Microplastic and natural sediment in bed load saltation: material does not dictate the fate

Microplastic (MP) pollution is a well document threat to our aquatic and terrestrial ecosystems, however, the mechanisms by which MPs are transported in river flows are still unknown. The transport of MPs and natural sediment in aquatic flows could be somewhat comparable, as particles are similar in size. However, it is unknown how the lower density of MPs and their different material properties impact their transport dynamics. To answer this, novel laboratory experiments on bed load saltation dynamics in an open-channel flow, using high-speed camera imaging and the detection of 11,035 individual saltation events were used to identify the similarities and differences between spherical MPs and spherical natural sediments transport. The tested MPs varied in terms of size and material properties (density and elasticity). Our analysis shows that the Rouse number accurately describes saltation length, height, transport velocity and collision angles equally well for both MPs and natural sediments. Through statistical inference, the distribution functions of saltation trajectory characteristics for MPs were analogous to natural sediment with only one sediment experiment (1.4 % of cases) differing from all other plastic experiments. Similarly, only nine experiments (9.3 % of cases) showed that collision angles for MPs differed from those of natural sediment experiments. Differences observed in terms of restitution become negligible in overall transport dynamics as turbulence overcomes the kinetic energy lost at particle-bed impact, which keeps particle motion independent from impact. Overall, spherical MP particles behave similarly to spherical natural sediments in aquatic environments under the examined experimental conditions. This is significant because there is an established body of knowledge for sediment transport that can serve as a foundation for the study of MP transport.


Introduction
Plastic polymers have played an influential role in shaping human activities since the 50s -70s of the past century (Andrady 2017;Geyer 2020) but their widespread use has also resulted in environmental contamination (Napper et al. 2020;Andrade et al. 2021;Lofty et al. 2022).
Toxicologists are increasingly concerned about the potential harm caused by plastics to human health (Dick Vethaak and Legler 2021; Koelmans et al. 2022) and just recently, plastics have been found in human blood (Leslie et al. 2022), lung tissue (Jenner et al. 2022) and even the placenta (Li et al. 2018;Amereh et al. 2022;Ragusa et al. 2022).Plastics can also have detrimental effects to ecosystems on land (Browne et al. 2013;Wu et al. 2022) and water This article has been accepted for publication in Water Research.Please cite the published version: Lofty, J., Valero, D., Wilson, C.A.M.E., Franca, M.J., & Ouro, P. (2023).Microplastic and natural sediment in bed load saltation: material does not dictate the fate.Water Research, 120329.https://doi.org/10.1016/j.watres.2023.120329This study comparatively investigates the bed load saltation motion of spherical MPs and spherical natural sediments, which is the dominant mode of bed load transport (Wiberg and Smith 1987;Sekine and Kikkawa 1992).The material-depending variables were isolated and their influence on the transport of MPs, in terms of bed load saltation motion, was examined.
Three different plastic materials in two sizes were considered, and their transport dynamics in bed load saltation were compared to experiments with amber particles, which possess similar mechanic properties to natural sediments.For direct comparability, MPs and amber were the same shape and size and experiments were undertaken using same flow conditions, methodology and analysis protocols; thus, reducing potential sources of bias.Hydrodynamic experiments were conducted in an open channel flume with controlled discharge and velocity, with discrete particle movement tracked via a high-speed camera to characterise particle behaviour.Therefore, conclusions on differences and similarities between MPs and natural sediments, uniquely from the material comparison perspective, could be drawn.

Experimental setup
Experiments were conducted in a 10 m long, 0.3 m wide, 0.3 m deep open channel flume with a longitudinal slope of 1/1000 (Figure 1).Rough sediment beds, made of quartz sand particles glued to plastic boards with thickness of one particle, covered the floor of the flume over the first 6 m of length, thereafter, the flume was floored with a smooth metal plate.Two sets of upstream roughness boards were tested: a first set consisted of uniformly graded sand particles with sand roughness (  ) of 1.86 mm (based on median particle diameter d50) and geometric This article has been accepted for publication in Water Research.Please cite the published version: Lofty, J., Valero, D., Wilson, C.A.M.E., Franca, M.J., & Ouro, P. (2023).Microplastic and natural sediment in bed load saltation: material does not dictate the fate.Water Research, 120329.https://doi.org/10.1016/j.watres.2023.120329standard deviation of the grain size (  = √ 84  16 ⁄ ) of 1.22, while a second set of boards consisted of uniformly graded sand particles with a   of 2.76 mm and   of 1.24.
Figure 1.Experimental setup, instrumentation and particle coordinate system.

Flow conditions
Six steady near-uniform flow conditions for the two sets of fixed roughness beds were established in the flume.Table 1 presents their discharge (), flow depth (), depth-averaged velocity (), Froude number (F = /√), Reynolds number (Re =   ⁄ ), where  is the kinematic viscosity of the fluid, shear velocity ( * ) based on the log-law equation and detailed velocity measurements and friction Reynolds number (Re * =  *   /) .Flow was recirculated using a pump and measured using an electromagnetic flowmeter (± 0.3%), while the flow depth was controlled by a weir gate.Uniformity of the flow was tested to determine whether channel flow accelerations were negligible and had little influence on the transport of particles.The streamwise mean velocity profile was computed through spatial and time averaging.These profiles are presented for all flow conditions in Figure S1 for completeness.The law of the wall for transitionally rough beds (based on Re * in Table 1) was fitted to the velocity profiles, allowing the estimation of the shear velocity  * (Pope 2000, Eq. 7.121): where  is the time-averaged streamwise velocity,  is the distance from the bed and  is a constant.The constant  was determined through Fig. 7.24 of Pope (2000), which yielded values of 8.5 for channel beds that were considered hydraulically rough (Re  For comparison, amber particles were also used in the experiments, which have been considered as a proxy for low density natural sediments in riverine transport experiments (Shields 1936;Rouse 1939) (Table 2).2), but are analogous to common natural sediments formed from rocks (Table 3).
Table 3. Material properties of amber compared to other common natural sediments (Ji et al. 2002;Cardaerlli 2008).velocity of each MP and amber particle was then calculated as the mean velocity of the falling particle.Εach test was repeated five times for each particle.
Given the particle settling velocity () and the shear velocity ( * ) for each flow condition, the Rouse number  can be calculated as (Rouse 1939): where = 0.41 is the von Kármán constant and β is a parameter that adjusts the assumption of parabolic eddy diffusive which is assumed to be equal to unity. determines the shape of the Rouse profile, which is a theoretical concentration profile of particles in turbulent flows and is also used to determine the mode at which particles are transported by the flow.A value of  > 2.5 usually indicates a particle is transported as bed load, while a value of  between 0.8 -2.5 indicates that a particle is in suspended mode of transport (Dey 2014;Cowger et al. 2021).In this study, Rouse numbers ranged between 5 to 24, which suggests that all particles should predominately be in bed load transport for all flow conditions analysed.For completeness, Shields numbers ranged between 0.027 -0.512 and Reynolds particle numbers ranged between 66.3 -161.1 for all flow conditions and particles used.

Experimental procedure
Figure 1 shows the experimental setup.Between 15 to 20 particles of the same diameter and polymer (Error!Reference source not found.),for every combination of bed roughness and flow condition (Table 1), were manually released in the upstream region at bed height at the centreline of the flume, one at a time.The particles travel for three meters before moving into the observation window, where the high-speed camera recorded at 90 frames per second captured the motion of the particles with a field of view of 0.5 × 0.18 m 2 .The release point was deemed far enough upstream so that the particle movement captured by the camera was unaffected by the initial conditions or any disturbance developed at the release point, and the particles achieved a steady bed load motion before passing through the observation window.
An example video of a 5 mm PA particle moving over a 1.86 mm roughness bed is provided within the supplementary material.At the downstream end of the experimental area, MPs were collected by a sediment trap composed of a mesh sheet (Figure 1).11,035 individual saltation events were observed throughout all experiments.

Video post-processing and saltation characterisation
Video recordings were analysed using open-source software in Fiji (ImageJ2) (Gulyás et al.

2016)
, where the coordinates (xp, zp) of the centroid of the particles moving through the observation window were extracted at each frame of the recorded videos.The coordinates were then used to calculate the mode of particle transport of the particles: rolling/sliding, saltation and suspension.For the saltation events, particle trajectory characteristics and particle-bed collision characteristics were determined.Examples of a 5 mm PA particle trajectory on a 2.76 mm roughness bed, under three different flow condition are shown in Figure 3 and the different transport modes (rolling/sliding, saltating or suspended transport) are highlighted.
The different modes of particle transport were determined by an algorithm coded in R statistical software (Lofty 2023) implementing the following algorithm.Looking at a particle, the trajectory from impact to impact with the bed is considered an individual event.Saltation events are events in which the centroid of the particle exceeded a height of   / away from the average height of the sediment bed.If the particle did not reach such heights during the event, then it is classified as rolling/sliding, and involves the particle moving majorly in contact with the bed.Conversely, suspension events were determined as those in which the particle is reverted upwards during the falling limb of the trajectory, i.e., the particle moves upwards again in the middle of the falling trajectory and before the collision with the bed (Figure 3C, green trajectory), which can only be driven by turbulent forces that keep the particle in suspension (Abbott et al. 1977).To characterise the trajectories of the saltation events in each recording, the saltation length (  ) was calculated as the distance between two successive saltation collisions with the bed, the saltation height (  ) was calculated as the maximum height of the saltation event relative to the height of the bed, and the saltation transport velocity (  ) was calculated as the saltation length divided by the saltation event duration (Figure 4A).Both   and   were made dimensionless by the particle diameter , while   was made dimensionless by the shear velocity  * .
To characterise the collision dynamics, the average inwards (αin) and outwards (αout) collision angles (prior and posterior), relative to the bed horizontal, were calculated using the xp, zp coordinates immediately before and after the particle impact (Figure 4B).Similarly, the streamwise inwards (  |  ) and outwards (  |  ) velocity and the vertical inwards (  |  ) and outwards (  |  ) velocities were also calculated.The inwards (  |  ) and outwards (  |  ) collision velocity magnitude was calculated as: (3) Where   and   are streamwise and vertical velocities of the particles, calculated through central differences.

Bed load mode of transport
A total of 1,665 individual MP and amber particle runs were recorded, with multiple modes of bed load (rolling/sliding and saltation) and suspended load transport observed.Figure 5 shows the percentage of time that particles spent either in saltation or in rolling/sliding mode of transport.All particles were mobile for all experimental conditions, which meant no particles were in repose, while suspension events remained below 5% for all particles.Given the low correlation between mode of bed load transport and the Rouse number ( 2 = 0.223, Figure S2), a dimensional analysis approach was undertaken to evaluate the relationship between the different physical particle properties (,   , ), fluid motion ( * ) and bed properties (  ) and the percentage of the MP and amber particles transported as bed load.Figure 5 shows that a the second, accounts for how the bed roughness impacts bed load transport dynamics and it is defined by the relative roughness to the size of the particle (   ⁄ ), similar to conclusions about hiding/exposure effects made by Waldschläger and Schüttrumpf (2019b) for the initiation of motion for MPs.
The improved fit of this parameter suggests that the mode of transport is not only determined by flow and turbulence (Rouse-based considerations) but also by bed roughness and MP particle size.Hence, particles are more likely to saltate as the ratio between the roughness and MPs diameter decreases or as the strength of turbulence acting upon the particle increases.The collapse of transport mode in Figure 5 for all particles indicates that material properties did not significantly influence this parameter.and less than 5% of all particles were in suspension mode of transport, thus omitted from this analysis.

Particle trajectory characteristics
Figure 6A-C shows the main descriptors of the particle trajectories (saltation length   , saltation height   and saltation transport velocity   ), which are computed as an average for each material for a certain flow condition and bed roughness.Lower Rouse numbers, indicating higher relative turbulent forces, yielded higher values of   ,   and   , suggesting a stronger particle-flow coupling.For the same flow condition, particles with a lower settling velocity (lighter and/or smaller particles) can reach higher regions in the water column with higher   It is observed that with decreasing Rouse number, distributions show larger dispersion (also observable in Figure 6A-C), regardless of the material, likely due to the relatively stronger diffusive effect of turbulence upon the particles' trajectory.
Conversing theories on sediment trajectory probability density functions have been previously proposed;   and   have both been suggested to fit a Gaussian (Hu and Hui 1996b) and Gamma (Lee et al. 2000;Lee et al. 2010;Roseberry et al. 2012) distribution, while   has been proposed to fit an exponential (Fathel et al. 2015;Shim and Duan 2019), which is also positively skewed, and Gaussian distributions (Hu and Hui 1996b;Lee et al. 2000;Lee et al. 2010).In this study, two functions were considered to describe the frequency distributions of   ,   and   across particle materials: Gaussian (N), as data may fall symmetrically around the mean, and Gamma (Γ), to take into consideration the potential skewness, or long tails, that data may show.This may be expected since particle trajectory characteristics should be physically bounded to zero; e.g., no negative saltation jump lengths or heights should be expected.
A one-sample Kolmogorov-Smirnov (K-S) test was performed to identify whether the sampled data is drawn from either a Gaussian or a Gamma function (Wilks 2006).For that purpose, a Gaussian and a Gamma function are fitted to the data via maximum likelihood estimation.Provided that the probability density functions are identified for MP and amber particles' saltation characteristics, the following question can be addressed: are MPs' trajectories different from those of natural sediments (using amber particles as a proxy)? Figure 6G compiles the results of the K-S tests for each trajectory characteristic, flow condition, roughness configuration and particle material.Each point in Figure 6G indicates if the data distribution is consistent with a Gaussian and/or Gamma function (p > 0.05).
Figure 6G can be seen as a similarity matrix in which the points in the same cell indicate that similar probability density functions may be expected for the different particles under the same flow conditions and bed roughness.For instance, considering particle velocity Up, for the lower   bed and a shear velocity of 0.0257 m/s, Gaussian and Gamma distributions can explain the data of all types of particles analogously; however, looking at the distribution of saltation height Hp, we see that only the Gamma function explains this data for all materials, while the Gaussian distribution only explains the amber particles.Altogether, it can be observed that distribution From visual inspection of Figure 6, no apparent difference in behaviour can be observed between MPs and amber particle's trajectory characteristics from an averaged ( and   , it is concluded that a power law function minimised relative bias (Bennett et al. 2013), the most accurate across all materials (Table S1).For completeness, the power law functions are included for Figure 6A-B.

Saltation trajectory shape
The shape of the saltation trajectory for particles in bed load is asymmetrical, with the falling limb being larger than the rising limb (Hu and Hui 1996a;Lee et al. 2000).Results suggest that  2 ≈ 1.5 1 and that the ratio of  2 / 1 is independent of the Rouse number, particle material and bed roughness, indicating that the shape of saltation trajectory is alike between MPs and amber particles.The maximum trajectory height (  ) is roughly 40 % of the total trajectory length, which is consistent with previous studies for natural sediment (Hu and Hui 1996a;Lee et al. 2000).The shape of the particle saltation trajectories can also be evaluated by identifying the relationship between the average   and   as shown in Figure 7A.  grew linearly with   , with both increasing with decreasing Rouse number.This occurred for both MPs and amber particles.For visualization purposes, Figure 7B illustrates the trajectory ranges observed for 5 mm MPs and amber particles trajectories, accounting for the average minimum, maximum and values for   and   .It is seen that the amber particles saltation is longer and higher than the MP particles, which is explained by the larger  values, mainly caused by the smaller settling velocity of the amber particles.

Collision angles
Particle collision characteristics are key features of particle saltation as they provide information on the energy loss during impact and rebound dynamics (Zeeshan Ali and Dey 2019; Pähtz et al. 2020).Figure 8A shows the average inwards collision angle (αin) against the outwards angle (αout) for the MP and amber particles, as defined in for average angles).On average, αout is larger than αin for all particles, flow conditions and bed roughnesses, suggesting that particles are directed upwards after collision.that, αout has larger dispersion compared to αin, likely to be a result of particle impact with a heterogeneous roughened bed, which leads to a wider range of outward angle possibilities.
Conversely, αin is mostly a consequence of the shape of the particle trajectory that was found to be relatively uniform across flow conditions (Section 3.3).This difference in distribution between αin and αout angles is similar to results found by previous experimental research for natural sediments (Lee et al. 2000).
A similar analysis to Section 3.2 was undertaken to identify whether sampled αin and αout data is drawn from either a Gaussian or a Gamma function using K-S tests, and if there are differences between probabilistic descriptors of plastic and natural sediments.Figure 8D compiles the results of the K-S tests for αin and αout for each flow condition and bed roughness, showing where data is consistent with a Gaussian and/or Gamma distribution (p > 0.05).The results show that distribution functions of the collision angle from amber experiments were analogous to the plastic materials for 57 % of all cases, with only nine cases (9.3 %) differing from all comparable plastic experiments.
At lower Rouse numbers, the particleflow coupling is stronger and a particle reaches higher water levels in the water column, with larger flow velocities, thus is carried by the flow for a longer distance and this elongates the particle trajectory, causing a flatter αin and αout angles.
Based on the observations in Figure 8E and F, αin and αout appear to be well described by the Rouse number.To quantify their relationship, a linear model is provided that is fitted to the data using least squares error minimisation and keeps relative bias and uncertainty (Bennett et al. 2013) similar across materials (Table S2) and is shown in Figure 8E-F for completeness.A analysis similar to the one presented in Section 3.2 and 3.4 was undertaken to identify where

Restitution coefficient
Provided that part of the inward velocity is lost during the particle-bed impact, it is convenient to define a restitution coefficient (), which can be also defined in terms of its streamwise (  ) and vertical (  ) components: The restitution coefficient is directly connected to the loss of kinetic energy during impact (Schmeeckle et al. 2001;Zeeshan Ali and Dey 2019).Figure 10 shows the average restitution coefficients by flow condition in A) the streamwise direction   , B) vertical direction   and C) magnitude  for all particles, plotted against the Rouse number .Values of  are lower than 1 outlining that kinetic energy is lost during particle-bed collision, while values of   are larger than 1, suggesting that kinetic energy is transferred from the streamwise to the vertical direction during collisions, as also indicted by results from the αin and αout collision angles (Figure 8A).Values of   ,   and  for amber particles are similar to previous studies observations for natural sediment (Hu and Hui 1996a;Niño and García 1998).to amber particles at impact, which have higher Young's modulus than the MP particles (Table 2) and remain more elastic.This study presents the results of novel laboratory experiments on bed load saltation dynamics using high-speed camera imaging and the detection of 11,035 individual saltation events to identify the similarities and differences between bed load transport dynamics of MPs and natural sediments.Our findings support the following conclusions: • Saltation trajectory characteristics of MPs are analogous to natural sediments, as distribution functions for MPs were the same as natural sediment with only one amber experiment (1.4 % of cases) differing from all other plastic materials (Figure 6).In all cases, the Rouse parameter could explain saltation length, height and transport velocity equally for all materials tested.
• Inwards and outwards collision angles were well described by the Rouse number, with negligible material influence (Figure 8).Only nine experiments (9.3 % of cases) showed that distribution functions of impact angles for MPs differed from all natural sediment experiments.
• Differences in terms of restitution become negligible in overall transport dynamics as turbulence outweighs the kinetic energy loss during particle-bed collisions and keeps particle motion independent from impact (Figure 10).
To conclude, spherical MP particles behave similarly to spherical natural sediments in aquatic environments, within experimental uncertainty.Potential differences, due to particle shape will need to be tested separately in future studies since shape-deviations from sediments are significant and have yet to be investigated.The findings of this study are important because there is a well-recognised body of literature in bed load transport (van Rijn 1984;Garcia 2008;Lofty, J., Valero, D., Wilson, C.A.M.E., Franca, M.J., & Ouro, P. (2023).Microplastic and natural sediment in bed load saltation: material does not dictate the fate.Water Research, 120329.https://doi.org/10.1016/j.watres.2023.120329 Ancey 2020) that can be directly applied to the description of MP transport in rivers within the range of flow conditions herein tested.

A
two-dimensional particle image velocimetry (PIV) system was used for water velocity measurements in the vertical plane for each of the uniform flow conditions and bed roughnesses.The PIV system employed a high-speed Baumer VLXT-50M.I camera, able to capture images of 2448 × 2048 px 2 in size at a sampling frequency of 140 frames per second, synced with a stroboscope via a wave generator.The camera was set at 4 m downstream of the flume inlet, in the observation window (Figure1) and captured images of 2000 × 700 px 2 (0.50 × 1.85 m 2 ) in size at a sampling frequency of 120 frames per second for an interval of 30 seconds.The images were analysed using the MATLAB open-source PIV software, PIVlab (Thielicke and Sonntag 2021).

Figure 2 .
Figure 2. Photographs of the 5 mm and 3 mm MP and the 5 mm amber particles used in experiments on roughened beds with   of A) 1.86 mm and B) 2.76 mm.

Figure 3 .
Figure 3. Trajectories of a 5 mm PA particle travelling over a roughened bed with a   of 2.76 mm at increasing shear velocity A)  * = 0.0238 m/s (P = 12.5), B)  * = 0.0288 m/s (P = 10.4) and C)  * = 0.0322 m/s (P = 9.27).Rolling/sliding, saltation and suspension events are highlighted in different Figure 4. A) Definitions of particle trajectory characteristics and B) Definitions of particlebed

Figure 5 .
Figure 5. Percentage of particles in saltation or rolling/sliding mode of transport against a modified Rouse number     * ⁄ , with the solid line indicating a linear fitting.No particles were in repose

Figure 6 .
Figure 6.Average saltation trajectory characteristics, in terms of A) saltation length   , B) saltation height   , and C) saltation velocity   plotted against the Rouse number .Error bars represent the

Figure 6 D
Figure 6 D -F shows the frequency distribution of   ,   and   , considering all individual Figure A-C) or frequency (Figure 6D-F) point of view.Therefore, a simple model based on  should be able to capture saltation dynamics, regardless of the particle material.Following careful consideration of four regression models (linear, power law, exponential and logarithmic functions), fitted to the data via least squares error minimisation for expected values of   , Figure 7. A) The average saltation length   plotted against saltation height   with point size Figure 4. Results indicate that αin ranged from 4.8 -15.5° whilst values of αout ranged between 12.2 -32.6° (minmax, Lofty, J., Valero, D., Wilson, C.A.M.E., Franca, M.J., & Ouro, P. (2023).Microplastic and natural sediment in bed load saltation: material does not dictate the fate.Water Research, 120329.https://doi.org/10.1016/j.watres.2023.120329

Figure 8 .
Figure 8. A) Inwards collision angle αin plotted against the outwards angle αout and respective kernel density plots for 5 mm particles for B) αin and C) αout for each flow condition  * and bed roughness   , using a kernel density bandwidth following suggestions of Silverman (1986).D) Summary of the K-S

Figure 9
Figure9shows the average inwards and outwards collision velocities in the streamwise (  |  ,
Figure 10.Restitution coefficients for PA, CA, POM and amber particles for the A) streamwise   , B) vertical   direction and C) magnitude .

Table 2 .
Properties of the MP and amber particles used in the experiments including their diameter ,
Valero, D., Wilson, C.A.M.E., Franca, M.J., & Ouro, P. (2023)drawn from either a Gaussian or a Gamma function using K-S tests, and whether those are different for MPs and natural sediments.For completeness, Figure S5 shows a compilation of the K-S test results for   |  ,   |  and   |  ,   |  , respectively, showing where data was consistent with a Gaussian and/or Gamma function (p > 0.05).The results show that the distribution functions of the collision velocities were analogous to all plastic materials in 90 % of cases for   |  and   |  , This article has been accepted for publication in Water Research.Please cite the published version:Lofty, J.,Valero, D., Wilson, C.A.M.E., Franca, M.J., & Ouro, P. (2023).Microplastic and natural sediment in bed load saltation: material does not dictate the fate.Water Research, 120329.https://doi.org/10.1016/j.watres.2023.120329and 66 % of cases for   |  and   |  , with only three cases from all collision velocities (1 %) differing from all corresponding plastic experiments.