Вы находитесь на странице: 1из 11

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/237188911

Modelling cohesive sediment transport in rivers

Article  in  Canadian Journal of Civil Engineering · February 2011


DOI: 10.1139/l04-043

CITATIONS READS
25 186

2 authors, including:

B. G. Krishnappan
Government of Canada
214 PUBLICATIONS   1,562 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Modelling of cohesive sediment transport in the Glenmore Reservoir in Calgary, Alberta, Canada View project

All content following this page was uploaded by B. G. Krishnappan on 15 February 2016.

The user has requested enhancement of the downloaded file.


749

Numerical modelling of cohesive sediment


transport in rivers
David H. Willis and B.G. Krishnappan

Abstract: Techniques available to practicing civil engineers for numerically modelling cohesive mud in rivers and estu-
aries are reviewed. Coupled models, treating water and sediment as a single process, remain research tools but are usu-
ally not three-dimensional. The decoupled approach, which separates water and sediment computations at each model
time step, allows the three-dimensional representation of at least the bed and the use of well-proven, commercial, nu-
merical, hydrodynamic models. Most hydrodynamic models compute sediment transport in suspension but may require
modification of the dispersion coefficients to account for the presence of sediment. The sediment model deals with the
sediment exchange between the water column and the bed using existing equations for erosion and deposition. Both
equations relate the sediment exchange rates to the shear stress in the bottom boundary layer. In real rivers and estuar-
ies, a depositional bed layer is associated with a period of low flow and shear, at slack tide for example, whereas in
numerical models a layer is defined by the model time step. The sediment model keeps track of the uppermost layers
at each model grid point, including consolidation and strengthening. Although numerical hydrodynamic models are
based strongly on physics, sediment models are only numerical frameworks for interpolating and extrapolating full-
scale field or laboratory measurements of “hydraulic sediment parameters,” such as threshold shear stresses. Calibration
and verification of models against measurement are therefore of prime importance.
Key words: cohesive sediment, mathematical modelling, settling velocity, erosion, resuspension, deposition, fluid mud,
bed layers.
Résumé : Cet article passe en revue les techniques disponibles dont disposent les ingénieurs civils praticiens pour la
modélisation numérique des boues cohésives dans les rivières et les estuaires. Des modèles couplés, traitant l’eau et les
sédiments comme étant un seul processus, demeurent des outils de recherche mais ne sont pas normalement tridimen-
sionnels. L’approche découplée, qui sépare les calculs de l’eau et des sédiments à chaque pas de calcul du modèle, per-
met une représentation tridimensionnelle d’au moins le lit et l’utilisation de modèles hydrodynamiques numériques,
industriels et bien éprouvés. La majorité des modèles hydrodynamiques calculent le transport des sédiments en suspen-
sion, mais peuvent demander une modification des coefficients de dispersion afin de tenir compte de la présence des
sédiments. Le modèle sur les sédiments traite de l’échange de sédiments entre la colonne d’eau et le lit en utilisant les
équations existantes pour l’érosion et la déposition. Les deux équations relient les taux d’échange des sédiments à la
contrainte de cisaillement dans la couche limite du fond. Dans les vrais estuaires et rivières, une couche de déposition
sur le lit est associée à une période de faible débit et de cisaillement, par exemple à l’étale de courant, alors que dans
les modèles numériques, une couche est identifiée par le pas de calcul du modèle. Le modèle sur les sédiments suit
l’évolution des couches supérieures à chaque point grille du modèle, incluant la consolidation et le renforcement. Bien
que les modèles numériques soient basés en grande partie sur la physique, les modèles sur les sédiments ne sont que
des cadres numériques pour l’interpolation et l’extrapolation des mesures en laboratoire ou à plein échelle sur le terrain
des « paramètres hydrauliques des sédiments », tels que les seuils de contraintes de cisaillement. L’étalonnage et la vé-
rification des modèles par rapport aux mesures sont donc d’une grande importance.

Mots clés : sédiments cohésifs, modélisation mathématique, vitesse de sédimentation, érosion, remise en suspension,
déposition, boues fluides, couches du lit.
[Traduit par la Rédaction] Willis and Krishnappan 758

Received 18 August 2003. Revision accepted 30 April 2004. Published on the NRC Research Press Web site at http://cjce.nrc.ca on
1 October 2004.
D.H. Willis. David H. Willis and Associates Limited, Ottawa, ON K1K 1X4, Canada.
B.G. Krishnappan.1 National Water Research Institute, Burlington, ON L7R 4A6, Canada.
Written discussion of this article is welcomed and will be received by the Editor until 28 February 2005.
1
Corresponding author (e-mail: Bommanna.Krishnappan@ec.gc.ca).

Can. J. Civ. Eng. 31: 749–758 (2004) doi: 10.1139/L04-043 © 2004 NRC Canada
750 Can. J. Civ. Eng. Vol. 31, 2004

1. Introduction where c is the mean concentration of sediment in suspen-


sion; t is the time; ui and ui are the mean and instantaneous
Transport processes of fine-grained, cohesive sediments flow velocities, respectively; ui′ and c′ are the fluctuating ve-
are significantly different from those of coarse-grained, locity components and fluctuating sediment concentration,
cohesionless sediments. The main difference is in the way respectively; w is the vertical velocity component; ws is the
the particles interact. For example, fine-grained sediment settling velocity of the sediment particle; and S(x, y, z) is the
particles in the silt and clay size classes have a tendency to sediment source or sink within the solution domain other
form agglomerations of particles called flocs, whereas than the boundaries, where x and y are the horizontal spatial
coarse-grained particles in the sand and gravel size classes components and z is the vertical spatial component.
behave as individual particles. While the size and the spe- The boundary conditions are as follows:
cific weight of the individually transported cohesionless sed-
iment particles are well defined and invariant with the flow [2] (w − w s) c + w ′c′ = 0 (at the free surface)
field, the same properties of the cohesive sediment floc are a [3] − w sc + w ′c′ = q d + q e (at the bed)
function of the flow field (van Leussen 1988; Krishnappan et
al. 1992; Lau and Krishnappan 1994a; Krishnappan and where qd and qe are fluxes due to deposition and erosion, re-
Engel 1997). These parameters have to be determined before spectively; and w′ is the fluctuating velocity component in
the modelling of the cohesive sediment transport can be at- the vertical direction.
tempted. In other words, the size and density of the cohesive The turbulent flux of sediment is usually determined using
sediment floc themselves become dependent variables, the eddy diffusivity concept:
thereby increasing the complexity of the cohesive sediment ∂c
transport models (Krishnappan and Marsalek 2000). In addi- [4] ui ′ c′ = −Γsi
tion, the agglomeration or flocculation of the cohesive sedi- ∂x i
ment depends on a number of factors such as particle where Γs is the diffusion coefficient.
mineralogy, the electrochemical nature of the flowing me- If we assume that the presence of sediment does not have
dium, biological factors such as bacteria and other organic an effect on the flow velocity components, then the flow
material, and the hydrodynamic properties of the flow field field can be calculated using a hydrodynamic model inde-
and hence requires a multidisciplinary approach to model- pendently (decoupled approach). In such cases, to close
ling cohesive sediment transport processes. eq. [1], we need to know ws, the expressions for qe and qd at
Past research on cohesive sediment transport was moti- the bed, and the sediment diffusion coefficient, Γs. Equa-
vated by the need for better estimates of soil erosion, shoal- tions [1]–[4] are equally valid for cohesionless sediment.
ing of channels and harbours, and dredging requirements for The difference between a cohesive and a cohesionless sedi-
navigational purposes; as a result, researchers were mainly ment transport model arises because of the differences in the
concerned with the cohesive sediment transport in the estu- specification of the parameters ws, qe, and qd for the two
aries and coastal areas. Recently, there is an additional moti- types of sediments.
vation prompted by the need for a better understanding of Models based on the decoupled approach are valid only
transport of pollutants that are attached to cohesive sedi- for low concentrations of the cohesive sediment in the flow
ment. A majority of highly toxic and persistent chemicals (i.e., for concentrations up to about 300 mg/L). For higher
entering a river system from agricultural, industrial, and mu- concentrations (i.e., for concentrations of the order of 1 g/L
nicipal sources have high affinity for fine particles and are and higher), the presence of sediment has an effect on the
transported mostly in association with the cohesive sedi- turbulence of the flow and modifies the turbulent fluxes of
ment. Therefore, many existing ecosystem models dealing momentum and concentration, eddy diffusivity, eddy viscos-
with contaminant transport, fate, and bioaccumulation in ity, and the bottom shear stress. For such flows a coupled
aquatic environments invariably include a cohesive sediment approach is needed. Teisson et al. (1991) proposed two dif-
transport component and require a better understanding of ferent approaches for modelling such flows. The first
the cohesive sediment transport processes. In this paper, an approach is based on a Reynolds stress model, which is ca-
overview of the knowledge base required for modelling co- pable of dealing with stratified flows. The second approach
hesive sediment transport and the associated contaminants in is based on a two-phase flow model, which takes into ac-
river flows is reviewed. count the interaction between the fluid and sediment in a di-
rect way. Le Hir (1997) investigated the capabilities of the
1.1. Governing equation two-phase flow models to predict the fluid-mud flows in the
The transport characteristics of cohesive sediment in a Loire estuary in France.
flow field can be described in terms of a sediment mass bal- Although the coupled models have the advantage of mini-
ance equation. A form of the equation that was used by mizing the empiricism in the cohesive sediment transport
Teisson (1997) is adopted here. In this form, the equation models and predict the vertical profile of sediment concen-
and the boundary conditions are as follows: tration from the water surface to the bed, they are still con-
sidered as “research” models and have not yet been applied
for engineering problems. For engineering applications, a
∂c ∂ c ∂ (w − w s) c ∂ (u′ i c′) multilayer concept (Le Normant 2000; Willis and Crook-
[1] + ui + =− + S(x, y, z),
∂t ∂x i ∂z ∂x i shank 1997) is used and equations are solved for water col-
umn, sediment in suspension, fluid mud, and consolidated
xi = x, y, z and ui = u, v, w sediment deposits taking into consideration the net fluxes

© 2004 NRC Canada


Willis and Krishnappan 751

among different layers. In Willis and Crookshank (1997), a tion, breaking up and resuspending flocs before they can set-
multilayer cohesive sediment transport model was applied to tle through the boundary layer onto the bed. There is there-
Cumberland Basin in the Bay of Fundy, Liverpool Bay, and fore a maximum threshold shear for deposition. The
Miramichi inner bay in Canada. In the present paper, the threshold shears are properties of the sediment. The shear
state-of-the-art of formulating multilayer cohesive sediment stress on the bed is a property of the flow and the resistance
transport models for rivers and estuaries is examined by re- to flow of the bedforms. Bed shear also correlates with the
viewing the physical processes involved in different layers level of turbulence of the flow.
and their governing equations. Our numerical hydrodynamic models then need to calcu-
late shear stress rather accurately. Most models are very
2. Hydrodynamic models good at calculating flow velocity and water level, and also
shear stress if they know the bed roughness. On sediment
Numerical hydrodynamic models form the basis of the co- beds, however, the definition of roughness is circular: ripples
hesive sediment transport models. Since it is the water that are generated by the flow, are properties of the flow and sed-
transports the cohesive sediment and contaminants, we need iment, and are the roughness that resists the flow. There is a
to have a good understanding of the flow. The cohesive sedi- suspicion that the ripples that form on sediment beds offer
ment transport is most important in river estuaries, where the minimum resistance to the flow that forms them, but this
fresh and salt water mix under the influence of tides and has yet to be proven.
ocean currents and the river flow itself. The hydrodynamic In steady unidirectional flow, bed shear (τ) is calculated by
models, therefore, need to be complex: a minimum of two-
dimensional (2-D), often with layers in the third vertical di- [5] τ = ρ wghI
mension; increasingly, fully three-dimensional (3-D). where ρ w is the density of water, g is the acceleration due to
gravity, h is the water depth, and I is the water surface or en-
2.1. Governing equations and processes ergy slope. Bed shear is often written with units of velocity as
Hydrodynamic models such as TELEMAC-3D (Hervouet
2000) solve the full Navier–Stokes equations. The presence [6] u* = τ / ρ w = ghI
of a free surface is accounted for with the help of an appro-
priate boundary condition. The pressure distribution is nor- in which u* is the shear velocity or “u-star.”
mally assumed to be hydrostatic. The models also have to Most numerical hydraulic models ask for a Chézy or
deal with the effects of vertical density resulting from tem- Manning expression of the roughness of the bed at each cal-
perature, suspended sediment, and salinity variations. The culation point. The Chézy coefficient might better be called
effects of wind stress on the free surface, heat exchange with a “smoothness” coefficient, relating the bed shear to the
the atmosphere, and Coriolis forces are normally included. mean flow velocity as

2.2. Flow resistance: viscosity, boundary layer, [7] ν = C CHu *


roughness, and shear where ν is the mean velocity averaged over depth, and CCH
Laminar flow is uninteresting in real rivers and estuaries is the dimensionless Chézy smoothness coefficient.
where the flow is fully turbulent. Even at the turn of the tide “Steady uniform flow” and “mean velocity averaged over
(near high water and low water, when the mean flow veloc- depth” have little meaning in a 3-D or layered 2-D model of
ity over the depth is zero), there is residual turbulence, par- an estuary, where salt water continues to flood upstream
ticularly in the mixing layer between the salt water, which near the bed while fresh water ebbs on the surface. Never-
continues to flow upstream near the bed, and the fresh water theless, eq. [7] remains useful as a definition of both CCH
flowing seaward near the surface. and ν.
There can still be a thin laminar boundary layer over a The Chézy coefficient is related to the bed roughness by
rare smooth bed, in which water velocity increases linearly,
at constant shear stress defined by viscosity, from zero at the [8] C CH = 5.75 log10(11h / r)
bed to the familiar logarithmic turbulent velocity profile a
small distance above the bed. Otherwise, the viscosity of the where r is a physical dimension of the bed roughness. In
water is not a major parameter for the flow. It will come into wave-generated sand ripples, for example,
its majority when we consider settling of suspended sedi-
ment (see Sect. 3.1.1, eq. [20]). [9] r ≈ 25R2/λ
Most natural rivers and estuaries are not smooth but have where R is the ripple height, trough to crest; and λ is the rip-
rocks and ripples projecting through the laminar sublayer, ple wavelength, crest to crest. On flat beds,
directly resisting the turbulent flow above it. The resistance
is again expressed as a shear stress, by the bed on the flow- [10] r ≈ 2D90
ing water and equally and oppositely by the water on the
bed. where D90 is the (coarse) grain size of which 90% of the bed
Bed shear is the most important flow property for model- is smaller.
ling cohesive sediment transport (Krishnappan 1990, 1991). We can see why the Chézy coefficient is called a smooth-
Shear is obviously the forcing function for erosion of the co- ness coefficient: the rougher the bed, the larger the rough-
hesive bed, when shear exceeds the (minimum) threshold ness, r, and the smaller the value of CCH; and the smoother
shear of the bed material. Less obviously it affects deposi- the bed, the smaller the roughness and the larger the value of

© 2004 NRC Canada


752 Can. J. Civ. Eng. Vol. 31, 2004

CCH. In practice, a very smooth bed has a CCH > 30, a very ical solution techniques. Excessive numerical dispersion will
rough bed CCH < 10, and a fairly smooth bed (muddy bed) smear the concentration profiles and yield negative values
CCH ≈ 25. for concentrations, which in turn can alter the nature of the
Other roughness measures can be derived from the Chézy source and sink terms; sources can become sinks and vice
coefficient as the Weisbach friction factor versa.
[11] f = 2.83 / C CH
2.5. Other estuarine processes: waves and tides, wetting
and the Manning roughness and drying
Both wind waves and tides are water waves. The differ-
[12] n = (1.49h1/ 6)/(C CH g1/ 2) ences lie in the frequencies and modes of propagation. Most
Canadian tides are semidiurnal (i.e., two high and two low
2.3. Representations of turbulence waters per day, with the tidal period between highs or lows
Turbulence plays an important role in maintaining the fine of about 12 h and 24 min). Wind waves, on the other hand,
sediment in suspension and in promoting the flocculation of have periods of the order of 10 s and frequencies of about
the fine suspended sediment. For maintaining the sediment 0.1 Hz.
in suspension, the turbulent diffusive flux counteracts the Most numerical hydraulic models used for cohesive sedi-
settling flux of the suspended sediment. To promote the floc- ment transport in rivers can handle tidal variations. Tidal flow
culation of the fine suspended sediment, the turbulent fluctu- varies slowly enough that during a time step it can almost be
ations cause the suspended particles to collide and overcome considered steady flow; in other words, acceleration terms can
the repulsive forces among particles due to like charges of be neglected. Detailed wind wave modelling requires not only
ions surrounding the particles. the acceleration terms, but also much smaller time-step and
The turbulence also restricts the growth of the flocs by grid spacing than are practical for large estuaries.
limiting the largest floc that can form under a certain turbu- Wind waves are nevertheless important in stirring up and
lence level. The breakage of the larger and fragile flocs is suspending cohesive sediments and contaminants. It is there-
due to the high intensity of particle collisions because of tur- fore common practice to augment the computed bed shears
bulence. The governing parameter for these two mechanisms for wave activity averaged over a wave period, before pro-
is the shear rate, G, of the turbulent flow, which is related to ceeding to the erosion and deposition calculations.
the turbulent energy dissipation rate, ε, and the kinematic The relevant equations are as follows: (i) bed shear stress
viscosity, ν, as follows: due to waves (τ w)
[13] G = (ε / ν)1/2
[15] τ w = ( fw / 4)ρu2b
A representation of turbulence that will facilitate the compu-
tation of the shear rate, G is desirable in a hydrodynamic (ii) maximum orbital velocity at the bed (ub)
flow model suitable for handling cohesive sediment trans-
port. The logical choice that will meet this requirement is [16] ub = πH/[T sinh(2πh/L)]
the k–ε turbulence model, where k is the turbulent kinetic en-
ergy. The drawback of the k–ε turbulence closure, however, where H is the wave height, trough to crest; and T is the
is the extra computation efforts needed to solve the addi- wave period; (iii) wave length in shallow water (L)
tional two equations for k and ε.
[17] L = T(gh)1/2
2.3.1. Eddy diffusivity
The eddy diffusivity needed to solve the mass balance (iv) wave friction factor (fw)
equation (eq. [1]) can be calculated by relating it to the eddy
viscosity and a turbulent Schmidt number. The eddy viscos- [18] fw = exp[–5.98 + 5.21(do/2r)–0.19]
ity, in turn, can be calculated from the turbulence closure re-
and (v) maximum orbital excursion at the bed (do)
lationships. For example, in the case of k–ε turbulence
closure, the eddy viscosity, νt , is related to k and ε as fol- [19] do = π H/[2 sinh(2π h/L)]
lows:
[14] νt = cµ k 2 / ε A surprising number of numerical tidal models remain
weak in wetting and drying of the boundaries during a tidal
where cµ is an empirical constant. cycle. During the ebb, the model should remove boundary
elements from the calculation as they emerge above the fall-
2.4. Advection, diffusion, and dispersion of solutes: ing water level. On the flood, elements should be returned to
salinity and pollution the calculation as they are covered. Many models retain all
The concentration distribution of neutrally buoyant dis- the boundary elements in all the calculations but reduce the
solved substances can be calculated using hydrodynamic water depth of “dry” elements to a few millimetres. This
models without additional computational efforts, because of fudge can seriously affect the model’s conservation of water
the similarity in the form of the equations describing the and sediment. It becomes very important on models of cohe-
mass and the momentum transport (advection–diffusion sive sediment transport, such as in the Bay of Fundy where
equation). Care should be taken, however, to minimize the mudflats above low water occupy a significant fraction of
numerical dispersion errors that are inherent in many numer- the modelled area.

© 2004 NRC Canada


Willis and Krishnappan 753

2.6. Calibration and verification of models drag and buoyancy just balance the weight. This property of
All hydraulic models, physical, numerical, and especially the sand grain is known as its “fall velocity” or “settling ve-
sedimentary, need to demonstrate that they can reproduce locity.” If the grain were spherical, its settling velocity
the past and present before they are used for projections into would be given by Stokes’ law:
the future. This demonstration is divided into at least two
phases, namely calibration and verification. [20] w s = [(ρ s − ρ w)gD]/18ρ wv

2.6.1. Calibration where ws is the turbulent settling velocity in still water, D is


With calibration, adjustment of model parameters (rough- the equivalent spherical diameter of the grain, ρ w is the mass
ness and turbulence) is allowed until the model adequately density of the water, and ρ s is the mass density of the sedi-
reproduces a comprehensive set of field measurements. Sta- ment. Buoyancy of the grain is taken care of in the following
tistical techniques as in design of experiments (Willis and term:
Crookshank 1997) can be employed for this process.
[21] ∆ = (ρ s − ρ w)/ ρ w
2.6.2. Verification
The calibrated model must then be verified against an in-
where ∆ is the submerged specific gravity of sediment.
dependent comprehensive set of field measurements. If it
does not adequately reproduce the verification dataset with- Equation [20] is more used in reverse to compute the
out further parameter adjustment, we must return to the cali- equivalent spherical diameter of an irregular grain or floc,
bration phase until the model reproduces both datasets. based on its settling velocity measured in a settling column.
But the numerical modeller can more easily work with just
the measured settling velocity, ws. Settling velocity contains
3. Sediment in suspension all the important information about the suspended sediment
(grain size, shape, and density) in a single number. Equation
It is traditional to speak of at least three modes of sedi- [20] is then used by the model to compute settling velocities
ment transport by water as follows: (i) bedload, in which the for different water temperatures and densities and by input
moving sediment is supported at all times by the bed be- and output routines to produce grain sizes for the customers.
neath it, rolling over a layer of stationary sediment; (ii) sal-
tation (a fancy word for “jumping”), in which a grain travels
in a series of jumps off the bed into the flow and back onto 3.2. Cohesive sediment: silt and clay
the bed where its impact dislodges another jumping grain; Most estuarine muds are far from cohesionless. They are
and (iii) suspension, in which grains are supported in the cohesive, or sticky. Sediment particles glom together on the
flow by the upward component of turbulence and travel with bed and in suspension. The cohesive forces between parti-
the velocity of the surrounding water. cles exceed the forces of weight and gravity.
None of these three modes fully describes sediment trans- The divide between cohesive and cohesionless sediment,
port over a rippled bed. It is partly bedload while the grains between silt and sand, is usually defined as 60 µm
are being dragged up the back of a ripple, from which they (0.06 mm). Nevertheless, sand as large as 120 µm can still
are launched in a turbulent eddy (neither saltation nor sus- exhibit cohesion in salt water. Cohesion dominates in mix-
pension, but a finite whorl with a more or less horizontal tures of 75% sand and 25% silt and clay (Torfs 1997).
axis) until the eddy settles back onto the next ripple. It is far
easier to lump all sediment transport into the single process 3.2.1. Cohesion: electrical and biological processes
called suspension. Even those favouring differentiation re- The electrical attraction between small sediment particles
gard bedload as the lower limit of suspended load. In any can be demonstrated in a laboratory settling column. When
case, the transport of cohesive sediment and contaminants is the column is filled with clean distilled water, the particles
always in suspension, both traditional and comprehensive settle slowly, individually. Introduce an electrolyte, like sea-
definitions. water, and the particles glom together into larger, less dense
aggregations called flocs (see Sect. 3.2.2).
3.1. Cohesionless sediment: a grain of sand Electrical forces aren’t the only forces of cohesion be-
How a grain of sand behaves in flowing water is governed tween cohesive sediment particles. Diatoms living in most
entirely by its physical properties: size, shape, and mineral estuarine muds excrete a mucous that binds the sediment, of-
density in water. A free-body diagram can be made of a ten loosely, into “fluid mud” (Sect. 4) like the diatomaceous
grain of sand (Fig. 1) in which bed shear, buoyancy, and lift earth, bentonite, used in construction and drilling. Paterson
and drag by flowing water are the disturbing forces, and the (1997) has detailed the biological components of cohesion.
weight of the grain is the only restoring force. While the Amos et al. (1988) have shown the importance of drying
grain is at rest on the bed, or moving as bedload, there are during low water in strengthening mudflats. Le Hir et al.
also reactive forces from the bed beneath it. (2000) have shown the opposite effect: European mudflats
that lose strength when exposed to air.
3.1.1. Settling velocity From the modeller’s point of view, the only fact about co-
A sand grain in suspension is supported by drag forces hesion is that it must be measured in some detail, using nat-
balancing its weight, due to the upward vertical component ural water, sediment, and organisms (Krishnappan 1993).
of turbulence. In still water, with no turbulence, the same The model is only able to interpolate and extrapolate mea-
sand grain will settle at a terminal velocity for which the surements, not to predict cohesion.

© 2004 NRC Canada


754 Can. J. Civ. Eng. Vol. 31, 2004

Fig. 1. Free body diagram of (a) a cohesionless sand grain on the bed and (b) a cohesive floc in suspension.

3.2.2. Variable settling velocity: flocculation spectively; D is the floc diameter in µm; and b and c are em-
When electrical cohesion is in full effect, a cohesive sedi- pirical constants determined as b = 0.0015 and c = 1.7.
ment settling model can be formulated as follows: (i) silt, Using eq. [22], an expression for the settling velocity of
clay, and fine sand are transported to the estuary in dispersed Fraser River sediment flocs was derived by Krishnappan
suspension in the fresh water of the river; (ii) on meeting the (2000) as follows:
more conductive sea water in the estuary, the fines agglom-
erate into larger flocs, in a “turbidity maximum;” and [23] ws = (1.65/18) exp(–0.0015D1.7)gD2/ν
(iii) the flocs then settle onto the bed and the tidal flats.
Equations [22] and [23] are plotted in Fig. 2. The fall ve-
This is still not bad, in the geological sense of “model.”
locity increases with an increase in floc size until the size
The turbidity maximum is an estuarial fact that our numeri-
reaches a value of 55 µm. With a further increase in floc
cal model must reproduce during calibration and verification
size, the fall velocity actually decreases and reaches asymp-
(Sect. 2.6). We must nevertheless be wary of the implied as-
totically to zero when the size of the floc is about 175 µm
sumption that because flocs are bigger than their constituent
and the density of the floc is close to that of water. The max-
particles, they will settle quicker, at a faster settling velocity.
imum fall velocity of about 0.7 mm/s was calculated for the
We could predict the settling velocity of a floc, using floc size of about 55 µm.
something like eq. [20], if we knew its size, shape, and den-
sity; like snowflakes, however, every floc is different. With 3.2.3. Resuspension and erosion
snowflakes and cohesionless grains of sand, we can at least As the floc settles in the estuary it catches up with and
assume a relatively constant, uniform density; the density of gloms onto other particles and flocs, increasing in size and
a floc, however, depends on how much and what density of decreasing in density. On the way to the bed it will have to
water it contains. There may be so much fresh water trapped pass through a zone of high shear stress near the bed and
in the floc that its density will approach that of the salt wa- perhaps another zone where fresh water and salt water mix.
ter. As the name implies, a floc resembles a flake, with two It is probable that the shear will be strong enough to break
approximately equal dimensions and a third much smaller the weak bonds between particles in the floc and to resus-
dimension. Flocs and flakes tend to settle with the shortest pend the smaller particles. The sediment thus exhibits a
axis vertical, maximizing the horizontal area and vertical threshold shear for deposition, τs. When the bed shear, τ, ex-
drag as they settle. A floc can settle very slowly, even in still ceeds τs, the floc cannot reach the bed but is broken up and
water. resuspended.
Settling velocity is therefore the most important hydraulic Even when the floc settles on the bed, it will probably be
property of cohesive sediment to be modelled. As with sand resuspended before it can consolidate, the next time bed
(Sect. 3.1.1), the settling velocity summarizes size, shape, shear exceeds the sediment’s threshold shear for erosion, τc .
and density of the flocs and is much easier to measure than The erosion threshold is a function of the bulk density of the
any of them. The complexity of the cohesive bond between surface of the bed according to the Migniot relation:
particles (Sect. 3.2.1), however, means that the settling ve-
locity of the flocs must be measured in natural water with [24] τc = mρ on
living bugs, usually in the field.
In a recent study, Lau and Krishnappan (1994a) developed where ρ o is the bulk submerged density of the fresh deposit,
a relationship between the density of the flocs and its size and m and n are empirical constants.
for the sediment particles from the Fraser River in Canada. The rate of erosion is given by Ariathurai and Arulan-
The form of the relationship is as follows: andan (1978):

[22] ρ f − ρ w = ρ s exp(−bDc ) [25] dM/dt = M2(τc − τ)/ τc for τ > τc

where ρ f , ρ w, and ρ s are densities of the sediment floc, wa- where M is the mass of sediment on the bed, and M2 is the
ter, and parent material (the last one measured in water), re- rate of erosion at τ = 2τc .

© 2004 NRC Canada


Willis and Krishnappan 755

Fig. 2. Floc density and fall velocity as a function of floc size.

Once again, the model depends on measurement of prop- ment per litre is often taken as the lower limit) is so great
erties, usually in full-scale annular flume tests (Krishnappan that pore water is trapped, unable to drain upwards as the
1993; Krishnappan and Engel 1997): τs, ρb , m, n, and M2. sediment settles or downwards through the static bed. Pore
pressures build up to support the weight of the sediment.
3.2.4. Deposition Bentonite, diatomaceous earth used as drilling mud, is a
If the floc makes it through the bottom boundary layer (if good example of a fluid mud. In fluid mud it is not just the
bed shear, τ, is less than the deposition threshold, τs, for the mucous of the diatoms lubricating the sediment, but also
sediment), then it will be deposited on the bed at a rate their shells interlocking to block the passage of water mole-
(Krone 1962) cules and prevent further consolidation. Fluid mud forms in
[26] dM/dt = Cws(τs − τ)/ τs for τ < τs low points (pits and troughs) in the bed at still water and
then flows under the influence of bed shear and slope on the
where C is the concentration of sediment in suspension, and subsequent floods and ebbs.
ws is the settling velocity of the floc. Fluid mud is a non-Newtonian fluid, which only tells you
The modeller requires measurements not only of τs but what it is not. It does not behave like water, in any number
also of the bulk submerged density of the fresh deposit, ρ o, of interesting ways. The most common way is the existence
which should be approximately the same as the density of of thresholds, slope or shear, below which the fluid mud re-
the flocs. Once again, these measurements need to be made mains stably in place. When a layer of fluid mud flows, it
in the laboratory or in the field, using the natural water, sedi- will probably be at uniform shear stress and laminar by defi-
ment, and organisms. nition. The act of flowing allows drainage, so the layer can
consolidate (settle) or dilate (resuspend) as it flows.
3.2.5. Adsorbed contaminants Modelling fluid mud requires a coupled 3-D (Le Hir
Because of their high specific surface areas, the fine sedi- 1997) or layered 2-D hydraulic model. Once again, the mod-
ments readily adsorb contaminants that are present in the eller requires good measurements of properties of the fluid
water column. For some of the hydrophobic contaminants mud, especially the conditions leading to its formation and
such as polycyclic aromatic hydrocarbons (PAHs) and pesti- destruction. The properties to be measured depend on the
cides, the concentration of the contaminant in sediment far rheology of the fluid mud to be modelled and may include
exceeds that in the water column. The transport of sediment- shear thresholds, viscosity, density, drainage, dilation, and
associated contaminants can be modelled using a fine- consolidation.
sediment transport model once the partition coefficient and
the rate of adsorption for the contaminant are established.

4. Fluid mud 5. Riverbed processes


A layer of fluid mud results from “hindered settling,” i.e., When a layer of cohesive sediment is deposited on the
when the concentration of flocs near the bed (25 g of sedi- bed, without resuspension or formation of fluid mud, the
© 2004 NRC Canada
756 Can. J. Civ. Eng. Vol. 31, 2004

model raises the bed elevation (decreases the water depth) Fig. 3. Bed sediment layers.
by the initial thickness of the layer.

5.1. Layers and consolidation


Most of the complexity of numerical cohesive bed sedi-
ment models comes from the bookkeeping of layers of sedi-
ment (Fig. 3) below the bed surface. A typical numerical
hydraulic model has 10 000 calculation points on a 100 ×
100 grid. A new layer of deposited sediment, a few milli-
metres thick, can be generated in a sedimentary model time
step, typically about a half hour. Twenty-five layers can then
be deposited in a semidiurnal tidal period.
The layers do not necessarily correspond to real physical
layers of sediment: most often, the thickness of a numerical
layer is determined by the length of the model time step over
which it was deposited. Some reduction in bookkeeping ef-
fort can usually be achieved by merging similar adjacent lay-
ers. Deposition will only take place for about an hour at
each still water for about four numerical layers per tidal cy-
cle, some of which may be eroded during the intervening
ebb and flood. Nevertheless, 25 is not an excessive number 5.3. Mudflats: wetting and drying
of layers to keep track of. What happens to the erodibility, τc , of a cohesive sedi-
For each layer at each grid point, we need to store infor- ment riverbed when it is exposed to the air on tidal flats at
mation on the composition, time of deposition, initial thick- low water? Comparisons to pottery clays suggest that it will
ness, and initial density, giving already a million pieces of strengthen if allowed to bake in the sun for a few hours
information to be stored and sorted. As a fresh layer is de- (Amos et al. 1988). Fresh rainfall may perhaps weaken it;
posited on the top of the stack of 25, the lowest layer must mudflat erosion has even been observed during heavy rain-
join the fully consolidated sub-bed. When a top layer is fall in the Annapolis Basin.
eroded, the 24 lower layers move up the stack, all at 10 000
grid points. 5.4. Mixtures of sand and mud
The consolidation must then be computed as water drains In Cumberland Basin, Willis and Crookshank (1997)
from the sediment. The Terzaghi relationship from founda- found cohesionless sand deposited in places on what were
tion engineering is as follows: otherwise mudflats. The model of Willis and Crookshank
then needed to deal with both cohesive (mud) and
[27] ρ t +1 = 0.964 ρt (∆tC T / P 2) 0.415 cohesionless (sand) phases. This was accomplished by dupli-
cating the Ariathurai and Arulanandan equation (eq. [25])
and Krone equation (eq. [26]), with calibration coefficients
where ρt is the bulk submerged density at time t, ∆t is the
for sand and τs ≈ τc , a physical constant based on grain size,
time step (t + 1) – t, CT is the Terzaghi consolidation coeffi-
say from the Shields’ curve. The percentage of sand then be-
cient, and P is the length of drainage path and is approxi-
comes another parameter to be stored for each bed layer
mately equivalent to the depth below the bed surface.
(Sect. 5.1).
This is another example of an empirical framework for When does a mixture of sand and mud begin to behave
interpolating and extrapolating measured rates of consolida- cohesively, that is, like a “pure” mud? We included this criti-
tion, through the laboratory determination of the consolida- cal percentage of sand as a calibration parameter in the
tion coefficient, CT. Cumberland Basin model and found a value of about 80%
Consolidation of all layers, and the resulting lowering of sand by weight or mass. In a series of laboratory experi-
the bed elevation, is updated every model time step at each ments, Torfs (1997) came to the same conclusion, with the
grid point. The model needs to know when to stop the con- result that 80% sand became fixed in subsequent models.
solidation process in a layer: when its bulk density reaches Twenty percent mud is all that’s required for a mixture to
the maximum, or full consolidation, ρ c , another property of behave cohesively.
the sediment to be measured, most likely in the soils labora- A modelling approach was proposed for mud and sand
tory. mixtures by Chesher and Ockenden (1997) based on labora-
tory and field studies carried out by Williamson and
5.2. Erosion Ockenden (1992) and Williamson (1993a, 1993b) for mud–
The erosion threshold of the layer exposed on the surface sand mixtures. These studies have shown that the sand con-
of the bed is calculated by the Migniot relation (eq. [24]), tent increased the mud consolidation rate and the shear
and the rate of erosion is given by the Ariathurai and strength of the bed and decreased the erosion rate of the
Arulanandan equation (eq. [25]). The model needs to check mud, perhaps due to the armouring effect of the sand frac-
for complete erosion of the surface layer and switch to the tion (Torfs 1997). Once suspended in the water column, mud
next layer to be exposed. It will also lower the bed elevation and sand appear to act independently with independent de-
and increase the water depth by the amount of erosion. position rates. In an earlier study, Kamphuis (1981) had

© 2004 NRC Canada


Willis and Krishnappan 757

shown that the presence of sand particles increased the ero- Krishnappan, B.G., and Engel, P. 1997. Critical shear stresses for
sion rate of a preconsolidated clay bed by the abrasive ac- erosion and deposition of fine suspended sediments in the Fraser
tion. More work is needed to quantify the erosion rate of the River. In Cohesive Sediments: Proceedings of the 4th Nearshore
sand–mud mixtures. and Estuarine Cohesive Sediment Transport Conference,
INTERCOH’94, Wallingford, UK, 11–15 July 1994. Edited by
N. Burt, R. Parker, and J. Watts. John Wiley & Sons, Chichester,
6. Summary UK. Paper 21, pp. 279–288.
Krishnappan, B.G., and Marsalek, J. 2000. Flocculation and trans-
Numerical models of cohesive sediment and contaminant port of cohesive sediments collected from a stormwater deten-
transport are extensions of numerical hydraulic models. The tion pond. In Building Partnerships — 2000 Joint Conference on
hydraulics is based on much firmer physical principles than Water Resource Engineering and Water Resources Planning and
the sediments. In fact, sediment models are no more than nu- Management, Minneapolis, Minn., 30 July – 2 August 2000.
merical frameworks for interpolation and extrapolation of Edited by R.H. Hotchkiss and M. Glade. [CD-ROM]. American
measurements. Society of Civil Engineers (ASCE), Reston, Va.
The numerical hydraulic model will compute the transport Krishnappan, B.G., Madsen, N., Stephens, R., and Ongley, E.D.
of sediment in suspension, including adsorbed contaminants, 1992. A field instrument for measuring size distribution of sus-
and of pollutants and salinity in solution. It therefore must pended sediments in rivers. In Proceedings of the 8th IAHR
accurately model turbulent diffusion and dispersion, includ- Congress (Asia and Pacific Regional Division), Pune, India, 20–
ing the downward biasing of sediment settling velocity, and 23 October 1992. International Association of Hydraulic Engi-
will probably need to model the third, vertical, dimension. neering and Research (IAHR). pp. F71–F81.
Like all hydraulic models, numerical models of cohesive Krone, R.B. 1962. Flume studies of the transport of sediment in es-
sediment and contaminant transport require calibration and tuarial shoaling processes. Hydraulic Engineering Laboratory
verification against at least two independent sets of measure- and Sanitary Engineering Research Laboratory, University of
ments. Extra emphasis must be given to calibrating and veri- California, Davis, Calif.
Lau, Y.L., and Krishnappan, B.G. 1994a. Comparison of particle
fying the sediment component: sediment transport models
size measurements made with a water elutriation apparatus and
are almost completely empirical.
a Malvern particle size analyzer. NWRI Contribution 94-82, Na-
All measurements must be done at full scale, either in the tional Water Research Institute (NWRI), Burlington, Ont.
field or in the laboratory using sediment, water, and living Lau, Y.L., and Krishnappan, B.G. 1994b. Does reentrainment occur
organisms from the site to be modelled. during cohesive sediment settling? ASCE Journal of Hydraulic
Engineering, 120(2): 236–244.
References Le Hir, P. 1997. Fluid and sediment “integrated” modelling appli-
cation to fluid mud flows in estuaries. In Cohesive Sediments:
Amos, C.L., Van Wagoner, N.A., and Daborn, G.R. 1988. The in- Proceedings of the 4th Nearshore and Estuarine Cohesive Sedi-
fluence of subaerial exposure on the bulk properties of fine- ment Transport Conference, INTERCOH’94, Wallingford, UK,
grained intertidal sediment from Minas Basin, Bay of Fundy. 11–15 July 1994. Edited by N. Burt, R. Parker, and J. Watts.
Estuarine, Coastal and Shelf Science, 27: 1–13. John Wiley & Sons, Chichester, UK. Paper 30, pp. 417–428.
Ariathurai, R., and Arulanandan, K. 1978. Erosion rate of cohesive Le Hir, P., Roberts, W., Cazaillet, O., Christie, M., Bassoullet, P.,
soils. ASCE Journal of the Hydraulics Division, 104(HY2): and Bacher, C. 2000. Characterization of intertidal flat hydrody-
279–283. namics. Continental Shelf Research, 20: 1433–1459.
Chesher, T.J., and Ockenden, M.C. 1997. Numerical modelling of Le Normant, C. 2000. Three dimensional modelling of cohesive
mud and sand mixtures. In Cohesive Sediments: Proceedings of sediment transport in Loire Estuary. Hydrological Processes,
the 4th Nearshore and Estuarine Cohesive Sediment Transport 14(13): 2231–2243.
Conference, INTERCOH’94, Wallingford, UK, 11–15 July Paterson, D.M. 1997. Biological remediation of sediment erodibility:
1994. Edited by N. Burt, R. Parker, and J. Watts. John Wiley & ecology and physical dynamics. In Cohesive Sediments: Proceed-
Sons, Chichester, UK. Paper 28, pp. 395–402. ings of the 4th Nearshore and Estuarine Cohesive Sediment
Hervouet, J.M. 2000. TELEMAC modelling system: an overview. Transport Conference, INTERCOH’94, Wallingford, UK, 11–15
Hydrological Processes, 14: 2209–2210. July 1994. Edited by N. Burt, R. Parker, and J. Watts. John Wiley
Kamphuis, J.W. 1981. Erosion of cohesive sediment by water car- & Sons, Chichester, UK. Paper 14, pp. 215–229.
rying suspended sand. Report ES 4, Queens University, Teisson, C. 1997. A review of cohesive sediment transport models.
Kingston, Ont. In Cohesive Sediments: Proceedings of the 4th Nearshore and
Krishnappan, B.G. 1990. Modelling of settling and flocculation of Estuarine Cohesive Sediment Transport Conference,
fine sediments in still water. Canadian Journal of Civil Engi- INTERCOH’94, Wallingford, UK, 11–15 July 1994. Edited by
neering, 17: 763–770. N. Burt, R. Parker, and J. Watts. John Wiley & Sons, Chichester,
Krishnappan, B.G. 1991. Modelling of cohesive sediment trans- UK. Paper 36, pp. 367–382.
port. In Proceedings of the International Symposium on the Teisson, C., Simonin, O., Galland, J.C., Lawrence, D., and Fritsch,
Transport of Suspended Sediments and its Mathematical D. 1991. Numerical modelling of cohesive sediment transport:
Modelling, Florence, Italy, 2–5 Sept. 1991. Officine Grafiche past experience and new research axes. In Proceedings of the In-
Tecnoprint, Bologna, Italy. pp. 433–448. ternational Symposium on the Transport of Suspended Sedi-
Krishnappan, B.G. 1993. Rotating circular flume. ASCE Journal of ments and its Mathematical Modelling, Florence, Italy, 2–5
Hydraulic Engineering, 119(6): 658–767. September 1991. Officine Grafiche Tecnoprint, Bologna, Italy.
Krishnappan, B.G. 2000. In situ size distribution of suspended par- pp. 475–493.
ticles in the Fraser River. ASCE Journal of Hydraulic Engi- Torfs, H. 1997. Erosion of mixed cohesive/non-cohesive sediments
neering, 126(8): 561–569. in uniform flow. In Cohesive Sediments: Proceedings of the 4th

© 2004 NRC Canada


758 Can. J. Civ. Eng. Vol. 31, 2004

Nearshore and Estuarine Cohesive Sediment Transport Confer- m empirical constant in Migniot relation (eq. [22])
ence, INTERCOH’94, Wallingford, UK, 11–15 July 1994. M mass of sediment on bed
Edited by N. Burt, R. Parker, and J. Watts. John Wiley & Sons, M2 rate of erosion at τ = 2τ c
Chichester, UK. Paper 16, pp. 245–252. nManning roughness (eq. [12])
van Leussen, W. 1988. Aggregation of particles, settling velocity of nempirical constant in Migniot relation (eq. [22])
mud flocs. In Physical processes in estuaries. Edited by J. P length of drainage path
Dronkers and W. van Leussen. Springer-Verlag, New York. qd sediment flux due to deposition
Williamson, H.J. 1993a. Tidal transport of mud/sand mixtures. qesediment flux due to erosion
Field trials at Blue Anchor Bay. HR Wallingford Report SR333, rphysical dimension of bed roughness (eq. [9])
HR Wallingford Ltd., Wallingford, Oxon, UK. R ripple height, trough to crest
Williamson, H.J. 1993b. Tidal transport of mud/sand mixtures.
S(x, y, z)sediment source or sink
Field trials at Clevedon, Severn Estuary. HR Wallingford Report
t
time
SR372, HR Wallingford Ltd., Wallingford, Oxon, UK.
Twave period
Williamson, H.J., and Ockenden, M.C. 1992. Tidal transport of
mud and sand mixtures: Laboratory tests. HR Wallingford Re- u, ν horizontal velocity components
port SR257, HR Wallingford Ltd., Wallingford, Oxon, UK. ub maximum orbital velocity at bed (eq. [16])
Willis, D.H., and Crookshank, N.L. 1997. Modelling multiphase uiinstantaneous flow velocity
sediment transport in estuaries. In Cohesive Sediments: Pro- ui′fluctuating velocity component
ceedings of the 4th Nearshore and Estuarine Cohesive Sediment uitemporal mean flow velocity
Transport Conference, INTERCOH’94, Wallingford, UK, 11–15 u* shear velocity, “u-star”
July 1994. Edited by N. Burt, R. Parker, and J. Watts. John vspatial mean velocity, averaged over depth
Wiley & Sons, Chichester, UK. Paper 27, pp. 383–394. w vertical velocity component
w′fluctuating velocity component in the vertical direction
ws settling velocity of sediment or floc
List of symbols
x, y horizontal spatial coordinates
b empirical constant (eq. [22]) zvertical spatial coordinate
c empirical constant (eq. [22]) εturbulent energy dissipation
c mean concentration of sediment in suspension λripple wavelength, crest to crest
c′ fluctuating sediment concentration νkinematic viscosity of water
cµ empirical constant (eq. [2]) νteddy viscosity due to turbulence
C concentration of sediment in suspension ρw density of water
CCH dimensionless Chézy smoothness coefficient (eq. [8]) ρb bulk submerged density of bed surface layer
CT Terzaghi consolidation coefficient (eq. [25]) ρf density of sediment floc
do maximum wave orbital excursion at bed (eq. [19]) ρo bulk submerged density of fresh deposit
D equivalent spherical diameter of grain or floc ρs mass density of sediment
D90 the (coarse) grain size of which 90% of the bed is smaller ρs submerged density of parent material (eq. [22])
(eq. [10]) ρtbulk submerged density at time t
f Weisbach friction factor (eq. [11]) τbed shear
fw wave friction factor (eq. [18]) τccritical (threshold) bed shear for erosion
g acceleration due to gravity τscritical (threshold) bed shear for deposition
G shear rate (eq. [13]) τw bed shear due to waves, averaged over wave period
h water depth (eq. [15])
H wave height, trough to crest ∆ submerged specific gravity of sediment
I water surface or energy slope ∆t time step
k turbulent kinetic energy Γ diffusion coefficient
L wavelength, crest to crest

© 2004 NRC Canada

View publication stats

Вам также может понравиться