.g., 2004–2003), as this is consistent with the order the elevations in each survey has to be subtracted from each other for erosion to be negative and deposition to be positive. Interestingly, none of these floods even exceed the 2 year recurrence interval (peak on 1 December 2006 is 125.3 m3s−1 at Feshie Bridge) and our study period does not capture any rarer high magnitude floods. For comparison, flows exceeded c. 140 m3s−1 at Feshie Bridge during our study period once versus the three times during the 1976–1981 Ferguson and Werritty  study period. Ferguson and Werritty  estimated that overbank flooding in the study reach during the first 3 years of their study period occurred sometime between 16 and 51 times (depending on the threshold assumed), and a similar analysis we performed shows that overbank flooding occurred in the study reach between 38 and 79 times in our study (Table 1).
Table 1. Flow Statistics From Feshie Bridge
Flow Statistics for Study Period
Mean Flow (m3s−1)
Peak Mean Daily flow (m3s−1)
Peak Instantaneous Flow (m3s−1)
2nd high flow (m3s−1)
3rd high flow (m3s−1)
4th high flow (m3s−1)
POT (28 m3s−1) Low Bankfull
POT (42 m3s−1) High Bankfull
POT (127 m3s−1) 2 year RI
Total Flow > Bankfull (m3)
Flow Recurrence Intervals
Return Period (years)
 For inferring the relative number and magnitude of floods during our study epochs, instantaneous peak flows were used. The top part of Table 1 includes mean flows, ranked peak flows for each epoch as well as peaks over threshold that counted the number of storms in each year over 28 m3s−1 (a low bankfull estimate), 42 m3s−1 (a high bankfull estimate) and 127 m3s−1 (the 2 year recurrence interval flow). The bottom row of Table 1 shows the standard flow recurrence intervals (using a Log-Pearson Type III analysis of maximum daily flows). For context, the average instantaneous peak annual discharge is 135.8 m3s−1, and the 2 year recurrence interval flood is 127.0 m3s−1over 19 years of record. As our survey frequency is annual, the changes captured between topographic surveys integrate the effects of a range of competent flow events with varying magnitudes and frequencies. From Figure 3, it is clear that the 2007–2006 and 2005–2004 epochs boasted significantly more potentially competent floods as well as notably higher peak flows. Based on peak instantaneous discharges, all four epochs were less than the long-term average, with the 2004–2003 and 2005–2004 epochs experiencing a nearly identical range of flows.
 To explore how braiding mechanisms are expressed through changes in sediment storage over a geomorphically active (but by no means extreme) 5 year period in the braided river Feshie, we relied on high-resolution repeat topography to produce DEMs of difference (DoDs). Here we first review the acquisition of topographic data and DEM construction. We then describe a simple topology applied to those DEMs to record an expression of braiding behavior. To infer the braiding mechanisms from DoDs, we needed to distinguish those changes resulting from geomorphic processes from those changes arising simply due to noise in our DEM data. Therefore, we also briefly review a methodology outlined by Wheaton et al. [2010a] that robustly addresses this geomorphic change detection problem. Finally, we describe the analysis and interpretation of DoDs used to track braiding mechanisms as expressed through changes in sediment storage with spatial segregation or masking [Wheaton, 2008; Wheaton et al., 2010b].
3.1 Topographic Surveys, DEMs and Channel Node Topology
 Repeat topographic surveys were conducted with between two and five Leica System 1200 rtkGPS rovers operating simultaneously and a base station occupying a benchmark set on a control network we established using the British National Grid projection. Raw topographic data were filtered and points with positional errors > 5 cm were removed. Concurrent 1 m resolution DEMs were derived from triangulated irregular networks (TINs) of the processed rtkGPS topographic surveys of each site (between 30,000 and 50,000 points per survey; average density 0.3 points/m2). The DEMs used here are the same reported in Wheaton et al. [2010a], and they provide a record of geomorphic change (behavior) in a series of individual DEMs. Detrended DEMs were used to help differentiate areas accessed by various stage flows, and were derived by normalizing raw elevation data with respect to a smoothly sloping plane set to the surrounding valley slope and elevated to the mean valley elevation.
 From the detrended DEMs, we derived an inventory of braid channel nodes using Ferguson's [1993, pp. 75–76, Figure 2] topology. Nodes inventoried included (i) confluence junctions between two braid channels, (ii) diffluences or bifurcations where a single channel splits into two, and (iii) channel heads. Ferguson  showed that channel heads can form either by head-cutting/incision from flows over a bar top or braid plain, or from partial choking of an anabranch. We independently identified and digitized (as points) all braid nodes for each of the five DEMs. In simple terms, braiding is maintained when the number of confluences is roughly matched by the number of channel heads and/or bifurcations [Kleinhans et al., 2012]. For example, a channel that only has confluences from tributaries is not braided; a channel that has more bifurcations than confluences is a distributary system, like a delta or fan [Jerolmack and Mohrig, 2007; Kleinhans et al., 2012]. We classified all channel heads, diffluences and confluences (high and low stage) that were resolved within a 1 m resolution DEM (i.e., generally > 2 m in width). We used this topology to assess the persistence of braiding conditions and inform our later classification of braiding mechanisms.
3.2 Quantifying Geomorphic Changes—DoD
 The Geomorphic Change Detection 5.0 (GCD) software (http://gcd.joewheaton.org) was used to compute the difference between sequential DEMs and conduct a spatially variable uncertainty analysis to robustly distinguish real changes from noise [Wheaton et al., 2010a]. Briefly, this approach employed a spatially variable, probabilistic, and minimum Level of Detection (minLoD) to account for errors propagated from the individual DEMs into the DoD (Figure 4). This was achieved by first estimating errors in each DEM on a cell-by-cell basis with a fuzzy inference system (FIS). The FIS accounts for the tradeoff between the completeness of sampling coverage (point density used as proxy) and topographic complexity (slope used as proxy) while keeping track of instrument-reported 3D GPS point quality. The FIS estimated a spatially distributed metric of surface reliability expressed as a vertical elevation error for each DEM. Basic error propagation was then used to incorporate errors from each concurrent 1 m resolution DEM into the DoD calculations. This propagated error term was used to define the probability that elevation changes measured between two successive DEMs were real by calculating a t score to compare the DoD differences against the minLoD defined by the propagated error [see Brasington et al., 2003; Lane et al., 2003]. An additional probability was estimated based on the spatial coherence of erosion and deposition. This was approximated using a moving window majority-filter determined over a 15 m interval (chosen to reflect average channel width). In essence, the probability of DoD change being real was increased where it was spatially coherent (i.e., if an erosion cell was surrounded by other erosion cells), and decreased where it was fragmented (e.g., a checkerboard of erosion and deposition). These two sets of probabilities of observed change—one reflecting the vertical magnitude (FIS), the other the spatial context (spatial coherence filter)—were combined using Bayes' Theorem. The DoDs were then thresholded at a 95% confidence interval, so that only changes estimated as having 95% or higher probability of being real were included in the budget.
 The volumetric change in storage is calculated by multiplying all elevation changes in the DoD by the cell area and accounting separately for erosion and deposition areas. The quantities described below can all be calculated for any area of interest: for example, the entire reach or just a defined sub-area of the reach. The net change in sediment storage (ΔVDoD) is defined as the sum of all the deposition volumes minus the sum of all the erosion volumes:
ΔVDoD over some time period Δt (epoch) represents the sediment budget (or expression of conservation of mass, assuming that ΔVDoD , and all have equal bulk densities):
where and are the volumetric bedload fluxes into and out of the control volume (typically a study reach). By contrast to the net change in sediment storage (ΔVDoD), we also have the total bulk change in sediment storage (∑ VDoD), which is simply the sum of the erosion and deposition change in storage volumes as opposed to the difference. In this study, we do not use bedload flux data and therefore only report net and total changes in sediment storage. However, the summed volumes of erosion (∑ VErosion) and deposition (∑ VDeposition) both spatially integrate the net changes in sediment storage over the course of the epoch. In terms of the alluvial sediment store that makes up the valley fill, ∑ VErosion represents withdrawals from storage whereas ∑ VDeposition represents deposits to that storage.
 Volumetric percentages of the total bulk change in sediment storage (∑ VDoD), are calculated separately for the erosion or deposition volumes of interest divided by ∑ VDoD (i.e., erosion plus deposition). The net change in sediment storage (ΔVDoD) is the difference of erosion and deposition volumes (negative when erosional, positive when aggradational). The percent imbalance is the departure from a condition of 50-50% split between erosion and deposition, and is calculated as the volumetric percentage of the bigger of erosion and deposition terms, minus %. For example, an imbalance of −20% indicates that total erosion for a given period represented 70% of the topographic change in the study reach.
 The methods described above are useful for establishing confidence that “real changes” were reliably being distinguished from noise. However, here we still report our uncertainty associated with the volumetric estimates of erosion and deposition as ± one standard deviation of our volumetric error. This is approximated by using the propagated spatially variable error estimates from the FIS error surfaces and converting these to volumetric errors. The thresholded DoD is treated as the mean estimate (as in a probability distribution function), and the propagated errors (sum of squares in quadrature) are an approximation of a standard deviation.
3.3 Classification of Morphodynamic Signatures
 Quantitative DoD segregation helps exploit topographic time series in a manner that facilitates a more mechanistic explanation of the signatures resulting from change in sediment storage dynamics [Wheaton et al., 2010b]. Here we consider a morphodynamic signature to be a distinct mechanism of erosion and/or deposition that leads to a consistent morphological response (e.g., creation of a specific geomorphic unit). DoD segregation involves the creation of spatial masks (i.e., polygons) of different categories to segregate the reach-scale DoDs (thresholded at a 95% confidence interval) into separate totals for ∑ VDoD, ΔVDoD, ∑ VErosion and ∑ VDeposition volumes of statistically significant storage changes for each polygon area. Using a mutually exclusive classification of the entire reach ensures that storage changes are not counted multiple times. We were primarily interested in quantifying the contribution of different braiding mechanisms to the dynamism of our study reach, but we also were interested in the importance of other morphodynamic signatures of change in driving the behavior of the Feshie reach over the study period. Mutually exclusive polygons were digitized around individual erosional and depositional units on the thresholded DoDs and then classified manually using a mix of the DEMs, aerial photographs, bar classifications (e.g., Figure 5), and field observations from the beginning and end of each epoch. Below, we describe each morphodynamic signature for which we derived polygons and the process of inferring them from DoDs.
3.3.1 Braiding Mechanisms
 All four of the mechanisms identified by Ashmore  and Ferguson  (see section 1.0 above) actually result from a mix of erosion and deposition, but central bar development and transverse bar conversion are predominantly the result of deposition creating mid-channel bars; whereas chute cutoff and lobe dissection are typically considered primarily the result of erosion converting a bank-attached bar to a mid-channel bar and multiplying the number of mid-channel bars respectively. All four processes involve the formation and/or reshaping of mid-channel bars and as such it was helpful to map all bars on each DEM to help distinguish processes.
 Central bar development results in the formation of a roughly symmetrical, elongated medial bar without an avalanche face [Ferguson, 1993]. These were easily identified on the detrended DEMs based on a tear-drop bar shape and position. We examined DoDs for concentrated zones of mid-channel deposition resulting in either new or expanded central bars.
 The building blocks of most bar forms are transverse unit bars, or thin sheets of sand, gravel or cobble—one to three grains thick [Rice et al., 2009]. These lobate unit bars combine to create larger bar forms and often are distinguished by their tear-drop and presence of avalanche slip faces at their fronts [Ashmore, 1991]. Transverse bar conversion is a process by which these transverse unit bars accumulate in the middle of the channel and force division of flow around themselves. These lobes were easily identified from the DEMs based on their position downstream of chutes and/or confluence pools and their bulbous shape and steep faced fronts [Ferguson, 1993]. We examined DoDs for expanses of deposition downstream of such features.
 Chute cutoff occurs where bank-attached alternate bars or point bars experience headcut incision in the form of a chute as flow takes a short-cut over the bar top, along the channel margin [Ashmore, 1991]. The result is that the detachment of the bar from the bank and a mid-channel diagonal bar forms. We examined sequences of DEMs for bank-attached bars becoming diagonal bars and examined the DoDs for evidence of chute erosion along a channel margin and deposition or expansion of the diagonal bar. This is a common mechanism in the Feshie and was well documented and described by Ferguson and Werritty .
 Lobe dissection involves erosion of multiple chutes (as opposed to a single chute) across an existing lobate bar surface [Ferguson, 1993]. We looked on the DEMs for emergence of chutes across lobate bar surfaces, and multiple, narrow chutes of erosion in the DoDs.
3.3.2 Other Morphodynamic Signatures
 In addition to these braiding mechanisms, we also saw evidence of bank erosion, bar edge trimming, channel incision, confluence pool scour, overbank gravel/cobble sheets, and lateral (bank-attached) bar development. The braided rivers literature provides passing mention of each of these mechanisms, but they are not considered primary mechanisms of braiding. As mentioned earlier, readily erodible banks are a prerequisite for the high width-to-depth ratios that facilitate braiding, but of themselves are not a mechanism of braiding. We were interested in the importance of these lesser recognized processes in contributing to the net changes in sediment storage (both ΔVDoD and ∑ VDoD). For example, bank erosion and bar edge trimming can produce considerable volumes of locally sourced sediment, which might be a key contributor to the extensive development of mid-channel bars (e.g., central bar development, transverse bar conversion).
 Both bank erosion and bar edge trimming appear as narrow, elongated bands of erosion oriented streamwise in the DoD. For bank erosion, these reflect lateral erosion into either the banks of less active surfaces on the braid plain (e.g., islands) or terraces on the edge of the braid plain; whereas for bar edge trimming, these represent lateral erosion along the margins of active bar surfaces. Both tend to be associated with topographic steering of the flow into these features and tend to result in planform convexities on the edges of banks or bars. The relative height of the surface being eroded (reflected in magnitude of erosion) was also used to help distinguish between the two.
 Confluence pool scour, where anabranches rejoin and convergent flow forces the scour of a pool, is a hallmark of braided rivers [Ashmore and Parker, 1983]. We looked for discrete zones of scour at confluences resulting in the formation, maintenance or growth of a pool.
 We identified channel incision as the down-cutting of an entire existing anabranch, typically from one diffluence to the next confluence. Such channel incision was relatively low magnitude (i.e., < 50 cm) and generally represented an adjustment to a new local base-level being set at the downstream junction, which typically shortened the anabranch length and steepened its profile.
 Overbank gravel or cobble sheets delineate areas of overbank deposition onto braidplain surfaces as described by Ferguson and Werritty  characterized by at least decimeter-thick gravel/cobble deposits burying heather, moss and grass vegetation on the braidplain. These bedload sheets are similar to unit bars but only occur in larger braidplain inundating floods, are relegated to the braidplain, and generally produce much larger more aerially extensive deposits. These deposits are differentiated from regular overbank deposition of finer suspended load material (typical floodplain deposits) that are often below minimum levels of detection. We classified potential minor floodplain deposition and micro-topographic scour of the braid plain in a “questionable or unresolved change” category.
 In individual anabranches of the braided network, the anabranch channels start off behaving like low-sinuosity single-thread channels and begin to form bank-attached alternate bars and point bars on inside bends. There, we see lateral (bank-attached) bar development, which is easily differentiated from the other forms of bar development included in the braiding mechanisms and described above. Such lateral bar development is more common in sub-bankfull floods, which are confined to the anabranches and eventually gives way in larger floods to chute cutoff braiding mechanisms and conversion of bank-attached bars to mid-channel bars.
 Although not explicitly included as a primary “braiding mechanism”, much of the braided river literature discusses the importance of avulsions in braided rivers [Leddy et al., 1993]. We identified avulsions by the disappearance of a channel diffluence or bifurcation and then used the DoDs to track what mechanisms led up to the avulsion and classify the avulsion type.
4 Results and Interpretations
4.1 DEMs and Topological Classification of Braided Channel Nodes
 The five DEMs used for change detection are shown in the top of Figure 6. Overlaid on top of the DEMs are the locations of braid channel nodes including channel heads, bifurcations and confluence junctions. The nodes include both active and relict features preserved in the detrended topography. The number of confluences varies through time between 82 and 124 with an average of 105. However, only 3–5 of these are near-symmetric confluences of sub-equal larger channels with year-round flow. The majority of confluences (75%) are paired with small channel heads leaving only 25% of those associated with bifurcations of channels such as chute cutoffs around diagonal bars. In epochs with a greater duration of braidplain inundating flows (e.g., 2004 and 2007; Table 1) the overall number of channel nodes decreases, potentially reflecting simplification of the braid plain network during larger floods.
4.2 Reach Scale Change in Sediment Storage
 The year-to-year changes in sediment storage (∑ VErosion and ∑ VDeposition) are shown as pie charts in Figures 7a–7d, with the volumetric erosion, deposition, and net change values reported in Table 2. Figures 7a–7d show the thresholded DoDs from which these sediment storage budgets were derived, as well as the raw and thresholded (post uncertainty analysis) volumetric elevation change distributions. The imbalance in the ΔVDoD was consistently calculated as degradational, ranging between −5% in 2004–2005 and −16% in 2003–2004. The total volumetric imbalance was −7%, giving an average annual net loss of 1064 ± 2767 m3 of sediment and total net loss of 4640 ± 11,057 m3 over the entire 5 year study period.
Table 2. Summary of Volumetric and Areal Storage Change by Erosion, Deposition, Net Change (Deposition-Erosion) and Total Change for Each Epoch
Vol Error as %
Vol Error as %
Volume Percentage Imbalance
Areal Percentage Imbalance
Areal % of Total Reach
Epoch values are totaled and averaged in the bottom rows.
 The large error bars on the net estimate (Figure 8), are the result of rather conservative volumetric estimates for erosion and deposition, which are then propagated into the difference calculation. Over the whole study period, the total erosion volume estimate was 18,612 ± 9777 m3 and the total deposition volume was 14,356 ± 4989 m3. The ±65% average volumetric error for erosion is notably higher than the ±40% average volumetric error for deposition; due to the fact that much of the erosion took place in steeper areas where lower survey point density may influence the quality of topographic survey points (i.e., where there is increased noise in the survey data). Given this uncertainty, we can only conclude that the sign of net change (that is, the balance between erosion and deposition) is indeterminate and that the reach is most likely roughly in equilibrium.
 The volumetric elevation change distributions in Figure 7 all have bimodal distributions with the erosional fractions (red left-hand peak) consistently larger the depositional fractions (blue right-hand peak). The grey, original unthresholded distribution shows the magnitude of volume which was thresholded out because it did not meet the 95% confidence criteria (§3.2). This large excluded area conceptually illustrates why the absolute volumetric estimates can be so uncertain despite relatively high spatial confidence (i.e., >95% probability) regarding the location and direction of geomorphic change. However, the percentage erosion and deposition volumes (pie charts) are fairly insensitive in this case to the exclusion/thresholding of that volume [see Wheaton et al., 2010a, Figure 12]. When absolute values are used, both the erosional and depositional distributions are right-skewed and both are long-tailed. Both the erosional and depositional fractions of the elevation change distributions exhibit relatively narrow, low magnitude peaks: typically 20–25 cm for deposition; 35–40 cm for erosion.
4.3 Morphodynamic Signatures of Change in Sediment Storage
 Figures 9a–9d shows the maps of the DoD segregation in terms of our classification of morphodynamic signatures. These include primary braiding mechanisms (central bar development, transverse bar conversion, chute cutoff and lobe dissection) as well as six other mechanisms of change and a questionable or unresolved changes category. The pie charts in Figures 9e–9h show the resulting percentages of total volumetric change in storage for each epoch broken down by each mechanism. Figure 10a uses the same results but plots their absolute magnitudes epoch-by-epoch to highlight both variability, in absolute amounts of work done, as well as consistency in the dominance of specific mechanisms. In the next two subsections, we quantify the relative importance of each signature in contributing to the reach dynamics.
 The four main braiding mechanisms accounted for 61% of the total volumetric change in the reach (Figure 10b), but varied between as little as 46% in 2007–2006 and as much as 82% in 2005–2004. Bank erosion was found to be as or more important a mechanism in changes in sediment storage as most of the individual braiding mechanisms, and emerged as the most important “secondary” mechanism (17% of total change). In every epoch, the four primary braiding mechanisms constituted net aggradation with an average ΔVDoD of +1127 m3 and total ΔVDoD of +4506 m3, which equates to average 11% aggradational imbalance. Figure 10a shows that central bar development, transverse bar conversion and chute cutoff were consistently the most important depositional mechanisms of change, but lobe dissection and chute cutoff were not always the most prevalent erosion mechanisms, falling behind bank erosion, channel incision and confluence pool scour in low water years (Figure 9). Figure 11 illustrates the signatures of volumetric elevation changes for each mechanism in a characteristic low-water epoch (2004–2003) and characteristic high-water epoch (2007–2006).
 Given that braiding mechanisms constitute a majority (61%) of the sediment dynamics of the reach and are consistently aggradational (11% aggradational imbalance), either upstream external sources of sediment or other erosional mechanisms within the reach must be providing the excess sediment supply. Our best estimate for the overall reach budgets are on average a 9% degradational imbalance (Figure 7). To make up this 20% volumetric discrepancy, the other mechanisms of change must be net degradational. Below, we describe the morphodynamic signatures found in the Feshie for each of the mechanisms of change.
4.3.1 Individual Mechanisms of Change
 Chute cutoff was the most prevalent braiding mechanism both on average and in the first three epochs, accounting for 23% of the total volumetric change (∑ VDoD) over the study period (more than any other mechanism) and as much as 40% in 2006–2005. Ferguson  casts chute cutoff and lobe dissection as a primarily erosional mechanism that leads to the conversion of a bank-attached bar (either point bar or alternate bar) to a mid-channel bar (diagonal bar). We found here that although chute cutoff was a mix of erosion (carving the chute along the bank) and deposition (building up the new diagonal bar) as described by Ferguson and Werritty , in three of the four epochs, the deposition outpaced erosion. Figures 11c and 11g show the shape of elevation change distributions for chute cutoffs in a relatively low water epoch (2004–2003) and relatively high water epoch (2007–2006). These elevation change distributions show the erosional fraction in red with a more dispersed distribution reflecting the range of the bar thicknesses which the chute cutoff channel dissected, whereas the depositional fraction exhibits a more concentrated peak around lower magnitude (c. 25 cm) deposition and growth of the diagonal bar surface.
 By contrast, lobe dissection shows a consistently net degradational ECD (Figures 11a and 11e) with a dispersed distribution, again reflecting the range of bar thicknesses of the lobate transverse bars that the chutes carved through. Lobe dissection accounted for 13% of the total volumetric change (∑ VDoD) in the reach (Figure 10b), but varied between as little as 5% in 2007–2006 and as much as 24% in both 2005–2004 and 2006–2005 (Figures 9e–9h).
 Both central bar development and transverse bar conversion show consistent predominance of deposition (Figure 10a), but in most years, these are not as prevalent forms of deposition as diagonal bar development (i.e., chute cutoff). The notable exception to this was the 2007–2006 epoch (Figure 9h), in which transverse bar conversion accounted for 27% of total volumetric change (∑ VDoD) for the epoch. Central bar development accounted for 6% of the total volumetric change (∑ VDoD) over the whole study period (Figure 10b), and varied between as little as 0% in 2006–2005 and as much as 9% in 2005–2004 (Figures 9e–9h) with a yearly average of 5%. Change due to transverse bar conversion was consistently larger and accounted for 19% of the total volumetric change (∑ VDoD) in the reach over the entire study (Figure 10b), constituted as little as 9% in 2005–2004 (Figures 9e–9h) and averaged 15% each year.
 Bank erosion accounted for 17% of the total volumetric change over the entire study period, ranging from 24% in 2007–2006 to 3% in 2006–2005, with an annual average of 13%. Bank erosion was almost exclusively associated with flow forcing and subsequent channel widening around mid-channel bars, while relatively rare lateral bar development did not produce any detectable bank erosion over the whole study period. These contrast with single-thread channels where lateral bars (e.g., point-bars) are associated with forcing pools and bank erosion on the outer bank of meanders. In the Feshie study reach, bank erosion was forced primarily by lobate bars (i.e., transverse bar conversion) in 2006–2005 and 2007–2006, diagonal bars in 2004–2003 (i.e., chute cutoff), and longitudinal bars in 2004–2005 (i.e., central bar development).
 Channel incision was only present in isolated anabranches in 2004–2003 and 2007–2006, with no evidence for systemic downcutting of all the low flow anabranches over the study period. When channel incision did take place, it produced modest quantities of sediment accounting for 5% of the total volumetric change (∑ VDoD) in the reach over the whole study period (Figure 10b), 13% in 2004–2003 and 8% in 2007–2006 (Figures 9e–9h). The location of channel incision is consistently located in portions of the main-flow channel that have recently formed from avulsions connecting a higher upstream channel to a lower paleo channel, which then reactivates as the main channel. These newly formed main channels carve new paths that make their way to the paleo-channels along over-steepened routes. Since the path of the new channels is generally shorter in length than the old channel, they are over-steepened, and these channels adjust by incising over their short distances to the new base-level control imposed by the paleo-channel to which they connect. There is no evidence of channel incision persisting in one location for more than a year, suggesting that the adjustment is able to take place rapidly (i.e., on an event or a few event timescales).
 Among the defining features of braided rivers are the presence of anabranches with confluences and diffluences [Ashmore, 1991]. The confluences are commonly the locations of scour pools, which Ashmore  and Ashworth  have shown are often the source zone for either central bar development or transverse bar conversion in the bar immediately downstream of the scour pool. Confluence pool scour accounted for 6% of the total volumetric change (∑ VDoD) in the reach over the whole study period (Figure 10b), ranging between <1% in 2005–2004 and 11% in 2007–2006 (Figures 9e–9h) with an annual average of 6%. With the notable exception of the most prominent confluence scour pool in the middle of the reach, most of the volume of confluence pool scour came from either the creation of a new scour pool or discrete episodes of accentuation of an existing pool. Some of the persistent confluence pools (Figure 6) seem to be defined more by their lack of major changes (persistence due to lack of change), as they are zones with adequate convergent flow so as to consistently pass the sediment delivered to them and deposit it downstream at the next bar where flows paths diverge [Ashworth, 1996].
 Overbank gravel or cobble sheets on the floodplain/braidplain accounted for 4% of the total volumetric change (∑ VDoD) in the reach over the whole study period (Figure 10b). Overbank gravel or cobble sheets only formed in the years with the highest magnitude floods, accounting for only 1% in 2005–2004 and 8% in 2007–2006 of the total epoch volumetric change (Figures 9f and 9h). Despite their relatively minor volumetric contribution to the change in sediment storage, these sheets blanket very large expanses of the reach (Figures 9b and 9d), leaving a disproportionate impression of their significance from an aerial perspective. Ferguson and Werritty  also reported seeing occasional decimeter thick gravel/cobble sheets on the Feshie, but did not quantify their relative importance. These sheets do not occur every year or every flood, but when large floods occur, gravel/cobble sheets are an important sink of sediment that typically bury the annual grasses, heather shrubs and mosses attempting to colonize the braid plain (see also the expanses of exposed gravels in overbank bars on aerial photos in Figure 2).
 The six “other” mechanisms of change accounted for 35% of the total volumetric change in the reach (Figure 10b), but varied between as little as 7% in 2006–2005 and as much as 52% in 2007–2006. The questionable or unresolved changes in the reach made up the other unaccounted 4% of the total over the study period, accounting for on average 8% of the changes each year. More precise survey methods such as ground-based LiDaR [Vericat et al., 2012; Williams et al., 2011] could better constrain the unresolved fraction of the budget.
 In every epoch, the six other morphodynamic signatures constituted net degradation with an average ΔVDoD of −2060 m3 and total ΔVDoD of −8243 m3, which equates to, on average, a 36% degradational imbalance. This strong tendency towards net degradation amongst the other mechanisms of change helps explain the missing sediment supply that leads to the net aggradation amongst the primary braiding mechanisms. Given the strength of this discrepancy and previous work in gravel-bed rivers, suggesting sediment travel distances are often on the order of the bar/pool spacing [Pyrce and Ashmore, 2003], it seems that these other mechanisms may be providing much of the sediment supply locally to facilitate the creation of major mid-channel bars and maintain braiding. Figure 10a shows that other than occasional overbank gravel or cobble sheets in large floods, the confluence pool scour, bank erosion, channel incision, and bar edge trimming mechanisms are indeed comprised of predominantly erosion. Overall, bank erosion contributes the most supply, but different mechanisms dominate year to year (Figure 10b).
 During the study period, only one major avulsion was observed (in the 2007–2006 epoch). Ferguson [1993, pp 78–79, Figure 5] described avulsions as a “mechanism characteristic of braiding” and reviewed three primary modes of avulsion. In the first mode (termed constriction and overflow), flow is constricted in a bend promoting overflow on the outside bend, which then connects to a topographically lower inactive braid channel (Figure 12a). In the second mode (termed bank erosion), flow is again constricted, but bank erosion on the outside bend provides the connection between the active channel and inactive braid channel (Figure 12b). In the third mode (termed choking), one of the anabranches at a bifurcation is experiencing incision and the other experiencing aggradation, which eventually leads to choking and abandonment of the aggrading anabranch, and the incising anabranch becomes the primary conduit of flow (Figure 12c).
 Figure 12 illustrates the mechanisms revealed on the Feshie by our change detection that ultimately led to an avulsion in comparison to the three conceptual models of avulsion summarized by Ferguson . A mix of braiding mechanisms combined and culminated in conditions that facilitated this avulsion event. The avulsion took place on a subtle outside bend of the main anabranch. In 2004–2003, bank erosion on the outside right bend of an active anabranch and chute cutoff of a point bar began to occur. In 2005–2004, the resulting diagonal bar expanded in the active channel and the channel widened significantly. In 2006–2005, the primary anabranch continued to aggrade as the diagonal bar continued to grow. Finally in 2007, the primary anabranch became so choked with sediment from continued bar development that the active anabranch plugged and the formerly inactive channel, which started to receive overflow in 2006, incised and became the primary anabranch.
 It is interesting to note that the choking avulsion mechanism that we witnessed here was also the most common anabranch avulsion mechanism that Leddy et al.  documented (70%–75% of all avulsions). The sequence we documented is also consistent with observations by Ferguson and Werritty  of diagonal bar development in the Feshie. Ferguson and Werritty  originally termed the choking avulsion an “interruption” of normal diagonal bar development. This “normal development” was later clarified by both Ashmore  and Ferguson  as chute cutoff—a primary braiding mechanism. So in the Feshie, where normal diagonal bar development (chute cutoff) is prevalent, there are increased opportunities for the process to be “interrupted” by an avulsion.
 We set out to perform one of the first empirical field tests of whether the four classic braiding mechanisms that appear repeatedly in the braided river literature are most responsible for maintaining the dynamism of a braided gravel-bed river. Here we consider the classic braiding mechanisms to be central bar development, transverse bar conversion, chute cutoff, and lobe dissection [Ashmore, 1991; Ferguson, 1993]. It is plausible that these braiding mechanisms are what are required to initiate braiding, but that some other fluvial processes and morphodynamic signatures are what maintain a reach's dynamism—i.e., its regular turnover due to erosion and deposition. However, insights from flume observations [Ashmore, 1982, 1991; Ashworth, 1996] suggest that these braiding mechanisms do occur with regularity over the evolution of braided channel. But what role do morphodynamic signatures like bank erosion and confluence pool scour that are not limited to just braided environments play in the overall dynamics and change in sediment storage of a braided reach?
 Analysis using the classification of morphodynamic signatures we identified revealed that in the Feshie, one of the four main braiding mechanisms (chute cutoff) dominated across the monitored flow range, although the volumetrically dominant process does vary between epochs (Figure 9). In three of the four epochs, chute cutoff dominates the volumetric change, followed by lobe dissection (Figure 9). In the highest flow epoch (2007–2006), the dominant braiding mechanism switches to transverse bar conversion, largely in response to channel widening accommodated by bank erosion. This raises the prospect of flow driven sequences of process dominance that ultimately may control braiding and dynamism.
5.1 Under-Appreciation of Bank Erosion in Facilitating Braiding
 Bank erosion is regarded as a major mechanism of channel change in single-thread channels and much research exists on the topic [Darby et al., 2007], but its role in braided rivers has been overshadowed in the literature by other braiding mechanisms [Mueller, 2012]. We found bank erosion to be as or more important of a mechanism than two of the four braiding mechanisms in terms of changes in sediment storage, and it emerged as the most important “secondary” mechanism (accounting for 17% of total change in storage). Indeed recent flume work has emphasized how critical vegetation is in providing bank cohesion to prevent braiding [Braudrick et al., 2009; Tal and Paola, 2007]. However, this work stops short of explaining just how important bank erosion is in braided rivers as a primary geomorphic mechanism of change and an essential local sediment supply source to help feed the deposition required for the function of braiding mechanisms that produce bars (e.g., central bar development, transverse bar conversion and chute cutoff). However, our data provide among the first empirical evidence of the relative importance of bank erosion in a braided river.
5.2 Role of Flow Magnitude and Variability in Braiding Mechanisms
 The roles of flow magnitude and variability were not an explicit focus of this study, and such an analysis is easier to untangle with event-based repeat topographic data. Within the braided river literature, the role of flow variability in the maintenance of braiding remains contested [Bertoldi et al., 2010]. Flume studies demonstrate that braiding mechanisms can occur at steady discharges perhaps unintentionally promoting the view that braiding is largely independent of flow variability [Ashmore, 1991; Ferguson, 1993]. Lane  suggested that the role of flow variability was to promote different mechanisms of bar evolution at different high flows as elevated bar surfaces become progressively more inundated and dissected by erosion. Bertoldi et al.  demonstrated morphodynamic dependence on discharge from cross-section measurements in a large, braided gravel-bed river. Thus, a greater range of high flows need not necessarily produce a greater volume of geomorphic change than a system with constrained flow variability (e.g., diurnal), but can produce a greater diversity in the mechanisms that bring about those geomorphic changes.
 At discharges below bankfull stage, flows are confined to the primary anabranch channels, and channel morphodynamics will be a product of low-flow channel geometry. At flows greater than bankfull, hydraulic patterns and sediment dynamics will be only partially influenced by the low-flow channels, with the flow geometry being controlled by floodplain or braidplain geometry. In order to generate flow division through either dissection of bar surfaces or aggradation of bars within a channel, there must be sufficient flow depth to (a) access the bar surface and generate sufficient shear to entrain and transport bar materials and (b) permit accumulation of unit bars above the general bed level. Bertoldi et al.  identify significant differences in sediment volume associated with the transition from flow confinement in primary channel networks below bankfull, and the increase in flow width and activation of secondary channel networks at flows above bankfull.
 To provide a simple test of the hypothesis that reach-scale changes in the volume of erosion or deposition were a function of flow variability, we undertook an analysis of total flow volume above a bankfull threshold during the epoch between surveys. Sear  explained this response at the event scale in terms of the duration of flow over some critical threshold, and concluded that sediment yields for flood events were best explained in terms of the total event power. In the absence of field measured hydraulic data to estimate total event power for the study reach, we consider flow volumes in excess of bankfull. We calculated the total flow volume greater than bankfull for the reach (∑Q − Qbf ). Since flow was sampled at consistent 15 min intervals, this metric is simply the sum of all 15 min values of Q > Qbf divided by duration above the same Qbf threshold (we chose the median between low and high estimates of Qbf in Table 1). Thus, this metric varies as a function of both the magnitude and duration of flows over Qbf for the period between topographic surveys (Figure 13).
 The total number of mechanisms involved in the evolution of the channel pattern and the reach dynamics changes with ∑Q − Qbf. At the lowest values of ∑Q − Qbf, six out of the 10 mechanisms operate, of which two contribute 64% of the total volumetric change. As ∑Q − Qbf increases, the number of mechanisms rises through eight, to a total of nine out of 10. The range of mechanisms also varies with ∑Q − Qbf . We had assumed that as duration over bankfull increased, there would be an increase in the processes of bar surface dissection [cf Lane et al., 2008], such as chute cutoff and lobe dissection, and an increase in central bar development and transverse bar conversion as d/d50 increased sufficiently for bar formation. In fact, dissection processes remained important throughout our study period, with the exception of the highest ∑Q−Qbf, when depositional processes of transverse bar conversion, and erosional processes of bank erosion, confluence pool-scour and chute cutoff dominated volumetric changes in storage. We postulate this might be because topographic steering of flows in smaller floods was more effective at “carving up” and sculpting heterogeneity in the braidplain; whereas bigger floods might carry with them higher sediment loads and topographic steering is muted such that they just “paste” big sheets of sediment across the braidplain, which are subsequently reworked by falling-limb flows and smaller floods.
 We deduce from this analysis that the total volume of change scales to the magnitude of flow magnitude and variability. Total flow volume above bankfull is also important in switching process dominance towards (in this case) mechanisms of channel widening and transverse bar building. At lower discharges, different braiding mechanisms still dominate, but these are confined to areas immediately surrounding the existing anabranch network, and overall less volumetric change (work) is undertaken.
5.3 Implications for Other Braided Rivers
 There are at least four conceptual implications of our findings which can be extended to other braided rivers. First, we have tied the conceptually powerful notion of dynamism, thought to be the hallmark of a braided river, to a measurable metric—change in sediment storage. In so doing, we learned that when combined, classic braiding mechanisms [e.g., Ashmore, 1991; Ferguson, 1993] appear to be responsible for a bulk of the dynamism, a finding that we speculate is likely to be true in most braided rivers where braiding is maintained—regardless of flow regime. Secondly, the importance of bank erosion and other erosional morphodynamic signatures in supplying sediment locally to feed these braiding mechanisms is critical. For gravel-bed braided rivers, we surmise that local changes in storage dynamics due to erosion are the primary supply of sediment for the extensive deposition that takes place regularly to maintain active mid-channel bar surfaces. Third, we found that chute dissection is not only an erosional process, but appears to promote significant deposition with the growth of the resultant mid-channel diagonal bar. This form of bar development has been previously identified [Ferguson and Werritty, 1983], but the relative roles in dynamism of carving the chute versus building the bar are underappreciated. Finally, we found that the four common braiding mechanisms in combination were net aggradational, despite requiring both significant amounts of erosion and deposition to occur. Since the defining geomorphic units of braided rivers are active mid-channel bars, most persistently braided rivers must have braiding mechanisms with net aggradation change in sediment storage to maintain such bar forms. However, whether or not the entire braided reach is in a state of net aggradation, dynamic equilibrium or net degradation will be explained by the prevalence of the other morphodynamic signatures of change (e.g., bank erosion, bar sculpting, channel incision) in relationship to the upstream supply of sediment to the reach.
 From a methodological perspective, this study has demonstrated a tractable and robust workflow for exploiting topographic time series to interrogate the mechanisms responsible for the dynamism of a reach. The DoD segregation techniques are facilitated by the GCD software, and our methods here describe specific techniques for identifying and interpreting geomorphic changes as specific braiding mechanisms and other modes of adjustment. Future studies on other braided rivers may examine whether or not our empirical observations hold of braiding mechanisms (i) dominating, (ii) being primarily depositional and (iii) being fed largely by a locally sourced supply of sediment from (in order of importance) bank erosion, confluence pool scour, channel incision, and bar edge trimming. If these findings hold, they provide a strong basis for testing the validity of morphodynamic simulation models by looking for these morphodynamic signatures as emergent properties in the results.
 We sought to classify the morphodynamic signatures of change responsible for driving the dynamism of a braided gravel-bed river, and in so doing test presumptions about the relative importance of classic braiding mechanisms versus alternative sediment exchange processes. Recent methodological progress in repeat topographic surveying and geomorphic change detection has opened the door to testing long-standing concepts and theories about how braided rivers function, but we asserted that these data and techniques have not been adequately exploited to progress our understanding of how braided rivers function. We recast the concept of geomorphic dynamism here in terms of erosion and deposition that are manifested topographically by changes in sediment storage. Using geomorphic change detection methods, we examined variability year to year in the morphodynamic signatures that produced that change on a braided reach of the River Feshie via annual resurveys.
 Our analyses indicated that the four classic braiding mechanisms frequently described in the literature [Ashmore, 1991; Ferguson, 1993] were in combination responsible for a majority of the dynamism in the reach (61% of total change), and these produced a net aggradational change in storage. However, even though bank erosion is not itself a braiding mechanism, it should be considered a key mechanism in facilitating braiding and, in this study, was the third most important individual mechanism of change. Chute cutoff appeared to be the most common mechanism of the four which accounted for braiding. We also found that the channel appeared to maintain its braided character in a dynamic equilibrium state over the span of this study, as suggested by topological classification of channel heads, bifurcations, and confluences and braiding indices. In contrast to the four classic braiding mechanisms, other morphodynamic signatures of sediment scour and deposition accounted for 35% of the total topographic change in the reach. While the four classic braiding mechanisms (dominated by chute cutoff) were usually aggradational in nature, these other mechanisms of change were typically erosional. Subsequent analyses indicate that flow magnitude has an impact on the braiding processes that act on the reach, but perhaps more importantly, the time that flows exceeded a bankfull level appear to influence how regularly these processes occur. With an increase in the duration that discharge exceeded bankfull came a corresponding increase in the diversity of braiding processes.
 This research marks one of the first cases in which flume-based observations of braided river morphodynamics have been tested empirically in the field via high-resolution data collection techniques. As such, the workflow presented here may be extended to other braided channels. The findings deserve further study, particularly our conclusion that the classic braiding mechanisms described previously in flume work appear to be of critical importance in shaping streams at the reach scale. Furthermore, the relative importance of different suites of morphodynamic signatures could be used as a basis for testing and verifying the performance of morphodynamic models.
 The authors are grateful to the numerous individuals who assisted in the data collection on the Feshie over an 8 year span from Aberystwyth University, University of Southampton, University of Cambridge, and Glasgow University. In particular, Clare Cox, Rebecca Hodge, and Richard Williams were instrumental in the surveying work. This research was primarily funded by a PhD studentship for the lead author at the University of Southampton, paid for jointly by the School of Geography and Centre for Ecology and Hydrology, with international fees covered by an Overseas Research Studentship from universities in UK. Further support for analysis was provided by the National Science Foundation (Award 1147942). The fieldwork was supported by a Horton Hydrology Research Grant from the American Geophysical Union, the NERC Geophysical Equipment Facility, the British Society for Geomorphology, and an Aberystwyth University Research Fund Award. Further, staff time, travel, and equipment were generously provided by the Institute of Geography and Earth Sciences at Aberystwyth University as well as the Department of Geography at University of Cambridge. Vericat's involvement in the manuscript preparation was supported by a Ramon y Cajal Fellowship (RYC-2010-06264) funded by the Spanish Ministry of Science and Innovation. The conceptual background for the manuscript benefited from helpful discussions with Michael Church and Susannah Erwin. Extensive constructive reviews from Murray Hicks, Editor Alex Densmore and an anonymous reviewer of earlier versions of the manuscript were extremely helpful in improving the focus and clarity of the manuscript. In particular, the exhaustive reviews of Associate Editors Rob Ferguson and John Buffington were instrumental in restructuring the paper. Finally, we thank Andrew Fomil for his patience in accommodating our requests for time extensions on the revisions.