You are on page 1of 200

Flood Studies Update

Technical Research Report

Volume III
Hydrograph Analysis
Kieran OConnor, Monomoy Goswami and Duncan Faulkner

Derived from Technical Research Reports by


NUI Galway and JBA Consulting

Volume I Rainfall Frequency


Volume II Flood Frequency Estimation
Volume III Hydrograph Analysis
Volume IV Physical Catchment Descriptors
Volume V River Basin Modelling
Volume VI Urbanised and Small Catchments
Volume III Hydrograph Analysis

Abstract
This volume presents methods of constructing the hydrograph to accompany a flood peak of
given return period at gauged and ungauged sites in Ireland. The peak flow is typically
derived by the methods of flood frequency estimation presented in Volume II.

The preferred route to constructing the hydrograph is based on the analysis of hydrograph
widths. Flood hydrographs are made comparable by characterising them by so-called
hydrograph widths. For example, W75 represents the duration in hours over which the flow
exceeds 75% of the peak value.

Flood hydrographs for many rivers in Ireland typically have a relatively complex shape, with
subsidiary peaks and undulations. These reflect the general pattern of successive periods of
heavy rainfall leading to the flood but are also moderated by features of the river network.

The aim of hydrograph width analysis is to construct a simplified flood hydrograph shape that
is characteristic of the catchment. Two approaches are considered. In one, a particular
parametric form is imposed on the hydrograph shape. The shape found most useful is a
modified version of the Gamma distribution. In the other approach, the characteristic
hydrograph is selected by direct analysis (and averaging) of hydrograph widths.

Regression-based expressions allow hydrograph descriptors at ungauged sites to be estimated


from physical catchment descriptors (PCDs). Using the parametric approach, a flood
hydrograph with unit peak is constructed as a continuous curve. This is rescaled by the
relevant peak flow to obtain the required design flood hydrograph.

A standalone software package, with graphical user interface, called HWA (Hydrograph
Width Analysis), was created both as a research tool and as an aid to constructing design
flood hydrographs at any site, ideally based on existing or updated flow data.

Although some variation of hydrograph width (and hence hydrograph shape) was noted, no
systematic pattern of variation with peak flow magnitude, season of occurrence or pre-event
flow value could be established. As expected, arterial drainage was typically found to lead to
narrower and peakier hydrographs.

Urbanised catchments are not well represented in the 89 stations subjected to hydrograph
width analysis. This is one of several areas of application where the Interactive Bridge
Invoking the Design Event Method extends the reach of FSU methods appreciably. The
HWA and IBIDEM software packages are available through the FSU Web Portal.

Research on Flood Event Analysis is briefly summarised in Appendix B.

Office of Public Works 2014

ii
Volume III Hydrograph Analysis

Further information about the research


FSU Technical Research Reports (TRRs) are available in their original
form for researchers and practitioners who seek additional information
about a method. The original TRRs sometimes document exhaustive
application of a method to many catchments. In others, additional options
are reported.

Inevitably, the relevance of the original TRRs is influenced by OPW


decisions on which methods to implement, and how best to arrange and
support them. Readers who consult the original TRRs will notice editorial
re-arrangements and compressions, and occasional changes in notation and
terminology. These were judged necessary to enhance understanding and
use of the FSU methods amongst general practitioners. More significant
changes are labelled explicitly as editorial notes.

iii
Volume III Hydrograph Analysis

Contents
i
Abstract ii
Contents iv
Notation xi
Symbols xi
Subscripts xii
Abbreviations and descriptor names xii
Glossary of terms xiii
1 Introduction 1
1.1 Overview 1
1.2 The goal and premise of hydrograph width analysis 2
1.3 Catchment selection 3
1.4 Physical catchment descriptors (PCDs) 4
1.5 The characteristic hydrograph 5
1.6 HWA software 5
2 Processing the flow data for HWA 6
2.1 Data screening and checking 6
2.1.1 Data handling 6
2.1.2 Missing flow data 6
2.1.3 Scrutiny of annual maximum flood peaks 6
2.1.4 Stations affected by arterial drainage 7
2.2 Defining the time-window of the flood hydrograph 7
2.3 Selection of flood hydrographs 8
2.4 Numbering of flood hydrographs 9
2.5 Seasonal distribution of flood events 9
2.6 Filtering of selected hydrographs 10
2.6.1 Desire for broadly unimodal hydrographs 10
2.6.2 Decoupling the main flood response within a complex flood event 10
2.6.3 Discarding the complex segments 10
3 Deriving the characteristic hydrograph at gauged sites 12
3.1 Standardising the flood hydrographs 12
3.2 Calculation of hydrograph widths at particular exceedance levels 12
3.3 Procedures for constructing the characteristic hydrograph 13
3.4 Split-sample and whole-sample calibration 15
3.5 Deriving the median hydrograph 16
3.5.1 Basic method 16
3.5.2 Anomalies in the derived median hydrograph 17
3.5.3 Improving the derived median hydrograph 17
4 The parametric approach 19
4.1 Objectives 19
4.2 General approach 19
4.3 UPO-Gamma model for the characteristic hydrograph 20
4.3.1 Gamma distribution 20
4.3.2 Peak of Gamma distribution 20

iv
Volume III Hydrograph Analysis

4.3.3 Gamma model with peak at time zero 20


4.3.4 Gamma model with unit peak at time zero 21
4.3.5 Formulation in terms of hydrograph rise time Tr 21
4.3.6 Families of hydrographs constructed using the model 21
4.3.7 Example application of UPO-Gamma model 22
4.4 UPO-ERR-Gamma model for the characteristic hydrograph 23
4.4.1 Formulation 23
4.4.2 Method of fitting 24
4.5 Method of fitting the parametric model 24
4.5.1 Objective function 25
4.5.2 Optimisation scheme 26
4.5.3 Performance evaluation 27
4.6 Reproduction of flood hydrographs of verification events 28
4.7 Other methods 30
5 Performance of methods at gauged sites 31
5.1 Introduction 31
5.2 Relative performance in verification compared to that in calibration 32
5.3 Complexity of hydrographs at Stations 06011 and 34018 33
5.4 Variability in hydrograph widths at some stations 34
5.5 Attenuated response at some stations 38
5.6 General guidance 38
5.7 Results of whole-sample calibration 39
5.7.1 Derived median hydrograph and its descriptors 39
5.7.2 UPO-ERR-Gamma model and its parameters 40
5.7.3 Stations where the flood hydrograph recedes faster than it rises 43
5.8 Characteristic hydrographs on the River Suir 44
5.9 Hydrograph width analysis at gauged sites a summary 45
5.10 Flood hydrographs having sustained peaks 46
5.10.1 Where the hydrograph shape reflects the temporal pattern of rainfall 46
5.10.2 Where the hydrograph shape is characteristic of the station 46
6 Constructing the characteristic hydrograph at ungauged sites 48
6.1 Introduction 48
6.1.1 Links with other parts of the FSU 48
6.1.2 Assumptions and difficulties 48
6.2 Selection of dependent variables (DVs) 49
6.3 Selection of independent variables (IVs) 50
6.4 Additional notes 51
6.4.1 Treatment of Gamma shape parameter n 51
6.4.2 Software used for the regression analysis 52
6.5 Correlation studies 52
6.5.1 Inter-correlations between PCDs 52
6.5.2 Individual correlations between DVs and initially selected IVs 56
6.5.3 Choosing a subset of PCDs to use as IVs 56
6.5.4 Checking the Normality of the DVs 57
6.5.5 Final selection of the independent variables; a note on the use of BFI 58
6.6 The regression method used 58
6.7 Illustrative results: Estimating W75 when BFI available 58

v
Volume III Hydrograph Analysis

6.7.1 Regression models and their performance evaluation 58


6.7.2 Checking the possible influence of collinearity 60
6.7.3 Checking the logical consistency of the model 60
6.7.4 Additional checks 62
6.8 Recommended models for use at ungauged sites 67
6.8.1 The final models 67
6.8.2 Model performance 69
6.8.3 Additional notes on the regression models 70
7 Ancillary investigations 71
7.1 Variation of hydrograph width with peak flow 71
7.2 Variation of hydrograph width with pre-event minimum flow 73
7.3 Variation of hydrograph width with time of year 74
7.4 Effect of arterial drainage on hydrograph widths 75
8 Constructing the characteristic flood hydrograph 79
8.1 Topics covered 79
8.2 Features of the methods 79
8.3 Allowances in design flood hydrographs for pre-event flow 80
8.3.1 Substitution approach 80
8.3.2 Terminology: baseflow or pre-event flow? 81
8.3.3 Choosing the pre-event flow 81
8.4 Estimation of volume of flow 82
8.4.1 Basic method 82
8.4.2 Non-parametric case 82
8.4.3 Parametric case 82
8.5 Deriving the characteristic hydrograph at a gauged site 83
8.6 Estimating the characteristic hydrograph at an ungauged site 84
8.6.1 Using the UPO-ERR-Gamma model 84
8.6.2 Using the parabolic curves method 84
8.6.3 Using IBIDEM 84
8.7 Parabolic curves method 84
8.7.1 Overview 84
8.7.2 Details of method 85
8.7.3 Examples 86
8.7.4 Application at an ungauged site 86
8.8 Constructing the design flood hydrograph 87
8.9 Software 87
8.10 Selection and use of the pivotal catchment 87
8.10.1 Overview 87
8.10.2 Selection of the pivotal catchment 88
8.10.3 Recommended procedure for data transfer 89
8.10.4 Example 89
8.10.5 Further discussion of choice of method and of pivotal catchment 92
8.10.6 Urbanised catchments 93
9 IBIDEM 94
9.1 The idea of IBIDEM 94
9.1.1 Reminder of hydrograph estimation by FSU methods 94
9.1.2 Hydrograph estimation by the FSR design event method 94

vi
Volume III Hydrograph Analysis

9.1.3 Basic idea of bridge between the FSR and FSU methods 95
9.2 How IBIDEM fits hydrographs 96
9.3 General approach to the optimisation 97
9.3.1 First Tp and then SPR 97
9.3.2 Use of horizontal fitting 98
9.3.3 Deriving Tp by optimising the fit to the FSU flood hydrograph 98
9.3.4 Deriving SPR by matching the required peak flow 100
9.4 Additional IBIDEM options 100
9.4.1 Flood frequency 100
9.4.2 Sensitivity to storm duration 100
9.4.3 Sensitivity to model parameters 100
9.4.4 Sensitivity to changes in urbanisation 101
9.5 Further details of the software 102
9.5.1 Inputs 102
9.5.2 Graphical displays 102
9.5.3 Display options 105
9.5.4 Goodness-of-fit measures 105
9.5.5 Tabular display 107
9.5.6 Export of results 108
9.6 Testing 108
9.6.1 Choice of test sites 108
9.6.2 Estimation of FSU hydrograph shapes 108
9.6.3 Estimation of peak flows 110
9.6.4 Rainfall depth-duration frequency tables 110
9.7 Results 111
9.7.1 Suir at Caher Park 111
9.7.2 Owenboy at Ballea 112
9.7.3 Lagan-Glyde at Aclint 112
9.7.4 Anner at Clonmel 114
9.7.5 Tributary to Tolka at Finglas 115
9.7.6 Illustration of effect of fitting threshold 116
9.7.7 Summary 117
9.8 Additional opportunities provided by IBIDEM 117
9.8.1 Strengths and limitations 117
9.8.2 Urban adjustment to design hydrographs 119
9.8.3 Supplying input hydrographs to river models 119
9.8.4 Allowances for projected land-use change 120
Acknowledgements 121
References 121
Appendices 123
Appendix A Gauges used in Hydrograph Width Analysis 123
Appendix B Prcis of UCC research on flood event analysis 128
Appendix C Performance of HWA methods on verification events 130
Appendix D HWA results and their estimates from PCDs 167
Appendix E Application of the HWA software 171

vii
Volume III Hydrograph Analysis

E1 Troubleshooting the installation 171


E2 Graphical User Interface (GUI) concepts of HWA software 173
Appendix F Further details of IBIDEM 183
F1 Method of optimising Tp 183
F2 Method of fitting SPR 184
F3 Checks and validation of outputs 185

Maps
Map 1.1: Stations used in hydrograph width analysis 4
Map 5.1: Catchment of Station 25014 Silver at Millbrook 37
Map 9.1: Location of test catchments 109

Figures
Figure 1.1: Screen-shot of start-up window of HWA software 5
Figure 2.1: Hydrographs of flood events displayed within a common window of 275 hours 8
Figure 2.2: Time series of flood events at Station 07009 Boyne at Navan Weir 8
Figure 2.3: Seasonal distribution of flood events at Station 07009 Boyne at Navan Weir 9
Figure 2.4: Decoupling the main components of Events 1, 7 and 24 at Station 07009 11
Figure 3.1: Standardised flood hydrographs for Events 1, 5, 7 and 24 at Station 07009 13
Figure 3.2: Methods of constructing a characteristic hydrograph 15
Figure 3.3: Median hydrograph for Station 07009 Boyne at Navan Weir 16
Figure 3.4: Median hydrographs with irregularities 17
Figure 3.5: (a) Smoothed median hydrograph; (b) Truncated median hydrograph 18
Figure 4.1: UPO-Gamma hydrograph for Tr=50 and different values of shape parameter n 22
Figure 4.2: UPO-Gamma hydrograph for n=3 and different values of scale parameter Tr 22
Figure 4.3: UPO-Gamma curve fitted to median hydrograph at Station 07009 23
Figure 4.4: Exponential replacement recession (ERR) for different values of parameter C 24
Figure 4.5: UPO-ERR-Gamma curve fitted to median hydrograph at Station 07009 25
Figure 4.6: Performance of UPO-ERR-Gamma model on verification events, Station 07009 29
Figure 4.7: Verification performance of UPO-ERR-Gamma calibrated in five versions 30
Figure 5.1: Performance of median hydrograph method across 37 Grade A1 stations 31
Figure 5.2: Performance of UPO-ERR-Gamma method across 37 Grade A1 stations 31
Figure 5.3: Comparison of model performance across 37 Grade A1 stations 32
Figure 5.4: Performance in verification compared to that in calibration 33
Figure 5.5: Varied hydrograph shapes at Station 34018 Turlough at Castlebar 34
Figure 5.6: Wide and narrow-peaked hydrographs at Station 24013 Deel at Rathkeale 35
Figure 5.7: Wide and narrow-peaked hydrographs at Station 25006 Brosna at Ferbane 35
Figure 5.8: Wide and narrow-peaked hydrographs at Station 25014 Silver at Millbrook 36
Figure 5.9: Wide and slanted hydrographs at Station 25025 Ballyfinboy at Ballyhooney 36
Figure 5.10: Rainfall-runoff behaviour in 30 June 1986 flood at Station 25014 37
Figure 5.11: Attenuated hydrographs at Station 25017 Shannon at Banagher 38
Figure 5.12: Median hydrograph for Station 07009 Boyne at Navan Weir (whole sample) 39
Figure 5.13: Summary index, s, of hydrograph skewness (89 Grade A1 + A2 stations) 40
Figure 5.14: Characteristic hydrograph for Station 35002 Owenbeg at Billa Bridge 43
Figure 5.15: Characteristic hydrograph for Station 30005 Robe at Foxhill 44
Figure 5.16: Characteristic hydrographs for four stations on the River Suir 44
Figure 5.17: UPO-ERR-Gamma characteristic hydrographs for four stations on the Suir 45
Figure 5.18: Hydrographs and rescaled characteristic hydrograph for Deel at Rathkeale 47
Figure 6.1: Matrix plot of PCDs that in part represent catchment size (89 stations) 55
Figure 6.2: Normality plots of log-transformed width descriptors and model parameters 57
Figure 6.3: Normality plot of standardised residuals for 5-variable model for nW75 62
Figure 6.4: Plot of standardised residuals for 5-variable model for nW75 63
Figure 6.5: Median hydrographs at: (a) Station 15002 and (b) Station 35071 68

viii
Volume III Hydrograph Analysis

Figure 6.6: Derived and modelled values of W75, W50, n, Tr and C (BFI unavailable case) 69
Figure 7.1: Variation of hydrograph width with peak flow at Station 07009 71
Figure 7.2: Variation of hydrograph width with peak flow at Station 07010 72
Figure 7.3: Patterns of variation of hydrograph width with peak flow (at six stations) 72
Figure 7.4: Slope of W75 trend with peak flow magnitude (Grade A1 stations) 73
Figure 7.5: Variation of hydrograph width with pre-event minimum flow (Station 07009) 74
Figure 7.6: Slope of W75 trend with pre-event minimum flow (Grade A1+A2 stations) 74
Figure 7.7: Plot of flood peak against time of year (Station 07009 Boyne at Navan Weir) 75
Figure 7.8: Circular plot of W50, W75 and W90 against time of year (floods at Station 07009) 75
Figure 7.9: Characteristic hydrographs for four sites affected by arterial drainage 78
Figure 8.1: UPO-ERR-Gamma characteristic hydrograph at Stn 07009 by Table 6.7 models 80
Figure 8.2: As Figure 8.1 but with pre-event flow substituting for first part of hydrograph 81
Figure 8.3: Example of parabolic curves method (Station 07009 treated as ungauged) 85
Figure 8.4: Parabolic hydrographs for four stations on the Suir 86
Figure 8.5: Upper hydrographs transferred from Derrycahill to Rookwood 91
Figure 8.6: Derived median and UPO-ERR-Gamma hydrographs for Stns 26002 and 26005 92
Figure 9.1: Design inputs to FSR rainfall-runoff method of flood frequency estimation 95
Figure 9.2: Illustration that FSR T-year peak flow varies with Tp as well as with SPR 97
Figure 9.3: Horizontal fitting by comparing hydrograph widths 98
Figure 9.4: Relationship between peak flow Qpeak and peak rapid response qpeak 98
Figure 9.5: Double-peaked hydrograph 99
Figure 9.6: Display of fitted and imported hydrographs 103
Figure 9.7: Display of how a variable changes with return period 103
Figure 9.8: Display of hydrographs for multiple storm durations 104
Figure 9.9: Display of how a variable changes with storm duration 104
Figure 9.10: Display of sensitivity to an increase in URBEXT 105
Figure 9.11: FSR hydrograph fitted to UPO-ERR-Gamma hydrograph 106
Figure 9.12: As Figure 9.11 but with fitting threshold raised to 60% of peak flow 107
Figure 9.13: Example of IBIDEM tabular display 107
Figure 9.14: Suir at Caher Park 100-year hydrograph fit 111
Figure 9.15: Owenboy at Ballea 100-year hydrograph fit 112
Figure 9.16: Lagan-Glyde at Aclint derived median hydrograph from HWA software 113
Figure 9.17: Lagan-Glyde at Aclint 100-year hydrograph fits 113
Figure 9.18: Anner at Clonmel 100-year hydrograph fit 114
Figure 9.19: Tributary of Tolka at Finglas 100-year hydrograph fit 115
Figure 9.20: Urban adjustment to hydrograph for Tolka tributary at Finglas test site 119
Figure B.1: Catchment-average unit hydrographs standardised by area 129
Figure C.1: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 06011 130
Figure C.2: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 06012 131
Figure C.3: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 06013 132
Figure C.4: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 06014 133
Figure C.5: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 06026 134
Figure C.6: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 07007 135
Figure C.7: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 07009 136
Figure C.8: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 07010 137
Figure C.9: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 07012 138
Figure C.10: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 09001 139
Figure C.11: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 14004 140
Figure C.12: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 14006 141
Figure C.13: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 14007 142
Figure C.14: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 14011 143
Figure C.15: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 14018 144
Figure C.16: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 15005 145
Figure C.17: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 23002 146
Figure C.18: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 24013 147

ix
Volume III Hydrograph Analysis

Figure C.19: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 25003 148
Figure C.20: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 25006 149
Figure C.21: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 25014 150
Figure C.22: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 25017 151
Figure C.23: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 25025 152
Figure C.24: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 25027 153
Figure C.25: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 25030 154
Figure C.26: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 26007 155
Figure C.27: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 26008 156
Figure C.28: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 26012 157
Figure C.29: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 26019 158
Figure C.30: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 27002 159
Figure C.31: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 29001 160
Figure C.32: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 29011 161
Figure C.33: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 30004 162
Figure C.34: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 30005 163
Figure C.35: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 34018 164
Figure C.36: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 36010 165
Figure C.37: Verification of median hydrograph + UPO-ERR-Gamma methods at Station 36015 166
Figure E.1: Main application window of the HWA program showing a message-box 174
Figure E.2: Data window for entering data 175
Figure E.3: Data windows for displaying graphical and tabular outputs 176
Figure E.4: A dialog window showing a standard windows file-opening dialog-box 176

Boxes
Box 4.1: The choice between horizontal and vertical fitting 26
Box 4.2: Editorial note on performance measures 28
Box 6.1: Collinearity 49
Box 6.2: The role of ALLUV in the hydrograph-width models 61

Tables
Table 3.1: Widths of exceedance for four floods at Station 07009 Boyne at Navan Weir 14
Table 5.1: Outcome of whole-sample calibration at 89 Grade A1 + A2 stations 40
Table 6.1: Summary statistics of dependent variables (DVs) selected for regression analysis 50
Table 6.2: Some summary statistics of the IVs initially selected 51
Table 6.3: Correlation matrix of selected IVs and DVs at (up to) 89 stations 53
Table 6.4: Stepwise regression results for modelling hydrograph width descriptor nW75 59
Table 6.5: Coefficient and collinearity statistics for selected model for nW75 60
Table 6.6: Recommended models when BFI available 66
Table 6.7: Recommended models when BFI unavailable 66
Table 7.1: Some leading PCDs of stations on the River Fane 73
Table 7.2: Stations studied for the effect of arterial drainage on hydrograph widths 76
Table 7.3: Pre- and post-drainage values of hydrograph width descriptors/parameters 77
Table 8.1: Selected PCDs for Suck at Rookwood 89
Table 8.2: Data transfers to Suck at Rookwood using parabolic curves method 91
Table 9.1: Some details of the applications to two ungauged test catchments 109
Table 9.2: Design flows (m3s-1) for the five test catchments 110
Table 9.3: IBIDEM input variables for the test catchments 110
Table 9.4: Summary of IBIDEM results for five test catchments (100-year flood case) 116
Table 9.5: Sensitivity to fitting threshold (ungauged site on River Anner) 118
Table A.1: Stations used in Hydrograph Width Analysis 123
Table A.2: Details of the flow data used (see also Table 7.2) 125
Table B.1: Stations subjected to rainfall-runoff analysis 128

x
Volume III Hydrograph Analysis

Table D.1: Hydrograph width analysis results for all 89 stations 167
Table E.1: Details of toolbar buttons 182
Table F.1: Example of iteration to find best-fitting value of Tp 184
Table F.2: Checks and outputs 186

Notation
Symbols
Approximately
AF Adjustment factor converting (e.g.) quantile estimates from fixed to sliding duration
ANSF Average non-separated flow (m3s-1 per km2)
ARF Areal reduction factor (in estimating design depth of catchment rainfall)
BF Baseflow in FSR rainfall-runoff method (m3s-1); equivalent to pre-event flow Q0
C Recession parameter of UPO-ERR-Gamma model (hours)
CWI Catchment wetness index (in FSR design event method)
D Duration (in hours) of design storm in FSR design event method
f(x) Probability density function
F(x), F Cumulative distribution function
g Ordinate of Gamma distribution
I(n,x) Incomplete Gamma function
K Scale parameter of Gamma distribution (hours)
n Natural logarithm
m Number of (% of peak flow) levels at which hydrograph width evaluated when fitting
n Shape parameter of Gamma distribution (and of UPO-ERR-Gamma model)
N Number of years of record, sample size
P Precipitation depth (mm)
PR Percentage runoff (in FSR design event method)
Q Flow (m3s-1)
Q0 Pre-event flow (m3s-1)
QT T-year peak flow (m3s-1)
QMED Median annual flood (m3s-1)
r2 Coefficient of determination
s Hydrograph skewness or eccentricity parameter, e.g. in parabolic curves method
s(t) S-curve (i.e. cumulative response curve)
SPR Standard percentage runoff (in FSR design event method)
T Return period (years)
Tflood Return period of flood (in FSR design event method)
Tr Rise-time (= translation parameter) of UPO-ERR-Gamma model
Train Return period of rainfall (in FSR design event method)
Tp Time-to-peak of unit hydrograph (in FSR design event method)
Tp(0) Time-to-peak of instantaneous unit hydrograph (in FSR design event method)
Vc(p) Semi-dimensionless volume of characteristic hydrograph above p% of peak flow
(hours)
VD(p) Volume of design flood hydrograph above p% of peak flow (m3s-1 hours; m3)
Var Variance
w Weighting function
W Hydrograph width (hours)
W50 Hydrograph width (hours) at 50% of peak flow
W75 Hydrograph width (hours) at 75% of peak flow
y Ordinate of Gamma distribution standardised to have a unit peak
(n) Gamma function

xi
Volume III Hydrograph Analysis

Subscripts
p Value at peak; percentage of peak
I Value at point of inflection

Abbreviations and descriptor names


AEP Annual exceedance probability
ALLUV Proportion of extent of floodplain alluvial deposit
AM Annual maximum
AMRE Average value of mean relative error
AREA Catchment area (km2)
BFI Baseflow index (see Volume IV)
CEH Centre for Ecology and Hydrology
CFRAM Catchment Flood Risk Assessment and Management
CSV Comma-separated values (file format)
DDF Depth-duration-frequency
DMH Derived median hydrograph
DV Dependent variable
ERR Exponential replacement recession
FEH Flood Estimation Handbook
FSE Factorial standard error
FSR Flood Studies Report
FSSR Flood Studies Supplementary Report
FSU Flood Studies Update
GIS Geographic information system
GUI Graphical user interface
HWA Hydrograph Width Analysis; also name of standalone software package for HWA
IBIDEM Interactive Bridge Invoking the Design Event Method
IH Institute of Hydrology, former name of CEH Wallingford
IV Independent variable
LH Left hand
MA Moving average
MRE Mean relative error
NERC (UK) Natural Environment Research Council
NSE Nash-Sutcliffe efficiency
OPW Office of Public Works
PCD Physical catchment descriptor (see Volume IV)
POT Peaks-over-threshold
PR Percentage runoff
RH Right hand
RMSE Root mean square error
SAAR Standard average annual rainfall (mm) the FSU uses 1961-90 as the standard period
SE, se Standard error
tsf Tab-separated format
UK United Kingdom
UPO Unit peak at origin
URBEXT Urban extent: fraction of catchment classified as urban
WP Work Package (within the FSU research programme)

xii
Volume III Hydrograph Analysis

Glossary of terms
Term Meaning
Annual
Probability of one or more exceedances in a year of a preset rainfall depth (in a
exceedance
given duration)
probability AEP
Annual Peak-over-threshold (POT) series comprising the largest N events in N years of
exceedance series record
Annual maximum Time series containing the largest value in each year (12-month period) of record
series for a particular duration
Comparison of a models predictions with actual data, and adjustment of its
Calibration
parameters to achieve a better fit with reality
Semi-dimensionless hydrograph defined to represent the characteristic shape of
Characteristic
flood hydrographs. Ordinates of the characteristic hydrograph are standardised
hydrograph
so that the peak value is 1.0. Abscissae indicate the time (in hours)
Coefficient of
Proportion of variation accounted for by (e.g.) a regression model
determination r2
Confidence Bounds within which a population parameter is estimated to lie with a stated
interval (usually %) confidence; used to indicate the reliability of an estimate
Easting and Coordinates of a location expressed as distance eastwards and distance
Northing northwards from a fixed datum (i.e. reference point)
Eccentricity Parameter summarising the skewness of the upper hydrograph
Genetic algorithm An optimisation (or calibration) method based on global or heuristic searching
nth root of the product of a sample of n values of a positive variable such as
Geometric mean
rainfall depth
Formula specifying the increase of a defined extreme (e.g. peak flow) with return
Growth curve period; provides the factor by which the index flood is multiplied to estimate the
T-year flood
Interpolation Any method of computing new data points from a set of existing data points
Parameter representing value subtracted from or added to a variable x to translate
Location
the graph of its probability distribution along the x-axis. The location of the
parameter
UPO-ERR-Gamma model is determined by the time of the peak flow.
For a given duration (e.g. 24 hours), a time series of independent events
Peak-over-
(abstracted from the period of record) that exceed a preset threshold; the series
threshold (POT)
retains the magnitudes (in mm) and dates of the peak exceedances, together with
series
their times of occurrence; successive POT rainfall events must not overlap
Residual Observed value minus the value estimated by a model
Average number of years between years with rainfalls exceeding a certain value.
T is the inverse of the annual exceedance probability (AEP). Thus, a 50-year
Return period T return period corresponds to an AEP of 0.02. The return period is a basic
component of the depth-duration-frequency model used to calculate a rainfall
depth of the desired frequency.
Parameter controlling the spread of a distribution; e.g. scale parameter Tr
Scale parameter controls the width of the characteristic hydrograph in the UPO-ERR-Gamma
model
Parameter controlling the shape of a distribution; e.g. shape parameter n controls
Shape parameter
the shape of the hydrograph in the Gamma part of the UPO-ERR-Gamma model

xiii
Volume III Hydrograph Analysis

Term Meaning
A measure of the departure from symmetry of a distribution; the hydrograph
Skewness skewness descriptor s is the mean ratio of the width under the rising limb of the
hydrograph to the total hydrograph width at that level.
Standard
Measure of dispersion (i.e. variation) of values about their mean
deviation
Estimated standard deviation of a sample statistic such as the mean, i.e. the
Standard error
standard deviation of the sampling distribution of the statistic
Unimodal Having one maximum e.g. on its probability density function or hydrograph
Verification (or Assessment or confirmation of a derived models performance by reference to
validation) additional data (i.e. data not used in calibration of the model)

xiv
Volume III Hydrograph Analysis

1 Introduction

1.1 Overview
Need for hydrograph information

Until recent years, practitioners in Ireland have typically used methods of flood frequency
estimation based on the Flood Studies Report (FSR). The Flood Studies Update (FSU)
builds on the methodologies of the FSR (NERC, 1975) by using updated databases of Irish
hydrometric data and by applying GIS tools for the computation of Physical Catchment
Descriptors (PCDs). While the estimation of peak flows (see Volume II) is of general
importance, in a proportion of cases it is also necessary to construct the flood hydrograph
associated with the T-year peak flow. The requirement is clearest where a flooding problem
or a flood alleviation scheme is sensitive to prolonged high flows.

Because of the relative richness of hydrograph data in Ireland and the relative paucity of
rainfall data (Reed, pers. comm., 2006), it was not envisaged that rainfall records would be
used in the FSU for flood hydrograph estimation. The hydrological analyses reported here
therefore have the objective of establishing methodologies for estimating design flood
hydrographs from recorded flow data only. PCDs are used to develop regression-based
estimates of such hydrographs so that they can be constructed at ungauged sites.

In many applications, the design flood hydrograph corresponding to a specified return period
is required at a particular site. The site of interest referred to as the subject site may be
gauged or ungauged. Whereas the design peak flow of a given return period is obtained in
the case of a gauged site by statistical frequency analysis of flood peak data, complementary
methods are required to produce the characteristic hydrograph to be associated with that peak
flow. In the FSU, the requirement is met by Hydrograph Width Analysis.

Some earlier methods

Reed and Marshall (1999) list three approaches to defining the design hydrograph: adjusting
the FSR rainfall-runoff model parameters, borrowing a standard hydrograph shape from the
FSR rainfall-runoff method, and applying a simplified model of hydrograph shape. The first
two approaches are taken forward in the FSU by development of IBIDEM (see Chapter 9). In
the third approach, the upper part of the hydrograph beneath a flood peak of the required
return period is synthesised by a quadratic function of W50 defined as the width of the
hydrograph (measured in hours) at 50% of the peak flow. If required, the lower part of the
hydrograph is sketched subjectively. At gauged sites, W50 is estimated from the analysis of
observed flood hydrographs. At ungauged sites, a regression-based estimate relating W50 to
the unit hydrograph time-to-peak is used. In the FEH method (Reed and Marshall, 1999), the
upper part of the flood hydrograph is constrained to be symmetric about its peak. This
approach is taken forward in the parabolic curves method (see Section 8.7). This exploits
additional hydrograph width information and permits the upper hydrograph to be asymmetric.

Archer et al. (2000) develop a non-parametric method for the synthesis of design flood
hydrographs. This is based on direct analysis of the shape of flood hydrographs observed at a
site. Archer et al. analyse hydrographs for flood events drawn from the annual maximum
series. They note the durations of exceedance of selected percentages of the peak flow,

1
Volume III Hydrograph Analysis

distinguishing the elements before and after the peak flow. For each exceedance percentage
point in turn, the median duration is noted across the N annual maximum events. A
hydrograph is thereby derived which is non-dimensional with respect to discharge. The
resulting characteristic hydrograph is then applied to the peak flow of given return period to
synthesise the required design flood hydrograph. By distinguishing periods before and after
the peak flow, the characteristic hydrograph is not constrained to be symmetric.

Archer et al. suggest that the characteristic hydrograph shape thus derived provides a more
realistic basis for generating a design flood hydrograph. The method is claimed to be simpler
and quicker, and not to require the separate assessment of baseflow and storm runoff (Archer
et al., 2000). However, their method is applicable only at gauged sites.

Approach adopted

The Hydrograph Width Analysis (HWA) reported below takes the Archer et al. method as its
starting point. The method presented in Chapters 2 and 3 is applied to flood hydrographs
from 89 gauging stations. Using physical catchment descriptors (PCDs) developed in
Volume IV, the method is generalised in Chapter 6 to allow synthesis of a characteristic
hydrograph at ungauged sites.

Some notes on the structure of volume

Volume III is largely based on HWA research undertaken at NUI Galway as Work Package
3.1 of the Flood Studies Update. Later chapters discuss the application of methods to design
flood hydrograph construction at gauged and ungauged sites. The IBIDEM software package
(see Chapter 9) developed by JBA Consulting extends both the applicability of the HWA
methods and the case-by-case interpretation of T-year flood estimates developed using
Volume II methods.

Station 07009 Boyne at Navan Weir is used as the primary example in illustrating the HWA
procedures. There was no special reason for choosing this station for demonstration
purposes. Other gauged catchments are used where appropriate to illustrate particular
features of HWA. In addition, testing of IBIDEM considers three gauged and two ungauged
sites.

An overarching requirement was that the HWA methods developed needed to be simple
enough to give scope to generalise their use at ungauged (as well as gauged) sites.

1.2 The goal and premise of hydrograph width analysis


The primary goal of the hydrograph width analysis (HWA) research was to devise a
methodology to flesh out the hydrograph beneath a given peak value. The aim was to
define the shape of the design flood hydrograph based on typical flood hydrographs which
have occurred. A subsidiary objective was to investigate the extent to which specific factors
influence the typical shape of the flood hydrograph. Ancillary studies reported in Chapter 7
examine the influence on flood hydrograph shape of the flood magnitude, its season of
occurrence, the pre-event flow and effects arising from arterial drainage works.

The premise of HWA is that hydrographs of floods occurring at a particular station are
broadly similar in character and that the typical shape of the hydrograph of a future flood

2
Volume III Hydrograph Analysis

embodied in the design flood hydrograph can be expected to reflect the general features of
those which already occurred. Assessment of similarity is subjective, being largely based on
visual judgement.

It is sometimes found that, because of prolonged rainfall at the peak-inducing intensity, a few
flood hydrographs at a station can exhibit a highly flattened and prolonged peak segment.
Use of the median hydrograph width as in Archer et al. (2000) ensures that such atypical
flood hydrographs do not receive undue weight in the analysis. Nevertheless, a statistical
measure is required to summarise the degree of similarity between synthesised and observed
flood hydrographs.

Whereas the shapes of the upper parts of observed flood hydrographs (e.g. the parts with
flows above 50% of the respective peak flows) are often found to be generally similar, those
of the lower parts tend to vary widely. The variation reflects a number of factors including
the occurrence of preceding and/or following floods subsidiary to the main event.

If only the upper part of the hydrograph of the design flood is deemed important, a procedure
to model the lower part is not required. However, in a proportion of applications, the
complete hydrograph is required.

1.3 Catchment selection


Detailed flow data are required for hydrograph width analysis, and flow data at 15-minute
interval were obtained for 90 gauging stations operated by the OPW. Physical catchment
descriptors (PCDs) were not initially available for Station 39008. The study therefore
considered a network of 89 stations.

The stations chosen were selected with regard to the quality of flow data expected. Grade A1
and Grade A2 are the highest categories of rating curve and water level measurement
reliability (see Appendix A1.1 of Volume II). The selected stations comprise 37 stations
graded A1 and 52 graded A2. Their catchment areas range in size from 23 to 7980 km2, with
a median value of 285 km2. The 89 catchments are identified in Map 1.1 and in Table A.1 of
Appendix A.

Hydrograph data were supplied as 15-minute data in a time-series format. Each *.tsf datafile
held the date and time of occurrence, the flow (in m3s-1) and a quality code against the
measurement. Table A.2 indicates the period of flow data abstracted, the completeness of
15-minute data across that period, the number of annual maximum (AM) values represented,
and the median of the annual maxima (i.e. QMED) across that period. Also shown is the
period during which any arterial drainage works were carried out on the catchment. For
stations affected by drainage works, the main HWA used the post-drainage record only.

3
Legend
! Stations considered in the "Hydrograph Width Analysis"

Volume III
-
Hydrograph Analysis

! Station location
!
39009
(The boundaries shown
mark the Hydrometric
Areas of Ireland)

!
35071

35005
35002! 36015
! !
34001 ! 36027 6012
! 35001 36021 ! ! 36019 !
34009 ! ! 36010 6011
! 36011 !
! 26012 ! 6014
34018 ! !
! 26009 7033 6026 ! 6013
26008! ! ! 7004
! 7006 7012
26022! 26019 !
30005 ! 7011! !
! ! ! 7009
26002 26021 7010
30007
!
! !
7002 !
26005 ! 7007
! !
!
26007 9001
30004 ! !
30061 29001 2500625016 14004
! ! ! ! ! 14011
!! ! !
29004 29011 ! !
25017 14009! 14006
25025 14007
!
!
25030 25029
27002 ! ! 15005
27001 ! ! !
! 25027
15003
25001 16004 16001 ! ! 11001
! ! !
! !
25005 ! 16003 ! 14018 !
2400825003! 15002
! !
23001 ! !! 24082 ! 16008 15006!
! 23002 24013 16005!
!
16009
!
23012
! 18004
! 18005
!
22071
!

19001
!

0 20 40 80 Kilometers

Map 1.1: Stations used in hydrograph width analysis

1.4 Physical catchment descriptors (PCDs)


Physical catchment descriptors (PCDs) at 216 gauged locations on rivers and lakes in Ireland
were supplied by the OPW. Details are given in Volume IV. That volume includes three
special PCDs developed in the FSU research: FAI, BFIsoil and FLATWET.

The special descriptor FLATWET was available and, in certain circumstances, plays a role in
constructing the characteristic hydrograph at an ungauged site. However, the FAI and BFIsoil
descriptors had not been developed at the time of the hydrograph width research. To allow

4
Volume III Hydrograph Analysis

consideration of soil permeability and other storage effects on hydrograph widths, median
values of the baseflow index (BFI) were supplied for 198 stations. BFI values were
unavailable for ten of the 89 stations selected for HWA. Consequently, Stations 07007,
07011, 14004, 15002, 25017, 26009, 29001, 36010, 36027 and 39009 had to be omitted from
those analyses requiring BFI.

1.5 The characteristic hydrograph


The aim of the research is to allow the user to construct the design hydrograph of a given
return period. The hydrograph represents flows in m3s-1. Following Archer et al. (2000), the
need is met by defining a semi-dimensionless flood hydrograph. This has time coordinates in
hours but the peak flow is standardised to be 1.0. The terminology adopted in the FSU is to
call this the characteristic hydrograph.

At gauged sites, the characteristic hydrograph is constructed by direct analysis of hydrograph


data observed in large floods, i.e. by hydrograph width analysis. At ungauged sites, the
characteristic hydrograph is synthesised from PCDs by methods presented in Chapter 6.
Typically, the required design hydrograph in m3s-1 is obtained by scaling up the characteristic
hydrograph by the T-year flood peak estimated using Volume II methods.

1.6 HWA software


Written in Visual Basic and Fortran, the standalone software package HWA provides an
interactive tool for Hydrograph Width Analysis, displaying data and results in tabular and
graphical forms. Outputs can be saved under user-defined filenames for future access and
use. A screen-shot of the HWA start-up window is shown in Figure 1.1. Technical details
appear later in Appendix E. The software is available to practitioners through the data and
software download module of the FSU Web Portal.

Figure 1.1: Screen-shot of start-up window of HWA software

5
Volume III Hydrograph Analysis

2 Processing the flow data for HWA


This chapter describes the processing procedures by which flow data were screened and flood
events selected and filtered for hydrograph width analysis.

2.1 Data screening and checking

2.1.1 Data handling


For gauging stations in the OPW network, river-flow data at 15-minute interval are held in
datafiles having the .tsf filename extension. For many stations, these files contain more than
a million items of data.

Scanning such long data series for the purpose of identifying and selecting flood events is
both tedious and restrictive within commonly used software applications. Prior to
introduction of the 2007 version, Microsoft Excel had a limit of 65,536 rows. Even in the
2007 version, the row limit is too small to accommodate 15-minute data series of 30 years or
longer.

Application-specific programs were therefore written in Fortran to accept data in the .tsf
format and to search the data series for flood events. Visual Basic was used to create a
standalone user-interface for displaying flood hydrographs with interactive graphics.
Additional programs were written to analyse the identified flood hydrographs and to select
and implement generally suitable methods for final adoption. The development resulted in
the HWA software package (see Appendix E).

2.1.2 Missing flow data


The HWA research considered flow data from 89 Grade A1 or A2 stations. There are
occasional data gaps in the 15-minute flow records. A typical reason given is that the chart
was missing. Percentages of missing data are listed in Table A.2 for each station. Eight
stations had more than 10% of data missing but only Stations 25005 and 26021 had more than
20% of data missing. The median proportion of missing data was 2.9%.

Complete hydrograph data were therefore not always available for all flood events. In a few
cases it proved possible to analyse the rising or receding limb of the flood event where the
gap in flow data affected only the opposite limb of the hydrograph.

2.1.3 Scrutiny of annual maximum flood peaks


Annual Maximum (AM) flood peaks were extracted and scrutinised for anomalies. In some
cases, the existence of a very large value in the AM series called for special investigations.
This occasionally led to the discard of doubtful data.

The index flood adopted in the FSU is QMED, the median of the AM floods. The QMED
value corresponding to the period of record used in HWA is given in the penultimate column
of Table A.2 (see Appendix A).

6
Volume III Hydrograph Analysis

2.1.4 Stations affected by arterial drainage


For information, the final column of Table A.2 identifies those of the 89 stations which have
experienced arterial drainage within or prior to their overall period of record. The
hydrological and spatial descriptors of the FSU catchments (see Volume IV) were assembled
relatively recently, and generally reflect the post-drainage characteristics of such catchments.
Consequently, the main HWA for these catchments was carried out using hydrographs from
the post-drainage period only.

Stations 07007, 07010, 23002, 26012 and 30004 had 15-minute flow data available for both
pre and post-drainage periods. These five catchments are used specifically in Section 7.4 to
examine the effect of arterial drainage on the typical shape of flood hydrographs.

2.2 Defining the time-window of the flood hydrograph


The convention adopted was to position the time origin at the peak of the hydrograph. Thus,
each flood hydrograph is presented with its peak occurring at t = 0, with its rising and
receding limbs shown on the LH and RH sides of the peak respectively.

In some instances, due to prolonged rainfall of peak-inducing intensity, an observed flood


hydrograph exhibits a flattened (i.e. sustained or persistent) peak. In such cases, the first
occurrence of the peak flow is considered as the origin of the time axis for the purpose of
hydrograph width analysis. Thus, in the case of sustained peak flow, the receding limb of the
flood hydrograph includes the peak flow over the period during which it persists after its first
occurrence.

The user of the HWA software specifies the start and end of the flood hydrograph by
indicating the time intervals from the peak back to the start of the hydrograph and from the
peak forward to the end of the hydrograph. The sum of these two time intervals defines the
time-base (or window) that embraces the flood peak and the flood hydrograph about it.

It is convenient to adopt a common time-base when displaying a number of flood


hydrographs for a particular station. The user of the HWA software guesses values for the
time intervals (before and after the flood peak). Each of the observed flood hydrographs is
then displayed within a window of common size. Finalising values of the time intervals
requires a number of trials, the objective being to contain all the selected flood hydrographs
within the final common window.

Figure 2.1 provides an example of flood hydrographs for three events displayed within a
common window. For this station, the 275-hour window comprises 50 hourly time-steps on
the rising limb of the hydrograph and 225 hourly time-steps on the receding side. This
reflects that flood hydrographs for Station 07009 Boyne at Navan Weir are generally
observed to rise (from the level of the pre-event flow to the peak) over about two days and to
recede (from the peak to the level of the baseflow) over about nine days.

7
Volume III Hydrograph Analysis

Figure 2.1: Hydrographs of flood events displayed within a common window of 275 hours

2.3 Selection of flood hydrographs


The final selection of flood hydrographs was made by applying a peaks-over-threshold (POT)
criterion. This ensures that the hydrograph width analysis gives due weight to the largest
floods recorded. Where the annual maximum series comprises N floods (i.e. representing N
water-years of record), the criterion applied was to select the N largest flood events from this
period. This special case of a POT series is known as the annual exceedance series.

Figure 2.2 illustrates the annual exceedance series for Station 07009 and shows how it relates
to the annual maximum (AM) series. The AM events are marked by small green circles, with
the broken green line drawn to indicate the QMED of 134.8 m3s-1. Events in the annual
exceedance series are marked by blue circles.

There are 29 annual maxima at this station. Some 19 of these 29 flood events are in the
annual exceedance series. The annual exceedance series includes ten further floods (from
flood-rich years) in place of the ten smallest annual maxima (in flood-poor years).

The blue line marks the level of the 30th largest flood in the annual exceedance series and
defines the threshold above which the annual exceedance series has been extracted to yield 29
events in 29 years. [Editorial note: For reasons that are not entirely clear, the 30th largest
flood has been included in the annual exceedance series at this station. The dates shown
along the x-axis of Figure 2.2 are correct but are not very helpful. The tick-marks are spaced
at an interval of 50675 15-minute periods i.e. about every 528 days.]

Figure 2.2: Time series of flood events at Station 07009 Boyne at Navan Weir

8
Volume III Hydrograph Analysis

An exception was made in two cases where the number of flood events would otherwise have
been very small. Stations 14011 and 25003 had hydrograph data for only five and seven
water-years respectively. The number of events selected in these cases was arbitrarily
doubled: i.e. using ten events at Station 14011 and 14 at Station 25003.

2.4 Numbering of flood hydrographs


As part of the selection process, flood events are numbered according to their ranking in the
annual exceedance series. Thus, Event 1 is the largest flood event and Event 10 is the tenth
largest flood event. The event numbering carries through to the flood hydrographs.

[Editorial note: This is thoroughly effective in ensuring that hydrographs of the largest
floods attract particular attention in analysis. It remains to be seen whether the device leads
to confusion when flow records are updated by the extraction of hydrographs for later events.]

2.5 Seasonal distribution of flood events


The circular plots of Figure 2.3 illustrate the seasonal distribution of flood events at
Station 07009. The angular position denotes the calendar date. Again, the green circles mark
the AM events and the blue circles denote the annual exceedance events. The plots in Figure
2.3 are identical except that the radial axis is marked in m3s-1 in the LH diagram and in
multiples of QMED in the RH diagram.

The mean time-of-year of floods is found by plotting the centroid (analogous to the centre of
mass) of the data points. The centroids in Figure 2.3 are marked by filled circles: green for
the AM series and blue for the annual exceedance series. The mean time-of-year of flood
events at Station 07009 is seen to be early January.

Figure 2.3: Seasonal distribution of flood events at Station 07009 Boyne at Navan Weir

[Editorial note: Radial positions in Figure 2.3 indicate the magnitudes of the flood events.
Calculation of the mean time-of-year of flooding has been weighted by flood magnitude.]

9
Volume III Hydrograph Analysis

2.6 Filtering of selected hydrographs

2.6.1 Desire for broadly unimodal hydrographs


Essentially unimodal (i.e. single-peaked) flood hydrographs are to be favoured when deriving
the characteristic hydrograph. [Editorial note: The characteristic hydrograph is to be applied
to design flood peaks derived by Volume II methods to yield the T-year flood hydrographs
required in many flood risk assessments. Such design events are hypothetical floods. It is
therefore reasonable to allow the characteristic hydrograph itself to be stylised. In essence,
the aim is to construct a design flood hydrograph that represents the typical catchment flood
response to one heavy rainfall event rather than to a succession of events.]

It was observed that many high-ranking flood events across the 89 stations had complex (i.e.
multi-peaked) hydrographs. At some stations, very few single-peaked floods could be
identified. This precluded the option of deriving the characteristic hydrograph from only
those flood events with broadly one-peaked hydrographs. Techniques for decoupling
complex floods were devised to overcome this difficulty.

2.6.2 Decoupling the main flood response within a complex flood event
A complex flood event is one having multiple peaks. Typically, it represents the catchment
flood response to more than one period of heavy rainfall. Several approaches were
investigated for decoupling (i.e. isolating) the main flood response within a complex event.

One approach constructed a master recession curve from segments of the receding limbs of
single-peaked hydrographs observed for the station. An analogous master rising curve was
similarly constructed from segments of the rising limbs of single-peaked hydrographs at the
station. Whilst dealing with some individual stations adequately, the approach failed to
generalise for use across all stations. There was a specific concern that the approach led to
bias in the characteristic hydrograph.

The approach ultimately adopted was simply to discard the complex segments of the flood
hydrograph: retaining only the broadly unimodal part.

2.6.3 Discarding the complex segments


Only that part of the observed complex hydrograph embracing the largest peak is considered
relevant for deriving a generalised shape of the design flood hydrograph. Parts of the
hydrograph that can be visually associated with flood responses before/after the largest-
peaked flood response are discarded. The decoupling is done subjectively, using interactive
features of the HWA software.

The decoupling is illustrated for three flood events at Station 07009 Boyne at Navan Weir.
Figure 2.4 shows the hydrographs for the largest, 7th largest and 24th largest floods analysed.
The cyan-coloured sections of the plotted hydrographs indicate the parts discarded. The grey-
coloured section indicates the decoupled component of the main flood response. This is the
hydrograph used in the subsequent Hydrograph Width Analysis.

10
Volume III Hydrograph Analysis

Figure 2.4: Decoupling the main components of Events 1, 7 and 24 at Station 07009

11
Volume III Hydrograph Analysis

3 Deriving the characteristic hydrograph at gauged sites


Methods are described for deriving the characteristic hydrograph at a gauged site. Two
groups of methods are distinguished: parametric and non-parametric. The non-parametric
approach derives the characteristic hydrograph by a statistical averaging of hydrographs. It is
non-parametric because it does not make any particular assumptions about the shape of the
hydrograph.

In contrast, the parametric approach specifies a formulaic shape for the characteristic
hydrograph. Its basic aim is to represent the hydrograph by a smooth curve expressible in
terms of a small number of parameters. Although other strategies are possible, it was found
most effective to fit the parametric form to the characteristic hydrograph first obtained using
the non-parametric method. The parametric approach is therefore taken up in Chapter 4.

3.1 Standardising the flood hydrographs


The flood hydrographs comprise isolated single-peaked floods and decoupled (unimodal)
segments of more complex flood events (see Section 2.6). Collectively, these might be
termed filtered hydrographs. Hereafter they are chiefly just called flood hydrographs.

The flood hydrographs are standardised by dividing the flow ordinates by the magnitude of
the peak flow. Each standardised flood hydrograph thus has a peak value of 1.0. This
corresponds to the 100th percentile of the peak flow, in that 100% of the hydrograph is less
than or equal to this flow. [Editorial note: This terminology is correct when applied at the
peak but not otherwise. The authors refer to percentiles of the peak flow when they ought to
refer to percentages of the peak flow. It was not practical for editing to correct the
mislabelling in every case. Users of the HWA package should therefore interpret percentile
of peak flow as meaning percentage of peak flow.]

The ordinate scale of the standardised hydrograph is percentages of the peak flow. The time
scale of the hydrograph is unaltered, with the abscissa in hours. Thus, the standardised flood
hydrograph is semi-dimensionless. As noted previously, the time origin is taken at the time
of the peak flow.

Flow levels at increments of 5% of the peak flow are identified to the extent that a particular
hydrograph allows. Figure 3.1illustrates this for four hydrographs at Station 07009 Boyne at
Navan Weir. Events 1, 7 and 24 are decoupled from complex flood events; Event 5 is an
isolated unimodal hydrograph.

3.2 Calculation of hydrograph widths at particular exceedance levels


The hydrograph width at a given percentage of the peak flow is defined as the time during
which the flow at a station exceeds the flow corresponding to that percentage of the peak
flow. The total width of exceedance is a time in hours. Following Archer et al. (2000), the
width is divided into two components: one on the rising limb and another on the receding
limb. As seen in Figure 3.1, the width component at a particular percentage of the peak flow
is sometimes available on one limb but not on the other.

12
Volume III Hydrograph Analysis

Figure 3.1: Standardised flood hydrographs for Events 1, 5, 7 and 24 at Station 07009

In characterising the hydrographs, the first step is to choose a series of percentages. If the
objective is to derive only the upper part of the characteristic hydrograph, percentages above
and including 50% of the peak flow are considered sufficient. For other purposes, a fuller
description of the hydrograph is required. The research chose to use percentages at 5%
intervals from 95% down to 5%. In addition, some use was made of the width at 98% of the
peak flow.

The precise value of the flow that corresponds to a selected percentage of the peak will not
generally appear in the 15-minute record of flow data for the event. Linear interpolation is
used to compute the time at which a particular percentage of the peak flow occurs. The
required hydrograph widths can then be found.

For many selected events, hydrograph widths at the lower percentages are generally
unavailable. The widths of exceedance at selected percentages are shown in Table 3.1 for the
sample events of Figure 3.1. For Event 24, the width of exceedance at 65% of the peak flow
is 15.49 hours on the rising side but is undefined on the receding side. [Editorial note: The
terms hydrograph width and width of exceedance are used interchangeably. There is an
unexplained discrepancy between Table 3.1 and Figure 3.1 in the hydrograph widths on the
receding limb of Event 1.]

In certain applications it can be helpful to focus on hydrograph widths at one or two fixed
percentages of the peak flow. There is particular interest here in W 75 and W50, which denote
the hydrograph widths at 75% and 50% of the peak flow.

3.3 Procedures for constructing the characteristic hydrograph


A major element of the research was to develop procedures for constructing the characteristic
hydrograph. The principal methods considered in the research are summarised in Figure 3.2.
The applicability of a particular procedure depends on whether the subject site is gauged or

13
Volume III Hydrograph Analysis

ungauged and whether the complete characteristic flood hydrograph or only its upper half is
of interest. The methods ultimately recommended are highlighted.

Table 3.1: Widths of exceedance for four floods at Station 07009 Boyne at Navan Weir
Width of exceedance (hours)
Percentage

Event 1 Event 5 Event 7 Event 24


Receding

Receding

Receding

Receding
Rising

Rising

Rising

Rising
Total

Total

Total

Total
98 2.84 2.44 5.29 2.81 3.65 6.45 4.24 2.25 6.49 2.96 7.01 9.97
95 5.23 4.25 9.48 4.36 6.54 10.91 7.77 5.32 13.09 5.39 10.22 15.61
90 8.32 6.60 14.92 6.45 9.60 16.05 9.64 8.44 18.07 8.24 14.41 22.65
85 14.11 8.75 22.86 7.84 12.16 20.00 10.78 11.55 22.33 10.56 18.17 28.73
80 16.63 10.34 26.97 9.02 14.49 23.51 10.94 14.23 25.17 12.41 21.98 34.39
75 18.02 10.04 17.18 27.22 11.60 17.34 28.94 13.89 25.83 39.72
70 19.29 10.79 21.06 31.85 12.67 20.68 33.35 15.49
65 20.37 12.20 26.81 39.01 14.27 23.44 37.72 17.41
60 21.07 12.97 33.67 46.64 15.14 27.12 42.26
55 21.79 13.64 40.54 54.17 15.82 30.84 46.66
50 22.54 14.41 47.18 61.59 17.00 35.80 52.79
45 23.51 15.16 53.70 68.86 18.71 43.59 62.31
40 24.42 15.85 61.55 77.39 22.45 56.43 78.88
35 25.33 16.61 73.71 90.32 72.50
30 26.44 17.50 110.40 127.90 97.82
25 28.50 19.11 136.67 155.78 151.27
20 44.84 22.69 200.38 223.07
15
10
5

The present chapter discusses the non-parametric approach in which a semi-dimensionless


flood hydrograph is derived broadly following Archer et al. (2000). Observed (isolated or de-
coupled) flood hydrographs are selected. The characteristic hydrograph is constructed to be
unimodal and to have exceedance widths before and after the peak that take the average of the
corresponding exceedance widths in the sample hydrographs. In the FSU hydrograph width
research, the recommended approach was to use the median value. The characteristic
hydrograph obtained in this way is referred to as the derived median hydrograph or just the
median hydrograph.

14
Volume III Hydrograph Analysis

Subject site

Gauged site Ungauged site

Non-parametric method based Parametric method


on widths of exceedance by curve fitting
If complete hydrograph required For upper hydrograph alone
Apply Gamma curve with Estimate width parameters
Derived Derived exponential replacement W75 and W50 from PCDs*.
median mean recession, estimating parameters Adopt suitable value of
hydrograph hydrograph n, Tr and C from PCDs* eccentricity parameter (s)

Hayashi et Negative Inverse Gamma curve with Gamma curve


Gamma Construct Apply modified
al. (1986) Binomial Gaussian algebraic with exponential
curve parabolic Gamma model
curve curve curve replacement replacement
Sketch curves that with parameters
recession recession
subjectively respect n, Tr and C that
required yield required
values of values of W75,
Fit to Fit to Fit to event Fit to event W75, W50 & s W50 and s
derived derived hydrographs hydrographs Fit to event
median mean individually; individually; hydrographs
collectively
hydrograph hydrograph adopt median adopt mean Alternate formulations are given.
parameter values parameter values One version requires BFI. BFIsoil
might be used but was not available
at the time of the HWA research.

Figure 3.2: Methods of constructing a characteristic hydrograph

In the parametric approach, the characteristic hydrograph is obtained by fitting an algebraic


formula to the observed hydrographs or to the characteristic hydrograph previously obtained
by the non-parametric approach. The coefficients and constants in the model form the
parameters. The choice of model structure is a matter of judgement but is aided by an
objective measure of how well the modelled hydrograph fits the characteristic hydrograph
derived by the non-parametric approach. The recommended parametric model is the Gamma
curve with exponential replacement recession. This UPO-ERR-Gamma model is introduced
in Section 4.4.

3.4 Split-sample and whole-sample calibration


The 37 Grade A1 stations were used for testing and comparison of the developed methods.
At each station, three flood hydrographs were reserved for verification, with the remaining
events used in calibration. This procedure is referred to as split-sample calibration. To avoid
bias, one large, one medium and one small event were enrolled as the verification events.
[Editorial note: Other parts of the Technical Research Report refer to validation rather than
verification. The terms are interchangeable.]

As detailed in Section 2.4, the M hydrographs at a station are numbered so that the one with
the largest flood peak is Event 1 and the one with the smallest flood peak is Event M. To
select the three verification events, the M events were divided into three equal or near-equal
groups characterising the ranges of large, medium and small flows. In the case of Station
07009 Boyne at Navan Weir, there are 30 flood events in the annual exceedance series.
Events 1 to 10 are deemed large flood events, Events 11 to 20 are deemed medium events and
Events 21 to 30 are deemed small events. A middle-ranking flood event in each group was
adopted as a verification event. For Station 07009, Events 5, 15 and 24 were thus selected as
verification events. The remaining 27 events were used in calibration.

15
Volume III Hydrograph Analysis

In the first phase of analysis, the most promising approaches were identified based on how
well the derived characteristic hydrograph fitted the three verification events. The test was
made across all 37 Grade A1 stations.

Having identified the most suitable non-parametric and parametric methods, these were
applied across the whole set of flood events at all 89 stations in a second phase of analysis.
This procedure is referred to as whole-sample calibration. Ultimately, the final estimates of
the characteristic hydrograph were based on all selected flood events at the particular station.
Two best methods were carried forward: one based on the non-parametric approach and the
other based on the parametric approach.

3.5 Deriving the median hydrograph

3.5.1 Basic method


The preferred method in the non-parametric approach is to take median values of the various
component widths of the standardised hydrographs: the averaging being taken across the
group of events being analysed. As noted earlier, widths are defined at various percentages of
the peak flow, and components before and after the time of peak flow are distinguished.
Component hydrograph widths are not available at all percentages. Thus, the median is taken
across those hydrographs that provide a component hydrograph width at the relevant
percentage of the peak flow.

The characteristic hydrograph is constructed so that its component widths at all percentages
correspond to the relevant median value. The resulting hydrograph is referred to as the
derived median hydrograph. The example in Figure 3.1 is for the split-sample calibration at
Station 07009. Thus the hydrograph is the median of 27 standardised hydrographs. Three of
the 30 flood events at this station have been withheld for use in verification. The embedded
table notes the values of the hydrograph widths at 75%, 50% and 25% of the peak.

100 Percentage Hydrograph width (hours)


90 of peak
Percentage of peak flow

flow On rising limb On receding limb Total


80
70 75% 10.49 17.32 27.81
60 50% 15.15 38.75 53.90
50 25% 24.11 135.43 159.54
40
30
20
10
0
-50 -37.5 -25 -12.5 0 56.25 112.5 168.75 225
Time in hours (relative to time of peak f low)
Figure 3.3: Median hydrograph for Station 07009 Boyne at Navan Weir

16
Volume III Hydrograph Analysis

3.5.2 Anomalies in the derived median hydrograph


Unrealistic kinks appear at some lower percentages in the rising/receding limb of the derived
median hydrograph. These appear because of the changing (i.e. reducing) number of events
across which the median is drawn at lower percentages of the peak. In the case of Station
07009, the kinks are barely discernible. But, for many stations, the anomalies are unrealistic
and rather distracting e.g. Station 07007 Boyne at Aqueduct in Figure 3.4a. The kinks reflect
both the non-availability of widths at lower percentages in some events and the wide variation
in those widths that are available at lower percentages.

(a) Station 07007 Boyne at Aqueduct (b) Station 14006 Barrow at Pass Bridge

Figure 3.4: Median hydrographs with irregularities

3.5.3 Improving the derived median hydrograph


Two techniques were applied to moderate irregularities in the derived hydrographs. These
were implemented interactively using options developed within the HWA software.

In some cases, a central 3-term moving-average filter was applied one or more times to
smooth the pattern of variation down the offending limb of the hydrograph. Where
smoothing failed to produce an acceptable hydrograph, the lower parts were simply
discarded. Sometimes the lower parts of the derived median hydrograph were discarded;
sometimes the lower parts were discarded after smoothing.

Three applications of the moving-average filter led to a satisfactory outcome for Station
07007 (see Figure 3.5a). At Station 14006, the best that could be achieved was to truncate the
median hydrograph by discarding parts below 20% of the peak flow (see Figure 3.5b).

There was concern that repeated use of a moving-average filter was unduly arbitrary and
subjective. The primary technique recommended for removing unacceptable features of the
derived median hydrograph is therefore to truncate the hydrograph by simply discarding the
lower parts that exhibit inconsistent widths.

Derived median hydrographs for the 37 Grade A1 stations can be glimpsed in Appendix C.
These are based on the initial split-sample calibration. Final results for the whole-sample
calibration at all 89 Grade A1 + A2 stations are summarised later in Table 5.1.

17
Volume III Hydrograph Analysis

(a) Station 07007 Boyne at Aqueduct (b) Station 14006 Barrow at Pass Bridge

Figure 3.5: (a) Smoothed median hydrograph; (b) Truncated median hydrograph

18
Volume III Hydrograph Analysis

4 The parametric approach


The parametric approach specifies a formulaic shape for the characteristic hydrograph. The
favoured model has three parameters.

4.1 Objectives
Generation of the median hydrograph from observed flood events at a gauged site is relatively
straightforward using the HWA software to implement the non-parametric approach
described in Section 3.5. That the median hydrograph is regularly spaced on the standardised
flow axis rather than on the time axis is a minor inconvenience. This can be overcome by
linear or other interpolation. Where the non-parametric approach is truly limiting is in the
difficulty of generalising the method to allow the characteristic hydrograph to be constructed
at an ungauged site.

In the parametric approach, the characteristic hydrograph is represented by an algebraic form


involving two or three parameters only. This leads to design hydrographs that are smoothly
varying and therefore more likely to be intuitively acceptable. However, the chief prize is
that a parametric form for the characteristic hydrograph gives much greater scope for
generalisation of the model for application at ungauged sites.

The two or three parameters of the algebraic form can be related to physical catchment
descriptors (PCDs) available for all sites: gauged and ungauged. [Editorial note:
Generalisation and automation of procedures can, for example, assist applications in
Catchment Flood Risk Assessment and Management (CFRAM).] A further benefit is that the
functional form makes it possible to define the flood hydrograph in its entirety, including the
low parts at the beginning and end. This aids the calculation of flood volumes and supports
other applications in which the whole hydrograph is required.

4.2 General approach


The general approach taken was to fit parametric curves to the median hydrographs derived in
Section 3.5. The following models were considered in the research:

Cubic polynomials (fitted separately before and after the peak);


The Negative Binomial distribution curve;
The Inverse Gaussian distribution curve;
The Hayashi et al. (1986) curve;
Various formulations and extensions of the Gamma distribution curve.

These models were subject to extensive exploration, although not all methods were applied to
all catchments in the 89-station dataset. In general, the parametric curves were fitted to the
median hydrograph in its native semi-dimensionless form i.e. with flows in percentages of the
peak and times in hours before and after the peak.

In the case of statistical distributions such as the Inverse Gaussian and Gamma, the model is
customarily standardised to have unit volume under the curve rather than a unit peak. It is a
matter of algebraic manipulation to come up with a variant that meets the unit peak criterion.

19
Volume III Hydrograph Analysis

This is now illustrated for the UPO-Gamma curve, which denotes a Gamma curve
reformulated to have a Unit Peak at the Origin (abbreviated to UPO). The parametric model
ultimately recommended is a modification of the UPO-Gamma curve.

4.3 UPO-Gamma model for the characteristic hydrograph

4.3.1 Gamma distribution


The Gamma distribution rises from the origin at x = 0 and encloses a unit volume under the
curve. It has the functional form:

n 1
1 x x
g f(x) exp where x 0 4.1
K n K K

The model has two parameters: the shape parameter n and the scale parameter K. (n) is the
Gamma function, defined by standard formulae or tables.

The Gamma distribution has a long history of application in rainfall-runoff modelling as the
Nash-cascade model for the so-called instantaneous unit hydrograph (Nash, 1957), and in
flood routing as the Kalinin-Milyukov routing method (Kalinin and Milyukov, 1957).
Examples appearing later (see Figure 4.1 and Figure 4.2) confirm the hydrograph-like curves
generated by the distribution.

4.3.2 Peak of Gamma distribution


It can be shown that the peak of the Gamma distribution occurs at:

x p Kn 1 4.2

and, when n > 1, takes the value:


gp
1
n 1n 1 exp n 1 4.3
K n
The Gamma curve is not hydrograph-like if n 1. For example, when n = 1, it degenerates to
an exponential recession. Thus, its application here is limited to the case n > 1.

4.3.3 Gamma model with peak at time zero


For the peak to occur at time zero, the distribution needs to be shifted left by xp units, i.e. by
K(n-1). Thus, the Gamma distribution with peak at time zero is given by:

1 x Kn 1 x Kn 1
n 1

g exp where x + K(n-1) 0 4.4


K n K K

This simplifies to:

20
Volume III Hydrograph Analysis

n 1
1 x x
g n 1 exp n 1 where x + K(n-1) 0 4.5
K n K K

4.3.4 Gamma model with unit peak at time zero


The requirement is for a model that can be fitted to the characteristic hydrograph. Typically,
this will be the median hydrograph derived in Section 3.5. The required model therefore has
a peak of 1.0 at time zero.

Dividing Equation 4.3 by Equation 4.5, we obtain the Gamma distribution with unit peak at
time zero:
n 1
x x
y g g p 1 exp 4.6
Kn 1 K

This provides the UPO-Gamma model for the characteristic hydrograph, where UPO denotes
Unit Peak at Origin. y denotes the flow as a proportion of the peak flow. The hydrograph
rises from y = 0 at time x = -K(n-1) to y = 1 at time x = 0, and recedes thereafter.

4.3.5 Formulation in terms of hydrograph rise time Tr


The formulation preferred here replaces K with Tr/(n-1), where Tr denotes the rise time of the
Gamma distribution (see Equation 4.2). The UPO-Gamma model is then:

x n 1
n 1
x
y 1 exp 4.7
Tr Tr

The model rises from 0 at time x = -Tr to 1 at time x = 0. [Editorial note: The Tr parameter
is sometimes referred to as the hydrograph rise time and sometimes as the translation
parameter. This is because moving the time origin from the beginning of the hydrograph to
the peak of the hydrograph has translated the time axis by Tr units.]

4.3.6 Families of hydrographs constructed using the model


Figure 4.1 shows hydrographs constructed using the UPO-Gamma model that all have a rise
time of Tr = 50 hours. The effect of varying the parameter n is seen and its role as a shape
parameter confirmed.

Figure 4.2 shows hydrographs that all have n = 3. The effect of varying the parameter Tr is
seen and its role as a scale parameter confirmed. The hydrograph rises over time Tr,
measured in hours.

[Editorial note: Some notational changes have been made. HWA uses q to denote the
standardised flow, i.e. the hydrograph with unit peak. This has been changed to y to avoid
clashing with the use of q in IBIDEM to denote the rapid response element of the total
hydrograph in m3s-1.]

21
Volume III Hydrograph Analysis

Percentage of peak flow

Time in hours (relative to time of peak flow)


Figure 4.1: UPO-Gamma hydrograph for Tr=50 and different values of shape parameter n
Percentage of peak flow

Time in hours (relative to time of peak flow)


Figure 4.2: UPO-Gamma hydrograph for n=3 and different values of scale parameter Tr

4.3.7 Example application of UPO-Gamma model


Figure 4.3 shows the outcome of fitting the UPO-Gamma model to the median hydrograph
for Station 07009 Boyne at Navan Weir derived in Section 3.5. The model has two
parameters: the shape parameter n and the rise-time parameter Tr. The use of only two
parameters limits how well the model can fit a derived hydrograph.

It can be seen in Figure 4.3 that the model fits the rising limb of the median hydrograph at
Station 07009 rather well, and also respects the general shape of the upper hydrograph.
However, it provides a poor representation of the lower part of the receding limb.

22
Volume III Hydrograph Analysis

100 Median hydrograph


90 UPO-Gamma model
Percentage of peak flow

80
70
60
50
40
30
20
10
0
-50 -37.5 -25 -12.5 0 56.25 112.5 168.75 225
Time in hours (relative to time of peak
Figure 4.3: UPO-Gamma flow)
curve fitted to median hydrograph at Station 07009

Experience in applying the model across 89 stations found the behaviour in Figure 4.3 to be
rather typical, with the modelled hydrograph receding too quickly. This limitation was
addressed by developing a version of the model offering an extended recession. This is the
UPO-ERR-Gamma model.

4.4 UPO-ERR-Gamma model for the characteristic hydrograph


The UPO-ERR-Gamma model replaces the lower part of the recession of the UPO-Gamma
model by an exponential recession. ERR denotes Exponential Replacement Recession. An
example is shown in Figure 4.5 below.

4.4.1 Formulation
The UPO-ERR-Gamma model follows the UPO-Gamma model of Section 4.3 until its point
of inflection xI on the receding limb of the hydrograph. It can be shown that the inflection
occurs at:

x I Tr n 1 where n > 1 4.8

Here, x again denotes time relative to the time of peak flow. From Equation 4.7, the
hydrograph ordinate at the inflection point is given by:

n 1

y I 1
1

n 1

exp n 1 4.9

Thereafter, the UPO-ERR-Gamma model replaces the Gamma curve by an exponential


recession. The recession curve is defined by:

23
Volume III Hydrograph Analysis

x x I
y y I exp 4.10
C

where parameter C controls the rate of recession. It is the time over which the recession
descends to a proportion e-1/C of its value. The recession decays to 0.5 of its value over the
time 0.693C, the so-called recession half-life.

Figure 4.4 provides an impression of the family of shapes supported by the exponential
replacement recession. The case illustrated is where the point of inflection on the Gamma
curve occurs 15 hours after the peak flow and at 60% of its value.

Shapes of the exponential recession curve for different values of the


parameter C C= 25
70
x o = 15, y o = 60
of peak flow

C= 50
60
C= 75
Ordinate of the recession

C= 100
curve (in percentile)

50
C= 125
40 C= 150
C= 175
Percentage

30
C= 200
20 C= 225
10
C= 250
C= 275
0 C= 300
0 50 100 150 200 250 300
Time (hr)
Figure 4.4: Exponential replacement recession (ERR) for different values of parameter C

The coupling of the exponential recession replacement to the Gamma curve means that
hydrographs constructed using the UPO-ERR-Gamma model exhibit a kink at the join. This
reflects a discontinuity in the derivative of the modelled hydrograph. This undesirable feature
was considered acceptable given the improved performance achieved by the UPO-ERR-
Gamma model in comparison to the UPO-Gamma model. If required, the kink could be
moderated by local smoothing.

4.4.2 Method of fitting


Various methods of fitting the UPO-ERR-Gamma model were explored. That ultimately
favoured was to fit the model to the median hydrograph by optimising the three parameters
(n, Tr and C) simultaneously using a genetic algorithm (see next section). The method is
implemented within the HWA software package. Figure 4.5 shows the outcome for Station
07009 Boyne at Navan Weir.

4.5 Method of fitting the parametric model


Fitting the UPO-ERR-Gamma model to the characteristic hydrograph requires a criterion of
what is considered a good fit and a means of seeking the parameter values that provide the
best fit by this measure. The criterion is the objective function and the means of seeking the
best parameters is the optimisation scheme.

24
Volume III Hydrograph Analysis

4.5.1 Objective function


The objective function adopted is the so-called least-squares criterion. The best fit is judged
the one that minimises the sum of squared differences between modelled and observed values
of the feature of interest.

100
Median hydrograph
90 UPO-ERR-Gamma model
80
Percentage of peak flow

70 Kink at join of Gamma and exponential replacement recession

60
50
40
30
20
10
0
-50 -37.5 -25 -12.5 0 56.25 112.5 168.75 225
Time relative to time of peak flow (hours)
Figure 4.5: UPO-ERR-Gamma curve fitted to median hydrograph at Station 07009

For applications in rainfall-runoff modelling, the feature of interest is typically ordinates of


the flow hydrograph along its length. In the HWA research, the feature of interest is the
hydrograph width at various percentages of the peak value. In either application, a weight
can be applied to prioritise a good fit in the part of the hydrograph of most interest. The
model fitting is then said to have been done by weighted least-squares.

In fitting the UPO-ERR-Gamma curve to a standardised hydrograph, the objective is to


optimise its three parameters (n, Tr and C) so as to match the known widths of exceedance as
closely as possible in the least-squares sense. The standardised hydrograph being fitted may
be that of a single flood event or a characteristic hydrograph such as the median hydrograph
derived in Section 3.5.

The objective function used for fitting the UPO-ERR-Gamma curve is the weighted sum-of-
squares:
N
S w i y i y i
2
4.11
i 1

Here, yi is the ith ordinate of the standardised hydrograph, i is the corresponding ordinate
modelled by the UPO-ERR-Gamma, wi is a weighting factor, and S is the sum of squares of
the differences between yi and i across ordinates 1 to N. The summation takes place across
all reference ordinates for which the hydrograph is defined, the reference ordinates
corresponding to end-points to which the hydrograph width (or a component of the
hydrograph width) at 98%, 95%, 90%, 85%, 5% of the peak flow is defined. If, for
example, the widths of exceedance on the rising side are available at percentiles 98, 95, 90,
85, 80, 75, 70, 65, 60, 55, 50 and 45 and the widths of exceedance on the receding side are
available at percentiles 98, 95, 90, 85, 80, 75, 70, 65, 60, 55, 50, 45, 40, 35 and 30 then N in

25
Volume III Hydrograph Analysis

Equation 4.11 will be 12 + 1 + 15 = 28. Box 4.1 explains why some users may be puzzled by
this choice of objective function.

Equation 4.11 includes the weight wi. The HWA software supports three weighting systems:

Greater weight to the fitting of the widths at higher percentages of the peak flow, by
setting: wi = (yi)2;
Greater weight to the fitting of the widths at lower percentages of the peak flow, by
setting: wi = (1/yi)2;
Even weighting, by setting: wi = 1.

The first option was adopted in the research, i.e. assigning greater weight to the fitting of
widths at higher percentages of the peak flow.

Box 4.1: The choice between horizontal and vertical fitting


Editorial note: Models such as the UPO-ERR-Gamma have a defined functional form,
albeit a moderately complicated one. It ought to have been possible to fit the hydrograph
width model using an objective function defined in terms of horizontal departures of the
modelled hydrograph from the observed hydrograph. Instead, the HWA authors chose an
objective function based on vertical fitting of the modelled hydrograph to the observed
hydrograph.

Although the difference may not have changed results very appreciably, it is confusing to
have evaluated the comparative performance of different models (in Section 4.5.3 below)
using a measure based on horizontal departures and yet fitted the hydrograph width
models by minimising vertical departures.

Unsurprisingly, the developers of IBIDEM (see Chapter 9) chose a different course. They
chose an objective function that fitted hydrographs from the FSR rainfall-runoff method to
FSU design hydrographs by minimising horizontal departures.

Practitioners are encouraged to make full use of the HWA and IBIDEM software. Visual
display of hydrographs is integral to both packages and should allow users to obtain
effective results. Researchers are to be encouraged to focus on horizontal fitting in any
further development of hydrograph width analysis.

4.5.2 Optimisation scheme


The Genetic Algorithm (GA) was chosen as the optimisation scheme for fitting the UPO-
ERR-Gamma curve. GAs are search algorithms based on the mechanics of natural selection
and genetics. The method was developed by John Holland, colleagues and students at the
University of Michigan (Holland, 1975). The method combines survival of the fittest among
string structures with a structured yet randomized information exchange to form a search
algorithm with some of the innovative flair of human search (Goldberg, 1989). The
University of Michigan research aimed to:
Abstract, and rigorously express, the adaptive processes of natural systems;
Design artificial systems that retain the important mechanisms of natural systems.

26
Volume III Hydrograph Analysis

Genetic algorithms aim to be robust and to balance the efficiency and efficacy necessary for
survival in many different environments. In each generation, a new set of artificial creatures
(strings) is created using bits and pieces of the fittest of the old strings. However, the
occasional new part is tried for good measure (Goldberg, 1989). [Editorial note: A few
details of the procedure implemented are given in the original research report but these are
not specific to the particular application. The method is said to be based on Duan (2003).]

In hydrological applications, the chief advantage of the genetic algorithm over more
conventional methods of optimisation is that the search is globally oriented. This can help
when fitting rather complex models to hydrological problems. [Editorial note: If the
surface being searched (to find the minimum of the objective function) is irregular, gradient
methods of optimisation may lead the user to parameter values corresponding to a local
minimum of the objective function. In contrast, genetic algorithms are equipped to leap clear
and are more likely to find the parameter values corresponding to the global minimum of the
objective function parameters.] Wang (1991) finds the GA to be a robust and efficient
method for calibrating conceptual rainfall-runoff models.

4.5.3 Performance evaluation


Several measures were considered for judging the relative performance of one method (of
modelling the characteristic hydrograph) over another, including the root mean square error
(RMSE) and the Nash-Sutcliffe efficiency coefficient (Nash and Sutcliffe, 1970). The
measure ultimately adopted was the Mean Relative Error (Elshorbagy et al., 2000). Whereas
the Nash-Sutcliffe efficiency and the RMSE are inflated by squared terms, the range of
variation of the values of the MRE was generally found to be small.

In the application here, the MRE at a given percentage pj of the peak flow is defined as:

1
iN j W
W

i, j j
MREp j 4.12
Nj i 1 Wi, j

where Wi,j is the exceedance width of the ith observed hydrograph, j is the exceedance width
given by the model and Nj is the number of observed hydrographs for which the width is
defined. In each case, the subscript j denotes the value relevant at percentage pj of the peak
flow. The measure was evaluated at (up to) 20 percentages of the peak flow, with p1 = 98, p2
= 95, p3 = 90,, p19 = 10 and p20 = 5. For reasons explained in Section 3.2, not all events
have hydrograph widths defined at all percentages of the peak flow. The lower the value of
MRE, the more efficient is the parametric model at reproducing the observed hydrograph
widths (e.g. the widths of the characteristic hydrograph derived by the Section 3.5 method).

In practice, the quality of model fit at and above a given percentage of the peak flow is of
interest. The Average MRE was therefore defined as:
1 k M
AMREp j Average MREp j MREpk,j 4.13
M k 1

where k counts over different percentages of the peak flow at and above pj%. For example,
when the quality of fit at and above 50% of the peak flow is of interest:

27
Volume III Hydrograph Analysis

AMRE50
1
MRE50 MRE55 MRE90 MRE95 MRE98 4.14
11
To simplify matters a little, assessments focused on three reference percentiles corresponding
to the fit above 75%, 50% and 25% of the peak flow. The lower the value of AMRE75, the
better is the fit of the model to the upper part of the hydrograph above 75% of the peak flow.

Box 4.2: Editorial note on performance measures


An array of measures is used to assess performance. Mention of the Nash-Sutcliffe
efficiency (NSE) is perhaps justified by its widespread use in hydrological modelling.
However, it is poorly adapted to the case of modelling hydrograph widths. More aptly, the
root mean square error (RMSE) in hydrograph width estimation is used to good effect in
IBIDEM (see Chapter 9, in particular Section 9.5.4).

The authors preference for the Equation 4.12 measure (MRE) is uncomfortable. When
defining the mean relative absolute error, it is more usual for the denominator to be taken
as the minimum absolute value of the terms being differenced in the numerator. The
disadvantage of the Equation 4.12 measure is that it does not treat modelled and observed
values even-handedly. n consequence, large modelled values of the hydrograph width
(i.e. j) degrade the evaluated performance much more than small values do. ndeed,
large values of j can lead to values of MRE (and AMRE) that exceed 1.0.

4.6 Reproduction of flood hydrographs of verification events


An independent test of the relative merit of competing parametric models was possible by
exploiting the three events reserved for verification at each station (see Section 3.4). Figure
4.6 illustrates the performance achieved with the UPO-ERR-Gamma model for Station 07009
Boyne at Navan Weir. The model reproduces the observed flood hydrographs quite well for
Events 5 and 15, and the fit is especially good above 50% of the peak flow. In the case of
Event 24, only the upper part of the hydrograph was available, this being the main flood
response decoupled from a complex event. The fit of the UPO-ERR-Gamma model is poorer
for this event but considered acceptable given that the flood is not a major one, being only the
24th largest of the 30 events at Station 07009.

When evaluated across all 37 Grade A1 station, the UPO-ERR-Gamma model was judged to
provide the best overall performance (amongst all the parametric models considered) in
modelling the upper parts of the hydrographs reserved for verification. Modelled and
observed hydrographs for these 111 verification events are shown in Figure C.1 to Figure
C.37 of Appendix C. Fits achieved by the non-parametric approach are also shown.

28
Volume III Hydrograph Analysis

100 Median hydrograph


90 UPO-ERR-Gamma model

Percentage of peak flow


Event 5 hydrograph
80
70
60
50
40
30
20
10
0
-50 -25 0 56.25 112.5 168.75 225
Time relative to time of peak f low
(hours)

100 Median hydrograph


90 UPO-ERR-Gamma model
Percentage of peak flow

Event 15 hydrograph
80
70
60
50
40
30
20
10
0
-50 -25 0 56.25 112.5 168.75 225
Time relative to time of peak f low
(hours)

100 Median hydrograph


90 UPO-ERR-Gamma model
Percentage of peak flow

Event 24 hydrograph
80
70
60
50
40
30
20
10
0
-50 -25 0 56.25 112.5 168.75 225
Time relative to time of peak f low
(hours)
Figure 4.6: Performance of UPO-ERR-Gamma model on verification events, Station 07009

29
Volume III Hydrograph Analysis

4.7 Other methods


The research considered many models, and multiple ways of fitting them. The Gamma-UPO-
ERR parametric model was calibrated extensively. The primary versions assessed were:

v1 Fitting to the derived median hydrograph (as recommended and illustrated above);
v2 Fitting to the derived mean hydrograph;
v3 Taking the median of parameter values obtained when fitting to flood hydrographs for
individual events;
v4 Taking the mean of parameter values obtained when fitting to flood hydrographs for
individual events;
v5 Fitting to the event hydrographs collectively.

Further stratification of the study took the objective function as minimising Average MRE
values calculated above 50% of the peak flow, above 75% of the peak flow and above other
percentages. The reader is referred to the original research report and its many appendices.

[Editorial note: It is the practice in applied research to prune back the number of methods,
and to consider in detail only the most promising sub-variations. Some investigators are,
however, reluctant to judge one method better than another without exhaustive analyses.]

Figure 4.7 provides an example of the richness of material generated in the research. A good
method constructs the upper hydrographs of the verification events with low error. The
performance measure AMRE is the average value of the Mean Relative Error in reproducing
upper hydrograph widths (see Section 4.5.3). The recommended version (v1) is seen to
perform best in verification across the 37 Grade A1 stations; v3 performs next best.

2.0 Gamma I median Gamma I mean Gamma II median Gamma II mean Gam
1.8
1.6 v1 v2 v3 v4 v5
MRE 50

1.4
AMRE

1.2
1.0
0.8
0.6
0.4
0.2
0.0
6011
6012
6013
6014
6026
7007
7009
7010
7012

9001
14004
14006
14007
14011
14018
15005
23002
24013
25003
25006
25014
25017
25025
25027
25030
26007
26008
mean Gamma II median
2.0 Gamma IIGamma
mean I median
Gamma III Gamma I mean Gamma IIStation
median no. Gamma II mean Gam
1.8
1.6 v1 v2 v3 v4 v5
1.4
MRE50

1.2
AMRE

1.0
0.8
0.6
0.4
0.2
0.0
14007
14011
14018
15005
23002
24013
25003
25006
25014
25017
25025
25027
25030
26007
26008
26012

26019
27002
29001
29011
30004
30005
34018
36010
36015
6011
6012
6013
6014
6026
7007
7009
7010
7012

9001
14004
14006
14007
14011
14018
15005
23002
24013
25003
25006
25014
25017
25025
25027
25030
26007
26008

Station no. Station no.


Figure 4.7: Verification performance of UPO-ERR-Gamma calibrated in five versions

30
Volume III Hydrograph Analysis

5 Performance of methods at gauged sites

5.1 Introduction
The performance of the recommended methods is now examined. The recommended method
in the non-parametric approach is the median hydrograph of Section 3.5. The recommended
method in the parametric approach is the UPO-ERR-Gamma model developed in Section 4.4.
The initial assessment is made across the 37 Grade A1 stations.

It is helpful to assess the methods jointly. Where poor performance is identified, it is then
possible to distinguish whether this arises more from the choice of method or more from the
variable nature of hydrographs at a particular station.

Figure 5.1 and Figure 5.2 summarise the performance of the median hydrograph and UPO-
ERR-Gamma methods in calibration and in verification across the 37 Grade A1 stations. The
measure shown is AMRE50.

AMRE denotes the average MRE. This is a double averaging. The mean relative error
(MRE) in modelling hydrograph widths is averaged across available segments of the
hydrograph above 50% of the peak flow, and is also averaged across the available flood
hydrographs. For the verification results, the averaging is across the three events withheld for
the purpose. For the calibration, the averaging is across the remaining flood hydrographs.

1.2 Calibration Verification


1.0
0.8
Calibration Verification
AMRE50
MRE

0.6
0.4
0.2
0.0
6011
6012
6013
6014
6026
7007
7009
7010
7012
9001
14004
14006
14007
14011
14018
15005
23002
24013
25003
25006
25014
25017
25025
25027
25030
26007
26008
26012
26019
27002
29001
29011
30004
30005
34018
36010
36015
Figure 5.1: Performance of median hydrograph
Station no. method across 37 Grade A1 stations

1.2 Calibration Verification


1
Calibration Verification
MRE 50

0.8
AMRE

0.6
0.4
0.2
0
6011
6012
6013
6014
6026
7007
7009
7010
7012
9001
14004
14006
14007
14011
14018
15005
23002
24013
25003
25006
25014
25017
25025
25027
25030
26007
26008
26012
26019
27002
29001
29011
30004
30005
34018
36010
36015

Figure 5.2: Performance of UPO-ERR-Gamma method


Station no. across 37 Grade A1 stations

A small value of AMRE50 indicates that the method typically represents the upper hydrograph
well. An arbitrary reference line AMRE50 = 0.4 is drawn across the figures to aid
discrimination of unusual cases. It is seen that AMRE50 lies below 0.4 at most stations.

Although only just discernible in Figure 5.1 and Figure 5.2, the median hydrograph method
outperforms the UPO-ERR-Gamma method. This is more clearly seen in Figure 5.3.

31
Volume III Hydrograph Analysis

1.2

Station 34018
AMRE50 for median hydrograph

1.0

Station 06011
0.8

0.6

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
AMRE50 for UPO-ERR-Gamma
Figure 5.3: Comparison of model performance across 37 Grade A1 stations

In calibration, the median AMRE50 (across the 37 stations) is 0.227 for the median
hydrograph method and 0.246 for the Gamma model. The difference is more noticeable in
verification; the median AMRE50 is 0.200 for the median hydrograph method (better than in
calibration!) and 0.252 for the Gamma model. The superior performance of the median
hydrograph method is to be expected given its greater flexibility to accommodate different
shapes of characteristic hydrographs.

A number of factors influence these results. Some stations exhibit wide variation in
hydrograph shapes across events. These are characterised by relatively high AMRE in
calibration. Prime examples are Stations 06011 and 34018. A further factor is that AMRE is
less well defined at stations for which the flood hydrograph is typically complex. AMRE50
nominally measures the average error in estimating hydrograph widths across the upper half
of the hydrograph. However, in practice, AMRE is evaluated across such widths as are
available after filtering of the hydrograph (see Section 2.6).

5.2 Relative performance in verification compared to that in calibration


The median hydrograph method performs better in verification than in calibration at 23
stations and worse at 13 stations, with Station 07010 too close to call. The UPO-ERR-
Gamma method performs better in verification than in calibration at 21 stations and worse at
15 stations, with Station 34018 too close to call.

As illustrated in Figure 5.4, the improvement in performance is most evident at Stations


06011, 06012, 14006, 14007, 14018, 26007 and 36015 (for both methods) and at Station
26012 (for the median hydrograph method). A deterioration in performance in verification is
most evident at Stations 07012 and 25003 (for both methods) and at Station 25017 (for the
median hydrograph method).

32
Volume III Hydrograph Analysis

4
25017 25003
for median hydrograph method

2 07012
AMRE50Verif / AMRE50Calib

25014
25006
24013
14004
25025
36010
06014
27002
1 30004 26008
3401807010
06026
15005
25030
07007
25027
29001
06013
23002 14011
26007 09001
26019
29011
07009
0.5 14006 30005
36015 14018
14007
06011 26012

0.25
06012

0.125
0.125 0.25 0.5 1 2 4
AMRE50Verif / AMRE50Calib
for UPO-ERR-Gamma model

Figure 5.4: Performance in verification compared to that in calibration

[Editorial note: With many events typically used in calibration, there is considerable scope
for wide variation in hydrographs at a particular station to degrade model performance. With
only three events used for verification at each station, improved performance may in part be a
matter of chance. Three samples may show broad conformity to the calibrated model or wide
departures from it. Nevertheless, these are good results, since one would normally expect
performance in verification to be inferior to performance in calibration.

The unexpectedly strong performance in verification evident in Figure 5.4 may reflect that the
selection of verification events (see Section 3.4) was not fully automated. With over 2800
flood events across 89 catchments, the selection was a very demanding task in itself. Stations
occasionally exhibit individual hydrographs of a highly unusual shape. There may have been
reluctance instinctive rather than conspiring to allow patently unusual events to join the
select few allocated to the verification set.]

5.3 Complexity of hydrographs at Stations 06011 and 34018


The recommended methods perform poorly at Stations 06011 Fane at Moyles Mill and 34018
Castlebar at Turlough. Other methods considered in the research (see Sections 4.2 and 4.7)
also performed poorly on these catchments. A particular difficulty is that flood hydrographs
at these stations are generally complex. At Station 06011, the total width of exceedance is
undefined below 65% of the peak flow in all the hydrographs studied. The restriction is even
more severe at Station 34018, where none of the hydrographs defines the total width of
exceedance below 80% of the peak flow. The hydrographs of four of the eight largest floods
at this station are illustrated in Figure 5.5.

33
Volume III Hydrograph Analysis

Event 1 Event 2
Flow (m3s-1)

Flow (m3s-1)
Event 6 Event 8

Flow (m3s-1)
Flow (m3s-1)

Figure 5.5: Varied hydrograph shapes at Station 34018 Turlough at Castlebar

A further difficulty is that upper segments of the hydrographs at Stations 06011 and 34018
have a mix of sharp-peaked and rounded shapes. The characteristic hydrographs derived by
the median and UPO-ERR-Gamma methods are gently rounded (see Figure C.1 and Figure
C.35) yet some individual hydrographs have upper segments that are sharp-peaked.

The variability in upper-hydrograph shape is particularly marked at Station 34018 Castlebar


at Turlough (see Figure 5.5). This may reflect special features of the catchment that provide
particular scope for temporal and spatial patterns in rainfall to lead to varied and complex
hydrograph shapes. [Editorial note: The name Castlebar at Turlough suggests that geolog-
ical features may in part account for the complexity and variability of hydrographs at Station
34018. The authors note the large loughs in the catchment and suggest that the layout
increases the sensitivity of response timings to the spatial pattern of rainfall.]

5.4 Variability in hydrograph widths at some stations


Hydrographs at stations such as 24013 Deel at Rathkeale, 25006 Brosna at Ferbane, 25014
Siler at Millbrook and 25025 Ballyfinboy at Ballyhooney exhibit wide variability in widths.
As examples, two relatively wide and two relatively narrow hydrographs at each of these
stations are reproduced in Figure 5.6 to Figure 5.9. In the case of Station 25025, the two
narrower hydrographs are rather curiously slanted.

In the context of such wide variations in the shapes of flood hydrographs, it is desirable that
the reasons be investigated and explanations sought in terms of catchment features and/or the
meteorological conditions leading to the observed floods at these stations. Station 24013 is
investigated further in Section 5.10.2.

34
Volume III Hydrograph Analysis

Event 2 Event 11
Flow (m3s-1)

Flow (m3s-1)
Event 20 Event 27
Flow (m3s-1)

Flow (m3s-1)

Figure 5.6: Wide and narrow-peaked hydrographs at Station 24013 Deel at Rathkeale

Event 1 Event 4
Flow (m3s-1)

Flow (m3s-1)

Event 3 Event 31
Flow (m3s-1)

Flow (m3s-1)

Figure 5.7: Wide and narrow-peaked hydrographs at Station 25006 Brosna at Ferbane

35
Volume III Hydrograph Analysis

Event 7 Event 23
Flow (m3s-1)

Flow (m3s-1)
Event 9 Event 21
Flow (m3s-1)

Flow (m3s-1)

Figure 5.8: Wide and narrow-peaked hydrographs at Station 25014 Silver at Millbrook

Event 5 Event 13
Flow (m3s-1)

Flow (m3s-1)

Event 2 Event 31
Flow (m3s-1)

Flow (m3s-1)

Figure 5.9: Wide and slanted hydrographs at Station 25025 Ballyfinboy at Ballyhooney

36
Volume III Hydrograph Analysis

Twelve flood events at Station 25014 Silver at Millbrook were studied as part of FSU work
on Flood Event Analysis (see Appendix B). Six events were used to derive an average unit
hydrograph, which was tested on six further events. One of these was the 30 June 1986 flood,
which corresponds to the second part of Event 9 (see lower left diagram in Figure 5.8).

The University College Cork analysis in Figure 5.10 suggests that the catchment responded
particularly quickly to rainfall in this event. DR denotes the direct runoff from rainfall. A
possible explanation for the unusually narrow hydrograph is that the flood arose from heavy
rainfall on part of the lower catchment only. The catchment configuration is somewhat
unusual (see Map 5.1 based on UCC research).

Figure 5.10: Rainfall-runoff behaviour in 30 June 1986 flood at Station 25014

Map 5.1: Catchment of Station 25014 Silver at Millbrook

37
Volume III Hydrograph Analysis

5.5 Attenuated response at some stations


In the case of (e.g.) Station 25017 Shannon at Banagher, the flood hydrographs are
exceedingly attenuated (i.e. dampened or flat), with only very slow rises to and recessions
from the peak. The peak segment is maintained for several days, and sometimes represents a
sustained high flow with no outstanding peak. Four such hydrographs are reproduced in
Figure 5.11.

Event 1

Event 4
Flow (m3s-1)

Flow (m3s-1)

Event 10
Flow (m3s-1)

Flow (m3s-1)

Event 7

Figure 5.11: Attenuated hydrographs at Station 25017 Shannon at Banagher

For a station characterised by such very flat flood hydrographs, only a very small number of
percentile widths are available for calibrating a model of the characteristic hydrograph. It is
therefore unsurprising that a derived hydrograph produced from such a small number of ill-
defined events may not reproduce the flood hydrograph of a verification event very well.

5.6 General guidance


On the basis of the above, it is concluded that, for reliability in producing a design
hydrograph:

Standardised hydrographs should be derived for as many flood events as possible;


Where these differ widely from event to event, the reasons should be explored on the
basis of the physical characteristics of the catchment and of the flood-producing
rainfall;

38
Volume III Hydrograph Analysis

A characteristic hydrograph fitted to very varied hydrographs, or one developed from


flood hydrographs that are highly attenuated (i.e. dampened or flat), should be
applied with caution when producing design flood hydrographs.

5.7 Results of whole-sample calibration


Having identified best non-parametric and parametric methods i.e. the derived median
hydrograph and the UPO-ERR-Gamma model based on 37 Grade A1 stations, the methods
were applied to the entire set of 89 stations in the study. All available events were now used:
hence, the term whole-sample calibration.

5.7.1 Derived median hydrograph and its descriptors


Figure 5.12 shows the outcome of the whole-sample calibration for Station 07009 Boyne at
Navan Weir.

100 Percentage Hydrograph width (hours)


90 of peak
flow On rising limb On receding limb Total
80
Percentage of peak flow

70 75% 10.46 17.29 27.75


60 50% 14.94 39.20 54.14
50 25% 23.38 136.67 160.05
40
30
20
10
0
-50 -37.5 -25 -12.5 0 56.25 112.5 168.75 225
Time relative to time of peak flow (hours)
Figure 5.12: Median hydrograph for Station 07009 Boyne at Navan Weir (whole sample)

The table embedded in Figure 5.12 notes the values of the hydrograph widths at 75%, 50%
and 25% of the peak. Comparison with Figure 3.3 reveals that, for this station, the median
hydrograph is little changed after inclusion of the three events previously reserved for
verification.

Table 5.1 summarises the outcome of hydrograph width analysis applied to the full set of 89
stations. Three width descriptors are presented in the central columns. Two are the widths of
exceedance W75 and W50 at 75% and 50% of the peak flow. Hydrographs are insufficiently
defined to establish these width descriptors for Stations 30061 and 34018; additionally, W50 is
undefined for Stations 06011, 25017 and 34001.

The third width descriptor is the mean ratio, s, of the width on the rising side to the total
width. This average value is calculated across all available widths of all the available flood
hydrographs. The ratio s summarises the skew of the hydrograph, with a value of 0.5
indicating broad symmetry about the peak. Figure 5.13 shows the histogram of values of s

39
Volume III Hydrograph Analysis

across the 89 stations analysed. It is seen that s is appreciably less than 0.5 at most stations.
This reflects that floods typically rise more quickly than they fall. For Station 07009,
s = 0.345.

7
<-- s = 0.5 (symmetric)
6
5
Frequency

4
3
2
1
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
s = mean ratio of width before peak to total hydrograph width
Figure 5.13: Summary index, s, of hydrograph skewness (89 Grade A1 + A2 stations)

[Editorial note: Ratios (such as s) should be averaged by taking the geometric mean not by
taking the arithmetic mean. Although not of great significance here, the difference can be
important in particular hydrological applications.]

5.7.2 UPO-ERR-Gamma model and its parameters


The UPO-ERR-Gamma was applied to the entire set of 89 stations, fitting the model by
optimisation using the genetic algorithm (see Section 4.5.2) to the derived median hydrograph
for the particular station. The optimised parameter values appear in the final three columns of
Table 5.1. These are values of the shape parameter n, the rise-time or translation parameter Tr
and parameter C of the exponential replacement recession.

The C parameter could not be calibrated for Station 30061. The Gamma curve fitted to the
derived median hydrograph for this Grade A2 station implies a point of inflection on the
receding limb at 71% of the peak flow, some 30 hours after the peak. For this station, the
median hydrograph does not define after-the-peak hydrograph widths below 90% of the peak
flow. Thus, it was not possible to fit the exponential replacement recession.

Table 5.1: Outcome of whole-sample calibration at 89 Grade A1 + A2 stations


Station Station Median hydrograph method UPO-ERR-Gamma model
number grade W75 W50 s n Tr C
06011 A1 114.35 N/A 0.374 1.300 29.690 360.834
06012 A1 136.25 243.02 0.325 2.457 93.311 280.948
06013 A1 41.99 70.87 0.354 7.782 59.989 102.855
06014 A1 95.87 154.30 0.394 3.014 89.910 113.849
06026 A1 82.47 146.16 0.390 4.998 102.176 140.966
07001 A2 18.36 34.20 0.377 5.006 23.309 38.360
07002 A2 46.74 93.14 0.351 2.770 32.736 151.959
07004 A2 110.14 171.87 0.382 3.014 101.979 116.780

40
Volume III Hydrograph Analysis

Station Station Median hydrograph method UPO-ERR-Gamma model


number grade W75 W50 s n Tr C
07006 A2 15.42 30.38 0.331 7.782 20.874 58.149
07007 A1 26.22 52.53 0.355 3.745 26.837 69.875
07009 A1 27.75 54.14 0.345 5.276 33.040 95.526
07010 A1 30.42 119.06 0.348 3.458 25.784 222.317
07011 A2 108.21 170.42 0.344 2.988 92.964 116.780
07012 A1 27.75 49.10 0.375 6.111 36.407 88.197
07033 A2 46.41 84.10 0.372 6.111 60.818 105.054
09001 A1 11.21 22.64 0.394 8.399 21.751 22.970
11001 B 7.54 10.84 0.456 26.356 25.241 5.380
14004 A1 48.17 74.35 0.447 5.406 66.484 53.751
14006 A1 57.46 83.02 0.487 4.710 75.828 41.292
14007 A1 9.34 17.14 0.487 21.412 31.787 11.243
14009 A2 26.27 46.20 0.378 3.884 27.746 46.422
14011 A1 45.96 89.57 0.361 5.276 54.734 116.780
14018 A1 65.08 121.78 0.350 2.944 57.728 116.780
15001 A2 15.74 28.49 0.434 9.417 30.458 23.703
15002 A2 8.12 19.23 0.428 17.635 24.966 21.504
15003 A2 5.77 8.79 0.489 10.541 12.422 6.113
15005 B 60.06 114.00 0.417 5.267 83.234 106.520
15006 A2 19.95 35.64 0.517 13.829 53.250 22.970
16001 A2 30.61 46.41 0.359 5.554 40.156 39.826
16002 A2 58.04 102.32 0.411 8.338 91.266 114.581
16003 A2 39.53 77.32 0.216 2.527 16.148 188.604
16004 A2 63.79 102.80 0.429 6.390 90.244 99.191
16005 A2 13.76 21.02 0.541 28.192 51.750 5.380
16008 A2 88.08 146.96 0.243 3.153 44.158 410.670
16009 A2 39.50 92.10 0.360 6.111 49.988 163.685
18004 A2 38.18 73.27 0.327 4.998 43.894 105.054
18005 A2 15.63 26.35 0.427 24.604 52.750 15.641
19001 A2 26.34 44.94 0.245 2.788 12.675 112.383
22071 A2 96.29 190.22 0.348 1.883 50.563 243.571
23001 A2 6.98 11.84 0.442 12.536 16.654 5.380
23002 A1 5.81 9.27 0.528 30.265 22.999 3.182
23012 A2 11.67 18.29 0.474 18.226 32.131 11.243
24001 A2 18.24 27.88 0.360 4.441 19.161 35.429
24008 A2 19.08 30.54 0.363 3.884 19.404 28.100
24013 A1 19.66 28.77 0.570 12.333 48.750 6.846
24082 A2 18.66 27.51 0.316 5.137 21.420 38.360
25001 A2 15.51 29.51 0.460 9.452 31.941 35.429
25003 A1 15.38 29.57 0.343 4.302 16.506 50.087
25005 A2 14.00 33.17 0.314 4.998 13.136 69.875

41
Volume III Hydrograph Analysis

Station Station Median hydrograph method UPO-ERR-Gamma model


number grade W75 W50 s n Tr C
25006 A1 35.86 63.91 0.396 5.276 42.419 94.060
25014 A1 17.63 30.05 0.393 8.965 30.830 36.162
25016 A2 20.26 36.19 0.332 3.884 18.404 62.546
25017 A1 231.91 N/A 0.442 1.500 109.750 374.759
25025 A1 85.33 187.22 0.285 4.136 111.286 222.317
25027 A1 7.66 14.25 0.438 19.409 23.416 19.305
25029 A2 15.87 29.50 0.416 8.886 28.700 50.087
25030 A1 70.22 122.12 0.395 4.223 81.857 94.060
26002 A2 118.67 163.98 0.462 6.581 186.500 39.826
26005 A2 136.04 209.83 0.418 3.597 148.511 84.533
26007 A1 156.97 257.02 0.445 3.040 159.806 107.985
26008 A1 106.69 206.81 0.419 3.875 124.500 177.610
26009 A2 26.52 38.47 0.486 8.469 49.000 14.175
26012 A1 177.27 285.94 0.400 3.849 195.500 257.496
26019 A1 72.81 114.01 0.432 3.362 71.390 70.608
26021 A2 52.68 179.79 0.275 3.484 24.799 527.933
26022 A2 41.75 74.82 0.404 4.998 53.640 66.943
27001 A2 14.87 20.26 0.385 4.998 17.722 11.243
27002 A1 270.72 493.96 0.430 5.276 429.250 187.871
29001 A1 49.94 86.19 0.429 5.328 74.750 46.422
29004 A2 76.87 130.05 0.526 2.770 79.500 76.471
29011 A1 138.65 242.46 0.458 4.014 182.750 114.581
30004 A1 44.33 68.49 0.453 5.302 59.439 46.422
30005 A1 39.87 56.46 0.562 7.399 77.000 22.970
30007 A2 38.52 59.31 0.549 6.111 67.750 20.771
30061 A2 N/A N/A 0.447 2.910 41.901 N/A
34001 A2 95.52 N/A 0.416 2.788 58.347 829.752
34009 A2 19.67 26.03 0.559 9.095 39.699 8.312
34018 A1 N/A N/A 0.315 1.274 35.584 187.871
35001 A2 103.62 154.99 0.435 3.875 116.937 105.787
35002 A2 7.62 11.39 0.577 17.765 24.011 0.25
35005 A2 38.18 85.88 0.340 4.998 36.804 187.871
35071 A2 123.67 210.68 0.321 4.223 99.464 456.843
36010 A1 75.15 155.52 0.440 3.040 73.102 152.692
36011 B 316.02 613.41 0.371 2.214 295.250 229.646
36015 A1 40.08 67.19 0.485 6.189 68.000 35.429
36019 A2 295.51 491.91 0.421 2.483 254.500 302.202
36021 A2 3.11 4.90 0.462 14.698 7.655 2.449
36027 A2 332.71 539.54 0.376 2.605 300.500 213.522
39009 A2 45.10 86.66 0.348 9.104 74.098 115.314

42
Volume III Hydrograph Analysis

[Editorial note: The parameter values of Table 5.1 reveal some unusual features of the
optimisation. Excessive decimal places are shown to confirm that parameter n takes certain
preferential values, with n = 4.998 at six stations and n = 6.111 at a further four stations.
Parameter C also has preferential values, with C = 116.780 at four stations and the values
5.380, 11.243, 27.970, 35.429, 46.422 and 187.871 each appearing at three stations. While
the fits achieved are fully satisfactory, the preferential values suggest that results obtained
with the genetic algorithm are not always fully optimised.]

5.7.3 Stations where the flood hydrograph recedes faster than it rises
Flood hydrographs at Station 35002 Owenbeg at Billa Bridge typically fall more steeply than
they rise. In consequence, the derived median hydrograph in Figure 5.14 also has this feature.
The station has the most skewed hydrograph (s = 0.577) of any in the 89-station dataset.
Reference to Figure 4.1 indicates that this is not a shape that the Gamma family of curves can
accommodate. When forced to do so, the exponential replacement recession in the adopted
UPO-ERR-Gamma is unreasonably steep, with C = 0.25 hours (see Figure 5.14). [Editorial
note: In essence, the ERR part of the model is attempting to compensate for overestimation
(by the Gamma curve) of the hydrograph width component that occurs after the peak.]

100 Median hydrograph


90 UPO-ERR-Gamma model
Percentage of peak flow

80
70
60
50
40
30
20
10-25 -18.75 -12.5 -6.25 0 18.75 37.5 56.25 75
Time in hours (relative to time of peak flow)
Figure 5.14: Characteristic hydrograph for Station 35002 Owenbeg at Billa Bridge

Station 30005 Robe at Foxhill also exhibits hydrographs that typically fall more steeply than
they rise. However, the feature is less pronounced and the UPO-ERR-Gamma model might
be considered just about acceptable (see Figure 5.15).

Overall, the UPO-ERR-Gamma curve was found to model the characteristic hydrograph
adequately at most of the 89 stations. One station having a somewhat unusual characteristic
hydrograph shape is Station 16008 Suir at New Bridge, investigated in the next section.

43
Volume III Hydrograph Analysis

100 Median hydrograph


90 UPO-ERR-Gamma model
Percentage of peak flow

80
70
60
50
40
30
20
10-75 -56.25 -37.5 -18.75 0 43.75 87.5 131.25 175
Time relative to time of peak flow (hours)
Figure 5.15: Characteristic hydrograph for Station 30005 Robe at Foxhill

5.8 Characteristic hydrographs on the River Suir


Multiple gauging stations allow one to see how the characteristic hydrograph changes down
the river system. Figure 5.17 presents the outcome of applying the recommended methods to
four stations on the Suir. The timescale is the same in each graph. Of particular interest is
that the width of hydrographs does not appear to widen appreciably as floods pass down the
river system, except around New Bridge. The steepening of the rising limb is particularly
noticeable at New Bridge and Caher Park.

Median hydrograph Median hydrograph


UPO-ERR-Gamma model UPO-ERR-Gamma model

16004 Thurles 16002 Beakstown


229 km2 486 km2

Median hydrograph Median hydrograph


UPO-ERR-Gamma model UPO-ERR-Gamma model

16008 New Bridge 16009 Caher Park


1090 km2 1583 km2

Figure 5.16: Characteristic hydrographs for four stations on the River Suir

The period of record used in the HWA is substantial. The analysis spans 1954 to 2000 at all
four sites, and somewhat longer periods at Stations 16008 and 16009 (see Table A.2). The
effects evident in Figure 5.17 are therefore not thought to be an artefact of the data samples.

44
Volume III Hydrograph Analysis

The UPO-ERR-Gamma model represents well the steepening of the rising limb (see Figure
5.17). However, the model has difficulty with the unusual shape of the median hydrograph at
Station 16008 Suir at New Bridge. Small values of n and Tr are needed to represent the rising
limb and the UPO-ERR-Gamma model cannot accommodate the appreciable hunch on the
receding limb.

The Suir example provides a strong reminder of the very substantial hydrograph records held
for many Irish rivers, and the scope for hydrograph width analysis to complement the
statistical analysis of flood peaks in Volume II.

1.00
16004 Suir at Thurles
16002 Suir at Beakstown
16008 Suir at New Bridge
Characteristic hydrograph

0.75
16009 Suir at Caher Park

0.50

0.25

0.00
-96 -72 -48 -24 0 24 48 72 96 120 144 168 192 216
Time relative to time of peak flow (hours)
Figure 5.17: UPO-ERR-Gamma characteristic hydrographs for four stations on the Suir

5.9 Hydrograph width analysis at gauged sites a summary


Table 5.1 has summarised the hydrograph width analysis (HWA) undertaken for the 89 Grade
A1 + A2 stations. The results presented comprise:

Three hydrograph-width descriptors (W75, W50 and s) summarising the upper part of
the median hydrograph derived using the recommended non-parametric method;
Three parameters (n, Tr and C) of the UPO-ERR-Gamma model, which is the
recommended parametric model for the characteristic hydrograph.

The design flood hydrograph of required return period can be produced by scaling up the
ordinates of the characteristic hydrograph by the relevant peak flow derived by Volume II
methods.

It is desirable that the HWA results are updated and extended as new data become available.
Hydrograph width analysis is assisted by provision of the HWA software (see Section 1.6 and
Appendix E). It is anticipated that users will upload their HWA results to the FSU Web
Portal: both to facilitate their own applications and to share the hydrograph-width descriptors
and parameters values obtained at gauged sites with the wider community.

45
Volume III Hydrograph Analysis

5.10 Flood hydrographs having sustained peaks


Sometimes flood hydrographs at a station are observed to have a sustained crest segment in
which the flow varies little around the peak. Such flood hydrographs may or may not be
representative of the typical flood hydrographs occurring at a station. Two cases are
discussed.

5.10.1 Where the hydrograph shape reflects the temporal pattern of rainfall
During a storm event, the continuation of heavy rainfall at the peak-inducing intensity for a
considerable period of time may cause the resulting flood hydrograph to exhibit a sustained
peak. [Editorial note: This behaviour may arise when a rain-producing weather system
becomes relatively fixed, e.g. with moist air continually fed into a particular zone where
topography induces heavy rainfall. The weather system is sometimes termed a seeder-feeder
mechanism. The topographic effect is referred to as orographic enhancement.] Although
these situations can give rise to a flood of unusually large volume, such events are unlikely to
be a regular occurrence at any particular station.

The premise of hydrograph width analysis is to represent only the characteristic shape of the
hydrograph. Where only one or two events exhibit the sustained hydrograph peak, it is proper
that they do not unduly influence the characteristic hydrograph. This is the case for the
recommended non-parametric method, which takes the median (rather than the mean) of
hydrograph widths.

5.10.2 Where the hydrograph shape is characteristic of the station


At some stations, relatively flat-peaked flood hydrographs are a characteristic feature
attributable to floodplain storage or other features upstream. A prime example is Station
24013 Deel at Rathkeale. A floodplain effect is inherently threshold-sensitive, so it is prudent
to study actual hydrographs (in m3s-1) as well as standardised hydrographs.

Filtered hydrographs for all 31 events at Station 24013 are overlain in Figure 5.18. The
hydrographs have been translated to synchronise their peaks but not otherwise adjusted. It is
seen that a slowly rising crest segment is indeed typical of flood hydrographs at this site. The
characteristic hydrographs shown in the figure have been scaled up by multiplying the median
standardised hydrograph and the UPO-Gamma standardised hydrograph by the median
hydrograph peak of 116.75 m3s-1.

The non-parametric method is seen to perform well, with the thick blue line in Figure 5.18
representing the tendency for the crest segment of Deel at Rathkeale flood hydrographs to rise
slowly to a peak before dropping away. The UPO-Gamma model is incapable of representing
this effect.

[Editorial note: Confusingly, the thick black line in Figure 5.18 shows the UPO-Gamma
model rather than the recommended UPO-ERR-Gamma. As explained in Section 5.7.3, the
exponential replacement recession can compensate a little for the intrinsic unsuitability of the

46
Volume III Hydrograph Analysis

Gamma curve in cases where the characteristic hydrograph falls more rapidly than it rises.
The relevant recession is marked by the black broken line superposed on Figure 5.18.]

Figure 5.18: Hydrographs and rescaled characteristic hydrograph for Deel at Rathkeale

47
Volume III Hydrograph Analysis

6 Constructing the characteristic hydrograph at ungauged sites

6.1 Introduction
A major part of research was to generalise Hydrograph Width Analysis (HWA) to allow the
characteristic hydrograph to be constructed at ungauged sites. The original research report
provides a very detailed description of the work. A prcis is presented here. Holder (1985)
provides a primer on the use of multiple regression in hydrology.

6.1.1 Links with other parts of the FSU


The generalisations are based on the physical catchment descriptors (PCDs) presented in
Volume IV. There is some commonality with the Volume II work on generalising a model of
the index flood QMED. Both studies use multiple regression analysis, and develop linear
models between logarithmic transforms of the variable of interest and logarithmic transforms
of the PCDs. Individual researchers favour particular notations, terminology and refinements
of the basic techniques. Editing has sought to unify the descriptions.

Whereas the flood frequency research generalises one variable only (QMED), the HWA
research considers three descriptors (W75, W50 and s) and three model parameters (Tr, n and
C). A generic term is needed and these six are referred to below as the dependent variables
(DVs) of the regression study. Having adopted this terminology, it is natural to refer to the
PCDs as the independent variables (IVs).

6.1.2 Assumptions and difficulties


The following standard assumptions of multiple linear regression are noted:

The relationship between the DV and the IVs is linear;


The expected value of the arithmetic mean of the errors is zero (i.e. unbiased);
The error variance is constant across the range of values of the IVs (i.e. homo-
scedasticity);
The errors associated with one set of IV values are uncorrelated with those of another
(i.e. independence);
The errors are normally distributed with zero mean and constant variance (i.e.
normality) while normality is necessary for the t-tests to be valid, estimation of the
model coefficients requires only that errors be identically and independently
distributed;
The model is properly specified (i.e. all relevant IVs are included and all irrelevant
ones discarded).

Other issues to be considered are:

The avoidance of over-determination. An over-determined or over-parameterised


model is one which fits too many free parameters in relation to the number of data
samples available. An adequately determined model is needed if the derived model is
to generalise sensibly to estimate the DV at ungauged sites.

48
Volume III Hydrograph Analysis

Checking that individual observations do not exert undue influence on the model.
Checking for collinearity amongst the IVs (see Box 6.1). If IVs are highly
correlated, the regression coefficients are unlikely to be robust; values are likely to be
unduly sensitive to changes such as the addition of a station to the calibration dataset.

It can readily be appreciated that, in hydrological applications, complete satisfaction of the


assumptions and full resolution of the above issues cannot be achieved. Indeed, some of the
assumptions are expressly contradicted:

Some subsets of the PCDs are known to exhibit high collinearity;


The sample size of stations is unlikely to be large enough to be able to generalise a
model that includes all factors thought physically relevant;
The homogeneity and representativeness of samples is often difficult to ascertain.

Box 6.1: Collinearity


Collinearity refers in a strict sense to the presence of exact linear relationships within a set
of variables. Typically, these are a set of candidate explanatory (i.e. predictor) variables in
a regression-type model. In statistical usage, collinearity also refers to near-collinearity,
i.e. when variables are close to linearly related.

In a multiple regression with collinearity, least-squares regression coefficients are highly


sensitive to very minor changes in the input data. The least-squares problem or the dataset
is said to be ill-conditioned. Some or all of the regression coefficients are likely to be
meaningless.

A typical approach to overcoming collinearity is to simplify the problem, e.g. by retaining


only one of the subset of variables that are highly correlated. Relatively arbitrary
decisions as to which variables to retain and which to remove are sometimes
unavoidable and inevitably influence the final model achieved.

6.2 Selection of dependent variables (DVs)


Regression models are developed for three hydrograph-width descriptors (W75, W50 and s)
and three model parameters (Tr, n and C). These respectively characterise the recommended
non-parametric and parametric approaches to HWA. These six hydrograph descriptors are
the dependent variables (DVs) of the regression analyses.

Four of the variables W75, W50, Tr and C are times in hours. The other two are dimensionless,
with the mean ratio s taking values between 0 and 1 and n being a real number > 1.0. Values
of the six DVs obtained at 89 gauging stations are given in Table 5.1 and form the basic
HWA data used. These data are further summarised in Table 6.1.

For reasons noted in Section 5.7, values of W75, W50 and C could not be derived for a few
stations. The six variables were n-transformed prior to the main model-building, where n
denotes the natural logarithm.

49
Volume III Hydrograph Analysis

Table 6.1: Summary statistics of dependent variables (DVs) selected for regression analysis
Dependent No. of Geom. Arith.
Unit Minimum Median Maximum CV
variable stations mean mean
W75 h 87 3.11 39.87 332.71 38.70 63.31 1.09
W50 h 84 4.90 72.07 613.41 65.07 106.36 1.11
s 89 0.22 0.40 0.58 0.40 0.41 0.18
n 89 1.27 5.00 30.27 5.44 7.04 0.86
Tr h 89 7.66 49.99 429.25 49.38 69.16 1.00
C h 87 2.45 84.53 829.75 63.46 116.04 1.14

6.3 Selection of independent variables (IVs)


In principle, some 36 PCDs might have been considered as independent variables (IVs) in the
regression analysis. Those available included such items as the Eastings and Northings of the
station location and of the catchment centroid. It was judged prudent to use only a selection
of them in order to minimise the expected effects of collinearity (see Box 6.1).

The number of stations for which the dependent variable is available (i.e. 84 to 89 in this
research) is a limiting factor on the number of IVs that can be effectively supported in a
model. Tabachnick and Fidell (2001) and Brace et al. (2003) are amongst those putting
forward rules of thumb for the maximum number of IVs that should be considered when
searching for a best model. The limit adopted here was to allow no more than 12 Vs in the
main model-building.

Table 6.2 presents a brief description and some summary statistics for the 19 PCDs
considered initially. Eighteen of these are explicitly discussed in Volume IV and need no
further introduction here. The additional PCD was CGDIST: the distance in km from
catchment centroid to catchment outlet (i.e. the station location). When used in conjunction
with other measures of catchment size, CGDIST can help to represent catchment shape.
Some aspects of hydrograph shape are expected to reflect catchment shape.

[Editorial note: The flood attenuation index FAI and the permeability descriptor BFIsoil were
unavailable at the time of study. The omission of FAI is unfortunate, given that upper
hydrograph shape is expected to be influenced by floodplain effects. Non-availability of
BFIsoil was dealt with here by developing model variants according to whether a gauged BFI
value is available. It will be noted from Volume IV that BFIsoil coincides with gauged BFI at
the 166 stations used in its calibration. Of the two indices of arterial drainage, ARTDRAIN
was favoured over ARTDRAIN2. The opposite choice was made in modelling QMED in
Volume II, so it is important for FSU users to distinguish the two descriptors of arterial
drainage. Both S1085 and TAYSLO definitions of mainstream slope were retained but
S1085 ultimately proved the more useful.]

The 19 PCDs were n-transformed prior to the main model-building, where n denotes the
natural logarithm. Where the lower range of a particular descriptor can take a value of zero,
1.0 is added to the value prior to the n-transformation. This applies to the fractions
ALLUV, ARTDRAIN, FOREST, PASTURE, PEAT and URBEXT.

50
Volume III Hydrograph Analysis

Table 6.2: Some summary statistics of the IVs initially selected


# of
Notation Brief description Unit Min Median Max Mean CV
stns
ALLUV Alluvial fraction 89 0.0029 0.0349 0.0977 0.0370 0.53
AREA Catchment area km2 89 23.41 292.67 7980.41 624.76 1.60
Fraction of area mapped as
ARTDRAIN benefiting from arterial 89 0.00 0.0015 0.3669 0.0576 1.45
drainage
BFI Baseflow index 79 0.27 0.61 0.83 0.60 0.20
Distance from catchment
CGDIST km 89 4.83 11.72 54.19 13.86 0.58
centroid to gauging station
km/
DRAIND NETLEN/AREA 89 0.39 0.96 1.81 0.99 0.29
km2
Index of flood attenuation by
FARL 89 0.66 1.00 1.00 0.94 0.09
reservoirs and lakes
FOREST Forested fraction 89 0.0161 0.0751 0.5619 0.0934 0.87
FLATWET Wetness index 89 0.54 0.62 0.73 0.63 0.07
MSL Mainstream length km 89 13.92 38.14 214.61 47.65 0.64
NETLEN Length of upstream network km 89 23.99 293.54 6428.92 580.01 1.47
PASTURE Pasture fraction 89 0.2319 0.8134 0.9738 0.7737 0.22
PEAT Peat fraction 89 0.00 0.0527 0.4826 0.0994 1.09
Average annual potential
SAAPE mm 89 449.60 503.96 546.98 498.22 0.05
evapotranspiration
SAAR Average annual rainfall mm 89 783.26 1023.31 2101.83 1068.66 0.19
Number of segments in
STMFRQ 89 7.00 288.00 5490.00 576.66 1.48
upstream river network
Mainstream slope (excluding
S1085 m/km 89 0.00021 0.00206 0.01867 0.00277 1.02
top 10% and bottom 15%)
TAYSLO Taylor-Schwarz stream slope m/km 89 0.00001 0.00106 0.01308 0.00175 1.25
URBEXT Urban fraction 89 0.00 0.0064 0.0553 0.0081 0.99

6.4 Additional notes

6.4.1 Treatment of Gamma shape parameter n


One of the six DVs the shape parameter n is required to take a value >1 if the Gamma
distribution is to provide a hydrograph that rises to a peak rather than immediately decays.
This case did not arise in fitting the UPO-ERR-Gamma at any of the 89 stations analysed.
The minimum value found was 1.29. Most of the modelling used the DV in the form n(n).
However, it was prudent in the final modelling to use the DV in the form n(n-1). This
guaranteed that hydrographs generated by the model at ungauged sites always rise to a peak
before decaying.

51
Volume III Hydrograph Analysis

6.4.2 Software used for the regression analysis


A one-sentence summary of the goal of multiple regression is to identify the fewest IVs
necessary to predict a DV where each IV predicts a substantial and independent segment of
the variability in the DV (Tabachnick and Fidell, 2001). The SPSS statistical software
package was used for: exploring inter-correlations, developing regression equations,
assumption-checking and making additional diagnostic checks. Details and examples are
given in books such as Pallant (2001), Tabachnick and Fidell (2001) and Brace et al. (2003).

6.5 Correlation studies


Correlations amongst the PCDs were examined, both in their native form and after
logarithmic transformation. Those most strongly correlated were examined further. The
correlation matrix of the 19 IVs and six DVs is given in Table 6.3. This shows the ordinary
(Pearson) correlation coefficient for the n-transformed variables.

6.5.1 Inter-correlations between PCDs


Matrix plots showing pairwise scatter-plots were extensively examined. Figure 6.1 illustrates
the extent to which AREA, MSL, NETLEN, STMFRQ and CGDIST compete to represent
catchment size. These strong correlations are understandable. [Editorial note: STMFRQ
denotes the number of streams in the catchment. This is one greater than the number of
stream junctions. Larger catchments tend to have more stream junctions. Thus, STMFRQ in
the FSU is correlated with catchment size. In contrast, the FSR (NERC, 1975) defined
STMFRQ as the number of junctions per unit area.]

Using a Pearson correlation with absolute value greater than 0.6 as an arbitrary guide, other
strong correlations evident in Table 6.3 are briefly discussed.

The strong inverse correlation between mainstream slope (represented by nS1085) and
variables indicative of catchment size (e.g. nAREA, nMSL and nCGDIST) is principally a
function of topography: steep catchments are inevitably rather small. [Editorial note: Such
correlations reflect both physical properties and the available network of stations. When all
134,000 FSU ungauged catchments are considered, the inverse correlation between nS1085
and nAREA is weaker, with r = -0.41 (as opposed to r = -0.66 in Table 6.3). Many small
catchments in Ireland have mild stream-slopes but the majority are not sufficiently important
to be gauged.]

Another grouping of PCDs is PASTURE, PEAT and SAAR. The strong inverse correlation
(r = -0.84) between n(1+PASTURE) and n(1+PEAT) appears straightforward. Catchments
dominated by more peaty formations tend to have fewer managed pastures. Also, the land-
cover classifications are mutually exclusive; in cases where most of a catchment is classed as
pasture, there is little available to be classed as peatlands. The evolution of peatlands in the
form of blanket and raised bogs is typically attributed to high rainfall combined with poor
drainage. In the Irish context, it is therefore understandable that a high value of n(1+PEAT)
is often associated with a high value of nSAAR (r = 0.54), as well as with a low value of
n(1+PASTURE). The strong inverse correlation between n(1+PASTURE) and nSAAR
(r = -0.70) is consistent with drier areas being more readily put to pasture.

52
Volume III Hydrograph Analysis

Table 6.3: Correlation matrix of selected IVs and DVs at (up to) 89 stations

ARTDRAIN

FLATWET

PASTURE
STMFRQ

URBEXT
TAYSLO
NETLEN

FOREST
DRAIND
PCD

CGDIST

ALLUV
SAAPE
AREA

SAAR
FARL

PEAT
S1085
or variable

MSL

BFI
n AREA 1
n MSL 0.94 1 Row above uses abbreviated names; full name is in first column
n NETLEN 0.96 0.93 1
n STMFRQ 0.88 0.85 0.97 1 Red denotes Pearson correlation -0.6 or 0.6
n DRAIND -0.14 -0.04 0.15 0.32 1
Orange indicates other correlations significant at 1%
n CGDIST 0.83 0.89 0.80 0.71 -0.11 1
n FARL -0.24 -0.26 -0.28 -0.33 -0.14 -0.15 1 Green indicates correlations significant at 5%
n(1+ARTDRAIN) 0.22 0.22 0.21 0.16 -0.02 0.26 0.01 1
n S1085 -0.66 -0.65 -0.56 -0.44 0.36 -0.60 0.21 -0.18 1
n TAYSLO -0.25 -0.23 -0.22 -0.18 0.11 -0.23 0.07 -0.11 0.44 1
n FLATWET -0.03 0.05 0.03 0.10 0.23 0.03 -0.48 0.00 -0.13 -0.10 1
n SAAR -0.15 -0.06 0.02 0.19 0.59 -0.12 -0.41 -0.20 0.32 0.12 0.59 1
n SAAPE -0.02 -0.09 -0.01 0.01 0.03 -0.09 0.39 -0.02 0.28 0.10 -0.83 -0.24 1
n(1+URBEXT) -0.05 -0.10 -0.08 -0.08 -0.11 -0.02 -0.04 0.18 0.06 -0.01 -0.13 -0.08 0.15 1
n(1+FOREST) -0.25 -0.18 -0.14 -0.07 0.36 -0.21 0.04 -0.26 0.37 0.21 0.09 0.49 -0.05 -0.26 1
n(1+PEAT) -0.03 -0.02 0.01 0.06 0.14 -0.12 -0.22 -0.05 -0.04 -0.05 0.52 0.54 -0.43 -0.02 0.42 1
n(1+PASTURE) 0.11 0.10 0.04 -0.07 -0.23 0.18 0.27 0.16 -0.16 -0.12 -0.49 -0.70 0.30 0.06 -0.58 -0.84 1
n(1+ALLUV) 0.07 0.04 0.12 0.08 0.16 0.03 0.38 0.13 0.28 0.08 -0.58 -0.26 0.56 0.03 0.06 -0.41 0.40 1
n BFI 0.42 0.38 0.29 0.23 -0.45 0.38 -0.39 0.22 -0.49 -0.24 0.08 -0.27 -0.16 0.08 -0.48 -0.14 0.15 -0.19 1
n W75 0.33 0.32 0.24 0.22 -0.31 0.23 -0.64 -0.16 -0.49 -0.31 0.41 0.04 -0.40 -0.09 -0.31 0.07 -0.09 -0.48 0.63
n W50 0.32 0.31 0.22 0.20 -0.32 0.22 -0.63 -0.15 -0.48 -0.36 0.36 0.00 -0.35 -0.10 -0.32 0.01 -0.04 -0.44 0.67
n s -0.05 0.00 -0.04 -0.03 0.01 -0.06 0.14 -0.13 0.04 0.05 0.16 0.09 -0.22 -0.12 0.19 0.22 -0.15 -0.17 -0.36
n n -0.26 -0.22 -0.19 -0.16 0.22 -0.23 0.53 -0.09 0.35 0.26 -0.32 -0.02 0.29 -0.14 0.30 -0.04 0.03 0.32 -0.48
n Tr 0.24 0.26 0.16 0.16 -0.27 0.12 -0.43 -0.27 -0.42 -0.25 0.36 0.04 -0.39 -0.13 -0.20 0.10 -0.11 -0.47 0.47
n C 0.37 0.32 0.30 0.30 -0.22 0.32 -0.56 0.04 -0.40 -0.22 0.28 -0.01 -0.26 0.04 -0.32 -0.02 -0.04 -0.27 0.74

53
Volume III Hydrograph Analysis

(This page is intentionally blank.)

54
Volume III Hydrograph Analysis

AREA
AREA

AREA
AREA

n AREA
MSL

MSL
NETLEN NETLEN

NETLEN NETLEN

n MSL
MSL
DRAIND STRMFRQ MSL

DRAIND STRMFRQ

n NETLEN
DRAIND STRMFRQ
CGDIST STRMFRQ

n STMFRQ
DRAIND

CGDIST

n CGD ST

n n n n n
CGDIST
CGDIST
Width

Width
W75

W75

AREA MSL NETLEN STMFRQ CGDIST


Figure 6.1: Matrix plot of PCDs that in part represent catchment size (89 stations)

AREA MSL
AREA NETLEN
MSL STRMFRQ DRAIND DRAIND
NETLEN STRMFRQ CGDIST CGDIST
Width Width
Width
Width

[Editorial note: From Table 6.2, PASTURE is seen to be the dominant classification,W75with a W75
W75
W75

mean of 77.4% of catchment land-cover as opposed to 9.9% for PEAT and 12.7% to other
classifications. When all 134,000 FSU ungauged catchments are considered, the mean
values of PASTURE, PEAT and other are 57.3%, 23.8% and 18.9% respectively. The
AREA AREA
MSL MSL
NETLEN NETLEN STRMFRQ
STRMFRQ DRAIND DRAINDCGDIST CGDIST
Width Width
corresponding mean values for SAAR are 1069 mm for the 89 catchments studied and W751293 W75
mm for the 134,000 FSU ungauged catchments. Thus there appears to be a bias towards
gauging drier and more agriculturally productive catchments than is the Irish norm. This
emphasises the importance of the research undertaken to generalise methods of estimating the
characteristic hydrograph that take full account of catchment properties.]

The strong inverse correlation (r = -0.83) between nSAAPE and nFLATWET reflects that
these descriptors derive respectively from standardised estimates based on climate data of
potential evaporation and soil moisture. FLATWET is the proportion of the time for which
soils can be expected to be typically quite wet. FLATWET is expected to be greater in areas
where rainfall is high but the potential for evaporation (indexed by SAAPE) is relatively low.

[Editorial note: SAAR is the pre-eminent climatological descriptor in many hydrological


applications. The lack of correlation between nSAAR and the hydrograph-width variables
(see bottom six rows of Table 6.3) is striking.]

55
Volume III Hydrograph Analysis

The strong inverse correlation between nS1085 and nCGD ST (r = -0.60) is more
inscrutable. It may reflect that streams tend to be steeper on catchments where the drainage
pattern is fan-shaped (with a more compact catchment) than when the drainage pattern is
elongated (with a relatively large CGDIST).

Interestingly, the correlation between nS1085 and nTAYSLO is not especially high
(r = 0.44), indicating that these are rather different measures of mainstream channel slope.

6.5.2 Individual correlations between DVs and initially selected IVs


The correlations between each of the six DVs and the 19 initially selected IVs are shown in
the bottom rows of Table 6.3. There are two stand-out sets of associations:

nW75 and nW50 are highly negatively correlated with nFARL;


nW75, nW50 and nC are highly correlated with nBFI.

These are consistent with physical interpretations that:

Lakes and other water bodies (consistent with a smaller value of nFARL) attenuate
the flood passing down the river system, tending to lead to hydrographs that are more
prolonged than otherwise (consistent with larger values of nW75 and nW50);
Catchments that are relatively permeable (consistent with a larger value of nBFI)
tend to lead to hydrographs that are more prolonged than otherwise (consistent with
larger values of nW75 and nW50).

6.5.3 Choosing a subset of PCDs to use as IVs


In view of the strong correlations (see Section 6.5.1) amongst various groups of PCDs, it was
decided that a subset would suffice for the main modelling. For reasons discussed in
Section 6.3, the main regression study would use no more than about 12 IVs.

The most suitable IVs in each of three competitive groups were chosen by favouring those
most strongly correlated with the six DVs under study. The relevant correlations are shown
in the bottom six rows of Table 6.3.

From inspection, it can be seen that:

The magnitudes of the correlations associated with nAREA are generally a little
larger than those associated with nMSL, nNETLEN, nSTMFRQ or nCGDIST;
The magnitudes of the correlations associated with n(1+PASTURE) and
n(1+PEAT) are larger than those associated with nSAAR;
The magnitudes of the correlations associated with nFLATWET are evenly
balanced with those associated with nSAAPE.

Accordingly, nAREA, n(1+PASTURE) and nFLATWET were selected as the


representative variables, and the seven other PCDs discarded. This achieved the target of
allowing no more than 12 IVs in the regression modelling.

56
Volume III Hydrograph Analysis

6.5.4 Checking the Normality of the DVs


The log-transforms yield DVs that are acceptably Normal, satisfying one of the requirements
for linear regression (see Section 6.1.2). The Gamma shape parameter n is somewhat less
convincing than the others in this respect (see Figure 6.2).
(if Normally distributed )

nW75 nW50
Expected value

Observed value Observed value


(if Normally distributed )

n s n n
Expected value

Observed value Observed value

nTr n C
(if Normally distributed )
Expected value

Observed value Observed value

Figure 6.2: Normality plots of log-transformed width descriptors and model parameters

Reading across the 4th-last row of Table 6.3, the hydrograph width descriptor n s is seen to
be no more than weakly correlated with the available IVs. Although an attempt was made to
model this index of hydrograph skewness, the best that could ultimately be achieved was to
adopt a fixed value of s = 0.40. From Table 6.1 it is seen that this is very close to both the
arithmetic and geometric means of s, i.e. to the arithmetic means of s and n s. It proved
possible to develop useful regression models for the other five DVs.

57
Volume III Hydrograph Analysis

6.5.5 Final selection of the independent variables; a note on the use of BFI
The 12 variables finally selected as IVs were (in no particular order):

nAREA, nDRAIND, nFARL and n(1+ARTDRAIN);


nS1085, nTAYSLO, nFLATWET and n(1+URBEXT);
n(1+FOREST), n(1+PASTURE), n(1+ALLUV) and nBFI.

BFI is not strictly a PCD. Rather, it is a hydrological index derived by the analysis of daily
mean flow data. At the time of the hydrograph width research, the mechanism by which BFI
would be modelled at ungauged sites was unclear. Accordingly, two sets of generalisations
of the hydrograph width variables W75 and W50 and the UPO-ERR-Gamma model parameters
n, Tr and C were developed according to whether a BFI value is/isnt available.

6.6 The regression method used


Various forms of multiple regression are possible. That adopted here was stepwise linear
regression. In stepwise regression, IVs are introduced into the regression model one variable
at a time. At each step, the IV offering the greatest reduction in the objective function (least-
squares or weighted least-squares) is added into the model. But IVs already included in the
model may also be deleted at any step where they no longer contribute significantly to
regression (Tabachnick and Fidell, 2001). The stepwise method results in a parsimonious
model i.e. one frugal in its use of parameters. The default settings in SPSS were used in
respect of the statistical criteria for including or removing an IV. Accordingly, the thresholds
used were 0.05 (5% significance) for inclusion of an IV and 0.10 (10% significance) for
removal of an IV, the F-test statistic being used to assess the significance of the departure
from the null hypothesis (that there is no linear relationship between the IV and the DV).

[Editorial note: For datasets of the moderate size considered here, computer power is
typically such that best subsets regression rather than stepwise regression can be used.
Indeed, the best subsets approach is adopted for the Volume II modelling of QMED. Many
researchers feel more comfortable with the stepwise approach, believing it to provide more
defined safeguards against over-determination of models. An over-determined model is one
which fits too many free parameters in relation to the number of data samples available. In
many cases, the different techniques lead to the same final model.]

6.7 Illustrative results: Estimating W75 when BFI available


Generalisation of the hydrograph width descriptor W75 provides an insight into the methods.
The regression is based on linear least-squares, with nW75 taken as the DV. The stepwise
regression analysis is summarised in Table 6.4, which reports intermediate results for 1, 2, 3,
and 4-variable models as well as the final 5-variable model.

6.7.1 Regression models and their performance evaluation


r2 indicates the proportion of variation explained by the regression model (i.e. by the IVs). It
is often termed the coefficient of determination. There is progressive improvement in r2 as
additional IVs are allowed into the model. The final 5-variable model explains about 72% of
the variability in nW75. Explaining this much of the variability is regarded as quite good.

58
Volume III Hydrograph Analysis

It is inevitable that r2 increases as additional IVs are allowed into the model. By this
criterion, a model using all 12 available IVs would be judged best.

To avoid over-determination, it is necessary to examine the adjusted r2. The adjustment takes
into account the number of degrees of freedom consumed by having to fit additional
parameters in the linear least-squares regression. The adjusted r2 provides a reasonable
estimate of how well the regression model might be expected to fit another dataset drawn
from the same population. The adjusted r2 value of 0.701 indicates that the 5-variable model
provides a reasonably good fit.

Table 6.4: Stepwise regression results for modelling hydrograph width descriptor nW75
Factorial
Standard
No. standard
No. of IV Adjusted error of
of IV added r2 2 error of
catchments removed r estimate
IVs estimate
(SEE)
(FSE)
1 77 nBF .399 .391 .788 2.20
2 77 n(1+ALLUV) .575 .564 .668 1.95
3 77 nFARL .643 .629 .616 1.85
4 77 n(1+ARTDRA N) .686 .669 .581 1.79
5 77 nS1085 .721 .701 .552 1.74

The standard error of estimate (SEE) is the standard deviation of the residuals or errors in
predictions by the model: in this case, in the estimates of nW75. This means that on the
assumption that the model residuals are normally distributed the actual value of nW75 is
expected to lie within 0.552 of the predicted value in about 68% of cases.

Because our principal interest is in estimating W75 rather than nW75, the factorial standard
error (FSE) is generally more relevant. The FSE is just the exponential of the SEE. For the
final (5-variable) model, the FSE is e0.552 = 1.74. This means that (in about 68% of cases) the
actual value of W75 is expected to lie within the factorial range 0.57 75 to 1.74 75 where
75 is the estimated value of the width descriptor W75 obtained from the regression model
and 0.57 is the reciprocal of 1.74. This confidence interval is considered reasonably good.

For the scenario when BFI is available, the recommended model is:

nW75 = 3.548 + 1.861 nBFI 12.199 n(1+ALLUV) 3.946 nFARL


6.1
3.324 n(1+ARTDRA N) 0.246 nS1085

Exponentiating both sides of the equation, the model can be written:

W75 = 34.74 BFI1.86 (1+ALLUV)-12.20 FARL-3.95 (1+ARTDRAIN)-3.32 S1085-0.25 6.2

Table 6.5 confirms that all coefficients are significant at the 0.05 level (t statistic> 1.96).
The standardised coefficients ( in the table) highlight the relative contribution of each
descriptor to explaining the variation in nW75. nBF is seen to be the most important
predictor. This vindicates the decision to develop models for the case when this variable is
available.

59
Volume III Hydrograph Analysis

Table 6.5: Coefficient and collinearity statistics for selected model for nW75
Variance
Standard
t inflation
Term/regressor Coefficient error of Tolerance
value statistic factor
coefficient
(VIF)
Constant 3.548 0.68 5.20
nBFI 1.861 0.34 0.43 5.53 0.660 1.52
n(1+ALLUV) -12.199 3.82 -0.23 -3.19 0.767 1.30
nFARL -3.946 0.94 -0.30 -4.20 0.751 1.33
n(1+ARTDRAIN) -3.324 0.92 -0.24 -3.62 0.903 1.11
nS1085 -0.246 0.08 -0.22 -2.95 0.708 1.41

6.7.2 Checking the possible influence of collinearity


In order to assess the possible impact of collinearity (see Box 6.1), the variance inflation
factor (VIF) was also studied. This statistic is the reciprocal of the tolerance, which in turn
denotes the proportion of the variance in a given catchment descriptor that cannot be
explained by the other regressors.

High VIF values (i.e. small tolerances) indicate that a large amount of the variance in one
regressor can be explained by the other regressors. VIF thus indexes the impact of
collinearity (amongst the regressors) on the stability of the multiple regression model. VIF
values are (by definition) greater than or equal to 1. Whilst only a guide, VIF values greater
than 10 are often regarded as indicating serious problems of collinearity. In weaker models,
values above 2.5 may sometimes be a cause for concern.

It is seen from the final column of Table 2.7 that VIF is less than 1.6 for all regressors.
Collinearity is therefore judged not to be a problem with the selected model.

6.7.3 Checking the logical consistency of the model


Logical consistency is often the overriding factor in the final choice of a regression model. Is
the model consistent with what we know about catchment flood behaviour? The 5-variable
model for nW75 appears credible in this respect:

W75 increases with permeability and storage (indicated by a larger value of BFI);
W75 decreases with larger values of FARL (a larger value of FARL is associated with
reduced attenuation of flood water by storage elements such as lakes and reservoirs;
flood hydrographs in such a catchment can be expected to have a higher peak and a
narrower width);
W75 decreases with larger values of ARTDRAIN, consistent with a flashier response
after arterial drainage works (i.e. a flood hydrograph with a higher peak and a
narrower width);
W75 decreases with larger values of S1085, indicating a flashier response from steeper
catchments;
W75 decreases when the proportion of land-cover classed as alluvial (ALLUV) is
larger (more alluvium in a catchment indicates a lower potential of the catchment for

60
Volume III Hydrograph Analysis

accepting rainfall; the resulting flood response is therefore expected to be faster, with
the flood hydrograph expected to be narrower than that in a catchment having less
alluvium).

The last feature is not wholly convincing, because of the high magnitude of the exponent of
1+ALLUV in the Equation 6.2 model for W75. The role of ALLUV in the hydrograph-width
models is discussed further in Box 6.2.

Box 6.2: The role of ALLUV in the hydrograph-width models


Editorial note: Discussion is warranted of an uncomfortable feature of the Equation 6.2
model for W75. This is the high magnitude (-12.20) of the exponent of 1+ALLUV.
ALLUV denotes the proportion of the catchment classed as alluvium in the Teagasc
classification of soils (see Volume IV).

The effect might conceivably reflect that alluvial areas lie close to the river network and
tend to contribute runoff quickly. However, the attribution of physical effects to results
obtained by regression is often hazardous.

From Table 6.2 it is seen that alluvial fractions for the 89 catchments studied range from
0.00 to 0.10. Thus, the modelled factorial effect of the alluvium on W75 ranges from
1.00-12.20 = 1.000 (when ALLUV = 0.00) to 1.10-12.20 = 0.313 (when ALLUV = 0.10).
This may just be reasonable.

PCDs for ungauged sites were not available at the time of the hydrograph width research.
98.3% of the 134,000 ungauged catchments supported by the FSU have a value of
ALLUV within the range 0.00 to 0.10. However, 0.25% of catchments are classified as
having ALLUV 0.20. For these cases, the W75 model (Equation 6.2) implies that
classification of land as alluvium reduces W75 by a factor of more than 9, since
(1+0.20)-12.20 = 0.108 1/9. This does not seem reasonable.

The urban fraction (URBEXT) is the only classification of land cover to play a major role
in the Volume II procedure for estimating the T-year flood peak at an ungauged site. It is
notable that ALLUV plays no role. Thus, an ungauged catchment classified as having an
unusually large alluvial fraction will be modelled as generating hydrographs that are
exceedingly narrow but not especially high-peaked. This appears physically unrealistic.
Should the catchment under study be mapped as having an unusually large alluvial
fraction, it will be prudent for the user to make special checks.

The Equation 6.2 model for W75 has not in fact been implemented. The models
implemented through the FSU Web Portal are shown later in Table 6.7. Only one of these
the model for estimating Tr includes the 1+ALLUV term. Moreover, the exponent of
1+ALLUV in that model is somewhat less severe (-8.83) than for the Equation 6.2 model.
Users are nevertheless to be encouraged to make special checks should they be using the
model to estimate Tr on a catchment for which ALLUV0.20.

The role of ALLUV in the hydrograph-width models may warrant further exploration.

61
Volume III Hydrograph Analysis

6.7.4 Additional checks


Additional checks were made on the sensitivity of the 5-variable model to the inclusion of
particular catchments in the calibration. The statistics examined include a leverage statistic
(to identify catchments whose observed values of the selected DV influence the regression
model more than others) and Cooks distance (which measures how much the model
coefficients change when a particular catchment is dropped from the analysis). Some weak
evidence was found that Station 26021 Inny at Ballymahon might have undue influence.
However, rules of thumb indicated that this was not so marked as to warrant its exclusion.

The model residuals (i.e. prediction errors) were tested for normality. The probability-
probability (P-P) plot shown in Figure 6.3 shows some departure from a perfect 1:1 line but
not enough to judge the model inadequate.
(if Normally distributed)
Expected value

Observed cumulative probability


Figure 6.3: Normality plot of standardised residuals for 5-variable model for nW75

Model residuals were also tested for homoscedasticity. This requires that the standard
deviations of errors of prediction are approximately equal for all predicted values of the DV.
Homescedastic derives from the Greek for equal scatter. Homoscedasticity is exhibited
when the plot of residuals displays a cloud of dots and the band enclosing the residuals is
approximately equal in width at all values of the predicted DV.

A lack of homoscedasticity (i.e. heteroscedasticity) is characterised by a pattern such as a


funnel shape, indicating greater errors associated with larger predicted values. This may arise
when there is an interaction between an IV and a variable not in the regression model, or
when some IVs are skewed while others are not. No serious violation of the assumption of
homoscedasticity is evident in the Figure 6.4 plot of standardised residuals against the
standardised predicted values. The 5-variable model for nW75 is therefore considered
acceptable.

62
Volume III Hydrograph Analysis

Regression standardised
predicted value

Regression standardised residual


Figure 6.4: Plot of standardised residuals for 5-variable model for nW75

63
Volume III Hydrograph Analysis

(This page is intentionally blank.)

64
Volume III Hydrograph Analysis

(This page is intentionally blank.)

65
Volume III Hydrograph Analysis

Table 6.6: Recommended models when BFI available


Hydrograph # of Adj. Stations identified as possible Eqn
FSE Recommended model
descriptor stations r2 outliers* #

W75 77 0.701 1.74 26021 W75 = 34.74 BFI1.86 FARL-3.95 (1+ALLUV)-12.20 6.3
(1+ARTDRAIN)-3.32 (S1085/1000)-0.25
W50 75 0.735 1.70 None W50 = 63.05 BFI2.11 FARL-4.55 (1+ALLUV)-10.24
6.4
(1+ARTDRAIN)-3.17 (S1085/1000)-0.25
06011, 11001, 14007, 16005,
n 79 0.377 2.02 n = 1 + 2.90 BFI-1.12 FARL4.37 6.5
18005, 25027, 34018

Tr 79 0.493 1.77 26021 Tr = 54.98 BFI1.32 (1+ALLUV)-13.08 (1+ARTDRAIN)-3.70 (S1085/1000)-0.20 6.6


C 77 0.630 2.16 16005 C = 310.75 BFI3.44 FARL-4.88 6.7
*Assignment based on a test of
standardised deleted residuals
Table 6.7: Recommended models when BFI unavailable
Hydrograph # of Adj. Stations identified as possible
FSE Recommended model Eqn #
descriptor stations r2 outliers*

W75 87 0.675 1.79 15002, 35071 W75 = 31.28 DRAIND-0.88 FARL-5.85 (1+FOREST)-2.86
6.8
(1+ARTDRAIN)-2.92 FLATWET3.12 (S1085/1000)-0.27
W50 84 0.675 1.80 35071 W50 = 34.02 DRAIND-0.95 FARL-6.59 (1+FOREST)-2.83
6.9
(1+ARTDRAIN)-2.60 FLATWET2.44 (S1085/1000)-0.29
n 89 0.412 1.99 06011, 11001, 14007, 34018 n = 1 + 4.78 DRAIND0.63 FARL5.46 (1 + FOREST)2.46 6.10

Tr 89 0.213 2.03 26021, 30061 Tr = 13.85 DRAIND-0.48 FARL-2.54 (1+ARTDRAIN)-3.16


6.11
(S1085/1000)-0.25 (1+ALLUV)-8.83
C 87 0.485 2.42 16008, 23002, 35071 C = 11.78 DRAIND-0.97 FARL-7.65 (1+FOREST)-3.70 AREA0.26 6.12
*Assignment based on a test of
standardised deleted residuals

66
Volume III Hydrograph Analysis

6.8 Recommended models for use at ungauged sites

6.8.1 The final models


The techniques of Sections 6.6 and 6.7 were applied to develop models for the hydrograph
width measures/parameters: W75, W50, n, Tr and C. The recommended models are presented
in Table 6.6 (for the case when BFI is available) and Table 6.7 (for the case when BFI is
unavailable). These models (see tables on previous page) are now jointly discussed:

When judged by the factorial standard error (FSE), the models using BFI (Table 6.6)
generally outperform those that do not (Table 6.7). This shows the value of BFI in
helping to model hydrograph shape. (Precise comparisons are inhibited because of
the somewhat different datasets used in Table 6.6 and Table 6.7. BFI values were
available for only 79 of the 89 stations.)
The likely importance of storage attenuation and the effectiveness of FARL in its
characterisation are highlighted by the appearance of the descriptor in all but one of
the ten models.
Mainstream slope S1085 appears in the models of the three main characteristic times
W75, W50 and Tr. The exponent of S1085 is reassuringly similar in all cases.
[Editorial note: For a reason known only to the research contractor, values of S1085
were divided by 1000 prior to the regression modelling of hydrograph widths. This
explains why the term S1085/1000 appears in the models for W75, W50 and Tr shown
in Table 6.6 and Table 6.7. This led to difficulties for the contractor testing IBIDEM
at two ungauged sites (see Sections 9.7.4 and 9.7.5) but should have no impact for the
user.]
When BFI is unavailable, the importance of FARL is heightened and DRAIND also
becomes indispensable. Both descriptors appear in all five models. FOREST also
then proves helpful (appearing in four of the models). The exponents of DRAIND
and FOREST indicate that hydrographs are narrower when drainage and forest cover
are more extensive. This likely reflects the faster conveyance of water, noting that
afforestation is often accompanied by drainage works such as ditching.
There is reassuring consistency in the models derived for W75 and W50. The same
terms appear in both models. Moreover, the exponents are broadly comparable.
Except for parameter C, the FSEs are in the range 1.7 to 2.0. This is a moderately
good performance by hydrological standards. [Editorial note: The excellent
performance achieved in estimating QMED on rural catchments (FSE = 1.37, see
Volume II) suggests that the task of modelling typical peak flows on an ungauged
catchment is rather easier than that of modelling typical hydrograph widths. This
likely reflects the highly attenuated nature of many flood hydrographs in Ireland. It
may also reflect the varied duration and often complex pattern of rainfalls that give
rise to flooding.]
The term least well-estimated is parameter C of the UPO-ERR-Gamma model. This
determines the rate of decay of the exponential replacement recession. The poor
performance in modelling C is neither surprising nor especially worrying. The
parameter has no influence on the peak segment of the hydrograph.
The model for C in the BFI unavailable case is the only occasion on which a PCD
indicative of catchment size appears in the hydrograph width modelling. It is also
notable that general wetness indexed by SAAR does not appear in any model.
67
Volume III Hydrograph Analysis

However, the wetness descriptor FLATWET proves useful in modelling W75 and W50
in the BFI unavailable case.
The analyses identify a number of stations that are possible outliers. Those implicated
in estimation of the parameters of the UPO-ERR-Gamma model may not warrant
much concern. It is inevitable that the limited family of hydrograph shapes supported
by the UPO-ERR-Gamma model (see Figure 4.1, Figure 4.2 and Figure 4.4) will not
suit all stations.
The sites implicated as possible outliers in the modelling of W75 and W50 are Stations
15002, 26021and 35071. These are all Grade A2 stations. Station 26021 had by far
the largest proportion of missing flow data of any of the 89 stations analysed (see
Table A.2). The poor performance of the PCD-based models at Stations 15002 and
35071 may warrant further investigation. The characteristic hydrographs at these
stations (see Figure 6.5) are relatively well defined and, in the case of Station 15002
Nore at Johns Bridge, rather finely shaped.

100 Median hydrograph


90 (a) Station 15002 Nore at Johns Bridge
Percentage of peak flow

80
70
60
50
40
30
20
10
0
-50 -37.5 -25 -12.5 0 50 100 150 200
Time in hours (relative to time of peak flow)

100 Median hydrograph


90 (b) Station 35071 Lareen at L. Melvin
Percentage of peak flow

80
70
60
50
40
30
20
10
0
-150 -112.5 -75 -37.5 0 75 150 225 300
Time in hours (relative to time of peak flow)
Figure 6.5: Median hydrographs at: (a) Station 15002 and (b) Station 35071

68
Volume III Hydrograph Analysis

6.8.2 Model performance


The quality of performance achieved by the regression models in the BFI unavailable case is
encapsulated in Figure 6.6. The plots show observed and predicted values of the width
descriptors (W75 and W50) and the UPO-ERR-Gamma parameters (n, Tr and C) for the 89-
station dataset. The supporting data are presented in Table D.1 of Appendix D.

W75 (hours) W50 (hours)

FSE = 1.79 FSE = 1.80

FSE = 1.99

Tr (hours) C (hours)

FSE = 2.03 FSE = 2.42

Figure 6.6: Derived and modelled values of W75, W50, n, Tr and C (BFI unavailable case)

Note the logarithmic scale of the plots. The factorial standard error (FSE) provides a one-
number summary of the performance achieved. While many values are predicted within a
factor of two, some for the C parameter are seen to be in error by up to a factor of ten.

69
Volume III Hydrograph Analysis

It is confirmed that the generalisation for application at ungauged sites is more successful for
the width descriptors W75 and W50 than for parameters of the UPO-ERR-Gamma model.

It is concluded that, for the BFI unavailable case, the UPO-ERR-Gamma model may not be
sufficiently reliable to construct the characteristic flood hydrograph at ungauged sites. An
alternative is to use the parabolic curves method (see Section 8.7) with the width descriptors
W75 and W50 estimated by Equations 6.8 and 6.9. This synthesises the upper hydrograph
only.

In view of the better performance of the regression models derived for the BFI available case,
consideration might be given to using any or all of the models from Table 6.6 at ungauged
sites by substituting BFIsoil for BFI. This possibility is discussed in Section 8.6.

6.8.3 Additional notes on the regression models


The Gamma shape parameter n is dimensionless. Both FARL and BFI are themselves
dimensionless. So it is encouraging that the regression model is dimensionally correct in the
BFI available case (Equation 6.5).

The other four hydrograph descriptors (W75, W50, Tr and C) are all characteristic times in
hours. None of the PCDs available to the generalisation research had a dimension involving
time. Thus, it is not possible for the regression models to be strictly dimensionally correct.

70
Volume III Hydrograph Analysis

7 Ancillary investigations
The research checked for systematic variations in the shape of the flood hydrograph with:

The magnitude of the peak flow;


The pre-event minimum flow;
Season;
Arterial drainage in the catchment.

7.1 Variation of hydrograph width with peak flow


The variation of hydrograph width with peak flow was checked for all 89 stations studied.
Figure 7.1 presents a typical outcome for Station 07009 Boyne at Navan Weir. There is no
particular trend evident towards wider or narrower hydrographs according to the magnitude
of the flood peak.

[Editorial note: The broken lines connect widths for different events. This is to help visual
detection of any underlying trend. Formal tests for linear trend were also undertaken using
ordinary least-squares regression.].

Labels mark % of peak flow at which width abstracted


Width of exceedance (hours)

Peak flow (m3s-1)

Figure 7.1: Variation of hydrograph width with peak flow at Station 07009

For a few stations (e.g. Station 07010 Blackwater at Liscartan shown in Figure 7.2), the
graphs display generally positive slopes. However, for most other stations, the parts of the
graphs corresponding to the two or three highest recorded floods actually show negative
slopes, in contrast to the positive slopes for the points corresponding to the floods of lower
magnitudes. Some examples of such inconsistently varying slopes are shown in Figure 7.3.

A composite trend-line of the form y = mx + c was fitted to the plots of hydrograph width
against the magnitude of the peak flow, y being the total width (in hours), x the peak flow (in
m3s-1), m the slope and c the intercept. At each station in turn, trends were examined across
all hydrograph widths collectively, and for the reference widths W75 and W50 individually.

71
Volume III Hydrograph Analysis

Width of exceedance (hours)

Labels mark % of peak flow at which width abstracted

Peak flow (m3s-1)


Figure 7.2: Variation of hydrograph width with peak flow at Station 07010

Labels mark % of peak flow at which width abstracted


Station 06026 Lagan-Glyde at Aclint Station 09001 Ryewater at Leixlip
Width of exceedance (hours)

Peak flow (m3s-1) Peak flow (m3s-1)

Station 14004 Figile at Clonbulloge Station 14011 Slate at Rathangan


Width of exceedance (hours)

Peak flow (m3s-1) Peak flow (m3s-1)

Station 24013 Deel at Rathkeale Station 36015 Finn at Anlore


Width of exceedance (hours)

Peak flow (m3s-1) Peak flow (m3s-1)

Figure 7.3: Patterns of variation of hydrograph width with peak flow (at six stations)
72
Volume III Hydrograph Analysis

Checks for trend were extensive but generally inconclusive. Figure 7.4 shows the slope of
the trend in W75 for the 37 Grade A1 stations. The stations are arranged by catchment size,
with the smallest (Station 34018) on the left and the largest (Station 25017) on the right.
While there is some variation, no marked pattern with catchment size is evident.
Slope m of the trend-line Y=mX+c fitting the available total w idths at 75 percentile of selected flood events plotted against the respective peak
flow s for each station
20 9000
9000
Slope of trend-line

15 Catchment area (km2) 8000


8000
s )

10 7000
7000
3 -1

Catchment area
5 6000
6000
(hours/m

(km 2)
0 5000
5000
Slope m

6026

6012

9001

6011

6014

6013

7007

7010

7009

7012
34018

14007

25027

29001

25025

14011

25014

36015

30005

26019

14004

25030

26008

29011

15005

25003

24013

26012

27002

23002

30004

36010

14006

26007

25006

14018

25017
-5 4000
4000
-10 3000
3000
-15 Station # 2000
2000
-20 1000
1000
-25 00
Figure 7.4: Slope of W75 trend with peak flowStation no.
magnitude (Grade A1 stations)

Although the pattern of trend was not generalised, some weak support was found to suggest
that hydrographs may tend to become narrower in larger events on steep catchments (e.g.
high S1085) with fewer lakes (i.e. high FARL) than otherwise. The opposing slopes of the
trends at Stations 06011 (Fane at Moyles Mill) and 06012 (Fane at Clarebane) evident in
Figure 7.4 were explored. The reason for the marked difference is unclear, although the
mainstream slope (S1085) is indeed much lower at the downstream site (see Table 7.1),
suggesting that the intervening area is rather flat.

Table 7.1: Some leading PCDs of stations on the River Fane


AREA S1085 FARL
Station # Station name Location
km2 m/km

06012 Fane at Clarebane Upstream 162.8 0.0052 0.874


06011 Fane at Moyles Mill Downstream 229.2 0.0027 0.831

7.2 Variation of hydrograph width with pre-event minimum flow


The variation of hydrograph width with peak-event minimum flow was checked for all 89
stations. For the purpose of HWA, the pre-event minimum flow is taken as the minimum
flow on the rising side of the flood hydrograph that lies within the time-window defined in
Section 2.2.

Figure 7.5 presents the outcome for Station 07009 Boyne at Navan Weir. No particular trend
is evident towards wider or narrower hydrographs according to the magnitude of the pre-
event flow. No systematic trend or pattern in the variation of hydrograph widths with pre-
event minimum flow could be identified across the 89 stations.

Although no pattern of trend could be generalised, some weak support was found to suggest
that hydrographs may tend to become wider in sparser river networks (i.e. with lower values
of DRAIND) or more permeable catchments (i.e. with higher values of BFI) when the pre-
event minimum flow is high than when it is low. A more obscure effect was a possible
tendency for hydrographs to become narrower when the pre-event minimum flow is high
in catchments where the alluvial fraction (ALLUV) is higher (see Figure 7.6). In the figure,

73
Volume III Hydrograph Analysis

the 89 Grade A1+A2 stations are ordered according to the fraction of the catchment mapped
as alluvium. See also the discussion of ALLUV in Section 6.7.3.

Labels mark % of peak flow at which width abstracted


Width of exceedance (hours)

Pre-event minimum flow (m3s-1)


Figure 7.5: Variation of hydrograph width with pre-event minimum flow (Station 07009)
Slope of trend-line

12 ALLUV 0.16
16
10
ALLUV
TREND_SLOPEW75_QPreMin 0.14
(hours/m3s-1)

14
TREND_SLOPEW75_QPreMin

8
0.12
12
6
4 0.10
10

ALLUV
2 0.08
8
0
0.06
6
-2
-4
0.04
4

-6 0.02
2

-8 0.00
0
29004
35001
14004
6012
26022
6011
35071
27002
30005
26009
34009
16001
25025
7002
7004
7033
29011
30007
26021
7011
29001
30004
34001
27001
26002
26007
7001
26005
36011
18005
35005
15003
36015
26019
22071
16002
39009
7010
36019
36010
23012
7012
26008
7007
35002
7009
23002
16004
25030
14006
7006
14011
15005
14009
36027
15002
6014
15006
11001
16008
25014
16009
6026
25016
16005
9001
6013
24013
15001
14018
23001
25006
19001
18004
16003
25027
25029
25001
24008
25003
24001
24082
25005
14007
Station No.
Station #
Figure 7.6: Slope of W75 trend with pre-event minimum flow (Grade A1+A2 stations)

7.3 Variation of hydrograph width with time of year


The variation of hydrograph width with time of year was checked for all 89 stations using
both regular and circular plots. The seasonal distribution of floods at Station 07009 Boyne at
Navan Weir is illustrated in Figure 7.7. It can be seen that most floods occur in the winter
half-year from October to March, with the three largest in November to January. However,
floods sometimes occur in the summer half-year.

The circular plot of Figure 7.8 indicates that wider hydrographs at this station are generally
associated with the winter half-year (right half of diagram).

74
Volume III Hydrograph Analysis

Peak flow (m3s-1)

Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep

Figure 7.7: Plot of flood peak against time of year (Station 07009 Boyne at Navan Weir)

Radial scale marks Station 07009 Boyne


hydrograph width in hours at Navan Weir

Figure 7.8: Circular plot of W50, W75 and W90 against time of year (floods at Station 07009)

However, a systematic pattern could not be established across the 89 stations studied. In
essence, the widest flood hydrographs tend to be in winter but this is in any case the season
where large floods are most prevalent.

7.4 Effect of arterial drainage on hydrograph widths


According to Table A.2, catchments associated with 19 of the 89 stations have undergone
arterial drainage. The drainage schemes were carried out over different periods for different
catchments. Of the 19 stations, 13 are Grade A1 and six are Grade A2. Only data for the
post-drainage periods have been used in the main HWA reported earlier. An ancillary study
investigated the effect of arterial drainage on the shape of the flood hydrographs of the
affected stations.

75
Volume III Hydrograph Analysis

Hydrograph data for the pre-drainage periods of five of the 13 Grade A1 stations were used.
The stations selected and periods considered are listed in Table 7.2, together with gauged
values of the index flood, QMED.

It is seen that the effect of arterial drainage is to increase QMED appreciably. This is to be
expected in the post-drainage period because of the improved conveyance characteristics of
the channel network. The one exception highlighted in Table 7.2 is that QMED is virtually
unchanged for Station 23002.

[Editorial note: Judging from the relevant PCDs, arterial drainage on the Feale is not at all
extensive (ARTDRAIN = 0.001 and ARTDRAIN2 = 0.002). The results for this station
therefore appear irrelevant to exploring the effect of arterial drainage on hydrograph widths,
and further discussion is omitted. As noted in Volume II, Catchment 23002 generates some
of the largest floods ever gauged in Ireland. The Feale has been studied by Martin et al.
(2000) amongst others.]

Table 7.2: Stations studied for the effect of arterial drainage on hydrograph widths
Pre-drainage Post-drainage
# QMED # events # QMED # events
Station Period Period
years (m3s-1) used years (m3s-1) used

01/10/1969 09/04/1979
07007 Boyne
to 3 23.72 7* to 25 35.41 25
at Aqueduct
30/09/1972 05/04/2004
07010 14/11/1952 08/12/1986
Blackwater to 29 50.40 29 to 17 69.61 17
at Liscartan 30/09/1981 21/05/2003
18/10/1946 01/10/1959
23002 Feale
to 12 373.74 12 to 46 371.84 46
at Listowel
30/09/1958 06/01/2006
01/10/1957 01/01/1991
26012 Boyle
to 24 36.95 24 to 11 46.72 11
at Tinacarra
30/09/1981 22/01/2002
04/08/1951 01/10/1964
30004 Clare
to 6 43.54 14** to 41 89.83 41
at Corrofin
30/09/1957 01/10/2005
* Event 4 discarded at Station 07007
** Events 2 and 9 discarded at Station 30004

The procedures recommended in Chapters 3 and 4 were applied to derive characteristic


hydrographs for flood hydrographs drawn from the pre-drainage period. Resulting values of
the width descriptors (W75 and W50) and UPO-ERR-Gamma model parameters (n, Tr and C)
are shown in Table 7.3, together with the corresponding results for the post-drainage period
(taken from HWA results and their estimates from PCDs

Table D.1).

76
Volume III Hydrograph Analysis

Table 7.3: Pre- and post-drainage values of hydrograph width descriptors/parameters


Pre-drainage Post-drainage
Station W75 W50 n Tr C W75 W50 n Tr C

07007 67.00 N/A 2.20 48.56 70.61 26.22 52.53 3.74 26.84 69.88
07010 76.77 158.23 1.85 44.96 217.82 30.42 119.06 3.46 25.78 222.32
26012 287.76 N/A 1.72 182.25 211.32 177.27 285.94 3.85 195.50 257.50
30004 100.49 213.04 2.76 110.75 140.97 44.33 68.49 5.30 59.44 46.42

The resulting characteristic hydrographs are shown in Figure 7.9. Note that the pre-drainage
and post-drainage hydrographs are drawn to the same timescale but that this differs from
station to station.

Generally, the width descriptors are smaller in the post-drainage period, indicating narrower
flood hydrographs. This corresponds to a quicker passage of flood flow and is consistent
with the expected impact of arterial drainage, which Bhattarai and OConnor (2004)
summarise as having the objective of achieving a reduction of the extent and duration of
flooding, by inducing faster runoff in the river channel, with higher peaks and shorter
recessions of the discharge hydrograph.

Values of the Gamma shape parameter n are found to increase in all cases after arterial
drainage. As illustrated in Figure 4.1, higher values of n correspond to a peakier hydrograph.
Values of the Gamma rise-time parameter Tr are expected to be shorter after drainage. This
is very much the finding at Stations 07007, 07010 and 30004. However, the post-drainage
period yields a somewhat larger value of Tr at Station 26012. [Editorial note: Reference to
part (c) of Figure 7.9 indicates that the increased value of Tr at Station 26012 is a product of
parameter interaction. The pre-drainage value n = 1.72 leads to an abrupt start to the
modelled hydrograph whereas the post-drainage value n = 3.85 provides a gentler (and
consequently earlier) start. The value of Tr has changed in compensation.]

The research confirms that the effect of arterial drainage is generally to yield peakier
hydrographs of reduced width. The swifter response ties in with the basic requirements of
drainage schemes.

77
Volume III Hydrograph Analysis

Pre-drainage period Post-drainage period

(a) Station 07007 Boyne at Aqueduct

-50 -25 0 56.25 112.5 168.75 225 -50 -25 0 56.25 112.5 168.75 225
Time in hours relative to time of peak flow Time in hours relative to time of peak flow

(b) Station 07010 Blackwater at Liscartan

-75 -37.5 0 56.25 112.5 168.75 225 -75 -37.5 0 56.25 112.5 168.75 225
Time in hours relative to time of peak flow Time in hours relative to time of peak flow

(c) Station 26012 Boyle at Tinacarra

-175 -87.5 0 125 250 375 500 -175 -87.5 0 125 250 375 500
Time in hours relative to time of peak flow Time in hours relative to time of peak flow

(d) Station 30004 Clare at Corrofin

-125 -62.5 0 87.5 175 -125 -62.5 0 87.5 175


Time in hours relative to time of peak flow Time in hours relative to time of peak flow

Figure 7.9: Characteristic hydrographs for four sites affected by arterial drainage
78
Volume III Hydrograph Analysis

8 Constructing the characteristic flood hydrograph

8.1 Topics covered


This chapter provides guidance in constructing the characteristic flood hydrograph. The
material covers a number of topics including:

Recommended methods at gauged sites;


Motivation for and description of the parabolic curves method of constructing the
upper hydrograph (Section 8.7);
Recommendations at ungauged sites;
Recommendation to use the pivotal catchment approach (Section 8.10).

The chapter explains the methods chosen for implementation through the FSU Web Portal.
The parabolic method was devised by NUI Galway as part of their HWA research. Its
description here is simplified.

The characteristic hydrograph is typically needed to construct a design hydrograph to


accompany the T-year (peak) flood estimate derived by Volume II methods. The design
flood hydrograph is obtained by scaling up the ordinates of the characteristic flood
hydrograph by the T-year flood flow, QT. In effect, the methods paint in a hydrograph below
the flood peak. Aspects of the synthesis of the T-year design flood hydrograph are therefore
also discussed.

8.2 Features of the methods


The non-parametric and parametric approaches to constructing the characteristic hydrograph
have strengths and weaknesses. It is helpful to recognise these to understand the methods
chosen for implementation. Some difficulties can be circumvented by use of the IBIDEM
method presented in Chapter 9.

Non-parametric method

Incompleteness of the lower part of the characteristic hydrograph generated using the non-
parametric approach presents a problem in some applications. Many flood analysts use
hydraulic modelling techniques that require a complete hydrograph as input. The FSU
recommendation is to sketch in the lower part of the hydrograph subjectively or to use
IBIDEM (see Chapter 9).

Parametric method

The recommended parametric method is the UPO-ERR-Gamma model of Section 4.4. The
characteristic hydrograph generated in this way is continuous and complete. A minor
unrealistic feature is that the gradient of the receding limb of the hydrograph changes
abruptly at the point where the Exponential Replacement Recession takes over from the
Gamma curve (e.g. Figure 8.1). Local smoothing of the hydrograph around the join is
recommended should the feature be found troublesome.

79
Volume III Hydrograph Analysis

It must be recognised that the UPO-ERR-Gamma model performs poorly for unusual sites at
which the flood hydrograph characteristically falls more steeply than it rises.

100
Percentage of peak flow

75

50

25

0
-24 0 24 48 72 96
Time in hours (relative to time of peak flow)
Figure 8.1: UPO-ERR-Gamma characteristic hydrograph at Stn 07009 by Table 6.7 models

A more general defect is that the UPO-ERR-Gamma portrays the flood hydrograph as rising
from a pre-event flow of zero (e.g. Figure 8.1). This is unrealistic for Irish rivers. Ways of
incorporating a non-zero pre-event flow are now discussed.

8.3 Allowances in design flood hydrographs for pre-event flow

8.3.1 Substitution approach


The approach recommended by NUI Galway for inclusion of pre-event flow is to:

Step 1 Estimate an appropriate pre-event flow, Q0 (see Section 8.3.3);


Step 2 Substitute Q0 for any ordinate of the design flood hydrograph that falls below Q0.

This substitution approach has not been implemented through the FSU Web Portal, where the
general recommendation is to use IBIDEM (see Chapter 9). However, the substitution
approach may be useful in some applications, especially those focused on a specific site.

Figure 8.2 shows an example where the pre-event flow represents 10% of the peak flow. The
substitution approach tends to lead to design hydrographs that rise abruptly from the pre-
event flow. However, the approach has the merit that the modelled hydrograph widths are
unchanged by the adjustment. Some hydraulic modelling may not cope well with the abrupt
change of gradient at the start of the flood hydrograph. Should the feature be found
troublesome, the user can deal with this by local smoothing of the hydrograph around the
join.

80
Volume III Hydrograph Analysis

8.3.2 Terminology: baseflow or pre-event flow?


The pre-event flow is denoted here by Q0. Many users will refer to this as the baseflow
allowance. However, this terminology can lead to misunderstanding. In the absence of
heavy rainfall, baseflow decays with time rather than remains constant.

100
Percentage of peak flow

75

50

25

0
-24 0 24 48 72 96
Time in hours (relative to time of peak flow)
Figure 8.2: As Figure 8.1 but with pre-event flow substituting for first part of hydrograph

Additionally, there is scope for the less experienced user to confuse the baseflow allowance
with the baseflow index (BFI). BFI is an index of catchment permeability and storage
developed from daily mean flow data. It is defined as the proportion of the long-term river
flow deriving from subsurface flows or from other delayed responses to rainfall. BFI takes a
value between 0 and 1. Further details of BFI are given in Volume IV.

Obscure terminology is nothing new. When formulating a procedure for constructing design
flood hydrographs, the FSR (NERC, 1975) coined the term average non-separated flow
(ANSF) to represent baseflow. The FSR decision to model ANSF in units of m3s-1 per km2
(i.e. as a standardised baseflow) rather than in m3s-1 added to confusion.

The description in this chapter refers chiefly to the pre-event flow. The FSU advocates use of
IBIDEM (see Chapter 9), which builds a bridge between FSU and FSR methods. Inevitably,
IBIDEM adopts the FSR terminology of referring to the pre-event flow as the baseflow. The
two terms should therefore be treated as synonymous.

8.3.3 Choosing the pre-event flow


The FSU has not researched pre-event flows. An approach possible at gauged sites will be to
fix the pre-event flow at the median of pre-event flows noted for past flood events. However,
the complexity of hydrographs at some stations may defy definition of a pre-event flow.

An alternative is to estimate pre-event flow using elements of the FSR rainfall-runoff method.
IBIDEM (see Chapter 9) specifically accommodates this and is applicable at ungauged as
well as gauged sites.

81
Volume III Hydrograph Analysis

When arbitrary choices are made, it is prudent to check sensitivities. The user is therefore
urged to test the sensitivity of results to the pre-event flow assumed. Flood risk assessments
and the design of flood alleviation works ought not to be greatly impacted by the detail of the
lower part of the hydrograph. If the final results are found to be sensitive to the pre-event
flow assumed, this may reflect weaknesses in the hydraulic modelling.

8.4 Estimation of volume of flow


Having derived the characteristic hydrograph, the widths of exceedance at given percentages
of the peak flow of a design flood are known. The flow volume above a given percentage of
the peak of a design flood can be calculated if required.

8.4.1 Basic method


The design flood hydrograph is produced by scaling the ordinates of the characteristic flood
hydrograph by the design peak flow QT of return period T. The volume VD of the design
flood hydrograph above any given flow Q can be estimated by:

Step 1 Expressing the flow Q as a percentage p of the peak flow;


Step 2 Evaluating the semi-dimensionless volume Vc of the characteristic hydrograph
above percentage p of the peak flow; remarkably, Vc is typically measured in hours;
Step 3 Multiplying the semi-dimensionless volume Vc by QT to obtain the volume VD of
the design flood hydrograph above percentage p of the peak flow; VD is typically
measured in m3s-1 hours or in m3.

8.4.2 Non-parametric case


In the non-parametric approach, a numerical method such as Simpsons rule can be used to
integrate the relevant area (i.e. above the flow of interest and below the flood hydrograph) to
obtain the required volume of flow. Surprisingly, it can be convenient to work out the
volume in horizontal slices, using the widths that define the hydrograph. Where necessary,
the hydrograph width is taken to vary linearly between percentages of the peak flow at which
the width is expressly tabulated. The HWA software provides this option.

8.4.3 Parametric case


In the parametric approach, theoretical expressions can be derived for the volume represented
by the area under the UPO-ERR-Gamma curve and above a given flow. When Equation 4.1
is used as an instantaneous unit hydrograph, the cumulative area under the Gamma defines
the so-called S-curve response (Nash, 1957):

s(t) = I(n, t/K) 8.1

where I(n, x) is the Incomplete Gamma function defined by:

x
In, x
1
n 0
e v v n 1dv 8.2

82
Volume III Hydrograph Analysis

and (n) is the Gamma function. Numerical algorithms for the Gamma and Incomplete
Gamma functions are available in standard packages. The volume under the hydrograph
between particular times is obtained as the difference between the S-curve ordinates at those
times, appropriately rescaled.

The volume of flow sometimes has to be computed in two parts: before and after the point of
inflection at which the exponential replacement recession becomes active in the UPO-ERR-
Gamma model. The relevant formulae are incorporated in the HWA software. [Editorial
note: Unless the HWA software is being applied, it is likely to be more convenient to
evaluate volumes using a numerical method such as Simpsons rule.]

8.5 Deriving the characteristic hydrograph at a gauged site


The non-parametric approach of Chapter 3 is recommended for producing the characteristic
hydrograph at a gauged site with long records of 15-minute flow data. The steps to be
followed are now summarised. The HWA software supports many of these steps. Further
detail is presented in Chapters 2 and 3 above.

Step 1 A set of single-peaked flood hydrographs is extracted from the entire record at a
gauged site. The number of flood hydrographs abstracted is a matter of judgement
but use of the annual exceedance series i.e. a Peaks-Over-Threshold (POT) series
yielding an average of one event per year is recommended.
Step 2 At some stations, many of the flood hydrographs will be found to be multi-peaked,
and a sufficient number of isolated single-peaked floods cannot be identified. In such
cases, some complex floods must be decoupled. In essence, the hydrograph is filtered
to retain only the unimodal part at its core. The procedure is described in Section 2.6.
Step 3 In some cases, either because of error in the flow data or peculiarity in the flow
generation process for the particular event, it may not be possible to identify a usable
filtered hydrograph. Such problematic events are best discarded.
Step 4 Each extracted hydrograph is standardised by dividing the flow ordinates by the flood
peak. The standardised hydrograph thereby attains a peak value of 1.0.
Step 5 For each of a number of reference percentages of the peak flow, the widths of
exceedance on the rising and receding limbs of the hydrograph are abstracted. The
widths are measured in hours. The reference percentages of the peak flow used in the
HWA research are: 98, 95, 90, 85, 80, , 10 and 5%. All such widths are abstracted
where available. In some cases, the width at (e.g.) 75% of the peak flow is available
on one limb of the hydrograph but not on the other.
Step 6 At each reference percentage of the peak flow, the medians of the widths on the rising
and receding sides are separately calculated.
Step 7 The two median widths (at each reference percentage of the peak flow) are plotted as
horizontal lines on a graph on either side of the peak, with width as the abscissa and
percentage of peak flow as the ordinate. The time origin of the graph is set at the time
of the peak flow. [Editorial note: The occurrence of flat-topped hydrographs
sometimes presents a problem. The HWA software adopts the first of exactly equal
peak ordinates as the peak of the hydrograph. If required, users can subvert this
convention by increasing the value of the most central ordinate (of a number of
exactly equal peak ordinates) by a very small amount.]
Step 8 The median hydrograph is constructed as a segmented line passing through the LH
extremities of the horizontal lines, passing through the peak of 1.0 at time 0.0, and
83
Volume III Hydrograph Analysis

passing through the RH extremities of the horizontal lines. The median hydrograph
thus defined is adopted as the characteristic hydrograph of the station.

8.6 Estimating the characteristic hydrograph at an ungauged site

8.6.1 Using the UPO-ERR-Gamma model


Based on the research reported in Chapters 4 and 6, the UPO-ERR-Gamma model can be
used to estimate the characteristic hydrograph at an ungauged site. The steps required are:

Step 1 Abstract the relevant PCDs.


Step 2 Estimate the curve descriptors n, Tr and C using the equations given in Table 6.7.
Step 3 Construct the characteristic hydrograph using Equations 4.7 to 4.10. The graph can be
drawn in Microsoft Excel or other such application software. It is a unit peak at origin
(UPO) model, i.e. the hydrograph is constructed to have a peak value of 1.0 at time
0.0. Thus the time origin is at the time of the hydrograph peak.

Station 07009 Boyne at Navan Weir is again used as the example. The characteristic
hydrograph shown in Figure 8.1 is that derived from PCDs using the equations of Table 6.7.
[Editorial note: This agrees almost perfectly with that derived earlier by HWA (Figure 4.5).
The exceptionally good performance is not typical of that achieved on the 89 catchments as a
whole. Estimation from PCDs is generally prone to considerable error, as evidenced by the
FSEs shown in Table 6.7.]

Experienced users may wish to consider at Step 2 the alternate models given in Table 6.6
which require a value of the baseflow index. The default at an ungauged site is to adopt
BFIsoil as the estimate of BFI.

8.6.2 Using the parabolic curves method


If only the upper hydrograph is required, it is possible to use the parabolic curves method.
This is introduced in Section 8.7.

8.6.3 Using IBIDEM


IBIDEM (described in Chapter 9) is capable of constructing the design flood hydrograph in
full. The package can be guided to match the characteristic hydrograph derived using the
UPO-ERR-Gamma model or that obtained by the parabolic curves method.

8.7 Parabolic curves method

8.7.1 Overview
The parabolic curves method provides a way of estimating the upper hydrograph based on
values of W75 and W50 derived in the non-parametric approach. The method is one of a
number of techniques developed by NUI Galway and can be applied at ungauged as well as
84
Volume III Hydrograph Analysis

gauged sites. The method is more sophisticated than the equivalent FEH technique (Reed
and Marshall, 1999). The parabolic curves method uses hydrograph widths at both 50% and
75% of the peak flow. Moreover, the upper hydrograph is not constrained to be symmetric
about the peak flow. An example is shown in Figure 8.3.

1.00
Flow as proportion of peak flow

W75
0.75

W50
0.50

Hydrograph descriptors
W75 = 36.69 hours
0.25
W50 = 68.65 hours
s = 0.40 (i.e. 40% of width occurs before peak)

0.00
-48 -36 -24 -12 0 12 24 36 48 60
Time in hours (relative to time of peak flow)

Figure 8.3: Example of parabolic curves method (Station 07009 treated as ungauged)

The method is defined by three descriptors: the widths at 75% and 50% of the peak flow (i.e.
W75 and W50 measured in hours) and the eccentricity parameter s. The eccentricity (or
skewness) parameter is a coefficient defining the proportion of the width that occurs before
the time of the peak flow. For the case illustrated in Figure 8.3, s = 0.40; this is the default
value for applications at ungauged sites.

8.7.2 Details of method


The description here differs from that given by NUI Galway. The upper hydrograph is
synthesised in two parts by parabolas passing through the peak flow at (0, 1). The relevant
equations can be written:

y = 1 + bx + cx2 on rising limb 8.3


and:
y = 1 + Bx + Cx2 on receding limb 8.4

The coefficients b, c, B and C are determined from W75, W50 and s.

At first, it appears that the problem is underspecified, with four unknown coefficients and
only three hydrograph descriptors. However, the eccentricity applies at both 75% and 50% of
the peak flow. The solution is therefore fully determined.

NUI Galway adopted a root-solving approach. However, an explicit solution is possible:

85
Volume III Hydrograph Analysis

2W75 W50 2W75 W50


2 2 on
y 1 x 2 x2 rising 8.5
4sW75W50 W75 W50 4s W75W50 W75 W50 limb
and
2W75 W50 2W75 W50
2 2 on
y 1 x x2 receding 8.6
41 s W75W50 W75 W50 41 s W75W50 W75 W50
2
limb

8.7.3 Examples
Stations on the Suir provide a convenient example of the range of shapes that the parabolic
method supports. The values of W75, W50 and s used are taken from the HWA results
summarised in Table D.1 of Appendix D. The upper hydrographs yielded by the parabolic
curves method are shown in Figure 8.4. The unusual shape for Station 16009 arises because
W75 is less than half W50.

1.00
16004 Suir at Thurles
16002 Suir at Beakstown
Flow as proportion of peak flow

16008 Suir at New Bridge


0.75
16009 Suir at Caher Park

0.50

0.25

0.00
-96 -72 -48 -24 0 24 48 72 96 120 144 168 192 216
Time in hours (relative to time of peak flow)

Figure 8.4: Parabolic hydrographs for four stations on the Suir

These characteristic hydrographs can be compared with those given earlier: for the full non-
parametric method (in Figure 5.16) and for the UPO-ERR-Gamma model (in Figure 5.17).

8.7.4 Application at an ungauged site


Based on the research reported in Chapters 3 and 6, the parabolic curves method can be used
to estimate the upper half of the characteristic hydrograph at an ungauged site. The steps
required are:

Step 1 Abstract the relevant PCDs.

86
Volume III Hydrograph Analysis

Step 2 Estimate the hydrograph width descriptors W75 and W50 using the equations given in
Table 6.7.
Step 3 Construct the upper half of the characteristic hydrograph using Equations 8.5 and 8.6.
The graph can be drawn in Microsoft Excel or other such application software. The
hydrograph is constructed to have a peak value of 1.0 at time 0.0. Thus the time origin
is at the time of the hydrograph peak.

Figure 8.3 (shown earlier) reports the outcome for Station 07009 Boyne at Navan Weir,
treating it as an ungauged site. Note that at an ungauged site the eccentricity (i.e.
skewness parameter) takes the fixed value s = 0.40.

Experienced users may wish to consider at Step 2 the alternate models given in Table 6.6
which require a value of the baseflow index. The default at an ungauged site is to adopt
BFIsoil as the estimate of BFI.

8.8 Constructing the design flood hydrograph

In principle, the design flood hydrograph for a given T-year peak flow (derived using
Volume II methods) is obtained by scaling up the ordinates of the characteristic flood
hydrograph by the appropriate factor. If only the upper part of the design flood hydrograph is
required, the procedure is complete.

Where the complete design flood hydrograph is required, the user has a number of options:

To sketch in the lower part of the design hydrograph subjectively;


To use IBIDEM (see Chapter 9);
To adopt the parametric method of constructing the characteristic hydrograph, i.e.
using the UPO-ERR-Gamma model of Section 4.4 (but see notes in Section 8.2).

8.9 Software
Implementation gives the user access to the extensive features of the HWA and IBIDEM
packages. These packages are supplied for use offline to the FSU Web Portal. Some
technical details of the HWA software are given in Appendix E. IBIDEM is presented in
Chapter 9 with further details in Appendix F.

Experienced users will be able to consider and develop additional options for constructing the
design flood hydrograph in particular applications.

8.10 Selection and use of the pivotal catchment


8.10.1 Overview
The concept of a pivotal catchment is introduced in Volume II. The pivotal catchment is the
gauged catchment judged to be most relevant to the specific flood estimation problem at an
ungauged site. The concept is fundamental to application of FSU methods at ungauged sites.

87
Volume III Hydrograph Analysis

The FSU Web Portal is designed to steer all users to select and use a pivotal catchment. This
is fundamental to applying Volume II methods for estimating the index flood QMED at an
ungauged site.

Less experienced users may wish to become familiar with the pivotal catchment concept by
exploring its use in QMED estimation in Volume II. In principle, the pivotal catchment
approach is followed whenever the characteristic hydrograph is needed at an ungauged site.

Selection and use of a pivotal catchment promotes the effective use of gauged flood data,
even at ungauged sites. The summary description is as follows. First, the performance of the
ungauged method of Section 8.6 method is assessed by checking how it performs for the
pivotal catchment. Second, the correction factor required to make the ungauged method
perform well at the pivotal site is transferred (fully or partially) to adjust the estimate at the
ungauged site. The overall procedure is referred to as a data transfer.

8.10.2 Selection of the pivotal catchment


The user assesses the most appropriate data transfer by making a reasoned selection of a
pivotal catchment. This is the gauged catchment judged to be most relevant to the specific
flood estimation problem.

The pivotal catchment is the users assessment of the most relevant catchment on which to
base a data transfer. Where flood data are available from a gauge sited upstream or
downstream of the subject site, this will often be selected as the pivotal catchment. In other
cases, the selection is likely to be more precarious and will hinge on the users judgement of
catchment similarity.

An algorithmic judgment of catchment similarity is likely to give weight to differences in a


few leading factors e.g. catchment size (represented by AREA), catchment wetness
(indexed by SAAR) and catchment permeability (indexed by BFI or BFIsoil) and to neglect
all other factors. This is not a safe approach.

A particular feature present in one catchment and absent from another may lead to strong
differences in their flood behaviour. Arterial drainage (indexed by ARTDRAIN or
ARTDRAIN2) is perhaps the most notable such feature. Volume II reports evidence that
BFIsoil and ARTDRAIN2 are important in characterising the post-drainage flood response of
a catchment, whilst the descriptors DRAIND and S1085 are more important in characterising
the response of undrained catchments. These findings may help the experienced user to
judge which PCDs to examine closely when judging hydrological similarity for the purpose
of selecting the pivotal catchment.

Other notable features to consider when assessing catchment similarity are the extent of
urbanisation (indexed by URBEXT) and the presence of large lakes (indexed by FARL).
Research reported in Volume II endorses the recommendation to favour geographical
closeness as well as similarity in key PCDs such as FARL when selecting a pivotal
catchment for use in a particular flood estimation problem.

88
Volume III Hydrograph Analysis

8.10.3 Recommended procedure for data transfer


Selection of the pivotal catchment is a demanding task that calls for reasoning and
judgement. Some judgements will be unsettling because of the impact that the data transfer
has on the final results.

In some cases not least on small catchments the pivotal catchment selected may not be
wholly convincing. This is not a reason to abandon making a data transfer. However, it may
justify making only a partial transfer. An approach to making a partial transfer is included in
Step 4 of the procedure now illustrated by worked example.

8.10.4 Example
The methodology for estimating the characteristic hydrograph at an ungauged site is
illustrated for the Suck at Rookwood. This corresponds to Station 26002 but is treated here
as an ungauged site. The following provides a broad guide to deriving an estimate of the
characteristic hydrograph at this location.

The worked example is for use of the parabolic curves method. The same principles apply to
use of the UPO-ERR-Gamma model. However, with three parameters to consider, data
transfers are likely to be rather complicated to execute for that model.

Step 1 Confirm location: It is important to confirm the location of the subject site and to
check the centroid of its catchment. The site location determines the PCDs extracted
in Step 2. The centroid is relevant to judging the nearness of the subject catchment to
available gauged catchments, when the user is undertaking the important task of
selecting the pivotal catchment (see Section 8.10.5). t is the users responsibility to
check that the FSU digital data provide a fair representation of the real conditions.

Step 2 Derive catchment descriptor information: Table 8.1 lists the PCDs needed in the
version of the parabolic curves method illustrated here. The PCDs derive from the
digital datasets made available with the FSU.

Table 8.1: Selected PCDs for Suck at Rookwood


PCD value unit PCD value unit
AREA 641.45 km2 FOREST 0.080
2
DRAIND 0.799 km/km FARL 0.979
S1085 0.500 m/km FLATWET 0.690
ALLUV 0.025 mm ARTDRAIN 0.000

Step 3 Apply hydrograph width models: From Table 6.7 we have for the parabolic curves
method:

W75 = 31.28 DRAIND-0.88 FARL-5.85 (1+FOREST)-2.86


(1+ARTDRAIN)-2.92 FLATWET3.12 (S1085/1000)-0.27

W75 = 31.28 (0.799)-0.88 (0.979)-5.85 (1.081)-2.86 (1.0002)-2.92 (0.690)3.12 (0.500/1000)-0.27


= 84.42 h

89
Volume III Hydrograph Analysis

W50 = 34.02 DRAIND-0.95 FARL-6.59 (1+FOREST)-2.83


(1+ARTDRAIN)-2.60 FLATWET2.44 (S1085/1000)-0.29
= 34.02 (0.799)-0.95 (0.979)-6.59 (1.081)-2.83 (1.0002)-2.60 (0.690)2.44 (0.500/1000)-0.29
= 142.30 h

[Editorial note: The estimates of W75 and W50 differ very slightly from those shown
in the centre-right columns of Table D.1 of Appendix D. This is because, in the
recommended models, the exponents have been rounded to 2 decimal places.]

Step 4 Transfer data from gauged locations to improve model prediction at subject site:

The general procedure is to infer an adjustment factor, AdjFac, by reference to the


performance of the PCD-based model at a nearby gauged site. Thus, the adjustment
factor for W75 would be:
W75,gauged
AdjFacW75 8.7
W75,PCD
The adjustment is then partially or fully transferred to the subject site:

8.8
W75,adj AdjFac W75 W75,PCD
h

The typical procedure is to apply a full transfer by setting the exponent h to 1.0. If
W75 is found to be 20% greater than the PCD-based estimate, it is assumed that the
model will be similarly in error at the subject site. Thus, the estimate of W75 at the
subject site is adjusted by multiplying by 1.20.

The exponent h can be thought of as the hardness of the data transfer. h = 1 denotes a
full (or hard) transfer. A partial (or softer) transfer might set h = 0.5. In this case,
if W75 is found (at the donor site) to be 20% greater than given by the PCD model, the
estimate of W75 at the subject site is adjusted by multiplying by a factor of 1.200.5 or
1.095.

Much skill attaches to deciding which of several possible donor catchments is pivotal to
improving estimation at the subject site. With gauged sites upstream (Station 26006) and
downstream (Station 26005) of the subject site, the choice might not be clear-cut for the Suck
at Rookwood. However, it transpires that HWA has not been undertaken for Station 26006
Suck at Willsbrook. Thus, the data transfer illustrated here is from Station 26005 Suck at
Derrycahill, which drains an area 69% greater than that at Rookwood.

Table 8.2 summarises the data transfer from the Suck at Derrycahill (Station 26005) to the
Suck at Rookwood. It is seen that hydrograph widths are appreciably underestimated at the
donor site. Use of a hard data transfer assumes that the same relative error will occur when
applying the PCD models at Rookwood and adjusts the hydrograph widths accordingly.

When HWA results for the Suck at Rookwood itself are examined (lighter-shaded rows in
Table 8.2), it is found that the PCD method does indeed underestimate hydrograph widths
there. However, it transpires that, the data transfer from Derrycahill is too strong. In this
instance, a soft transfer with h = 0.5 would work better (see RH column of table). However,
the user will not know this for a subject site that is truly ungauged! The upper hydrographs

90
Volume III Hydrograph Analysis

transferred (see Figure 8.5) are somewhat pointy in comparison to the notably flat-topped
hydrograph (black curve) that the parabolic method yields directly at Rookwood.

Table 8.2: Data transfers to Suck at Rookwood using parabolic curves method
Hydrograph width
Data HWA PCD Implied
Hard Soft
Method transfer factorial
HWA transfer transfer
from adjustment
with h = 1 with h = 0.5
hours hours
Width at 75% of hydrograph peak (W75)
No data
84.42
transfer
Downstream Station
136.04 86.15 1.58 133.4 106.1
donor 26005
Analysis of Station
118.7
gauged data 26002
Width at 50% of hydrograph peak (W50)
No data
142.30
transfer
Downstream Station
209.83 147.63 1.42 202.1 169.6
donor 26005
Analysis of Station
163.98
gauged data 26002

1.00
Proportion of peak flow

0.75

0.50

Parabolic curves method


HWA at 26002
0.25 Soft transfer (26005 to 26002)
Hard transfer (26005 to 26002)

0.00
-144 -96 -48 0 48 96 144 192
Time in hours (relative to time of peak flow)

Figure 8.5: Upper hydrographs transferred from Derrycahill to Rookwood

91
Volume III Hydrograph Analysis

[Editorial note: Station 26002 Suck at Rookwood provides an example where the parabolas
fitted to the rising and receding limbs have turning points just before and just after the
nominal peak value of 1.0. This feature is detectable in Figure 8.5 by close scrutiny of the
crest segment of the hydrograph shown in black. Where the feature arises, it is reasonable to
reduce any ordinate that exceeds 1.0 to the peak value of 1.0, thereby producing a flat-topped
hydrograph.]

For completeness, HWA results for the two catchments are shown in Figure 8.6. The
characteristic hydrograph at Derrycahill (Station 26005) is relatively smoothly represented by
the derived median hydrograph (of Section 3.5), which in turn is especially well modelled by
the UPO-ERR-Gamma (of Section 4.4). It is confirmed that hydrographs at Rookwood are
rather narrower than at Derrycahill, although the difference is almost entirely in the receding
limb.

Figure 8.6: Derived median and UPO-ERR-Gamma hydrographs for Stns 26002 and 26005

8.10.5 Further discussion of choice of method and of pivotal catchment


While the pivotal catchment principle applies to transferring a characteristic hydrograph in
much the same as it does to transferring a gauged value of QMED in Volume II, a number of
important differences arise in practice:

In the index flood case, there is the one flood measure QMED to be evaluated and
transferred. In the characteristic hydrograph case there are several possibilities.
There is the choice between use of the UPO-ERR-Gamma model (with three
parameters open to adjustment) and use of the parabolic curves method (with two
width descriptors open to adjustment). The use of IBIDEM (see Chapter 9) brings
additional options.
In the index flood case there is essentially the one PCD-based method of estimating
QMED, albeit with important adjustments for urbanised catchments. In the
characteristic hydrograph case, there is a choice between models that use BFI and
models that do not.

The suitability of a donor reflects many factors. Donors on the same river system are likely
to be most suitable. However, the typical width and shape of hydrographs will be influenced
by any notable feature that intervenes between the subject site and the donor site. The most
obvious features are lakes or reservoirs, which are likely to modify the width and shape of
hydrographs appreciably. This is confirmed by the appearance of FARL in all ten PCD
models presented in Table 6.6 and Table 6.7.

92
Volume III Hydrograph Analysis

Different users will make different judgements. Some will argue that the presence of FARL
in the models allows the transfer to proceed. Others may argue that site-specific features
mean that the lake has a stronger effect than represented in the necessarily generalised model.

Confidence in the relevance of the data transfer is weakened if a special feature intervenes
that is not represented in the model, e.g. if a significant tributary of wholly different character
joins the river between the donor and subject sites. If it is the most suitable donor available,
the pragmatist will allow the data transfer in part: e.g. setting h to 0.5 (or less).

Finally, it should be noted that it may be appropriate or necessary to choose one gauged site
as the pivotal catchment in QMED estimation and another in estimation of the characteristic
hydrograph.

8.10.6 Urbanised catchments


Large-scale urbanisation affects flood magnitudes and catchment response times very
appreciably. Flood magnitudes are generally increased and flood response times (and
hydrograph widths) are compressed, often markedly. Great care is therefore required in
making data transfers to a subject catchment that is heavily urbanised. In similar vein, a
notably urbanised catchment should not normally be chosen as the pivotal station if the
subject catchment itself is largely rural.

Experienced users may find it helpful to apply IBIDEM. This allows the merit of particular
data transfers to be interpreted in terms of the parameters of the FSR rainfall-runoff method
and the urban adjustments to that method presented in FSSR16 (IH, 1985).

93
Volume III Hydrograph Analysis

9 IBIDEM

9.1 The idea of IBIDEM


IBIDEM stands for Interactive Bridge Invoking the Design Event Method. The idea is to
provide a bridge between the FSU method of estimating a design flood hydrograph and the
FSR design event method that it replaces. Two parameters of the FSR rainfall-runoff model
(time to peak and standard percentage runoff) are chosen by optimisation so that the design
hydrograph synthesised by the FSR method matches that produced by the FSU procedures.

This chapter explains how IBIDEM is structured and illustrates its use on five example
catchments. The tests indicate that IBIDEM is helpful in assessing design flood hydrographs
produced using the FSU procedures and can help the experienced user to judge whether a
design hydrograph is consistent with the properties expected of the particular catchment.
IBIDEM is supplied as a standalone software package downloadable through the FSU Web
Portal. Some technical details are given in Section 9.5 and Appendix F. The package was
developed by JBA Consulting.

9.1.1 Reminder of hydrograph estimation by FSU methods


The T-year peak flow is typically estimated as the product of an index flood and a growth
curve. The methods are described in Volume II. The required design hydrograph is
constructed around the peak flow by applying a hydrograph shape taken from the HWA
presented in earlier chapters. There are two main methods:

At gauged sites, the characteristic hydrograph is built up using widths extracted from
observed hydrographs at given percentages of the peak flow. This is the non-
parametric approach of Chapter 3, and can be executed using the HWA software.
At ungauged sites, the parametric approach of Chapter 4 can be applied. Parameters
of the UPO-ERR-Gamma model are estimated from PCDs using the equations
presented in Chapter 6. Application of the parabolic curves method of Section 8.7 is
also supported.

IBIDEM also provides the user with a number of additional options.

9.1.2 Hydrograph estimation by the FSR design event method


Some users may not be very familiar with the FSR design event method. A brief summary is
attempted here. The relevant volume of the Flood Estimation Handbook provides further
details (Houghton-Carr, 1999).

The T-year design hydrograph is constructed as the output to the unit hydrograph/losses
rainfall-runoff model. The FSR design event method combines four inputs: the temporal
profile, duration, and rarity of the rainfall event and the pre-event catchment wetness. The
first three define the rainfall input (to the rainfall-runoff model), whilst the fourth defines the
initial condition (of the rainfall-runoff model). These inputs take specific values according to
particular rules (see Figure 9.1). The rules reflect some of the general properties of the
catchment and its climate.
94
Volume III Hydrograph Analysis

Rainfall profile Rainfall duration Rainfall rarity Pre-event wetness

Profile = constant D = D (SAAR, Tp) Train = Train ( Tflood) CWI = CWI (SAAR)

Figure 9.1: Design inputs to FSR rainfall-runoff method of flood frequency estimation

It is necessary to adopt suitable values for the parameters of the rainfall-runoff model itself.
For the FSR unit hydrograph/losses model, the parameters are:

The standard percentage runoff, SPR;


The unit hydrograph time-to-peak, Tp;
The standardised baseflow, known as the average non-separated flow, ANSF.

On all but highly permeable catchments, the last parameter tends to be relatively unimportant.

Two other factors play a role in the model: the catchment area (AREA), and the areal
reduction factor (ARF). ARF is applied to estimate the design catchment rainfall from the
design rainfall depth at a typical point within the catchment. In application here, the rainfall
depth-duration-frequency model is taken from Volume I rather than from FSR methods.

9.1.3 Basic idea of bridge between the FSR and FSU methods
The aim of IBIDEM is to link the FSU method of T-year hydrograph estimation to the FSR
rainfall-runoff method it replaces. The Tp and SPR parameters of the rainfall-runoff model
are chosen so that the design hydrograph synthesised by the FSR method matches that
produced by the FSU procedures. The approach offers several gains:

Whereas the HWA methods of Chapter 3 construct only the upper parts of the design
hydrograph, IBIDEM synthesises the entire hydrograph. This allows the user to look
at runoff volumes (e.g. for assessing flood storage) and to route flood hydrographs,
as they do when using the FSR rainfall-runoff method.
A link with rainfall is made. By noting the percentage runoff (PR) and the rainfall
duration (D) implied by IBIDEM, the user is able to check whether the FSU design
hydrograph has properties consistent with that expected of the catchment.
Those with particular experience of the FSR design event method are able to interpret
the Tp and SPR parameters of the rainfall-runoff model to which the FSU design
hydrograph is said to be equivalent. Based on experience or further guidance, users
can vary these values to test sensitivities and to investigate the possible effects of
catchment change on the design flood hydrograph.

95
Volume III Hydrograph Analysis

9.2 How IBIDEM fits hydrographs


IBIDEM fits an FSR rainfall-runoff hydrograph to match the shape and peak flow of an
imported FSU design hydrograph. The fitting is achieved by adjusting the time to peak (Tp)
and standard percentage runoff (SPR) parameters of the FSR rainfall-runoff method.

IBIDEM implements all parts of the FSSR16 version of the rainfall-runoff method (IH,
1985), other than estimation of Tp and SPR from catchment characteristics. Values of Tp
and SPR are instead derived by fitting the implied FSR hydrograph to the imported FSU
hydrograph.

In running the rainfall-runoff method, IBIDEM undertakes the following steps:

Step 1 Imports the physical catchment descriptors AREA, SAAR and URBEXT; URBEXT
is needed for the urban adjustment to the percentage runoff.
Step 2 Finds values of SPR and Tp by optimisation. In some options, adjustments are made
by the user.
Step 3 Selects an appropriate data interval T based on Tp, adopting a convenient value
such as 0.25 hours or 1 hour.
Step 4 Calculates the design rainfall duration D from Tp and SAAR, and evaluates the areal
reduction factor (ARF) from AREA and D.
Step 5 Constructs a triangular unit hydrograph with time-to-peak Tp.
Step 6 Evaluates the design rainfall depth from D and a user-supplied flood return period (or
set of return periods). The user-supplied flood return period (Tflood) is linked to a
rainfall return period (Train) according to the FSR design package used. [Editorial
note: The design package is a prescribed a set of rules to be followed when using the
FSR rainfall-runoff method. One package corresponds to winter conditions and is
typically applied on catchments that are largely rural.]
Step 7 IBIDEM requires the rainfall depth for a typical (i.e. average) point in the catchment.
The depth is multiplied by ARF (Step 4) to obtain the catchment-average rainfall (P)
of required return period. [Editorial note: The user obtains rainfall depths for a set
of durations and return periods through the FSU Web Portal. These are supplied to
IBIDEM in the form of a table of rainfall depths, with durations (0.25 to 600 hours)
in rows and return periods (2 to 200 years) in columns. IBIDEM calculates the
required rainfall depth by interpolation, using linear interpolation between durations
and logarithmic interpolation between return periods.]
Step 8 Distributes the rainfall depth according to a standard temporal profile taken from the
FSR. The most commonly used profiles are the so-called 75% winter and 50%
summer profiles. To meet the project specification, the default setting in IBIDEM is
to use the 75% winter rainfall profile: even on an urbanised catchment. The
alternative summer profile can be chosen by the user.
Step 9 Calculates the percentage runoff PR from SPR, rainfall depth P and the catchment
wetness index CWI, applying an urban adjustment if necessary. The urban
adjustment is amended to use URBEXT from the FSU rather than URBAN from the
FSR. The relevant formula is:
PR = PRrural (1.0 0.47 URBEXT) + 70 (0.47 URBEXT) 9.1
(The factor 0.47 arises as the product of 0.30 and 1.567. The factor 1.567 back-
converts URBEXT to be compatible with the URBAN index used in the FSR.)
96
Volume III Hydrograph Analysis

Step 10 Applies the PR to the total rainfall profile to obtain the net rainfall profile, i.e. the
portion of rainfall that generates rapid response runoff.
Step 11 Convolves the unit hydrograph with the net rainfall profile to give the rapid response
hydrograph.
Step 12 Adds baseflow to give the total runoff hydrograph. Baseflow is calculated from
AREA, SAAR and CWI using a standard equation from FSSR16.

9.3 General approach to the optimisation

9.3.1 First Tp and then SPR


The two parameters to be fitted (Tp and SPR) affect the hydrograph in distinct ways. SPR
affects the magnitude of the flows but does not have any effect on timings. In cases where
baseflow forms only a minor element, the flood magnitude is roughly proportional to SPR
(see the example in Figure 9.2).

The Tp parameter affects both the timing and the magnitude of the flows. A shorter Tp alters
the T-year flood magnitude in three ways:

It forces an increase in the peak of the unit hydrograph (to maintain the same volume
of flow in a shorter time);
It shortens the design storm duration, hence increasing the rainfall intensity for a
given return period;
The resulting change in the design rainfall depth affects the percentage runoff via the
DPRRAIN term (see Section F2 of Appendix F).

Figure 9.2 provides an example of the kind of effect that Tp and SPR have on the peak of the
design flood hydrograph in the FSR rainfall-runoff method.

25
25
(m3 s-1)

20
20
(m3/s)

SPR = 20%
flowflow


15
15
SPR = 20%
peak

10
10 SPR
SPR ==40%
40%
Peak
T-year

55

00
00 22 44 66 8 10
10
Tp (hours)
Tp (hours)

Figure 9.2: Illustration that FSR T-year peak flow varies with Tp as well as with SPR

Because SPR has no effect on hydrograph timings, it is convenient to optimise Tp first. SPR
is adjusted in a later step to give a peak flow that exactly matches the peak of the imported
FSU design hydrograph. For rather intricate reasons with advantages outweighing
disadvantages this strategy is in fact preferable to that of optimising SPR and Tp jointly.

97
Volume III Hydrograph Analysis

9.3.2 Use of horizontal fitting


IBIDEM fits the FSR rainfall-runoff hydrograph standardised so that its ordinates are
expressed as percentages of the peak flow to the characteristic hydrograph derived by FSU
methods. The process uses horizontal fitting, in which differences in hydrograph widths are
minimised between the FSU and FSR representations. This is in contrast to the conventional
vertical fitting of hydrographs in terms of differences in flows.

The set-up is illustrated in Figure 9.3. The FSU hydrograph is built up from median
hydrograph widths at various percentages of the peak flow, using the non-parametric
approach of Chapter 3. [Editorial note: IBIDEM refers to this as the empirical approach.]
Fitting is carried out for the portion of the hydrograph above a threshold. This accommodates
the feature that FSU hydrograph does not cover the full range of flows down to zero.
FSU hydrograph Sample FSR hydrograph

Threshold for fitting Threshold for fitting

Figure 9.3: Horizontal fitting by comparing hydrograph widths

9.3.3 Deriving Tp by optimising the fit to the FSU flood hydrograph


Tp is obtained by optimising the shape of the FSR rainfall-runoff hydrograph so that it best
matches the FSU characteristic hydrograph. These semi-dimensionless hydrographs are
expressed as a percentage of the peak flow.

A complication is that baseflow (in the FSR rainfall-runoff method) is defined as a fixed
amount in m3s-1 rather than as a proportion of the peak flow. The difficulty is overcome by
fitting the widths of the rapid response parts of the hydrographs, i.e. after subtracting the
(fixed) baseflow BF from the FSU hydrograph (Figure 9.4).
Flow (m3 s-1)

qpeak Qpeak

BF
Time (hours)
Figure 9.4: Relationship between peak flow Qpeak and peak rapid response qpeak
98
Volume III Hydrograph Analysis

The detailed procedure is:

Step 1 Calculate the baseflow BF (in m3s-1) from the FSSR16 method [Editorial note:
FSSR16 provides an equation for ANSF, defined as the baseflow per unit area i.e. in
m3s-1 per km2. Thus, BF = ANSFAREA.]
Step 2 Subtract BF from the FSU hydrograph. The peak flow Qpeak is thereby reduced to the
response peak qpeak.
Step 3 Express the threshold for fitting as a % of qpeak:
%qthreshold = %Qthreshold - 100(BF/Qpeak)
This meets the IBIDEM requirement that the user specifies the fitting threshold as a
percentage of the total peak flow, %Qthreshold. The user is prompted to raise the
threshold should the percentage first entered yield a threshold flow that is less than or
equal to BF.
Step 4 Calculate a set of m widths WFSU of the FSU response hydrograph for percentage
flows between %qthreshold and 100%, at an interval of 1% (see LH side of Figure 9.3).
Step 5 Run the FSR rainfall-runoff method with an arbitrary (fixed) SPR and an initial guess
for Tp.
Step 6 Calculate a set of widths WFSR of the FSR response hydrograph at the same
percentages as in Step 4 (see RH side of Figure 9.3).
Step 7 Evaluate the objective function: [WFSU(i) - WFSR(i)]2 for i = 1 to m.
Step 8 Vary Tp and repeat Steps 5 to 7 until the objective function is minimised (see
Section F1 of Appendix F for details of the optimisation method).

The non-parametric method of Chapter 3 may sometimes produce hydrographs with more
than one peak (Figure 9.5). Within IBIDEM, these cases are treated by excluding the time
when flow is below the relevant percentage of the peak when evaluating the width of the
hydrograph. This pragmatic approach ensures that IBIDEM represents the total duration of
the hydrograph but without being unduly affected by the separation of the peaks. The output
from IBIDEM is always a single-peaked hydrograph.

IBIDEM discounts this portion of


the total width of the hydrograph

Figure 9.5: Double-peaked hydrograph

99
Volume III Hydrograph Analysis

9.3.4 Deriving SPR by matching the required peak flow


After Tp has been optimised, the FSR hydrograph is scaled to fit the relevant peak flow,
which will typically have been derived by Volume II methods. SPR is calculated by working
out the factor by which the hydrograph has to be multiplied to match the FSU peak flow.
The details of how IBIDEM does this are given in Section F2 of Appendix F.

9.4 Additional IBIDEM options


The IBIDEM software provides a number of additional options. The principal ones are now
summarised. Graphical displays are illustrated later, chiefly in Section 9.5.2.

9.4.1 Flood frequency


The Flood frequency option allows users to optimise a set of hydrographs for different return
periods. To allow maximum flexibility of options, users are required to import a separate
FSU flood hydrograph for each return period. IBIDEM allows up to seven return periods to
be analysed together. A separate pair of Tp and SPR values is fitted at each return period,
using the method described in Sections 9.2-9.3.

9.4.2 Sensitivity to storm duration


The terms design rainfall and design storm are used interchangeably. The Sensitivity to storm
duration option and further options are made available once IBIDEM has performed the
hydrograph fit for a single return period. The fitted values of Tp and SPR are retained (i.e. no
further optimisation is carried out) and the FSR rainfall-runoff method is re-run for five trial
storm durations. A feature of the FSR rainfall-runoff method is that the volume of the flood
hydrograph increases as the storm duration increases.

Default durations for the trials are 0.5D, 0.5D, D, 2D and 2D where D is the duration
resulting from the Tp value found in the optimisation. The user can change the trial durations
if desired.

In addition to testing sensitivities, this option may be helpful in river modelling applications
(see Volume V) where a longer-than-normal storm is being applied to a tributary catchment
in order to generate a T-year flood further down the river system.

9.4.3 Sensitivity to model parameters


The Sensitivity to model parameters option allows the user to re-run the rainfall-runoff
method (after the initial optimisation stage), with altered values of the parameters Tp and/or
SPR. If Tp is altered, the storm duration is automatically updated. If SPR is altered, PR is
updated. Other settings remain unchanged.

The option allows the user to explore the possible impact of land-use change, by adjusting Tp
and/or SPR to represent conditions before and after the change. An implicit assumption is
that the user trusts application of the FSR rainfall-runoff method to represent the particular
land-use change (such as agricultural drainage or tree planting) adequately.
100
Volume III Hydrograph Analysis

9.4.4 Sensitivity to changes in urbanisation


The Sensitivity to changes in urbanisation option allows the user to re-run the rainfall-runoff
method for an altered value of URBEXT. IBIDEM calculates revised values for Tp and PR
to reflect the change in URBEXT. Because of the joint use of newer and older technologies,
the procedure is rather intricate.

Back-converting URBEXT

The first step is to back-convert the revised URBEXT to the equivalent FSR descriptor of
urban extent (URBAN). This is done using:
URBAN = 1.567 URBEXT 9.2
In the FSR rainfall-runoff method method, URBAN affects Tp, PR and the choice of design
event package (winter or summer).

Choice of design package

Were IBIDEM to choose the design package automatically based on the degree of
urbanisation, it would be possible for a small increase in URBEXT to lead to an abrupt
change from use of a winter design event to use of a summer design event, with potentially a
large change in the design flood hydrograph. To avoid this possibility and to allow greater
flexibility the choice of design event package is set manually by the user.

Effect on PR

IBIDEM calculates a new PR from SPR using the urban adjustment (Equation 9.1) from Step
9 of the Section 9.2 procedure.

Effect on Tp

Tp for the base condition will have been found during the initial run. IBIDEM updates this
for the revised value of URBAN by invoking part of the FSSR16 model for estimating Tp(0).
The factor representing the urban effect on response times in the FSSR16 model is the term
(1+URBAN)-2.2.

The Tp value obtained in the initial run of IBIDEM is therefore updated for the altered level
of urbanisation using:
(1 URBANrevised )2.2
Tp0revised = Tp(0) 9.3
(1 URBAN) 2.2

Tp is converted to and from Tp(0) as required, using: Tp = Tp(0) + T/2 where T is the data
interval (see Section 9.2). Equation 9.2 is used as required to convert current and projected
values of URBEXT to the corresponding values of URBAN.

Summary

Although intricate, this option allows users to investigate the impacts of urban development
on design flood hydrographs and peak flows.
101
Volume III Hydrograph Analysis

9.5 Further details of the software

9.5.1 Inputs
The IBIDEM software requires the following inputs in all cases:

Catchment descriptors AREA, SAAR and URBEXT: typed in on the first screen
(the FSU descriptor of urbanisation is the fractional urban extent URBEXT, and takes
a value between 0 and 1);
Design flood hydrograph derived from FSU procedures: supplied as a CSV file
(this gives pairs of time and flow values);
Rainfall frequency information from FSU procedures: supplied as a CSV file
giving a table of design rainfall depths at an average point in the catchment (the
tabulated depths are for a standard set of durations and return periods);
Return period associated with the FSU hydrograph (default is 2 years);
Threshold flow to be used in fitting: expressed as a percentage of the peak flow
(default is 50%).

These data are validated during the input process and the user notified of any exceptions (see
Section F3 of Appendix F. In particular, the user is warned if:

The FSR baseflow exceeds part of the imported FSU hydrograph. In this case,
the user needs to import a hydrograph with a higher minimum flow (e.g. by removing
the first or last few values) or to adjust the baseflow.
The threshold falls below the lowest flow value in the rising/receding limb of the
imported FSU hydrograph. In this case, the user needs to raise the threshold flow
used for the fitting or to import a more complete hydrograph.

Additional inputs are required for some of the optional functionality:

For the flood frequency option: Hydrographs for multiple return periods;
For the option to vary baseflow: Value for baseflow;
For the sensitivity to model parameters option: New value for Tp or SPR or both;
For the sensitivity to changes in urbanisation option: New value for URBEXT.

9.5.2 Graphical displays


IBIDEM displays the hydrograph fit, showing the imported FSU hydrograph, the fitted FSR
hydrograph and the threshold flow used for fitting. Figure 9.6 provides an example. The two
hydrographs are aligned so that they peak at the same time.

When the flood frequency option is selected, the user can choose the return period for which
the fit is displayed. Alternatively, it is possible to display a graph showing how a particular
variable changes with return period. The user can select any one of peak flow, percentage
runoff, SPR, Tp, rainfall depth or runoff volume. The example in Figure 9.7 shows how the

102
Volume III Hydrograph Analysis

runoff volume expressed as a depth in mm across the catchment changes with return
period.

FSR hydrograph FSU derived


fitted by IBIDEM median hydrograph

Figure 9.6: Display of fitted and imported hydrographs

Figure 9.7: Display of how a variable changes with return period

When the sensitivity to storm duration option is selected, the user can plot either a graph
showing multiple hydrographs i.e. one for each duration plus the imported FSU hydrograph
(e.g. Figure 9.8) or a graph showing how a particular variable changes with return period.

The example in Figure 9.9 shows how the peak flow varies with the design storm duration.
This provides a way of identifying the critical duration that the bridge to the FSR rainfall-
runoff method implies for the catchment. The figure also illustrates the manner in which
103
Volume III Hydrograph Analysis

options are selected on-screen in IBIDEM. Q denotes the peak flow, PR the percentage
runoff, P the rainfall depth and V the hydrograph volume.

Figure 9.8: Display of hydrographs for multiple storm durations

Figure 9.9: Display of how a variable changes with storm duration

When the sensitivity to model parameters or the urbanisation option is selected, IBIDEM
plots the imported FSU hydrograph, the original fitted FSR hydrograph and the altered FSR
hydrograph resulting from the changed parameter(s). Figure 9.10 illustrates this for a case
where URBEXT increases from 0.00 to 0.20.

104
Volume III Hydrograph Analysis

Figure 9.10: Display of sensitivity to an increase in URBEXT

9.5.3 Display options


Various options are provided for graph layout and units (the option name is in italics):

Hydrographs can be plotted with the vertical axis showing either % of peak flow
(default) or m3s-1.
Hydrographs can be plotted with the time origin either at the peak (default) or at the
start of the FSR hydrograph.
Flow units can be either m3s-1 (default) or mm/hr.
Runoff volume units can be either mm equivalent of catchment runoff (default) or m3
or cumec-hours. A cumec-hour is the volume represented by a flow of 1 m3s-1
sustained for one hour, i.e. 3600 m3.
Plots of variables against return period can have a horizontal axis showing the
Gumbel reduced variate, the Logistic reduced variate or the natural logarithm of
return period. In each case, a subsidiary axis shows the return period in years. These
are return periods on the annual maximum scale. Thus, the 50-year event corresponds
to a value with an annual exceedance probability of 0.02. [Editorial note: Frequency
statements should be treated with some degree of caution. Although the peak of the
FSU flood hydrograph is nominally of the stated frequency, the frequency assignment
does not strictly transfer to quantities (such as the flood volume) derived by invoking
the bridge to the FSR rainfall-runoff method.]

9.5.4 Goodness-of-fit measures


IBIDEM summarises the goodness of fit of the FSR rainfall-runoff hydrograph to the FSU
design hydrograph in two measures: the root mean square error (RMSE) and the Nash-
Sutcliffe efficiency (NSE). These are calculated in quite different ways:
105
Volume III Hydrograph Analysis

RMSE is calculated as part of the IBIDEM fitting process. It is the root mean square
error in terms of hydrograph width (measured in hours) for the upper portion of the
hydrograph over which the fitting is carried out. It indicates how well the FSR and
FSU hydrographs match in terms of their widths. A small value of RMSE indicates a
good fit. The RMSE obtained using IBIDEM is always the minimum possible given
the shape of the imported FSU hydrograph and the family of shapes that the FSR
rainfall-runoff hydrograph can take.
NSE is a dimensionless measure of hydrograph fit calculated in the conventional (i.e.)
vertical direction. It is a measure of the goodness of fit in terms of flow over the
duration of the imported FSU hydrograph. Values close to 1.0 indicate an excellent
fit. Negative values of NSE indicate that a better fit could be achieved using the mean
flow. The statistic is calculated independently of the fitting done by IBIDEM, and
will not usually take the minimum possible value.

[Editorial note: The Nash-Sutcliffe efficiency is used to good effect in generalising a model
for the baseflow index BFI (see Volume IV). However, NSE is problematic to interpret in
some of the cases arising in IBIDEM. Because it gives no special weight to the quality of fit
around the peak, NSE is not ideal for evaluating the match to the upper hydrograph. The
measure has the minor merit of being independent of the method of fitting used in IBIDEM.]

In the case of Figure 9.11, the fit of the FSR hydrograph is judged to be very poor, with
NSE = -0.70. The difficulty arises largely because the receding limb of the FSR hydrograph
is much steeper than that of the FSU hydrograph once the inflection point four hours after the
peak has been passed. Beyond that time, there is a long period when the FSR hydrograph is
much lower than the FSU hydrograph. The recession limb of the imported hydrograph has
less influence on the IBIDEM fit if the threshold used is raised from 50% to 60% (see Figure
9.12). The NSE becomes a respectable +0.70.

NSE evaluated
across this period
NSE = 0.70

FSR hydrograph Characteristic


fitted by IBIDEM hydrograph by
UPO-ERR-Gamma

Figure 9.11: FSR hydrograph fitted to UPO-ERR-Gamma hydrograph

106
Volume III Hydrograph Analysis

NSE
across
here NSE = +0.70

Figure 9.12: As Figure 9.11 but with fitting threshold raised to 60% of peak flow

[Editorial note: The take-home messages are: (i) Give greater weight to the RMSE measure
than NSE, (ii) Consider the sensitivity of IBIDEM fits to the threshold chosen and (iii) Fitting
horizontally rather than vertically has real merit in hydrograph width modelling!]

9.5.5 Tabular display


Below the graph, IBIDEM tabulates the parameter values and other variables (as shown in
Figure 9.13). After the hydrograph fitting is carried out, the variables shown are: flow return
period, rainfall return period, baseflow (BF), fitted Tp, fitted SPR, PR (based on SPR and
other terms including URBEXT), time-step, storm duration (calculated from Tp and SAAR),
rainfall depth (calculated from storm duration and return period), peak flow (taken from the
imported FSU hydrograph), runoff volume, RMSE and Nash-Sutcliffe efficiency.

Figure 9.13: Example of IBIDEM tabular display

107
Volume III Hydrograph Analysis

Slightly different versions of the table appear in some options. For example, it is not
appropriate to show the goodness-of-fit statistics within the Sensitivity to storm duration
option. The order of the rows in each version of the table remains the same but the variable
that the user has changed (between columns) is highlighted to aid interpretation. For
example, the row containing the flow return period is highlighted in Figure 9.13. Another
feature is that variables not changing between columns are indicated in grey. Baseflow does
not vary with return period, so the BF row is shaded grey in this example.

9.5.6 Export of results


IBIDEM allows export of a summary report in .CSV format, including a record of the FSU
and FSR hydrographs, PCDs used in the calculations, fitted parameters and measures of the
goodness of fit. A different version of the report is available in the Sensitivity to storm
duration option.

9.6 Testing

9.6.1 Choice of test sites


IBIDEM has been tested on a range of catchments and a variety of FSU hydrographs. Design
hydrographs on gauged catchments will usually be based on hydrograph width analysis, with
the characteristic hydrograph derived by the non-parametric method of Chapter 3. At
ungauged sites, the characteristic hydrograph is likely to be based on the UPO-ERR-Gamma
model of Section 4.4 or the parabolic curves method of Section 8.7.

The five test catchments (see Map 9.1) are:

Station 16009 Suir at Caher Park a large rural catchment (1602 km2);
Station 19001 Owenboy at Ballea a small to medium-sized rural catchment
(103 km2);
Station 06026 Lagan-Glyde at Aclint a medium-sized rural catchment (144 km2);
An ungauged site on a medium to large-sized rural catchment (443 km2) on the Anner
(a tributary of the Suir which it joins at Clonmel);
An ungauged site on a small urbanised catchment (8 km2) on a tributary of the Tolka
at Finglas.

9.6.2 Estimation of FSU hydrograph shapes


For the three test sites with flow data, the HWA software was applied to derive the
characteristic hydrograph by the non-parametric method of Chapter 3.

At the two ungauged sites, the characteristic hydrograph was constructed using a version of
the Chapter 6 procedure, i.e. estimating parameters of the UPO-ERR-Gamma model from
PCDs. The alternative method of constructing the upper hydrograph by the parabolic curves
method was also tested. Some details are reported in Table 9.1.

[Editorial note: Testing of IBIDEM was undertaken before HWA recommendations were
finalised, and before it was possible to estimate BFI at ungauged sites using the BFIsoil model
108
Volume III Hydrograph Analysis

of Volume IV. It was therefore necessary to borrow BFI values: from Station 15001 for the
Anner ungauged site and from Station 08005 for the Tolka tributary. The differences in
technique are not thought to compromise the integrity of the testing of IBIDEM.]

Map 9.1: Location of test catchments

Table 9.1: Some details of the applications to two ungauged test catchments
Variable Unit River Anner Tributary of Tolka at Finglas
Physical catchment descriptors
BFI (see editorial note above) 0.51 0.52
FARL 0.999 1.000
ALLUV 0.047 0.000
ARTDRAIN 0.000 0.014
S1085 m/km 3.4 16.1
Parameters of the UPO-ERR-Gamma model
n 7.35 7.23
Tr hours 9.68 12.62
C hours 30.80 32.77
Width descriptors for use of the parabolic curves method
W75 hours 4.18 4.90
W50 hours 7.03 7.57
s (eccentricity parameter) 0.40 0.40

109
Volume III Hydrograph Analysis

9.6.3 Estimation of peak flows


For the test sites with flow data, QMED was estimated as the median of the annual maximum
flows. At the two ungauged test sites, QMED was estimated from PCDs. [Editorial note:
Testing of IBIDEM was undertaken before the Volume II model for estimating QMED from
PCDs was finalised. Indeed, feedback from IBIDEM testing on the tributary of the Tolka led
to changes in the urban adjustment model finally recommended for QMED estimation.]

Design flows for other return periods were estimated by applying a flood growth curve based
on the FSR regional growth curve for Ireland. Due adjustment was made by dividing the T-
year flood growth factor by the 2-year flood growth factor for the different index variable
used in the FSR method. The design flows are shown in Table 9.2. It should be noted that
applications of IBIDEM will generally use design flows estimated by Volume II procedures.

Table 9.2: Design flows (m3s-1) for the five test catchments
Return period (years)
Catchment
2 5 50 100 200
16009 Suir at Caher Park 162 204 303 334 366
19001 Owenboy at Ballea 15.4 19.4 28.8 31.7 34.7
06026 Lagan-Glyde at Aclint 12.9 16.3 24.2 26.6 29.2
Anner at Clonmel 61.5 77.6 115 127 139
Tributary of Tolka at Finglas 1.62 2.05 3.03 3.34 3.66

9.6.4 Rainfall depth-duration frequency tables


IBIDEM requires a table of design rainfall values for a typical point in the catchment. The
user will obtain this through the FSU Web Portal. For the purpose of testing, tables of values
based on the Volume I rainfall frequency procedure were transferred to IBIDEM in CSV files
supplied by Met ireann. In each case (i.e. for each of the test catchments in turn) the table
of design rainfall depths was supplied for a 2-km grid point close to the catchment centroid.
Table 9.3 summarises the main input variables used in the IBIDEM testing.

Table 9.3: IBIDEM input variables for the test catchments


AREA SAAR Characteristic Approx. centroid
Catchment URBEXT
(km2) (mm) hydrograph Easting Northing
Suir at
1602 1079 0.009 200000 146000
Caher Park
Derived median
Owenboy
106 1176 0.018 hydrograph of 162000 62000
at Ballea
Chapter 3
Lagan-Glyde
144 1072 0.007 310000 224000
at Aclint
Anner
443 986 0.003 UPO-ERR-Gamma 224000 134000
at Clonmel
model (Section 4.4
Tributary of + pilot of Chapter 6) 321000
8 734 0.527 240000
Tolka at Finglas
110
Volume III Hydrograph Analysis

The heavily urbanised nature of the tributary of the Tolka at Finglas is to be noted.

9.7 Results
The sections below discuss the hydrograph fit obtained with IBIDEM for each of the test
catchments in turn. Although testing considered a range of return periods, the examples
shown are for the 100-year flood. Results are summarised later in Table 9.4.

9.7.1 Suir at Caher Park

For Station 16009 Suir at Caher Park, the HWA used hydrographs from 54 flood events. For
this catchment, the prescribed baseflow by the FSR rainfall-runoff method was 51.0 m3s-1.
This was slightly lower than the minimum flow in the imported hydrograph. After IBIDEM
provided a warning, the baseflow was reduced to 50 m3s-1 to overcome the difficulty.

The IBIDEM 100-year hydrograph is shown in Figure 9.14, based on the default fitting
threshold of 50% of the peak flow. The FSU hydrograph has a rapid rising limb but a much
slower recession limb. The FSR rainfall-runoff method results in only a limited range of
hydrograph shapes, and cannot capture this feature of the FSU hydrograph. However, the
RMSE criterion used by IBIDEM to optimise the fit ensures that the widths of the two
hydrographs are broadly similar, on average, for flows above the threshold.

The RMSE for this site is 12.7 hours: the largest for any of the test sites (see Table 9.4). This
reflects the propensity of the catchment to produce flood hydrographs of long duration.
Those interested in the Suir should also refer to Section 5.8. It is notable that the FSR
hydrograph overestimates widths near the peak and underestimates widths at lower flows.
The Nash-Sutcliffe efficiency (NSE) is 0.50, indicating a fairly good fit in terms of flow.

The fitted Tp is 52.8 hours and the SPR is 36.7%. These values seem reasonable for this
large catchment of moderate permeability (BFI = 0.63).

IBIDEM
hydrograph Derived median
hydrograph RMSE = 12.7 hours

NSE = 0.50

Figure 9.14: Suir at Caher Park 100-year hydrograph fit

111
Volume III Hydrograph Analysis

9.7.2 Owenboy at Ballea


For Station 19001 Owenboy at Ballea, the HWA used hydrographs from 35 flood events.
The IBIDEM 100-year hydrograph is shown in Figure 9.15, based on the default fitting
threshold of 50% of the peak flow. As for the Suir at Caher Park, the FSU hydrograph is
more skewed than the FSR one, in that it rises rapidly and falls more slowly.

The RMSE for this site is 6.1 hours. The Nash-Sutcliffe efficiency is -0.48. The poor
performance in terms of NSE arises because the statistic is calculated as an average over the
duration of the FSU hydrograph, i.e. from about 10 hours before the peak to 55 hours after.
The statistic could be improved by entering a higher value for baseflow (BF) or by raising the
threshold used in fitting.

The fitted Tp is 31.1 hours and the SPR is 29.9%. These values seem reasonable for this
moderate-sized catchment of moderate permeability (BFI = 0.64), although the Tp is perhaps
rather long, as can be seen from the hydrograph plot. However, a shorter Tp would give a
narrower hydrograph and hence a poorer fit in terms of hydrograph widths.

IBIDEM Derived median RMSE = 6.1 hours


hydrograph hydrograph
NSE = -0.48

Figure 9.15: Owenboy at Ballea 100-year hydrograph fit

9.7.3 Lagan-Glyde at Aclint

Station 06026 Lagan-Glyde at Aclint used hydrographs from 31 flood events and provides an
example of a case where basic use of HWA yields a derived median hydrograph with time
reversals. One of those on the rising limb is at quite a high level, between 70 and 75% of
the peak flow (see Figure 9.16). Optionally, these can be removed using interactive features
within the HWA software.

The fitting method used in IBIDEM can in fact cope with such time reversals, as shown in the
LH plot of Figure 9.17. However, users may be reluctant to present a hydrograph in which
time appears to run backwards! The anomaly can be avoided by editing the FSU hydrograph
within the HWA software or en route to IBIDEM.

112
Volume III Hydrograph Analysis

Time reversal

Figure 9.16: Lagan-Glyde at Aclint derived median hydrograph from HWA software

IBIDEM Derived median IBIDEM DMH edited to


hydrograph hydrograph hydrograph remove time reversal

Figure 9.17: Lagan-Glyde at Aclint 100-year hydrograph fits

Incompleteness of the imported hydrograph prompts IBIDEM to issue a warning where


appropriate. A typical message is: Threshold flow is less than 100 yr input hydrograph
start/end values, please adjust. Many users would instinctively raise the fitting threshold
from the default value of 50% to 75%. However, as shown in the RH plot of Figure 9.17, it is
permissible to adopt an intermediate threshold at which the hydrograph width (in the
imported FSU hydrograph) is defined on one limb of the hydrograph but not the other. This
allows the fitting to exploit the width information on the receding limb between 70 and 75%
of the peak flow.

IBIDEM gives a good fit to the rising and falling limbs of the FSU hydrograph, above the
threshold value used for fitting. The RMSE is 11.6 hours and the Nash-Sutcliffe efficiency is
0.64, indicating a moderately good fit in terms of flow.

The fitted Tp is 97.3 hours. This is a surprisingly long time-to-peak (of the unit hydrograph)
given the modest size of the catchment (144 km2). The typically slow flood response may in
part reflect the influence of loughs in the catchment (FARL is 0.91).

The fitted SPR is 48.8%. The BFI for this catchment is 0.66, indicating a moderately
permeable catchment. The fitted SPR value is surprisingly high for such a catchment. The
high value of SPR is partly explained by the long time-to-peak which tends to produce a
subdued hydrograph with a relatively low peak. IBIDEM has increased SPR in
compensation, in order to ensure that the FSU peak flow is matched.
113
Volume III Hydrograph Analysis

9.7.4 Anner at Clonmel


Two sets of calculations were carried out for this ungauged catchment: one with the
characteristic hydrograph constructed using the UPO-ERR-Gamma model and the other using
the parabolic curves method. These are shown in the Figure 9.18, along with hydrographs
fitted by IBIDEM.

The FSR hydrograph fitted by IBIDEM to the FSU hydrograph in UPO-ERR-Gamma form is
shown in the left-hand side of Figure 9.18. The fit is seen to be fairly good, the main defect
being that the FSR rainfall-runoff method cannot reproduce the sudden change of gradient on
the falling limb when the Gamma curve is replaced by the exponential recession. The RMSE
is 1.9 hours. The Nash-Sutcliffe efficiency is a very poor -0.70. This reflects the departure
of the IBIDEM and UPO-ERR-Gamma hydrographs in the early and (especially) later part of
the period over which the fitting is made. (As explained in Section 9.5.4, NSE is evaluated
from vertical differences throughout the period for which the FSU hydrograph exceeds the
fitting threshold.)

UPO-ERR-Gamma Parabolic curves


IBIDEM hydrograph IBIDEM hydrograph
hydrograph hydrograph

Figure 9.18: Anner at Clonmel 100-year hydrograph fit

IBIDEM provides a very good fit to the FSU hydrograph in parabolic form (see RH side of
Figure 9.18), with RMSE = 1.0 hours and NSE = 0.85.

Interpretation of the Tp and SPR values fitted by IBIDEM is revealing for this catchment.
The fitted Tp is 2.8 hours for the parabolic hydrograph and 7.1 hours for the Gamma
hydrograph. These times to peak (particularly 2.8 hours) seem unreasonably short for this
medium to large rural catchment of 443 km2. Fitted SPR values are 0.4% for the parabolic
hydrograph and 2.3% for the Gamma hydrograph. IBIDEM helpfully warns that these SPR
values are suspiciously low. An SPR of 0.4% is unreasonably low, because it implies
virtually none of the storm rainfall typically becomes rapid response runoff. The very small
value is probably due in part to the need to compensate for the excessively small value of Tp.
IBIDEM has had to decrease SPR in order to ensure that the FSU peak flow is matched.

[Editorial note: Multipliers in the PCD-based models for the hydrograph width parameter
Tr and the descriptors W75 and W50 supplied for testing were much too small due to the use
of non-standard units for the mainstream slope descriptor S1085. A feature of the FSR
rainfall-runoff method is that underestimation of Tp leads to underestimation of SPR also. It
is helpful that IBIDEM helpfully warns the user when SPR values are suspiciously low. See
also the editorial note in Section 9.7.5.]
114
Volume III Hydrograph Analysis

9.7.5 Tributary to Tolka at Finglas


Two sets of calculations were carried out for this urbanised but ungauged catchment: one
with the characteristic hydrograph constructed using the UPO-ERR-Gamma parametric
model and the other using the width descriptors W75 and W50 and the parabolic method.
These are shown in the Figure 9.18, along with hydrographs fitted by IBIDEM.

Because this small catchment is heavily urbanised, the urban catchment design package was
applied within IBIDEM. This means that the 100-year return period rainfall was used to
synthesise the 100-year flood hydrograph, and the 50% summer rainfall profile was adopted.

As on the River Anner, the fitted hydrograph matches the FSU shape fairly well for the UPO-
ERR-Gamma hydrograph and very closely for the parabolic hydrograph. RMSE and NSE
values can be seen in the final columns of Table 9.4, which summarises results for all five
test catchments.

IBIDEM Parabolic
IBIDEM UPO-ERR-Gamma hydrograph hydrograph
hydrograph

Figure 9.19: Tributary of Tolka at Finglas 100-year hydrograph fit

Fitted Tp values are 9.3 hours for the UPO-ERR-Gamma hydrograph and 4.2 hours for the
parabolic hydrograph. These values seem suspiciously long for a small heavily urbanised
catchment.

For reasons explained in Section 9.3.1, there is interaction between the Tp and SPR
parameters that IBIDEM obtains when fitting the FSR rainfall-runoff method to the design
flood hydrograph that the user has derived by FSU methods. IBIDEM always respects the
peak flow of the imported FSU hydrograph. Whilst the parabolic curves method appears
preferable to the UPO-ERR-Gamma model in terms of the Tp values resulting for this
catchment, this preference is reversed when the SPR values are considered (see final columns
of Table 9.4). The fitted SPR for the parabolic hydrograph is 6.3% which is unusually low.

[Editorial note: As discussed in the editorial note in Section 9.7.4, the developer was
supplied with incorrect multipliers in the models for the hydrograph width parameter Tr and
the descriptors W75 and W50. Because the PCD-based models had grossly underestimated
hydrograph width at both sites used in testing (i.e. the Anner and the Tolka tributary), the
IBIDEM developers understandably cast around for a possible explanation. Indeed they
expressed surprise that none of the regression models for Tr, W75 or W50 includes a term that
reflects catchment size. Such PCDs were considered in the generalisation research of

115
Volume III Hydrograph Analysis

Chapter 6 but found useful only once: AREA appears in the regression model for parameter
C of the UPO-ERR-Gamma model in the case where BFI is unavailable (see Table 6.7).

A characteristic feature of flood response times in Ireland and Great Britain is that catchment
size often plays a much smaller role than the hydraulically-minded expect. Although
mainstream length (MSL) appeared in the FSSR16 model for Tp(0), it did so only to a
modest exponent of 0.23. This means that the estimated value of Tp(0) doubles only if the
mainstream is 20 times longer. Catchment size is a very poor guide to flood response time!

It is reiterated that the poor results obtained in IBIDEM testing on the Anner and the Tolka
tributary arose largely because the developers had been supplied with PCD models for Tr,
W75 and W50 that had faulty multipliers.]

Table 9.4: Summary of IBIDEM results for five test catchments (100-year flood case)
* *
Anner Tolka
Variable Suir Owenboy Lagan
Gamma Parabolic Gamma Parabolic
Rainfall return
140 140 140 140 140 100 100
period (years)
BF (m3s-1) 50 3.69 4.56 12.85 12.85 0.13 0.13
Tp (hours) 52.8 31.1 97.3 7.1 2.8 9.3 4.2
SPR 36.7 29.9 48.8 2.3 0.4 25.3 6.3
PR 46.8 40.2 60.7 10.9 6.2 38.1 22.1
Time-step
1 1 1 1 0.25 1 0.25
(hours)
Storm duration
111 69 203 15 5.75 17 7.25
(hours)
Rainfall depth
122.2 126.9 149.4 102.1 70.8 85.2 64.8
(mm)
Peak flow
334.0 31.7 26.6 126.9 126.9 3.3 3.3
(m3s-1)
Volume (mm
of catcht 84.6 69.9 141.3 15.0 5.9 24.4 10.7
runoff)
RMSE (hours) 12.68 6.09 11.58 1.88 1.02 1.69 0.48
NSE 0.50 -0.48 0.64 -0.70 0.85 0.03 0.88
*
Editorial note: Test results for the Anner and Tolka catchments are in error due to incorrect
multipliers in the PCD-based models originally supplied for Tr, W75 and W50

9.7.6 Illustration of effect of fitting threshold


Tests were carried out using a wide range of thresholds. The effect of varying the threshold
is illustrated for the ungauged site on the River Anner, for an FSU hydrograph constructed
using the UPO-ERR-Gamma model of hydrograph width. The fitting threshold is marked on
each plot in Table 9.5, and a commentary provided.

116
Volume III Hydrograph Analysis

[Editorial note: The option to vary the fitting threshold is an exceedingly valuable feature of
IBIDEM. It allows the user to focus on the hydrograph features most relevant to their
application. In some cases, only the upper hydrograph may be relevant; in others, the entire
hydrograph is important. It is important that the user inspects and interprets values of Tp and
SPR thoroughly. Otherwise, a more hydrologically-informed analyst may later judge the
choice of fitting threshold to have been made as a matter of convenience.]

9.7.7 Summary
The tests led to improvements in IBIDEM and to timely feedback to other parts of the FSU
research. IBIDEM gave broadly sensible results on the three gauged catchments. Tp and
SPR values were within expected ranges on the Suir and Owenboy. The inferred Tp was
longer than expected for the Lagan-Glyde, with SPR taking on a high value in compensation.

As noted above, results for the two ungauged catchments were undermined by the supply of
PCD-based models for Tr, W75 and W50 with incorrect multipliers.

Even when the correct models are used for estimating the characteristic hydrograph from
PCDs, users will find cases where the Tp and SPR values inferred when IBIDEM fits the
FSR rainfall-runoff method to the imported FSU hydrograph are unrealistic with perceived
properties of the catchment. There is no simple recipe to deal with such cases. It is largely a
matter of experience.

It is further emphasised that the FSU recommendation is to base flood estimates wherever
possible on data transfers rather than on PCDs alone. Implementation of the FSU research
stresses the importance of choosing a pivotal catchment so that estimates at ungauged sites
gain from knowledge of flood behaviour at gauged sites. Section 8.10 discusses the
selection and use of the pivotal catchment in the context of estimating the characteristic
hydrograph at an ungauged site.

9.8 Additional opportunities provided by IBIDEM

9.8.1 Strengths and limitations


IBIDEM provides a way of assessing FSU outputs using a structured model of hydrograph
formation. The test results above illustrate that the software can help in detecting hydrograph
shapes or peak flows that appear to be inconsistent with properties of a catchment.

A limitation is that IBIDEM relies on the assumptions made in the FSR design event
approach. These may not always be appropriate. The design event method derives a flood
hydrograph from a single combination of inputs (rainfall depth, rainfall duration, rainfall
profile and catchment wetness index). This combination does not always result in a
hydrograph peak of the required return period. Thus, the design event used by IBIDEM may
not always be relevant to the supplied FSU design hydrograph. The implication for users is
that, while IBIDEM can provide a useful diagnostic test of the FSU design hydrograph, the
method is not a cure-all.

117
Volume III Hydrograph Analysis

Table 9.5: Sensitivity to fitting threshold (ungauged site on River Anner)

Threshold
Performance
used in Hydrograph plot Comment
statistics
fitting

Threshold

RMSE = 0.31 hr Good fit to widths for


90%
small segment above
of peak flow threshold
NSE = -1.29

Threshold

RMSE = 0.19 hr Very good fit to widths


70%
above threshold (hence
of peak flow small RMSE)
NSE = -1.42

RMSE = 1.88 hr Larger RMSE indicates


50%
Threshold inferior fit to widths but
smaller NSE indicates
of peak flow NSE = -0.70
better fit in terms of
flow

RMSE = 5.80 hr
Continuation of trends
30% noted above ( when
of peak flow 50% of peak flow used
Threshold NSE = -0.50
as threshold)

118
Volume III Hydrograph Analysis

9.8.2 Urban adjustment to design hydrographs


The options described in Section 9.4 provide additional functionality. A potential further
application of IBIDEM is in estimating improved design hydrographs on urban catchments.

One possibility is to use FSU methods to derive the design flood hydrograph as if the
catchment were entirely rural, by omitting the urban adjustment to QMED. The hydrograph
would then be imported to IBIDEM, with URBEXT set to zero. After carrying out the
fitting, the user would select the option to test sensitivity to urbanisation, and enter the correct
value of URBEXT. The adjusted FSR hydrograph obtained is this way provides an
alternative allowance for urbanisation. A potential advantage of this approach is that it
adjusts the hydrograph widths for urbanisation, not just the peak flow. This is possible
because the FSR rainfall-runoff method incorporates an urban adjustment to both Tp and PR.

Figure 9.20 illustrates the outcome for the heavily urbanised ungauged tributary of the Tolka
at Finglas considered in Section 9.7.5. The adjusted hydrograph has a peak flow over four
times larger than the original one, and the time to peak is much shorter. The increase in peak
flow is caused by two effects: the increase in percentage runoff and the decrease in time to
peak (and consequently storm duration). The increase in peak flow thus synthesised is much
greater than that provided by the Volume II urban adjustment to QMED. [Editorial note:
Details of this example may have been compromised by the supply of PCD-based models for
Tr, W75 and W50 with incorrect multipliers.]

Imported FSU hydrograph


Calculated FSR hydrograph
Adjusted FSR hydrograph

Figure 9.20: Urban adjustment to hydrograph for Tolka tributary at Finglas test site

9.8.3 Supplying input hydrographs to river models


IBIDEM also has potential in deriving inflows for hydrodynamic river models, which often
require hydrographs resulting from design storms with durations different from the critical

119
Volume III Hydrograph Analysis

duration of the particular subcatchment. These hydrographs can be generated using the
option to test sensitivity to storm duration.

Hydrographs generated by IBIDEM are likely to be useful for flood storage and flood routing
studies because they cover the full range of flows. Although the non-parametric method of
Chapter 3 is strongly recommended for use at gauged sites, the characteristic hydrograph it
produces does not extend all the way down to zero flow. IBIDEM provides a structured
alternative to sketching in the lower part of the hydrograph by hand.

9.8.4 Allowances for projected land-use change


IBIDEM provides a route to assessing the effect of future changes in urbanisation or other
land use by varying the parameters of the rainfall-runoff method. The relevant options have
been discussed in Sections 9.4.3 and 9.4.4.

120
Volume III Hydrograph Analysis

Acknowledgements
The hydrograph width analysis was undertaken at the Department of Engineering Hydrology
in the National University of Ireland Galway, principally by Kieran OConnor and Monomoy
Goswami. Samiran Das undertook some testing of the methods.

IBIDEM was developed by JBA Consulting: principally by Zo Whiteman and Duncan


Faulkner.

The flood event analysis summarised in Appendix B was undertaken at University College
Cork.

The help of organisations and individuals who gather and curate hydrograph data is gratefully
acknowledged. Peter Newport of OPW is especially thanked for supplying extensive data.

Volume III was edited by Duncan Reed of DWRconsult.

References
Archer, D., Foster, M., Faulkner, D. and Mawdsley, H. 2000. The synthesis of design flood
hydrographs. Proc. Flooding: risks and reactions. CIWEM/ICE Conference, London,
October 5, 2000, 4557.
Bhattarai, K.P. and OConnor, K.M., 2004. The effects over time of an arterial drainage
scheme on the rainfall-runoff transformation in the Brosna catchment. Physics and Chem. of
the Earth, 29: 787794.
Brace, N., Kemp, R. and Snelgar, R., 2003. SPSS for psychologists: a guide to data analysis
using SPSS for Windows. Lawrence Erlbaum Associates.
Brune, A. 2007. Comparison of the hydrological behaviour of various Irish catchments.
M.Eng.Sc. thesis, Dept of Civil and Environmental Engg University College Cork, 200pp.
Duan, Q. 2003. Global optimization for watershed model calibration. In: Calibration of
watershed models (eds Duan, Q, Gupta, H.V., Sorooshian, S., Rousseau, A.N. and Turcotte,
R.), Water Science and Application 6, American Geophysical Union, Washington, 89104.
Elshorbagy, A., Simonovic, S.P. and Panu, U.S. 2000. Performance evaluation of artificial
neural networks for runoff prediction. J. Hydrol. Engg, 5: 424427.
IH, 1979. Short cut to unit hydrograph convolution. Flood Studies Supplementary Report 9,
Institute of Hydrology, Wallingford, UK, 6pp.
IH, 1985. The FSR rainfall-runoff model parameter estimation equations updated. Flood
Studies Supplementary Report 16, Institute of Hydrology, Wallingford, UK, 5pp.
Goldberg, D.E. 1989. Genetic Algorithms in search, optimization and machine learning.
Addison-Wesley, 432pp.
Hayashi T., Nagamine, Y. and Nishida, A., 1986. A vibration model to describe the lactation
curve of a dairy cow. Japanese J of Zootechnical Science, 57: 471478.
Holder, R.L. 1985. Multiple regression in hydrology. Institute of Hydrology, Wallingford,
UK, 147pp.

121
Volume III Hydrograph Analysis

Holland, J.H. 1975. Adaptation in natural and artificial systems. Ann Arbor: University of
Michigan Press.
Houghton-Carr, H. 1999. Restatement and application of the Flood Studies Report rainfall-
runoff method. Volume 4, Flood Estimation Handbook, CEH Wallingford, 288pp.
Kalinin, G.P. and Milyukov, P.I. 1957. O raschete neustanovivshegosya dvizheniya vody v
otkrytykh ruslakh (On the computation of unsteady flow in open channels). Met. i
Gydrologia Zhurnal, 10: 1018.
Martin, J., OKane, J.P. and Javan, M. 2000. Computer modelling for flood alleviation in the
lower Feale catchment. Proc. Water in the Celtic world: managing resources for the 21st
century, Aberystwyth, 3-7 July 2000, BHS Occasional Paper No. 11, 163172.
Nash, J.E. 1957. The form of the instantaneous unit hydrograph. In: Proc. Toronto General
Assembly, IASH Publ. No. 45, 114121.
Nash, J.E. and Sutcliffe, J.V., 1970. River flow forecasting through conceptual models, Part
1: A discussion of principles. J. Hydrol., 10: 282290.
NERC 1975. Flood Studies Report (5 volumes). Natural Environment Research Council,
London, UK.
Pallant, H., 2001. SPSS Survival Manual. A step by step guide to data analysis using SPSS
for Windows. Open University Press, Buckingham, the UK.
Reed, D.W. and Marshall, D.C.W. 1999. Defining a design hydrograph. Chapter 10,
Volume 3, Flood Estimation Handbook, CEH Wallingford: 5962.
Tabachnick, B. G. and Fidell, L. S. 2001. Using multivariate statistics. 4th ed., Allyn and
Bacon, Needham Heights, Massachusetts, 966pp.
Wang, Q.J., 1991. The Genetic Algorithm and its application to calibrating conceptual
rainfall-runoff models. Water Resources Research, 27(9), 2467-2471.

122
Volume III Hydrograph Analysis

Appendices

Appendix A Gauges used in Hydrograph Width Analysis


Table A.1: Stations used in Hydrograph Width Analysis
Station
Station River Name Area (km2) River Basin District
grade
06011 Fane Moyles Mill A1 229.2 Eastern
06012 Fane Clarebane A1 162.80 Eastern
06013 Dee Charleville A1 309.1 Eastern
06014 Glyde Tallanstown A1 270.4 Eastern
06026 Lagan (Glyde) Aclint A1 148.4 Eastern
07001 Tremblestown Tremblestown A2 151.3 Eastern
07002 Deel Killyon A2 285.0 Eastern
07004 Kells Blackwater Stramatt A2 245.7 Eastern
07006 Moynalty Fyanstown A2 177.5 Eastern
07007 Boyne Aqueduct A1 441.2 Eastern
07009 Boyne Navan Weir A1 1658.2 Eastern
07010 Blackwater Liscartan A1 699.7 Eastern
07011 Kells Blackwater ODalys Br A2 281.7 Eastern
07012 Boyne Slane Castle A1 2460.3 Eastern
07033 Kells Blackwater Virginia Hatchy A2 124.9 Eastern
09001 Ryewater Leixlip A1 209.6 Eastern
11001 Owenavorragh Boleany B 155.1 South-Eastern
14004 Figile Clonbulloge A1 268.9 South-Eastern
14006 Barrow Pass Bridge A1 1063.6 South-Eastern
14007 Stradbally Derrybrock A1 118.6 South-Eastern
14009 Cushina Cushina A2 68.4 South-Eastern
14011 Slate Rathangan A1 162.3 South-Eastern
14018 Barrow Royal Oak A1 2419.4 South-Eastern
15001 Kings Annamult A2 444.3 South-Eastern
15002 Nore Johns Br A2 1644.1 South-Eastern
15003 Dinin Dinin Br A2 299.2 South-Eastern
15005 Erkina Durrow Foot Br B 379.4 South-Eastern
15006 Nore Brownbarn A2 2418.3 South-Eastern
16001 Drish Athlummon A2 135.1 South-Eastern
16002 Suir Beakstown A2 485.7 South-Eastern
16003 Clodiagh Rathkennan A2 243.2 South-Eastern
16004 Suir Thurles A2 228.7 South-Eastern
16005 Multeen Aughnagross A2 84.0 South-Eastern

123
Volume III Hydrograph Analysis

Station
Station River Name Area (km2) River Basin District
grade
16008 Suir New Bridge A2 1090.3 South-Eastern
16009 Suir Caher Park A2 1582.7 South-Eastern
18004 Awbeg Ballynamona A2 310.3 Southern
18005 Funshion Downing Br A2 378.5 Southern
19001 Owenboy Ballea A2 103.3 Southern
22071 L. Leane Tomies Pier A2 557.7 Southern
23001 Galey Inch Br A2 191.7 Mid-Western
23002 Feale Listowel A1 646.8 Mid-Western
23012 Lee (Kerry) Ballymullen A2 61.6 Mid-Western
24001 Maigue Croom A2 770.2 Mid-Western
24008 Maigue Castleroberts A2 806.0 Mid-Western
24013 Deel Rathkeale A1 438.8 Mid-Western
24082 Maigue Islandmore A2 762.8 Mid-Western
25001 Mulkear Annacotty A2 647.6 Shannon
25003 Mulkear Abington A1 399.1 Shannon
25005 Dead Sunville A2 192.6 Shannon
25006 Brosna Ferbane A1 1162.8 Shannon
25014 Silver Millbrook A1 164.4 Shannon
25016 Clodiagh Rahan A2 275.2 Shannon
25017 Shannon Banagher A1 7980.4 Shannon
25025 Ballyfinboy Ballyhooney A1 161.2 Shannon
25027 Ollatrim Gourdeen A1 118.9 Shannon
25029 Nenagh Clarianna A2 292.7 Shannon
25030 Graney Scarrif A1 280.0 Shannon
26002 Suck Rookwood A2 641.5 Shannon
26005 Suck Derrycahill A2 1085.4 Shannon
26007 Suck Bellagill A1 1207.2 Shannon
26008 Rinn Johnstons Br A1 280.3 Shannon
26009 Black Bellantra Br A2 98.2 Shannon
26012 Boyle Tinacarra A1 519.9 Shannon
26019 Camlin Mullagh A1 253.0 Shannon
26021 Inny Ballymahon A2 1098.8 Shannon
26022 Fallan Kilmore A2 61.9 Shannon
27001 Claureen Inch Br A2 46.7 Mid-Western
27002 Fergus Ballycorey A1 564.3 Mid-Western
29001 Raford Rath-gorgin A1 115.5 Western
29004 Clarinbridge Clarinbridge A2 121.4 Western
29011 Dunkellin Kilcolgan A1 354.1 Western
124
Volume III Hydrograph Analysis

Station
Station River Name Area (km2) River Basin District
grade
30004 Clare Corrofin A1 699.2 Western
30005 Robe Foxhill A1 237.8 Western
30007 Clare Ballygaddy A2 469.9 Western
30061 Corrib Estuary Wolfe Tone Br A2 3136.1 Western
34001 Moy Rahans A2 1974.8 Western
34009 Owengrave Curraghbonaun A2 117.1 Western
34018 Castlebar Turlough A1 95.4 Western
35001 Owenmore Ballynacarrow A2 299.4 North-Western
35002 Owenbeg Billa Br A2 88.8 North-Western
35005 Ballysadare Ballysadare A2 639.7 North-Western
35071 L. Melvin Lareen A2 247.2 North-Western
36010 Annalee Butlers Br A1 771.7 North-Western
36011 Erne Bellahillan B 320.5 North-Western
36015 Finn Anlore A1 153.1 North-Western
36019 Erne Belturbet A2 1491.8 North-Western
36021 Yellow Kiltybarden A2 23.4 North-Western
36027 Woodford Bellaheady A2 333.8 North-Western
39009 Fern O/L Aghawoney A2 1974.8 North-Western

Table A.2: Details of the flow data used (see also Table 7.2)
Flow data studied Period of
% of 15-min # of AM QMED
Station arterial drainage
From To data missing events (m3s-1)
works
06011 01/10/1972 01/02/2001 1.6 29 15.45
06012 01/10/1972 11/01/2004 0.7 32 12.5
06013 29/10/1975 21/12/2004 1.5 30 27.75
06014 23/10/1975 22/10/2002 2.2 28 21.06 1950-57
06026 01/01/1972 01/02/2001 1.6 29 12.33 1950-57
07001 21/05/1975 30/09/2001 3.7 26 19.95 1971-73
07002 01/10/1970 01/09/2004 17.4 30 18.88
07004 26/10/1982 30/09/2005 0.9 23 22.56
07006 05/11/1956 01/10/2005 9.8 46 19.8
07007 09/04/1979 05/04/2004 2.1 25 35.41 1973-78
07009 05/11/1976 03/10/2005 0.8 30 144.91
07010 08/12/1986 21/05/2003 2.7 17 69.61 1982-86
07011 21/12/1983 30/09/1998 2.9 15 31.95 1980-82
07012 01/10/1986 01/01/2006 0.5 20 261.05 1969-86
07033 10/01/1980 01/10/2005 0.9 26 13.32
09001 15/10/1956 01/01/2006 5.8 50 33.77
125
Volume III Hydrograph Analysis

Flow data studied Period of


% of 15-min # of AM QMED
Station arterial drainage
From To data missing events (m3s-1)
works
11001 07/03/1972 17/11/2005 4.2 34 45.7
14004 01/01/1972 01/07/2001 0.1 29 21.7
14006 01/01/1972 02/01/2006 0.4 34 78.34
14007 01/02/1980 01/10/2001 1.5 22 15.21
14009 01/01/1980 01/11/1999 2.7 20 6.78
14011 27/09/2000 12/11/2004 3.0 5 11.08
14018 01/01/1972 10/10/2005 1.5 34 147.91
15001 01/01/1972 01/08/2005 1.9 33 84.51
15002 01/10/1965 24/08/2001 1.3 36 198.19
15003 01/01/1972 01/05/2005 3.2 33 145.27
15005 01/01/1972 21/03/2005 1.3 33 27.68
15006 01/01/1972 01/01/2006 0.6 34 294.38
16001 01/10/1972 28/04/2005 7.2 33 15.41
16002 01/10/1954 21/11/2001 1.4 48 53.14
16003 01/10/1954 21/11/2001 1.4 48 29.07
16004 01/10/1954 13/06/2000 1.5 46 21.03
16005 01/10/1954 30/09/2001 16.0 47 20.4
16008 01/10/1954 01/06/2005 4.3 51 92.02
16009 14/01/1940 30/09/2004 0.5 64 163.63
18004 01/01/1972 23/04/2005 1.1 33 31.02
18005 01/01/1972 29/07/2002 0.2 30 54.83
19001 01/10/1972 01/01/2005 3.7 33 15.47
22071 01/10/1973 30/09/2004 5.2 31 105.20
23001 01/01/1960 20/12/2004 2.8 45 105.04
23002 01/10/1959 06/01/2006 7.6 46 371.84 1951-59
23012 04/04/1974 31/12/1991 4.6 18 15.88
24001 21/10/1976 01/01/2006 0.9 30 107.84 1975-76
24008 28/11/1973 01/01/2006 6.2 33 117.24
24013 01/10/1972 23/04/2003 4.9 31 108.99
24082 26/10/1977 21/02/2001 6.2 25 138.93 1975-76
25001 20/10/1977 01/01/2004 7.5 27 125.27
25003 03/04/1995 01/01/2002 4.2 7 66.71
25005 01/10/1972 01/04/1999 65.1 11 29.17
25006 05/09/1953 26/09/2005 1.7 52 81.04 1948-53
25014 01/10/1972 19/09/2005 3.2 33 16.52 1948-56
25016 01/10/1951 29/05/2005 3.1 54 24.31 1948-53
25017 02/01/1989 20/03/2003 2.1 14 481.44

126
Volume III Hydrograph Analysis

Flow data studied Period of


% of 15-min # of AM QMED
Station arterial drainage
From To data missing events (m3s-1)
works
25025 01/10/1972 17/08/2003 6.9 31 9.85
25027 01/01/1972 15/01/2002 2.9 30 23.5 1955-65
25029 01/01/1972 01/01/2005 9.7 33 52.81
25030 01/10/1972 15/10/2005 4.3 34 41.66
26002 01/10/1972 01/11/2005 7.0 34 55.14
26005 01/10/1954 01/01/2003 1.7 49 94.56
26007 01/10/1972 13/08/2003 2.6 31 85.95
26008 26/09/1979 08/08/2003 3.7 24 23.69
26009 01/10/1972 18/02/2002 0.4 30 13.02
26012 01/01/1991 22/01/2002 3.0 11 46.72 1982-92
26019 16/09/1953 21/01/2002 4.4 49 21.04
26021 01/10/1972 31/07/2003 73.0 10 64.84
26022 01/01/1972 22/01/2002 10.1 30 6.35
27001 01/10/1972 23/04/2003 8.1 31 20.03
27002 03/05/1954 31/12/2005 1.8 52 32.22
29001 07/10/1957 01/01/2001 20.0 36 13.03
29004 10/07/1973 02/01/1986 1.8 13 11.24
29011 11/02/1983 06/01/2003 2.6 20 29.26
30004 01/10/1964 01/01/2005 9.9 41 89.83 1958-64
30005 01/10/1978 31/12/2004 10.3 26 37.89 1973-78
30007 20/11/1974 30/06/2005 3.6 31 64.13
30061 01/10/1950 12/02/2004 6.1 54 228.44
34001 01/10/1972 01/01/2006 1.8 37 174.27 1960-71
34009 01/01/1972 01/01/2000 4.5 28 28.05
34018 12/08/1976 27/04/2004 3.6 28 11.51 1960-71
35001 08/11/1955 01/10/1997 3.7 43 31.34
35002 19/01/1972 12/11/2002 3.1 31 52.88
35005 01/01/1946 30/09/2004 17.2 51 74.92
35071 04/12/1974 15/01/2003 4.0 29 26.69
36010 01/10/1972 01/11/1998 0.3 27 61.61
36011 01/10/1972 01/07/1998 2.1 26 18.32
36015 29/10/1956 01/02/2001 0.9 45 22.41
36019 19/12/1597 01/09/1998 0.3 41 88.75
36021 14/03/1978 01/12/1999 0.6 22 23.59
36027 08/08/1974 23/10/1992 0.5 19 25.12
39009 05/09/1972 09/01/1982 0.0 10 43.24

127
Volume III Hydrograph Analysis

Appendix B Prcis of UCC research on flood event analysis


University College Cork (UCC) analysed rainfall-runoff behaviour for a range of Irish
catchments. The research was intended to remind FSU users that whilst river floods in the
temperate climate of Ireland generally arise from heavy rainfall there are important
differences between catchments in the conditions that give rise to flooding.

Some details of the 12 catchments studied are given in Table B.1. Nine of the catchments are
listed in ascending order of size. The remaining three form a nested set on the Munster
Blackwater. The catchments range in size from 15 km2 for the Dripsey at Coachford to 1605
km2 for the Nore at Johns Bridge. Mainstream slopes (S1085) range from 0.32 m km-1 for
the Fergus at Ballycorey to 10.3 m km-1 for the Dripsey. Average annual rainfalls range from
913 mm for the Boyne at Trim to 1470 mm for the Dripsey.

[Editorial note: Brune (2007) expands the work by considering a further ten catchments,
chiefly in the north-west and in the greater Dublin area.]

Table B.1: Stations subjected to rainfall-runoff analysis


S1085 Sites from which
AREA SAAR
Station (m hourly rainfall data Comment
(km2) (mm)
km-1) taken
Dripsey at
15 10.3 1470 Dripsey Upland
Coachford
19001 Owenboy at
106 3.79 1248 Cork Airport 100% rural
Ballea
25014 Silver at
165 5.55 992 Birr Rural
Millbrook
07002 Deel at 1975, post-
285 12.72 960 Mullingar
Killyon drainage
06013 Dee at Dundalk +
307 2.37 1096
Charleville Weir Bailiborough
27002 Fergus at Karst +
562 0.32 1252 Shannon Airport
Ballycorey minor lakes
16008 Suir at
1120 0.96 1030 Bansha* + Dundalk*
Newbridge
07005 Boyne at 1975, post-
1282 0.43 913 Mullingar + Navan
Trim drainage
15002 Nore at Kilkenny + Coon +
1605 0.85 979 Up to 2002
Johns Bridge Oakpark
18050 Blackwater Nested
245 3.9 1456 Millstreet + 32**
at Duarrigle (upper)
18048 Blackwater Millstreet + Nested
881 2.7 1356
at Dromcummer Freemount + 32** (middle)
18006 Blackwater Millstreet + Free + Nested
1186 2.1 1303
at Mallow Mallow + 32** (lower)

UCC research catchment, subcatchment of Station 19028 Dripsey at Dripsey
*
Daily data only
**
Special network of 32 raingauges in Blackwater catchment, monitored by UCC for 2005-2006

128
Volume III Hydrograph Analysis

Volume IV presents physical catchment descriptors (PCDs) developed for general use in the
FSU. UCC attempted a more detailed description of the catchments. The overall goal was to
illustrate rather than codifying flood event analysis in Ireland.

Twelve flood events were analysed for each of the 12 catchments. The times series data
analysed are river flows and rainfall depths drawn from several years of record. The main
flood event analysis undertaken was based on the unit hydrograph method but did not
consider volumetric aspects of the rainfall-runoff response to any great degree. The main
output was discussion of the catchment-average unit hydrographs (UHs) derived for the 12
catchments. The UHs shown in Figure B.1 have been standardised by dividing by catchment
area to facilitate inter-catchment comparisons.

Figure B.1: Catchment-average unit hydrographs standardised by area

The 12 catchments were tentatively allocated to five catchment types:

Type 1 The lowest UH peak magnitude and longest UH duration response was for the Fergus
at Ballycorey, with an area-normalised UH peak of 1.510-3 m3s-1 mm-1 km-2 and a
UH timebase (i.e. the duration of flood response to a unit net rainfall) of 12 days;
Type 2 The second lowest UH peak magnitude (710-3 m3s-1 mm-1 km-2) and a timebase of
36 to 60 hours, as in the four rivers: the Deel, the Suir, the Dee and the Boyne;
Type 3 The third lowest UH peak magnitude (1510-3 m3s-1 mm-1 km-2) and a timebase of
12 to 30 hours, as in the four rivers: the Dripsey, the Owenboy, the Silver and the
Blackwater at Mallow;
Type 4 The fourth lowest UH peak magnitude (2010-3 m3s-1 mm-1 km-2) and a timebase
of 24 hours, as in the two rivers: the Nore and the Blackwater at Dromcummer;
Type 5 The highest UH peak magnitude (2710-3 m3s-1 mm-1 km-2) and a timebase of ~18
hours, as in the Blackwater at Duarrigle.

Some consideration was also given to artificial neural network (ANN) analysis, and to the
potential application of ANN models in flood warning.

129
Volume III Hydrograph Analysis

Appendix C Performance of HWA methods on verification events

Figure C.1: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


06011
130
Volume III Hydrograph Analysis

Figure C.2: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


06012

131
Volume III Hydrograph Analysis

Figure C.3: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


06013

132
Volume III Hydrograph Analysis

Figure C.4: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


06014

133
Volume III Hydrograph Analysis

Figure C.5: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


06026

134
Volume III Hydrograph Analysis

Figure C.6: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


07007

135
Volume III Hydrograph Analysis

Figure C.7: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


07009

136
Volume III Hydrograph Analysis

Figure C.8: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


07010

137
Volume III Hydrograph Analysis

Figure C.9: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


07012

138
Volume III Hydrograph Analysis

Figure C.10: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


09001

139
Volume III Hydrograph Analysis

Figure C.11: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


14004

140
Volume III Hydrograph Analysis

Figure C.12: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


14006

141
Volume III Hydrograph Analysis

Figure C.13: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


14007

142
Volume III Hydrograph Analysis

Figure C.14: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


14011

143
Volume III Hydrograph Analysis

Figure C.15: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


14018

144
Volume III Hydrograph Analysis

Figure C.16: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


15005

145
Volume III Hydrograph Analysis

Figure C.17: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


23002

146
Volume III Hydrograph Analysis

Figure C.18: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


24013

147
Volume III Hydrograph Analysis

Figure C.19: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


25003

148
Volume III Hydrograph Analysis

Figure C.20: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


25006

149
Volume III Hydrograph Analysis

Figure C.21: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


25014

150
Volume III Hydrograph Analysis

Figure C.22: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


25017

151
Volume III Hydrograph Analysis

Figure C.23: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


25025

152
Volume III Hydrograph Analysis

Figure C.24: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


25027

153
Volume III Hydrograph Analysis

Figure C.25: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


25030

154
Volume III Hydrograph Analysis

Figure C.26: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


26007

155
Volume III Hydrograph Analysis

Figure C.27: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


26008

156
Volume III Hydrograph Analysis

Figure C.28: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


26012

157
Volume III Hydrograph Analysis

Figure C.29: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


26019

158
Volume III Hydrograph Analysis

Figure C.30: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


27002

159
Volume III Hydrograph Analysis

Figure C.31: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


29001

160
Volume III Hydrograph Analysis

Figure C.32: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


29011

161
Volume III Hydrograph Analysis

Figure C.33: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


30004

162
Volume III Hydrograph Analysis

Figure C.34: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


30005

163
Volume III Hydrograph Analysis

Figure C.35: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


34018

164
Volume III Hydrograph Analysis

Figure C.36: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


36010

165
Volume III Hydrograph Analysis

Figure C.37: Verification of median hydrograph + UPO-ERR-Gamma methods at Station


36015

166
Volume III Hydrograph Analysis

Appendix D HWA results and their estimates from PCDs


Table D.1: Hydrograph width analysis results for all 89 stations

Estimates in the RH columns derive from PCD regression models that do not use BFI, i.e. from the models reported in Table 6.7

From HWA From PCD models of Table 6.7 (no BFI)


Width descriptors UPO-ERR-Gamma params Width descriptors UPO-ERR-Gamma params
Station # River Station Name
W75 (h) W50 (h) n Tr (h) C(h) W75 (h) W50 (h) n Tr (h) C(h)
06011 Fane Moyles Mill 114.35 1.30 29.69 360.83 131.72 2.67 99.81 220.11
06012 Fane Clarebane 136.25 243.02 2.46 93.31 280.95 143.58 275.96 2.35 92.91 270.81
06013 Dee Charleville 41.99 70.87 7.78 59.99 102.86 20.32 37.60 5.63 24.52 53.83
06014 Glyde Tallanstown 95.87 154.30 3.01 89.91 113.85 36.71 69.03 4.21 35.31 88.45
06026 Lagan (Glyde) Aclint 82.47 146.16 5.00 102.18 140.97 28.95 53.96 4.43 29.03 67.84
07001 Tremblestown Tremblestown 18.36 34.20 5.01 23.31 38.36 24.88 45.71 5.44 34.50 48.64
07002 Deel Killyon 46.74 93.14 2.77 32.74 151.96 41.90 82.28 4.14 42.49 92.89
07004 Kells Blackwater Stramatt 110.14 171.87 3.01 101.98 116.78 134.85 274.14 2.38 88.18 274.11
07006 Moynalty Fyanstown 15.42 30.38 7.78 20.87 58.15 18.91 33.47 6.51 27.49 36.21
07007 Boyne Aqueduct 26.22 52.53 3.75 26.84 69.88 20.42 40.55 5.86 27.13 56.14
07009 Boyne Navan Weir 27.75 54.14 5.28 33.04 95.53 26.51 52.30 5.41 34.31 91.07
07010 Blackwater Liscartan 30.42 119.06 3.46 25.78 222.32 47.19 89.45 4.08 48.25 119.06
07011 Kells Blackwater ODalys Br 108.21 170.42 2.99 92.96 116.78 123.25 246.43 2.57 85.30 238.80
07012 Boyne Slane Castle 27.75 49.10 6.11 36.41 88.20 33.89 65.87 4.98 40.20 116.29
07033 Kells Blackwater Virginia Hatchy 46.41 84.10 6.11 60.82 105.05 65.21 118.23 3.93 66.92 81.00
09001 Ryewater Leixlip 11.21 22.64 8.40 21.75 22.97 28.54 52.78 5.56 44.47 50.93
11001 Owenavorragh Boleany 7.54 10.84 26.36 25.24 5.38 18.13 34.51 6.52 36.26 35.12
14004 Figile Clonbulloge 48.17 74.35 5.41 66.48 53.75 53.65 100.57 5.04 96.62 65.75
14006 Barrow Pass Bridge 57.46 83.02 4.71 75.83 41.29 27.45 50.07 5.91 49.03 69.84
14007 Stradbally Derrybrock 9.34 17.14 21.41 31.79 11.24 32.10 59.83 5.06 32.68 52.78
14009 Cushina Cushina 26.27 46.20 3.88 27.75 46.42 32.62 59.38 5.84 53.95 35.04
14011 Slate Rathangan 45.96 89.57 5.28 54.73 116.78 61.11 121.30 4.21 86.32 82.24
14018 Barrow Royal Oak 65.08 121.78 2.94 57.73 116.78 46.55 88.89 5.66 68.27 93.39

167
Volume III Hydrograph Analysis

From HWA From PCD models of Table 6.7 (no BFI)


Width descriptors UPO-ERR-Gamma params Width descriptors UPO-ERR-Gamma params
Station # River Station Name
W75 (h) W50 (h) n Tr (h) C(h) W75 (h) W50 (h) n Tr (h) C(h)
15001 Kings Annamult 15.74 28.49 9.42 30.46 23.70 20.33 35.66 6.87 33.68 42.10
15002 Nore Johns Br 8.12 19.23 17.64 24.97 21.50 31.62 58.18 6.91 60.23 58.77
15003 Dinin Dinin Br 5.77 8.79 10.54 12.42 6.11 14.41 25.68 8.36 41.24 27.00
15005 Erkina Durrow Foot Br 60.06 114.00 5.27 83.23 106.52 36.07 65.31 5.58 54.18 59.22
15006 Nore Brownbarn 19.95 35.64 13.83 53.25 22.97 31.07 56.89 6.73 56.72 68.11
16001 Drish Athlummon 30.61 46.41 5.55 40.16 39.83 28.69 52.40 7.11 69.12 29.22
16002 Suir Beakstown 58.04 102.32 8.34 91.27 114.58 31.72 57.04 6.47 58.76 48.16
16003 Clodiagh Rathkennan 39.53 77.32 2.53 16.15 188.60 18.62 32.01 7.03 27.76 34.51
16004 Suir Thurles 63.79 102.80 6.39 90.24 99.19 28.94 51.57 6.44 49.91 39.89
16005 Multeen Aughnagross 13.76 21.02 28.19 51.75 5.38 10.77 18.20 9.93 26.89 14.46
16008 Suir New Bridge 88.08 146.96 3.15 44.16 410.67 31.21 55.98 6.74 52.23 55.12
16009 Suir Caher Park 39.50 92.10 6.11 49.99 163.69 30.29 54.38 6.92 52.13 57.84
18004 Awbeg Ballynamona 38.18 73.27 5.00 43.89 105.05 23.77 41.61 8.21 41.88 28.22
18005 Funshion Downing Br 15.63 26.35 24.60 52.75 15.64 27.52 47.41 6.73 48.80 41.96
19001 Owenboy Ballea 26.34 44.94 2.79 12.68 112.38 32.67 52.61 6.50 32.46 31.84
22071 L. Leane Tomies Pier 96.29 190.22 1.88 50.56 243.57 103.53 202.15 2.51 67.99 286.23
23001 Galey Inch Br 6.98 11.84 12.54 16.65 5.38 12.30 21.01 9.13 21.59 20.64
23002 Feale Listowel 5.81 9.27 30.27 23.00 3.18 15.08 25.04 10.05 37.07 24.21
23012 Lee (Kerry) Ballymullen 11.67 18.29 18.23 32.13 11.24 14.00 21.82 8.57 24.43 17.01
24001 Maigue Croom 18.24 27.88 4.44 19.16 35.43 20.86 37.55 6.39 22.81 55.25
24008 Maigue Castleroberts 19.08 30.54 3.88 19.40 28.10 21.15 38.13 6.28 23.27 57.71
24013 Deel Rathkeale 19.66 28.77 12.33 48.75 6.85 21.06 37.01 6.75 27.70 43.26
24082 Maigue Islandmore 18.66 27.51 5.14 21.42 38.36 20.76 37.37 6.39 22.70 55.06
25001 Mulkear Annacotty 15.51 29.51 9.45 31.94 35.43 9.48 16.65 10.11 19.85 23.86
25003 Mulkear Abington 15.38 29.57 4.30 16.51 50.09 10.65 18.62 8.96 18.87 25.76
25005 Dead Sunville 14.00 33.17 5.00 13.14 69.88 16.17 28.73 7.11 20.71 31.80
25006 Brosna Ferbane 35.86 63.91 5.28 42.42 94.06 33.85 66.59 4.64 28.77 109.20
25014 Silver Millbrook 17.63 30.05 8.97 30.83 36.16 10.69 19.40 7.27 16.99 29.60
25016 Clodiagh Rahan 20.26 36.19 3.88 18.40 62.55 16.72 31.01 5.68 22.13 52.82
25017 Shannon Banagher 231.91 1.50 109.75 374.76 285.21 2.35 142.58 719.78

168
Volume III Hydrograph Analysis

From HWA From PCD models of Table 6.7 (no BFI)


Width descriptors UPO-ERR-Gamma params Width descriptors UPO-ERR-Gamma params
Station # River Station Name
W75 (h) W50 (h) n Tr (h) C(h) W75 (h) W50 (h) n Tr (h) C(h)
25025 Ballyfinboy Ballyhooney 85.33 187.22 4.14 111.29 222.32 41.39 74.47 4.98 64.07 58.50
25027 Ollatrim Gourdeen 7.66 14.25 19.41 23.42 19.31 18.66 33.56 6.35 24.45 34.51
25029 Nenagh Clarianna 15.87 29.50 8.89 28.70 50.09 18.27 32.82 6.36 24.67 43.40
25030 Graney Scarrif 70.22 122.12 4.22 81.86 94.06 27.49 52.55 5.74 54.81 46.96
26002 Suck Rookwood 118.67 163.98 6.58 186.50 39.83 84.62 142.63 5.47 87.48 69.54
26005 Suck Derrycahill 136.04 209.83 3.60 148.51 84.53 86.15 147.63 5.46 93.33 80.02
26007 Suck Bellagill 156.97 257.02 3.04 159.81 107.99 88.32 151.33 5.44 94.60 83.12
26008 Rinn Johnstons Br 106.69 206.81 3.88 124.50 177.61 110.70 200.38 3.61 80.01 115.85
26009 Black Bellantra Br 26.52 38.47 8.47 49.00 14.18 58.62 98.28 4.76 62.44 53.74
26012 Boyle Tinacarra 177.27 285.94 3.85 195.50 257.50 187.00 366.65 2.98 122.83 202.74
26019 Camlin Mullagh 72.81 114.01 3.36 71.39 70.61 64.94 109.55 5.85 72.53 48.30
26021 Inny Ballymahon 52.68 179.79 3.48 24.80 527.93 185.77 397.77 2.47 108.40 384.16
26022 Fallan Kilmore 41.75 74.82 5.00 53.64 66.94 75.77 129.73 4.18 79.86 64.81
27001 Claureen Inch Br 14.87 20.26 5.00 17.72 11.24 10.02 16.57 11.91 33.86 9.06
27002 Fergus Ballycorey 270.72 493.96 5.28 429.25 187.87 153.55 316.40 2.59 139.23 292.53
29001 Raford Rath-gorgin 49.94 86.19 5.33 74.75 46.42 29.52 48.72 7.48 51.37 25.59
29004 Clarinbridge Clarinbridge 76.87 130.05 2.77 79.50 76.47 58.11 99.97 4.70 77.98 60.86
29011 Dunkellin Kilcolgan 138.65 242.46 4.01 182.75 114.58 50.90 88.36 5.18 66.23 65.87
30004 Clare Corrofin 44.33 68.49 5.30 59.44 46.42 49.13 84.48 5.33 46.89 75.02
30005 Robe Foxhill 39.87 56.46 7.40 77.00 22.97 44.37 74.76 5.67 43.71 50.26
30007 Clare Ballygaddy 38.52 59.31 6.11 67.75 20.77 53.24 89.71 5.35 50.11 67.04
30061 Corrib Estuary Wolfe Tone Br 2.91 41.90 1.56 168.16
34001 Moy Rahans 95.52 2.79 58.35 829.75 103.33 3.52 64.53 197.49
34009 Owengrave Curraghbonaun 19.67 26.03 9.10 39.70 8.31 32.28 49.18 7.44 38.86 25.88
34018 Castlebar Turlough 1.27 35.58 187.87 2.43 66.80 196.27
35001 Owenmore Ballynacarrow 103.62 154.99 3.88 116.94 105.79 111.98 188.17 5.12 111.33 61.88
35002 Owenbeg Billa Br 7.62 11.39 17.77 24.01 10.56 15.38 13.72 23.90
35005 Ballysadare Ballysadare 38.18 85.88 5.00 36.80 187.87 69.53 116.80 5.34 68.86 68.27
35071 L. Melvin Lareen 123.67 210.68 4.22 99.46 456.84 28.70 44.00 7.54 42.51 30.62
36010 Annalee Butlers Br 75.15 155.52 3.04 73.10 152.69 107.09 196.64 3.30 76.90 183.22

169
Volume III Hydrograph Analysis

From HWA From PCD models of Table 6.7 (no BFI)


Width descriptors UPO-ERR-Gamma params Width descriptors UPO-ERR-Gamma params
Station # River Station Name
W75 (h) W50 (h) n Tr (h) C(h) W75 (h) W50 (h) n Tr (h) C(h)
36011 Erne Bellahillan 316.02 613.41 2.21 295.25 229.65 195.91 380.64 2.46 107.26 277.04
36015 Finn Anlore 40.08 67.19 6.19 68.00 35.43 54.95 90.55 4.92 51.43 57.30
36019 Erne Belturbet 295.51 491.91 2.48 254.50 302.20 234.64 468.71 2.20 109.37 540.31
36021 Yellow Kiltybarden 3.11 4.90 14.70 7.66 2.45 5.89 8.84 18.78 22.48 3.67
36027 Woodford Bellaheady 332.71 539.54 2.61 300.50 213.52 233.97 476.48 2.30 112.89 314.24
39009 Fern O/L Aghawoney 45.10 86.66 9.10 74.10 115.31 59.63 102.78 4.04 49.19 81.79

170
Appendix E Application of the HWA software

E1 Troubleshooting the installation


The installation process may fail because of a number of conflicts or inadequacies in the
computer in which the HWA program is to be installed. If that happens, before exiting the
installation process, the installation wizard will display a message informing the user on the
cause of failure to install. The user may then take appropriate action to satisfy the
requirements to successfully install the program. Some of the most likely causes of failure of
the installation process and possible remedies are given below.

E1.1 Lack of storage space


The users computer may not have adequate disk space for installing all files required for
running the program. This problem may be solved by creating enough space in the disk either
by removing unnecessary or temporary files and folders, or by adding extra physical disk
space by upgrading to higher spcifications. When enough disk space is available, the
installation may be tried again.

E1.2 One or more files exist in the destination folder


The installation process may fail because of the existence of one or more files in the
destination folder. This problem may be solved by emptying the destination folder, either by
deleting the files, if not useful, or by moving the useful files to some other folder or folders.
The installation process may then be tried again.

E1.3 An out-of-date system file exists in the computer


When the Version of a dll (Dynamic Link Library) file (Oleaut32.dll) is older than that
required by the program for successful installation, the following message appears and the
installation process fails:

Setup cannot continue because some system files are out of date on your system. Click
OK if you would like Setup to update these files now. You will need to restart
Windows before you can run Setup again. Click Cancel to exit Setup without
updating system files.

When the user selects the OK button on the message box thus displayed, the Setup.Exe
program installs a newer version of the required dll file, which is compatible with the
installation program. In order to update the file to the correct version, the operating system
must be restarted by rebooting the computer. After rebooting, the application, the Setup.Exe
program is to be re-run to install.

E1.4 Out-of-date system files exist in the computer (multiple errors)


During installation, the installation program delays the replacement of the in-use system files,
until rebooting takes place, by saving the new files as temporary files in the Temp folder. In
order to replace the existing files with the .tmp files to complete the installation process, the

171
system uses a replacing and renaming operation. If something interferes with this operation,
then the in-use system files are not updated. Hence, once the computer is rebooted and the
installation program is restarted, the same error message appears and the installation process
fails again. The two most common causes for this to happen are that the .tmp files are deleted
or that the Temp folder is on a different drive or partition from the operating system. By
default, the operating system is installed to either the Windows or the Winnt folder.

The following stops may resolve this problem:

a) Copy the TEMP and TMP environment variables to a folder that is in the same drive
partition as the Windows system files. To do this, open a command prompt window and
type the following at the prompt:

Set TMP = C:\TEMP

Set TEMP = C:\TEMP

This will save the TEMP and TMP environment variables to a folder named Temp that
resides on the C: drive. The folder must exist prior to carrying out these steps. Once
these environment variables are set, the application should then install.

b) If the Autoexec.bat file contains the following line (or similar):

If exists c:\temp\*.tmp del c:\temp\*.tmp

this is to be commented out by placing REM in front of it.

c) Disable any antivirus software (or other memory resident programs) and try running Setup
again. Often the best way to accomplish this is to run Setup in Safe Mode. It may also be
necessary to copy all of the Setup files to a temporary folder on the hard drive disk and
run Setup.exe from there.

d) Left-over files from a failed Setup attempt can also cause this problem. If found, delete
the msftqws.pdw subfolder and its contents from the Temp folder. Also look for
Setup1.exe in the Windows or Winnt folder and any *.CAB files from previous installs, and
delete them. This should be done after each failed install.

e) Some logon scripts can cause this problem, so try to run Setup before logging on to the
network if the computer is connected to a network.

E1.5 Existence of out-of-date system files in the computer (on non-upgraded


operating systems)
If the operating system in the users machine is out of date, older versions of the system files
may cause problems during installation of the program and the installation process may fail.
The file that causes this problem is Oleaut32.dll. In this case, the same message, as given in
Section E1.3, appears repeatedly during each successive installation attempt even after
carrying out the modifications suggested in Sections E1.3 and E1.4.

172
In order to solve this problem, the user has to install the latest Versions of system files in the
computer. Microsoft provides freely downloadable programs called Service Packs for this
purpose. These are available in the Download Centre (or the Support Section) of the
Microsoft website (www.Microsoft.com). The user has to download the relevant Service
Pack and install it in his/her computer before installing the HWA program. The Service
Packs replace the older Versions of the system files by the latest (updated) ones, and add
additional components which were not previously available for proper functioning of the
system. It is expected that after successful installation of Service Packs and hence upgrading
of the operating system, the HWA program can be installed properly.

E1.6 The program icon is not created and the program is not listed under the
Program menu
During the last phase of installation of the HWA program, the Installation Wizard may issue a
message saying, An error occurred trying to create a program icon for HWA. This
message box displays Abort, Retry, and Ignore buttons. The Abort button terminates the
installation process, the Retry button does not help in resolving the problem, and repeats the
message after each click, while the Ignore button finishes the installation process, and the
message saying the HWA Setup was completed successfully appears. But, while searching for
the HWA Program in the Program list under the Start task bar item, the program may not be
found. This problem may be caused by some error in the required System files. In order to
find a solution for this problem, the user may locate the HWA.EXE application file in the
folder where the program was installed in the users hard-disk (usually, by default,
C:\Program Files\HWA). Double clicking the .Exe file in the My Computer pane opens the
HWA program and the program is then expected to work properly. The user may create a
Shortcut to the application file and place the shortcut on the Desktop for quick access to the
program.

E2 Graphical User Interface (GUI) concepts of HWA software

E2.1 Basic GUI components


In the Graphical User Interface (GUI) environment, the user interacts with the HWA Package
using the keyboard, the mouse, the windows, the menus, the tool bar buttons and the
command buttons. A window is usually a facility for entering input information, which is
surrounded by a border and moveable around the screen. The main application window in the
HWA Package is shown in Figure E.1.

As HWA is a Windows-based program, other Windows-based applications may also be


operated at the same time without necessarily closing the HWA program. The main window
of the HWA program can be minimised or maximised or used in its normal size, during
working with other Windows applications. The mouse enables the user to point at and to
select objects to work on, or to perform actions, by clicking the LH mouse button. The
keyboard is used for entering input information for running the HWA program, to move
between various objects, or to perform particular actions. Menus in the menu bar are used to
perform certain actions which may be accessed by clicking with the mouse or pressing the
relevant key on the keyboard. The buttons in the tool bar at the top of a window provide a
visual representation of and quick access to the intended tasks, which the buttons are meant to

173
link. Whenever the mouse pointer is placed above a button in the tool bar, a short description
(ToolTip text), briefing the intended task to be performed by the button, is displayed. To
invoke an effect on a button or a menu with the keyboard keys, identical to that produced by
clicking a button or menu with a mouse pointer, the user has to hold down the ALT key and
press the key for the letter (i.e. character) shown underlined (i.e. underscored) in the caption
of the button or menu. The buttons captioned Continue or OK or Close in any window may
be invoked by pressing the Return key on the keyboard, while the buttons captioned Back or
Cancel may be activated by pressing the Esc (Escape) key. A message box shown in the
window in Figure E.1 is used to instruct, inform or warn the user as appropriate.

Menu bar Toolbar Toolbar Top title Close button


Menu item button Maximize button
Minimize button

Message-box Main application window


Figure E.1: Main application window of the HWA program showing a message-box

The HWA GUI uses three types of windows. These are:

Main application window


Data window
Dialog window

Some of the important features of these windows are now described.

174
Main application window

The main application window is the one shown in Figure E.1. It appears as soon as the user
starts running the program. This window is comprised of a Multiple Document Interface
(MDI) form which includes a menu, a toolbar, a child window area, and a status bar. The
title bar contains the title of the program as well as the maximise, minimise and close buttons.
All other windows required for entering data or displaying outputs reside within this MDI
window.

Data windows

Data windows are the windows containing either data-entry fields or displays of tabular and
graphical outputs. The data window in Figure E.2 contains information labels, six data-entry
boxes, three pairs of radio (i.e. option) buttons, one check box and three command buttons
which are to be used to enter input information for running the HWA program.

The data window in Figure E.3 contains information labels, a list-box showing event
numbers, a picture-box exhibiting the hydrograph of the flood event shown highlighted in the
list box, one grid with a vertical scroll bar providing the ordinate number, date, time and
discharge related to the displayed hydrograph in a tabular form, one pair of radio (option)
buttons, two check boxes, 18 command buttons and one frame shown disabled. In the case of
such a data-entry window, the cursor is automatically placed at the topmost data-entry field
when the data window first appears. The user can directly enter data by using the keyboard
and/or mouse. In order to specify data in other fields, the user navigates by using the (Up),
(Down), (Left), and (Right) buttons, by pressing the Tab button on the keyboard, or by
clicking the mouse after placing the cursor into that field where the data is to be entered.

Figure E.2: Data window for entering data

175
Figure E.3: Data windows for displaying graphical and tabular outputs

Dialog window

Dialog windows are used in HWA to either request information from the user or provide
information to the user. Usually an ellipsis () after the caption of a menu or a button
indicates that invoking the intended action results in a dialog window. A dialog window is
also displayed when some command buttons, e.g. that captioned Browse, are clicked, or
when a data-entry field that requires a filename is double-clicked. Figure E.4 shows a dialog
window for specifying a filename for storing the results summary.

Figure E.4: A dialog window showing a standard windows file-opening dialog-box

176
E2.2 Typical GUI features of the windows displaying hydrographs
In the case of a data-display window containing a hydrograph, different GUI components are
provided for effective display of relevant information. For illustration, some features of the
window displaying flood hydrographs are described below. These features are common to
other data-display windows as well.

i Displaying the hydrograph of an event from a list of selected events: In the


window displaying the observed flood hydrographs, event numbers of all selected
observed floods are provided in a list-box on the left hand side of the window. By
default, one item in the list, usually the first one, is shown highlighted by a blue shade,
and the corresponding hydrograph is displayed in a picture-box provided on the right
hand side of the list-box. The user may scroll through the list using the vertical scroll
bar provided alongside the list-box to choose any item. The list of events can also be
navigated using the (Up) and (Down) keys. When the user clicks on an event
number in the list box, that event number is highlighted and the existing hydrograph is
replaced by that corresponding to the highlighted event number.
ii Displaying hydrograph in different units: By default, time is displayed in hours on
the time-axis, i.e. the abscissa, of the hydrograph. However, given that the OPW data
files hold data in 15-minute time intervals, an option is provided to display time in 15-
minute intervals as well. A pair of radio (option) buttons, one captioned Hour and the
other captioned No. of 15 minute intervals (data interval in OPW data file) is provided
for this purpose. The user can view the hydrograph in either of the two time units by
clicking the respective radio button.
iii Showing data values in a grid: By default, the tabular values are not displayed.
However, a command button captioned Show tabular values of hydrograph is
provided at the bottom on the right hand side of the window. When this button is
pressed, a grid appears on the right hand side of the picture-box and the caption of the
button changes to Hide tabular values of hydrograph. The values related to the
displayed flood hydrograph are shown in that grid in tabular form, using rows and
columns. By default, one of the rows, usually the top one, is shown highlighted by a
yellow shade. When the user clicks on any other row, the highlighter moves to the
row thus clicked. The (Down) and (Up) keys on the keyboard can also be used to
move the highlighter from one row to the next, either downward or upward. The user
may scroll through rows using the vertical scroll bar provided alongside the grid.
iv Showing values on the hydrograph at any data point: When the user clicks on any
point inside the picture-box, an indicator appears at that point as a grey-coloured
vertical line. Note that for some graphical outputs in other windows, e.g. those
displaying derived flood hydrographs, an indicator having a light green colour instead
of grey, is displayed. The picture-box shown in Figure E.3 shows this line at the 62nd
hour, i.e. the 248th time-step in the 15-minute interval, on the receding side of the
hydrograph. A label having a light yellow background simultaneously appears at the
top right of the window displaying relevant information for the time-step
corresponding to the clicked data point, and a check-box provided on the left hand
side of the label is shown checked. The user can hide the label containing the
information by unchecking the check-box. The user can place the indicator in the
picture-box by clicking at any point for which s/he wishes to see the value.
Alternatively, the or the keys may be pressed to move the indicator by one

177
time-step at a time, either to the left or to the right, in order to show the values at the
desired data point. If, as described in iii) above, the grid containing the values of the
hydrograph is exhibited, then the highlighter on this grid automatically moves to the
row containing the information related to the data-point, either clicked on the
hydrograph, or reached by pressing the or the arrow buttons on the keyboard.
When the user double-clicks on the picture-box, the indicator and the label disappears.
v Changing the background colour: If the user wishes to change the background
colour of the hydrograph, for example for the purpose of presentation and/or printing,
the command button BackGround Colour needs to be clicked. The standard
Windows colour-palette displaying all colours, hues and shades, appears. The user
may choose any colour and press the OK button provided on the palette. The graph is
redrawn with the chosen colour on the background.
vi Copying a graph onto the clipboard: By pressing the command button Copy Graph,
a graph can be copied onto the clipboard. The Graph, thus copied, can then be pasted
in other Windows application programs such as MS-Word or MS-Excel.
vii Copying tabular values onto the clipboard: By pressing the command button Copy
table, the table can be copied onto the clipboard. The values, thus copied, can then be
pasted into other Windows application programs. In the case of MS-Excel, the copied
values are pasted in separate rows and columns; the user can thus carry out any further
analysis or displays using facilities of MS-Excel to augment those provided within the
HWA program.
viii Saving a graph in a picture file: The hydrograph displayed in the picture-box can be
saved as a picture file in bitmap format for later use, e.g. for preparing a report, or for
presentation. This is achieved by clicking the command button Save graph As when
a standard Windows file-saving dialog box appears. The user can use the controls on
the dialog box, thus displayed, and specify a filename to save the hydrograph as a
bitmap in the chosen folder.
ix Printing a graph using a printer: The hydrograph displayed in the picture-box can
be printed using a printer connected to the users computer by clicking on the
command button Print graph, when a standard Windows printing dialog box
appears. The user can select the desired settings before printing the hydrograph.

In addition to the above-mentioned GUI components, the data-display window showing the
hydrographs of the observed flood events supports other specific features. Some of these are
evident from diagrams presented in the main body of the volume (e.g. Figure 5.18).

E2.3 Options on start-up of the HWA program


When the user clicks the Continue button on the start-up window, a message box appears as
shown in Figure E.1. This message provides information about two options of entering input
information into the program.

Under the first option, the user may enter all relevant input information into data-entry fields
in a window by using keyboard and mouse. Each data-entry field is provided with an
appropriate label to indicate the specific type of data that the field is supposed to receive from
the user. This option is essentially required when the user applies the program to a gauged
station for the first time. In the data-entry window, a data-entry field is provided in which the
user can specify the name of a results summary file. The default filename extension of such a

178
file is *.sum. If no filename is specified, a default filename WP3_1Sys.sum is used to hold the
results summary of the Hydrograph Width Analysis each time the program is run.

The second option can be used only if the HWA program was run at least once before with
the data of the gauged station and the results summary had been saved in a user-specified file.
Under this option, the user has to specify the name of the previously-saved results-summary
file in a data-entry field. In order to do this, s/he can browse for the required file, either by
clicking on the Browse button, or by double-clicking the data-entry field. The resulting
window shows the list of directories and folders which can be searched for the required file,
as in any other Windows-based program. Once the desired file is located, the user can either
select the file by clicking on its name and press the OK button, or double-click on the
filename. The window containing the list of directories and folders automatically closes and
the filename appears in the data-entry field. When the user presses the Continue button,
relevant information is extracted by the program from the specified results summary file and
displayed in the data-entry window. Thus, the user is saved the effort of entering input
information each time the program is re-run with data of a particular station. Of, course, the
user can make changes and deletions in the data-entry fields after these are displayed in the
window, and then run the program with the changed or modified set of information. The
default system filename of WP3_1Sys.sum can be specified if the user wishes to resume the
previous analysis but had omitted to save this in a named file.

E2.4 Menu bar items


As highlighted in Figure E.1, the main application window of the HWA program contains
nine menu items on the menu bar. The menu and sub-menu captions provide a brief
indication of the intended tasks. Although a mouse can be used for pointing to and clicking
on a menu item, the keyboard keys can also be used for invoking a similar effect. In order to
do this, the user has to press the key for the letter (i.e. character) underlined in the caption of
the menu or sub-menu item when the user holds down the ALT key. Some initially disabled
menu bar items are later enabled. For example, the menu items Results summary and
Hydrographs of flood events are enabled once hydrographs of observed flood events have
been extracted by the program from the record of discharge data. Similarly, the Derived
hydrographs menu item is enabled once the derived semi-dimensionless flood hydrographs
are produced by the program. A brief description of the menu bar items is provided below.

File menu

This menu is activated by clicking File on the menu bar or by keying ALT+F. A list of sub-
menus appears having two enabled items (Set working directory and Open) and three disabled
items.

The sub-menu Set working directory facilitates setting of the working directory in the case
when the user wants to use data files from and store output files into a folder other than the
one in which the HWA Package is installed. When clicked, this menu item brings up a
standard dialog window, showing the list of drives, folders and files, which may be used to
nominate/select the working directory before running a program.

The Open sub-menu has two further sub-menus (Existing file and New document) which
can be used for either opening an existing file or a new file in ASCII format and for carrying
out editing, printing and saving operations as required. The Save As, Close and Print

179
sub-menus under the File menu are enabled once the user opens an existing file or a new
document.

Analysis options menu

This menu is activated by clicking Analysis options on the menu bar or by keying ALT+A.
Sub-menus Start from scratch and Use results file stored during an earlier run then appear.

By clicking the sub-menu Start from scratch, the user can enter relevant input information
into data-entry fields in a window by using the keyboard and mouse (see Section E2.2). The
sub-menu Use results file stored during an earlier run can only be used if the HWA program
was run at least once before, and the data and results summary saved in a user-specified file.

Results summary menu

This menu is activated by clicking Results summary on the menu bar or by keying ALT+R. It
is enabled after the HWA program extracts the hydrographs of observed flood events from the
record and displays these in a data window. When the user clicks on this menu, a data
window appears showing the summary of results in an ASCII formatted file. It may be noted
that the current version of the HWA program retains all options and procedures developed
and tested in the course of the hydrograph-width research. All outputs of the Hydrograph
Width Analysis are stored in the results summary file, which can be quite large.

Hydrographs of flood events menu

This menu is activated by clicking Hydrographs of flood events on the menu bar or by keying
ALT+G. It is enabled after the HWA program extracts the hydrographs of observed flood
events from the record and displays these in a data window. If a data window other than that
containing the observed flood hydrographs is shown active within the main application
window, a click on this menu activates and brings to the front the window containing the
flood hydrographs.

Derived hydrographs menu

This menu is activated by clicking Derived hydrographs on the menu bar or by keying
ALT+D. It is enabled after the HWA program successfully produces the derived median and
mean flood hydrographs as one of the outputs of the Hydrograph Width Analysis. If a data
window other than that containing the derived flood hydrographs is shown active within the
main application window, a click on this menu activates and brings to the front the window
containing the derived flood hydrographs.

Modified Gamma hydrographs menu

This menu is activated by either clicking Modified Gamma hydrographs on the menu bar or
by keying ALT+y. It is enabled after the HWA program successfully produces the modified
Gamma hydrographs as one of the outputs of the Hydrograph Width Analysis. If a data
window other than that containing the modified Gamma hydrographs is shown active within
the main application window, a click on this menu activates and brings to the front the
window containing the modified Gamma hydrographs.

180
Window menu

The Window menu contains four sub-menus and a list of child windows which are open
within the main application window. It can be activated by clicking the menu title or by
keying ALT+W. The first sub-menu is New Window, which allows the user to open a new
document window for creating a new ASCII formatted file, if required. Any number of such
windows can be opened. The other sub-menu items are Cascade, Tile Horizontal and Tile
Vertical, which may be used to arrange a number of opened windows on the screen by any of
the three different styles as available in any standard Windows application. When one or
more child windows are kept opened during a run of HWA, a list (by title) of the windows
appears at the bottom of the Window menu. The user may bring up any of these child
windows by clicking the relevant title.

Help menu

This menu option is disabled in the current version of the HWA software program.

Exit menu

In order to close the session, the user can click Exit in the menu bar or key ALT+x. The user
is guided through the closure procedure.

E2.5 Toolbar buttons


As shown in Figure E.1, the main application window of the HWA program contains eleven
buttons on the toolbar. ToolTips provide a brief indication of the intended tasks and appear
when the mouse hovers on the relevant button. Some buttons are disabled during data entry
operations. The toolbar functions are summarised in Table E.1.

E2.6 Closing the HWA program


The expected options are provided to close the HWA program in a number of different ways.

181
Table E.1: Details of toolbar buttons
(Letters in red show the relevant keyboard shortcuts when depressed with the ALT key.)
Menu or sub-
menu item
Icon ToolTip text Status
performing the
task
Set working File: Set working
Enabled throughout.
directory directory
Open new File: Open:
Enabled throughout.
document Existing file
File: Open: New
Open existing file Enabled throughout.
document
Enabled at the beginning. Disabled when the
option of analysing by entering data on the
Start analysing Analysis options:
screen is chosen by clicking this button.
from scratch Start from scratch
Enabled again when this option of analysing
is closed.
Enabled at the beginning. Disabled when the
Use results file Use results file option of analysing by reading data from a
stored during an stored during an pre-saved results summary file is chosen by
earlier run earlier run clicking this button. Enabled again when this
option of analysing is closed.
Becomes enabled once hydrographs of
Results summary Results summary observed flood events are extracted and
displayed.
Becomes enabled once hydrographs of
Hydrographs of Hydrographs of
observed flood events are extracted and
flood events flood events
displayed.
Becomes enabled once semi-dimensionless
Derived Derived
derived hydrographs are produced and
hydrographs hydrographs
displayed.
Disabled at the beginning. Becomes enabled
Modified Gamma Modified Gamma
after semi-dimensionless modified Gamma
hydrographs hydrographs
hydrographs are produced and displayed.
Disabled throughout in the current version of
Help Help
the HWA software.
Exit Exit Enabled throughout.

182
Appendix F Further details of IBIDEM
F1 Method of optimising Tp
Optimising Tp involves minimising the objective function:

[WFSU(i) - WFSR(i)]2 for i = 1 to m.

With only one variable to adjust, the minimisation is straightforward. IBIDEM adopts a
simple approach based on the method of bisection.

The starting point for the search is an initial guess for Tp. This is set as the time elapsing
from the first ordinate of the input FSU hydrograph to its peak value. It is not necessary to
start with a particularly good guess. The search range is then set as follows:

Minimum = 0.001 hours


Maximum = 10 times initial guess

This range of extreme values is likely to far exceed the realistic range in which the best
parameter value lies. However, with automated calculations, there is little penalty in adopting
such a wide range as a precautionary measure.

At each iteration, IBIDEM considers five trial values of Tp denoted by Tp[1], Tp[2], Tp[3],
Tp[4] and Tp[5]. For the first iteration, the five trial values are:

Tp[1] = 0.001 i.e. the minimum value which will be considered


Tp[2] = (Tp[1] + Tp[3]) /2 i.e. half way between Tp[1] and Tp[3]
Tp[3] = initial guess
Tp[4] = (Tp[3] + Tp[5]) /2 i.e. half way between Tp[3] and Tp[5]
Tp[5] = 10 x initial guess i.e. the maximum which will be considered

Each of the five trial values of Tp is used to create an FSR hydrograph shape. The value,
Tp[n], that gives the best fit (i.e. the smallest value of the objective function) is used as the
central value for Tp in the next iteration. Trial values for Tp are reassigned as follows:

Tp[1] = Tp[n-1] i.e. updates lower bound of range in which optimum lies
Tp[2] = (Tp[1] + Tp[3]) /2 i.e. half way between Tp[1] and Tp[3]
Tp[3] = Tp[n] i.e. best value from previous iteration
Tp[4] = (Tp[3] + Tp[5]) /2 i.e. half way between Tp[3] and Tp[5]
Tp[5] = Tp[n+1] i.e. updates upper bound of range in which optimum lies

The iterative procedure continues until the range within which the optimum value is known to
lie becomes acceptably small. The criterion used is Tp[5] - Tp[1] < 0.01 hours. At that point,
the value giving the best fit is adopted as the optimum value of Tp.

Consider, as an example, the case where the ultimate best-fitting value is Tp = 3.0 hours and
the initial guess is Tp = 5.0 hours. Table F.1 shows the trial values used in the first five

183
iterations. The best-fitting trial value of Tp (shown in red) is adopted as the central value
Tp[3] in the next iteration.

Table F.1: Example of iteration to find best-fitting value of Tp


Iteration First Second Third Fourth Fifth
Tp[1] 0.001 0.001 1.25 2.50 2.81
Tp[2] 2.50 1.25 1.87 2.81 2.96
Tp[3] 5.00 2.50 2.50 3.12 3.12
Tp[4] 27.5 3.75 3.12 3.43 3.27
Tp[5] 50.0 5.00 3.75 3.75 3.43

It is seen that, after five iterations, the Tp value (2.96 hours) is already close to the ultimate
best-fitting value of 3.0 hours.

F2 Method of fitting SPR


Having determined Tp so as to fit the characteristic hydrograph as closely as possible,
IBIDEM calculates SPR by matching the peak of the FSR response hydrograph, qpeak, to the
required response peak from the FSU hydrograph, qFSUpeak, which is found by subtracting BF
from the peak FSU flow.

The calculation of SPR follows from the fact that the response hydrograph is directly
proportional to the percentage runoff, PR. In the general case of a part-urbanised catchment,
PR = PRrural (1.0 0.47 URBEXT) + 70 (0.47 URBEXT) F.1
This applies the FSSR16 urban adjustment, but using the FSU index URBEXT rather than the
FSR index URBAN. See Step 9 of Section 9.2.

Simplifying Equation F.1 gives:


PR = PRrural (1 0.47 URBEXT) + 32.9 URBEXT F.2
The term PRrural is composed of three parts:
PRrural = SPR + DPRCWI + DPRRAIN F.3
Here, SPR is standard percentage runoff, DPRCWI is dynamic percentage runoff attributable to
catchment wetness and DPRRAIN is dynamic percentage runoff attributable to event rainfall.
The latter two quantities depend solely on catchment descriptors (SAAR, which controls
CWI) and the design rainfall depth, which is a function of storm duration, which can be
determined from Tp.

It is possible to calculate the response hydrograph peak qpeak without carrying out a full
convolution (see Houghton-Carr, 1999), by using the short-cut method described in FSSR9
(IH, 1979):
qpeak = RC.(PR/100).(P/D).AREA F.4

184
where:

RC is a routing coefficient which depends on D/Tp (a function of SAAR);


P is the depth of the design storm (after application of ARF);
D is the duration of the design storm.

FSSR9 presents a graph showing how RC varies with D/Tp based on the 75% winter rainfall
profile. A similar graph is presented by Houghton-Carr (1999), which includes a line
corresponding to the 50% summer rainfall profile.

Based on Equations F.2 to F.4, IBIDEM calculates the value of SPR that yields the required
qpeak. It is found from:

100.D.q peak
- 32.9 URBEXT
F.5
SPR RC.AREA.P DPR CWI - DPR RAIN
1 - 0.47 URBEXT

IBIDEM evaluates RC from a digitised version of the graph in Figure 3.10 of Houghton-Carr
(1999), with the y axis quantities divided by 10 to correct a mistake in the FEH. The fact that
RC is found graphically introduces a slight uncertainty to the calculation of SPR, but this has
been found to be small.

F3 Checks and validation of outputs


IBIDEM checks the outputs and provides error, warning or information messages when
necessary. A list of the checks made is in Table F.2.

185
Table F.2: Checks and outputs
Criterion Status Comment Notes
This could happen, for example, when the peak of
the input hydrograph is unrealistically low and the
catchment is urbanised. After the urban
component of the PR is calculated by IBIDEM, it
is sometimes found necessary to use a negative
SPR in order to match the input peak flow. A
Inferred SPR is
SPR 0 Error typical cause will be that the imported FSU peak
not positive
flow is too low, e.g. due to a failure to incorporate
an adequate urban adjustment in QMED.
Alternatively, the catchment may not be well
represented by the structure of the FSR rainfall-
runoff model or by the composition of the design
event used as the input to the model.
This may be valid if the catchment is highly
Inferred SPR is permeable. But it could otherwise be caused by the
0<SPR10 Warning
unusually low problems mentioned above, e.g. an unrealistically
low input peak flow.
Inferred SPR
implies a notably
10<SPR25 Information
permeable
catchment
Inferred SPR
implies a notably
50<SPR60 Information
impermeable
catchment
This may be valid if the catchment is extremely
Inferred SPR is
60<SPR100 Warning impermeable. But it could otherwise be caused by
unusually high
an unrealistically high imported peak flow.
This could happen when the peak of the input
Inferred SPR is
SPR>100 Error hydrograph is unrealistically high given the nature
more than 100%
and size of the catchment.
Inferred Tp is not This cannot happen, given the procedure used to
Tp0 Error
positive determine Tp. However, the check is still made.
This should not happen. PR is back-calculated
Inferred PR is using Equation F.4, which should not yield a
PR 0 Error
not positive negative result as all the other variables in the
equation can only take positive values.
Inferred PR is
75<PR100 Warning
unusually high
Inferred PR is
PR>100 Error This would typically happen only if SPR>100.
more than 100%

186

You might also like