The effects of kinetic sorting on sediment mobility on steep slopes

Experimental arrangement

The experimental setup consisted of a 6-m-long, 0.1-m-wide tilting flume. We chose to use a narrow flume in order to prevent the formation of bedforms (lateral bars) and we did not add lateral wall roughness in order to allow efficient lateral observations of how the bed evolved. Because of the large width-to-depth ratios used (W/d > 5), no side wall effects were expected for the hydraulics (Song et al., 1995). In addition, another 50 h of experiments (not shown here) have confirmed that all observations presented in this paper are reproduced similarly when the same sediment mixture and flow conditions are used with rough walls (obtained by gluing sediments on the glass walls, as shown in a video available as Supplementary Material to this paper).

The flow rate at the flume inlet was controlled by a constant head reservoir and measured using an electromagnetic flow-meter. The choice of a feed instead of a recirculating experiment is important (Parker and Wilcock, 1993). A recirculating experiment is more likely to represent the long-term evolution of an infinite channel in interaction with the transported sediment. Given that the objective of the present work was to study the effects of grain sorting, it was essential to maintain a constant grain size distribution at the flume entrance, and therefore a feed experiment was used. This choice is also consistent with mountain streams, especially in the upstream parts of the catchment where sediments are fed directly by hill slope processes. The sediment feeding device was composed of a feeding tank and a conveyor belt whose velocity was controlled during all the experiments with a special tachometer device (see Recking et al. 2009 for a more detailed description). The sediment mixture used for the experiments was obtained by mixing several nearly uniform mixtures whose median size was 1.0 mm, 2.1 mm, 4.9 mm, 9.0 mm, 12.5 mm and 18.7 mm; the final grain size distribution (noted hereafter GSD) respected a similitude with natural material (as described in Recking, 2013a) and was characterized by D₁₆ = 1.8 mm, D₃₀ = 2.1 mm, D₅₀ = 3.2 mm, D₈₄ = 9.0 mm and D_m = 6.0 mm (arithmetic mean diameter). The largest grains of the grain size distribution (corresponding to the 9.0 mm, 12.5 mm and 18.7 mm mixtures) were painted with three different colours in order to facilitate the experimental observations. To measure local and mean bed slope (difference between the bed height at the flume entrance and the bed height outlet divided by the total bed length), we used seven staff gauges placed along the flume side (with the same procedure as described in Recking et al., 2009).

The outlet bedload transport was sampled with a bucket every 5 min (the sample was dried and weighed) and also measured with a light table and a customized video system (Frey et al., 2003): at the outlet of the flume, the mixture of sand, gravel and water was forced to flow on a tilted transparent ramp placed above an illuminated waterproof box. A video camera was placed above the plate and operated in back-lighted mode. Software with specific libraries was developed to grab and save series of several images, which were processed with a segmentation algorithm (WIMA software, Ducottet, 1994) for computing the transport rate (see Frey et al., 2003 for greater detail).

To determine the surface GSD, digital images were collected (every 3 h) from approximately 1.5 m above the bed. The camera was attached to a mechanical arm moving along the channel and two kinds of shots were taken. A first shot captured about 10 cm × 1 m of the bed and was used to have a qualitative estimate of the bed armouring. A second shot captured about 10 × 15 cm of the bed and was used to compute the local bed surface grain size distribution, with a Wolman count procedure (Wolman, 1954); 100 stones were counted for each sample using a digital regular step grid. The run had to be stopped before each GSD measurement, and no impact of these stops could be detected when analysing the slope and bedload signals. In fact, when the flow was stopped, the sediments ceased to move within a few seconds, i.e. within a timeframe much smaller than the timeframe which characteristically controls the bed morphodynamics.

Experimental conditions

A very steep 12% slope was used in order to be representative of mountain streams and also to investigate flow conditions not already tested in previous experiments (Recking, 2006; Recking et al., 2009). Each run was characterized by a transport stage τ^*/τ_c^* (Church and Zimmerman, 2007), defined as the ratio between the mean Shields stress and the critical Shields stress. Since the objective was to investigate the effect of kinetic sorting, it was important to consider two situations. In Run 1 the flow was not strong enough to entrain the coarser fractions of the mixture and consequently with negligible kinetic sorting, whereas in Run 2 the flow was strong enough to entrain the coarser fraction of the sediment mixture, producing kinetic sorting. We considered diameter D₈₄ (9 mm) as representative of the coarse fraction and evaluated the flows leading to τ₈₄^*/τ_c84^* < 1 and τ₈₄^*/τ_c84^* > 1, respectively. We computed the hydraulic conditions with the following flow resistance equation developed with a large data set (including poorly sorted sediments) and particularly well adapted for steep slope flows (Recking et al., 2008b):

(1)

with

(2)

(3)

where α_RL is a roughness layer coefficient taking into account deviation from the logarithmic profile on small relative depth flows and α_BR is a bedload roughness coefficient taking into account additional flow resistance due to bedload. The results of the computations are presented in Table 1.

Table 1. Experimental conditions

	Run 1 (Static sorting)	Run 2 (Kinetic sorting)
Q (L/s)	0.3	0.55
Qs (g/s)	6	60
U (m/s)	0.28	0.33
d (cm)	1.1	1.7
d/D84	1.2	1.9
	0.072	0.098
	0.084	0.084
	< 1 (0.86)	>1 (1.18)
Duration (h)	110	92

We also estimated the critical Shields stress with the following equation taking into account the effects of changing hydraulics with the slope (Recking et al., 2008b):

(4)

The strength of this equation stems from its having been obtained by another independent study using a different method. Whereas Equation 4 was obtained by investigating the effect of bedload transport on flow resistance (Recking, 2006, 2008). Shvidchenko et al. (2001) obtained the same equation (coefficient, 0.11–0.2, varying with grain diameter, and exponent, 0.278) using their own flume measurements.

The computed values are presented in Table 1 and indicate that discharges of Q = 0.3 L/s and 0.55 L/s produce transport stages of τ₈₄^*/τ_c84^* = 0.86 and 1.18, respectively. These values are of course dependent on the equations used for their computation; this is why the theoretical approach was complemented with additional tests (by running water over a bed composed of 9 mm of uniform material), which confirmed no sediment movement for Q = 0.3 L/s and material entrainment for 0.55 L/s.

A bedload equation proposed in Recking et al. (2008b) was also used (with the mean diameter of the sediment mixture) to estimate the associated solid discharge for a 12% slope. The values obtained for Q = 0.3 L/s and 0.55 L/s were 8 g/s and 62 g/s, respectively. Each run was started with these computed values. Since we wanted self-made beds, the runs were started with slopes milder than the 12% theoretical equilibrium slope. For instance, Run 1 was started with a 9% slope and it required 200 h for the bed to reach equilibrium (when the mean bed slope stabilized and the average outlet solid discharge equalled the input value). Finally, the conditions were adjusted by trial and error and for each discharge Q = 0.3 L/s and 0.55 L/s, the solid discharge was 6 g/s and 60 g/s, respectively, for a mean bed slope of 12.2%. The total durations for Run 1 and Run 2 (started once the bed was considered at equilibrium) were 110 h and 92 h, respectively.

It is worth noting an important characteristic of these experiments, that is, for a given run, the flow and solid discharge were maintained constant at the inlet during the entire run. This implies that patterns and associated fluctuations observed in the flume are directly linked to the process of transport, and not to variations in the inlet values. This was a necessary condition with regard to the objectives of this study, which was to identify the effects of kinetic sorting on bedload transport and associated bed morphology.

Large fluctuations of the slope and bedload transport

A dynamic equilibrium was attained for Run 1 and Run 2, with a constant mean bed slope and outlet solid discharge, but with fluctuations measured around these mean values (the bed never stabilized). The main differences observed between Run 1 and Run 2 concern the amplitude of these fluctuations, occurring at different time and space scales (Figure 2). The slope fluctuations were larger in Run 2, where on five occasions (noted E1 to E5, separated by a time step of approximately 15 h), the slope attained a peak at the final stage of a long aggradation process, followed by sharp and rapid bed erosion. On the other hand, bedload fluctuations (with reference to the inlet bedload transport value) were larger in Run 1. Considering the mean input bedload values Q_s-mean (6 g/s and 60 g/s for Run 1 and Run 2, respectively), the instantaneous bedload Q_s measured at the flume outlet produced Q_s/Q_s-mean ratios as high as 10 for Run 1, whereas these ratios were smaller than 3 for Run 2.

Bed armouring in Run 1 and Run 2

Despite similar bed surface armouring could be measured in both runs, as shown by the pictures and the grain size distributions in Figure 3, the bed dynamic was very different. In Run 1 the armour was generalized to the entire channel surface and persistent throughout the experiment (except very short erosions occurring at the flume entrance as discussed later), whereas in Run 2 it usually developed over short distances, was very unstable and periodically destroyed, leading to local erosion. Only in a few circumstances, corresponding to the final stage of a long aggradation process preceding peaks E1 to E5, did the bed armour develop over the entire channel bed in Run 2. This bed state was always associated with reduced sediment mobility and minimum bedload transport.

A second difference between Run 1 and Run 2 is that in Run 1 the bed developed transverse clusters, forming step-like structures (but no clear step–pool sequences were observed), whereas in Run 2 the coarse sediments interacted in a more random pattern. Despite its permanent armour and transvers clusters, Run 1 produced an efficient transfer downstream of the injected sediment (no bed aggradation or erosion over the long term); this was possible by cluster destabilization and progressive migration, from place to place, of the largest elements, as was already described for step pool destruction (Rosport and Dittrich, 1995; Church and Zimmerman, 2007; Curran, 2007, 2012)

Vertical sorting in Run 2

In addition to bed surface armouring, very efficient vertical sorting was observed in Run 2. This could be observed visually through the transparent wall, but also by sampling the bed. The bed subsurface was sampled after manually removing the coarse surface over one grain diameter. Figure 4 presents two bed pictures of Run 2, taken at the same location with and without the bed surface, and the associated grain size distributions. It appears visually that the subsurface is composed essentially of fine sediments, which was confirmed by sieving. The thickness of the subsurface layer varied between one and several time the sand diameter in Run 2, whereas it was quasi-inexistent in Run 1.

Bedload sheet production in Run 1 and Run 2

In both runs, the bed armour was periodically destroyed, giving place to local erosion and downstream propagation of sediments, taking the form of bedload sheets (Iseya and Ikeda, 1987; Recking et al., 2009). These successive bed armouring and erosion episodes were responsible for high (with regards to the experiment duration and its critical time scales) frequency fluctuations of the upstream slope, with periods of about 15 min and 1 h for Run 1 and Run 2, respectively.

One important result is that bedload sheets were observed only where the coarse material was mobile (i.e. when kinetic sorting was possible): in Run 1 (with τ₈₄^*/τ_c84^* <1) they were limited to the upstream part of the channel only (over a distance of approximately 0.5 m), close to the feeding device, where the coarse grains were mobile because of kinetic energy associated with the injection; in Run 2 (with τ₈₄^*/τ_c84^* >1) bedload sheets were observed in all parts of the channel.

This capacity to produce bedload sheets has strongly impacted the dynamics of bedload transport measured at the flume outlet. In Run 2, bedload fluctuations were strongly controlled by bedload sheets, as shown in a video available as supplementary material to this paper. This was not the case for Run 1 where bedload sheets produced in the flume upstream part rapidly disappeared when interacting downstream with the bed armour, and almost never attained the flume outlet. Actually, in Run 1, bedload fluctuations were more the consequence of streamwise sorting, with increased mobility of sediments every time a cluster was destabilized immediately upstream of the outlet section.

Substantial bed erosion in Run 2

On five occasions, Run 2 developed complete bed surface armouring as the final stage of a long aggradation process (peaks E1 to E5 in Figure 2). This bed armour was systematically destroyed, replaced by considerable bed erosion affecting the entire flume length. Erosion events always started in the form of a bedload sheet at the upstream end of the flume (described above), before propagating downstream. The process of armour destruction was very similar to what has been described in other reports (Iseya and Ikeda, 1987; Kuhnle and Southard, 1988; Recking et al., 2009): by reducing bed roughness, fine materials transported in bedload sheets increase the transport efficiency of coarse material, which contributes to destabilizing the nearby local armour by impacting the gravel at rest, and the armour is destroyed from place to place in the downstream direction. This phenomenon is illustrated by a video provided as supplementary material to this paper (armour destruction in a bimodal material corresponding to Run 1 and Figure 5 in Recking et al., 2009). After each large erosion event, the resulting bed surface was composed essentially of fine sediments; but this bed state was always ephemeral, replaced by new clusters and bed armouring (new aggradation). Substantial bed erosion is shown in Figure 5, featuring the three associated bed states (from left to right): complete armour, transitional bed (armour destruction propagating from upstream to downstream) and finally, eroded bed. The slope change resulting from such erosion was always very high (3% in Figure 5). The associated GSDs are shown in Figure 6 and vary a great deal around the mean GSD (input mixture).

Slope fluctuations

The amplitude of slope fluctuations is larger in Run 2 (between 11.3 and 14.5%) than in Run 1 (between 11.8 and 12.8%): this was an unexpected result, in contradiction with Recking et al. (2009), who observed a reduced amplitude of slope fluctuations with increasing flow conditions. However, in the experiments of Recking et al. (2009) the flow conditions were such that kinetic sorting was always possible, as in Run 2. The absence of kinetic sorting in Run 1 explains permanent armour, with no large bed erosion events, and small bed slope fluctuations limited successive local bed erosion at the flume upstream end only.

The slope signals measured for Run 1 and Run 2 also differ in shape. Both signals comprise a high frequency associated with bedload sheet production at the flume upstream end, but in Run 2 these high frequencies are superposed on low frequencies, absent in Run 1. The low frequencies of Run 2 in Figure 2 actually correspond to rapid episodes of slope increase and decrease (peaks). We identified five of these peaks in Figure 2, noted E1 to E5. Figure 7 plots the frequency distributions for the mean bed slope S and each local slope S₁₂ (upstream) to S_7v (downstream) for Run 2. Whereas the local slopes could vary between 8% and 17%, the resulting mean bed slope is constrained between 11 and 15%. This result is possible only if there is no correlation between the local slopes, which was confirmed by a correlation matrix showing a maximum correlation coefficient of 0.3 between two subsequent sections of the channel, and a decreasing coefficient when the distance between the two sections was increased. However, this absence of correlation was not true during the peaks preceding the large bed erosion events (peaks E1 to E5, Figure 2): each of them coincided with simultaneous maximum local slopes (between 13 and 17%), as indicated in Table 2.

We used wavelet analyses to confirm these observations since they have recently been shown in several studies to be powerful tools to analyse bedload transport measurement time series (Singh et al., 2011; Singh et al., 2012). The advantage compared with the classic Fourier transform is that it has an easy space/time interpretation, showing where and at which time scale/frequency the energy of the signal is concentrated. We used the classical discrete decomposition using the Daubechies (1992) wavelet family. Three vanishing moments were chosen, since they led to optimal significant decomposition level numbers with regard to the run lengths, namely seven and eight decomposition levels for Runs 1 and 2, respectively. These levels correspond to the temporal scales over which the temporal signal is analysed, so that the magnitude of the signal as a function of these scales is simply the discrete wavelet spectrum returned by the transformation (Figure 8, top). The fraction of variance at each time scale has also been used as an average of the wavelet spectrum similar to the Fourier spectrum, so as to highlight the predominant time scales in each of the two runs (Figure 8, bottom).

Figure 8 presents the results for the mean bed slope analyses. In the bottom subplots, the x-axes represent time and the y-axes the percentage of variance at each time scale considered (the residual variance, i.e. the part of the signal not taken into account by decomposition, is 5 and 14% for Run 1 and Run 2, respectively, which is very small and, hence, considered satisfactory). Figure 8 (bottom) indicates that the maximum energy corresponds to the medium frequencies for Run 1 and Run 2, but that Run 2 is clearly characterized by more energy in the low frequencies (time scales ~2–5 h), which corresponds to the large bed erosion events in Run 2. The peak events in Run 2 can be traced back on the wavelet spectrum (Figure 8, top). Where a Fourier transform would have indicated the existence of low frequencies, the wavelet analysis shows that the peaks not only correspond to the low frequencies, but they also correspond to a concomitance of maximum energy in several levels of the decomposition. This is consistent with the concomitance of maximum slopes attained in each part of the flume before erosion, as described above.

Figure 9 presents the wavelet decomposition for the upstream and the downstream part of the flume. It shows that the low frequencies are more pronounced for the local downstream slope, implying that the complex slope fluctuation would reflect the behaviour of the different parts of the canal eroding and aggrading at different rates.

We conclude from the above results that high rates of bedload sheet production in the flume upstream part impose its high frequencies on the slope signal; these bedload sheets progressively feed the downstream part of the flume, which aggrades slowly, imposing its low frequencies on the slope signal. In the long term, substantial erosion occurs when aggradation is complete for the entire bed (simultaneous maximum local slopes for each part of the flume).

Bedload fluctuations

Figure 10 plots the fractional transport measured at the flume outlet for Run 1 and Run 2. For the sake of clarity, the signal was reduced to 4 h and to two sediment classes considering diameters above and below 3 mm (D₅₀), and different scales were used for the y-axes (but same time scale on the x-axes).

Figure 10 shows that in Run 1, bedload fluctuations took the form of individual peaks, with a short time duration not exceeding a few minutes. Each peak corresponded to individual cluster destruction immediately upstream of the outlet section, and no clear correlation can be seen with the slope. In Run 2, transport events were longer than in Run 1 (up to 0.5 h) and are typical of what has already been described for bedload sheets (Kuhnle and Southard, 1988): (1) bedload transport correlates well with slope fluctuation (it drops to almost zero each time the bed aggrades and is maximum during slope decrease), and (2) coarse gravel (D > 3 mm) usually peaks shortly before the smaller materials (D < 3 mm) at the beginning of each pulse, which is consistent with a gravel peak controlled by impact effects at the leading edge of the migrating bedload sheets. These peaks correspond to the passage of bedload sheets described above and are shown in the videos provided as Supplementary Material.

Wavelet decomposition (shown in Figure 11 for Run 2) indicates similar results for both Run 1 and Run 2: high frequencies (almost no information in levels higher than 4) and an exponential shape of the variance ratios as a function of the time scale. Actually, the output solid discharge in Run 1 was controlled essentially by step–pool destruction occurring at the downstream end of the flume, and producing a quasi-binary response of the transport signal (zero or peak transport). This explanation also holds for Run 2, where no correlation could be observed between the transport intensity and amplitude of the slope fluctuation: transport rate in Run 2 alternates between near zero and a maximum value (associated with transport in bedload sheets as shown in Figure 10 and in the videos provided as Supplementary Material). The wavelet analysis also showed a negative correlation between the output solid discharge and the mean bed slope. In other words, an eroding bed was associated with a peak solid discharge and inversely, an aggrading bed was associated with a low solid discharge, which seems intuitive.

Comparison with other observations made on steep slopes

The results presented above should be placed within the general context of research dealing with steep slope flume experiments. The Run 1 experiment is close to many other step–pool experiments (Rosport and Dittrich, 1995; Church and Zimmerman, 2007; Curran, 2007, 2012), showing poor mobility of coarse sediments (τ^*/τ_c^* < 1). On the other hand, Run 2 was characterized by constant sediment feeding, a natural mixture comprising a high percentage of sand (nearly 30%), a transport stage in the range 1 < τ^*/τ_c^* < 2 and a long duration (92 h): it corresponds to an experimental condition that has rarely (if at all) been considered (except in Recking et al., 2009, but with a strictly bimodal mixture).

Many steep slope experiments were conducted with a high shear stress ratio τ^*/τ_c^* > 2 (Smart and Jaeggi, 1983; Rickenmann, 1991; Recking et al., 2008a). It was shown that such flow conditions produce near equal mobility of the transported material and are associated with no bedload or slope fluctuations (Recking et al., 2009). On the other hand, step–pool flume experiments, all conducted on steep slopes with poorly sorted materials, did not report such large fluctuations, with superposed periods and bedload sheet production (Grant and Mizumura, 1992; Curran and Wilcock, 2005a; Comiti et al., 2009; Zimmermann, 2009). This can be explained by many factors including insufficient duration (most experiments published in the literature lasted less than 10 h), absence of constant sediment feeding and coarsening of the sediment mixture (the proportion of sand of our mixture was 30%). Only the experiments from Suzuki et al. (1998) and Ikeda and Iseya (1988), conducted on steep slopes with a large grain size distribution, produced high levels of bed erosion similar to that described here; however these experiments were too short (a few hours) to measure the complex fluctuation pattern described in Run 2.

Many reports have been published during the last two decades concerning the hydraulics, transported sediment and morphodynamics of step–pool streams (exhaustive reviews were reported by Chin and Wohl, 2005; Church and Zimmerman, 2007; Comiti and Mao, 2012). Because of the difficulty of measuring exceptional flow events capable of moving large boulders composing the steps, most of these studies have investigated the dynamics associated with relatively low flow conditions (τ^*/τ_c^* < < 1 for boulders); also considering the short time scale for field observation, the phenomenon described here could not be observed in these studies.

Mechanisms responsible for fluctuations

Some studies have proposed a statistical description of bedload fluctuations at the grain scale (Ancey et al., 2006; Ancey et al., 2008; Heyman et al., 2013), but none of the experiments conducted with nearly uniform material (Smart and Jaeggi, 1983; Rickenmann, 1991; Recking et al., 2008a) has described the large bedload and slope fluctuations we are considering here, whatever the Shields stress τ^*/τ_c^* ratio used. This was an argument put forward by Recking et al. (2009) for concluding that these fluctuations are the consequence of grain sorting only. Given the imposed flow conditions (fixed inlet discharge) and kinetic sorting, how can we explain the existence of quasi-periodical and finite amplitude fluctuations?

Local bed shear stress controls sediment mobility. Unfortunately, it was not possible to measure precisely the bed shear stress or the hydraulics in these steep slope experiments with small relative depth. In addition, even though the runs were performed with a constant discharge, the slope, grain size distribution and bedload transport always changed with time. As a consequence, the fractional transport could not be matched appropriately to the bed shear stress. However, shear stress alone does not explain the fluctuations observed because whatever the bed state and slope, stopping the sediment feeding always produced a rapid selective transport of fine material, as well as coarsening and stabilization of the bed surface. Only the association with sediment forcing leads to large fluctuations.

Using the Bagnold equation (Bagnold, 1966), Recking et al. (2009) showed that when the feeding rates (Q, Qs) are maintained constant, energy conservation imposes that the product of transport efficiency e and slope S is also maintained constant:

(5)

This function supports the existence of periodical slope and bedload fluctuations, as the efficiency term e (which depends on the GSD of the bed surface and of the transported sediments) changes throughout the experiment under the effect of grain sorting. The value of e is maximum in bedload sheets, where the transport of coarse material is very efficient. Actually, higher mobility of coarse gravels when transported within a mixture was demonstrated nearly a century ago by Gilbert (1914). Other later studies also demonstrated that injection of fine sediments can increase gravel transport efficiency for a given flow condition (Raudkivi and Ettema, 1982; Ferguson et al., 1989; Wilcock et al., 2001; Curran and Wilcock, 2005b) or destabilize a previously armoured and immobile bed (Jackson and Beschta, 1984; Cui et al., 2003; Venditti et al., 2010).

On the basis of Equation 5, Recking et al. (2009) proposed a general scenario for mechanisms leading to fluctuations: during aggradation, kinetic sorting produces an efficient downward migration of fine materials and bed surface armouring. For the given feeding rates, the low transport efficiency leads the slope to aggrade with respect to Equation 5. Inevitably, because aggradation cannot be infinite, it stops when a critical (maximum) slope is attained (or, considering the constant flow, when the flow power is strong enough for transporting the injected sediments). Then the injected fine material is no longer embedded (kinetic sorting stops) and is entirely transported once the armour interstices are filled. Reducing the bed roughness increases the transport efficiency of coarse material, and the slope erodes for this new efficiency following Equation 5. In this process, displacement of coarse material also contributes to destabilizing the nearby local armour by impacting the gravel at rest (Iseya and Ikeda, 1987; Kuhnle and Southard, 1988). This phenomenon is reproduced from place to place in the downstream direction, as long as the local slope is higher than required for the given transport rate efficiency. After erosion, the bed surface is composed of fine sediments essentially, but the reduced flow competence associated with the reduced slope rapidly leads to reduced mobility of coarse gravel (which usually is captured in small sediment waves), and the formation of new clusters, which progressively evolve into a new bed armour. From this point, the low transport rate efficiency leads to new bed aggradation, following Equation 5. As a consequence, for a given sediment mixture, the constancy of the eS product (Equation 5) maintains the slope fluctuations between two limiting values.

The above conceptual model explains local bed aggradation and erosion, as well as bedload sheets production. Extending it to the entire flume length explains the multi-scale bed erosions and associated superposed periods of the slope signal (Recking, 2013b).

Consequences for field applications

The authors acknowledge that what has been presented here was produced in a flume, and that more work is needed to confirm these results, especially in the field where the time scales necessary for observing similar processes would be much longer. However, more and more people live in mountain areas, and because catastrophic events due to mountain streams are generated not only by water discharge, but also by the transported solids, we believe that it is important not to neglect some of the aspects developed here.

The first consequence of these experiments is that evaluation of the degree of bed surface armouring may not be sufficient to predict the bed stability of mountain streams. Indeed, both Run 1 and Run 2 produced armoured beds; both runs provided efficient transfer downstream of the injected material, but over the long term only Run 1 had a permanent armour, whereas Run 2 (allowing kinetic sorting) produced substantial erosion.

The second consequence of these experiments is that a catastrophic flood event may not be necessary to produce significant bed erosion: such erosion was obtained in Run 2 with a constant flow and τ₈₄^*/τ_c84^* ~ 1. Large events are known to control the morphology of mountain streams by destroying chain forces and allowing the availability for transport of finer sediments stored in the bed. These events able to move boulders are rare, with return periods estimated to be ~20–50 years (Whittaker and Jaeggi, 1982; Grant et al., 1990; Chin, 1998; Lenzi, 2001; Turowski et al., 2009; Molnar et al., 2010; Recking et al., 2012). Between two large events, the bed stores materials coming from the watershed. The experiments reported herein suggest that the channel will behave differently depending on whether the hydrological regime permits flows such that on average τ₈₄^*/τ_c84^* < 1 or τ₈₄^*/τ_c84^* > 1. If τ₈₄^*/τ_c84^* < 1 the bed will store material until the next large event occurs. If τ₈₄^*/τ_c84^* > 1, the bed will aggrade (with a velocity depending on the rate of the sediment supply) and substantial erosion can occur in response to internal bed stability criteria, even for low to moderate flow events.

A third consequence of these experiments is that for a given slope, sediment mixture and flow condition, a large set of bedload transport rates can exist. When fine sediments are mobilized in the bedload layer, they smooth the bed and enhance the transportation of coarse sediments. These effects were observed in channels connected to active hillslope processes (Recking, 2012; Recking et al., 2012) or when fine sediments stored in the bed are released in the recovery period following a large flow event (Gintz et al., 1996; Lenzi, 2001; Lenzi et al., 2004; Turowski et al., 2009). Here we demonstrated that because fine sediments cannot be stored indefinitely in the bed by kinetic sorting, they can also periodically participate in transport and produce peak bedload transport.