Sunteți pe pagina 1din 275

ESTIMATION OF ROUGHNESS COEFFICIENTS

IN OPEN CHANNELS





Nguyen, Thu Hien

Submitted in total fulfilment of the requirements
of the degree of Doctor of Philosophy





July 2006



Department of Civil and Environmental Engineering
The University of Melbourne


Abstract
An accurate estimation of Mannings n roughness coefficient is of primary importance
in any hydraulic study involving open-channel flows. However, the estimation of this
coefficient for a natural channel is not a trivial task as it depends on many factors such
as surface roughness, vegetation condition, cross-sectional shape, channel irregularity
and flow conditions. The literature shows that there is no universal method of
selecting the n value. Further study in this area is still required to improve the quality
of the estimation of Mannings n.
This study focuses on aspects of the estimation of Manning's n in open channels using
flow measurements for both steady and unsteady flows. For steady flow, the method of
using two-point velocity data to estimate Mannings n is reinvestigated. Two new
formulae for estimating this coefficient are derived. The sensitivity analyses of the
relative error in n and the relative errors in measurements are theoretically and
experimentally investigated. The proposed formulae are applied and verified for a
number of rivers where the method is applicable. They are also compared with
different current empirical formulae. It is suggested that this method can be used as a
means to estimate the roughness coefficients for wide streams where two-point
velocity data are available.
For unsteady flow, the roughness coefficient(s) embedded in the momentum equation
needs to be estimated either by trial and error methods or by an automatic calibration
approach. The latter approach is also known as the inverse problem or the roughness
identification problem. This part of the study focuses on this approach where the main
study objectives are to investigate the modelling factors affecting the quality of the
identified roughness coefficient(s) and to extend the method to channels with
compound sections and varying roughness between the main channel and the
floodplain.
A roughness identification model for unsteady flow applicable to compound channels
is developed. The implicit finite difference Pressmann scheme is adopted to solve the
iii

Saint-Venant equations. The compound channel is treated as a divided channel section
in which for any depth the conveyance of the compound section is then the sum of the
main channel and floodplain conveyances. The algebraic equation system is linearised
and solved by using the double sweep algorithm. The objective function of least
square errors between observed and simulated data was chosen for this inverse
problem, which is solved using the Powell algorithm. Manning's n is considered for
two cases: constant and stage dependent.
Synthetic data are used to investigate the modelling factors affecting the quality of the
identified roughness coefficients and the performance of the model for compound
channels. This investigation shows that understanding these modelling factors is very
important in avoiding an unidentifiable inverse problem and to improve the quality of
the identified parameters. Several recommendations to improve the quality of the
identified roughness coefficient are made. For the compound channel case, the results
show that the model can identify both the roughness coefficients of the main channel
and the floodplain. The quality of the identified floodplain roughness coefficient is
usually poorer than the main channel. However, it can be improved considerably when
the depth on the floodplain becomes significant compared to the main channel depth.
Moreover, in cases when the cross-sections along the reach do not change much, an
alternative approach of identifying the conveyance K is suggested.
The roughness identification model is applied to two natural rivers, the Goulburn River
in Victoria, Australia and the Duong River in the Red River delta, Vietnam. Depending
on the data availability, the characteristics of the cross-sections, the flow variation in
the main channel and the floodplain, the performance of the model is investigated and
several problems related to the roughness coefficients of the study reaches are
explored.
iv

Declaration
This is to certify that:
i. The thesis comprise only my original work towards the PhD except where
indicated in the Preface,
ii. Due acknowledgement has been made in the text to all other material used,
iii. The thesis is less than 100,000 words in length, exclusive of tables, maps,
bibliographies, appendices and footnotes.


Nguyen, Thu Hien
July 2006
v

Preface
A substantial portion of the work described in this thesis has been published in the
papers listed below.
Nguyen, H. T., and Fenton, J. D. (2004). Identification of roughness in open
channels, Proceedings of the 6th International Conference on Hydro-Science and
Engineering, Brisbane, Australia, The University of Mississippi, May 31-June 3, 2004,
CD-ROM.
Nguyen, H. T., and Fenton J. D. (2004). Using two-point velocity measurements to
estimate roughness in streams, Proceedings of 4th Australian Stream Management
Conference, Launceston, Tasmania, 19-22 October 2004, 445-450.
Nguyen, H. T., and Fenton, J. D. (2005). Identification of roughness for flood routing
in compound channels, Proceedings of the 31th IAHR Congress on Water Engineering
for the future: Choices and Challenges, Seoul, Korea, September 11-16, 2005, CD-
ROM.
Nguyen, H. T., and Fenton, J. D. (2005). Identification of roughness in compound
channels, Proceedings of the International Congress on Modelling and Simulation -
Advances and Applications for Management and Decision Making MODSIM05,
Melbourne, Australia, December 12-15, 2005, CD-ROM.

vi

Acknowledgements
I gratefully acknowledge the excellent supervision and support provided by my
supervisors, Professor John D. Fenton and Associate Professor Roger Hughes. They
have saved no efforts in providing the support, supervision, understanding and advice
throughout the time of my study at the University of Melbourne. Special thanks and
appreciation also go to Professor Tom McMahon, Ass. Professor Nichola Haritos and
Dr. Andrew Western for their assistances and helpful discussions during my supervisor
was going for study leave.
I wish to extend my thanks to Mr. Joska Shepherd for his efforts in setting up the
experimental arrangements of the study. I also appreciate Dr. Yebegaeshet T. Zerihun
for his helpful advice regarding the use of the ADV instrument.
I also would like to thank Ms Barbara Dworakovski, Thiess Environmental Services,
Pty Ltd. for providing me the data for the rivers in Australia. Special thanks to Dr.
Tony Ladson for the roughness information of the rivers in Victoria, Australia. I owe
many thanks to Dr. Murray Hicks and staff in NIWA, New Zealand for the velocity
measurement data of the rivers in New Zealand. Special thanks to Dr. Mike
Stewardson for the topography data of the Goulburn River in Victoria, Australia. I also
would like to thank the staff of the Bureau of Meteorology and Hydrology and The
Institute of Water Resources Planning, Vietnam for their help in providing
hydrological and topographical data for the Duong River in Vietnam.
The financial support of Australian Development Scholarship (ADS) is gratefully
acknowledged.
I am indebted to my family: my parents and my brother. I thank them for their
continual support and encouragement throughout my years of education and I sincerely
dedicate this thesis to them. I extend deep gratitude to my husband and my two
daughters for their love, support, patience and understanding during the course of the
PhD study.
vii
Table of contents
Table of contents
Abstract. .................................................................................................iii
Declaration.................................................................................................. .v
Preface... ................................................................................................ vi
Acknowledgements ...................................................................................vii
Table of contents.......................................................................................viii
List of Figures............................................................................................xii
List of tables ............................................................................................xvii
List of notations ......................................................................................... xx
Chapter 1 Introduction............................................................................ 1
1.1 Description of the study ...................................................................................... 1
1.1.1 General ...................................................................................................... 1
1.1.2 Background of the research ...................................................................... 2
1.1.3 Scope and objectives................................................................................. 5
1.2 Research methodology ........................................................................................ 5
1.3 Outline of the thesis............................................................................................. 6
Chapter 2 Methods of estimating roughness coefficients in open
channels................................................................................... 9
2.1 Introduction......................................................................................................... 9
2.2 Flow resistance in open channels ........................................................................ 9
2.3 Methods of estimating Manning's n for steady flow in open channels ............ 12
2.3.1 Direct measurement method ................................................................... 12
2.3.2 Using the tables or guidelines ................................................................. 14
2.3.3 Photographic comparisons...................................................................... 15
2.3.4 Fitting calculated water surface profile to observed levels..................... 16
2.3.5 Field measurements of velocity distribution........................................... 16
2.3.6 Empirical formulae ................................................................................. 18
viii
Table of contents
2.3.7 Lump roughness coefficient model......................................................... 22
2.4 Estimating roughness coefficient for unsteady flow in open channels ............. 23
2.4.1 Trial and error method ............................................................................ 24
2.4.2 Automatic calibration methods............................................................... 25
2.5 Equivalent roughness coefficient for compound channels................................ 28
2.6 Variation of Manning's n with flow.................................................................. 30
2.7 Summary of key reviews................................................................................... 33
Chapter 3 Estimation of Roughness Coefficients Using Velocity Data
................................................................................................ 35
3.1 Introduction....................................................................................................... 35
3.2 Background ....................................................................................................... 36
3.2.1 Velocity distribution in turbulent flow ................................................... 36
3.2.2 Current method using velocity data to estimate Mannings n ................ 39
3.3 Proposed method using velocity data to estimate Mannings n........................ 41
3.3.1 Using average equivalent roughness....................................................... 41
3.3.2 Using average shear stress ...................................................................... 43
3.4 Sensitivity analysis............................................................................................ 44
3.4.1 Sensitivity analysis.................................................................................. 44
3.4.2 Experimental work and analysis ............................................................. 46
3.5 Application to natural rivers.............................................................................. 57
3.5.1 Data......................................................................................................... 58
3.5.2 Results and discussion ............................................................................ 59
3.6 Comparing the performance of the proposed method with other empirical
equations............................................................................................................ 65
3.7 Summary and conclusion .................................................................................. 71
Chapter 4 Roughness Identification Model Development for
Unsteady Flow...................................................................... 73
4.1 Introduction....................................................................................................... 73
4.2 General discussion............................................................................................. 73
4.3 Hydraulic sub-model ......................................................................................... 76
4.3.1 Governing equations of 1-D unsteady flow in open channels ................ 76
4.3.2 Numerical methods for unsteady flow computation............................... 77
ix
Table of contents
4.3.3 Implicit Pressmanns scheme.................................................................. 81
4.3.4 Compound channel models..................................................................... 89
4.3.5 Identification of roughness functions...................................................... 94
4.3.6 Identification of conveyance function .................................................... 96
4.4 Optimisation sub-model .................................................................................... 97
4.4.1 Objective functions ................................................................................. 97
4.4.2 Optimisation methods ........................................................................... 101
4.5 Summary ......................................................................................................... 104
Chapter 5 Numerical experiments on the roughness identification
problem............................................................................... 105
5.1 Introduction..................................................................................................... 105
5.2 Generation of synthetic data............................................................................ 106
5.2.1 Cross-sectional data .............................................................................. 106
5.2.2 Flow data............................................................................................... 107
5.3 Model performance and preliminary results ................................................... 111
5.4 Modelling factors affecting the quality of the identified roughness
coefficient.................................................................................................... 115
5.4.1 Effect of weighting factor () ............................................................... 116
5.4.2 Effect of computational grid sizes ........................................................ 122
5.4.3 Effect of different combinations of boundary conditions..................... 126
5.4.4 Effect of errors in initial conditions...................................................... 139
5.5 Roughness identification in compound channels ............................................ 142
5.5.1 Model performance............................................................................... 142
5.5.2 Factors affecting the quality of the identified roughness coefficients in
compound channels............................................................................... 144
5.6 Identification of roughness functions.............................................................. 148
5.7 Identification of conveyance functions ........................................................... 154
5.8 Summary and conclusion ................................................................................ 159
Chapter 6 Application of the roughness identification problem to
natural rivers ...................................................................... 162
6.1 Introduction..................................................................................................... 162
6.2 Goulburn River................................................................................................ 162
x
Table of contents
6.2.1 General description of study area ......................................................... 162
6.2.2 Data....................................................................................................... 164
6.2.3 Application, results and discussions ..................................................... 166
6.3 Duong River .................................................................................................... 183
6.3.1 General description the study area........................................................ 183
6.3.2 Data....................................................................................................... 185
6.3.3 Application, results and discussions ..................................................... 186
6.4 Summary and conclusion ................................................................................ 196
Chapter 7 Conclusions and Recommendations................................. 199
7.1 Summary ......................................................................................................... 199
7.2 Conclusions ..................................................................................................... 200
7.3 Recommendations for further research ........................................................... 206
References ........................................................................................ ..209
Appendix A Measured and computed roughness coefficients using
velocity data of the 14 rivers................................................ 224
Appendix B Solution algorithm for a general river network .................. 226
Appendix C Powells method.................................................................. 231
Appendix D Preliminary results of the identified roughness coefficients
using synthetic data.............................................................. 235
Appendix E Data for the Goulburn River ............................................... 240
Appendix F Data for the Duong River.................................................... 248
xi
List of Figures
List of Figures
Figure 3.1 Two-point velocity measurement at a cross-section ................................... 39
Figure 3.2 Theoretical relationship between relative errors in estimated n and relative
errors in x with different depths for a channel with 035 . 0 = n ................... 45
Figure 3.3 Theoretical relationship between relative errors in estimated nand relative
errors in x for different roughness channels with a flow depth of 2 m.......... 46
Figure 3.4 The experimental set-up diagram (not to scale). ......................................... 47
Figure 3.5 The two roughness types were carried out in the experiment. .................... 48
Figure 3.6 Velocity time series measurement at 7 . 1 = z cm of the gravel bed flume of
8.5 cm water depth. ....................................................................................... 50
Figure 3.7 Average SNR and COR values and recorded flags for measurement at
cm of the gravel bed flume of 8.5 cm water depth. .......................... 50 7 . 1 = z
Figure 3.8 Variation of velocity profiles at different locations along the flume .......... 51
Figure 3.9 Measured velocity profiles.......................................................................... 53
Figure 3.10 Experimental relationships between relative errors in x and relative errors
in estimated n and corresponding theoretical lines for wire mesh roughness
with different depths ..................................................................................... 56
Figure 3.11 Experimental relationships between relative errors in x and relative errors
in estimated n and corresponding theoretical lines for the gravel bed with
different depths ............................................................................................. 56
Figure 3.12 Experimental relationships between relative errors in x and relative errors
in estimated n and corresponding theoretical lines for the same depth with
different types of roughness.......................................................................... 57
Figure 3.13 Computed n obtain from velocity data using average equivalent roughness
versus measured n ......................................................................................... 62
Figure 3.14 Computed n obtain from velocity data using average shear stress versus
measured n .................................................................................................... 62
Figure 3.15 Computed n obtained from different equations versus measured n for
the 12 selected rivers..................................................................................... 68
xii
List of Figures
Figure 4.1 Roughness identification procedure............................................................ 75
Figure 4.2 Numerical Pressmanns scheme.................................................................. 81
Figure 5.1 The cross-sections of the model channels................................................. 107
Figure 5.2 Effect of different parameter in Equation (5.1) on the size and shape of the
upstream hydrograph: (a) effect of
p
t , (b) effect of
p
Q , (c) effect of ... 109
Figure 5.3 Noise contamination of the observed stages and discharges at the
observation gauge at the middle of the channel .......................................... 111
Figure 5.4 Variation of the mean of identified parameters with the number of samples
..................................................................................................................... 114
Figure 5.5 Identified roughness coefficient versus with different time steps for
synthetic data with 700 =
p
Q m
3
/s, 8 =
p
t hours, = n 0.035, = S 0.0004,
= x 1 km.................................................................................................... 118
Figure 5.6 Identified roughness coefficient versus with different space steps for
synthetic data with 700 =
p
Q m
3
/s, 8 =
p
t hours, = n 0.035, =
0
S 0.0004,
5 . 0 = t hour ............................................................................................... 118
Figure 5.7 Identified roughness coefficient versus with different channel slopes for
synthetic data with 700 =
p
Q m3/s, 8 =
p
t hours, = n 0.035........................ 120
Figure 5.8 Identified roughness coefficient versus with different channel roughness
for synthetic data with 700 =
p
Q m3/s, 8 =
p
t hours, =
0
S 0.0004 ............. 120
Figure 5.9 Identified roughness coefficient versus with different peak discharges for
synthetic data with 8 =
p
t hours, = n 0.035, =
0
S 0.0004............................ 121
Figure 5.10 Identified roughness coefficient versus with different time to peak
discharge for synthetic data with 700 =
p
Q m
3
/s, = n 0.035, =
0
S 0.0004 121
Figure 5.11 Identified roughness coefficient versus grid sizes................................... 124
Figure 5.12 Sensitivity of computed discharge and stage hydrographs to the value of
roughness coefficient at the middle of the channel with different
combinations of boundary conditions ......................................................... 132
Figure 5.13 Variation of the means of the identified n and their confidence intervals of
95% with noise level 1 . 0 = for the Q-Z combination.............................. 134
Figure 5.14 Variation of the means of the identified n and their confidence intervals of
95% with noise level 1 . 0 = for the Z-Z combination.............................. 135
xiii
List of Figures
Figure 5.15 Variation of the means of the identified n and their confidence intervals of
95% with noise level 1 . 0 = for the Q-Qz combination ........................... 136
Figure 5.16 Variation of the means of the identified n and their confidence intervals of
95% with noise level 1 . 0 = for the Z-Qz combination............................ 137
Figure 5.17 Comparison of the computed stage hydrographs at the cross-section of
30 km with different combinations of boundary conditions................. 140 = x
Figure 5.18 Identified n versus the starting time to calculate the objective function for
the case of the Q-Z combination................................................................. 141
Figure 5.19 Variation of identified n with the starting time to calculate objective
function and the slope of the channel (the errors in initial stage are +0.5 m)
..................................................................................................................... 141
Figure 5.20 The means of identified roughness coefficients and their range at a 95%
confidence level with different noise levels................................................ 144
Figure 5.21 The true roughness and identified roughness curves with noise-free data
..................................................................................................................... 150
Figure 5.22 The true roughness and identified roughness curves when data contained
noise (5 noisy data samples) ....................................................................... 151
Figure 5.23 The true roughness curves and identified roughness curves with noise-free
data.............................................................................................................. 152
Figure 5.24 The true roughness curves and identified roughness curves when data
contained noise (5 noisy data samples)....................................................... 153
Figure 5.25 The true conveyance and identified conveyance curves with noise-free
data.............................................................................................................. 155
Figure 5.26 The true conveyance and identified conveyance curves by using the
polynomial function when data contained noise (5 noisy data samples) .... 157
Figure 5.27 The true conveyance and identified conveyance curves by using the power
function when data contained noise (5 noisy data samples) ....................... 157
Figure 5.28 Observed and simulated hydrographs computed from the identified
roughness functions and the identified conveyance functions.................... 158
Figure 6.1 Study reach of the Goulburn River and the gauging stations along the reach
..................................................................................................................... 163
Figure 6.2 Observed and simulated hydrograph at different gauging stations for the
flood event 21/07 - 06/08/1978 using the identified roughness coefficient 168
xiv
List of Figures
Figure 6.3 Observed and simulated hydrograph at different gauging stations for the
flood event 04/09 - 25/09/1979 using the identified roughness coefficient 169
Figure 6.4 Verification of the identified roughness coefficient by the independent flood
event 27/09 - 09/10/1978 ............................................................................ 170
Figure 6.5 Verification of the identified roughness coefficient by the independent flood
event 29/07 - 05/08/1980 ............................................................................ 171
Figure 6.6 Identified roughness coefficient with different combinations of boundary
conditions and observed data types............................................................. 173
Figure 6.7 Observed and simulated hydrographs for the flood event 27/09
09/10/1978 using the identified roughness coefficient from Z-Z boundary
conditions and the observed stages at Shepparton Golf Club used in the
objective function........................................................................................ 174
Figure 6.8 Identified roughness curves with different flood events at the Shepparton
Golf Club gauge: (a) Linear function of stage, (b) Quadratic function of stage
..................................................................................................................... 176
Figure 6.9 Verification of the identified roughness coefficients by the independent
flood event 08/08 - 31/08/1978................................................................... 179
Figure 6.10 Verification of the identified roughness coefficients by the independent
flood event 25/06 05/07/1981 .................................................................. 180
Figure 6.11 Identified roughness coefficients obtained from different flood events . 181
Figure 6.12 Identified roughness function of the main channel with different flood
events at the Shepparton Golf Club gauge: (a) Linear function of stage, (b)
Quadratic function of stage......................................................................... 183
Figure 6.13 Duong River and the gauging stations along the river ............................ 185
Figure 6.14 Observed and simulated stage hydrograph at different gauging stations by
using the identified roughness coefficients obtained from the flood itself . 189
Figure 6.15 Observed and simulated stage hydrograph at different gauging stations by
using the average values of the identified roughness coefficients obtained
from the 1996 and 1997 floods ................................................................... 190
Figure 6.16 Identified roughness coefficients obtained from different flood events . 191
Figure 6.17 Identified floodplain roughness functions with different flood events at
Ben Ho gauge: (a) Linear function of stage, (b) Quadratic function of stage
..................................................................................................................... 192
xv
List of Figures
Figure 6.18 Variation of the roughness coefficient of the main channel with time
during the flood season ............................................................................... 195
Figure B.1 Model of a river network.......................................................................... 226
Figure C.1 Contours of the function values to show the progress of Powells method
..................................................................................................................... 233
Figure C.2 Flow chart for Powells method ............................................................... 234
Figure E.1 Ten representative cross-sections of the Goulburn River......................... 241
Figure E.2 The computational scheme of the Goulbourn River................................. 243
Figure E.3 Observed stages for flood event 21/07-06/08/1978.................................. 244
Figure E.4 Observed stages for flood event 08/08-31/08/1978.................................. 244
Figure E.5 Observed stages for flood event 27/09-09/10/1978.................................. 245
Figure E.6 Observed stages for flood event 04/09-25/09/1979.................................. 245
Figure E.7 Observed stages for flood event 28/09-31/10/1979.................................. 246
Figure E.8 Observed stages for flood event 29/07-05/08/1980.................................. 246
Figure E.9 Observed stages for flood event 25/06-05/07/1981.................................. 247
Figure E.10 Rating curves at Shepparton and Loch Garry gauges............................. 247
Figure F.1 Sixteen representation cross-sections of the Duong River........................ 249
Figure F.2 The computational scheme of the Duong River........................................ 251
Figure F.3 Observed stages and discharges for flood event 14/8-31/8/1995 ............. 252
Figure F.4 Observed stages and discharges for flood event 16/8-30/8/1996 ............. 252
Figure F.5 Observed stages and discharges for flood event 22/7-6/8/1997 ............... 253
Figure F.6 Observed stages and discharges for flood event 26/7-5/8/1998 ............... 253
xvi
List of Tables
List of tables
Table 2.1 Comparison of calculated and estimated values of Mannings n (after
Sargent (1979))................................................................................................15
Table 2.2 Some empirical formulae in Stricklers form (all
x
d is computed in metres)
.........................................................................................................................19
Table 2.3 Some empirical formulae in Limerinos form ( R and
x
d are in m) ..............20
Table 2.4 Some other empirical formulae .....................................................................21
Table 3.1 Characteristic data of experimental runs .......................................................53
Table 3.2 Computed roughness coefficients from the full velocity profiles .................54
Table 3.3 Brief description of the bed and banks for the selected rivers ......................60
4 Summary of the main hydraulic and roughness characteristics of the studied Table 3.
Table 3.
Table 5.
Table 5.
Table 5.
Table 5.
Table 5.
Table 5.7 Different combinations of boundary conditions..........................................128
rivers............................................................................................................61
Table 3.5 Mean relative error of the computed Manning's n for the selected rivers .63
Table 3.6 Some applicable equations for the selected river..........66
7 Range of the difference between and measured n and estimated n from
different equations and the corresponding values of MAE and MRE .............70
Table 4.1 Classification of mathematical models (after Cunge (1975))........................74
Table 5.1 The parameters of the upstream hydrographs used in many test cases .......108
2 The range and average values of the identified n of 50 samples with different
noise levels (the true value is 0.035) .............................................................113
3 The means of the identified roughness coefficients of 10 samples and their
range with a 95% confidence level (the true value is 0.035).........................114
4 Effect of different grid sizes on the identified roughness coefficient (the true
value is 0.035) ...............................................................................................123
5 Relative errors in the identified roughness coefficient with different time
steps related to channel characteristics..........................................................125
6 Relative errors in the identified roughness coefficient with different time
steps related to the characteristics of flood events ........................................126
xvii
List of Tables
Table 5.8 Identified roughness coefficient from difference boundary combinations (the
true value = n 0.035)......................................................................................129
Table 5.9 Identified roughness coefficient using a single rating curve as a downstream
boundary condition (the true value = n 0.035) ..............................................130
Table 5.10 Errors in the identified n with different combinations of boundary
conditions and different types of observed data at the middle cross-section of
the channel when data contained noise ( 1 . 0 = ) .........................................133
Table 5.11 Means of identified roughness coefficients and their confidence intervals of
95% with different noise levels .....................................................................143
Table 5.12 Identified roughness coefficients for noise-free data with different peak
discharges (the true values of
c
n and
f
n are 0.028 and 0.042 respectively)145
Table 5.13 Errors in the identified roughness coefficients of the main channel and
floodplains when data contain noise ( 05 . 0 = ) ...........................................146
Table 5.14 Identified roughness coefficients for noise-free data with different
observation intervals (the true values of
c
n and
f
n are 0.028 and 0.042
respectively) ..................................................................................................147
Table 5.15 Errors in the identified roughness coefficients of the main channel and
floodplain when data contain noise ( 05 . 0 = ).............................................148
Table 5.16 Identified coefficients of roughness function when data contain noise ....150
Table 5.17 Computation times by using the roughness functions and the conveyance
functions ........................................................................................................156
Table 6.1 Observed peak discharge
p
Q and stage
p
Z
Z
of the selected flood events....165
Table 6.2 Different combinations of boundary conditions..........................................173
Table 6.3 Root mean square errors (RMSE) for different roughness models .............177
Table 6.4 Identified roughness coefficients and RMSEs with different flood events.182
Table 6.5 Observed peak discharge
p
Q and stage
p
of the selected flood events....186
Table 6.6 Identified roughnesses and RMSE with different flood events...................193
Table A.1 Measured and computed roughness coefficients (using velocity data) ......224
Table D.1 Identified roughness coefficients from different noisy data samples .........235
Table D.2 The calculated
2
for 50 samples of noise level = 0.05 .........................238
Table D.3 The calculated
2
for 50 samples of noise level = 0.10 .........................238
xviii
List of Tables
Table D.4 The calculated
2
for 50 samples of noise level = 0.15 .........................238
Table D.5 The calculated
2
for 50 samples of noise level = 0.20 .........................239
.6 Identified roughness coefficients from different noisy data samples for the Table D
Table E.1 Thalweg elevation at each cros
Table F.1 Thalweg elevation at each cross-section in the Duong River .....................248
compound channel.........................................................................................239
s-section in the Goulburn River from
Shepparton to Loch Garry .............................................................................240
xix
List of Notations
List of notations
A = wetted area
s
= water surface width of the inundated area B
B = water surface of the flow
C = Chezys roughness coefficient
c = wave celerity
Cr = Courant number

x
s
D = flow depth
x
E = relative error in x (x is ratio of the velocity at two-tenths the depth to that at eight-
tenths the depth)
n
= relative error in estimated n using velocity measurements from the experiment E
d = height of roughness
d = the bed material diameter such that x percent of material by weight is smaller
Fr = Froude number
f = Weisbachs roughness coefficient
g = acceleration due to gravitation
K = conveyance
k = equivalent sand roughness height
xx
List of Notations
L = reach length
l = characteristic length known as the mixing length
n = Manning's roughness coefficient
c
n = roughness coefficient of the main channel
n
comp
= computed roughness coefficient
e
= equivalent roughness coefficient n
f
n = roughness coefficient of the floodplain
n
meas
= measured roughness coefficient that is determined using the direct method
P = wetted perimeter
Q = discharge
b
= initial discharge of the inflow hydrograph function (for the synthetic data) Q
= friction slope
= bed slope
= water surface slope
p
Q = peak discharge of the inflow hydrograph function (for the synthetic data)
q = lateral inflow per unit length of channel
R = hydraulic radius
Re = Reynolds number
k
Re = roughness Reynolds number
f
S
0
S
w
S
xxi
List of Notations
u = velocity at a point
t = time variable in the Saint-Venant equation
p
= time to peak discharge of the inflow hydrograph function (for the synthetic data) t
2 . 0
u = velocity at two-tenths the depth
8 . 0
u = velocity at eight-tenths the depth
= cross-sectional average shear velocity
= shear velocity
V = cross-sectional average velocity
= to that at eight-tenths the depth or
space variable in the Saint-Venant equation
= observed discharge or stage in the objective function
= simulated discharge or stage in the objective function
z = distance of a point from the solid surface
= constant of integration in the vertical velocity distribution equation
q
u = the velocity component in x direction of the lateral inflow in the Saint-Venant
equations
*
U
*
u
x ratio of the velocity at two-tenths the depth
O
Y
S
Y
0
z
= water stage/water surface elevation
= bed level at a cross-section the level that the discharge equal to zero (cease level)
= elevation of the floodplain
Z
b
Z
f
Z
xxii
List of Notations
Z
0
= minimum water level starting to calculate the roughness function
= parameter of the inflow hydrograph function (for the synthetic data)
tor

ulated stage/discharge
onstant
of the synthetic data
= kinematic viscosity
= weighting coefficient in the objective function
MAE = Mean Absolute Error
MRE = Mean Relative Error
RMSE = Root Mean Square Error
= momentum correction fac
t = computational time step
x = computational space step
= weighting coefficient of Pressmanns scheme
= error between the observed and sim
= von Krmns turbulent c
= mass density of the fluid
= standard deviation to indicate the noise/error level
0
= shear stress on the bed of the flow in the channel.

xxiii
Chapter 1
Chapter 1
Introduction
1.1 Description of the study
1.1.1 General
The knowledge of the roughness/friction coefficient is very important throughout
water engineering. While flow resistance in full pipe flow is one of the most
extensively studied in hydraulic engineering, many difficulties still remain in the
estimation of roughness coefficients for this area in open channels.
In the case of full pipe flow, the friction coefficient depends on relative roughness and
Reynolds number (Nikuradses experiments). The well-known Moody diagram
(Moody 1944) to determine the friction coefficient was developed from the work by
Colebrook (1938-1939). It has been applied with considerable success for determining
pipe flow resistance. However, flow resistance in open channels has been more
difficult to quantify because it depends on many factors such as surface roughness,
cross-sectional shape, vegetation conditions, channel irregularity and flow conditions.
The common roughness coefficients used in practice are Mannings n, Darcy-
Weisbachs friction coefficient f and Chezys coefficient C. When using these
coefficients for open channel resistance, they are theoretically interchangeable and
equivalent. However, for a given type of rough boundary surface, the f or C varies with
flow depth whereas n remains nearly unchanged (Yen 2002). This is the reason for the
popularity of Mannings formula in the open channel hydraulic area. This has made
1
Chapter 1 Introduction
Mannings roughness coefficient n a valuable tool for assessing the effects of open
channel flow resistance.
Over the years, many organisations working on water field have interested in
improving the estimation of roughness coefficients such as US Geological Survey
(USGS), US Army Corps of Engineering (USACE), Land and Water Australia (LWA)
and DSIR Marine and Freshwater in New Zealand. There have been numerous studies
that have made important contributions to open channel flow resistance. As mentioned
above, the value of Mannings n depends on many factors. Although those factors are
identifiable, their individual contributions and their interactions to the value of
roughness coefficient are difficult, if not impossible, to quantify. To select a value of
Manning's n, all of their effects are lumped together into a single coefficient. This is
really a matter of intangibles. At our present stage of knowledge, there is no exact
method of selecting the n value. There are many problems still remaining and yet the
amount of research is almost negligible. Further study in this area is still required.
1.1.2 Background of the research
At present, several methods for estimating Manning's n have been developed
depending on the data availability and application conditions.
For steady flow, when discharges, water stages, friction slopes and some cross-sections
are measured, by using the steady flow equation, the value of the roughness coefficient
can be directly determined. This method is known as the direct method and considered
as a most accurate method to estimate n. It is described and used in Barnes (1967) and
Hicks and Mason (1991). However, this method is time consuming and costly.
Therefore, in current practice this coefficient is usually estimated from
qualitative/empirical methods and empirical formulae.
When flow and cross-section data or surface bed material data are not available, the
qualitative/empirical methods are usually used. For these methods, tables, guidelines
or photographs of channel reaches of known n values can be used to estimate n (e.g.
Chow 1959; French 1985; Urquhart 1975; Barnes 1967; Hicks and Mason 1991; Coon
1998). Although these methods are rather simple and can quickly obtain the
2
Chapter 1 Introduction
preliminary value of n, they are highly uncertain and subjective. Using these methods,
different investigators can provide different values of this roughness coefficient.
Many empirical formulae obtained from experimental flume and natural channels to
estimate the roughness coefficient have been proposed. These are usually in Stricklers
form (see French 1985) or Limerinos form (e.g. Limerinos 1970; Bray 1979). For the
formulae in the Stricklers form, information on the particle size distribution curve of
surface bed material is required. These data are not usually available for many rivers.
For formulae in Limerinos form, besides information on the particle size distribution
curve of surface bed material, the cross-section data (in terms of hydraulic radius) is
also required.
Some others types of empirical formulae come indirectly from slope-area empirical
formulae and are combined with Manning's equation to obtain the value of Manning's
n (e.g. Lacey, 1946; Riggs, 1976; Jarrett, 1984; Dingman and Sharma, 1997). Usually,
these formulae require the information about cross-sections (in terms of hydraulic
radius) and water surface slope. However, their practical applicability and accuracy are
still questionable.
In many rivers, a simple method to measure stream flow is to measure velocity in
several verticals at two-tenths and eight-tenths the depth. From the logarithmic law of
velocity distribution, it can be seen that the velocity distribution depends on the
roughness height, which may be related to Mannings n. If the distribution is known
then the value of n can be determined without any other information such as surface
bed material or hydraulic characteristics and water surface/friction slope of a reach.
Chow (1959) and French (1985) described this method of estimating n for wide rough
channels. However, this is not well established and yet to be applied in practice.
Hence, it is necessary to reinvestigate this method. If it can be shown satisfactory for
practical applications, it will provide an option of estimating Manning's n in wide
channels where two-point velocity observations have been made.
For unsteady flow, although the parameters estimated by the above methods can be a
satisfactory representation of steady behaviour, the values may not be satisfied in these
flow conditions because Mannings n may vary with discharge or depth (Khatibi et al.
3
Chapter 1 Introduction
1997). In this case, there seems to be no approach available to modellers other than to
estimate a roughness coefficient using numerical models. The value of the roughness
coefficient can be obtained either by using trial and error or automatic optimisation
methods.
The earlier method is a traditional one. It mainly involves visual comparisons between
observed data and simulated ones to estimate the values of parameters. It suffers from
subjectivity and inefficiency. To overcome subjectivity problems and to achieve
efficiency, the later methods may be applied. The problem of identifying the values of
roughness coefficient embedded in the unsteady flow equations is referred to as the
inverse problem or the roughness identification problem. These methods involve a
systematic iterative procedure that seeks to uncover the optimum values of parameters
by minimising a chosen error criterion/objective function.
There is a wide range of objective functions and optimization methods used for the
inverse problem (e.g. Becker and Yeh 1972, 1973; Wiggert et al. 1976; Fread and
Smith 1978; Wormleaton and Karmegam 1984; Atanov et al. 1999; Ramesh et al.
2000; Ding et al. 2004). However, studies on the roughness identification problem in
open channels are still sparse. They have not been sufficient to establish a clear
understanding of the factors affecting the identified parameters in open-channel
problems. At present, only some factors have been considered by Khatibi et al. (1997).
Some other factors related to the modelling problem, such as values of weighting
coefficient, computational grid sizes, boundary condition combinations and errors in
initial conditions have not been considered. Moreover, the applications of this
problem to natural rivers are limited and only the roughness coefficient of the main
channel has been considered Wormleaton and Karmegam (1984). In flood routing in
natural rivers, many channels have compound sections and the values of roughness
coefficients in main channel and floodplains are usually different. Therefore, it is
necessary to extend the method to overbank flow where floodplain roughness
coefficient will obviously have to be considered.
4
Chapter 1 Introduction
1.1.3 Scope and objectives
This study will focus on problems related to the estimation of the roughness coefficient
(Manning's n) for one-dimensional flow in open channels. The study includes two
main parts. The first part is for steady flow and the second part is for unsteady flow.
For steady flow, the main objective is to reinvestigate and extend the method of
estimating Mannings n using two-point velocity measurements. This work is based on
a theory of logarithmic velocity distribution in fully turbulent flow for rough channels.
Thus, it can be only applied to wide channels where velocity distribution following
logarithmic distribution is assumed throughout the depth.
For unsteady flow, this research concentrates on aspects related to the inverse problem
or the roughness identification problem. The governing equations of one-dimensional
unsteady flow can be derived from the principles of conservation of mass and
momentum. The resulting equations are hyperbolic, non-linear differential equations
known as the Saint-Venant equations. The roughness coefficient, as embedded in the
momentum equation, cannot be measured directly and therefore needs to be estimated
by optimisation methods. There are two main objectives here. The first one is to
investigate modelling factors affecting the quality of the identified parameter. The
second one is to extend the roughness identification problem to compound channels
and to channels where the roughness varies with water stage.
1.2 Research methodology
In order to obtain the research objectives, the following research methodology has
been carried out:
For steady flow
Review the limitations of the current method;
Derive the equations to estimate Manning's n using two-point velocity data;
Set up the experiment to verify the sensitivity analysis;
Apply the proposed method to natural rivers; and
5
Chapter 1 Introduction
Assess the performance of the proposed method by comparing the computed
Manning's n with the measured n and with the values computed by the other
empirical formulae.
For unsteady flow
Briefly review the numerical methods for solving one-dimensional unsteady
flow in open channels and the optimisation methods applied to the inverse
problem in this area;
Develop an appropriate roughness identification model applicable to compound
channels;
Test the performance of the model using synthetic data for both a single
channel and a compound channel;
Investigate the effect of modelling factors affecting the quality of the identified
roughness coefficient by using synthetic data;
Apply the roughness identification model to some natural rivers, especially for
the rivers with inundated floodplains.
1.3 Outline of the thesis
This thesis consists of seven chapters. Following the introduction, Chapter 2 briefly
discusses flow resistance in open channels and factors affecting the roughness
coefficient in open channels. Then current methods of estimating the roughness
coefficient Mannings n in both steady and unsteady flows are reviewed. Their
advantages and limitations are also discussed. Several aspects of estimating the
roughness coefficients for compound channels including the main channel and
floodplain are raised.
In Chapter 3, the method of using two-point velocity data to estimate Manning's n in
open channels for steady flow is reinvestigated and extended. Firstly, the current
method is reviewed. Then proposed formulae for estimating the roughness coefficient
of the cross-section using velocity gauging data will be derived. The sensitivity
analysis in terms of the effect measurement errors in depths and velocities to the
relative errors in estimated n is investigated and verified using experimental tests. The
6
Chapter 1 Introduction
proposed method of estimating Manning's n is then applied to 14 rivers in Australia
and New Zealand. The performance of the proposed method is assessed by comparing
the computed n with the measured values of this roughness coefficient. The method is
also compared with some applicable empirical formulae.
Chapter 4 concentrates on developing an appropriate model for identifying roughness
coefficient Mannings n for unsteady flows in open channels. This model consists of
two sub-models, a hydraulic sub-model and an optimisation sub-model. Firstly, the
chapter provides the theory, background and numerical techniques for solution of the
Saint-Venant equations. Secondly, a hydraulic sub-model is developed for a non-
prismatic channel with a compound section, lateral inflow, and offstream storage.
Thirdly, different optimisation techniques used to identify the roughness coefficient in
open channels are briefly reviewed. Then, a direct non-linear optimisation algorithm is
adopted for the optimisation sub-model. Finally, a roughness identification model is
developed by combining the hydraulic sub-model and the optimisation sub-model.
This model will be used to investigate factors affecting the quality of the identified
roughness coefficient and applied to some natural rivers in the following chapters.
Chapter 5 addresses the investigation of different modelling factors affecting the
quality of the identified roughness coefficient. Synthetic data is generated where the
observed data are considered for both cases with and without noise or errors. Firstly,
some numerical experiments are done to test the performance of the model (developed
in Chapter 4). Then by using the synthetic data, the roughness identification model is
applied to investigate the modelling factors that affect the quality of the identified
roughness coefficient. These factors include the weighting coefficient of the numerical
scheme, computational grid size, combinations of boundary conditions and errors in
the initial conditions. Moreover, the problem is extended to compound channels. Some
additional factors affecting the quality of the identified roughness coefficients in
compound channels are considered. Furthermore, the problem is also extended to the
case of roughness variation with the water stage where the roughness coefficient may
be formulated as a function of the water stage. Finally, an alternative conveyance
function is tested for prismatic channels or channels where cross-sections do not
change much along the channel.
7
Chapter 1 Introduction
In Chapter 6, the application of the roughness identification model to natural rivers is
performed. The model is applied to two natural rivers. The first river is the Goulburn
River in Victoria, Australia and the second is the Duong River in the North of
Vietnam. Both of them have extensive floodplains. Depending on the data availability,
the characteristics of the cross-sections, and the flow variation in the main channel and
the floodplain, the performance of the model is investigated. Several aspects related to
the roughness variation with water stage and time are also explored.
Finally, Chapter 7 summarises the major conclusions from this thesis and outlines
recommendations for further research in this area.
8
Chapter 2
Chapter 2
Methods of estimating roughness
coefficients in open channels
2.1 Introduction
This chapter focuses on the different methods of estimating the roughness coefficients
in open channels. Before reviewing these methods, firstly the flow resistance and
roughness coefficients used in open channels are briefly discussed. Then, current
methods of estimating the roughness coefficient (Mannings n for this study) for both
steady and unsteady flows are reviewed. Their strengths and weaknesses are analysed.
For unsteady flow, the necessities of extending the roughness estimation problem to
compound channels and to the channels where the value of the roughness coefficient
varies with flows are then discussed. Finally, key points are summarised.
2.2 Flow resistance in open channels
Real fluid flows in general and water flow in particular are always subject to hydraulic
resistance and energy dissipation. For flow computation, the resistance effect is
involved in flow formulae in terms of resistance/roughness coefficients. The most
commonly used formulae are:
Chezys equation:
RS C V = (2.1)
9
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
Darcy-Weisbachs equation:
RS
f
g
V
8
= (2.2)
and Mannings equation:

2 / 1 3 / 2
1
S R
n
V = (2.3)
y flows it is the friction/resistance slope ( ); g e acceleration
From Equations (2.1), (2.2) and (2.3), the roughness coefficients can be related as:

where V is the cross-sectional average velocity, n, f and C are the Manning, Weisbach
and Chezy roughness coefficients (or resistance coefficients), respectively; R is the
hydraulic radius; S is the slope: for uniform flow it is the bed slope (
0
S ) and for non-
uniform and unstead
f
S is th
due to gravitation.
C R V 8
g g n f
R gS
f
= = =
6 / 1
(2.4)
quation (2.3) can be rearranged Mannings e to give the friction slope:
2 3 / 4 2
2
3 / 4
2
K
Q Q

R A
Q Q n
R
V V n
S
f
= = = (2.5)
where Q is the discharge; A is the wetted cross-section area; K is the conveyance
( n AR K /
3 / 2
= ).
Chezys equation and Mannings equation are used widely for flow computation in
open channels. The Darcy-Weisbach equation is mainly used in the field of pipe flow
problems but has also been recommended to be used in open channel flows (Task
Force Committee of US Army Corps of Engineers 1963). From a theoretical point of
view, these three flow equations are equally appropriate for flow computation. The
primary difficulty in using any of these equations in practice is accurately estimating
an appropriate value of these roughness coefficients.
10
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
Concerning the preference of using Mannings n, Darcy-Weisbachs f or Chezys C
coefficients for open channel resistance, in Equation (2.4) theoretically these
coefficients are interchangeable and equivalent. If the value of one resistance
coefficient is determined, the corresponding values of the other resistance coefficients
can be computed. However, the flow in natural rivers is usually fully rough and
turbulent. For rigid boundaries of open channels, f and C vary with flow depth whereas
n remains nearly unchanged (French 1985 and Yen 2002). It can be argued that in view
of possible changing bed-forms and the composite-rough nature of a river, it is
unlikely n can be a constant for different depth. For example, some field studies on the
roughness coefficients of some natural channels (Butler et al. 1978; Robbins 1976;
Sargent 1979) showed that n probably varies with flow depth. Nevertheless, the
popularity of the Manning formula in the open channel area may be an indication of
less variability of n as compared to f. This makes Mannings n become a valuable tool
to assess the effects of open channels flow resistance. It is also the reason why
Mannings equation is the most reliable and widely used in practice for flow
computation in open channels (Henderson 1966 and French 1985). Therefore, with the
consideration of estimating roughness coefficients in open channels, Mannings n is
chosen as the roughness parameter for the rest of this thesis. The symbol n and the
roughness coefficient are used interchangeable to indicate this roughness coefficient.
One of the difficulties involved in open channel flow computation is accurately
estimating an appropriate value of Manning's n. This is because its value depends on
many factors. The major factor is channel surface roughness, which is determined by
size, shape and distribution of the grains of material that line the bed and sides of the
channel (the wetted perimeter). Five other main factors are channel surface
irregularity, channel shape variation, obstructions, type and density of vegetation, and
degree of meandering (Cowan, 1956). Some additional factors that affect energy loss
in a channel are depth of flow, seasonal changes in vegetation, amount of suspended
material and bed load, and changes in channel configuration due to deposition and
scouring. All these factors were described in detail by Chow (1959). Rouse (1965)
critically reviewed hydraulic resistance in open channels on the basis of fluid
mechanics. He classified flow resistance into four main components: (i) surface
resistance, (ii) form resistance, (iii) wave resistance from free surface distortion, and
11
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
(iv) resistance arising from flow unsteadiness. The effect of flow unsteadiness on
Manning's n was also discussed by Sturm (2001) and Yen (2002).
Therefore, to select a value of Manning's n, we actually combine all of the effects of
many factors affecting the resistance to flow into a single coefficient. Furthermore, this
is really a matter of intangibles. As a result, there are a number of methods developed
to estimate Manning's n. In the next sections, the current methods of estimating the
value of this roughness coefficient are briefly described. Their strengths and
weaknesses are also discussed.
2.3 Methods of estimating Manning's n for steady flow in
open channels
The estimation of Manning's n for a given flow in a given channel reach involves a
substantial measure of engineering judgment. In practice, depending on the availability
of data, there are different methods of estimating this roughness coefficient in open
channels. These are discussed in the following subsections.
2.3.1 Direct measurement method
If the other parameters in Mannings formula can be evaluated by measurement or
observation, then the value of Manning's n can be calculated. This was the basis for
derivation of the roughness parameter (e.g. Barnes 1967; French 1985; Hick and
Mason 1991). According to Arnold et al. (1988) the ideal characteristics of a reach for
applying this method are:
it is straight;
its length is at least five times its width;
it has uniform cross-sections or is converging;
its flow is contained without overflow; and
it has straight entrance and exit conditions, with no backwater effects.

The formula to calculate Manning's n by the direct method is:
12
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
( ) ( ) ( )
( ) ( )

2 1
2
3 / 2
1
3 / 2
, 1
2
, 1 ), 1 ( 1 1
1

+
=

=

N
i
i i
i i
N
i
i i i i N N
AR AR
L
h k h h Z Z
Q
n (2.6)
where Q is the water discharge, N is the number of cross-sections (with the N
th
cross-
section being furthest upstream), Z

is the water surface e vation, le is the
velocity head, V is average velocity,
g V h 2 /
2
=
i i
h
, 1
is the change in velocity head between
sections the 1 i and i, k head loss coefficient of a contracting or expanding reach (k is
assumed to be equal to zero for uniform or contracting reaches and 0.5 for expanding
reaches (Hicks and Mason 1991)), A is the wetted channel cross-sectional area, R is the
hydraulic radius, and
i i
L
, 1
is the distance between sections 1 i and i.
The discharge can be derived from velocity measurements or rating curves. The cross-
sectional area and wetted perimeter (hence the hydraulic radius) can be derived from
stage observation and cross-sectional surveys (usually more than two cross-sections in
each case). In practice, a survey covering a length of 1 km or more along the channel
may be required to provide a reasonable estimate of the slope.
Manning's n determined from this method is usually considered as an accurate value. It
is considered as a measured Manning's n and used as a basis to assess the performance
of the other methods (e.g. Jarrett 1984; Sargent 1979; Coon 1998). However, this
method is time consuming and costly because it requires surveying not only the water
stages, discharges, and some cross-sections along the reach but also the slope.
stressed that care should be taken to
relate to a steady flow situation, that is,
Moreover, for applying this method, it should be
ensure as far as possible that the observations
to a situation in which conditions do not change significantly with time. Roughness
parameter values derived from data relating to unsteady flows are likely to be
erroneous.
13
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
2.3.2 Using the tables or guidelines
When both flow and cross-section data and surface bed material data are not available,
usually the tables or guidelines to estimate Manning's n are used. Most text books on
hydraulics and many other publications dealing with free-surface flow present tables
showing the values of roughness coefficients for channels of various descriptions.
Typical examples are the tables presented by Chow (1959), Henderson (1966) and
French (1985). For each kind of channel the minimum, normal and maximum values
of Manning's n are presented.
Urquahart (1975) provided a detailed guideline of how to estimate the value of
In general, this method is the most simple, rough and quick selection of the n value to
. However, such tables might be assumed only to offer the
asis for a preliminary estimation of the value of Manning's n in a given channel reach.
Moreover, in most cases, the extent to which the reliability of the tabulated values has
bee a
comparison between the computed Manning' with
the values estimate hows figures fo s of sim n. From
the table re signific ifferences values,
especiall

Manning's n for a given channel. This method involves the selection of a basic n value
for uniform, straight and regular channel in a native material and then modifing this
value by adding correction coefficients. These coefficients are added by a critical
consideration of the effects of some other factors such as vegetation, channel
irregularity, obstructions and channel alignment. In this process, it is critical that each
factor be considered and evaluated independently. It is suggested that the turbulence of
the flow can be used as a measure or indicator of degree of retardation; i.e. factors
which induce a greater turbulence should also result in an increase in Manning's n.
be used in a given problem
b
n established is not explicitly stated. For example, Sargent (1979) made
s n of 5 rivers in United Kingdom
d from C r the channel ilar descriptio
it can be seen that there a ant d between these
y for the first two rivers.

14
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
Table 2.1 Comparison of calculated and estimated values of Mannings n (after
Sargent (1979))
Station Estimated n Calculated n
River Almond at Craigiehall 0.04 0.0183-0.0255
Water of Leith at Murrayfield 0.03-0.04 0.0193-0.0257
River Esk at Musselburgh 0.03 0.0265-0.0438
River Tyne at East Linton 0.035-0.050 0.0382-0.0490
River Tyne at Spilmersford 0.03-0.04 0.0254-0.0291

2.3.3 Photographic comparisons
In the absence of a satisfactory quantitative procedure to determine the Manning's n for
a channel, this visual assistance method can be applied. The principle of this method is
that based on the photographs and geometric and hydraulic descriptions of a wide
range of channel reaches of known n values, the n value of a certain reach can be
adopted for a different reach having recognisably similar hydraulic characteristics. At
present, there are several documents available with photographs of a wide range of
with n values ranging from 0.024 to 0.075. More than 150 colour
photos from 78 streams in New Zealand with a wide range of mean flow (from 0.1 to
ith floodplains, it is necessary to refer to Arcement and
Schneider (1989).
typical channels, accompanied by brief descriptions of the channel conditions, the
hydraulic parameters, and the corresponding n values. For example, Chow (1959)
presented 24 black and white photographs of channel reaches with n value ranging
from 0.012 to 0.150. Barnes (1967) presented approximately 100 colour photos from
50 natural channels
353 m3/s), slope (from 0.00001 to 0.042), and bed material (from silt to boulders and
bedrock) were also presented by Hicks and Mason (1991). A similar visual aid was
also employed by the U.S. Geological Survey (Arcement and Schneider 1989).
In these documents, the values of Manning's n of the channels were determined using
the direct measurement method, as discussed in Section 2.3.1. Measurements were
generally limited to flow below bankfull level. To estimate the roughness coefficients
for compound channels w
15
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
Although this method is very simple and can quickly be used to obtain value of
In principle, estimation of the value of Manning's n by this method involves the
ethod was applied for a number of streams in
south-east Australia (e.g. Goyen 1982; Keller 1982; Mitchell 1982; Morris 1982;
measurements of velocity distribution
calculate the roughness coefficient from Mannings equation. However, with
the logarithmic law of velocit
Manning's n for a channel, this method is subjective and in some cases, the accuracy is
limited. Coon (1998) showed that many channels appearing similar upon visual
inspection actually have unrecognisable differences that could substantially affect the
value of the roughness coefficient. For example, water surface slope is one of the
factors that can have an influence on the n values (e.g Bray 1979; Jarrett 1984;Riggs
1976; Coon 1998). However, it is impossible to accurately estimate the water surface
slope from a photograph.
2.3.4 Fitting calculated water surface profile to observed levels
calculation of water surface profiles using a range of assumed values of the roughness
coefficient. The water surface profiles are calculated at a given discharge for steady
gradually-varied flow. For this method, it is necessary that a reliable estimate of the
discharge should be available. The value of n yielding the water profile which best fits
the observed water levels then might be expected to represent a reasonable estimate of
the value of n for the channel. This m
Sheehan 1982).
For applying this method, the quantity of data required, and the effort and cost
involved in its acquisition may clearly be of significant magnitude. Moreover, because
the water profiles are calculated for steady cases, care should be taken when applying
to the cases of data which may relate to unsteady conditions (i.e. the use of the peak
discharge and the flood marks along a reach). In order to overcome this limitation,
unsteady flow models may be applied.
2.3.5 Field
The practical measurement of discharge is usually made by taking velocity
measurements in several verticals at two-tenths and eight-tenths the depths. Although
discharges are known, usually the slope is not known and hence one cannot directly
reference to y distribution (Keulegan 1938), it can be
16
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
seen that the velocity distribution depends on the roughness height, which may be
related to Mannings n. If the distribution is known then the value of n can be
determined.
This concept was used by Boyer (1954) and Langbein (1940) for estimating the value
of n. The procedure for estimating n by using velocity measurements is presented by
Chow (1959) and French (1985). The formula to calculate Manning's n was derived as:

) 95 . 0 ( 57 . 5
) 1 (
6 / 1
+

=
x
D x
n (2.7)
where
8 . 0 2 . 0
/ u u x = ,
2 . 0
u and
8 . 0
u the velocities at two-tenths and eight-tenths the
depth and D is the depth of flow calculated in metres.
This equation gives the value for n for a wide channel with logarithmic velocity
distribution. When this equati n is ap ied to actual streams, the value of D may be
taken as the mean depth. French and McCutcheon (1977) found that this method
provided reasonable estimates of n in a study of the Cumberland River, Tennessee.
However, French (1985) indicated that although the above equation provided a
reasonable estimate of n, additional verification of the method was still required. Chow
(1959) also remarked that if this method can be shown as sa
o pl
tisfactory for practical
determining the roughness coefficient in
ade.
more
general formulae to calculate Manning's n and verify the method by applying it to
anning's n is known.
applications, it will provide an easy way of
streams where the velocity observations have been m
In practice, the velocities are measured at many vertical lines at a certain cross-
sections, but which values of
2 . 0
u and
8 . 0
u should be used in Equation (2.7) to
calculate n is not clear. Also, the velocity distribution at a vertical line reflects only the
roughness characteristic at the vicinity of this vertical. It indicates that the formula for
calculating n has not been well established. Therefore, it is necessary to derive
some natural rivers where the measured M
17
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
2.3.6 Empirical formulae
There are disagreements between researchers regarding the conditions under which
S
value of coefficients a and b (French 1985). The most common material diameters
used in these empirical fo re d
50
, d
75
and d
90
. However, there is no clear
explana Hey et al. 1982). Simons and Senturk (1977)
found t alue of n are different from the ones computed by the Meyer-
Peter and Muller formula (1948) and they are not affected very much by variation of
d
90
. Also fr vers, Bray (1979) concluded that the
correla n is very low and that the selection of different
grain d not affect the result. In a plot of Mannings n versus d
5
hese
rivers, Bray (1979) found that there was considerable scatter of data points and that the
Besides the methods described above, there are a number of empirical formulae
developed for estimating Manning's n. These formulae are usually based on the results
obtained from experimental flumes or natural channels. They can be classified into
three main groups: the group with Stricklers form (Strickler 1923), the group with
Limerinos form (Limerinos 1970), and the group with other empirical equations
obtained from equating Mannings equation with other empirical flow formulae.
Empirical formulae with Stricklers form:
For this group, the value of n is associated with the surface roughness of a channel
boundary. These formulae are in the following form:

b
x
ad n = (2-8)
where
x
d is the bed material diameter such that x percent of material by weight is
smaller and a and b are constants. The earliest formula of this type was proposed by
Strickler in 1923 and is mentioned in many textbooks (e.g. Chow 1959; French 1985;
Henderson 1966). Most researchers have accepted the constant b to be equal to 1/6.
From the statistical analysis of 67 gravel bed rivers, Bray (1979) showed that the
variability of this coefficient is very low. Table 2.2 shows some formulae of different
authors in Stricklers form for estimating Manning's n.
tricklers experiment was conducted, the definition of d
x
and its unit, as well as the
rmulae we
tion or reasons for their selection (
hat the measured v
om a study of 67 gravel bed ri
tion coefficient between d
x
and
iameters does
0
for t
18
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
Stricklers formula gave a lower estimate of n. However, Hey et al.
that d
90
is a good indication of bed roughness.
Table pirical formulae in Stricklers form (all is com uted in metres)
Author(s) Equation
(1982) indicated
2.2 Some em
x
d p
Strickler (1923) (cited in Yen 1991)
6 1
50
0474 . 0 d n =

Keulegan (1938)
6 1
50
0395 . 0 d n =

Henderson (1966)
6 1
75
038 . 0 d n =

Meyer-Peter and Muller (1948)
6 1
90
0385 . 0 d n =

6 1
65
0416 . 0 d n =

Irmay (1949)
6 1
90
0249 . 0 d n =


6 1
50
0594 . 0 d n =

Bray (1979)
6 1
65
0569 . 0 d n =


6 1
90
0523 . 0 d n =

Raudkivi (1976)
16 . 0
65
041 . 0 d n =

6 1
50
047 . 0 d n =

Subramania (1982) and Garde and Raju (1978)
6 1
Chen (1991a)
50
0397 . 0

d n =

Empirical formulae with Limerinos form:
This type of equations is derived from the existing semi-logarithmic formulations of
grave n ed in
the following form:

l bed rivers, where the Darcy-Weisbach friction coefficient f ca be express

+ =
x
d
R
C C
f
log
1
2 1
(2.9)
and from that Manning's n can be obtained by:
19
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels

+
=
x
d
R
C C
R
n
log
113 . 0
2 1
6 / 1
(2.10)
where R is the hydraulic radius, is the bed material diameter such that x percent of
material by weight is sm and are constants.
Table 2.3 Some empirical formulae in Limerinos form (
x
d
1
C
2
C aller and
R and are in m)
x
d
Author(s) Equation
Limerinos (1970)

+
=
84
6 1
log 03 . 2 16 . 1
113 . 0
d
R
R
n

50
d

+ log 36 . 2 248 . 0
R
=
6 1
113 . 0 R
n

Bray (1979)

+
=
65
6 1
log 28 . 2 608 . 0
113 . 0
d
R
R
n

+
=
6 1
log 16 . 2 26 . 1
113 . 0
d
R
R
n

90
Griffiths (1981)

+
=
50
6 1
log 98 . 1 76 . 0
113 . 0
d
R
R
n

Phillips and Ingersoll (1998)

+
=
50
log 23 . 2 46 . 1
113 . 0
d
R
R
n

6 1

Similar to Stricklers ifferent bed material
diameters in their formulae of Limerinos form equations. The most common
diameters employed in their formulae were d
50
, d
84
and 2.3 shows some
empiri n the form of Limerinos equation. By comparing with formulae in
the Strickler form, Bray (1979) indicated that the equations in Limerinos form
perform better because for their determination of Manning's n is stage dependent.
form equations, different authors used d
d
90
. Table
cal formulae i
20
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
Using the power formulations, Manning's n is also estimated in a similar way (e.g.
Bray 1979; Charlton et al. 1988; Chen 1991b; Griffiths 1981). Also, there are some
other g ulae proposed by other authors (e.g. Afzalimehr
and An . 1988), where the friction coefficient is not only the
function of the relative roughness but also of the Froude number, a sediment mobility
parame l form factor that can improve the prediction of the
friction
Other empirical equations
In these equations, n was not originally the subject of the empirical equations. These
le 2.4 Some other empirical formulae
ravel bed semi-logarithmic form
ctil 1998; Colosimo et al
ter and a cross-sectiona
coefficient.
formulae of n are usually obtained by relating Manning equation with other empirical
flow formulae (Lang et al. 2004). Table 2.4 shows several formulae from these types.
Tab
Author Equation*
Lacey (1946)
6 1
0928 0 S . n =

Riggs (1976)
w
S log . .
w
.
S R A
.
n
056 0 45 0 3 2 33 0
55 1
1
+
=

Froehlich (1978) (in Coon, 1998)
30 . 0
44 . 0
50
14 . 0
289 . 0

=

B
R
d
R
R n

Bray (1982) Equation 6.34
21 . 0 067 . 0
125 . 0
w
S R n =
Bray (1982) Equation 6.16
177 . 0
104 . 0
w
S n =

Jarrett (1984)
16 0 38 0
32 0
. .
f
R S . n

=

Sauer (1990) (in Coon 1998)
08 0
18 0
3048 0
1 0. n = 1
.
.
w
.
R
S


Dingman and Sharma (1997)
w
S log . .
w
. .
S R A
.
0543 0 5 0 267 0 173 0
564 1
1
+
n =

* where n is the roughness coefficient, d
x
is the intermediate particle diameter for which x% of
the sampled intermediate particle diameters are smaller (in m), A is the wetted channel cross-
sectional area (in m
2
), R is the hydraulic radius (in m),
w
S is the water surface slope; and B is
the top width of the stream.
21
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
In brief, using empirical equations to estimate the roughness coefficient is less data
dependent compared with the direct method and is an easy way to obtain the value of
n. However, due to the fact that all the empirical equations were derived from the
particular geographic and hydraulic conditions of their respective research sites, there
gree of uncertainty in the application of em
Manning's n for other sites. Therefore, for a certain site, its conditions should be
her methods
and together with calibration of n to discharge data.
is a high de pirical formulae to estimate
similar to those from which the equations were derived. An assessment of different
empirical equations by Coon (1998) showed that no one equation could provide an
accurate estimation of n. Lang et al. (2004) also indicated that estimating n by using
empirical formulae is fraught with danger. They concluded that these formulae can
only be used as one tool to estimate Manning's n in conjunction with ot
2.3.7 Lump roughness coefficient model
The methods of estimating Manning's n by tables, photographic comparisons or
empirical formulae described above may or may not include energy losses that result
from the factors other than particle size and distribution of bed material. If Manning's n
does not represent all major factors affecting its value, it should be treated as a lumped
model which accounts for cumulative effects of these factors. The following formula
proposed by Cowan (1956) shows how some of primary factors contribute to the
lumped value of n:

5 4 3 2 1 0
) ( m n n n n n n + + + + = (2.11)
where
0
n is a basic value for straight, uniform and smooth channels in the natural
material,
1
n is the correction value for surface irregularity on the channel sides and
bottom,
2
n is the correction value for variation in the shape and the size of cross-
section,
3
n is the correction value for obstructions,
4
n is the correction value for
vegetation, and
5
m is the correction value for meandering of the channel.
Cowans treatment of n is based on its nature. Some base values and limits for the
correction values above have been suggested by Chow (1959), Jarrett (1985) and
Arcement and Schneider (1989). Such a treatment is still highly subjective because
22
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
objective identification of the individual factors is not possible. Caution must be
exercised to ensure that the effect of each factor is not duplicated. Moreover, there is
no physical basis for the algebraic additions of quantities called . In addition, as
small s with top widths mostly less than 18 m. Therefore,
g's n for a
2.4 Estimating the roughness coefficients for unsteady flow
ation was developed for uniform flow. Although its
application to steady non-uniform flow as well as unsteady flow has not been
theoretically justified (Yen 1973), the traditional empirical Manning equation is
i
indicated by Coon (1998), the correction values were developed from an analysis of
- to medium-size channel
n
using these correction factors for large channels is questionable.
In brief, this review of the current methods of estimating Manning's n for steady flow
in open channels indicates that the direct method is the most accurate method.
However, this is expensive and time-consuming because of its data requirement. In the
absence of a satisfactory quantitative procedure to determine the Mannin
channel, empirical methods can be applied. Although these methods are very simple
and can quickly obtain the value of the roughness coefficient, they are highly
subjective and susceptible to gross misjudgement. Using velocity measurements to
estimate n may provide a more objective method. However, this method is not well
established and needs further examination in the context of natural rivers.
For unsteady flow, although the parameters estimated by the above methods can be a
satisfactory representation of steady behaviour, the values may not be satisfied in
unsteady flow conditions. The following section will review the methods of estimating
Manning's n for this case.
in open channels
The unsteady one-dimensional flow equations in open-channels can be derived from
the principles of conservation of mass and momentum. The resulting equations are
hyperbolic, non-linear differential equations known as the Saint-Venant equations. The
friction slope evaluated from the roughness coefficient as embedded in the momentum
equation cannot be measured directly and therefore needs to be estimated.
It is argued that the Mannings equ
23
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
assumed to be suitable and practical for unsteady flow (Lai 1986; Zoppou 1984).
used
in practice. In this method, the roughness coefficients are estimated by relying on
ficient in the next trials for improving the performance. Fread
(1998) described the procedure and suggested the way to adjust the trial value of
This trial and error procedure includes
downstream boundary was suggested); (ii) estimate Manning's n; (iii) run the model to
However, as mentioned in Section 2.1, for unsteady flow the roughness coefficient in
this case includes additional effects of other factors that contribute to energy losses due
to unsteadiness (Rouse 1965; Yen 2002). Moreover, Manning's n may vary with flow
depth and the value obtained by the methods presented in Section 2.3 can, therefore, be
an inappropriate presentation of unsteady flow cases where the flow often occurs at
greater depths (Khatibi et al. 1997).
Therefore, there seems to be no appropriate approach available to modellers other than
to estimate the global resistance coefficient which encompasses all these effects by
using numerical models. Generally, calibration is a process of a systematic change in a
set of parameters so that the observed prototype behaviour is reasonably reproduced.
This can be achieved by using either trial and error methods to estimate the roughness
coefficient or optimisation techniques to identify it. Both of these are discussed in the
following subsections. The later approach is generally referred to as the parameter
identification problem (Khatibi et al. 1997), and for this study it is referred to as the
roughness identification problem.
2.4.1 Trial and error method
The trial and error method is a more traditional and common calibration method
visual comparisons between observed data and simulated results in terms of stages
and/or discharges through a trial and error procedure. Posada and Bras (1982) argued
that this visual fitting procedure is an approximate least square fit, because it attempts
to minimise the deviations between the predicted and observed data.
The main problem with this method is deciding on the amount of change in the value
of the roughness coef
roughness coefficient in the FLDWAV model.
six steps: (i) use the observed flow hydrograph as the upstream boundary condition
and select an appropriate downstream boundary (an observed stage hydrograph at the
24
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
obtain the simulated stage hydrograph at the upstream boundary and elsewhere; (iv)
compare the simulated stages with observed ones; (v) adjust n: if the computed stages
are higher than the observed, increase the estimated n value and vice versa; (vi) repeat
ith (1978) used a
modified Newton-Raphson search technique for estimating the roughness parameter as
step (iii) and (iv) until the simulated and observed stages are approximately the same.
Several useful suggestions for calibration were also discussed by Cunge (1975).
Although this method utilizes unsteady flow models to estimate the roughness
coefficient, it suffers from two main disadvantages: (a) it is subjective because such an
approach involves intuition, experience and human judgement, hence different results
may be obtained by different users; (b) it is inefficient, tedious and time-consuming
especially when more than one parameter need to be identified. To overcome these
problems, automatic optimisation methods may be applied which are discussed in the
following subsection.
2.4.2 Automatic calibration methods
These methods involve a systematic iterative procedure that seeks to uncover the
optimum values of the roughness coefficients by minimizing a chosen error
criterion/objective function. These methods are also known as the inverse problem or
the roughness identification problem.
There is a wide range of objective functions and optimisation methods applied to the
roughness identification problem. Becker and Yeh (1972, 1973) used the influence
coefficient approach by minimising the sum of squares of differences between
observed data and numerically simulated values to estimate the parameters. Wiggert et
al. (1976) employed a conjugate gradient method and formulated the objective
function by using the sum of the absolute difference between observed and simulated
stages and discharges at intermediate sections. Fread and Sm
a function of stage and discharge. They minimised the sum of the absolute value of the
difference between observed and computed stages and discharges. Their method
required breaking down the river into a number of single channel reaches before
calibrating each reach separately. Wormleaton and Karmegam (1984) formulated the
objective function in terms of relative errors using both depth and discharge values and
25
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
identified the parameters with the influence coefficient algorithm and also a nonlinear
least square technique. A pattern search algorithm and minimization of sum square
errors and sum of absolute errors were used by Zoppou (1984) and applied to several
rivers. Khatibi et al. (1997) and Jun et al. (2005) used a nonlinear least square
technique by a modified Gauss-Newton method. Atanov et al. (1999) introduced a
ion methods can be categorised into
two main groups: gradient methods and direct search methods. The gradient methods
e quantity and quality of
observation data are usually insufficient. Thirdly, the modelling factors, even in one
variational approach of Lagrangian multipliers using a least square errors criterion to
estimate the roughness coefficient. However, the algorithm can be applied only to
simple prismatic channels. The Sequential Quadratic Programming Algorithm was
used by Ramesh et al. (2000) to minimise the objective function based on the least
square error criterion. Recently, a Limited-memory quasi-Newton method was used by
Ding et al. (2004) to identify Mannings n in shallow water flows.
There are some comparisons on the methods of optimisation used for the inverse
problem in open channels. In general, the optimisat
(e.g. Wiggert et al. 1976, Fread and Smith 1978, Khatibi et al. 1997, Ramesh et al.
2000; Ding et al. 2004) are more efficient than direct search methods (e.g. Wu 1976;
Zoppou 1984). However, the direct search methods are simple because they do not
need to compute the gradient of objective function. Also, the main advantage of these
methods is that they can overcome the scaling problem that may occur when the
derivatives, with respect to a particular parameter, are considerably smaller than the
others (Zoppou, 1984).
The progress of the inverse problem is blocked by several inherent difficulties (Sun
1994) and the roughness identification problem is not an exception. Firstly, the inverse
problem is often improperly posed, i.e. its solution may be non-unique and unstable
with respect to observation errors. Secondly, in practice, th
particular numerical scheme and algorithm to solve the resulting algebraic equations
often affect the quality of identified parameters. If these difficulties are not considered,
a satisfactory solution of the inverse problem can never be obtained by changing only
performance criteria and optimisation algorithms.
26
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
Khatibi et al.(1997) used synthetic data to investigate several factors affecting the
quality of the identified friction parameter including: the selection of an objective
function, the data errors, the choice of observation station, the effect of the channel and
actors
y Jun et al. (2005) for the
ess coefficient in the in-bank channel.
flow for compound channels. For one-dimensional models, there are two
main approaches. The basic difference of these approaches is the treatment of the main
some aspects of downstream boundary conditions on the identified parameters.
Nevertheless, that study has not been sufficient to establish a clear understanding of
the factors affecting the identified parameters in open-channel simulation problems.
Moreover, the roughness identification problem is based on numerical models. Even
in a certain numerical model, there are several factors that could affect the quality of
identified parameter such as the value of weighting coefficient in the finite difference
scheme, computational grid size, combinations of boundary conditions and errors in
initial conditions. However, at present, no study has focused on the modelling f
affecting the quality of the identified roughness coefficients.
For unsteady flow, the application of the roughness identification problem to natural
rivers is still sparse. Optimisation techniques have been used to identify roughness
coefficient by Wiggert et al. (1976) for the James River, by Davidson et al. (1978) for
the Upper Delaware River Estuary, by Fread and Smith (1978) for the Mississippi
River and Mississippi-Ohio-Cumberland-Tennessee River system, by Wormleaton and
Karmegam (1984) for the River Wye, by Wasantha Lal (1995) for the Upper Niagara
River, by Ding et al. (2004) for the East Fork River and b
Han River.
These studies have just considered the roughn
However, in flood routing in natural rivers, many channels have compound sections
and the values of the roughness coefficients in the main channel and floodplains are
usually different. As indicated by Wormleaton and Karmegam (1984) this problem
needs to extend to over-bank flow where the roughness coefficient of the floodplain
will obviously have to be considered. This indicates the necessity of extending the
roughness identification problem for compound channels where not only the roughness
coefficient of the main channel but also of the floodplain needs to be identified.
To deal with this problem, it is necessary to choose an appropriate method to compute
unsteady
27
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
channel-floodplain system. The first approach is treating the channel as a single
channel and employing an equivalent roughness coefficient formula for the whole
section (single channel approach). The second one is treating the channel as a
compound channel by subdividing the channels into single channels each with their
own conveyance properties (compound channel approach).
g section will describe differen
roughness coefficient for compound channels.
d channels
sistance. Unfortunately, it
is not easy to develop an equivalent roughness coefficient formula to replace the actual
ented by Yen
(2002). Some of the most common formulae and their assumptions are described
er of
subsections and assumed that each part of the area has the same mean velocity equal to
locity of the whole cross-section. Th
) was obtained by the following equation:

The followin t methods used to estimate the equivalent
2.5 Equivalent roughness coefficient for compoun
For this approach, a compound channel may be replaced with a single channel of
equivalent roughness coefficient offering the same friction re
channel with an equivalent channel. This has consequently led to the development of a
number of methods based on different assumptions (Chow 1959). A detailed
discussion of the equivalent roughness coefficient formulae was pres
below.
Horton (1933) was the first to propose a formula for the equivalent roughness
coefficient in a channel cross-section. He divided the whole section into N numb
the mean ve e equivalent roughness coefficient
(
e
n
3 / 2
2 / 3
n P
N

3 / 2
P
e
where
1
n
i
i i

=
=
(2.12)
P the wetted perimeter of the whole cross-section,
i
P and are the wetted
perimeter and Manning's n respectively for any arbitrary subsection i.
This approach has little regard to natural channels with large floodplains where the
he main channel.
i
n
flow is much smaller than that of t
28
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
Einstein and Banks (1950) and Pavlovskii (1931) assumed that the total force resisting
the flow is equal to the sum of the forces resisting the flow developed in each
subsection. By this assumption, the equivalent roughness coefficient is:

2 / 1
2 / 1
1
2
n P
n
N
i
i i
e

=

=
(2.13)
low is equal to the sum of the
discharges in each subdivided area. Thus, the equivalent roughness coefficient is:
P
Lotter (1933) assumed that the total discharge of the f

=
=
N
i i
i i
e
n
R P
PR
n
1
3 / 5
3 / 5
(2.14)
where R and
i
R are hydraulic radii of the whole section and subsection i.
By assuming a logarithmic velocity distribution with depth, and also the friction slope
of each subsection is the same as that the whole cross-section, Krishnamurthy and
Christensen (1972) derived the following formula:

=
=
N
i i
N
i
i i i
e
D P
n D P
n
1
2 / 3
1
2 / 3
) ln(
ln (2.15)
where D
i
is depth of i subsection.
Cox (1973) proposed the following formula by simply weighting the roughness
coefficient according to the area of subdivision:

A
A n
n
i
i i
e

N
=
=
1
(2.16)
where A the wetted area of the whole cross-section,
i
A and
i
n are the wetted area and
Manning's n respectively for any arbitrary subsection i.
29
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
Several studies have investigated the performance of different formulae in estimating
the equivalent roughness coefficient. Motayed and Krishnamuthy (1978) evaluated the
formulae (2.12) to (2.15) by comparing Mannings n computed from the velocity data
of 36 natural streams in Maryland, Georgia, Pennsylvania and Oregon. Their results
ion 2.14) could predict the

coefficient value at bankfull level. There are some discrepancies in the values of the
the roughness conditions of bed and banks are equal and
regular, the value of Manning's n may vary little at all stages. In this case, the use of a
indicated that the formula proposed by Lotter (Equat
composite roughness coefficient with the least error. Zoppou (1984) used the
experimental results of James and Brown (1977) to compare the performance of those
formulae. His results also showed that Equation (2.14) produces the best results and
has almost eliminated the discontinuity in discharge-stage curve at bankfull level.
Yen (2002) investigated the variation of
e
n with depth of a simple compound channel
obtained from 17 formulae for calculating the equivalent roughness coefficient. His
results showed that all the formulae produce an abrupt change in the roughness
equivalent roughness coefficient obtained from different formulae. For the formulae
where the radius or area of the subsection are used, the value of
e
n depends not only on
the formulae used but also on the way to divide the cross-section. He stressed that in
order to assess these formulae, it is necessary to verify them with accurate
experimental data.
The above discussion on the different equivalent roughness coefficient formulae
indicates two main features. The first one is that there is a discontinuity of the
roughness curve at bankfull level. The second one is that there are discrepancies in the
values of the equivalent roughness coefficient obtained from these formulae. These
features would be disadvantages of the single channel approach for unsteady flow
computation in compound channels. Accordingly, it is not an appropriate approach for
the roughness identification problem for unsteady flow in compound channels.
2.6 Variation of Manning's n with flow
In some river reaches, if
constant n appears to be satisfactory. However, in other cases a variable resistance
coefficient is preferable or necessary because of the significant changes in roughness
30
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
characteristics on the bed and banks of a reach. In general, the value of Manning's n in
many streams decreases with increasing water stage or discharge. However, the value
of Manning's n may be large at high stages for the channels with rough banks and
dense vegetation or in the case of high flow with the change of bedforms.
A number of field studies have been reported suggesting how the value of a roughness
u
River between Memphis and Fulton, Tennessee
showed that n value decreases with increasing stage, but it starts to increase at near and
coefficient varies. The study of Robbins (1976) on the hydraulic characteristics of the
Lower Mississippi River showed that the values of Manning's n for this river reach is
almost constant when the discharge is greater than 22,000 cubic metres per second, but
it increases considerably with decreasing discharge. Butler et al. (1978) investigated
the roughness coefficient variation with flow for 3 reaches of an urban stream during
low and medium flow. The result indicated that Mannings n values vary widely with
flow. The value of n was of order of 100% greater at low flows than at high flows.
They also suggested that further work is required to extend the data to cover higher
flow studies. Sargent (1979) calculated the values of n using the direct method for 5
rivers in the United Kingdom. His results showed that as discharge increased,
Manning's n was found to decrease rapidly and to approach an asymptotic value as
bankfull conditions were reached. Stewardson and Anderson (2002) proposed a
relationship of Mannings n and discharge based on an analysis of data from Hick and
Mason (1991). Their formula was tested for some Australian streams in Lang et al.
(2004).
The experimental study of Valentine et al. (2001) carried out at a large laboratory scale
show that roughness coefficient values increase with increasing hydraulic radius. This
could be attributed to an increasing bedform effect which was consistent with their
observation of two-dimensional dunes. Also, from the field measurements during the
1998 flood of the Rhine River in the Netherlands, Julien et al. (2002) showed that the
value of Manning's n increases with discharge. The change in bedform roughness
height and friction factors were attrib ted to the increase in the value of Manning's n.
Several studies show that there is an abrupt change in the value or the variation trend
of Manning's n at bankfull level. For example, the curve of n value versus stage given
by Lane (1951) of the Mississippi
31
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
above bankfull level. Richards and Hollis (1980) examined variations in Manning's n
with discharge in concrete and earth reaches of the River Brent in north London. Their
results also showed that values of n fell with increased discharge but rose as flow
approached bankfull. They highlighted the problem of characterising channel
roughness with a single n value. Nalluri and Judy (1985) indicated that Manning's n is
found to be widely affected by type and shape of roughness elements and variation
more than 100% with depth in most cases. Their study showed that below bankfull
level, the variation in Manning's n is found to be insignificant with the average value
of the main channel. However, above bankfull level, n increases with increasing depth.
These studies showed the inaccuracy in adopting one n value for such conditions in the
computation of unsteady flow in open channels.
identification problem that allow for the
roughness coefficient varying with stage or discharge are very sparse. The variation of
At present, studies on the roughness
roughness coefficient with discharge was studied by Fread and Smith (1978) for the
Mississippi River and Mississippi-Ohio-Cumberland-Tennessee River system. In their
model, n variation could be described as a continuous piece-wise linear function. The
total range of possible discharges was divided into a number of strata
i
Q . Each stratum
is associated with a value of n in an ) (Q n piece-wise linear function. More recently,
Jun et al. (2005) developed a model to study the roughness variation of the Han River.
hness
coefficient in a compound channel as a single roughness function. In this case, the
In their study, the roughness variation with flow was considered as a step-wise
function and a power function of discharge.
As mentioned in Section 2.5, in flood routing, many rivers have compound channels.
However, the above studies only focused on the variation of the main channel
roughness coefficient. The roughness characteristics of the main channel and the
floodplain are usually different and there is an abrupt change in the roughness value at
bankfull level. Hence, it is unreasonable to express the variation of roug
variation of the roughness coefficients should be formulated as a discontinuous
function (for the single channel approach) or two separate functions (for the compound
channel approach) of stage or discharge, one for the main channel and the other for the
32
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
floodplain. The later approach is preferable because it can explicitly identify the
roughness coefficient of the floodplain.
2.7 Summary of key reviews
ighted. From
this literature review, the following key points were drawn.
s n obtained from these methods
mic law of
velocity distribution can be used to estimate the value of Manning's n in wide channels
(Chow, 1959). However, this method has not been well-established and needs further
examination for natural rivers. If it can be shown to be satisfactory for practical
application, it will provide a simpler and more objective way to estimate the roughness
coefficient in rivers where velocity data are available. Therefore, in this study, Chapter
3 will focus on further investigation into this approach to estimate Manning's n in
streams.
For unsteady flow, the values of roughness coefficient(s) are identified by using
modelling tools. The traditional method uses trial and error. It suffers from subjectivity
and by being inefficient and time-consuming. To overcome these disadvantages,
automatic optimisation methods may be applied. This approach is also known as the
roughness identification problem which will be focused on in this study.
In this chapter, some aspects on the resistance flow and roughness coefficients in open
channel were discussed. The strength and weakness of different methods of estimating
Manning's roughness coefficient n have been reviewed for both steady and unsteady
flow. The requirements of extending the roughness identification problem to
compound channels and the channel with roughness variation were highl
For steady flow, the direct method can provide the most accurate estimation of
roughness coefficient. However, this method is expensive and time-consuming because
of its data requirement. In the absence of a satisfactory quantitative procedure to
determine the value of Manning's n for a channel, the qualitative or empirical methods
are usually applied. These methods are very simple and can quickly obtain the value of
Manning's n for a channel. They are, however, subjective and the accuracy of the
estimated n is very limited. The values of Manning'
can only be considered as preliminary values for any flow computation purpose.
Using two-point velocity measurements with reference to the logarith
33
Chapter 2 Methods of Estimating Roughness Coefficients in Open Channels
The literature survey on the roughness identification problem in open channels showed
that several factors that could affect the quality of the identified roughness coefficient,
especially modelling factors have not been investigated. Moreover, in flood routing in
pound sections and the values of roughness
main channel and floodplains are usually different. However,
current studies have just considered the roughness coefficient in the in-bank channel.

problems will be dealt with in Chapters 4, 5 and 6.


natural rivers, many channels have com
coefficients in the
This indicates the necessity to extend the method to compound channels. These

34
Chapter 3
Chapter 3
Estimation of Roughness Coefficients Using
Velocity Data
3.1 Introduction
Experimental observations have shown that the logarithmic velocity profile adequately
represents the vertical velocity variations in many open channels (French 1985;
Keulegan 1938). Recently, some surveys in natural rivers using an acoustic Doppler
current profiler also indicate that the velocity profile in the vertical for several channels
tion not only for the region near the bed but also over most of
lude the observed velocity profiles of the Sacramento River
onzalez and Melching 1996)
n if it can be taken as a dominating factor that affects the velocity distribution.
f the distribution is known, the value
(Chow 1959).
n/D
1/6
(D is the mean depth) and the ratio of velocities at two-tenths and
ip between
Mannings n and the velocity distribution.
has a logarithmic distribu
the depth. Examples inc
(Gilbert 1997), the Chicago Sanitary and Ship Canal (G
and the Hurunui River (Nikora and Smart 1997).
For fully rough wide open channels, the logarithmic law of velocity distribution
depends on the roughness height (Keulegan 1938), which may be related to Mannings
Therefore, i of Mannings n can be determined
The study of Boyer (1954) on several streams in the North-western U.S. and the
Mississippi River showed a good agreement between the theoretical relationship curve
between
eight-tenths the depth. This suggests that there is a general relationsh
35
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
Chow (1959) and French (1985) described the method using two-point velocities at
two-tenths and eight-tenths the depth to estimate the value of Manning's n. However,
hat the method should be used
verification. The relative effects of measurement errors and the errors in the estimated
measurements are taken at several verticals and the roughness condition along the
value of the roughness coefficient at this local vertical. The general formula
to obtain Manning's n for a whole cross-section needs to be derived.
In this chapter, general formulae to estimate Manning's n of the cross-section using
ta will be derived. T
estimated n will be investigated and verified using experimental tests. The proposed
estimate Manning's n to some natural rivers in New
Zealand and Australia and compared with applicable empirical formulae. Before
the main parts of this chapter, the t
distribution in turbulent flow and the current method using two-point velocity to

locity distribution in turbulent flow
The velocity distribution of uniform turbulent flow in streams can be derived by using
Prandtls mixing length theory (Schlichting 1960). Based on this theory, the shear
stress at any point in a turbulent flow moving over a solid surface can be expressed as:
they noted t with caution and that is requires more
roughness coefficient have not yet been investigated. Moreover, in practice velocity
cross-section change from point to point. The velocities at each vertical can only
reflect the
velocity gauging da hen the effect of measurement errors to
method will be then applied to
approaching heoretical background of velocity
estimate Manning's n will be discussed.
3.2 Background
3.2.1 Ve
2
2

dz
(3.1)

=
du
l
where is the mass density of the fluid, l is the characteristic length known as the
mixing length, u is velocity at a point, and z is the distance of a point from the solid
surface.
For the region near the solid surface, Prandtl introduced two assumptions:
36
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
a. The mix proportional to z,

z l
ing length is
=
(3.2)
where is known as von Krmns turbulent constant. The value of determined
is equal to the shear stress on the bed

from many experiments is 0.4 (Schlichting 1960).


b. The shear stress
0
of the flow in the
channel.
From these two assumptions, Equation (3.1) can be written as
z
dz
du

=
0
1
(3.3)
Integrating Equation (3.3) gives

0
0
ln
1
z
z
u

= (3.4)
where is the constant of integration.
0
z
is represented as a bed shear velocity
*
u It is also known that
0
the bed shear stress
defined by

=
0
*
u (3.5)
on (3.4) can be written Thus Equati

0
*
ln
z
z u
u

= (3.6)
Equation (3.6) indicates that the velocity distribution in the turbulent region is a
logarithmic function of the distance z. This is commonly known as the Prandtl-von
Krmn universal velocity distribution law. The constant of integration, z
0
, is of the
37
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
same order o viscous sub-layer thickne hether f magnitude as the ss. It is a function of w
the boundary is hydraulically smooth or rough depending on the roughness Reynolds
*
u k
s k
, where k
s
is the ss height,
kinematic viscosity) (Schlichting 1960). Below a roughness Reynolds number of 5, the
h turbulent.
number ( Re equivalent sand roughne is the = /
flow is hydraulically smooth, between 5 and 70, the flow is transitional and above 70,
the flow is fully roug
If the boundary is hydraulically smooth, then z
0
depends only the kinematic viscosity
and shear velocity or

*
0
u
m
z

= (3.7)
where m is a coefficient and equal to approximately 1/9 for a smooth surface
(Keulegan 1938). Substitution of Equation (3.7) into Equation (3.6) yields


=
* *
9
ln
zu u
u (3.8)
When the boundary surface is hydraulically rough, the viscous sub-layer is disrupted
by roughness elements. The viscosity is no longer important, but the height of
roughness elements becomes very influential in determining velocity profile. In this
ase depends only on the roughness height, usually expressed in terms of equivalent
roughness k
s


0
z c
s
mk z =
0
(3.9)
where, in this case, m is a coefficient approximately equal to 1/30 for sand grain
ughness (Keulegan 1938). Substituting Equation (3.9) for z
0
in Equation (3.6) yields

ro
s
k
French (1985) indicated that Equations (3.8) and (3.10) adequately represent the
vertical velocity profile in a w
z u 30
u ln
*
= (3.10)
ide channel of an unstratified flow. Using the Prandtl-
38
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
von Karman universal velocity distribution law s data and Nikuradses data and Bazin
Keulegan (1938) derived equations for mean velocity of turbulent flow in open

channels as

+ =
*
*
ln 5 . 2 25 . 3
RU
U
V
for smooth channels (3.11)
s
k
R
U
V
ln 5 . 2 25 . 6
*
+ =
39

and are cross-sectional mean velocity and shear velocity respectively and
ams, the flow is usually fully rough-turbulent, and the logarithmic
law of velocity distribution depends on the roughness height (Equations (3.10) and
e taken as the dominating factor that aff
roughness height and shear velocity are related to Mannings n. Hence, if this
determined. The next section
reviews the current method and highlights its limitations.
3.2.2 Current method using velocity data to estimate Mannings n
ly made by king velocity
Figure 3.1). These measurements, which are computed to give the mean velocities in
calculate discharge, can be use
French 1985) as described below.

for rough channels (3.12)
*
U where V
R is hydraulic radius.
In natural wide stre
(3.12)) can b ects the velocity distribution. The
distribution is known, the value of Mannings n can be
The simple measurement of stream flow is usual ta
measurements in several verticals at two-tenths and eight-tenths of the depth (see
verticals to d to estimate Mannings n (Chow 1959;



Figure 3.1 Two-point velocity measurement at a cross-section
i
D
u
i
u
8 . 0
i 2 . 0

Chapter 3 Estimation of Roughness Coefficients Using Velocity data
Let
2 . 0
u be the velocity at two-tenths the depth, that is, at a distance 0.8D from the
bottom of a channel, where D is the depth of the flow. Using Equation (3.10) the

velocity may be expressed as

s
k
D u
u
24
ln
*
2 . 0

= (3.13)
Similarly, let
8 . 0
u be the velocity at eight-tenths the depth, then

s
k
D u
u
6
ln
*
8 . 0

= (3.14)
Eliminating from the two equations above gives
*
u

1
ln

792 . 1 178 . 3
=
x k
s
(3.15)
where
8 . 0 2 . 0
/ u u x = .
x D
for g Substituting Equation (3.15) in Equation (3.12) the rou h channels with D R = and
simplifying yields
1
) 95 . 0 ( 78 . 1
*

+
=
x
x
U
V
(3.16)
ula,

R V
/ 3 2
= gRS U =
*
n / S , and friction velocity Combining Mannings form
gives

n
g n
U 13 . 3
*
D R V
6 / 1 6 / 1
= = (3.17)
n m and g is the gravitational accelera
Equating the right-hand sides of Equations (3.16) and (3.17) and solving for n gives
where D is i tion ( / 81 . 9 s m g = ).
2
40
Chapter 3 Estimation of Roughness Coefficients Using Velocity data

) 95 . 0 ( 57 . 5
) 1 (
6 / 1
+

=
x
D x
n (3.18)
This equation gives the value for Manning's n for fully-rough flow in a wide channel
with a logarithmic vertical velocity distribution. It is suggested that when this equation
actual streams, the value of D may be taken as the mean depth (
1959; French 1985). However, which values of and (see Figure 3.1) that
at each vertical line can reflect only the local shear stress or
roughness at that point and their values along the cross-section are changed from point
erefore, it is necessary to derive more general formulae to calcula
roughness coefficient along the perimeter of the cross-section. This is described in the
3.3 Proposed method using velocity data to estimate
Mannings n
rage shear stress.
a ghness k
s
of a cross-section varies from point to point.
Using the two velocities at each vertical strip the local value of the equivalent
can be calculated and then the average value can
At each location substituting the values of and into Equation (3.10) then
is applied to Chow
2 . 0
u
8 . 0
u
should be taken to calculate the values of x in Equation (3.16) are not clear. Moreover,
the velocity distribution
to point. Th te the
following section.
This section will derive formulae to calculate Manning's n using two-point velocity
measurements. The formulae will be derived based on the assumption that the velocity
distribution is the logarithmic distribution. Two different parameters will be used to
derive the formulae for estimating n. The first parameter is the cross-sectional average
equivalent roughness which is extended from the above current method and the second
one is the cross-sectional ave
3.3.1 Using average equivalent roughness
The value of the equiv lent rou
roughness k
si
be obtained.
i 2 . 0 i 8 . 0
eliminating
*
u gives
u u
41
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
1
792 . 1 178 . 3
ln

=
i
i
si
i
x
x
k
D
(3.19)
where
i
is the depth at strip i D
th
and x is the ratio of velocitie at two-tenths and
eight-tenths the depth of the strip (
i i i
u u x
8 . 0 2 . 0
/
s
i
= ).

Rearranging Equation (3.19) gives

1
exp
i
i
x
Thus, the cross-sectional average equivalent roughness can be obtained:

=
792 . 1 178 . 3
i
si
x
k (3.20)

D

=
i
P
i si
s
P k
k (3.21)
where and are equivalent roughness and perimeter length over which is
spectively at vertical measurem t i
th
.
s
si
k
i
P
si
k
applicable re en
Replacing the value of k in Equation (3.12) by the mean value (Equation (3.21)) gives
s
k
R
U
25 . 6
*
=
V
ln 5 . 2 + (3.22)
Furthermore, equating the right-hand sides of Equations (3.22) and (3.17) and solving
for Manning's n yields

+
=
s
k
R
g
R
n
ln 5 . 2 25 . 6
6 / 1
(3.23)
This is the equation to calculate Manning's n for a cross-section using average
equivalent roughness.
42
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
3.3.2 Using average shear stress
s found to have a wide
range (varying from 3.23 to 16.92 (Keulegan, 1938)) and the value of 6.25 is its mean
Manning's n. In this approach, this roughness coefficient can be directly calculated by
strip i
th
substituting the values of and of the depth
Equation (3.10) and solving for yields
Formula (3.23) is obtained using Keulegans equation for rough channels (Equation
(3.12)). However, in thi equation the additive constant actually
value (Chow 1959). There is another approach that can be used to estimate the value of
using the average shear stress and the discharge through the cross-section without
using Keulegans equation.
For vertical
. 0 i
into
i
u
2 . 0 i
u
8
D
i
u
*
) ( 289 . 0
) 4 ln(
8 . 0 2 . 0 i i
i
u u
u u
u =
8 . 0 2 . 0 * i i

= (3.24)
Having shear velocity
i
u
*
at the bottom of that vertical, we assume that all such
measurements across the channel can be used in a similar manner. From the definition
of the Darcy friction factor f (Chanson 1999)

2
2
* 0
8 8 u
f =

=
(3.25)
2
) / ( ) / ( A Q A Q

where Q is the discharge of the flow, A is cross-section area,
2
* 0
u = is the mean
shear stress around the perimeter, is the fluid density, and

i i
P u
P
u

=
2
*
2
*
1
(3.26)
As the main objective of velocity measurement is to measure Q accurately, and as the
area A can also be determined from the measurements, one can calculate a numerical
value for the Darcy friction factor or Manning's n using only the information from this
velocity data.
43
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
f
g
R
n
C
8 1
6 / 1
= = From Chezys coefficient we simply obtain n as

g
u
A Q
R
g
f
R n
2
*
6 / 1
6 / 1
) / ( 8
= = (3.27)
This is the equation to calculate Manning's n for a cross-section using average shear
stress.
3.4 Sensitivity analysis
In practice, errors from measurements cannot be avoided. The errors of velocity
h s
obtained from this data. Therefore, it is necessary to investigate the effect of
measurement errors on the quality of the estimated n. In this section, firstly, the
analytical formula of the relationship between relative errors in estimated n and
relative measurement errors in depth (D) and in velocity ratio ( ) is derived. Secondly
some experimental data are used to verify this theoretical formula.
3.4.1 Sensitivity analysis
For an analysis of sensitivity, to simplify the problem, Manning's n at a single strip is
considered. For this case, the equation to estimate Manning's n is identical to Equation
(3.18):

measurement data will affect the quality of the estimated roug ne s coefficient
x

) 95 . 0 ( 57 . 5 +
=
x
n
) 1 (
6 / 1
D x
coefficient Furthermore, considering errors of the roughness ( n ), depth ( ) and the
ratio of two velocities ( ) in the three quantities, to first order:
D
x
x
x
n
D
D
n
n

= (3.28)
From Equations (3.18) and (3.28), the relationship between the relative error in n and
the relative errors in D and x is obtained as
44
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
x
x
x x
x
D
D n
n

+
+

=
1

) 1 )( 95 . 0 (
95 . 1
6
(3.29)
n the estimated n the relative errors in
x are plotted against the relative errors in n with different values of depth and the
that the relationship of the relative errors in n and
relative errors in x (the ratio of velocity at two-tenths the depth to that at eight-tenths
Equation (3.29) indicates that the relative error in n is always equal to 1/6 of the
relative error in depth D, while it is expected to be more sensitive to the relative errors
in x because of the term 1 x in the denominator.
In order to see the effect of errors in x on errors i
roughness coefficient (see Figure 3.2 and Figure 3.3). These relationships were
calculated from the depth range of 0.5 m to 4 m and with a roughness coefficient range
of 0.02 to 0.05. These are the common ranges of depth and the value of Manning's n in
natural streams.
From these figures it can be seen
the depth) depend on the depth and the roughness of streams. The smoother and deeper
a stream is, the more sensitive the relative error in n is to the relative error in x. This
indicates that the application of the two-point velocity method should be used with
caution in relatively smooth deep rivers. However, this finding needs to be verified
using the experiments that are discussed in the next section.
0
1
2
3
4
5
6
7
8
9
10
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Relative error in x (%)
R
e
l
a
t
i
v
e

e
r
r
o
r

i
n

n

(
%
)
D=0.5 m
D=1.0 m
D=2.0 m
D=3.0 m
D=4.0 m

Figure 3.2 Theoretical relationship between relative errors in estimated and relative
errors in with different depths for a channel with
n
x 035 . 0 = n
45
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
0
4
8
12
16
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Relative error in x (%)
R
e
l
a
t
i
v
e

e
r
r
o
r

i
n

n
(
%
)
n=0.020
n=0.025
n=0.030
n=0.035
n=0.040
n=0.045
n=0.050

Figure 3.3 Theoretical relationship between relative errors in estimated and relative
errors in for different roughness channels with a flow depth of 2 m.
3.4.2 Experimental work and analysis
The experimental work was conducted to measure the velocity profiles in the turbulent
boundary with different flow depths and roughnesses. The flow conditions were fully
rough turbulent. The main objective of the experiment is to investigate how sensitive
the relative error in the estimated value of Manning's n is to relative error in the ratio
of the velocities at two-tenths and eight-tenths of the depth, using two-point velocity
data wi
n
x
th different roughnesses and depths.
3.4.2.1 Experimental equipment
The experimental runs were conducted using a laboratory flume in the Michell
Hydraulic Laboratory, Department of Civil and Environmental Engineering at the
University of Melbourne. The water was supplied to the flume from a constant head
tank. Thus the supply always allowed steady conditions to be maintained. The inflow
to the flume was controlled by a valve in the main supply line. Figure 3.4 shows the
general arrangement of the experimental set-up.
The flume was 7100 mm long, 500 mm wide and 3800 mm deep. It was completely
made of plexiglass and had an adjustable bed slope. Water entered the flume from a
115 mm pipe, which discharged horizontally into a turbulent suppression tank that was
situated at the upstream end of the flume. A screen was provided inside the turbulent
46
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
suppression tank near the entrance of this pipe to dampen the turbulence generated by
the incoming flow into the tank.
The experiment was conducted using two different types of roughness. The first type
method of roughening has been used in the past of roughness was wire mesh. Such a
for simulating the bed roughness in free flow surface (e.g. Rajaratnam et al. 1976 and
Zerihun 2004). A piece of mild steel wire screen with mesh size 6.5 mm square was
tautly stretched over the entire surface of the flume bed and dots of superglue were
used to fix the screen flat on the bed. The diameter of the wire from which the screen
was made was 0.76 mm. The second type of roughness of the bed was gravel with the
sieve analysis of =
50
d 16.5 mm, =
84
d 19.5 mm and =
90
d 20.0 mm (see Figure 3.5).



50 cm Roughen bed
370 cm 340 cm
Turbulence
suppression tank
PLAN VIEW








Figure 3.4 The experimental set-up diagram (not to scale).
Roughness element
Water supply pipe
Roughen bed
LONGITUDINAL SECTION
ADV instrument
38 cm
Outflow to
measuring tank
47
Chapter 3 Estimation of Roughness Coefficients Using Velocity data

Figure 3.5 The two roughness types were carried out in the experiment.
3.4.2.2 Discharge m
ade with the
i
vertically and horizontally in the plane perpendicular to the flow. The exact
easurement
The discharge for all the tests was determined by the measurement m
laboratorys measuring tank system. Flow from the downstream end of the flume was
normally conveyed by a small concrete canal to the sump of the laboratory for
recirculation. During the tests for a given length of time this flow was diverted directly
into the measuring tank which was located below the floor level of the laboratory for
the purpose of measuring the volume rate of the flow. A centrifugal pump was used to
empty this tank after each test run. The discharge measuring system consists of the
measuring tank with plan dimensions of 2000 mm by 1500 mm and a depth of 4500
mm, and a manometer inclined at an angle of 56.62
o
(to the horizontal) to measure the
depth of water in the tank. During the tests, a tank filling time longer than one minute
was used to minimise errors associated with the starting and stopping of the stopwatch.
For each test, two readings were taken for each parameter related to the d scharge
measurements and the average of the resulting discharge values was used in the
analysis.
3.4.2.3 Velocity measurement
The vertical velocity profiles were measured by an Acoustic-Doppler Velocimeter
(ADV) of a two-dimensional (2-D), side-looking probe manufactured by SonTek Inc.
The probe was installed in the flume by a coordinate meter allowing motion of the
probe
48
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
position of the measurement volume was easily established by the coordinate meter
and the distance from the flume wall was varied in a controlled manner.
The two-dimensional side-looking probe that was used in this experimental work had
the advantage of measuring the velocity very close to the bed. For this version of the
instrument, the sample volume is located 50 mm (horizontally) away from the tip of
the probe. This enables measurements to be taken without interfering with the flow.
The main objective of velocity profile measurement was to determine Manning's n by
using the whole velocity profile and the two-point velocity method. For these
purposes, velocity observations were done at closely spaced sections so that they could
accurately describe the actual flow profile. The duration of each velocity measurement
was tly
e
set between 60 and 65 s. Since the velocity of the flow was measured indirec
from the scatter, seeding materials were added in each experiment to improve the
quality of observed data by decreasing the Doppler noise.
In order to obtain reliable results, WinADV32 software (Wahl 2000) was used to filter
the poor quality data from the data file based on the criteria of signal to noise ratio
(SNR) and the correlation (COR) parameter. The manufacturers recommendation of
SNR and COR values of not less than 15 and 70 respectively were used to filter the
data. The time-averaged local velocities at different levels were also calculated using
this software. Figures 3.6 and 3.7 show an example of the time series velocity and the
associated average values of COR and SNR parameters of the gravel bed of the flum
for the flow depth of 8.5 cm. The measurement point was at a height of 1.7 cm above
reference level. In Figure 3.7 the lower and upper curves are for variation of the
average value of SNR and COR parameters respectively.

49
Chapter 3 Estimation of Roughness Coefficients Using Velocity data

Figure 3.6 Velocity time series measurem 7 . 1 ent at = z cm of the gravel bed flume of
8.5 cm water depth.

Figure 3.7 Average SNR and COR values and recorded flags for measurement at
7 . 1 = z cm of the gravel bed flume of 8.5 cm water depth.
3.4.2.4 Determination of the location of velocity measurement
It is known that the turbulent boundary layer is expected to be fully developed at a
distance of about 50-70 times the depth from the entrance of the flume (Balachandar et
al. 2002). The length of flumes in laboratories is usually limited. In order to reduce the
layer, a roughness element should
be placed at the entrance. As a result, the turbulence boundary layer will develop at the
s
urpose, a trip wire of 1 cm diameter located higher than the roughened surface was
placed at the entrance of the flume (see Figure 3.4). Also, in order to determine the
appropriate position of the ADV instrum where the turbulent boundary is fully
developed, some preliminary measurement profiles of the maximum depth ( cm)
of all test runs alo re 3.8 shows the
velocity profiles at the m s of some cross-sections along the flume,
wher ance the e ce p the flu From e
length of the full development of turbulent boundary
very beginning of the channel and the total length of the zone for the full development
of boundary layer can be shortened (Chow 1959 and Balachandar et al. 2002). For thi
p
ent,
9 = D
ng the centre line of the flume were conducted. Figu
iddle vertical line
from e L is the dist ntran oint of me. the figure, it can b
50
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
seen that the velocity pr s at atio ofile the loc n 5 . 3 = L m and 5 . 4 = L m wer d to
be almost identical. Thi icate from ocat
e foun
s ind s that the l ion 5 . 3 = L m a nearly uniform
flow state was attained and the turbulent boundary layer was considered fully
de eloped. Therefore, t ocati v he l on 7 . 3 = L m was c to m e th city
profiles for all test runs.
hosen easur e velo

0
0.3
0.4
0.5
0.6
0.7
0.8
0
Z
/
d

.9
1
L=2.0m
L=2
0.1
0.2
0 10 20 30 40 50 60 70 80
Velocity (cm/s)
.5m
L=3.5m
L=4.5m

Figure 3.8 Variation of velocity profiles at different locations along the flume
or evaluating the water depth accurately, the reference level needs to be defined. This
position of the bottom allows agreement between the measured velocity distribution
nd logarithmic distribution law. According to Hinze (1975), there is uncertainty in
defining the position of the plane
F
a
0 = z of the rough wall and no generally accepted
many studies have tried to find the position of the
ference plane. For a wire screen with a wire diameter d
w
, Furuya and Fijita (1967)
and Rajaratnam et al. (1976) obtained values of 0.4d
w
to 0.8d
w
for the location of the
rigin below the surface of the wire screen. Therefore, in this experiment the reference
level of 0.5d
w
below the surface of the wire screen was adopted.
For the gravel bed, Kamphuis (1974) consistently showed that for
definition is available. However,
re
o
4 . 0 = the straight
line of best fit would indicate the virtual bottom to be 0.7d
90
above the top of the
contact surface. From this, the virtual bed is 0.7d
90
above the plane where the particles
are fastened. Some other researchers (e.g. Bayazit (1976), Zippe and Graf (1983))
suggested that the reference level would lie at a distance below the top of the elements,
51
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
having a diameter d, equal to 0.15 to 0.27d. Therefore, in th ex
ference level for the gravel roughness was taken as 0.75d
is periment, the
re
90
(15mm) above the
surface of the flume.
3.4.2.5 Scope of the experiment
As mentioned above, the purpose of conducting the experiment at different
depths/discharges on different roughnesses is to assess the sensitivity of relative error
in x to relative error in n. Two series of experiments were performed with two types of
roughness. All experimental tests were for the case of fully rough turbulent flow. The
values of Reynolds number Re ( = / Re VR , where V is the average flow velocity, R
is the hydraulic radius and tic viscosity) ranged from 15000 to 30000
and the value of roughness Reynolds num
is the kinema
ber = / Re Vd
k
Re (
k
, where is the heigh d t
of roughness) ranged from 71 to 902. The depth of the flow was ranged from 6.4 to 9.0
cm. The minimum value of the depth was restri d to e e ADV instrument
could measure the velocities at two-tenths and eight-tenths the depth because this
instrument can no easure the es at the points below 8 mm from the bed
surface or less tha mm from surface. T upper valu depth were
restricted by / D B in orde ure the v at the ine was not
affected by the dip-phenomenon (the maximum velocity appears below the water
surface) and the velocity distrib logarithmi distribution for the whole depth
ezu and Rodi 1985 and Yang et al. 2004). The increment of depth between each
experiment ranged from 0.4 to 0.6 cm. These gradual increments of depth in each
different flow depths and roughnesses.

cte nsure that th
t m velociti
n 12 the free he es of the
5 > r to ens elocity central l
ution is a c

(N
experiment enable us to obtain the data with different depths to achieve the purposes of
the experiment. The depths and discharges together with other corresponding
parameters in the tests with the two roughness types are given in Table 3.1.
3.4.2.6 Experimental results and discussion
Eleven test runs were conducted (see Table 3.1) in order to investigate the effect of
relative errors in x to relative errors in n with

52
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
Table 3.1 Characteristic data of experimental runs
type
Depth
(cm)
Q
(l/s) (cm/s)
Surface
V
Re Re
k
Fr
6.4 13.70 42.81 19150 71.0 0.540
7.2 16.68 46.33 22472 73.4 0.551
7.5 17.85 47.59 23729 77.4 0.555
8.1 20.29 50.10 26279 79.0 0 62 .5
8.5 21.93 51.60 27919 80.4 0.565
Wire mesh
76 . 0 =
w
d mm

9.0 24.12 53.61 30072 82.5 0.571
6.5 10.87 33.44 15124 772.2 0.419
7.0 12.34 35.76 16855 803.6 0.435
7.5 14.07 37.53 18714 838.2 0.438
8.0 15.90 39.27 20597 871.2 0.441
Gravel bed
5 . 16
50
= d mm

8.5 17.67 41.58 22500 902.6 0.455

For each test, firstly the whole velocity profiles were first measured at every 2 or 3 mm
Figure 3.9 Measured velocity profiles
intervals. Secondly the velocities at two-tenths and eight-tenths the depth were
independently measured 30 times at the central vertical line. All measured velocity
profiles were approximately logarithmic distributions. Figure 3.9 shows two examples
of the velocity profiles with the flow depth of 8.5 cm for the wire screen roughness and
the gravel roughness.






Depth =8.5 cm, wire mesh roughness
u(z) = 13.389(ln(z)) + 36.19
R
2
= 0.978
0
10
20
30
40
50
60
70
0 1 2 3 4 5 6 7 8 9
Distance from the bed (cm)
V
e
l
o
c
i
t
y

(
c
m
/
s
)
Depth =8.5 cm, gravel roughness
Distance from the bed (cm)
u(z) = 14.069(ln(z)) + 25.469
R
2
= 0.9764
0
10
20
30
40
50
60
0 1 2 3 4 5 6 7 8 9
V
e
l
o
c
i
t
y

(
c
m
/
s
)
53
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
In order to assess the effect of error in x to error in n, firstly the true values of
Manning's n should be determined. Initially, the direct method (see Section 2.3.1) was
applied to calculate the true values of Manning's n of the rough bed of the flume. The
water depths were measured at three locations along the flume (at m,
m and m). Unfortunately, because the length of the flume was not long
enough, the distances between these locations were too close. As a result, a small error
in measured depths could produce a large error in computed values of Manning's n,
and hence the computed values for n obtained from this data were found not to be
accurate. However, as mentioned in Section 3.2, if the velocity distribution is known,
the value of Mannings n can be determined. Therefore, in these experiments, this
method was used as an alternative to the direct method of estimating the true values of
exa d.
The results are shown in Table 3.2.
Table 3.2 Computed roughness coefficients from the full velocity profiles
Wire mesh Gravel bed
1 . 2 = L
7 . 3 = L 7 . 4 = L
n. For each flow depth, the whole velocity profile was measured (see Figure 3.9 as
mples). From these velocity profiles the true values of Manning's n were compute
Depth (cm) n Depth (cm) n
6.4 0.02186 6.5 0.02807
7.2 0.0 .9 0.02794 2175 6
7.5 0.02168 7.5 0.02785
8.1 0.02165 8.1 0.02773
8.5 0.02160 8.5 0.02765
9 0.02150

On the other hand, for each flow depth, 30 independent measurements of velocities
were taken at two-tenths and eight-tenths the depth. From these measurements, 30
values of x and 30 values of n were computed using the two-point velocity formula
(Equation 3.18). From 30 independent measurements of velocities at two-tenths and
eight-tenths of the depth 30 values of x, 30 estimated values of Manning's n were
computed by using two-point velocity formula (Equation 3.18). Then the relative
errors in x and n were calculated as follows:
54
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
% 100
x
x x
E
i
x

= (3.30)
and
% 100
n
n n
E
i
n

= (3.31)
where x
i
is the ratio of
2 . 0
u and
8 . 0
u of i
th
measurement; n
i
is the estimated Manning's n
by using two-point velocity method of i
th
measurement; x is the mean value of x; n is
the roughness coefficient computed from the whole velocity profile;
x
E and
n
E are the
relative errors in x
i
and in n
i
of i
th
measurement.
Fig s ive
erro i lts for the wire mesh and the gravel bed
resp t
betwee
This c
coeffic , the more sensitive the relative errors in estimated n
are t
The rel
with th ) but with the two types of roughness (see Figure
3.12 the
in n are to the relative errors in x. However, this figure also indicates
that the rougher a channel, the higher the relative error in x, which results in a higher
the
smoother and deeper a stream, the more sensitive the relative error in estimated n is to
ure 3.10 and 3.11 show the relationships between relative errors in x and relat
rs n n obtained from the experimental resu
ec ively. From these figures, it can be seen that there is very good agreement
n experimental results and the corresponding theoretical lines (Equation 3.29).
onfirms that when using two-point velocity data to estimate the roughness
ient, the greater the depth
to he relative errors in x.
ative errors in x were also plotted against the relative errors in n for the cases
e same depth ( 5 . 7 = D cm
). This figure shows clearly that the smoother a channel, the more sensitive
relative errors
relative error in n.
In brief, the sensitivity analysis indicates that the errors in the estimated n are very
sensitive to the errors in x. The theoretical and experimental work shows that
the relative error in x. However, for rougher channels with shallow depth, the errors in
velocity measurement may be higher because of higher disturbance of roughness
elements. Accordingly, the relative errors in x are also higher, which will result in
55
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
higher relative errors in n. Therefore, this method should be used with caution and
needs to be verified with natural streams.
2
4
6
8
10
12
14
r
e
l
a
t
i
v
e

e
r
r
o
r

i
n

n

(
%
)
D=6.4 cm
D=7.5 cm
D=9.0 cm
0
0 1 2 3 4 5 6
relative error in x (%)

Figure 3.10 Experimental relationships between relative errors in x and relative errors
in estimated n and corresponding theoretical lines for wire mesh roughness with
different depths

2
r
4
6
8
10
12
14
e
l
a
t
i
v
e

e
r
r
o
r

i
n

n

(
%
)
D=6.5 cm
D=7.5 cm
D=8.5 cm
0
0 1 2 3 4 5 6 7 8
relative error in x (%)

Figure 3.11 Experimental relationships between relative errors in x and relative errors
in estimated nand corresponding theoretical lines for the gravel bed with
different depths
56
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
0
2
4
6
8
10
12
0 1 2 3 4 5 6 7 8
relative error in x (%)
r
e
l
a
t
i
v
e

e
r
r
o
r

i
n

n

(
%
)
14
D=7.5 cm - gravel bed
D=7.5 cm - wire mesh

Figure 3.12 Experimental relationships between relative errors in x and relative errors
in estimated and corresponding theoretical lines for the same depth with different
types of roughness
Application to natural rivers
French and McCutcheon (1977) applied the method using velocity measurement data
estimate Manning's n for the Cumberland River and found that this method provided
asonable estimates. However, they noted that the methodology needs additional
verification. Therefore, in this study, the proposed methods described in Section 3.3 to
estimate the roughness coefficient in open channels using two-point velocity gauging
data were applied to some natural rivers. In order to assess the performance of the
method, the selected rivers should satisfy the following preliminary conditions:
A wide channel (width/depth ratio >9) with a simple trapezoidal geometry;
Two-point gauging data are available;
Manning's n calculated from the direct measurement method (as described in
Section 2.3.1 of Chapter 2) considered as the measured n of the rivers, are
available. These values will be considered as a basis to assess the performance
of the proposed method;
The values of discharges obtained from velocity gauging data should be
identical or as close as possible to the discharges used to calculate n from direct
measurement method; and
n
3.5
to
re
57
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
e.
3.5.1 Data
With reference to the conditions mentioned above, some rivers in New Zealand and
Australia where the measured Manning's n (computed by the direct method) available
were chosen. The rivers were selected based on two sources: Hicks and Mason (1991)
for the rivers in New Zealand and Ladson et al. (2006) for the rivers in Australia.
For New Zealand rivers, Hicks and Mason (1991) have presented 78 rivers where
roughness coefficients have been computed for each river with different discharges.
However, only 40% of them (32 rivers) had velocity gauging data available. For the
other rivers, the discharges were computed from the rating curves. Moreover, for the
rivers with velocity gauging data, many of them had only one velocity measured at
each vertical line (at six-tenths of the depth). There are two main reasons for this. The
first one is that in some rivers the water depth is too small to take two-point velocity
measurements. The second one is that the velocities were measured in high flow
conditions. For these conditions, measuring only one-point velocity could reduce the
errors because of the effects of logs/debris and the rapid rate of change of the water
level. This could reduce the time of measurements and also limit dangers under these
flow conditions (Pete Mason and John Porteous, National Institution of Water and
Atmospheric Research Ltd., New Zealand, personal communication, 2005).
Furthermore, several rivers presented by Hicks and Mason are not wide channels.
The velocity data of the rivers in New Zealand was obtained from National Institution
of Water and Atmospheric Research Ltd., New Zealand (NIWA). Unfortunately, the
gauging cards of some rivers were missing and only the gauging cards of 12 rivers
could be obtained, of which only 10 rivers can be considered to be wide channels to
apply for the proposed method.
For the rivers in Australia, Ladson et al. (2006) computed the values of Manning's n
directly from field measurement similar to Hicks and Mason (1991) for 4 rivers in
Victoria, Australia. They indicated that all the discharges used to determine the
Manning's n were determined from the applicable rating table. These 4 rivers were
selected to assess the performance of the proposed method. For this purpose, the
The roughness characteristics and cross-sections of rivers should be stabl
58
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
discharges corresponding to the velocity data should be the same as or as close as
possible to the discharges used by Ladson et al. (2006) for calculating the value of the
measured n. The velocity measurements at the gauges of those rivers were supplied by
Thiess Environmental Services Pty Ltd.
Finally, 14 rivers were chosen to verify the performance of the proposed method
including 4 rivers in Victoria, Australia and 10 rivers in New Zealand. All the
discharges for which the data were used to estimate the roughness coefficient were
within-bank flows. Brief descriptions of the bed and banks of those rivers are
presented in Table 3.3. The main hydraulic and roughness characteristics of these
rivers are summarised in Table 3.4. More detailed descriptions of these rivers are
available in Hicks and Mason (1991) and Ladson et al. (2006).
3.5.2 Results and discussion
velocity-gauging cards. The m n (obtained from the direct
measurement method in Hicks and Mason (1991) and Ladson et al. (2006)) and the
computed n for these rivers by the two proposed formulae (Equations 3.23 and 3.27)
u
sectional perimeter respectively are presented in Appendix A. Figures 3.13 and 3.14

The proposed method was applied to the 14 rivers described above including 68
easured Manning's
sing the average equivalent roughness and the average shear stress along the cross-
plot the computed roughness coefficients against the measured ones. From these
figures it can be seen that generally there is a relatively good agreement between the
computed and measured values. Table 3.5 summaries the results of computed
roughness coefficients in terms of mean relative errors (ARE) of the computed
roughness coefficients with respect to the measured ones for the 14 rivers.




59
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
Table 3.3 Brief description of the bed and banks for the selected rivers
No. River name Location Description
1 Acheron
trees and grass.
1
at Taggerty
Bed consists of gravel and cobbles. Banks are
lined with some overhanging
2
Merriman
Cr
at e West
Bed material is mainly fine sands and clay. Banks
and dense
etatio
eek
1

Stradbrok are lined with overhanging trees
willow and veg n.
3 Mitta
1

jie Bed consists of large gravel es. Banks
ar with overhanging
Mitta
at Hinnomu
Bridge
and cobbl
trees. e lined
4 Tam
1

of
reek
B sists of gravel an les. Banks are
li h overhanging tree rass.
bo
at downstream
Ramrod C
ed con d cobb
ned wit s and g
5 Gre
B bbled. Right ban mposed of bed
r bles and soil, w se bush cover.
L is composed of s, silt and soil,
an hick bush cover w es overhanging
th el.
y
2
at Dobson
ed is co
ock, cob
k is co
ith spar
eft bank cobble
d has t
e chann
ith tre
6 Pou
B ists of an eroded ow. Banks are
li grass and bro ar the control
st
tu
2
at Ford
ed cons lava fl
ned with
ructure.
om. Ne
7
Oa
Ca
ert
A l man-made canal. Bed and banks are of
c d gravel and bou ith area of silt
a growth. Near the culvert.
kden
nal
2

at Oakden culv
arge
ompacte lders, w
nd weed control
8 On e
2

B sists of gravel a les. Farmland
g ne both banks.
garu at Taringamotu
ed con
rasses li
nd cobb

9 Rangitaiki
2
at Te Teko
overh the
reach ion by
operation.
B rial consists of s h some gravel.
B terial is mainly s sand. Willows
ang the banks along more than half
. Lower bank are kept clear of vegetat
water level fluctuations associated with hydro
ed mate and wit
ank ma ilt and
10 Tahunatara
2
at Ohakuri Road
Bed is composed with pumi
formed of silt and are grasse
ce sand. Banks are
d down to the mean
water level.
11 Waipapa
2
at Ngaroma Rd.
Bed material is bed rock along entire reach.
Vegetation overhangs both banks.
12 Wanganui
2
(1) at Te Whaiau Canal
A man-made channel. Bed and banks consist of
sand and gravel. Banks vegetated with grasses
and flaxes. Near the control structure.
13 Wanganui
2
(2) at Wairehu Canal
A man-made channel. Bed and banks consist of
uniformly-sized cobbles. Broom grows along the
banks.
14 Whirinaki
2
at Galatea
Bed material is mainly cobbles but with some
boulders and patches of sand gravel. Banks
formed of similar material are lined with
overhangin willows and a dense scrub of
juvenile w s.
60
g
illow
Note:
1
indicates rivers in Australia;
2
indicates rivers in New Zealand;
Source: Hicks and Mason (1991) and Ladson et al. (2006).

Chapter 3 Estimation of Roughness Coefficients Using Velocity data

61
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
0.06
0.07
0.08
0.09
0.02
0.03
0.04
0.05
a
t
e
d
0
0.01
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
E
s
t
i
m

n
Measured n
Poutu Wanganui 1 Wanganui 2 Grey Ongarue
Oakden Canal Rangitaiki Tahunatara Waipapa Whirinaki
Acheron Tambo Mitta Mitta Merriman Agreement line

Figure 3.13 Computed n obtain from velocity data using average equivalent roughness
versus measured n
0
0.01
0.02
0.03
0.04
E
s
t
i
m
0.05
a
t
e
0.06
0.07
0.08
0.09
d

n
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Measured n
Poutu Wanganui 1 Wanganui 2 Grey Ongarue
Oakden Canal Rangitaiki Tahunatara Waipapa Whirinaki
Acheron Tambo Mitta Mitta Merriman Agreement line

Figure 3.14 Computed n obtain from velocity data using average shear stress versus
measured n
From Table 3.5 it can be seen that th varies from
river to river. N reaches have a mean relative error with respect to
the me rs ve some comm
e le od to provide better re ers, their widths are
e quality of the computed Manning's n
ine of these 14 river
asured n less than 25%. These rive ha on features that could
nab the proposed meth sults. For some riv
62
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
greater than 30 m, such as the Grey River, the Ongarue River, the Rangitaiki River, the
M a the Tambo River (see Table 3.4). For these rivers, the
percentage of wetted perimeter with vegetation is usually less than 25% of the total
channel wetted perimeter. Therefore, the effec icant (Coon
1998). For the other rivers, their bank vegetation is less dense than the others like
Acheron River and the Tahunatara River (see Table 3.3). This limits the effect of
vegetation which can distort the velocity distribution. Or for the Waipapa River and
the Wh eir aspect ratios (B/D) are very large (greater than 20) (see
Table 3.4). This ratio is one of important factors to ensure the velocity distribution is
a ox mic distribution.
T le puted Manning's n for the selec
results are listed in ascending order of MRE)
Mean relative error (MRE) * (%)
itt Mitta River, and
t of vegetation is insignif
irinaki River, th
ppr imately logarith
ab 3.5 Mean relative error of the com ted rivers (the
No. River
Equation (3.23) Equation (3.27)
1 Mitta Mitta
1
8.49 7.73
2 Tahunatara
2
16.23 12.31
3 Tambo
1
10.18 19.97
4 Whirinaki
2
17.38 12.46
5 Ongarue
2
20.47 13.73
6 Acheron
1
20.33 19.51
7 Grey
2
25.03 16.68
8 Rangitaiki
2
23.56 24.93
9 Waipapa
2
24.71 24.23
10 Merriman Creek
1
24.48 44.76
11 Wanganui
2
1 30.75 29.12
12 Oakden Canal
2
38.23 33.40
13 Poutu
2
57.54 24.26
14 Wanganui
2
2 65.27 56.77
Note: * Mean relative error (MRE) is defined as ( ) % 100 . / / N n n n
meas meas comp
, where
and are computed and measured Manning's n, and N is number of computed n;
1

indicates the rivers in Australia and
2
indicates the rivers in New Zealand

comp
n
meas
n
63
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
The other 5 rivers that gave poor results in computed Manning's n are Oakden Canal,
the Poutu River, both the Wanganui Rivers and Merriman Creek. These rivers have
some main features that may be the reasons of failure in estimating n. Firstly, they are
near the control structures such as Oakden Canal, the Poutu River, the Wanganui (1)
River (see Table 3.3). Secondly, all these rivers have dense bank vegetation with
smaller aspect ratios and narrower widths compared to the other rivers such as the
Poutu River ( = B 7-10 m), the Wanganui (2) River ( = B 11-14 m) and Merriman
Creek ( = B 16-26 m). These features may distort the velocity distribution and produce
from the velocity data used to calculate this coefficient were not
identical. If the conditions of the selected rivers are rather stable but these discharges
tical this can still produce som difference between the compute
measured values of Manning's n. For example, for the cases of the four rivers in
es used to calculate the measured n were obtained from the
rating curves (Ladson et al. 2004). And the last reason is that the n values obtained
y measurement data could represent only the average cross-sectiona
while the measured n obtained from direct method represent the roughness coefficients
ults indicated that the wide channel condition (aspect ratios
greater than 10) is not enough to ensure to obtain a reliable result for Manning's n from
e proposed method by using velocity data. Although experimental observations have
shown that the logarithmic velocity distribution adequate to represent the vertical
ariation in many wide channels (Chow 1959 and French 1985), there are a number of
rt this distribution such as non-uniform flow conditions (Carling
1992), flow stratification (French 1985), extreme bed roughness (Jarrett 1990) and
the poor results in the computed n.
The discrepancies between the measured and computed Manning's n may also be
attributed to some others reasons. The first one is measurement errors which are
unavoidable. For calculating the measured values of n, the data required include water
stages, discharges, cross-sections and friction slope. As a result, not only the computed
n obtained from velocity data suffers from measurement errors (as analysed in Section
3.4) but the measured n obtained from the direct method also suffer from them. The
second one is that the discharges used to calculate Manning's n by direct method and
the discharges
are not iden e d and
Australia, the discharg
from velocit l ones
of the computed reaches.
The analysis of the res
th
v
factors that can disto
64
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
bank drag (Carling 1992). By considering the quality of the computed n with the
conditions of the rivers applied in this study and referring the factors mentioned above,
e following applicable conditions for the proposed method are suggested:
A wide channel (width/depth ratio>10) carrying unstratified flow;
A uniform flow or quasi-uniform and steady condition without overflow;
Straight channel with a simple trapezoidal and uniform cross-sections;
A small scale roughness, i.e., the uniform flow depth is much higher than the
characteristic size of the particles arranged on the channel boundary D/d
50
>7
(Bathurst et al. 1981);
No significant vegetation in the channel bed and banks. If the bank vegetation
is significant, the percentage of vegetation should be less than 25% (usually
satisfied for channel with a width larger than 30m) (Coon 1998);
A stable channel in natural equilibrium with no dominant bedform features;
and
No effect of control structures or obstacles that may affect the velocity
distribution.
The above discussion on the performance of the proposed method by comparing the
omputed Manning's n with the measured ones shows that there is a relatively good
ent between these values. The reasons for some discrepancies between these
values have been analysed. The applicable conditions for the proposed method have
een made. However, in order to assess the significance of the proposed method it is
necessary to compare it with other published equations. This will be discussed in the
ext section.
3.6
with other empirical equations
Numerous empirical equations for estimating Manning's n have been considered.
Based on the hydraulic and roughness characteristics of the selected rivers (see Tables
.3 and 3.4) and the applicable conditions of the equations (discussed in Coon (1998)
and Lang et al. (2004)), 10 empirical equations for estimating Manning's n were
th
c
agreem
b
n
Comparing the performance of the proposed method
3
65
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
selected to compare with the proposed method (see Table 3.6). Although it is
impossible to consider all equations that have been developed to estimate n, they are
Table 3.6 Some applicable equations for the selected rivers
No Investigator/ Author Equation*
1 Strickler (1923) (cited in Yen (1991)) (3.32)
6 / 1
50
0747 . 0 d n =
2 Henderson (1966)
6 / 1
75
3048 . 0
031 . 0

=
d
n (3.33)
3 Limerinos (1970)

+
=
84
6 / 1
log 03 . 2 16 . 1
113 . 0
d
R
R
n (3.34)
4 Riggs (1976)
w
S
w
S R A n
log 056 . 0 45 . 0 3 / 2 33 . 0
55 . 1
1
+
= (3.35)
5 Bray (1979)

+
=
50
6 / 1
log 36 . 2 248 . 0
113 . 0
d
R
R
n (3.36)
6 Bray (1982)
21 . 0 067 . 0
0 . 8
1
w
S R n = (3.37)
7 Bray (1982) (3.38)
177 . 0
104 . 0
w
S n =
8 Griffiths (1981)

+
=
50
6 / 1
log 98 . 1 76 . 0
113 . 0
d
R
R
n (3.39)
9 Sauer (1990) (cited in Coon (1998))
08 . 0
18 . 0
3048 . 0
11 . 0

=
R
S n
w
(3.40)
10
w
S log . .
w
.564 1

Dingman and Sharma (1997)
. .
S R A n
0543 0 5 0 267 0 173 0
1
+
=
(3.41)
* where n is the roughness coefficient, d
x
is the bed material diameter such that x percent
of material by weight is smaller (in m), A cross-sectional area (in
m
2
), R is the hydraulic radius (in m); S
w
is the water-surface slope and B is the top width of
the stream.
is the wetted channel

66
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
representative of the many forms that have been proposed. These equations were
chosen based on the application conditions which are as close as possible to the
conditions of the selected rivers in this study. It can be seen that many of these
empirical formulae required bed material information (i.e. the value of
x
d ). However,
these data are not available for the Merriman Creek and the Waipapa River. Thus, in
the comparison of these formulae with the proposed method, these two rivers were
excluded.
In order to compare the performance of the different methods for estimating
Mannings n, the computed n were plotted against the measured ones (see Figures
3.15a and 3.15b). From these figures it can be seen that the relationship points of
com d
the agreement line and are less biased than the other equations. However, it was found
that it is more e to compare the ent eq s by
using the range of difference between c asure and
criteria. In this study, the mean absolu sed b (19 d the
mea MRE) used by Afz til (1998) were chosen. They
are

puted and measured n obtained from the proposed equations are scattered aroun
indicativ the performance of differ uation
omputed and me d values some error
te error (MAE) u y Coon 98) an
n relative error ( alimehr and Anc
defined as


meas comp
n n =
N
MAE
1
3.42)
and

(


meas comp
n n
=
N
1

t method and N is the number
of computed n.
meas
n
MRE (3.43)
where n
comp
is the computed Manning's n by using a certain equation, n
meas
is the
measured Manning's n that is determined using the direc


67
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
68


















Figure 3.15a Computed n obtained from different equations versus measured n
for the 12 selected rivers

Henderson (1966)
0.06
0.08
e
d

n
0
0 0.02 0.04 0.06 0.08
Measured n
0.02
0.04
C
o
m
p
u
t
Strickle 1923)
0.06
0.08
e
d

n
r (
0.04
m
p
u
t
0.02
C
o
0
0 0.02 0.04 0.06 0.08
Measured n
Limerinos (1970)
0
0.02
0.04
0.06
0.08
0 0.02 0.04 0.06 0.08
Measured n
C
o
m
p
u
t
e
d

n
Riggs (1976)
0.00
0.02
0.04
0.06
0.08
0 0.02 0.04 0.06 0.08
Measured n
C
o
m
p
u
t
e
d

n
Bray (1979)
0
0.02
0.04
0.06
0.08
0 0.02 0.04 0.06 0.08
Measured n
C
o
m
p
u
t
e
d

n
Bray (1982)
(Equation 3.37)
0
0.02
0.04
0.06
0.08
0 0.02 0.04 0.06 0.08
Measured n
C
o
m
p
u
t
e
d

n
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
69
Figure 3.15b Computed n obtained from different equations versus measured n
for the 12 selected rivers


















Griffiths (1981)
0
0.02
0.04
0.06
0.08
C
o
m
p
u
t
e
d

n
0 0.02 0.04 0.06 0.08
Measured n
Bray (1982)
(Equation 3.38)
0
0.02
0.04
0.06
0.08
C
o
m
p
u
t
e
d

n
0 0.02 0.04 0.06 0.08
Measured n
Sauer (1990)
0
0.02
0.04
0.06
0.08
0 0.02 0.04 0.06 0.08
Measured n
C
o
m
p
u
t
e
d

n
Dingman and Sharma(1997)
0
0.02
0.04
0.06
0.08
0 0.02 0.04 0.06 0.08
Measured n
C
o
m
p
u
t
e
d

n
Velocity - k
s
0
0.02
0.04
0.06
0.08
0 0.02 0.04 0.06 0.08
Measured n
C
o
m
p
u
t
e
d

n
Velocity -
0
0
0.02
0.04
0.06
0.08
0 0.02 0.04 0.06 0.08
Measured n
C
o
m
p
u
t
e
d

n
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
Table 3.7 presents a range of difference between computed and measured n and the
values of MAE and MRE obtained from different equations. From this table it can be
seen that no single equation was capable of accurately estimating n values for all
r and the more symmetric of the difference
ted and measured values as well as the smaller the values of MAE
and MRE indicate the better the performance of the equation. As shown in Table 3.7,
82), Sauer (1990), and
Dingman and Sharma (1997) performed better than the others for these study cases. It
the proposed method
are smaller than the ones obtained from the other methods. This indicates that the
proposed method can provide a better estimation of Manning's n in comparison with
the empirical equations.
e difference between and measured n and estimated n from
different equations and the corresponding values of MAE and MRE
discharges of all the channels. The narrowe
between the compu
for the empirical equations, the equations proposed by Bray (19
also can be seen that the values of MAE and MRE obtained from
Table 3.7 Range of th
Method Range of differences MAE MRE
Strickler (1923) -0.035 to 0.008 0.0103 0.275
Henderson (1966) -0.037 to 0.005 0.0120 0.304
Limerinos (1970) -0.032 to 0.016 0.0096 0.301
Riggs (1976) -0.021 to 0.017 0.0091 0.285
Bray (1979) -0.035 to 0.015 0.0099 0.286
Bray (1982) (Equation 3.37) -0.033 to 0.012 0.0103 0.269
Bray (1982) (Equation 3.38) -0.031 to 0.014 0.0100 0.266
Griffiths (1981) -0.033 to 0.014 0.0092 0.272
Sauer (1990, in Coon 1998) -0.029 to 0.016 0.0082 0.244
Dingman and Sharma (1997) -0.020 to 0.020 0.0084 0.261
Velocity -
s
k (Equation 3.23) -0.023 to 0.019 0.0077 0.246
Velocity -
0
(Equation 3.27) -0.018 to 0.017 0.0069 0.209
Note: A negative difference indicates an underestimation of the computed n value by the
equations. A positive difference indicates an overestimation.
equations obtained from the average equivalent Comparing the two proposed
roughness (k
s
method) and the average shear stress along the wetted perimeter (
0

method), the values of MAE and MRE for the k
s
method were 0.0077 and 0.2466
70
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
respectively, while these values for the
0
method were 0.0069 and 0.2094
respectively. Especially, the difference range of the computed and measured Manning's
n obtained from the
0
method is narrower and less biased. These results indicate that
the formula obtained from average shear stress is superior. This formula can avoid the
use of Keulegans equation (Equation 3.12) for rough channels where the first constant
in this eq
3.7 Summ clusi
In this chap ere pr e g th s coefficient
Manning's n measurem the average
based on the average shear stress along the
cross-section nsitiv sis was performe nvestigate the
ensitivity of errors in estimated n to measurement errors. Experimental work was also
conducted to verify the theoretical findings. The results indicated that the smoother and
The proposed formulae were applied to 14 wide rivers (the ratio between width and
uation actually varies in a wide range as mentioned above.
ary and con on
ter, two formulae w oposed for stimatin e roughnes
using velocity ents. The first one is based on
equivalent roughness and the second one is
perimeter. A se ity analy d to i
s
deeper the river, the more sensitive the error in n is to the error in the ratio of velocity
at two-tenths the depth to that at eight-tenths the depth. However, for rougher channels
with shallower depth, the errors in velocity measurement may be higher because of the
greater disturbance of the roughness elements. These roughness elements may also
cause higher relative errors in x and thus result in higher relative errors in n.
depth being greater than 9) in Australia and New Zealand. In these rivers, the values of
Manning's n determined from the direct measurement method (in Hicks and Mason
(1991) and Ladson et al. (2006)) were considered as the measured roughness
coefficients. These were the basis for assessing the ability of the proposed method
using velocity data. There is relatively good agreement between the computed
roughness coefficients obtained from the proposed method and the measured ones. The
reasons for considerable discrepancies between the computed and measured values of
Manning's n in some rivers were analysed. The analysis shows that the wide channel
condition is not enough to ensure reliable results of the estimated roughness
coefficients. Therefore, the applicable conditions for the proposed method were
suggested. For the two proposed formulae, the formula based on average shear stress
gave better results.
71
Chapter 3 Estimation of Roughness Coefficients Using Velocity data
The proposed formulae for estimating Manning's n using velocity data were also
compared with some other applicable empirical equations. Although velocity
measurement errors were unavoidable or the assumption of logarithmic velocity
distribution could be violated, the proposed method provided similar or better results

than other empirical equations. It is suggested that the proposed method can be used as
a means to estimate the values of Manning's n for wide streams where two-point
velocity data are available, which is the case for many existing stream-gaugings.

72
Chapter 4
Chapter 4
Roughness Identification Model
Development for Unsteady Flow
4.1 Introduction
As discussed in Chapter 2, for unsteady flow computation in open channels, the
roughness coefficient(s) as embedded in the momentum equation cannot be measured
directly and therefore also needs to be estimated. However, the methods of estimating
roughness coefficients for steady flow are not adequate for unsteady flow cases. In
these cases, modelling methods should be applied by either using a trial and error
by an automatic optimisation approac
the roughness identification problem or the inverse problem and forms the focus of
ation problem is discussed.
Secondly, the theoretical background and the model development of the roughness
4.2 General discussion
A mathematical model generally comprises three types of data: physical system data,
input data, and output data. In a model for unsteady flows in open channels, the
approach or h. The later approach is known as
this chapter. Firstly, an overview on the roughness identific
identification problem for unsteady flow in open channels are briefly reviewed,
including hydraulic models and optimisation models. Finally, an appropriate model for
identifying roughness coefficients extended for compound channels is developed. This
model will be used as a tool to investigate some factors affecting the quality of the
identified roughness coefficients and applied to some natural rivers in the following
chapters.
73
Chapter 4 Roughness Identification Model Development for Unsteady Flow
phy
such as cross-sections data, slope and structures. The input data are represented by the
init Finally, the output data is represented by the simulated
hyd ge (1975) classified problems
rela
oblem Input System Output
sical system is represented by its roughness features and geometric configurations
ial and boundary conditions.
rographs or flow profiles along the channels. Cun
ted to the mathematical model into three categories, as shown in Table 4.1.
Table 4.1 Classification of mathematical models (after Cunge (1975))
Type of pr
Forward problem
(Simulation model)
Given Given Unknown
Inverse problem
Given Unknown Given
(Identification model)
Reverse problem
(Detection model)
Unknown Given Given

The forward problem is the one most often found in open channel flow hydraulics
which is known as flood routing or simulation problems. In contrast, the reverse
problem is seldom encountered in open channel flow problems except for the gate
1978; Falvey
produce these observed hydrographs. This procedure
atic optimisation approaches.
By finding the unknown parameters the system is calibrated or identified. The trial and
stroking problem introduced by Wylie (1969) (e.g. Bodley and Wylie
1987; Falvey and Lunning 1979; Wylie 1969; Wylie et al. 1993) and reverse level pool
routing proposed by Zoppou (1998).
The inverse problem is also referred to as the calibration problem (Cunge 1975). In this
case, the models determine unknown physical parameters. For this problem, the input
data and observation hydrograph(s) (which is the output data) are known. One
calibrates the model trying to re
can be done by using either a trial and error or an autom
error method is based on visual comparisons between the computed and observed
hydrographs to estimate the parameters. This approach suffers from subjectivity and
inefficiency. To overcome these problems, automatic optimisation methods may be
applied. This approach involves a system iterative procedure to identify the optimum
values of the parameters by minimising a chosen objective function. Khatibi et al.
(1997) refers it as the parameter identification problem. For the consideration of
74
Chapter 4 Roughness Identification Model Development for Unsteady Flow
estimating the roughness coefficients for unsteady flow, this approach is known as the
roughness identification problem and the roughness coefficients obtained are called the
identified roughness coefficients.
Figure 4.1 illustrates the roughness identification procedure. The procedure starts with
initial estimated parameters and performs a complete simulation run. The objective
function is evaluated by comparing the observed data against the simulated ones by the

model. If the value of the function is above the prescribed tolerance value, the process
is continued iteratively through computing a correction to the parameters by using an
optimisation algorithm.





Initial roughness values
START







Figure 4.1 Roughness identification procedure
HYDRAULIC
MODEL
Simulated data
Objective
function value
OPTIMISATION MODEL
Required accuracy
reached
Input data
Yes
No
N
e
w

i
t
e
r
a
t
i
o
n

Observed data
Identified roughness values
STOP
Fitting roughness
values
75
Chapter 4 Roughness Identification Model Development for Unsteady Flow
From this figure it can be seen that the roughness identification model includes two
main components: a hydraulic sub-model and an optimisation sub-model. In the next
sections, firstly, the hydraulic modelling methods to solve the Saint-Venant equations
are briefly reviewed and evaluated and a chosen method to develop the hydraulic sub-
model for compound channels is described. The variation in the value of the roughness
coefficient with stage is also considered. An alternative method of using conveyance
function is suggested and described. Secondly, for the optimisation sub-model, the
objective functions and the automatic methods are briefly reviewed and then an
appropriate objective function and an optimisation algorithm are adopted for the
optimisation sub-model of the roughness identification model.
4.3 Hydraulic sub-model
4.3.1 Governing equations of 1-D unsteady flow in open channels
The unsteady one-dimensional open-channel flow equations can be derived from the
principles of conservation of mass and momentum resulting in equations known as the
Saint-Venant equations. These equations written in terms of stage and discharge as
dependent variables are as follows:
q
x
Q
t
Z
B =

(4.1)
0 2
0 2
2
2
2
= +

q u gAS S
A
B Q
x
Z
A
B Q
gA
x
Q
A
Q
t
Q
q f
(4.2)
where A is the wetted cross-sectional area; Q is the discharge; Z is the water stage or
surface water elevation; q is the lateral inflow per unit length of channel; B is the
channel width at the surface water; is the momentum correction factor; g is the
gravitational acceleration;
0
S is the channel bed slope; S
f
is the friction slope, given by
the Manning equation
2
/ K Q Q S
f
= ; K is the conveyance of channel, n AR K /
3 / 2
= ; n
is the Mannings roughness coefficient; u
q
is the x direction velocity component of the
lateral inflow; x and t are the space and time variables respectively.
These equations are derived on the basis of the following assumptions:
76
Chapter 4 Roughness Identification Model Development for Unsteady Flow
a) The wave profile varies gradually and the hydrostatic pressure distribution
prevails,
b) The bed slope is small,
c) The velocity across the wetted area is uniform,
annel is approximately straight and the changes
numerical methods. In the next section, some common numerical methods for solving
rived.
and physically more straightforward and perhaps more direct and accurate
d) The longitudinal axis of the ch
in cross-section shape are gradual,
e) The resistance coefficient for steady uniform flow is considered applicable to
unsteady flow,
f) Body forces other than gravity are neglected, and
g) A horizontal transverse water surface profile is assumed.
The governing equations of one-dimensional unsteady flow in open channels are a pair
of first order of hyperbolic partial differential equations which are too complex to be
solved by analytical methods. It is possible, however, to find approximate solutions,
i.e. stages and discharges at a certain number of points in the time space domain using
Saint-Venant equations will be briefly reviewed.
4.3.2 Numerical methods for unsteady flow computation
In the current technical literature, there are several numerical methods for the solution
of hyperbolic partial differential equations (Chaudry 1993; Lai 1986; Cunge et al.
1980; Liggett and Cunge 1975). The methods most often used are characteristics, finite
difference and finite element. While the method of characteristics is based on the
characteristic form of the equations, the other ones are based on the partial differential
equations as originally de
4.3.2.1 The method of characteristics
The method of characteristic uses the total differential equations along the
characteristics which are transformed from the basic flow equations. In this method,
there are two main types of numerical schemes which are the Grid of Characteristics
(GC) and the Specified Time Interval (STI). These schemes are described in detail in
Abbott (1979), Chaudry (1993) and Liggett and Cunge (1975). The GC scheme is
theoretically
77
Chapter 4 Roughness Identification Model Development for Unsteady Flow
in many cases. In this scheme, however, both the original topography data and the final
er and Hamilton (1967) and Fang
(1968).
The STI scheme can avoid the interpolation of results at a number of (x,t) points. The
ethod was devised by Hartree (1952) and later made more widely known by Lister
(1960). With this scheme, the grid of points is predetermined and only dependent
ariables are to be computed at each unknown point. The scheme also requires
interpolation. The accuracy problems related to interpolation are analysed in Jolly and
Yevjevich (1971). Moreover, the STI scheme permits the use of the extrapolation
rocedures to increase the order of accuracy with little extra computational work. This
scheme has been applied to several rivers and estuaries by Lai (1967), Baltzer and Lai
968), Lai (1976), Schaffranek and Lai (1996); to the dam-break problem by (Chen
1980) and Chen and Ar d to flood routing by Yevjevich (1970),
Schmitz (1983), Lai (1988)).
l advantage of the characteristics methods is their c
easily with supercritical flow, which is, however, rarely encountered in river modelling
variable grid characteristic is often used. In practice, it is more convenient to use
ethods for flood o g based on finite diffe
in the section below.

t wide used for
solving Saint-Venant equations. Depending on finite difference schemes the methods
ethods, the finite difference e
variable which can be computed explicitly from the grid function at the previous time
results need to be interpolated. For this reason, the GC scheme is unsuitable and
inconvenient to use in the cases of non-prismatic or multiple-reach computations.
Hence, the application of GC scheme is not widely used. Some examples can be found
in the literature, such as Amein (1966), Fletch
m
v
p
(1
mbruster (1980) an
The principa apability of dealing
problems. A notable exception is the study of the dam-break problem for which the
numerical m r utin rence methods, as discussed
4.3.2.2 Finite difference methods
Numerical methods based on finite difference methods are the mos ly
can be categorised into two main classes: explicit and implicit finite difference
methods.
In explicit m quations contain only one unknown grid
78
Chapter 4 Roughness Identification Model Development for Unsteady Flow
level. The most popular explicit schemes include the Lax scheme the leap-frog
scheme. Another well-known explicit scheme with the second order of accuracy is the
me. These schemes are described in detail in Liggett an
(1975). They have been applied by several authors such as by Balloffet (1969) for the
Orinoco Delta and the Sinu River, by Dronkers (1969) for the Lek River, by Kamphuis
(1970) for the St. Lawrence River, by McDowell and Prandle (1972) for the Hooghly
River and by Davidson et al. (1978) for the Upper Delaware River Estuary system.
h the explicit schemes are simpler in solving the finite difference equations,
stability
considerations known as the Courant condition.
implicit finite difference schemes, the r ulting finite difference equations cannot be
solved directly for unknown grid variables on the forward time level, and the system of
fference
methods are their unconditional stability and no restrictions on the computational time
step due to mathematical stability required by explicit schemes. There are several
plicit schemes that are described in Cunge et al. (1980), Liggett
and Lai (1986), such as Preissmanns scheme (Preissmann 1961), Abbott-Ionescus
scheme (Abbott and Ionescu 1967), Vasilevs scheme (Vasiliev et al. 1965). The
application of implicit finite difference schemes is widespread. For example, Dronkers
ed them to the Lek River, Am
Kamphuis (1970) to the St. Lawrence River, Chaudry and Contractor (1973) for
simulation of a surge to the Smith River, Fread and Smith (1978) to the Mississippi
iver, Zoppou (1984) to five natural rivers: the Chattahooc
the Yarra River, the Mitta Mitta River and the Neuse River, Nguyen and Kawano
(1995) to the Red River network in Vietnam, Schaffranek and Lai (1996) to Threemile
Slough, Hsu et al. (2003) for flash flood routing in the Tanshui River in Taiwan.
Many studies have shown the advantages of implicit finite difference schemes (e.g. Jin
schemes: Preissmann, Abbott-Ionescu and Valisiev. Their study showed that the first
Lax-Wendroff sche d Cunge
Althoug
they are all restricted in computational time steps because of mathematical
In es
equations must be solved. The main advantages of the implicit finite di
im and Cunge (1975)
(1969) appli ein and Fang (1970) to the Neuse River,
R hee River, the Mary River,
and Fread 1997; Price 1974; Amein and Fang 1970). The main advantages of these
schemes are that they are unconditionally stable and flexible and can handle large
variations in channel geometry. Skeels and Samuels (1989) examined the practical
limits for stability and both amplitude and the phase errors for three numerical
79
Chapter 4 Roughness Identification Model Development for Unsteady Flow
two schemes are both superior to the last one and when the Froude number increases
then Preissmanns scheme is eferable. Fread (1974) also indicated that this scheme
has considerable flexibility in the choice of space and time step.
pr
4.3.2.3 Finite element methods
ethods developed originall
their application spread to many other engineering fields including unsteady open-
s. In the finite element methods, the solution of the dependent
s differential equations is approximated by some continuous functions with
unknown coefficients. The weighted residual techniques are applied to force the
residual or error to zero, produced by a set of approximate functions, and solved for the
unknown interpolation coefficient. For non-linear problems the Galerkin weighted
residual technique (Cooley and Moin 1976; Smith and Cheng 1976) is the most often
employed, where the set of weighting coefficients is the same as the set of
interpolation functions. Finite differences are used to discretise the time derivatives.
The finite element methods are generally considered to be suitable in handling
irregular shapes. However, because most schemes use finite differences for the time
derivatives, its application to the time dependent problem is not always clear.
Moreover, at this stage they do not offer any major advantage as compared with other
numerical methods for dealing with one-dimensional flow problems (Chaudry 1993;
Lai 1986). The application of finite element techniques in solving the Saint-Venant
equations is a relatively recent development and only a few cases have been reported
nd o
flow in p t nit
section.
Finite element m y in the structural engineering field. Then
channel flow problem
variable in
(e.g. Sen and Garg 1998; Katopodes 1984; Watt et al. 1980; Cooley a M in 1976;
Smith and Cheng 1976).
From this brief overview on numerical modelling methods for one-dimensional
unsteady open channels, the im lici fi e difference methods show their
advantages over the other methods, of which Preissmanns scheme has proven to be
efficient and superior in terms of the stability, accuracy and flexibility in the choice of
space and time step. Therefore, in this study Preissmanns scheme is adopted to solve
the Saint-Venant equations. The detail of the numerical method is described in the next
80
Chapter 4 Roughness Identification Model Development for Unsteady Flow
4.3.3 Implicit Pressmanns scheme
This section describes Preissmanns scheme for solving a finite difference equation.


Figure 4.2 and Equations (4.3a) to (4.3c) below show the actual discretisation of
dependent variables f and its derivatives with respect to time t and space x according
to the Preissmanns scheme at point M.

t
1 + j
t ) 1 (


M
t
2 / x
j
2 / x
i 1 + i x
Figure 4.2 Numerical Pressmanns scheme
) ( )]
j
i
j
i
j
f f + +
+
+
1
1
) 1 ( ) (4.3a) ( [
i
j
i
f f t x f + =
+
+
1
1
2
1
) , (


x
f f
x
f f
x
f
j
i
j
i
j
i
j
i

+
+ +
+ 1
1 1
1
) 1 ( (4.3b)

t
f f f f
t
f
i i i i

+ +
2
1 1

j j j j
+
+ + 1 1
(4.3c)
d out tests to investigate the effects of
where is a weighting coefficient. Liggett and Cunge (1975) and Quinn and Wylie
(1972) carrie a
schem is numerically stable when
nd indicated that the difference
e 1 5 . 0 and it is always unstable when .
Replacing the variables and their partial derivatives with the above formulas for the
governing Equations (4.1) and (4.2) gives
5 . 0 <
81
Chapter 4 Roughness Identification Model Development for Unsteady Flow
( ) ( )
0 ) 1 (
2
1
1 1
1

+
Q

1
1
1
1
=

+
+ + +
+
+
+

q
x
Q Q
x
Q
t
Z Z Z Z
B
j
i
j
i
j
i
j
i
j
i
j
i
j
i
j
i
(4.4)
( ) ( ) ( )

( ) ( )
( ) 0
) 1 (
2
2
1
1 1
1
2
=

+
+ +
+

) 1 (
2
2
1
1 1
1 1
1 1
1
+
+

+
+
+ +
+ +
+ +
0
2 2



S
A x A
gA (4.5)

+ gA
+
q u
K
Q Q
B Q Z Z Z Z B Q
x A
q
j
i
j
i
j
i
j
i
Q Q Q Q Q
t
Q Q Q Q
j
i
j
i
j
i
j
i
j
i
j
i
j
i
j
i
where:
( ) ( )
j
i
j
i
j
i
j
i
B B B B B =

2
; +

+ +

+
+ +
+ 1
1 1
1
2
1

( ) ( )
j
i
j
i
j
i i
q q q +

+ +
+
+
+ 1
1
1
2
1
2
;
j
q q

=
+1

+ =


+ +

+


+
+
+
+
+
+
j
i
j
i
j
i
j
i
j
i
j
i
A
Q Q Q Q Q
1
1
1 1
1
;

j
i
j
i
A A A A
1 1
1
) 1 ( 2

;
2
1
2
2
1
2
1
1

2 2
1
2
2
1
2
1 1
1
2
2


+ +
+ +
+
j j
j
i
j
i
Q
g
B Q B Q
gA gA
B Q
gA
+

+


+
+
+
j
i
j
i
j
i
j
i
i i
A
B
A
B Q
A gA
A A A

+ i i
A
2
1
2

+ +
+
j j j
i
j
i
B Q
A
B Q
A
B Q
A
B Q
A
B Q
2 2
1
2
2
1
1
2
2
2
2
2
1
2
;

;
2
2 2
1

+
j
i
j
i
K
gA
K
gA
1
2
1
1
2
1
1
1
2
1
1
2

+
+
+
+
+
+
+

j
i
j
i
j
j
i
j
i
j
i
Q Q Q Q
Q Q
gA
K
gA gA

Q Q Q Q


i
K K

82
Chapter 4 Roughness Identification Model Development for Unsteady Flow
( ) ( ) ( ) ( ) ( ) ( ) ( )
j j
q u q u q u +
i
q
i
q
j
i
q
j
i
q q
q u q u

+ +

=
+
+ +
+ 1
1 1
1
2
1
2
The finite difference approximation leads to a system of two non-linear algebraic
term of , , for r
points (nodes), there are N segments and there will be a system of 2N such equations
di l
term of unknowns ,
.
equations in eve y segment. For N+1 computation
1 + j
i
Z ,
1 + j
i
Q
1
1
+
+
j
i
Z
1
1
+
+
j
i
Q
for 2(N+1) unknowns. With the ad tiona of two boundary conditions the number of
equations is equal to the number of unknowns. The equations are first linearised in
1 + j 1 + j
i
Z
i
Q , = i 1, 2,,N in a manner similar to the method used by
4) for the iteration process. For t i
th
be written as follows:
Nguyen (199 he segment the linearised system may

i
j
i i
j
i i
j
i i
j
i i
e Q d Z c Q b Z a = + + +
+
+
+
+
Q
+ + 1
1
1
1
1 1
(4.6)

' 1
1
1
1
1 1
i
j
i i
j
i i
j
i i i
e Q d Z c b Z a = + + +
+ j
i +
+
+
+
(4.7)
+

where

= = B
t
x
c a
i i
2
;
1 = =
i i
d b ;
( ) ( ) x q Q Q Z Z
t
e
j
i
j
i
j
i
j
i i

+ +

=
+ +

2
1 1
;
x B 1 1
x A
gA c a
i i

= =

2
;
B Q
2
1
2
1
2
2
2
1
+
+

=
j
i
j
i i
K
Q
gA
x A
Q
t
b ;

1
1
2
1
1
2
2
2
1

=
i
A
Q
t
d
+
+
+
+

j
i
j
i
K
Q
gA
x
;
83
Chapter 4 Roughness Identification Model Development for Unsteady Flow
( ) ( ) ( )
. ) (
2 2

+

2
1
.
1
1 0
2
2
+
+


) 1 ( ) 1 (
2
2
1
2
2
1 1 + +

+
=
Z Z
A
B Q
gA
x
Q Q
A
Q
t
Q
e
j
i
j
i
j
i
j
i
j
i
j
i
i

K
Q Q
gA
K
Q Q
gA S
A
B Q
x
Q
j
i
j
i
j
i
j
i

In the first iteration the values are set to known values of , and the
uted. Then, the linearised system of Equations
(4.6) and (4.7) can be solved to give a second approximation of the unknowns at the
new tim j ( ,
+ q u
q
1 + j
i
Z ,
1 + j
i
Q
j
i
Z
j
i
Q
coefficients
i
a ,
i
b ,
i
c ,
i
d are comp
e t + ) 1 , and coefficients
i
a ,
i
b
i
c ,
i
d are then updated and the linear
u ishing the third approximation. Nguyen (1994)
indicated that to improve the convergence of iteration procedure the next
me step
system can be solved again, f rn
approximation of unknowns at the new ti t j + ) 1 ( can be taken as the average
value of two successive iteration values. He showed that in most cases three iterations
t, so that there is no need for further iterations
technique was adopted.
spare matrix
that is amenable for fast computer treatment. Two methods which have been shown to
rred by many investigators ar
method and Newtons iteration method (Newton-Raphson method) (Lai 1986). In this
research, the double sweep method was adopted.
4.3.3.1 Solution algorithm
The double sweep algorithm applies only to subcritical flows. The algorithm applied in
is similar t
are sufficien . Therefore, in this study this
As can be seen the system of Equations (4.6) and (4.7) lead to a banded
be very efficient and reliable and prefe e the double sweep
this research o the algorithm applied in Nguyen (1994). For simplicity, the
superscripts of unknowns at time t j + ) 1 ( will be dropped from the following
analysis. The algorithm for a single reach with (N+1) nodes and N segments is
described below.
For each segment, there are two equations with four unknowns as given in Equations
tions for the i
th
segment gives: (4.6) and (4.7). Rewriting these equa
Q b Z a
i Di i Di i Ui i Ui i
e Q d Z c = + + + (4.8)
84
Chapter 4 Roughness Identification Model Development for Unsteady Flow

'
i Di i
e Z c = + (4.9)
i Di Ui i Ui i
Q d Q b Z a + +
e the upstream and downstream of a segment.
ns for any segment gives:
Z H Q
where subscripts U and D denot
Adding and subtracting these two equatio

i Di i
J Z I (4.10)
Ui i Ui
= +

i Di i Ui i Di
G Z M Z L Q + = (4.11)
where:
i i i i
i i i i
i
d b b d
d a a d
H


= ;
i i i
d b
i
i i i
i
b d
d c c d
I


= ;
i i i i
i
d b b d
i i i i
d e e d
J

=
i i i i
i i i i
i
d b b d
b a a b
L


= ;
i i i i
i i i i
i
d b b d
b c c b
M


= ;
i i i i
i i i i
i
d b b d
b e e b
G


=
At any node the compatibility condition must be satisfied by the continuity condition
and the common water level.
At the i
th
node between the segment (i-1)
th
and segment i
th
of a single channel we have:
Ui Di
Q Q =
1
(4.12)
1 1
=
i Ui
Z Z ,
i Ui Di
Z Z Z = =
1
Substituting Equations (4.10) and (4.11) with and
Z Z into Equation (4.12) we can obtain:
1 + i Di

i i i i i i i i i
J G Z I Z H M Z L
=
+ = + + +
+ 1 1 1 1 1
) ( (4.13)
or we can write:

i i i i i i i
D Z C Z B Z A = + +
+ 1 1
(4.14)
These equations contain only Z variables where:
85
Chapter 4 Roughness Identification Model Development for Unsteady Flow
1
=
i i
L A ;
i i i
H M B + =
1
;
i i
I C = ;
i i i
J G D + =
1

Hence, for single channels with one boundary condition given at each end, there is
actually a system of ) 1 ( + N equations for ) 1 ( + N unknown Z (see Equation (4.14)).
The coefficient matrices contain elements only in the tridiagonal band along the main
diagonal. This system of linear algebraic equations must be solved for all
computational points for every time step during the period of computation. In this
model, the double sweep method is used to solve this system.
Introducing the auxiliary relation

i i i i
F Z E Z + =
1
(4.15)
and substituting it into Equation (4.14) for grid point i , one obtains an expression
relating
i
Z and
1 + i
Z

i i i
i i i i i
i
E A B
F A D Z C
Z
+
+
=
+
) (
1
(4.16)
By analogy with Equation (4.15) the recurrence relationships can be obtained

i i i
i
i
E A B
C
E
+

=
+1
;
i i i
i i i
i
E A B
F A D
F
+

=
+1
(4.17)
To initialise a forward sweep, the
2
E and
2
F values are evaluated from the equivalent
of Equation (4.17) and the first equation of the Equation (4.14)

1
1
2
B
C
E = ;
1
1
2
B
D
F = (4.18)
where
1
B ,
1
C and
1
D are determined depending on different types of boundary
condition discussed later in this section. The forward sweep now can be carried on to
compute coefficients
i
E and
i
F for = i 3, 4, until the downstream boundary
1 + = N i by applying the recurrence relations of Equation (4.17). At the last point, by
86
Chapter 4 Roughness Identification Model Development for Unsteady Flow
solving Equation (4.15) together with the last equation of the system (4.14) the value
of
1 + N
Z can be obtained by eliminating
N
Z :

1 1 1
1 1 1
1
+ + +
+ + +
+
+

=
N N N
N N N
N
E A B
F A D
Z (4.19)
Then a return sweep is made by using Equation (4.16) to compute
N
Z ,
1 N
Z
1
Z .
Then the discharges are computed from Equations (4.10) and (4.11).
As mentioned before, the double sweep algorithm can be applied only for subcritical
flow. For this case, each boundary requires one external boundary condition. There are
three types of boundary conditions which are (i) ) (t Z Z = , (ii) ) (t Q Q = , and (iii)
) (Z Q Q = . The coefficients A, B, C and D at boundary nodes are formulated
depending on the type of boundary conditions as follows:
At the upstream boundary
(i) a discharge hydrograph is given, i.e. ) (
1 1
t Q Q
U
=
Substituting
1 ) 1 (
Q Q
U
= in Equation (4.10) and comparing it with Equation (4.14) give
the formulas to calculate the coefficients at the upstream boundary:
0
1
= A ;
1 1
H B = ;
1 1
I C = ;
1 1 1
Q J D = (4.20)
(ii) a stage hydrograph is given, i.e. ) (
1 1
t Z Z
U
=
Comparing with Equation (4.14) gives
0
1
= A ; 1
1
= B ; 0
1
= C ;
1 1
Z D = (4.21)
(iii) a rating curve is known, i.e. ) (
1 1
Z f Q
U
=
Linearising the boundary condition then
87
Chapter 4 Roughness Identification Model Development for Unsteady Flow
( )
j j
U
Z Z
dZ
df
Z f Q
1 1
1
1 1
) ( + = (4.22)
Equating Equation (4.22) and Equation (4.10) and comparing it with Equation (4.14)
gives the formulas to calculate the coefficients at the upstream boundary:
0
1
= A ;
1
1 1
dZ
df
H B + = ;
1 1
I C = ;
j j
Z
dZ
df
Z f J D
1
1
1 1 1
) ( + = (4.23)
At the downstream boundary
(i) a discharge hydrograph is given, i.e. ) (
1
t Q Q
N DN +
=
Substituting
1 +
=
N DN
Q Q in Equation (4.11) and comparing it with Equation (4.14) we
can have the formulas to calculate the coefficients at the downstream boundary as
follows:

N N
L A =
+1
;
N N
M B =
+1
; 0
1
=
+ N
C ;
N N N
G Q D + =
+ + 1 1
(4.24)
(ii) a stage hydrograph is given, i.e. ) (
1 1
t Z Z
U
=
Comparing with Equation (4.14) gives
0
1
=
+ N
A ; 1
1
=
+ N
B ; 0
1
=
+ N
C ;
1 1 + +
=
N N
Z D (4.25)
(iii) a rating curve is given, i.e. ) (
1 +
=
N DN
Z f Q
Linearising the boundary condition then
( )
j
N N
N
j
N DN
Z Z
dZ
df
Z f Q
1 1
1
1
) (
+ +
+
+
+ = (4.26)
Equating Equation (4.26) and Equation (4.11) and comparing it with Equation (4.14)
gives the formulas to calculate the coefficients at the downstream boundary:
88
Chapter 4 Roughness Identification Model Development for Unsteady Flow
N N
L A =
+1
;
1
1
+
+
=
N N
df
M B ; C 0
1
=
+ N
dZ
N
; and

j
N
N
j
N N N
dZ
df
1
1
1 1 +
+
+ +
+ =
In natural rivers, many rivers have a looped or branched network and structures such as
weirs, bridges and sluices. A system of algebraic linear equations is developed for the
river network also using the segm
Z Z f G D ) ( (4.27)
ent equations (i.e. Equations (4.10) and (4.11)) and
the compatible condition at a node. The resulting equation has unknowns of only the
f
sparse. The algorithm to solve this equation system is described in Appendix B.
r details about numerical modelling for flow in compound channe
discussed in the next section.
uting in natural rivers, many channels have compound sections and the
values of the roughness coefficient in the main channel and the floodplains are usually
e more accurate results, the roughness coefficients of th
channel and floodplains should be considered separately. However, studies on the
ction is to select and develop a one-dimensional
(1D) model for compound channels that is appropriate for the roughness identification
4.3.4.1 One-dimensional models for flood routing in compound channels
ional models for flood routing in compo
single channel models, off-stream models, compound (or composite) channel models,
models is in the treatment of flow between floodplains and a main channel. In this
water level at each node. The coefficient matrix o this linear equation system is very
Some furthe ls are
4.3.4 Compound channel models
In flood ro
different. In order to hav e main
inverse problem of estimating the roughness coefficients in compound channels are
still limited. The purpose of this se
problem. The subsection below briefly reviews current floodplain models then the
selected model for compound channels is described.
One-dimens und channels can be classified into
and separate channel models (Zoppou 1984). The basic difference between these
section, some main features of these models are briefly reviewed and from that an
89
Chapter 4 Roughness Identification Model Development for Unsteady Flow
appropriate d for the roughness identification problem floodplain model will be selecte
for compound channels.
oach, there is no distinction between the main channel and
the floodplain. The assumptions of these models are that the water surface across the
fferent. For the single channel approach,
ulae for computation of the
Section 2.5 in Chapter 2. This
ed with the main channel velocity. For
these models, only the roughness coefficient of the main channel is considered. This
researchers (e.g. Rashid and Chaudhry

time than that experienced by the observed flows. For these cases, the compound
For the single channel appr
whole cross-section is horizontal and the mean velocity occurs in both the main
channel and the floodplains. Generally, the roughness coefficient values in the main
channel and the floodplain are significantly di
the equivalent roughness coefficient is used. Several form
equivalent roughness coefficient were discussed in
technique was used by several authors (e.g. Rashid and Chaudhry 1995; Zoppou 1984;
Fread 1976; Wiggert 1976). Zoppou (1984) indicated that if the value of the roughness
coefficient is formulated as a constant or a simple function of depth the simulated
discharge will produce apparent drop when there is an abrupt change in water surface
width. The study of Fread (1976) also showed that this approach could encounter
computational problems caused by a sudden change in water surface width at the
bankfull stage. Hence, this method is appropriate only for rivers with no abrupt
changes in the width of cross-sections.
Offstream models are appropriate for rivers only when the flow velocity in the
floodplain is restricted by vegetation or large obstacles. For these cases, the floodplain
velocity is considered to be negligible compar
approach has been used successfully by many
1995; Fread 1976; Ackermann and Shi-igai 1976; Tingsanchali and Ackermann 1976;
Kamphuis 1970; Balloffet 1969; Dronkers 1969; Baltzer and Lai 1968). However,
these models are not appropriate to apply to cases when the flow velocity in the
floodplain is considerable compared with the main channel velocity or the floodplains
have large cross-sectional areas that contribute to the significant conveyance to the
channel system. As a result, the computed flows have greater attenuation and travel
channel models may be more appropriate for the simulation of floodplain flows.
90
Chapter 4 Roughness Identification Model Development for Unsteady Flow
Compound channel models assume that the friction slope is common to main channel
and floodplain flows, that the water surface is horizontal across any cross-section of
the channel and that the total discharge can be distributed among the main channel and
overbank flow in proportion to their individual conveyances. The conveyance of a
compound section is computed using the divided section method in which for any
depth the conveyance is the sum of the main channel and floodplain conveyances. For
iation
several researchers such as
Gradowczyk and Ivanissevich (1971), Fread (1976), Ackermann and Shi-igai (1976),
is that how to avoid the computational instabilities
which would arise with very small depths and dry bed condition. The solutions for
een these approaches have been made in the literature. Fread
(1976) compared the single channel model, the composite model and the offstream
ith no sinuosity. He indicated that the
offstream storage model produced the greatest attenuation and the greatest travel time
odels gave the similar results. Bousmar and Zech (1999) also
these models, the knowledge of both the roughness coefficients of the main channel
and the floodplain are required. The procedure is only different from the single channel
approach with an equivalent roughness coefficient by considering the var of
velocity distribution across the compound section by the momentum correction
coefficient. This approach has been applied by
Zoppou (1984) and Bousmar and Zech (1999).
When the assumption of a horizontal water surface is violated separate channel models
may be an alternative approach to improve the accuracy of flood routing. These
models treat the main channel and floodplain as separate parallel channels and the flow
in each channel is routed individually. As a result, the water levels between these
channels will usually be different. This difference in water surface elevation is
distributed between channels as lateral flows. The method by which this distribution is
carried out is the distinguishing feature between these models. For example, the
EMBER model was developed by Hydraulic Research, Wallingford (Samuels 1979)
and the USDALL-74 model was developed by Yen (1974) or the models were
proposed by Grijsen and Meijer (1980) and Radojkovic (1976). Separate channel
models usually suffer from two main difficulties. The first one is how to apportion the
boundary conditions. The second
these problems were discussed by Zoppou (1984).
Some comparisons betw
storage model for a hypothetical channel w
while the other two m
91
Chapter 4 Roughness Identification Model Development for Unsteady Flow
made a comparison between a single channel model and a compound channel model.
Their results showed that the compound channel model can reproduce appropriate
erforms better than the single c
the assumptions implied in offstream storage model were found to be inappropriate for
As mentioned above, the main purpose of this chapter is to develop an appropriate
e conveyances of these channels (Yen et al. 1974) which

d need to be identified. This
s of the model are that (i) the friction slope is the same for all channels,
results and p hannel model. The study of Zoppou (1984)
about floodplain models applied for the Mitta Mitta and Neuse Rivers indicated that
the simulation of floodplain flows in those rivers. He also showed that the instabilities
of the single channel model caused by abrupt changes in the channel width can be
overcome by using the compound channel models.
model to identify the roughness coefficients in compound channels where not only the
value of the main channel roughness coefficient but also of the floodplain one needs to
be identified. Of all the above approaches, only the compound channel approach and
separate channel approach can consider these roughness coefficients separately.
However, for separate channel models, one of the main difficulties is how to apportion
the inflow discharge to the main channel and the floodplain channels. Usually, this
distribution is based on th
depend on their roughness coefficient values. However, for the roughness
identification problem, these values are unknown an
indicates that the compound channel approach is the most appropriate one to develop a
hydraulic sub-model for the roughness identification problem. Moreover, in natural
rivers, some parts of the floodplain at some segments of a river reach actually act as
storage areas. Therefore, in order to develop a more general model for flood routing in
compound channels in this study, a compound channel model combined with offstream
storages (if any) will be adopted.
4.3.4.2 The chosen compound channel model
The model considers the main channel and floodplain as separate channels. The
assumption
f
S
(ii) the flow in each channel is one-dimensional, and (iii) the water surface across the
whole cross-section is horizontal. The difference in the values of the roughness
coefficients between the main channel and floodplains are taken into consideration the
92
Chapter 4 Roughness Identification Model Development for Unsteady Flow
conveyance of the compound section together with using the momentum correction
coefficient.
The principal assumption in this technique is that
f
S is constant for all channels, then

( )
2 2 2
K
Q Q
K
Q Q
K
Q Q
S
i
i
i i
f
= = =

(4.28)
where Q
i
and Q are the discharge of sub-section i and the total discharge of the
section,
i
K is the conveyance of subsection i for the main channel and floodplains
defined as

3 / 2
1
i i
i
i
R A
n
K = (4.29)
where , and are the area, hydraulic radius and Mannings roughness
i
A
i
R
i
n
coefficient of subsection i respectively.
Applying the continuity equation with storage (if any) and the momentum equation for
each channel and summing up for the whole cross-section, the resulting continuity and
momentum equations for the compound channels system becomes
q
x
Q
t
Z
B
s
=

(4.30)

( )
0 2
2
2
2
2
= +

2 0


q u
Q Q
gA S
A
B Q
x
Z
A
B Q
gA
x
Q
A
Q Q
(4.31)
K
t
q
i
where
s
is the water surface width of the inundated area, B B is the water surface of the
flow, s the momentum correction factor defined as i
( )


= =

=
i i
Q QV

i
i
i
i i
A
K
K
A
V
V Q V Q
2
2
(4.32)
93
Chapter 4 Roughness Identification Model Development for Unsteady Flow
where is the mass density,
i
V and V are the velocities of sub-cross-section and whole
cross-section respectively.
Equations (4.30) and (4.31) can be solved for the unknowns Q and Z at all the cross-
sections (nodes) in the channel using the implicit finite difference scheme and double
tion 4.3.3.
Judy 1985; Higginson and Johnston 1991; Julien et al.
2002). For these cases, a variation in the roughness coefficient is preferable or
ness coefficient in the main
channel has been considered. In this section, the roughness identification problem is
the cases where the values of Manning's n var
both single channels and compound channels.
The variation of the roughness coefficient with flow can be approximately expressed
atical function of stage or dis
o
n decrease the number of unknowns
for the identification problem and (iii) it can extrapolate the value of the roughness
ence above the observed flow.
The variation curves of Manning's n with stage or discharge as studied by Robbins
s on or power
function of flow depth or discharge is likely to be appropriate to represent its variation.
sweep algorithm described in Sec
4.3.5 Identification of roughness functions
In some reaches, if the roughness characteristics of bed and banks are regular, the
value of roughness coefficient may vary little at all stages. The use of a constant n for
these cases appears to be satisfactory. However, a number of studies have been shown
that in several rivers, the values of n vary considerably with stages/discharges as
discussed in Chapter 2 (e.g. Butler et al. 1978; Robbins 1976; Sargent 1979; Richards
and Hollis 1980; Nalluri and
necessary. However, much of previous work in the roughness identification problem
has dealt with constant parameter and only the rough
extended for y with stages/discharges for
4.3.5.1 Choice of roughness functions
as a mathem charge. The advantages of expressing the
variation of roughness coefficient as a function are that (i) it is convenient t
incorporate it into a computer program, (ii) it ca
coefficient with a certain confid
(1976), Sargent (1979), Butler et al. (1978), Richards and Hollis (1980) and
Higginson and Johnston (1991) show that econd order polynomial functi
94
Chapter 4 Roughness Identification Model Development for Unsteady Flow
Lai (1986) based on the field data of the U.S. Geological Survey (USGS) also
indicated that quadratic polynomial function is an appropriate function that can
represent the variation of the roughness coefficient with flow.
Regarding the selection of the roughness function type, Zoppou (1984) chose a
putation. Moreover, the power function suffers
a mathematical and physical problem when stage reaches bankfull. Accordingly, the
oefficient:
roughness function as a power function of flow depth y (
b
ay n = ). Jun et al. (2005)
expressed the roughness coefficient as a power function of discharge Q (
b
aQ n = ).
Baltzer and Lai (1968) and Lai et al. (1978) treated the roughness coefficient as a
second order of polynomial function of discharge (
2
cQ bQ a n + + = ).
In this study, the variation of roughness coefficient with flow was chosen as a
quadratic polynomial function of the water stage. This type of function was chosen
because for unsteady flow, a rating curve is a loop rating curve. For a given discharge,
there are two different values of water stage corresponding to raising and falling
phases. Thus, the value of roughness coefficient treated as stage dependent is more
appropriate than discharge dependent especially for compound channel cases, because
for these cases, with a certain discharge the floodplain part may or may not be
inundated corresponding to falling stage or raising stage. As a result, the floodplain
may or may not be included in the com
depth on the flood plain is equal to zero and hence the value of the floodplain
roughness coefficient at this stage could become zero or infinite depending on whether
the value of power is positive or negative respectively. Therefore, using a polynomial
function can avoid this problem.
The roughness coefficients of the main channel and floodplains are formulated as two
separate polynomial functions of stage, one for main channel and the other for
floodplains as follows:
For the main channel roughness c

2
0 2 0 1 0
) ( ) ( Z Z a Z Z a a n
c
+ + = (4.33)
For the floodplain roughness coefficient:
95
Chapter 4 Roughness Identification Model Development for Unsteady Flow

2
2 1 0
) ( ) (
f f f
Z Z b Z Z b b n + + = (4.34)
where
c
n and
f
n are the roughness coefficients of the main channel and of the
floodplains respectively, Z is the water stage,
0
Z is the minimum water level or bed
level at a certain cross-section at which the cross-section characteristics are initially
tabulated in the input data, is elevation of the floodplains, and
f
Z
1 0 2 1 0
, , , , b b a a a and
a compound channel are
presented as two separate quadratic roughness functions, the number of unknowns
2
b are coefficients of the roughness functions that need to be identified.
The hydraulic model was programmed for the general cases where the roughness
coefficients can be considered as a constant, first order or second order of a polynomial
function. For the case where the roughness coefficients of
becomes six. This will significantly increase the computation time. To overcome this
disadvantage, a conveyance function may be used. The subsection below will describe
this alternative approach.
4.3.6 Identification of conveyance function
The conveyance K can also represent friction effects of a channel. It is related to
ughness coefficients.
conveyance function could provide several advantages. Firstly, for this alternative
Manning's n and cross-section property (see Equations (4.28) and (4.29)). For
compound channels, the conveyance of the whole section is the sum of the main
channel and floodplain conveyances. Cunge et al. (1980) gave a simple example of the
variation of the roughness coefficient and the conveyance for a cross-section with
abrupt changes in width and roughness. This example indicated that while the
equivalent roughness value varies discontinuously, the variation of conveyance with
depth is much smoother. Therefore, it is plausible to test an alternative method of using
a single conveyance function instead of using two roughness functions for compound
channels.
It can be argued that using a conveyance function cannot directly and explicitly
identify the values of the main channel and floodplain ro
However, the main purpose of identifying roughness coefficients is to compute
unsteady flow. Thus, for compound channels with roughness variation, using a
96
Chapter 4 Roughness Identification Model Development for Unsteady Flow
method, only one function needs to be identified instead of two separate roughness
functions. Therefore, this can reduce the number of unknowns. As a result, the
computation time could be decreased considerably. Secondly, by using this alternative
method, there is no need to be overly concerned with the effect of momentum transfer
and the compensation of conveyances between the main channel and the floodplain,
because they are all summed in the total conveyance that will be identified by the
inverse model. Thirdly, for cross-sections with abrupt changes in roughness and width,
the corresponding conveyance function is still a continuous curve. Hence, an
extrapolation of the conveyance using a conveyance function may give a higher degree
of confidence compared to the case of using roughness functions.
Therefore, for this alternative approach, instead of identifying the coefficient of two
roughness functions, the coefficients of a conveyance function are identified. The
appropriate types of conveyance functions are described in the subsection below.
4.3.6.1 Choice of conveyance functions
Several types of conveyance functions were tested for compound cross-section with
abrupt change in width and non-uniform roughness. It was found that a cubic function
r function of depth are the m
variation of conveyance with depth for compound cross-section with roughness
(4.36)
and a powe ost appropriate functions to represent the
variation as follows:

3 2
) ( ) ( ) (
b b b
Z Z c Z Z b Z Z a K + + = (4.35)
b
b
Z Z a K ) ( =
where
b
Z is the bed elevation or the level that the discharge equal to zero (cease level),
and a, b and c (for the cubic function) or a and b (for the power function) are the
unknown coefficients of the conveyance functions that need to be identified.
4.4 Optimisation sub-model
This section continues to develop the roughness identification model. The model
includes the hydraulic sub-model developed in the previous sections and an
97
Chapter 4 Roughness Identification Model Development for Unsteady Flow
optimisation sub-model (see Figure 4.1 in Section 4.2). In the sub-sections below, the
objective functions and optimisation techniques used in literature for the inverse
problem in open channels are reviewed. Then an optimisation sub-model of the
identification roughness model is developed.
4.4.1 Objective functions
An objective function is a numerical criterion to measure the difference between the
model response and the observed prototype behaviour. This function is minimised by
optimisation methods to identify the model parameters. This is a collective measure of
individual errors, . For unsteady flow, an individual error is calculated by comparing
observed flow values with their corresponding simulated values in terms of discharge
(or velocity) and/or stage (or depth). There are a wide range of objective functions.
The most common ones based on the error criterion which are used in literature can be
classified into three groups:
Sum of least squares: According to this norm, the parameters of a model are obtained
by minimising the sum of square errors. It is defined as:
( ) =

= =
J
j
Sj Oj
J
j
j
Y Y
1
2
1
2
min min (4.37)
Sum of the absolute errors: According to this norm, the parameters of a model are
obtained by minimising the sum of absolute errors. It is defined as:
=

= =
J
j
Sj Oj
J
j
j
Y Y
1 1
min min (4.38)
Minimax: According to this norm, the parameters of a model are sought by
minimising the maximum error. It is defined as:


J J
Y Y max min max min (4.39)

= =
Sj Oj
j
j
j 1 1

where is the observed discharge or stage, is the simulated discharge or stage at
time j
th
, J is number of observation times,
Oj
Y
Sj
Y
is the weighting coefficient.
98
Chapter 4 Roughness Identification Model Development for Unsteady Flow
Several researchers used objective functions which are a combination of stage (or
depth) and discharge (or velocity) such as Becker and Yeh (1972), Wormleaton and
Karmegam (1984), Ding et al. (2004). Individual errors can also be relative errors
with respect to observed data or with respect to simulated one.
The weighting coefficients may be included in the objective functions (Bennett 1975;
Sorooshian and Dracup 1980). These weighting coefficients used have a significant
influence on the optimised parameters. However, the application of a weighting
scheme must not be in conflict with the theory of maximum likelihood (Sorooshian
and Dracup 1980).
Errors or noise are inevitable in mathematical modelling for flow computation in open
channels. The source of these errors comes from measurement and data processing
ctive function should be made carefully
The least squares technique is the best choice to obtain the least biased values of the
parameters when the distribution of the errors is normally distributed (Sun 1994). The
ordinary least square objective function minimises the sum of the difference between
stages that are contained in channel (cross-section and channel slope) data and flow
(stage and discharge) data. Also, in mathematical modelling the other error sources are
model errors and numerical errors. Model errors are associated with imperfections of
the governing equation and some restricted assumptions to simplify the physical
processes. Numerical errors include rounding errors and truncation errors related to the
finite different methods. These errors and noise result in the calculated flow values
differing from their true values.
In the absence of the above errors, minimisation of the objective function would lead
to the exact values of the parameters, after which the objective function would be zero.
However, because these errors are unavoidable, minimisation of the objective function
will lead to the identification of optimum values of the parameters. It is shown that the
selection of objective functions is one of the factors affecting the quality of identified
parameters. Theoretically, the choice of an obje
so that the distribution of residuals does not violate the statistical assumptions implied
in the objective function, otherwise biased parameter estimations may result
(Sorooshian and Dracup 1980).
99
Chapter 4 Roughness Identification Model Development for Unsteady Flow
the computed and observed values. The identified parameters will not be biased to the
true values if the assumptions implicit in this objective function are satisfied. These
it is only used when the observation errors are small or there is no
outlier. Moreover, Khatibi et al. (1997) argued that this criterion is not a good choice
ecause there is no evidence that the maximum error will occur at the peak flow. This
will affect the accuracy of the computed peak discharge or stage, which is very
portant in many design problems.
With regard to the choice of objective function for the inverse problem in open
channels, the sum of least squares has been used by many researchers (e.g. Becker and
Yeh 1972; Bennett 1975; Wu 1976; Davidson et al. 1978; Wormleaton and
Karmegam 1984; Zoppou 1984; Wasantha Lal 1995; Ramesh 2000; Atanov 1999. The
sum of the absolute difference has been used by Rao et al. (1976), Wiggert et al.
(1976), Fread and Smith (1978) and Zoppou (1984). Minimax has been used by Yeh
and Becker (1973), Wormleaton and Karmegam (1984), Chu and Yeh (1980) and
Wasantha Lal (1995). Usually, the reasons for the choice of a particular objective
function are not clear. They are mainly based on its ease of application and previous
experience (Zoppou 1984). The identified parameters are validated using a visual
comparison between the observed and simulated response only.
However, Khatibi et al. (1997) indicated that the selection of objective functions is one
of the factors affecting the quality of the identification problem. Nguyen and Fenton
assumptions include: (i) errors have zero mean and constant variance and (ii) errors are
mutually uncorrelated (Clarke 1973; Diskin and Simon 1977). When the errors are
autocorrelated the weighted/generalised least square technique has been shown to be
mathematically equivalent to the maximum likehood (Williams and Yeh 1983).
If the errors are distributed exponentially, the identified parameters will be equivalent
to the maximum likelihood estimate if the objective function is formulated by the sum
of the absolute errors. When outliers are suspected in observation data, using this
objective function the negative effects of outliers is not as much as the least square
objective function (Sun 1994 and Williams and Yeh 1983).
For the minimax technique, the maximum of absolute errors is minimised. Sun (1994)
indicated that
b
im
100
Chapter 4 Roughness Identification Model Development for Unsteady Flow
(2004) investigated the effect of three main types of objective function (minimising
sum of square errors, sum of absolute errors and maximum error) and showed that the
least square objective function had the best performance. Khatibi et al. (1997)
nt least square errors of: absolute errors, relative errors with
values, and relative errors with respect to simulated values. Their
study indicated that the least square of absolute errors provide the most appropriate
n the
identified parameter. Zoppou (1984) considered the violation of the assumptions of the
ariance assumption and autocorrelation
assumption for the Mary and Yarra Rivers. His study indicated that the ordinary least
square model produced similar or better results compared to the non-linear least square
model for the case studies despite the gross violations of the assumptions implied in
. (2000) used synthetic data containing auto-correlated data
a higher autocorrelation level the confidence
Automatic methods or optimisation methods are mathematical tools which can be
method with the sum of the absolute error objective function was used by Wiggert et
considered three differe
respect to observed
results while the others using relative errors seem to induce an undue bias i
least squares criterion both in constancy of v
the model. Khatibi et al
errors also confirmed that although at
interval widens, the least square of absolute errors is capable of producing acceptable
results at high autocorrelation levels. Therefore, the least square of errors between
observed and simulated stages/discharges is chosen in this study as follows:
( )

= =
=
J
j
Sj Oj
J
j
j
Y Y
1
2
1
2
min min (4.40)
4.4.2 Optimisation methods
applied to minimise a chosen objective function. Throughout this process, the
roughness coefficients or any other parameters can be identified.
The literature survey showed that a number of iterative automatic methods have
already implemented in the inverse problem for unsteady flow in open channels. For
example, the influence coefficient algorithm in conjunction with the least squares
approach was introduced by Becker and Yeh (1972, 1973) and the influence
coefficient algorithm in association with linear programming was used by Yeh and
Becker (1973) and Wormleaton and Karmegam (1984). The conjugate gradient
101
Chapter 4 Roughness Identification Model Development for Unsteady Flow
al. (1976). The modified Newton-Raphson technique was employed by Fread and
Smith (1978). Davidson et al. (1978), Wormleaton and Karmegam (1984), Khatibi et
al. (1997) and Jun et al. (2005) used a non-linear least square technique by a modified
ing a least square
erro
Program et al. (2000). The quasi-Newton method
(The
Gauss-Newton method also known as the Levenberg-Marquardt method. Atanov et al.
(1999) introduced a variational approach of Lagrangian multipliers us
rs criterion to estimate roughness coefficients. The Sequential Quadratic
ming Algorithm was used by Ramesh
Limited memory Broyden, Fletcher, Goldfarb and Shanno with Bound constraints
(L-BFG
All e they utilise the
values of first derivatives. There is another optimisation approach known as direct
ed by Wu (1976), and a pattern search
ppou (1984).
approaches or global
search methods (e.g. Solomatine 1995; Wu and Simpson 2001) can be used in
ation problem. For example, a Genetic Algorithm
(1999) and Adaptive Clustering Covering
Optimisation with a Local search (ACCOL) algorithm (Solomatine 1995) was used by

S-B)) was used by Ding et al. (2004).
of the above methods belong to the gradient methods becaus
search methods that also can be applied to the parameter identification problems. For
these methods, the trial values of the coefficients are determined by a simple strategy,
which are usually based on the previous changes to the coefficients. These methods are
simple because they do not need to compute the numerical gradients of objective
functions for the unknowns, which may be too complicated to evaluate. The main
disadvantage of this approach is that the convergence towards the optimum is very
slow and may require large amounts of computation. The direct search methods have
been applied in the inverse problem in open channels by some authors. For example,
the Rosenbrock and Simplex methods was us
developed by Hooke and Jeeves was used by Zo
Several other optimisation methods widely applied in hydrology such as shuffle
complex evolution (Duan et al. 1993), genetic and evolutionary
hydraulics for parameter identific
(GA) was used by Sanchez and Westphal
Nguyen and Fenton (2004) for identifying roughness coefficient. However, these
methods are usually very inefficient because they require a large number of iterations.

102
Chapter 4 Roughness Identification Model Development for Unsteady Flow
In comparison, the gradient methods are more efficient than the direct search methods
because of the higher speed of convergence. However, they are more complicated
depth on the floodplain is small.
For this case, the derivatives of the objective function with respect to the floodplain
roughness coefficient are considerably smaller than the main channel roughness
coefficient. This can be overcome by using direct search methods. For these reasons, in
this study, a direct search method was chosen.
Based on the chosen least square objective function, a direct search optimisation
algorithm proposed by Powell (1965) and improved by (Acton 1970) is adopted for the
optimisation sub-model. This method is an extension of the basic pattern search
method. It is the most widely used direct search method of conjugate directions (Rao
1996). It is an efficient method for finding the minimum of a function of several
variables without calculating derivatives. A conjugate directions method will minimise
a quadratic function in a finite number of steps. Since a general non-linear function can
be approximated reasonably well by a quadratic function near its minimum, a
conjugate directions method is expected to speed up the convergence of even general
nonlinear objective functions. Also, the upper and lower constraints are introduced in
the model to restrict the coefficients to physically realistic values. The brief description
of this algorithm is described in Appendix C. A subroutine based on the Powell
method w
s identification model was developed by combining the hydraulic
isation sub-model described above.
The computer code for the model was written in FORTRAN 77/90.
because they require gradients of the objective function for the unknowns. Moreover,
for the inverse problem in open channels, the main disadvantage of the gradient
methods, as indicated by Zoppou (1984) is that these methods exhibit scaling problems
that may occur when the derivatives with respect to a particular parameter are
considerably smaller than the others. This is the case for the roughness identification
problem applying to compound channels when the
as obtained from Press et al. (1992).
Finally, the roughnes
sub-model described in Section 4.3 and the optim
103
Chapter 4 Roughness Identification Model Development for Unsteady Flow
4.5 Summary
This chapter offered a brief review about the hydraulic modelling and the inverse
problem in open channels. The roughness identification model was developed
o sub-models called the hydrau
model. For the hydraulic sub-model, an appropriate model for one-dimensional
ction was chosen and Powells algorithm was adopted. This
roughness identification model will be applied in Chapter 5 to investigate the
mod en
channels. The performance of the model for compound channels will also be
investigated. Then the model will be applied to the natural riv ed
in Chapter 6.


including tw lic sub-model and the optimisation sub-
unsteady flow with compound channels was developed. The variation of roughness
coefficient value with stage was also considered. An alternative method of using a
conveyance function was suggested and described. For the optimisation sub-model, the
least square objective fun
elling factors that affect the quality of the identified roughness coefficients in op
ers that will be discuss
104
Chapter 5
Chapter 5
Numerical experiments on the roughness
identification problem
5.1 Introduction
fect the quality of the identified parameters. If these
difficulties are not considered, the satisfactory solution of the inverse problem can
ance criteria and optimisation algorithms.
tions and the values
The progress of the inverse problem in open channels has been faced with several
inherent difficulties because of different factors. Firstly, the inverse problem is often
improperly posed, i.e. its solution may be non-unique and unstable with respect to the
data error. Secondly, the quantity and quality of observation data are usually
insufficient. Thirdly, the modelling factors, even for a particular numerical scheme and
solving algorithm, often af
never be obtained by changing only perform
Several factors related to the choice of objective function, data errors and the location
of gauge sites for a single channel have been considered by Khatibi et al. (1997).
However, no studies address the effect of modelling factors on the quality of the
identified roughness coefficient. Moreover, as indicated in the previous chapter, in
flood routing in natural rivers, many channels have compound sec
of roughness coefficients in the main channel and floodplains are usually different.
This indicates the need to extend the roughness identification problem to compound
channels. Furthermore, the values of the roughness coefficient, in many cases in
natural rivers, vary with flow where the roughness coefficient may be formulated as a
105
Chapter 5 Numerical Experiments on Roughness Identification Problem
function of water stage. It is necessary to test the ability of the model to identify the
rough ess function( n s).
y using the model developed in Chap
objectives related to the roughness identification problem as follows:
of t
f ent (Manning's n);
Extends the roughness identification problem to a compound channel;
ance of the model developed in
Chapter 4 needs to be tested. The synthetic data generation and the performance of the
model are presented in the next sections.
5.2 Generation of synthetic data
The synthetic data include channel and cross-section data and flow data. The details of
generation of these data are described in the subsections below.
5.2.1 Cross-sectional data
For a channel with a single cross-section:
A model channel has a length of 40 km. The cross-section has a trapezoidal shape with
a width of 50m and sideslope of 1.5 (see Figure 5.1a). In many test cases, the slope
and the true roughness coefficient n of the channel are 0.0004 and 0.035 respectively,
otherwise unless specified.
Therefore, b ter 4, this chapter addresses several
Investigates the modelling factors affecting the quality he identified
roughness coef ici
Considers the variation of the roughness coefficients with water stage; and
Tests an alternative method by using a conveyance function.
In order to obtain the above study objectives, the synthetic data are used. Due to the
fact that in the field the true value of Mannings n is not known, the advantage of
using synthetic data is that it is possible to make comparisons between the estimated n
with the true value n. This can provide the abilities to evaluate the parameter
identifiability of the model and investigate factors that affect the quality of the
identification problem. Moreover, before carrying out the numerical experiments
related to the objectives of this study, the perform
0
S
106
Chapter 5 Numerical Experiments on Roughness Identification Problem
For a
ue values of the
c
n
f
)

channel with a compound cross-section:
A model channel has a length of 40 km with a slope of 0.0004. The tr
main channel roughness coefficient (n ) and the floodplain roughness coefficient (
are 0.028 and 0.042 respectively, which are selected for the whole channel reach. The
sideslopes of the cross-section for the main channel and the floodplains are 1 vertical
and 1.5 horizontal. The lateral slope of the floodplains is 0.005. The geometry of the
compound cross-section is illustrated in Figure 5.1b.
Figure 5.1 The cross-sections of the model channels (not to scale)
5.2.2 Flow data
5.2.2.1 Generation of boundary conditions
In this study, the effect of different combinations of boundary conditions will be
considered. There are three common types of boundary conditions at upstream and
down stream ends of the channel which are (i) ) (t Z Z = , (ii) ) (t Q Q = , and (iii)
.
1.5
1.0
50
a. Single cross-section
b. Compound cross-section
0.5
50
4.5
100
100
1.5
1.5
1.0
1.0
) (Z f Q =
107
Chapter 5 Numerical Experiments on Roughness Identification Problem
When the upstream boundary condition is a discharge hydrograph, it is a synthetic
hydrograph generated by the equation similar to Khatibi et al. (1997) and Ramesh et
al. (2000), as

+ =
p p
b p b
t
t
t
t
Q Q Q t Q 1 exp ) ( ) ( (5.1)
where is the init e to peak
discharge, an st es ameters for
the upstream hydrographs for the single channel case an nd channel case
are shown in Table 5.1.
Table 5.1 The parameters ydrographs any test cases
Parameter
(m /s) (m /s) (hours)
b
Q ial discharge,
p
Q

is the peak discharge,
p
t is the tim
d is a con ant. For many test cas , the values of these par
d the compou
of the upstream h used in m
b
Q

3
p
Q

3
p
t

Total simulated
time (hours)
For the single channel
200 700 8 5 48
For the compound channel
200 1500 24 5 96

However, when considering the effect of characteristics of flood events, these
parameters will be changed. For these cases, the parameters Q
p
, t
p
or in Equation
(5.1) will be specified. Figure 5.2 shows the effects of different parameters on the
shape and size of an upstream hydrograph. This figure shows clearly that by changing
p
Q and
p
t the size and shape of the upstream discharge hydrograph are changed
considerably compared with changing. Hence, these two parameters will be used for
the numerical tests when considering the effect of characteristics of flood events on the
quality of identified parameters. For all numerical test cases the value of is 5.

108
Chapter 5 Numerical Experiments on Roughness Identification Problem
100
200
300
400
500
600
700
800
m
3
/
s
)
0 10 20 30 40 50
Time (hours)
D
i
s
c
h
a
r
g
e

(
tp = 4 hours
tp = 8 hours
tp = 16 hours
(a)
(b)
100
200
300
400
500
600
700
800
900
1000
1100
0 10 20 30 40 50
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Qp = 500 m3/s
Qp = 700 m3/s
Qp = 1000 m3/s

(c)
100
200
0 10 20 30 40 50
300
400
500
600
700
800
s
c
h
a
r
g
e

(
m
3
/
s
)
Time (hours)
D
i
= 5
= 10
= 15

Figure 5.2 Effect of different parameter in Equation (5.1) on the size and shape of the
upstrea drograph: (a) (b) effec of
The stage hydrograph at the upstream end and the stage and discharge hydrographs at
the downstrea d of the model channel are the raphs at the
pstream end and at the middle of a 80 km long channel respectively. This 80 km
m hy effect of
p
t , t of Q , (c) effect
p

m en simulated hydrog
u
109
Chapter 5 Numerical Experiments on Roughness Identification Problem
channel has the same channel properties as the 40 km channel. The upstream boundary
condition in this case is the discharge hydrograph (Equation (5.1)) and the downstream
boundary is a steady uniform rating function. The computational time and step are 0.5
hour and 1 km respectively.
5.2.2.2 Generation of observed data
The observation data are the simulated stage/discharge with a true value of the
roughness coefficient(s) of the channel at a node located at one or several sections of
the model channel. Assuming that n is no longer known, these observation data,
together with the geometric data and the boundary conditions both upstream and
downstream ends are sufficient to formulate the inverse problem.
As discussed in Chapter 4, flow measurement data and geometry data usually contain
errors or noise. In mathematical modelling, the other error sources are model errors
and numerical errors. It is generally believed that all these source of errors are
normally distributed (Khatibi et al. 1997). In order to mimic the real data and to
investigate the effect of data errors on the identified roughness coefficient, normally
distributed noise is introduced into the simulated observed discharges or depths (noise
free) as follows:
+ = (5.2)
o
Y
Y
o o
n
o
Y Y
where
n
o
Y is the observed depth/discharge data with noise level , ) , ( = N is a
random error/noise term sampled from a normal distribution of zero mean ( 0 = ) and
with a standard deviation of .
The noise level is measured by the magnitude of the standard deviation . For each
noise level, noise samples are generated with zero mean and standard deviation of
by using the random data generation tool from EXCEL. For all the test runs when the
effect of data errors is considered, the range of noise level is from 0.05 to 0.2. The
corresponding maximum relative error could be up to 15% and 50% for the noise level
of 0.05 and 0.2 respectively. Figure 5.3 shows the examples of the observed stages and
discharges when the observed data are contaminated with two random noise samples at
110
Chapter 5 Numerical Experiments on Roughness Identification Problem
the d oise levels is to reconfirm
the f
perform model developed in Chapter 4.
mi dle station of the reach. The use of wide range of n
ef ect of data noise on the quality of the identified parameter and to evaluate the
ance of the
16
9
11
12
13
14
15
0 10 20 30 40 50
Time (hours)
S
t
a
g
e

(
m
)
10
noise free
= 0.05
= 0.20

100
200
300
400
500
600
700
800
900
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
0 10 20 30 40 50
noise free
= 0.05
= 0.20

Before conducting the numerical experiment to obtain the study objectives mentioned
Figure 5.3 Noise contamination of the observed stages and discharges at the
observation gauge at the middle of the channel
5.3 Model performance and preliminary results
above, it is necessary to validate the performance of the roughness identification model
111
Chapter 5 Numerical Experiments on Roughness Identification Problem
which was developed in Chapter 4. Thus, in this section some preliminary numerical
experiments were conducted similar to the work done by Kha et al. (1997).
In this case, the model channel illustrated in Figure 5.1a with a constant roughness
coefficient of 0.035 for a whole channel, wa
tibi
s used to demonstrate the performance of
the model. The upstream boundary condition was the discharge hydrograph and the
model
downstream boundary condition was the stage hydrograph. The observed data were the
simulated stages at the middle cross-section of the channel. For all computations, the
initial value of the roughness coefficient was assumed to be 0.02, which was quite
different from the true value. Also, in order to restrict the values of Manning's n
between the physical meanings, the bound constraints of the roughness coefficient in
the inverse were set as 0 . 1 001 . 0 n for all numerical tests in this study.
The roughness coefficient was computed with different cases of the observed data with
and without noise. When the noise was contained in the observed data, four different
noise levels of 05 . 0 = , 0.10, 0.15 and 0.20 were considered. For each noise level, 50
ne. The computed results also indicate that the
higher the noise level the wider the range of identified parameters. However, these

samples of observed data were generated using Equation (5.2) where the random noise
) , 0 ( = N was generated by different random seedings. The computed results of the
identified roughness coefficient with different noise levels are shown in Table D.1 in
Appendix D and a summary of the identified roughness coefficients including the
average values and the maximum and minimum values are given in Table 5.2. From
this table it can be seen that for the case of the noise-free observed data, the roughness
identification model led to the true value of the roughness coefficient. When data noise
was included, for any given noise level, the individual values of the identified
coefficient deviated from the true o
identified values scattered around the true value and the average values were very
close to the true one, even with a high noise level. This proved the good performance
of the roughness identification model developed in Chapter 4.

112
Chapter 5 Numerical Experiments on Roughness Identification Problem
Table 5.2 The range and average values of the identified n of 50 samples with
different noise levels (the true value is 0.035)
Noise level
Range of identifie n
(50 samples)
Average of identified n
(50 samples)
d
0.00 0.03500 0.03500
0.05 0.03389 0.03596 0.03501
0.10 0.03301 0.03680 0.03495
0.15 0.03238 0.03767 0.03502
0.20 0.03103 0.03858 0.03499

When observed data contains noise the ed from each
sample generally deviates from the true value, it is difficult to evaluate the behaviour
of the identified roughness coefficients. eans and their confidence
int e
perf l.
From the results of 50 samples of noisy data for each noise level, the statistical
tion of the mean value of identified parameters with different
sample sizes for the noise level of 0.1. The figure shows that the mean varies very little
u
e
identified coefficients obtain
Thus, their m
ervals applied by Khatibi et al. (1997) could be useful parameters to evaluate th
ormance of the mode
distribution of the errors induced in identified roughness coefficients was tested using
the Chi-square (
2
) goodness-of-fit test as described by Haan (2002). The results of
fitting a normal distribution to the identified roughness coefficients (see Section D.2 in
Appendix D) showed that at a 5% significance level it is significant to hypothesise that
the identified coefficients are normally distributed. This justifies using Students t
distribution for calculation of the confidence intervals of the identified parameters for a
small sample size. The mean values of the identified parameters with different sample
sizes (the number of samples for each noise level) are calculated. Figure 5.4 illustrates
an example of the varia
with sample size. For the sample size of 10, the deviation of the mean value from the
true one is less than 1%. This indicates that sing the small sample sizes can represent
reasonably the population mean of identified parameters. Therefore, in order to reduc
the amount of computation, a sample size of 10 is used when the data noise is
considered.
113
Chapter 5 Numerical Experiments on Roughness Identification Problem
32
33
34
35
36
37
38
0 10 20 30 40 50
N samples umber of
T
h
e

m
e
a
n
s

o
f

n

(
1
0
-
3
)

Figure 5.4 Variation of the mean of identified parameters with the number of samples
Table 5.3 shows the mean values of identified roughness coefficients and their range
w
it can be seen e to the true
one. The ranges of the mean values at a 95% confidence level contain the true value.
As expected, the lower the noise level, the narrower the confidence interval, indicating
Table 5.3 The means of the identified roughness coefficients and of 10 samples and
their range with a 95% confidence level (the true value is 0.035)
)
n level
ith a 95% confidence level by using 10 samples for each noise level. From this table
that the mean values of a small sample size are very clos
the more accurate the ability to identify the value of the roughness coefficient.
Noise level
Mean of identified n
(10 samples
Range of the mean of
identified n at a 95%
co fidence
0.05 0.03499 0.03455 0.03542
0.10 0.03501 0.03448 0.03553
0.15 0.03519 0.03434 0.03604
0.20 0.03513 0.03405 0.03622

In this section, several numerical experiments were conducted. The performance of the
model for identifying of the roughness coefficients was evaluated using synthetic data
for a hypothetical open channel reach. The computed results indicated that when
observed data were free from noise, the model can very accurately identify the value of
the roughness coefficient. When observed data were contaminated with noise, although
114
Chapter 5 Numerical Experiments on Roughness Identification Problem
the individual values of the identified roughness coefficient deviated from the true
value, they scattered around the true value and the average value was very close to the
true one, even with a high noise level. This indicates the good performance of the
model. Therefore, the model can be applied to investigate other factors that affect the
quality of the identified roughness coefficient. The results also showed that the mean
values varied very little with the sample size and that the identified values are normally
distributed. This justifies the use of Students t distribution for calculation of the
confidence intervals of identified n for a small sample size. This can reduce the amount
of computation. For further numerical tests related to data errors in this study, in order
vels of to reduce the amount of computation, only noise le 05 . 0 = and/or 0.1 and the
Weighting coefficient ,
mean of small samples size (10 samples) with a confidence level of 95% will be used.
The following section will investigate the modelling factors affecting the quality of the
identified roughness coefficient.
5.4 Modelling factors affecting the quality of the identified
roughness coefficient
Every finite difference scheme used to solve partial differential equations is an
approximation to these equations. The accurate representation of the partial differential
equations is not only dependent on the choice of the finite difference scheme, but also
the choice of grid sizes, combinations of boundary conditions and errors in initial
conditions, and, in some schemes, on the value of weighting coefficient(s) of the
computational scheme. Since the values of the roughness coefficient n for unsteady
flow are obtained with the use of a numerical model, the identified coefficients must
also depend on the modelling factors. In this section, several modelling factors related
to the Preissmann scheme are investigated, including:
Computational grid sizes,
Different combinations of boundary conditions, and
Errors of initial condition related to different combinations of boundary
conditions.
115
Chapter 5 Numerical Experiments on Roughness Identification Problem
5.4.1 Effect of weighting factor ()
In the Preissmann scheme or four-point box scheme, is a weighting factor applied to
the finite difference approximations when solving the unsteady flow equations (see
Equation 4.3 in Chapter 4). The study of Liggett and Cunge (1975) on the stability and
is accuracy of this scheme showed that it is numerically stable when 1 5 . 0 < and it
always unstable for . They indicated that when the Courant number
(
5 . 0 <
x t V c Cr + = / , where c is the wave celerity and B gA c / = ) equals 1 and
5 . 0 = the numerical solution gives the most accurate results. When 5 . 0 , the
attenuation or damping of the different components is important. However, when Cr is
different from 1, which is the case for practical computations, oscillations are found for
the values of near 0.5 and disappear for 66 . 0 . The larger values of can produce
greater artificial damping/num ooth out any trace of physical
dispersion. The influence of the curacy of the computation
was exam by Fre (19 ho c a ewhat
decrease as it departs from 0.5 and approaches 1. However, the value of
provides t ost s lity. I applica is scheme, different values of
erical diffusion and can sm
weighting factor on the ac
ined ad 74), w oncluded that the ccuracy tends to som
0 . 1 =
he m tabi n the tion of th
have been used. For exampl values e, the of 0 66 . is suggested by Liggett and
Cunge (1975) and Cunge et al. (1980) and the value of = 0.6 - 1.0 is suggested in the
HECRAS document (U.S. Ar orps of s 2002). A as used by
Baltzer an Lai (19 hile f 0.5 wa ein and Fang (1970) in their box
scheme m l.
sing larger values of increases numerical diffusion; as a result, it will also affect the
results of the identified n, but by how much? This analysis is certainly very important
my C Engineer of 1.0 w
d 68) w a o s used Am
ode
U
for interpreting the effect of on the quality of the roughness identification problem.
To answer this question, several numerical experiments have been conducted.
For all test cases, the upstream boundary condition was the discharge hydrograph and
the downstream boundary condition was the stage hydrograph. The observed data were
the stage hydrograph at the middle cross-section of the channel. The computational
space and time steps for observed data were = x 1 km and = t 0.5 hour respectively
(except for the case when the effect of related to grid size is considered). Also, in
116
Chapter 5 Numerical Experiments on Roughness Identification Problem
order to investigate the pure effect of on the identified n the data errors were not
considered for this case.
The observed data were simulated stage hydrograph at the intermediate section with a
value of 6 . 0 = . Then the roughness identification model ran with different values of
varying from 0.55 to 1.0. Because the observed data were obtained from a simulation
with the true value of the roughness coefficient and the value of 6 . 0 = , the values of
the identified ith were identical with the true roughness. These results do
not mean that
n w 6 . 0 =
6 . 0 =
difference between the identified
provides the most accurate results. However, the relative
n for the case of 6 . 0 = or with the true value with
other cases enables us to assess the effect of on the quality of the identified
roughness coefficients.
The plots of the identified n against different values of are shown in Figures 5.5 to
5.10. For different numerical experiment scenarios, the computed results indicate that
larger values of weighting coefficient can make the identified n deviate from the true
value. Also, as expected, all the identified n with 6 . 0 > are over-estimated compared
to the true v e numerical
diffusion effect of : the larger the value of , the more the numerical diffusion.
The i sid
nts in terms of
flood peak and time to flood peak.
alue. This over-estimation in the identified n is attributed to th
However, the effect of on the quality of the identified roughness coefficient is, in
general, insignificant. The largest error in the identified n for these numerical tests was
only 0.0004 (relative error is 1.2%).
nvestigation also con ered the effect of on the quality of identified roughness
coefficient related to: (i) computational grid sizes, (ii) characteristics of channel
including roughness and slope, and (iii) characteristics of the flood eve
Regarding the grid sizes, different values of space and time steps within an acceptable
limit (see Section 5.4.2) were considered. These were 1 km, 2 km and 4 km, and 0.5
hour, 1 hour and 2 hours respectively. The computed results are shown in Figures 5.5
and 5.6. From these figures it can be seen that the effect of related to the time steps
t on identified roughness coefficients is more pronounced than the effect of related
117
Chapter 5 Numerical Experiments on Roughness Identification Problem
to the space steps x . The larger the value of t , the more over-estimated the
identified n when using a higher value of . However, the identified n was almost
identical for difference values of x (see Figure 5.6).
34
34.9
35.0
35.1
0. 0 0. 1
e
n
t
i
f
i
0
35.3
35.4
35.5
-
3
)
35.2
e
d

n

(
1
.8
0.5 0.6 7

.8 9
I
d
= 0.5 h
= 1.0 h
= 2.0 h

t

Figure 5.5 fied ro ss coeff ent vers ith d t tim for
synthetic data with
3
Identi ughne ici us w ifferen e steps
700 = m
p
Q /s, 8 =
p
t = n 0.035, = S 0.0004, = x 1 km hours,

34.98
35.00
35.02
35.04
35.06
35.08
35.10
35.12
0.5 0.6 0.7 0.8 0.9 1

I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
= 1 km
= 2 km
= 4 km

x

s
Figure 5.6 Identified roughness coefficient versus with different space steps for
ynthetic data with 700 =
p
Q m
3
/s, 8 =
p
t hours, = n 0.035, =
0
S 0.0004, 5 . 0 = t hour

118
Chapter 5 Numerical Experiments on Roughness Identification Problem
Figures 5 and 5 of on the identified roughness coefficient
related to channel characteristics in terms of channel slope and the value of the channel
roughness coefficient. For the channel slope, the e ducted
with three different slopes of 0.001, 0.0004 and 0.0001. From 5.7 e seen
that th ffec on th ified ess c nt is signi for the
channels with sm
where the values of the channel roughness coefficient vary fr 5 to 0.045. Figure
ffect of on the iden ied
roughness coefficient.
.7 .8 illustrate the effect
numerical test runs w re con
Figure it can b
e e t of e ident roughn oefficie more ficant
aller slope. For the channel roughness, four cases were considered
om 0.01
5.8 indicates that the rougher the channel, the greater the e tif
To consider the effect of on the identified n related to the characteristic of flood
events, different values of peak discharge (
p
Q ) of 500, 700 and 1000 m
3
/s and the time
to peak discharge (
p
t ) of 4, 8, 16 hours were considered. The higher the value of
p
Q
the faster the rise of the flood, and the smaller the value of
p
t , the shorter the period
waves of the flood (see also Figures 5.2a and 5.2b). From Figures 5.9 and 5.10, it can
be seen that when using a higher value of the deviation of identified n from the true
values corresponds more with the flood events with shorter waves or with faster rise,
i.e. the floods with higher values of
p
Q or smaller values of
p
t .


119
Chapter 5 Numerical Experiments on Roughness Identification Problem
35.15
So = 0.0001
So = 0.0004
So = 0.001
34.95
35.00
35.05
35.10
I
d
e
n
t
i
f
e
d

n

(
1
0
-
3
)
34.90
0.5 0.6 0.7 0.8 0.9 1


Figure 5.7 Identified roughness coefficient versus with different channel slopes for
synthetic data with 700 =
p
Q m
3
/s, 8 =
p
t hours, = n 0.035.

-0.1
0.0
0.1
0.2
0.5 0.6 0.7 0.8 0.9 1

%

d
e
v
i
a
t
e
d

f
r
o
m

t
h
0.3
0.4
u
e

e

t
r
n
n = 0.015
n = 0.025
n = 0.035
n = 0.045

Figure 5.8 Identified roughness coefficient versus with different channel roughness
for synthetic data with 700 =
p
Q m
3
/s, 8 =
p
t hours, =
0
S 0.0004
120
Chapter 5 Numerical Experiments on Roughness Identification Problem
34.90
34.95
35.00
35.05
35.10
35.15
35.20

I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
Q = 500 m3/s
Q = 700 m3/s
Q =1
0.5 0.6 0.7 0.8 0.9 1
000 m3/s

Figure 5.9 Identified roughness coefficient versus with different peak discharges for
synthetic data with 8 =
p
t hours, = n 0.035, =
0
S 0.0004
35.15
35.20
-
3
)
34.95
35.00
35.05
35.10
0.5 0.6 0.7 0.8 0.9 1

I
d
e
n
t
i
f
i
e
d

n

(
1
0
= 4 hours
= 8 hours
=16 hours
t
p
t
p
t
p

Figure 5.10 Identified roughness coefficient versus with different time to peak
discharge for synthetic data with 700 =
p
Q m
3
/s, = n 0.035, =
0
S 0.0004
In summary, this section has investigated the effect of the weighting coefficient on
the identified roughness coefficient. The results show that using improper values of
can
over-estimated the identified r i his is attributed to numerical
da se o use
cause biases in the value of the identified n. The larger the value of , the more
oughness coeff cient. T
mping/dispersion becau f the of 5 . 0 > (Cunge e
experim that th o he id tified r ghn i
t al., 1980). The numerical
ents showed e effect f on t en ou ess coeff cient also
121
Chapter 5 Numerical Experiments on Roughness Identification Problem
depends on the grid size, channel characteristics and the characteristics of flood events.
The effect on identified roughness coefficient is more pronounced in the following
cases: (i) longer computational time steps, (ii) rougher and smaller slope channels, and
(iii) the flood events with higher peak discharges or shorter time to peak. However,
this study indicates that the effect of on the identified roughness coefficient is
insignificant compared to the effect of data errors (investigated in Section 5.3).
Nevertheless, in order to limit the effect of on the identified roughness coefficient,
the lower values of should be employed, especially for the rougher or low gradient
reaches and for flood events with a rapid rise (high peak discharge or short time to
peak). Using smaller time steps is one appropriate solution to reduce the effect of to
the biases of the identified roughness coefficient for the flood routing problem.
5.4.2 Effect of computational grid sizes
The implicit difference schemes in general and the Preissmann scheme in particular
have a key advantage in that they are unconditionally stable and avoid restrictions on
the computational time step for a weighting coefficient of 5 . 0 > (Liggett and Cunge
1975). Hence, in the computation of unsteady flow with these schemes, using the
larger computational grid sizes (time and space steps) in the model may be preferable
to reduce the computational time. However, as indicated by Cunge et al. (1980) the
grid sizes are one of the most important factors affecting the accuracy of the solution
and should not be blindly selected at arbitrary values. Computational time and space
steps should be limited to adequately describe the prototype and to achieve an
r
vestigated.
different space and tim , the channel
m d in F ployed. The b s at the
acceptable level of numerical accuracy. Moreover, although implicit models are
theoretically stable they can become suddenly unstable because of improper selection
of grid sizes (Fread 1974; Liggett and Woolhiser 1967). In the roughness identification
problems, the roughness coefficient values are identified by minimising the difference
between simulated and observed stages or discharges. Therefore, additional er ors
attributable to the discretisation must also affect the identified parameters and therefore
need to be in
In order to evaluate the effect of grid sizes on the identified roughness coefficient,
e steps were considered. For these numerical tests
odel illustrate igure 5.1a was em oundary condition
122
Chapter 5 Numerical Experiments on Roughness Identification Problem
u m wnstream bou daries were the discharge hydrograph and the stage
hy ograph ec he d th r di yd t
the middle ss-s th l. putational tim ac r
the bserve ta t r
pstrea and do n
dr resp tively. T observed ata were e stage o scharge h rograph a
cro ection of e channe The com e and sp e steps fo
o d da were = 0.5 hou and = x 1 ect e s
identificati odel ran with t v
km resp ively. Th roughnes
on m differen alues of time steps ( x ) ranged 5
ho to 4.0 rs e s
from 0.2
ur hou and spac teps ( t ) 2 4 k is n
order to see the pure effect of , t of rs w on
of 1 km, km and m. In th section, i
grid sizes he effect data erro as not c sidered.
Table 5.4 shows the values of the identified roughness coefficient with the time
steps 2 t hours and space steps 4 x km and the corresponding maximum
Courant number (Cr
max
). From this table it can be seen that if the grid sizes are kept
within acceptable limits, the grid sizes and the Courant number do not significantly
affect the results of the identified roughness coefficient. The maximum error of
estimated n is only 0.0004 (relative error is only 1.2 %) for the case of 4 = x km and
2 = t hours.
Table 5.4 Effect of different grid sizes on the identified roughness coefficient
(the true value is 0.035)

(km) (h)
Estimated n Estimated n
x t
Cr
max
(observed stage) (observed discharge)
1 0.5 16.2 0.035000 0.035001
1 1.0 32.3 0.035039 0.034986
1 2.0 64.6 0.035102 0.035272
2 0.5 8.1 0.034999 0.035063
2 1.0 16.2 0.035046 0.035024
2 2.0 32.3 0.035111 0.035312
4 0.5 4.0 0.034998 0.035054
4 1.0 8.1 0.035035 0.035118
4 2.0 16.2 0.035114 0.035410

However, as expected, when the grid sizes increase to a certain limit, the identified
roughness coefficient starts to deviate from the true value. Figure 5.11 clearly
illustrates the effect of grid sizes on the quality of the identified roughness coefficient.
From this figure it can be seen that for these numerical tests, when 2 > t hours, the
123
Chapter 5 Numerical Experiments on Roughness Identification Problem
identified roughness coefficient deviated considerably from the true value. The biases
of the identified roughness coefficient because of large grid sizes can be explained by a
number of reasons. Firstly, when the computational grids increase to a certain limit, the
grid sizes cannot describe accurately the prototype. For these numerical tests, it can be
seen that if the time steps are too large ( 2 > t hours for these tests), the hydrograph
cannot be properly characterised and the peak of the hydrograph can be ignored.
Secondly, when the computational steps chosen are large, the resulting truncation
errors also become large which will cause th ditional errors in simulated stages and
l
Fi
e ad
discharges. As a result, the roughness coefficient that can be identified by minimising
the difference between the observed and simu ated data will be biased.






gure 5.11 Identified roughness coefficient versus grid sizes
The effect of grid sizes on the identified roughness coefficient related to the channel
characteristics and the characteristics of flood events were also investigated. Several
test runs were conducted with different values of channel roughness, channel slope,
peak discharge and time to peak. The parameters for the typical case were = n 0.035,
=
0
S 0.0004, 700 =
p
Q m
3
/s and =
p
t 8 hours. It should be noted that, in these
numerical tests, when considering the effect of a certain parameter, only the value of
this parameter is changed and all the other parameters of the model are kept
unchanged.
26
28
30
32
34
36
38
I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
Table 5.5 shows the computed results with different time steps for the cases of
different values of channel slope and channel roughness. From this table it can be seen
0 0.5 1 1.5 2 2.5 3 3.5 4
t (hours)
= 1 km
= 2 km
= 4 km
x
x
x
26
28
30
32
34
36
38
I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
0 0.5 1 1.5 2 2.5 3 3.5 4
t (hours)
= 1 km
= 2 km
= 4 km
x
x
x
124
Chapter 5 Numerical Experiments on Roughness Identification Problem
that the slope of a channel plays an important role in the choice of grid sizes. In order
to obtain a reasonable roughness coefficient value for the greater slope channels,
maller grid sizes should be chosen. Even in the case of a channel slope of 0.001 in
is study with grid sizes of
s
1 > t th , the computation process was interrupted.
Accordingly, the roughness coefficient could not be obtained. This can be explained by
oting that in these channels the hydraulic properties are changing with respect to
space and time rapidly. When the computational step chosen is too large, the resulting
uncation error may be so large that the computed depths or discharges are totally
nrealistic, e.g., the computed flow depths are less than critical depth or negative. This
causes instabilities in the computed hydrograph or an execution of the hydraulic model
rogram to abort. Therefore, for channels with high slope ( 0.001), the choice of
computation grid sizes should be done with caution.
Table 5.5 Relative errors in the identified roughness coefficient with different time
steps related to channel characteristics
Channel slope Channel roughness
n
tr
u
S p
t

(h)
0.0001 0.0004 =
0
S =
0
S =
0
S 0.001
= n 0.025 = n 0.035 0.045 = n
0.25 0.027 0.011 0.022 0.022 0.011 0.010
0.5 0.001 0.002 0.001 0.001 0.002 0.001
1 0.242 0.039 11.169 0.267 0.039 0.038
2 0.451 0.776 - 4.284 0.776 0.285
3 1.051 3.589 - 6.47 3.589 0.372
4 4.385 21.681 - 21.092 21.681 31.658
Note: Error percentage of are presented that are calculated as n % 100 . /
true true comp
n n n
Regarding the characteristics of flood events in terms of peak discharge and time to
peak, the numerical tests were conducted with different values of and . These
were 500, 700 and 1000 m
3
/s and 4, 8 and 16 hours respectively. Table 5.6 shows the
computed results for different time steps with different values of peak discharge and
t
rou
discharges. However, these behaviours with me to peak were
different. For floods with very short time to peak, a large computational time step
could produce significant errors. Therefore, in order to limit the effect of grid sizes on
p
Q
p
t
ime to peak. The results show that the behaviours of the errors in the identified
ghness coefficient with computational grid size were similar with different peak
different values of ti
125
Chapter 5 Numerical Experiments on Roughness Identification Problem
the q ood
eve
T el e e it
s lated to the cteristics of vents
(m k (hours)
uality of the identified parameter the smaller time step should be chosen for fl
nts with shorter time to peak discharge.
able 5.6 R ative errors in th
teps re
identified roughn
chara
ss coefficient w
flood e
h different time
Peak discharge
3
/s) Time to pea
t

(h) 500 =
p
Q 7 =
p
Q 00 1000 =
p
Q =
p
4 t =
p
t 8 1 =
p
t 6
0.25 0.01 02 0 00 0.011 0. 5 .014 0.011 0. 4
0 0. 00 0 2 00 .5 003 0.002 0. 1 .001 0.00 0. 1
1 0. 22 0 034 128 0.039 0. 5 .132 0.039 0.
2 0 0.8 2 026 .526 0.776 .577 0.776 0.
3 3 .59 20 496 .061 3.589 3 8 .206 3.589 1.
4 21 1 1.304 30 1 .58 .702 21.68 2 .127 21.68 5 1
Note: Error percentage of re n are p sented that are calculated as % 0 10 .
true
n n
In brie e nu l e e w e on d o he
. The
hydrographs with smaller
space step, then additional cross-sections should be added in the area where the
ifferences occur. Also, the grid sizes can be selected based on the criteria given by
Samuels (1985) and Fread and Lewis (1993).
Effect of different combinations of boundary conditions
In the roughness identification model, one of the important input data for the hydraulic
ub-model is the boundary conditions. It is important to establish whether different
/
true
n
comp

f, th merica xperim nts sho that th selecti of gri sizes is ne of t
important factors that affect the quality of identified roughness coefficient
selection of grid sizes should be made with caution, especially for cases with high
slope channels or flood events with short time to peak. Generally, smaller grid sizes
can provide more accurate results and there is less chance for the occurrence of
numerical difficulties or instabilities; however, smaller grid sizes result in more
computational time to obtain the solution. In order to choose appropriate grid sizes, it
is suggested that before running the roughness identification model, one should try to
run the hydraulic model first with a smaller time step to see if the behaviour of the
results (profiles and hydrographs) change significantly. If the results do change
significantly, then the original time step should be decreased. Similarly, for space step,
if there are some significant differences in the simulated
d
5.4.3
s
126
Chapter 5 Numerical Experiments on Roughness Identification Problem
types of boundary conditions would affect the quality of the identified roughness
coefficient. To answer this question, several numerical test runs were conducted and
e results are analysed with different combination of boundary conditions in the
ubsections below.
.4.3.1 Different combinations of boundary conditions
In this study, only subcritical flows are considered where the boundary conditions are
and downstream ends. As discussed in Chapter 4, there are
ree common types of external boundary conditions specified at boundaries. These
are: (i) discharge as a function of time (a discharge hydrograph,
th
s
5
required at both upstream
th
) (t Q Q = ), (ii) water
tage as a function of time (a stage hydrograph, ) (t Z Z = s ), and (iii) discharge as a
nction of water stage (a rating curve, ) (Z f Q = fu ).
lthough the choice of a combination of boundary conditions in unsteady flow
computation is usually influenced by the availability and accuracy of the boundary
ata, several studies which analysed the problems associated with different boundary
conditions (e.g. Zoppou and O'Neill 1981; Cunge et al. 1980; Verwey 1975) indicated
at certain combinations of boundary conditions must be used with caution.
Moreover, in the inverse problem, the roughness coefficient is identified through an
iteration process with a wide range of roughness coefficient values (for this study, the
ound constraints of the roughness coefficient were set as ). It is
closely the effect of different combinations of boundary
conditions on the quality of the identified roughness coefficient.
As indicated by Verwey (1975), the use of a rating function as a boundary condition
r unsteady flow is strictly incorrect mathematically because the use of this function
type of boundary condition as an upstre ads to numerical
instability in mathematical models. Cunge et al. (1980) also noted that this type of
boundary condition cannot be used as an external upstream boundary because the flow
ould increase or decrease without bound. However, the use of a rating function at the
ownstream boundary may be acceptable although the use of a single value of rating
function may introduce additional errors, especially for the rivers with small slopes.
A
d
th
0 . 1 001 . 0 n b
necessary to investigate more
fo
forces various terms in the equation of motion to vanish. He showed that using that
am boundary condition le
w
d
127
Chapter 5 Numerical Experiments on Roughness Identification Problem
Therefore, in this investigation, from three common types of boundary condition, six
combinations of boundary conditions were considered as shown in Table 5.7.
Table 5.7 Different combinations of boundary conditions
No. Combination Upstream boundary Downstream boundary
1 Q-Z
) (t Q Q =

) (t Z Z =
2 Z-Z
) (t Z Z = ) (t Z Z =
3 Q-Q
) (t Q Q =

) (t Q Q =

4 Z-Q
) (t Z Z =

) (t Q Q =

5 Q-Qz
) (t Q Q =

) (Z f Q =

6 Z-Qz
) (t Z Z =

) (Z f Q =


5.4.3.2 Effect of different combinations of boundary conditions
To investigate the effect of different combinations of boundary conditions on the
uality of identified roughness coefficient, the channel model illustrated in Figure 5.1a
was employed. The slope of channel is 0.0004 and the channel roughness coefficient is
.035. The boundary conditions and observed data at the intermediate sections were
the manner described in Section 0. The computation time and space steps
of 0.5 hour and 1 km respectively were applied while simulating the
observed data and boundary condition data nd while solving the inverse problem.
The noise-free observed data at the middle channel were used to identify the roughness
coefficient with different combinations of boundary conditions. The computed results
are summarised in Table 5.8. From this ta e it can be seen that the computed results
are very accurate for the Q-Z and Z-Z combinations. For the Q-Qz and Z-Qz
combinations, there is a very slight difference in the value of the identified roughness
rating curve at th ndition will be
vestigated in more detail later on in this section. One important finding from this
investigation is that the inverse model fails to obtain the roughness coefficient when
sing a discharge hydrograph at the downstream boundary condition (for the
combinations Q-Q and Z-Q).
q
0
generated in
= t = x
a
bl
coefficient that can be attributed to the additional errors because of the use of a single
e downstream boundary. The effect of this boundary co
in
u
128
Chapter 5 Numerical Experiments on Roughness Identification Problem
Table 5.8 Identified roughness coefficient from different boundary combinations
(the true value = n 0.035)
Boundary combination Q-Z Z-Z Q-Q Z-Q Q-Qz Z-Qz
Identified n 0.03500 0.03500 - - 0.03497 0.03492

To explore the reasons behind this failing, test runs were undertaken with different
values of the roughness coefficient (30% smaller and larger than the true one) by using
the hydraulic model for the Q-Q and Z-Q combinations. The computed results
indicated that when the assigned roughness coefficient value was the larger one, after
some computational steps, there were computational instabilities and the program was
interrupted. At the simulation time before the program was interrupted, the simulated
depths and discharges in some parts of the channel became very small or had negative
values. These results are also supported by the finding by Cunge et al. (1980) that a
discharge hydrograph specified as a down m boundary condition must be applied
with care. This problem can be explaine by noting that when a higher value of
roughness coefficient is applied to the channel, the channel capacity will be decreased.
The discharges imposed at the downstream e capacity of the
channel. When an assigned value is large at a certain level, this may cause some parts
of the channel to become dry (dewatered). This phenomenon will cause computational
instabilities. In the inverse model, the roughness coefficient is identified through the
automatic iteration procedure with a wide range of roughness coefficient values. The
dewatered problem, when a high value of the roughness coefficient is assigned (in the
iteration procedure), is the reason for the failure in the roughness identification
procedure. Hence, for the inverse problem to identify the roughness coefficient, a
discharge hydrograph at the downstream boundary should not be used.
roughness coeffic p annel. Different
values of channel slope of 0.0001, 0.0004 and 0.001 were investigated. The roughness
coefficient was identified using observed data at different locations of observation
stations from the upstream boundary at
strea
d
boundary will exceed th
The effect of using a single rating curve at the downstream boundary on the identified
ient is believed to de end on the slope of the ch
= x 20 km (at the middle of the channel),
= x 30 km and = x 40 km (at the downstream boundary). The computed results are
shown in Table 5.9.
129
Chapter 5 Numerical Experiments on Roughness Identification Problem
Table 5.9 Identified roughness coefficient using a single rating curve as a downstream
boundary condition (the true value = n 0.035)
Observed data Q-Qz combination Z-Qz combination
Type Station*
0001 . 0
0
= S

0004 . 0
0
= S

001 . 0
0
= S

0001 . 0
0
= S

0004 . 0
0
= S

001 . 0
0
= S


= x 20
0.03453 0.03496 0.03501 0.03478 0.03495 0.03501
Q
= x 30
0.03336 0.03491 0.03498 0.03432 0.03493 0.03499

= x 40
0.02965 0.03474 0.03492 0.03198 0.03485 0.03493

= x 20
0.03473 0.03497 0.03502 0.03451 0.03492 0.03500
Z
= x 30
0.03437 0.03494 0.03500 0.03332 0.03489 0.03499
= x 40 0.03204 0.03489 0.03494 0.02959 0.0347 0.03491
Note: * indicates distance from the upstream boundary in km.
From Table 5.9 it can be seen that using these combinations of boundary conditions,
the identified roughness coefficient values deviated from the true one. For the Q-Qz
ombination, the use of the observed stage data to identify the roughness coefficient
gave better results than the use of observed discharge data, while for the Z-Qz
combination, the use of observed discharges were better. As expected, the effect of a
ingle value of rating function decreases when the slope of the channel increases. This
effect is also decreased when the observation stations are far from the downstream
boundary. The computed results indicate that the effect of these combinations on the
uality of the identified roughness coefficient can be negligible for the cases of steeply
sloping channels ( ). For channels with a small slope (
c
s
q
001 . 0
0
S 0001 . 0 S ), in order
to reduce the effect of this boundary condition, the observation station should not be
close to the downstream boundary. This effect can also be reduced by prolonging the
length of the reach (Singh et al. 1997). However, this will increase the amount of
computation, especially for the roughness identification problem where the hydraulic
5.4.3.3 Identifiability of the roughness coefficient related to different
sub-model is run many times through the iteration process.
combinations of boundary conditions and types of observed data
This section will investigate the ability to identify the roughness coefficient with
different boundary conditions (Q-Z, Z-Z, Q-Qz and Z-Qz) related to the type of
observed data (stages or discharges). The unknown roughness coefficient is identified
by minimising the difference between the computed data and observed data. When
130
Chapter 5 Numerical Experiments on Roughness Identification Problem
data contain no noise, the n values converge to the true value of n. However, when data
contain noise, the identifiability will then depend on the response of computed
stages/discharges with respect to the changes in the value of n. It is expected that if the
type of data is not sensitive to roughness coefficient, the use of these data for the
formulation of the objective function could lead to an unidentifiable inverse problem.
Therefore, in order to avoid any pitfalls because of using improper observed data for
the roughness identification problem, it is necessary to assess the response of
computed stage and discharge with the change in the value of the roughness coefficient
Table 5.10 shows the values of the identified roughness coefficient when data contain
for different combinations of boundary conditions.
Figure 5.12 illustrates the discharge and stage hydrographs at the cross-section at the
middle channel with different values of n related to different combinations of
boundary conditions. From this figure it can be seen that the response of computed
stages/discharges to a change of roughness coefficient are different for different
combinations. For the Q-Z and Q-Qz combinations, the stage hydrographs are more
sensitive to the changes of n than the discharge hydrographs, while for the Z-Z and Z-
Qz combinations, the discharge hydrographs are more sensitive. From these results it is
anticipated that for the Q-Z and Q-Qz combinations, using stage data for the
formulation of the objective function could give better quality of the identified
roughness coefficient. But for the Z-Z and Z-Qz combinations, the discharge data
could provide greater potential for a reliable roughness coefficient evaluation.
noise ( ) 1 . 0 = by using different combinations of boundary conditions and different
observed data types (stages or discharges). From this table it can be clearly seen that
the errors in the identified roughness coefficient are different with different
combinations and observed data types. At this cross-section for the combinations Q-Z
and Q-Qz, using observed stage data for identifying the roughness coefficient gave
much better results than using observed discharge data, while for the cases of the Z-Z
and Z-Qz combinations using the discharge data for identifying the roughness
coefficient gave better results. These results indicate that the selection of the observed
data type has a particularly important effect on the quality of the identified roughness
coefficient.
131
Chapter 5 Numerical Experiments on Roughness Identification Problem

132
of computed discharge and stage hydrographs to the value of
roughness coefficient at the middle of the channel with different combinations of
boundary conditions

















Figure 5.12 Sensitivity
10
11
12
13
14
15
0 10 20 30 40 50
Time (h)
S
t
a
g
e

(
m
)
n = 0.0350
n = 0.0315
n = 0.0385
150
250
350
450
550
650
750
0 10 20 30 40 50
Time (h)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
n = 0.0350
n = 0.0315
n = 0.0385
10
11
12
13
14
15
0 10 20 30 40 50
Time (h)
S
t
a
g
e

(
m
)
n = 0.0350
n = 0.0315
n = 0.0385
100
200
300
400
500
600
700
0 10 20 30 40 50
Time (h)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
n = 0.0350
n = 0.0315
n = 0.0385
a. Q-Z combination
d. Z-Qz combination
b. Z-Z combination
c. Q-Qz combination
10
0 10 20 30 40 50
Time (h)
11
13
e

(
m
)
12
S
t
a
g
14
15
n = 0.0350
n = 0.0315
n = 0.0385
700
300
400
i
s
c
h
a
r
g
e
500
600
D

(
m
3
/
s
)
100
200
0 10 20 30 40 50
Time (h)
n = 0.0350
n = 0.0315
n = 0.0385
10
0 10 20 30 40 50
Time (h)
11
12
13
14
15
S
t
a
g
e

(
m
)
n = 0.0350
n = 0.0315
n = 0.0385
250
D
150
0 10 20 30 40 50
Time (h)
350
450
550
650
750
i
s
c
h
a
r
g
e

(
m
3
/
s
)
n = 0.0350
n = 0.0315
n = 0.0385
Chapter 5 Numerical Experiments on Roughness Identification Problem
Table 5.10 Errors in the identified n with different combinations of boundary
conditions and different types of observed data at the middle cross-section of the
channel when data contained noise ( 1 . 0 = )
Combination
Q-Z
Combination
Z-Z
Combination
Q-Qz
Combination
Z-Qz
Data
Sample
Z Q Z Q Z Q Z Q
1 -0.96 16.94 -6.09 1.24 -1.05 13.92 -6.32 1.09
2 -1.55 -16.30 -24.28 -0.69 -1.64 -16.42 -24.51 -0.84
3 4.47 1.79 16.71 -1.53 4.38 1.65 16.49 -1.68
4 0.08 11.88 19.77 0.36 0.00 11.72 19.54 0.22
5 -0.97 -5.79 -2.13 0.85 -1.05 -5.93 -2.35 0.71
6 -3.56 -8.81 -7.93 3.20 -3.65 -8.94 -8.16 3.06
7 0.87 7.54 8.43 -0.97 0.78 7.39 8.20 -1.11
8 0.52 16.60 20.91 0.96 0.43 13.58 20.68 0.81
9 -2.65 1.34 -1.89 4.19 -2.51 1.25 -2.03 3.96
10 -0.03 -1.93 -12.02 -1.85 -0.12 -2.07 -12.25 -2.00
Note: Error percentages in n are calculated as % 100 . / ) (
true true comp
n n n . Values greater than
10% are in bold.
The above results have demonstrated the ability to identify the roughness coefficient
with different combinations of boundary conditions related to data types at a particular
location. In order to explore more systematically the roughness identification abilities
of these combinations with different locations of observation gauges, some further
numerical tests were conducted. Figures 5.13 to 5.16 show the variation of the mean
identified n and their confidence intervals of 95% with noise level 1 . 0 = for
combinations Z-Q, Z-Z, Q-Qz and Z-Qz respectively. These figures demonstrate
clearly that the behaviour of the identified n is not only dependent on the quality of

data (as discussed in Section 5.3), the type of boundary conditions and the observed
data type, but also the location of the observation station.



133
Chapter 5 Numerical Experiments on Roughness Identification Problem
134










a. Using observed discharge data to identify n






Figure 5.13 Variation of the means of the identified n and their confidence intervals of
95% with noise level 1 . 0 = for the Q-Z combination



b. Using observed stage data to identify n
20
40
60
80
100
120
0 5 10 15 20 25 30 35 40
Distance from upstream boundary (km)
I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
mean
upper limit
lower limit
true n 10% error

20
25
30
35
0 10 15 20 25 30 35 40
Distance from upstream boundary (km)
I
d
e
n
t
i
f
i
e
d

n
40
45
0
-
3
)
50
5

(
1
Chapter 5 Numerical Experiments on Roughness Identification Problem




Figure 5.14 Variation of the means of the identified n and their confidence intervals of












95% with noise level 1 . 0 = for the Z-Z combination



b. Using observed stage data to identify n
a. Using observed discharge data to identify n
25
35
40
0 5 10 15 20 25 30 35 40
Distance from upstream boundary (km)
I
d
e
n
t
i
f
i
e
d

n
55
30
50

(
1
45
0
-
3
)
mean
upper limit
lower limit
true n 10% error

20
40
60
80
100
120
140
160
0 5 10 15 20 25 30 35 40
Distance from upstream boundary (km)
I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
135
Chapter 5 Numerical Experiments on Roughness Identification Problem


136

igure 5.15 Variation of the means of the identified n and their confidence intervals of
95% with noise level









a. Using observed discharge data to identify n
20
40
0 5 10 15 20 25 30 35 40
Distance from upstream boundary (km)
I




F
1 . 0 = for the Q-Qz combination



b. Using observed stage data to identify n
60
80
n
t
i
f
i
e
d

n

(
1
0
100
120
d
e
mean
upper limit
lower limit
-
3
)
true n 10% error

20
25
30
35
0 5 10 15 20 25 30 35 40
Distance from upstream boundary (km)
I
d
e
n
t
i
f
i
e
d

n
40
45
50

(
1
0
-
3
)
Chapter 5 Numerical Experiments on Roughness Identification Problem










a. Using observed discharge data to identify n
25
30
35
40
0 5 10 15 20 25 30 35 40
Distance from upstream boundary (km)
I
d
e
n
t
i
f
i
e
d

n

(
1
0
45
55
-
3
)
mean
upper limit
lower limi
50


intervals of




t
true n 10% error

100
120
d

n

(
1
0
-
3
)
20
40
60
80
0 5 10 15 20 25 30 35 40
Distance from upstream boundary (km)
I
d
e
n
t
i
f
i
140
e
160
b. Using observed stage data to identify n
Figure 5.16 Variation of the means of the identified n and their confidence
95% with noise level 1 . 0 = for the Z-Qz combination
For the Q-Z and Q-Q combinations, Figures 5.13 and 5.15 show that using stage
observed data generally gives very good quality in the identified coefficient when the
location of observation gauge is far enough from the downstream boundary (here about
one- fourth of the length of the channel). However, for the Q-Z combination when the
observation location is near the downstream boundary, using observed stage data gave
slightly poorer quality in the identified n (see Figure 5.13b). Figures 5.13a and 5.15a
137
Chapter 5 Numerical Experiments on Roughness Identification Problem
indicate that for these combinations, when using observed discharge data to identify
roughness coefficient, the means of the identified parameters deviated dramatically
from the true value and the confidence interval was greatly extended when the
observation gauge is located in the half upstream part of the channel. This indicates
that for these combinations, using discharge data with the observation gauge located in
the upstream part leads to an unidentifiable problem. This problem becomes more
ta, the means of the identified roughness coefficients were very close to
the true one and the confidence interval lies within
severe when the location of observation gauge approaches the upstream boundary.
For the combinations Z-Z and Z-Qz, Figures 5.14a and 5.16a show that by using
observed discharge data, the inverse model performed much better than by using
observed stage data especially for the combination Z-Z. For the cases of using
discharge da
10% error of the true value for all
ga
import erved
stage d the results e very poor. Figures 5.14b and 5.16b
illus the mean identified n dramatically deviated from
the true on terval when the
pstream part of the channel,
election
uge locations. This indicates that for this observed data, the gauge location is not an
ant factor affecting the quality of identified n. By contrast, when the obs
ata were used to identify n wer
trates clearly that for these cases,
e. The true value even lay outside the confidence in
observation stations located less than 5 km from the upstream boundary. These results
indicate that if the observation gauge is located in the u
using stage data for these combinations to identify the roughness coefficient causes the
unidentifiable problems.
In brief, the results obtained from the numerical experiments in this section indicated
that the s of a combination of boundary conditions is an important factor
affecting the quality of the identified roughness coefficients. These effects relate very
closely to the choice of the observed data type and locations of the observation gauge.
Improper choice of these combination factors could lead to unidentifiable problems.
The use of observed discharge data for the Q-Z and Q-Qz combinations and stage data
for the Z-Z and Z-Qz combinations when observation gauges are located in upstream
part of the channel should not be used for the identification problem. It is suggested
that for the Q-Z and Q-Qz combinations, observed stage data should be chosen, while
for the Z-Z and Z-Qz combinations, observed discharge data should be chosen.
However, for the Q-Z combination, when the observation station is very close to the
138
Chapter 5 Numerical Experiments on Roughness Identification Problem
downstream boundary, using observed discharge could provide better results than
using the observed stage.
5.4.4 Effect of errors in initial conditions
For unste nd stage
every node along the channel before the unsteady-flow computations can begin. In
practice, these ly vided thr w urface
profile computation or from inform of disch s an es at the es
along the channels. Hence rrors in l conditions are always unavoidable. This
secti inves the of th rs in i nd on the identified
roug ss coe t. A m to lim effect of these errors is then proposed.
The errors in initial cond cause in th uted charges. As a
result, the identified roughness coefficient biases from the lue. F tely,
from e theo chara tics, an ors in stima initial tions
propagate to i but t rrors w ome insignificant time. A rical
experiment was conducted where errors of 0.5 m in initial depths were introduced at
he of error in initial conditions with six different combinations of
boundary conditions (see Table 5.7). From this figure it is clear that for the four
) and Zoppou and O'Neill (1981). Therefore, even if the roughness
mputed
roughness coefficient would be in error if the initial conditions are incorrectly

ady flow modelling, the initial values of flow a must be specified at
initial values are usual pro ough a steady-flow ater-s
the ation arge d stag gaug
, the e initia
on tigates effects e erro nitial co itions
hne fficien ethod it the
itions errors e comp stages/dis
true va ortuna
th ry of cteris y err the e te of condi
nfinity hese e ill bec with nume
some cross-section along the channel. F gure 5.17 shows the computed stages that
indicate t effects
i
combinations Q-Z, Z-Z, Q-Qz and Z-Qz, the effects of errors in the initial conditions
become negligible after a certain number of time steps. However, this is not the case
for the Q-Q and Z-Q combinations where the errors in the initial conditions remain in
the reach for the whole computational time. These results also agree with those of
Cunge (1969
identification problem could work for these boundary combinations the co
estimated. This again shows the weakness of using the Q-Q and Z-Q combinations for
the roughness identification problem. The results again confirm that a discharge
hydrograph specified at the downstream boundary is not suitable for the roughness
identification problem.
139
Chapter 5 Numerical Experiments on Roughness Identification Problem
7.0
10.5
no error in the initial cond.
Q-Z boundary cond. and error in the initial cond.
8.
8.5
9.0
10.0
S
t
a
g
e

(
m
)
Z-Z boundary cond. and error in the intial cond.
Q-Q boundary cond. and error in initial cond.
Z-Q b y cond. and error in the initial cond.
9.5
oundar
Q-Qz boundary cond. and er the initial cond. ror in
Z ary cond. a he ini -Qz bound nd error in t tial cond.
0
7.5
0 10 20 30 40 50
Time (h)

Figure 5.17 Comparison of the computed stage hydrographs at the cross-section of
= x 30 km with different combinations of boundary conditions
From the results presented in Figure 5.17 it is expected that the effect of errors in
initial conditions on the identified roughness coefficient will be dissipated after a
number of computational time steps for the combinations Z-Q, Z-Z, Q-Qz and Z-Qz.
Figure 5.18 shows one example of the variation in the value of the identified roughness
coefficient with the starting time to calculate the objective function for the case of the
Q-Z combination. From the figure it can be seen that the identified roughness
coefficient converged quickly to the true value by increasing the starting time to
calculate the objective function. It reconfirms that to eliminate the effect of errors in
the initial conditions, the starting time to calculate the objective function can be lagged
after only a period of computation.
The starting time to minate the effect errors in the
initial condition of the channel (or friction losses). Figure
5.1 time to calculate the objective function can
be shorter for the channel with a higher s This is due to the influence of initial
conditions which are progressively d through friction (Cunge et 80).
Henc he erro initial tions a sipated er for ls wi gher
slope.
calculate the objective function to eli
s relates closely to the slope
9 demonstrates clearly that the starting
lope.
amped al. 19
e, t rs in condi re dis fast channe th a hi
140
Chapter 5 Numerical Experiments on Roughness Identification Problem
30
Initial stage 0.5m error=-
32
I
d
34
38
40
e
Initial stage 0.5m error=+
true
36
n
t
i
f
i
e
d

n

(
1
0
-
3
)
0 2 4 6 8 10 12
Starting time to calculate objective function (hours)
n

Figure 5.18 Identified n versus the starting time to calculate the objective function for
the case of the Q-Z combination
34
35
36
37
38
39
40
41
0 2 4 6
I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
S=0.001
S=0.0004
S=0.0001
true n
8 10 12
Starting time to calculate the objective function (hours)

the starting time to calculate objective
rrors in initial stage are +0.5 m)
results in this section indicat
entified roughness coefficients. However,
this problem can be eliminated by starting to calculate the objective function after a
mputation. The higher the channel slope the shorte
before starting to calculate the objective function. It is suggested that for channels with
Figure 5.19 Variation of identified n with
function and the slope of the channel (the e
In brief, the e that the errors in initial conditions are
unavoidable and can cause biases in the id
period of co r the time necessary
141
Chapter 5 Numerical Experiments on Roughness Identification Problem
a high slope ( 001 . 0
0
S ) the time before starting to calculate the objective function is
only 3-5 hours. For channels with a small slope ( 0001 . 0
0
S ), this time should be
longer. The time of 10-20 hours is suggested for these channels.
5.5 Roughness identification in compound channels
In flood routing in natural rivers, many channels have compound sections and the
values of the roughness coefficients in the main channel and in the floodplains are
usually different. In order to achieve more accurate results, the roughness of main
channel and floodplains should be considered separately. In this section, the roughness
monstrate the
performance of the model. The true values of the roughness coefficients of the main
channel ( ) and of the floodplain ( ) are 0.028 and 0.042 respectively. The
upstream boundary condition was a discharge hydrograph and the downstream
boundary condition was a stage hydrograph. The observed data were the simulated
stage at the middle cross-section of the channel. For all computations, the initial values
of both the roughness coefficients were assumed to be 0.02, which were quite different
from the true ones.
The roughness coefficients were computed with different cases for the observed data
with and without noise. When noise was contained in the observed data, four different
noise levels of , 0.10, 0.15 and 0.20 were considered. For each noise level, 10
samples were computed. The computed results of the identified roughness coefficients
with different noise levels are shown in Table D.6 in Appendix D, and the summary of
the results
upper and lower limits at a 95% confidence level are given in Table 5.11. Similar to
identification problem is extended to compound channels. The performance of the
model developed in Chapter 4 is assessed and certain factors related to the quality of
the identified roughness coefficients are investigated.
5.5.1 Model performance
In this case, the model channel, illustrated in Figure 5.1b, was used to de
c
n
f
n
05 0. =
including the mean values of the identified roughness coefficients and their
the single channel, the computed results indicate that for the case of the noise-free
observed data, the identified roughness values converged to the true ones. However,
142
Chapter 5 Numerical Experiments on Roughness Identification Problem
when data contain noise, for any given noise level, the individual values of the
identified roughness coefficient differed from the true ones. The computed results also
indicate that the higher the noise level is the wider the range of identified parameters.
However, these identified values scatter around the true ones and the average values
are very close to the true ones, even at a high noise level.
Ta of
95% with different noise levels
ain channel roughnes
ble 5.11 Means of identified roughness coefficients and their confidence intervals
s n Floodplain roughness
c

f
n M
Noise
level Lower
limi
Upper
limit
Lower
limit
Upper
limit
Mean Mean
t
0.05 0.02800 0.02774 0.02827 0 7 0.04042 04353 .0419 0.
0.10 0.02802 0.02748 0.02855 0 0.03887 04523 .04205 0.
0.15 0.02802 0.02724 0.02881 0 0.03759 04728 .04244 0.
0.20 0.02802 0.02699 0.02905 0 0.03651 04925 .04288 0.
No 0.0280 0.0420026 noise 002* *
Note: the true values of and are 0.028 and 0.042 respectively. * for noise-free data, the
results show up to 7 decimals to indicate the accuracy of the identified roughness coefficients
obtained from the roughness identification model.
Figure 5.20 also depicts the m
their upper and lower limits using 10 samples for each
identified roughness coefficient in the floodplain.

c
n
f
n
ean values of the identified roughness coefficients and
at a 95% confidence level by
noise level. From this figure it can be clearly seen that the mean values are very close
to the true ones. Their ranges at a 95% confidence level contain the true values of the
roughness coefficients. As expected, the lower the noise level, the narrower the
confidence interval, indicating the more accurate the ability to identify the roughness
coefficient. It can be seen from this figure that for the main channel, the range of the
means of the identified roughness coefficients at a 95% confidence interval are
significantly narrower than for the floodplain. This indicates that the quality of the
identified roughness coefficient in the main channel is better than the quality of the


143
Chapter 5 Numerical Experiments on Roughness Identification Problem







Figure 5.20 The means of identified roughness coefficients and their range at a 95%
confidence level with different noise levels
In summary, in this section, several numerical experiments were conducted to assess
the ability of the model to identify roughness coefficients for compound channels. The
computed results indicated that when observed data were free from noise, the model
can identify very accurately the values of roughness coefficients of the main channel
and of the floodplain. When observed data were contaminated with noise, although the
individual values of id
25
30
35
40
45
50
55
I
d
n
t
i
f
i
e
d

n

(
1
0
-
3
)
0 0.05 0.1 0.15 0.2
e
Noise level
mean of the identified
mean of the identified
true roughness coefficients
n
c
n
f
entified roughness coefficients deviated from the true value,
they scattered around the true value and their means were very close with the true one
even with high noise level. The higher the noise level, the higher is the dispersion level
in identified parameters (the wider the ranges of the means of identified and at a
95% confidence level). For the sa odel can provide a better quality
c f
me noise level, the m
n n
for the main channel roughness coefficient than the floodplain one.
5.5.2 Factors affecting the quality of the identified roughness
coefficients in compound channels
In Sections 5.3 and 5.4, several factors affecting the quality of the identified roughness
coefficient were investigated. However, for compound channels, beside those factors,
there are other typical factors that affect the quality of the identified roughness
coefficients. This section will investigate factors affecting the quality of the identified
144
Chapter 5 Numerical Experiments on Roughness Identification Problem
roughness coefficients, including the effects of the inundated level on the floodplain in
term of peak discharge and observation interval.
5.5.2.1 Effect of the inundated level in terms of peak discharges
In s
coefficient ( is iden ile the floodplains are
k discharge (m3/s)
the roughness identification problem, the value of the floodplain roughnes
tified based on the observed data wh
f
n )
inundated. Hence, it is believed that the inundated level and duration in the floodplain
must affect the quality of the identified roughness coefficients. The inundated level in
the floodplain depends on the inflow peak discharge. The larger the peak discharge,
the higher the inundated level and the longer the flooded time. In order to investigate
the effect of inundated level on the quality of the identified roughness coefficients,
several numerical tests were conducted with three inflow peak discharges of 700,
1000, and 1500 m
3
/s for the same model channel (Figure 5.1b) with the same
observation interval of 1 hour. The corresponding maximum depths and flooded
duration in the floodplains at the observation gauge (at the middle section) of these
floods were 0.86, 1.64 and 2.64 m, and 25, 34 and 41 hours respectively.
Table 5.12 shows the identified roughness coefficients of the main channel (
c
n ) and
floodplain (
f
n ). The computed results indicated that when there was no noise
contained in the observed data, the model could obtain very accurate values of
roughness for the compound channel. The identified roughness coefficients converged
to the true values even when the flooded level in floodplains was small as in the case
of a peak discharge of 700m
3
/s.
Table 5.12 Identified roughness coefficients for noise-free data with different peak
discharges (the true values of
c
n and
f
n are 0.028 and 0.042 respectively)
Pea
p
Q =
p
Q 700 =
p
Q 1000 1500 =
p
Q
Main channel identified coeff.
c
n 0.0280003 0.0280008 0.0280002
Floodplain
0.0420070 0.0419866 0.0420026
f
n identified coeff.
Note: the results show up to 7 decimals to indicate the accuracy of the identified roughness
coefficients obtained from the roughness identification model.
145
Chapter 5 Numerical Experiments on Roughness Identification Problem
However, as may be expected for the data with noise, the individual identified
roughness coefficient deviated from the true value. Table 5.13 presents the computed
roughness coefficients of the main channel and floodplain for the data with the noise
level of 05 . 0 = . From this table it can be seen that while the identified roughness
coefficients of the main channel for different floods were quite close to the true one
(the m error was less than 2%), the identified roughness coefficients of the
e re alue of floodplain roughness; (iii) for higher
water levels in the e floodplain part becomes more
im becomes more sensitive to the stages/discharges.
Table 5.13 Errors in the identified roughness coefficients of the main channel and
floodplains when data contain noise (
aximum
floodplain were deviated more from the true one. The computed results indicate that
the larger the peak discharge (i.e. the higher the flooded level and the longer the
inundated time), the more accurate the identified floodplain roughness coefficient. This
can be attributed to: (i) the value of
f
n is less sensitive to stages and discharges than
the value of
c
n when the water depth in the floodplain is small; (ii) for flood events
with small peak discharge, the inundated time is too short, which mean that when data
contain noise the number of observation data points during flooding time is too few
and is not enough to reflect th al v
floodplain, the conveyance of th
portant and the value of
f
n
05 . 0 = )
Error in the identified (%) Error in identified (%)
c
n
f
n
Sample
700 1000 =
p
Q =
p
Q =
p
Q 1500 =
p
Q 700 =
p
Q 1000 1500 =
p
Q
1 1.18 1.07 2.00 -2.41 -9.79 -6.44
2 -1.09 -0.27 -0.56 6.67 -0.94 0.10
3 -0.77 -0.69 -1.43 13.28 12.19 8.66
4 -1.75 -1.16 -0.92 26.46 7.37 4.11
5 1.32 1.57 0.76 -22.17 -12.67 -4.75
6 0.04 -0.14 0.29 -3.68 -2.16 -2.95
7 1.35 0.91 0.58 -20.07 -7.38 -4.84
8 -1.46 -1.37 -1.02 29.87 9.43 3.34
9 1.32 1.38 2.00 -18.78 -6.38 -3.40
10 -1.05 -0.92 -1.61 29.96 8.30 5.52
Note: % error in identified n is defined as % 100 . / ) (
true true comp
n n n ; Values greater than 10%
are in bold. Peak discharge
p
Q in m
3
/s
146
Chapter 5 Numerical Experiments on Roughness Identification Problem
Therefore, the selection of flood events is one of important factors that affects the
quality of identified roughness values of a river reach, especially for reaches with
floodplains. It is suggested that using observed data of flood events with higher peak
discharges could provide more accurate results on the identified floodplain roughness
coefficient.
5.5.2.2 Effect of data availability in terms of observation intervals
To investigate the effect of observation intervals on the quality of identified
n of e pl
parameters, several intervals with the same inflow hydrograph of 1500 =
p
Q m
3
/s were
considered including 1 hour, 2 hours and 4 hours. Table 5.14 shows the computed
roughness coefficients of the main channel a d th flood ain obtained from the
model. The results in the table illustrate that when data have no noise the identified
roughness coefficients converged to the true values with no matter what the value of
the observation interval.
Table 5.14 Identified roughness coefficients for noise-free data with different
observation intervals (the true values of
c
n and
f
n are 0.028 and 0.042 respectively)
Observation interval
ob
I
=
ob
I 1 hour =
ob
I 2 hours = 4 hours
ob
I
Main channel identified coeff. 0.0280002 0.0280003 0.0280003
c
n
Floodplain identified coeff.
f
n
0.0420026 0.0419974 0.0420002
Note: the results show up to 7 decimals he accuracy of the identified roughness
coefficients obtained from the roughness identification model.
S dual identified
v lts in Table 5.15 sho t although the
identified roughness coefficients of the main channel were rather accurate (the
aximum error was less than 3.2%) the identified roughness coefficients of the
floodplain deviated much more from the true values. From this table it can be seen that
to indicate t
imilar to the previous cases, when data contained noise, the indivi
alues deviated from the true ones. The resu w tha
m
for the same flood event with the same noise level, the smaller observation intervals
provide a better quality of the identified parameters. For example, when the
observation interval was 1 hour the relative maximum error of the identified
f
n for 10
samples was less than 9%. But when the observation interval was 4 hours the relative
maximum error of the identified
f
n was up to 15%. The results obtained from these
147
Chapter 5 Numerical Experiments on Roughness Identification Problem
numerical tests indicate that the quality of the identified roughness coefficient of the
floodplain could be improved by using the observation data of flood events with a
higher frequency of observation.
Table 5.15 Errors in the identified roughness coefficients of the main channel and
floodplain when data contain noise ( 05 . 0 = )
Error in the identified (%) Error in the identified (%)
c
n
f
n
Sample
=
ob
I 1 2 =
ob
I =
ob
I 4 =
ob
I 1 =
ob
I 2 4 =
ob
I
1 2.00 2.19 -1.19 -6.44 -7.61 0.75
2 -0.56 0.64 0.00 0.10 -5.93 -7.83
3 -1.43 -1.00 -0.96 8.66 11.05 14.83
4 1 -0.92 -1.70 -1.87 4.11 1.72 11.3
5 0.76 1.02 -0.79 -4.75 -5.28 2.99
6 0.29 0.78 0.00 -2.95 -3.79 -2.97
7 0.58 -0.41 2.75 -4.84 -2.12 -13.68
8 -1.02 -0.87 3.20 3.34 4.26 -14.49
9 2.00 1.03 0.80 -3.40 -2.53 3.97
10 -1.61 -1.68 -0.56 5.52 3.71 0.98
Note: % error in identified n is defined as % 100 . / ) (
true true comp
n n n ; Observation interval
in hours. Values greater than 10% are in bold.
5.6 Identification of roughness functions
During a flood event, the water level in a river usually varies in a wide range. If the
roughness conditions in the river are changed considerably, the roughness coefficients
for this case are stage dependent. In order to investigate the ability of the model to
identify the roughness coefficients when their values varies with stage, several
maximum water stage is below the bankfull level) and a compound channel case.
Two model channels with the cross-sections shown in Figure 5.1 were used for a
single channel and a compound channel. The upstream boundary condition was a
discharge hydrograph as defined in Equation (5.1), and the parameter values of the
hydrograph are shown in Table 5.1. The downstream boundary condition is a stage
hydrograph. The observed water stages at the intermediate section of the channel were
ob
I
numerical tests were conducted for both cases: a single channel case (when the
148
Chapter 5 Numerical Experiments on Roughness Identification Problem
used to identify the roughness function. The downstream boundary condition and the
observed data were generated with the true roughness function(s). Identical initial and
boundary conditions were applied while obtaining the simulated observation data and
while solving the roughness identification model.
The true roughness functions were presented as functions of stage as follows:
For the main channel roughness coefficient:
( ) ( )
2
0 0
0001 0 0015 0 032 0 Z Z . Z Z . . n
c
+ = (5.3)
For the floodplain roughness coefficient:
( ) ( )
2
0 0
001 0 002 0 043 0 Z Z . Z Z . . n
f
+ = (5.4)
5.6.1.1 For a single channel
For single channels, the true roughness function of the channel was defined by
Equation (5.3). The identification procedure starts with initial assumed coefficients in
Equation (4.33). For all computations, for simplicity, the initial value of n was
assumed to be a constant and equal to 0.02. It means that the initial assumed values of
coefficients
1
a ,
2
a ,
3
a were 0.02, 0.0 and 0.0 respectively. Several test runs were
conducted for the cases when the observed data were with and without noise. When
noise was contained in the observed data, a noise level of 5 . 0 = was considered.
case of the noise-free observed data, the
unction converged to the true ones. The

The computed results indicate that for the
identified coefficients of the roughness f
computed coefficients
0
a ,
1
a and
2
a of the roughness function were 0.031997,
-0.0014977 and 0.0000995. Figure 5.21 shows the true roughness function and the
identified roughness function when the data contain no noise. From this figure it can
be seen that the computed roughness curve is identical with the true one.
149
Chapter 5 Numerical Experiments on Roughness Identification Problem
1
2
3
0.020 0.025 0.030 0.035 0.040 0.045
Roughness n
D
e
p
4
t
h
5
6
7

(
m
)
True n
Identified n

Figure 5.21 The true roughness and identified roughness curves
with noise-free data
To n
observe
coeffic
of the r
were d e 5.22 shows that the identified roughness
cur
Tab
co sider the case when data contain noise, five noisy samples at a noise level of
were generated from different seeding random numbers and added to the true
d data, as described in Section 5.2.2.2. As expected, for this case the identified
ients deviated from the true ones. Table 5.16 presents the identified coefficients
oughness function for five noisy observed data sets. Although these coefficients
ifferent from the true ones, Figur
5 . 0 =
ves deviated around but were still close to the true curve.
le 5.16 Identified coefficients of roughness function when data contain noise
Identified coefficients (10
-3
)
Sample
0
a

1
a

2
a

1 31.92 -1.82 0.20
2 30.29 0.58 -0.35
3 32.06 -0.88 -0.16
4 33.06 -2.26 0.28
5 31.85 -1.06 -0.06
True value (10
-3
) 32.00 -1.5 0.10

150
Chapter 5 Numerical Experiments on Roughness Identification Problem
7
1
2
3
4
5
6
0.020 0.025 0.030 0.035 0.040 0.045
Roughness n
D
e
p
t
h

(
m
)
true n
n 1
n 2
n 3
n 4
n 5

Figur
The ho er
stages of the observed data. This figure shows clearly that in general the identified
entioned above, two separate roughness functions
(5.3) and (5.4). Similar to the single
e 5.22 The true roughness and identified roughness curves when data contained
noise (5 noisy data samples)
rizontal dash lines in Figure 5.22 indicate the minimum and maximum wat
roughness inside the range of observed stage were much closer than the outside of this
range. The higher deviation of the identified roughness curve from the true one in the
regions outside of the observed stages indicates that the extrapolation of the roughness
coefficient in these regions is highly uncertain. Therefore, care should be taken when
using the identified roughness function obtained from a certain flood for routing other
larger flood events.
5.6.1.2 For a compound channel
For compound channels, as m
(Equations (4.33) and (4.34)) need to be identified one for the main channel and one
for the floodplain.
Several test runs were conducted to investigate the ability to identify the roughness
functions of the inverse model for the compound channel. For this case, the true
roughness functions were defined by Equations
channel case, the initial values of the roughness coefficients of the main channel and
floodplain were assumed to be constants of 0.02 and 0.03 respectively. This means that
151
Chapter 5 Numerical Experiments on Roughness Identification Problem
the initial coefficients of the main channel roughness function (
1
a ,
2
a ,
3
a ) were (0.02,
0.0, 0.0) and the initial coefficients of the floodplains (
1
b ,
2
b ,
3
b ) were (0.03, 0.0,
0.0). The observed data were considered for two cases: with and without noise. When
the noise was contained in observed data, the noise level of 5 . 0 = was considered.
The results indicate that when there is no noise contained in the observed data, the
model can properly identify the roughness functions. Although there seems to be a
difference in the coefficients between the identified function and the true functions
(Equations (5.3) and (5.4)), the computed roughness curves are very close to the true
curves especially for the main channel, as illustrated in Figure 5.23. From this figure it
also can be seen that, when the water is above bankfull level, the computed roughness
curves deviated slightly from the true ones. This may be attributed to the compensation
of the conveyance between the main channel and the floodplains because when the
water level is greater than bankfull level at a cross-section the two roughness values
can offset each other.
0
1
2
3
4
5
6
7
8
m
)
0.020 0.025 0.030 0.035 0.040 0.045 0.050 0.055 0.060
Roughness n
D
e
p
t
h

(
n true
n identified
bankfull level

Figure 5.23 The true roughness curves and identified roughness curves
with noise free-data
To consider the case when data contain noise, five noise samples of were
generated from different seeding random numbers. Figure 5.24 shows the true
roughness curves and computed roughness curves n
1
, n
2
, n
3
, n
4
and n
5
with five noisy
n
c
n
f
5 . 0 =
152
Chapter 5 Numerical Experiments on Roughness Identification Problem
observed data sets respectively. From this figure it can be seen that for the main
channel, the roughness curve can be identified rather accurately (the maximum
deviation from the true value was less than 8%). The trends of the identified roughness
. However, for the floodplain, the roughness
re from the true one. The computed results also indicate that there
was some compensation between the values of the main channel and floodplain
curves were very close to the true one
curves deviated mo
roughness coefficients.
0
1
2
3
8
4
D
e
p
t
h

(
5
6
7
m
)
0.020 0.025 0.030 0.035 0.040 0.045 0.050 0.055 0.060
Roughness n
n true
n 1
n 2
n 3
n 4
n 5
bankfull level
ness curves when data
)
l numerical test runs were conducted to investigate the
ability to identify the roughness function(s) for a single channel and a compound
channel the model can identify the
ied roughness curves
were scattered around the true ones. However, the results indicate that the inverse
model can identify the roughness curve more accurately for the main channel than for
n
f
n
c

Figure 5.24 The true roughness curves and identified rough
contained noise (5 noisy data samples
In brief, in this section severa
channel. The results indicated that for the single
roughness function properly even when data contain noise. The results also indicates
that the extrapolation of the roughness coefficient value outside of the observed range
of depth should be done with caution because outside the observation depth range the
identified roughness curve usually deviates more from the true curve. For the
compound channel, although the identified roughness functions for the main channel
and floodplain were not the same with the true ones, the identif
153
Chapter 5 Numerical Experiments on Roughness Identification Problem
the floodplain. Also, there is some compensation of the roughness values between the
main channel and the floodplain. When data contain errors, the roughness function for
the floodplain can be obtained with a higher degree of uncertainty compared to the
roughness function of the main channel. Accordingly, the constant value of floodplain
ughness should be enough for practical purpose. This will be verified for the real
ata in Chapter 6. Moreover, for compound channels with roughness variation, the
number of unknowns is large because both roughness functions need to be identified.
order to avoid this disadvantage, an alternative method, described in Section 4.3.6 in
Chapter 4, will be tested in the following section, where a conveyance function will be
entified.
Identification of conveyance function
order to investigate the ability of the model to identify the conveyance function of
erical tests were conducted. The channel geometry and roughness
characteristics as well as the boundary conditions and observed data were used from
e case of Section 5.6.1.2. In this case, a conveyance function is considered for two
cases: a cubic polynomial function (Equation (4.35) in Section 4.3.6) and a power
nction (Equation (4.36)). It will be identified instead of two roughness functions.
conditions were applied while obtaining the simulated
observation data and while solving the inverse model.
Firstly, the model was tested for the case where the data were free from noise. Figure
is indicates that the cubic polynomial function performs
ro
d
In
id
5.7
In
stage, several num
th
fu
Identical initial and boundary
5.25 shows the true conveyance curve which is calculated from the true roughness
functions (Equations (5.3) and (5.4)) and the identified conveyance curves for both
cases using the cubic polynomial function and the power function. From this figure it
can be seen that there is a good agreement between the identified conveyances and the
true one. This figure also illustrates that the identified conveyance obtained from the
case using the polynomial function was closer to the true one compared to the case
using the power function. Th
better than the power function.
154
Chapter 5 Numerical Experiments on Roughness Identification Problem
8000
m
3
/
s
)
0
20000
40000
0 1 2 3 4 5 6 7 8 9
Depth (m)
C
o
n
v
60000
e
y
a
n
c
e
0
100000

(
120000
K true
K identified using the polynomial function
K identified using the power function

Figure 5.25 The true conveyance and identified conveyance curves
with noise-free data
One of the main advantages of the alternative method by using the conveyance
function instead of using the roughness functions for compound channels is that the
effects of variation of friction and cros

s-section can be presented by a single function.
spectively of the computation time by using
the roughness functions. However, although the computation time of using the power
ct of data noise on the identified conveyance function, five noise
samples of generated from different seeding random numbers were used.
identified conveyance curves were very close to each other and to the true one. Again,
This can reduce the number of unknown parameters that need to be identified and thus
reduce the computation time. Table 5.17 presents the computation times for identifying
the roughness functions and the conveyance functions. The inverse model using
conveyance functions converged much faster towards the optimum. From this table, it
can be seen that the computation time by using the polynomial and power conveyance
functions are only a half and one third re
function is smaller than using the polynomial function, the identified conveyance is
usually poorer, as mentioned above.
To investigate the effe
5 . 0 =
Figures 5.26 and 5.27 show the true conveyance computed from the true roughness
functions and the identified conveyance functions
1
K ,
2
K ,
3
K ,
4
K and
5
K corresponding to five noisy observed data sets by using the polynomial function
and the power function respectively. Amazingly, although data contains noise the
155
Chapter 5 Numerical Experiments on Roughness Identification Problem
these figures illustrate the better performance of the polynomial function. The results
indicate that the model could identify the conveyance functions even when data
contain noise. The results for these illustrative problems indicate the potential
app a oblems.
Table es by using the roughness functions and the conveyance
Function Computation time (s)
lic bility of the conveyance identification model for practical pr
5.17 Computation tim
functions
Roughness functions (Eqs. (4.33) and (4.34))
284
Polynomial conveyance function (Eq. (4.35))
148
Power conveyance function (Eq. (4.36))
92

To illustrate more of the performance of the conveyance identification model, Figure
5.28 t he
computed hydrographs using the identified roughness functions and the identified
conveyance functions at 10 km fr the dow eam
hy raph cat ect c
identifying param
were generated by the given ess f , the rap ned e
pol om ct very to t nes rfo of
the e co ce fu is a s go he d
rou ne the con e fun the aph ly
dev e . H , for al pu thes ma e
acc ta se e is t, as ned above.
shows the observed hydrographs (using the rue roughness functions), t
om
ion were no
nstr
t involved
end of the reach. These observed
in the obj drog s at this lo ive fun tion during
eters. From this figure it can be seen that although the observed data
roughn unctions hydrog hs obtai from th
yn ial conveyance fun ion are to close he true o . The pe rmance
id ntified polynomial nveyan nction lmost a od as t identifie
gh ss functions. For power veyanc ction, hydrogr s slight
iat d from the true one owever practic rpose, e results y still b
ep ble. Also, for this ca the computation tim the leas mentio

156
Chapter 5 Numerical Experiments on Roughness Identification Problem
0
20000
40000
0 1 2 3 4 5 6 7 8 9
Depth (m)
C
o
n
v
e
y
60000
n
c
e

(
80000
a
m
3
/
s
100000
120000
)
K true
K 1
K 2
K 3
K 4
K 5

Figure 5.26 The true conveyance and iden ified convey t ance curves by using the
polynomial function when data contained noise (5 noisy data samples)

0
20000
40000
60000
0 1 2 3 4 5 6 7 8 9
Depth (m)
C
o
n
v
e
y
a
n
c
e
80000
100000
120000
3
/
s
)

(
m
K true
K 1
K 2
K 3
K 4
K 5

Figure 5.27 The true conveyance and identified conveyance curves by using the power
function when data contained noise (5 noisy data samples)

157
Chapter 5 Numerical Experiments on Roughness Identification Problem
8
9
10
11
12
S
t
a
g
e

(
m
)
Z observed
Z - roughness functions
Z - K polynomial function
Z - K power function
7
6
0 20 40 60 80 100
Time (hours)

100
300
500
700
900
1100
1300
1500
0 20 40 60 80 100
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Q observed
Q - roughness functions
Q - K polynomial function
Q - K Power function

Figure 5.28 Observed and simulated hydrographs computed from the identified
roughness functions and the identified conveyance functions
In brief, the results obtained from this section indicate that the alternative approach of
sing the conveyance function is a good alternative for identifying parameters for
flood routing purposes. Two conveyance function types were tested: a cubic
olynomial function and a power function. The results indicated that the computation
me for the power function was shorter, but the performance of the cubic polynomial
function is better. The results of these illustrative problems indicate the potential
pplicability of the model for practical problems for compound channels where the
cross-section has abrupt changes in width and roughness.
u
p
ti
a
158
Chapter 5 Numerical Experiments on Roughness Identification Problem
5.8 Summary and conclusion
this chapter, the performance of the roughness identification model developed in
Chapter 4 was tested by using synthetic data where the true roughness coefficient(s) is
efined. As in the field the true value of the roughness coefficient is not known, this is
data make it possible to have comparisons
between the identified with the true . This can provide the abilities to evaluate the
erformance of the model and assess factors affecting the quality of the identified
roughness coefficient.
Firstly, preliminary numerical experiments were conducted to investigate the abilities
f the model to identify the roughness coefficient. The results indicated that when data
ontain no noise the model can identify the roughness coefficient very accurately.
When data contain noise, the identified roughness deviated from the true one, but it
cattered around the true one and was very close to it.
econdly, using this model, several numerical test runs were conducted to investigate
the effects of the modelling factors on the quality of identified roughness coefficient.
hese factors include the effects of weighting coefficient , computational grid sizes,
ifferent combinations of boundary conditions, and errors of initial condition related to
different combinations of boundary conditions. The results indicated that
nderstanding these modelling factors is very important to avoid unidentifiable
problems and to improve the quality of the identified parameters. Some conclusions
nd recommendations drawn from these investigations are as follows:
The effect of on the identified roughness coefficient is insignificant compared
to the effect of data errors. However, in order to limit this effect, the lower
values of are suggested especially for the rougher or low gradient reaches and
for the flood events with a rapid rise. Using smaller time steps is one good
solution to reduce effect of on the biases of the identified roughness
coefficients.

ld be
appropriate to describe the prototype and avoid numerical instabilities,
In
d
the advantage of using synthetic data. These
n n
p
o
c
s
S
T
d
u
a
The selection of grid sizes is one of the important factors that affect the quality
of identified roughness coefficient. The selection of grid sizes shou
159
Chapter 5 Numerical Experiments on Roughness Identification Problem
especially for the cases with high slope channels or flood events with a short
time to peak.
This study indicated that using a discharge hydrograph at a downstream
boundary is not appropriate for the roughness identification problem. Four
combinations of boundary conditions suggested for the inverse problem are Q-
Z, Z-Z, Q-Qz and Z-Qz. However, for channels with a small slope, there is
some deviation from the true value when the Q-Qz and Z-Qz combinations are
used to identify the roughness coefficient. Hence, in general, the boundary
condition combinations Q-Z and Z-Z are the most appropriate for the
roughness identification problem.
The effects of a certain combination of boundary conditions relate very closely
to the choice of observed data type and locations of the observation gauge.
Improper choice of these combination factors can lead to unidentifiable
problems. The use of observed discharge data for the Q-Z and Q-Qz
combinations and stage data for the Z-Z and Z-Qz combinations when the
observation gauge is located in the upstream part of the channel should not be
used for identification problem. However, for the Q-Z combination, when the
observation station is located close to the downstream boundary, using
observed discharge could provide better results than using observed stage.
The biases in the value of the identified roughness coefficient due to the errors
in initial conditions can be eliminated by starting to calculate the objective
function after a few computational time steps. The higher the channel slope the
shorter the time necessary before starting to calculate the objective function.
he identification problem was also extended to compound channels, for this case two
arameters need to be identified. The computed results indicated that when observed
data were free from noise, the model can identify very accurately the values of
ughness coefficients of the main channel and the floodplains. When observed data
with noise, although the individual values of the identified roughness coefficient
eviated from the true value, they scatter around the true value and the average value
w
quality an in
the floodplain.
T
p
ro
d
as very close with the true one even at a high noise level. At the same noise level, the
of the identified roughness coefficient in the main channel was better th
160
Chapter 5 Numerical Experiments on Roughness Identification Problem
The quality of the identified roughness coefficients of compound channels is also
dependent on the selection of flood events. In order to improve the quality of the
entified roughness of the floodplain, flood events with higher peak discharges and
ith higher frequency of observation could provide more accurate results of the
identified roughness coefficient.
For the cases of the roughness variation with depth, the roughness function(s) was
rmulated as quadratic polynomial function(s). The coefficients of the roughness
functions were identified. The results indicate that, when data were free from noise, the
entified roughness curves were identical or very close to the true values. As
xpected, when data contain noise, although the identified roughness curves were
biased from the true ones they scattered around the true roughness curves. The results
lso indicate that the main channel roughness curve could be more accurately
identified than the floodplain roughness curve. Also, there is some compensation in the
alues of the roughness coefficients between the main channel and the floodplains.
oreover, in cases when the channel is a prismatic channel or the cross-sections along
uch, the more practical concept of conveyance
id
w
fo
id
e
a
v
M
K the channel do not change m was
tested. The conveyance variation with stage was formulated as a cubic polynomial
nction or a power function. The conveyance identification model converged to the
optimum values faster than the roughness identification model when two roughness
nctions need to be identified. The results indicated that the cubic polynomial
med better than the power conveyance function for both data with and
without noise. However, the power function required less computation time.
All the findings from this chapter were obtained from synthetic data. The application
f the roughness identification problem to natural rivers is still limited. Therefore, in
the next chapter, the roughness identification model will be applied to natural rivers
specially for cases with compound channels.
fu
fu
function perfor
o
e

161
Chapter 6
Chapter 6
Application of the roughness identification
6.1 Introduction
In Chapter 5, several factors affecting the quality of the identified roughness
oefficient (Mannings n) have been investigated using synthetic data.
Recommendations to improve the quality of the identification problem have been
rawn. The study indicated the applicability of the model to natural rivers. Therefore,
is chapter, the model will be applied to two natural rivers especially for the cases
with compound channels. The first river is the Goulburn River, Victoria, Australia and
e second one is the Duong River in the North of Vietnam.
.2 Goulburn River
.2.1 General description of study area
ent consists of an extensive network of streams in central
Victoria extending from the high range of the Great Dividing Range near Woods Point
the Murray River near Echuca. The catchment area covers an area of 16,192 km
2
,
accounting for 7.1% of the total area of the state of Victoria (VIC). The mean annual
ischarge from the catchment is 3,040 million cubic metres, which is 13.7% of the
tota
problem to natural rivers
c
d
in th
th
6
6
The Goulburn River catchm
to
d
l runoff in the state (Rural Water Commission of VIC 1987).
162
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
The study reach belongs to the Lower Goulburn River part. This reach is 37.8 km long
from the Shepparton gauge (405204) to Loch Garry gauge (405276), as shown in
There are extensive levee works confining overbank flooding. Levees have tended to
Figure 6.1.










Figure 6.1 Study reach of the Goulburn River and the gauging stations along the reach
The study reach has a very gentle slope with an average bed slope of 0.00005 or
5cm/km. The present course of the Goulburn River within the study area, natural
levees and creek systems reflect a long history of change (Rural Water Commission of
VIC 1987). The river channel is meandering. The bedload is mud and clay. The
floodplain near the main channel is state forest with extensive area of woodlands. The
forest along this reach has an overstorey of River Red Gum and an understorey of
tussock and wallaby grasses.
G
o
u
l
b
u
r
n
R
i
v
e
r
G
o
u
l
b
u
r
n
R
i
v
e
r
follow the meanders and interfere with flood flows across the bends in the main river
channel. Levees have also been constructed to block important distributary flow paths,
resulting in higher flood levels elsewhere in the flood plain or the main channel.
163
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
Incorporated into the levee banks are flood outlet structures, the most important of
which is the Loch Garry regulator.
6.2.2 Data
6.2.2 tional
The cross sectional data of the Goulburn Rive ere obtained from
carried out by the State Rivers and W
1981. The distances between every cross section range from 1.0 to 2.0 km. The cross
ection data surveyed in 1977 are mainly for the main channel part between distances
136.8 km to 182.8 km from the Murray River. The cross section data surveyed in 1981
re mainly for the flood plain from the junction with the Murray River to 135.5 km
from this river. For this survey, the main channel cross section was surveyed using a
ounding method. The cross-sectional data of this study reach were extracted from
ese two sources between distances 120.63 km to 157.9 km from the Murray River
including 27 surveyed cross-sections.
The works consists of an earthen embankment around Loch Garry at an average height
f about 2 m to a maximum of 3 m. The embankment extends for approximately 10
km (Rural Water Commission of VIC 1987). A complete mapping coverage of the
tudy area at scale of 1:10,000 with 0.5 m contour interval was used to complete the
oodplain part of the cross section data survey in 1977 and the area of the Loch Garry
storage.
The details of the positions and bed elevations of the cross sections are given in Table
.1. Ten representative cross sections are illustrated in Figure E.1 in Appendix E.
6.2.2.2 Flow data
The flow data were supplied by Thiess Environmental Services Pty Ltd. There are
three gauges located in this reach which are the Shepparton gauge (405204), the
Shepparton Golf Club gauge (405217) and the Loch Garry gauge (405276). For the
gauging stations at Shepparton and Loch Garry, the data are available for both stage
and discharg .
.1 Cross-sec data
r w the detailed surveys
ater Supply Commission in May 1977 and April
s
a
s
th
o
s
fl
E
e. The discharge data of these gauges are obtained from the rating curves
164
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
But for the Shepparton Golf Club gauge, only measured stage is available. The
observation times for these gauges are as follows:
The Shepparton gauge (405204): from June 1921 to date,
The Shepparton Golf Club gauge (405271): from June 1977 to 1982 (only stage
data), and
The Loch Garry gauge (405276): from February 1978 to date.
Based on the availability of flow data for this reach as well as the time of the cross
ection survey, different major flood events which occurred during 1978 to 1981 were
chosen for this study. Because numerous levee failures occurred during July 1981,
nly the flood event from 25/6/1981 to 5/7/1981 was considered. Seven flood events
hich have been selected for this study are shown in Table 6.1.
Table 6.1 Observed peak discharge and stage of the selected flood events
No. Flood event
at
Shepparton
(m
3
/s)
at
Shepparton
(m)
at
Shepparton
Golf Club
(m)
at
Loch
Garry
(m
3
/s)
at
Loch
Garry (m)
s
o
w
p
Q
p
Z
p
Q
p
Z
p
Z
p
Q
p
Z
1 21/07 - 06/08/1978 290.11 109.30 108.61 257.81 105.35
2 08/08 - 31/08/1978 366.01 109.86 109.34 329.50 105.98
3 27/09 09/10/1978 338.08 109.71 109.08 298.21 105.78
4 04/09 25/09/1979 302.59 109.41 108.78 275.30 105.54
5 28/09 - 31/10/1979 522.24 110.35 109.90 438.99 106.47
6 29/07 05/08/1980 191.18 107.90 107.31 184.20 104.07
7 25/06 05/07/1981 447.29 110.18 109.69 335.82 106.02
Note: Stage follows AHD (Australian Height Datum)
he operation of the Loch Garry regulator is governed by a set of rules relating to the
oval of bars in the regulator back to the stage height in the Goulburn River at the
Shepparton gauge. The trigger for the operation of the Loch Garry regulator is 110.487
AHD at the Shepparton gauge (Rural Water Commission of VIC 1987). From the
stage heights at this gauge, the operation of the Loch Garry regulator could be
d
Tab ts.
T
rem
m
etermined. The peak flows and the water stages at the Shepparton gauge, as shown in
le 6.1, indicate that the Loch Garry regulator did not operate for these flood even
165
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
The instantaneous stage hydrograph at these gauges of the selected floods, as well as
the rating curves at Shepparton and Loch Garry gauges, are shown in Figures E.3 to
E.10 in Appendix E.
6.2.3 Application, results and discussions
Based on the topographical data, the river reach was divided into 32 nodes (cross
sections) with the lengths of segments ranging from 900m to 1950m. The scheme of
the modelled river reach is shown in Figure E.2 in Appendix E. In order to obtain a
good quality of the identified roughness coefficient, the modelling parameters were
also chosen based on the recommendations from Chapter 5: a weighting factor of
= 0.7, a computational time step of = t 0.5 hour and a computational period before
starting to calculate the objective function of =
0
T 10 hours.
6.2.3.1 Identification of the roughness coefficient for the main channel
Four flood events No. 1, 3, 4 and 6 (see Table 6.1) were chosen for identification and
verification of the main channel roughness coefficient for the study reach of the
Goulburn River. These are all inbank floods except for the downstream part of the
reach where Loch Garry acts as a storage area.
Because the roughness conditions along the reach are very similar, the main channel
roughness coefficient was considered as a constant for the whole reach. The flood
events from 21/07 to 06/08/1978 and from 04/09 to 25/09/1979 were used to identify
the roughness coefficient. The discharge hydrograph at Shepparton and the stage
reproduced the prototype not only for the gauge where the observed data were included
hydrograph at Loch Garry gauges were the upstream and downstream boundary
conditions respectively. The stage data at Shepparton Golf Club gauge were used to
identify the roughness coefficient of the reach.
The identified roughness coefficients obtained from these flood events were 0.0544
and 0.0570 respectively. Figures 6.2 and 6.3 illustrate the observed and simulated
hydrographs at Shepparton Golf Club as well as the other two gauges, Shepparton and
Loch Garry, where the observed data were not included in the objective function. From
these figures it can be seen that the simulated hydrographs have been successfully
166
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
in the objective function but also for the other gauges. There are only some
discrepancies in the amplitude of the discharge hydrographs at the Loch Garry gauge.
his may be attributed to the quality of topography data at Loch Garry storage which
ere not surveyed directly but obtained from the contour map as well as the use of
discharge data obtained from a single rating curve. The results also indicate that the
onsideration of the roughness coefficient as a constant for the whole reach is
acceptable for this river reach.
It is imperative that the identified parameters should be verified using at least one
dependent event which differs significantly from the floods used during the
entification phase. Hence, two flood events from 27/09 to 09/10/1978 and from29/07
to 05/08/1980 were used to verify the identified roughness coefficient for this river
ach. The average value of the two identified roughness coefficients obtained above
was used to simulate the flow for these flood events. Figures 6.4 and 6.5 illustrate their
observed hydrographs and simulated hydrographs at different gauging stations. From
ese figures it can be seen that there is good agreement between the observed and
simulated stage and discharge hydrographs at all gauge stations. Both the phase and
amplitude of the floods were reasonably reproduced. Similarly to the identification
hase, for these flood events there are some discrepancies in the phase and amplitude
of the hydrographs especially at Loch Garry gauge because of the limitation of the
topographical data at Loch Garry and the use of rating curves for obtaining the
discharge data. For the flood from 29/07 05/08/1980, where the maximum water
level was below the elevation of Loch Garry, the observed peak discharge at this gauge
h
identified roughness coefficient.

T
w
c
in
id
re
th
p
to
as been successfully simulated. The verification results indicate a good quality of the




167
Chapter 6 Application of Roughness Identification Problem to Natural Rivers






102
103
104
105
106
107
108
109
0 50 100 150 200 250 300 350 400 450
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated







Figure 6.2 Observed and simulated hydrograph at different gauging stations for the
a. At Shepparton Golf Club Gauge
109
110





flood event 21/07 - 06/08/1978 using the identified roughness coefficient

102
103
104
105
107
108
50 1 0 200 2 350 400
Time (ho
S
t
m
)
106
a
g
e

(
Observed
0 00 15 50 300 450
urs)
Simulated
b. At Shepparton Gauge
300
0
50
150
200
250
0 50 100 150 200 250 300 350 400 450
Time (hours)
D
i
s
c

(
m
3
/
s
)
Observed
Simulated
100
h
a
r
g
e
c. At Loch Garry Gauge
168
Chapter 6 Application of Roughness Identification Problem to Natural Rivers






102
103
104
105
106
107
108
109
110
0 100 200 300 400 500
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated
a. At Shepparton Golf Club Gauge





102
103
104
105
106
107
108
109
110
S
t
a
g
e

(
m
)
0 100 200 300 400 500
Time (hours)
Observed
Simulated







b. At Shepparton Gauge
c. At Loch G
0
50
100
150
200
250
300
350
0 100 200 300 400 500
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Observed
Simulated
Figure 6.3 Observed and simulated hydrograph at different gauging stations for the
flood event 04/09 - 25/09/1979 using the identified roughness coefficient

arry Gauge
169
Chapter 6 Application of Roughness Identification Problem to Natural Rivers

a. At Shepparton Golf Club Gauge
b. At Shepparton Gauge
c. At Loch Garry Gauge
102
103
104
105
106
107
108
109
110
0 50 100 150 200 250 300 350
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated
102
103
104
105
106
107
108
109
110
111
0 50 100 150 200 250 300 350
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated
0
50
100
150
200
250
300
350
0 50 100 150 200 250 300 350
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
observed
Simulated

















Figure 6.4 Verification of the identified roughness coefficient by the independent
flood event 27/09 - 09/10/1978

170
Chapter 6 Application of Roughness Identification Problem to Natural Rivers


a. At Shepparton Golf Club Gauge
b. At Shepparton Gauge
c. At Loch Garry Gauge
102
103
104
105
106
107
108
0 50 100 150 200
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated
102
103
104
105
106
107
108
109
0 50 100 150 200
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated
0
40
80
120
160
200
0 50 100 150 200
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Observed
Simulated

















Figure 6.5 Verification of the identified roughness coefficient by the independent
flood event 29/07 - 05/08/1980
171
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
Combinations of boundary conditions
The results obtained in Chapter 5 using synthetic data indicated that the choice of
boundary condition combination and observed data type have a significant effect on
the quality of the identified roughness coefficient. These factors are investigated
further in this section by using real data of this river reach. Based on the data
availability, the four different combinations of boundary conditions which are
suggested in Chapter 5 were considered (see Table 6.2). The four flood events above
were used to identify the roughness coefficient. Figure 6.6 shows the values of the
identified roughness coefficient obtained from different combinations of boundary
conditions with different observed data types at different gauges used in the objective
function. The more consistent their values obtained from different floods, the better the
identifiability of a combination. From the figure it can be seen that for the Q-Z and Q-
Qz combinations, the identified values obtained from different flood events were much
more consistent by using the observed stage data than the observed discharge data,
while for the Z-Z and Z-Qz, using the observed discharge data seems much more
superior. Moreover, the computed results indicated that by using observed discharge
data for the Q-Z and Q-Qz combinations or observed stage data for the Z-Z and Z-Qz
combinations to identify roughness coefficient, the simulated hydrographs calculated
from the identified value of the roughness coefficient could only reproduce the
prototype at the gauge where the observed data were used in the objective function but
not at the other gauges. Figure 6.7 shows the observed and simulated hydrograph as an
example for the flood event 27/9-9/10/1978 using the Z-Z combination where the stage
data at the Shepparton Golf Club gauge were used in the objective function to identify
the roughness coefficient. From this figure it can be seen that although the simulated
stages at this gauge are very close to the observed one (Figure 6.7a), the other
discharge hydrographs at Shepparton and Loch Garry gauges deviated significantly
from the observed ones (Figure 6.7b and c). This indicates the poor quality of the
identified parameter when using this combination with the observed stage used in the
objective function.
Therefore, although the true value of the roughness coefficient of the reach is not
known, the consistency of the identified values obtained from the combinations of
eit d her Q-Z or Q-Qz using stage data or Z-Z and Z-Qz using discharge data coul
172
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
ensure that these values are reasonable values of the roughness coefficient of the reach.
These results reconfirm the findings obtained from the synthetic data in Chapter 5.
Table 6.2 Different combinations of boundary conditions
Combination Upstream boundary Downstream boundary
Q-Z
Discharge hydrograph Q(t) Stage hydrograph Z(t)
Z-Z
Stage hydrograph Z(t) Stage hydrograph Z(t)
Q-Qz
Discharge hydrograph Q(t) Rating curve Q(Z)
Z-Qz
Stage hydrograph Z(t) Rating curve Q(Z)






0
10
20
30
40
50
60
70
80
Z1
Observed data
I
d
e
n
t
i
f
e
d

n

(
1
0
-
3
)
Q31 Z8
0
10
20
30
40
50
60
70
80
Q1
Observed data
I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
Q31 Z8
a. Q-Z combination b. Z-Z combination




0
10
20
30
40
70
Ob ata
n
t
i
f
i
e
d
)



Figure 6.6 Identified roughness coefficient with different combinations of boundary
conditions and observed data types
0
Z1
10
20
30
60
70
Observed
I
n
d
e
n
t
i
f
i

(
1
0
-
3
)
40
50
e
d

n
80
data
Z31
80
50
60

n

(
1
0
-
3
Q1
served d
I
d
e
Z31 Z8 Q31 Z8 Q31
c. Q- bi -Qz tion Qz com nation d. Z combina
Flood event 8/197 od eve 09/10
Flood event 0 /09/1979 lood eve 05/08
Z1 a are t age at Sh auge
8 bser Shep b gaug
31 and Q31 a ed st ge at gaug
21/07 - 06/0 8 Flo nt 27/09 - /1978
4/09 - 25 F nt 29/07 - /1980
nd Q1
is the o
he observed st
ved stage at
and discharge
parton Golf Clu
epparton g
e Z
Z re the observ age and dischar Loch Garry e
173
Chapter 6 Application of Roughness Identification Problem to Natural Rivers






a. At Shepparton Golf Club Gauge
b. At Shepparton Gauge












Figure 6.7 Observed and simulated hydrographs for the flood event 27/09
09/10/1978 using the identified roughness coefficient from Z-Z boundary conditions
and the observed stages at Shepparton Golf Club used in the objective function
c. At Loch Garry Gauge
102
103
104
105
106
107
108
109
110
0 50 100 150 200 250 300 350
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated
0
100
250
s
c

(
m
3
/
s
300
350
)
Observed
Simulated
150
200
h
a
r
g
e
50
D
i
0 50 100 150 200 250 300 350
Time (hours)
0
50
100
150
200
250
300
350
400
0 50 100 150 200 250 300 350
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Observed
Simulated
174
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
The limitation of using a single rating curve as the downstream boundary condition for
the roughness identification problem for rivers with mild bed slopes has been
illustrated in Chapter 5. Although the slope of the study reach is very small with
average slope of 5cm/km, Figure 6.6 does not show any significant advantage in using
Variation of the roughness coefficient with stage
a flood event. For the selected
flood events shown in Table 6.1, the range of variation in water stage was from 5 m to
e Equation 5.3). The discharge and stage hydrographs were chosen as the
tic polynomial
hness curves show
clearly the general behaviour of the roughness variation with stage for this river reach.
a stage hydrograph as the downstream boundary condition over using a single rating
curve. This is due to the fact that the discharge data at Shepparton and Loch Garry
gauges were obtained from the single rating curves. Therefore, it should be noted that
using a stage hydrograph as the downstream boundary condition for these cases will
not overcome the limitation of using a single rating curve.
The Goulburn River is a very deep river (see the cross-sections in Figure E.1 in
Appendix E) with a large variation in water level during
7 m. Accordingly, if there is any considerable variation of the roughness coefficient
with water stage, the formulation of the roughness coefficient as a function of stage
could improve the flood simulation for this river reach. In this study, the variation of
the roughness coefficient was modelled as linear and quadratic polynomial functions of
stage (se
upstream and downstream boundary conditions respectively. The observed stages at
the Shepparton Golf Club gauge were used in the least square objective function. The
four flood events above were chosen to identify the roughness functions.
The identified roughness functions for different flood events are shown in Figure 6.8
where the roughness curves were formulated as linear and quadra
functions of stage. As expected, the identified roughness curves obtained from
different events cannot be identical because of several reasons such as measurement
errors and the change of the roughness coefficient with time. However, they are very
close to each other, especially in the range of observed stages of the floods (from 102
m to 109 m at Shepparton Golf Club gauge). The identified roug
It shows that the roughness coefficient of the study reach increases slightly with depth.
175
Chapter 6 Application of Roughness Identification Problem to Natural Rivers









ified roughness curves with different flood events at the Shepparton
stage, (b) Quadratic function of stage





Figure 6.8 Ident
Golf Club gauge: (a) Linear function of
20
30
40
50
60
70
80
99 100 101 102 103 104 105 106 107 108 109 110 111
Stage (m)
I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
In order to compare the performance of different roughness models the root mean
square errors (RMSE) between the observed and simulated hydrographs were used.

( )
N
RMSE
i
Y Y
N
Si Oi
=

number of observed stages or discharges.
=
1
(6.1)
where
Oi
Y and
Si
Y are the observed and simulated stages/discharges respectively, N is
20
30
40
60
70
80
99 100 101 102 103 104 105 106 107 108 109 110 111
n
t
i
1
0
-
3
)
50
f
i
e
d

n

(
I
n
d
e
Stage (m)
21/07-06/08/1978 27/09-09/10/1978
04/09-25/09/1979 29/07-05/08/1980
(a)
(b)
176
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
Table 6.3 presents RMSEs between the observed and computed stages at Shepparton
Golf Club gauge of the different roughness models. It can be seen that by allowing
stage variation of the roughness coefficient, the RMSEs are smaller. This indicates that
s were almost the same. These results
ind s a
constant should be enough.
ab ean s
MSE of d t roughne del (m)
the model with the roughness functions could reproduce the prototype better. However,
due to the variation of the roughness coefficient with stage of this study reach is not
significant (see Figure 6.8), the RMSEs of the different roughness models were not
much different. Especially for the cases when the roughness function was formulated
as linear and quadratic functions, the RMSE
icate that for this river reach, con ideration of the roughness coefficient as
T le 6.3 Root m square errors (RMSE) for different roughnes models
R ifferen ss mo
Flood event
= n const ) (Z n n = ) (Z n n = 2
nd
order 1
st
order
21 2 0.283 /07 - 06/08/1978 0.287 0.28
27 0 /09 - 09/10/1978 0.264 0.26 0.260
04/09 - 25/09/1979 0.250 0.246 0.246
29/07 - 05/08/1980 0.371 0.335 0.328

6.2.3.2 Identification of roughness coefficients for the compound channel
When the water stage rises above the bankfull level, the cross sections include the
main channel and the floodplain. Generally, the roughness coefficients of the main
channel ( ) and floodplain ( ) are significantly different. In order to have more
ain channel and floodplain should
e
c f
accurate results, the roughness coefficients of the m
n n
be considered separately. The roughness identification problem for compound
channels developed in Chapter 4 was applied for this river reach. For the selected
floods in the Goulburn River from 1978-1981, only three flood events (No. 2, 4 and 7,
as showed in Table 6.1) caused flooding in the surrounding floodplains. These flood
events were chosen to investigate the performance of the mod l.
For these floods, the floodplains were not inundated entirely. The maximum depths on
the floodplain of all these floods are much shallower compared to the main channel
177
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
depth. They range from 0.35 m to 0.90 m at the Shepparton Golf Club gauge for the
different events. Moreover, as indicated in Chapter 5, for the floods with shallow
depths on the floodplain, there is high degree of uncertainty in the identified roughness
coefficient of the floodplain. Therefore, in this case study, the consideration of the
floodplain roughness coefficient as a constant should be adequate and reasonable. The
roughness coefficients of the compound channel are considered for two cases. For the
e floodplain are
floods in the previous

first case, both roughness coefficients of the main channel and th
considered as constants. For the second one, the main channel roughness coefficient is
a function of stage and the floodplain one is a constant.
Following the suggestions obtained from Chapter 5, the largest flood from 28/09 to
31/10/1979 will be used to identify the roughness coefficients and should provide
reliable values for the roughness coefficient for the smaller floods. The roughness
coefficients of the main channel and floodplain identified from this flood event were
0.055 and 0.078 respectively. It can be seen that the value of the main channel
roughness coefficient obtained from this flood event is very close to the values of the
main channel roughness coefficient obtained from the inbank
sub-section.
The two flood events 08/08 - 31/08/1978 and 25/06 05/07/1981 were used to verify
the identified roughness for this river reach. Figures 6.9 and 6.10 show the observed
hydrographs and simulated hydrographs at different gauging stations for these flood
events by using the identified roughness coefficients obtained from the flood event
28/09 - 31/10/1979. These figures show that the simulated stage and discharge
hydrographs at all gauging stations were reasonably reproduced for the observed
floods. Similar to the previous cases, there are, however, some small discrepancies in
the phase and the amplitude of the hydrographs at the Loch Garry gauge. Again, this
may be attributed to the quality of the cross section data on the floodplain part which
were not surveyed directly but obtained from the contour map, and the discharge data
was obtained from single rating curves.

178
Chapter 6 Application of Roughness Identification Problem to Natural Rivers










Figure 6.9 Verification of the identified roughness coefficients by the independent
flood event 08/08 - 31/08/1978








a. At Shepparton Golf Club Gauge
b. At Shepparton Gauge
c. At Loch Garry Gauge

101
102
0 50 100 150 200 250 300 350 400 450 500 550 600
Time (hours)
103
104
105
106
107
108
109
110
S
t
a
g
e

(
m
)
Observed
Simulated
0
50
100
150
200
250
0 50 100 150 200 250 300 350 400 450 500 550 600
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
300
350
400
/
s
)
Observed
Simulated
102
103
104
105
106
107
108
109
110
111
0 50 100 150 200 250 300 350 400 450 500 550 600
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated
179
Chapter 6 Application of Roughness Identification Problem to Natural Rivers

















Figure 6.10 Verification of the identified roughness coefficients by the independent
flood event 25/06 05/07/1981
a. At Shepparton Golf Club Gauge
b. At Shepparton Gauge

c. At Loch Garry Gauge
102
103
104
105
106
107
108
109
110
111
0 50 100 150 200 250 300
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated
103
104
105
106
107
108
109
110
111
0 50 100 150 200 250 300
Time (hours)
S
t
a
g
e

(
m
)
Observed
Simulated
0
50
100
150
200
250
300
350
400
0 50 100 150 200 250 300
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Time (hours)
Observed
Simulated
180
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
To investigate the effect of the floodplain depth on the identified roughness
coefficients, these coefficients were identified independently from the three flood
vents. Figure 6.11 shows their identified values obtained from these three flood
vents. From this figure, it can be seen that for the main channel the values of the
identified roughness coefficient are very consistent. However, for the floodplain the
alues are very different from each other. It can be argued that the discrepancies in the
values of the identified roughness coefficient of the floodplain may be because of the
hange in the roughness condition of the floodplains. However, even with a big
ifference in the values of the identified floodplain roughness coefficient, using the
values of the identified roughness coefficients from the flood event 28/09 - 31/10/1979
simulate the other flood events, the simulated hydrographs still agreed very well
with the prototype as shown in Figures 6.9 and 6.10. These results indicate that for the
ases when the floodplain depths are very shallow, the value of the floodplain
e simulated hydrographs. As a result, the
values of the identified floodplain roughness coefficient are highly uncertain. This can
e explained by the fact that when the depth on the floodplain is much smaller
compared with the main channel depth, the conveyance of the floodplain is not
portant for the whole conveyance of the cross section. Moreover, the very high
ss coefficient obtained from the flood event 08/08 -
31/08/1978 can also be attributed to the very shallow floodplain depth where the
oodplain may act as off-stream storage.
e
e
v
c
d
to
c
roughness coefficient is not sensitive to th
b
im
value of the floodplain roughne
fl
0
20
40
I
d
e
n
t
i
60
80
100
120
140
Main channel roughness
f
i
e
d


n

(
1
0
-
3
)
08/08-31/08/1978
28/09-31/10/1979
25/06-05/07/1981
Floodplain roughness

Figure 6.11 Identified roughness coefficients obtained from different flood events
181
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
For the second case, the main channel roughness coefficient is formulated as a function
of water stage and the floodplain one is a constant. Table 6.4 shows the identified
ughness coefficients obtained from different roughness models and corresponding
MSEs between observed and simulated stages at the Shepparton Golf Club gauge
calculated by using the identified roughness coefficients obtained from the flood
vents itself and from the flood event 28/09 - 31/10/1979. From Table 6.4 and Figure
6.12 it can be seen that the values of the identified roughness coefficient of the main
hannel are very consistent. They are also very similar with the values obtained from
the cases for inbank floods (see Figure 6.12). However, for the floodplain, there are
very large discrepancies in the values of the identified roughness coefficient obtained
fro c
, there is a high degree of
uncertainty on the value of the identified floodplain roughness coefficient. Table 6.4
ro
R
e
c
m these three floods. This reconfirms the conclusion obtained by using syntheti
data in Chapter 5 that when the floodplain depths are small
also shows that there are some improvements in RMSEs when the main channel
roughness coefficient was considered as a function of stage. However, because its
variation with stage is not significant as shown in Figure 6.12, the differences of
RMSEs for these cases are very small. Therefore, consideration of the roughness
coefficients as constants should be enough to simulate the flow for this river reach.
Table 6.4 Identified roughness coefficients and RMSEs with different flood events
Identified roughness
Flood event
Depth*
(m)
c
n
f
n
RMSE 1
(m)
RMSE 2
(m)
const 0.0545 0.1559 0.245 0.257
1
st
order Fig. 6.12a 0.1354 0.243 0.254
08/08 -
0.38
31/08/1978
2
nd
order Fig. 6.12b 0.1297 0.242 0.254
const 0.0549 0.0779 0.229 0.229
1
st
order Fig. 6.12a 0.0695 0.223 0.223
28/09 -
31/10/1979
0.94
2
nd
order Fig. 6.12b 0.0708 0.222 0.222
const 0.0554 0.0858 0.424 0.445
1
st
order Fig. 6.12a 0.0793 0.395 0.401
25/06 -
05/07/1981
0.74
2
nd
order Fig. 6.12b 0.0796 0.395 0.402
Note: * indicates the maximum depth on the floodplain at the Shepparton Golf Club gauge.
RMSE 1 and RMSE 2 are RMSEs calculated from the identified roughnesses obtained from
the flood itself and from the flood 28/09 - 31/10/1979 respectively.
182
Chapter 6 Application of Roughness Identification Problem to Natural Rivers












Figure 6.12 Identified roughness function of the main channel with different flood
events at the Shepparton Golf Club gauge: (a) Linear function of stage, (b) Quadratic
function of stage
Duong River
.3.1 General description the study area
Th
This river is one of the main distributaries
30
40
50
60
n
t
i
f
i
e
d

n

(
1
0
-
3
)
99 100 101 102 103 104 105 106 107 108 109 110
Stage (m)
I
d
e
08/08-31/08/1978
28/09-31/10/1979
25/06-05/07/1981
30
40
50
60
99 100 101 102 103 104 105 106 107 108 109 110
Stage (m)
I
d
e
n
t
i
f
i
e
d


n

(
1
0
-
3
)
08/08-31/08/1978
28/09-31/10/1979
25/06-25/07/1981
(a)
(b)
6.3
6
e Duong River is situated in the Red River Delta, in the northern part of Viet Nam.
of the Red River which conveys the water to
the Thai Binh River. It starts from Xuan Canh (Hanoi) and flows into the Thai Binh
River at Pha Lai with a length of 67 km. The Duong River has an important effect on
the Thai Binh River system. In the flood season, the Duong River conveys a great
amount of water to this system. According to observation data over many years, the
discharge conveyed by Duong River from Hong River provides 70% discharge of the
183
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
Thai Binh River at Pha Lai. In the dry season, the discharge of the Duong River at
Phalai is about 200 to 300 m
3
/s and this is an important source supplying the Thai Binh
River.
The flood season is from June to October. Every year, the floodplains are flooded
extensively for a significant time, especially from middle of July to end of August.
Therefore, not only the value of the main channel roughness coefficient but also the
floodplain one plays an important role for the flood routing problem. This river is
especially chosen to test the performance of the roughness identification model for the
river with compound sections.
The study reach is 61.71 km long from Thuong Cat to Pha Lai. Figure 6.13 shows the
com h au e th n
located in this reach: Thuong Cat, Ben Ho and Pha Lai.
Th ope e stu h is r 1 The of t in
channel ranges from 250 to The th ns from to
1500 it wide war strea floo atio es fr to
7 m sea level. The d is ep d el s fr to
-12 m
There is an extensive dyke s long h co erba odin s
an av height g f 5 m yke s te o fo e
me interferes wit flow be ma nnel.
The description of roughness conditions from the topography data surveyed in 1996
puted reac and the g ging stations along it. There ar ree gaugi g stations
e average sl of th dy reac 0.0001 o 0 cm/km. width he ma
450 m. width of e floodplai ranges 200
m and
above the
ns to ds down m. The dplain elev n rang om 5
river be rather de with the be evation om -1
.
ystem a t e river nfining ov nk flo g. It ha
erage
anders and
rangin rom 4 to . The d system ha nded t llow th
h flood s across nds in the in cha Levees
have been also constructed to block important distributary flow paths resulting in
higher flood levels elsewhere in the floodplain or in the main channel.
for this river shows that the roughness conditions in the main channel and in the
floodplains of the river are different. The main channel is an alluvial channel while the
floodplains have bushes, trees and small houses. The roughness conditions of the main
channel and of the floodplains are very similar along the channel.
The topographical and hydrological data and computed results are discussed in the
following subsections.
184
Chapter 6 Application of Roughness Identification Problem to Natural Rivers



6.3.2.2 Flow data





Figure 6.13 Duong River and the gauging stations along the river
6.3.2 Data
6.3.2.1 Cross section data
Cross section data were obtained from the Institute for Water Resources Planning and
Management (Viet Nam) which were surveyed in the dry season of the year 1996.
There were 33 cross sections measured along the river. The distances between every
cross section range from 0.5 to 4.0 km. The cross-sectional data of this river were
rather complete with compound cross sections including both of the main channel and
the floodplains.
The details of the bed elevations and the positions of the cross sections are given in
Table F.1. Ten representative cross sections illustrated in Figure F.1 in Appendix F.
There are three gauging stations located in this reach which are Thuong Cat, Ben Ho
and Pha Lai (see Figure 6.13). For the Thuong Cat gauge, the flow data are available
for both stages and discharges. The discharge and stage data at this gauge were
185
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
measured independently. However, for the Ben Ho and Pha Lai gauges, only measured
stage data are available.
The four recent highest flood events in the years of 1995, 1996, 1997 and 1998 which
occurred during the validity of the cross section survey will be considered to identify
the roughness coefficients of the main channel and floodplains. For these flood events,
the floodplains are extensively flooded. The observed peak discharges and stage of the
selected flood events are shown in Table 6.5.
Table 6.5 Observed peak discharge
p
Q and stage
p
Z of the selected flood events
No. Flood event
Observed
p
Q
at Thuong Cat
(m
3
/s)
Observed
p
Z
at Thuong Cat
(m)
Observed
p
Z at Ben Ho
(m)
Observed
p
Z
at Pha Lai
(m)
1 14/08-31/08/1995 5650 11.12 8.58 6.61
2 16/08-30/08/1996 6120 11.80 9.02 6.52
3 22/07-06/08/1997 4870 10.63 8.06 5.87
4 26/07-05/08/1998 4910 10.49 7.42 4.72
Note: the starting elevation of the floodplain at Ben Ho is 6m
Moreover, some other small flood events which occurred before and after these main
ood events also used to investigate the variation of the roughness coefficient during
and discharges at these gauges were
supplied by the Bureau of Meteorology and Hydrology of Vietnam. The observed
tage and discharge hydrographs at the gauging stations are shown in Figures F.3 to
F.6 in Appendix F.
6.3.3 Application, results and discussions
The river reach was divided into 33 nodes (cross sections) with the lengths of the
seg e
computed scheme of the river re .2. Similar to the case study of
fl
the flood season. The instantaneous stages
s
ments ranging from 500m to 3800m based on the variation of cross sections. Th
ach is shown in Figure F
the Goulburn River, the modelling parameters were also chosen following the
suggestions drawn from Chapter 5: a weighting factor of = 0.7, a computation time
step of = t 1.0 hour and a computational period starting to calculate the objective
function =
0
T 10 hours. The upstream and downstream boundary conditions are the
discharge hydrograph at Thuong Cat and the stage hydrograph at Pha Lai respectively.
186
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
The roughness coefficients were identified by using the observed stage at Ben Ho.
Because along the reach there is only one gauge that the observed data can be used to
identify roughness coefficients and, as indicated above, the roughness conditions along
the computed reach are very similar, it is assumed that there is no variation in the value
of the roughness coefficients along the reach.
Identification of the roughness coefficients for the compound channel
As mentioned above, to identify the roughness coefficients of the compound channel,
he roughness coefficients were modelled for two cases.
The first case is a conventional case, where the roughness coefficients of the main
constants. The second case is the one that
allows the roughness coefficients to vary with the water stage.
floodplain, respectively. Figure 6.14 shows the observed and
t indicated that the consideration of no
variation in the roughness values along the reach is acceptable.
the four recent highest flood events during the years of 1995, 1996, 1997 and 1998
were chosen (see Table 6.5). For these flood events, the water depths on the floodplain
are significant compared with the main channel depth, which range from 1.4 to 3.0 m.
Therefore, it is expected that the quality of the identified roughness coefficients would
be better for this river reach. T
channel and floodplain are considered as
Firstly, the roughness coefficients of the main channel (
c
n ) and floodplain (
f
n ) are
considered as constants. For the identification phase, two flood events 16/08 -
30/08/1996 and 22/07 - 06/08/1997 were selected. The identified roughness values
obtained from these flood events are 0.0315 and 0.0340 for the main channel and
0.0601 and 0.0619 for the
simulated stage hydrographs at Ben Ho and Thuong Cat by using the identified
roughness coefficients obtained from these flood events. From this figure it can be
seen that the simulated stage hydrographs using identified roughness coefficients are
matched very well to the observed ones for all the flood events at these gauging
stations. It should be noted that only the observed stage data at Ben Ho were used in
the objective function during optimisation. The agreements of the observed and
simulated stage hydrograph at Thuong Ca
To verify the identified roughness coefficients for this river reach, the flood events
14/08-31/08/1995 and 26/07-05/08/1998 were used. The observed and simulated
187
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
hydrographs by using the average values of the identified roughness coefficients above
at different gauging stations are shown in Figure 6.15. From this figure it can be seen
that there is good agreement between the observed stages and the simulated ones at all
gauging stations in terms of both the phase and amplitude of the floods. The
verification results indicate a good quality of the identified roughness coefficients.
The consistency in the values of the identified roughness coefficients which are
are
rather consistent. These indicate that there is much improvement in the identified
independently obtained from different flood events is also one of the criteria to indicate
the quality of the identified coefficient. Figure 6.16 illustrates the identified roughness
coefficients independently from the four flood events. It should be noted that the
maximum water depths on the floodplain of these flood events were significant when
compared to the main channel depth. From this figure it can be seen that not only the
identified roughness coefficient of the main channel but also the floodplain ones
floodplain roughness coefficient compared with the case of the Goulburn River, where
the water depths on the floodplain were small (see Figure 6.11). This indicates that,
when the depths of floodplain are larger, the quality of identified roughness coefficient
is much improved. These results were supported by the finding obtained from Chapter
5 by using synthetic data for identifying roughness coefficients in compound channels.
In the following subsection, further investigation into the cases where the roughness is
allowed to vary with the water stage will be discussed.
188
Chapter 6 Application of Roughness Identification Problem to Natural Rivers



12
13 Observed
Simulated













Figure 6.14 Observed and simulated stage hydrograph at different gauging stations by
using the identified roughness coefficients obtained from the flood itself


6
7
8
9
0 50 100 150 200 250 300 350 400
Time (hours)

S
t
a
g
e

(
10
11
m
)
Ben Ho
Thuong Cat Thuong Cat
a. Flood event 16/08-30/8/1996
b. Flood event 22/07-06/08/1997
12 Observed
7
8
9
10
0 50 100 150 200 250 300 350 400
Time (hours)
S
t
a
g
e

(
m
)
11
Simulated
Thuong Cat
Ben Ho
189
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
















a. Flood event 14/08-31/08/1995
b. Flood event 26/07-05/08/1998
6
7
Figure 6.15 Observed and simulated stage hydrograph at different gauging stations by
using the average values of the identified roughness coefficients obtained from the
1996 and 1997 floods


8
S
t
9
11
12
0 50 100 150 200 250 300 350 400 450
Time (hours)
a
g
e

(
10
m
)
Observed
Simulated
Ben Ho
Thuong Cat
5
6
7
8
9
10
0 50 100 150 200 250 300
Time (hours)
S
t
a
g
e

(
m
)
11 Observed
Simulated
Ben Ho
Thuong Cat
190
Chapter 6 Application of Roughness Identification Problem to Natural Rivers






Figure 6.16 Identified roughness coefficients obtained from different flood events
Variation of the floodplain roughness coefficient with stage
For the Duong River, due to the fact that from the beginning of the flood season (from
June) the water level is almost at bankfull level, constant values were adopted for the
main channel roughness c
20
30
40
50
60
70
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
I
d
14/08-31/08/1995
16/08-30/08/1996
22/07-06/08/1997
0
10
1 2
27/07-05/08/1998
Main channel roughness Flood plain roughness
oefficient to avoid any uncertainties with the extrapolation
s, if the variation of the floodplain
roughness is significant, formulating the floodplain roughness coefficient as functions
e quality of the simulated hydrographs.
igure shows that except for the roughness curve obtained
from the flood event 26/07-05/08/1998 (the maximum depth on the floodplain is
r depth on the floodplain. There is not much
are
dis
of
above the bankfull stage of the polynomial function for the main channel roughness
coefficient. However, the variation of water depths on the floodplain for the selected
events is rather large (up to 2 to 3 m except for the 1998 flood event (see Table 6.6))
and the inundated time on the floodplains was lasting for the almost the whole duration
of the flood events. Therefore, for these case
of stage could improve th
Figure 6.17 illustrates the identified roughness curves obtained from different
roughness models. This f
smallest compared to the other floods), the value of the floodplain roughness
coefficient tends to increase with the wate
difference between the identified roughness curves for the case when roughness curves
formulated as linear and quadratic polynomial functions. Also, there are still some
crepancies in the identified roughness curves. However, the rather consistent trend
the roughness curves indicates that the ability to identify the variation of the
191
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
rou
the
ch ery consistent.
This indicates that these identified values of the main channel roughness coefficient












Figure
at Ben
Similarly t
roughness
Ben Ho g
coefficient
are presented in Table 6.6. As expected, for this study case, because the variation of
ghness coefficient for the floodplain could be improved when the water depth on
floodplain is significant. For the identified roughness coefficient of the main
annel, it can be seen from Table 6.6 that its identified values are v
are reliable.
6.17 Identified floodplain roughness functions with different flood events
Ho gauge: (a) Linear function of stage, (b) Quadratic function of stage
o the Goulburn River, in order to compare the performance of different
models for this river, the RMSEs between observed and simulated stages at
auge were used. The calculated RMSEs using the identified roughness
s obtained from the flood event itself and the flood event 16/08-30/08/1996
(a)
40
5 6 7 8 9 10
Stage (m)
70
45
55
65
d

n
60

(
1
0
-
3
)
14/08-31/08/1995
16/08-30/08/1996
50
I
d
e
n
t
i
f
i
e
22/07-06/08/1997
26/07-05/08/1998
B
a
n
k
f
u
l
l

l
e
v
e
l

(b)
40
45
50
55
60
65
5 6
70
7 8 9 10
I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)
Stage (m)
14/08-31/08/1995
16/08-30/08/1996
22/07-06/08/1997
26/07-05/08/1998
B
a
n
k
f
u
l
l

l
e
v
l


e
192
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
the floodp
there are s urves
obtaine f
roughness
very small
RMSEs ar
functions f
could not
coefficient
Table 6
lain roughness with stage is not significant (as shown in Figure 6.12) and
ome discrepancies in the trend of identified floodplain roughness c
d rom different flood events, there is no advantage in using floodplain
curves. This table shows that the differences of RMSEs for these cases are
or even for the flood events 14/08-31/08/1995 and 22/07-06/08/1997 their
e almost unchanged. The calculated RMSEs by using identified roughness
rom the flood event 16/08-30/08/1996 for simulating the other flood events
show any advantages over the case where the flood plain roughness
are considered as a constant.
.6 Identified roughness coefficients and RMSE with different flood events
Identified roughness coefficients
Flood event
Depth*
(m)
c
n
f
n
RMSE 1
(m)
RMSE 2
(m)
0.0332 const 0.0622 0.069 0.115
0.0334 1
st
order Fig. 6.17a 0.069 0.096
14/08
31/08/19
-
95
2.58
0.0335 2
nd
order Fig. 6.17b 0.068 0.098
0.0315 const 0.0601 0.060 0.060
0.0326 1
st
order Fig. 6.17a 0.057 0.057
16/08
30/0 9
-
3.02
8/1 96
0.0327 2
nd
order Fig. 6.17b 0.056 0.056
0.0340 const 0.0619 0.048 0.122
0.0341 1
st
order Fig. 6.17a 0.047 0.124
22/07
06/08/19
-
97
2.06
0.0342 2
nd
order Fig. 6.17b 0.047 0.123
0.0337 const 0.0554 0.071 0.099
0.0333 1
st
order Fig. 6.17a 0.069 0.117
26/07
05/08/19
-
98
1.42
0.0335 2
nd
order Fig. 6.17b 0.070 0.120
Note: * indicates the
RMSE 2 ar
flood itself
Moreover,
as function
computatio
computatio
coefficient were con
floodplain roughness coefficient, as indicated in Chapter 5, these results also show that
when the floodplain roughness conditions do not change significantly with depth, the
maximum depth on the floodplain at the Ben Ho gauge. RMSE 1 and
e RMSEs calculated from the identified roughness coefficients obtained from the
and from the flood 16/08-30/08/1996 respectively.
the study also found that when the roughness coefficients were formulated
s of stages, the number of identified parameters increases. As a result, the
n time increases considerably. For example, for this case study, the
n time was about four times greater than the case where the roughness
s sidered as constants. Due to the uncertainty of identifying the
193
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
consideration of the roughness coefficient of the floodplain as a constant is the most
appropriate solution for flood routing purposes.
Variation o
The variati
in bedform
Engineers
consists of
and rigid c
With diffe
changed. Especially during high flood events, the flow velocity significantly increases
with increasing discharge and
flow changes. Therefore, in the flood season,
is the most
The mech
bedforms a
effects of these factors on the roughness coefficient are very difficult, if not
possible to investigate the variation of the
roughness coefficient during the flood season by using the roughness identification
model i
can be arg
variation o
roughness
year, using
times durin
variation o
The Duong River is an alluvial river. Unlik
where the maxim
the main floods in the Red River system
usually occu
August. Therefore, this ri
f the roughness coefficient with time in the Duong River
on of the roughness coefficient of a river could be attributed to the variation
s, the changes in vegetation or water temperature (U.S. Army Corps of
1993; Chow 1959). For the main channel of alluvial rivers, their bed
movable particles/sediments. The major difference between these channels
hannels concerning flow resistance is the formation of mobile bedforms.
rent flow conditions, bedforms and hence the roughness coefficients are
the bedforms also change significantly with time as the
the variation in bedforms of these rivers
dominant factor affecting the variation of the roughness coefficients.
anism of sediment transport, interaction process and the formation of
re complex and they are not within the scope of this study. Moreover, the
impossible, to quantify. However, it may be
to dentify the value of roughness coefficient at different time of the season. It
ued that by using a single flood season, it may not be enough to show the
f the roughness coefficient with time because of the error in the identified
coefficient. However, if the flow regime of a river is fairly consistent every
the identified roughness obtained from different flow events at different
g the flood season for several years could provide a general picture of the
f the roughness coefficient with time during the flood season.
e the flow regime of the Goulburn River,
um flood events can occur at different times from June to November,
in general and the Duong River in particular
rs in very similar times of the year, from the middle of July to the end of
ver was chosen as a case study for using the roughness
194
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
identifi i
flood seaso
Using seve in flood events of the years
from 1995 to 1998, the values of roughness coefficients at different time were
hness coefficient of the main
channel in the Duong River at different times in the flood season from June to October.
From t f
in the figu
lowest at
slightly an
(from the e
coefficient
its value is
Th
be esults are
also supported by the experi
indicated that with an in
increases. It is specified that this increase is due to an increasing bedform effect which
cat on model to investigate the variation of the roughness coefficient during the
n.
ral flow events before, during and after the ma
identified. Figure 6.18 illustrates the computed roug
his igure it can be seen that the roughness values vary with time. The trend line
re shows clearly that the roughness coefficient values of the channel are
the beginning of the flood season. The roughness coefficient increases
d reaches its highest value at the time when the main flood events occur
nd of July to the end of August). After the main flood events, the roughness
value of the main channel slightly decreases. At the end of the flood season,
still a bit larger than the values at the beginning of the flood season.






0
5
10
15
20
25
30
35
40
0 30 60 90 120 150
I
d
e
n
t
i
f
i
e
d

n

(
1
0
-
3
)

Figure 6.18 Variation of the roughness coefficient of the main channel with time
during the flood season
e variation of the roughness coefficient of the main channel in the Duong River can
attributed to the variation of the bedform during the flood season. These r
mental study of Valentine et al. (2001). Their study
crease in hydraulic radius the roughness coefficient value
Time
1-June 1-July 1-Augst 1-Sept. 1-Oct. 1-Nov.
1995 Roughness coeff. identified from
1996 the highest flow of the year
1997 Trend line
1998
195
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
is consiste
the higher
to the hig sporting the
sedime
Institute of
density are
River, the
volume.
In brief, th
Duong Riv
that this m o estimate the variation of the
roughness coefficient during the flood season
at the same
6.4 Sum
In this cha
applied to
River in th the characteristics of the cross-
section n
the model
reaches we
For the Goulburn River, seven flood events
from 1978 to 1981, during the validity of the
inbank floods and three overbank floods.
coefficient from
The simulated hydrographs using the iden
observed hydrographs at different gaugi
performance of the m
The effect
types used in the objective function was also investigated. The results indicate that, for
the Q-Z and Q-Qz combinations, stage data should be used in the objective function to
nt with observed two-dimensional dunes. Beside the changes in bedforms,
value of roughness coefficient during the main floods may also be attributed
her energy and momentum losses for picking up and tran
nt of high flows. The study on the sediment of the of Red river system (The
Water Resources Research 2004) indicated that the sediment discharge and
very different between the dry season and the flood season. For the Duong
sediment volume in the flood season occupies 92% of the annual sediment
is subsection investigates the variation of the roughness coefficient of the
er using the roughness identification model. The result analysis indicates
ethod can be used as a simple means t
for rivers where maximum flows occur
period each year.
mary and conclusion
pter, the inverse problem of identifying roughness coefficients has been
two natural rivers: the Goulburn River in Victoria, Australia and the Duong
e Red River Delta, Viet Nam. Depending on
s a d water variation in the main channel and the floodplain, the performance of
was investigated and several problems related to the roughness of the study
re explored.
were considered, all of which occurred
cross section survey. They included four
The values of the identified roughness
one or two flood events were verified using the other flood events.
tified value reasonably reproduced the
ng stations. This indicated the good
odel for the real river.
of the boundary condition combinations incorporated with the observed data
196
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
identify e
data should
Chapter 5.
stage hydro
data were tion of using a
single r n
The three e
River were chosen to identify the roughness
floodplains. However, for these floods, the wa
small. The
reach indicates that there is
roughness coefficient by using flood events
There is a
during a fl oughness coefficient
with w r
events we
coefficient
RMSEs in
when the r
when roug
For the D
bankfull le
investigate
compound The four highest flood events from 1995 to 1998 were selected to
identify and verify the identified roughness coefficients for this river. The simulated
produced the observed
d that the values of the identified floodplain roughness coefficient
were much more consistent compared with the case of the Goulburn River. The
th roughness coefficient, while for the Z-Z and Z-Qz combinations, discharge
be used. This reconfirms the findings obtained from the synthetic data in
In addition, one thing should be noted: for rivers with mild slope, using a
graph as the downstream boundary condition for the cases where discharge
obtained from rating curves could not overcome the limita
ati g curve.
vents which caused flooding in the surrounding floodplains of the Goulburn
coefficients for the main channel and
ter depths on the floodplains are very
results and analysis of the identified roughness coefficients for this river
a high degree of uncertainty in the identified floodplain
with shallow depths on the floodplain.
large variation of water stages in the main channel of the Goulburn River
ood event. An attempt to identify the behaviour of the r
ate stage was performed. The identified roughness curves from different flood
re very close to each other. These curves showed that the roughness
value of this river reach slightly increases with increasing stage. The
dicated that there is some small improvement in the simulated hydrograph
oughness coefficient was considered as the function of stage compared to
hness coefficients were considered as a constant.
uong River, during the flood season the water levels are mainly above
vel with relatively high depths on the floodplains. This river was selected to
some aspects of the roughness identification problem for the channel with
sections.
hydrographs using the identified coefficients successfully re
hydrographs at different gauging stations. This indicates the good quality of the
identified roughness coefficients.
The results showe
197
Chapter 6 Application of Roughness Identification Problem to Natural Rivers
agreement of the simulated and observed hydrographs and the consistency in the
es of the identified floodplain roughness coefficient proved the ability to identify
roughness coefficient for the floodplain by using the flood events with higher
ths.
variation of the floodplain roughness coefficient with flow was also considered
re the roughness functions were formulated as first and second order polynomial
tions of water stage. The computed results indicated that although there were still
e discrepancies in the identified roughness curves, the model could identify the
ation trend of the floodplain roughness coefficient. However, there is no significant
rovement in RMSEs compared to where the roughness coefficients were
sidered as constants. These results showed that due to the fact that there is a higher
ree of uncertainty in identification of the floodplain roughness curve, when t
valu
the
dep
The
whe
func
som
vari
imp
con
deg he
roughness condition of the floodplain does not change much, the consideration of the
be s
The
channel roughness coefficient for an alluvial channel during the flood season by using
the roughness identification model. This study shows that this method could provide a
vari
floo
the
roug
floodplain roughness coefficient as a constant could reduce the computation time but
atisfactory for the practical purposes of flood routing.
Duong River was chosen as a case study to investigate the variation of the main
general picture of the variation of the roughness coefficient with time. The trend of the
ation of the identified roughness coefficient indicated that, at the beginning of
d season, the roughness coefficient has the smallest value. It increases and reaches
highest value during the main flood events. After the main flood events, the
hness coefficient slowly decreases.
198
Chapter 7
Chapter 7
Conclusions and Recommendations
7.1 Summary
Various aspects related to the estimation of roughness coefficient Mannings n using
flow data measurement for both steady flow and unsteady flow were considered in this
work.
For steady flow, the method of using two-point velocity data to estimate Manning's n
was reinvestigated and extended. Two formulae to estimate Mannings n in terms of
the average equivalent roughness and average shear stress were derived. The
sensitivity of the error of the estimated n to the error in the ratio of measured velocities
at two-tenths and eight-tenths of the depth were analysed theoretically and then
verified by experimental data. The proposed method was applied to 14 rivers. For these
rivers, the measured values of Mannings n are available. This enabled the assessment
of the performance of the proposed formulae. The computed results obtained from
these formulae were also compared with the ones obtained from a number of current
empirical formulae.
For unsteady flow, a roughness identification model was developed that allowed the
identification of the roughness coefficients for both single channel and compound
channels. The implicit finite difference four-point Pressmann scheme was adopted to
solve the Saint-Venant equations. For compound channels, the channel cross-section
was treated as a divided channel method, in which for any depth the conveyance of the
compound section is the sum of the main channel and floodplain conveyances. The
199
Chapter 7 Conclusions and Recommendations
algebraic equation system was linearised and solved by using the double sweep
algorithm. The objective function of least square errors between observed and
ge was chosen for this inverse problem and solved by the
Powell algorithm. The model was also extended to the case that can allow the variation
channels were formulated as polynomial functions of stage.
The m coefficient is
affect t
investigation, several recommendations were made to improve the quality of the
or the
terms o the cross-
convey
and the Duong River in the Red River delta, Vietnam
perform igated and several problems related to the
7.2 Conclusions
conclusions were drawn:
velocit
by its application to 14 wide streams and was also compared with a number of other
be used as a means to estimate roughness coefficients for wide streams where two-
simulated discharge/sta
of roughness coefficients with stage. For this case, the roughness coefficients in
odel was tested with synthetic data where the true roughness
known. Then by using the model with synthetic data, several modelling factors that
he quality of the identified roughness coefficient were investigated. From this
roughness identification problem. The performance of the model was tested f
case of compound channels. The variation of the roughness coefficient with the flow in
f function of stage was also considered. Moreover, in cases when
sections along the reach do not change much, the more practical concept of
ance K was used.
The model was applied to two natural rivers, the Goulburn River in Victoria, Australia
. Depending on the characteristics
of the cross-sections, water variation in the main channel and the floodplain, the
ance of the model was invest
roughness of the study reaches were explored.
Based on the results obtained from Chapters 3, 5 and 6, the following major
A. The method to estimate the roughness coefficient Manning's n by using two-point
y measurements
1. The performance of the proposed formulae to estimate the Manning's n was assessed
applicable empirical equations. The results showed that the proposed method can
200
Chapter 7 Conclusions and Recommendations
point velocity data are available and the application conditions are satisfied. For the
proposed formulae, the formula obtained from average two shear stress gave better
erro d that the smoother and deeper the river,
the d
Prei icient.
The
size inations of boundary conditions, and the errors in initial
of t
t of
ct of data errors.
lead to a biased value of the identified coefficient or failure of the
of grid
running the roughness identification model, the grid size should be tested
results. The applicable conditions for the method are also suggested.
2. The sensitivity analysis of the effect of velocity measurement errors on the relative
rs in the estimated Manning's n indicate
the more sensitive the error in n is to the error in the ratio of velocity at two-tenths
epth to that at eight-tenths the depth.
B. The roughness identification problem for unsteady flow in open channels
1. Effects of different modelling factors related to the implicit finite difference
ssmanns scheme on the quality of the identified roughness coeff
se factors included the effects of the weighting coefficient , computational grid
s, different comb
condition related to different combinations of boundary conditions. Investigations
hese factors by using synthetic data led to the following conclusions:
Using higher values of could provide an over-estimation of the roughness
coefficient. Hence, values of from 0.65 to 0.75 were suggested especially
for the rougher or low gradient reaches and for flood events with rapid rise.
Using smaller time steps is an appropriate solution to reduce the effec
on the biases of the identified roughness coefficient. However, the effect of
on the identified roughness coefficient is insignificant compared with the
effe
The selection of grid sizes is one of the important factors that affect the
quality of identified roughness coefficient. Using a large grid size could
identification problem due to numerical instabilities. The selection
size should be appropriate, especially for the cases with high slope channels
or flood events with a short time to the peak. It is suggested that before
201
Chapter 7 Conclusions and Recommendations
by running the hydraulic model using a smaller grid size. If the results
change significantly, then the original time step should be decreased.
the quality of the roughness
t be used. Four combinations of boundary conditions were suggested for
am one (Q-Z); (ii) stage hydrographs
e at
eam
boundary and a rating curve at downstream boundary (Z-Qz). However, for
ficient from the true value when using the Q-Qz and Z-Qz
m.
f observation gauge. Improper choice of
jective
y when the
es in
tion after a period of computation.
) and 10-20 hours for the channels with small slope ( )
The choice of a combination of boundary conditions is one of the most
important modelling factors affecting
identification problem. Using a discharge hydrograph as the downstream
boundary condition led to failure of the identification problem and should
no
the inverse problem: (i) a discharge hydrograph at the upstream boundary
and a stage hydrograph at the downstre
as both upstream and downstream boundary conditions (Z-Z); (iii) a
discharge hydrograph at the upstream boundary and a rating curv
downstream boundary (Q-Qz); and (iv) a stage hydrograph at the upstr
channels with a small slope, there was some deviation of the identified
roughness coef
combinations. Hence, in general, the boundary condition combinations Q-Z
and Z-Z are the most appropriate for the roughness identification proble
The effects of a certain combination of boundary conditions on the quality
of the identified roughness coefficient relate very closely to the choice of
observed data type and locations o
these combination factors can lead to unidentifiable problems or ill-
condition. Observed discharge data should not be used in the ob
function for the Q-Z and Q-Qz combinations and observed stage data
should not be used for the Z-Z and Z-Qz combinations, especiall
observation gauges are located in the upstream part of the channel.
In practice, the errors in initial conditions are unavoidable. The bias
identified roughness coefficient due to these errors can be eliminated by
starting to calculate the objective func
The higher the channel slope the shorter the time required before starting to
calculate the objective function (3-5 hours for the channels with high slope
( 001 . 0 S 0001 . 0 S
are suggested).
202
Chapter 7 Conclusions and Recommendations
2. Roughness identification problem for compound channels
For this case two roughness coefficients need to be identified, one for the
can
individual values of the identified roughness coefficients deviated from the
the true ones. For the same noise level, the quality for the
d data.
k discharges (i.e. longer inundated
For the cases when the value of the roughness coefficient varied with depth,
were identified.
ta
gher deviation of the identified roughness curve from the true

rger flood events.
main channel and the other for the floodplain. The computed results
indicated that when observed data were free from noise, the model
identify very accurately these values of roughness coefficients. When
observed data contains noise even with high noise level, although the
true value, they scattered around the true value and the average values were
very close to
identified roughness coefficient of the main channel was better than the
floodplain one.
The quality of the identified floodplain roughness coefficient is also
dependent on the selection of flood events and the quality of observe
In order to improve the quality of the identified floodplain roughness
coefficient, flood events with higher pea
time and higher depth on the floodplain) and with higher frequency of
observation should be chosen.
3. Identification of the roughness functions
the roughness function(s) was formulated as quadratic polynomial
function(s). The coefficients of the roughness functions
The results indicated that when data were free from noise, the identified
roughness curves were identical or very close to the true ones. When da
contain noise, although the identified roughness curves were biased from
the true ones, they were scattered around the true roughness curves.
There is a hi
one in the regions outside the range of observed stages. This indicates that
there a higher degree of uncertainty in extrapolating the value of the
roughness coefficient in these regions Therefore, the identified roughness
function obtained from a certain flood should be used with care for routing
other la
203
Chapter 7 Conclusions and Recommendations
The results indicated that the main channel roughness function could be
more accurately identified than the floodplain roughness function. Also,
there is some compensation of the values of the roughness coefficients
4. Alte nction
roughness conditions, the conveyance variation with depth is much
ctions because this can reduce the number of unknown parameters. This
tudy indicated
um
function and a power function. The cubic function gave better results.
former one because fewer parameters need to be identified. The results
for practical problems for compound channels where the cross-sections
The
Goulburn River in Victoria, Australia and the Duong River in the Red River delta,
the study reaches were explored which led to the following conclusions:
were verified using other independent flood events. The simulated
between the main channel and the floodplains.
rnative method of using a conveyance fu
Where the cross-section of a channel has abrupt changes in width and
smoother than the variation of the roughness coefficient. It is reasonable to
use a single conveyance function instead of using two separate roughness
fun
alternative method could be applied when the channel is prismatic or the
cross-sections along the channel do not change much. This s
that the model using a conveyance function converged to the optim
values much faster than using roughness functions and also provided
reasonable results.
The conveyance variation with stage was formulated as a cubic polynomial
However, the power function required less computational time than the
indicated the potential applicability of the conveyance identification model
along the channel do not change much.
5. Application of the roughness identification model to natural rivers
roughness identification model has been applied to two river reaches, the
Vietnam. Several aspects related to the roughness identification problem applied to
The roughness coefficients were identified from one or two flood events.
They
204
Chapter 7 Conclusions and Recommendations
hydrographs using the identified roughness coefficients reasonably
reproduced the observed hydrographs at different gauging stations. This
observed data types used in the objective function was also investigated for
ss
observed data. The results reconfirm the findings obtained from the
in the objective function to identify the roughness coefficient, while for Z-Z
The results obtained from the Goulburn River showed that for rivers with a
condition for the cases where discharge data were obtained from rating
rivers, the water stage
ooped rating curves
should be used to adjust the discharge data.
degree of uncertainty in the identified roughness coefficient for the
ever, when the water depth on the
roved (for the case of the Duong River). This can
hole conveyance of the cross-section. And for this
flood events with a higher depth on the
indicated the good performance of the model for real rivers.
The effect of the boundary condition combinations incorporated with the
the real river. The consistency in the values of the identified roughne
coefficient and the ability to reproduce the prototype are two criteria used
to assess the performance of a combination of boundary conditions and
synthetic data. For Q-Z and Q-Qz combinations, stage data should be used
and Z-Qz, discharge data should be used.
mild or flat slope, using a stage hydrograph as the downstream boundary
curves could not overcome the limitation of using a single rating curve as
the downstream boundary condition. Thus, for those
and discharge should be measured independently or l
For the compound channel cases, the results indicated that there was a high
floodplain by using flood events with shallow depth on the floodplain (as in
the case of the Goulburn River). How
floodplains becomes significant, the identified floodplain roughness
coefficients are much imp
be explained as follows: when the depth on the floodplain is much smaller
compared with the main channel depth, the conveyance of the floodplain is
not important for the w
case, the value of floodplain roughness coefficient is not sensitive to the
observed data. Therefore, the
floodplain should be chosen to identify the floodplain roughness
coefficient.
205
Chapter 7 Conclusions and Recommendations
The ability to identify the roughness function of the main channel was
illustrated by the case study of the Goulburn River. The relative agreement
ability to identify the roughness curve of the main channel.
ion. The roughness functions were formulated as first and
ot mean square errors (RMSEs)
ss coefficient as a constant should
efficient with time in alluvial channels is
e unsteadiness and the
coefficient from different events during the flood season for several years,
trend of the variation of the roughness coefficient with time. The results
oefficient for alluvial river during the flood
7.3 R tions for further research
channels in this study, some recommendations for further research in this area are
in the magnitudes and the trends of the main channel roughness curves
obtained from different flood events indicated that the model has a good
The Duong River was chosen to assess the ability to identify the floodplain
roughness funct
second order polynomial functions of the water stage. The computed results
indicated that there were discrepancies in the trend of identified roughness
curves. Also no improvement in the ro
compared with the case where the roughness coefficients were considered
as constants. The results showed that there is high uncertainty in
identification of the floodplain roughness curves. It is suggested that
consideration of the floodplain roughne
be satisfactory for practical purposes of flood routing.
The variation of the roughness co
very difficult to quantify because it depends on several factors such as the
change in bedforms, sediment transport, th
turbulence of high floods. However, the case study for the Duong River
showed that by using the model to identify the values of the roughness
these values of the identified roughness coefficient could provide a general
from this study indicated that this method can be used as a means to assess
the variation of the roughness c
season.
ecommenda
From several aspects studied on the estimation of roughness coefficients in open
listed as follows:
206
Chapter 7 Conclusions and Recommendations
1. For using two-point velocity measurements data, Manning's n obtained from this
earch is restricted to wide channels where the assumption of velocity res
distribution following logarithmic distribution is valid. In natural streams, many
cas
vel stribution law (Sarma et al. 2000). This
law he velocity profile at
If te width, they could be
sim
ability of using these data to estimate Manning's n for rivers with a finite width.
2. Fo
coefficients between the main channel and floodplain could be observed for the
cou l to two-dimensional flow
3. The study case for the Duong River indicated that for alluvial channels, there is a
sea nt with time for
sed f large floods. The
La
understand these effects on the variation of the roughness coefficient.
4. The
to s random functions
the
streams have a finite width (i.e. the ratio of depth and width is less than 10). In this
e, the maximum velocity occurs at some distance below the free surface and the
ocity distribution follows the binary di
distribution combines the logarithmic law of the inner region with the parabolic
of the outer region. Currently, it is possible to measure t
each vertical line such as by using an Acoustic Doppler Current Profiler (ADCP).
these measurements are available for rivers with a fini
used to define the limit of the inner region and then estimate Manning's n by the
ilar method proposed in this study. Further study is necessary to investigate the
r compound channels, the compensation problem of the identified roughness
case of identifying roughness functions, as indicated in Chapter 5. This problem
ld be improved by extending the sub-hydraulic mode
computation models.
considerable variation in the value of the roughness coefficient during the flood
son. The variations of the main channel roughness coefficie
these channels can be attributed to several factors: the movable bedforms,
iment transportation, unsteadiness and the high turbulence o
interaction of two or more of these factors could further affect channel-energy loss.
boratory and field investigations and research on this area can be useful to
observation error is always a random variable. Moreover, it is often convenient
treat physical parameters of roughness coefficients a
especially for the case of compound channels. Therefore, it is necessary to study
roughness identification problem as a stochastic approach, in which the theory
207
Chapter 7 Conclusions and Recommendations
of Ridge Analysis or the Bayesian theory can be applied that have not been
sidered in this study. con
208
References
References
Acton, F. S. (1970). Numerical Methods That Work, Harper & Row, New York.
Arcement, G. J., and Schneider, V. R. (1989). Guide for selecting Manning' roughness
Arnold, P. E., Holland, G., McKerchar, A. I., and Soutter, W. R. (1988). Water
Atanov, G. A., Evseeva, E.G. and Meselhe, E. A. (1999). "Estimation of roughness
Balloffet, A. of floods and tides in open channels",
Baltzer, R. A., and Lai, C. (1968). "C

Abbott, M. B. (1979). Computational Hydraulics: Elements of the Theory of Free
Surface Flows, Pitman Publishing Ltd., London.
Abbott, M. B., and Ionescu, F. (1967). "On the numerical computation of nearly
horizontal flows", Journal of Hydraulic Research, 5(2), 97-117.
Ackermann, N. L., and Shi-igai, H. (1976). "River basin model for simulating flood
control alternatives", Proc. Symposium on inland waterways for navigation,
flood control and water diversions, River's 76.
Afzalimehr, H., and Anctil, F. (1998). "Estimation of Gravel-Bed River Flow
Resistance", Journal of Hydraulic Engineering, 124(10), 1054-1058.
Amein, M. (1966). "Streamflow routing on computer by characteristics", Water
Resources Research, 2(1), 123-130.
Amein, M., and Fang, C. S. (1970). "Implicit flood routing in natural channels",
Journal of the Hydraulic Division, 96(HY11), 2481-2500.
coefficients for natural channels and flood plains, U.S. Geological Survey
Water-Supply Paper 2339, 38 p.
Resources Survey Hydrologist's Field Manual, DSIR Water Sciences Division,
Wellington.
profile in trapezoidal open channel", Journal of Hydraulic Engineering, 125(3),
309-312.
Balachandar, R., Blakely, D., and Bugg, J. (2002). "Friction velocity and power law
velocity profile in smooth and rough shallow open channel flows", Canadian
Journal of Civil Engineering, 29(2), 256-266.
(1969). "One-dimensional analysis
Journal of the Hydraulic Division, 95(HY4), 1429-1451.
omputer simulation of unsteady flow in
waterways", Journal of the Hydraulic Division, 94(HY4), 1083-1117.
209
References
Barnes, H. H. (1967). Roughness characteristics of natural channels, U.S. Geological
Survey Water-Supply Paper 1849, Washington, D. C.
Bathurst, J. C., Li, R. M., and Simons, D. B. (1981). "Resistance equation for large
Bayazit, M. (1976). "Free surface flow in a channel of large relative roughness",
Journal of Hydraulic Research, 14(2).
Becker
Bennett, J. P. (1975). "General model to simulate flow in branched estuaries", Proc.
Bodley, W. E., and Wylie, E. B. (1978). "Control of transients in series channel with
Bousm omentum Transfer for Practical Flow
Computation in Compound Channels", Journal of Hydraulic Engineering,
Boyer, M. C. (1954). "Estimating the Manning coefficient from an average bed
Bray, D. I. (1982). "Flow resistance in gravel bed rivers". In Gravel-bed rivers :
Butler, D., Rock, S. P., and West, J. R. (1978). "Friction coefficient variation with flow
ations,
Oxford, UK, 101-125.
Chanson, H. (1999). The Hydraulics of Open Channel Flow, Arnold, London.
Charlto
Chaudry, M. H. (1993). Open Channel Flow, Prentice-Hall, Englewood Cliffs, New
Jercy.
scale roughness", Journal of Hydraulic Division, ASCE, 107(12).
, L., and Yeh, W. W. G. (1972). "Identification of parameters in unsteady open
channel flows", Water Resources Research, 8(4), 956-965.
The symposium on modelling techniques. - Modelling '75,, 643-662.
gates", Journal of the Hydraulic Division, 104(10), 1395-1407.
ar, D., and Zech, Y. (1999). "M
125(7), 696-706.
roughness in open channels", Transaction of American Geometry Union, 35(6),
957-961.
Bray, D. I. (1979). "Estimating average velocity in gravel-bed rivers", Journal of
Hydraulic Division, 105(HY9), 1103-1122.
Fluvial Processes, Engineering, and Management, R. D. Hey, J. C. Barthurst,
and C. R. Thorne, eds., John Willey and Sons, Chichester, UK, 109-137.
in an urban stream", Journal of Institution of Water Engineers and Scientists,
32(3).
Carling, P. A. (1992). "In-stream hydraulics and sediment transport". In The Rivers
Handbook, P. Calow and G. E. Petts, eds., Blackwell Scientific Public
n, D., Copertino, V. A., and Veltri, M. (1988). "The hydraulic geometry of
some gravel rivers in Britain", Report IT 180, Hydraulic Research Station,
Wallingford, England.
210
References
Chaudry, M. H., and Contractor, D. N. (1973). "Application of the implicit method to
surges in open channels", Water Resources Research, 9(6), 1605-1612.
Chen, C. L. (1980). "Laboratory verification of a dam-break flood model", Journal of
Chen, resistance in open channels: Manning's
formula revisited". In Channel Flow Resistance: Centennial of Manning's
Chen, C. L., and Armbruster, J. T. (1980). "A dam-break wave model: Formulation
Chow,
Chu, W.-S., and Yeh, W. W.-G. (1980). "Parameter identification in estuarine
Clarke, R. T. (1973). "A review of some mathematical models used in hydrology with
Colebr ith particular reference to
the transition region between the smooth and rough pipe laws", Journal of
Colosim "Friction factor evaluation in
gravel bed rivers", Journal of Hydraulic Engineering, 114(8), 861-876.
Cooley
Coon, W. F. (1998). Estimation of roughness coefficients for natural stream channels
Cowan
Cox, R. G. (1973). "Effective hydraulic roughness for channels having bed roughness
Engineers Waterways Experiment Station, Vicksburg, Miss.
Cunge, J. A. (1969). "Discussion of "Computer simulation of unsteady flows in
waterways" by R.A. Baltzer and C. Lai"", Journal of the Hydraulic Division,
95(2), 758-761.
the Hydraulic Division, 106(HY4), 535-556.
C. L. (1991a). "Power law of flow
Formula, B. C. Yen, ed., Water resources publication, Colorado, USA, 206-
240.
Chen, C. L. (1991b). "Unified theory on power laws for flow resistance", Journal of
Hydraulic Engineering, 117(3), 371-389.
and verification", Journal of the Hydraulic Division, 106(HY5), 747-767.
V. T. (1959). Open Channel Hydraulics, McGraw-Hill, New York, U.S.A.
modelling", Proc. The 17th International Conference on Coastal Engineering,
Sydney, 434-435.
observations on their calibration and use", Journal of Hydrology, 19, 1-20.
ook, C. F. (1938 -1939). "Turbulent flow in pipes, w
Institution of Civil Eng., 11, 133-156.
o, C., Copertino, V. A., and M., V. (1988).
, R. L., and Moin, S. A. (1976). "Finite element solution of Saint-Venant
equations", Journal of the Hydraulic Division, 102(HY6), 759-775.
with vegetated banks, U.S. Geological Survey Water-Supply Paper 2441.
, W. L. (1956). "Estimating hydraulic roughness coefficients", Journal of
Agricultural Engineering, 37(7), 473-475.
different from bank roughness", Misc. Paper H-73-2, U.S. Army Corps of
211
References
Cunge, J. A. (1975). "Applied mathematical modelling of open channel flow". In
Chapter 10 of Unsteady Flow in Open Channels, Vol. 1, K. Mahmood and V.
Yevjevich, eds., Water resources publications, Fort Collins, CO., 367-405.
Cunge, Aspects of
Computational River Hydraulics, Pitman Advanced Pub. Program, Boston.
Davids

estuarial river systems", Water Resources Research, 14(5), 777-789.
Ding, Y ness
Coefficients in Shallow Water Flows", Journal of Hydraulic Engineering,
Dingman, S. L., and Sharma, K. P. (1997). "Statistical development and validation of
Diskin, M. H., and Simon, E. (1977). "Procedure for selection of objective functions
Dronkers, J. J. (1969). "Tidal computations for rivers coastal areas, and seas", Journal
Duan, 993). "A shuffled complex evolution
approach for efficient global minimization", Journal of Optimization Theory
Einstei id resistance of composite roughness",
Transaction of American Geophysical Union, 31(4), 603-610.
Falvey
habilitation and Automation of
Irrigation Water Delivery Systems, New York, N.Y., 176-179.
Falvey
laimation, Washington, D. C.
Carolina State University, Raleigh,
North Carolina.
Fletche el",
Journal of Engineering Mechanics, 93(EM3), 45-62.
Fread,
tory, National Weather Service, Silver Spring,
Maryland.
J. A., Holly, F. M., Jr, and Verwey, A. (1980). Practical
on, B., Vichnevestsky, R., and Wang, H. T. (1978). "Numerical techniques for
estimating best distribution of Manning's roughness coefficients for open
., Jia, Y., and Wang, S. S. Y. (2004). "Identification of Manning's Rough
130(6), 501-510.
discharge equations for natural channels", Journal of Hydrology, 199, 13-35.
for hydrologic simulation models", Journal of Hydrology, 34, 129-149.
of the Hydraulic Division, 95(HY1), 29-77.
Q., Gupta, V. K. and Sorooshian S. (1
and its Applications, 76(3), 501-521.
n, H. A., and Banks, R. B. (1950). "Flu
, H. T. (1987). "Philosophy and implementation of gate stroking", Proc.
Symposium on Planning, Operation, Re
, H. T., and Lunning, P. C. (1979). "Gate stroking", Report REC-ERC-79-7,
U.S. Department of the Interior, Bureau of Rec
Fang, C. S. (1968). "Mathematical solution of complete equations of unsteady flow in
open channels", PhD Dissertation, North
r, A. G., and Hamilton, W. S. (1967). "Flood routing in an irregular chann
D. L. (1974). "Numerical properties of implicit four-point finite different
equations of unsteady flow", HRL-45, NOAA Tech, Memo NWS HYDRO-18,
Hydrologic Research Labora
212
References
Fread, D. L. (1976). "Flood routing in meandering rivers with flood plains", Proc.
Symposium on Inland Waterways for Navigation, Flood Control and Water
Diversions, River's 76, Colorado State University, Fort Collins, Colorado, 16-
Fread, D. L. (1991). "Flood routing model and Manning n". In Channel flow
Fread, S FLDWAV Model: Theoretical description, Hydrologic
Research Laboratory, National Weather Services, NOAA, Silverspring.
Fread,
it nonlinear dynamic routing models", Proc. ASCE
National Hydraulic Engineering Conference, San Francisco, California, 1569-
Fread, D. L., and Smith, G. F. (1978). "Calibration technique for 1-D unsteady flow
French, R. H. (1985). Open Channel Hydraulics, McGraw Hill, New York, U.S.A.
French
of boundary generated turbulence: Field
verification." Technical report No. 39, Environmental and Water Resources
Furuya, Y., and Fijita, H. (1967). "Turbulent boundary layers on wire screen
Garde, R. J., and Raju, K. G. R. (1978). Mechanics of Sediment Transportation and
Gilbert les and Longitudinal Dispersion", Journal
of Hydraulic Engineering, 123(9), 816-820.
Gonzalez, J. A., and Melching, C. S. (1996). "Analysis of open-channel velocity
measurements collected with an Acoustic Doppler Current Profiler", Proc. The
Goyen, A. G. (1982). "Case study: Sullivans Creek, Canberra", Proc.
Gradowczyk, M. H., and Ivanissevich, L. (1971). "A model for flood computation in
35.
resistance: Centennial of Manning's Formula, B. C. Yen, ed., Water resources
publications, Colorado, USA, 421-435.
D. L. (1998). NW
D. L., and Lewis, J. M. (1993). "Selection of x and t computational steps
for four-point implic
1573.
models", Journal of the Hydraulic Division, 104(7), 1007-1044.
, R. H., and McCutcheon, S. C. (1977). "The stability of a two layer flow
without shear in the presence
Engineering, Vanderbilt University, Nashville, Tann.
roughnesses", Bulletin, Japan Society of Mechanical Engineering, 10(37).
Alluvial Stream Problems, Wiley Eastern, New Delhi.
, V. B. (1997). "Stream Velocity Profi
1
st
International Conference on New/Emerging Concepts for Rivers, Chicago,
Illinois, USA.
Forum/Workshop on Energy Losses Parameters for Flow Profile Computation
Programs, Melbourne.
the Panama area", Proc. 14th International Congress, IAHR, Paris, France,
199-204.
213
References
Griffiths, G. A. (1981). "Flow resistance in coarse gravel bed rivers", Journal of
Hydraulic Division, ASCE, 107(HY7), 899-918.
Grijsen, J. G., and Meijer, T. J. G. P. (1980). "On modelling of flood flows in large
river systems with flood plains", Publication No. 227, Delft hydraulics
laboratory.
Haan, C
Hartree, D. R. (1952). "Some practical methods of using characteristics in the
calculation of non-steady compressible flows", Rep. LA-HU-1, Los Alamos
Sci. Lab.
Henderson, F. M. (1966). Open Channel Flow, Macmillan Publishing Co., Inc., New
York, U.S.A.
Hey, R. D., Bathust, J. C., and Thorne, C. R. (1982). Gravel Bed Rivers: Fluvial
Processes, Engineering, and Management, John Willey and Sons Ltd.
Hicks, D. M., and Mason, P. D. (1991). Roughness Characteristics of New Zealand
Rivers, Water Resources Survey DSIR Marine and Freshwater.
Higginson, N. N. J., and Johnston, H. T. (1991). "Estimation of n in alluvial streams".
In Channel Flow Resistance: Centennial of Manning's Formula, B. C. Yen, ed.,
Water resources publications, Colorado, USA, 195-205.
Hinze, J. O. (1975). Turbulence, McGraw-Hill, New York, USA.
Horton, R. E. (1933). "Separate roughness coefficients for channel bottom and sides",
Engineering News Record, 111(22), 652-653.
Hsu, M.-H., Fu, J.-C., and Liu, W.-C. (2003). "Flood routing with real-time stage
correction method for flash flood forecasting in the Tanshui River, Taiwan",
Journal of Hydrology, 283(1-4), 267-280.
Irmay, S. (1949). "On steady flow formulae in pipes and channels", Proc. IAHR 3rd
Congress, Grenoble, French.
James, M., and Brown, B. J. (1977). "Geometric parameters that influence floodplain
flow", Research report H-77-1, U.S. Army Engineers Waterways Experiment
Station, Vicksburg, Miss.
Jarrett, R. D. (1984). "Hydraulics of high-gradient streams", Journal of Hydraulic
Engineering, 110(11), 1519-1539.
Jarrett, R. D. (1985). "Determined of roughness coefficients for streams in Colorado",
Report 85-4004, U.S. Geological Survey Water Resources Investigation.
. T. (2002). Statistical Methods in Hydrology, Iowa State Press, Iowa, USA.
214
References
Jarrett, R. D. (1990). "
Water Resources
Hydrological and hydraulic research in mountain streams",
Research, 26, 419-429.
Jolly, J. P., and Yevjevich, V. (1971). "Amplification criterion of gradually varied
C
Julien, P. Y., Klaassen, G. J., Ten Brinke, W. B. M., and Wilb 02).
stu Rhi ng 199 l of
aulic E ing, 1 12),
ng, and K . S. ( "Un flow
1th C ss, I eoul, K rea, 868- 7.
Kamphuis, J. W. (1 Math al tidal study of St. Lawrence River", Journal of
, 96(HY3), 643-664.
Kamphuis, J. W. (1 ete ion of sand roughness for fixed beds", Journal of
, 2(1 3-20
Ka . (19 "A d ive G n sc ,
lic En ing, , 450
Keller, R. J. (1982). "Case study: Conflu urray Rivers at
ill" For rksh Ene ss eters low
, Me e, F1
Keulegan, G. H. (1938). "Laws of turbulence flow in open channels", l of
, 21, 707-741.
Khatibi, R. H., Wi , J. J. R., and leaton, P. R. (1997). "Identification
em of open-channel friction param Journal of Hydraulic
, 123(12), 1078-1088.
Khatibi, R. H., Wor on, P nd W s, J. J. R. (2000). "Parameter quality
", al of H draulic R earch,
(6), 447-4
Krishnamurthy, M., and Christ B. A ). "E ent rou hness for shallow
", J of H ic Di 98(H 2257-2 63.
Ladson A. R., Lan M., S G. M derso G., an Rutherfur , I. D.
lo tance r Au riv stralia Journal o Water
, ( ew).
Jin, M., and Fread, D. L. (1997). "Dynamic flood routing with an explicit and implicit
numerical solution schemes", Journal of Hydraulic Engineering, 123(3), 166-
173.
single peak waves", Hydrology paper No. 51, Colorado State University, Fort
ollins, Colorado.
ers, A. W. E. (20
8 flood", Journa "Case
Hydr
dy: Bed re
ngineer
sistance of
28(
ne River duri
1042-1050.
Jun, K. ., Hwa S Y. J., im, J 2005). steady model with variable
roughness coefficient", Proc. 3 ongre AHR, S o 87
970). " ematic
the Hydraulic Division
974). "D rminat
Hydraulic Research 2), 19 3.
topodes, N 84). issipat alerki heme for open-channel flow"
Journal of Hydrau gineer 110(4) -466.
ence of little Murray and M
Swan H , Proc. um/Wo op on rgy Lo Param fo F r
Profile Computation Programs lbourn -F10.
Journa
Research of the National Bureau of Standards
lliams Worm
probl eters",
Engineering
mleat . R., a illiam
conditions in open-channel inverse problem Journ y es
38 58.
ensen, . (1972 quival g
channels ournal ydraul vision, Y12), 2
, g, S. mart, ., An n, B. d d
(2006). "F w resis in fou stralian ers", Au n f
Resources in revi
215
References
Lai, C. ( e multiple-
re gical Surv.
Prof. Paper 575-D, D273-D278.
Lai, 76). " putational aspe n dy
simulat by t tho a s
posium o nstead in Cha Univ of N tle-
pon-Tyne, nd, Pa 1, 1-1
Lai, C. 986). "Num al modeling of unsteady open-channel flow". In Advances in
ydro-scien C. Ye Acad ress 63.
La ). "Com ensive d of teris dels w si on",
lic En ing, 114(4), 1074-1097.
Lai, C., Schaffranek, R. W., and Baltzer, (1978). "An operational system for
plimentin lation ls, a tudy . 26th Annual Hydraulics
l Conf e, University of Maryland, College park, M
Lane, E. W. (1951). cussio lope rge f e for a uvial stre ms, by
E.C Schnackenberg", Pr , Wellington,
Lang, S., Ladson, A And B. (2 A re of empirical equations for
ating s rough nd th lication to four streams in Victoria",
n Jo l of W sour 1), 6
Langbein, W. B. (19 Deter on of ng's n from vertical velocity curves",
, 618-620.
L and ge, J. 975) eric thods of solution of the
steady flow uation ,
ater Resources Public ort C Col 89-181
Ligg ., and W ser, D 967) erenc tions sha ater
Jo of Hy c Div 93(E 9-71.
19 eterm n of nnin ficie measured bed
in al ch ", U.S. Geological Survey Journal of Research,
3), 285-29
and H. S. Wilf, eds., Academic Press, New York, 95-164.
f channels with different
tific Research Institute of
Hydraulic Engineering, 9, 238-241 (in Russian).
1967). "Computation of transient flows in rivers and estuaries by th
ach method of characteristics", Geological survey research, Geolo
C. (19 Some com cts of one and two-dim
d h
ensio al unstea
International flow
Sym
ion he me of c rac ic terist ", . Proc
er ity n U y Flow Open nnels, s ewcas
U Engla per D 2.
(1 eric
H ce, B. n, ed., emic P , 161-2
i, C. ( 988 1 preh metho charac tics mo fo flo r mulati
Journal of Hydrau gineer
R. A.
im g simu mode case s ", Proc
Specia erenc aryland, 415-454.
"Dis n on s
oc. New Zealand Institution of Engineers
discha ormula ll a
435-438.
., and erson, 004). " view
estim tream ness a eir app
Australia urna ater Re ces, 8( 9-82.
40). " minati Manni
Transactions of American Geophysical Union
iggett, J. A., Cun A. (1 . "Num al me
un eq s". In Chapter 4 of Unsteady Flow in Open Channels
W ation, F ollins, orado, .
ett, J. A oolhi . A. (1 . "Diff e solu of the llow-w
equation", urnal drauli ision, M2), 3
Limerinos, J. T. (
r ughness
70). "D inatio t a he M g coef nt f om r
o natur annels
4( 1.
Lister, M. (1960). "The numerical solution of hyperbolic partial equations by method
of characteristics". In Mathematical Method for Digital Computers, A. Ralston
Lotter, G. K. (1933). "Consideration on hydraulic design o
roughness of walls", Transaction, All-Union Scien
216
References
McDowell, D. M., and Prandle, D. (1972)
Journal of the W
. "Mathematical model of River Hooghly",
aterways Harbours and Coastal Engineering Division,
Meyer-Peter, E., and Muller, R. (1948). Proc. 3rd
Meeting of IAHR, Stockholm, 39-64.
Mitchell, D. (1982). "Case study: Werribee River, Werribee", Proc. ACADS
ters for Flow Profile Computation
s, Melbourne.
Motayed, A. K., and Krishnamuthy, M. (1978). "Composite roughness of nature
channels", Proc. The 26th Conference of Hydraulic division speciality -
Verification of mathematical and physical models in hydraulic engineering,
Uni. Of Maryland - Maryland.
Nalluri, C., and Judy. (1985). "Interaction between main channel and flood plain
flow", Proc. 21st IAHR congress, Melbourne, Australia, 337-382.
Nezu, I., and Rodi, W. (1985). "Experimental study on secondary currents in open
channel flow", Proc. 21st IAHR Congress, Melbourne, Australia, 115-119.
Nguyen, H. T., and Fenton, J. D. (2004). "Identification of roughness in open
channels", Proc. 6th International Conference on Hydro-Science and
Engineering, Bris
Nguyen, Q. K., and Kawano, H. (1995). "Sim
channel networks", Journal of Hydraulic Engineering, 121(10), 744-750.
n Russian).
"Verification o
ted natural and constructed stream channels in A
Paper 1584: U.S. Geological Survey.
98(WW2), 225-242.
"Formulas for bed-load transport",
Forum/Workshop on Energy Losses Parameters for Flow Profile Computation
Programs, Melbourne.
Morris, G. (1982). "Case study: Wimmera River, Horsham", Proc. ACADS
Forum/Workshop on Energy Losses Parame
Program
bane, Australia, CD-ROM.
Nguyen, N. K. (1994). "Practical guideline for development of the numerical
modelling for unsteady flow and salt concentration", The Institution for Water
Resources Planing and Management, Vietnam (in Vietnamese).
ultaneous solution for flood routing in
Nikora, V. I., and Smart, G. M. (1997). "Turbulence characteristics of New Zealand
gravel-bed Rivers", Journal of Hydraulic Engineering, 123(9), 764-773.
Pavlovskii, N. N. (1931). "On a design formula for uniform movement in channels
with nonhomogenous walls", Transaction, All-Union Scientific Research
Institute of Hydraulic Engineering, 3, 157-164 (i
Phillips, J., and Ingersoll, T. (1998). f roughness coefficients for
selec rizona", Professional
217
References
Posada, P. J. R., and Bras, R. L. (1982). "Automatic parameter estimation of large
conceptual rainfall-runoff model: a maximu lik ih d a ro ", Report No.
267, Ralph M. Parsons Laboratory for Wa R ou es nd drodyn ics,
MIT.
m el oo pp ach
ter es rc a Hy am
Powell, M. J. D. (1965). "A method for minimising a sum of squares of non-linear
Preissm
Press, W. H., Teukolsky, S. A., Vetterling, W. T., and Flannary, B. P. (1992).
sity
numerical methods fo
, Journal of the Hydraulic Division, 100(7), 879-89
Quinn, F. R., and Wylie, E. B. (1972). "Transient analysis of the Detroit River by the
plic
.
Symposium on inland waterways for navigation, flood control and water
Rajaratnam Beltaos, S. (1976). "Roughness effects in
rectangular free overfall", Journal of the Hydraulic Division, ASCE, 102(HY5),
14.
Ra
Rao, S. S.
Rao, V. S., Contractor, D.
iver's 76, Colorado State University,
Fort Collins, Colorado, 421-433.
(199
flood plains", Journal of Hydrology, 171(1-2), 75-91.
i, A on
functions without calculating derivatives", The Computer Journal, 7, 303-307.
ann, A. (1961). "Propagation des intumescences dans les canaux et rivieres",
Proc. First congress of the French association for computation, Grenoble,
France, 433-442.
"Minimization or Maximization of Functions". In Chapter 10 of Numerical
Recipes in Fortran - The Art of Scientific Computing, Cambridge Univer
Press.
Price, R. K. (1974). "Comparison of four r floods routing", ASCE
Proc. 9.
im it method", Water Resources Research, 8(6), 1461-1469.
Radojkovic, M. (1976). "Mathematical modelling of rivers with flood plains", Proc
diversions, River's 76, Colorado State University, Fort Collins, Colorado, 56-
64.
, N., Muralidhar, D., and
599-6
mesh, R., Datta, B., Bhallamudi, S. M., and Narayana, A. (2000). "Optimal
estimation of roughness in open channel flows" Journal of Hydraulic
Engineering, 126(4), 299-303.
(1996). Engineering Optimization -Theory and Practice, A Wiley-
Interscience Publication, New York.
N., and Tiyamani, C. (1976). "Optimal river cross sections
for flood routing", Proc. Symposium on Inland Waterways for Navigation,
Flood Control and Water Diversions, R
Rashid, R. S. M. M., and Chaudhry, M. H. 5). "Flood routing in channels with
Raudkiv . J. (1976). Loose boundary hydraulic, Pergam Press, Oxford.
218
References
Richards, G., and Hollis, G. E. (1980). "Relationship of roughness coefficient
Manning's n and discharge in an urban river", Journal of Institution of Water
vigation, Flood Control and Water
River's 76, Colorado State University, Fort Collins, Colorado,
1209-1228.
Rouse, H. (1965). "Critical analys Journal of the
nagement
study. Phase 3 - Shepparton to Kanyapella - Main report", Cameron McNamara
PTY. Ltd. ACIL Australia PTY. Ltd.
G. (1979). "EMBER - numerical model of an em
Hydraulics Research Station, Walingford, England.
Cambridge University, Cambridge, England.
Sanchez, D., and W
Schaffranek, R. W., and Lai, C. (1996). "Friction-term response to boundary-condition
Schmitz, G., and Edenhofer, J. (1983). "Flood routing in the Danube River by the new
Implicit Method of Characteristics (IMOC)", Proc. International Conference
rd
Engineers and Scientists, 34, 357-360.
Riggs, H. C. (1976). "A simplified slope area method for estimating flood discharges
in natural channels", U.S. Geological Survey Journal of Research, 4(3), 285-
291.
Robbins, L. G. (1976). "Roughness characteristics of the lower Mississippi", Proc.
Symposium on Inland Waterways for Na
Diversions,
is of open-channel resistance",
Hydraulic Division, 91(HY4), 1-25.
Rural Water Commission of VIC. (1987). "Lower Goulburn flood plain ma
Samuels, P. banked river", IT 183,
Samuels, P. G. (1985). "Models of open channel flow using Preissmann's Scheme",
estphal, J. A. (1999). "Optimized calibration for unsteady flow
modeling using a genetic algorithm", Proc. The 29
th
Annual Water Resources
Planning and Management Conference, Tempe, Arizona, USA.
Sargent, R. J. (1979). "Variation of Manning's n roughness coefficient with flow in
open river channels", Journal of the Institution of Water Engineers and
scientists, 33(290-294).
Sarma, K. V. N., Prasad, B. V. R., and Sarma, A. K. (2000). "Detailed Study of Binary
Law for Open Channels", Journal of Hydraulic Engineering, 126(3), 210-214.
type in flow models", Journal of Hydraulic Engineering, 122(2), 73-81.
Schlichting, H. (1960). Boundary Layer Theory, McGraw-Hill, New York.
on Application of Mathematical Modelling, Mitteilungen des Inst. fur
Meereskunde der 3 , University of Hamburg, West Germany.
219
References
Sen, D. J., and Garg, N. K. (1998). "Efficient Solution Technique for Dendritic
Channel Networks Using FEM", Journal of Hydraulic Engineering, 124(8),
Computation Programs, Melbourne.
Simons, D. B., and Senturk, F. (1977). Sediment Transport Technology, Water
ort Collins, Colo.
Singh, A. K., Porey, P. D., and Ranga Raju, K. G. (1997). "Criterion for location of
dynamic flood routing", Journa
n
schemes modelling open channel flow", Proc. International Conference on

am erkin
technique", Proc. The 15th Coastal Engineering Conference, Honolulu,
Hawaii, 108-111.
Sorooshian, S., and Dracup, J. A. (1980). "Stochastic parameter estimation procedures
for hydraulic rainfall runoff models: correlated and heteroscedastic error
rch, 16(2), 430-442.
Solomatine, D.P. (1995). "The use of global random search methods for models
a
Hydraulic Research, London, September 1995.
es Symposium, May 20-23, 2002, Melbourne: Institution of Engineers,
a.
der geschwindigkeitsformel und der
rauhigkeitszahlen fur strome, kanale und geschlossene leitungen", Mitteilungen
idgenossischen Amtes fur Wasserw
slated as "Some contributions to the p
roughness factor in rivers, canals, and closed conduits).
Sturm
-Hill Publishing Co.,
New Dehli.
831-839.
Sheehan, D. B., G. (1982). "Case study: Ovens and King River, Wangaratta", Proc.
ACADS Forum/Workshop on Energy Losses Parameters for Flow Profile
resources Publications, F
downstream control for l of Hydrology, 196(1-
4), 66-75.
Skeels, C. P., a d Samuels, P. G. (1989). "Stability and accuracy analysis of numerical
Computational Modelling and Experimental Methods in Hydraulics, June 13-
16, 1989, Dubrovnik, Yugoslavia, 149-157.
Smith, L. H., and Cheng, R. T. (1976). "Tidal stre flow solved by Gal
cases", Water Resources Resea
c libration", Proc. The 26th Congress of the International Association for
Stewardson M. J. and Anderson B. G. (2002). "Variation in the flow resistance of
natural channels with discharge", Proc. The 27th Hydrology and Water
Resourc
Australi
Strickler, A. (1923). "Beitragezur frage
des E irtschaft, 16, Bern, Switzerland,
(Tran roblem of the velocity formula and
, T. W. (2001). Open Channel Hydraulics, McGraw-Hill, New York, NY 10020.
Subramania, K. (1982). Flow in Open Channels, Tata McGraw
220
References
Sun, N.-Z. (1994). Inverse Problems in Groundwater Modeling, Kluwer Academic,
Dordrecht , Boston.
Task Force Committee of US Army Corps of Engineers. (1963). "Friction factors in
open channels", Journal of Hydraulic Division, 89(HY2), 97-143.
Tingsanchali, T., and Ackermann, N. L. (1976). "Effects of overbank flow in flood
computations", Journal of the Hydraulic Division, 102(HY7), 1013-1025.
U.S. Army Corps of Engineers (2002). HEC-RAS River Analysis - User's Manual -
Version 3.1.
U.S. Army Corps of Engineers (1993). Engineering and Design - River Hydraulics,
http://www.usace.army.mil/publications/eng-manuals/em1110-2-1416.
Urquahart, W. J. (1975). Hydraulics: Engineering Field Manual, U.S. Department of
Agriculture. Soil Conservation Service, Washington, D. C.
Valentine, E. M., Benson, I. A., Nalluri, C., and Bathurst, J. C. (2001). "Regime theory
and the stability of straight channels with bankfull and overbank flow", Journal
of Hydraulic Research, 39(3), 259-268.
Vasiliev, O. F., Temnoyeva, T. A., and Shugrin, S. M. (1965). "Numerical method for
the calculation of unsteady flows in open channels", Izv. Akad. Nauk SSSR
Mechanics, No. 2, 17-25, (in Russian).
Verwey, A. (1975). "On external and internal boundary conditions and theirs influence
on net flow and mean water levels in systems of channels", Proc. The 16th
Congress of IAHR, S. Paulo, Brazil, 26-33.
Wahl, T. L. (2000). "Analyzing ADV Data Using WinADV", Proc. 2000 Joint
Conference on Water Resources Engineering and Water Resources Planning &
Management, Minneapolis, Minnesota, USA.
Wasantha Lal, A. M. (1995). "Calibration of riverbed roughness", Journal of
Hydraulic Engineering, 121(9), 664-670.
Watt, C. S., Hobbs, J. M., and Boldy, A. P. (1980). "Hydraulic transients following
valve closure", Journal of the Hydraulic Division, 106(HY10), 1627-1640.
Wiggert, J. M., Taylor M. R., and Contrator, D.N. (1976). "Optimization of an implicit
scheme flow routing model", Proc. International Symposium on Unsteady flow
in open channels, University of Newcastle-on-Tyne, BHRA fluid Engineering,
Cranfield, Bedford, England.
The Institute of Water Resources Research. (2004). "Sediment balance study of Red
River and Thai Binh River System. Component", Report of Project KC-08-11,
The Institute of Water Resources Research, Hanoi, Vietnam. (in Vietnamese).
221
References
Williams, B. J., and Yeh, W. W.-G. (1983). "Parameter estimation in rainfall-runoff
models", Journal of Hydrology, 63, 373-393.
Wormleaton, P. R., and Karmegam, M. (1984). "Parameter optimization in flood
routing", ASCE Proc., Journal of the Hydraulic Division, 102(12), 1799-1810.
Wu, C. M. (1976). "Movable bed flood routing in open channel and its calibration",
Proc. Symposium on Inland Waterways for Navigation, Flood Control and
Water Diversions, River's 76, Colorado State University, Fort Collins,
Colorado, 36-55.
Wu, Z. Y., and Simpson, A. R. (2001). "Competent genetic-evolutionary optimization
of water distribution systems", Journal of Computing in Civil Engineering,
ASCE, 15(2), 89-101.
Wylie, E. B. (1969). "Control of transient free-surface flow", Journal of the Hydraulic
Division, 95(1), 347-361.
Wylie, E. B., Streeter, V. L., and Suo, L. (1993). Fluid transients in systems, Prentice-
Hall, Inc., Englewood Cliffs, New Jersey.
Yang, S.-Q., Tan, S.-K., and Lim, S.-Y. (2004). "Velocity distribution and dip-
phenomenon in smooth uniform open channel flows", Journal of Hydraulic
Engineering, 130(12), 1179-1186.
Yeh, W. W.-G., and Becker, L. (1973). "Linear programming and channel flow
identification", Journal of the Hydraulic Division, 99(11), 2013-2021.
Yen, B. C. (1973). "Open channel flow equations revisited", Journal of the
Engineering Mechanics Division, 99(EM5), 979-1009.
Yen, B. C. (1991). "Hydraulic resistance in open channels". In Channel Flow
Resistance: Centennial of Manning's Formula, B. C. Yen, ed., Water resources
publications, Colorado, USA, 1-135.
Yen, B. C. (2002). "Open Channel Flow Resistance", Journal of Hydraulic
Engineering, 128(1), 20-39.
Yen, C. L., Comer, G. H., and Holtan, H. N. (1974). "USDALL-74 flood routing
method", Plant physiology institute report No. 7, U.S. Department of
Agriculture, Agriculture research service.
Yevjevich, V., and Barnes, A. H. (1970). "Flood routing through storm drains". In
Hydrology paper Nos. 43 and 46, Colorado State University, Fort Collins,
Colorado.
"A one-dimensional flow model for flow over
trapezoidal profile weirs", Proc. 6th International conference on Hydro-
Science and Engineering, Brisbane, Australia, CD-ROM.
Zerihun, Y. T., and Fenton, J. D. (2004).
222
References
Zippe, H. J., and Graf, W. H. (1983). "Turbulent boundary layer flow over permeabl
and non-permeable rough surface", Journal of Hydraulic Research, 21(1).
Zoppou, C. (1984). "Numerical modelling of unsteady flows in rivers and flood
plains", PhD thesis, University
and O'Neill, I. C. (1981). "Num ethods and boundary conditions
for the solution of unsteady flow problems", Proc. Conference on Hydraulics
in Civil Engineering, Sydney, Australia, 16-24.

e
of Melbourne, Melbourne, Australia.
Zoppou, C., erical m
223
Appendices
Appendix A
Measured and computed roughness coefficients using
velocity data of the 14 rivers
Table A.1 Measured and computed roughness coefficients (using velocity data)
(Rivers are listed in ascending order of their average absolute errors)
Computed n
Average relative
error
b
(%) River
Discharge
(m3/s)
Measured
n
a
Eq. (3.23) Eq. (3.27) Eq. (3.23) Eq. (3.27)
Mitta Mitta River
1
28.51 0.043 0.041 0.041 8.48 7.73
36.88 0.039 0.035 0.038
41.09 0.044 0.035 0.034
61.94 0.041 0.037 0.04
89.53 0.041 0.040 0.037
102.73 0.045 0.047 0.045
115.79 0.043 0.043 0.043
127.29 0.043 0.040 0.034
144.73 0.046 0.054 0.049

Tahunatara
2
2.93 0.049 0.036 0.039 16.23 12.31
9.97 0.034 0.029 0.037
18.10 0.031 0.029 0.028
36.00 0.029 0.034 0.026
Tambo
1
129.00 0.041 0.040 0.040 10.17 19.96
428.43 0.042 0.033 0.030
701.66 0.045 0.048 0.032
Whirinaki
2
34.3 0.044 0.039 0.036 17.38 12.46
35.3 0.046 0.052 0.051
61.8 0.039 0.045 0.045
64.0 0.037 0.048 0.039
Ongarue
2
10.5 0.050 0.033 0.032 20.46 13.73
14.8 0.034 0.028 0.029
35.1 0.030 0.028 0.027
35.8 0.028 0.027 0.028
41.7 0.035 0.028 0.029
241.0 0.022 0.031 0.023
Acheron
1
3.17 0.047 0.036 0.037 20.33 19.51
15.74 0.034 0.032 0.034
19.10 0.042 0.035 0.040
22.82 0.045 0.038 0.032
26.50 0.043 0.027 0.027
34.55 0.046 0.043 0.034
42.55 0.046 0.026 0.035
72.94 0.043 0.037 0.037 (cont.)
224
Appendices
Table A.1 Measured and computed roughness coefficients (cont.)
Computed n
Average relative
error
b
(%) River Discharge
Measured
n
a
Eq. (3.23) Eq. (3.27) Eq. (3.23) Eq. (3.27)
Grey
2
73 0.031 0.035 0.031 25.03 16.67
116 0.026 034 0.035 0.
217 0.026 0.032 0.031
334 0.025 0.034 0.032
917 0.030 035 0.032 0.
1110 0.026 0.034 0.029
Rangitaiki
2
47.5 0.050 027 0.032 23.55 24.92 0.
74.0 0.042 0.027 0.031
98.0 0.046 028 0.030 0.
107.0 0.044 0.037 0.034
120.0 0.044 043 0.037 0.
144.0 0.043 044 0.037 0.
Waipapa
2
12.5 0.040 031 0.028 24.71 24.23 0.
22.9 0.034 044 0.046 0.
38.5 0.027 0.033 0.025
Merriman
1
Creek 8.56 0.076 0.054 0.040 24.48 44.76
15.12 0.079 0.064 0.052
31.31 0.080 059 0.039 0.
36.49 0.080 0.061 0.043
Wanganui
2
(2) 6.15 0.024 025 0.026 30.75 29.12 0.
at Wairehu canal 6.22 0.025 0.033 0.033
9.89 0.023 026 0.025 0.
26.5 0.022 0.034 0.035
31.9 0.024 0.036 0.033
Oakden
2
Canal 4.9 0.037 0.037 0.034 38.23 33.39
8.0 0.031 045 0.042 0.
13.9 0.030 049 0.047 0.
20.5 0.027 0.039 0.036
Poutu
2
2 3 24.26 .31 0.017 0.027 0.021 57.5
5.80 0.016 0.025 0.02
Wanganui
2
(1) 6.79 0.022 0.032 0.028 65.26 56.76
at Te Whaiau Canal 10.70 0.018 0.03 0.029
12.40 0.019 0.034 0.033
13.00 0.020 0.034 0.033
Note:
1
indicates rivers in Australia;
2
indicates rivers in New Zealand;
a
Measured roughness
values are obtained from Hick and Mason (1991) and Ladson et al. (2006);
b
Average absolute
error is defined as ( ) N n n n
meas meas comp
/ /

where N is the number of computed n for each


river, n
meas
and n
comp
are measured n and computed n respectively.

225
Appendices
Appendix B
Solution algorithm for a general river network
B.1 The equation system for a general river network
ogether at nodes.
Nodes are intersection of one, two or several river segments. Each segment links two
node f a
river network.

Figu o iv k
In the network, nod re cod at s, o total number of
nodes). The coding is princi itr fr th f nodes on the
plane. However, it is advised e e
Segments are basic te diffe m he th enant equation
system is develope r ever nt ib ti 1 of Chapter 4
(Equations (4.8) an .9)). F e f e ions (4.10) and
(4.11) in Chapter r the e re d points of i
th

segment are renumbered as (B B ow
J Z I
A model river is divided into many segments. These segments link t
s: the upstream node and the downstream node. Figure B.1 shows an example o
re B.1 M del of a r er networ
es a ed by n ural array from 1 t N (N is
pally arb ary and ee from e place o
to code th m in som order.
fini rence ele ents of t system, e Saint-V
d fo y segme as descr ed in Sec on 4.3.3.
d (4 rom thes equations or a segm nt, Equat
4 fo discharg s at upst am and ownstream
.1) and ( .2) as foll s
i Di i U i Ui
Z H Q
i
+ = (B.1)
and
i Di i Ui i Di
G Z M Z L Q + = (B.2)
1
2
3
j-1
j
l
i
k
k+1
m
ii
iii
L
M
226
Appendices
w
downstream points of a segment respectively, , , , and are
coefficients deter the for e desc in Se 4.3.3
These equations become the nt eq or ment, by which
the discharges at the upstream d ns de ment are linear
functions of water l on bo ns an tion at nodes links them to
form an equation system for t n
Write the water bal equa d od se some coming
segments and some ving s ( e e We have:

here Z is the stage, Q is the discharge, subscripts U and D denote the upstream and
i i i
H I J ,
i
L ,
i
M
i
G
mined by mula ribed ction .1.
fundame al linear uations f one seg
node an the dow tream no of a seg
evel th sectio . The bal ce condi
he whole etwork.
ance tion at no es. For n e i, inter ction of
lea segment see Figur B.1 as an xample).
( )
t
i

Z Z
i i

F Q
j
e com
D
+

(B.3)
where is the sum of lat t ce nt f) node i, is
the area of the free ace of ed an ly to node i
Substituting an by E (1
of water level. The common f

Q =

Q
L
leav
U
e

L
Q
i
F eral in/ou flow con ntrated (i o/or out o
surf submerg plain (if y), open adjacent
U
Q d Q quation ) and (2) into Equation (3) we get an equation
orm is
D
i l il i ii k ik j ij
b Z a Z a Z a Z a = + + + + ...... (B.4)
The water balance equation at each node provides one linear equation relating to the
unknown water level of the node, and that of neighbouring nodes.
Equation (B.4) is common for every node, including the boundary nodes. The
coefficients at the boundary nodes are determined depending on the type of the
boundary condition as described in Section 4.3.3.1 of Chapter 4. All the nodes of the
whole system provide a linear equation system:
[ ]{ } { } B Z A = (B.5)
where is coefficient matrix with ( [ ] A N N ) elements, { } Z is the vector of unknown
water stages, is vector of free terms with N elements. { } B
227
Appendices
B.2 Solution of the linear equation system
The matrix A in the Equation System (B.5) has the following main features: (i) It is
very sparse; (ii) The non-zero elements lay symmetrically with the main diagonal; and
(iii) the main diagonal elements are predominant. This linear equation system is solved
by elimination method. The problem is how to select a special way of elimination
appropriate to the speciality of the equation system cited above in order to minimize
the computation.
The process of solution used in VRSAP model (Vietnam River System and Plain
model) (Nguyen 1994) is adopted. It consists of two steps.
Step 1: Classically eliminating, the unknown from the range i, we have:

i
Z
ii
i
l
ii
il
k
ii
ik
j
ii
ij
i
a
b
Z
a
a
Z
a
a
Z
a
a
Z + = ....... (B.6)
The substitution of by (B.6) into the range j, k, , l must bring to these range the
new non-zero elem , , , , .that may not exist before.
Such enlargement of the range could be avoided if the elimination follows a logical
order. The elimination should begin from the ranges having fewest non-zero elements,
corresponding to the nodes at the boundary point such as nodes 1, L and M in Figure
B.1. After that, the ranges having only 3 non-zero elements, corresponding to the
intermediate nodes in the branches, such as k, k+1, m in Figure B.1 can be then
eliminated, without making the remain system more complicated. This procedure make
the remained matrix become smaller and smaller and more and more dense.
Step 2: After the fulfilment of the step 1, the system of linear equations has only
unknowns of water level at junction nodes. This equation system can be solved by any
classical methods. In this study the Gauss elimination method is adopted.
The return sweep finally uses Equation (B.6) and gives as solutions, following the
inverse order, from step 2 to step 1 with very limited number of computation. Finally,
the discharges are computed using Equation (B.1) and (B.2).

i
Z
ents
ik
a
jk
a
kj
a
kl
a
lj
a
lk
a
i
Z
228
Appendices
Formulate the formulae for structure segments
In a river reach, there may be some structures along the channel such as bridges, weirs
or orifices/sluices. In this case, special segments namely as structures segments are
used. For structure segments, the formulae for weir, sluice or orifice can be used such
as:
For a free flow weir:
2 / 3
) ( 2
w D
Z Z g mb Q = (B.7)
For a submerged flow weir: ) ( 2 ) (
D U w D
Z Z g Z Z b Q = (B.8)
where m and are the discharge coefficients of free flow weir and submerged flow
weir respectively which depends on geometric characteristics of weir, b is the width of
weir, g is the acceleration due to gravitation, and are the upstream and
downstream water elevations of weir respectively, and is the crest elevation.
For a free flow orifice/sluice:

U
Z
D
Z
w
Z
) ( 2
O U O
Z Z g A Q = (B.9)
For submerged orifice/sluice: ) ( 2
D U O
Z Z g A Q = (B.10)
where is the discharge coefficient of orifice, is the area of orifice/sluice, and
are the upstream and downstream water elevations of orifice respectively, and
is the elevation of the centre of (for orifice) or the water elevation of the contracted
cross-section.
The equation of continuity here is evidently:

O
A
U
Z
D
Z
O
Z
O
A
Q Q Q
D U
= = (B.11)
The above non-linear formulae of structures and Equation (B.11) can also be
transformed into linear equations in the forms of Equations (B.1) and (B.2). For
example, for a structure segment with free flow weir we have:
2 / 1
) ( 2
w Ui i i
Z Z b g m H L = =
229
Appendices
0 = =
i i
I M
w w Ui i i
Z Z Z b g m J G
2 / 1
) ( 2 = =
Therefore, for a river with structures, all segments of the river including structure
segments are linked together by continuity equation at nodes. This also produces a
linear equation system in the form of Equation system (B.5).
230
Appendices
Appendix C
Powells method
C.1 General
Powells method is an extension of the basic pattern search method that is named by the
author who first discovered (Powell, 1964). It is an efficient and most widely used method
for direct search approach to find the minimum of a function of several variables without
calculating derivatives.
Powell (1964) proved that this method is a method of conjugate directions
1
. For a
quadratic function, it will minimize the function in a finite number of steps. Thus, it is a
quadratically convergent
2
. Since a general nonlinear function can be approximate
reasonably well by a quadratic function near its minimum, a conjugate directions metho

d
is expected to speed up the quadratic convergence of even general nonlinear objective
functions. The subsection below will describe the algorithm of Powells method.
C.2 Algorithm
In order to illustrate the basic idea of Powells method, an example of a two-variable
function is considered. The progress of Powells method is illustrated in Figure 1. In this
figure, starting with point 1, the function is first minimised once along each of the
coordinate direction
1
x and
2
x and point 3 is obtained. Then the corresponding pattern
direction
1
S is defined. After that, the function is minimised along direction
1
S . This leads

1
Conjugate directions: let A=[A] be a NN symmetric matrix. A set of N non-zero vectors (or directions)
{S
i
} is said to be conjugate (more accurately A-conjugate) if
S
i
T
AS
j
= 0 for all i j , i =1, 2, , N, j = 1, 2,,N.

2
A method is quadratic convergent if a minimisation method, using exact arithmetic, can find the minimum
point in N steps while minimising a quadratic function in N variables.



231
Appendices
232
to point 4. For the next cycle of minimisation, one of the coordinate directions (the
direction in the present case) is discarded in favour of the pattern direction. Thus the
function is minimized along and and point 6 is obtained. Then a new pattern
direction is generated as shown in the figure. For the next cycle of minimization,
one of the previous used directions (the direction in this case) in favour of the
newly generated pattern direction. Then, by starting from point 7, the function is
minimized along direction and , thereby obtaining points 8 and 9, respectively.
This procedure is continued until the desired minimum point is found.
The general flow chart of this method is described in Figure 2. In this figure, N is
number of variables, X
0
is the point at the beginning of each minimisation cycle, is
the coordinates of i
th
search direction of each minimisation cycle, X
i
is the location of
the new point obtained from each minimisation step along direction.
There is a problem with Powells quadratically convergent algorithm. The progress
can break down before the minimum point is found. This is because the search
directions may become linearly dependent or almost linearly dependent during
numerical computation (Rao 1996, Press et al. 1992). For this case, one of the
solutions to fix this problem is trying to find a few good directions along narrow
valleys instead of N necessarily conjugate directions (Acton 1970). The algorithm of
modified Powells method to minimise a nonlinear function f is described by Press et
al. (1992). The basic idea of this version is still to take X
N
-X
0
as a new direction. The
change is to discard the old direction along which the function f made its largest
decrease. Although this direction is the best of the previous iteration (minimisation
cycle), it is also likely to be a major component of the new added direction, so
dropping it provides the best chance of avoiding a build-up of linear dependence. Press
et al. (1992) also included a couple of exceptions where the old set of directions will
be used for the next iteration. The upper and lower limits of parameter are introduced
following the external penalty function method as described in (Rao 1996).
A subroutine of Powells method with this version obtained from Press et al. (1992) is
adopted for the optimisation model in this study. It is written in FORTRAN 77.
1
x

2
x
1
S
2
S
2
x
1
S
2
S
i
u
i
u
Appendices
233


Figure C.1 Contours of the function values to show the progress of Powells method






2
3
4
5
6
7
8
9
1
1
x
Contour of the
response surface
1
x
Appendices
234


Figure C.2 Flow chart for Powells method













Move X
N
to the minimum along
direction u
N
call this point X
0
No
Set i = 1
Start with X
0
Save position X
0
Move X
i-1
to the minimum along
direction u
i
call this point X
i
Set u
i
= u
i+1
, i = 1 to N
and u
N
= X
N
-X
0
Stop
Set i = i+1
Is i = N+1?
Is X
0
optimum?
Is X
i
optimum?
Yes
Yes
Yes
No
No
Set u
i
equal to the coordinate unit
vectors 1 = i to N
Appendices
Appendix D
Preliminary results of the identified roughness
coefficients using synthetic data
D.1 Identified roughness coefficient for the single channel
Table D.1 Identified roughness coefficients from different noisy data samples
Identified n with noise level
Sample
0.05 = = 0.10 = 0.15 = 0.20
1 0.03480 0.03624 0.03478 0.03325
2 0.03476 0.03414 0.03447 0.03587
3 0.03580 0.03419 0.03687 0.03491
4 0.03508 0.03522 0.03689 0.03749
5 0.03479 0.03372 0.03328 0.03508
6 0.03437 0.03604 0.03505 0.03689
7 0.03521 0.03478 0.03614 0.03587
8 0.03512 0.03592 0.03556 0.03369
9 0.03526 0.03490 0.03399 0.03284
10 0.03500 0.03631 0.03484 0.03540
11 0.03577 0.03402 0.03489 0.03587
12 0.03458 0.03524 0.03580 0.03594
13 0.03464 0.03539 0.03426 0.03497
14 0.03514 0.03347 0.03267 0.03103
15 0.03434 0.03380 0.03495 0.03421
16 0.03566 0.03572 0.03430 0.03474
17 0.03490 0.03429 0.03767 0.03765
18 0.03555 0.03667 0.03421 0.03318
19 0.03499 0.03400 0.03238 0.03757
20 0.03578 0.03479 0.03580 0.03822
21 0.03454 0.03575 0.03400 0.03485
22 0.03521 0.03481 0.03496 0.03605
23 0.03525 0.03680 0.03377 0.03404
24 0.03395 0.03483 0.03698 0.03196
25 0.03440 0.03433 0.03556 0.03490
To be continued in the next page


235
Appendices
Table D.1 Identified roughness coefficients from different noisy data samples (cont.)
Identified n with noise level
Sample
0.05 = = 0.10 = 0.15 = 0.20
26 0.03534 0.03407 0.03366 0.03281
27 0.03466 0.03603 0.03357 0.03409
28 0.03581 0.03421 0.03662 0.03571
29 0.03448 0.03499 0.03576 0.03398
30 0.03493 0.03582 0.03326 0.03542
31 0.03544 0.03575 0.03594 0.03655
32 0.03495 0.03487 0.03485 0.03485
33 0.03592 0.03680 0.03767 0.03878
34 0.03498 0.03493 0.03493 0.03490
35 0.03471 0.03433 0.03400 0.03398
36 0.03456 0.03407 0.03362 0.03318
37 0.03565 0.03603 0.03671 0.03698
38 0.03465 0.03421 0.03385 0.03365
39 0.03500 0.03501 0.03502 0.03505
40 0.03550 0.03582 0.03614 0.03674
41 0.03393 0.03315 0.03277 0.03152
42 0.03529 0.03551 0.03587 0.03594
43 0.03523 0.03533 0.03580 0.03587
44 0.03473 0.03442 0.03412 0.03404
45 0.03565 0.03601 0.03682 0.03694
46 0.03565 0.03599 0.03635 0.03689
47 0.03475 0.03442 0.03418 0.03409
48 0.03484 0.03472 0.03464 0.03444
49 0.03476 0.03444 0.03426 0.03413
50 0.03389 0.03301 0.03238 0.03103
Mean 0.03500 0.03496 0.03502 0.03499
Standard
deviation
0.00051 0.00092 0.00135 0.00179

D.2 Chi-square ( ) goodness-of-fit test on the identified roughness
coefficients
The chi-square test was chosen to test the how the errors induced in the identified
roughness coefficients are distributed. It is one of the most commonly used tests for
goodness of fit of a data set to theoretical frequency distributions (Haan 2002). This
test makes a comparison between actual number of tested data and the expected
2

236
Appendices
number of the data (expected according to the distribution under test) that fall in the
class intervals. The test statistic is calculated from the relationship:

( )

=
k
i i
i i
c
E
E O
1
2
2
(D.1)
Where k is the number of class intervals, and O
i
and E
i
are the observed and expected
number of the data in the i
th
class interval. The distribution of s a chi-square
distribution with degree of freedom, where p is the number of parameter
estimated from the data. The hypothesis that the data are from the specified
distribution is rejected at significant level if
(D.2)
In this test, the 50 identified roughness coefficients for each noise level (see Table D.1)
are divided into 7 class intervals. The data are hypothesised following normally
distributed ( ). The class intervals are selected with the probably 1/7 (the
expected nu er of each interval is the same). The computed results of are shown
in Tables D.2 to D.5. The tabulated value of is 9.49 (at 5% significant level).
From these tables it can be seen that all the co puted are much smaller than the
tabulated value of . This indicates that at 5% significant level, it is significant to
hypothesise that the identified roughness coefficients are normal distributed.






2
c
i
1 p k

2
1 , 1
2

<
p k c
2 = p
mb
2
c

2
4 , 95

m
2
c

2
4 , 95

237
Appendices
Table D.2 The calculated for 50 samples of noise level 0.05
Class
number
Lower
bound
Upper
bound
Expected
number
Observed
number
Calculated
2
=
2
c

1 0 0.03446 7.143 6 0.183
2 0.03446 0.03472 7.143 8 0.103
3 0.03472 0.03491 7.143 8 0.103
4 0.03491 0.03510 7.143 7 0.003
5 0.03510 0.03529 7.143 8 0.103
6 0.03529 0.03555 7.143 4 1.383
7 0.03555

7.143 9 0.483
50 50 2.361

Table D.3 The calculated for 50 samples of noise level 0.10
Class
number
Lower
bound
Upper
bound
Expected
number
Observed
number
Calculated
2
=
2

1 0.00000 0.03398 7.143 6 0.183
2 0.03398 0.03444 7.143 10 1.143
3 0.03444 0.03479 7.143 7 0.003
4 0.03479 0.03512 7.143 6 0.183
5 0.03512 0.03548 7.143 7 0.003
6 0.03548 0.03594 7.143 5 0.643
7 0.03594 0.00000 7.143 9 0.483
50 50 2.641

Table D.4 The calculated for 50 samples of noise level 0.15
Class
number
Lower
bound
Upper
bound
Expected
number
Observed
number
Calculated
2
=
2

1 0.00000 0.03361 7.143 8 0.103
2 0.03361 0.03427 7.143 8 0.103
3 0.03427 0.03478 7.143 4 1.383
4 0.03478 0.03526 7.143 7 0.003
5 0.03526 0.03577 7.143 6 0.183
6 0.03577 0.03643 7.143 9 0.483
7 0.03643 0.00000 7.143 8 0.103
50 50 2.361

238
Appendices
Table D.5 The calculated for 50 samples of noise level 0.20
Class
number
Lower
bound
Upper
bound
Expected
number
Observed
number
Calculated
2
=
2

1 0.00000 0.03309 7.143 7 0.003
2 0.03309 0.03398 7.143 6 0.183
3 0.03398 0.03467 7.143 9 0.483
4 0.03467 0.03531 7.143 7 0.003
5 0.03531 0.03600 7.143 8 0.103
6 0.03600 0.03690 7.143 5 0.643
7 0.03690 0.00000 7.143 8 0.103
50 50 1.521

D.3 Identified roughness coefficients for the compound channel
Table D.6 Identified roughness coefficients from different noisy data samples for the
compound channel
Noise level
0.05
Noise level
0.10
Noise level
0.15
Noise level
0.20 = = = =
Sample
n
c
n
f
n
c
n
f
n
c
n
f
n
c
n
f
1 0.02856 0.03929 0.02914 0.03661 0.02971 0.03482 0.03022 0.03301
2 0.02784 0.04204 0.02767 0.04207 0.02749 0.04234 0.02732 0.04239
3 0.02760 0.04564 0.02724 0.04973 0.02689 0.05455 0.02656 0.05801
4 0.02774 0.04373 0.02745 0.04555 0.02715 0.04770 0.02681 0.05049
5 0.02821 0.04000 0.02844 0.03801 0.02865 0.03618 0.02888 0.03446
6 0.02808 0.04076 0.02818 0.03948 0.02828 0.03836 0.02836 0.03716
7 0.02816 0.03997 0.02833 0.03799 0.02851 0.03612 0.02870 0.03449
8 0.02771 0.04340 0.02744 0.04500 0.02719 0.04645 0.02693 0.04831
9 0.02856 0.04057 0.02913 0.03938 0.02964 0.03827 0.03012 0.03765
10 0.02755 0.04432 0.02712 0.04671 0.02671 0.04957 0.02630 0.05282
Note: n
c
and n
f
are identified roughness coefficients for the main channel and floodplain
respectively.
239
Appendices
Appendix E
Data for the Goulburn River
E.1 Cross-section data
Table E.1 Thalweg elevation at each cross-section in the Goulburn River from
Shepparton to Loch Garry
Note: Cross-sections 1-21 from survey data in 1977 and cross-sections from 21-25 from survey
data in1981. * indicates the distance from the junction with the Murray River.

Cross-
section
Distance
(km)
Bed level
(m)
Note
1 157.9 98.86 At Shepparton
2 157.4 98.4
3 156.3 97.82
4 155.1 98.03
5 154.7 99.91 Anabranch U/S at 154.7
6 154.7 98.04
7 154.7 98.98 Anabranch D/S at 154.7
8 153.5 96.13
9 152.5 95.94
10 151.4 96.46 At Golf Club- Shepparton
11 150.1 97.43
12 149.5 95.9
13 148.1 95.57
14 146.8 97.25
15 145.5 97.84
16 143.8 99.32
17 142.2 97.34
18 140.5 97.66
19 138.8 98.39
20 137.3 97.36
21 136.8 97.12
22 135.5 97.97
23 132.97 97.32
24 129.07 96.32
25 125.83 96.47
26 122.81 96.56
27 120.63 96.80 At Loch Garry
240
Appendices

















Figure E.1 Ten representative cross-sections of the Goulburn River (the number under
the figure indicates the distance from the junction of the Goulburn River and Murray
River)

241
Appendices


















Figure E.1 Ten representative cross-sections of the Goulburn River (Cont.)

242
Appendices
243


















Figure E.2 The computational scheme of the Goulbourn River

Anabranch
Shepparton
Shepparton Golf Club
Loch Garry
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Appendices
E.2 Flow data








Shepparton
102.0
104.0
106.0
108.0
110.0
0 100 200 300 400
Time (hours)
S
t
a
g
e

(
m
)
Shepparton Golf Club
102.0
104.0
106.0
108.0
110.0
0 100 200 300 400
Time (hours)
S
t
a
g
e

(
m
)
Loch Garry
98.0
100.0
102.0
104.0
106.0
0 100 200 300 400
Time (hours)
S
t
a
g
e

(
m
)
244
Figures E.3 Observed stages for flood event 21/07-06/08/1978









Figure E.4 Observed stages for flood event 08/08-31/08/1978
Shepparton
102.0
104.0
106.0
108.0
110.0
0 100 200 300 400 500 600
Time (hours)
S
t
a
g
e

(
m
)
Shepparton Golf Club
102.0
104.0
106.0
108.0
110.0
0 100 200 300 400 500 600
Time (hours)
S
t
a
g
e

(
m
)
Loch Garry
98.0
100.0
102.0
104.0
106.0
0 100 200 300 400 500 600
Time (hours)
S
t
a
g
e

(
m
)
Appendices
245








Figure E.5 Observed stages for flood event 27/09-09/10/1978









Figure E.6 Observed stages for flood event 04/09-25/09/1979

Shepparton
102.0
104.0
106.0
108.0
110.0
0 50 100 150 200 250 300 350
Time (hours)
S
t
a
g
e

(
m
)
Shepparton Golf Club
102.0
104.0
106.0
108.0
110.0
0 50 100 150 200 250 300 350
Time (hours)
S
t
a
g
e

(
m
)
Loch Garry
98.0
100.0
102.0
104.0
106.0
0 50 100 150 200 250 300 350
Time (hours)
S
t
a
g
e

(
m
)
Shepparton
102.0
104.0
106.0
108.0
110.0
0 100 200 300 400 500
Time (hours)
S
t
a
g
e

(
m
)
Shepparton Golf Club
102.0
104.0
106.0
108.0
110.0
0 100 200 300 400 500
Time (hours)
S
t
a
g
e

(
m
)
Loch Garry
98.0
100.0
102.0
104.0
106.0
0 100 200 300 400 500
Time (hours)
S
t
a
g
e

(
m
)
Appendices
246








Figure E.7 Observed stages for flood event 28/09-31/10/1979









Figure E.8 Observed stages for flood event 29/07-05/08/1980

Shepparton
102.0
104.0
106.0
108.0
110.0
0 200 400 600 800
Time (hours)
S
t
a
g
e

(
m
)
Shepparton Golf Club
102.0
104.0
106.0
108.0
110.0
0 200 400 600 800
Time (hours)
S
t
a
g
e

(
m
)
Loch Garry
99.0
101.0
103.0
105.0
107.0
0 100 200 300 400 500 600 700 800 900
Time (hours)
S
t
a
g
e

(
m
)
Shepparton
102.0
104.0
106.0
108.0
0 50 100 150 200
Time (hours)
S
t
a
g
e

(
m
)
Shepparton Golf Club
102.0
104.0
106.0
108.0
0 50 100 150 200
Time (hours)
S
t
a
g
e

(
m
)
Loch Garry
98.0
100.0
102.0
104.0
0 50 100 150 200
Time (hours)
S
t
a
g
e

(
m
)
Appendices
247








Figure E.9 Observed stages for flood event 25/06-05/07/1981







Figure E.10 Rating curves at Shepparton and Loch Garry gauges

Shepparton
103.0
105.0
107.0
109.0
111.0
0 50 100 150 200 250 300
Time (hours)
S
t
a
g
e

(
m
)
Shepparton Golf Club
103.0
105.0
107.0
109.0
111.0
0 50 100 150 200 250 300
Time (hours)
S
t
a
g
e

(
m
)
Loch Garry
99.0
101.0
103.0
105.0
107.0
0 50 100 150 200 250 300
Time (hours)
S
t
a
g
e

(
m
)
At Shepparton
0
100
200
300
400
500
600
100 102 104 106 108 110 112
Stage (m)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
At Loch Garry
0
100
200
300
400
500
96 98 100 102 104 106 108
Stage (m)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Appendices
248
Appendix F
Data for the Duong River
F.1 Cross-section data
Table F.1 Thalweg elevation at each cross-section in the Duong River
Cross-
section
Distance*
(km)
Bed level
(m)
Note
1 0.00 -7.09 Red River and Duong River junction
2 2.30 -5.7 at Thuong Cat
3 4.60 -2.08
4 5.70 -4.84
5 6.70 -10.26
6 7.70 -10.3 at Duong Bridge
7 9.40 -5.83
8 12.55 -4.73
9 15.55 -1.89
10 16.25 -12.09
11 17.67 -2.53
12 19.42 -1.32
13 19.97 -4.72
14 20.92 -3.11
15 21.51 -10.81
16 22.82 -1.88
17 26.02 -5.2
18 27.97 -1.03
19 30.82 -4.96 at Ben Ho
20 34.62 -7.73
21 38.17 -10.32
22 39.52 -6.54
23 42.32 -7.28
24 44.22 -5.21
25 45.47 -3.02
26 46.77 -12.53
27 47.92 -2.31
28 49.84 -5.52
29 52.34 -2.73
30 56.69 -7.52
31 58.74 -4.47
32 61.04 -4.79
33 64.01 -4.79 at Pha Lai
Note: * indicates the distance from the junction of the Red River and the Duong River
Appendices
249
Km 2.30
-8
-4
0
4
8
12
16
0 200 400 600 800
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 4.60
-8
-4
0
4
8
12
16
0 200 400 600 800
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 6.70
-12
-8
-4
0
4
8
12
0 200 400 600 800
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 9.40
-12
-8
-4
0
4
8
12
0 300 600 900 1200
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 12.55
-4
0
4
8
12
0 500 1000 1500 2000 2500
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 17.67
-4
0
4
8
12
0 500 1000 1500 2000 2500
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 19.42
-4
0
4
8
12
0 400 800 1200 1600 2000
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 21.51
-12
-8
-4
0
4
8
12
0 300 600 900 1200 1500
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Figure F.1 Sixteen representation cross-sections of the Duong River (Km indicate the
distance from the junction of the Red River and the Duong River.)
Appendices
250
Km 26.02
-6
-2
2
6
10
0 400 800 1200 1600 2000
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 30.82
-6
-2
2
6
10
0 200 400 600 800 1000 1200
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 34.62
-8
-4
0
4
8
12
0 200 400 600 800 1000 1200
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 39.52
-8
-4
0
4
8
12
0 400 800 1200 1600 2000
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 42.32
-8
-4
0
4
8
12
0 500 1000 1500 2000 2500 3000
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 47.92
-4
0
4
8
12
0 300 600 900 1200 1500
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 52.34
-4
0
4
8
12
0 300 600 900 1200 1500
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Km 58.74
-6
-2
2
6
10
0 200 400 600 800 1000 1200
Distance (m)
E
l
e
v
a
t
i
o
n

(
m
)
Figure F.1 Sixteen representative cross-sections of the Duong River (Km indicate the
distance from the junction of Red River and the Duong River.) (Cont.)


Appendices
251


















Figure F.2 The computational scheme of the Duong River

Thuong Cat
Ben Ho
Pha Lai
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
Duong Bridge
Appendices
252
F.2 Flow data








Figure F.3 Observed stages and discharges for flood event 14/8-31/8/1995








Figure F.4 Observed stages and discharges for flood event 16/8-30/8/1996

Thuong Cat
8.0
9.0
10.0
11.0
12.0
0 100 200 300 400
Time (hours)
S
t
a
g
e

(
m
)
Thuong Cat
2000
3000
4000
5000
6000
0 100 200 300 400
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Ben Ho
6.0
6.5
7.0
7.5
8.0
8.5
9.0
9.5
0 100 200 300 400
Time (hours)
S
t
a
g
e

(
m
)
Ben Ho
4.0
4.5
5.0
5.5
6.0
6.5
7.0
0 100 200 300 400
Time (hours)
S
t
a
g
e

(
m
)
Thuong Cat
9.0
9.5
10.0
10.5
11.0
11.5
0 100 200 300 400 500
Time (hours)
S
t
a
g
e

(
m
)
Thuong Cat
2000
3000
4000
5000
6000
0 100 200 300 400 500
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Ben Ho
6.0
6.5
7.0
7.5
8.0
8.5
9.0
0 100 200 300 400 500
Time (hours)
S
t
a
g
e

(
m
)
Pha Lai
4.0
4.5
5.0
5.5
6.0
6.5
7.0
0 100 200 300 400 500
Time (hours)
S
t
a
g
e

(
m
)
Appendices
253








Figure F.5 Observed stages and discharges for flood event 22/7-6/8/1997









Figure F.6 Observed stages and discharges for flood event 26/7-5/8/1998
Thuong Cat
9.0
9.5
10.0
10.5
11.0
0 100 200 300 400
Time (hours)
S
t
a
g
e

(
m
)
Thuong Cat
2500
3500
4500
5500
0 100 200 300 400
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Ben Ho
6.5
7.0
7.5
8.0
8.5
0 100 200 300 400
Time (hours)
S
t
a
g
e

(
m
)
Pha Lai
4.5
5.0
5.5
6.0
6.5
0 100 200 300 400
Time (hours)
S
t
a
g
e

(
m
)
Thuong Cat
7.0
8.0
9.0
10.0
11.0
0 50 100 150 200 250 300
Time (hours)
S
t
a
g
e

(
m
)
Thuong Cat
2000
3000
4000
5000
6000
0 50 100 150 200 250 300
Time (hours)
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)
Ben Ho
5.5
6.0
6.5
7.0
7.5
0 50 100 150 200 250 300
Time (hours)
S
t
a
g
e

(
m
)
Pha Lai
3.0
3.5
4.0
4.5
5.0
0 50 100 150 200 250 300
Time (hours)
S
t
a
g
e

(
m
)

S-ar putea să vă placă și