Professional Documents
Culture Documents
ABSTRACT
This chapter considers some further practical issues in river engineering and suggests
how the CES-AES software might be used to investigate them. The issues considered
relate to river ecology, sediment transport, geomorphology, trash screens and blockages to culverts. A brief overview is given of 1-D, 2-D and 3-D approaches to modelling
river flows, and sets the CES-AES software in this context. The chapter concludes with
some high level flow charts that indicate how the software might be adapted for other
specic research purposes.
A useful overview of existing legislation, policy and guidance supporting this for
drainage channels is provided in Buisson et al. (2008). Perhaps most important in the
context of European and UK flood risk management are:
Table 5.1 Example habitat information for various UK channel species (adapted from Natural England) (See colour plate section).
The CES enables users to describe the local vegetation and roughness cover
and to resolve the local depths and velocities whereas more traditional approaches
(e.g., PHABSIM) are based on section-averaged calculations of flow and velocity
(e.g., Chezy, Manning). This more localised information can be used with habitat
information, for example:
Figure 5.1 Variation of bed shear stress and friction factor with velocity and bedforms for a sand bed
river (after Raudkivi, 1998).
Since the local bed shear stress is mainly governed by the local depth of flow, H, the
distribution is naturally related somewhat to the geometry of the cross section. At high
stage, the bedforms change, in this example by generally a single category, and become
dunes, plane bed, antidunes and plane bed again. The effect of this on resistance,
and hence the H v Q relationship, is profound. Since dunes have a high resistance,
and plane bed (i.e., transitional flow at Froude Numbers approaching unity) have a
relatively low resistance, as shown in Figure 5.1, the corresponding H v Q relationship
is seen to flatten out at high flows, leading to a more or less constant stage of 27 feet.
This is much lower than might have been anticipated had only the low to medium flow
data, up to say 75,000 cfs (cubic feet per second) [i.e., 2,124 m3 s1 ], been used in a
regression equation and then subsequently used to extrapolate the H v Q relationship
up to flows of say 300,000 cfs [i.e., 8,495 m3 s1 ]. That estimate would then have
been higher than that which actually occurred, arising from the significant proportion
of the wetted perimeter experiencing plane bed (i.e., low resistance) conditions at high
stages. This illustrates the care that needs to be taken when estimating resistance for
sand bed rivers and the difficulties in extrapolating H v Q relationships without taking
into account all the relevant physical factors.
There is a large amount of literature on the subject of sediment mechanics,
including the flow conditions associated with initiation of motion (e.g., Shields, 1936;
Liu, 1957), equilibrium channel conditions (e.g., Ackers, 1992a; Bettess & White,
1987; Blench, 1969; Cao, 1995; Cao & Knight, 1996; Chang, 1988, White et al.,
1982), sediment transport rates and concentrations (e.g., Du Boys, 1879; Einstein,
1942; Bagnold, 1966; Engelund & Hansen, 1966; White, 1972; Ackers & White
1973; Chang, 1988; Ackers, 1990), bedforms and resistance (e.g., van Rijn, 1982 &
Figure 5.2 Stage-discharge relationship and bedforms in the Padma River, Bangladesh, at low and high
discharges (after Shen, 1971).
1984; White et al., 1980) and general behaviour as given in various textbooks (e.g.,
Chang, 1988; Garde & Ranga-Raju, 1977; Raudkivi, 1998; Shen, 1971; Simons &
Senturk, 1992; Thorne et al., 1997; Yalin & da Silva, 2001; Yang, 1987).
With regard to sediment transport rates, most methods typically rely on a resolution of the boundary flow conditions, using bed shear stress, shear velocities, velocity
and velocity gradients adjacent to the channel bed as well as the sediment properties
such as sediment size, density and fall velocity. This section is now aimed at demonstrating the use of the CES in estimating sediment transport rates, using one of the
more recent sediment transport approaches by Ackers & White (1973) and Ackers
(1990 & 1993c).
In 1973, Ackers and White first proposed their total material load formulae, which
are based on physical considerations and dimensionless analysis. It assumes that for
sediment moving as a bed load, transport correlates with grain shear stress, and the
grain shear stress becomes less signicant while total shear stresses become more signicant for finer sediments. Three dimensionless quantities, the sediment transport
parameter Ggr , the particle mobility number Fgr and the particle size number Dgr , are
defined as
mac
Fgr
qs h u nac
Ggr =
= Cac
1
(5.1)
qd50 U
Aac
(1nac )
U
unac
Fgr =
(5.2)
32 log10 (10h/d50 )
gd50 (/s 1)
g(/s 1) 1/3
Dgr = d50
(5.3)
v2
where qs (m3 s1 m1 ) is the volume of sediment transported per second per unit channel width and s is the sediment density. The coefcients in these formulae, which vary
with Dgr , were originally based on over 1000 experiments. More recently, these have
been updated to include additional data (Ackers, 1990 & 1993), and these coefcients
are shown in Table 5.2 according to the value of Dgr . A review of the performance
of several sediment transport formulae, including that of Ackers & White, is given in
Chang (1988).
It should be noted that nearly all sediment transport equations have been derived
for inbank flow conditions, typically in a single channel. Their application to overbank
flows therefore requires special treatment, as some studies suggest (e.g., Atabay, 2001;
Atabay et al., 2004; Atabay et al., 2005; Ayyoubzadeh, 1997; Brown, 1997; Knight
et al., 1999; Knight & Brown, 2001; Ikeda & McEwan, 2009; Tang & Knight, 2001).
An example is now given to illustrate how the CES can be of use in sediment transport
investigations, for both inbank and overbank flows.
The CES shear velocity and unit flow rate outputs may be used to solve these
formulae at each point in the cross-section, to evaluate the lateral distribution of qs .
Then by lateral integration of qs , the total sediment transport rate Qs (m3 s1 ) may be
obtained, where
B
Qs =
qs dy
(5.4)
Table 5.2 Coefficients to be used in the Ackers & White sediment transport
equations, (5.1)(5.3).
Coefficient
Dgr > 60
nac
mac
Aac
Cac
0
1.78
0.17
0.025
It is thus possible to determine the unit sediment charge xs (ppm) i.e. the unit mass
flux of sediment per unit mass of flow per unit time from
qs s
(5.5)
106 (ppm)
xs =
q
and the sediment charge Xs (ppm) i.e. the total mass flux of sediment per mass flux of
fluid per unit time from
Qs s
(5.6)
106 (ppm)
Xs =
Q
Sediment transport is highly non-linear and using an average main channel velocity such as that derived from Divided Channel Methods (DCM) or Ackers coherence
(COHM) approach (Section 2.5), rather than the depth-averaged velocity distribution, may result in substantially different sediment transport rates. In particular,
the average main channel velocity does not capture the reduced sediment transport
rates at the channel banks, tending to over-predict the average velocity and hence the
overall sediment transport. The CES methodology improves the local shear velocity
approximation from the more general formula (Eq. (2.31) in Section 2.2.3),
(5.7)
U = gRSf
1200
1000
800
600
400
200
0.2
0.4
0.6
0.8
1.2
1.4
Figure 5.3 The predicted sediment charge Xs for the FCF Phase A Series 2 channel with a d50 of
0.8 mm (after Mc Gahey, 2006).
Some work has already been attempted on determining channel shape and bank
erosion for inbank flows, as illustrated by Cao & Knight (1998) and Yu & Knight
(1998). Geomorphological issues related to overbank flows are dealt with more fully
in Ikeda and McEwan (2009), as are sediment processes, computer simulations and
design considerations. Work in the FCF on the behaviour of graded sediments in meandering channels is described by Wormleaton et al. (2004 & 2005). For information
on scour at bridges, see Melville & Coleman (2000).
There are many causes of blockage, but all are a combination of a source of debris
that can cause a blockage and a means of trapping that debris. Table 5.3 provides a
list of the most common factors.
The quantitative assessment of blockage potential remains difcult. There is some
empirical evidence to support predictive relationships for certain types of blockage.
Accumulations of large, oating woody debris (drift) were analyzed by the US Geological Survey (Diehl, 1997) in a Federal Highway Administration study based on 2,577
reported drift accumulations and field investigations of 144 drift accumulations in
large, wooded catchments. The maximum width of drift accumulations for any bridge
span is about equal to the length of large logs delivered to the bridge. Figure 5.4 shows
data used to estimate a design log length that could be used in assessing blockage
risk for a given structure. This empirical data is valuable but it is also specic to the
particular environment, blockage mechanism and bridge types studied.
Stream slope
Urbanisation
Vegetation
Sinuosity of stream
Upstream structures
Channel width
Trash screens
Access to the channel
Time since last flood
Upstream channel maintenance regime
Upstream land use and riparian management
High rainfall intensity
Width/height
Opening ratio
Shape
Skew
Length
Existence of piers/multiple culverts
Trash screens
Trash booms
Availability of skilled operators
Auto screen raking
Figure 5.4 Effective width of drift-blocked spans outside the Pacific Northwest (Diehl, 1997).
In the UK, a Blockage Risk Assessment method was developed for the Environment Agency South West Region in 19978. This method is reviewed by Benn et al.
(2004). Its main elements are:
The study identied several major problems with identifying blockage risk. One
is the lack of data and the second is that the probability distributions of the variables
describing the inuence of blockage are not independent, and the associated probabilities are subjective. This problem is not unique to blockage, and requires a consistent
risk management framework aligned with the treatment of other risk issues.
Research underway within the Flood Risk Management Research Consortium
(FRMRC2, http://www.floodrisk.org.uk) is gathering further field data on blockage
and seeking to develop a probabilistic method for predicting the onset and degree of
blockage potential at pinch points in the flooding system, such as bridges and culverts.
Trash screens (for example see Figure 5.5) are often placed over the entrance of
a culvert either to reduce the amount of debris entering the culvert barrel, where it
could cause blockage, or as a security screen to prevent unauthorised access. While
a trash screen may prevent blockage within the culvert barrel, where it is difficult to
clear debris, it may simply move the location of the blockage from the barrel of the
culvert or inlet to the screen.
Figure 5.5 Trash screen on the River Sheaf in Sheffield. The screen can be cleared by a mechanical
grab suspended from the gantry visible above the inlet. (Photo: JBA Consulting, Maltby
Land Surveys).
2 U2
Usc
mc
2g
(5.8)
where Umc is the section average velocity in the main channel, Usc is the average
velocity through the spaces between bars, g is acceleration due to gravity and Ksc
is an energy loss coefcient. The average velocity through the screen can be computed as Usc = Q/Asc where Q is the discharge and Asc is the flow area between
the bars. CIRIA (1997) recommended Ks 1.5. Representation of trash screens and
blockage in hydraulic models remains an area of on going research. Experimental
measurements and numerical modelling indicate that head loss coefcients increase
with increasing blockage and that it may be possible to identify dimensionless relationships between flow variables for trash screens (Tsikata et al., 2009) which may
assist with the calibration of simplied empirical models for 1-D analysis.
2-D models
3-D models
ISIS
MIKE11
HEC-RAS
SOBEK
InfoWorks RS 1D
NOAH
ISIS2D
MIKE21
RMA2
TELEMAC2D
InfoWorks RS 2D
DIVAST
SSIM
LISFLOOD
TUFLOW
Research codes
CFX
PHOENICS
FLUENT
TELEMAC3D
DELFT FlOW3D
Research codes
Research codes
showing the range of some standard models and the extent of progress in Computational Fluid Dynamics (CFD) over the last few decades. Whereas 1-D codes were
just starting in the 1960s, today they are commonplace and 2-D & 3-D codes are
now frequently used in solving practical problems. This does not imply that 2-D &
3-D codes have superseded the use of all 1-D models, as each type of model has
its own particular use, functionality and purpose. As seen later on, 1-D models are
particularly useful in dealing with many river issues since rivers themselves are, by
nature, long single-thread systems that convey water in a predominately downstream
direction. It is also shown that the results obtained from higher dimensionality 3-D
models are not necessarily more accurate and, furthermore, involve considerable effort
and cost in obtaining them (Wright et al., 2004). The purpose here is to identify
where the CES-AES software fits into this framework and to consider briefly the
issues that arise when using software of different spatial dimensionality, i.e., 1-D,
2-D or 3-D. Before discussing these issues an overview of the equations used in each
spatial dimension is given, with particular emphasis placed on 1-D unsteady flow
models, so that a direct comparison may be made between these and the CES-AES,
even though it contains some 2-D and 3-D flow characteristics and is for steady
flow only.
i One-dimensional model equations
As noted in Section 2.2.4, the 1-D equation for flow in an open channel relates the
gravitational and frictional forces with the local and convective accelerations. Rather
than expressing them in terms of h and U as functions of x and t, as in Eqs. (2.36)
(2.38), in 1-D river modelling codes it is customary to express the continuity and
momentum equations in terms of A and Q as functions of x and t. These are known
as the St Venant equations, and may be expressed in the form:
A Q
+
= q
t
x
Q2
h
Q
+
+ gA
+ S f So = 0
t
x
A
x
(5.9)
(5.10)
where the symbols have their usual meaning, and where q = lateral inflow/outflow per
unit length and = momentum correction coefficient. Equations (5.9) & (5.10) thus
constitute two equations for two unknowns (A & Q) and may be solved numerically
to give A and Q as functions of x and t. The area, A, and discharge, Q, are then
known at all points along the river and with time. Examples of unsteady flow where
such results might be useful are those involving flood simulations along rivers (flood
routing) or tidal motion in long narrow estuaries. Knowing A and Q for all x and t,
allows the depth, h, or water level, , to be determined since the A v h relationship
is known at every cross-section. Likewise, the velocity, U, may be calculated at every
time step and cross section, since U = Q/A. The 1-D nature of the equations thus
reflects the mainly one-dimensional flow in a river or estuary. An assessment of the
relative importance of the various terms in Eq. (5.10) is given by Knight (1981), based
on measurements in the Conwy estuary.
The classication of 1-D flows based on Equations (2.36) and (2.38) in Chapter 2
may now be claried and linked to the St Venant equations above. Reverting to
U and h as variables, then Eq. (2.36) gives the following classication for 1-D flows:
Sf = So
1 U
h U U
x
g x
g t
(5.11)
So x gSo x
gSo t
kinematic wave |
diffusion wave |
full dynamic wave |
For the purposes of comparing these two classications, based on Equations (5.11)
& (5.12), with the St Venant equations, (5.9) & (5.10), consider 1-D flow in a frictionless, horizontal prismatic rectangular channel, with no lateral inflow, so that Sf = 0,
So = 0, q = 0 and A = bh, where b = constant. Then, by ignoring the convective
acceleration term (the second term in Eq. (5.10)), elimination of the terms A/t and
h/x by cross differentiation reduces the pair of equations to a single equation, giving
the water surface elevation, , as:
2
2
gh 2 = 0
2
t
x
(5.13)
This represents a simple progressive wave, since a solution to Eq. (5.13) is of the
form:
= a cos(kx ct)
(5.14)
(5.15)
or
Q
Q
+c
=0
t
x
(5.16)
(5.17)
dQ
1 dQ
=
dA
B dh
(5.18)
The solution to Eq. (5.17), the kinematic wave equation, provided c is constant, is
Q = f (x ct)
(5.19)
where f is any function. Eq. (5.19) is thus seen to represent any shape of wave, travelling in a positive x direction with a speed or celerity, c, given by Eq. (5.18), without
change in shape. This involves the inverse of the gradient of the stage-discharge relationship, 1/(dh/dQ), and the water surface width, B. Since the CES can provide the
stage-discharge relationship for rivers with any prismatic cross-sectional shape, this
maybe demonstrates another use to which the software may be put, as described later
in Section 5.4.3.
However, in reality the effect of friction tends to attenuate and delay the passage of
a flood wave, so that the simple wave form suggested by Eq. (5.19) becomes distorted
as the wave translates (moves) along the river. The non-uniformity of the cross-section
shape with stage is also particularly important in this respect. To account for this, the
first two terms in Eq. (5.12) may be combined with Eq. (5.9) to give the so-called
convective-diffusion equation, usually presented in the standard form as
Q
2Q
Q
+c
=D
t
x
x2
(5.20)
in which c = kinematic wave speed, given again by Eq. (5.18) and D = diffusion
coefficient, given by
D=
Q
(2BSo )
(5.21)
It follows from Eq. (5.20) that the discharge in a channel during a flood event has
the characteristics of a wave that now translates and attenuates. It should be noted
that in the context of river engineering, both c and D in (5.20) are functions of the
discharge Q, making the solution of Eq. (5.20) somewhat difficult. However, Cunge
(1969) has shown that the Variable Parameter Muskingum-Cunge (VPMC) method
is similar to the Preissmann finite-difference scheme used in solving the St Venant
equations (Preissmann, 1961). VPMC uses 4 points in x-t, like the Preissmann box
scheme but the leading term of the numerical truncation error is tuned to match the
diffusion term. This then relates Equations (5.9) & (5.10) with (5.20) and allows for
kinematic wave routing methods to be used in practice for channels with bed slopes
greater than around 0.002 (See Knight, 2006c, for further details). It also gives a
theoretical basis for varying the travel time constant, K, and the distance weighting
parameter, , in the basic routing equations used in the original Muskingum method
(Bedient & Huber, 1988; Shaw, 1994):
dS
dt
S = K[ I + (1 ) O]
IO=
(5.22)
(5.23)
(5.24)
hUd2
h
hUd
hUd Vd
+
+ gh
= gh(Sox Sfx )
+
t
y
x
x
(5.25)
hVd2
hUd Vd
h
hVd
+
+
+ gh
= gh(Soy Sfy )
t
x
y
y
(5.26)
with
Sfx =
n2 Ud Ud2 + Vd2
7
h /3
and Sfy =
n2 Vd Ud2 + Vd2
7
h /3
(5.27)
where Ud and Vd are the depth-averaged velocities in the x and y directions respectively. Other terms can be included to represent the effects of turbulence. If the
turbulent terms are omitted they are in effect lumped together with the numerical
truncation errors and any other calibration coefcient, rather than truly reecting the
physical process. The determination of n is therefore not straightforward and this
raises a number of issues, as discussed by Cunge et al. (1980), Samuels (1985) and
Morvan et al. (2008).
Equations (5.24)(5.26) thus constitute three equations for three unknowns
(Ud , Vd and h) and may be solved numerically to give these as functions of x, y and t. In
this case, rather than use cross-section data, a mesh of {x, y} co-ordinates are required
to represent the river and any associated oodplains. The individual velocity vectors
may be combined to give the magnitude and direction of the depth-averaged velocity
at any point with time. This is one advantage of such a model compared with the CESAES, in which the direction of flow is assumed to be always in the streamwise direction.
A 2-D model is therefore more suitable where details of flows over oodplains are
required, or where the geometry of a river is such that the flow is complex, as in the
case of bends or braided rivers for example. However, the 2-D model, as expressed
in Equations (5.24)(5.26), cannot deal with secondary flows, plan form vorticity or
Reynolds stresses, which the CES-AES can do in an approximate, but effective manner.
iii Three-dimensional model equations
In 3-D models, the three time-averaged RANS equations, together with one equation
for mass conservation, are used determine all three velocity components {UVW} at
every point {xyz}, together with the depth, h, as a function of {xy} in plan form, with
time, t. In essence, 4 equations are solved to give the 4 unknowns {U, V, W & h} as
functions of {x, y, z & t}. A 3-dimensional mesh is therefore now required to represent the model river, with special treatment being taken for the free surface (initial
water level throughout the domain, as well as how it is controlled during the computation). See Ikeda & McEwan (2009), Nezu and Nakagawa (1993), Rodi (1980) and
Versteeg & Malalaskera (1995) for further details.
Such models are inevitably mathematically complex, and also require a large number of ancillary equations to deal with the turbulent structures, in what is known as the
turbulence closure problem. It is referred to as a problem because of the difculty
in describing turbulence in fluid flows, based on our incomplete knowledge of the
physical reality, as well as our inability to describe such turbulence in a mathematical
set of equations that are universally agreed. Turbulence closure relates the unknown
turbulent processes to the larger scale resolved motions. For further information on
turbulent energy, cascades and spectra see Tennekes & Lumley (1972) and Versteeg &
Malalaskera (1995).
The turbulence closure issue has been addressed by many researchers, usually by
formulating an equation (or set of equations), that purports to represent the turbulence.
These additional equations may be of many types. The simplest is a zero equation
model, based either on the eddy viscosity principle, first proposed by Boussinesq in
1877, or the mixing length hypothesis, first proposed by Prandtl in 1925. The CES-AES
in fact uses the same eddy viscosity approach to deal with one aspect of turbulence,
as already indicated in Eq. (2.59) or Eq. (3.18). Other types of equations to deal
with turbulence closure problem are: a two-equation model, involving differential
equations that represent the relationship between energy production and dissipation;
a Reynolds stress model that accounts for all three internal turbulent stresses; or, an
algebraic stress model that in a somewhat simpler manner accounts for the anisotropy
of the Reynolds stresses. Finally, some have even suggested that the time-averaged
RANS equations may not even be the right starting place for certain types of problem
and that double-averaged equations (i.e., averaging over time and space) should be
used instead (Nikora et al., 2008). As an alternative to RANS based models, large
eddy simulation (LES) models or Direct Numerical Simulation (DNS) may be used to
actually account for the individual turbulent fluctuations. The present position is best
summed up by the authors of two books, written some 70 years apart, as cited in
Nakato & Ettema (1996, pages 457458) by one of the present authors:
Were water a perfectly non-viscous, inelastic fluid, whose particles, when in
motion, always followed sensibly parallel paths, Hydraulics would be one of the
most exact of the sciences. But water satises none of these conditions, and the
result is that in the majority of cases brought before the engineer, motions and
forces of such complexity are introduced as bafe all attempts at a rigorous solution. This being so, the best that can be done is to discuss each phenomenon on
the assumption that the fluid in motion is perfect, and to modify the results so
obtained until they fit the results of experiments, by the introduction of some
empirical constant which shall involve the effect of every disregarded factor. It is
worthwhile here impressing on the student of the science that, apart from these
experimentally-deduced constants, his theoretical results are, at the best, only
approximations to the truth, and may, if care be not taken in their interpretation,
be actually misleading. On the other hand, it may be well to answer the criticism
of those who would cavil at such theoretical treatment, by pointing out that the
results so obtained provide the only rational framework on which to erect the
more complete structure of hydraulics.
(A.H. Gibson, 1908)
Turbulent motions contribute signicantly to the transport of momentum, heat
and mass in most flows of practical interest and therefore have a determining
inuence on the distribution of velocity, temperature and species concentration
over the flow field. It is the basic task of engineers working in the field of fluid
mechanics to determine these distributions for a certain problem, and if the task is
to run the model at all. For example, in 1-D unsteady flow models, the boundary
condition upstream is normally that of an input hydrograph {Q v t} and the boundary
condition downstream is frequently that of a fixed depth or a {h v Q} relationship.
Initial conditions will also be required throughout the domain, e.g., starting values
for depths and discharges (or velocities) at all cross sections. In 2-D models, similar
requirements apply, but at open boundaries the velocity field must be specified as well,
which may initially be unknown. The same applies in 3-D models where wall functions
are used to mimic the boundary conditions at mesh points very close to the wall, and
cyclic boundary conditions might be applied at open ends to improve convergence and
reduce the run time. For further information on boundary conditions, see Cunge et al.
(1980), Samuels (1985) and Vreugdenhil (1989).
The numerical approach used in solving any governing differential equation, possibly based on finite-difference, finite-element or finite-volume methods, also needs
to be appreciated and understood. The consistency, convergence, stability and accuracy of any adopted numerical method needs to be known, and possibly also the
techniques used in investigating them. Such techniques are often based on Fourier
series in complex exponential form, but are clearly beyond the scope of this book.
See Abbot & Basco (1989), Cunge et al. (1980) and Preissmann (1961) for further
information.
In 1-D models, it should be noted that even the governing St Venant equations may
be discretized in different ways, and that this will inevitably affect the numerical results,
as shown by Whitlow and Knight (1992). In 2-D and 3-D models it is imperative to
vary the mesh size and to observe at what point of resolution any two solutions agree.
In 3-D models, it is not only important to vary the mesh size, but also to check on
key parameters at strategic points in the mesh, monitoring them at certain time steps
as the computation proceeds. This is because the numerical procedure may take many
thousands of iterations to fully converge to a solution, even when solving for a steady
flow case. For example, the turbulent intensity or Reynolds stress at a certain point
in the mesh is influenced by values at adjacent points in the mesh. Since the solution
technique is iterative, the parameters at a single point (e.g., 3 velocity components,
{UVW}, mean pressure, 3 turbulent intensities, 3 Reynolds stresses, etc.) may take a
long time to converge at that particular point, and at every other point in the mesh
(often thousands). The numerical solution for the whole domain is then not complete
until similar convergences are obtained at all points in the mesh, to a pre-determined
and specified accuracy for each particular variable or parameter.
From the preceding paragraph it is clear that the data required for different types
of model varies according to their dimensionality and complexity. A 1-D model will
generally only require cross section data at specified intervals, chosen to reflect the
terrain, as indicated by Samuels (1990), and more recently by Casellarin et al. (2009).
Channel roughness (and its representation) is arguably the single most important issue
to resolve prior to successful modelling. In 1-D models, the use of channel cross sections
makes it relatively easy to compare results with known mean velocities, since U = Q/A.
In 2-D models, the data required for calibration increases significantly, since two
components of velocity are required at every point in the mesh, giving both magnitude
and direction locally. Furthermore, in 2-D models the water surface itself may no
longer be considered as planar, as for example occurs in bend flow or in the vicinity
of bridges. In 3-D models the data requirements multiply even more significantly,
with mean and turbulent parameters required at every location in an {xyz} co-ordinate
system.
In most practical case studies there is simply not enough hydraulic or turbulence
data at sufciently detailed spatial and temporal scales to be meaningful. This makes
the calibration stage particularly difcult, and reliance has then to be placed on the
software users experience. The dimensionality of the model also affects the run time
and data handling. Modern CFD packages usually come with a number of post processing packages, enabling the results to be analysed in several ways. The use of
graphs and charts to plot secondary flow cells, strength of vorticity, shear stresses
on internal elements, colours to visualize velocity fields within each cross section,
etc., make it possible to undertake assessments and any comparisons quickly and efciently. Comparisons may not always be between measured and numerical results,
on account of the scarcity of measured data, but often between different computer
runs with differing flow or geometric features. Not to be forgotten are comparisons
with any analytical results, if they are available, and with any suitable benchmark
laboratory data.
The concept of equifinality is an important one in model calibration, as noted
by Beven & Freer (2001), who state that
. . . the concept of equifinality is that the uncertainty associated with the use
of models in prediction might be wider, than has hitherto been considered, since
there are several (many?) different acceptable model structures or many acceptable
parameter sets scattered throughout the parameter space all of which are consistent
in some sense with the calibration data. The range of any predicted variables is
likely to be greater than might be suggested by a linearised analysis of the area
of parameter space around the optimum. This suggests that the predictions
of all the acceptable models should be included in the assessment of prediction
uncertainty, weighted by their relative likelihood or level of acceptability.
The use of models outside their calibration range also raises issues in assessing
uncertainty. See Beven, (2006), Hall & Solamatine (2008), Samuels (1995) and Sharifi
et al. (2009a & b) for further details.
The overall structure of the software from a user perspective is shown in Figure 5.6.
The software incorporates 3 main elements, the roughness advisor, the conveyance
generator and the backwater calculator. Afflux is calculated as part of a backwater
as this provides the necessary downstream water level for the afflux method. Data
specific to the site of interest is stored in two data files. The .RAD file contains the
user specific data relating to vegetation, substrate and irregularity for each of the
roughness zones of interest. This file is saved by the user and stores the output of
the roughness calculations carried out within the roughness advisor. The .GEN file
contains the geometrical data for the channel and any bridge and culvert structures,
as well as information on which roughness zones are used for calculating the section
conveyance. Together the .GEN and .RAD files contain all of the data relevant to a
particular site.
The raw data for the roughness advisor is provided from a number of databases
that capture the outputs of the roughness review carried out for the original conveyance
estimation project. These databases are in a simple .CSV format and this was selected
to allow users the flexibility to edit and update the files if they have access to improved
or alternative roughness data, though for most purposes the data should be considered
as fixed.
Figures 5.7, 5.8, 5.9 and 5.10 show the flow charts that define the operation of
the calculation engines for roughness, conveyance and backwater.
The majority of users of the CES-AES will run the software through the standard
user interface. However, it was anticipated in both the conveyance and afflux projects
that the source code for the calculation methods would be made available for the purposes of further research. The software design allows for this by separating all the
main control and calculation code from the user interface. This is essential, as the user
interface contains compiled proprietary code that cannot be distributed. The current
structure of the software when compiled is shown in a simplied form in Figure 5.11.
The main software procedures are all handled by the Convey.DLL module, including
file access, data structure and manipulation, roughness calculation, backwater calculation and interfacing to the conveyance and afflux calculation modules. This structure
effectively allows a programmer with access to the source code to replace the standard
user interface with a simple interface of his or her own, and run all of the underlying
calculation code.
The Afflux calculation is accessed through an interface (AEData). This structure
was adopted as it allowed the afflux calculator to be developed independently from
the CES software, with interaction taking place through the agreed software interface.
In practice, the overall structure shown is more complicated at the level of
the source code itself, with the code sub-divided into classes in line with good
software development procedures. The code has been written in three different languages. The core conveyance calculation engine (ConveyCalcs) is currently written
in C for maximum portability, whilst the afflux interface and engine are written in
VB. All other parts of the software are written in C++. It is probable that future
updates to the software may result in changes to this structure though these will
not remove the capability of running the calculation modules from outside the user
interface.
Customisation of the software can be carried out at three levels. Firstly, the user
can modify the default parameters that are used in the conveyance calculation. The
Advanced Option box shown in Figure 5.12 provides a means for users to alter many
of the model parameters via the CES-AES User Interface (Table 5.5).
Default value
Allowable range
Temperature
Number of depth intervals
Minimum depth used in calculation
15 C
25
Lowest bed elevation
0.24
100
>0
100
Any depth below maximum
entered elevation
0.10.5
500
1
0.001
20
1.5
Off
Secondly, the user may modify the default roughness databases used by the
software. These changes might take the form of adjustments to the default roughness values, incorporation of new photographs or addition of new materials and
roughnesses. The file formats are fully documented and this documentation can be
downloaded from the CES-AES Website.
Thirdly, it will be possible for researchers to investigate changes to the core
calculation routines in the software through modication of the software source code.
For example, the management models used to implement cutting of vegetation might
be modied in the light of improved knowledge of the re-growth rates of various vegetation types. The existing cutting routines are found in the CManagementModel class
and the C++ code that implements this is available as ManagementModel.cpp.
At the time of writing, the mechanism by which source code for the CES and AES
calculation may be made available had yet to be finalised. However, it is expected
that it will be made available under licence to other software houses and bona fide
researchers. Further information will be available through the CES-AES website at
www.river-conveyance.net.