Wood dispersal in braided streams: Results from physical modeling

Authors


Abstract

[1] Wood is a key feature of riverine systems, playing a significant role in their morphodynamics and ecology. Wood dynamics have been widely investigated in single-thread streams, but limited information is available about wood transport and deposition in large multithread rivers. In this work, we used a large (3 × 25 m) physical model to provide a quantitative description of wood dispersal processes in braided systems. Deposition patterns were characterized in terms of downstream distribution and accumulation size and linked to wood piece properties (diameter, length, and presence of a rootwad) and flow stage. Statistical analysis of observed wood patterns showed that relative log diameter strongly reduces wood mobility, with travel distance dropping for diameters exceeding 50% of median channel depth. Wood tended to form small, sparse accumulations, with large log jams occurring for large log length, complex piece shape, and moderate flow discharge. Flow stage had a dual effect on mobility as both channel conveyance and the availability of retentive sites increased with discharge for the tested range of flow conditions. Additionally, wood deposition was strongly linked to bed morphology. In particular, 40–60% of transported wood was stored at bar apex, with more than 30% of wood deposited on the first bar downstream of the input point, highlighting the crucial role of local-scale morphology in wood dispersal.

1. Introduction

[2] In-channel wood is widely recognized as one of the driving factors of fluvial morphology and ecology [Keller and Swanson, 1979; Gurnell et al., 2000a; Piégay, 2003; Gurnell, 2013]. Deposited wood affects local river morphology promoting scour and pool formation as well as sediment accumulation. Wood also enhances habitat diversity and may promote island initiation [Nakamura and Swanson, 1993; Abbe and Montgomery, 1996; Kollmann et al., 1999; Gurnell et al., 2001; Gurnell and Petts, 2002; Montgomery and Piégay, 2003; Swanson, 2003; van der Nat et al., 2003; Gurnell et al., 2005].

[3] The geomorphic role of wood and its in-stream distribution depend on input processes, channel morphology, and hydrological regime. Past research has shown that the relative relevance of these factors changes along the river system, resulting in distinct downstream trends in wood accumulation style [Keller and Swanson, 1979; Gurnell et al., 1995; Marcus et al., 2002; Comiti et al., 2006]. In particular, the log length to channel width ratio has been identified as a primary driver of wood dispersal dynamics [Piégay and Gurnell, 1997; Gurnell and Sweet, 1998]. Wood mobility in small to medium streams is strongly limited by the large relative size of logs. On the contrary, in “large” rivers [see Gurnell et al., 2002], logs are much shorter than typical channel width and can travel a long distance before deposition. In these systems, wood dispersal patterns are driven by the spatial distribution of flow velocity and the availability of potential retention sites, in turn governed by the combination of discharge and morphological structures.

[4] Field observations on large rivers show a strong relationship between bank erosion, bar morphology, and wood accumulation. Bars are primary wood retention sites and trees originating from bank erosion are often found on the first emergent morphological structures downstream of the input point [Abbe and Montgomery, 1996; Piégay et al., 1999; Abbe and Montgomery, 2003; Lassettre et al., 2008]. The relevance of morphological processes on wood dispersal is further enhanced in multithread rivers, where flow depth is generally small and the vast occurrence of sediment bars and islands provides both wood sources and retention sites [Piégay et al., 1999; Gurnell et al., 2000a, 2001; Bertoldi et al., 2013]. Moreover, braided rivers are characterized by specific width/stage relationships [van der Nat et al., 2002; Ashmore and Sauks, 2006; Welber et al., 2012], where relatively small increases of water depth are associated with major widening of wet area. As a consequence, wood is more dispersed in braided rivers and accumulations are smaller [Abbe and Montgomery, 2003; Lassettre et al., 2008; Bertoldi et al., 2013].

[5] A physical model of a braided river was designed with the aim to investigate wood dynamics, accumulation styles, and mobility over freely evolving bed morphologies. Previous laboratory studies addressed wood dispersal and its governing factors, including wood density, log length, diameter and initial orientation, wood input rate, and flow discharge [Braudrick et al., 1997; Braudrick and Grant, 2000, 2001; Bocchiola et al., 2006a, 2006b; Crosato et al., 2013]. However, few morphological conditions have been tested so far, notably not multithread systems.

[6] In the present study, physical modeling was carried out with the aim of simulating wood dispersal processes occurring in large, gravel bed, braided rivers, rather than reproducing a specific field case [Ashmore, 1991; Paola et al., 2009]. The specific objectives of this study can be summarized as follows: (i) providing a quantitative description of wood dispersal and accumulation style in braided streams under controlled conditions; (ii) investigating the factors governing wood transport, focusing on wood piece properties and flow discharge; and (iii) exploring the relationship between bed morphology and wood dynamics at bar and reach scale, identifying retention sites and wood transport spatial scales.

2. Methods

2.1. Flume Setup

[7] Experiments were performed in a 3 m wide, 25 m long flume located at the Hydraulics Laboratory of the University of Trento (Figure 1a). The flume was filled with well-sorted sand with a median grain size (d50) of 1.03 mm. A software-controlled recirculating pump and a helical screw conveyor were used for water and sand supply, respectively. Output bedload was collected in a filtering crate and automatically weighted. A laser profiler mounted on a carriage moving on high-precision rails provided bed elevation data. Surveys covered a 2.5 m wide, 22 m long area with a spacing of 0.01 and 0.1 m along the transversal and longitudinal coordinates. A 4 m high, movable metal frame carrying a digital camera allowed the acquisition of series of vertical images covering the entire flume length.

Figure 1.

(a) View of the laboratory flume. Flow is toward the camera; (b) size range of the dowels used for flume simulations.

[8] Flow discharge for braided network formation and longitudinal slope were set to 1.8 l/s and 0.01, respectively. Average sand output observed during a set of preliminary model runs (1.5 g/s) was assigned as sand input rate. The initial condition was represented by a straight, 0.22 m wide rectangular channel carved along the flume centerline. Networks developed freely under constant water and sediment supply and evolved toward a reach-scale steady state configuration characterized by (i) the balance between time-averaged sand input and output fluxes and (ii) the stabilization of space-averaged braiding index and flow width [Bertoldi et al., 2009]. This steady state configuration was achieved after about 20 h. A total of five different braided networks were employed for this study.

2.2. Network Properties

[9] The morphological and hydraulic properties of the simulated braided networks were reconstructed from imagery and bed topography data. Relevant dimensionless parameters (Shields stress, Froude number, width to depth ratio of single anabranches) are summarized in Table 1. Data derived for two reaches of the braided Tagliamento River (northeast Italy) are provided for comparison.

Table 1. Morphological and Hydraulic Parameters and Wood Properties of Flume-Scale Networks and Two Braided Reaches of the Tagliamento River
 Laboratory FlumeTagliamento at Cornino (Lower Montane)Tagliamento at Forni di Sotto (Upper Montane)
  1. a

    Figures represent median values.

Discharge (m3/s)0.00181100125
Slope0.010.0030.02
Median grain size (m)0.0010.040.10
Froude numbera0.86a0.60a0.95a
Median Shields number0.0340.0350.045
90th perc. Shields number0.0800.0710.113
Anabranch width/depth31a246a18a
Log diameter (m)0.002 ÷ 0.0060.15a0.09a
Log length (m)0.04 ÷ 0.1214.0a2.3a
Log diameter/median anabranch depth0.36 ÷ 1.090.17a0.19a
Log diameter/90th perc. anabranch depth0.142 ÷ 0.4280.0930.098
Log diameter/d502 ÷ 63.80.9
Log length/median channel width0.15 ÷ 0.450.070.18

[10] Four different flow stages ranging from 50% to 100% of network-formative discharge were selected for wood dispersal modeling, and frequency distributions of anabranch width and depth were computed to provide information for wood piece scaling. Statistics were computed only on upstream-connected anabranches as they represent the portion of the network where wood transport can occur. Results are summarized in Figure 2 for each of the four discharges.

Figure 2.

Network properties: (a) anabranch width; (b) flow depth distribution; and (c) braiding index expressed as box plots. Whiskers extend to the 10th and 90th percentiles.

[11] Total wetted width increased with discharge due to (i) the widening of single anabranches and (ii) the opening of new channels, as observed by Mosley [1982]. Median channel width (wm, Figure 2a) ranged between 20 cm at the lowest flow and 27 cm in formative conditions. The local deepening of existing channels (on average 3 mm between the minimum and maximum discharge) was counterbalanced by the generation of new, shallow anabranches. As a consequence, channel depth distribution showed minimal changes, with a median value of 5.5 mm (dm, Figure 2b).

[12] Finally, the reach-averaged braiding index was computed by counting upstream-connected anabranches intersected by a set of cross sections (“channel count index”) [see Egozi and Ashmore, 2008]. Braiding index (Figure 2c) slightly increased with discharge, ranging from less than 2 for lower values of discharge (0.9 and 1.2 l/s) to 2.5 for higher flows (1.5 and 1.8 l/s).

2.3. Wood Modeling

[13] Cylindrical wooden dowels were used to simulate in-channel wood at flume scale and four values of length and diameter were selected to investigate log size variability (see Figure 1b and Table 2). Wood piece size distributions computed for the aforementioned reaches of the Tagliamento River were used as reference values for dowel design (see Table 1 and Bertoldi et al. [2013]).

Table 2. Combinations of Wood Properties and Flow Conditions Used for Model Runs
Wood and Flow PropertiesNumber of Runs
D (mm)L (cm)X (%)Q (l/s)
  1. a

    These groups of runs are the same.

2, 3a, 4, 6801.840
34, 6, 8a, 1201.840
380a, 25, 50, 1001.840
3800.9, 1.2, 1.5, 1.840
38500.9, 1.2, 1.5, 1.840

[14] Dowel diameter was chosen on the basis of grain size and ranged between 2 and 6 times the median bed particle size. On the Tagliamento River, the median log diameter to d50 ratio is equal to 3.8 in the lower montane reach and smaller (0.9) in the upper course. Dowel length values were selected on the basis of wetted channel width distributions of flume-scale networks with the aim of reproducing “large” river conditions. The highest value of length (12 cm) corresponds to the 14th percentile of channel width, while the others (4–8 cm) are lower than the 3rd percentile.

[15] Finally, cross-shaped elements were added to a fraction of the dowels to model the effect of rootwads (see Figure 1b). The width of the cross-shaped element (12 mm) was set to four times the log diameter on the basis of a linear relationship derived from Tagliamento River data.

[16] Dowels were manufactured using commercially available wood rods. Wood density was measured in dry conditions and at regular intervals after immersion in water and was found to increase rapidly during the first hour of immersion and very slowly afterward. Chestnut wood was selected for log stems because its wet density (0.63 ± 0.04 kg/dm3, measured after immersion in water for 1 h) is similar to that of typical riparian species (Salix sp., Alnus glutinosa, Populus nigra) as reported by Thévenet et al. [1998]. Walnut wood was used for rootwads because it ensured higher durability and slightly higher density (0.70 ± 0.03 kg/dm3). Tree roots are generally denser than stem wood and rootwads often incorporate soil and/or gravel and cobbles, which increase their weight [MacVicar et al., 2009]. Dowels were immersed in water for 1 h before use, to minimize density fluctuations during model runs.

2.4. Structure of Model Runs

[17] Wood dispersal simulations were conducted by adding cohorts of 48 dowels to fully formed, wood-free braided networks under steady flow conditions. Sets of model runs were designed to separately investigate the influence of four external factors on wood mobility, namely (i) dowel length (L); (ii) dowel diameter (D); (iii) percentage of elements with roots (X); and (iv) dispersal discharge (Q). The combinations of values of the four factors used for flume simulations are summarized in Table 2. For each combination, 10 replications of the experiment were performed in order to take into account the spatial and temporal variability of braided networks. For this reason, a different wood input point was adopted for each model run within a set of replications. Moreover, runs were performed on different network realizations in order to maximize the effect of morphological variability. Wood input points were selected only along anabranches with detectable bedload, because bank erosion, which is the primary wood supply mechanism in partly confined and unconfined rivers (erosion—the primary wood supply mechanism in partly confined—and unconfined rivers [Nakamura and Swanson, 1993; Gurnell and Sweet, 1998; Piégay, 2003]), is expected to occur only along active channels. Overall, a total of 180 model runs were performed.

[18] Dowels were manually dropped into the selected anabranch at a random orientation to the flow. Input frequency was set to one element every 4 s to prevent piece-to-piece interaction during transport (thus to have “uncongested transport” as defined by Braudrick et al. [1997]). This choice also resulted in very short runs (lasting for approximately 4 min), so that changes in bed configuration within a single model run could be assumed as negligible.

[19] After the last element of the cohort had stopped, discharge was lowered to a very low value (about 0.2 l/s) to keep the sand bed saturated and wood deposits were mapped by visual inspection on a 5 × 5 cm grid. Dowels touching each other were considered as a log jam, thus setting the minimum size of a jam to two logs in analogy with Comiti et al. [2006]. For each deposition site, the number of logs (accumulation size) was recorded and travel distance was computed as the difference between the longitudinal coordinate of the site and that of the input point. Logs exiting the flume (less than 2% of overall wood input) were excluded from all subsequent analyses.

2.5. Data Analysis

[20] Wood dispersal patterns were described in terms of frequency distributions of travel distance and accumulation size. As a first step, distributions were computed for each set of 10 replicated runs and qualitatively compared. Subsequently, median travel distance, the percentage of isolated dowels, and the average size of the three largest jams were computed for individual runs. Data referring to sets of replicated runs were organized in samples (referred to as “groups”).

[21] One-way analysis of variance (ANOVA) and Tukey post hoc tests [see Steel et al., 1997] were used to test the hypothesis that log diameter and length, the percentage of elements with roots and discharge produce statistically significant differences in wood dispersal patterns. The nonparametric Kruskal-Wallis test and Nemenyi post hoc test [see Hollander and Wolfe, 1999] were used in case of nonnormal, heteroscedastic samples. However, the results of parametric and nonparametric tests were very similar. For two-sample comparisons, the t-test or (whenever needed) the equivalent nonparametric Mann-Whitney test were used. Outliers were defined in accordance with McGill et al. [1978]. Statistical analysis was conducted using the software R, version 2.15.2.

3. Results

3.1. Observations on Wood Transport and Deposition Styles

[22] Dowels aligned with flow over a very short distance (a few log lengths) regardless of initial orientation, as observed also by Braudrick and Grant [2001], and pieces with rootwads most commonly floated with these at their upstream side. Along channel bends, dowels were pushed against the outer bank where they could be trapped parallel to the water margin. No obvious differences in dowel motion were observed between pieces with and without roots flowing through deep channel sections (fully floating conditions as defined by Bocchiola et al. [2008]). Upon encountering fast-flowing, low-depth areas such as bar crests, dowels came into contact with the bed surface while still subject to significant drag. In this case, dowels without roots tended to rotate until approximately perpendicular to flow and then roll along the bar surface until (i) stopping in a low-velocity area or (ii) meeting a deeper section, where they usually resumed their flow-parallel position. The presence of roots significantly influenced dowel motion over shallow flow areas. When the tips of the roots touched the bed, the dowel invariably aligned itself parallel to flow with rootwads upstream and was slowly pushed downstream by the drag force exerted on the rootwad.

[23] Distinctive deposition styles were observed in model runs, which reproduced wood dispersal patterns and processes in large rivers. Typical log orientation and jam structure found at flume scale (Figure 3) mimic those observed at field scale and reported, for example, by Piégay and Gurnell [1997] for the wandering Drôme River (France) and by Gurnell et al. [2000b] and Bertoldi et al. [2013] for the braided Tagliamento River (Italy). In particular, jams often developed from individual logs deposited parallel to flow by deposition of additional orthogonal members, as observed by Abbe and Montgomery [1996] in a large meandering river.

Figure 3.

Examples of wood deposition patterns: (a) jam on a mid-channel bar (L = 12 cm, D = 3 mm, X = 0); (b) logs scattered across bars (L = 4 cm, D = 3 mm, X = 0); (c) log accumulation at bar head (L = 8 cm, D = 3 mm, X = 50%). Dotted lines represent flow direction.

[24] Wood deposits were often found at the upstream end of emerged or submerged bars (“bar apex”) [see Abbe and Montgomery, 1996, 2003] and along concave banks (see Figures 3a and 3c, respectively). Braudrick et al. [1997] also observed wood deposition by gradual accretion on submerged bars for low to moderate wood supply rates. The vast majority of jams comprised a limited number of pieces, while accumulations of more than 10 logs occurred only for long dowels (L = 12 cm) or cohorts with a high percentage of elements with roots (50% and 100%). Longer dowels showed a tendency to form jams (Figure 3a) while smaller elements were often found in small groups or scattered across bar surfaces (Figure 3b).

[25] The complexity and high spatial variability of bed morphology determined relevant differences in wood distribution and aggregation size within groups of run repetitions. The occurrence of many diffluence sites provided shallow areas and favorable locations for wood deposition. On the other side, small changes in water distribution between anabranches were responsible of widespread variations in wood deposition pattern. In the following, results are presented in terms of median travel distance and jam size distribution for each set of runs and the role of wood piece properties, flow discharge, and bed morphology is discussed.

3.2. Wood Properties and Dispersal

[26] Figure 4 shows the cumulative frequency distribution of travel distance computed over sets of 10 runs (480 logs) for tested values of log characteristics and discharge. In all cases, about 50% of dowels are found a short distance downstream of the input section (less than 4 ÷ 6 m). Wood dispersal increases for decreasing values of log diameter (Figure 4a) and is helped by high discharge (Figure 4d), while mobility is strongly limited by the combination of moderate discharge and presence of roots (Figure 4e).

Figure 4.

Cumulative distributions of travel distance computed over groups of 10 runs, for different values of (a) log diameter; (b) log length; (c) percentage of logs with roots; (d) discharge, all logs without roots; and (e) discharge, 50% of logs with roots.

[27] Statistics of median travel distance values computed for individual runs (48 logs) are presented in Figure 5 as box plots. Median travel distance ranges from 1 to 10 m, showing a large scatter due to the observed high variability of single runs. However, statistically significant (p < 0.05) between-group differences occur. In particular, ANOVA highlighted an effect of diameter on wood mobility (F(3,36) = 5.63, p = 0.003). Tukey tests (Table 3) showed a significant reduction of median travel distance between small (2 ÷ 3 mm) and large diameters (4 ÷ 6 mm). The percentage of logs with roots also appears to play a role (ANOVA F(3,36) = 2.99, p = 0.043), but Tukey tests show a statistically significant effect for only one group comparison (see Table 3).

Figure 5.

Median travel distance of log cohorts with different values of (a) log diameter; (b) log length; and (c) percentage of logs with roots. Data for sets of 10 runs are presented as box plots with dots indicating outliers; whiskers extend to the minimum and maximum values. Significant ANOVA p values are shown in bold.

Table 3. Tukey Test p Values for Median Travel Distance, Percentage of Isolated Logs, and 90th Percentile of Travel Distance pa
Median Travel DistancePercentage of Isolated Logs90th Percentile of Travel Distanceb
D (mm)PX (%)pL (cm)PX (%)pQ (l/s)p
  1. a

    Significant p values are shown in bold.

  2. b

    For cohorts of dowels without roots.

2–30.9150–250.7064–60.6300–250.8430.9–1.20.867
2–40.0260–500.0284–80.0500–500.9990.9–1.50.034
2–60.0090–1000.3674–12<0.0010–1000.0100.9–1.80.007
3–40.11325–500.2626–80.45225–500.7911.2–1.50.177
3–60.04525–1000.9406–120.00225–1000.0011.2–1.80.047
4–60.97450–1000.5748–120.10250–1000.0131.5–1.80.921

[28] Finally, travel distance peaks for intermediate values of piece length (8 cm), while both short (4 cm) and especially long dowels (12 cm) show a distinctively low mobility. No significant difference in travel distance associated with dowel length was detected by ANOVA but the Kruskal-Wallis test yielded a contrasting result. Nonparametric post hoc comparisons identified a significant decrease in travel distance between 8 cm long and 12 cm long elements (Nemenyi test p = 0.018). It must be noted that very high within-group variability occurs for high values of length and intermediate values of root prevalence. Direct observations suggest that, for the case of mixed dowel cohorts (25% and 50% of logs with roots), log input order might influence deposition patterns. In the present study, care was taken in adding dowels with and without roots in random order. However, when logs with roots were among the first to enter the flume, their influence appeared to be comparatively larger. Results for other percentiles of travel distance closely mirror those found for the median.

[29] The effect of dowel dimensions on jam size distribution was investigated in terms of (i) (un)likeliness to form a group (expressed as the percentage of isolated logs) and (ii) relative abundance of accumulations of a given size.

[30] The percentage of isolated logs is reported in Figure 6 for individual runs. Overall, 30–60% of wood in a model run is deposited as single logs. Median values for groups of 10 replications vary between 40% and 50%, with the exception of the case of very long dowels and 100% of elements with roots. Log diameter appears to have no influence on the proportion of isolated logs (Figure 6a). In contrast, a decreasing trend can be observed for increasing values of dowel length and proportion of elements with roots (Figures 6b and 6c). A significant difference is highlighted for the largest values of both length and percentage of wood with roots (ANOVA for L: F(3,36) = 9.51, p = <0.001; ANOVA for X: F(3,36) = 6.77, p = <0.001; see Table 3 for Tukey test outputs). The average size of the three largest jams of each run was also computed and was found to range between 3 and 15 logs. The influence of log characteristics on this parameter is essentially identical to that observed for the proportion of isolated logs, further pointing out the role of piece length and roots on accumulation style.

Figure 6.

Percentage of isolated logs for different values of (a) log diameter; (b) log length; and (c) percentage of logs with roots. Box plot specifications as in Figure 5.

[31] In order to investigate frequency distributions of jam size, accumulations were classified into four classes (jams formed by 2, 3, 4 to 9 and more than 9 logs) and the proportion of wood in each jam size class was computed (Figures 7a–7c). Small accumulations (2 ÷ 3 logs) store invariably more than 30% of wood deposited as jams. This percentage does not depend on dowel diameter, but it decreases from 70% to 30% with increasing dowel length. Large jams with more than 9 logs are particularly relevant (more than 40% of all wood in jams) for 12 cm long dowels and when all logs have rootwads.

Figure 7.

Wood distribution over jam size classes for different values of (a) log diameter; (b) log length; (c) percentage of logs with roots; (d) discharge, all logs without roots; and (e) discharge, 50% of logs with roots.

3.3. The Role of Flow Discharge

[32] The investigation of the role of flow conditions was carried out in analogy with that of piece properties. Frequency distributions of travel distance computed for each set of 10 runs show an increase in travel distance with discharge (Figures 4d and 4e). For all flow conditions, skewness increases and standard deviation decreases if elements with roots are present, showing that rootwads reduce wood mobility. However, the effect of high discharge is not very strong and is comparable to the decrease in mobility determined by the presence of roots.

[33] The median value and 90th percentile of travel distance were computed for individual runs (48 logs) and are presented as box plots in Figure 8. No significant effect of discharge on median travel distance was found (Figures 8a and 8b), while the 90th percentile significantly increases with discharge if roots are absent (ANOVA: F(3,36) = 5.71, p = 0.003; see Figures 8c and 8d and Table 3 for Tukey test outputs). A t-test was carried out on selected percentiles of the distribution (10th, 25th, 50th, 75th, and 90th) to test the combined effect of discharge and roots on travel distance. As shown in Table 4, the influence of roots changes with discharge. Their presence induces a significant reduction of all percentiles for wood dispersed at low flow, while at higher discharge, it influences only the tail of the distribution (higher percentiles). The complex relationship between flow depth, velocity, and discharge in braided networks (see Figure 2) is likely the main cause of this complex pattern, as discussed in section 4.3.

Figure 8.

Travel distance of log cohorts for different values of discharge and percentage of logs with roots: median, (a) for X = 0% and (b) for X = 50%; 90th percentile: (c) for X = 0% and (d) for X = 50%. Box plot specifications as in Figure 5.

Table 4. t-Test p Values for a Set of Percentiles of Transport Distance, as a Function of Flow Dischargea
PercentilesFlow Discharge
0.9 l/s1.2 l/s1.5 l/s1.8 l/s
  1. a

    Significant p values are shown in bold.

10th0.0230.9960.0550.007
25th0.0360.4140.1850.112
50th0.0140.3610.7030.095
75th0.0490.4090.0620.004
90th0.0560.5410.0470.008

[34] The influence of discharge on jam size distribution was also analyzed. No significant effect of flow conditions on either the percentage of isolated logs or the average size of the largest jams was detected. Discharge also exerts a weak control on jam size distribution (Figures 7d and 7e). Small accumulations (2 ÷ 3 logs) store approximately 40% of wood regardless of flow stage. However, the combination of low flow conditions (up to 1.2 l/s) and presence of dowels with roots (Figure 7e) determines an increase in the relative relevance of large jams (more than 9 logs) from less than 20% to almost 40% of transported wood. This effect is comparable to that observed for the longest dowels (L = 12 cm) and the highest percentage of elements with roots (X = 100%).

3.4. Braid Bar Pattern and Wood Dispersal Pattern

[35] Wood dispersal patterns observed at flume scale clearly show that bar apex areas and concave channel banks are preferential sites for wood deposition, confirming laboratory observations by Braudrick et al. [1997]. In particular, 30–40% of all deposition sites are located at bar apex and store 40–60% of all transported wood. ANOVA and t-test analyses showed no statistically significant effect of either discharge or presence of roots on these figures, suggesting that the retentiveness of bar apex sites is large regardless of flow conditions and log shape.

[36] The strong association between braid bars and wood deposits suggests a relationship between bar length and the longitudinal variability of wood volume. This hypothesis was tested on a subset of 10 runs conducted under identical flow and bed morphology conditions and using the whole range of dowel sizes and shapes. In this set of runs, bed topography showed a regular series of bars, with a characteristic length of approximately 2 m, as measured from the bed topography surveys. The aggregated wood deposition map of the 10 runs was resampled over a 20 × 20 cm grid (Figure 9a). The longitudinal distribution of wood was computed at two different spatial scales by aggregating accumulations first over one quarter of bar wavelength (0.5 m) and then over an entire wavelength (2 m).

Figure 9.

(a) Map of wood deposition across the network; (b) downstream distribution of wood aggregated over ¼ of bar wavelength; and (c) downstream distribution of wood aggregated over one bar wavelength.

[37] The relationship between bar and wood patterns is highlighted by histograms presented in Figure 9. At bar scale (Figure 9b), wood retention peaks at the apex of the four main braid bars identified in the network. Approximately 80% of wood is deposited on the upstream half of each bar. Furthermore, at reach scale (Figure 9c), the first bar immediately downstream of the wood input point hosts more than 30% of dowels and this percentage steadily decreases downstream. In this case, more than 50% of wood is deposited on the first two bars and only about 15% travels beyond the fourth bar.

4. Discussion

4.1. Modeling Wood Dispersal in Braided Systems

[38] Wood dispersal processes observed in the present flume experiments are generally in good agreement with those reported from flume and field studies. Limited information is available regarding transport style, especially at field scale (video monitoring being a promising method, see MacVicar et al. [2009] and MacVicar and Piégay [2012]). However, laboratory simulations by Braudrick and Grant [2001] show a similar behavior in terms of dowel orientation and interaction with submerged bars.

[39] With respect to wood deposition, results of the present study reproduce dispersal patterns occurring in wide gravel-bed rivers, especially in terms of typical accumulation sites and style. The present laboratory-scale simulations showed significant wood retention at characteristic sites (bar apex areas and channel margins), thus closely mirroring field-scale patterns observed in wandering [Piégay et al., 1999], and braided systems [Gurnell et al., 2002; Wyżga and Zawiejska, 2005].

[40] Laboratory experiments highlighted the key role of jams, as they store 40 ÷ 80% of all transported wood. Manners and Doyle [2008] note that the relative relevance of jams depends on wood supply and transport mechanisms. While the spatial density of wood jams was evaluated in various riverine contexts [Piégay and Gurnell, 1997; Gurnell et al., 2000b; Andreoli et al., 2007; Curran, 2010], limited data are available on the proportion of wood stored in jams, and direct comparisons are hindered by the use of different definitions and metrics. In a survey carried out on the large, meandering Lower Roanoke River, Moulin et al. [2011] found that slightly more than 50% of wood was deposited as jams.

[41] Jam size distribution shows the typical characteristics of large braided rivers with relatively low wood input rate and uncongested transport. Logs tend to form small accumulations, generally comprising less than 10 pieces (as observed, for example, by Bertoldi et al. [2013]) as opposed to the accumulations of hundreds of pieces occurring in single-thread systems characterized by intense wood supply and/or large relative log size [Marcus et al., 2002; Andreoli et al., 2007; Manners and Doyle, 2008].

[42] The dominant role of close-range retention is a clear example of the effect of local-scale network structure on wood transport. Bertoldi et al. [2013] provided quantitative data on the dispersal of wood originating from eroding banks in two reaches of the braided Tagliamento River. They reported that about 20 ÷ 50% of eroded trees were deposited on the first bar downstream of input location and these values are comparable with those observed in the present study. A strong link between wood source and wood sink areas has been clearly identified in both single-thread [Lassettre et al., 2008] and braided systems [Pettit et al., 2005]. The importance of local-scale bed morphology was also highlighted by Braudrick and Grant [2001], who noted that channel properties (width and depth) can be a good predictor of wood mobility only when their local values are taken into account. In the case of a highly dynamic braided river, local depth is extremely variable in space and time. The large differences in dispersal patterns observed between replications of the same experiment are most likely associated to the spatial organization of bars and channels immediately downstream of wood input points. As a consequence, wood deposition patterns can be predicted only in statistical terms.

[43] Wood travel distance can be expressed in terms of the mean diffluence spacing as defined by Ashmore [2001]. This parameter represents a longitudinal spatial scale of braided rivers that takes into account the spacing of central bars and therefore the distance of the most efficient depositional sites. In the present study, the mean diffluence spacing computed at formative discharge is about 2.6 m. Therefore, median wood travel distance ranges between approximately 1 and 2.5 times this diffluence spacing.

4.2. Log Characteristics and Wood Mobility

[44] In the present study, clear relationships between piece properties and wood mobility and aggregation style were identified, despite significant residual variance caused by spatial and temporal variations of bed morphology.

[45] Wood diameter appears to be the dominant factor governing travel distance in braided rivers. Early observations by Bilby and Ward [1989] linked log diameter with minimum water depth required for wood flotation. More recently, Braudrick et al. [1997] suggested relative log diameter (defined as the ratio of diameter to mean channel depth) as a key driver of wood mobility. In the present work, median water depth dm (5.5 mm) was used as a more robust statistic to compute relative log diameter D/dm (see Figures 5a and 6a).

[46] Significantly higher travel distance was observed for the two smallest diameters (D/dm = 0.36 and D/dm = 0.54) in comparison with the largest two (D/dm = 0.73 and D/dm = 1.09). Our results confirm field observations showing that wood mobility in large rivers drops when relative log diameter exceeds 0.5 [e.g., Abbe and Montgomery, 2003; Curran, 2010]. It is worth noting that relative log diameter computed from Tagliamento River data is quite small, especially in lower montane sections (Table 1). This may suggest that wood is highly mobile at high flow, with larger transport rate on the rising limb of floods [MacVicar et al., 2009] whereas it deposits mostly during the falling limb of the hydrograph, when the relative log diameter is smaller and a larger number of preferential sites for deposition are available (as observed by Bertoldi et al. [2013]).

[47] It is important to underline that minimum flow depth required for log flotation (“buoyant depth”) [see Braudrick and Grant, 2001] is also influenced by wood density (ρw), which in turn depends on a number of factors including tree species, climatic conditions, decay status, and especially water absorption [Millington and Sear, 2007; MacVicar et al., 2009; Curran, 2010]. All dowels used in the present study essentially share the same density because the slightly heavier rootwads account for only 2% of the total piece volume. However, qualitative considerations can be drawn on the basis of analytical models of incipient log motion. In particular, Braudrick et al. [1997] propose an analytical relationship for buoyant depth as a function of log diameter and wood density for a cylindrical log lying over a flat surface. Buoyant depth for dowels used in the present study corresponds to 0.89D. The cited relationship shows that buoyant depth increases with both log diameter and density. Therefore, a significant reduction of wood mobility can be expected for heavier logs, as occurs for particularly large pieces. As a result, present findings on the role of log diameter may be generalized considering buoyant depth as integrative of both diameter and density, as suggested by Braudrick and Grant [2001]. Finally, it is important to note that the observed drop in log mobility occurs for values of median water depth approximately equal to 2D, that is, slightly more than two times the buoyant depth. This suggests that the local variability of bed topography at both grain and bar scale has a major role in wood transport. Localized shallow flow areas can favor wood retention even when median flow depth should guarantee full transport.

[48] As opposed to log diameter, piece length was expected to exert a weak control on travel distance because the physical model was designed to reproduce “large” river conditions. In the present study, relative log length L/wm [see Braudrick et al., 1997] was computed using median channel width (wm = 26.7 cm). Resulting values of L/wm are 0.15, 0.22, 0.30, and 0.45 for the four tested log lengths (Figures 5b and 6b) and a statistically significant drop in travel distance was found only between the latter two. Differences in mobility are more clearly explained if dowel length is compared with percentiles of channel width. Considering runs with the longest wood dowels (L = 12 cm), 14% of the upstream connect anabranches were narrower than the wood pieces, while for the other logs, this percentage drops to less than 3%. This suggests that the limited mobility of large logs may be related to an increased probability of interaction with channel banks. However, is important to note that no dowels were found spanning a narrow channel because hydraulic steering rapidly aligns pieces with the flow during transport [Braudrick and Grant, 2001]. On the Tagliamento River, the median value of relative log length ranges between 0.07 (in the lower montane reach) and 0.18 (in the upper montane reach), suggesting that logs longer than 0.3 times the median channel width are probably very rare and therefore that element length plays no role in wood retention, especially in large piedmont braided rivers.

[49] Interestingly, travel distance peaks for intermediate values of length. This behavior may confirm the observations of Braudrick and Grant [2001]. The authors suggested that longer logs are more likely to encounter local low depth, low-velocity areas (and therefore stop), but they also experience a wider range of water velocity and depths with respect to shorter logs. The unimodal relationship between length and travel distance may arise from the balance between these two mechanisms.

[50] Element length also enhances jam formation as the percentage of wood deposited as isolated logs drops especially for the longest elements (L/wm = 0.45; Figure 6b). Gurnell et al. [2002] observed that longer logs are more likely to accumulate in jams because they are more prone to contact with previously deposited wood.

[51] In the present study, we explored the influence of root boles on log mobility and jam formation. Although the relevant influence of piece shape is widely recognized, little quantitative information on its effects is found in the literature. At the laboratory scale, cylindrical dowels have been used in the vast majority of studies, with the notable exceptions of Braudrick and Grant [2000], who tested incipient motion of logs with disc-shaped root boles, and Gippel et al. [1996] and Schmocker and Hager [2011] who used irregularly shaped pieces with roots and branches. Moreover, limited information on the proportion of logs with roots is available. This variable is strongly related to forest properties and supply mechanisms and shows significant fluctuations in the field. Montgomery and Abbe [2006] reported that the vast majority of large logs (key members of jams) found in the large, unmanaged Queets River has a root bole. In contrast, Moulin et al. [2004] characterized wood trapped in a reservoir in the French Alps and found that only 10% of pieces had roots. The authors also noted that elements with roots are less mobile and therefore less likely to be found at large distances. Moreover, wood pieces tend to lose their roots (and branches) during transport or because of weathering [Abbe and Montgomery, 1996; Piégay et al., 1999; Millington and Sear, 2007].

[52] Our results show that the presence of roots limits log mobility, especially for wood dispersed by moderate flows, and enhances jam formation. Direct observation of wood transport during experiments highlighted that the presence of rootwads affects log mobility mainly by preventing rolling, which is common in the case of cylindrical dowels transported downstream on top of bar surfaces.

4.3. The Role of Discharge

[53] It is worth noting that flow conditions tested in the present study are limited to values lower than formative discharge, as the objective was the investigation of depositional patterns. Previous experimental and field analysis showed that wood is more likely to deposit at low to medium flows, when relative log size is larger and depositional sites (as bar apex and channel banks) are more abundant [Bertoldi et al., 2013].

[54] Statistics of network properties reported in Figure 2 show that both anabranch width and braiding index increase with discharge. The first parameter can be interpreted as a measure of the network ability to disperse buoyant particles (e.g., wood), while the latter represents morphological complexity and therefore network retentiveness. A larger number of anabranches is associated with higher proportion of diffluences and extended shoreline length and therefore to an increased availability of major wood deposition sites such as bar apex areas and channel margins.

[55] These contrasting mechanisms provide an explanation for the observed weak relationship between travel distance and flow conditions. At low to moderate flow, discharge is mostly concentrated in a single channel, generating a limited number of potential deposition sites. On the contrary, at high flow, the increase in local water depth and velocity is counterbalanced by the higher retentiveness of the network.

[56] In the present study, a significant positive effect of discharge was found only for high percentiles of the downstream distribution of wood, implying that long travel paths occur mainly for wood dispersed at high flow. However, within-group variability is large. The occurrence of a highly retentive bar immediately downstream of the input point may offset the effect of higher discharge.

[57] Furthermore, the relationship between discharge and braiding index is complex, particularly when considering a wide range of flow conditions [Mosley, 1982; van der Nat et al., 2002; Malard et al., 2006; Doering et al., 2007; Welber et al., 2012]. Small to moderate floods (<1 year recurrence interval) determine high braiding index, while the number of anabranches (and therefore of retentive sites) drops again for higher flows, due to channel merging. This implies that very large events may be needed to observe long-range dispersal.

5. Conclusions

[58] The experimental results presented in this paper show that physical modeling can provide valuable quantitative insights on wood mobility and deposition in braided networks, provided that relevant morphological processes of these river systems are reproduced. We analyzed the effect of wood piece properties (i.e., diameter, length, and presence of roots) and flow discharge on wood deposition patterns, expressed in terms of both travel distance and jam size distribution, through the analysis of a total of 180 runs. Statistical analysis highlighted the primary role of log diameter on travel distance and of log length and presence of roots on the tendency to form large jams. The effect of discharge is less clear, due to the contrasting effect of increased flow depth and velocity and increased retentiveness of networks characterized by a higher number of anabranches.

[59] The experiments pointed out the strong relationship between wood dispersal and bed morphology. In particular, we observed that wood deposition patterns are determined by presence and form of sediment bars. At the bar scale, a high proportion of wood is trapped at bar apex locations, while at reach scale, the first bars downstream of wood input points store the majority of pieces.

[60] Finally, our results indicate that an experimental protocol based on a significant number of test repetitions is of primary importance to infer statistically sound conclusions on wood dynamics in gravel-bed rivers, as a consequence of the complex interaction among bed morphology, flow field, and wood transport observed in these systems.

Acknowledgments

[61] The authors gratefully acknowledge funding from the CARIPARO foundation, which supported the research reported in this paper. Discussions with Angela Gurnell, Hervé Piégay, and Jonathan Laronne greatly helped in experiments design and data analysis. The experiments have been performed thanks to invaluable support of Lorenzo Forti, Marco Redolfi, Martino Salvaro, and Sandra Zanella. We would also like to thank Francesco Comiti and two anonymous reviewers for comments and input on this manuscript. These results are part of the PhD thesis of M. Welber.

Ancillary