Professional Documents
Culture Documents
by
JIWESHWAR SINHA
(Reg. No.: D 09 CE 111)
Under Supervision of
Dr. P.L. Patel, Dr. B.K. Samtani & Dr. S.K. Das
II
III
IV
Acknowledgement
I would like to take this opportunity to first and foremost thank God
for being my strength and guide during this research work. Without Him,
I would not have had the wisdom and the physical ability to do so.
I am deeply grateful to all my research supervisors Dr. P. L. Patel, Dr.
S.K. Das and Dr. B.K. Samtani for their constant support and guidance.
Dr. Patels continuous encouragement helped me to complete this research
work in time. Dr. Das was always available for technical discussions
throughout this work.
I am thankful to Dr. D.C. Jinwala, Dr. J. Banerjee, Dr. S M Yadav
and Dr. V.L. Manekar for evaluating my work as examiners in Research
Progress & Pre-synopsis Seminar Committee and for their guidance to
improve the quality of thesis.
I would like to express my sincere thanks to Dr. M.K. Sinha,
Director, CWPRS & Shri M.D. Kudale, Additional Director, CWPRS for
providing the departmental infrastructure facilities during this research
work. I am also thankful to Smt. V.M. Bendre, Ex. Director, CWPRS for
permitting me to pursue higher studies.
My special thanks are due to Dr. Shailesh B. Patel and Shri Viraj
Loliyana for their help during the research work and writing of this thesis.
I would also like to thank Dr. Prafulkumar V. Timbadiya, Dr. U.C.
Roman, Dr. P. Chandra, Dr. A.K. Bagwan, S/Smt. Seema Shiyekar,
Shivani Sahu, Komal Vighe, Vaibhavi Roy, Miss Kavya, S/Shri V.B.
Sharma, S.N. Jha, Priyank Sharma, Nitin Bharadiya, Lalit Pal, Anil,
Akshay, Hasmukh Patel for extending all kinds of support during the
research work.
I would also like to express my gratitude towards my mother Smt.
Bachchi Sinha, In-laws Smt. Premlata Sinha & Shri Ayodhya P. Sinha,
Elder brother Shri Jitendra Sinha, and My wife Sudha, son Navneet &
daughter Smriti for providing me inspiration and enthusiasm at all level
to work hard. Sincere thanks to all my family members without their
support it is not possible to pursue higher studies.
Last but not least I would like to thank all known and unknown
persons for their kind co-operation, help and motivation during the
research.
Jiweshwar Sinha
VI
Abstract
In nature, rivers are made up of a centrally located main channel and floodplains on one
or both sides of it. The river systems are being overloaded with pollutants due to rapid
land use changes in their respective catchments. A thorough understanding of the flow
interaction between the main channel and floodplains is essentially required for river
management and related works. Also, the prediction of hydraulic characteristics in
natural river systems is required in assessment of their pollutant transport capacity along
the flow. The natural river can be represented as a two layered system in the form of
upper layer and lower layer, and a horizontal interface in between them where convective
exchange between the layers is allowed. A two-layer two-dimensional (2-D)
mathematical model is developed for hydrodynamics and solute transport in compound
open-channel using Cartesian equations for non-uniform grid system. Turbulent exchange
across the interface is treated empirically. Turbulent terms are dealt with turbulence
closure schemes under large eddy simulation (LES) approach, such as, standard
Smagorinsky (SS) scheme and dynamic subgrid scale (DSGS) scheme. In order to obtain
high accuracy and oscillation-free solutions, two step predictor-corrector explicit scheme,
second-order accurate in space and time, and slope limiter function minmod are
employed for solving the governing equations.
Performance of the developed hydrodynamic and solute transport models have been
assessed for turbulent flows in wide symmetric and narrow asymmetric compound
channels. A normal asymmetric compound channel conveying shallow and deep flows
was also considered for the aforesaid purpose. The comparison of primary velocity and
bed shear stress was performed based on past experimental data from literature using
statistical error measuring index, root-mean-square error (RMSE). A number of other
flow characteristics, such as secondary velocity, shear stress at horizontal interface,
subgrid scale (SGS) turbulent shear stresses in upper and lower layers, have been
simulated in aforesaid models. The simulated result on flow parameters like interfacial
VII
shear stress and vertical flow velocity gives insight about the overall momentum
exchange between the main channel and the floodplain of straight compound openchannel. The analysis of model results revealed the existence of horizontal vortices near
the junction of the main channel and the floodplain where momentum exchange takes
place. The dominance of shear stresses at the junction of the main channel and floodplain
is existed irrespective of channel cross-sections. The model reproduces the existence of
one inflection point in case of shallow flows and two inflection points in case of deep
flows. The significant difference in layered average primary velocities (in the range of 14% to +12% of Ub) between lower and upper layers is present near the junction of the
channel. Horizontal isolines plots of secondary currents (ranging between 0.02% of Ub to
1.1% of Ub); bed (varying between1.15
0.08 N/m2 to 0.4 N/m2) shear stresses indicated the formation of vortices near the
junction of the main channel and the floodplains. Variation of dynamic SGS coefficient,
C, (in the range of 0.02 to 0.145) for turbulent flows in wide symmetric and narrow
asymmetric compound channels, poses slight effects in flow characteristics in the
aforesaid channels.
Performance of solute transport model is also assessed using past experimental data on
symmetric and asymmetric compound open-channels. Transport processes are
investigated through injection of inert pollutant at various locations in compound openchannels. The model performs better for pollutant injected at other than shear layer
locations. The significant differences in distribution of solute concentration between the
computed and experimental results were observed due to under estimation of secondary
currents near the junction of main channel and floodplain. The performance of transport
model is better at downstream locations as compared to upstream locations due to better
mixing of pollutants. Further, the presence of vertical velocity component at horizontal
interface causes higher rate of distribution of concentrations vertically as compared to in
horizontal directions.
VIII
Contents
Acknowledgement
Abstract
VII
Contents
IX
List of Figures
XII
List of Tables
XVI
XVII
Chapter
Name
Introduction
Page
No.
1-6
1.1
General
1.2
1.3
1.4
Research objectives
1.5
Thesis organisation
Literature review
7-21
2
2.1
General
2.2
2.2.1
2.2.2
2.3
14
18
Concluding remarks
21
IX
22-52
3.1
General
22
3.2
22
3.2.1
Hydrodynamic equations
23
3.2.2
27
Method of solution
29
3.3.1
30
3.3.2
33
3.3.3
38
39
3.4.1
40
3.4.2
40
3.5
44
3.6
Predictor-corrector scheme
49
3.7
Stability conditions
50
3.8
Concluding remarks
52
3.3
3.4
53-117
4.1
General
53
4.2
Experimental data
53
4.2.1
53
4.2.2
54
56
56
56
69
98
98
4.3
4.3.1
4.3.2
107
4.4
Concluding remarks
116
5.1
5.2
118-131
118
118
5.2.1
118
5.2.2
119
120
5.3.1
121
5.3.2
127
Concluding Remarks
131
5.3
5.4
Conclusions
132-135
6.1
General
132
6.2
Hydrodynamic model
132
6.3
134
6.4
134
References
136
Research publications
142
XI
List of Figures
Figure
No.
Title
Page
No.
3.1
23
3.2
30
3.3
31
3.4
32
3.5
35
3.6
36
3.7
46
3.8
47
3.9
51
4.1
54
4.2
55
4.3
56
4.4(a)
59
4.4(b)
59
4.5
60
4.6
61
4.7
62
4.8
62
XII
4.9
64
4.10
65
4.11
67
4.12
68
4.13
69
4.14
72
4.15
73
4.16
74
4.17
75-76
4.18
4.19
78-79
4.20
80
4.21(a)
82
4.21(b)
82
4.22
84
4.23
85
4.24
XIII
77
86-87
4.25
88-91
4.26
92-94
4.27
94-95
4.28
96-97
4.29
99
4.30
100
4.31
100
4.32
4.33
103
4.34
104
4.35
105
4.36
106
4.37
107
4.38
108
108
4.39
101-102
4.40
109-110
4.41
111-112
XIV
4.42
112-113
4.43
114
4.44
115
5.1
122-123
5.2
124
5.4
125-126
126
127
5.5(b)
128
5.6(a)
129
5.6(b)
129
5.7(a)
130
5.7(b)
130
XV
List of Tables
Table
No.
Title
Page
No.
4.1
55
4.2
57
4.3
58
4.4
70
4.5
81
5.1
XVI
120
Latin Letters
A
A1
A
bl
b
bu
C
cf
cl
cu
dr
ds
dv
Dx
Dy
E
Eh
Ec
F
Fc
Fv
g
G
Gh
Gc
hl
XVII
hu
h (=hu+hl)
k
Ke
lo
Lij
Mij
substituted parameter for different STS and SGS stresses at test filter level,
in Eq.(3.56a);
n
nl
nu
nx
ny
Ni
Nj
sl
so
su
S
Sc
Sh
Sij
Sx
ul
um
uo
uu
u*
u'
U
Ub
Uc
Uh
U*
XVIII
vl
vm
vo
vu
wo
Zb
Greek Letters
XIX
Superscripts
p
*
p+1
Subscripts
b
c
h
i
ij
k
l
l
L
m
o
p
r
at bed
convective grid cell
hydrodynamic part
in ith direction (i=x, y)
ith face and jth direction
kth layer (k=u, l)
lower layer
laminar flow
left side of face of grid cell
mean value
at horizontal interface
grid cell point
face of grid cell
XX
R
s
u
v
Abbreviations
AGC
CFL
DNS
DSGS
ENO
Exp.
FLDA
FVM
LDA
LES
MUSCL
RANS
RMSE
SGS
SS
STS
TVD
XXI
Chapter 1
Introduction
1.1 GENERAL
The natural rivers are, invariably, compound in cross-sections, consisting of deep main channel
flanked by floodplains on either or both sides to accommodate high flows during the floods. The
rivers have become more prone to pollutant and sediment overloading due to extensive
urbanisation across the globe. The characterisation of hydraulic parameters is of immense
importance in developing transport models of organic and inorganic pollutants for selection of
water intake locations, and prediction of erosion and deposition of sediments in natural channels.
Better understanding and complete control over transport phenomena are the essentially required
for many industries such as water sector, oil and chemical sectors and thermal sector in the
interest of environmental and economic benefits.
The flow velocity in the floodplains is lower than main channel due to shallow water depth. As a
result of the velocity gradient, shearing occurs at the interface between the main channel and the
floodplains, leading to the flow patterns most of which are characterized by large-scale vertical
structures with vertical axes (macro-vortices). Mixing and transport processes of solutes in
compound channels are mainly controlled and regulated by the exchange of mass and
momentum between the main channel and the lateral floodplains. Coherent structures are
commonly recognized as major agents of mass transport, owing to their ability to trap mass and
inhibit exchanges among different flow regions.
though give the precise results, may have their limited applications while applying the same for
long reach of natural rivers due to their extensive computational requirements. The 2-D models,
under such circumstances, may be very useful for monitoring the flow and transport
characteristics of natural channels with large reaches. Brief description of recent work
undertaken in the foregoing theme is included in following paragraphs.
Numerous works on straight compound channel flow have been reported in recent past using
analytical methods (Tang and Knight, 2008; Yang et al., 2013), laboratory experiments
(Rajaratnam and Ahmadi 1981; Knight and Hamed 1984; Tominaga and Nezu 1991; Nugroho
and Ikeda, 2007; Fernandes et al., 2014) and numerical simulations (Naot et al., 1993; Thomas
and Williams, 1995; Sofialidis and Prinos, 1998; Nugroho and Ikeda, 2007; Joung and Choi,
2008; Cater and Williams, 2008; Kara et al., 2012; Xie et al., 2013) in investigation of flow
characteristics and mechanism of lateral momentum transfer from main channel to floodplains.
With the advent of high-speed computers, the study on two and three-dimensional flows in open
channels had experienced surge of interest in recent years. A number of numerical studies on
compound open-channel hydraulics, mainly based on the Reynolds-averaged Navier-Stokes
(RANS) equations, have been reported in the literature. Naot et al. (1993), Pezzinga (1994), and
Sofialidis and Prinos (1998) solved three-dimensional form of equations and showed that their
models are capable of predicting reasonably well the time-averaged primary flow including bed
shear stress distributions for different compound open-channel geometries. Pezzinga (1994) and,
Sofialidis and Prinos (1998) were able to reproduce the primary and secondary flows quite well
using nonlinear turbulence closure schemes. However, the bed shear stress predictions were less
accurate (Pezzinga, 1994), due to implementation of wall functions. The large eddy simulation
(LES), an alternative to RANS modelling, resolves the large-scale turbulence, and hence, allows
an accurate computation of flow characteristics. Thomas and Williams (1995) were the first to
employ LES in three-dimensional flow to study the flow and turbulence characteristics in a
compound open-channel. Recently, Cater and Williams (2008) and Kara et al. (2012) also
2
successfully applied the LES using 3-D equations in the simulation of compound open-channel
flows. Practical applicability of commercially available 3-D hydrodynamic models in practical
engineering problems is due to their extensive computational efforts and generic utilisation.
Owing to such difficulties, simplified two-layer models, in which each layer can be considered
two-dimensional, vertically homogeneous, and depth-averaged with a horizontal interface, could
be very handy in resolving the practical issues of compound channels.
Chau and Jin (1995) developed a two-layered, two-dimensional model based on finite difference
approach using CFD-based numerically generated boundary fitted orthogonal coordinate system,
and used to simulate density stratified unsteady flows in a natural water-body with complex
topography. Later on, Jin et al. (1998) applied the model to simulate turbulent flows in a
sinusoidal meandering compound open channel. In both, above-mentioned numerical studies
(Chau and Jin, 1995; Jin et al., 1998), grid block technique was used wherein the boundary
fitted orthogonal curvilinear grids were generated in the physical plane, and the same was
transformed into rectangular grids in the computational plane resulting in complicated
mathematical formulations with additional transformation relations. Prandtls mixing length
approach was used for modelling turbulent shear stress terms.
The 2-D depth averaged numerical models have also been in use in past due to their less
computational efforts in prediction of inert pollutant transport with relevant turbulent closure
schemes in compound open channel flows. Djordjevic (1993) presented a two dimensional
depth-averaged numerical model using the operator-splitting approach for unit Schmidt number
for solute transport in both streamwise and lateral directions in steady open channel flows. Wood
and Liang (1989) developed a 2-D semi-analytical model using eigen-function expansion
technique for prediction of tracer concentration in compound open channel flows. Chatila and
Townsend (1998) developed a 2-D finite-difference based model for simulating inert pollutant
transport in compound open-channels using constant eddy-viscosity model, and accounted for
lateral momentum transfer (LMT) feature at horizontal interface of main channel and
floodplains. Fraselle et al. (2008) carried out experimental studies in a symmetric compound
laboratory flume for diffusion and dispersion of tracer. Fraselle et al. (2010) compared their
results with commercial CFD software TELEMAC-2D and showed the role of turbulent eddies
in solute transport.
Existing numerical models have their own advantages and disadvantages in prediction of
secondary flows and solute concentration in compound open-channels. Most of the reported
studies have focused their attention mainly on vertical or inclined interface at the junction of
main channel and floodplains whereas a little attention has been paid to the flow characteristics
along the horizontal interface. Also, the past studies did not analyze the influence of flow depth
ratio and cross-sectional channel characteristics (wide or narrow) on transverse mixing of
pollutants.
Very few publications deal with the transport of pollutants in rivers with compound sections. It
can be argued that a 3-D model is required for more precise simulation of the flow field.
However, the effort and cost in developing a 3-D model would outweigh the improvement
gained over a 2-D model simulation. Thus, a 2-D model is generally considered to be a
reasonable compromise between simplified 1-D simulation and difficult 3-D models.
Invariably, the river systems are compound in cross-sections, and being overloaded with
pollutants due to rapid land use changes in their respective catchments. The prediction of
hydraulic characteristics in natural river systems is required in assessment of their pollutant
transport capacity along the flow. Due to rapid advancement in computational fluid dynamics
(CFD) and its popularity vis--vis physical modelling, the writer is prompted to take up the
problem of hydrodynamics and solute transport in compound channel using two layer 2-D
models with LES closure scheme.
Keeping view the research gap and suitability of two layer 2-D model in prediction of hydraulic
and transport characteristics in compound open-channel, present study has been planned with
following research objectives:
a) Development of two layer 2-D hydrodynamic numerical model using LES approach for
straight compound open-channels.
b) Validation of developed numerical model using past experimental data on symmetric and
asymmetric compound channels.
c) Development of two layer 2-D solute transport model using LES approach for straight
compound open-channel.
d) Validation of solute transport model using past experimental data on symmetric and
asymmetric compound open-channels.
The brief description of PhD Thesis, comprising six chapters, are described in following
paragraphs:
Chapter 1 provides the background information and brief literature review on hydrodynamics
and transport phenomena involved in compound channel flow. The motivation, research
objectives and thesis structure are described.
In Chapter 3, theoretical aspects dealing with two-layered two dimensional mathematical models
and their numerical solutions for hydrodynamics and transport of inert solute are presented. Two
types of turbulence schemes, viz., standard Smagorinsky (SS) and dynamic subgrid scale
(DSGS) closure schemes, under large eddy simulation (LES) technique, are discussed.
In Chapter 4, numerical model results on hydrodynamics are presented in both symmetric and
asymmetric compound open-channel flows. The results obtained from both standard
Smargorinsky (SS) and dynamic subgrid scale (DSGS) model are discussed in the chapter. The
performance of present numerical model is explored with respect to laboratory data from
literature. Further, flow properties at horizontal interface of the channels are also explored in
detail.
In Chapter 5, numerical model results are presented on inert solute transport in both symmetric
and asymmetric compound open-channel flows. The results obtained from dynamic subgrid scale
(DSGS) model coupled with hydrodynamic equations are presented in this chapter. The
performance of developed numerical model is explored with respect to laboratory data from
literature.
In Chapter 6, the key conclusions derived from foregoing study on hydrodynamics and transport
numerical models are presented. Also, the directions for future work on selected theme are
included.
Chapter 2
Literature Review
2.1
GENERAL
In nature, rivers are made up of a centrally located main channel and one or two floodplains
extending laterally away from it. A thorough understanding of the flow interaction between the
main channel and the floodplains is essentially required for river management and related works.
Since 1960s, compound channels have been studied extensively through both laboratory and
numerical model studies. Previous studies related with hydrodynamic and solute transport
phenomena in compound open-channel flows are presented in this chapter.
2.2
In this section, different phenomena associated with hydrodynamic and solute transport model of
straight compound open channels are reviewed.
Prior to early sixties, very little attention was paid on complex flow patterns which exist between
the main channel and the floodplains in compound channels, but, later on developments have led
to a clearer understanding of hydraulic mechanism involved at the level of model studies. Prandtl
(1925) proposed a logarithmic velocity profile for velocity distribution in pipe flow using mixing
length hypothesis. Subsequently, Von Karman (1930) proposed a velocity distribution model
7
using similarity hypothesis. A general form of velocity distribution law was developed by
Prandtl (1932) which is generally considered as the Prandtl-Von Karman velocity law. Chow
(1959) discussed open channel hydraulics problems and related structures in his classical
literature that describe fundamental principles in open-channels. Numerous researchers have
studied in detail about the flow phenomenon in compound open channels both, experimentally
and numerically, afterwards. Further, in pursuance of analysing the complex flow features in
straight compound channel, various types of laboratory channels were used under different
conditions. Similarly, in order to fully analyse the flow phenomena of straight compound open
channels, numerous numerical methods were proposed. In furtherance of these researches, the
laboratory produced results have been compared with simulated results of straight compound
open-channel flows. Based on these researches, it is established that the flow structure in straight
compound open channels is extremely complex in nature which induces significant mass and
momentum exchange, and formation of vortices. One can obtain one-dimensional (1-D) SaintVenant equations by integrating along the depth and width from the Navier-Stokes equations (NS); and, similarly, by integrating along the depth and assuming hydrostatic pressure distribution
in vertical scale, two-dimensional (2-D) shallow water equations (SWE) are obtained. Since the
basic flow equations describing the open channel flows are non-linear and coupled and, as a
result, it is difficult to solve them analytically. However, only in small number of cases, it is
possible to obtain an analytical solution of the Navier-Stokes equations under certain
assumptions. In order to solve or close turbulent shear stress terms, several model are considered
and these are (i) Boussinesq eddy viscosity approach (ii) Mixing length hypothesis (iii) Zero
equation models (iv) One equation models (v) Two equation models (vi) Multi-equation models
(vii) Sub-grid-scale (SGS) models (viii) Direct numerical simulation (DNS) and extension
thereof.
The hydrodynamic behaviour of compound channel flow is composed of the averaged wall
shear stress and the continuous mass exchange between the floodplain and the main channel.
The intensity of vortex formed near the junction of the floodplain and the main channel
corresponds to continuous mass exchange between them. In addition to above, a fully
developed turbulent flow in compound open-channels is specified by the momentum transfer
from the faster flow in the main channel to the slower flow in the floodplain. The shear layer is
8
developed at interface of the main channel and floodplain, and the secondary currents are
generated at the corners of channel cross-sections due to bed-generated turbulence and freesurface effects. The presence of such co-existing phenomena results into a very complex flow
field in straight compound open channels.
The hydrodynamics of two stage open channels of symmetric type, was first investigated in the
laboratory by Sellin (1964) wherein momentum transfer mechanism between the deep main
channel and the shallow floodplains along with the presence of vortices and their effect on
velocity and discharge were demonstrated using flow visualisation technique. He described
phenomenon of formation of vertical vortices due to transfer of fluid from the main channel to
the floodplains with high momentum. The studies were performed in a 4.5 m long symmetrical
compound open-channel. He was not able to estimate overbank flow which depends upon
empirical weighing of roughness number requiring sizable amount of experimental data. He also
opined that numerical expression for the effect of momentum transfer mechanism depends upon
the effect of variations of scale, cross-sectional shape, and floodplain and channel roughness. He
inferred that lateral momentum exchange from the main channel to the flood plains was
significantly increased which, ultimately, caused reduction in the main channel conveyance and
increase in flow resistance in the channel. In addition to the vertical vortices due to the
anisotropy of turbulence, generation of secondary currents was observed in the longitudinal
direction which played an important role in the momentum exchange, especially near the
junction.
Zheleznyakov (1971) carried out investigations into the problem of main channel and floodplain
interaction using a 5.2 m wide laboratory flume having parabolic shaped flood channels of 0.45
m and 0.6 m width on either side. The tests were performed for discharges ranging from 0.0069
cumecs to 0.0514 cumecs under bed slopes ranging from 1/1000 to 1/2000. He reported the
effect of turbulent flow mixing between the main channel and the floodplains, and inferred that a
significant reduction of flow occurs in the main channel. The formation of vortices with vertical
axes and turbulent eddies at the junction of main channel and floodplain were also observed by
9
him. Based on his experiments, it was also inferred that interaction of flows between the main
channel and the floodplain is inversely dependent on the depth of floodplains as the interaction
of flow between the main channel and the floodplain becomes weaker with increasing water
depth. He named such a momentum-exchange phenomenon as the kinematic effect. Similar
observations were also reported by Myers and Elsaway (1975), and Myers (1978).
Ghosh and Jena (1971) investigated shear stress distribution for rough and smooth sidewalls in
symmetric compound flume of length 8.5 m, main channel width 0.203 m and floodplains width
0.076 m. They obtained boundary shear distribution in the flume for different flow depths using
combined technique of Preston tube and Patel calibration (Patel, 1965). They observed that
boundary shear in the flume gets redistributed due to roughening of sidewalls and bed of the
channel. Such redistribution of boundary shear was also reported by Ghosh and Mehta (1974)
who conducted laboratory experiments with different compound channels.
Rajaratnam and Ahmadi (1970, 1981) carried out laboratory tests in a symmetrical compound
channel of 18.29 metres long with main channel width of 0.2032 metres flanked by two
floodplains, each 0.508 metres wide to manifest interaction mechanism in the channel. They
developed equations based on their experimental results to express velocity and shear stress
profiles in main channel and floodplain. They also showed that mixing zone is extended by a
distance of 6 times the bankfull depth across main channel and floodplain.
Price (1975) showed that turbulent shear occurring between the main channel and the floodplain
is dependent on difference of mean velocity between them. Based on his experiments carried out
in a laboratory flume, he opined that shear layers should be considered as vertical and be
regarded as rough surfaces.
Myers (1978) measured shear stress distributions around the periphery of a complex channel,
consisting of a deep section and one flood plain and the results were used to quantify the
momentum transfer due to interaction of flows between the main channel and floodplain.
10
In order to explore the interaction of flows and shear forces at different levels between the main
channel and the floodplains, Knight and Demetriou (1983) conducted experimental studies on
symmetrical rectangular compound open channel flume of length 15 m, at a constant bed slope
of 9.66 x 10-4, having 0.152 m wide and 0.076 m deep centrally located channel along with 0.229
m wide and 0.076 m high two floodplains on either sides. Flow discharge and velocity were
measured using a differential type Venturi meter, and a Novar stream flow miniature propeller
current meter, respectively. Shear stresses were measured using a Preston tube. Percentage of
flows carried by different sections and percentages of shear force distributions at different
interfaces of the channel were calculated for different relative water depths [(H-h)/H=0.5, 0.4,
0.33, 0.25, 0.2, 0.1] and relative widths [B/b=4, 3, 2, 1] of the main channel and the floodplain.
Here, water depth in main channel, H is 0.152 m; water depth in floodplain, h is 0.076 m, total
width of channel, B is 0.305 m, and width of main channel, b is 0.076 m. Among different
interfaces, they also calculated apparent shear forces and discharges in terms of percentage along
the horizontal interface. The horizontal interface was considered by dividing the whole channel
section into two parts; one part as the lower main channel and the other part as the upper main
channel and floodplains taken together. It was evident from the experiments that, at low relative
water depths and high relative channel widths, values of shear force were positive which
indicated retardation of flow in the lower main channel due to the flow in upper region. At high
relative water depths, the values of shear forces turned out to be negative indicating that the flow
in the lower main channel is accelerated due to flow of the upper region.
Knight and Hamed (1984) carried out experimental studies and presented boundary shear stress
distributions at different sections of a symmetric compound channel to quantify momentum
transfer mechanism in terms of apparent shear force acting on channel interface. He studied the
progressive roughness of floodplains in six steps to observe the influence of differential
roughness between floodplains and main channel on lateral momentum transfer process.
Empirical equations were also presented for shear force on floodplains in terms of percentage of
total shear force. Supplementary relationships were established for apparent shear forces on
vertical, inclined and horizontal interfaces within the cross section. The influence of momentum
transfer between sub-areas on vertical and lateral distribution of longitudinal velocity was also
assessed.
11
Formation of vertical vortices and helical secondary currents at the junction of main channel and
floodplains was also observed by Shiono and Knight (1989) in their experiments carried out in
the SERC Flood Channel Facility (FCF) at Hydraulics Research Station, Wallingford, UK
wherein they had undertaken secondary current measurements using Laser Doppler Anemometer
(LDA). With experimental observations, they inferred the importance of geometry of wide
symmetrical compound channel in influencing the structure of secondary currents in main
channel. Shiono and Knight (1989) further showed that a relatively smaller secondary cell of
clockwise is formed near left sidewall of main channel apart from two larger secondary cells in
upper and corner region of main channel. The FCF, a large scale facility having channel length
and width of 60 m and 10 m, respectively, was constructed in 1986 for experimental
investigations of overbank flows in rivers. It was configured as rigid or mobile bed channel as
per requirement. Phase A of the facility mainly included on straight and skewed fixed bed
compound channels flows, and experimental results were presented by Ackers (1993) who
introduced the coherence concept in account of the interaction effect between the main channel
and the floodplains.
Tominaga and Nezu (1991) carried out experimental studies in a laboratory flume composed of
painted iron plates channel bed and smoother glass plates side walls to investigate secondary
current patterns, turbulent structure of flows such as turbulent intensities and Reynolds stresses
in straight asymmetric compound open-channel of length 12.5 m and width 0.4 m, under various
relative water depth ratios, and reported a 3-D data set using Fiber-optic Laser Doppler
anemometer (FLDA). The data set was used by several researchers (Naot et al., 1993; Kara et al.,
2012; Xie et al., 2013) for examining the validity of their numerical model of compound openchannel flows. They reported turbulent flow characteristics of the channels for different water
depth ratios, and showed the presence of strong inclined secondary current on momentum
transport generated near the junction of the main channel and the floodplain. They inferred that
secondary currents along with vertical vortices in the longitudinal direction, played significant
role in the momentum exchange near the junction region. They also reported the strength of
secondary current as about 4% of the maximum longitudinal velocity (Umax) whereas other
researchers (Naot et al., 1993; Nezu, 1996), reported it as 2-3% of the maximum longitudinal
12
velocity (Umax). They also inferred that component of turbulence intensity increases near the
junction of the main channel and floodplain.
Nezu and Nakayama (1997) experimentally obtained detailed information about 3-D flow
structures in compound open channel using both Particle Imaging Velocimetry (PIV) and LDA.
In their works, they observed presence of strong interaction between upward flows and
horizontal vortices at the junction of the main channel and floodplain.
Nezu et al. (2004) developed a new dual-layer PIV system to analyze the velocity distribution at
different elevations simultaneously, and investigated 3-D coherent properties of horizontal
vortices, convection mechanism and transverse momentum exchange between the main channel
and the floodplain for shallow and deep water depths.
Sanjou et al. (2010) conducted an experimental study on a large scale compound open channel
flume using acoustic Doppler velocimetry (ADV) to reveal space correlation properties and
velocity fluctuations between the main channel and floodplains. Based on their experimental
observations, they concluded that two layer model can be very useful to understand the spanwise
flow features and interaction between the upper and the lower layers in the main channel.
Stocchino and Brocchini (2010) presented a statistical analysis of the properties of quasi-2D
macro-vortices which form at the interface of a compound open channel under quasi uniform
flow condition, and showed that these structures remain constant in the streamwise direction.
Stocchino et al. (2011) conducted extensive experimental studies for showing Lagrangian mixing
process and quasi-2D macro-vortices properties in a uniform compound open channel flow by
assuming a shallow water conceptual model for varying physical flow parameters like flow depth
ratio rh [=H/(H-h)] and Froude number. Here, H=total flow depth and h=main channel flow
depth.
Fernandes et al. (2014) carried out experimental studies in laboratory compound channel to
investigate longitudinal flows and, corresponding mixing and boundary layers for six uniform
flows generated in different locations of the channel. They reported effects of shallowness and
13
floodplain roughness on longitudinal streamwise velocities and lateral shear stresses. They also
reported relationship of mixing layers developed with the growth rate for different roughness in
channel sections.
Hydrodynamic behaviour of flows in four straight compound channels was reported by Naot et
al. (1993) using 3-D steady state model in conjunction with k- turbulence model and algebraic
stress model developed by Naot and Rodi (1982). Special wall functions were considered in
treatment of channel walls. They inferred that smaller secondary currents influence flow
behaviours, such as velocity and boundary shear stress in channels. Simulated results, such as
streamwise velocity, secondary flow, shear stress and mass exchange between the main channel
and the floodplains were compared with experimental results of Tominaga and Nezu (1991).
They also inferred that large differences in Reynolds number of flows between the floodplain
14
and main channel were one of the reasons for discrepancies occurred between the model results
and the experimental observations.
Pender and Manson (1994) implemented a k- turbulence model in a set of 3-D flow equations
and tested the same against the results obtained from McKeogh and Kiely (1989). The model
was used for predicting water levels and flow velocities in a compound channel.
Smagorinsky (1963) derived an eddy viscosity subgrid scale (SGS) model based on a Boussinesq
approximation and obtained Smagorinsky coefficient. The classical Smagorinsky model is valid
only if scales of size on the order of filter width are within the inertial range, which are larger
than the dissipative scales, yet smaller than the scales that contain of the energy. Large eddy
simulation (LES) of classical Smagorinsky model along with the application of van-driest
damping function for solid walls was first applied by Thomas and Williams (1995) to simulate
complex turbulent structures in a straight compound channel under Reynolds numbers of 42,000
and 43,000. They reported reasonable match of time averaged streamwise flow velocity and
secondary flows with measured data of the SERC Flood Channel Facility at Hydraulics Research
Station, Wallingford, UK, but failed to represent surface depression effect in the channel. They
reported generation of bottom shear stresses, turbulence intensities and turbulent kinetic energy
of same magnitude as that of experimental results but locations of occurrence of their peaks were
different. They inferred that, due to isotropic nature of Smagorinsky (1963) model, the model
was not able to capture secondary flows in channels; and, due to coarse mesh resolution, the
model over-predicted mean velocities by about 8% near the mid of the main channel.
viscosity. The turbulent exchange was treated empirically across the interface of the flow
domain. The hydrodynamic model was applied for Tolo Harbour, Hong Kong. They reported
velocity hodographs showing different patterns of tidal circulations at various positions in each
layer. Based on trends of computed Lagrangian pathline they inferred that tidal excursion was
dependent on types of tide in flow domain. In order to capture the interaction between turbulence
and mixing across the interface, they suggested for use of more sophisticated turbulence model.
They also suggested incorporation of transport equations in their model as future scope of work.
Their models were used by Jin et al. (1998) for flow simulation in a meandering compound open
channel.
Nadaoka and Yagi (1998) reported characteristic of large eddies in compound open-channel
under different flow depth conditions using Sub-Depth Scale, SDS-2DH model. In model, eddyviscosity approach, corresponding to Sub-Depth Scale turbulence, was used to model small-scale
turbulences and reproduce mechanism of eddy generation. Wavelength of vortex obtained from
model was compared with experimental results. They also predicted average flow velocity and
bed shear stress profiles in shear layer. It was reported that the model under predicted the flow
velocities near mid of the main channel for relative depths < 0.15 and over predicted the same for
relative water depths > 0.20. It was also reported that flow velocities on floodplains was underpredicted for high relative water depth conditions.
Sofialidis and Prinos (1998) modified low-Reynolds, nonlinear k- turbulence model developed
by Craft et al. (1993) to predict simple flows, and applied the same to study compound open
channel flow with low relative depths (the ratio of water depth on the floodplain to the total
depth). They reported the prediction of turbulent parameters such as turbulent shear stresses and
turbulent kinetic energy and, flow interaction between main channel and floodplain. The model
was not able to generate secondary currents of desired strength in open channel. They stressed
the use of nonlinear stress-strain relationship for representing anisotropy of Reynolds stresses at
the free surface and dissipation equation for prediction of turbulent kinetic energy. Sofialidis and
Prinos (1999) further reported development of k- model, and applied it in straight compound
channel with different relative flow depths. They applied redistribution mechanism for normal
turbulent stresses and simulated eddy viscosity, t, near free surface. They reported success in
16
reproduction of streamwise velocity, secondary currents and turbulent shear stresses from the
developed model but the model was unable to simulate depression of the maximum velocity near
free surface. They concluded that, with increase in relative flow depth, lateral shear at interface is
intensified for an increased range of flow depth in floodplain. Further, the presence of weaker
secondary currents causes the velocity-maxima below the free surface and turbulence levels are
enhanced in the region of interaction.
Jin et al. (1998) applied two-layer two-dimensional mathematical model of Chau and Jin (1995)
for unsteady flows in a sine-generated meandering compound channel, and evaluated the change
of flow patterns such as water level, velocity vector, Lagrangian pathline and friction factor in
terms of water depth ratio between the main channel and the floodplains. Empirical treatment
was proposed for approximation of turbulent exchange across the horizontal interface. They
inferred that shallower floodplain flow was affected severely by meandering flow of main
channel, but it was independent of water depth ratios greater than 0.9.
Shi et al. (2000) reported development of 3-D model for Large Eddy Simulation (LES) of
turbulent flow in symmetric straight compound channels using Series-A experimental
programme of the UK FCF, and compared the results with algebraic stress model. They reported
computational difficulties in using these models and inferred that the models are helpful in
replicating physical processes like intrinsic mechanisms of flow and accurately predicting the
flow features including discharge capacity in straight compound channels. They also inferred
that their models are capable of determining the impact of flow on other processes such as scour
and pollution transport.
Othman and Valentine (2006) developed 3-D model in Cartesian coordinate system for uniform
flow in compound open channel of the UK FCF using non-linear k- (NKE) turbulence model
and used Reynolds stress components for generation of secondary currents. The SIMPLE
technique was used to approximate pressure terms. They reported prediction of primary mean
velocity, secondary currents, bulging of contours at bottom corner of the main channel,
inclination of contours near free surface towards the channel centre and, depression of the
17
maximum velocity below free surface along with formation of vortices in the main channel and
floodplains.
Zarrati et al. (2008) derived semi-analytical equations for streamwise vorticity along with
secondary Reynolds stresses for shear stress distribution in straight open channels of different
cross-sections, such as rectangular, trapezoidal and compound cross-sections. They reported
evaluation of relative shear stress distribution equation along the width of different channel
cross-sections using experimental data, and investigated the effect of secondary flows due to
shear layer between the main channel and floodplains.
Kara et al. (2012) reported model results of large eddy simulation (LES) of turbulent flow in an
asymmetric compound open channel with deep and shallow flow depths and the results were
compared with experimental data of Tominaga and Nezu (1991). They also reported the effects
of water depth ratio of main channel and floodplain on primary and secondary flows, secondorder turbulence statistics and anisotropy of turbulence. They demonstrated that anticlockwise
rotated vortex pair along horizontal axes caused velocity bulge at interface of the main channel
and the floodplain. They also reported that resulting secondary currents as well as primary and
secondary Reynolds stresses contributed to the streamwise momentum transport.
Xie et al. (2013) employed LES approach with dynamic SGS model using partial cell treatment
technique in investigating turbulent structure, large-scale vortical structures and instantaneous
secondary flows in an asymmetric compound open-channel which proved to be significant
parameters in flow resistance and sediment transport fluxes. They reported comparison of mean
velocity and boundary shear stress distributions, secondary currents, and turbulence statistics
with experimental data of Tominaga and Nezu (1991). They inferred that vortical structures can
cause significant lateral exchange of mass and momentum in compound channels.
Flow characteristic of the rivers during the flood, and transport of sediment and pollutant in
natural systems are the sole reasons to study the channel with compound sections. Heat and mass
18
transfer induced transport phenomena, such as convection and diffusion play a vital role in
human life. Complete understanding of transport phenomena is essential for many industrial
applications like aerodynamic, flows in furnaces, heat exchangers, and chemical reactors. In
nature, both advection and diffusion are responsible for transport processes to occur in natural
rivers or channels with high water. Hydrodynamic processes in such rivers tend to be complex
due to complex geometry and bed roughness conditions. Numerous research works have been
undertaken till date to understand this complex hydrodynamic behaviour. However, a few works
were only undertaken to reveal the effect of hydrodynamic process on mass transport. The
traditional approach to investigate such processes is based on observations and experiments
which have limited scope of obtaining complete information about the processes. The complex
transport phenomena expressed in terms of complex mathematical equations can only be solved
using numerical methods. Research works related to solute transport in straight compound openchannels undertaken till date are reviewed and presented in following paragraphs.
Lin and Shiono (1995) developed 3-D model by solving Navier-Stokes equations in conjunction
with linear and nonlinear k- turbulent schemes to investigate transport processes of solute in
compound open channels. They reported comparison of flow velocity and the solute transport
rates with experimental works undertaken by Wood and Liang (1989). With the help of both
linear and non-linear k- turbulent approaches they showed influence of secondary currents on
the mixing processes in the compound open-channel.
Simoes and Wang (1997) reported three-dimensional numerical model in conjunction with two
algebraic eddy viscosity schemes to simulate time-dependent turbulent flows and transport of
dissolved materials in compound open channels. Accuracy along with advantages and drawbacks
of the two models were reported by them based on comparison of simulated results with
experimental data.
Chatila and Townsend (1998) presented 2-D finite-difference mathematical model using depthaveraged versions of 3-D Navier-Stokes equations in conjunction with advection-diffusion
equation and constant eddy-viscosity model to simulate inert pollutant transport in compound
open channels. They employed Platzman's space staggered scheme and double time step
19
operation in their numerical solution. They reported a comparison of model results with
laboratory data obtained from continuous-injection dye tracer experiments in asymmetrical
compound open channel at Hydraulics Laboratory, University of Ottawa.
Shiono et al. (2003) reported numerical model studies using two turbulence closure schemes, viz.
k- model and algebraic stress model (ASM) for prediction of solute transport injected at three
injection points near the water surface for deep flow depth in a narrow asymmetric compound
open channel of length 20 m and width 0.2 m. They put forward a comparison of simulated
results with experimental work of Shiono and Feng (2003) undertaken in laboratory flume using
sophisticated measuring equipments such as laser Doppler anemometer (LDA) and laser induced
fluorescence (LIF). Fluorescence dye (Rhodamine 6G) was used as solute material injected at 13
m from the channel entrance and measured at 1 m downstream of the injection point. They
reported a skewed distribution of solute on floodplain and inferred that algebraic stress model
(ASM) performed better than k- model. They investigated the cause of skewed distribution
through the variations of secondary flow and eddy diffusivity. They also inferred that, in deep
flow case, the effect of secondary currents on peak concentration in shear layer was significant.
They further reported the presence of secondary current cells of large magnitude in upper region
and smaller magnitude in left corner of the main channel. They demonstrated that, in solute
concentration distribution, the role of turbulent Schmidt number was significant in comparison to
eddy viscosity.
Laboratory experiments for solute transport in a wide symmetric compound open-channel were
undertaken by Fraselle et al. (2008). Total length and width of channel were 10 m and 1.2 m
respectively. Width of main channel and floodplains on either side was equal to 0.4 m each. A
constant discharge of 0.01855 m3/s in compound channel at longitudinal slope 0.001 resulted rise
of water level up to a height of 0.0756 m in main channel for a subcritical flow condition
(Fr=0.4). Flow measurements were undertaken using a pitot tube. Fraselle et al. (2010) reported
depth-averaged primary flow velocity profile of the experiments while comparing results of
numerical models of Telemac-2D software for two turbulence schemes, i.e., k- and Elder
schemes. Diluted common salt (NaCl) was used as solute material. Total four experiments were
undertaken for transport of solute by injecting solute material at different places in main channel
20
21
Chapter 3
3.1 GENERAL
In this chapter, the governing equations used in formulating hydrodynamic and transport flow
models in straight rectangular compound channel are presented. A rectangular compound
channel has been represented in the form of a two layer system separated along the horizontal
interface at junction of the main channel and the floodplains, considering lower main channel as
bottom or lower layer, and upper main channel and floodplains taken together as top or upper
layer. Basic governing equations for lower and upper layers are described separately.
Subsequently, the partial differential equations describing the transport of inert pollutants in the
compound channels are presented. Also, the numerical scheme and its stability and relevant
boundary conditions being used in solution of aforesaid equations are presented.
The turbulent flows in natural channels can be solved using simplified two-layer twodimensional models. Such considerations simplify the analysis and yields reasonably good
results. Two-dimensional shallow water flow equations are obtained by integrating threedimensional equations of motion along the depth individually for each layer. Derivation of twolayered two-dimensional equations for body-fitted coordinate system was presented by Chau and
Jin (1995) and Jin et al. (1998). These equations have been transformed into Cartesian coordinate
system and used in present study for simulation of hydraulic variables and inert pollutant
22
transport. The schematic of two-dimensional and two layered flows system of a typical
compound channel is shown in Fig. 3.1.
Conservation of mass and momentum of an incompressible fluid over depth of each layer can be
expressed in terms of partial differential equations, are derived in terms of depth-averaged
equations. For such two layered system, the position of interface is considered horizontally in
between upper and lower layers. In shallow water flows, where interface level is at or lower than
upper layer bed, the lower layer vanishes and only upper layer exists.
The two layered two dimensional hydrodynamic equations for shallow water flows in Cartesian
co-ordinate system with consideration of mass and momentum exchange between the layers are
derived in the same way as suggested by Abbott (1979) for depth-averaged formulations. The
layer averaged continuity and momentum equations for each layer are described in following
paragraph. The detailed derivation of the equations is available elsewhere (Chau and Jin, 1995).
and gravitation. This assumption is valid for gradually varied flow. Although some
details are lost in the vicinity of sharp gradients if the shallow water equations are used.
The overall results are adequate for engineering purposes (Cunge, 1975).
2. Density in each layer is constant and equal, i.e., variation in density between the layers is
neglected.
3. Coriolis force is negligible.
Equations for upper and lower layers can be expressed using Eqs. (3.1) and (3.2) respectively as
(3.1a)
(3.1b)
(3.1c)
(3.2a)
(3.2b)
(3.2c)
Here, subscripts u and l designate the variables corresponding to upper and lower layers,
respectively. In other subscripted variables, symbols appearing as o, s and b denote the
position at interface, surface and bed, respectively. Here, hk, uk and vk represent filtered water
depth, streamwise and spanwise layered average velocity, respectively in kth layer;
average filtered turbulent stress at ith face in jth direction in kth layer;
and
is layered
are filtered
shear stress in ith direction at surface, interface and bed respectively; uo, vo and wo are filtered
velocity at interface in streamwise, spanwise and vertical directions respectively; = mass
24
density of water; so= channel bed slope in streamwise direction; and g = gravitational
acceleration.
interface of the two layers appearing in governing equations (Eqs. 3.1-3.2) is expressed
empirically which is similar to that adopted in Chau and Jin (1995).
The filtered bed shear stresses are represented empirically similar to depth averaged method as
;
where,
(3.3)
where,
,
(3.5)
and total depth of water, h=hu+hl ; and cf , the coefficient of bed resistance, can be determined
using
(3.6)
where, n is the Mannings roughness coefficient.
Here, horizontal interface is considered at the top of lower layer.
The filtered shear stresses at interface between the two layers are approximated empirically as
;
where,
(3.7)
; and
and
25
The filtered horizontal interfacial velocities in streamwise and spanwise directions can be
expressed approximately (Chau and Jin, 1995) as
,
(3.9)
Here, vertical interface velocity, wo, considered positive upward, resulting due to convective
exchange between the layers, is determined from the layer-averaged continuity equation (Eq.
3.1a or 3.2a) of either of the layers after obtaining flow velocities and water surface levels of
respective layers.
Eqs. (3.1) and (3.2) can be written in divergence or strong conservative form as
(3.10)
where Uh, Eh, Gh, and Sh are row vectors, defined as
In Eq. (3.10), Uh is the state vector, Eh and Gh are the vectors representing flux terms and Sh is
the vector representing source term.
26
According to Fick's laws of diffusion developed by Adolf Fick in 19th century (Mehrer and
Stolwijk, 2009), the solute flux due to diffusion is proportional to the solute concentration
gradient, and the rate of change of solute concentration at a point in space is proportional to the
second derivative of its concentration with space.
Mathematically, for molecular processes, flux is specified by Ficks law, and diffusion equation
can be written in Cartesian coordinate in two-dimension as
(3.11)
where, Dx and Dy are the diffusion coefficients in x- and y-directions, respectively. In case of
molecular diffusion, Dx=Dy=D.
The processes of advection, caused due to mean motion of the fluid, and diffusion are separate,
however, additive in nature. The rate of mass transport by the velocity through a unit area in
perpendicular direction (x-direction) is given by the quantity (u.c). The total rate of mass
transport per unit area in perpendicular direction, sum of advection and diffusive fluxes, in two
dimensions is represented as
(3.12)
While substituting Eq. (3.12) into the equation for conservation of mass, we obtain
(3.13)
Eq. (3.13) is generally referred to as advection-diffusion equation.
Using the above analogy and following Ficks law with negligible molecular motion in
comparison to turbulent motion, two layered two-dimensional solute transport equations for inert
solute in a compound channel can be written as
Upper layer:
27
(3.14)
Lower layer:
(3.15)
where,
ck
sk
Ke
rate of exchange of solute concentration from one layer to other (per sec)
Combining Eq. (3.10) and Eq. (3.16) [i.e. all equations containing hydrodynamic and transport
terms together], final form of governing equation in divergence form can be expressed as
(3.17)
where, U, E, G, and S are row vectors. U is the state vector, E and G are the vectors representing
flux terms and S is the vector representing source term; and can be expressed as
28
The governing equations expressed in two layered two dimensional form (Eq. 3.17) are a set of
non-linear, hyperbolic partial differential equations, and their solutions can be obtained using the
numerical methods. Finite volume method, among the other available numerical methods, is
suitable choice, for the solutions of non-linear partial differential equations as they can be
implemented with ease even for complex unstructured domains. In present study, finite volume
method, with explicit scheme, is used for solutions of the governing equations. For discretisation
of various terms of the governing equations of both layers, both in space and time, method of
lines approach is employed. This approach has lot of flexibility as flux integrals are evaluated
29
using grid volume technique and the resulting time-dependent equations are solved from a
known initial values. Further, for numerical approximations, different accuracy level of spatial
and temporal derivatives can be adopted. Brief description of finite volume method,
discretisation of governing equations, turbulent closure schemes, appropriate boundary
conditions and stability consideration of numerical scheme are described in following
paragraphs.
Finite volume method (FVM) utilises conservation laws through integral formulation of NavierStokes equations. It was first employed by McDonald in 1971 for simulation of 2-D inviscid
flows. The governing equations, in such method, are discretized by dividing physical space into a
number of non-rectangular control volumes as shown in Fig. 3.2. The surface integral flux terms
are approximated by the sum of fluxes crossing individual faces of control volume. The accuracy
of spatial discretisation depends on a particular scheme with which fluxes are evaluated.
The FVM is based on direct discretisation of conservation laws such as mass, momentum and
energy which enable it to compute correctly even weak solution of the governing equations. In
present study, cell-centred approach of FVM is used. In this approach, entire flow domain is first
30
divided into a set of cells. Each cell is identified by the corresponding centre point and the flow
quantities are stored at the centroid of grid cells.
Fig. 3.3 shows a non-orthogonal elemental cell (i,j) and its immediate neighbours. The governing
equations are integrated using finite volume technique on each of cells covering the whole flow
domain. After the application of Gauss divergence theorem, Eq. (3.17) can be written in integral
form, as under,
(3.18)
where, F = flux terms at cell centre.
The 1st term represents the integral of time evolution of function over the area of cell. The 2nd
term is the total normal flux through cell boundaries and the 3rd term represents the integral of
source quantities over the volume of grid cells.
The scalar product (F.n) can be expressed in terms of components in X and Y directions as
F.n = E.nx + G.ny
(3.19)
31
Assuming the vector U to be uniform over a 2-D grid cell with unit depth, Eq. (3.18) can be
written as
(3.20)
where,
The surface integral in Eq. (3.20) is approximated by sum over the four walls of grid cell as
(3.21)
where, drr is length of each wall of grid cell that contours the cell (i,j), and Fr is numerical flux
through the cell face r that contour the cell (i,j)
In order to approximate surface integral by summing over all four walls of grid cells, geometrical
quantities of grid cell in terms of its surface area, its face area and unit normal vectors are
calculated.
In present study, 2-D flow problem is considered on XY plane having symmetric with respect to
Z-direction. Because of symmetry, and in order to obtain physical units for flow quantities, depth
of each grid cell is set equal to one resulting into the grid cell as two dimensional with its volume
equal to its area.
32
In 2-D, faces of control volume are given by straight lines and therefore, unit normal vector is
constant along them. During approximation process, fluxes at cell face are integrated based on
the product of area of face A and the corresponding unit normal vector n and the face vector s.
(3.23)
Face vectors of the grid cell are given by the relations
(3.24a)
(3.24b)
(3.24c)
(3.24d)
The unit normal vector at face m is obtained from Eq. (3.23) as
(3.25)
where,
and
are only computed and stored for each grid cell. In order to
save memory and reduce the number of operations, the face vectors
and
The procedure for discretisation of each term of governing equations is followed separately as
discussed in following paragraphs:
33
can be expressed as
(3.26)
Evaluation of numerical flux at a face of grid cell (i,j) between nodes (i+1,j) and (i,j) is carried
out as
(3.28)
where, is the positive coefficient. FR = f(UR) is the flux computed using the information from
the right side of cell face and, FL = f(UL) is the flux computed using the information from the left
side of cell face.
34
Fig. 3.5 Left and right state at cell face [(i+1/2,j) & (i,j+1/2)]
The UR and UL are obtained using a MUSCL (Monotone Upwind Scheme for Conservation
Laws) approach as
,
and
(3.29)
where, subscript (i,j) is the value at node (i,j) and subscript (i+1/2,j) is the value at interface
between nodes (i,j) and (i+1,j). There are several ways to determine
and
using
slope limiter procedures (Yee, 1989; Alcrudo et al., 1992). The slope limiter procedure is
generally used to suppress non-physical oscillation of the solution. The minmod limiter
procedure has been adopted in present study. According to this,
and
(3.30)
b,
0,
if a.b < = 0
Positive coefficient in Eq. (3.28) is determined using the maximum value (for all the grid
points) of the largest eigen value of Jacobian of the system of equations (Nujic, 1995) such as
i=1 to Ni
(3.31)
35
and
j=1 to Nj
where
in which Ni, Nj are the total number of grid points in X- and Y-directions respectively. Vi,j is the
resultant velocity and
for 2D cases.
While computing diffusive flux, the non-uniform nature of grid cell is treated by adopting an
interpolation scheme based on averaging of the flow variable of the grid cell with the area of the
diagonally opposite cell (Sharma and Eswaran, 2003). These are subsequently summed up as the
contribution to the viscous fluxes as given in Eq. (3.32).
37
is the SGS eddy viscosity at kth layer approximated using turbulence closure schemes
such as zero equation, one equation, two equations, large eddy simulation (LES) and direct
numerical simulation (DNS). In present study, LES approach is adopted for closure of turbulence
terms appearing in governing equations for both layers. Two schemes, viz. Standard
Smagorinsky (SS) and dynamic subgrid scale model (DSGS) model under large eddy simulation
(LES) technique have been used in present study.
(ii) Turbulent terms in Transport equations
Considering convective terms of transport equations in X-direction as
(3.41)
Averaging over the area of grid cell and neglecting small terms, Eq. (3.41) can be written as
(3.42)
where subscript k represents the kth layer of flow domain
38
In order to model correlations between turbulent fluctuation of velocity and scalar quantity
transport turbulence terms, viz.
and
Eq. (3.43), eddy diffusivity coefficient, t is introduced using Boussinesq approach similar to that
adopted in hydrodynamic equations as
(3.44)
and,
(3.45)
where,
(3.46)
and,
is Schmidt number, generally, varies between 0.5 to 1.0 (Djordjevic, 1993; Lin and
is turbulent eddy viscosity in ith direction at kth
(3.47b)
Turbulence is a frenzied process in nature in which flows are generally indiscriminate and
therefore, in order to model such a system it is necessary to employ statistically averaged flow
variables. Shears stress is generally generated due to friction from the bed and banks of flow
domain resulting into development of turbulence and change in the mean flow field. The
modelling of turbulent flows requires proper treatment for the length and time scales over which
turbulent eddies occur.
39
Flow variables are generally separated into resolved and unresolved parts while using Large
Eddy Simulation (LES) in investigating the turbulent eddies in compound open channel flows
(Lesieur et al., 2005). The resolved parts (large scale) affecting the turbulent diffusion of
momentum are computed through the numerical solution of the conservation equations whereas
the unresolved parts (small scale) are modelled using different subgrid models. The LES
technique has been utilised successfully in capturing the unsteady nature of large eddies. It has
also been proved that LES can model the flow behaviours better than the Reynolds Averaged
Navier-Stokes (RANS) approach. Details of subgrid models can be found elsewhere including in
Lesieur et al. (2005). In LES generally central schemes are applied and the subgrid scale (SGS)
term is treated with explicit model.
Time-dependent equations for large-scale motion are generally resolved directly, however,
small-scale turbulence, that cannot be resolved with the chosen numerical grid, must be
approximated by a model, known as subgrid scale (SGS) model. In this approach, large scales
turbulence is solved using time-dependent equations along with suitable numerical grids and
small scale turbulence is approximated through modelling technique.
, to strain rate
velocity field. The Smagorinsky model for kth layer can be expressed as
40
(3.48)
The SGS viscosity term,
where,
is the SGS coefficient, hereafter called as model parameter or coefficient. The parameter C is a
calibration parameter and its optimal value depends upon type of flow, the Reynolds number and
the discretisation scheme. Smagorinsky model having the limitation that, for
, it is
Similarly, expanding Eqs. (3.47) in terms of flow velocities in streamwise and spanwise
directions, can be expressed as
(3.50a)
and,
(3.50b)
41
The dynamic model is developed by the introduction of a test filter with a length scale larger
than original filter and, generally, a factor of two is adopted for the same. The stresses obtained
due to test filter are referred to as subtest scale (STS) stresses, and are modelled similar to that of
the SGS stresses. All parameters are calculated separately for each layer before averaging C
value for lower and upper layers in spanwise direction.
The
anisotropic
part
of
the
SGS
stress
at
grid-filter
scale
is
given
by
and
and
(3.52b)
where,
and
refers to the strain rate tensors at the grid- and test- filter levels for each layer
and
can be written as
(3.53)
(3.54)
It is worth mentioning here that grid-filtered strain rates can be obtained locally from nonequilibrium conditions, as mentioned above; and test-filtered strain rates can be determined
directly from gradient of test-filtered velocity field, which can, in turn, be obtained by explicit
application of test-filter on grid-filtered velocity field.
The unknown SGS stress at each filter level can be related by Germano identity as
(3.55)
42
where, Lij is the resolved turbulent stress for each layer, and upon substituting Eq. (3.51) and Eq.
(3.52) into Eq. (3.55), an expression relating Lij and other known variables are obtained with
model coefficient C being the only unknown (Eq.3.56).
(3.56a)
where,
(3.56b)
Since, the tensor expressions, i.e., Eq. (3.56) leads to five independent equations containing C, its
solution can be obtained using a least square minimization approach as (Eq.3.57)
(3.57)
where, < - > represents the spatial averaging in homogenous directions so as to stabilize the
computations following the summation convention of the repeated indices. In this study,
streamwise direction, both in upper and lower layers of compound channel, is considered
homogenous as variation of flow velocities in this direction is analogous.
In DSGS method, the model parameter C, is allowed to vary locally in the numerical solution
through internal calculations while adjusting the level of viscosity (i.e. tk 0 ) in the flow field.
The procedure is initiated by specifying a value of C, and subsequently, after each time level the
flow field is examined, and a new value of C for each grid cell is determined. Therefore, the
parameter C is continuously updated as solution proceeds with time. In present study, dynamic
model has been developed by introduction of a test filter, replacing the grid filter with a length
scale of two times larger than that the length scale of the grid filter. The stresses in this range are
referred to as subtest scale (STS) stresses, and are modelled separately for lower and upper layers
which is similar to that of SGS stresses. In order to calculate the subtest scale (STS) stresses, the
test-filtered quantities are determined using the grid filtered variables. Among the several
available approaches, the discrete trapezoidal filter approach has been adopted for an efficient
implementation of test-filtered quantities, as they are used to accommodate with underlying
rectangular grid structure (Premnath et al., 2009). The procedure of filtering involves application
of one dimensional filter individually in each of the two coordinate directions in both the layers.
For detailed description, the work of Premnath et al. (2009) can be referred.
43
In present study, Lij and Mij are calculated separately for lower and upper layers at each grid cell
before taking their summative average in streamwise direction. In order to keep positive value of
C,
is imposed for lower and upper layers in model domains. Subsequently, average of C
both for lower and upper layers are calculated for each grid cell along streamwise direction.
Thus, the same value of C is obtained at grid cells in spanwise direction for both the layers.
However, at each grid cell for rest of the floodplain portion in spanwise direction at upper layer,
the value of C is obtained based on summative average in streamwise direction.
Test Filters
There are several approaches available to obtain the test-filtered quantities directly from gridfiltered variables. Discrete trapezoidal filter approach is one of them applied in this work as they
naturally accommodate with the underlying non-uniform grid structure and thereby allowing an
efficient implementation. They involve applying one-dimensional filter successively in each of
two coordinate directions to obtain test filtered values
from
In order to calculate test-filter quantities from the grid-filtered variables, one layer of ghost cell
around the model domain is considered whose values are interpolated from the respective inner
grid cells.
3.5
The finite volume grid is chosen such that the boundaries of the flow domain coincide with the
cell faces of the control volume cells. Boundary conditions are specified at all the outer cell faces
of the flow domain. Four types of boundaries (i) no flow boundary, i.e., solid wall boundary, (ii)
inflow boundary, i.e., upstream boundary (iii) outflow boundary, i.e., downstream boundary and
(iv) free surface boundary, i.e., top boundary are encountered in present study. In order to close
the problem and solve the governing equations suitable initial and boundary conditions are
44
required to be specified. The initial conditions determine the state of the flow at the time t = 0.
Further, for a faster final solution, initial condition should be chosen closer to the likely solution
and therefore, it is important that the initial solution should satisfy the governing equations. The
proper imposition of boundary condition is crucial as the accuracy of the solution depends on a
proper physical and numerical treatment of boundaries. In addition to that, the robustness of the
model and the convergence speed of the solution are considerably affected by the boundary
conditions. The treatment of various boundaries of the flow domain is presented in following
paragraphs.
(ii)
Downstream boundary
Outlet boundaries should be as far downstream of the region of interest as possible. The flow
should be directed out of the domain over the entire outlet cross-section and be parallel. The
simplest approximation at the outlet boundary is that of zero gradients along the grid lines. For
convective flux, a first order upwind approximation is used implicitly. For getting higher
accuracy, higher-order and one-sided finite difference approximations of the derivatives at the
outlet boundary can be used. Both convective and diffusive fluxes can be expressed in terms of
the variable values at inner nodes. In present study, Neumann condition is imposed at the
downstream boundaries for both lower and upper layers, i.e.,
(3.59)
45
and
(3.60)
Nezu and Rodi (1986) for open channels flow. Here, up represents tangential velocity component
along wall boundary as shown in Fig. 3.8.
, is represented as
(3.62)
where, w is wall shear stress, and over bar represents time-averaged value.
In order to resolve time-averaged streamwise velocity, Up, near wall boundary, similarity rule
suggested by Shiono et al. (2003) is used in present study as
(3.63)
and,
(3.64)
47
Here, up and
Above hypothesis associate wall stresses with the streamwise velocity at half cell distance from
the wall. In order to calculate slip velocity at given instant, time-averaged velocity,
calculated at y/2 away from the wall. Mean wall shear velocity,
is
(3.62). Instantaneous shear stress, , is calculated using Eq. (3.63) and Eq. (3.64). Eddy viscosity
is approximated using SGS Reynolds stresses under LES scheme.
Finally, velocity gradient,
where
Finally, slip velocity ub at the wall boundary of kth layer is evaluated from up and
as
(3.66)
Flow variables at the centre of ghost cell are extracted from the known slip velocity at wall
boundary. Based on known flow variables at the centre of ghost cell, derivatives of diffusive flux
are calculated and imposed as boundary condition at the side wall.
Boundary conditions for solute transport model were considered in such a way that there was no
loss of material through the solid boundaries of flow domain, i.e., by considering solute
concentration in terms of
(iv)
water depth is assumed in upper layer which lies over the lower layer at time t=0. This
assumption simplifies the overcoming of the numerical singularity in computation.
In order to solve the hyperbolic partial differential governing equations, a two-step predictorcorrector scheme developed by Nujic (1995) for solving shallow water flow equations was used.
The procedure adopted in the scheme was based on Total Variation Diminishing (TVD) and
Essentially Non-Oscillating (ENO) schemes. The high resolution scheme was later on modified
by Singh and Bhallamudi (1997) for two-dimensional case to account for the non-zero source
term in the continuity equations. This scheme, unlike many other classical second order accurate
schemes, such as the MacCormack scheme, is non-oscillatory even when sharp gradients in the
flow variables are present. The main advantages of this scheme are its simplicity, ease of
implementation and more robust than the classical MacCormack scheme. Also, the explicit
scheme is second-order accurate in both space and time. The results obtained by using this
method are satisfactory for many fluid applications. The brief description of the method is
included in following paragraph:
(i)
Predictor part
discretisation
(3.67)
In which, superscripts p and * refer to value at known time level, t, and predicted value at
unknown time level, t+t, respectively. Here, t is computational time step. Eq. (3.17) should be
interpreted component wise for the vector U. Eq. (3.67) gives the predicted values of flow
variables, viz., h, u and v for both the layers at unknown time level, p+1, at each node (i,j) of
flow domain.
49
(ii)
Corrector part
The vector U at unknown time level, p+1, and node (i,j), i.e.,
(3.68)
Following Singh and Bhallamudi (1997),
determined from
and
and
in predictor step (Eq.3.69). The procedure cited above results in better numerical stability.
(3.69a)
(3.69b)
Computation of
and
in Eq. (3.68) is similar that of F and S in Eq. (3.67) except that the
3.7
STABILITY CONDITIONS
The proposed finite volume scheme is an explicit scheme and, therefore, computational time
step, t is computed dynamically at every time step using the Courant-Friedrichs-Lewy (CFL)
condition proposed by Alcrudo et al. (1993) (Eq.3.70).
(3.70)
Herein, dr(i,j) are the whole set of distance between every centre point of grid cell (i,j) and those
of its four adjacent cells. The t is chosen dynamically in numerical model, such that Eq. (3.70)
is satisfied at all nodes of lower and upper layers. The (Vi,j)k is the resultant velocity at grid cell
(i,j) in kth layer and (hi,j)k is the water depth at grid cell (i,j) in kth layer.
The complete methodology including implementation of selected numerical method in
estimation of flow variables of compound channel using developed SS and DSGS models is
shown through a flow chart, see Fig. 3.9. In Fig. 3.9, T represents the last time step of simulation.
50
51
In this chapter, governing equations of fluid flow and solute transport in a rectangular compound
open-channel in the form of a two-layer system have been presented. Different terms of
governing equations, such as convective, diffusive, source are discretised following the method
of lines and finite volume approach. The approaches followed in present study are second
accurate both in time and space. Turbulent stress terms are dealt with turbulence closure schemes
of standard Smagorinsky (SS) and dynamic subgrid scale (DSGS) model under large eddy
simulation (LES) technique. The discretised equations have been solved using two steps
predictor-corrector approach. Boundary conditions for the treatment of open and sidewall
boundaries of model domain along with stability conditions useful for a numerical model are
discussed in detail. At the end, various steps involved in the development of the models (SS and
DSGS models), in present study, are sequenced through a flowchart.
52
Chapter 4
4.1 GENERAL
The hydrodynamic model developed based on governing equations presented in Subsection 3.2.1
in Chapter 3, has been used to simulate the flow conditions in wide symmetric and narrow
asymmetric compound channels reported in the literature (Fraselle et al., 2008; Shiono and Feng,
2003). The observed experimental data of these channels have been used to assess performance
of the hydrodynamic model. In addition to this, two cases of laboratory experiments undertaken
by Tominaga and Nezu (1991) conveying shallow and deep flows in an asymmetric compound
channel is also considered to assess the performance of hydrodynamic model. In hydrodynamic
model, two turbulence closure schemes under LES technique, viz. standard Smagorinsky model
(SS model) and dynamic subgrid scale model (DSGS model) are employed. Also, the
comparative performance of SS model and DSGS model using compound channels data are
presented.
experiments is included in Fig. 4.1, wherein position of upper and lower layers along with their
horizontal interface is also depicted. Total length (L) and width (B) of symmetric compound
channel were 10 m and 1.2 m respectively. The brief description of channel geometry along with
key flow characteristics employed during the experiments are in included in Table 4.1. The
reported flow measurements were undertaken at a test section 0.6L downstream of the channel
inlet. Further, Fraselle et al. (2010) reported observed depth-averaged primary flow velocity in
aforesaid experiments while comparing the results of Telemac 2-D for two turbulence schemes,
i.e., k- and Elder.
The experimental set up consisting of 20 m long tilted narrow asymmetric compound channel
was developed by Shiono and Feng (2003). The schematic diagram of asymmetric compound
channel and the position of upper and lower layers along with horizontal interface is depicted in
Fig. 4.2. The dimensions of the channel along with key flow characteristics of observed data is
included in Table 4.1. In experiment, sophisticated measuring equipment such as laser Doppler
anemometer (LDA) was used to measure flow velocities at a test section 14 m downstream of
channel inlet. The experiments were also performed by Tominaga and Nezu (1991) in a tilting
laboratory flume of asymmetric compound section having 12.5 m length and 0.4 m width (see
Fig. 4.3). Channel bed was constructed with a painted iron plate and the sidewalls were made of
glass.
54
Flow velocity were measured using a two-color fiber-optic laser Doppler anemometer (FLDA) at
a test section 7.5 m downstream from the channel entrance after the establishment of fully
developed uniform flow conditions.
Fig. 4.2 Schematic of asymmetric compound channel (Shiono and Feng, 2003)
Experiments were undertaken for different flow depth ratios; out of which two cases, viz. Exp.S2 and Exp.S-3 of flow depth ratios 0.5 and 0.25 respectively, were referred herein for validation
of proposed hydrodynamic model. In both the cases, width of main channel as well as floodplain
of the channel were each equal to 0.2 m. Schematic diagram of the compound channel and the
position of upper and lower layers, and horizontal interface are depicted in Fig. 4.3. Channel
section and flow characteristics employed in the experiments are presented in Table 4.1.
Table 4.1 Compound channel geometries and hydraulic characteristics of experimental data
Upper lower
Channel type
depth
Mean
bulk
Slope
flow vel.
S0
Depth
ratio,
rh
Channel
hu
width
bl
depth
hl
(m)
(m)
(m)
(m)
(m/s)
x10-3
1.2
0.025
0.4
0.0506
0.369
1.0
3.0
wide/
shallow
0.2
0.055
0.1
0.055
0.237
0.5
2.0
narrow/
deep
Exp.
S-2
0.4
0.04
0.2
0.04
0.349
0.64
2.0
normal/
deep
Exp.
S-3
0.4
0.06
0.2
0.02
0.288
0.56
4.0
normal/
shallow
Symmetric
(Fraselle et al.,
2008)
Asymmetric
(Shiono and Feng,
2003)
Asymmetric
(Tominaga
and Nezu,
1991)
width
bu
Lower layer
55
(hu hl )
]
hu
/flow
type
As mentioned earlier, two turbulence closure schemes, viz. standard Smagorinsky and dynamic
SGS are employed in the proposed hydrodynamic model. Model simulations were carried out for
compound open-channel experiments reported in Section 4.2 and their corresponding results are
presented in following subsections:
First of all, grid dependence test were performed for given flow conditions in symmetric
compound open-channel (Fraselle et. al., 2008) having length to width ratio equal to 8.3 (wide
channel). Thereafter, sensitivity analysis was undertaken for two important parameters employed
56
in the hydrodynamic model. Among the two parameters; one is SGS coefficient, C, used as
model parameter in turbulent schemes and other is a constant, , appearing during the
discretisation of fluxes of governing equations. Based on the optimised number of grid cells and
the values of parameters, the hydrodynamic model under standard Smagorinsky turbulence
closure scheme (SS model) has been validated for primary flow velocities.
Table 4.2 Grid cells used in wide symmetric compound channel for grid dependence test and
verification of SS model
Channel
type
Symmetric
500x60
500x20
0.02x0.02
1.0
RMSE
of
primary
flow
velocity
(m/s)
0.028
500x70
500x26
0.02x(vary)
1-1.3
0.026
500x80
500x32
0.02x0.015
1.33
0.026
Run
No.
Aspect
ratio
(x/y)
Other
characteristic
Re = 5.5x104
Fr = 0.43
The grid cells giving least RMSE equal to 0.026, i.e., Run No.2, were chosen for further
simulation of other flow parameters. For symmetric compound channel, 500x70 grid cells in
upper layer and 500x26 grid cells in lower layer were used in streamwise x spanwise directions
respectively. In streamwise direction, uniform grid spacing was used whereas, in spanwise
direction, non-uniform grid spacing near side walls was adopted in upper and lower layers of the
channel.
Root mean square error (RMSE) is a frequently used measure of differences between values
predicted by a model and observed in an experiment. It, generally, represents sample standard
deviation of differences between predicted and observed values. The RMSE, expressed in unit, is
57
a good measure of accuracy, but only to compare errors of different models for a particular
variable and not between variables, as it is scale-dependent and quantifies spread of residuals.
Expected value of RMSE is zero indicating good performance of model. The RMSE is defined as
(4.1)
where, Ue is the experimentally observed longitudinal flow velocity (m/s)
Us is the model computed longitudinal flow velocity (m/s)
N is number of observations in spanwise direction
1
2
3
4
5
0.1
0.065
0.2
0.1
0.1
(Alpha)
0.5
0.5
0.5
0.2
1.0
From Fig. 4.4(a), for =0.5 and C varying from 0.065 to 0.2, it is seen that there is a little
variation exists in simulated primary flow velocity in the vicinity of floodplain walls, whereas in
Fig. 4.4(b), for C=0.1 and varying from 0.2 to 1.0, small variation occurs in flow velocities at
the junction of the main channel and the floodplain. Based on least root mean square error
(RMSE), optimum values of C (=0.1) and (=0.5) were used further for model validation and
extraction of other simulated flow parameters in the channel.
58
C=0.1
C=0.2
C=0.065
0.5
Fp
0.4
0.3
0.2
Half of Mc
0.1
0
0
0.1
0.2
0.3
0.4
0.5
0.6
Alpha=0.5
Alpha=1.0
Alpha=0.2
0.5
Fp
0.4
0.3
0.2
Half of Mc
0.1
0
0
0.1
0.2
0.3
0.4
0.5
0.6
conditions derived from grid dependency test (Table 4.2). Near solid walls both in lower and
upper layers, finer grid cells were adopted considering the limiting criteria (y+>30) based on loglaw rule. The value of y+=47 (see, Eq.3.61) has been adopted in foregoing simulation. The
schematic diagram of computational cells in upper and lower layers used in the simulation is
shown in Fig. 4.5.
Fig. 4.5 Computational grid cells and boundaries in symmetric compound channel for SS model
The Mannings roughness coefficients for lower and upper layers were considered as 0.01 and
0.013 respectively and turbulence coefficient, C=0.1 was adopted for entire model domain. The
model was run for steady state discharge of 0.01855m3/s and total water depth of 0.0756m.
Being an explicit model, computational time steps were dynamically calculated based on
60
stability criterion; and after stabilizing the uniform flow condition, a time step of the order of
0.002 sec was found adequate for Courant number equivalent to 0.25. Flow was treated as fully
developed when successive bed shear stress plots had converged at about 75000 time steps. The
simulation was further continued for 25000 time steps to obtain time averaged results which are
presented in following paragraphs
SS model
Half of Mc
Fp
25
RMSE for Uda=0.026 m/s
Uda/U*
20
15
10
5
0
0
0.1
0.2
0.3
0.4
0.5
0.6
Fig. 4.7 Computed flow field in symmetric compound channel (SS model)
At junction of the main channel and the floodplain in upper layer, a sharp rise in gradient of flow
velocity is seen due to significant difference in primary flow velocities exist between the main
channel and the floodplains which can be explained based on calculations of percentage
difference between primary flow velocities of two layers with respect to mean bulk velocity of
channel
16
Mc
12
8
4
0
-4
0.4
0.5
0.6
0.7
0.8
62
In Fig. 4.8, ul and uu represent layer averaged primary flow velocities in lower and upper layers,
respectively, and Ub (=Q/A) represents the mean bulk flow velocity in the same direction. From
Fig. 4.8, it is seen that maximum difference between primary velocities of two layers, of the
order of +12%, is noticed near the junction of the main channel and the floodplain.
It is also observed that, in rest of the interface, the velocity difference between two layers is very
small to the tune of -3%. The change in sign (from ve at mid section of interface to +ve at
junction edges) of velocity difference suggests that considerable reduction in primary velocities
in upper layer may be due to the effect of side walls of main channel and shallower flow depth in
floodplains. Such decrease in primary flow velocities (velocity-dip) near the junction was also
reported by Tominaga and Nezu (1991), and described its occurrence due to secondary currents
that transport low-speed momentum from the junction edge towards the free surface. The higher
velocity in lower layer vis--vis upper layer in the main channel near the junction, clearly
signifies the influence of floodplain in modifying the flow characteristics in the main channel.
simulated near the junction of main channel and floodplain. The bed shear stress decreases
drastically in the main channel to 0.602
N/m2
at the mid span of the main channel. Such sudden decrease in bed shear stress from maximum
value in the floodplain to a minimum value in the main channel at the junction is responsible for
the vortex like structures at the same location. Similar bed shear stress pattern are also observed
in isolines plot shown in Fig. 4.9(b).
63
1.6
Half of Mc
Fp
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
0.1
0.2
0.3
0.4
0.5
0.6
6.05
6
5.95
0
0.1
0.2
0.3
0.4
0.5
0.6
spanwise (v0)
0.004
0.003
0.002
0.001
-1E-17
-0.001
-0.002
-0.003
-0.004
Vertical (w0)
Mc
0.4
0.5
0.6
0.7
Channel width at Interface (m)
(a) Velocities profile
0.8
6.05
5.95
0.4
0.5
0.6
0.7
0.8
However, in literature (Nezu and Rodi, 1986; Tominaga et al., 1989; Tominaga and Nezu, 1991;
Joung and Choi, 2008), the maximum magnitude of the secondary currents observed (point
observation) near the interface in various channels were reported in the range of 2.5-5% of Ub. It
indicates that the strength of secondary currents is under predicted in SS model. It is evident
from the figure that both velocities attain their maximum values near the junction of the main
channel and floodplain. Spanwise interface velocity shows gradual rise in magnitude from zero
to its maximum value (0.0027 m/s) near the junctions.
The rise in magnitude of the vertical interface velocity is sharp before attaining its maximum
(0.0029 m/s) value little away from junction edges. The interfacial vertical velocity is +ve at both
ends of the main channel. It is also evident from the figure (see Fig. 4.10a) that both the interface
velocities become zero at midpoint of the channel. At left end of main channel wall, interfacial
span-wise velocity is +ve while, at right end, it is ve. Isolines plot of secondary current
magnitude at interface is shown in Fig. 4.10(b). The sign of interfacial spanwise flow velocities
represents direction of secondary current at interface. It is clearly seen from the figure (see Fig.
4.10c) that secondary currents are clockwise at left end and anticlockwise at right end of main
channel. Secondary circulation pattern are opposite in nature in both half of the lower layer with
zero exactly at mid width of the channel, indicating that secondary interface velocities are
directed towards the junction of the main channel and the floodplains. Such phenomenon can
also be seen in vector plots (see Fig. 4.10c) of secondary currents, indicating the existence of
vertical interface velocity directed towards upper layer. This implies that transfer of momentum
generally occurs towards floodplains in a symmetric compound channel.
wherein densely spaced isolines near the junctions indicate sharp rise in interfacial shear stress at
the junction of the main channel and floodplain, see Fig. 4.11(b).
0.5
Mc
0 (N/m2)
0.4
0.3
0.2
0.1
0
0.4
0.5
0.6
0.7
0.8
6.05
5.95
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
normalised with square of mean shear velocity (U*) and the resulting profile is plotted along
width of upper layer, see Fig. 4.12(a). In the figure, peak values (0.052 U*2), with opposite in
sign are simulated at the junctions in upper layer of the compound channel. This signifies
67
reversal of flow along with transfer of momentum at these regions. In rest of the upper layer
width, the value of (
opposite nature at the junctions are also depicted in isoline plots in Fig. 4.12(b).
0.06
Fp
Mc
Fp
0.04
0.02
0
-0.02
-0.04
-0.06
0
0.2
0.4
0.6
0.8
1.2
0.2
0.4
0.6
0.8
1.2
contour plots for lower layer of the channel is shown in Fig. 4.13(a-b). These values were
calculated from the layered average flow velocities of lower layer. From Fig. 4.13(a-b), it is seen
that distribution of turbulent shear stress is same and opposite in nature about the midpoint in the
lower layer, and maximum values (0.01 U*2) are observed near junctions edges where transfer of
momentum generally takes place. Isolines plot of the SGS turbulent shear stresses in lower layer
68
(see Fig. 4.13b), confirms circulation forming at the junction and related processes of momentum
transfer.
0.02
Mc
0.01
-0.01
0.4
0.5
-0.02
0.6
0.7
0.8
6.05
5.95
0.4
0.5
0.6
0.7
0.8
used by Shiono et al. (2003) for validation of their numerical models. Tominaga and Nezu
(1991) conducted several laboratory experiments in an asymmetric compound channel having
length to width ratio equal to 31.25 (normal channel), under various steady state discharges. Two
experimental cases, namely, S-2 and S-3 falling in the category of deep and shallow flows,
respectively are considered in present study. Their experimental results were also used by several
researchers for validation of their numerical models (Kara et al., 2012). Similar to symmetric
compound channel, firstly, grid dependence test was undertaken. Thereafter, hydrodynamic SS
model with optimal grid has been used for simulation of flow characteristics in asymmetric
compound channel used by Shiono and Feng (2003) and Tominaga and Nezu (1991). The
detailed description of such simulation is described in following paragraphs:
Table 4.4 Grid cells used in narrow asymmetric compound channel for grid dependence test and
verification of SS model
Grid cells used in model domain
RMSE
Run
(Longitudinal x Transverse)
Aspect
of
Other
Channe No.
ratio
primary characteristic
Total Nos. (in
l type
flow
layer)
Spacing
(x/y) velocity
(m)
Upper
Lower
Asymmetric
(Shiono
& Feng,
2003)
2000x10
2000x5
0.01x0.02
0.5
0.03
2000x20
2000x10
0.01x0.01
1.0
0.021
2000x40
2000x20
0.01x0.005
2.0
0.023
Re = 3.7x104
Fr = 0.23
From Table 4.4, it is seen that RMSE is minimum for grid corresponding to 2000 x 20
(streamwise X spanwise) in upper layer and 2000 x 10 (streamwise X spanwise) in lower layer.
However, the grid corresponding to Run-3 (RMSE=0.023 m/s) has been used, in present study,
for further simulation to enable better resolution near the wall of the model domain. Uniform
70
grid spacing has been adopted in streamwise direction, however, non-uniform grid spacing was
adopted in spanwise direction for better capturing of the flow near the wall.
71
SS model
25
Uda/U*
20
15
Mc
10
Fp
RMSE=0.019 (for Uda)
5
0
0
0.05
0.1
0.15
0.2
From the figure it is seen that maximum difference of the order of -14% occurs near the junction
of the main channel and the floodplain. It can also be observed that velocity difference is
gradually rising from -3% at common wall of lower and upper layers up to the junction edge.
The -ve sign of velocity difference shows that primary velocity in upper layer is more than that
of lower layer throughout the channel span. Further, higher ve value of velocity difference near
the junction edge of the main channel and floodplain suggests that the effect of side wall and
floodplain on the primary velocity in upper layer is minimal. And, also, the higher velocity at
upper layer near the junction edge confirms /causes the presence of two inflection points.
72
0
Mc
-2
-4
-6
-8
-10
-12
-14
-16
0
0.02
0.04
0.06
0.08
0.1
While comparing the shallow flow condition (see, Fig. 4.8 for symmetric compound channel)
with deep flow condition (Fig. 4.15), clearly demonstrates that flow conditions in the floodplain
play dominant role in affecting the flow conditions in the main channel for shallow flow
conditions.
The situation is reversed for deep flow condition wherein the flow velocity in upper layer is
consistently higher from lower layer in the main channel.
N/m2) of
bed shear stress occurs in the floodplain near the junction of the main channel and the floodplain.
Also, another peak of 1.198
vortex like structures are clearly seen on both sides of junction in the isolines plot of bed shear
stress, see Fig. 4.16(b).
73
1.4
1.2
1
0.8
0.6
Fp
Mc
0.4
0.2
0
0
0.05
0.1
0.15
0.2
14.05
14
13.95
0
0.05
0.1
0.15
0.2
generated in the channel is of the order of 0.015% (see Fig. 4.17c) of its mean bulk flow velocity
(Ub). It is worth mentioning that secondary currents reported herein are computed from the layer
averaged velocities of the channel. It is seen that, in the channel, the strength of secondary
currents is highly under predicted by the model. It represents only the qualitative value of the
currents.
Vertical (w0)
5E-05
Mc
4E-05
3E-05
2E-05
1E-05
0
0.02
0.04
0.06
0.08
Spanwise (v0)
14
13.95
0
0.02
0.04
0.06
0.08
0.1
0.1
76
0.4
Mc
0 (N/m2)
0.3
0.2
0.1
0
0.02
0.04
0.06
0.08
0.1
14.05
14
13.95
0
0.02
0.04
0.06
0.08
0.1
From profile plot of interfacial shear stress (see Fig. 4.18a), it is seen that peak value (0.33 N/m2)
occurs at the right end of interface near the junction of the main channel and the floodplain. It is
seen that there is gradual rise in its magnitude from left end of interface to the junction edge.
However, as junction is approached, the rise in interfacial shear is sharp. Such behaviour of rise
towards the junction is due to existence of momentum transfer from lower layer to upper layer at
foregoing locations of the channel. Further, formation of densely spaced isolines at the ends of
interface, near the junction, can be clearly seen in contour plot of interface shear stress (see Fig.
4.18b).
shear velocity (U*) and the resulting profile across width of upper layer is shown in Fig. 4.19(a).
From the figure, maximum value (0.073 U*2 N/m2) of turbulent shear stress can be observed near
left wall of upper layer in the main channel whereas, at junction of the main channel and the
floodplain, negligibly small value is observed.
0.08
Fp
Mc
0.06
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
0
0.05
0.1
Channel width (m)
(a) Profile plot
78
0.15
0.2
14.05
14
13.95
0
0.04
0.08
0.12
0.16
0.2
Apart from this, near the right wall of upper layer in floodplain, a shear stress value of the order
of 0.058U*2 N/m2 has been simulated in the SS model. It is also seen that nature of shear stresses
at both the ends are opposite to each other, see isolines plot of turbulent shear stress in Fig.
4.19(b).
contour plots for lower layer of the channel is shown in Fig. 4.20(a-b).
From the shear stress profile (Fig. 4.20a), it is seen that a peak shear stress value (0.07 U*2 N/m2)
occurs near the left wall of lower layer whereas, second peak value (0.04U*2 N/m2), opposite in
nature, is simulated near the junction of the main channel and the floodplain where transfer of
momentum is affected. In isolines plot (see Fig. 4.20b), circulation formed at both the ends of
lower layer is seen whereas, in upper layer plot (see Fig. 4.20b), such formation is absent at the
junction of the main channel and floodplain. The model under predicts the magnitude of shear
stresses, however, qualitatively it represents flow features in the compound channel.
79
0.08
Mc
0.06
0.04
0.02
0
-0.02
-0.04
-0.06
0
0.02
0.04
0.06
0.08
0.1
14.05
14
13.95
0
0.02
0.04
0.06
0.08
0.1
80
Channel
type
Asymmetric
Run
No.
Aspect
ratio
RMSE of
primary flow
velocity
(x/y)
(m/s)
625x25
625x13
0.02x0.02
1.0
0.01
625x31
625x16
0.02x0.015
1.3
0.009
625x48
625x24
0.02x0.01
2.0
0.008
For simulation of asymmetric compound channel, number of grid cells under Run No.3 with
RMSE=0.008 m/s were adopted wherein 625x48 grid cells in upper layer and 625x24 grid cells
in lower layer were used in streamwise x spanwise directions respectively. It is to be noted that,
in streamwise direction, uniform grid spacing was selected whereas in spanwise direction nonuniform grid spacing was adopted in upper and lower layers of both the channels.
81
Similar flow conditions and channel geometries of the channel were adopted for each
combination. The primary flow velocity profile across width of asymmetric compound channel
obtained for each combination is shown in Figs. 4.21(a-b).
C=0.1
0.5
C=0.2
Mc
C=0.065
Fp
0.4
0.3
0.2
0.1
0
0
0.1
0.2
0.3
0.4
Alpha=0.5
Alpha=1.0
Alpha=0.2
0.5
Mc
Fp
0.4
0.3
0.2
0.1
0
0
0.1
0.2
0.3
0.4
82
From Fig. 4.21(a), for =0.5 and C varying from 0.065 to 0.2, it is seen that there is a little
variation exists in simulated primary flow velocity in the region of floodplain as well as in the
vicinity of walls, whereas in Fig. 4.21(b), for C=0.1 and varying from 0.2 to 1.0, significant
variation is observed in main channel and floodplain of the channel. However, in rest of the
channel, negligibly small effect can be observed due to variation in aforesaid parameters. In
further analysis, an average value of C=0.1 and =0.5 have been taken to simulate the flow
parameters.
83
From Fig. 4.22, it is observed that velocities are reasonably in good agreement (RMSE for
Uda=0.019 m/s and 0.017 m/s for deep and shallow flows respectively) except near the junction
of the main channel and the floodplain. In deep flows, two inflection points are simulated, see
Fig. 4.22(a), whereas in shallow flows, only one inflection point occurs near the junction of the
main channel and the floodplain, see Fig. 4.22(b).
SS model
Mc
Fp
Uda/U*
25
20
15
10
RMSE for Uda =0.019 m/s
5
0
0
0.1
0.2
0.3
0.4
Mc
Uda/U*
25
SS model
Fp
20
15
10
5
0
0
0.1
0.2
Channel width (m)
0.3
0.4
For both the cases, at junction of the main channel and floodplain, a rise in gradient of flow
velocity is seen due to significant difference in primary flow velocities existing between the main
channel and floodplain which can be explained based on calculation of percentage difference
between primary velocities of two layers with respect to mean bulk velocity of channel
, see Fig. 4.23(a-b).
0
Mc
-2
-4
-6
-8
-10
0
0.05
0.1
0.15
0.2
6
Mc
4
2
0
-2
-4
-6
0
0.05
0.1
0.15
0.2
From Fig. 4.23(a-b), it is seen that the maximum difference has been found to be of the order of
-9% for deep flows (Fig. 4.23a) and +4% for shallow flows (Fig. 4.23b) near the junction of the
main channel and the floodplain. In deep flows (Exp.S-2, Fig. 4.23a) condition, the primary flow
velocity in upper layer has been found higher than lower layer throughout the main channel.
However, such difference is increased at the junction of main channel and floodplain. The -ve
sign of velocity difference shows that primary velocity in upper layer is more than lower layer.
Further, higher ve value of velocity difference near the junction edge of the main channel and
floodplain suggests that the effect of side wall and floodplain on the primary velocity in upper
layer is minimal. Also, the higher velocity in upper layer near the junction edge confirms the
presence of two inflection points. On the other hand, in shallow flows (Exp.S-3, Fig. 4.23b)
condition, the primary velocity in upper layer is higher than lower layer at left end, and, the trend
is getting reversed at the junction edge of the main channel and floodplain. In other words, the
effect of floodplain is dominant in affecting the flow characteristics of main channel in shallow
flow condition vis--vis deep flow condition.
SS model
1.6
1.2
0.8
Mc
Fp
0.4
0
0
0.1
0.2
Channel width (m)
0.3
0.4
7.55
7.5
7.45
0
0.1
0.2
0.3
0.4
SS model
1.6
RMSE for b=0.027 N/m2
1.2
0.8
0.4
Mc
Fp
0
0
0.1
0.2
Channel width (m)
0.3
0.4
7.55
7.5
7.45
0
0.1
0.2
0.3
0.4
87
From Figs. 4.24(i) and 4.24 (ii), it is seen that bed shear stresses are in general agreement
(RMSE for
=0.027 N/m2 for both deep and shallow flows) with experimental results except at
few places in the main channel where it is over-predicted the experimental values. The minor
differences between experimental and numerical results can be attributed to the layered average
assumption being applicable in the main channel. Further, in deep flows case, peak value (1.2 )
is observed at two places across the channel width, i.e., at the junction of the main channel and
the floodplain, and at mid section of the main channel. From the peak, its value decreases
towards the walls of both main channel and floodplain, and attains a minimum value of the order
of 0.6 . In shallow flows, the peak (1.55 ) occurs at mid section of the main channel, and a
minimum value of the order of 0.5
b) show isolines of bed shear stresses at test section of the channel in which formation of vortex
like structures are clearly visible near the junction of the main channel and the floodplain.
Spanwise (v0)
Vertical (w0)
0.0004
Mc
0.0003
0.0002
0.0001
0
0
0.05
0.1
0.15
88
0.2
7.55
7.5
7.45
0
0.05
0.1
Channel width at Interface (m)
89
0.15
0.2
Spanwise (v0)
0.0008
Vertical (w0)
Mc
0.0006
0.0004
0.0002
0
0
0.05
0.1
0.15
0.2
7.55
7.5
7.45
0
0.05
0.1
Channel width at Interface (m)
90
0.15
0.2
direction of secondary current at interface. It is clearly seen from Figs. 4.25(i-c) and 4.25(ii-c)
that secondary currents are anticlockwise at right end of the main channel which follows the
trends reported by Tominaga and Nezu (1991). The variation of secondary currents across the
width of main channel at horizontal interface is also indicated in their respective isolines plots,
i.e., Figs. 4.25(i-b) and 4.25(ii-b) respectively. From Figs. 4.25(i-a) and 4.25(ii-a), it is clearly
seen that spanwise interfacial velocity in shallow flow condition is higher than deep flow
condition, particularly at junction of the main channel and floodplain.
0.4
Mc
0 (N/m2)
0.3
0.2
0.1
0
0
0.05
0.1
0.15
0.2
7.55
7.5
7.45
0
0.05
0.1
0.15
0.2
0.1
Mc
0 (N/m2)
0.08
0.06
0.04
0.02
0
0
0.05
0.1
Interface width (m)
93
0.15
0.2
7.55
7.5
7.45
0
0.05
0.1
0.15
0.2
shear velocity (U*) and the resulting profile plot across upper layer width is shown in Figs.
4.27(i-ii).
0.015
Fp
Mc
0.01
0.005
0
-0.005
-0.01
0
0.1
0.2
0.3
Upper layer width (m)
(a) Profile plot
94
0.4
7.55
7.5
7.45
0
0.1
0.2
0.3
0.4
0.02
0.01
0
-0.01
-0.02
-0.03
-0.04
-0.05
Mc
0.1
Fp
0.2
0.3
Upper layer width (m)
0.4
7.55
7.5
7.45
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
occurs at the junction edge. The variation of SGS shear stresses in upper layer clearly signifies
the effect of walls in modifying the pattern of shear stresses in the channel. The isolines plots
(Figs. 4.27(i-b) and 4.27(ii-b)) also indicate the existence of circulation at the junction and both
ends of upper layer. The significant sudden drop in shear stress at the junction of shallow flow
clearly signifies the major influence of floodplain in flow characteristics in the main channel and
transfer of momentum at the junction of the main channel and floodplain.
isolines plots for lower layer of the channel is shown in Figs. 4.28(i-ii).
0.015
Mc
0.01
0.005
0
-0.005
-0.01
-0.015
0
0.05
0.1
0.15
Lower layer width (m)
0.2
7.55
7.5
7.45
0
0.05
0.1
0.15
0.2
0.04
0.03
0.02
0.01
0
-0.01
-0.02
-0.03
-0.04
Mc
0.05
0.1
0.15
Lower layer width (m)
0.2
7.55
7.5
7.45
0
0.05
0.1
0.15
0.2
circulations formed at both ends of lower layer confirms processes of mass and momentum
transfer from these regions of the channel.
The hydrodynamic model developed with dynamic SGS turbulence closure has been validated
using the data of both types of compound channels, i.e., deep and shallow flows. The same sets
of compound channel data are used in validation of DSGS model as were used in Section 4.3.1
for validation of SS model. In all the cases, same computational grid cells are adopted.
In standard Smagorinsky (SS) model, constant value of C was applied for the whole flow domain
and in order to model the flows near the solid walls, damping factor was used. However, in LES
several potential difficulties do occur in case of shear flows and flow near the solid walls due to
selection of a constant Smagorinsky coefficient. A dynamic model for C, as proposed by
Germano et al. (1991) to account for asymptotic behaviour of the subgrid scale stresses existing
near the walls, has been used in DSGS model. In dynamic model, coefficient C is allowed to
vary locally to adjust eddy viscosity in the flow field.
From the figure, it is seen that, at sidewalls of the channel, the value of C reduces to minimum as
eddy viscosity is minimal close to the walls due to occurrence of anisotropy turbulence (Rodi,
1980). Further, away from the wall in the floodplain (Fp), its value rises steeply in symmetric
channel reaching a maximum value (0.138), and then, decreases slowly before attaining its
minimum value (0.02) near the wall of lower layer.
98
0.16
Fp
Half of Mc
0.2
0.3
0.4
Half of channel width (m)
0.5
0.12
0.08
0.04
0
0
0.1
0.6
Fig. 4.29 Variation of SGS coefficient across wide symmetric compound channel (DSGS model)
In the main channel of symmetric compound channel, the value of model parameter rises again,
and regains to its second peak at mid of the channel. Being a symmetric compound channel,
variation of the model parameter would also be symmetric. The parameter varies within the
range of 0.02 to 0.138 throughout the channel during the model simulations (Fig. 4.45) after
attaining steady state flow condition. The variation of model parameter in both parts of the
channel is in the recommended range of Rodi (1980), i.e., C 0.1 . Thus, it is seen in the model
simulation that variation of model parameter C is highly dependent on grid spacing of the flow
domain. The variation of model parameter C across the width of compound channel, underlines
the importance of taking dynamic SGS parameter instead of taking a constant value throughout
the simulation.
The depth averaged flow characteristics in the main channel and floodplain of corresponding
channels using DSGS model are described in flowing paragraphs:
99
floodplain where model under predicted flow velocities. It is also observed from the figure that
due to shallow flows in channel only one inflection point occurs near the junction of the main
channel and the floodplain.
30
25
Uda/U*
DSGS model
20
15
10
Half of Mc
Fp
5
0
0
0.1
0.2
0.3
0.4
0.5
0.6
8
4
0
-4
0.4
0.5
0.6
0.7
0.8
difference between primary velocities of two layers with respect to mean bulk velocity of
channel
From the figure, it is seen that the difference is found to be similar to that of obtained in SS
model case. The pattern of primary velocity prediction from DSGS model is similar to SS model
(see Fig. 4.6), however, former predicts the flow characteristics slightly better than latter model.
1.6
Fp
1.4
Half of Mc
1.2
1
0.8
0.6
0.4
0.2
0
0.1
0.2
0.3
0.4
101
0.5
0.6
6.05
6
5.95
0
0.1
0.2
0.3
0.4
0.5
0.6
N/m2 to 1.37
N/m2) at midpoint of
N/m2) in the bed shear
stresses at junction edge is due to formation of vortex like structure as can be seen clearly in
isolines plot of Fig. 4.32(b).
102
Spanwise (v0)
Vertical (w0)
0.004
0.003
0.002
0.001
0
-0.001
-0.002
-0.003
-0.004
Mc
0.4
0.5
0.6
0.7
0.8
6.05
5.95
0.4
0.5
0.6
0.7
0.8
0 (N/m2)
0.4
0.3
0.2
0.1
0
0.4
0.5
0.6
0.7
0.8
6.05
5.95
0.4
0.5
0.6
0.7
0.8
with square of shear velocity (U*), is shown in Fig. 4.35(a). From the figure, it is clearly seen
that maximum values (0.05 U*2) of SGS shear stresses occur at the junction of main channel and
floodplain. The SGS shear stresses at both the junctions and floodplains are opposite in nature
due to transfer of momentum across the floodplains from the main channel.
0.06
Fp
Mc
Fp
0.2
0.4
0.6
0.8
Upper layer width (m)
0.04
0.02
0
-0.02
-0.04
-0.06
0
1.2
6.05
6
5.95
0
0.2
0.4
0.6
0.8
1.2
The isolines plot (Fig. 4.35b) also depicts the formation of vortices due to existence of opposite
circulation pattern at both junctions of the symmetric compound channel. The variation of SGS
105
turbulence shear stresses across upper layer simulated from SS model (Fig. 4.12a) and DSGS
model (Fig. 4.35a) is due to the turbulence coefficient C varying across the channel width.
; see
0.02
Mc
0.01
0
-0.01
-0.02
0.4
0.5
0.6
0.7
Lower layer width (m)
0.8
6.05
5.95
0.4
0.5
0.6
0.7
0.8
From the figure, it is seen that distribution of turbulent shear stress is symmetrical about the
midpoint of the lower layer, and maximum values (0.015U*2), opposite in nature, are observed
near junctions of the main channel and floodplain where transfer of momentum in the floodplains
106
are affected. The isolines plot (Fig. 4.36b) confirms the formation of circulation pattern at either
end of lower layer wherein transfer of momentum takes place from main channel to floodplain.
In comparison to SS model (see Fig. 4.13a), SGS turbulent shear stress at lower layer was
simulated marginally higher in DSGS model (see Fig. 4.36b) due to variation of turbulence
parameter C across the channel width.
4.3.2.2 Asymmetric compound channel
(i) Spanwise variation of model parameter C
Fig. (4.37) shows spanwise variation of model parameter, C, in asymmetric compound channel
flow. The values of C were determined locally during the model simulation while adjusting the
level of eddy viscosity in flow field.
0.16
Mc
Fp
0.12
0.08
0.04
0
0
0.05
0.1
0.15
0.2
DSGS model
25
Mc
Fp
Uda/U*
20
15
10
RMSE for Uda=0.019 m/s
5
0
0
0.05
0.1
Channel width (m)
0.15
0.2
Fig. 4.38 Comparison of primary flow velocity in narrow asymmetric compound channel (DSGS
model)
0
Mc
-2
-4
-6
-8
-10
-12
-14
0
0.02
0.04
0.06
0.08
0.1
108
From the figure, it is apparent that computed flow velocities (RMSE for Uda=0.019 m/s) in both
main channel and floodplain follow the trend of the measured data, though slight differences
between simulated and experimental values remain near the junction of the channel. Similar to
SS model, it is seen that model over predicts the flows in the main channel and under predicts the
same in the floodplain.
The depth-averaged primary velocities in lower layer (ul) and upper layer (uu) were computed
separately. The differences between ul and uu, expressed as
significant at the junction of the main channel and floodplain, see Fig. 4.39. Here, Ub represents
mean bulk velocity in the channel. From the figure, it is seen that the differences between ul and
uu have been found to be of the order of 13% near the junction of the main channel and the
floodplain. Similar order of differences between the velocities, ul and uu, have been observed in
SS model, also.
1.6
Mc
Fp
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
0.05
0.1
Channel width (m)
0.15
0.2
14.05
14
13.95
0
0.05
0.1
0.15
0.2
the junction of the main channel and floodplain. In comparison to SS model, more bed shear
stresses have been simulated in the channel. Formation of vortex like structure, at the junction of
the main channel and floodplain, is clearly seen in Fig. 4.40(b), showing isolines plot of bed
shear stresses in the channel.
spanwise and vertical flow velocities at interface are simulated lesser in magnitude in DSGS
model (Fig. 4.41).
Spanwise (v0)
Vertical (w0)
0.00003
Mc
0.000025
0.00002
0.000015
0.00001
0.000005
0
0
0.02
0.04
0.06
0.08
0.1
14.05
14
13.95
0
0.02
0.04
0.06
0.08
0.1
0.4
Mc
0 (N/m2)
0.3
0.2
0.1
0
0
0.02
0.04
0.06
0.08
0.1
14.05
14
13.95
0
0.02
0.04
0.06
0.08
0.1
shear velocity (U*) and the resulting profile across width of upper layer is shown in Fig. 4.43(a).
From the figure, maximum value (+0.027 U*2) of turbulent shear stress is simulated near left wall
of upper layer in the main channel, and a value of -0.01U*2 has been simulated at the junction of
main channel and floodplain. The reverse trend of shear stresses in upper layer is reported in the
floodplain due to transfer of flow and momentum from main channel to floodplain. In
comparison to SS model (see Fig. 4.19), more distributed variation of SGS turbulent shear
113
stresses in upper layer are simulated in DSGS model (Fig. 4.43) due to dynamic variation of
model coefficient C in latter across the width of the channel.
0.04
Mc
Fp
0.02
0
-0.02
-0.04
0
0.05
0.1
0.15
0.2
14.05
14
13.95
0
0.05
0.1
0.15
0.2
isolines plots across the width of lower layer of the channel were obtained as shown in Fig. 4.44
(a-b).
114
0.04
Mc
0.02
-0.02
-0.04
0
0.025
0.05
0.075
0.1
14.05
14
13.95
0
0.02
0.04
0.06
0.08
0.1
channel and the floodplain where transfer of momentum generally takes place. Further, in
comparison to SS model (see Fig. 4.20), more distributed variation of SGS turbulent shear
stresses in lower layer are simulated in DSGS model (Fig. 4.44).
The performance of hydrodynamic models such as SS model and DSGS model developed using
turbulence closure schemes under LES approach were assessed for turbulent flows in a wide
symmetric and a narrow asymmetric compound channels. In addition, a normal asymmetric
compound channel conveying shallow and deep flows was also considered. The grid dependence
tests have been undertaken to arrive best comparisons of grids before simulations of hydraulic
characteristics in symmetric and asymmetric compound channels using developed SS and DSGS
models. The sensitivity analysis of model parameters C and has been undertaken, and it is
found that average values of C and equals to 0.1 and 0.5 respectively are adequate for
simulation of hydraulic characteristics using developed model. The SS model has been able to
correctly simulate the primary velocity in symmetric and asymmetric compound channels with
RMSE in the range 0.017 m/s to 0.026 m/s. Significant difference in primary velocity of lower
and upper layers
floodplain in a wide symmetric channel with shallow flow conditions. The situation is reversed
for a narrow asymmetric compound channel with deep flow conditions where primary velocity
of upper layer is consistently higher than lower layer, and significant difference
has been simulated at the junctions of the main channel and floodplain in the order of -14%. The
simulated trend is also reported for shallow vis--vis deep flow conditions in a normal
asymmetric compound open channel. The simulated bed shear stresses in symmetric and
asymmetric compound channels for shallow and deep flow conditions have shown sudden
changes at junctions of the main channel and floodplain due to existence of significant secondary
currents at that location. The existence of secondary currents (combination of spanwise and
vertical velocity components along the horizontal interface) at the junction of the main channel
and floodplain is responsible for transfer of momentum from main channel to the floodplain.
116
Shear stresses at horizontal interface of both the layers, and averages shear stresses in both upper
and lower layers are simulated, and the same give clear signature of significant secondary
currents at the junctions of the main channel and floodplain. The model parameter, C, has been
found to have significant variations across the channel width which depends on the variation of
grid sizes and flow conditions. The DSGS model, while varying the value of C across the width,
has been found to give marginally better results vis--vis SS model in simulating hydraulic
characteristics such as flow velocities, turbulent shear stresses in compound channel flows.
117
Chapter 5
5.1 GENERAL
The transport characteristics of inert solute in lower and upper layers, and along the horizontal
interface between the layers are investigated in compound open channel flows using dynamic
subgrid scale (DSGS) closure scheme in large eddy simulation (LES) model. The performance of
transport model coupled with hydrodynamic model have been assessed using experimental data
of symmetric and asymmetric compound open-channel flows reported by Fraselle et al. (2008)
and Shiono and Feng (2003) respectively.
The solute measurements in symmetric and asymmetric compound channels carried out by
Freselle et al. (2010) and Shiono and Feng (2003) respectively, are used for validation of the
performance of DSGS model in respective compound open-channels. Schematics of symmetric
and asymmetric compound cross-sections are included in Figs. 4.1 and 4.2 respectively in
Chapter 4.
Channel geometry and flow characteristics of a wide symmetric compound channel, used for
laboratory experiment of solute transport by Fraselle et al. (2008), are presented, see Fig. 4.1 and
Table 4.1, in Chapter 4. Two sets of experimental data for solute transport, as reported by
118
Fraselle et al. (2008), in symmetric compound channel were used in the present study. In first
experiment, named as Exp.I-1, injection point of solute as point source was located at mid of the
main channel at y=0.6 m and 0.0375 m above the main channel bed, and in the second
experiment, named as Exp.I-4, the solute was injected as line source near the junction of main
channel and floodplain at y=0.415m in a vertical line along depth of the main channel. In both
the experiments, solute was injected into the flow at longitudinal distance x=2 m from the
channel inlet and measuring sections were considered at four different cross-sections along
streamwise direction located at downstream of the channel inlet, i.e., at a distance of 3 m, 4 m, 6
m and 9 m from the inlet section across the channel depth. Diluted common table salt (NaCl) at
flow rate of 1 g/s was applied as tracer material. For respective measurements of concentration in
the channel, specially designed conductivity meter (Fraselle et al., 2008) were used. Dye
injection and measurement locations in experimental runs in symmetric compound channel are
detailed in Table 5.1.
119
Table 5.1 Solute injection and measurement locations in simulation experiments in wide
symmetric and narrow asymmetric compound channels
Channel
Channel
type
length,
L
Exp.
runs
(m)
Symmetric
(Fraselle et
al., 2008)
mid of lower
layer
Cs & Qs
12.55 g/l
0.0797 l/s
I-4
0.415
mid of both
lower &
upper layers
Cs & Qs
16.74 g/l
0.0597 l/s
C-1
C-2
0.05
0.1
mid of upper
layer
Cs & Qs
2500 ppb
54 ml/min
10.0
20.0
Injection
concentration
(Cs) & flow
rate
(Qs)
0.6
I-1
Asymmetric
(Shiono
and Feng,
2003)
2.0
C-3
13.0
0.15
Simulated
locations d/s
of channel
inlet
(m)
3, 4, 6 & 9
14.0
The solute transport model coupled with hydrodynamic model developed for governing
equations for solute transport, presented in Subsection 3.2.2 of Chapter 3, is used to simulate the
transport of passive materials in wide symmetric and narrow asymmetric compound channels
reported in the literature (Fraselle et al., 2008; Shiono and Feng, 2003).
As the transport model proposed in present study is two layered two dimensional in nature, the
vertical locations of solute injection and observation points in respect of each numerical
experimental set were placed at mid depth in respective layers of both channels. Turbulent
Schmidt number, t , was considered as 0.5 for the compound channel. After attaining the steady
state condition, model simulations were continued further for about 25000 time steps in
symmetric and 30000 time steps in asymmetric compound open-channels for acquiring time
averaged results.
120
In the numerical experiments of solute transport, grid spacing similar to that of hydrodynamic
model (see Subsection 4.3.1.1a in Chapter 4), i.e., streamwise x spanwise, 500x70 grid cells in
upper layer and 500x26 grid cells in lower layer for symmetric channel, has been adopted. The
predicted layered average solute concentrations were converted into depth averaged value for
comparing the same with experimental data. The data for symmetric compound channel were
available in terms of their depth averaged value (Fraselle et al., 2008; Fraselle et al., 2010).
121
122
123
In experiments, the solute was injected as point source in Exp.I-1 at a vertical distance of 0.0375
m above the main channel bed while in Exp.I-4, solute was injected as a line source along the
depth of the whole main channel as mentioned in Table 5.1. In Figs. 5.3(a-c), significant
differences in distribution of solute concentration between the computed and the experimental
results are also due to under estimation of secondary currents by the present model near the
1.2
0.8
0.4
0
2
0.8
0.6
0.4
2
Fig. 5.2 Isolines plot of depth-averaged solute concentration (g/l) in layers of wide symmetric
compound channel for Exp.I-1
Figs. 5.2(a-b) and 5.4(a-b) show the simulated concentration distributions, both in upper and
lower layers, for experiments I-1 and I-4 respectively. From both the figures, it is evident that
mixing of solute is significant in upper layer even though the solute was injected in lower layer
(Exp.I-1). The existence of vertical velocity component causes higher rate of distribution of
124
125
2.5
3.5
4.5
5.5
6.5
6.5
0.8
0.6
0.4
2
2.5
3.5
4.5
5.5
In the numerical experiments of solute transport in asymmetric compound channel, grid spacing
similar to that of hydrodynamic model (see Subsection 4.3.1.2 in Chapter 4), i.e., streamwise x
spanwise, 2000x40 grid cells in upper layer and 2000x20 grid cells in lower layer were adopted.
The predicted depth averaged solute concentrations in upper layer are compared with
experimental data. The experimental data is available in literature as point observed value, i.e., at
the level of 0.105 m above the channel bed (Shiono and Feng, 2003; Shiono et al., 2003).
Fig. 5.5(a) Solute concentration profile in narrow asymmetric compound channel for Exp.C-1
127
The injection point in the numerical experiment has been located at 13 m downstream of inlet
channel at midpoint of upper layer. The differences between observed and simulated
concentration profiles for all these experimental sets are due to, (a) vertical locations of injection
point in physical and numerical experiments are different, (b) measurements of concentration in
physical experiments are located at the same vertical positions as location of injection points
while the simulated results are reported for depth averaged concentration of upper layer, (c) The
existence of the 3-D effects in the narrow channel, are not well reproduced in 2-D depth
averaged model.
Figs. 5.5(b), 5.6(b) and 5.7(b) show the simulated depth-averaged concentration isolines, both for
upper and lower layers, for experiments C-1, C-2 and C-3 respectively. From the figures, it is
apparent that concentration of solute is predominant in upper layer only as the injection point is
located in upper layer and existence of vertical velocity, though magnitude is very small, at the
interface.
0.2
0.1
0
13
13.5
14
14.5
0.1
0.05
0
13
13.5
14
14.5
128
Fig. 5.6(a) Solute concentration profile in narrow asymmetric compound channel for Exp.C-2
0.2
0.1
0
13
13.5
14
14.5
0.1
0.05
0
13
13.5
14
14.5
129
Fig. 5.7(a) Solute concentration profile in narrow asymmetric compound channel for Exp.C-3
0.2
0.1
0
13
13.5
14
14.5
0.1
0.05
0
13
13.5
14
14.5
The performance of solute transport model developed in present study using dynamic SGS
scheme has been assessed for experimental data of symmetric and asymmetric compound
channels from literature. Transport process has been investigated through injection of inert
pollutant at various locations in compound open-channels. The model performs better for
pollutant injected at other than shear layer locations. Significant differences in distribution of
solute concentration between the computed and the experiments results are due to under
estimation of secondary currents near the junction of main channel and floodplain. However, the
performance of the model is better at downstream locations as compared to upstream locations
due to better mixing of pollutants in the said locations. Further, the presence of vertical velocity
component at horizontal interface causes higher rate of distribution of concentrations vertically
as compared to in horizontal directions.
131
Chapter 6
Conclusions
6.1 GENERAL
A two-layer 2-D hydrodynamic and solute transport model for turbulent flows in straight
compound channel is developed using turbulence closure schemes based on LES approach. The
turbulent closure schemes employed are standard Smagorinsky (SS) scheme and dynamic
subgrid scale (DSGS) scheme. The governing equations are solved using finite volume method
based on a two step predictor-corrector scheme. The developed model is tested with the
experimental data of symmetric and asymmetric straight compound channels from literature. The
major findings of foregoing study are presented in the following paragraphs:
a) The model result on flow parameters like interfacial shear stress and vertical flow
velocity gives insight about the overall momentum exchange between the main channel
and the floodplain of straight compound open-channel.
b) Based on sensitivity analysis of model parameters C and , it is found that average
values of C and equals to 0.1 and 0.5 respectively are adequate for simulation of
hydraulic characteristics using developed model.
132
c) The significant difference in layered average primary flow velocities between lower and
upper layers, i.e., of the order of 10-14% of mean bulk velocity (Ub), has been simulated
near the junction of the compound channels.
to 1.35
spanwise and vertical velocities are analyzed at the horizontal interface between the
lower layer and the upper layer. The analyses revealed the existence of horizontal
vortices near the junction of the main channel and the floodplain where momentum
exchange takes place.
f) The developed model has been able to simulate satisfactorily the depth averaged primary
velocity in compound open channels with RMSE in the range 0.017 m/s to 0.026 m/s.
The model is able to reproduce one inflection point in case of shallow flows in wide
symmetric compound channel while two inflection points in case of deep flows in narrow
asymmetric compound channel.
g) The maximum secondary currents are generated (between 0.02% of Ub to 1.1% of Ub)
near the junction of the main channel and the floodplain; the direction of the currents
indicates the possible trend of mass and momentum transfer from lower layer to upper
layer. Horizontal contour plots of secondary currents, bed and interfacial shear stresses
indicate the formation of vortices near the junction of the main channel and the
floodplains.
h) The variation of depth averaged dynamic model parameter, C, in the compound channels
has been found in the range of 0.02 to 0.145. The model parameter, C, depends on the
variation of grid sizes and flow conditions.
133
i) The DSGS model has been found to give marginally better results vis--vis SS model in
simulating hydraulic characteristics such as flow velocities, turbulent shear stresses in
compound open channel flows.
a) The DSGS model has been able to predict the peak solute concentrations irrespective of
point of injection in the compound channel cross-section. However, spanwise distribution
of concentration across the junction of the main channel and the floodplain, depends upon
the magnitude of secondary currents.
b) The developed model has been able to simulate the concentration of solute in both
symmetric (Exps.I-1, I-4) and asymmetric (Exp.C-1, C-2, C-3) compound open channels
with RMSE ranging from 0.01 g/l to 0.05 g/l and 0.73 ppb to 1.19 ppb respectively.
a) Performance of present model can be improved further by adding additional terms at the
junction of the main channel and floodplain, responsible for generating secondary
currents at those regions.
134
b) Empirical relations can be modified for efficient turbulent exchange across the horizontal
interface between the layers.
c) Performance of hydrodynamic model can be considered for density driven flows while
giving due consideration to the density of fluids.
d) The developed model can be made more realistic by incorporating pressure gradient
terms using SIMPLE like procedure.
e) Active pollutants term can be added in governing equations for model simulation to
tackle real world problems.
135
References
Abbott, M.B. (1979). Computational Hydraulics: Elements of the theory of free surface flows,
Ashgate, Boston.
Ackers, P. (1993). Flow formulae for straight two-stage channels, J. Hyd. Res., 31(4), 509
531.
Alcrudo, F. and Navarro, P.G. (1993). A high resolution Godunov type scheme in Finite
volumes for the 2D shallow water equations, Int. J. Num. Meth. Fluids, 16, 489-505.
Bardina, J., Lyrio, A., Kline, S.J., Ferziger, J.H. and Johnston, J.P.(1981). A prediction method
of planar diffuser flows, J. Fluids Engg., 103, 315-321.
Cater, J.E., and Williams, J.J.R. (2008). Large eddy simulation of a long asymmetric compound
open channel, J. Hyd. Res., 46(4), 445-453.
Celik, I.B. (1999). Introductory Turbulence Modeling. Lectures Notes, West Virginia University
Mechanical & Aerospace Engineering Dept.
Chatila, J.G. and Townsend, R.D. (1998). Modelling of pollutant transport in compound openchannels, Canadian Water Res. J., 23(3), 259-271.
Chau, K.W. and Jin, H.S. (1995). Numerical solution of two-layer, two-dimensional tidal flow
in a boundary fitted orthogonal curvilinear co-ordinate system, Int. J. Num. Meth. Fluids, 21,
1087-1107.
Chow, V.T. (1959). Open Channel Hydraulics, MacGraw-Hill, NewYork.
Cokljat, D. and Younis, B. (1995). Second order closure study of Open channel flows, J. Hyd.
Engg., ASCE, 121(2), 94-107.
Craft, T.J., Suga, K. and Launder, B.E. (1993). Extending the applicability of eddy viscosity
models through the use of deformation invariants and non-linear elements, Proc 5th Int. Symp.
on Refined Flow Modelling and Turbulence Measurements, 125132.
Cunge, J. (1975). Two dimensional modelling of flood plains, in: Mahmood K. and Yevjevich
V., eds., Unsteady flow in open channels, Water Resources Publications, Fort Collins, Colorado,
705-762.
Djordjevic, S. (1993). Mathematical model of unsteady transport and its experimental validation
in compound open channel flow, J. Hyd. Res., 31(2), 229248.
136
Fernandes, J.N., Leal, J.B. and Cardoso, A.H. (2014), Longitudinal development of compound
channel flows, 3rd IAHR European Congress, Porto, Portugal.
Fraselle, Q., Arnould, T., Lissoir, X., Bousmar, D. and Zech, Y. (2008). Investigating diffusion
and dispersion in compound channels using low cost tracer, Proc. River Flow 2008 conf. on
Fluvial Hydraulics, Cesme, Izmir, Turkey, September 3-5, 1, 529-538.
Fraselle, Q., Bousmar, D., and Zech, Y. (2010). Diffusion and dispersion in compound
channels: The role of large scale turbulent structures, Proc. 1st European Congress, IAHR,
Ediburge, UK, May 4-6, No.FPIVb.
Germano, M., Piomelli, U., Moin, P. and Cabot, W. H. (1991). A dynamic subgrid scale eddyviscosity model Physics of Fluids, A3(7), 17601765.
Ghose, S. and Kline, S.J. (1978). The computation of optimum pressure recovery in twodimensional diffusers, J. Fluid Engg., 100, 419-426.
Ghosh, S. N. and Jena, S. B. (1971). "Boundary shear distribution in open compound channel",
Proc. I.C.E, 49.
Ghosh, S. N. and Mehta, P.J. (1974). "Boundary shear distribution in a channel with varying
roughness distribution", Proc. I.C.E., 57.
Hosoda, T., Sakurai, T., Kimura, I., and Muramoto, Y. (1999). 3-D computations of compound
channel flows with horizontal vortices and secondary currents by means of non-linear k-
model, J. Hydros. Hyd. Engg., JSCE, 17(2), 87-96.
Jin, H.S., Egashira, S. and Liu, B.Y. (1998). Characteristics of meandering compound channel
flow evaluated with two-layered, 2-D method, J. Hydros. Hyd. Engg., 16(1), 87-96.
Joung, Y. and Choi, S.U. (2008). Investigation of twin vortices near the interface in turbulent
compound open-channel flows using DNS data, J. Hyd. Engg., ASCE, 134(12), 1744-1756.
Kara, S., Stoesser, T. and Sturm, T.W. (2012). Turbulence statistics in compound channels with
deep and shallow overbank flows, J. Hyd. Res., 50(5), 482-493.
Kawahara, Y., Fukuoka, S., Ikeda, S., Pender, G. and Shimizu, Y. (2008). Chapter 4: Computer
simulation, Flow and sediment transport in compound channels: The experiences of Japanese
and UK research, S. Ikeda & I.K. McEwan (Eds.), IAHR Monograph Series, 167-234
Knight, D.W and Demetriou, J.D. (1983). Floodplain and main channel flow interaction J.
Hyd. Engg., ASCE, 109(8), 1073-1092
Knight, D.W. and Hamed, M.E., (1984). Boundary shear in symmetrical compound channels,
J. Hyd. Engg., 110(10), 1412-1430
137
Launder, B.E. and Spalding, D.B. (1974). The numerical computation of turbulent flows,
Comp. Meth. Applied Mech. Engg., 3, 269-289.
Lesieur, M., Metais, O. and Comte, P. (2005). Large eddy simulation of turbulence. Cambridge
University Press, Cambridge.
Lin, B. and Shiono, K. (1995). Numerical modelling of solute transport in compound channel
flows, J. Hyd. Res., 33(6), 773-788.
McKeogh, E.J. and Kiely, G. (1989). Experimental study of the mechanisms of flood flow in
meandering channels, Proceedings of the 23rd Int. Ass. Hydraulics Res. (IAHR), Congress,
Ottawa.
Mehrer, H. and Stolwijk, N.A. (2009). Heroes and highlights in the history of diffusion,
diffusion-fundamentals.org, 11(1), 1-32.
Myers, W.R.C. and Elsawy, E.M. (1975). "Boundary shear in channel with floodplain", J.
Hydraulic Div., ASCE, 101(7), 933-946.
Myers, W.R.C. (1978). "Momentum transfer in a compound channel", J. Hydraulic Res., IAHR,
16(2), 139-150.
Nadaoka, K. and Yagi, H. (1998). "Shallow-water turbulence modeling and horizontal largeeddy computation of river flow", J. Hydraulic Engg., ASCE, 124(5), 493-500.
Naot, D. and Rodi, W. (1982). Calculation of secondary currents in channel flow, J. Hydraulic
Div., ASCE, 108(8), 948968.
Naot, D., Nezu, I., and Nakagawa, H. (1993). Hydrodynamic behaviour of compound
rectangular open channels, J. Hydraulic Engg., ASCE, 119(3), 390-408.
Nezu, I. (1996). Experimental and numerical study on 3-D turbulent structures in compound
open-channel flows. Flow modelling and Turbulence measurements, edited by C.J. Chen et al.,
Balkema, 65-74.
Nezu, I. and Nakayama, T. (1997). Space-time correlation structures of horizontal coherent
vortices in compound channel flows by using particle-tracking velocimetry, J. Hydraulic Res.,
IAHR, 35(5), 191- 208.
Nezu, I. and Rodi, W. (1986). Open channel flow measurements with a laser Doppler
anemometer, J. Hydraulic Eng., 112(5), 335-355.
Nezu, I., Sanjou, M. and Goto, K. (2004). Transition process of coherent vortices in depthvarying unsteady compound open-channel Flows. Shallow Flows (ed. G.H. Jirka & W.S.J.
Uijttewaal), Balkema, 251-258.
138
Nugroho, E.O. and Ikeda, S. (2007). Comparison study of flow in a compound channel:
Experimental and numerical method using large eddy simulation SDS-2DH model, ITB J.
Engg. Sc., 39B(2), 67-97.
Nujic, M. (1995) Efficient implementation of non-oscillatory schemes for the computation of
free-surface flows, J. Hydraulic Res., IAHR, 33(1), 101-111.
Othman, F. and Valentine, E.M. (2006). Numerical modelling of the velocity distribution in a
compound channel, J. Hydrol. Hydromech., 54(3), 269279
Patel, V.C. (1965). Calibration of the Preston tube and limitations on its use in pressure
gradients, J. Fluid Mech., 23(1), 185-208.
Pender, G. and Manson, J.R. (1994). Developments in three-dimensional numerical modelling
of river flows, Proc. 2nd Int. Conf. River Flood Hyd., 22-25 March, York, UK, Wiley
Pezzinga, G. (1994). Velocity distribution in compound channel flows by numerical
modelling, J. Hydraulic Engrg., 120(10), 11761198.
Prandtl, L. (1925). Bericht ber Untersuchungen zur ausgebildeten Turbulenz, Z. Angew.
Math, Meth., 5, 136-139.
Prandtl, L. (1932). Recent Results of Turbulent Research, translation by National Advisory
Committee for Aeronautics, TM No. 720 (originally published in German in 1932).
Premnath, K.N., Pattison, M.J. and Banerjee, S. (2009). Generalized lattice Boltzmann equation
with forcing term for computation of wall-bounded turbulent flows, Physical Review, E79 (2,
Part 2).
Price, R.K. (1975). "A mathematical model for river developments Theoretical development",
Report No. INT 127, December.
Rajaratnam, N. and Ahmadi, R.M. (1970). "Interaction between main channel and floodplain
flows", J. Hyd. Div., ASCE, 105(5), 573-588.
Rajaratnam, N. and Ahmadi, R.M. (1981). "Hydraulics of channels with floodplains", J. Hyd.
Res., IAHR, 19(1), 43-60.
Rodi, W. (1980). Turbulence Models and Their Application in Hydraulics, Delft, Netherland,
IAHR.
Sanjou, M., Nezu, I. and Itai, K. (2010). Space-time correlation and momentum exchanges in
compound open-channel flow by simultaneous measurements of two-sets of ADVs, River Flow
2010 - Dittrich, Koll, Aberle and Geisenhainer (eds), Bundesanstalt.
139
Thomas, T.G. and Williams J.J.R. (1995), Large eddy simulation of turbulent flow in an
asymmetric compound open channel, J. Hyd. Res., 33(1), 27-41.
Tominaga, A., Ezaki, K., Nezu, I. and Nakagawa, H. (1989). Three-dimensional turbulent
structure in straight open channel flows, J. Hydraulic Res., 27, 149173
Tominaga, A. and Nezu, I. (1991). Turbulent structures in compound open-channel flow, J.
Hyd. Engg., ASCE, 117(1), 2140.
Von Karman, T.H. (1930). Mechanische aehnlichkeit und turbulenz, Proc. Third Intern.
Congress for Appl. Mech., Stockholm.
Wilcox, D.C. (1994) Turbulence modelling for CFD, DCW Industries, Inc., California.
Wood, I.R. and Liang, T. (1989). Dispersion in an open channel with a step in the crosssection, J. Hyd. Res., 27(5), 587601.
Xie, Z., Lin, B. and Falconer, R.A. (2013). Large-eddy simulation of the turbulent structure in
compound open-channel flows, Adv. Water Res., 53, 66-75.
Yang, K., Nie, R., Liu, X. and Cao, S. (2013). Modelling depth-averaged velocity and boundary
shear stress in rectangular compound channels with secondary flows, J. Hyd. Engg., ASCE,
1329(1), 76-83.
Yee, H. (1989). A class of high resolution explicit and implicit shock-capturing methods,
NASA Report N89-25652.
Yuan, M., Song, C.S. and He, J. (1988),Numerical analysis of turbulent flow in a twodimensional non-symmetric plane-wall diffuser, J. Fluids Engg,, 113, 210-215.
Zarrati, A.R., Jin, Y.C. and Karimpour, S. (2008). A semi-analytical model for shear stress
distribution in simple and compound open channels, J. Hyd. Engg., ASCE, 134(2), 205-215.
Zheleznyakov, G.V. (1971). "Interaction of channel and floodplain streams", Proc. 14th Int.
Congr., IAHR, Paris.
141
Research Publications
1. Development of two-layered model for compound open-channel flow Journal of
Hydraulic Engineering, ISH, 2014, Vol. 20, Issue 3, pp.250-262; published by Taylor and
Francis, UK.
2. Modelling of solute transport in a two-layered compound open channel flows by
using dynamic SGS closure scheme submitted to the Journal of Hydrodynamics,
Elsevier. (Paper submitted in May 2016 and reported under review).
3. Flow characteristics at horizontal interface of asymmetric compound open
channel Proceedings of 21st International Conference on Hydraulics, Water Resources
and Coastal Engineering, HYDRO-2016 International, CWPRS, Pune, Dec.8-10, 2016.
4. Hydrodynamic characteristics of flows in a two-layered compound open channels
by using dynamic SGS model National Conference on Water Resources & Flood
Management with special reference to Flood Modelling, October 14-15, 2016. SVNIT,
Surat, pp.WRF-24-1 to WRF-24-16.
5. Development of two-layered model for compound open channel flow Proceedings
of National Conference on Hydraulics and Water Resources, HYDRO-2012, IIT
Bombay, India, Dec.7-8, 2012, pp.395-404.
142