betaFeedback

WorksPhysics and Astronomy

A co-spectral budget model links turbulent eddies to suspended sediment concentration in channel flows

Abstract

The vertical distribution of suspended sediment concentration (SSC) remains a subject of active research given its relevance to a plethora of problems in hydraulics, hydrology, ecology, and water quality control. Much of the classical theories developed over the course of 90 years represent the effects of turbulence on suspended sediments (SS) using an effective mixing length or eddy diffusivity without explicitly accounting for the energetics of turbulent eddies across scales. To address this gap, the turbulent flux of sediments is derived using a co-spectral budget (CSB) model that can be imminently used in SS and other fine particle transport models. The CSB closes the pressure-redistribution effect using a spectral linear Rotta scheme modified to include isotropoziation of production and interactions between turbulent eddies and sediment grains through a modified scale-dependent de-correlation time. The result is a formulation similar in complexity to the widely used Rouse's equation but with all characteristic scales, Reynolds number, and Schmidt number effects derived from well-established spectral shapes of the vertical velocity and accepted constants from turbulence models. Finally, the proposed CSB model can recover Prandtl's and Rouse's equations under restricted conditions.

Key Points

  • A suspended sediment concentration (SSC) equation for turbulent flows is proposed and tested.

  • The equation is derived from a co-spectral budget that accounts for energy distribution in all eddy sizes.

  • The effects of Reynolds number and a scale-dependent Schmidt number on SSC are explicitly described.

Introduction

In his classic treatise on sediment transport, Hans Albert Einstein (HAE) presented a definition of suspended sediments (SS) and the role of turbulence in maintaining suspension as follows [21]:

”The characteristic definition of a suspended solid particle is that its weight is supported by the surrounding fluid during its entire motion. While being moved by the fluid, the solid particle, which is heavier than the fluid, tends to settle in the surrounding fluid. If the fluid flow has only horizontal velocities, it is impossible to explain how any sediment particle can be permanently suspended. Only if the irregular motion of the fluid particles, called turbulence, is introduced can one show that sediment may be permanently suspended.”

This operational definition is now standard in textbooks and research articles alike [18, 26, 19]. Despite some 80 years of research, the dominant factors controlling suspended sediment concentration (SSC) in streams continue to draw interest due to its multiple connections to ecosystem benefits and water quality degradation issues [55, 49, 56, 17, 32, 31, 75]. High SSC can intercept photosynthetically active radiation necessary for sustaining submerged aquatic plants in lakes and rivers. The presence of high SSC is also related to eutrophication and corollary water quality issues [85, 41], clogging of gills of fish and other aquatic organisms, accelerating the denitrification process [48].

In certain cases, sediments provide necessary nutrients to aquatic plants and are of primary significance to sustaining nearshore ecosystems such as floodplains and marshes. Their role in element-cycling has been highlighted in several studies [50, 54] as well. Another issue is the connection between SSC and micro/nano-plastics in saline environments. Recent work has shown that SS can promote polystyrene nano plastics settling in the presence of saline conditions, prompting further interest in SSC distribution in natural waters (Y. Li et al., 2019).

Even in the most idealized flow condition with a balance between the gravitational settling flux and the vertical turbulent sediment flux, the description of SSC remains a recalcitrant problem. A model for the turbulent vertical flux is required and is often derived using Reynolds’ analogy [18] where eddies are assumed to transport momentum and SS similarly. This analogy was the cornerstone of the well-celebrated Rouse’s formula [67] that assumes sediment diffusivity is proportional to eddy viscosity. Since the early work of O’Brien [60], Prandtl and von Kármán [80], these analogies have spawned numerous theories and closure models for the mixing length [78, 58, 6, 7, 18]. However, these models make no explicit contact with turbulent eddies and their associated kinetic energy distribution in the vertical direction. It is precisely the scale-wise vertical turbulent kinetic energy component that maintains sediments in suspension [71, 52, 18] as noted by HAE.

The turbulent vertical flux of SS is directly modeled here from the spectrum of turbulent eddies thereby providing a new perspective on Reynold’s analogy, the multiple length scales involved in describing SSC, and the emergence of Reynolds, Rouse, Schmidt, and Stokes numbers when linking eddy viscosity with eddy diffusivity for SS. The role of the Reynolds number has been introduced in prior studies as a damping correction to the mixing length [77, 81, 57] whereas the Rouse number is operationally used in the classification of sediment load.

The proposed approach uses a co-spectral budget model (CSB) derived from an approximated Navier-Stokes equation in spectral form for the Reynolds stress and SS turbulent flux. It uses a spectral Rotta scheme modified to include the isotropization of the production term for the pressure decorrelation effect [40] and a Schmidt number effect similar in form to van Rijin’s bulk formulation [79] for linking the fluid and particle velocity decorrelation time scales, explicitly made here scale-dependent. The newly proposed formulation and a simplified solution derived from it are tested with several published experiments that span a wide range of flow conditions and grain properties (diameter and density). A comparison against the widely-used Rouse formula is featured and discussed.

Theory

Definitions and General Considerations

As a starting point to review models for SSC profiles in streams, a prismatic rectangular channel with constant width BB and bed slope SoS_o is considered. The flow is assumed to be steady and uniform with constant water depth HH and flow rate QQ. For small slopes, a balance between gravitational and frictional forces for a length segment Δx\Delta x along the flow direction xx yields

ρ(BHΔx)gSo=2τs(HΔx)+τo(BΔx),(1)\rho(BH\Delta x)gS_o = 2\tau_s(H\Delta x) + \tau_o(B\Delta x), \tag*{(1)}

where τs\tau_s is the side stress, τo\tau_o is the bed stress, gg is the gravitational acceleration, and ρ\rho is the fluid density. This expression can be re-arranged as

u∗2=τoρ=gHSo(1+2HBτsτo)−1,(2)u_*^2 = \frac{\tau_o}{\rho} = gHS_o\left(1 + \frac{2H}{B}\frac{\tau_s}{\tau_o}\right)^{-1}, \tag*{(2)}

where u∗u_* is the friction (or shear) velocity. For the case where τs=τo\tau_s = \tau_o, u∗2=gRhSou_*^2 = gR_hS_o with Rh=H(1+2H/B)−1R_h = H(1 + 2H/B)^{-1} being the hydraulic radius. However, in many SS laboratory experiments, the channel bed is covered with sediments whereas the channel sides remain smooth to permit optical access. This difference in roughness between sides and bed leads to τs/τo≪1\tau_s/\tau_o \ll1. This assumption can be combined with H/B≤1H/B \le1 usually selected to minimize secondary circulation to result in τo/ρ≈gHSo\tau_o/\rho\approx gHS_o. This approximation is adopted throughout. Fully turbulent flow conditions are also assumed to prevail so that the bulk Reynolds number Reb=UbH/ν>500Re_b = U_bH/\nu> 500, where ν\nu is the kinematic viscosity and UbU_b is the bulk or depth-averaged velocity given as

Ub=QBH≈1H∫0Hu‾(z) dz,(3)U_b = \frac{Q}{BH} \approx\frac{1}{H}\int_0^H \overline{u}(z)\,dz, \tag*{(3)}

where u‾(z)\overline{u}(z) is the mean velocity at vertical distance zz from the channel bed (positive upwards), and overline indicates ensemble-averaging usually determined from time averaging. For such a flow, the Reynolds-averaged mean continuity equation for SSC in steady and planar homogeneous flow at high RebRe_b yields [65]

∂C‾(z)∂t=0=−∂∂z[w′C′‾−wsC‾−Φ(z)],(4)\frac{\partial\overline{C}(z)}{\partial t}=0=-\frac{\partial}{\partial z}\left[\overline{w'C'}-w_s\overline{C}-\Phi(z)\right], \tag*{(4)}

where tt is time, C=C‾+C′C=\overline{C}+C' is the instantaneous volumetric SSC in the flow, primed quantities are the fluctuating component, ww is the instantaneous vertical velocity component with w‾=0\overline{w}=0 (assuming water is of constant ρ\rho), w′C′‾\overline{w'C'} is the turbulent vertical flux that requires a closure model, wsw_s is the terminal velocity of sediment grains, and Φ(z)\Phi(z) arises from particle inertia. In the regime where particle inertia is weak, to a leading approximation, Φ(z)\Phi(z) is given by [23, 65]

Φ(z)=τp[CDw′Dt]‾=τpC‾∂σw2∂z+τp[C′Dw′Dt]‾,(5)\Phi(z)=\tau_p\overline{\left[C\frac{Dw'}{Dt}\right]}=\tau_p\overline{C}\frac{\partial\sigma_w^2}{\partial z}+\tau_p\overline{\left[C'\frac{Dw'}{Dt}\right]}, \tag*{(5)}

where τp=ws/g\tau_p=w_s/g is a particle time scale, σw2=w′w′‾\sigma_w^2=\overline{w'w'} is the vertical velocity variance at zz and D(.)/DtD(.)/Dt is the material derivative (local and advective) along a fluid particle trajectory. The Φ(z)\Phi(z) is the sum of a turbophoretic effect that arises due to finite ∂σw2/∂z\partial\sigma_w^2/\partial z in inhomogeneous flows such as channels [64, 70, 36] and a turbulent concentration-vertical acceleration interaction terms. In equation 4, the overall significance of Φ(z)\Phi(z) at any zz depends on a local Stokes number St(z)=τp/τK(z)St(z)=\tau_p/\tau_K(z) where τK(z)=[ν/ϵ(z)]1/2\tau_K(z)=[\nu/\epsilon(z)]^{1/2} is the Kolmogorov time scale formed by the local turbulent kinetic energy dissipation rate ϵ(z)\epsilon(z) and ν\nu as reviewed elsewhere [10]. An associated length scale to τK\tau_K is η=(ν3/ϵ)1/4\eta=(\nu^3/\epsilon)^{1/4}, which is the Kolmogorov micro-scale representing eddy sizes impacted by viscous effects at zz. Upon defining the Kolmogorov velocity as vk=η/τKv_k=\eta/\tau_K, the Kolmogorov micro-scale Reynolds number Rek=vkη/ν=1Re_k=v_k\eta/\nu=1, meaning that both turbulence and viscous effects are equally important at scales commensurate to η\eta [74]. In the limit St→0St\to0, the particle vertical velocity is given by the sum of the local vertical fluid velocity minus wsw_s, and Φ(z)\Phi(z) can be ignored relative to the turbulent flux at zz, an assumption routinely invoked in operational models for SSC. To allow for a ’bulk’ Stokes number StbSt_b to be formulated, thereby facilitating comparisons across experiments, τK,b=(ν/ϵb)1/2\tau_{K,b}=(\nu/\epsilon_b)^{1/2} is proposed where εb\varepsilon_b is the over-all bulk dissipation rate in clear water. Thermodynamic considerations require that the work per unit mass per unit time to move clear water at UbU_b is (gSo)Ub(gS_o)U_b. For steady-state conditions (i.e. turbulent kinetic energy is stationary), this mechanical work produces turbulence that is then dissipated by the action of viscosity leading to an increase in the internal energy of the fluid. Hence,

εb=(gSo)Ub;τK,b=νεb;andStb=(wsg)−1τK,b−1.(6)\varepsilon_b=(gS_o)U_b;\quad\tau_{K,b}=\sqrt{\frac{\nu}{\varepsilon_b}};\quad\text{and}\quad St_b=\left(\frac{w_s}{g}\right)^{-1}\tau_{K,b}^{-1}. \tag*{(6)}

It is assumed that Φ\Phi is small and can be ignored when Stb≪1St_b\ll1 (although, more precisely, Φ\Phi can only be ignored when max⁡[Stb,St]≪1\max[St_b,St]\ll1). Another estimate of bulk Stokes number is St+=τp(u∗/H)St_+=\tau_p(u_*/H) ([28, 27]), where (H/u∗)(H/u_*) is presumed to represent an outer-layer eddy turnover time. Noting that gSo=u∗2/HgS_o=u_*^2/H, the two bulk Stokes numbers can related using Stb=St+(Reb)1/2St_b=St_+(Re_b)^{1/2}. A critique for using St+St_+ as a bulk Stokes number measure have been discussed elsewhere ([28, 65]).

With regards to the terminal sediment velocity, a simplified expression for wsw_s that recovers many prior formulae ([73, 31]) is used here and is given by (Cheng, [15])

ws=νds[(25+1.2ds2(ρs−ρρgν2))2/3−5]3/2,(7)w_s=\frac{\nu}{d_s}\left[\left(\sqrt{25+1.2d_s^2\left(\frac{\rho_s-\rho}{\rho}\frac{g}{\nu^2}\right)}\right)^{2/3}-5\right]^{3/2}, \tag*{(7)}

where ρs\rho_s is the sediment grain density (with ρs/ρ>1\rho_s/\rho>1), and dsd_s is the sediment grain diameter. This wsw_s is smaller than the Stokes settling velocity (wstw_{st})

wst=118gν(ρs−ρρ)ds2,(8)w_{st}=\frac{1}{18}\frac{g}{\nu}\left(\frac{\rho_s-\rho}{\rho}\right)d_s^2, \tag*{(8)}

except when wstds/ν≪1w_{st}d_s/\nu\ll1. The comparison between the two settling velocities is shown in Figure 1 for reference. Since wstw_{st} only applies to creeping flow past a sphere, equation 7 is used as it covers a wider range of wsds/νw_sd_s/\nu.

Figure 1. Comparison between the empirical sediment settling velocity wsw_s used here and the Stokes settling velocity wstw_{st} for different sediment to fluid density ratios. The one-to-one line is shown. The comparison between wsw_s and wstw_{st} for the data sets explored here is also featured as inset.

The mode of sediment transport is operationally related to wsw_s and some measure of the strength of turbulence based on bulk flow properties. One such measure is the Rouse number RR or ’unit’ Rouse number R∗R^* given by

R∗=1κwsu∗,R=1βR∗;(9)R^*=\frac{1}{\kappa}\frac{w_s}{u_*},\quad R=\frac{1}{\beta}R^*; \tag*{(9)}

where κ=0.41\kappa=0.41 is the von Kármán constant and β=Sc−1\beta=Sc^{-1} is an inverse turbulent Schmidt number (ScSc). The Rouse number is routinely used for classifying sediment load: R>R> 2.5 for bedload, 0.8<R<2.50.8 < R < 2.5 for SS, and R<0.8R < 0.8 for washload. To solve for C‾\overline{C}, models linking w′C′‾\overline{w'C'} to C‾\overline{C} as well as estimates for ScSc (and Φ\Phi, though this is ignored here) are required in equation 4, and those models are to be briefly covered.

Conventional Formulations and Revisions

Conventional approaches (including Rouse and O’Brien) for modeling SSC begin by ignoring Φ(z)\Phi(z) and employing a gradient-diffusion approximation (or some non-Fickian revision to it) given as

w′C′‾=−DsdC‾dz,(10)\overline{w'C'} = -D_s \frac{d\overline{C}}{dz}, \tag*{(10)}

where DsD_s is the sediment turbulent diffusivity. To estimate Ds(z)D_s(z), existing theories approximate Ds(z)D_s(z) by νt/Sc\nu_t/Sc or βνt\beta\nu_t, where νt\nu_t is the turbulent or eddy viscosity (νt/ν≫1\nu_t/\nu\gg1). When the mixing length hypothesis is further invoked to model νt\nu_t as a product of a characteristic length and velocity, it yields

νt=lo(lo∣du‾dz∣),(11)\nu_t = l_o \left(l_o \left|\frac{d\overline{u}}{dz}\right|\right), \tag*{(11)}

where lol_o is a generic mixing length to be externally supplied that can vary with zz. Dimensional analysis and similarity theory represent

du‾dz=−u′w′‾(z)lo(z),(12)\frac{d\overline{u}}{dz} = \frac{\sqrt{-\overline{u'w'}(z)}}{l_o(z)}, \tag*{(12)}

where u′u' is the longitudinal velocity fluctuation, and u′w′‾\overline{u'w'} is the momentum turbulent flux at height zz that can be estimated from the mean momentum balance using [18]

−u′w′‾(z)u∗2=(1−zn),(13)\frac{-\overline{u'w'}(z)}{u_*^2} = (1-z_n), \tag*{(13)}

where zn=z/Hz_n=z/H is the normalized water depth. With this estimate of u′w′‾(z)\overline{u'w'}(z), it follows directly that

du‾dz=u∗lo(1−zn)1/2; νt=u∗lo(1−zn)1/2; Ds=βu∗lo(1−zn)1/2.(14)\frac{d\overline{u}}{dz}=\frac{u_*}{l_o}(1-z_n)^{1/2};\ \nu_t=u_*l_o(1-z_n)^{1/2};\ D_s=\beta u_*l_o(1-z_n)^{1/2}. \tag*{(14)}

These expressions ensure that as zn→1z_n\to1, du‾/dz→0d\overline{u}/dz\to0, νt→0\nu_t\to0, and Ds→0D_s\to0. For zn≪1z_n\ll1 but z+>50z^+>50 (i.e. above the buffer layer) where z+=zu∗/νz^+=zu_*/\nu is a normalized distance in wall units [62] that can also be interpreted as a local Reynolds number (ResRe_s), lol_o is constrained by the channel bottom so that lo=κzl_o=\kappa z. In this case, du‾/dz≈u∗/(κz)d\overline{u}/dz\approx u_*/(\kappa z) and u‾(z)\overline{u}(z) varies logarithmically with zz, νt=κzu∗\nu_t=\kappa zu_*, and Ds=βκzu∗D_s=\beta\kappa zu_* (i.e. linear in zz).

As zn→1z_n\to1, the largest eddies are restricted by HH so that lo∝Hl_o\propto H instead of zz. Combining these two arguments using lo=κz(1−zn)1/2l_o=\kappa z(1-z_n)^{1/2} yields the quadratic diffusivity profile reported in a number of stream flow studies [24] and direct numerical simulations (DNS) of stratified atmospheric flows on inclined planes [25]. Assuming β=Sc−1=1\beta=Sc^{-1}=1, the SSC profiles associated with the linear and quadratic Ds(z)D_s(z) are

C‾(zn)C‾b={(znzn,b)−R∗,linear diffusivity, Prandtl’s power law,(zn1−zn1−zn,bzn,b)−R∗,quadratic diffusivity, Rouse’s formula,(15)\frac{\overline{C}(z_n)}{\overline{C}_b}= \begin{cases} \left(\dfrac{z_n}{z_{n,b}}\right)^{-R^*}, & \text{linear diffusivity, Prandtl's power law},\\ \left(\dfrac{z_n}{1-z_n}\dfrac{1-z_{n,b}}{z_{n,b}}\right)^{-R^*}, & \text{quadratic diffusivity, Rouse's formula}, \end{cases} \tag*{(15)}

where C‾b\overline{C}_b is a reference concentration at height zn,b=zb/Hz_{n,b}=z_b/H and R∗=RR^*=R when setting β=1\beta=1. The R∗R^* in equation (15) is commonly replaced by a fitted RR (or β\beta is no longer unity) as discussed elsewhere [55, 18]. The analysis using fitted RR is termed here as ’fitted’ Rouse’s formula. Other models for lol_o have been introduced but only two are singled out for illustrating differences in approaches to adjusting conventional formulations (usually for κz\kappa z): (i) lo=κzVn(zn)l_o=\kappa zV_n(z_n), where Vn=1−exp⁡(−z+/26)V_n=1-\exp(-z^+/26) (labeled as the van Driest damping function); (ii) lo=κz(1−zn)m1l_o=\kappa z(1-z_n)^{m_1}, where

m1=12[1+ae(C‾CR)],(16)m_1=\frac{1}{2}\left[1+a_e\left(\frac{\overline{C}}{C_R}\right)\right], \tag*{(16)}

CRC_R is some reference concentration and aea_e is an empirical coefficient [76, 52, 12]. In the second case, the mixing length is assumed to vary with SSC and recovers lo=κz(1−zn)1/2l_o=\kappa z(1-z_n)^{1/2} only for clear water. However, in the presence of sediments, m1m_1 varies with znz_n (and RR). In the first case, deviations from a linear mixing length is made to dependent on z+z^{+} (instead of HH), which is appropriate in the viscous and buffer regions of smooth boundary layers. Another revision to equation 10 is to re-cast turbulent transport in fractional derivatives to emphasize its non-Fickian aspect [58]. In this approach, the fractional order becomes a parameter that must be determined from experiments depending on how SS trajectories deviate from Brownian trajectories [72]. In practice, the order of the fractional derivative is set as a ’free’ parameter and must implicitly include the ScSc effect. This approach is not pursued further here.

Turbulent Stress and SS Flux Budgets

Simplified turbulent stress and SS flux budgets are now considered. For a stationary and planar homogeneous flow in the absence of subsidence (w‾=0\overline{w}=0), these budgets reduce to

∂w′u′‾∂t=0=−w′w′‾∂u‾∂z−∂w′w′u′‾∂z+p′∂u′‾∂z−ϵwu,(17)\frac{\partial\overline{w'u'}}{\partial t}=0=-\overline{w'w'}\frac{\partial\overline{u}}{\partial z}-\frac{\partial\overline{w'w'u'}}{\partial z}+p'\frac{\partial\overline{u'}}{\partial z}-\epsilon_{wu}, \tag*{(17)}
∂w′C′‾∂t=0=−w′w′‾∂C‾∂z−∂w′w′C′‾∂z+p′∂C′‾∂z−ϵwc−ws(w′∂C′∂z‾),\frac{\partial\overline{w'C'}}{\partial t}=0=-\overline{w'w'}\frac{\partial\overline{C}}{\partial z}-\frac{\partial\overline{w'w'C'}}{\partial z}+p'\frac{\partial\overline{C'}}{\partial z}-\epsilon_{wc}-w_s\left(\overline{w'\frac{\partial C'}{\partial z}}\right),

where p′p' is the turbulent pressure, ϵwu\epsilon_{wu} and ϵwc\epsilon_{wc} are molecular destruction terms assumed to be small when compared to the pressure-decorrelation terms at high Reynolds numbers [40]. The turbulence-particle interaction term requires closure that may be achieved by commencing with a local decomposition given by,

ws(w′∂C′∂z‾)=ws[(∂w′C′‾∂z)−(C′∂w′∂z‾)].(18)w_s\left(\overline{w'\frac{\partial C'}{\partial z}}\right)=w_s\left[\left(\frac{\partial\overline{w'C'}}{\partial z}\right)-\left(\overline{C'\frac{\partial w'}{\partial z}}\right)\right]. \tag*{(18)}

When assuming Φ(z)=0\Phi(z)=0 in equation 4 (i.e. no particle inertia), w′C′‾=wsC‾\overline{w'C'}=w_s\overline{C} thereby allowing one of the two terms in the difference shown in equation 18 to be linked to variables that are explicitly modeled. The other term (i.e. C′∂w′/∂z‾\overline{C'\partial w'/\partial z}) still necessitates a closure. A heuristic model that maintains maximum simplicity is to set

C′∂w′∂z‾=b1∂w′C′‾∂z,(19)\overline{C'\frac{\partial w'}{\partial z}}=b_1\frac{\partial\overline{w'C'}}{\partial z}, \tag*{(19)}

where b1b_1 is a positive or a negative constant. Upon setting w′C′‾=wsC‾\overline{w'C'}=w_s\overline{C}, this heuristic closure model yields [33],

ws(w′∂C′∂z‾)=ws[(∂wsC‾∂z)−b1(∂wsC‾∂z)]=α′ws2∂C‾∂z,(20)w_s\left(\overline{w'\frac{\partial C'}{\partial z}}\right)=w_s\left[\left(\frac{\partial w_s\overline{C}}{\partial z}\right)-b_1\left(\frac{\partial w_s\overline{C}}{\partial z}\right)\right]=\alpha' w_s^2\frac{\partial\overline{C}}{\partial z}, \tag*{(20)}

where α′=1−b1\alpha'=1-b_1 is a constant. When ∣b1∣≪1\lvert b_1\rvert\ll1, then α′=1\alpha'=1 and

ws(w′∂C′∂z‾)=ws2∂C‾∂z.(21)w_s\left(\overline{w'\frac{\partial C'}{\partial z}}\right)=w_s^2\frac{\partial\overline{C}}{\partial z}. \tag*{(21)}

Whether b1b_1 or α′\alpha' are strictly closure constants independent of sediment and/or flow conditions cannot be a priori ascertained. To do so requires another scaling analysis based on different assumptions and approximations. In this proposed scaling analysis, C′C' is assumed to vary with a turbulent quantity such as σc\sigma_c, and w′w' to vary with σw\sigma_w. Hence,

(C′∂w′∂z)‾=AF[σc(z)]∂σw∂z=AF[w′C′‾u∗F1(zn)]∂σw∂z.(22)\overline{\left(C'\frac{\partial w'}{\partial z}\right)}=A_F\left[\sigma_c(z)\right]\frac{\partial\sigma_w}{\partial z}=A_F\left[\frac{\overline{w'C'}}{u_*}F_1(z_n)\right]\frac{\partial\sigma_w}{\partial z}. \tag*{(22)}

where AFA_F is a flux-variance [1] similarity constant that can be positive or negative depending on the sign of the correlation coefficient between C′C' and ∂w′/∂z\partial w'/\partial z, and F1(zn)F_1(z_n) is an unknown dimensionless function describing the sediment concentration variance with znz_n above and beyond the w′C′‾\overline{w'C'} variations with znz_n. Since the goal is to determine the minimum governing variables impacting b1b_1 or α′\alpha' while assuming b1b_1 is independent of znz_n, equations (22) and (19) can be equated to yield

AF[w′C′(zn)‾u∗F1(zn)]∂σw∂z=b1∂w′C′(zn)‾∂z.(23)A_F\left[\frac{\overline{w'C'(z_n)}}{u_*}F_1(z_n)\right]\frac{\partial\sigma_w}{\partial z}=b_1\frac{\partial\overline{w'C'(z_n)}}{\partial z}. \tag*{(23)}

Re-arranging to infer b1b_1 results in

b1=AF[w′C′‾(∂w′C′‾∂z)−1][1u∗F1(zn)∂σw∂z].(24)b_1=A_F\left[\overline{w'C'}\left(\frac{\partial\overline{w'C'}}{\partial z}\right)^{-1}\right]\left[\frac{1}{u_*}F_1(z_n)\frac{\partial\sigma_w}{\partial z}\right]. \tag*{(24)}

With the assumption that b1b_1 is not dependent on znz_n, additional order of magnitude arguments must now be invoked to assess the sediment/flow variables that impact its magnitude: (i) ∂σw/∂z∼−u∗/H\partial\sigma_w/\partial z\sim-u_*/H (likely valid except near the channel bottom), (ii) F1F_1 is roughly a constant, (iii) ∂w′C′‾/∂z=ws∂C‾/∂z\partial\overline{w'C'}/\partial z=w_s\partial\overline{C}/\partial z, and (iv) w′C′‾/(∂C‾/∂z)∼−Ds,avg\overline{w'C'}/(\partial\overline{C}/\partial z)\sim-D_{s,\mathrm{avg}} where Ds,avg=(1/H)∫0HDs(z) dz∼u∗HD_{s,\mathrm{avg}}=(1/H)\int_0^H D_s(z)\,dz\sim u_*H. Inserting these order of magnitude arguments into equation (24) result in

b1∼sgn⁡(Af)u∗Hws[1u∗H]∼sgn⁡(Af)u∗ws.(25)b_1\sim\operatorname{sgn}(A_f)\frac{u_*H}{w_s}\left[\frac{1}{u_*H}\right]\sim\operatorname{sgn}(A_f)\frac{u_*}{w_s}. \tag*{(25)}

Equation (18) is used to suggest a pragmatic closure in equation (20) that applies to only one of two terms, and this one term itself is only one term in the overall flux budget. Given the interplay between these multiple terms, the overall model results for C‾\overline{C} may be robust to uncertainties in this closure vis-a-vis externally imposing ScSc or β−1\beta^{-1} directly on the eddy diffusivity as common in prior models.

Upon ignoring the flux transport terms (triple moments), and closing the pressure decorrelation terms using a linear Rotta scheme that accounts for the isotropization of the production yields

−(1−CI)σw2∂u‾∂z−ARw′u′‾τ=0,[−(1−CI)−α′ws2σw2]σw2∂C‾∂z−ARw′C′‾τ=0.(26)-(1-C_I)\sigma_w^2\frac{\partial\overline{u}}{\partial z}-A_R\frac{\overline{w'u'}}{\tau}=0,\qquad \left[-(1-C_I)-\alpha'\frac{w_s^2}{\sigma_w^2}\right]\sigma_w^2\frac{\partial\overline{C}}{\partial z}-A_R\frac{\overline{w'C'}}{\tau}=0. \tag*{(26)}

where τ\tau is a turbulent relaxation time scale, CI=3/5C_I = 3/5 is the isotropization of the production constant determined from rapid distortion theory [62], and AR=1.8A_R = 1.8 [40, 38] is the Rotta constant assumed to be the same for momentum and SS. It directly follows from these simplified budgets that a model of maximum simplicity for ScSc may be derived as

Sc−1(zn)=Dsνt=1+α(wsσw)2,(27)Sc^{-1}(z_n) = \frac{D_s}{\nu_t} = 1 + \alpha\left(\frac{w_s}{\sigma_w}\right)^2, \tag*{(27)}

where α=α′/(1−CI)\alpha= \alpha^\prime/(1-C_I), though α′\alpha^\prime or b1b_1 can vary themselves with u∗/wsu_*/w_s as noted earlier. It is necessary to point out that when α≥0\alpha\ge0, equation 27 is opposite to what is predicted by the so-called ’crossing-trajectories’ effect for heavy particles settling in a turbulent flow. The crossing trajectories arise when particle trajectories cross trajectories of fluid elements under the influence of gravity. This effect invariably forces particles to move from a region of highly correlated flow to another less correlated region [83]. In this manner, particles lose velocity correlation more rapidly than the corresponding fluid points and thus must disperse less. Thus, the crossing trajectories effect requires Sc>1Sc > 1 [16, 20].

The Co-spectral Budget Model

The models so far make no explicit contact with the phenomenon they perpetrate to represent: turbulent eddies and their energy distribution. The proposed approach here uses a co-spectral budget model (CSB) to achieve such a link. The CSB is derived from an approximated Navier-Stokes equation in a spectral form that links turbulent eddies of different sizes to w′C′‾\overline{w^\prime C^\prime}. The CSB derivation commences by noting that w′C′‾\overline{w^\prime C^\prime} and u′w′‾\overline{u^\prime w^\prime} both satisfy the normalizing properties,

−w′C′‾=∫0∞ϕwc(k) dk,−u′w′‾=∫0∞ϕwu(k) dk,(28)-\overline{w^\prime C^\prime} = \int_{0}^{\infty} \phi_{wc}(k)\,dk,\qquad-\overline{u^\prime w^\prime} = \int_{0}^{\infty} \phi_{wu}(k)\,dk, \tag*{(28)}

where ϕwc(k)\phi_{wc}(k) and ϕwu(k)\phi_{wu}(k) are the co-spectral density functions of the turbulent vertical velocity-turbulent sediment concentration and turbulent vertical-longitudinal velocities, respectively, and kk is the wavenumber or inverse eddy size. The co-spectral budgets associated with equation 18 have been derived elsewhere and simplify to [9, 13, 40, 38],

∂∂tϕwu(k)=0=Pwu(k)+Twu(k)+πwu(k)−2νk2ϕwu(k),(29)\frac{\partial}{\partial t}\phi_{wu}(k) = 0 = P_{wu}(k) + T_{wu}(k) + \pi_{wu}(k) - 2\nu k^2\phi_{wu}(k), \tag*{(29)}
∂∂tϕwc(k)=0=Pwc(k)+Twc(k)+πwc(k)−ν(1+Scm−1)k2ϕwc(k),(30)\frac{\partial}{\partial t}\phi_{wc}(k) = 0 = P_{wc}(k) + T_{wc}(k) + \pi_{wc}(k) - \nu(1+Sc_m^{-1})k^2\phi_{wc}(k), \tag*{(30)}

where Pwu(k)=(dw‾/dz)Eww(k)P_{wu}(k) = (\mathrm{d}\overline{w}/\mathrm{d}z)E_{ww}(k) and Pwc(k)=(dC‾/dz)Eww(k)P_{wc}(k) = (\mathrm{d}\overline{C}/\mathrm{d}z)E_{ww}(k) are the stress and flux production terms at kk, Eww(k)E_{ww}(k) is the vertical velocity spectrum satisfying the normalizing relation σw2=∫0∞Eww(k)dk\sigma_w^2 = \int_{0}^{\infty} E_{ww}(k)\mathrm{d}k, Twu(k)T_{wu}(k) and Twc(k)T_{wc}(k) are turbulent transfer terms, πwu(k)\pi_{wu}(k) and πwc(k)\pi_{wc}(k) are pressure-velocity and pressure-scalar decorrelation terms, and ScmSc_m is the molecular Schmidt number (not related to ScSc). Invoking a spectral-based Rotta model that includes the isotropization of the production as before, the pressure-scalar co-variance in k-space can be modeled as

πwu(k)=−AR1tww(k)ϕwu(k)−CIPwu(k),πwc(k)=−AR1tr(k)ϕwc(k)−CIPwc(k),(31)\pi_{wu}(k) = -A_R\frac{1}{t_{ww}(k)}\phi_{wu}(k) - C_I P_{wu}(k), \qquad\pi_{wc}(k) = -A_R\frac{1}{t_r(k)}\phi_{wc}(k) - C_I P_{wc}(k), \tag*{(31)}

where AR≈1.8A_R \approx1.8 and CI=3/5C_I = 3/5 are as before, tww(k)t_{ww}(k) and tr(k)t_r(k) are the decorrelation time-scale of the turbulent stress and particle concentration. A model of maximum simplicity is to assume that these two wavenumber dependent time scales are related using a wavenumber-dependent Sc(k)Sc(k) given by,

tr(k)=tww(k)Sc−1(k),with Sc−1(k)=1+α(ωsktwc)2,(32)t_r(k) = t_{ww}(k)Sc^{-1}(k), \qquad\text{with } Sc^{-1}(k) = 1 + \alpha(\omega_s k t_{wc})^2, \tag*{(32)}

where twc=min⁡(tww,fo tK,b)t_{wc} = \min(t_{ww}, f_o\,t_{K,b}) with fof_o being a constant (a plausibility argument to such twc(k)t_{wc}(k) representation is discussed later), ScSc is modeled in analogy to equation 27 albeit in a spectral form e.g. the local characteristic turbulent velocity is estimated by (ktwc)−1(k t_{wc})^{-1} using a one-way coupling approach [22], tww(k)∝ϵ−1/3k−2/3t_{ww}(k) \propto\epsilon^{-1/3}k^{-2/3} is interpreted as a characteristic time scale derived from dimensional analysis assuming ϵ\epsilon is the conserved quantity across the energy cascade of Eww(k)E_{ww}(k), and ϵ\epsilon is the turbulent kinetic energy dissipation rate. One plausible choice for the proportionality constant is Co−1/2C_o^{-1/2} so as to recover a Kolmogorov time scale in the inertial subrange, where Co=0.65C_o = 0.65 is the Kolmogorov constant for the vertical velocity component.

For scalewise integration, it is also necessary to maintain a bounded tr(k)t_r(k) as k→0k \rightarrow0 for any znz_n. We set tr(k)=tr(kc)t_r(k) = t_r(k_c) when k<kck < k_c, where kck_c is the smallest inverse length scale where Eww(k)E_{ww}(k) increases with increasing kk. The viscous-destruction terms are negligible when compared to the Rotta terms for kη≪1k\eta\ll1. Since Twu(k)T_{wu}(k) and Twc(k)T_{wc}(k) do not contribute to the net production or destruction of ϕwu(k)\phi_{wu}(k) and ϕwc(k)\phi_{wc}(k) but only redistribute them across scales (i.e. ∫0∞Twu(k)dk=0\int_{0}^{\infty} T_{wu}(k)\mathrm{d}k = 0, and ∫0∞Twc(k)dk=0\int_{0}^{\infty} T_{wc}(k)\mathrm{d}k = 0), they are ignored for simplicity [8]. Adopting these simplifications,

ϕuw(k)=(1−CIAR)dw‾dz[Eww(k)tww(k)],ϕwc(k)=(1−CIAR)dC‾dz[Eww(k)tr(k)].(33)\phi_{uw}(k) = \left(\frac{1-C_I}{A_R}\right)\frac{\mathrm{d}\overline{w}}{\mathrm{d}z}\left[E_{ww}(k)t_{ww}(k)\right], \qquad\phi_{wc}(k) = \left(\frac{1-C_I}{A_R}\right)\frac{\mathrm{d}\overline{C}}{\mathrm{d}z}\left[E_{ww}(k)t_r(k)\right]. \tag*{(33)}

To integrate these equations across kk and derive turbulent shear stress and sediment flux at any height znz_n, an expression for Eww(k)E_{ww}(k) is required. A model for Eww(k)E_{ww}(k) that captures known spectral features at an arbitrary znz_n is shown in Figure 2.

Figure 2. Left: A typical Eww(k)E_{ww}(k) at znz_n from the channel bottom. The very low wavenumber range are assumed to follow the Saffman spectrum (Eww(k)∝k2E_{ww}(k) \propto k^2) until kc=1/Hk_c = 1/H. The Saffman spectrum is then connected using a flat transition (i.e. wall effects introduce energy splashing) to the inertial subrange at ko∝1/zk_o \propto1/z where Eww(k)∝k−5/3E_{ww}(k) \propto k^{-5/3}. The black curves are extracted from measurements ([59]) with different flow conditions using ADV and do not resolve the viscous dissipation range in the vicinity of ke=1/ηk_e = 1/\eta or the presumed Saffman spectrum. Right: The σw2/u∗2\sigma_w^2/u_*^2 profile modeled from scale-wise integration of Eww(k)E_{ww}(k) and its simplified form (i.e. ignoring the Saffman contribution and extending the inertial subrange indefinitely to fine scales). The measured σw2/u∗2\sigma_w^2/u_*^2 profiles are from experiments described elsewhere ([63, 59, 30]). They include field experiments and wind-tunnel experiments over a wide range of roughness types and Reynolds number conditions. The direct numerical simulations (DNS) for a smooth channel (red) are also included for comparisons ([30]).

The Eww(k)E_{ww}(k) is now piece-wise approximated as

Eww(k)={Ekol(ko)kc−2k2,if 0≤k≤kcEkol(ko),if kc≤k≤koEkol(k),if ko≤k≤ke(34)E_{ww}(k) = \begin{cases} E_{kol}(k_o)k_c^{-2}k^2, & \text{if } 0 \leq k \leq k_c \\ E_{kol}(k_o), & \text{if } k_c \leq k \leq k_o \\ E_{kol}(k), & \text{if } k_o \leq k \leq k_e \end{cases} \tag*{(34)}

where kc=H−1k_c = H^{-1}, ko=(κz)−1k_o = (\kappa z)^{-1} and ke=η−1k_e = \eta^{-1} are three characteristic wavenumbers that mark the key transitions in Eww(k)E_{ww}(k) between HH and the characteristic eddy scales bounding the inertial subrange ([8, 40, 4] (S. Li & Katul 2019)), and Ekol(k)=Coε(z)2/3k−5/3E_{kol}(k) = C_o\varepsilon(z)^{2/3}k^{-5/3} is the Kolmogorov spectrum.

In the case of Ekol(k)E_{kol}(k), the transfer of energy across scales shapes the energy cascade and is necessary for obtaining the k−5/3k^{-5/3} scaling. The transfer of stress across scales, as given by Twu(k)T_{wu}(k), was ignored in the CSB model here. The inclusion of the transfer term in the energy cascade (indirectly specified by Eww(k)E_{ww}(k)) but not in the CSB may appear paradoxical. This is not so as the role and significance of the transfer terms are quite different when analyzing scale-wise energy and stress budgets [9]. In the inertial subrange where Ekol(k)∼k−5/3E_{kol}(k) \sim k^{-5/3}, a ϕuw(k)∼k−7/3\phi_{uw}(k) \sim k^{-7/3} has also been reported and confirmed in numerous boundary layer experiments and simulations of wall-bounded flows [62]. A balance between production and dissipation terms in the CSB model leads to a ϕuw(k)∼(du‾/dz)Eww(k)tww(k)\phi_{uw}(k) \sim(\mathrm{d}\overline{u}/\mathrm{d}z)E_{ww}(k)t_{ww}(k), which recovers the [(du‾/dz)1/3]k−7/3[(\mathrm{d}\overline{u}/\mathrm{d}z)^{1/3}]k^{-7/3} scaling in the inertial subrange. Inclusion of Twu(k)T_{wu}(k) necessarily leads to ϕuw(k)\phi_{uw}(k) that must deviate from a k−7/3k^{-7/3} scaling in the inertial subrange as discussed elsewhere (D. Li et al., 2015). Moreover, the constants emerging from a production balancing dissipation in the scale-wise CSB model for the inertial subrange, [(1−CI)/AR]Co1/2=0.18[(1-C_I)/A_R]C_o^{1/2}=0.18, does recover the accepted co-spectral similarity constant whose numerical value was determined at 0.150.15–0.160.16 from wind tunnel studies, atmospheric surface layer studies, and DNS [40]. For these reasons (i.e. Twu(k)T_{wu}(k) ignored within the inertial subrange) and because ∫0∞Twu(k)dk=0\int_{0}^{\infty}T_{wu}(k)\mathrm{d}k=0, Twu(k)T_{wu}(k) is ignored at all kk. This assumption is also compatible with ignoring the triple moments in equations 18.

The only remaining term needed to describe the magnitude of Eww(k)E_{ww}(k) at all kk is ϵ(z)\epsilon(z). A model of maximum simplicity is to relate ϵ(z)\epsilon(z) to the mechanical production Pwu(z)P_{wu}(z) of the turbulent kinetic energy budget using [62]

ϵ(z)=Pwu(z)ϕ(zn)=ϕ−1(zn)(−u′w′‾du‾dz)=ϕ−1(zn)u∗2(1−zH)du‾dz.(35)\epsilon(z)=\frac{P_{wu}(z)}{\phi(z_n)}=\phi^{-1}(z_n)\left(-\overline{u'w'}\frac{\mathrm{d}\overline{u}}{\mathrm{d}z}\right)=\phi^{-1}(z_n)u_*^2\left(1-\frac{z}{H}\right)\frac{\mathrm{d}\overline{u}}{\mathrm{d}z}. \tag*{(35)}

where ϕ(zn)\phi(z_n) is a modification function to account for the imbalance between the local mechanical production and local dissipation terms in the turbulent kinetic energy budget. For stationary and planar-homogeneous flow conditions without any mean vertical advection and in the absence of any transport terms, ϵ(z)≈Pwu(z)\epsilon(z)\approx P_{wu}(z) and ϕ(zn)≈1\phi(z_n)\approx1. While this estimate may be acceptable in the log-region describing u‾(z)\overline{u}(z), deviations near the channel bottom (ϕ(zn)>1\phi(z_n)>1) and near the water surface (ϕ(zn)<1\phi(z_n)<1) are expected. Hence, ϕ(zn)\phi(z_n) must be viewed as a depth-dependent function [42, 62] though its variation from unity is not considered here to maintain maximum simplicity. A plausibility argument for ignoring its variation from unity is that w′C′‾∝[ϕ(zn)]−1/3\overline{w'C'}\propto[\phi(z_n)]^{-1/3} (shown later), which makes the SSC calculations less sensitive to ϕ(zn)\phi(z_n) deviations from unity. This point is considered later in the context of modeling σw2(zn)\sigma_w^2(z_n) based on the assumed Eww(k)E_{ww}(k) shape.

Returning to the choice of twc=min⁡(tww,fotk,b)t_{wc}=\min(t_{ww},f_o t_{k,b}) and the choice fof_o, as zn→1z_n\to1, w′u′‾→0\overline{w'u'}\to0, Pwu(zn)→0P_{wu}(z_n)\to0, and thus ϵ→0\epsilon\to0 (i.e. no turbulence) near the free water surface. With ϵ→0\epsilon\to0, tww(k)→∞t_{ww}(k)\to\infty (along with τk→∞\tau_k\to\infty and ηk→∞\eta_k\to\infty). That tww(k)→∞t_{ww}(k)\to\infty is not problematic for the closure scheme of πwu(k)\pi_{wu}(k) and πwc(k)\pi_{wc}(k) as those terms are expected to decay near the free water surface and this decay remains compatible with tww(k)→∞t_{ww}(k)\to\infty. The problem of ϵ→0\epsilon\to0 arises in maintaining a finite Sc−1(k)Sc^{-1}(k) dominated by turbulent processes thereby necessitating a finite ϵ\epsilon in the calculation of Sc−1(k)Sc^{-1}(k) that cannot be readily inferred from Pwu(zn)P_{wu}(z_n). To ensure that the particle interaction time scale twct_{wc} remains bounded in Sc−1(k)Sc^{-1}(k), an adhoc minimal value of ϵ\epsilon, set to be 0.1% ϵb\epsilon_b, is proposed. This choice of minimal ϵb\epsilon_b prevents ϵ→0\epsilon\to0 as zn→1z_n\to1 in the Sc(k)Sc(k) formulation only. This minimal threshold set to ensure a finite ϵ\epsilon in Sc(k)Sc(k) (mainly near the free water surface) leads to fo=1000≈31f_o=\sqrt{1000}\approx31.

Results and Discussion

Co-spectral Budget Model

By scale-wise integrating ϕuw(k)\phi_{uw}(k) and using u∗2(1−zn)=∫0keϕuw(k) dku_*^2(1-z_n)=\int_0^{k_e}\phi_{uw}(k)\,\mathrm{d}k, the velocity gradient du‾/dz\mathrm{d}\overline{u}/\mathrm{d}z at zz is obtained as

du‾dz=Aπ−3/4ϕ1/4(zn)(1−zH)1/2[154−83(kcko)1/3−34(koke)4/3]−3/4(kou∗),(36)\frac{\mathrm{d}\overline{u}}{\mathrm{d}z}=A_{\pi}^{-3/4}\phi^{1/4}(z_n)\left(1-\frac{z}{H}\right)^{1/2}\left[\frac{15}{4}-\frac{8}{3}\left(\frac{k_c}{k_o}\right)^{1/3}-\frac{3}{4}\left(\frac{k_o}{k_e}\right)^{4/3}\right]^{-3/4}(k_o u_*), \tag*{(36)}

where Aπ=(1−CI)Co/AR≈0.18A_{\pi}=(1-C_I)\sqrt{C_o}/A_R\approx0.18, and the vertical velocity variance can be derived by scale-wise integrating Eww(k)E_{ww}(k) as,

σw2u∗2=52CoAπ−1/2ϕ−1/2(zn)[1−415kcko−35(keko)−2/3][154−83(kcko)1/3−34(keko)−4/3]−1/2(1−zn).(37)\frac{\sigma_w^2}{u_*^2}=\frac{5}{2}C_oA_{\pi}^{-1/2}\phi^{-1/2}(z_n)\left[1-\frac{4}{15}\frac{k_c}{k_o}-\frac{3}{5}\left(\frac{k_e}{k_o}\right)^{-2/3}\right]\left[\frac{15}{4}-\frac{8}{3}\left(\frac{k_c}{k_o}\right)^{1/3}-\frac{3}{4}\left(\frac{k_e}{k_o}\right)^{-4/3}\right]^{-1/2}(1-z_n). \tag*{(37)}

Likewise, the SSC turbulent flux is solved as

−w′C′‾=Aπϕ−1/3(zn)Ω(z)u∗2/3[(1−zH)du‾dz]1/3dC‾dz=−wsC‾,(38)-\overline{w'C'}=A_{\pi}\phi^{-1/3}(z_n)\Omega(z)u_*^{2/3}\left[\left(1-\frac{z}{H}\right)\frac{\mathrm{d}\overline{u}}{\mathrm{d}z}\right]^{1/3}\frac{\mathrm{d}\overline{C}}{\mathrm{d}z}=-w_s\overline{C}, \tag*{(38)}

with Ω(zn)\Omega(z_n) given by

Ω(zn)=∫0kcSc−1(kc)kc−8/3ko−5/3k2 dk+∫kckoko−5/3Sc−1(k)k−2/3 dk+∫kokeSc−1(k)k−7/3 dk.(39)\Omega(z_n)=\int_0^{k_c}Sc^{-1}(k_c)k_c^{-8/3}k_o^{-5/3}k^2\,\mathrm{d}k+\int_{k_c}^{k_o}k_o^{-5/3}Sc^{-1}(k)k^{-2/3}\,\mathrm{d}k+\int_{k_o}^{k_e}Sc^{-1}(k)k^{-7/3}\,\mathrm{d}k. \tag*{(39)}

Therefore, the turbulent Schmidt number Sc(zn)Sc(z_n) can be determined from the CSB model as

Sc(zn)=νtDs=Ω−1(zn)[154−83(kcko)1/3−34(koke)4/3]ko−4/3.(40)Sc(z_n)=\frac{\nu_t}{D_s}=\Omega^{-1}(z_n)\left[\frac{15}{4}-\frac{8}{3}\left(\frac{k_c}{k_o}\right)^{1/3}-\frac{3}{4}\left(\frac{k_o}{k_e}\right)^{4/3}\right]k_o^{-4/3}. \tag*{(40)}

Because the determination of ke=1/ηk_e = 1/\eta (where η=(ν3/ϵ)1/4\eta= (\nu^3/\epsilon)^{1/4}) requires an estimate of ϵ(zn)=Puw(zn)\epsilon(z_n) = P_{uw}(z_n) and thus an estimate of duˉ/dz\mathrm{d}\bar{u}/\mathrm{d}z, an iterative scheme is needed to determine duˉ/dz\mathrm{d}\bar{u}/\mathrm{d}z and kek_e at every znz_n from equation 36. Once determined, the Eww(k)E_{ww}(k), Sc(zn)Sc(z_n), w′C′‾\overline{w'C'} and the subsequent SSC profile can be computed at each znz_n by solving equations 37, 38, and 40 for σw2\sigma_w^2, w′C′‾\overline{w'C'} and ScSc. Since there is no analytical solution to this system, a numerical integration using a 3rd-order Adams–Bashforth method is employed.

Before proceeding to the analysis of SSC, an assessment of the assumed shape of Eww(k)E_{ww}(k), its transition wavenumbers, as well as the consequence of the assumption of ϕ(zn)≈1\phi(z_n) \approx1 is conducted in Figure 2. The predicted σw2/u∗2\sigma_w^2/u_*^2 and its simplified version using Eww1(k)E_{ww}^{1}(k) without the Saffman spectrum and assuming ke→∞k_e \to\infty are compared against two sets of experiments: (i) wind tunnel experiments conducted over a wide range of surface roughness types [63] and (ii) field experiments [59] of the sediment flow in the Balmoral Irrigation Canal (New Zealand). The wind-tunnel experiments used a hot-wire probe whereas the field experiments used acoustic Doppler velocity (ADV) measurements that do not resolve the viscous dissipation regime. As expected, the predicted σw2/u∗2\sigma_w^2/u_*^2 here exceeds the measurements because the spectral shapes assumed in Eww(k)E_{ww}(k) account for a much broader range of eddy sizes than the experiments interrogate. Specifically, the Saffman and dissipation ranges are not resolved by the flume experiments whereas the wind tunnel experiments resolve a limited dissipation range but are not conducted over a sufficiently long enough sampling period to cover the Saffman spectrum. Nonetheless, the model recovers key features of the (σw/u∗)2(\sigma_w/u_*)^2 profile: a rapid increase with znz_n near the surface, a peak at (σw/u∗)2=1.9(\sigma_w/u_*)^2 = 1.9, and a quasi-linear decline as zn→1z_n \to1. The peak (σw/u∗)2=1.9(\sigma_w/u_*)^2 = 1.9 is compatible with near-neutral atmospheric surface layer measurements (=1.8)(= 1.8) where lateral confinements of the flow are absent (unlike flumes and wind tunnels) and where H/ηH/\eta far exceeds those obtained in laboratory studies.

Now the comparisons of the CSB results with α\alpha temporarily set as a ’free’ parameter with (i) Prandtl’s power law solution and (ii) Rouse’s formula are shown in Figure 3. The computed SSC and ScSc profiles are also presented when the flow conditions and sediment properties are externally supplied. For Prandtl’s power-law and Rouse’s formula, the bulk Schmidt number was set to unity. However, the CSB model allows for a depth-dependent Sc(zn)Sc(z_n), which is set by α\alpha. When α=0\alpha= 0, Sc(zn)=1Sc(z_n) = 1 in the entire channel, consistent with equation 27. When α>0\alpha> 0, Sc(zn)Sc(z_n) varies with depth and is generally greater in the near-bed region and becomes smaller with increasing znz_n. However, because of the imposition of a finite ϵ\epsilon near the water surface (=0.001ϵb=0.001\epsilon_b), Sc(zn)Sc(z_n) increases back to near unity when zn→1z_n \to1. Rouse’s equation and CSB models exhibit different behavior near the water surface. Rouse’s equation yields a zero-concentration at zn=1z_n=1 whereas the CSB model does not. One advantage to the CSB approach is its ability to resolve the dependence of C‾/Cb‾\overline{C}/\overline{C_b} on Reynolds number. Using different ν\nu, variations in Re∗=u∗H/νRe_*=u_*H/\nu can be generated and their effects on CSB model predictions tracked. Recall that H/ηbH/\eta_b (modeled in the CSB) scales as Re∗3/4Re_*^{3/4}, and the effects of this scale separation on the shape of the vertical velocity spectrum, sediment flux co-spectrum, and the resulting C‾/Cb‾\overline{C}/\overline{C_b} profiles are explicitly determined. The effects of α\alpha are much more significant than the effects of Re∗Re_*, which is heuristically supportive for using Direct Numerical Simulation runs (lower Re∗Re_*) to further explore the CSB approach. As earlier noted, the implications of setting twc=min⁡(tww,fo tK,b)t_{wc}=\min(t_{ww},f_o\,t_{K,b}) with fo=1000f_o=\sqrt{1000} are most visible

Figure 3. The predicted SSC, ScSc, and DsD_s profiles based on the CSB model when setting ds=1d_s=1 mm, ρs=1.2\rho_s=1.2 g cm−3^{-3}, u∗=3u_*=3 cm s−1^{-1}, and Ub/u∗=10U_b/u_*=10. The reference position is at zn,b=0.01z_{n,b}=0.01. Different α\alpha values and Reynolds numbers (Re∗=u∗H/νRe_*=u_*H/\nu) are featured to illustrate overall sensitivity of the normalized SSC profile to these parameters. The Prandtl and Rouse model predictions of SSC are shown for reference in the top-left panel. The Reynolds number is varied by altering ν\nu.

on the Sc(zn)Sc(z_n) profile near the free water interface. Altering fof_o primarily modifies the thickness of the region near the water interface impacted by the imposed finite twct_{wc} (or finite ϵ\epsilon in the Sc(k)Sc(k) determination). However, the CSB model itself is not expected to be valid in this zone as the assumed shape of Eww(k)E_{ww}(k) is not realistic, the flux transport terms can be finite, and turbo-phoretic effects may also be large in this vicinity. In sum, predictions from the CSB model near the free water surface must be treated with skepticism and caution.

Recovery of the Rouse and Prandtl equations

Whether a Rouse equation can be recovered from the CSB model under certain simplifications is now examined. Any explicit model must include ScSc and approximations to equation (40). Assuming kc/ko→0k_c/k_o \to0 and ko/ke→0k_o/k_e \to0 in Ω(zn)\Omega(z_n) only (i.e. setting the area under the Saffman spectrum to zero that is then partially compensated for by extending the inertial subrange to ke→∞k_e \to\infty), the Schmidt number derived from equation (40) can be approximated as

Sc−1≈1+Bπ(wsu∗)2,with Bπ=15Aπ3Coα≈0.84α,(41)Sc^{-1} \approx1 + B_\pi\left(\frac{w_s}{u_*}\right)^2,\quad\text{with } B_\pi= \frac{\sqrt{15A_\pi}}{3C_o}\alpha\approx0.84\alpha, \tag*{(41)}

which directly recovers the quadratic model for Sc−1Sc^{-1} reported elsewhere [79, 7] as expected. With R≈0R \approx0, equation (41) indicate β=Sc−1→1\beta= Sc^{-1} \to1 thereby recovering Rouse’s original assumption (i.e. SS resemble passive scalars in this case). This estimate of β\beta also allows for the determination of the model coefficient α\alpha using a separate data set and model runs shown in Figure 4.

Model coefficient $\beta$ based on different formulae, experiments, and model runs

Figure 4. The model coefficient β\beta based on different formulae, experiments, and model runs. The experiments and model runs presented here are described elsewhere [35].

Figure 4 shows different predictions of β\beta, including β=1+2(ws/u∗)2\beta= 1+2(w_s/u_*)^2 [79] and β=1.3+3(ws/u∗)2\beta= 1.3 + 3(w_s/u_*)^2 [35] for model results that explicitly consider particle-fluid interactions. Moreover, with ScSc provided in equation (41), the SS diffusivity is derived as,

Ds(z)κzu∗=1Sc(1−zn)=[1+Bπ(wsu∗)2](1−zn)(42)\frac{D_s(z)}{\kappa z u_*} = \frac{1}{Sc}(1-z_n) = \left[1+B_\pi\left(\frac{w_s}{u_*}\right)^2\right](1-z_n) \tag*{(42)}

where κzu∗\kappa z u_* is the eddy viscosity in the log-region of u‾(z)\overline{u}(z). Depending on choices made for α\alpha or BπB_\pi, a number of empirical relations can be recovered including the widely used Rouse’s equation and variants on it [34]. For a given α\alpha, an analytical solution for the SSC can be derived and compared with published experiments. The SSC solution for an arbitrary α\alpha is given as

C‾(zn)C‾b=(zn1−zn1−zn,bzn,b)−R+.(43)\frac{\overline{C}(z_n)}{\overline{C}_b} = \left(\frac{z_n}{1-z_n}\frac{1-z_{n,b}}{z_{n,b}}\right)^{-R_+}. \tag*{(43)}

where the power exponent R+R_+ is defined as

R+=11+Bπ(ws/u∗)2wsκu∗.(44)R_+ = \frac{1}{1+B_\pi(w_s/u_*)^2}\frac{w_s}{\kappa u_*}. \tag*{(44)}

When α=0\alpha= 0 (or Bπ=0B_\pi= 0), a quadratic diffusivity profile [60] as well as Rouse’s formula [67, 66] for SSC given in equation (15) are recovered. Furthermore, in the limit of (zn≪1z_n \ll1) a linear diffusivity profile [80] along with the classic power law solution are also recovered from equation 43. The consequences on σw2\sigma_w^2 of setting the Saffman spectrum to zero and extending the inertial subrange to k→∞k \to\infty on σw2\sigma_w^2 are briefly discussed using Figure 2. As expected, these approximation over-estimate (σw/u∗)2(\sigma_w/u_*)^2 in the near-wall region and underestimate (σw/u∗)2(\sigma_w/u_*)^2 in the outer layer when compared to a Eww(k)E_{ww}(k) that accommodates the Saffman spectrum (i.e. large scale effects) but truncates the inertial subrange at 1/ke1/k_e. These effects cannot be readily ignored and may influence the choices made about α\alpha.

Comparison with Experiments

The CSB model given by equations (38) and (36) and its simplified version featured in equation (43) are compared with published experiments ([27, 78, 75]) summarized in Table 1. We assume Φ(z)=0\Phi(z)=0 thereby neglecting inertial effects for compatibility with operational models (e.g. the Rouse model). The comparisons are shown in Figure 5. For these experiments, all the reported parameters including measured u∗u_*, dsd_s, and ρs\rho_s and the fitted β\beta (needed for assessing the fitted Rouse equation) and α\alpha (needed for evaluating the numerical CSB model) are presented in Table 1.

Run(a)(b)(c)(d)(e)(f)
Flow Properties
HH (m)0.100.520.500.100.100.10
BB (m)−0.840.840.150.150.15
UbU_b (measured, m s−1^{-1})1.983.953.630.310.220.17
Reb=UbH/ν×10−4Re_b = U_bH/\nu\times10^{-4}18.71931702.92.11.6
Repa=u∗ds/νRe_{pa} = u_*d_s/\nu10316916616137
u∗u_* (cm s−1^{-1})7.6720.020.01.71.40.8
zn,b×102z_{n,b} \times10^2 (measured)6.32.63.05.03.85.6
zn,b+(=u∗zb/ν)z_{n,b}^+ (= u_*z_b/\nu)45627472988855345
Ub/u∗U_b/u_* (measured)25.819.418.218.215.721.3
Sediment Properties
ρs/ρ\rho_s/\rho1.052.652.651.201.201.20
dsd_s (mm)1.420.880.881.001.001.00
wsw_s (cm s−1^{-1})1.710102.92.92.9
Dimensionless Model Parameters
R∗=ws/(κu∗)R^* = w_s/(\kappa u_*)0.51.21.24.35.29.1
α\alpha27.714.516.30.60.40.2
β\beta (Rouse)1.31.61.82.22.22.8
R=ws/(βκu∗)R = w_s/(\beta\kappa u_*)0.40.80.71.92.43.2
β\beta (Prandtl)0.91.11.21.41.51.9
R=ws/(βκu∗)R = w_s/(\beta\kappa u_*)0.61.11.03.03.54.8
Stb(=ws/ggSoUb/ν)St_b (= w_s/g\sqrt{gS_oU_b/\nu})0.575.635.40.090.060.03
St+(=τpu∗/H)×103St_+ (= \tau_pu_*/H) \times10^31.34.04.10.50.40.2
Frd(=Ub/(ρs/ρ−1)gd)Fr_d (= U_b/\sqrt{(\rho_s/\rho- 1)gd})753330754
FrdcFr_{dc}344333
Ub/u∗U_b/u_* (CSB rough bed)12.415.215.613.014.013.5
Ub/u∗U_b/u_* (CSB smooth bed)27.233.133.523.523.722.4

Table 1. Summary of published experiments and parameters used in model-data comparisons. When setting Sc=1Sc = 1, not all runs are classified as SS (or 0.8≤R∗≤2.50.8 \le R^* \le2.5) even though sediments were reported as suspended. While StbSt_b is not very small for (b) and (c), St+≪1St_+ \ll1. Calculated densimetric and critical Froude numbers (FrdFr_d and FrdcFr_{dc}) are presented along with the roughness Reynolds number RepaRe_{pa}. All experiments lie in the fully-rough (Repa>100Re_{pa} > 100) or transitional (3<Repa<1003 < Re_{pa} < 100) regimes in fully-developed turbulence (Reb≥500Re_b \ge500).

Figure 5. The predicted SSC profiles normalized by CbC_b selected at the measurement height with the highest reported concentration. The panel labeling follows Table 1 with the top panel showing the comparisons from earlier measurement i.e. (a) [82, 27] and (b)-(c) [78], and the bottom panel showing the comparisons using recent experiments (d)-(f) [75]. For experiments in panels (a)-(c), the StbSt_b is not small (> 0.5).

In the experiments, the sediments covered the bed and were assumed to have reached an equilibrium state where equation (4) applies ([75]). The densimetric Froude number FrdFr_d and the critical densimetric Froude number FrdcFr_{dc} whose formulation is described elsewhere ([2, 3] (S. Li & Katul 2019)) are also presented in Table 1. In all cases, the UbU_b, ρs/ρ\rho_s/\rho, and d/Hd/H result in Frd>FrdcFr_d > Fr_{dc} meaning that sediments can be released from the bed and must be balanced by sediments depositing onto the bed. Thus, the experiments do not strictly abide by HAE’s definition of SS as sediments here are not remain permanently suspended. Across the experiments, the flow variables UbU_b and u∗u_* varied from 10 cm s−1^{-1} to 40 cm s−1^{-1} and 0.8 to 8 cm s−1^{-1}, respectively. However, Ub/u∗=(8/fdw)1/2U_b/u_*=(8/f_{dw})^{1/2}, related to the Darcy-Weisbach friction factor fdwf_{dw}, varied much less (15-25) as may be anticipated in fully rough flow over a channel bed covered by grains of similar dsd_s. The particle properties ρs/ρ\rho_s/\rho and dsd_s varied from 1.05 to 2.65 and 0.88 to 1.4 mm, respectively. The consequence of these variations is that the empirically derived settling velocity wsw_s is much smaller than the Stokes settling velocity as shown in the inset of Figure 1. Collectively, these experiments span wide-ranging particle sizes (in the SS range) and flow properties from different sources. The lowest measured sediment concentration near the channel bottom is close to the surface (zn,b∈[0.026,0.063]z_{n,b} \in[0.026, 0.063]) but remains above the buffer region z+=u∗zb/ν>30z^+ = u_*z_b/\nu> 30 as shown in Table 1. For some runs, the z+<100z^+ < 100 and wall-blockage effects (not considered here) can impact Eww(k)E_{ww}(k) and du‾(z)/dzd\overline{u}(z)/dz ([53]), which introduce obvious uncertainties. As shown in Table 1, experiments (a)-(c) are characterized by Stb>0.5St_b > 0.5, which may be indicative that Φ(z)\Phi(z) is not small. Experiments (d)-(f) are characterized by a small StbSt_b as assumed by the CSB and Rouse’s formula.

Figure 5 confirms that the fitted Rouse formula and fitted Prandtl formula (i.e. the RR model, allowing β\beta to be fitted) offer good agreements with some measurements (for (a)-(b) and (d)-(f) respectively) at all depths. Given that the simplified CSB model is identical to Rouse’s formula, an agreement between the fitted Rouses’s formula and the measurements can also be juxtaposed to the simplified CSB model. However, the numerical CSB model provides reasonable agreements for all the runs when allowing α\alpha to vary. Allowing α\alpha to be a free parameter has several advantages when compared to β\beta in the fitted Rouse equation. Setting β\beta as constant implies ScSc is constant at all znz_n while setting α\alpha as constant incorporates some of the local variations in ScSc with znz_n (albeit near the free water surface, maintaining a finite ϵ\epsilon can be problematic without adjustments). The impact of minor variations in particle sizes is shown in the shaded area: the particle sizes are increased/decreased by 20% to illustrate model sensitivity to dsd_s. Uncertainty in sediment composition (and thus dsd_s and wsw_s) can be a factor in determining SSC uncertainty but not in all cases (runs d,e,f). While the SSC model does not require uˉ\bar{u} (only du‾/dz\mathrm{d}\overline{u}/\mathrm{d}z), the predicted UbU_b from the CSB turbulent stress budget can be compared against measured UbU_b for a plausibility check. The modeled UbU_b requires u∗u_* along with a boundary condition specified here as u‾(zn,b)/u∗\overline{u}(z_{n,b})/u_* at zn,bz_{n,b}. A number of choices can be made about this boundary condition. Given that zn,bz_{n,b} is sufficiently distant from the wall, the most direct of those choices is the log-law for two end-member cases: (i) fully rough with an externally imposed surface roughness and (ii) hydrodynamically smooth. In both cases, the mean velocity at zn,bz_{n,b} is approximated as

u‾(zb)u∗=1κlog⁡(zbzo);u‾(zb)u∗=1κln⁡(zn,b+)+5,(45)\frac{\overline{u}(z_b)}{u_*} = \frac{1}{\kappa}\log\left(\frac{z_b}{z_o}\right); \qquad\frac{\overline{u}(z_b)}{u_*} = \frac{1}{\kappa}\ln\left(z_{n,b}^{+}\right)+5, \tag*{(45)}

where zoz_o is the momentum roughness length. The zoz_o can be related to dsd_s by zo≈ds/30z_o \approx d_s/30 where the grain diameter is assumed constant. In all cases, the roughness Reynolds number Repa=u∗ds/ν>3Re_{pa}=u_*d_s/\nu>3 but in some cases, the flow is not fully rough (i.e. transitional with 3<Repa<1003<Re_{pa}<100). For this reason, the CSB model forced by both rough and smooth surface boundary conditions at zn,bz_{n,b} are featured in Table 1. The agreement between measured and the range of CSB modeled Ub/u∗U_b/u_* for these two end-member cases appears reasonable. Runs (a) and (f) are closer to a smooth-wall case whereas runs (b), (c), and (e) are better approximated by a rough-wall boundary condition. Run (d) falls in-between these two end-member cases. While Run (f) had the smallest Repa=8Re_{pa}=8 and a near-smooth wall approximation may be justifiable, run (a) had an Repa>100Re_{pa}>100. We do not have a clear explanation as to why UbU_b in run (a) is better approximated by a smooth wall boundary condition.

An investigation of the relation between fitted α\alpha (and β\beta) and ws/u∗w_s/u_* is undertaken and shown in Figure 6. A near-linear relation between α−1\alpha^{-1} and ws/u∗w_s/u_* indirectly supports the heuristic closure adopted for C′w′∂w′/∂z‾\overline{C'w'\partial w'/\partial z} with some caveats.

Figure 6. The dependence of fitted α\alpha and β\beta on the ws/u∗w_s/u_*. The red, blue and black dashed lines show the fitted trend-lines of α−1\alpha^{-1} and β\beta from Rouse and Prandtl equations respectively. The cyan dashed line is β=1+2(ws/u∗)2\beta=1+2(w_s/u_*)^2 [79] extrapolated for large ws/u∗w_s/u_*.

In the regime ws/u∗≫1w_s/u_* \gg1, the closure model with b1∼sgn⁡(Af)u∗/wsb_1 \sim\operatorname{sgn}(A_f)u_*/w_s leads to an α−1∼−sgn⁡(Af)(1−CI)(ws/u∗)\alpha^{-1}\sim-\operatorname{sgn}(A_f)(1-C_I)(w_s/u_*) and β∼−sgn⁡(Af)/(1−CI)(ws/u∗)\beta\sim-\operatorname{sgn}(A_f)/(1-C_I)(w_s/u_*), both of which are negative unless sgn⁡(Af)\operatorname{sgn}(A_f) is negative. The relation in Figure 6 indicates a positive slope between fitted α−1\alpha^{-1} and ws/u∗w_s/u_*, suggesting that the coefficient AfA_f in the flux-variance similarity closure (i.e. equation (21)) is negative. More broadly, to what extend this closure is general and how robust are its results in the context of SSC profile predictions cannot be unpacked from the experiments here and is better kept for a future research topic.

Model Limitations

The treatment of suspended sediments as a dilute mixture is an obvious model limitation. This assumption requires particles to settle independently and that the solid volume can be ignored relative to the water volume. For the experiments considered here, this assumption is reasonable. Another restrictive assumption is setting Φ=0\Phi=0 [43, 14]. A Φ=0\Phi=0 also leads to C‾=w′C′‾/ws→0\overline{C}=\overline{w'C'}/w_s\to0 at zn→1z_n\to1, which may not be general. Given the large vertical gradients in σw2\sigma_w^2 near the channel bottom and near the free water surface, turbophoretic effects can be significant in these two regions [11, 29, 51, 86, 37, 14]. The turbophoretic effect act to increase the SS concentration near the water surface; however, the measurements here (runs a-c) suggest that for the Stb>1St_b>1 cases, the SS concentrations near the water surface experience a decline as zn→1z_n\to1 instead of an increase. This finding can be used to suggest that Φ=0\Phi=0 may be plausible as the turbophoretic term was shown to dominate PhiPhi near the water surface [65, 10]. The CSB budget formulation here (i.e. equation (30)) ignored the flux transfer term and their vertical variation. In the case of the turbulent stress, ignoring the flux transfer term (and its vertical gradients) altogether guarantees that the co-spectrum between w′w' and u′u' in the inertial subrange maintains a k−7/3k^{-7/3} scaling. This k−7/3k^{-7/3} scaling has been observed in numerous boundary layer studies reporting co-spectra thereby offering indirect justification for this assumption. The flux transport terms (i.e. the vertical gradients of triple moments in the Reynolds averaged equations) have also been ignored. These terms have been studied less for stress and sediment flux turbulent budgets compared to their turbulent kinetic energy budget counterparts. The work here highlights the need for an assessment of these terms relative to their mechanical production terms. The CSB model also assumes that the linear Rotta scheme (slow component) with an isotropization of production (rapid component) applies equally to SS and momentum fluxes without adjustments in constants (i.e. AR=1.8A_R = 1.8 and CI=3/5C_I = 3/5). Hence, any departure from these established constants must be absorbed by tww(k)/tr(k)t_{ww}(k)/t_r(k), which manifests itself as a Schmidt number effect (or α\alpha variations).

The assumed shape of Eww(k)E_{ww}(k) is also over-simplified and certainly not reflective of what is known about the energetics near the surface (z+<100z^+ < 100) such as wall-blockage. Moving away from the wall region itself, other ’shape issues’ arise. For example, near the spectral transition from inertial to viscous regimes, usually occurring at around kη≈0.1k\eta\approx0.1, Eww(k)E_{ww}(k) experiences a bottleneck that is absent here [69, 39]. Likewise, as kη>0.1k\eta> 0.1 and increases further into the viscous regime, Eww(k)E_{ww}(k) decays exponentially [62]. Hence, extending the inertial subrange to kη=1k\eta= 1 is not intended to capture all such mechanisms impacting the vertical velocity spectrum. Instead, it allows for some compensation of loss in energy due to censoring Eww(k)E_{ww}(k) at kη=1k\eta= 1 while introducing extra energy due to an expected overestimation of the extrapolated inertial subrange spectrum in this vicinity. On a more positive note, while the full details of the turbulent kinetic energy cascade across scales are not explicitly considered, their effects remain implicitly contained in the assumed shape of Eww(k)E_{ww}(k). As such, some of these effects can be accommodated (e.g. the bottleneck, viscous cutoff, etc...) by various revisions to Eww(k)E_{ww}(k) (e.g. including a bump around kη=0.1k\eta= 0.1, re-solving the viscous cutoff region using the Pao spectral shape or variants [62] on it, etc...).

It is to be noted that the co-spectral budget is integrated scale-wise, which means that the precise shape of Eww(k)E_{ww}(k) in the vicinity of kη≈1k\eta\approx1 is less crucial. Moving beyond the shape issues of Eww(k)E_{ww}(k) and focusing on its primary input variable ϵ(zn)\epsilon(z_n), the approach assumes turbulent kinetic energy production is balanced by its dissipation at every znz_n (i.e. ϕ(zn)=1\phi(z_n) = 1), which is certainly not realistic for all znz_n. However, as previously mentioned, deviations from unity in ϕ(zn)\phi(z_n) may be ameliorated by the sub-unity exponent (−1/3)(-1/3) dependence in the SSC budget. An exception to this statement is the particle time scale twc(k)t_{wc}(k) in Sc(k)Sc(k). A ϕ(zn)=1\phi(z_n) = 1 as zn→1z_n \to1 leads to an unbounded Sc−1(k)Sc^{-1}(k) and thus an uncertain DsD_s shape in the vicinity of the free surface. A plausible adjustment to the Sc−1(k)Sc^{-1}(k) calculations based on maintaining a minimal ϵ(=0.001ϵb)\epsilon(= 0.001\epsilon_b) was introduced here though this correction remains adhoc. Last, the turbulent SS flux from the CSB model(s) follows the same form as gradient-diffusion closure upon ignoring both - turbulent flux transport and scale-wise transfer terms. However, a key advantage here is that the effective diffusion coefficient DsD_s from the CSB model contains contributions from turbulent eddies and Schmidt numbers at all scales. The proposed Schmidt number (or α\alpha) is consistent with bulk Schmidt number formulations such as those by van Rijn’s and other one-way coupling schemes (i.e. particle transport does not impact the flow) when Sc<1Sc < 1 [7]. For dense mixture or other aeolian particles in the atmosphere, the particle Schmidt number can be larger than unity [16] implying other particle-fluid interaction models are required. When using the CSB model, the α\alpha used for the determination of the Schmidt number is treated as a single fitted parameter. Hence, the CSB model offers the same number of free parameters as the fitted Rouse equation. What was found here is that α−1\alpha^{-1} varies linearly with ws/u∗w_s/u_* when combining all the experiments. A plausibility argument as to why α\alpha depends on ws/u∗w_s/u_* was also offered. In some instances, the addition of a single fitted parameter may be desirable in hydraulic models as discussed elsewhere [61, 5, 68] (S. Li et al. 2019), but an increasing number of free model parameters does not necessarily lead to a better physical understanding. The sediment settling velocity estimated in equation 7 is commonly based on a mass-median-diameter from particle size distribution measurements, which however may not be an optimized characteristic size as shown by some in-situ measurements [84]. Large variations in dsd_s can have a substantial impact on SSC profiles, which may be more significant than models for α\alpha.

Conclusion

Operational modeling of SSC in turbulent flows continues to be a formidable challenge in hydraulics, hydrology, ecology, and water quality control. The work here establishes a new link between the spectrum of vertical velocity and SS turbulent flux, which was then used to arrive at expressions for the SSC profile. The spectrum of vertical velocity is characterized by multiple scaling regimes that include the Saffman spectrum (Eww(k)∼k+2E_{ww}(k) \sim k^{+2}), the ’energy splashing’ effect due to the presence of a wall (Eww(k)∼k0E_{ww}(k) \sim k^{0}), and the much-studied inertial subrange regime (Eww(k)∼k−5/3E_{ww}(k) \sim k^{-5/3}). Finite Reynolds effects are accommodated through a scale separation between zz and the Kolmogorov microscale η\eta terminating the scale-wise extent of the inertial subrange (as a first approximation). This dependence can be noted when considering the scaling argument ke/ko=z/η∼(zu∗/ν)3/4k_e/k_o = z/\eta\sim(zu_*/\nu)^{3/4} ([74]). Hence, increasing Res=(zu∗/ν)Re_s = (zu_*/\nu) by either increasing zz or u∗u_* leads to a widening of the scale-wise extent of the inertial subrange, which then impacts all subsequent expressions such as Ω(zn)\Omega(z_n) and du‾/dzd\overline{u}/dz. As such, the proposed model is responsive to finite Reynolds number, Schmidt number, and Rouse number effects. Prior ad-hoc efforts such as correcting lol_o by VnV_n (i.e. the van Driest damping function) can now be interpreted from this new spectral perspective (i.e. ResRe_s effects become large for small zz or u∗u_*). A simplified solution to the CSB model in which the Saffman spectrum is truncated but the inertial subrange is now extended to infinite wave-numbers (i.e. Res→∞Re_s \to\infty) was shown to recover earlier theories (e.g. Rouse’s formula). The fitted Rouse’s equation (and by extension the simplified CSB solution) also describes the measured SSC profiles in all the experiments considered here provided α\alpha (or β\beta) is allowed to vary with ws/u∗w_s/u_*. Thus, one of the main novelties here is to provide a spectral link between the energy distribution in eddies and the SSC shape. Interactions between turbulent eddies and suspended sediment grains at various heights were also proposed, resulting in a scale-dependent ScSc captured by a single parameter α\alpha that varies with ws/u∗w_s/u_*. Such ScSc variations were formulated in spectral space but recover expected bulk relations between RR and ScSc identified by other models, experiments, and simulation studies. When all these findings are taken together, future extension of this work must focus on upgrading the particle-turbulence interaction scheme and its signature in a scale-dependent Schmidt number. Such extension will benefit from targeted DNS runs where all the terms in the particle co-spectrum as well as Eww(k)E_{ww}(k) can be computed or determined. Likewise, an exploration of where the sediment flux transport term is significant relative to the mechanical production term and how to incorporate its effects can be undertaken from the aforementioned DNS runs.

Data Availability and Acknowledgements

All the data used were digitized from the published literature [27, 78, 35, 75]. SL was supported by a fellowship from the Nicholas School of the Environment at Duke University. GK and ADB acknowledge support from the U.S. National Science Foundation (NSF-AGS-1644382, NSF-AGS-2028633, NSF-IOS-1754893, and NSF-CBET-2042346).

References

  1. [1]Albertson, J. D., Parlange, M. B., Katul, G. G., Chu, C.-R., Stricker, H., & Tyler, S. (1995). Sensible heat flux from arid regions: A simple flux-variance method. Water Resources Research, 31(4), 969–973.DOI
  2. [2]Ali, S. Z., & Dey, S. (2017). Origin of the scaling laws of sediment transport. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 473(2197), 20160785.
  3. [3]Ali, S. Z., & Dey, S. (2018). Impact of phenomenological theory of turbulence on pragmatic approach to fluvial hydraulics. Physics of Fluids, 30(4), 045105.DOI
  4. [4]Ayet, A., & Katul, G. (2020). Scaling laws for the length scale of energy-containing eddies in a sheared and thermally stratified atmospheric surface layer. Geophysical Research Letters, 47(23), e2020GL089997.DOI
  5. [5]Battiato, I., & Rubol, S. (2014). Single-parameter model of vegetated aquatic flows. Water Resources Research, 50(8), 6358–6369.
  6. [6]Bombardelli, F., & Jha, S. (2009). Hierarchical modeling of the dilute transport of suspended sediment in open channels. Environmental Fluid Mechanics, 9(2), 207.
  7. [7]Bombardelli, F., & Moreno, P. (2012). Exchange at the bed sediments-water column interface. In Fluid Mechanics of Environmental Interfaces, 2nd ed.: Gualtieri, C., Mihailovic, D.T, Eds (pp. 221–253).DOI
  8. [8]Bonetti, S., Manoli, G., Manes, C., Porporato, A., & Katul, G. (2017). Manning’s formula and Strickler’s scaling explained by a co-spectral budget model. Journal of Fluid Mechanics, 812, 1189–1212.DOI
  9. [9]Bos, W., Touil, H., Shao, L., & Bertoglio, J.-P. (2004). On the behavior of the velocity-scalar cross correlation spectrum in the inertial range. Physics of Fluids, 16(10), 3818–3823.DOI
  10. [10]Bragg, A. D., Richter, D. H., & Wang, G. (2021). Mechanisms governing the settling velocities and spatial distributions of inertial particles in wall-bounded turbulence. Physical Review Fluids, 6(6), 064302.DOI
  11. [11]Caporaloni, M., Tampieri, F., Trombetti, F., & Vittori, O. (1975). Transfer of particles in nonisotropic air turbulence. Journal of Atmospheric Sciences, 32(3), 565–568.doi.org/10.1175/1520-0469(1975)032<0565:topina>2.0.co;2
  12. [12]Castro-Orgaz, O., Giráldez, J., Mateos, L., & Dey, S. (2012). Is the von kármán constant affected by sediment suspension? Journal of Geophysical Research: Earth Surface, 117(F4).DOI
  13. [13]Cava, D., & Katul, G. (2012). On the scaling laws of the velocity-scalar cospectra in the canopy sublayer above tall forests. Boundary-layer Meteorology, 145(2), 351–367.DOI
  14. [14]Chamecki, M., Van Hout, R., Meneveau, C., & Parlange, M. B. (2007). Concentration profiles of particles settling in the neutral and stratified atmospheric boundary layer. Boundary-layer Meteorology, 125(1), 25–38.DOI
  15. [15]Cheng, N.-S. (1997). Simplified settling velocity formula for sediment particle. Journal of Hydraulic Engineering, 123(2), 149–152.DOI
  16. [16]Csanady, G. (1963). Turbulent diffusion of heavy particles in the atmosphere. Journal of Atmospheric Sciences, 20(3), 201–208.
  17. [17]Dai, Z., Fagherazzi, S., Mei, X., & Gao, J. (2016). Decline in suspended sediment concentration delivered by the Changjiang (Yangtze) River into the East China Sea between 1956 and 2013. Geomorphology, 268, 123–132.DOI
  18. [18]Dey, S. (2014). Fluvial hydrodynamics. Springer.
  19. [19]Dey, S., & Ali, S. Z. (2020). Fluvial instabilities. Physics of Fluids, 32(6), 061301.
  20. [20]Duman, T., Trakhtenbrot, A., Poggi, D., Cassiani, M., & Katul, G. G. (2016). Dissipation intermittency increases long-distance dispersal of heavy particles in the canopy sublayer. Boundary-Layer Meteorology, 159(1), 41–68.DOI
  21. [21]Einstein, H. (1950). The bed-load function for sediment transportation in open channel flows (No. 1026). Technical Report, US Department of Agriculture, Soil Conservation Service, Washington DC.
  22. [22]Elghobashi, S. (1994). On predicting particle-laden turbulent flows. Applied Scientific Research, 52(4), 309–329.DOI
  23. [23]Ferry, J., & Balachandar, S. (2001). A fast Eulerian method for disperse two-phase flow. International Journal of Multiphase Flow, 27(7), 1199–1226.DOI
  24. [24]Fischer, H. B., List, J. E., Koh, C. R., Imberger, J., & Brooks, N. H. (2013). Mixing in Inland and Coastal Waters. Elsevier.
  25. [25]Giometto, M., Katul, G., Fang, J., & Parlange, M. (2017). Direct numerical simulation of turbulent slope flows up to Grashof number Gr = 2.1 × 10¹¹. Journal of Fluid Mechanics, 829, 589.
  26. [26]Green, M., & Coco, G. (2014). Review of wave-driven sediment resuspension and transport in estuaries. Reviews of Geophysics, 52(1), 77–117.DOI
  27. [27]Greimann, B., & Holly Jr, F. (2001). Two-phase flow analysis of concentration profiles. Journal of Hydraulic Engineering, 127(9), 753–762.DOI
  28. [28]Greimann, B., Muste, M., & Holly Jr, F. (1999). Two-phase formulation of suspended sediment transport. Journal of Hydraulic Research, 37(4), 479–500.DOI
  29. [29]Guha, A. (1997). A unified Eulerian theory of turbulent deposition to smooth and rough surfaces. Journal of Aerosol Science, 28(8), 1517–1537.DOI
  30. [30]Heisel, M., Katul, G., Chamecki, M., & Guala, M. (2020). Velocity asymmetry and turbulent transport closure in smooth-and-rough-wall boundary layers. Physical Review Fluids, 5(10), 104605.DOI
  31. [31]Huai, W., Yang, L., & Guo, Y. (2020). Analytical solution of suspended sediment concentration profile: Relevance of dispersive flow term in vegetated channels. Water Resources Research, 56(7), e2019WR027012.DOI
  32. [32]Huai, W., Yang, L., Wang, W.-J., Guo, Y., Wang, T., & Cheng, Y.-g. (2019). Predicting the vertical low suspended sediment concentration in vegetated flow using a random displacement model. Journal of Hydrology, 578, 124101.DOI
  33. [33]Huang, C.-W., Launiainen, S., Grönholm, T., & Katul, G. G. (2014). Particle deposition to forests: An alternative to K-theory. Atmospheric Environment, 94, 593–605.DOI
  34. [34]Hunt, J. (1954). The turbulent transport of suspended sediment in open channels. Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences, 224(1158), 322–335.
  35. [35]Jha, S. K., & Bombardelli, F. A. (2009). Two-phase modeling of turbulence in dilute sediment-laden, open-channel flows. Environmental Fluid Mechanics, 9(2), 237.DOI
  36. [36]Johnson, P. L., Bassenne, M., & Moin, P. (2020). Turbophoresis of small inertial particles: theoretical considerations and application to wall-modelled large-eddy simulations. J. Fluid Mech., 883, A27. doi: 10.1017/jfm.2019.865
  37. [37]Katul, G. G., Grönholm, T., Launiainen, S., & Vesala, T. (2010). Predicting the dry deposition of aerosol-sized particles using layer-resolved canopy and pipe flow analogy models: Role of turbophoresis. Journal of Geophysical Research: Atmospheres, 115(D12).DOI
  38. [38]Katul, G. G., & Manes, C. (2014). Cospectral budget of turbulence explains the bulk properties of smooth pipe flow. Physical Review E, 90(6), 063008.
  39. [39]Katul, G. G., Manes, C., Porporato, A., Bou-Zeid, E., & Chamecki, M. (2015). Bottlenecks in turbulent kinetic energy spectra predicted from structure function inflections using the von kármán-howarth equation. Physical Review E, 92(3), 033009.DOI
  40. [40]Katul, G. G., Porporato, A., Manes, C., & Meneveau, C. (2013). Co-spectrum and mean velocity in turbulent boundary layers. Physics of Fluids, 25(9), 091702.DOI
  41. [41]Kellogg, M. L., Smyth, A. R., Luckenbach, M. W., Carmichael, R. H., Brown, B. L., Cornwell, J. C., . . . Higgins, C. B. (2014). Use of oysters to mitigate eutrophication in coastal waters. Estuarine, Coastal and Shelf Science, 151, 156–168.DOI
  42. [42]Kim, J., Moin, P., & Moser, R. (1987). Turbulence statistics in fully developed channel flow at low Reynolds number. Journal of Fluid Mechanics, 177, 133–166.DOI
  43. [43]Kind, R. (1992). One-dimensional aeolian suspension above beds of loose particles—a new concentration-profile equation. Atmospheric Environment. Part A. General Topics, 26(5), 927–931.DOI
  44. [48]Liu, T., Xia, X., Liu, S., Mou, X., & Qiu, Y. (2013). Acceleration of denitrification in turbid rivers due to denitrification occurring on suspended sediment in oxic waters. Environmental Science and Technology, 47(9), 4053–4061.DOI
  45. [49]Long, C. M., & Pavelsky, T. M. (2013). Remote sensing of suspended sediment concentration and hydrologic connectivity in a complex wetland environment. Remote Sensing of Environment, 129, 197–209.DOI
  46. [50]Lupker, M., France-Lanord, C., Lavé, J., Bouchez, J., Galy, V., Métivier, F., . . . Mugnier, J. (2011). A Rouse-based method to integrate the chemical composition of river sediments: Application to the Ganga basin. Journal of Geophysical Research: Earth Surface, 116(F4).DOI
  47. [51]Marchioli, C., & Soldati, A. (2002). Mechanisms for particle transfer and segregation in a turbulent boundary layer. Journal of Fluid Mechanics, 468, 283.DOI
  48. [52]Mazumder, B., & Ghoshal, K. (2006). Velocity and concentration profiles in uniform sediment-laden flow. Applied Mathematical Modelling, 30(2), 164–176.DOI
  49. [53]McColl, K., Katul, G., Gentine, P., & Entekhabi, D. (2016). Mean-velocity profile of smooth channel flow explained by a cospectral budget model with wall-blockage. Physics of Fluids, 28(3), 035107.DOI
  50. [54]Mohtar, W. H. M. W., Lee, J. W., Azha, N. I. M., & Cheng, N.-S. (2020). Incipient sediment motion based on turbulent fluctuations. International Journal of Sediment Research, 35(2), 125–133.DOI
  51. [55]Muste, M., Yu, K., Fujita, I., & Ettema, R. (2005). Two-phase versus mixed-flow perspective on suspended sediment transport in turbulent channel flows. Water Resources Research, 41(10).DOI
  52. [56]Nazeer, S., Hashmi, M. Z., & Malik, R. N. (2014). Heavy metals distribution, risk assessment and water quality characterization by water quality index of the river soan, pakistan. Ecological Indicators, 43, 262–270.DOI
  53. [57]Nezu, I., & Azuma, R. (2004). Turbulence characteristics and interaction between particles and fluid in particle-laden open channel flows. Journal of Hydraulic Engineering, 130(10), 988–1001.
  54. [58]Nie, S., Sun, H., Zhang, Y., Chen, D., Chen, W., Chen, L., & Schaefer, S. (2017). Vertical distribution of suspended sediment under steady flow: existing theories and fractional derivative model. Discrete Dynamics in Nature and Society, 2017.DOI
  55. [59]Nikora, V. I., & Goring, D. G. (2002). Fluctuations of suspended sediment concentration and turbulent sediment fluxes in an open-channel flow. Journal of Hydraulic Engineering, 128(2), 214–224.DOI
  56. [60]O’Brien, M. (1933). Review of the theory of turbulent flow and its relation to sediment-transportation. EOS Transactions American Geophysical Union, 14(1), 487–491.DOI
  57. [61]Papke, A., & Battiato, I. (2013). A reduced complexity model for dynamic similarity in obstructed shear flows. Geophysical Research Letters, 40(15), 3888–3892.
  58. [62]Pope, S. (2000). Turbulent flows. Cambridge University Press, Cambridge, U.K.
  59. [63]Raupach, M. (1981). Conditional statistics of Reynolds stress in rough-wall and smooth-wall turbulent boundary layers. Journal of Fluid Mechanics, 108, 363–382.DOI
  60. [64]Reeks, M. (1983). The transport of discrete particles in inhomogeneous turbulence. J. Aerosol Sci., 14(6), 729 - 739. doi: https://doi.org/10.1016/0021-8502(83)90055-1
  61. [65]Richter, D., & Chamecki, M. (2018). Inertial effects on the vertical transport of suspended particles in a turbulent boundary layer. Boundary-Layer Meteorology, 167(2), 235–256.DOI
  62. [66]Rouse, H. (1937). Modern conceptions of the mechanics of fluid turbulence. Transactions of the American Society of Civil Engineers, 102(1), 463–505.DOI
  63. [67]Rouse, H. (1939). An analysis of sediment transportation in the light of fluid turbulence. (SCS-TP-25), 1–40.
  64. [68]Rubol, S., Ling, B., & Battiato, I. (2018). Universal scaling-law for flow resistance over canopies with complex morphology. Scientific Reports, 8(1), 4430.DOI
  65. [69]Saddoughi, S. G., & Veeravalli, S. V. (1994). Local isotropy in turbulent boundary layers at high reynolds number. Journal of Fluid Mechanics, 268, 333–372.
  66. [70]Sardina, G., Schlatter, P., Brandt, L., Picano, F., & Casciola, C. M. (2012). Wall accumulation and spatial localization in particle-laden wall flows. J. Fluid Mech., 699, 50–78.DOI
  67. [71]Scully, M. E., & Friedrichs, C. T. (2003). The influence of asymmetries in overlying stratification on near-bed turbulence and sediment suspension in a partially mixed estuary. Ocean Dynamics, 53(3), 208–219.DOI
  68. [72]Sun, L., Qiu, H., Wu, C., Niu, J., & Hu, B. X. (2020). A review of applications of fractional advection–dispersion equations for anomalous solute transport in surface and subsurface water. Wiley Interdisciplinary Reviews: Water, 7(4), e1448.DOI
  69. [73]Tan, G., Fang, H., Dey, S., & Wu, W. (2018). Rui-Jin Zhang’s research on sediment transport. Journal of Hydraulic Engineering, 144(6), 02518002.DOI
  70. [74]Tennekes, H., & Lumley, J. (1972). A first course in turbulence. MIT press, Boston Massachusetts, USA.
  71. [75]Tseng, C., & Tinoco, R. (2020). A two-layer turbulence-based model to predict suspended sediment concentration in flows with aquatic vegetation. Geophysical Research Letters, e2020GL091255.
  72. [76]Umeyaina, M. (1992). Vertical distribution of suspended sediment in uniform open-channel flow. Journal of Hydraulic Engineering, 118(6), 936–941.
  73. [77]Van Driest, E. (1956). On turbulent flow near a wall. Journal of the Aeronautical Sciences, 23(11), 1007–1011.
  74. [78]Vanoni, V. A. (1984). Fifty years of sedimentation. Journal of Hydraulic Engineering, 110(8), 1021–1057.
  75. [79]van Rijn, L. C. (1984). Sediment transport, part ii: suspended load transport. Journal of Hydraulic Engineering, 110(11), 1613–1641.
  76. [80]von Karman, T. (1934). Some aspects of the turbulence problem. In Proceedings of the Fourth International Congress of Applied Mechanics, Cambridge, England (pp. 54–91).
  77. [81]Wallin, S., & Johansson, A. (2000). An explicit algebraic Reynolds stress model for incompressible and compressible turbulent flows. Journal of Fluid Mechanics, 403, 89–132.DOI
  78. [82]Wang, X., & Qian, N. (1992). Velocity profiles of sediment-laden flow. International Journal of Sediment Research, 7(1), 27–58.
  79. [83]Wells, M., & Stock, D. (1983). The effects of crossing trajectories on the dispersion of particles in a turbulent flow. Journal of Fluid Mechanics, 136, 31–62.DOI
  80. [84]Williams, N., Walling, D., & Leeks, G. (2007). High temporal resolution in situ measurement of the effective particle size characteristics of fluvial suspended sediment. Water Research, 41(5), 1081–1093.
  81. [85]Yujun, Y., Zhaoyin, W., Zhang, K., Guoan, Y., & Xuehua, D. (2008). Sediment pollution and its effect on fish through food chain in the yangtze river. Inter-national Journal of Sediment Research, 23(4), 338–347.
  82. [86]Zhao, B., & Wu, J. (2006). Modeling particle deposition from fully developed turbu-lent flow in ventilation duct. Atmospheric Environment, 40(3), 457–466.DOI

Paper details

vlnc:W_ykyqcj6t
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents