Hydrologkal Sciences - Journal - des Sciences Hydrologiques, 32, 1, 3/1987

Dispersion of pollutants in rivers

W. CZERNUSZENKO
Institute of Meteorology and Water Management, ul. Podlesna 61, 01-673 Warsaw, Poland
ABSTRACT The problem under consideration is the mixing of conservative, nonbuoyant pollutants released into a river. Such pollutants are mixed with the water of a river by different mechanisms, i.e. molecular diffusion, turbulent diffusion, advection and dispersion. These mechanisms are explained. One of the most important is dispersion, which is derived and described in the paper. The mixing process in rivers is represented by the advection-dispersion equation. Analysing this process in irregular rivers is usually impossible because of the lateral advection term in the dispersion equation. However, this term disappears in a special curvilinear coordinate system. Such a system is introduced and a method of solution of the dispersion equation is presented. Two examples of the simulation of a mixing process in curvilinear channels are given.

La dispersion

des polluants

dans les

rivires

RESUME Le problme que l'on considre ici est celui du mlange des polluants conservatoires non-flottants, rejets dans une rivire. D'habitude la description de ce processus dans les rivires non amnages est impossible par suite d'un terme d'advection latrale dans l'quation de dispersion. Ce terme disparait dans un systme spcial de coordonnes curvilignes. Ce systme est prsent avec une mthode de solution de l'quation de dispersion propose. On a donn deux exemples de simulation du processus de mlange dans un canal courviligne.

INTRODUCTION
The problem of natural mixing in rivers is commonly treated as two-dimensional. In order to obtain a two-dimensional equation representing the variation of the depth-averaged concentration in both the transverse and longitudinal directions, the three-dimensional advection-diffusion equation is integrated over the water depth. It is generally assumed that Fickian diffusion theory can be applied to the dispersion process in rivers. This gives the two-

*Based on a paper presented at the Anglo-Polish shop, Jablonna, Poland, 17-21 September 1984. Hydrological Sciences Journal, vol.30, no.l, Open for discussion until 1 September 1987.

Hydrological Work(See report in p.165.)

59

60

W. Czemuszenko

dimensional advection-dispersion equation which is usually used for describing the pollution concentration field in rivers. The equation may be applied in the case of conservative and nonbuoyant pollutants which are vertically well-mixed in rivers. Further consideration is limited to the mixing associated with the ambient flow conditions rather than to any initial mixing which might be caused by energy in the discharged pollution.

BASIC EQUATIONS
A description of the mixing of pollutants in a river or open channel is most frequently based on the mass conservation law. The general form of this law in a differential form is: -r- + div q = 0 (1)

where f = the mass density, and q = the mass flux density vector. Mixing in a river sufficiently far from a pollution source will be considered. A pollutant in a river is displaced from a place of higher concentration to a place of lower concentration. In the case of turbulent flow the displacement occurs due to molecular diffusion, turbulent diffusion and advection. These three mass density fluxes are usually expressed in the forms: q = - K grad C -m m q t = - Ktgrad C q =C -a (2) (3) (4)

where : K m = coefficient of molecular diffusion; Kt = coefficient of turbulent diffusion; C = mean concentration at a point (it is also the mass density term in equation (1)); and u = mean velocity at a point. The two-dimensional case of pollution spreading in rivers is now under consideration. The several two-dimensional fluxes can be obtained by averaging quantities defined earlier. Thus in twodimensional approximations: - mass density is: C h = J* C dy (where two overbars means the depth-average operation) - pure advection flux density splits into two fluxes: (5)

u C h = u C+uyCy

(6 .)

Dispersion of pollutants in rivers

61

where: uy(x,y,z) = (x,y,z) - u(x,z) C y ( x , y , z ) = C ( x , y , z ) - C^x.z) - turbulent diffusion flux density. (9) flux density: (10) (7) (8)

h = - K grad C h -t t - molecular diffusion

~q h = - K grad C h ^ -m m

Thus the total two-dimensional mass flux density can be expressed

9total

HCB+l /Cy

3t +

ginjh

(11)

The term in curved brackets is called the dispersion flux density, D (see, e.g. Holley, 1969). According to equation (1) the law of conservation of mass in a two-dimensional approximation gives:

hC + div I uC h + D h I = 0

(12)

In applying Fickian diffusion theory to the dispersion process in turbulent open-channel flow, an exact analogy with the molecular diffusion process is assumed: D
i

= - E. dC/dx.
l l

(i = x, z)

(13)

where Ej = coefficient of dispersion in the i-direction. Equation (12) now takes the form:

d , d , _ 6 d , oC d . _ dC Trrh C + - ^ h u C + - ^ h w C = ^ - hE7r- + hE^rat dx dz dx

x

,, . . . (14)

dx

dz

dz

where : u, w = longitudinal and lateral components of depth-averaged velocity; and C = depth-averaged concentration. In equation (14) all over-bars are neglected for simplicity.

62

W. Czernuszenko

NUMERICAL SOLUTION
The lateral component of velocity, w, occurring in the above equation, greatly affects the mixing process, mainly for unregulated rivers with irregular transverse cross-sections. It is usually impossible to take this velocity into account in computations due to lack of data. Measurements of this component of velocity are not being made, and its calculation is associated with a very difficult problem of solution of the Navier-Stokes equation. These difficulties can be overcome by using an orthogonal curvilinear coordinate system which is composed of mutually orthogonal longitudinal and transverse coordinates. The longitudinal coordinate should align closely with the depth-averaged velocity vector. In this coordinate system w = 0, and the dispersion equation takes the form: 3C \ Sx"')

A + U

1
m m x z

/~ m hE Id z z dC \ 3x m 3x' \ x

3 3z

x h E z 3C m 3z

(15)

where: m x , mz = metric coefficients; and x', z' = longitudinal and lateral curvilinear coordinates. The metric coefficients, m x and m z , are introduced to calculate the distance between two points in the curvilinear coordinate system. Distances along a longitudinal coordinate curve from point A to point B are given by: L^ B = /? m x dx', and along a transverse coordinate curve from C to D by: L C D = I m z d z ' T h e values of m x and m z may vary from point to point and are in general functions of both x and z. For small distances these coefficients are constant, and then: L A Q = m (x'(B) - x'CA)) AB x
L

(16) (17>

CD

= m

(x

'(D)

x*(C))

The above relations are used to calculate m x and m z for a numerical grid. In engineering practice it is reasonable to assume that the depth and the metric coefficient, m z , does not vary very much in the longitudinal direction, nor does the coefficient, m x , in the lateral direction. If so, equation (15) takes the form:

3C 3t

u m
X

3C _ 1_ 3_ !* 9 3x' m 3x' m 3x'

X X

hm

3z' m

z 3c 3z

(18)

The above equation together with the initial condition: C(0,x,z) = C (x,z) o and the boundary condition: (19)

Dispersion of pollutants in rivers 3C/3z = 0 for z = 0 and z = B (20)

63

is solved by using the fractional step method (Yanenko, 1971). numerical solution consists of three steps:

The

Step
3C 3t

1.
u m

Solution
3C =0 dx'

of the pure advection

equation:
(21)

Step
3C 3t

2.
E

Solution
x 3 m 3m x

of the longitudinal

diffusion

equation:
(22)

c m 3x' x

Step
3C 3t

3.

Solution
1 hm
m 3 z 3z' m z z

of the lateral
3C 3z'

diffusion

equation:
(23)

The distribution of concentration obtained from one step is used as the initial concentration in the next step. The advection equation is solved by the two-point higher order method developed by Holly & Preissmann (1977). This method is conditional-stable (Courant condition) and it introduces slight errors in phase and amplitude. The diffusion equations are solved by use of an explicit difference numerical scheme. It is not a precise scheme, with a truncation error of the order of (At, Ax ) , but it was used especially for showing how the metric coefficients and the bed configuration affect the pollutant dispersion. The diffusion equation is approximated by a forward difference in the time direction and twice the central difference in the space direction. For diffusion in the x direction it gives the scheme: E At / \ n+l n x / n n n \ ,* C J. , = C. , + 2 C.1 , , - 2 JC , , + C -. k, (24) J 2 J+ k ,k 1 ' ' m Ax' \ ' J / x where subscripts j and k refer to the x-coordinate and k the zcoordinate respectively; and the superscript n refers to the time coordinate. The difference equation (24) is conditionally stable with the condition: E At
X

0.5

(25)

2 m Ax'

In the case of the lateral diffusion equation, the numerical scheme is more complicated:

64

W. Czernuszenko E At k k
c

2Az'

k-1

k \

-n- m k k+1 h, k n k+1

k ! ^
m

> k7

E At z 2Ax'
2 m

1 k-l \

k-1 m, k
m
+

_k "k

k-1 h.

n k-1

n(\-~ 1 k \ \ m, k

k+1 h k

- k h k-1

(26)

where m k = 0.5 (m^-i + m k ) , and the subscript j is omitted for simplicity. One can see how the depth and the metric coefficient distributions influence the concentration field from equation (26) . When variations of depth in a river are not large and the assumption that h is constant can be made, then equation (26) takes the form:

n+1 n C. = C, +

E At z Az' 2
m

k-1 k
m

k-l

'k+1

2C

n n '"k-l + C, n k k-1 m

(27)

In regular rivers the transverse component of velocity is rather small and the values of the metric coefficient are about unity. In this case, equation (27) takes a very simple form. Comparing the above equations, one can see differences between a straight, regular channel and an irregular one.

NUMERICAL SIMULATION
Calculations of the field of pollution concentration were carried out for experimental channels in which measurements of the velocity field and the elevation of the free water surface were made. Channel A (Fig.l) has a fixed uneven bed having a large scale configuration as in a natural river bed, but channel B has a rectangular cross section. The plan view geometry of these channels is the same. The values of the metric coefficients depend on the geometry of the curvilinear coordinate system. The preparation of the distribution of cumulative discharge, q c = / hudz (see Yotsukura & Sayre, 1976) for several measuring cross sections is required to establish the geometry of a set of longitudinal coordinate curves. The transverse coordinate curves are perpendicular to the longitudinal ones. The values of the metric coefficient are calculated for each cell of a numerical grid from equations (16) and (17). In the case of a circular shape for the longitudinal coordinates, the metric coefficient, m x , may be calculated from: m x = 1 + z/R (28)

Dispersion of pollutants in rivers

65

where : z = transverse coordinate (the inner bank of the river is chosen as the x-axis); and R = radius of curvature. The value of the metric coefficient, m , is difficult to express in any analytical form.
CROSS-SECTION

VELOCITY DISTRIBUTION

0
o

5.98 MINUTES

VELOCITY DIStRIBOTION

OUTLET: 21.8 GRAM

Fig. 1 Concentration distributions in meandering channels for different times (instantaneous source).

Simulations of the dispersion of polluted water introduced into channels A and B, both instantaneously and continuously, were carried out. In the case of channel A, the important role in mixing is governed by the irregular cross sections. For a river discharge of 0.463 m 3 s - 1 (average velocities of about 0.4 m s -1 ) the range of changes of m z is from 0.4 to 4. The iso-concentration lines are

66

W. Czemuszenko

shown in Fig.l. One can see the influence of the channel curvature on the dispersion process. The concentration distributions are not at all similar to Gaussian ones. In the case of channel B, with the same average velocities of 0.4 m s , the concentration distributions are more regular and the range of changes of m z is smaller, from 0.7 to 1.1. This range of changes may be used as a measure of irregularity of a river cross section. The rates of decrease of maximum concentration (percentage of initial concentration) in channels A and B are shown in Fig.2. In

40

30

CHANNEL A

CHANNEL B

z g

10

TIME , mln

Fig. 2 Maximum pollutant concentration in two different channels (instantaneous source).

the more regular channel B, concentrations are lower because of greater depths and velocities at the pollutant outlet bank. In the case of a continuous source, the field of pollution concentration is easy to achieve from the solution of equation (18) by its integration with respect to time. The simulation of dispersion in this case was carried out at an average velocity of about 0.2 m s - 1 . The distributions of concentration in a few cross sections of channels A and B are shown in Fig.3. The concentrations in channel A are much higher than those in channel B because of greater irregularities in channel A. The metric coefficient, m z , changes from 3.8 to 0.5 for channel A, and for channel B, from 1.45 to 0.9.

CONCLUSIONS
(a) The advection-dispersion equation may be solved in the case of irregular rivers with irregular cross sections where lateral velocities exist. In this case a special curvilinear coordinate system should be used. (b) The advection-dispersion equation in a curvilinear coordinate system has two additional parameters viz. the metric coef-

Dispersion of pollutants in rivers

CHANNEL B

67

DISTANCE FROM OUTLET BANK , m

Fig. 3 Concentration distributions at a number of channel cross-sections (continuous source),

The influence of m on the pollution concenficients m x and m z tration distribution in rivers is rather small and can be neglected, but the role of m z is significant.

REFERENCES
Holley, E.R. (1969) Unified view of diffusion and dispersion.
J. Hydraul. Div. ASCE 95 (HY2).

Holly, F.M. & Preissmann, A. (1977) Accurate calculation of transport in two dimensions. J. Hydraul. Div. ASCE 103 (HY11)
Yanenko, N.N. (1971) The lethod of Fractional Steps. Springer

Verlag, New York. Yotsukura, N. & Sayre, W.W. (1976) Transverse mixing in natural
channels. Received Wat. Resour. 1985; Res. 12(4). 26 August 1986. 30 December accepted