Bedload transport of a bimodal sediment bed

Authors

  • M. Houssais,

    Corresponding author
    1. Equipe de Dynamique des Fluides Géologiques, Institut de Physique du Globe, Sorbonne Paris Cité, Paris, France
    • Corresponding author: M. Houssais, Equipe de Dynamique des Fluides Géologiques, Institut de Physique du Globe, Sorbonne Paris Cité, 1 rue Jussieu, FR-75238 Paris CEDEX 05, France. (houssais@ipgp.fr)

    Search for more papers by this author
  • E. Lajeunesse

    1. Equipe de Dynamique des Fluides Géologiques, Institut de Physique du Globe, Sorbonne Paris Cité, Paris, France
    Search for more papers by this author

Abstract

[1] Despite several decades of investigations, accounting for the effect of the wide range of grain sizes composing the bed of rivers on bedload transport remains a challenging problem. We investigate this problem by studying experimentally the influence of grain size distribution on bedload transport in the simple configuration of a bimodal sediment bed composed of a mixture of 2 populations of quartz grains of sizes D1 = 0.7 ± 0.1 mm and D2 = 2.2 ± 0.4 mm, respectively. The experiments are carried out in a tilted rectangular flume inside which the sediment bed is sheared by a steady and spatially uniform turbulent flow. Using a high-speed video imaging system, we focus on the measurement of the average particle velocity and the surface density of moving particles, defined as the number of moving particles per unit surface of the bed. These two quantities are measured separately for each population of grains as a function of the dimensionless shear stress (or Shields number) and the fraction of the bed surface covered with small grains. We show that the average velocity and the surface density of moving particles obey the same equations as those reported by Lajeunesse et al. (2010) for a bed of homogeneous grain size. Once in motion, the grains follow therefore similar laws whether the bed is made of uniform sediment or of a bimodal mixture. This suggests that the erosion-deposition model established by Lajeunesse et al. (2010) for a bed of uniform sediment can be generalized to the case of a bimodal one. The only difference evidenced by our experiments concerns the critical Shields number for incipient sediment motion. Above a uniform sediment bed, the latter depends on the particle Reynolds number through the Shields curve. In the case of a bimodal bed, our experiments show that the critical Shields numbers of both populations of grains decrease linearly with the fraction of the bed surface covered with small grains. We propose a simple model to account for this observation.

1. Introduction

[2] The wide range of grain sizes found in most rivers, especially gravel bed rivers, complicates the problem of the prediction of bedload transport rate. As stated by Wilcock and Kenworthy [2002], grain size influences sediment transport in two different ways. For given flow conditions above a bed of homogeneous sediment, transport is controlled by the absolute size of grains, small grains being more mobile than large ones. However, when the sediment bed is a mixture of different grain sizes, relative size effects tend to increase the transport rate of larger grains and decrease the transport rate of smaller grains [Wilcock, 1993, 2001; Wilcock and Kenworthy, 2002]. This effect is very sensitive to the composition of the mixture which can change during transport and in response to variations in flow and sediment supply. Relative size effects thus influence the transport rate of each individual size and, consequently, the overall transport rate in a gravel bed river [Kuhnle and Southard, 1988; Wilcock and Crowe, 2003; Parker, 2008].

[3] A first approach to tackle this difficult problem consists of avoiding part of the difficulties associated with specifying individual size distribution by predicting the total transport rate as a function of a single representative grain size D, usually the median diameter D50of the grain-size distribution [Meyer-Peter and Müller, 1948]. Within the framework of this approach, the volumetric transport rate per unit river width qs is related to the flow shear stress τ by the Meyer-Peter and Müller [1948] equation,

display math

where c is a dimensionless coefficient, math formula is a dimensionless transport rate called the Einstein number and θ is a dimensionless shear stress called the Shields number. These two dimensionless numbers are defined by

display math
display math

where g is the gravitational acceleration, ρ is the fluid density, ρs is the sediment density and R = (ρsρ)/ρ. The critical Shields number θc is the value of θ below which sediment transport ceases [Shields, 1936; Meyer-Peter and Müller, 1948]. Equation (1) is practical because the only sediment data required are the representative size and the value of θc. It is however unable to predict changes of grain size distribution in the river.

[4] A second approach consists of discretizing the grain-size distribution into a finite numberN of populations of characteristic grain size Di and investigating how the transport rate of each population, qs,i, depends on the shear stress and the grain-size distribution at the bed surface [Parker, 2008]. It is usually observed that 1) qs,i increases with the proportion of grains belonging to the ith population within the surface layer, fi; 2) far from the threshold of motion, math formula where θi = τ/ρRgDi is the Shields number calculated for the grain size Di [Liu et al., 2008]. Consequently, many authors hypothesized a relation of the following form [Wilcock, 1988; Parker, 2008]:

display math

where math formula is the dimensionless transport rate of the grains of size Di and F(θi,θc,i) is a function capturing the complexity introduced by the grain-size distribution. Among other things,F(θi,θc,i) depends on the critical Shields number of incipient motion associated with the grains of size Di, θc,i, which depends in turn on the grain-size distribution at the bed surface. Within the framework of this approach, authors have therefore focused on the determination of bothF(θi,θc,i) and θc,i from experimental data and theoretical modeling.

[5] Parker et al. [1982], for example, proposed that

display math

where a, b, l, m and p are adjustable coefficients depending on the configuration investigated [Parker et al., 1982; Parker, 1990; Wilcock and Kenworthy, 2002; Wilcock and Crowe, 2003; Powell et al., 2001, 2003]. As the Shields number approaches the value of incipient sediment motion, sediment transport becomes sensitive to local bed heterogeneities and pavement effects, which may not be captured by (5). This led to the formulation of several transport laws specifically designed to describe sediment transport in the vicinity of the threshold [Parker et al., 1982; Parker, 1990; Wilcock and Kenworthy, 2002; Wilcock and Crowe, 2003].

[6] The prediction of the critical Shields number of the grains of size Di, θc,i, rests on a qualitatively well observed phenomenon: the ability of a grain of size Di to move with respect to some representative grain size Dc increases with the ratio Di/Dc [Parker, 2008]. This is usually formulated by a relation of the form

display math

where θc,c = τc,c/ρgDc is the critical Shields number of the representative grain size, τc,c is the value of the shear stress at which grains of size Dc start moving and Gis the so-called hiding function which incorporates relative size effects [Kirchner et al., 1990; Buffington et al., 1992].

[7] The frequently cited forms of the hiding function G(Di/Dc) are listed in Table 1. Most of them reduce to

display math

with 0 ≤ γ ≤ 1. Equation (7) presents two interesting limiting cases. The first one corresponds to γ = 0 which implies θc,i/θc,c = 1. In this case, a grain of given size Di within a mixture has exactly the same critical Shields number as it would have if the bed was composed entirely of grains of size Di. In such a scenario, the initiation of transport of sediment mixtures is highly selective, based on grain size. The second limiting case corresponds to γ = 1 which implies θc,i/θc,c = Di/Dc so that τc,i = τc,c where τc,i is the value of the shear stress at which grains of size Di start moving. In this limiting case, the effect of the mixture is to equalize the threshold of motion so that all grains are mobilized at the same absolute boundary shear stress, a configuration referred to as “equal mobility”. In practice, sediment mixtures appear to behave in between these two limiting cases.

Table 1. Examples of Hiding Functions Commonly Found in the Literaturea
Hiding FunctionReferenceRemark
math formulaEgiazaroff [1965]Theoretical modeling
math formula for math formulaAshida and Michiue [1973]Based on experimental data
math formula for math formula  
math formulaParker et al. [1982]Based on field data
math formulaPowell et al. [2001, 2003]Based on field data
math formula with math formulaWilcock and Crowe [2003]Based on experimental data

[8] The transport laws discussed so far rely on the fit of empirical sediment transport rate curves. Consequently the physical meaning of the coefficients involved in equations (5) and (7) is unclear. An alternative approach to the problem of bedload transport is to consider that qs,i can be written

display math

where ni (dimensions [L]−2) is the surface density of moving particles of the ith population defined as the number of particles of size Di in motion per unit bed area, Vi is their averaged velocity and δvi is the volume of an individual particle. A better insight in the problem of bedload transport can be gained from the separate measurements of Vi and ni and the determination of their dependence on the control parameters. This approach has motivated several experimental investigations of bedload transport at the grain scale and the development of statistical transport models [Einstein, 1937, 1950; Francis, 1973; Fernandez-Luque and Van Beek, 1976; Abbott and Francis, 1977; Lee and Hsu, 1994; Ancey et al., 2008; Ganti et al., 2009; Ancey, 2010; Furbish et al., 2012a, 2012b, 2012c; Roseberry et al., 2012].

[9] Lajeunesse et al. [2010], in particular, reported the results of an experimental investigation of the motion of bedload particles, under steady and spatially uniform turbulent flow above a flat sediment bed of uniform grain size. Using a high-speed video imaging system to investigate the trajectories of the moving particles, they found that: (1) the surface density of moving particles increases linearly with (θθc),

display math

where α = 4.6 ± 0.2 is a coefficient related to the erosion and deposition rate of bedload particles and σ = 1/D2 is the number of particles at repose per unit surface of the bed; (2) the average particle velocity increases linearly with math formula, with a finite nonzero value at threshold,

display math

where math formula is a characteristic sedimentation velocity and β = 4.4 ± 0.2 and Vc/Vs = 0.11 ± 0.03 are two fitting coefficients.

[10] As far as we know, experiments similar to those of Lajeunesse et al. [2010] have been carried out with homogeneous sediment beds only. In this paper, our objective is therefore to extend the experimental approach of Lajeunesse et al. [2010] to the case of a mixture of sediments of different sizes. To this end, we investigate experimentally bedload transport above a bed composed of a bimodal mixture of small and large grains sheared by a turbulent flow in a small experimental flume. We focus on the measurement of the surface density of moving particles and of their average velocity for each size fraction. The experimental results allow us to characterize how sediment transport depends on the proportion of small and large grains on the bed.

2. Experimental Apparatus and Procedure

2.1. Experimental Setup

[11] The experiments were conducted in a tilted laboratory flume of length 240 cm and width W = 9.6 cm (Figure 1a). The bottom of the flume was covered with a sediment bed composed of a mixture of small and large irregularly shaped quartz grains of density ρs = 2650 kg m−3 (Figure 1b). The average size of the small grains determined from sieve analysis was D1 = 0.7 ± 0.1 mm whereas the average size of the large grains estimated from image analysis was D2 = 2.2 ± 0.4 mm.

Figure 1.

(a) Scheme of the experimental setup. (b) Sediment bed (top view) for an experimental run of the series 2 (ϕ1 = 0.71). On this image, the large grains are dyed in black whereas the small ones are white.

[12] The bed, typically 10 centimeters thick, was flattened by sweeping a rake whose orientation and distance from the bottom of the flume were constrained by two rails parallel to the channel. The bed slope S was measured with a digital inclinometer (accuracy 0.1°).

[13] Once the bed was ready, water was injected by a pump at the upstream flume inlet with a constant flow discharge Qw measured with a flowmeter (accuracy 0.01 L/minute). The discharge was always high enough for the flow to form across the full width of the flume. To prevent any disturbance of the bed, water was not injected as a point source but rather it overflowed smoothly onto the river bed via a small reservoir (see Figure 1). The reservoir extended across the full width of the channel and therefore guaranteed a flow injection that was uniform across the channel width.

[14] Each experimental run consisted of the preparation of an initial flat bed composed of a bimodal mixture of grains. The sediment bed was then sheared by a steady and spatially uniform turbulent flow. The motion of the particles entrained by the flow was studied with a high speed camera as described in the next sections. The typical duration of an experimental run was a few minutes. Once the experimental run was over, a new bed was prepared and a new experiment was performed with different values of the control parameters which are the flow rate, the bed slope and the proportion of small and large grains.

2.2. Measurement of the Surface Fraction of Small Sediments

[15] The surface fraction of small sediments, ϕ1, defined as the fraction of the bed surface covered with small grains, was measured for each experimental run. This was achieved by using a camera positioned above the experimental setup to acquire images of the sediment bed. These images were used to measure the fraction of the bed surface covered with large grains, ϕ2, from a direct counting of the coarse grains within the field of view of the camera. The fraction of small grains was then deduced from ϕ1 = 1 − ϕ2.

2.3. Measurement of the Shields Number

[16] Because we did not feed sediment at the channel inlet, an erosion wave slowly propagated from the inlet toward the outlet of the flume. All our experiments were stopped well before this degradation wave had reached the middle of the flume where we performed our measurements so that it never interfered with our results. Indeed the slope of the sediment bed measured at the end of each experimental run was equal to the initial slope within the experimental accuracy.

[17] The sidewalls of the flume were made of glass of roughness negligible compared to that of the sediment bed. The water flow depth H measured with a ruler (accuracy ±1 mm) at three locations regularly spaced along the flume was constant along the section of the flume (within the experimental accuracy). The flow was therefore uniform. In these conditions, the shear stress on the sediment bed was estimated using the classical steady flow assumption

display math

where g is the acceleration of gravity, and Rh = HW/(2H + W) is the hydraulic radius. The Shields number for each grain size i is then given by:

display math

The uncertainties in τ and θi result from the uncertainties in the measurements on H and S.

2.4. Measurement of the Sediment Transport Rate

[18] Sediment particles transported by the flow settled out in a constant water level overflow tank located at the flume outlet. The tank rested on a high-precision scale (accuracy 0.1 g) connected to a computer that recorded the mass every 5 s. The total sediment discharge per unit channel widthqs was deduced from the sediment cumulative mass. The initiation of the flow was followed by a transient phase which lasted for about two minutes. After this transient, the transport rate reached a steady state. All the experimental measurements described hereafter were performed during this steady state regime and as long as the bed was flat. As discussed in the previous section, these experimental conditions facilitate the estimate of the flow shear stress.

[19] Our setup provides real time measurements of the total sediment transport rate only. The fractional transport rates for each grain size, qs,1 and qs,2, were estimated by interposing a grid between the flume outlet and the overflowing tank once the steady state regime had been reached. qs,1 and qs,2 were then estimated with an accuracy of typically 1% by sieving and weighing the sediments accumulated over the grid.

2.5. Characterization of the Motion of the Particles

[20] The motion of the particles entrained by the flow was observed using a high-speed camera (250 images/s, 1024 × 1024 pixels) positioned vertically above the bed. The grains entrained by the flow followed intermittent trajectories composed of a succession of periods of rest and periods of motion. Although qualitatively similar, the motion of small and large grains had to be studied following different experimental protocols as described in the next sections.

2.5.1. Measurement of the Surface Density of Moving Particles

[21] The surface densities of small and large moving particles, n1 and n2, were measured for each experimental run by counting manually the numbers of particles of each size moving downstream between two successive frames within the field of view, and averaging over a sufficiently large number of frames, typically several hundreds, for the mean to converge.

[22] Because of slight oscillations of the water surface, a resting grain may appear to be in motion. However, such an apparent displacement fluctuates back and forth around a fixed position so that the time integrated displacement of the corresponding grain is null. This is not the case if the particle actually moves. To determine reliably whether a particle is at rest or not, we therefore correlated the displacement of each examined particle on several successive images. Particles that appeared to slightly rock back and forth were considered at rest. This method turned out to be efficient but difficult to automatize. This is why moving particles were identified by one person. Although time consuming, this method was far better than the automated techniques we tried to implement.

2.5.2. Measurement of the Velocity of the Large Grains

[23] The images were also used to measure the velocities of the large grains. To do so, a small fraction of the large particles were dyed in black. This allowed us to track their position from frame to frame, with a time resolution of 0.004 s, using a particle tracking algorithm developed in our lab. The data were then processed to calculate the streamwise components of the large particle velocities, ux,2. The spatial resolution of the images was such that the diameter of the large particles was about 50 pixels and the size of the field of view of the order of 20 × 20 particle diameters. Under these conditions, we were able to determine the position of the center of mass of a particle on an image with a spatial resolution on the order of 2% of a grain diameter. The oscillations of the water surface, which are the main source of experimental error, degraded the accuracy of these measurements causing an apparent movement of particles at rest. The corresponding noise was calibrated for each experimental run by computing this apparent velocity. This allowed us to define a cutoff velocity in the range 1–3 cm s−1 depending on the water flow rate, below which the velocity measurement was considered too noisy to be taken into account.

[24] Measuring a large number (typically 7000 for each experimental run) of large particles instantaneous velocities allowed us to compute their experimental distributions. For the explored range of parameters, these distributions all exhibit a peak and are skewed to the large velocities (see Figure 2). They are well fitted by a chi-squared distribution:

display math

where Γ is the gamma function and kis a fitting coefficient, equal to the mean of the chi-squared distribution. The average velocity of the large particles,V2, was therefore computed for each experimental run from a fit of the experimental distribution of ux,2 by equation (13).

Figure 2.

PDF of the streamwise velocity component of the large particles, ux,2, measured for θ2 = 0.087 and ϕ1 = 0.71. The dotted line corresponds to a fit of the data by equation (13).

[25] For each experimental run, the transport rate of the large particles deduced from the average velocity and the surface density of moving particles, δv2n2V2 (see equation (8)), was compared to the transport rate directly measured with the scale at the flume outlet. As illustrated by Figure 3, the two methods led to similar values thus confirming the consistency of the experimental data.

Figure 3.

Transport rate of the large grains deduced from the average velocity and the surface density of moving particles, δ2n2V2 (see equation (8)), plotted as a function of the sediment transport rate directly measured with the scale at the flume outlet. Circles, squares and diamonds correspond to the experimental series 1, 2, and 4, respectively. The solid line corresponds to a linear relationship of slope 1.

2.5.3. Measurement of the Velocity of the Small Grains

[26] The noise introduced by the water surface oscillations did not prevent us from detecting whether a small grain is moving or not using the procedure described above. However, it did not allow us to measure the velocity of the small particles with a satisfying accuracy. As a result the automated technique used to measure the experimental distributions of the velocities of the large grains could not be employed with the small ones. The average velocity of the small particles, V1, was therefore computed from the measurement of the transport rate qs,1 and the surface density of moving particles, n1, using equation (8)

display math

3. Experimental Results

[27] We performed a total of 53 experiments with bed slopes ranging from 5.10−3 to 7.10−2, water discharges ranging from 20 to 62 L min−1 and flow depth ranging from 1 to 3 cm. These experimental runs are organized in 4 series, labeled from 0 to 3. Each series correspond to a different bed composition, namely ϕ1 = 1 (uniform bed of small particles), 0.92, 0.71 and 0.54, respectively. We also used an additional set of data, labeled 4, extracted from Lajeunesse et al. [2010] and corresponding to measurements performed with a uniform bed of large particles i.e. ϕ1 = 0. Altogether, these 5 experimental series cover a relatively wide range of parameters summarized in Table 2.

Table 2. Range of Parameters Explored for Each Series of Experimentsa
Seriesϕ1Sθ1θ2θc,1θc,2 math formula math formulaH/D1H/D2
  • a

    The series labeled 0 to 3 were performed by the authors, the data of the 4th series are extracted from Lajeunesse et al. [2010].

010.0052–0.0080.045–0.106-0.041 ± 0.004-16.4–25.1-18–34-
10.91 ± 0.010.0045–0.0320.041–0.3010.013–0.0960.044 ± 0.0030.006 ± 0.00115.7–42.249–13216–265–9
20.71 ± 0.050.0055–0.0330.055–0.2740.018–0.0870.07 ± 0.0050.012 ± 0.00218–40.356.5–126.116–255–9
30.54 ± 0.050.007–0.0350.051–0.3320.016–0.1060.068 ± 0.0040.017 ± 0.00117.4–44.354.4–138.915–305–10
400.01–0.068-0.02–0.15-0.023 ± 0.003-65–179-4–6

3.1. Phenomenology

[28] The images acquired with the fast camera reveal that the motion of the sediments entrained by the flow above a bimodal bed is qualitatively similar to the one reported for the case of a unimodal bed of sediment: 1) only a small fraction of the sediment bed particles is entrained by the flow; 2) these bedload particles exhibit intermittent behavior: periods of motion, called “flights” and characterized by a highly fluctuating velocity, alternating with periods of rest.

[29] However the relative proportion of small and large particles at the bed surface influences quantitatively the sediment transport rate. The total sediment flux measured by the scale at the flume outlet is plotted on Figure 4 as a function of the flow shear stress for two different experimental series. For a given shear stress, the total sediment transport rate varies with the surface fraction of fine sediments, ϕ1.

Figure 4.

Total sediment flux (g s−1) plotted as a function of the flow shear stress τ (N m−2) for the experimental series 2 (ϕ1 = 0.71, solid circles) and 4 (ϕ1 = 0, open triangles). The solid lines correspond to the prediction of equation (23).

3.2. Surface Density of Moving Particles

[30] By analogy with Lajeunesse et al. [2010], we chose to normalize the surface density of moving particles of a given size Di with respect to the number of static particles of the same size available per unit surface of the bed, math formula. We therefore introduce the dimensionless surface density of moving particles of size Di:

display math

The dimensionless surface densities of large particles in motion, math formula, measured for ϕ1 = 0.91 and ϕ1 = 0.54 (experimental series 1 and 3, respectively) are plotted as a function of the Shields number, θ2, on Figure 5. Two observations can be made: 1) math formula is 0 below a threshold value of the Shields number, which differs between the two series; 2) above this threshold, math formula increases linearly with θ2. These observations hold for all experimental series and both grain sizes. It seems therefore natural to define the critical Shields number of the grains of size Di, θc,i, from the threshold value below which math formula vanishes. A linear fit of math formula versus θi, i = 1,2, for each experimental series leads to the values of θc,i reported in Table 2. As commonly observed, the critical Shields number decreases when the proportion of fine sediments on the bed, ϕ1, increases. These variations will be discussed later.

Figure 5.

Dimensionless surface density of moving large particles, math formula, as a function of the Shields number, θ2, for ϕ1 = 0.91 (experimental series 1, circles) and ϕ1 = 0.54 (experimental series 3, triangles). The two dotted lines correspond to linear fits of the data.

[31] Another interesting observation can be made from Figure 5: the slope of the relationship between the surface density of moving grains and the Shields number is identical for the two series of data plotted on this graph. This result can be generalized to both grain sizes (i = 1, 2) and all experimental series. Indeed, plotting the dimensionless surface densities measured for the 5 experimental series, math formula, as a function of (θi − θc,i) for both grain sizes reasonably merges all data on the same line (Figure 6). A fit of this line leads to the linear relationship

display math

This result is consistent with the observations of Lajeunesse et al. [2010] who reported a similar linear relationship above a uniform sediment bed (see equation (9)), with a coefficient equal to 4.6 ± 0.2. As a matter of fact, the measurements made by Lajeunesse et al. [2010] with uniform sediment beds of size 1.1 and 5.5 mm fall on the same trend (crosses on Figure 6).

Figure 6.

Dimensionless surface densities of moving grains math formula versus (θi − θc,i) for both grain sizes (i = 1, 2) and the 5 experimental series (see Table 2). Circles, squares and triangles correspond to the experimental series 1, 2 and 3, respectively, with solid symbols for large particles and open ones for small particles. Open and solid diamonds correspond to the series 0 (ϕ1 = 1) and 4 (ϕ1 = 0), respectively. The dotted line is a linear fit of our data. The data measured by Lajeunesse et al. [2010] with uniform sediment beds of size 1.1 and 5.5 mm are represented by crosses.

3.3. Average Velocities

[32] As discussed in the experimental setup section, the average particle velocity of each size fraction, Vi, was estimated for each experimental run, either from a fit of the velocity distribution for the large particles or from the ratio of the sediment flux to the surface density of moving particles for the small ones. By analogy with Lajeunesse et al. [2010], we normalize the average particle velocities of each fraction with respect to their characteristic settling velocity math formula, thus introducing the dimensionless average particle velocity

display math

[33] As shown in Figure 7, the dimensionless particle velocities of both small and large grains converge onto a single line independent of the proportion of fine sediment, ϕ1, when plotted as a function of math formula. Thus the particle velocity can be written

display math

where Vc,iis the (non-zero) average particle velocity at the threshold. Fitting the data, the slope and the threshold velocity are found to be

display math
Figure 7.

Average dimensionless velocities versus math formula for both grain sizes (i = 1, 2) and the 5 experimental series (see Table 2). Circles, squares and triangles correspond to the experimental series 1, 2 and 3, respectively, with solid symbols for large particles and open ones for small particles. Open and solid diamonds correspond to the series 0 (ϕ1 = 1) and 4 (ϕ1 = 0), respectively. The dotted line is the linear fit of all the experiments. The data measured by Lajeunesse et al. [2010] with uniform sediment beds of size 1.15 and 5.5 mm are represented by crosses.

[34] Again, we recover a result consistent with the measurements performed by Lajeunesse et al. [2010] who found a linear relationship (see equation (10)) above a uniform sediment bed. The slope coefficient reported by Lajeunesse et al. [2010], 4.4 ± 0.2, is slightly smaller than the one measured in the present study. However, despite some dispersion, the velocity measurements of Lajeunesse et al. [2010] (crosses on Figure 7) are compatible with our data. It is therefore difficult to assess whether this difference of slope coefficient is significant or due to experimental uncertainties.

[35] Finally, the above result suggests that the velocity is discontinuous at the threshold of sediment transport, particles moving there with a nonzero velocity Vc. As pointed out by Lajeunesse et al. [2010], this result is supported by the experimental observation that, close to the threshold, a particle, once dislodged from the sediment bed, does not stop immediately but may travel over some distance. It is also consistent with the hysteretic nature of the threshold of motion first evidenced by Hjülstrom [1935] and with observations of the motion of a single particle on a fixed bed by Francis [1973] and Abbott and Francis [1977].

4. Discussion

4.1. Summary of the Experimental Results

[36] To summarize, we have investigated bedload transport of a bimodal sediment bed, composed of a mixture of two populations of quartz grains of size D1 = 0.7 and D2 = 2.2 mm, respectively. The particles of both populations are entrained by a steady and uniform turbulent flow above a flat topography. In this equilibrium regime, the erosion and deposition rates balance each other. The granulometric composition of the bed is characterized by the surface fraction of small grains, ϕ1, defined as the fraction of the bed surface covered with grains of the population 1.

[37] Our experimental results show that:

[38] 1. Particles begin to move for Shields number larger than a threshold θc,i, which was determined from the extrapolation to zero of the surface density of moving particles. This threshold differs from one population of grains to the other and decreases when the surface fraction of small grains increases (see Table 2).

[39] 2. Above the threshold Shields number, the dimensionless surface density of moving particles, math formula, increases linearly with (θi − θc,i) following equation (16) independent of the fraction of small grains (see Figure 6). This linear dependence is consistent with the measurements performed by Lajeunesse et al. [2010] who found a similar linear relationship (see equation (9)) above a uniform sediment bed, with a coefficient equal to 4.6 ± 0.2. These experimental investigations demonstrate that the dimensionless number of particles in motion per unit surface is independent of both the grain size and the surface fraction of small particles, at least for the range of parameters explored.

[40] 3. The dimensionless average particle velocity increases linearly with math formula following equation (18) independent of the fraction of small grains (see Figure 7). The slope of the linear law and velocity at threshold are given by β = 5.1 ± 0.2 and Vc,i/Vs,i = 0.11 ± 0.02. Again, we recover a result consistent with the measurements performed by Lajeunesse et al. [2010] and no significant dependence is noted with both the grain size and the surface fraction of small particles, at least for the range of parameters explored. This velocity law fully agrees with Lajeunesse et al. [2010, equation (13)] which arises from a simple force balance. The value of the coefficient β is close to that of previous investigations, which found β in the range 4.4–5.5 for an even larger range of grain sizes [Nino and Garcia, 1994].

4.2. Erosion-Deposition Model for a Bimodal Bed

[41] Our experimental observations suggest that the erosion-deposition model established byCharru [2006] and Lajeunesse et al. [2010] for a bed of uniform sediment can be generalized to the case of a bimodal one. These authors showed that the deposition rate scales as the surface density of moving particles divided by the time necessary for a particle to settle from a characteristic height equal to the grain size. Generalizing this argument to the case of a bimodal bed leads to

display math

where cd is a dimensionless coefficient. Similarly, the erosion rate is proportional to the number of particles at repose per unit surface of the bed

display math

where ce is a dimensionless coefficient. The term te,i is a typical hydrodynamic timescale which can be thought as the time needed for a particle at rest, submitted to the force exerted by the flow, to escape the small trough where it is trapped and reach some “escape velocity” of the order of the settling velocity [see Lajeunesse et al., 2010, equation (12)].

[42] Our experiments were performed above a steady state flat topography. In this equilibrium regime, the erosion and deposition rates balance each other so that

display math

Equation (22) is consistent with the experimental relation (16), with ce/cd × ρ/ρs = 4.2 ± 0.2 which leads to ce/cd = 11.1 ± 0.5. Note that this value of ce/cd is different from the one reported by Lajeunesse et al. [2010]. The contradiction is however only apparent. Indeed, Lajeunesse et al. [2010, equation 15] forgot the term ρ/ρs. As a result, the term ρ/ρs is included in their fitting coefficient ce. This being taken into account, the values of ce/cd are identical.

[43] Finally, the sediment flux of each size fraction can be computed from (18) and (22)

display math

from which we deduce the total sediment flux per unit of width, qm = qs,1 + qs,2. The mass fluxes of small and large particles, and consequently the total sediment flux, calculated from (23) are consistent with the transport rate measurements performed with the scale as illustrated on Figures 4 and 8.

Figure 8.

Transport rate (g s−1) of all particles (solid circles) and of the small particles only (open circles) plotted as a function of the flow shear stress τ (N m−2) for the experimental series 2 (ϕ1 = 0.71). Solid lines correspond to the predictions of equation (23).

4.3. Variation of the Critical Shields Number With the Surface Fraction of Small Grains

[44] Our experimental results demonstrate that, once in motion, the grains obey similar equations whether the bed is made of uniform sediment or of a bimodal mixture: in both cases the surface density of moving particles increases linearly with the Shields number and the average velocity increases linearly with the square root of the Shields number. The only difference evidenced by our experimental results concerns the value of the critical Shields number. Above a uniform sediment bed, the latter is uniquely determined from the particle Reynolds number through the Shields curve [Shields, 1936; Wiberg and Smith, 1987]. In the case of a bimodal bed, our experiments show that the critical Shields number for a given grain size depends also on the grain-size distribution at the bed surface as already observed in previous investigations [Wilcock, 1998]. This is illustrated in Figure 9 which displays the critical Shields numbers of both populations of grains, θc,1 and θc,2, as a function of the surface fraction of small grains, ϕ1. Two observations can be made: 1) the critical Shields number of the small grains is larger than that of the large grains, whatever the value of ϕ1; 2) the critical Shields numbers of both grain populations decrease linearly with ϕ1.

Figure 9.

Critical Shields number of the small (θc,1, open circles) and the large (θc,2, solid circles) grains as a function of the surface fraction of small grains ϕ1 on the bed. These critical Shields values and the corresponding error bars were obtained from a linear fit of the surface density of moving grains as discussed in section 3.2. Crosses and triangles correspond to measurements of the critical Shields number of a single large grain protruding above or partly buried in a bed of small grains, respectively. Stars correspond to measurements of the critical Shields number of a few small grains deposited on a bed of large grains. Dotted lines represent the equations (24) and (25).

[45] Despite a considerable amount of work partly summarized in the introduction, a mechanistic prediction of the critical Shields number of a sediment mixture remains an open problem which lies far beyond the scope of the present paper. Instead, we will now take advantage of the experimental observations to propose a simple statistical model that can describe the variation of the critical Shields numbers of the small and large grains composing a bimodal bed as a function of the surface fraction of small grains, ϕ1.

[46] Let us consider a test grain of size D2 at rest on a bimodal sediment bed characterized by its surface fraction of fine sediment, ϕ1. Our test grain has a probability ϕ2 = 1 − ϕ1 to repose on grains of size D2. In that case, it is expected to start moving when the Shields number reaches a value θc,2/2 equal to the critical Shields number for a bed composed exclusively of grains of size D2. The value of θc,2/2 is deduced from the experimental series 4 performed with a bed of uniform sediment of size D2 leading to θc,2/2 = 0.023 ± 0.003.

[47] Our test grain has also a probability ϕ1 to repose on grains of size D1. In that case, it is expected to start moving when the Shields number reaches a value θc,2/1 similar to the critical Shields number of a grain of size D2which would be at repose on bed composed exclusively of grains of size D1. Consequently, on average, we expect that the critical Shields number of the large grains is

display math

[48] Equation (24) predicts a linear relationship between θc,2 and ϕ1 which is consistent with the data plotted on Figure 9. To further test this equation, we need to estimate the value of θc,2/1. This was achieved by a series of specific experimental runs in which we prepared a sediment bed exclusively composed of small grains of size D1 onto which we added a single large grain of size D2. The critical Shields number of the latter was determined by increasing slowly the flow discharge until it was carried away. We performed several runs using different large grains to account for the slight variations of shape and size. Some of the experimental runs were conducted with the large grain half-buried in the bed (triangles onFigure 9). In other experimental runs, the large grain was posed above the sediment bed so that it fully protruded (crosses on Figure 9). Both protocols led to quasi-identical values (seeFigure 9) so that, on average, θc,2/1 = 0.011 ± 0.001. As shown on Figure 9, the critical Shields number of the large grains calculated from equation (24) using the experimental measurements of θc,2/1 and θc,2/2, is consistent with our experimental data.

[49] A similar reasoning leads to the following expression for the critical Shields number of the small grains:

display math

where θc,1/1 is the critical Shields number of a bed of uniform grain size D1 and θc,1/2 is the critical Shields number for a single small grain of size D1 at repose on bed composed exclusively of large grains of size D2. The experimental series 0 performed with a bed of uniform sediment of size D1 gives θc,1/1 = 0.041 ± 0.004. As for θc,1/2, it was estimated from specific experimental runs in which we deposited a few small grains of size D1 onto a bed exclusively composed of large grains of size D2. The critical Shields number of these latter was determined by increasing slowly the flow discharge until they were carried away, leading to θc,1/2 = 0.143 ± 0.014 (stars on Figure 9). The critical Shields number of the small grains calculated from equation (25) using the experimental measurements of θc,1/1 and θc,1/2 is plotted as a function of ϕ1 on Figure 9. Again, the prediction of this very simple model is in good agreement with the experimental data.

[50] According to the model developed in this section, the critical Shields number of each of the two populations of grains forming a bimodal sediment bed is a linear combination of two asymptotic configurations: a first one where the grains have all the same size and a second one where a grain of a given size rests on a bed exclusively composed of grains of the other size. Within the framework of this approach, complex hydrodynamics and granular effects (grain size, bed roughness, friction, lift and drag forces, etc.) are embedded in the values of the 4 coefficients θc,1/1, θc,1/2, θc,2/1 and θc,2/2.

5. Conclusion

[51] We have reported the results of an experimental investigation of bedload transport above a bimodal sediment bed, composed of a mixture of two populations of quartz grains of size D1 = 0.7 and D2 = 2.2 mm, respectively. The particles of both populations are entrained in a flume by a steady and uniform turbulent flow above a flat topography. In this equilibrium regime, the erosion and deposition rates balance each other. The granulometric composition of the bed is characterized by the surface fraction of small grains, ϕ1, defined as the fraction of the bed surface covered with grains of size D1. Using a high-speed video imaging system, we focus on the measurement of the average particle velocity and the surface density of moving particles, defined as the number of moving particles per unit surface of the bed. These two quantities are measured separately for each population of grains as a function of the dimensionless shear stress (or Shields number) and the fraction of the bed surface covered with small grains.

[52] Our experimental results show that the average velocity and the surface density of moving particles obey the same equations as those reported by Lajeunesse et al. [2010] for a bed of homogeneous grain size (see equations (16) and (18)). Once in motion, the grains follow therefore similar laws whether the bed is made of uniform sediment or of a bimodal mixture. The only difference evidenced by our experiments concerns the value of the critical Shields number. Above a uniform sediment bed, the latter depends on the particle Reynolds number through the Shields curve [Shields, 1936]. In the case of a bimodal bed, our experiments show that the critical Shields numbers of both grains populations decrease linearly with the surface fraction of small grains. We suggest that the critical Shields number of each of the two populations of grains forming a bimodal sediment bed is a linear combination of two asymptotic configurations: one in which the grains have all the same size and a second in which a grain of a given size rests on a bed exclusively composed of grains of the other size.

[53] Note that our results show that an addition of fine particles decreases the critical Shields number of the latter (see equation (25)) and therefore increases their transport rate. The case of large particles is however not so straightforward. Indeed, according to equation (24), an addition of fine particles decreases the critical Shields number of the large particles thus favoring their mobilization. In the meantime, increasing the proportion of fine particles in the sediment bed necessarily decreases the surface fraction of large particles and therefore decreases their mobilization rate as predicted from equation (23). The overall effect of an addition of fine sediments on the transport rate of large particles and, consequently, on the total sediment transport rate, depends on the balance between these two competing effects. According to equations (23), (24) and (25), several regimes are possible. A careful investigation of these regimes might help to better understand the increased mobilization of beds by the addition of fines particles reported for example by Wilcock and Crowe [2003] and Venditti et al. [2010]. This is a work in progress.

[54] Finally, our experimental study suggests that the erosion-deposition model established byLajeunesse et al. [2010]for a sediment bed of uniform grain size can be generalized to the case of a bimodal one, provided that the dependency of the critical Shields number with the surface fraction of small grains is taken into account. Further investigation should focus on the use of the bimodal erosion-deposition model for predicting bedforms and granulometric patch development in rivers.

Acknowledgments

[55] We thank Olivier Devauchelle for many fruitful discussions. We are grateful to H. Bouquerel, Y. Gamblin and A. Vieira for their technical assistance in designing and realizing the experimental apparatus. We gratefully acknowledge support by the Agence Nationale de la Recherche through contract NANR-09-RISK-004 / GESTRANS. This is IPGP contribution 3326.

Ancillary