This action might not be possible to undo. Are you sure you want to continue?

**numerical approach by high order ﬁnite volume methods
**

q

M.J. Castro Dı ´az

a

, E.D. Ferna´ndez-Nieto

b,

*

, A.M. Ferreiro

c

a

Dpto. Ana´ lisis Matema´ tico, Universidad de Ma´ laga, Campus Teatinos s/n, Ma´ laga, Spain

b

Departamento de Matema´ tica Aplicada I, Universidad de Sevilla, 41012 Sevilla, Spain

c

Dpto. Ana´ lisis Ecuaciones Diferenciales y Ana´ lisis Nume´ rico, Universidad de Sevilla, C/ Tarﬁa S/N, Sevilla, Spain

Received 5 May 2006; received in revised form 30 October 2006; accepted 10 July 2007

Available online 29 September 2007

Abstract

This paper is concerned with the numerical approximation of bedload sediment transport due to water evolution. For the hydrody-

namical component we consider Shallow Water equations. The morphodynamical component is deﬁned by a continuity equation, which

is deﬁned in function of the solid transport discharge. We present several deterministic models, such as Meyer-Peter & Mu¨ ller, Van Rijn

or Grass model. We also present an uniﬁed deﬁnition for the solid transport discharge, and we compare with Grass model. Both com-

ponents deﬁne a coupled system of equations that can be rewrite as a non-conservative hyperbolic system. To discretize it, we consider

ﬁnite volume methods with or without ﬂux limiters and high order state reconstructions. Finally we present several tests, where we

observe numerically the order of the numerical schemes. Comparisons with analytical solutions and experimental data are also presented.

Ó 2007 Elsevier Ltd. All rights reserved.

1. Introduction

The sediment can be deﬁned as a fragmented material

from rocks that has been formed by diﬀerent physical

and/or chemical process. The study of sediment transport

processes includes movement of rocks in a mountain as

material diﬀusion in water, among other processes. Trans-

port is caused by gravity eﬀects and by friction eﬀects with

the air or the ﬂuid containing the sediment.

Sediment transport is usually divided into three types:

bedload, saltation and suspension (see Fig. 1). Bedload

transport is deﬁned as the type of transport where sediment

grains roll or slide along the bed. Saltation transport is

deﬁned as the type of transport where single grains jump

over the bed a length proportional to their diameter, losing

for instants the contact with the soil. Sediment is suspended

when the ﬂux is intense enough such as the sediment grains

reach height over the bed.

In this paper we face the study of bedload sediment

transport. To model bedload sediment transport process

caused by the movement of a ﬂuid in contact with the sed-

iment layer, we consider a coupled model constituted by a

hydrodynamical component and a morphodynamical

component.

The hydrodynamical component is modeled by Shallow

Water equations, which is are used to study ﬂuid move-

ment in rivers, channel, coast areas, etc., while a sediment

transport equation, depending on solid transport ﬂux, is

considered to model the morphodynamical component.

In literature diﬀerent equations to model the solid trans-

port sediment ﬂux could be found: (Grass equation [25],

Meyer-Peter & Mu¨ ller’s equation [38], Van Rijn’s equation

[54–56], Nielsen’s equation [39], Kalinske [32,33], Einstein’s

equation [18,19,31,60], etc., generally obtained by empirical

methods. Among all these formulae, some are deterministic

formulae and others are based on probabilistic terms. In

this paper we only consider deterministic equations (see

Section 2.3).

0045-7930/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved.

doi:10.1016/j.compﬂuid.2007.07.017

q

This research was partially supported by Spanish Government

Research Projects BFM2003-07530-C02-01 and BFM2003-07530-C02-02.

*

Corresponding author.

E-mail addresses: castro@anamat.cie.uma.es (M.J. Castro Dı ´az),

edofer@us.es (E.D. Ferna´ndez-Nieto), anafefe@us.es (A.M. Ferreiro).

www.elsevier.com/locate/compﬂuid

Available online at www.sciencedirect.com

Computers & Fluids 37 (2008) 299–316

In most of them, except for Grass model, the movement

of the sediment is controlled by a physical parameter called

critical shear stress, usually determined experimentally.

Therefore, hydrodynamical and morphodynamical com-

ponents constitute a coupled system, with three unknowns:

h(x, t), the thickness of the ﬂuid, q(x, t), the discharge of the

ﬂuid and z

b

(x, t), the thickness of the sediment layer. This

coupled system can be written as a non-conservative hyper-

bolic system

oW

ot

þAðW Þ

oW

ox

¼ 0: ð1Þ

It is well known that standard methods that solve cor-

rectly systems of conservation laws can fail in solving sys-

tems of balance laws, specially when approaching

equilibria or near to equilibria solutions. Moreover, they

can produce unstable methods when they are applied to

coupled systems of conservation or balance laws. Many

authors have studied well-balanced numerical schemes for

balance laws or coupled systems. See [2–4,6,9,12,13,22,23,

26,27,37,40,43,44,47,52,57,58,61], etc.

One of these techniques consists on discretizing the con-

servative terms of the system by using a well-known solver

for homogeneous conservative systems and studying a dis-

cretization of the source term and/or the non-conservative

terms. On related papers authors study if with the proposed

discretization the numerical scheme exactly or approxi-

mately preserves a given stationary solution or a family

of equilibria. The form in which the discretization of the

source term and/or the coupled term is discretized depends,

in general, on the speciﬁc considered system, and, usually,

the extension to high order numerical schemes becomes a

diﬃcult task. In [12] it was shown that the technique of

Modiﬁed Equations can be helpful in this procedure.

For example, this technique has been used to extend

Roe method, Lax–Wendrof with ﬂux limiters or HLL

scheme among other classical schemes. Roe method was

studied in [4,9], a family of numerical schemes, including

numerical schemes using ﬂux limiters, was studied in [13].

In [34] Kurganov and Levy extend HLL scheme with state

reconstruction for Shallow Water equation with topogra-

phy as source term, using the water surface and the

discharge as unknowns. This method is known as central-

upwind. In [20] Fraccarollo et al. extend HLL method

for a system coupling Shallow Water equation with a sed-

iment layer in order to study intense sediment transport.

They introduce a lateralized discretization of the coupled

term, which consists on diﬀerent approximations to the left

and to the right. They also write the system in terms of the

water surface. They call this method LHLL.

Another technique consists on considering balance laws

or coupled systems as a particular case of the one-dimen-

sional quasilinear hyperbolic system (1). Once the system

is rewritten under this form, piecewise constant approxima-

tions of the solutions are considered, which are updated by

means of Approximate Riemann Solvers at the intercells.

The extension to high order numerical schemes is in this

case straightforward.

Following [41], this second procedure has been consid-

ered in this work. The main diﬃculty both from the math-

ematical and the numerical points of view comes from the

presence of non-conservative products, which makes diﬃ-

cult even the deﬁnition of weak solutions. Many papers

have been devoted to the deﬁnition and the stability of

non-conservative products and its application to the deﬁni-

tion of weak solutions of non-conservative hyperbolic sys-

tems: see [1,5,7,11,15,16,35,45,59], etc.

In this article we assume the deﬁnition of non-conserva-

tive products as Borel measures given by Dal Maso et al. in

[16]. This deﬁnition, which depends on the choice of a fam-

ily of paths in the phase space, allows to give a rigorous

deﬁnition of weak solutions of (1). Together with the deﬁ-

nition of weak solutions, a notion of entropy has to be cho-

sen, as the usual Lax’s concept or one related to an entropy

pair. The classical theory of simple waves of hyperbolic sys-

tems of conservation laws and the results concerning the

solutions of Riemann problems can be then extended to

systems (1).

The choice of the family of paths may be, in general, a

diﬃcult task. The introduction of such family of paths also

allows to give a general deﬁnition of Roe linearizations for

non-conservative systems. In [40] the authors propose a

general deﬁnition of well-balanced schemes to solve sys-

tems like (1).

Hudson in [29] proposed several numerical approxima-

tions of a system constituted by Shallow Water equations

and bedload sediment transport equation using Grass

model. He considers diﬀerent reformulations of the

problem.

Pen˜ a Gonza´lez in [42] proposed a numerical approxima-

tion of a non-coupled system formed by Shallow Water

equations and bedload sediment transport equation. He

applies a ﬁnite volumes scheme based on Roe scheme to

obtain the Shallow Water unknowns, thickness and dis-

charge of the ﬂuid. The approximated velocity and thick-

ness of the ﬂuid are used in bedload sediment transport

equation to get an approximation of thickness of the sedi-

ment layer. He also proposed diﬀerent techniques of mea-

surement of the thickness of the sediment layer. In

Section 6.3 a comparison with one of the experiments per-

formed in his work is shown.

In [20] a second order LHLL method for the computa-

tion of open channel ﬂows in conditions of rapid bed ero-

sion and intense sediment transport is presented. The

phenomena that govern these ﬂows are diﬀerent from those

Fig. 1. Types of sediment transport.

300 M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316

presented in bedload sediment transport, where the interac-

tion between the ﬂuid and the sediment is very weak.

One of the main diﬃculty to obtain a precise numerical

approximation for bedload sediment transport models is

that the interaction between ﬂuid and sediment is usually

very weak (see Section 2.5) and long time simulations are

needed. For this reason, as characteristic velocities of ﬂuid

and sediment are very diﬀerent, low order numerical

schemes are very diﬀusive (see Section 6.1). So, it is neces-

sary to use high order numerical schemes, at least bigger

than two to obtaining precise results. In this paper we

consider numerical schemes based on ﬂux limiters and

numerical schemes based on state reconstructions for

non-conservative hyperbolic systems (see [10]).

This paper is organized as follows: In Section 2 we pres-

ent the system of equations for diﬀerent morphodynamical

models, such as Meyer-Peter & Mu¨ ller, Van Rijn or Grass.

In Section 2.4 we present a uniﬁed formulation of the pre-

sented models, redeﬁning the solid transport discharge. In

Section 2.5 we perform a comparison with Grass model

when the critical shear stress is set to zero, studying the

type of interaction between the ﬂuid and the sediment. In

Section 2.6 we present the coupled system of equations.

We introduce a new variable that allows us to simplify

the system. We rewrite it as a non-conservative hyperbolic

system. In Section 3 we present ﬁnite volume methods for

non-conservative hyperbolic systems with and without ﬂux

limiters. In Section 4 we present high order ﬁnite-volume

methods using state reconstructions, concretely we have

used WENO reconstructions (see [28,30,49]). In Section 5

we present Roe matrix for the diﬀerent considered models,

in an approximated way, and how we treat the problem of

transition cells: transition with–without sediment and tran-

sition with–without movement. Finally in Section 6 we

present three numerical tests, we observe numerically the

order of the diﬀerent numerical schemes, we also compare

with an analytical solution for Grass model and ﬁnally we

compare with experimental data.

2. Sediment transport models in Shallow Water equations

In this section we present a system of equations that

models sediment transport in areas such us rivers, lakes

or coast. To solve the problem of sediment transport by

a ﬂuid, a coupled mathematical model can be used with a

hydrodynamical component and a morphodynamical

component.

• For the hydrodynamical component we consider Shal-

low Water equations, which simulates the movement

of a ﬂuid in rivers, channels, coastal areas, etc. (see

[52,9]),

• The morphodynamical component is described with a

transport equation, which is deﬁned by the solid trans-

port discharge. In Section 2.2 we present some experi-

mental formulae that can be considered.

Shallow Water system and the sediment transport equa-

tion form a coupled system that is described in Section 2.6.

2.1. Hydrodynamical model: Shallow Water equations

We consider a one-dimensional channel with variable

bottom and constant rectangular section. If the bottom is

ﬁxed, we have the equations

oh

ot

þ

oq

ox

¼ 0;

oq

2

ot

þ

o

ox

q

2

h

þ

1

2

gh

2

_ _

¼ gh

dH

dx

ÀghS

f

;

_

¸

¸

_

¸

¸

_

ð2Þ

where by x we denote the horizontal variable at the axis of

the channel and t is the time variable. By q(x, t) and h(x, t)

we denote the discharge and the height of the water column

respectively. g is the gravity constant and H(x) is the

bathymetry function respect to a ﬁxed reference level

(A

R

) (see Fig. 2a).

The term S

f

models the friction term, for Manning law

we have

S

f

¼

gg

2

juju

R

4=3

h

; ð3Þ

where g is the Manning’s coeﬃcient, R

h

is the hydraulic ra-

tio, which can be approximated by h and u is the velocity of

the ﬂuid ðuðx; tÞ ¼

qðx;tÞ

hðx;tÞ

Þ.

To study the problem of sediment transport it is neces-

sary to consider a sediment layer, z

b

and a ﬁxed bottom

(without sediment), z

f

= ÀH + A

R

(z

f

+ H = A

R

). In this

case the system of equation (2) can be written as follows

(see Fig. 2b):

oh

ot

þ

oq

ox

¼ 0;

oq

ot

þ

o

ox

q

2

h

þ

1

2

gh

2

_ _

¼ Àgh

oz

b

ox

þgh

dH

dx

ÀghS

f

:

_

¸

¸

_

¸

¸

_

ð4Þ

2.2. Morphodynamical model

The equation that describes sediment transport is a con-

tinuity equation and the expression of the conservation

sediment volume equation is given by

oz

b

ot

þn

oq

b

ox

¼ 0; ð5Þ

where n = 1/(1 À q

0

) and q

0

is the porosity of the sediment

layer. By q

b

= q

b

(h, q) we denote the solid transport

discharge.

2.3. Solid transport discharge

In this section we present the most popular formulae to

deﬁne the solid transport discharge, q

b

. The study of the

deﬁnition of the solid transport discharge can be seen as

a deterministic problem or a probabilistic one. For exam-

ple, deterministic methods have been proposed by Duboy

M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316 301

and Meyer-Peter [17], Meyer-Peter and Mu¨ ller [38]; and

probabilistic methods by Kalinske [32,33] and Einstein

[18,19]. Diﬀerent estimations of the solid transport dis-

charge q

b

have been obtained by empirical methods.

In what follows, we present some formulae for q

b

corre-

sponding to deterministic models. They were proposed for

granular and non-cohesive sediments. They were obtained

for stationary ﬂux in rivers. Nevertheless, it can also be

applied to ﬂuids in tides, by the eﬀects of the waves and the

currents, because the time response of the sediment is very

small in comparison with the period of the tide or the wave.

These formulae do not include any gradient pressure eﬀect to

model grain falling when the velocity of the ﬂuid is zero.

All models considered in this work, except Grass model,

have into account that the movement of the granular sedi-

ment begins when the shear stress (s) is bigger than a cer-

tain critical shear stress (s

c

).

Usually, shear stress can be written in terms of the

hydrodynamical unknowns, h and q, using the friction

term

s ¼ cR

h

jS

f

j; ð6Þ

where c denotes the speciﬁc weight of the ﬂuid c = gq,

being q the water density and S

f

is deﬁned by (3). Usually,

Manning’s coeﬃcient g over the sediment layer is not equal

to Manning’s coeﬃcient over the ﬁxed bottom.

Nevertheless, the shear stress is usually presented in a

non-dimensional form in q

b

. If we denote by s

Ã

and s

Ãc

the non-dimensional shear stress and the critical shear

stress, respectively, then

s

Ã

¼

s

ðc

s

ÀcÞd

i

; s

Ãc

¼

s

ðc

s

ÀcÞd

i

; ð7Þ

where c

s

denotes the speciﬁc sediment weight c

s

= gq

s

,

being q

s

the sediment density, and d denotes the sediment

grain size (diameter).

Using (6) and (7), s

Ã

can be written in function of the rel-

ative density of ﬂuids G = q

s

/q.

s

Ã

¼

gg

2

u

2

ðG À1Þd

i

R

1=3

h

:

To determine s

Ãc

many experiments have been performed in

diﬀerent works. Concretely, Shields proposed the well-

known Shields-diagram. He did the measurement for big-

ger diameters and after extrapolate.

Some of the most usual formulae for rivers are:

• Grass equation:

Grass [25] proposed the following formula for the solid

transport discharge:

q

b

¼ A

g

q

h

q

h

¸

¸

¸

¸

¸

¸

mgÀ1

; 1 6 m

g

6 4; ð8Þ

where the constant A

g

(s

2

/m) is usually obtained by

experimental data and takes into account the grain

diameter and the cinematic viscosity. The usual value

of the exponent m

g

is set to m

g

= 3, obtaining

q

b

¼ A

g

u

3

: ð9Þ

The coeﬃcient A

g

takes values between 0 and 1. When it

is closed to 0 the model reﬂects a weak interaction be-

tween the sediment and the ﬂuid.

Note that one of the main characteristics of this model is

that the critical shear stress is set to zero, so the sediment

movement begins at the same instant that beginning

ﬂuid motion.

• Meyer-Peter & Mu¨ ller’s equation:

Meyer-Peter & Mu¨ ller [38] (MP & M in what follows)

developed one of the most known formulae for the solid

transport discharge, based on median grain diameter d

50

(see [51]). Later, Chien [14] proved that original formula

can be reduced to the following expression:

q

b

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

ðG À1Þgd

3

i

_ ¼ sgnðuÞ8ðs

Ã

Às

Ãc

Þ

3=2

; ð10Þ

where s

Ãc

usually is set to 0.047.

This expression is one of the most used and it is applied

to rivers and channels with slope lower than 2%. A more

complex formula for non-rectangular cross-section

channels can be found in [46,48].

• Ferna´ ndez Luque & Van Beek equation (FL & V):

This formula is very similar to (10), and in fact, only one

coeﬃcient varies,

q

b

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

ðG À1Þgd

3

i

_ ¼ sgnðuÞ5:7ðs

Ã

Às

Ãc

Þ

3=2

: ð11Þ

• Van Rijn’s equation:

Van Rijn [54–56] developed the following formula for

the solid transport discharge:

z

f

h

A

R

H

h

A

R

H

z

f

z

b

(a) without sediment. (b) A fixed bottom and a sediment layar.

Fig. 2. Possible conﬁgurations of the bottom of the channel in function of the presence or not of a sediment layer.

302 M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316

q

b

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

ðG À1Þgd

3

i

_ ¼

0:005

C

1:7

D

d

i

h

_ _

0:2

s

1=2

Ã

ðs

1=2

Ã

Às

1=2

Ãc

Þ

2:4

; ð12Þ

where C

D

is the dragging coeﬃcient. Another formula

based on a complex theory mixing empirical results

and fundamental physics was proposed by Van Rijn

(see [51], for example).

• Nielsen formula:

Nielsen [39] developed the following formula:

q

b

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

ðG À1Þgd

3

i

_ ¼ sgnðuÞ12

ﬃﬃﬃﬃ

s

Ã

p

ðs

Ã

Às

Ãc

Þ: ð13Þ

In this case the usual value of s

Ãc

is set equal to

s

Ãc

= 0.05.

All these formulae have a range of application in func-

tion of grain size, the slope of the bottom, the Froude num-

ber and the relative density G. For example in the case of

the MP & M it is applied for a 0.4 6 d 6 29 mm, slopes

of the bottom smaller than 0.02 and 1.25 6 G 6 4.2 (see

[42]). In Section 6.3 we have used MP & M model in order

to compare numerical results with experimental data.

2.4. Uniﬁed formulation for solid transport discharge

formulae

The diﬀerent formulae of q

b

shown in the previous sub-

section can be written under a same structure as follows:

q

b

ðh; qÞ ¼ c

1

g

2

ðh; qÞðc

2

þc

3

g

1

ðh; qÞÞ

m

; ð14Þ

where m, c

1

, c

2

and c

3

are constant values and g

1

, g

2

are sca-

lar functions of h and q. For example,

• Grass model corresponds to

c

1

¼ A

g

; g

2

ðh; qÞ ¼ u; c

2

¼ 0; c

1

¼ 1;

g

1

ðh; qÞ ¼ juj; m ¼ m

g

À1:

• Meyer-Peter & Mu¨ ller’s model corresponds to,

c

1

¼ 8

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

ðG À1Þgd

3

i

_

; c

2

¼ À0:047; c

3

¼

cg

2

ðc

s

ÀcÞd

i

;

g

1

ðq; hÞ ¼

q

2

h

2

R

1=3

h

; g

2

ðq; hÞ ¼ sgnðuÞ; m ¼ 1:5:

2.5. Comparison of the diﬀerent models with Grass model

We observe that if we set s

Ãc

= 0 at the previous models,

then it is possible to interpret them as particular cases of

Grass model, for diﬀerent deﬁnitions of A

g

and m

g

. As

the value of A

g

represents the type of interaction between

ﬂuid and sediment, we can look for it in the diﬀerent

models.

We analyze in detail the MP & M’s model for s

*c

= 0.

The solid transport discharge in this case can be written

under the structure (9), where

A

g

¼

8g

2

cg

3

ðc

s

ÀcÞR

1=2

h

:

We observe that A

g

cannot be constant because it depends

on the hydraulic ratio. A

g

depends also on the speciﬁc sed-

iment weight, the speciﬁc water weight and Manning’s

coeﬃcient. Nevertheless, for s

Ãc

= 0, the MP & M’s model

does not depend on grain diameter d

i

. In fact, this is similar

for all considered models.

We consider the following values c = g Æ 1000,

c

s

= g Æ 2600, g = 0.0196. With this values, we obtain

A

g

%

3:623065897 Â10

À3

R

1=2

h

:

If the hydraulic ratio R

h

is approximated by h, and if

h % 10cm, we obtain A

g

% 8.163212495 · 10

À4

. For values

of h % 1 cm we obtain A

g

% 3.623065897 · 10

À3

. In any

case, we observe that MP & M always models a weak inter-

action between ﬂuid and sediment.

As shown in Table 1, all models considered here corre-

spond to a weak interaction between ﬂuid and sediment.

In fact, this is one of the main problems in order to obtain

precise numerical approximations. As the interaction

between the ﬂuid and the sediment is very weak, the velocity

of the ﬂuid is much greater than the velocity of the sediment.

Consequently, low order numerical methods do not capture

well the sediment evolution; they are usually very diﬀusive

(see test in Section 6.1). So, it is necessary to use high order

numerical schemes in order to avoid this phenomena.

2.6. Coupled model

The system of equations used in this work is obtained by

coupling Shallow Water equation (4) and the sediment

transport equation (5). The expression of this system is as

follows:

oh

ot

þ

oq

ox

¼ 0;

oq

ot

þ

o

ox

q

2

h

þ

1

2

gh

2

_ _

¼ Àgh

oz

b

ox

þgh

dH

dx

ÀghS

f

;

oz

b

ox

þn

o

ox

q

b

ðh; qÞ ¼ 0;

_

¸

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

¸

_

ð15Þ

where S

f

is the friction term modeling the drag eﬀects be-

tween the ﬂuid and the bed deﬁned by (3). Note that, if

the variable S = H À z

b

is deﬁned, as

oS

ot

¼ À

oz

b

ot

, (15) could

be rewritten as

Table 1

Comparison with Grass model

Formula A

g

m

g

Approximation

MP & M A

g

¼

8g

2

cg

3

ðc

s

ÀcÞR

1=2

h

3 3:62306 Â10

À3 1

R

1=2

h

Nielsen A

g

¼

12g

2

cg

3

ðc

s

ÀcÞR

1=2

h

3 5:43459 Â10

À3 1

R

1=2

h

Van Rijn A

g

¼

0:005g

2:2

g

3:4

ðGÀ1Þ

1:2

h

0:2

R

1:7=3

h

C

1:7

D

3.4 0:6751 Â10

À6 1

C

1:7

D

R

1:7=3

h

FL & V A

g

¼

5:7g

2

cg

3

ðc

s

ÀcÞR

1=2

h

3 2:58143 Â10

À3 1

R

1=2

h

M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316 303

oh

ot

þ

oq

ox

¼ 0;

oq

ot

þ

o

ox

q

2

h

þ

1

2

gh

2

_ _

¼ gh

oS

ox

ÀghS

f

;

oS

ot

Àn

oq

b

ox

¼ 0:

_

¸

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

¸

_

ð16Þ

Note that, system (16) can be written as an hyperbolic

system with a non-conservative term and a source term S

F

oW

ot

þ

oF ðW Þ

ox

¼ BðWÞ

oW

ox

þS

F

; ð17Þ

where

W ¼

h

q

S

_

_

_

_

; F ¼

q

q

2

h

þ

1

2

gh

2

Ànq

b

_

_

_

_

; BðW Þ ¼

0 0 0

0 0 gh

0 0 0

_

_

_

_

;

S

F

¼

0

ÀghS

f

0

_

_

_

_

: ð18Þ

Finally, system (17) can be written as a non-conservative

hyperbolic system

oW

ot

þAðW Þ

oW

ox

¼ S

f

; x 2 R; t > 0; ð19Þ

where

AðWÞ ¼ AðWÞ ÀBðWÞ; ð20Þ

and A(W) is the Jacobian matrix of F(W),

AðW Þ ¼

oF

oW

ðWÞ ¼

0 1 0

À

q

2

h

2

þgh 2

q

h

0

Àn

oq

b

oh

Àn

oq

b

oq

0

_

¸

_

_

¸

_: ð21Þ

Then, by the deﬁnitions of A(W) and B(W) we obtain that

AðW Þ ¼

0 1 0

À

q

2

h

2

þgh 2

q

h

Àgh

Àn

oq

b

oh

Àn

oq

b

oq

0

_

¸

_

_

¸

_: ð22Þ

The eigenvalues of the matrix AðWÞ can be obtained as

roots of the characteristic polynomial

PðkÞ ¼ k

3

þa

1

k

2

þa

2

k þa

3

; ð23Þ

where

a

1

¼ À2

q

h

; a

2

¼

q

2

h

2

Àgh 1 þn

oq

b

oq

_ _

and

a

3

¼ Àghn

oq

b

oh

:

The roots of this polynomial are

k

1

¼ 2

ﬃﬃﬃﬃﬃﬃﬃﬃ

ÀQ

_

cosðh=3Þ Àa

1

=3;

k

2

¼ 2

ﬃﬃﬃﬃﬃﬃﬃﬃ

ÀQ

_

cosððh þ2pÞ=3Þ Àa

1

=3;

k

3

¼ 2

ﬃﬃﬃﬃﬃﬃﬃﬃ

ÀQ

_

cosððh þ4pÞ=3Þ Àa

1

=3;

where Q ¼ ð3a

2

Àa

2

1

Þ=9, h ¼ arccosðR=

ﬃﬃﬃﬃﬃﬃﬃﬃﬃ

ÀQ

3

_

Þ and R ¼

ð9a

1

a

2

À27a

3

þ2a

3

1

Þ=54. The eigenvalues are real if

Q

3

+ R

2

< 0. In the case of Grass model, it can be proven

that all eigenvalues are real. Although in general we cannot

prove it for the other considered models, we have not

found any complex eigenvalue in our numerical tests.

The associated eigenvectors are

R

j

¼

1

k

k

q

h

Àk

k ð Þ

2

gh

À1

_

¸

¸

_

_

¸

¸

_

; j ¼ 1; 2; 3:

We note that for a ﬁxed bottom (q

b

= 0) the eigenvalues

are

q

h

þ

ﬃﬃﬃﬃﬃ

gh

_

;

q

h

À

ﬃﬃﬃﬃﬃ

gh

_

and 0:

By simplicity we will suppose that the system is strictly

hyperbolic, where k

i

5k

j

, if i 5j and all of them are real.

Moreover, we must remark that the non-conservative

product AðW Þ Á W

x

, has not sense as a distribution when

W presents a discontinuity. Following the theory devel-

oped by Dal Maso, et al. [16], can be deﬁned in a rigorous

way the non-conservative product, introducing a family of

paths in X.

Deﬁnition 1. A family of paths in X & R

N

is a local-

Lipschitz application

U : ½0; 1 ÂX ÂX ! X

that satisﬁes the following properties:

1. U(0;W

L

, W

R

) = W

L

and U(1;W

L

, W

R

) = W

R

, for all W

L

,

W

R

2 X.

2. For a bounded set O & X, there exists a constant k such

that

oU

os

ðs; W

L

; W

R

Þ

¸

¸

¸

¸

¸

¸

¸

¸

6 kjW

L

ÀW

R

j

for all W

L

, W

R

2 O and for almost all s 2 [0, 1].

3. For a given bounded set O & X, there exists a constant K

such that,

oU

os

ðs; W

1

L

; W

1

R

Þ À

oU

os

ðs; W

2

L

; W

2

R

Þ

¸

¸

¸

¸

¸

¸

¸

¸

6 KðjW

1

L

ÀW

2

L

j þjW

1

R

ÀW

2

R

jÞ;

for all W

1

L

, W

1

R

, W

2

L

, W

1

R

2 O and for almost all s 2 [0, 1].

Once a family of paths is chosen, U in X, given a func-

tion W 2 ðL

1

ðR ÂR

þ

Þ \ BV ðR ÂR

þ

ÞÞ

N

it is possible to

give a sense to the non-conservative product as a Borel

measure (see [40]), which is denoted by ½AðW ÞW

x

U

and

weak solutions of (19) are the functions satisfying the

equality

W

t

þ½AðWÞW

x

U

¼ S

F

:

The choice of the family of paths is very important, because

it determines the propagation speed of shocks. Neverthe-

less, in general, it is very diﬃcult to parametrice the

304 M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316

optimal choice, that must be related to the physics of the

problem. In practice, the simplest choice is given by a fam-

ily of segments,

Uðs; W

L

; W

R

Þ ¼ W

L

þsðW

R

ÀW

L

Þ; s 2 ½0; 1; ð24Þ

which corresponds to the non-conservative product deﬁni-

tion proposed by Volpert [59].

3. Finite volume method for non-conservative hyperbolic

systems

Friction term S

F

will be discretized in an semi-implicit

way (see [8]), so, in what follows we consider the system:

oW

ot

þAðW Þ

oW

ox

¼ 0; x 2 R; t > 0: ð25Þ

We consider a partition of the interval [0,L] in cells

deﬁned by I

i

¼ ½x

iÀ1=2

; x

iþ1=2

; i 2 Z. For the sake of simplic-

ity, we suppose that all of them have the same length Dx

and x

iþ

1

2

¼ iDx. x

i

= (i À 1/2)Dx is the center of the control

volume I

i

. Let Dt be the time step and t

n

= nDt.

Then we denote by W

n

i

an approximation on the mean

value of W over I

i

at time t = t

n

,

W

n

i

ﬃ

1

Dx

_

x

iþ1=2

x

iÀ1=2

Wðx; t

n

Þ dx:

To approximate W

n

i

we consider two diﬀerent types of

schemes, the generalized Roe method for non-conservative

hyperbolic systems and schemes with ﬂux limiters, which

are of second order in regular areas, at least for linear

problems. In the case of a non-conservative system it is also

necessary to deﬁne a generalization of Roe matrix. The def-

inition of Roe matrix is also associated to the choice of a

family of paths.

In [53] the following deﬁnition is introduced.

Deﬁnition 2. For a given family of paths W, a function

A: X ÂX !M

N

ðRÞ is a Roe linearization of system (25),

if it satisﬁes the following properties:

1. For each W

L

, W

R

2 X, A

W

ðW

L

; W

R

Þ has N real and dif-

ferent eigenvalues.

2. A

W

ðW ; W Þ ¼ AðW Þ, for all W2 X.

3. For each W

L

, W

R

2 X,

A

W

ðW

L

; W

R

ÞðW

R

ÀW

L

Þ

¼

_

1

0

AWðs; W

L

; W

R

Þ ½

oW

os

ðs; W

L

; W

R

Þ ds: ð26Þ

We observe that if AðW Þ is the Jacobian matrix of a regu-

lar ﬂux F(W), (26) is independent from the family of paths

and it is reduced to the usual Roe property

A

W

ðW

L

; W

R

ÞðW

R

ÀW

L

Þ ¼ F ðW

R

Þ ÀF ðW

L

Þ: ð27Þ

We will use the following notation: we denote

A

iþ1=2

¼ A

W

ðW

n

i

; W

n

iþ1

Þ ð28Þ

the Roe matrix associated to the states W

i

and W

i+1

, with

eigenvalues

k

iþ1=2

1

< k

iþ1=2

2

< Á Á Á < k

iþ1=2

N

;

and fR

iþ1=2

l

g

N

l¼1

is the base of the associated eigenvectors.

By K

iþ1=2

we denote the N · N matrix whose columns

are eigenvectors and by L

iþ1=2

, the diagonal matrix of

eigenvalues. We will also use the following matrices

L

þ

iþ1=2

, L

À

iþ1=2

, A

þ

iþ1=2

, A

À

iþ1=2

and jA

þ

iþ1=2

j:

L

Æ

iþ1=2

¼

ðk

iþ1=2

1

Þ

Æ

0 0

0 ðk

iþ1=2

2

Þ

Æ

0

0 0 ðk

iþ1=2

3

Þ

Æ

_

¸

¸

_

_

¸

¸

_

;

ðk

iþ1=2

k

Þ

Æ

¼

k

iþ1=2

k

Æjk

iþ1=2

k

j

2

;

A

Æ

iþ1=2

¼ K

iþ1=2

L

Æ

iþ1=2

K

À1

iþ1=2

;

and

jA

iþ1=2

j ¼ A

þ

iþ1=2

ÀA

À

iþ1=2

:

The numerical scheme progresses in time as follows:

once the approximations at time t

n

, W

n

i

, have been calcu-

lated, a Linear Riemann problem is considered at each

intercell x = x

i+1/2

with the associated Roe matrix A

iþ1=2

and the constant states W

n

i

and W

nþ1

i

, respectively. The

approximations at time t

n+1

, W

nþ1

i

, are obtained by averag-

ing in the cells the solutions of the Riemann problem. As in

the case of a system of conservation laws, some calcula-

tions allow to show that, under the hypothesis,

x

iÀ1=2

þk

iÀ1=2

1

Dt 6 x

i

6 x

iþ1=2

þk

iþ1=2

N

Dt; ð29Þ

the approximations at time t

n+1

can be obtained by the

formula:

W

nþ1

i

¼ W

n

i

À

Dt

Dx

ðA

þ

D;iÀ1=2

ðW

n

i

ÀW

n

iÀ1

Þ

þA

À

D;iþ1=2

ðW

n

iþ1

ÀW

n

i

ÞÞ; ð30Þ

where

A

Æ

D;iþ1=2

¼

A

iþ1=2

ÆD

iþ1=2

2

: ð31Þ

D

iþ1=2

is the viscosity matrix of the numerical scheme de-

ﬁned by

(a) Roe method:

D

iþ1=2

¼ jA

iþ1=2

j: ð32Þ

(b) Method with ﬂux limiters: the idea is to use a combi-

nation of Roe method (which is only of order one)

near discontinuities, and the generalization of Lax–

Wendroﬀ method (which is second order in space

and time, at least for linear systems) in regular areas.

The obtained scheme corresponds to deﬁne

D

iþ1=2

¼ jA

iþ1=2

j ÀK

iþ1=2

ðL

u

Þ

iþ1=2

K

À1

iþ1=2

; ð33Þ

with,

L

u

¼ Diag sgnðk

j

Þ À

Dt

Dx

k

j

_ _

k

j

uðr

j

Þ; j ¼ 1; . . . ; N

_ _

:

ð34Þ

M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316 305

The function u is the ﬂux limiter and

r

j

¼

a

j;i

0

a

j;i

; a

j;i

¼ ½K

À1

ðW

iþ1

ÀW

i

Þ

j

with

i

0

¼ i Àsgnðk

j

Þ: ð35Þ

Some classical examples of ﬂux limiters function are

(see [21,36]):

– Van Leer ﬂux limiter:

uðrÞ ¼

jrj þr

1 þjrj

:

– The family of ﬂux limiters functions deﬁned by

uðrÞ ¼ maxð0; minð1; brÞ; minðb; rÞÞ:

If b = 1 we have the minmod limiter, b = 2 corre-

sponds to superbee limiter.

Moreover it can be proven that scheme (30) is well

balanced.

Remark 1. In practice, to ensure (29) we use the CFL

condition

maxfjk

iþ1=2

l

j; 1 6 l 6 N; i 2 Zg

Dt

Dx

6 c

with 0 < c 6 1.

Taking into account that the system (17) has a conserva-

tive term and a non-conservative product, the numerical

scheme could be rewritten as follows:

W

nþ1

i

¼ W

n

i

þ

Dt

Dx

G

iÀ1=2

ÀG

iþ1=2

_ _

þ

Dt

2Dx

ðB

iÀ1=2

ðW

n

i

ÀW

n

iÀ1

Þ

þB

iþ1=2

ðW

n

iþ1

ÀW

n

i

ÞÞ ð36Þ

where

G

iþ1=2

¼

1

2

ðF ðW

n

i

Þ þF ðW

n

iþ1

ÞÞ À

1

2

D

iþ1=2

ðW

n

iþ1

ÀW

n

i

Þ:

ð37Þ

4. High order ﬁnite volume scheme based on state

reconstructions

There exist diﬀerent techniques to obtain high order

methods. A possibility is to use state reconstructions.

Firstly a state reconstruction operator P

t

is considered,

that is, an operator that associates, to a given sequence

{W

i

(t)}, two new sequences fW

À

iþ1=2

ðtÞg, fW

þ

iþ1=2

ðtÞg in such

a way that, whenever

W

i

ðtÞ ¼

1

Dx

_

Ii

Wðx; tÞ dx; 8i 2 Z

for some regular function W, then

W

Æ

iþ1=2

ðtÞ ¼ W ðx

iþ1=2

; tÞ þOðDx

p

Þ; 8i 2 Z:

For conservative hyperbolic systems it is enough with these

reconstructions, nevertheless for non-conservative terms it

is necessary to know the state reconstruction over the inter-

val I

i

. That happens because if AðWÞW

x

has an associated

ﬂux and we integrate it over the control volume, then only

the states at the boundary of I

i

(x

iÀ1/2

and x

i+1/2

) are nec-

essary. However if an associated ﬂux does not exist, we

must integrate AðWÞW

x

over all the interval.

Then, ﬁrstly, over each control volume I

i

, at each instant

t > 0, we deﬁne a regular function P

t

i

such that

lim

x!x

þ

iÀ1=2

P

t

i

ðxÞ ¼ W

þ

iÀ1=2

ðtÞ; lim

x!x

À

iþ1=2

P

t

i

ðxÞ ¼ W

À

iþ1=2

ðtÞ: ð38Þ

In [10] the following numerical scheme is proposed:

W

0

i

¼ À

1

Dx

_

A

þ

iÀ1=2

ðW

þ

iÀ1=2

ðtÞ ÀW

À

iÀ1=2

ðtÞÞ

þA

À

iþ1=2

ðW

þ

iþ1=2

ðtÞ ÀW

À

iþ1=2

ðtÞÞ

þ

_

x

iþ1=2

x

iÀ1=2

A½P

t

i

ðxÞ

d

dx

P

t

i

ðxÞdx

_

; ð39Þ

where P

t

i

is a regular function that veriﬁes (38). Applied to

our system (17) with a conservative part and a non-conser-

vative product, it can be rewritten as

W

0

i

¼

Dt

Dx

ð

¯

G

iÀ1=2

À

¯

G

iþ1=2

Þ þ

Dt

2Dx

ðB

iÀ1=2

Á ðW

þ

iÀ1=2

ÀW

À

iÀ1=2

Þ

þB

iþ1=2

Á ðW

þ

iþ1=2

ÀW

À

iþ1=2

ÞÞ þ

Dt

Dx

I

B;i

; ð40Þ

where

¯

G

iÀ1=2

¼

1

2

ðF ðW

À

iþ1=2

Þ þF ðW

þ

iþ1=2

ÞÞ

À

1

2

jA

iþ1=2

j Á ðW

þ

iþ1=2

ÀW

À

iþ1=2

Þ: ð41Þ

A

iþ1=2

is the Roe matrix and

I

B;i

¼

_

x

iþ1=2

x

iÀ1=2

B½

¯

P

t

i

d

dx

P

t

i

ðxÞ dx: ð42Þ

Eﬀectively, we see that the main diﬀerence between the

conservative and the non-conservative system is that the

conservative one is independent of the deﬁnition of P

t

i

, only

depends on W

Æ

iÆ1=2

. The consequences are that the order of

the numerical scheme depends upon the order of the state

reconstruction operator P

t

i

and its derivative. Concretely,

in [10] the following theorem is proven.

Theorem 1. If A is order C

2

with bounded derivatives and

A

W

is bounded. We suppose that the state reconstruction

operator of order p veriﬁes that for a sequence deﬁned by

W

i

¼

1

Dx

_

Ii

WðxÞ dx

for some regular function W,

P

i

ðx; W

iÀl

; . . . ; W

iþr

Þ ¼ W ðxÞ þOðDx

q

Þ; 8x 2 I

i

;

d

dx

P

i

ðxÞ ¼ W

0

þOðDx

r

Þ; 8x 2 I

i

:

Then (39) provides an approximation of at least order

a = min(p, q + 1,r + 1) for system (25) in the following sense:

306 M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316

_

A

þ

iÀ1=2

ðW

þ

iÀ1=2

ðtÞ ÀW

À

iÀ1=2

ðtÞÞ þA

À

iþ1=2

ðW

þ

iþ1=2

ðtÞ

ÀW

À

iþ1=2

ðtÞÞþ

_

x

iþ1=2

x

iÀ1=2

A½P

t

i

ðxÞ

d

dx

P

t

i

ðxÞ dx

_

¼

_

x

iþ1=2

x

iÀ1=2

AðWðx; tÞÞW

x

ðx; tÞ dx þOðDx

a

Þ ð43Þ

for any sufﬁciently smooth solution W.

Remark 2. For usual state reconstruction operators, we

have r 6 q 6 p and then, the order of (39) is c = r + 1 for

non-conservative systems, by while for conservative sys-

tems the order is p. So, the order of the numerical scheme

for non-conservative systems is usually smaller than for

conservative one.

In this paper we have used state operator reconstruction

deﬁned by WENO reconstructions with stencils of r points

(see [28,30,49]). So, we denote by r-WENO method the

scheme obtained by this state reconstruction. For this oper-

ator of reconstruction we obtain at the internal points of

the control volume only order r and q = r À 1 for their

derivatives, so, by previous theorem, the scheme is order r.

For time evolution, we have used the high order TVD

Runge–Kutta method proposed in [24,50].

For the numerical tests presented in Section 6 we use a

3-point Gauss quadrature formula to approximate integral

(42).

5. Roe matrix and numerical treatment of transitions close

the critical shear stress

In this section we study the problem of determining Roe

matrices for the coupled model formed by Shallow Water

equations and sediment transport deﬁned by the family

of solid transport discharge (deﬁned in Section 2.3). In

the particular case of Grass model it is possible to deter-

mine a Roe matrix, when the family of paths is deﬁned

by segments. Nevertheless the calculus of the Roe matrices

is complicated for the other models presented in this work,

and its implementation is very costly. So, in practice we use

an approximation of Roe matrix. In Section 5.3 we present

the numerical treatment that we perform in transition

interfaces with–without sediment, with–without movement

of the sediment layer.

5.1. Roe matrix for Grass model

In the particular case where Grass model is considered

and the family of paths is deﬁned by segments (24), Roe

matrix is (see [29]),

A

W

ðW

L

; W

R

Þ ¼

0 1 0

Àð~uÞ

2

þð~cÞ

2

2~u ð~cÞ

2

À~u

~

d

~

d 0

_

¸

_

_

¸

_; ð44Þ

where

~u ¼

ﬃﬃﬃﬃﬃ

h

R

p

u

R

þ

ﬃﬃﬃﬃﬃ

h

L

p

u

L

ﬃﬃﬃﬃﬃ

h

R

p

þ

ﬃﬃﬃﬃﬃ

h

L

p ;

~

h ¼

1

2

ðh

R

þh

L

Þ; ~c ¼

ﬃﬃﬃﬃﬃ

g

~

h

_

; ð45Þ

~

d ¼

A

g

n

ﬃﬃﬃﬃﬃ

h

R

p

þ

ﬃﬃﬃﬃﬃ

h

L

p _ _

ﬃﬃﬃﬃﬃ

h

L

p

h

R

þ

ﬃﬃﬃﬃﬃ

h

R

p

h

L

mgÀ1

k¼0

ðu

R

Þ

k

ðu

L

Þ

mgÀðkþ1Þ

: ð46Þ

5.2. Approximation of Roe matrix

In the previous subsection we have presented a Roe

matrix corresponding to Grass model. But for the other

models considered in this work, is not always possible to

obtain an explicit formula and its implementation is very

costly. So, in practice we use the following approximation

of Roe matrix:

A

W

ðW

L

; W

R

Þ ¼

0 1 0

Àð~uÞ

2

þð~cÞ

2

2~u ð~cÞ

2

Àn

oq

b

oh

ð~u;

~

hÞ Àn

oq

b

oq

ð~u;

~

hÞ 0

_

¸

¸

_

_

¸

¸

_

;

ð47Þ

with ~u,

~

h and ~c deﬁned by (46), that is, the mean Roe values

of u, h and c corresponding to Shallow Water equations.

Remark 3. We observe that as we use an approximated

Roe matrix, we cannot use directly schemes on function of

A

Æ

, we must use the schemes that explicitly depend on the

physical ﬂux function F(W), that is, scheme (36), (37) or

scheme (40), (41).

5.3. Numerical treatment of transitions close to the critical

shear stress

As we have seen in Section 2.3, in general, the movement

of the sediment layer starts when shear stress s is bigger

than a critical shear stress s

c

. So, when we use this type

of models, at interface x

i+1/2

, we could ﬁnd some of the fol-

lowing situations:

(st1) Transition with–without sediment: if x

i+1/2

is the inter-

face between I

i

and I

i+1

, with a sediment layer of

thickness z

b,i

such that the shear stress at I

i

is bigger

than the critical one (s

i

> s

c

), and there is not sedi-

ment at the control volume I

i+1

(Fig. 3a).

(st2) Transition with–without movement of the sediment: if

x

i+1/2

is the interface between I

i

and I

i+1

, with a sed-

iment layer with thickness z

bi

and such that the shear

stress is bigger than the critical one (s

i

> s

c

); and the

volume I

i+1

, with a sediment layer of thickness

z

b,i+1

where the shear stress is smaller than the critical

one (s

i+1

< s

c

) (Fig. 3b).

In order to avoid instabilities, it is necessary to change

the numerical scheme so that these situations are numeri-

cally well treated.

M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316 307

For example, Let us suppose that in the intercell x

i+1/2

we have the following situation (Fig. 3):

s

i

> s

c

;

s

iþ1

< s

c

;

s

iþ1=2

< s

c

;

ð48Þ

where s

i

is the shear stress at volume I

i

and s

i+1/2

is the

shear stress corresponding to the mean value of Roe. In

this case, we propose the following algorithm:

1. A

iþ1=2

¼ K

iþ1=2

L

iþ1=2

K

À1

iþ1=2

.

2. Let k

iþ1=2

3

be the characteristic velocity corresponding to

the sediment transport at the intercell. As s

i+1/2

< s

c

,

then k

iþ1=2

3

¼ 0. Next, we set

k

iþ1=2

3

¼ k

i

3

, where k

i

3

is the

characteristic velocity corresponding to the sediment

transport at the cell I

i

. Note that k

i

3

6¼ 0 as s

i

> s

c

.

3. The diﬀusion matrix D

iþ1=2

is computed using the mod-

iﬁed value

k

iþ1=2

3

instead of k

iþ1=2

3

¼ 0.

The algorithm can be summarized as follows: The values

s

i

, s

i+1

and s

i+1/2

are computed. If in the three cases the

shear stress is bigger than the critical shear stress we do

not perform any change. In the case of a transition we

change the diﬀusion matrix as described before.

6. Numerical test

In this section we present some numerical tests to vali-

date numerical schemes described in Sections 3 and 4.

We have considered an academic test with a known analyt-

ical solution, a test of order and an experimental test where

we compare with experimental measures.

6.1. Transport of parabolic sediment layer

This ﬁrst test consists on comparing a numerical solu-

tion with an asymptotic and analytical solution obtained

by Hudson and Sweby in [29], for Grass model when inter-

action constant A

g

(8) is smaller than 10

À2

. In this case, the

layer sediment ~z

b

is over all computational domain and

ﬂuid is moving slowly with a constant discharge

q = q

0

6 10. Under these hypothesis it is possible to obtain

the following analytical solution,

h ¼ A

r

Àz

b

ðx; tÞ; q ¼ q

0

being A

r

a ﬁxed level of reference, q

0

6 10 a constant value,

and

z

b

ðx; tÞ ¼

sin

2 pðx

0

À300Þ

200

_ _

if 300 6 x

0

6 500;

0 otherwise;

_

ð49Þ

where x

0

is the solution of equation

x ¼ x

0

þA

g

nm

g

q

mg

0

t A

r

Àsin

2 pðx

0

À300Þ

200

_ _ _ _

Àðmgþ1Þ

if 300 6 x

0

6 500;

x ¼ x

0

þA

g

nm

g

q

mg

0

tA

Àðmgþ1Þ

r

otherwise:

_

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

_

ð50Þ

The usual value of m

g

= 3 is considered.

This solution is valid to t < t

0

, being t

0

the instant in

which characteristics cross. In [29] it is estimated

t

0

¼ 238079:124 ÂA

À1

g

.

In the same way, it is veriﬁed that

~z

b

ðx; tÞ ¼ z

b

ðx; tÞ þc

0

;

where c

0

is a constant value, it is also a solution of the

problem.

To be able to compare numerical solution with analyti-

cal solution obtained in [29], without considering the

behavior of numerical scheme in zones with–without sedi-

ment, we have compared the solution for the case,

~z

b

ðx; tÞ ¼ z

b

ðx; tÞ þ0:1:

We consider a computational domain whose length is

L = 1000 m, discretized with 250 nodes. CFL condition is

set to 0.8. The sediment porosity is set to q

0

= 0.4 and

the constant A

g

of Grass formula (8) is set to 0.001 which

corresponds to a weak interaction.

The initial conditions are (see Fig. 4),

hðx; 0Þ ¼ 10 Àz

b

ðx; 0Þ; qðx; 0Þ ¼ 10;

z

b

ðx; 0Þ ¼

0:1 þsin

2 pðxÀ300Þ

200

_ _

if 300 6 x 6 500;

0:1 otherwise:

_

ð51Þ

As boundary condition, the ﬂux and the depth of the

sediment is imposed upstream, while free boundary condi-

tions are imposed downstream.

In Figs. 5–7 we compare analytical solution (continuous

line) and numerical solution obtained with the diﬀerent

schemes of Sections 3 and 4, corresponding to sediment

(a) with/without sediment. (b) With sediment in both cells, τ in cell

in cell

and τ

τ τ

Fig. 3. Possible cases of transition around the critical shear stress.

308 M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316

layer evolution at instant t = 238080 s; that is smaller than

the maximal time in which analytical solution (50) is valid.

We can observe that all numerical schemes show a good

sediment layer localization, being Roe scheme the most dif-

fusive (Fig. 5a). Moreover, the scheme that gives the best

approximation is high order generalized Roe scheme with

Weno state reconstructions of order 3, where time approx-

imation is made using Runge–Kutta of order 3 (Fig. 7a).

In Fig. 5 it is shown comparison between Roe method

and linearized Lax–Wendroﬀ method with ﬂux limiters.

In both methods we use Euler scheme for time discretiza-

tion. In Fig. 5a, that describes layer sediment evolution,

it is observed that Roe method is more diﬀusive than

method with ﬂux limiters, that is of second order in space

.

Fig. 4. Initial condition.

0 100 200 300 400 500 600 700 800 900 1000

9.9985

9.999

9.9995

10

10.0005

10.001

10.0015

t = 238080

(a) Sediment layer thickness. (b) Mass-flow.

0 100 200 300 400 500 600 700 800 900 1000

9.9985

9.999

9.9995

10

10.0005

10.001

10.0015

t = 238080

q roeclf

q roeslf

q roeclf

q roeslf

Fig. 5. Roe-Flux limiters (dotted line). Euler–Roe (dash line).

0 100 200 300 400 500 600 700 800 900 1000

0

0.2

0.4

0.6

0.8

1

t = 238080

sol exact

rk2w2sl

roecl

0 100 200 300 400 500 600 700 800 900 1000

9.9992

9.9994

9.9996

9.9998

10

10.0002

rk2w2sl

roecl

(a) Sediment layer thickness. (b) Mass-flow fluid.

Fig. 6. Roe-Flux limiters (dash line). Weno2-Rk2 (dotted line).

M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316 309

and time for linear problems. In Fig. 5b the discharge is

presented. We observe that it is near 10 in both cases and

almost constant in all domain, that is one of the hypothesis

used to develop the analytical solution (50).

In Fig. 6 it is compared ﬂux limiter scheme with Roe–

Weno of order 2. For time evolution it is used Euler for

ﬂux limiter scheme, because Lax–Wendroﬀ scheme ensures

order 2 in space and time. For scheme with Weno2 state

reconstructions it is used TVD Runge–Kutta method of

order 2. It must be observed that approximation of thick-

ness sediment layer is similar in both schemes.

Finally in Fig. 7 schemes with Weno2 and Weno3 state

reconstructions are compared. For time evolution it is used

Runge-Kutta2 and Runge-Kutta3, respectively. Higher

order scheme approximates better the sediment layer thick-

ness as expected.

6.2. Test of order

We make the following test to compare the order of the

proposed numerical schemes. We consider a one-dimen-

sional channel with ﬂat bed along the interval [À 10, 10],

and with the following initial condition (see Fig. 8):

qðx; 0Þ ¼ 0; hðx; 0Þ ¼ 2 À0:1e

Àx

2

; z

b

ðx; 0Þ ¼ 0:1 À0:01e

Àx

2

:

The exact solution for this problem is unknown, there-

fore we use as reference solution a numerical one obtained

with a very ﬁne mesh, of 5120 volumes.

This test has been made using Grass model (8) for solid

discharge. We consider a strong interaction between sedi-

ment layer and ﬂuid, concretely we impose A

g

= 0.3.

0 100 200 300 400 500 600 700 800 900 1000

0

0.2

0.4

0.6

0.8

1

t = 238080

sol exact

rk2w2sl

rk3w3sl

0 100 200 300 400 500 600 700 800 900 1000

9.9992

9.9994

9.9996

9.9998

10

t = 238080

rk2w2sl

rk3w3sl

(a) Sediment layer thickness. (b) Mass-flow fluid.

Fig. 7. Weno2-Rk2 (dotted line). Weno3-Rk3 (dash line).

2

0.0 0.0

0.1

0.098

0.096

0.094

0.092

0.09

1.5

1

0.5

0

-10 -5 0 5 10 -10 -5 0 5 10

(a) Initial condition. (b) Sediment layer thickness

Fig. 8. Test of order. Initial condition. Layer water thickness (h): dashed line. Sediment layer thickness (z

b

): continuous line. Depth (H): dotted line.

310 M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316

In Figs. 9 and 10 the diﬀerent results are presented, at

instant t = 0.5 s, obtained with the schemes: Euler–Roe

(described in Section 3), Roe–Weno2–RungeKutta2,

Roe–Weno3–RungeKutta3 (described in Section 4). Tables

2–4 shown the errors for variables h, q and z

b

in L

1

norm,

for t = 0.5. It must be observed that numerical schemes get

in this test the expected order: Euler–Roe order one, Roe–

Weno2–RungeKutta2 order two and Roe–Weno3–Run-

geKutta3 order three.

6.3. Comparison with experimental data

In this section we present an experiment where we sim-

ulate sediment layer evolution over a soil which is not

eroded. The numerical stationary solution is compared

with the experimental data obtained by the Hydraulic Lab-

oratory of Escuela Superior de Ingenieros de Caminos,

Canales y Puertos (A Corun˜ a University) over a channel

of 15 m long and 0.5 m width.

Pen˜a Gonza´lez details in his PhD Thesis [42] diﬀerent

techniques and toolkits used to develop this and other

experiments. In this particular case, measures were made

using a PIV technique and a Scanner 3D technique (in

[42] the details of the diﬀerent measurement techniques

are explained).

The experimental test was developed introducing a sand

layer in the central part of laboratory channel, and induc-

ing hydrodynamical conditions to erode the sand layer,

until a steady state is reached. The channel (Fig. 11) has

a very small slope of 0.052%. Sand layer was situated in

interval [4.5 m, 9 m], with a thickness of 4.5 cm; being

media diameter of the grain equals to 1 mm. During the

experiment the sand layer was eroded very slowly, extract-

ing material from upstream at same time that sand was

being transport downstream, creating a tongue of sand.

After approximately 8 h, a stationary state is reached and

the sand front was placed at the point x

s

= 12.15 m, so

the total material advance was 3.15 m. The ﬁnal thickness

of the sand layer was approximated equal to 3 cm along

all length, except in the transition areas at initial and ﬁnal

extremes.

The numerical simulation has been made using Meyer-

Peter & Mu¨ ller’s equation (10) for solid transport

discharge.

-10 -5 0 5 10 -10 -5 0 5 10

-10 -5 0 5 10

1.94

1.95

1.96

1.97

1.98

1.99

2

2.01

h (x,0.5) q (x,0.5)

z

b

(x,0.5)

Euler-Roe

Weno2-Rk2

Exacta

-0.2

-0.1

0

0.1

0.2

Weno2-Rk2

Weno3-Rk3

Exacta

0.09

0.092

0.094

0.096

0.098

0.1

Euler-Roe

Weno2-Rk2

Exacta

-3 -2.8 -2.6 -2.4 -2.2 -2 -1.8 -1.6

9.9e-2

9.92e-2

9.94e-2

9.96e-2

9.98e-2

1e-1

1.6 1.8 2 2.2 2.4 2.6 2.8 3

9.9e-2

9.92e-2

9.94e-2

9.96e-2

9.98e-2

1e-1

(a) Water layer thickness .

(b) Mass-flow .

(c) Sediment layer thickness .

Fig. 9. Order test. Comparison between the solution at t = 0.5 s. Fine mesh (continuous line). Euler–Roe (continuous line with dots). Weno2-Rk2

(continuous line with stars).

M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316 311

1.94

1.95

1.96

1.97

1.98

1.99

2

2.01

-10 -5 0 5 10 -10 -5 0 5 10

-10 -5 0 5 10

0.09

0.092

0.094

0.096

0.098

0.1

-0.2

-0.1

0

0.1

0.2

-3 -2.8 -2.6 -2.4 -2.2 -2 -1.8 -1.6

9.9e-2

9.92e-2

9.94e-2

9.96e-2

9.98e-2

1e-1

1.6 1.8 2 2.2 2.4 2.6 2.8 3

9.9e-2

9.92e-2

9.94e-2

9.96e-2

9.98e-2

1e-1

h (x,0.5) q (x,0.5)

z

b

(x,0.5)

(a) Water layer thickness .

(b) Mass-flow .

(c) Sediment layer thickness .

Fig. 10. Order test. Comparison with the ﬁne mesh solution in t = 0.5 s. Fine mesh solution (continued line), Weno2-Rk2 (continued line with stars) and

Weno3-Rk3 (continued line with diamonds).

Table 2

Errors and order: generalized Roe scheme

Nodes L

1

err h L

1

order h L

1

err q L

1

order q L

1

err z

b

L

1

order z

b

20 0.1070 0.4118 0.6148 · 10

À3

40 0.0701 0.6111 0.2810 0.5513 0.4218 · 10

À3

0.5433

80 0.0418 0.7441 0.1733 0.6971 0.2726 · 10

À3

0.6297

160 0.0234 0.8370 0.0985 0.8154 0.1639 · 10

À3

0.7335

320 0.0124 0.9066 0.05287 0.8977 0.1639 · 10

À3

0.85430

640 0.0064 0.9521 0.0274 0.9481 0.0907 · 10

À3

0.9168

Table 3

Errors and order: Weno2 + Runge Kutta 2

Nodes L

1

err h L

1

order h L

1

err q L

1

order q L

1

err z

b

L

1

order z

b

20 0.0968 0.4050 0.5338 · 10

À3

40 0.0521 0.8923 0.2226 0.8630 0.3073 · 10

À3

0.7966

80 0.0160 1.7043 0.0689 1.6919 0.1331 · 10

À3

1.2074

160 0.0047 1.7515 0.0203 1.7580 0.0418 · 10

À3

1.6702

320 0.0012 2.0000 0.0051 2.0025 0.0113 · 10

À3

1.8824

640 0.0003 2.0995 0.0012 2.0934 0.0027 · 10

À3

2.0303

312 M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316

The ﬁrst problem is to give an admissible initial condi-

tion for the laboratory experiment, because we only know

the position and the thickness of the sediment layer z

b

at

the steady state, but we have not information about initial

conditions. To build an initial condition we have proceeded

as follows:

a) Initial condition for ﬂux: we know by experimental

data that the average thickness of the sediment layer

at the steady state is about 3 cm. Moreover the thick-

ness of the water column over the sediment layer is

approximately 9.5 cm. Using the shear stress expres-

sion and the critical shear stress value, it is possible

to determine the critical discharge value, q

c

, for which

solid transport discharge is zero. We obtain the value

of q

c

= 0.0285 cm

2

/s, which we use as initial condi-

tion for the ﬂux.

b) As we have information about the thickness of the

water layer at the right extreme of the channel, and

therefore, ﬂuid surface elevation, g, a ﬁrst choice for

the initial condition over the thickness of the water

column is to consider hðx; 0Þ ¼ g ÀHðxÞ Àz

b

ðx; 0Þ.

However, for this initial condition we observe an

‘anomalous’ behavior of the evolution of the sedi-

ment layer. To avoid this problem, ﬁrstly we calculate

the free water surface using Bernoulli’s equation,

imposing a constant discharge q = q

c

along the chan-

nel and using as topography the function given by

HðxÞ ¼ HðxÞ Àz

b

ðx; 0Þ. That is, calculating the steady

solution for the topography HðxÞ ¼ HðxÞ Àz

b

ðx; 0Þ,

the discharge q = q

c

, and such that the thickness of

the water layer at the right extreme of the channel

is equal to the experimental measure.

Therefore, as we do not know the initial condition and

we must impose it using the technique explained before,

we can only compare the numerical and experimental solu-

tion at steady state. In Fig. 12 we show the initial condition

used in this numerical simulation.

For numerical simulation we have meshed the domain

with 250 nodes The CFL is set to 0.9. Sediment porosity

is set to 0.4. Friction between ﬂuid and bed is modeled

using a Manning’s law with coeﬃcient equal to 0.0125 over

the ﬁxed bed and 0.0196 over the sediment layer. As

boundary conditions we impose an incoming discharge

equal to 0.0285 m

2

/s upstream, while the water thickness

is imposed to 0.129 m downstream.

The model that we are using does not include pressure

forces, so the sediment does not fall by its own weight

due to gravity eﬀects. For this reason we cannot expect that

the numerical scheme could reproduce accurately the zones

where gravity eﬀects are relevant, as for example, the

advancing front of the sand layer, but it must reproduce

at least, the downstream sand slope and the median proﬁle

of the sediment layer. This behavior is reﬂected in Fig. 13,

that shows at 10 min, 40 min and 120 min, comparison

between experimental measures (line with dots) obtained

by a classic technique (see [42]), and the numerical

solutions obtained with Roe scheme without state

reconstructions (dash line) and scheme with Weno2 state

0m

15m

4.5m 9m

0.052%

4.5cm

0

.

5

m

Fig. 11. Sketch of the laboratory experiment.

0 2 4 6 8 10 12 14 16

-0.12

-0.1

-0.08

-0.06

-0.04

-0.02

0

0.0

Fig. 12. Initial condition imposed in numerical test. Layer water thickness

(h): dashed line. Sediment layer thickness (z

b

): continuous line. Depth (H):

dotted line.

Table 4

Errors and order: Weno3 + Runge Kutta 3

Nodes err L

1

h order L

1

h err L

1

q order L

1

q err L

1

z

b

order L

1

z

b

20 0.0398 0.2690 0.2689 · 10

À3

40 0.0115 1.7912 0.0479 2.4880 0.0656 · 10

À3

2.0333

80 0.0009 3.6751 0.0038 3.6307 0.0064 · 10

À3

3.3407

160 6.0 · 10

À5

3.8304 0.00026 3.858 4.5709 · 10

À7

3.8264

320 6.0 · 10

À6

3.3870 2.5334 · 10

À5

3.3973 4.2298 · 10

À8

3.4338

640 6.9 · 10

À7

3.1220 2.9077 · 10

À6

3.1231 4.8039 · 10

À9

3.1382

M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316 313

reconstructions (continuous line). In this ﬁgure it is possi-

ble to observe that the scheme of order one (Roe without

state reconstructions) is more diﬀusive than the scheme of

order two. Although we do not know the initial condition,

the numerical solution tends to the steady state. In the dif-

ferent instants, the graphics show that we are approximat-

ing the average of the thickness sediment layer. Moreover,

numerically it is proved that we conserve total mass of sed-

iment layer.

In Fig. 14 we compare the experimental measures of the

position of the sediment layer top at the steady state with

the numerical solutions obtained using diﬀerent numerical

schemes. In this ﬁgure line with dots represents experimen-

tal data, line with squares represents the solution obtained

0 2 4 6 8 10 12 14 16

0

0.01

0.02

0.03

0.04

0.05

t = 10 min

0 2 4 6 8 10 12 14 16

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

t = 40 min

4.2e-2

4.3e-2

4.4e-2

4.5e-2

4.6e-2

4.7e-2

4.8e-2

4.9e-2

5e-2

t = 10 min

(a) = 10 min.

(c) = 10 min. Detail of the sediment layar top

(d) = 40 min. Detail of the sediment layar top

(f) = 120 min. Detail of the sediment layar top

(b) = 40 min.

5 5.5 6 6.5 7 7.5 8 8.5

5.5 6 6.5 7 7.5 8

6 6.5 7 7.5 8 8.5

0.03

0.032

0.034

0.036

0.038

0.04

0.042

0.044

t = 40 min

0 2 4 6 8 10 12 14 16

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

t = 120 min

0.03

0.032

0.034

0.036

0.038

0.04

0.042

0.044

t = 120 min

fondo

Numerica

Numerica Euler

Experimental

fondo

Numerica Weno2+Rk2

Numerica Euler

Experimental

Numerica Weno2+Rk2

Numerica Euler

Experimental

NumericaWe

Numerica Euler

Experimental

no2+Rk2

fondo

Numerica W eno2+Rk2

Numerica Euler

Experimental

NumericaWeno2+Rk2

Numerica Euler

Experimental

Weno2+Rk2

(e) = 10 min.

Fig. 13. Sediment layer evolution at 10 min, 40 min and 120 min. Experimental measures (continuous line with dots). Euler–Roe (dashed line), Weno2-

Rk2 (continuous line).

314 M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316

by scheme with reconstruction Weno2, and line with dia-

monds represents the solution got by Roe scheme without

state reconstructions. In Fig. 14 it is observed that both

numerical solutions converge slowly to the steady state.

In this example, it is diﬃcult to say what numerical

scheme reproduce better the laboratory experiment, mainly

because the physical model does not include pressure

forces, so the sediment does not fall by its own weight

due to gravity eﬀects. Nevertheless, we could say that, in

general (see Fig. 13), the numerical scheme Rk2-Weno2

captures better the depth proﬁles of the sediment layer at

the initial part of the sediment layer, where gravity eﬀects

are not so relevant, while Euler–Roe scheme not. A more

complex physical model including the gravity eﬀects at

the sediment layer must be considered in order to repro-

duce better the laboratory experiment.

References

[1] Alouges F, Merlet B. Approximate shock curves of non-conservative

hyperbolic equations in one space dimension. Preprint, available at

<http://mahery.math.u-psud.fr/alouges/full_publi.html>.

[2] Alcrudo F, Benkhaldoun F. Exact solutions to the Riemann problem

of the Shallow Water equations with a bottom step. Comput Fluids

2001;30:643–71.

[3] Audusse E, Bouchut F, Bristeau MO, Klein R, Perthame B. A fast

and stable well-balanced scheme with hydrostatic reconstruction for

Shallow Water ﬂows. SIAM J Sci Comp 2004;25(6):2050–65.

[4] Bermu´ dez A, Va´zquez ME. Upwind methods for hyperbolic conser-

vation laws with source terms. Comput Fluids 1994;23(8):1049–71.

[5] Bianchini S, Bressan A. Vanishing viscosity solutions of nonlinear

hyperbolic systems. Ann Math 2005;161(1):223–342.

[6] Bouchut F. Nonlinear stability of ﬁnite volume methods for

hyperbolic conservation laws and well-balanced schemes for

sources. Birkha¨user; 2004.

[7] Bouchut F, James F. One-dimensional transport equations with

discontinuous coeﬃcients. Nonlinear Anal TMA 1998;32:891–933.

[8] Brufau Pilar. Simulacio´ n bidimensional de ﬂujos hidrodina´micos

transitorios en geometrı ´as irregulares. Thesis doctoral. Universidad

de Zaragoza; 2000.

[9] Castro MJ, Pare´s JMacı ´asyC. A Q-scheme for a class of systems of

coupled conservation laws with source term. Application to a two-

layer 1-D Shallow Water system. Math Mod Numer Anal

2001;35(1):107–27.

[10] Castro MJ, Gallardo JM, Pare´s C. High order ﬁnite volume schemes

based on reconstruction of states for solving hyperbolic systems with

nonconservative products. Applications to shallow water systems.

Math Comput 2006;75:1103–34.

[11] Cauret JJ, Colombeau JF, LeRoux AY. Discontinuous generalized

solutions of nonlinear nonconservative hyperbolic equations. J Math

Anal Appl 1989;139:552–73.

[12] Chaco´ n T, Domı ´nguez A, Ferna´ndez ED. A family of stable

numerical solvers for Shallow Water equations with source terms.

Comp Meth Appl Mech Eng 2003;192:203–25.

[13] Chaco´ n T, Domı ´nguez A, Ferna´ndez ED. Asymptotically balanced

schemes for non-homogeneous hyperbolic systems – application to

the Shallow Water equations. CR Acad Sci Paris, Ser I 2004;338

8:5–90.

[14] Chien N. The present status of research on sediment transport. Trans

ASCE 1956;121S:833–68.

[15] Colombeau JF, Heibig A. Nonconservative products in bounded

variation functions. SIAM J Math Anal 1992;23(4):941–9.

[16] Dal Maso G, LeFloch PG, Murat F. Deﬁnition and weak stabil-

ity of nonconservative products. J Math Pures Appl 1995;74:483–

548.

[17] Duboys, Meyer-Peter E. Etudes du re´gime du Rhone et de l’action

exerce´ par les eaux sur un lit a´ fond de graviers inde´ﬁniment

aﬀouillable. Ann Ponts et Chausse´es, Ser 1979;5(18):141–95.

[18] Einstein HA. Formulas for the transportation of bed load. Trans

ASCE 1942;107:561–75.

[19] Einstein HA. The bed load function for sediment transport in open

channel ﬂows, Technical Bulletin no. 1026. Washington, DC: U.S.

Department of Agriculture, Soil Conservation Service; 1950.

[20] Fracarollo L, Capart H, Zech Y. A Godunov method for the

computation of erosional Shallow Water transients. Int J Numer

Meth Fluids 2003;41:951–76.

[21] Godlewski E, Raviart PA. Numerical approximation of hyperbolic

systems of conservation laws. New York: Springer-Verlag; 1996.

[22] Gosse L. A well-balanced ﬂux-vector splitting scheme designed for

hyperbolic systems of conservation laws with source terms. Comp

Math Appl 2000;39:135–59.

[23] Gosse L. A well-balanced scheme using non-conservative products

designed for hyperbolic system of conservation laws with source

terms. Math Mod Meth Appl Sci 2001;11:339–65.

[24] Gottlieb S, W Shu C. Total variation diminishing Runge–Kutta

schemes. Math Comput 1998;67:73–85.

[25] Grass AJ. Sediments transport by waves and currents. SERC London

Cent Mar Technol, Report No. FL29; 1981.

[26] Greenberg JM, LeRoux AY. A well balanced scheme for the

numerical processing of source terms in hyperbolic equations. SIAM

J Numer Anal 1996;33:1–16.

[27] Greenberg JM, LeRoux AY, Baraille R, Noussair A. An introduction

to continuum mechanics. Academic Press; 1981.

[28] Hu Changqing, Shu CW. Weighted essentially non-oscillatory

schemes on triangular meshes. J Comput Phys 1999;150:97–127.

[29] Hudson J. Numerical techniques for morphodynamic modelling.

Thesis doctoral. University of Whiteknights; 2001.

[30] Jiang G, Shu CW. Eﬃcient implementation of weighted ENO

schemes. J Comput Phys 1996;126:202–28.

[31] Julien PY. Erosion and sedimentation. Cambridge University Press;

1998.

[32] Kalinske AA. Criteria for determining sand transport by surface

creep and saltation. Trans AGU 1942;23(2):639–43.

[33] Kalinske AA. Movement of sediment as bed load in rivers. Trans

AGU 1947;28(4):615–20.

[34] Kurganov A, Levy D. Central-Upwind schemes for the Saint-Venant

System. Math Mod Numer Anal 2002;36:397–425.

[35] LeFloch PG, Tzavaras AE. Representation and weak limits and

deﬁnition of nonconservative products. SIAM J Math Anal

1999;30:1309–42.

[36] LeVeque R. Numerical methods for conservations laws. Zurich: Bir-

khauser Verlag; 1990.

7.5

2.96e-2

2.98e-2

3e-2

3.02e-2

3.04e-2

3.06e-2

3.08e-2

7.6 7.7 7.8 7.9 8 8.1 8.2 8.3

Fig. 14. Comparison of the stationary solution for the sediment layer at

the interval [7.6 m, 8.5 m]. Experimental data (line with points). Euler–

Roe (line with rhombus) Rk2-Weno2 method (line with squares).

M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316 315

[37] LeVeque R. Balancing source terms and ﬂux gradients in high-

resolution Godunov methods: the quasi-steady wave-propagation

algorithm. J Comp Phys 1998;146:346–65.

[38] Meyer-Peter E, Mu¨ ller R. Formulas for bed-load transport. Report

on 2nd meeting on international association on hydraulic structures

research. Stockholm; 1948. p. 39–64.

[39] Nielsen P. Coastal bottom boundary layers and sediment transport.

Advanced series on ocean engineering, vol. 4. Singapore: World

Scientiﬁc Publishing; 1992.

[40] Pare´s C, Castro MJ. On the well-balanced property of Roe method

for nonconservative hyperbolic systems. Applications to Shallow

Water systems. ESAIM: M2AN 2004;38(5):821–52.

[41] Pare´s C. Numerical methods for nonconservative hyperbolic

systems: a theoretical framework. SIAM J Num Anal 2006;44(1):

300–21.

[42] Gonza´lez Enrique Pen˜ a. Estudio nume´rico y experimental del

transporte de sedimentos en cauces aluviales, Thesis Doctoral.

Universidade da Corun˜ a. Grupo de Ingenierı ´a del agua y del medio

ambiente; 2002.

[43] Perthame B, Simeoni C. A kinetic scheme for the Saint-Venant

system with a source term. Calcolo 2001;38(4):201–31.

[44] Perthame B, Simeoni C. Convergence of the upwind interface source

method for hyperbolic conservation laws. In: Hou T, Tadmor E,

editors. Proceedings of Hyp 2002. Springer; 2003.

[45] Raymond JP. A new deﬁnition of nonconservative products and

weak stability results. Boll Un Mat Ital B(7) 1996;10(3):681–99.

[46] Richardson EV, Simons DB, Julien PY. Highways in the river

environment. Design and training manual for US Department of

Transportation, Federal Highway Administration, publication no.

FHWA-HI-90-016. Washington, DC; 1990.

[47] Roe PL. Upwinding diﬀerence schemes for hyperbolic conservation

laws with source terms. In: Carasso C, Raviart PA, Serre D, editors.

Proceedings of the conference on hyperbolic problems. Springer;

1986. p. 41–51.

[48] Simons DB, Senturk F. Sediment transport technology. Littleton,

Colo: Water Resources Pub; 1977.

[49] Shu CW. Essentially non-oscillatory and weighted essentially non-

oscillatory schemes for hyperbolic conservation laws ICASE Report:

97–65; 1997.

[50] Shu CW, Osher S. Eﬃcient implementation of essentially non-

oscillatory shock capturing schemes. J Comput Phys 1998;77:439–71.

[51] Soulsby R. Dynamics of marine sands. A manual for practical

applications. Thomas Telford Publications, Thomas Telford Services

Ltd; 1997.

[52] Toro EF. Shock-capturing methods for free-surface shallow

ﬂows. Wiley; 2001.

[53] Toumi I. A weak formulation of Roe approximate Riemann Solver.

J Comput Phys 1992;102(2):360–73.

[54] Van Rijn LC. Sediment transport (I): bed load transport. J Hydraul

Div, Proc ASCE 1984;110:1431–56.

[55] Van Rijn LC. Sediment transport (II): suspended load transport.

J Hydraul Div, Proc ASCE 1984;111:1613–41.

[56] Van Rijn LC. Sediment transport (III): bed forms and alluvial

roughness. J Hydraul Div, Proc ASCE 1984;112:1733–54.

[57] Va´zquez-Cendo´ n ME. Estudio de Esquemas Descentrados para su

Aplicacio´ n a las Leyes de Conservacio´ n Hiperbo´ licas con Te´rminos

Fuente. Thesis doctoral. Universidad de Santiago de Compostela;

1994.

[58] Va´zquez-Cendo´ n ME. Improved treatment of source terms in upwind

schemes for the Shallow Water equations in channels with irregular

geometry. J Comput Phys 1999;148:497–526.

[59] Volpert AI. The space BV and quasilinear equations. Math USSR

Sbornik 1967;73(115):225–67.

[60] Yalin MS, Karahan E. Inception of sediment transport. J Hydraul

Div ASCE 1979;105(11):133–43.

[61] Zhou JG, Causon DM, Mingham CG, Ingram DM. The surface

gradient method for the treatment of source terms in the shallow-

water equations. J Comput Phys 2001;168:1–25.

316 M.J. Castro Dı´az et al. / Computers & Fluids 37 (2008) 299–316

On related papers authors study if with the proposed discretization the numerical scheme exactly or approximately preserves a given stationary solution or a family of equilibria.52. For example. using the water surface and the discharge as unknowns. which consists on diﬀerent approximations to the left and to the right. This coupled system can be written as a non-conservative hyperbolic system oW oW þ AðW Þ ¼ 0: ot ox ð1Þ It is well known that standard methods that solve correctly systems of conservation laws can fail in solving systems of balance laws. hydrodynamical and morphodynamical components constitute a coupled system.7. Once the system is rewritten under this form. The phenomena that govern these ﬂows are diﬀerent from those . usually. In [40] the authors propose a general deﬁnition of well-balanced schemes to solve systems like (1). with three unknowns: h(x. which depends on the choice of a family of paths in the phase space.22. the extension to high order numerical schemes becomes a diﬃcult task. this technique has been used to extend Roe method. and. extend HLL method for a system coupling Shallow Water equation with a sediment layer in order to study intense sediment transport.37.13.27. Lax–Wendrof with ﬂux limiters or HLL scheme among other classical schemes.3 a comparison with one of the experiments performed in his work is shown.35. They call this method LHLL. This method is known as centralupwind. which makes diﬃcult even the deﬁnition of weak solutions. Therefore. Castro Dıaz et al. / Computers & Fluids 37 (2008) 299–316 Fig. including numerical schemes using ﬂux limiters. in general.16. which are updated by means of Approximate Riemann Solvers at the intercells.40. Another technique consists on considering balance laws or coupled systems as a particular case of the one-dimensional quasilinear hyperbolic system (1). 26. ´ Pena Gonzalez in [42] proposed a numerical approxima˜ tion of a non-coupled system formed by Shallow Water equations and bedload sediment transport equation. the discharge of the ﬂuid and zb(x. Roe method was studied in [4. except for Grass model. Many papers have been devoted to the deﬁnition and the stability of non-conservative products and its application to the deﬁnition of weak solutions of non-conservative hyperbolic systems: see [1. This deﬁnition.J. They introduce a lateralized discretization of the coupled term. in [16].43. The choice of the family of paths may be. Moreover.12. In [34] Kurganov and Levy extend HLL scheme with state reconstruction for Shallow Water equation with topography as source term.45.5. He also proposed diﬀerent techniques of measurement of the thickness of the sediment layer.47. allows to give a rigorous deﬁnition of weak solutions of (1). the movement of the sediment is controlled by a physical parameter called critical shear stress.9.9]. In this article we assume the deﬁnition of non-conservative products as Borel measures given by Dal Maso et al. specially when approaching equilibria or near to equilibria solutions. In [20] Fraccarollo et al. Types of sediment transport. thickness and discharge of the ﬂuid. this second procedure has been considered in this work. See [2–4. The approximated velocity and thickness of the ﬂuid are used in bedload sediment transport equation to get an approximation of thickness of the sediment layer. a family of numerical schemes. on the speciﬁc considered system. One of these techniques consists on discretizing the conservative terms of the system by using a well-known solver for homogeneous conservative systems and studying a discretization of the source term and/or the non-conservative terms. t). etc. In [12] it was shown that the technique of Modiﬁed Equations can be helpful in this procedure. The introduction of such family of paths also allows to give a general deﬁnition of Roe linearizations for non-conservative systems. The main diﬃculty both from the mathematical and the numerical points of view comes from the presence of non-conservative products. t). Many authors have studied well-balanced numerical schemes for balance laws or coupled systems. piecewise constant approximations of the solutions are considered. was studied in [13]. They also write the system in terms of the water surface. the thickness of the sediment layer. Together with the deﬁnition of weak solutions.6.300 ´ M.58. He applies a ﬁnite volumes scheme based on Roe scheme to obtain the Shallow Water unknowns. In [20] a second order LHLL method for the computation of open channel ﬂows in conditions of rapid bed erosion and intense sediment transport is presented. etc. In most of them. a diﬃcult task. the thickness of the ﬂuid. as the usual Lax’s concept or one related to an entropy pair. a notion of entropy has to be chosen. In Section 6. Following [41]. q(x. in general. Hudson in [29] proposed several numerical approximations of a system constituted by Shallow Water equations and bedload sediment transport equation using Grass model. The extension to high order numerical schemes is in this case straightforward. The form in which the discretization of the source term and/or the coupled term is discretized depends. The classical theory of simple waves of hyperbolic systems of conservation laws and the results concerning the solutions of Riemann problems can be then extended to systems (1). usually determined experimentally.44.23. they can produce unstable methods when they are applied to coupled systems of conservation or balance laws.15.61]. t). 1.11.59].57. He considers diﬀerent reformulations of the problem.

in an approximated way. 2. In Section 2. In Section 2. 2a). low order numerical schemes are very diﬀusive (see Section 6. We rewrite it as a non-conservative hyperbolic system. g is the gravity constant and H(x) is the bathymetry function respect to a ﬁxed reference level (AR) (see Fig.2. zb and a ﬁxed bottom (without sediment).49]). • The morphodynamical component is described with a transport equation. concretely we have used WENO reconstructions (see [28. 2b): 8 > oh oq > þ ¼ 0. Shallow Water system and the sediment transport equation form a coupled system that is described in Section 2.6. t) and h(x. q) we denote the solid transport discharge. we have the equations 8 > oh oq > þ ¼ 0. for Manning law we have gg2 juju . deterministic methods have been proposed by Duboy . coastal areas. ¨ In Section 2. at least bigger than two to obtaining precise results. (see [52. as characteristic velocities of ﬂuid and sediment are very diﬀerent. The study of the deﬁnition of the solid transport discharge can be seen as a deterministic problem or a probabilistic one. where the interaction between the ﬂuid and the sediment is very weak. tÞ ¼ qðx. Rh is the hydraulic ratio.1. Castro Dıaz et al. we also compare with an analytical solution for Grass model and ﬁnally we compare with experimental data. which can be approximated by h and u is the velocity of the ﬂuid ðuðx. and how we treat the problem of transition cells: transition with–without sediment and transition with–without movement. This paper is organized as follows: In Section 2 we present the system of equations for diﬀerent morphodynamical models. < ot ox ð2Þ 2 o q2 1 2 dH > oq > À ghS f . which simulates the movement of a ﬂuid in rivers. In this case the system of equation (2) can be written as follows (see Fig. By q(x. If the bottom is ﬁxed.5) and long time simulations are needed. To solve the problem of sediment transport by a ﬂuid. Hydrodynamical model: Shallow Water equations We consider a one-dimensional channel with variable bottom and constant rectangular section. In Section 2. lakes or coast.´ M. zf = ÀH + AR (zf + H = AR). 2. which is deﬁned by the solid transport discharge.3. The term Sf models the friction term. For example.4 we present a uniﬁed formulation of the presented models. By qb = qb(h. < ot ox ð4Þ 2 1 2 ozb dH > oq o q > þ À ghS f : þ gh ¼ Àgh þ gh : ot ox h 2 dx ox 2.J. t) we denote the discharge and the height of the water column respectively.2 we present some experimental formulae that can be considered. In this paper we consider numerical schemes based on ﬂux limiters and numerical schemes based on state reconstructions for non-conservative hyperbolic systems (see [10]). hðx.30. it is necessary to use high order numerical schemes. channels. So. studying the type of interaction between the ﬂuid and the sediment. • For the hydrodynamical component we consider Shallow Water equations. ð3Þ Sf ¼ 4=3 Rh where g is the Manning’s coeﬃcient. For this reason.9]).tÞÞ. / Computers & Fluids 37 (2008) 299–316 301 presented in bedload sediment transport. In Section 4 we present high order ﬁnite-volume methods using state reconstructions. One of the main diﬃculty to obtain a precise numerical approximation for bedload sediment transport models is that the interaction between ﬂuid and sediment is usually very weak (see Section 2. a coupled mathematical model can be used with a hydrodynamical component and a morphodynamical component. ot ox ð5Þ where n = 1/(1 À q0) and q0 is the porosity of the sediment layer. þ þ gh ¼ gh : ox h 2 dx ot where by x we denote the horizontal variable at the axis of the channel and t is the time variable. The equation that describes sediment transport is a continuity equation and the expression of the conservation sediment volume equation is given by ozb oq þ n b ¼ 0.5 we perform a comparison with Grass model when the critical shear stress is set to zero. In Section 5 we present Roe matrix for the diﬀerent considered models.tÞ To study the problem of sediment transport it is necessary to consider a sediment layer. we observe numerically the order of the diﬀerent numerical schemes. Van Rijn or Grass.1). Finally in Section 6 we present three numerical tests. redeﬁning the solid transport discharge.6 we present the coupled system of equations. We introduce a new variable that allows us to simplify the system. such as Meyer-Peter & Muller. Sediment transport models in Shallow Water equations In this section we present a system of equations that models sediment transport in areas such us rivers. etc. Solid transport discharge In this section we present the most popular formulae to deﬁne the solid transport discharge. qb. Morphodynamical model 2. In Section 3 we present ﬁnite volume methods for non-conservative hyperbolic systems with and without ﬂux limiters.

H zf (b) A fixed bottom and a sediment layar. Possible conﬁgurations of the bottom of the channel in function of the presence or not of a sediment layer. Manning’s coeﬃcient g over the sediment layer is not equal to Manning’s coeﬃcient over the ﬁxed bottom. Concretely. When it is closed to 0 the model reﬂects a weak interaction between the sediment and the ﬂuid. Nevertheless. / Computers & Fluids 37 (2008) 299–316 AR h H AR h zb zf (a) without sediment. ´ • Fernandez Luque & Van Beek equation (FL & V): This formula is very similar to (10). only one coeﬃcient varies. Shields proposed the wellknown Shields-diagram. The usual value of the exponent mg is set to mg = 3.19]. and d denotes the sediment grain size (diameter).J. the shear stress is usually presented in a non-dimensional form in qb. using the friction term s ¼ cRh jS f j. h and q. Castro Dıaz et al. They were proposed for granular and non-cohesive sediments. ðcs À cÞd i sÃc ¼ s . 1 6 mg 6 4. If we denote by sÃ and sÃc the non-dimensional shear stress and the critical shear stress. ð6Þ where c denotes the speciﬁc weight of the ﬂuid c = gq. Chien [14] proved that original formula can be reduced to the following expression: qb qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ¼ sgnðuÞ8ðsÃ À sÃc Þ3=2 . then sÃ ¼ s . . A more complex formula for non-rectangular cross-section channels can be found in [46. Usually. being qs the sediment density. except Grass model. He did the measurement for bigger diameters and after extrapolate.33] and Einstein [18. Later. They were obtained for stationary ﬂux in rivers. and ¨ probabilistic methods by Kalinske [32.48]. have into account that the movement of the granular sediment begins when the shear stress (s) is bigger than a certain critical shear stress (sc). because the time response of the sediment is very small in comparison with the period of the tide or the wave. being q the water density and Sf is deﬁned by (3). respectively. by the eﬀects of the waves and the currents. Meyer-Peter and Muller [38]. In what follows.302 ´ M. Nevertheless.047. 2. and Meyer-Peter [17]. This expression is one of the most used and it is applied to rivers and channels with slope lower than 2%. ðcs À cÞd i ð7Þ Some of the most usual formulae for rivers are: • Grass equation: Grass [25] proposed the following formula for the solid transport discharge: q qmg À1 . so the sediment movement begins at the same instant that beginning ﬂuid motion. it can also be applied to ﬂuids in tides. sÃ can be written in function of the relative density of ﬂuids G = qs/q. Fig. sÃ ¼ gg2 u2 ðG À 1Þd i Rh 1=3 : To determine sÃc many experiments have been performed in diﬀerent works. we present some formulae for qb corresponding to deterministic models. Using (6) and (7). ð10Þ ðG À 1Þgd 3 i where sÃc usually is set to 0. obtaining q b ¼ Ag u 3 : ð9Þ The coeﬃcient Ag takes values between 0 and 1. qb qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ¼ sgnðuÞ5:7ðsÃ À sÃc Þ3=2 : ð11Þ ðG À 1Þgd 3 i • Van Rijn’s equation: Van Rijn [54–56] developed the following formula for the solid transport discharge: where cs denotes the speciﬁc sediment weight cs = gqs. based on median grain diameter d50 (see [51]). These formulae do not include any gradient pressure eﬀect to model grain falling when the velocity of the ﬂuid is zero. Note that one of the main characteristics of this model is that the critical shear stress is set to zero. shear stress can be written in terms of the hydrodynamical unknowns. ð8Þ q b ¼ Ag h h where the constant Ag (s2/m) is usually obtained by experimental data and takes into account the grain diameter and the cinematic viscosity. All models considered in this work. • Meyer-Peter & Muller’s equation: ¨ Meyer-Peter & Muller [38] (MP & M in what follows) ¨ developed one of the most known formulae for the solid transport discharge. Diﬀerent estimations of the solid transport discharge qb have been obtained by empirical methods. Usually. and in fact.

this is similar for all considered models.J. m ¼ mg À 1: c1 ¼ 1.163212495 · 10À4. m ¼ 1:5: h Rh 2.0196.05. where mg 3 3 3. > > ot ox > > > < oq o q2 1 ozb dH 2 ð15Þ þ À ghS f .623065897 · 10À3. all models considered here correspond to a weak interaction between ﬂuid and sediment. • Grass model corresponds to c1 ¼ Ag . ¨ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ cg2 c1 ¼ 8 ðG À 1Þgd 3 . (15) could ot ot be rewritten as Table 1 Comparison with Grass model Formula MP & M Nielsen Van Rijn FL & V Ag Ag ¼ Ag ¼ Ag ¼ Ag ¼ 8g2 cg3 1=2 ðcs ÀcÞRh 12g2 cg3 1=2 ðcs ÀcÞRh 0:005g g 1:7=3 ðGÀ1Þ1:2 h0:2 Rh C 1:7 D 5:7g2 cg3 1=2 ðcs ÀcÞRh 2:2 3:4 ð14Þ where m. we obtain Ag % 8. g = 0. this is one of the main problems in order to obtain precise numerical approximations. • Meyer-Peter & Muller’s model corresponds to.25 6 G 6 4. • Nielsen formula: Nielsen [39] developed the following formula: pﬃﬃﬃﬃ qb qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ¼ sgnðuÞ12 sÃ ðsÃ À sÃc Þ: ð13Þ ðG À 1Þgd 3 i In this case the usual value of sÃc is set equal to sÃc = 0. qÞðc2 þ c3 g1 ðh. c2 and c3 are constant values and g1. ox ox where Sf is the friction term modeling the drag eﬀects between the ﬂuid and the bed deﬁned by (3). / Computers & Fluids 37 (2008) 299–316 303 0:2 qb 0:005 d i 1=2 1=2 qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ¼ 1:7 sÃ ðsÃ À s1=2 Þ2:4 . We consider the following values c = g Æ 1000. The solid transport discharge in this case can be written under the structure (9). cs = g Æ 2600. þ gh ¼ Àgh þ gh > ot ox h 2 dx ox > > > > ozb o > : þ n qb ðh. qÞ ¼ juj.02 and 1. as oS ¼ À ozb . the speciﬁc water weight and Manning’s coeﬃcient. for example). we obtain Ag % 3:623065897 Â 10À3 Rh 1=2 : If the hydraulic ratio Rh is approximated by h. Consequently. As the interaction between the ﬂuid and the sediment is very weak. for diﬀerent deﬁnitions of Ag and mg. c2 ¼ À0:047. they are usually very diﬀusive (see test in Section 6. then it is possible to interpret them as particular cases of Grass model. qÞÞ . For values of h % 1 cm we obtain Ag % 3. All these formulae have a range of application in function of grain size.4 6 d 6 29 mm. g2 ðq. g1 ðh. As shown in Table 1. c3 ¼ . for sÃc = 0. the velocity of the ﬂuid is much greater than the velocity of the sediment. it is necessary to use high order numerical schemes in order to avoid this phenomena. We analyze in detail the MP & M’s model for s*c = 0. we observe that MP & M always models a weak interaction between ﬂuid and sediment.4. low order numerical methods do not capture well the sediment evolution. The expression of this system is as follows: 8 > oh þ oq ¼ 0. qÞ ¼ 0. we can look for it in the diﬀerent models. Coupled model The system of equations used in this work is obtained by coupling Shallow Water equation (4) and the sediment transport equation (5). With this values. In any case. i ðcs À cÞd i q2 g1 ðq. 2.3 we have used MP & M model in order to compare numerical results with experimental data. 2. m We observe that Ag cannot be constant because it depends on the hydraulic ratio. As the value of Ag represents the type of interaction between ﬂuid and sediment. In fact. Ag depends also on the speciﬁc sediment weight.´ M. the Froude number and the relative density G. c2 ¼ 0. slopes of the bottom smaller than 0. For example. Castro Dıaz et al. In Section 6. In fact. c1. qÞ ¼ u. and if h % 10cm. So. g2 ðh. qÞ ¼ c1 g2 ðh. hÞ ¼ sgnðuÞ. hÞ ¼ 2 1=3 . For example in the case of the MP & M it is applied for a 0. Note that. the slope of the bottom.6.1). Nevertheless. Ãc h 3 CD ðG À 1Þgd i ð12Þ Ag ¼ 8g2 cg3 ðcs À cÞRh 1=2 : where CD is the dragging coeﬃcient.4 3 Approximation 3:62306 Â 10À3 5:43459 Â 10À3 0:6751 Â 10À6 1 1=2 Rh 1 1=2 Rh 1 1:7=3 C 1:7 Rh D 1 1=2 Rh 2:58143 Â 10À3 . if the variable S = H À zb is deﬁned. Comparison of the diﬀerent models with Grass model We observe that if we set sÃc = 0 at the previous models.5.2 (see [42]). Another formula based on a complex theory mixing empirical results and fundamental physics was proposed by Van Rijn (see [51]. g2 are scalar functions of h and q. Uniﬁed formulation for solid transport discharge formulae The diﬀerent formulae of qb shown in the previous subsection can be written under a same structure as follows: qb ðh. the MP & M’s model does not depend on grain diameter di.

in general. The associated eigenvectors are 2 3 1 6 7 kk 7. can be deﬁned in a rigorous way the non-conservative product. h a3 ¼ Àghn q2 oqb a2 ¼ 2 À gh 1 þ n oq h oqb : oh and ð23Þ 1. it can be proven that all eigenvalues are real. The eigenvalues are real if 1 The roots of this polynomial are pﬃﬃﬃﬃﬃﬃﬃﬃ k1 ¼ 2 ÀQ cosðh=3Þ À a1 =3. system (16) can be written as an hyperbolic system with a non-conservative term and a source term SF oW oF ðW Þ oW þ ¼ BðW Þ þ SF . which is denoted by ½AðW ÞW x U and weak solutions of (19) are the functions satisfying the equality W t þ ½AðW ÞW x U ¼ S F : The choice of the family of paths is very important. h ¼ arccosðR= ÀQ3 Þ and R ¼ 1 ð9a1 a2 À 27a3 þ 2a3 Þ=54. by the deﬁnitions of A(W) and B(W) we obtain that 2 3 0 1 0 6 q2 2q Àgh 7: AðW Þ ¼ 4 À h2 þ gh ð22Þ 5 h Àn oqb oh Àn oqb oq 0 The eigenvalues of the matrix AðW Þ can be obtained as roots of the characteristic polynomial P ðkÞ ¼ k3 þ a1 k2 þ a2 k þ a3 . WR) = WR. there exists a constant k such that oU ðs. 2 S Ànqb 2 3 0 S F ¼ 4 ÀghS f 5: 0 0 BðW Þ ¼ 4 0 0 2 ð17Þ 3 0 0 0 gh 5. and A(W) is the Jacobian matrix of F(W). W 1 2 O and for almost all s 2 [0. there exists a constant K such that. Although in general we cannot prove it for the other considered models. 2. 1]. 3: Rj ¼ 6 4 q 5 2 ðhÀkk Þ À1 gh We note that for a ﬁxed bottom (qb = 0) the eigenvalues are q pﬃﬃﬃﬃﬃ q pﬃﬃﬃﬃﬃ þ gh. W L . ð19Þ ð21Þ Then. W 2 Þ L R L R os os 6 KðjW 1 À W 2 j þ jW 1 À W 2 jÞ.J. 3. for all WL. W R Þ 6 kjW L À W R j os for all WL. WR 2 X. because it determines the propagation speed of shocks. For a given bounded set O & X. given a funcN tion W 2 ðL1 ðR Â Rþ Þ \ BV ðR Â Rþ ÞÞ it is possible to give a sense to the non-conservative product as a Borel measure (see [40]). introducing a family of paths in X. / Computers & Fluids 37 (2008) 299–316 8 > oh oq > þ > > ot ox ¼ 0. For a bounded set O & X. system (17) can be written as a non-conservative hyperbolic system oW oW þ AðW Þ ¼ Sf . L L R R for all W 1 . pﬃﬃﬃﬃﬃﬃﬃﬃ k3 ¼ 2 ÀQ cosððh þ 4pÞ=3Þ À a1 =3. we must remark that the non-conservative product AðW Þ Á W x . W 2 . F ¼ 4 qh þ 1 gh2 5. Nevertheless. W 2 . [16]. W 1 . À gh and 0: h h By simplicity we will suppose that the system is strictly hyperbolic. pﬃﬃﬃﬃﬃﬃﬃﬃ k2 ¼ 2 ÀQ cosððh þ 2pÞ=3Þ À a1 =3. WR) = WL and U(1. j ¼ 1. W 1 . ot ox ox where 2 3 2 3 q h 2 W ¼ 4 q 5. Castro Dıaz et al. . A family of paths in X & RN is a localLipschitz application U : ½0. L R L R Once a family of paths is chosen. 2. U(0. Deﬁnition 1. if i 5 j and all of them are real. W 1 Þ À oU ðs. oU ðs. Following the theory developed by Dal Maso. W R 2 O and for almost all s 2 [0. where q a1 ¼ À2 . Moreover. 1 Â X Â X ! X that satisﬁes the following properties: Finally. In the case of Grass model. et al. it is very diﬃcult to parametrice the pﬃﬃﬃﬃﬃﬃﬃﬃﬃ where Q ¼ ð3a2 À a2 Þ=9. we have not found any complex eigenvalue in our numerical tests. > > < oq o q2 1 oS þ þ gh2 ¼ gh À ghS f .WL. 1].WL. U in X. > ot ox h 2 ox > > > oS > > À n oqb ¼ 0: : ot ox ð16Þ Note that. has not sense as a distribution when W presents a discontinuity. 0 0 ð18Þ Q3 + R2 < 0. 2 3 0 1 0 2 oF 6 q 2q 0 7: AðW Þ ¼ ðW Þ ¼ 4 À h2 þ gh 5 h oW oqb oqb Àn oq 0 Àn oh ð20Þ x 2 R. t > 0. where ki 5 kj. ot ox where AðW Þ ¼ AðW Þ À BðW Þ.304 ´ M.

iþ1=2 ðW iþ1 À W i ÞÞ. W Þ ¼ AðW Þ. is the base of the associated eigenvectors. . xiþ1=2 . with eigenvalues ð27Þ (b) Method with ﬂux limiters: the idea is to use a combination of Roe method (which is only of order one) near discontinuities. 3. WR 2 X. Aþ . if it satisﬁes the following properties: 1. Castro Dıaz et al. For a given family of paths W. a function A : X Â X ! MN ðRÞ is a Roe linearization of system (25).iÀ1=2 ðW i À W iÀ1 Þ Dx n n þ AÀ D. ð24Þ which corresponds to the non-conservative product deﬁnition proposed by Volpert [59]. ðkk iþ1=2 Æ iþ1=2 N fRl gl¼1 Þ ¼ kk iþ1=2 iþ1=2 j .J. Let Dt be the time step and tn = nDt. W R Þ ¼ W L þ sðW R À W L Þ. Deﬁnition 2. the simplest choice is given by a family of segments. For each WL. W n . LÀ . In [53] the following deﬁnition is introduced. LÆ ¼ 6 0 0 ðk2 Þ iþ1=2 4 5 iþ1=2 Æ 0 0 ðk3 Þ Æ jkk 2 À1 Æ ¼ Kiþ1=2 Liþ1=2 Kiþ1=2 . WR 2 X. W R Þ has N real and different eigenvalues. which are of second order in regular areas. and By Kiþ1=2 we denote the N · N matrix whose columns are eigenvectors and by Liþ1=2 . In the case of a non-conservative system it is also necessary to deﬁne a generalization of Roe matrix. ð29Þ the approximations at time t formula: W nþ1 ¼ W n À i i n+1 can be obtained by the Dt n n ðAþ D. a Linear Riemann problem is considered at each intercell x = xi+1/2 with the associated Roe matrix Aiþ1=2 and the constant states W n and W inþ1 . W L . are obtained by averagi ing in the cells the solutions of the Riemann problem. have been calcui lated. W R Þ ds: A½Wðs. N : ð34Þ . Then we denote by W n an approximation on the mean i value of W over Ii at time t = tn. AW ðW L . AW ðW . AW ðW L . W n Þ i iþ1 ð28Þ the Roe matrix associated to the states Wi and Wi+1. W R Þ ¼ os 0 jAiþ1=2 j ¼ Aþ À AÀ : iþ1=2 iþ1=2 The numerical scheme progresses in time as follows: once the approximations at time tn.L] in cells deﬁned by I i ¼ ½xiÀ1=2 . Uðs. As in the case of a system of conservation laws. The definition of Roe matrix is also associated to the choice of a family of paths. so. ot ox x 2 R. We will also use the following matrices þ Lþ . W R ÞðW R À W L Þ ¼ F ðW R Þ À F ðW L Þ: We will use the following notation: we denote Aiþ1=2 ¼ AW ðW n . W nþ1 . W L . . (26) is independent from the family of paths and it is reduced to the usual Roe property AW ðW L . For each WL. the diagonal matrix of eigenvalues. for all W 2 X. tn Þ dx: Wn ﬃ i Dx xiÀ1=2 To approximate W n we consider two diﬀerent types of i schemes. that must be related to the physics of the problem. . Lu ¼ Diag ÈÀ ð33Þ Á É Dt sgnðkj Þ À Dx kj kj uðrj Þ. / Computers & Fluids 37 (2008) 299–316 305 optimal choice. For the sake of simplicity. i 2 Z. 1. some calculations allow to show that. Aiþ1=2 Æ Diþ1=2 : 2 ð30Þ where AÆ D. Finite volume method for non-conservative hyperbolic systems Friction term SF will be discretized in an semi-implicit way (see [8]). xiÀ1=2 þ k1 iÀ1=2 Dt 6 xi 6 xiþ1=2 þ kN iþ1=2 Dt. respectively. W R ÞðW R À W L Þ Z 1 oW ðs. under the hypothesis. at least for linear problems. j ¼ 1. s 2 ½0. at least for linear systems) in regular areas. the generalized Roe method for non-conservative hyperbolic systems and schemes with ﬂux limiters. . 3. xi = (i À 1/2)Dx is the center of the control 2 volume Ii. 2. Z xiþ1=2 1 W ðx. AÀ iþ1=2 iþ1=2 iþ1=2 iþ1=2 and jAiþ1=2 j: 2 3 iþ1=2 Æ 0 0 ðk1 Þ 6 7 iþ1=2 Æ 7. t > 0: ð25Þ k1 iþ1=2 < k2 iþ1=2 < Á Á Á < kN iþ1=2 . iþ1=2 with. in what follows we consider the system: oW oW þ AðW Þ ¼ 0.´ M. and the generalization of Lax– Wendroﬀ method (which is second order in space and time. The i approximations at time tn+1. W L . The obtained scheme corresponds to deﬁne Diþ1=2 ¼ jAiþ1=2 j À Kiþ1=2 ðLu Þiþ1=2 KÀ1 . we suppose that all of them have the same length Dx and xiþ1 ¼ iDx.iþ1=2 ¼ ð31Þ Diþ1=2 is the viscosity matrix of the numerical scheme deﬁned by (a) Roe method: Diþ1=2 ¼ jAiþ1=2 j: ð32Þ ð26Þ We observe that if AðW Þ is the Jacobian matrix of a regular ﬂux F(W). In practice. AÆ iþ1=2 and We consider a partition of the interval [0.

in [10] the following theorem is proven. to a given sequence {Wi(t)}. minð1. The consequences are that the order of iÆ1=2 the numerical scheme depends upon the order of the state reconstruction operator P ti and its derivative. 8x 2 I i : dx 8x 2 I i . whenever Z 1 W i ðtÞ ¼ W ðx. aj. Then. Moreover it can be proven that scheme (30) is well balanced.i i0 ¼ i À sgnðkj Þ: ð35Þ Some classical examples of ﬂux limiters function are (see [21. at each instant t > 0. 8i 2 Z Dx I i for some regular function W. that is. ﬁrstly.36]): – Van Leer ﬂux limiter: jrj þ r uðrÞ ¼ : 1 þ jrj – The family of ﬂux limiters functions deﬁned by uðrÞ ¼ maxð0. to ensure (29) we use the CFL condition maxfjkl iþ1=2 val Ii. then only the states at the boundary of Ii (xiÀ1/2 and xi+1/2) are necessary. . we must integrate AðW ÞW x over all the interval. In practice. P i ðx.r + 1) for system (25) in the following sense: . Castro Dıaz et al.i rj ¼ 0 . . minðb. it can be rewritten as W 0i ¼ Dt e Dt e ð G iÀ1=2 À G iþ1=2 Þ þ ðBiÀ1=2 Á ðW þ À W À Þ iÀ1=2 iÀ1=2 Dx 2Dx Dt IB. 8i 2 Z: p For conservative hyperbolic systems it is enough with these reconstructions. . ¼ W ðxiþ1=2 . . Applied to our system (17) with a conservative part and a non-conservative product. q + 1. Taking into account that the system (17) has a conservative term and a non-conservative product. we deﬁne a regular function P ti such that x!xþ iÀ1=2 lim P ti ðxÞ ¼ W þ ðtÞ. ð40Þ þ Biþ1=2 Á ðW þ À W À ÞÞ þ iþ1=2 iþ1=2 Dx j. d P i ðxÞ ¼ W 0 þ OðDxr Þ. dx xiÀ1=2 ð39Þ where P ti is a regular function that veriﬁes (38). iÀ1=2 x!xÀ iþ1=2 lim P ti ðxÞ ¼ W À ðtÞ: iþ1=2 ð38Þ In [10] the following numerical scheme is proposed: W 0i ¼ À 1 Aþ ðW þ ðtÞ À W À ðtÞÞ iÀ1=2 iÀ1=2 iÀ1=2 Dx þ AÀ ðW þ ðtÞ À W À ðtÞÞ iþ1=2 iþ1=2 iþ1=2 ! Z xiþ1=2 d t t þ A½P i ðxÞ P i ðxÞdx . nevertheless for non-conservative terms it is necessary to know the state reconstruction over the inter- Then (39) provides an approximation of at least order a = min(p. 1 6 l 6 N . tÞ þ OðDx Þ. b = 2 corresponds to superbee limiter. We suppose that the state reconstruction operator of order p veriﬁes that for a sequence deﬁned by Z 1 Wi ¼ W ðxÞ dx Dx I i for some regular function W.306 ´ M. That happens because if AðW ÞW x has an associated ﬂux and we integrate it over the control volume. A possibility is to use state reconstructions. However if an associated ﬂux does not exist. W iþr Þ ¼ W ðxÞ þ OðDxq Þ. High order ﬁnite volume scheme based on state reconstructions There exist diﬀerent techniques to obtain high order methods. tÞ dx. If A is order C2 with bounded derivatives and AW is bounded. the numerical scheme could be rewritten as follows: Á Dt À Dt GiÀ1=2 À Giþ1=2 þ ðBiÀ1=2 ðW n À W n Þ W nþ1 ¼ W n þ i i i iÀ1 Dx 2Dx n n ð36Þ þ Biþ1=2 ðW iþ1 À W i ÞÞ where 1 1 Giþ1=2 ¼ ðF ðW n Þ þ F ðW n ÞÞ À Diþ1=2 ðW n À W n Þ: i iþ1 iþ1 i 2 2 ð37Þ 4.i ¼ dx xiÀ1=2 ð41Þ ð42Þ Eﬀectively. Theorem 1. rÞÞ: If b = 1 we have the minmod limiter. over each control volume Ii. Remark 1. then W Æ iþ1=2 ðtÞ where 1 e G iÀ1=2 ¼ ðF ðW À Þ þ F ðW þ ÞÞ iþ1=2 iþ1=2 2 1 À jAiþ1=2 j Á ðW þ À W À Þ: iþ1=2 iþ1=2 2 Aiþ1=2 is the Roe matrix and Z xiþ1=2 e d B½ P ti P ti ðxÞ dx: IB. W iÀl . an operator that associates.i ¼ ½KÀ1 ðW iþ1 À W i Þj with aj. / Computers & Fluids 37 (2008) 299–316 The function u is the ﬂux limiter and aj. i 2 Zg Dt 6c Dx with 0 < c 6 1. Concretely. brÞ.J. fW þ ðtÞg in such iþ1=2 iþ1=2 a way that. Firstly a state reconstruction operator Pt is considered. only depends on W Æ . we see that the main diﬀerence between the conservative and the non-conservative system is that the conservative one is independent of the deﬁnition of P ti . two new sequences fW À ðtÞg.i .

by while for conservative systems the order is p. W R Þ ¼ 4 Àð~Þ þ ð~Þ 2~ ð~Þ 5.30. So. in practice we use the following approximation of Roe matrix: 2 0 1 2~ u Àn oqb ð~. with a sediment layer of thickness zb. we have used the high order TVD Runge–Kutta method proposed in [24. and the volume Ii+1.´ M. In order to avoid instabilities. Roe matrix for Grass model In the particular case where Grass model is considered and the family of paths is deﬁned by segments (24). Roe matrix is (see [29]). scheme (36).2. we have r 6 q 6 p and then. with–without movement of the sediment layer. (37) or scheme (40). we denote by r-WENO method the scheme obtained by this state reconstruction. the movement of the sediment layer starts when shear stress s is bigger than a critical shear stress sc. We observe that as we use an approximated Roe matrix. ~ ¼ g~. the scheme is order r. when we use this type of models. is not always possible to obtain an explicit formula and its implementation is very costly. tÞÞW x ðx. the order of the numerical scheme for non-conservative systems is usually smaller than for conservative one. at interface xi+1/2. (st2) Transition with–without movement of the sediment: if xi+1/2 is the interface between Ii and Ii+1. W R Þ ¼ 6 Àð~Þ þ ð~Þ 4 u Àn oqb ð~. the order of (39) is c = r + 1 for non-conservative systems. Remark 2. In this paper we have used state operator reconstruction deﬁned by WENO reconstructions with stencils of r points (see [28.1. we could ﬁnd some of the following situations: (st1) Transition with–without sediment: if xi+1/2 is the interface between Ii and Ii+1. . and its implementation is very costly. Approximation of Roe matrix ð45Þ ð46Þ xiÀ1=2 for any sufﬁciently smooth solution W. So. Numerical treatment of transitions close to the critical shear stress As we have seen in Section 2.i+1 where the shear stress is smaller than the critical one (si+1 < sc) (Fig. and there is not sediment at the control volume Ii+1 (Fig. c h h u 2 hR þ hL Àpﬃﬃﬃﬃﬃ pﬃﬃﬃﬃﬃÁ mg À1 X k m Àðkþ1Þ ~ ¼ Ag n ﬃ hR þ ﬃﬃﬃﬃﬃhL pﬃﬃﬃﬃ p ðuR Þ ðuL Þ g : d hL hR þ hR hL k¼0 5. so. / Computers & Fluids 37 (2008) 299–316 307 Aþ ðW þ ðtÞ À W À ðtÞÞ þ AÀ ðW þ ðtÞ iÀ1=2 iÀ1=2 iþ1=2 iÀ1=2 iþ1=2 ÀW ¼ Z À iþ1=2 ðtÞÞþ xiþ1=2 Z xiþ1=2 A½P ti ðxÞ xiÀ1=2 d t P ðxÞ dx dx i a ! AðW ðx. So. we must use the schemes that explicitly depend on the physical ﬂux function F(W). with a sediment layer of thickness zb. In Section 5. In the particular case of Grass model it is possible to determine a Roe matrix. h and c corresponding to Shallow Water equations. that is.49]). tÞ dx þ OðDx Þ ð43Þ where pﬃﬃﬃﬃﬃ pﬃﬃﬃﬃﬃ qﬃﬃﬃﬃﬃ hR uR þ hL uL 1 ~ ¼ pﬃﬃﬃﬃﬃ pﬃﬃﬃﬃﬃ . we cannot use directly schemes on function of AÆ . it is necessary to change the numerical scheme so that these situations are numerically well treated. Roe matrix and numerical treatment of transitions close the critical shear stress In this section we study the problem of determining Roe matrices for the coupled model formed by Shallow Water equations and sediment transport deﬁned by the family of solid transport discharge (deﬁned in Section 2. ~ and ~ deﬁned by (46). in practice we use an approximation of Roe matrix. For usual state reconstruction operators.3). So. when the family of paths is deﬁned by segments. ~ u hÞ oq 0 3 6 2 2 c AW ðW L . hÞ u ~ oh 27 ð~Þ 7.3 we present the numerical treatment that we perform in transition interfaces with–without sediment. Castro Dıaz et al. that is.3. For time evolution. For this operator of reconstruction we obtain at the internal points of the control volume only order r and q = r À 1 for their derivatives. Nevertheless the calculus of the Roe matrices is complicated for the other models presented in this work. with a sediment layer with thickness zbi and such that the shear stress is bigger than the critical one (si > sc).3. 3a). c 5 0 ð47Þ with ~. 3b).J. in general. (41). the mean Roe values u h c of u.i such that the shear stress at Ii is bigger than the critical one (si > sc). But for the other models considered in this work. u c u c ð44Þ ~ À~d u~ d 0 In the previous subsection we have presented a Roe matrix corresponding to Grass model. Remark 3. 5. So. 5. ~ ¼ ðhR þ hL Þ. 2 3 0 1 0 6 2 2 27 AW ðW L . 5.50]. For the numerical tests presented in Section 6 we use a 3-point Gauss quadrature formula to approximate integral (42). by previous theorem.

The diﬀusion matrix Diþ1=2 is computed using the modiþ1=2 iþ1=2 iﬁed value 3 k instead of k3 ¼ 0. siþ1 < sc . 3. zb ðx. Possible cases of transition around the critical shear stress. iþ1=2 iþ1=2 2. ( 0:1 þ sin2 pðxÀ300Þ if 300 6 x 6 500. we set 3 k ¼ ki3 . In this case. 5–7 we compare analytical solution (continuous line) and numerical solution obtained with the diﬀerent schemes of Sections 3 and 4. ~b ðx. Under these hypothesis it is possible to obtain the following analytical solution. being t0 the instant in which characteristics cross. we have compared the solution for the case. and ( À300Þ sin2 pðx0200 if 300 6 x0 6 500. q0 6 10 a constant value. for Grass model when interaction constant Ag (8) is smaller than 10À2. (b) With sediment in both cells. Castro Dıaz et al. si+1 and si+1/2 are computed. CFL condition is set to 0. If in the three cases the shear stress is bigger than the critical shear stress we do not perform any change. > > x ¼ x0 þ Ag nmg qmg tAÀðmg þ1Þ > 0 r > > : otherwise: where si is the shear stress at volume Ii and si+1/2 is the shear stress corresponding to the mean value of Roe. the ﬂux and the depth of the sediment is imposed upstream. tÞ ¼ zb ðx. tÞ ¼ zb ðx. corresponding to sediment . 0Þ ¼ 0:1 otherwise: ð51Þ As boundary condition. Aiþ1=2 ¼ Kiþ1=2 Liþ1=2 KÀ1 . The algorithm can be summarized as follows: The values si. siþ1=2 < sc . Let us suppose that in the intercell xi+1/2 we have the following situation (Fig. where x0 is the solution of equation 8 Àðmg þ1Þ > > x ¼ x0 þ Ag nmg qmg t Ar À sin2 pðx0 À300Þ > 0 > 200 > < if 300 6 x0 6 500. g In the same way. In the case of a transition we change the diﬀusion matrix as described before.J. Let k3 be the characteristic velocity corresponding to the sediment transport at the intercell. q ¼ q0 ð50Þ The usual value of mg = 3 is considered. Numerical test In this section we present some numerical tests to validate numerical schemes described in Sections 3 and 4. 6. ð48Þ being Ar a ﬁxed level of reference. iþ1=2 iþ1=2 then k3 ¼ 0. The initial conditions are (see Fig. We have considered an academic test with a known analytical solution. 3): si > sc . it is veriﬁed that ~b ðx. Next. Note that ki3 6¼ 0 as si > sc. 0Þ ¼ 10 À zb ðx. qðx. 0Þ ¼ 10. 0Þ. it is also a solution of the problem. discretized with 250 nodes. To be able to compare numerical solution with analytical solution obtained in [29]. For example. Transport of parabolic sediment layer This ﬁrst test consists on comparing a numerical solution with an asymptotic and analytical solution obtained by Hudson and Sweby in [29]. 3. tÞ ¼ ð49Þ 0 otherwise. where ki3 is the characteristic velocity corresponding to the sediment transport at the cell Ii. As si+1/2 < sc. tÞ. hðx. without considering the behavior of numerical scheme in zones with–without sediment. tÞ þ c0 . tÞ þ 0:1: z We consider a computational domain whose length is L = 1000 m. In this case. τ τ τ in cell τ in cell and Fig. The sediment porosity is set to q0 = 0. the layer sediment ~b is over all computational domain and z ﬂuid is moving slowly with a constant discharge q = q0 6 10. a test of order and an experimental test where we compare with experimental measures.1.308 ´ M. In [29] it is estimated t0 ¼ 238079:124 Â AÀ1 . 4). 6.001 which corresponds to a weak interaction. while free boundary conditions are imposed downstream. This solution is valid to t < t0.8. we propose the following algorithm: 1. z where c0 is a constant value. In Figs. 200 zb ðx. h ¼ Ar À zb ðx. / Computers & Fluids 37 (2008) 299–316 (a) with/without sediment.4 and the constant Ag of Grass formula (8) is set to 0.

4. being Roe scheme the most diffusive (Fig. Initial condition.9995 9. t = 238080 sol exact rk2w2sl roecl 10. (b) Mass-flow fluid. Fig. Fig. (b) Mass-flow.9994 0 0 100 200 300 400 500 600 700 800 900 1000 9.4 0. 7a).0015 t = 238080 10. We can observe that all numerical schemes show a good sediment layer localization.001 10. Fig. 5 it is shown comparison between Roe method and linearized Lax–Wendroﬀ method with ﬂux limiters.001 10. Roe-Flux limiters (dotted line). Moreover. 5a. that is of second order in space 10.9985 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 (a) Sediment layer thickness.0005 10 10 9. 6. Roe-Flux limiters (dash line).9998 0.0005 10.8 9. the scheme that gives the best approximation is high order generalized Roe scheme with Weno state reconstructions of order 3. 5a). it is observed that Roe method is more diﬀusive than method with ﬂux limiters. Weno2-Rk2 (dotted line).J. / Computers & Fluids 37 (2008) 299–316 309 .9985 0 9. In Fig.0015 q roeclf q roeslf t = 238080 q roeclf q roeslf 10. In both methods we use Euler scheme for time discretization. where time approximation is made using Runge–Kutta of order 3 (Fig. that describes layer sediment evolution. Euler–Roe (dash line).9996 0.´ M. that is smaller than the maximal time in which analytical solution (50) is valid.999 9.0002 rk2w2sl roecl 10 1 0. layer evolution at instant t = 238080 s.2 9. 5.9992 0 100 200 300 400 500 600 700 800 900 1000 (a) Sediment layer thickness.9995 9.6 9. Castro Dıaz et al. . In Fig.999 9.

0Þ ¼ 0:1 À 0:01eÀx : 2 The exact solution for this problem is unknown. 10]. 6 it is compared ﬂux limiter scheme with Roe– Weno of order 2. and with the following initial condition (see Fig.092 0 -10 -5 0 5 10 0.8 9.5 0. 2 zb ðx.9992 0 100 200 300 400 500 600 700 800 900 1000 (a) Sediment layer thickness. that is one of the hypothesis used to develop the analytical solution (50). 7.9996 0. Weno2-Rk2 (dotted line). of 5120 volumes. Initial condition. 5b the discharge is presented.1 2 0.094 0.J. therefore we use as reference solution a numerical one obtained with a very ﬁne mesh. 0. 6. 0Þ ¼ 0. (b) Sediment layer thickness Fig. For time evolution it is used Runge-Kutta2 and Runge-Kutta3. . For time evolution it is used Euler for ﬂux limiter scheme. For scheme with Weno2 state reconstructions it is used TVD Runge–Kutta method of order 2.0 0.6 9. Test of order. Higher order scheme approximates better the sediment layer thickness as expected. (b) Mass-flow fluid. Castro Dıaz et al. hðx. / Computers & Fluids 37 (2008) 299–316 t = 238080 sol exact rk2w2sl rk3w3sl t = 238080 1 10 rk2w2sl rk3w3sl 0. Fig.2 0 0 100 200 300 400 500 600 700 800 900 1000 9. 8.0 0. 0Þ ¼ 2 À 0:1eÀx . respectively. Weno3-Rk3 (dash line). Finally in Fig. We observe that it is near 10 in both cases and almost constant in all domain.4 9.3.098 1. 7 schemes with Weno2 and Weno3 state reconstructions are compared. We consider a strong interaction between sediment layer and ﬂuid.9994 0. We consider a one-dimensional channel with ﬂat bed along the interval [À 10.096 1 0. concretely we impose Ag = 0.9998 0.09 -10 -5 0 5 10 (a) Initial condition. In Fig.310 ´ M. Depth (H): dotted line.2. 8): qðx. Sediment layer thickness (zb): continuous line. This test has been made using Grass model (8) for solid discharge. and time for linear problems. It must be observed that approximation of thickness sediment layer is similar in both schemes. because Lax–Wendroﬀ scheme ensures order 2 in space and time. Layer water thickness (h): dashed line. In Fig. Test of order We make the following test to compare the order of the proposed numerical schemes.5 0.

9 and 10 the diﬀerent results are presented.1 1.3. a stationary state is reached and the sand front was placed at the point xs = 12. zb (x. obtained with the schemes: Euler–Roe (described in Section 3).99 0. It must be observed that numerical schemes get in this test the expected order: Euler–Roe order one. except in the transition areas at initial and ﬁnal extremes.0.5) 0.2 Weno2-Rk2 Weno3-Rk3 Exacta 2 1. The experimental test was developed introducing a sand layer in the central part of laboratory channel. creating a tongue of sand. q (x.92e-2 9.98e-2 9.98e-2 9. The numerical simulation has been made using MeyerPeter & Muller’s equation (10) for solid transport ¨ discharge.97 -0. measures were made h (x. 9 m]. so the total material advance was 3.5 m.2 2.92e-2 9. Roe–Weno2–RungeKutta2.9e-2 -3 9.098 1e-1 1e-1 0. until a steady state is reached.6 -2.4 -2.4 2. Roe–Weno3–RungeKutta3 (described in Section 4).9e-2 -2.8 2 2. The channel (Fig.8 -1. Weno2-Rk2 (continuous line with stars).6 1. Canales y Puertos (A Coruna University) over a channel ˜ of 15 m long and 0. Tables 2–4 shown the errors for variables h.96e-2 0.94e-2 0. 9. After approximately 8 h. Fig.5 cm.09 -10 -5 0 5 10 (c) Sediment layer thickness .2 -2 -1. In this particular case.1 1.95 -0.6 1.8 3 0. extracting material from upstream at same time that sand was being transport downstream. for t = 0.5 s. / Computers & Fluids 37 (2008) 299–316 311 In Figs. 11) has a very small slope of 0.96e-2 9. at instant t = 0. q and zb in L1 norm.092 9. Comparison with experimental data In this section we present an experiment where we simulate sediment layer evolution over a soil which is not eroded. being media diameter of the grain equals to 1 mm. Fine mesh (continuous line). Order test.0. .15 m.094 9.1 (b) Mass-flow . Euler-Roe Weno2-Rk2 Exacta 0. 6.0.052%.96 1.8 -2. During the experiment the sand layer was eroded very slowly. The ﬁnal thickness of the sand layer was approximated equal to 3 cm along all length.6 2.2 1.5 m width.94 -10 -5 0 5 10 -10 -5 0 5 10 (a) Water layer thickness .5 s. Roe– Weno2–RungeKutta2 order two and Roe–Weno3–RungeKutta3 order three.´ M.5.01 Euler-Roe Weno2-Rk2 Exacta using a PIV technique and a Scanner 3D technique (in [42] the details of the diﬀerent measurement techniques are explained). Comparison between the solution at t = 0. Euler–Roe (continuous line with dots).J.096 9.5) 0.15 m. and inducing hydrodynamical conditions to erode the sand layer.98 0 1. The numerical stationary solution is compared with the experimental data obtained by the Hydraulic Laboratory of Escuela Superior de Ingenieros de Caminos.94e-2 9. ´ Pen Gonzalez details in his PhD Thesis [42] diﬀerent ˜a techniques and toolkits used to develop this and other experiments.5) 2. with a thickness of 4. Castro Dıaz et al. Sand layer was situated in interval [4.

7441 0.8 3 0.5) 2.0689 0.99 0.8 2 2.0003 L1 order h 0.7515 2.0000 2.6 -2.98e-2 9.01 0.2 1.9066 0.0934 L1 err zb 0.9e-2 -2.8923 1.9481 L1 err zb 0.0303 .5) (b) Mass-flow .6148 · 10 0.2074 1.5433 0.6 2.0995 L1 err q 0.6 1.92e-2 9.8154 0.96 1.0.98e-2 9.0051 0.98 0 1.9521 L1 err q 0.094 9.96e-2 9.6919 1.2726 · 10À3 0.6111 0.0012 L1 order q 0.1070 0.05287 0.0160 0.8630 1.0968 0.6702 1.2 2.97 -0.4050 0.7580 2.0124 0.5513 0.0274 L1 order q 0.85430 0.92e-2 9.5 s.95 -0. 0.09 -10 -5 0 5 10 (c) Sediment layer thickness .4118 0.6971 0.0012 0.4 -2. 10.J.096 9.7043 1.1733 0. / Computers & Fluids 37 (2008) 299–316 h (x.1 1.1 1. Castro Dıaz et al.7966 1.8 -1.092 9.94 -10 -5 0 5 10 -10 -5 0 5 10 (a) Water layer thickness . Fig.3073 · 10À3 0.0113 · 10À3 0.8 -2.8977 0.7335 0. zb (x.1639 · 10À3 0.5338 · 10À3 0.0701 0.96e-2 0.5) 1.0418 · 10À3 0.8370 0.2810 0.2226 0. Weno2-Rk2 (continued line with stars) and Weno3-Rk3 (continued line with diamonds).312 ´ M.94e-2 0. Fine mesh solution (continued line).0234 0.0027 · 10À3 L1 order zb 0.0.0521 0.0418 0.4218 · 10À3 0. Comparison with the ﬁne mesh solution in t = 0. Order test.0.6297 0.1331 · 10À3 0.8824 2.9168 Table 3 Errors and order: Weno2 + Runge Kutta 2 Nodes 20 40 80 160 320 640 L1 err h 0.2 -2 -1.0907 · 10À3 À3 L1 order zb 0.94e-2 9.0047 0.2 2 q (x.098 1e-1 1e-1 0.0025 2.0064 L1 order h 0.6 1. Table 2 Errors and order: generalized Roe scheme Nodes 20 40 80 160 320 640 L1 err h 0.0985 0.1 0.1639 · 10À3 0.0203 0.9e-2 -3 9.4 2.

129 m downstream.5709 · 10À7 4.1220 err L1 q 0.08 Fig.0285 m2/s upstream.0333 3.0064 · 10À3 4. and such that the thickness of the water layer at the right extreme of the channel is equal to the experimental measure. For numerical simulation we have meshed the domain with 250 nodes The CFL is set to 0.2690 0. 0Þ ¼ À H ðxÞ À zb ðx. Castro Dıaz et al.3973 3.´ M.9077 · 10À6 order L1 q 2.0038 0. . which we use as initial condition for the ﬂux.0398 0.0 · 10À5 6.0285 cm2/s. For this reason we cannot expect that the numerical scheme could reproduce accurately the zones where gravity eﬀects are relevant. Sediment porosity is set to 0.5cm -0.0479 0.6307 3. Therefore. Using the shear stress expression and the critical shear stress value. Sediment layer thickness (zb): continuous line. To build an initial condition we have proceeded as follows: a) Initial condition for ﬂux: we know by experimental data that the average thickness of the sediment layer at the steady state is about 3 cm.0009 6. Sketch of the laboratory experiment. because we only know the position and the thickness of the sediment layer zb at the steady state.6751 3. 0Þ. Layer water thickness (h): dashed line. That is. 0 2 4 6 8 10 12 14 16 Fig.0 · 10À6 6.2689 · 10À3 0.1 -0.5334 · 10À5 2. -0.3870 3. while the water thickness is imposed to 0. for this initial condition we observe an ‘anomalous’ behavior of the evolution of the sediment layer.858 3. As boundary conditions we impose an incoming discharge equal to 0. that shows at 10 min. so the sediment does not fall by its own weight due to gravity eﬀects. as we do not know the initial condition and we must impose it using the technique explained before. Depth (H): dotted line.0 0 -0. the discharge q = qc. and the numerical solutions obtained with Roe scheme without state reconstructions (dash line) and scheme with Weno2 state . the downstream sand slope and the median proﬁle of the sediment layer. 40 min and 120 min.0125 over the ﬁxed bed and 0. The model that we are using does not include pressure forces.8264 3. g However. 0Þ.9.00026 2. b) As we have information about the thickness of the water layer at the right extreme of the channel. We obtain the value of qc = 0. 13.02 m 0.0196 over the sediment layer. qc.5m 9m 15m -0. In Fig. comparison between experimental measures (line with dots) obtained by a classic technique (see [42]). To avoid this problem. we can only compare the numerical and experimental solution at steady state.4338 3. and therefore. but it must reproduce at least. a ﬁrst choice for g the initial condition over the thickness of the water column is to consider hðx.0656 · 10À3 0.2298 · 10À8 4. as for example.5 -0. calculating the steady solution for the topography H ðxÞ ¼ H ðxÞ À zb ðx.1382 0.052% 0m 4.J. / Computers & Fluids 37 (2008) 299–316 Table 4 Errors and order: Weno3 + Runge Kutta 3 Nodes 20 40 80 160 320 640 err L1 h 0.12 The ﬁrst problem is to give an admissible initial condition for the laboratory experiment. 12 we show the initial condition used in this numerical simulation.8039 · 10À9 313 order L1 zb 2.04 4. Friction between ﬂuid and bed is modeled using a Manning’s law with coeﬃcient equal to 0.3407 3. 12.1231 err L1 zb 0.9 · 10À7 order L1 h 1. but we have not information about initial conditions. for which solid transport discharge is zero.4.5 cm.0115 0. imposing a constant discharge q = qc along the channel and using as topography the function given by H ðxÞ ¼ H ðxÞ À zb ðx. Moreover the thickness of the water column over the sediment layer is approximately 9. Initial condition imposed in numerical test.06 0.4880 3. the advancing front of the sand layer.8304 3. 0Þ. ﬂuid surface elevation. it is possible to determine the critical discharge value. 11.7912 3. This behavior is reﬂected in Fig. ﬁrstly we calculate the free water surface using Bernoulli’s equation.

t = 40 min NumericaWeno2+Rk2 Numerica Euler Experimental t = 10 min 0.036 0.034 0.04 0.044 0. In Fig.04 NumericaWeno2+Rk2 Numerica Euler Experimental 6 6.038 0.6e-2 4.036 0.5 7 7.01 0.5 7 7.04 0.045 0.5e-2 4.02 0.4e-2 4. Euler–Roe (dashed line).038 0.01 0.04 0.5 (b) = 40 min.05 0. line with squares represents the solution obtained .5 8 8. the numerical solution tends to the steady state. Castro Dıaz et al.J.025 0.035 0.5 8 8.02 0.042 0.034 0. the graphics show that we are approximating the average of the thickness sediment layer. reconstructions (continuous line). Fig.9e-2 4.01 0.035 0.5 (d) = 40 min. Detail of the sediment layar top 0. In this ﬁgure it is possible to observe that the scheme of order one (Roe without state reconstructions) is more diﬀusive than the scheme of order two.314 ´ M.032 0. In the different instants.005 0 0 2 4 6 8 10 12 14 16 0 0 2 4 6 8 10 12 14 16 (a) = 10 min.042 0.7e-2 4. Weno2Rk2 (continuous line).5 (c) = 10 min. 40 min and 120 min.5 8 (f) = 120 min.015 0. 13. 14 we compare the experimental measures of the position of the sediment layer top at the steady state with the numerical solutions obtained using diﬀerent numerical schemes.025 0. Although we do not know the initial condition.2e-2 6 6. Experimental measures (continuous line with dots). / Computers & Fluids 37 (2008) 299–316 t = 10 min fondo NumericaWeno2+Rk2 Numerica Euler Experimental 0.015 0. Sediment layer evolution at 10 min.02 0.045 t = 40 min fondo Numerica Weno2+Rk2 Numerica Euler Experimental 0.03 0. numerically it is proved that we conserve total mass of sediment layer.04 0. Detail of the sediment layar top (e) = 10 min.044 0.03 0.03 0.3e-2 4.03 5 5.03 5. Moreover.5 6 Numerica We no2+Rk2 Numerica Euler Experimental 6. Detail of the sediment layar top t = 120 min fondo Numerica W eno2+Rk2 Numerica Euler Experimental t = 120 min 0.005 0 0 2 4 6 8 10 12 14 16 0.032 0.5 7 7. In this ﬁgure line with dots represents experimental data. 5e-2 4.8e-2 4.

Nevertheless. Nonlinear stability of ﬁnite volume methods for hyperbolic conservation laws and well-balanced schemes for sources.J. Erosion and sedimentation. Heibig A.75:1103–34. [18] Einstein HA.5 m]. Fernandez ED. Nonconservative products in bounded variation functions. A more complex physical model including the gravity eﬀects at the sediment layer must be considered in order to reproduce better the laboratory experiment. Birkhauser. Academic Press.3 Fig. Pares C. Comp Meth Appl Mech Eng 2003. Comput Fluids 2001.6 m. W Shu C.74:483– 548.25(6):2050–65. [21] Godlewski E. Sediments transport by waves and currents.32:891–933. SIAM J Sci Comp 2004. ´ ´ ´ [13] Chacon T. in general (see Fig. A well balanced scheme for the numerical processing of source terms in hyperbolic equations. LeFloch PG.math. Comput Fluids 1994.121S:833–68. 14 it is observed that both numerical solutions converge slowly to the steady state. Math Mod Meth Appl Sci 2001.2 8. 13). Shu CW. SIAM J Math Anal 1992. Application to a twolayer 1-D Shallow Water system. ´ ´ [4] Bermudez A.41:951–76. [30] Jiang G. Numerical approximation of hyperbolic systems of conservation laws. 2004.192:203–25. Bristeau MO. A well-balanced scheme using non-conservative products designed for hyperbolic system of conservation laws with source terms. Total variation diminishing Runge–Kutta schemes. ´ [10] Castro MJ. J Math Anal Appl 1989. Domınguez A. Asymptotically balanced schemes for non-homogeneous hyperbolic systems – application to the Shallow Water equations. The bed load function for sediment transport in open channel ﬂows. In this example. Trans AGU 1947.30:1309–42. Raviart PA. Merlet B. 1996. available at <http://mahery. Washington. . High order ﬁnite volume schemes based on reconstruction of states for solving hyperbolic systems with nonconservative products. while Euler–Roe scheme not. Comparison of the stationary solution for the sediment layer at the interval [7.126:202–28.u-psud. Comp Math Appl 2000.6 7.1 8. Cambridge University Press. [28] Hu Changqing. LeRoux AY.35(1):107–27.5 7.30:643–71. SERC London Cent Mar Technol. [2] Alcrudo F. 14. FL29. 1026. CR Acad Sci Paris. [24] Gottlieb S. Department of Agriculture. Soil Conservation Service. Murat F. 1990. LeRoux AY. A Godunov method for the computation of erosional Shallow Water transients. Math Mod Numer Anal 2002. [20] Fracarollo L. Ann Math 2005. [5] Bianchini S. One-dimensional transport equations with discontinuous coeﬃcients. Gallardo JM. / Computers & Fluids 37 (2008) 299–316 3. Deﬁnition and weak stability of nonconservative products. Klein R.98e-2 2. [33] Kalinske AA. A well-balanced ﬂux-vector splitting scheme designed for hyperbolic systems of conservation laws with source terms. Zech Y.5(18):141–95. Ser I 2004. ´ ´ [8] Brufau Pilar.338 8:5–90. Vazquez ME. 1950. James F. Thesis doctoral.8 7. J Math Pures Appl 1995. 1998. Report No.161(1):223–342. ´ [17] Duboys.39:135–59. Math Mod Numer Anal 2001.7 7. ´ ´ [9] Castro MJ. Ser 1979.96e-2 315 7. Formulas for the transportation of bed load. Fernandez ED. Pares JMacıasyC. Approximate shock curves of non-conservative hyperbolic equations in one space dimension. 2001. J Comput Phys 1999. Criteria for determining sand transport by surface creep and saltation. Applications to shallow water systems. Representation and weak limits and deﬁnition of nonconservative products. The present status of research on sediment transport. [15] Colombeau JF. [22] Gosse L. Euler– Roe (line with rhombus) Rk2-Weno2 method (line with squares). Trans ASCE 1956. Numerical methods for conservations laws.67:73–85.11:339–65. J Comput Phys 1996. Levy D. Experimental data (line with points). Noussair A. ´ ´ ´ [12] Chacon T. [19] Einstein HA. it is diﬃcult to say what numerical scheme reproduce better the laboratory experiment. An introduction to continuum mechanics. Math Comput 1998. [3] Audusse E.06e-2 3.28(4):615–20. [27] Greenberg JM. Simulacion bidimensional de ﬂujos hidrodinamicos ´ transitorios en geometrıas irregulares. Perthame B. Benkhaldoun F. Exact solutions to the Riemann problem of the Shallow Water equations with a bottom step.23(4):941–9. Shu CW. [11] Cauret JJ. where gravity eﬀects are not so relevant. Movement of sediment as bed load in rivers. University of Whiteknights. [16] Dal Maso G. Ann Ponts et Chaussees. so the sediment does not fall by its own weight due to gravity eﬀects. Colombeau JF. A Q-scheme for a class of systems of coupled conservation laws with source term. 1981. mainly because the physical model does not include pressure forces. Trans AGU 1942. Int J Numer Meth Fluids 2003. Etudes du regime du Rhone et de l’action ´ ´ ´ exerce par les eaux sur un lit a fond de graviers indeﬁniment ´ aﬀouillable.23(8):1049–71. 8.33:1–16. Capart H. Preprint. [25] Grass AJ.html>. Vanishing viscosity solutions of nonlinear hyperbolic systems.107:561–75. Central-Upwind schemes for the Saint-Venant System. 1981. [6] Bouchut F. [36] LeVeque R. New York: Springer-Verlag. the numerical scheme Rk2-Weno2 captures better the depth proﬁles of the sediment layer at the initial part of the sediment layer. [32] Kalinske AA. DC: U. Numerical techniques for morphodynamic modelling.08e-2 3.04e-2 3. SIAM J Numer Anal 1996. by scheme with reconstruction Weno2. Zurich: Birkhauser Verlag. Castro Dıaz et al. Nonlinear Anal TMA 1998.150:97–127. and line with diamonds represents the solution got by Roe scheme without state reconstructions. LeRoux AY. Bressan A. Eﬃcient implementation of weighted ENO schemes. In Fig.36:397–425. [29] Hudson J. [35] LeFloch PG. Domınguez A.fr/alouges/full_publi. A family of stable numerical solvers for Shallow Water equations with source terms. Universidad de Zaragoza. References [1] Alouges F.9 8 8. A fast and stable well-balanced scheme with hydrostatic reconstruction for Shallow Water ﬂows.´ M. Trans ASCE 1942. Weighted essentially non-oscillatory schemes on triangular meshes. SIAM J Math Anal 1999. [26] Greenberg JM.02e-2 3e-2 2. ¨ [7] Bouchut F. [34] Kurganov A. Discontinuous generalized solutions of nonlinear nonconservative hyperbolic equations. Baraille R.139:552–73. Technical Bulletin no. [14] Chien N. Bouchut F. we could say that. Tzavaras AE. 2000. Meyer-Peter E. Thesis doctoral.S. Upwind methods for hyperbolic conservation laws with source terms. Math Comput 2006. [23] Gosse L. [31] Julien PY.23(2):639–43.

ESAIM: M2AN 2004. Formulas for bed-load transport. Proc ASCE 1984. Tadmor E. ´ ´ [42] Gonzalez Enrique Pena.38(5):821–52. A manual for practical applications. In: Hou T. Simons DB. Applications to Shallow Water systems. Causon DM. 1977. Ingram DM.112:1733–54. Serre D. J Comput Phys 1998. 1986. Proceedings of the conference on hyperbolic problems.73(115):225–67.111:1613–41. [39] Nielsen P. 1997. Sediment transport (I): bed load transport. Senturk F. Essentially non-oscillatory and weighted essentially nonoscillatory schemes for hyperbolic conservation laws ICASE Report: 97–65. Highways in the river environment. Design and training manual for US Department of Transportation. ´ [40] Pares C. ´ ´ [57] Vazquez-Cendon ME. Balancing source terms and ﬂux gradients in highresolution Godunov methods: the quasi-steady wave-propagation algorithm.110:1431–56. J Hydraul Div.316 ´ M. Proc ASCE 1984. editors. A new deﬁnition of nonconservative products and weak stability results. Improved treatment of source terms in upwind schemes for the Shallow Water equations in channels with irregular geometry. editors. Wiley. J Comp Phys 1998. ´ ´ [58] Vazquez-Cendon ME. 2001. Thomas Telford Services Ltd. Proceedings of Hyp 2002. Singapore: World Scientiﬁc Publishing. Grupo de Ingenierıa del agua y del medio ˜ ambiente.J.10(3):681–99. On the well-balanced property of Roe method for nonconservative hyperbolic systems. J Comput Phys 1999. J Comput Phys 1992. Stockholm. Osher S. vol. Report ¨ on 2nd meeting on international association on hydraulic structures research. Colo: Water Resources Pub. Coastal bottom boundary layers and sediment transport. Eﬃcient implementation of essentially nonoscillatory shock capturing schemes. [51] Soulsby R. Karahan E. [53] Toumi I. SIAM J Num Anal 2006. Math USSR Sbornik 1967. A kinetic scheme for the Saint-Venant system with a source term. Julien PY. Sediment transport (III): bed forms and alluvial roughness. Washington. A weak formulation of Roe approximate Riemann Solver. Springer. ´ Universidade da Coruna. [60] Yalin MS. Estudio de Esquemas Descentrados para su ´ ´ ´ ´ Aplicacion a las Leyes de Conservacion Hiperbolicas con Terminos Fuente.44(1): 300–21. ´ [41] Pares C. 2003.77:439–71. [50] Shu CW. Estudio numerico y experimental del ˜ transporte de sedimentos en cauces aluviales. publication no. Raviart PA. 1994. Calcolo 2001. [44] Perthame B. [43] Perthame B. Littleton. J Hydraul Div. [45] Raymond JP. Advanced series on ocean engineering. Thesis doctoral. Castro MJ. J Hydraul Div ASCE 1979.38(4):201–31. 4. J Comput Phys 2001. The space BV and quasilinear equations. [46] Richardson EV.148:497–526. Universidad de Santiago de Compostela. The surface gradient method for the treatment of source terms in the shallowwater equations. [55] Van Rijn LC. p. Numerical methods for nonconservative hyperbolic systems: a theoretical framework. Proc ASCE 1984. 1997. 2002. FHWA-HI-90-016. Thesis Doctoral. [54] Van Rijn LC. [56] Van Rijn LC. Castro Dıaz et al. 1948. DC. Inception of sediment transport. [38] Meyer-Peter E. Thomas Telford Publications. [59] Volpert AI. / Computers & Fluids 37 (2008) 299–316 [48] Simons DB. Simeoni C. [61] Zhou JG. In: Carasso C. 41–51. 1992. Upwinding diﬀerence schemes for hyperbolic conservation laws with source terms.105(11):133–43. Boll Un Mat Ital B(7) 1996.146:346–65. Federal Highway Administration. Muller R. Dynamics of marine sands. Simeoni C. p. [52] Toro EF. Springer. Convergence of the upwind interface source method for hyperbolic conservation laws. [47] Roe PL. 39–64. Shock-capturing methods for free-surface shallow ﬂows. [37] LeVeque R. Sediment transport technology.168:1–25.102(2):360–73. [49] Shu CW. Mingham CG. 1990. Sediment transport (II): suspended load transport. J Hydraul Div. .

- Jurnal Optimal Control-Wavelat Method
- Jurnal Dynamic Programming-Economic
- Jurnal Kontrol Pada Sistem Inventori-produksi
- Jurnal Control-Mothod of Moment
- Jurnal Kontrol Tuberkulosis
- TPA-Barisan Dan Deret
- 08-Kalkulus 2 Bab 4 Integral Lipat
- Pra Kalkulus 2013.1-V2
- Anderson Chaplain Bmb Angio
- MateriFisikaSMAXSmt2
- vektor revisi kirim1
- Program Pendidikan Untuk Kesehatan
- 01 Intro
- Finite Vol 1
- Format Print Amplop Surat
- salmon.2002
- 6. Fisika Modern
- 0605046
- v6n2-9pd
- Tumor
- Intro
- Capuccio Elishmereni Agur Interleukin 21 2006
- amc18312
- amc1333

Sign up to vote on this title

UsefulNot useful- Logging Sedimentary Outcrops and Cores New 2by Zamzuri Zaidin
- SEDIMENTLOGY OF QUATERNARY DEPOSITS OF NARMADA RIFT VALLEY IN INCREASING ANTIQUITY WITH SPECIAL REFERENCETO HOMONID LOCALITY HATHNORA DISTRICT SEHORE MP INDIA.by IJAR Journal
- HEC-RAS 5.0 Release Notesby sk8_cartucho
- 1999 Dillard K characterization of fluvial deposits interbedded with flood basalt.pdfby Sagnik Basu Roy

- Sedimentation-Scour-Model.pdf
- 04 ST Types of Sediment Transport Student
- Sediment Transport in Open Channel
- Rem-q No. 03-2005 Mid Exam
- Characteristics of Sediment Concentration and Suspended Sediment Transport Due to Horizontal and Vertikal Asymmetric Waves
- zarrati-A-10-111-1-c01596e
- 2017 Lecture 01 Sedimentary Textures
- Oceanography Research Paper
- Silt trap
- Ch04
- Bulletin430_35-46
- 1D Mathematical Modelling of Debris Flow
- 3_texture_2012.pdf
- Sediment Transport Analysis With HEC-RAS 2
- Modern Sediment Accumulation Rates and Sediment Transport in the Monsoonal
- G.C. Sander; P.B. Hairsine; C.W. Rose; D. Cassidy; J.-y. Parlang -- Unsteady Soil Erosion Model
- Tierras Raras Pachuca.pdf
- Gomez 1998
- Leduc & Pilditch 2013 JEMBE
- 10-
- 004 Zarina Md. Ali Malaysia
- G433 (2010) review of sed structures_v2.pdf
- Ge 490601
- Sequence Stratigraphy 3
- Logging Sedimentary Outcrops and Cores New 2
- SEDIMENTLOGY OF QUATERNARY DEPOSITS OF NARMADA RIFT VALLEY IN INCREASING ANTIQUITY WITH SPECIAL REFERENCETO HOMONID LOCALITY HATHNORA DISTRICT SEHORE MP INDIA.
- HEC-RAS 5.0 Release Notes
- 1999 Dillard K characterization of fluvial deposits interbedded with flood basalt.pdf
- Modelling the Effects of Sediment Compaction on Salt Marsh Reconstructions
- Nease Feasibility Study200806
- CFF_2008