Professional Documents
Culture Documents
Modelling The Distribution of Velocity in A River Cross Section
Modelling The Distribution of Velocity in A River Cross Section
To cite this article: Derek G. Goring , Jeremy M. Walsh , Peter Rutschmann & Jürg Trösch (1997)
Modelling the distribution of velocity in a river cross‐section, New Zealand Journal of Marine and
Freshwater Research, 31:2, 155-162, DOI: 10.1080/00288330.1997.9516754
INTRODUCTION MODEL
Estimating the distribution of velocity across a river Mathematical development
cross-section is important for several practical Our model is based on the Navier-Stokes equations
reasons. For example, river engineers must estimate (e.g., see Rodi 1984), which are the general
the velocity to determine what size sediment will equations which govern the motion of a Newtonian
move and under what conditions; scientists studying fluid. We make the following assumptions about
fish habitats estimate the velocity distribution to flow in the cross-section:
determine what effect a change in the flow regime (1) the velocity gradient in the flow direction is
negligible, i.e., steady uniform flow is assumed;
(2) velocity vectors are orthogonal to the cross-
section, i.e., there are no velocity components
M95041 in the plane of the cross-section; and
Received 30 June 1995; accepted 27 February 1997 (3) the water level is horizontal across the section.
156 New Zealand Journal of Marine and Freshwater Research, 1997, Vol. 31
With these assumptions, and using Boussinesq's Model boundary
eddy-viscosity concept (Rodi 1984), the Navier-
Stokes equations reduce to a single equation:
gS (1)
Origin of the wall
where u(y,z) is the longitudinal velocity,^ andz are
horizontal and vertical components of distance in
the plane of the cross-section, g is the acceleration Representation of a bed roughness element
of gravity, S is the slope, and v is the total viscosity.
The total viscosity is the sum of the molecular and Fig. 1 Bed roughness element.
turbulent viscosities, but since the latter is orders
of magnitude higher than the former, the molecular
viscosity may be neglected. The total viscosity
therefore becomes equal to the turbulent viscosity factor/, which can take any value between 0 and 1
and using a mixing length hypothesis can be defined inclusive. Therefore, the distance from the model
boundary to the wall can be varied within the range
as: of one dgo grain diameter. Figure 1 shows the model
boundary and theoretical wall level for a typical
(2) boundary roughness element located at the bed.
Experimental data (Yalin 1977) pertaining to the
where /„, is the Prandtl mixing length, which is taken theoretical wall level, below the top of the
as: roughness elements 8, suggests that a range of 0.15—
0.3 times the equivalent sand grain roughness, e,
is appropriate. Thus, we expect/to take values
K is von Karman's constant, equal to 0.4 for all between 0.3 and 0.6.
homogenous fluids, and h(y) and z'(y) are total depth By substituting in the relationships for 8 and e
and distance above the bed, respectively, both (i.e., 8 =fd90 ande = 2 dgo), Eq. 4 can be rewritten
measured vertically from the theoretical wall level as a function of the parameter/, to give the velocity
at each vertical. at the model boundary, i.e.:
The described mixing length hypothesis strictly
applies only to planar boundaries. It is assumed that - ^ = 2.5 1n(30.1//2) (6)
the same equations will approximate the behaviour
of the boundary layer for an irregular boundary. where u¡, is the velocity at the boundary.
i s t h e 6, a velocity at the boundary .
Model boundary determined which is indépendant of e, except when
£ is a significant fraction of the total depth, in which
The bed co-ordinates are assumed to represent a case the effect of e on the near bed shear velocity,
surface tangential to the top of the bed-roughness
M* is significant. The effect of £ on fluid particles
elements. This surface defines the lower boundary
of the calculation domain (or the model lower within the cross-section is felt primarily through
boundary). Velocities at this boundary are the influence of e on the turbulent viscosity.
determined with a logarithmic wall function:
Mathematical model
— = 2.5 1n(30.15/e) (4) The simplified Navier-Stokes partial differential
where 8 is the distance to the wall, s is the equivalent equation, Eq. 1, is combined with the expression
sand grain roughness (taken as 2x dgo ), and u* is for turbulent viscosity, Eq. 2, and the resulting non-
the near-bed shear velocity which is approximated linear partial differential equation is transformed
by: to an integral equation using the standard Galerkin
approach (e.g., Prinos & Townsend 1984). The
«* = 4gM (5) integral equation is then discretised using an 8-node
The distance from the model boundary to the quadrilateral basis function (Rutschmann 1990).
theoretical wall (i.e., bed) level, S, is determined Linearisation of the integral equation is
by multiplying the dgo size of the bed material by a performed with the Picard iteration using an
Goring et al.—Velocity in a river cross-section 157
Fig. 2 Finite element mesh for
a typical cross-section, showing 4fl4
252 eight-node quadrilateral
elements.
48.2
48.0
47.8
47.6
Offset (m)
underrelaxation factor of 0.5. Starting with a penetration of the flow to below the top of the
logarithmic velocity distribution, where the velocity roughness elements as a proportion of the d^ grain
depends only on the position in depth and the sand size of the bed material.
grain roughness, a new solution is obtained by Figure 3A illustrates that, within the expected
solving the system of linearised integral equations limits o f / ( i . e . , 0.3-0.6), there is a moderate
over the calculation domain. Iteration is then sensitivity of velocity to this parameter. For
performed until a stable solution is obtained. The example, the difference in the velocity at 0.6 of the
linearised integral equations are solved using frontal depth obtained assuming/= 0.3, compared to that
elimination with full pivoting. assuming/= 0.6 is c. 31%. In comparison, Fig. 3B
shows that there is very little sensitivity to grain
Mesh generation size. Clearly, the most sensitive of the parameters
Using an automatic finite element mesh generation is water surface slope (see Fig. 3C), with velocity
program, the cross-section is divided into 252 varying as the square root of the slope. This agrees
elements with 843 nodes, as shown in Fig. 2. The with the empirical Manning's equation (Henderson
spacing between nodes increases logarithmically 1966) commonly used to estimate mean cross-
from the bed up to a semi-ellipse which sectional velocity:
encompasses the water surface. This ensures that
elements are sized appropriately to model velocity
gradients at all positions in the cross-section. where V is the mean, cross-sectional velocity, R is
the hydraulic radius, and n is Manning's coefficient,
related to the equivalent sand-grain roughness by
APPLICATION the following relationship: (Garde & Ranga Raju
Model sensitivity 1985)
Before presenting the results of applying the model n = £ 1 / 6 /24 (8)
to field data, the sensitivity of the model is examined Figures 4A and 4B show that the depth-
by applying it to flow in a wide, rectangular section, averaged velocities calculated from applying the
1 m deep by 200 m wide, with material of size dgo finite element method to a wide rectangular section
= 0.128 m (e= 0.256 m) and slope 5= 0.001. Figure compare closely with those calculated from the
3 shows the velocity distribution at mid width for Manning equation over a wide range of slopes and
various combinations of the parameters: the bed roughnesses.
equivalent sand grain roughness e, the water surface Accurate measurement of the water surface
slope S, and the factor/which determines the slope in the field is difficult. Figure 4B illustrates
158 New Zealand Journal of Marine and Freshwater Research, 1997, Vol. 31
Fig. 3 Velocity profiles at mid-
width of a wide, rectangular
section, showing the sensitivity
to: A, the proportion of penetra-
tion into the bed; B, equivalent;
sand grain roughness; and C,
slope.
S = 0.001
£ = 0.256 m
1.4 1.6
1.6
Velocity (m s"1 )
24 30 36 42 48 54 60
Distance downstream (m)
flow conditions and secondary currents. Thus, the deteriorates near the downstream end, and at
comparative results presented in the figures may Section 5, where the flow exhibits most distortion
be viewed, not as indications of how well the model in the cross-section. In this way, the model can be
performs, but indications of the proximity of the used to determine the non-uniformity of measured
actual flow to steady, uniform flow with no cross-sectional velocity data.
secondary currents. Viewed from this angle, Fig. 6, Of more practical importance than this,
7, and 8 indicate that for most of the reach the flow however, is the possibility of using the model to
essentially conforms to these assumptions, but enhance or, in some instances, even displace the
Goring et al.—Velocity in a river cross-section 161
m i i o.m \ 7
48.40
Offset (m)
Fig. 7 Comparison between velocity contours calculated by the model and measured point velocities for Section 3.
48.40
conveyance method for habitat hydraulics sections of the reach under consideration. This
studies. To determine the velocity distribution can be an expensive exercise which may not be
using the conveyance method, one must first justified for the particular application. However,
measure the velocity distribution in a section or for the finite element model described here, the
162 New Zealand Journal of Marine and Freshwater Research, 1997, Vol. 31
basic requirements are the cross-sectional shape, CONCLUSIONS
the roughness (i.e., the d90 of the bed material)
A finite element model which solves the simplified
and any two of the following: water level, flow,
Navier-Stokes equations and uses the Prandtl
or slope of the energy line. Thus, a primary
mixing length hypothesis provides an alternative
difference between the two methods is that the
means of estimating the velocity distribution across
conveyance method is empirical whereas the
finite element model is based on the physical laws a river cross-section. The model has been shown to
of conservation of mass and momentum; it is give reasonable results in a straight reach where
therefore deterministic and does not require field near-uniform flow conditions exist. Where
measurements of the velocity. Of course, the hydraulic conditions are different from those
velocity distributions thus produced may lack the assumed, the results are less satisfactory.
details of the actual velocity distribution such as The model differs from the well-known
are introduced by non-uniform flow and conveyance method (Mosley & Jowett 1983) in that
secondary currents (as illustrated in Fig. 6, 7, and it is deterministic, not empirical, and the results are
8). Nevertheless, if the purpose of the study is to two-dimensional, not one-dimensional.
estimate how the flow field changes with flow, The model has application to research on the
for example, thus affecting the habitat, the presence or absence of secondary currents in
differences between velocity distributions may be measured velocity distributions as well as habitat
more relevant than the absolute distributions. hydraulics studies where it provides an alternative
to the conveyance method.
The other primary difference between the
conveyance method and the finite element method
described here is that whereas the conveyance
method produces depth-averaged velocities across REFERENCES
the section and is thus one-dimensional, the finite Garde, R. J.; Ranga Raju, K. G. 1985: Mechanics of
element model produces a two-dimensional velocity sediment transportation and alluvial stream
field. problems. 2nd Ed., John Wiley & Sons, Inc., New
York.
For the results shown in Fig. 6, 7, and 8, the
water level and flow were known and the slope was Henderson, F.M. 1966: Open channel flow. MacMillan,
calculated from Manning's equation. However, an New York.
alternative approach would be to use the model to Moslev, M. P.; Jowett, I. G. 1985: Fish habitat analysis
iterate on slope until the desired flow was produced. using river flow simulation. New Zealand journal
Under most circumstances, the results would be of marine and freshwater research 19: 293-309.
essentially the same, as was illustrated in Fig. 4. Prinos, P; Townsend, R. D. 1984: Prediction of main
Nevertheless, this iterative method may be channel/flood plain flow interaction with FEM.
preferable in some circumstances, such as when Proceedings of the 5th International Conference
there is difficulty in solving Manning's equation on Finite Elements in Water Resources.
for a complex section. Berlington, Vermont, pp. 509-520.
If, for a particular application, the relationship Rodi, W. 1984: Turbulence models and their application
between flow and water level is unknown, one in hydraulics, a state of the art review.
solution is to use the valley slope as the slope of International Association of Hydraulic Research,
the energy line, then run the model with a series of A.A. Balkema, Rotterdam, 2nd ed. 104 p.
water levels to produce the velocity distribution for Rutschmann, P. 1990: Die Methode der finiten Elemente
the corresponding series of flows. Such an exercise in der Hydraulik. In: Halin, J. ed. ASIM
is unlikely to reproduce the actual velocities very Mitteilungen, Heft 19: 8-14. Zurich, Switzerland.
accurately, but will give a representative distribution Yalin, M. S. 1977: Mechanics of sediment transport. 2nd
and how it changes for different flows. Ed. Pergamon Press.