You are on page 1of 132

THE UNIVERSITY OF HULL

Flood plain modelling based on


TRMM 3b42 satellite rainfall.

Lessons learned from the Semois basin

being a Dissertation submitted in partial f~lfilment of


the requirements for the-

Degree of Master of Science

GIS & Environmental Modelling

in the University of Hull

by

Albert Grela, Agr Eng

September 2010
Flood plain modelling based on TRMM 3b42 satellite rainfall.

Lessons learned from the Semois basin

Summary

The extensive flood damages sustained worldwide cost approximately 20 billions f: a year.

These natural catastrophes have a certain level of predictability allowing preparedness,

mitigation and defences. The key to minimizing the damages lies in the precision of the

prediction of the flood extent and dynamic. The present study explores the error propagation

from satellite rainfall estimates into the modelling of the flood plain extent in the Semois

basin (Belgium). The area near Villers-sur-Semois (49.69° Lat N, 5.56° Long E, 5.8 ha) has

been modelled with a ID2D software suites (ISIS free) forced with discharge inputs from 3

different sources : the observed gauged flow, the simulated flow based on rain gauges data

and the simulated flow based on the 3b42 v6. multi-satellite product of the Tropical Rain

Monitoring Mission (TRMM) providing a rainfall rate every 3hours for a cell size of 0.25

deg. Precipitations were transformed into dischal'ge with the revitalised FSRlFEH rainfall-

runoff method which is part the Flood Estimation Handbook in UK. The flood plain based on

the 3b42 dataappears grossly overestimated (135% of the observed flow flood extent) or

underestimated (50%) depending on the event chosen, c1early dismissing the adequacy of the

dataset. The flood plain modelled with the observed discharge and the one modelled based on

the rain gauge data appear quite similar in their maximum extent but present a different

temporal dynamic. They present a good similarity with the flood risk map pl'epared by the

regional authorities. The modelled flow based on the rain gauges has a Nash Sutcliffe

coefficient of efficiency between 0.23.and 0.33 according to the event modelled.

Additional online presentation of the study :

http://web.me.comlalbert.grelal3b42_flood...J)lain/Welcome.html
Abstract

Abstract
The extensive flood damages sustained worldwide cost approximately 20 billions f a

year. These natural catastrophes have a certain level of predictability allowing

preparedness, mitigation and defences. The key to minimizing the damages lies in

the precision of the prediction of the flood extent and dynamic. The present study

explores the error propagation from satellite rainfall estimates into the modelling of

the flood plain extent in the Semois basin (Belgium). The area near Villers-sur-

Semois (49.69° Lat N, 5.56° Long E , 5.8 ,ha) has been modelled with a ID2D

software suites (ISIS free) forced with discharge inputs from 3 different sources: the

observed gauged flow, the simulated flow based on rain gauges data and the

simulated flow based on the 3b42 v6. multi-satellite product of the Tropical Rain

Monitoring Mission (TRMM) providing a rainfall rate every 3hours for a cell size of

0.25 deg. Rain were transformed into discharge with the revitalised FSRlFEH

rainfall-runoff method which is part the Flood EstimatIon Handbook in UK. The

flood plain based on the 3b42 data appears grossly overestimated (135% of the

observed flow flood extent) or underestimated (50%) depending on the event

chosen, clearly dismissing the adequacy of the dataset. The flood plain modelled

with the observed discharge and the one modelled based on the rain gauge data

appear quite similar in their maximum extent but present a different temporal

dynamic. They present a good similarity with the flood risk map prepared by the

regional authorities. The modelled flow based on the rain gauges has a Nash

Sutcliffe coefficient of efficiency between 0.23.and 0.33 according to the event

modelled.
Acknowledgements

Acknowledgements

My deepest gratitude is due to my supervisor Dr. Tim Bellerby, University of Hull who
corrected many ofmy misunderstanding regarding the 3b42 rain product and helped me to
focus my efforts torward sorne achievable aims.

1 would have remained clueless about flood modelling without the careful tutoring of Dr.
Matt Horris and Dr. Mark Bailes from Halcrow. They are great trainers and very
experienced modellers.

Professor Keith Beven from Lancaster University de serves a special mention here. His
books have been a real guiding light in the complex world of hydrological uncertainties. 1
like to express my admiration for the depth of his knowledge and his very pedagogic writing
style. .

1 got excellent response and supply of data and information from Eng. Phillipe Dierickx
from "Direction de la Gestion hydrologique intégrée, Service Public de Wallonie" and Eng.
Didier de Thysebaert "Direction des Cours d'Eau non navigables - Service Public de
Wallonie". They both de serve my sincere thanks.

Marina Thunus from "Direction de la Gestion hydrologique intégrée - Service Public de


Wallonie" and Cecile Motte "Cellule CARTO, Service Public de Wallonie" have been very
cooperative in the supply of cartographic data. Their contribution is acknowledged.

Dave Brooks, Alex Duke, John Goodman, Alan Wright, Geoff Watson, Martin Burton,
Gerry Stephenson from the Cottingham Flood Action group were excellent guide in the
intricate world of flood defence planning in UK. They provided me a lot of inspiration and
their courage and resilience was contagious. They kept me sane and focussed.

Professeur Philippe Lejeune from the University of Liège with whom 1 graduated 26 years
ago shared his experience with various data set. He provided sorne useful advice and
companionship.

Philippe Hellemans from Ge06 is an other classmate tumed GIS expert, he shared his
experience with GIS dedicated to crisis management. He helped me to remain practical and
realistic.

Last but not least my daughter Shivata and wife Ragini de serve my most profound gratitude
for their patience and encouragement.
Table of content

Content

LIST OF TABLES •....•....••••........•...•••..•.•••.....••....•••...•••..•.•.•....••..••...•.•••.•.•...•..•••...•.••.•.••••.•..........•.•.•...v

LIST OF FIGURES ............................................................................................................................. VII

LIST OF ABBREVIATIONS AND ACRONYMS ...................................................................................... IX

1 GENERAL INTRODUCTION ....................................................................................................... 1

1.1 WHO IS TALKING? ••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••.• 1

1.2 THE FLOOD PLAIN, THE RUNOFF AND THE SATElLITE ............................................................................. 2

2 LITERATURE REVlEW............................................................................................................... 4

2.1 THE PRAGMATIC REAlISM AND INSTRUMENTAlISM .............................................................................. 4

2.2 ENVIRONMENTAL MODELlING EQUIFINAlITY ...................................................................................... 4

2.3 FLOOD DAMAGES .......................................................................................................................... 6

2.4 FLOOD MODELlING ....................................................................................................................... 8

2.4.1 The Google Scholar metric .............................................................................................. 8

2.4.2 The hydrological connections ......................................................................................... 9

2.4.3 The hydraulic models .................................................................................................... 11

2.4.3.1 10 modelling - The de St-Venant equations .....................................................................11

2.4.3.2 20 shallow water equations ..............................................................................................14

2.4.3.3 10 versus 20 floodplain inundation modelling ................................................................. 16

2.4.4 2 D flood modelling packages ...................................................................................... 19

2.5 RAINFALL-RUNOFF MODEL ............................................................................................................ 22

2.5.1 The Google Scholar metric ........................................................................................... 22

2.5.2 The rationales of rainfall-runoff modelling .................................................................. 23

2.5.3 The perceptual model................................................................................................... 23

2.5.4 The conceptual model .................................................................................................. 25

2.5.5 The procedural model .................................................................................................. 26

2.5.6 The calibration ............................................................................................................. 26


Table of content

2.5.7 Validation /performance evaluation ............................................................................ 27

2.5.8 Rainfall-runoff model classification .............................................................................. 28

2.5.9 ReFH package ............................................................................................................... 30

2.5.9.1 Introduction ......................................................................................................................30

2.5.9.2 Modelling concepts ...........................................................................................................32

2.6 SATELLITE RAINFALl. .................................................................................................................... 39

2.6.1 The rain gouges deficiencies ........................................................................................ 39

2.6.2 The satellite rainfall concepts and instruments ........................................................... 42

2.6.3 The Tropical Rainfall Monitoring Mission .................................................................... 47

2.6.3.1 Concepts and instruments ................................................................................................47

2.6.3.2 TRMM data products ........................................................................................................50

2.6.3.3 3b42 rainfall product ........................................................................................................51

2.6.4 3b42 rainfall runoff modelling...................................................................................... 53

2.6.4.1 Google Scholar metric .......................................................................................................53

2.6.4.2 Results of rainfall-runoff modelling with 3642 data ......................................................... 54

2.6.5 3b42 and flood warning system ......................................:............................................. 56

3 METHODOLOGY.................................................................................................................... 59

3.1 RESEARCH QUESTION AND RATIONALES ........................................................................................... 59

3.2 THE FLOOD PLAIN OF INTEREST ...................................................................................................... 61

3.3 THE REACH OF INTEREST ............................................................................................................... 62

3.4 THE CATCHMENT OF INTEREST ....................................................................................................... 63

3.5 THE PERIOD STUDIED ................................................................................................................... 63

3.6 THE CHOICE OF PASSIVE-MICROWAVE- AND INFRARED-BASED SATElLITE RAINFALL DATA ........................... 64

3.7 THE CHOICE OF 10-20 FLOOD PLAIN MODElllNG SOFTWARE .............................................................. 64

3.8 THE CHOICE OF RAINFALL-RUNOFF MODELlING PACKAGE .................................................................... 64

3.9 A BRIEF INTRODUCTION TO THE REFH PACKAGE OPERATION & PARAMETERS PREPARATION ....................... 65

3.10 THE RAINFALL DATA SOURCES & PROCESSING .............................................................................. 68

3.10.1 Rainfall Gouge ......................................................................................................... 68

11
Table of content

3.10.2 3b42 data set. .......................................................................................................... 69

3.10.3 Aggregation of rain gauges hourly data to 3 hourly data ....................................... 69

3.10.4 Comparison between 3b42 pixel and individual rain gauges .................................. 69

3.10.5 8ias .......................................................................................................................... 72

3.10.6 Root Mean Square Error .......................................................................................... 72

3.10.7 Probability of detection ........................................................................................... 72

3.10.8 False alarm Rate ...................................................................................................... 72

3.10.9 Heidke Skill Score (HSS) ............................................................................................ 72

3.10.10 Inverse distance weighing (lDW) ............................................................................. 73

3.11 SOURCE OF DISCHARGE DATA ................................................................................................... 74

3.12 SOURCE OF EVAPOTRANSPIRATION DATA .................................................................................... 74

3.13 PREPARATION OF CROSS SECTIONS ............................................................................................. 75

3.13.1 Correction of LJDAR data ......................................................................................... 75

3.13.2 Generation of cross sections with ISIS Mapper ,...................................................... 76

3.14 GENERATION OF FLOOD EVENTS AND ANTECEDENT RAINFALL SETS..................................................... 77

3.15 CALIBRATION AND VALIDATION SET ............................................................................................ 80

3.16 RUNNING ISIS 10 WITH THE VARIOUS FLOOD EVENTS ................................................................... 80

3.17 RUNNING THE ISIS 10 20 WITH THE VARIOUS FLOOD EVENTS ........................................................ 81

4 RESULTS& DISCUSSIONS •..•...•.........•......•.•..•.••....•••..•....•.••...•.••................•...•..•...•.....•........... 82

4.1 COMPARISONS OF 3842 DATA WITH RAIN GAUGES ............................................................................ 82

4.1.1 8ias ............................................................................................................................... 82

4.1.2 Root Mean Square Error............................................................................................... 82

4.1.3 Probability of detection ................................................................................................ 83

4.1.4 Fa/se a/arm Rate ........................................................................................................... 83

4.1.5 Heidke Skill Score (HSS) ................................................................................................ 83

4.1.6 Correlations .................................................................................................................. 83

4.1.7 Individual gauges comparison performance review .................................................... 84

4.2 COMPARISONS OF INVERSE DISTANCE WEIGHTED RAINFALL FOR THE WHOLE SEMOIS WATERSHED .............. 85

111
Table of content

4.3 COMPARISONS OF INVERSE DISTANCE WEIGHTED RAINFAll FOR THE ST MARIE-SUR-SEMOIS CATCH MENT... 85

4.4 COMPARISONS OF RAINFALL DURING FLOOD PERIODS ......................................................................... 86

4.5 COMPARISON OF RAINFALl DURING INTER-flOOD PERIODS ................................................................. 87

4.6 COMPARISON OF MODEllED FlOWS ............................................................................................... 88

4.6.1 Introduction .................................................................................................................. 88

4.6.2 Total, Average, Peak, Minimum& Error ....................................................................... 89

4.6.3 Graphie presentation .................................................................................................... 91

4.7 COMPARISON OF 10 MODElUNG RESUlTS ....................................................................................... 93

4.7.1 Introduction .................................................................................................................. 93

4.7.2 Maximum stage and overfJow depth ........................................................................... 93

4.8 COMPARISON OF THE 10-20 MODElUNG RESUlTS ........................................................................... 99

5 CONCLUSIONS •••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••• 103

5.1 FLOOD PLAIN MODElllNG ........................................................................................................... 103

5.2 REFH ADEQUACY OUTSIDE UK .................................................................................................... 103

5.3 3842 VAUDITV UNDERTHE 50° LAT N...................................... ;................................................... l04

5.4 LIDAR DATA PROCESSING .......................................................................................................... 104

5.5 DATA PROCESSING CAPACITV ....................................................................................................... 105

5.6 PROPAGATION OF ERROR AND RECOMMENDATION FOR lOCAL OBSERVATIONS ...................................... 106

5.7 FURTHER WORK ........................................................................................................................ 106

6 BIBLIOGRAPHY ••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••• 108

Online presentation of the study at :


http://web.me.comlalbert.grela/3b42_flood--p1ain/Welcome.html

lV
List of tables

List of tables

Table 1 : Flood damages since 1985 ............................................................................ 7

Table 2 : Flood & inundation modelling - Google Scholar hits ................................... 8

Table 3 : Most common 2 D packages in UK (Néelz and Great, 2009) ..................... 21

Table 4 : Classification of inundation packages (Néelz and Great, 2009) ................. 22

Table 5 : Rainfall-runoffhits - Google Scholar ......................................................... 23

Table 6 : Meteorological satellites ............................................................................ .46

Table 7 : TRMM data level (NASDA, 2001 ) ............................................................. 50

Table 8 : TRMM Data products (NASDA, 2001) ...................................................... 50

Table 9 : TRMM data flow (NASDA, 2001) ............................................................. 51

Table 10 : Google Scholar hits for 3b42 & hydrological models, flood, runoff ........ 53

Table Il: 3b42 pixel paired with rain gauge .............................................................. 71

Table 12 : Flood and flood antecedent periods ............. ,..... :...................................... 78

Table 13 : Individual gauges comparison .................................................................. 82

Table 14 : Correlations coefficients with individual rain gauges .............................. 84

Table 15 : Five major flood events ............................................................................. 86

Table 16 : Accumulated rainfall during five floodlhigh flow events ......................... 86

Table 17 : Nbr of 3 hours period > 0 mm of acc. rainfall- Flood periods ................ 87

Table 18 : Inter-floodlhigh flow periods .................................................................... 87

Table 19 : Accumulated rainfall during five inter-flood/inter-high flow events ........ 88

Table 20 : Nbr of 3 hours period > 0 mm of acc. rainfall- inter flood periods ......... 88

Table 21 : Flow descriptors - one hour timestep ........................................................ 89

Table 22 : Flow descriptors 3hours timestep ............................................................. 90

Table 23 : Maximum stages and overflow depth ....................................................... 94

v
List of tables

Table 24 : Flood maximum extent near Villers-sur-Semois (m2) •••••••••••••••••••••••••••••• 99

Table 25 : Ratio ofFLl et FL2 flood extent with a depth superior to 30 cm and area

classified as high risk ............................................................................................... 102

Vi
List of figures

List of figures

Figure 1 : Taxonomy ofhydrological models ............................................................ 10

Figure 2 : A schematic outline of the different steps in the modelling process ......... 11

Figure 3 : A classification of process mechanisms in the response of hillslopes to

rainfalls. (a) Infiltration excess overland flow (Horton, 1933). (b) Partial area

infiltration excess overland flow (Betson, 1964). (c) Saturation excess overland

(Cappus~ 1960~ Dunne and Blac~ 1970). (d) Subsurface stormflow (Hursh~ 1936~

Hewlett, 1961). (e) Perched saturation and' throughflow (Weyman, 1970) ............... 25

Figure 4 : ReFH model concept diagram ................................................................... 34

Figure 5 : Cumulative distribution of soil moisture capacity ..................................... 35

Figure 6 : Shape of standard instantaneous unit hydrograph adopted in ReFH ......... 37

Figure 7 : GHCN-Monthly Coverage Map for Precipitation (Burroughs, 2008) ...... .42

Figure 8 : TRMM sensors chart .................................... ,..... ~ ...................................... 49

Figure 9 : Location of Villers-sur-Semois .................................................................. 62

Figure 10 : Location of the modelled reach ............................................................... 63

Figure Il : Rain gauges localisation .......................................................................... 70

Figure 12 : 3b42 pixel centroid and rain gauges location ......................................... 71

Figure 13 : Localisation of long section ..................................................................... 76

Figure 14 : Cross-section localisation ........................................................................ 76

Figure 15 : Long section before DEM correction ...................................................... 76

Figure 16 : Cross-section before DEM correction ..................................................... 76

Figure 17 : Long section after DEM correction ......................................................... 76

Figure 18 : Cross-section after DEM correction ........................................................ 76

Figure 19 : 3D view before DEM correction ............................................................. 76

vu
List of figures

Figure 20 : 3D view after DEM correction ................................................................ 76

Figure 21 : Generation of cross-sections in ISIS Mapper .......................................... 77

Figure 23 : Maximum stage 5 flood events ................................................................ 79

Figure 23 : FLI rain gauges modelled ........................................................................ 91

Figure 24 : FL1 3b42 modelled .................................................................................. 91

Figure 25 : FL2 rain gauges modelled ........................................................................ 92

Figure 26 : FL2 3b42 modelled .................................................................................. 92

Figure 27: FL1 Maximum stages Rain Gauged Modelled & Gauged Flow ............. 95

Figure 28 : FL 1 Maximum stages: 3b42 Modelled Flow & Gauged Flow .............. 96

Figure 29: FL2 - Maximum stages - Rain gauges modelled & Gauged Flow .......... 97

Figure 31 : FL2 Maximum stages 3b42 modelled & Gauged Flow ........................... 98

Figure 31 : FL 1 flood extent .................................................................................... 100

Figure 32 : FL2 flood extent .................................................................................... 100

Figure 33 : FLI Flood extent with a depth of at least 30 cm and official flood risk

map ........................................................................................................................... 101

Figure 34 : FL2 Flood extent with a depth of at least 30 cm and official flood risk

map ........................................................................................................................... 102

Vlll
List of abbreviations and acronyms

List of abbreviations and acronyms

Acronym Signification
Adjusted Geostationary Operational Environmental Satellite Precipitation
AGPI Index (AGPI)]
ALTBAR Mean altitude of catchment (metres above sea level)
AMSR-E Advanced Microwave Scanning Radiometer-EOS
AM SU Advanced Microwave Sounding Unit
AMSU-A Advanced Microwave Sounding Unit (on Aqua satellite)
AMSU-B Advanced Microwave Sounding Unit (on NOAA satellite)
ASPBAR Mean direction of aU drainage path slopes (bearing in degrees)
ASPVAR Invariability of slope d~ections

ASTER Advanced Spacebome Thermal Emission and Reflection Radiometer


AVHRR Advanced Very High Resolution Radiometer
BaseFlow Index derived using the HOST (Hydrology Of Soil Types)
BFIHOST classification
BL Baseflow recession constant (lag)
BR Baseflow recharge
CEH Centre for Ecology and Hydrology
CERES Clouds and the Earth's Radiant Energy System
CMax Maximum Soil Moisture Capacity
Cmorph NOAA CPC Morphing Technique
DK Daily Soil Moisture Decay rate
DMSP Defense Meteorological Satellite Program
Mean of distances along drainage paths between each 50m grid no de and
DPLBAR the outlet
DPSBAR Mean of an inter-nodal slopes, along drainage paths (mIkm)
DSS differential split-sample test
EOS-AM Earth Observing System launched 1998
EOS-PM Earth Observing System launched 2000
ESMA explicit soil moisture accounting
ESMR Electrically Scanning Microwave Radiometer
Index of flood attenuation due to reservoirs and lakes (1.0 indicates no
FARL attenuation)
FC Field Capacity
FD Finite difference
~:. _0 ..
FEH Flood Estimation Handbook
FSR Flood Study Report
GDEM Global Digital Elevation Model
GFM Global Flood Monitoring
GLUE Generalised Likelihood Uncertainty Estimation
GOES Geostationary Operational Environmental Satellites
GOMS Geostationary Operational Meteorological Satellite
GPCP Global Precipitation Climatology Project (GPCP)

lX
List of abbreviations and acronyms

GPCP Global Precipitation Climatology Project (GPCP)


GPCP Global Precipitation Climatology Project
GPM Global Precipitation Measurement
GPROF Graph Execution Profiler
GSMap_MVK Global Satellite Mapping of Precipitation with Micro Wave sounder
HEC Hydrologic Engineering Center
HIRS High Resolution InfraRed Sounder
HMS Hydrologic Modeling System
INSAT Indian National Satellite System
IR Infrared
IUH Instantaneous Unit Hydrograph
JAXA National Space Development Agency of Japan
LDP Longest drainage path
LiDAR Light Detection And Ranging
LIS Lightning Imaging Sensor
MODIS Moderate Resolution Imaging Spectroradiometer
MS multi-satellite
MSU Microwave Sounding Unit
NASA National Aeronautics and Space Administration in USA
NASDA National Space Development Agency of Japan
NOAA National Oceanic and Atmospheric Administration in USA
NSE Nash Sutcliffe coefficient of efficiency
PB proxy-basin test
PBDSS proxy-basin differential split-sample
PDM Probability Distributed Model
Precipitation Estimation from Remote Sensing Information using
PERSIANN Artificial Neural Network.
POES Polar Orbiting Environmental Satellites POES
PR Precipitation Radar
PROPWET Proportion of time when soil moi sture deficit was below 6mm
QI total observed flow at time t
RD Routing depth
ReFH Revitalised Flood Hydrograph
RMEDID Median annual maximum I-day rainfall (mm)
RMEDIH Median annual maximum I-hour rainfall (mm)
RMED2D Median annual maximum 2-day rainfall (mm)
RMS Root Mean Square
RMSE Root Mean Square Error
RW Region Wallonne
SAAR average annual rainfall
SG satellite-gauge combination
SMMR Scanning Multichannel Microwave
SMS Synchronous Meteorological Satellites
SPRHOST Standard Percentage Runoff derived using the HOST classification
SPW Service Public de Wallonnie

x
List of abbreviations and acronyms

SRFE Satellite Rainfall Estimate


SRTM Shuttle Radar Topography Mission
SS split-sample test
SSMII Special Sensor Microwavellmager
SSU Stratospheric Sounding Unit
TCI TRMM combined instrument
TIROS-I Television and Infrared Observation Satellite
TM! Microwave Imager
TMPA TRMM multi satellite precipitation analysis
TOVAS TRMM Online Visualization and Analysis System
Tp Time-to-peak
TRMM Tropical Rainfall Measuring Mission
UH Unit Hydrograph
UK United Kingdom
Uk KinkofIUH
Up Height ofIUH
URBEXT Extent of urban and suburban land coyer
USA United States of America
UTC Coordinated Univers al Time
VIRS Visible and Infrared Scanner

Xl
General introduction

1 General introduction

1.1 Who is talking?

Although the author of the present study is confined by academic rules to present his

work without using the first person, semantic and epistemological demands cannot

avoid the elucidation of the philosophical position used to formulate and credit the

present work. The distortion of discourse is entirely the responsibility of the

University's of Hull Department of Geogniphy. The elucidation of the philosophical

and personal element of a scientific study is recommended by a wide spectrum of

scholars ranging from Edgar Morin(Morin, 1977) and Michel Foucault (Foucault,

1976) to Keith Beven (Beven, 2009) and Noel Castree (Castree et al., 2005).

The Karl Poper's the ory ofverisimilitude (Popper, 1963) is totally embraced by the

present study . The intention of pragmatic realism is admowledged : the modelling

effort aims at representing as weIl as possible the reality but the limitations of

observations and computational capacities restrict the "fitting",- as long as the models

allows sorne useful predictions, they should be granted sorne scientific respectability.

(Beven, 2009)

Although a very minimal treatment of the uncertainty could be provided within the

timeframe of the present study, the environmental modelling and systematic

elucidation of the uncertainty should be pronounced in a same breath.(Beven, 2009).

The equifinality of hydrological modelling is also accepted as a rather reassuring

concept rather than a suggestion of inadequate science. (Beven, 2006, , 2009,

Savenije, 2001) .

1
General introduction

1.2 The flood plain, the runoff and the satellite

During the decade ending in 2008, the number of people affected by flood reached

1.037 billions according to the World Disaster report 2008(IFRCRCS, 2010). This

year, the flood in China and Pakistan will add 32 millions of displaced

people.(Reliefweb, 2010a, Reliefweb, 2010b). During the last 25 years, the average

annual flood damages exceeds 30 billions USD (Brakenridge, 2010). Prevention and

mitigation are crucial to face this natural disaster only second to food insecurity

(IFRCRCS, 2010). The capacity to model flood plain extent is a key element to

quantify the flood risk, design flood defences and prepare emergency plans. (Néelz

and Great, 2009). This modelling process presuppose the ability to estimate the

discharge in the water course and the surface runoff during a flood event.(Shaw et

al., 2010, Beven, 2004). This estimation is constraint by the availability of weIl

established stream gauges and in the absence of them many plans have been made

based on a simulated discharge through a rainfall-runoffmodel ((Kjeldsen and Jones,

2009, Kjeldsen, 2007, Kjeldsen et al., 2006). The recent availability of satellite

rainfall products and their growing historical depth offer a opportunity to

compensate the deficiencies of rain gauges as a prime of rainfall information

((Vazquez and Feyen, 2010, STISEN et al., 2010, Lohani et al., 2010, Harrison et al.,

2009, Sugiura et al., 2009, Valeriano et al., 2009, Su et al., 2008, Collischonn et al.,

2008a, Magome et al., 2008, Werner and Delft, 2005). Although many researches

highlight the importance of flood forecasting only a very few (Fotopoulos et al.,

2010, Magome et al., 2008, Kuszmaul et al., 2008, Hossain et al., 2007, HOSSAIN

et al., 2004) elaborate a real flood detection method and quantify its error and

uncertainty. The present study appears rather original as the impact of the error in

rainfall estimate will be carry over until the ca1culation of the flood extent. Such

2
,, '
General introduction

approach bears a certain level of irreconcilability as the flood plain modelling

requires extremely detailed elevation data while the satellite rainfall products are

meant to coyer large swat of land. The exercise offers the possibility to focus the

analysis of the errors into operational values such as the foreseeable extent of the

flood plain. The non-linearity of the rainfall-runoff relation and the specificity of the

topography of the flood plain prevent the estimation of the flood plain error based on

the rainfall error.(Beven, 2006, Beven, 2004, Beven, 1993, Beven, 1989). The

present study was extremely constrained by time, software availability, local

expertise and word count. The literature review in chapter 2 tries to explore briefly

three huge body of sciences and research : the flood modelling techniques, the

rainfall-runoff models and the satellite rainfall product. The methodology presents in

chapter 3 the selection of location and data sets which was based more on practical

constraints than refinement of scientific approach. The task to assemble and process

all the data required was huge and the experience with hydraulic modelling was nil.

A full year may have been spent on the subject without exhausting it. The results are

satisfactory in tenus of internai consistency but dismiss the opportunity to use the

Tropical Rainfall Monitoring Mission data for flood plain modelling in the 50° lat N.

But using a better rain sources (rain gauges) in a rainfall-runoffmode1 seems a

appropriate option for flood plain modelling. The future satellite rainfall products are

likely to provide much better estimates of rainfall and soi! moisture and should

improve the hydrological modelling particularly for the higher latitude. (NASA,

2010a, Beven, 2009). It is hoped that these improvements will translate in better

management of flood risks and decrease the fatalities observed so sadly during the

summer 2010.

3
Literature review

2 Literature review

2.1 The pragmatic realism and instrumentalism

The Karl Popper (Popper, 1963) states that the most that could be hoped from a

scientific the ory is a certain degree of verisimilitude. He rates the closeness to Truth

by two factors : truth and content. The more truths that a theory entails (other things

being equal) the closer it is to the truth.

The intention of pragmatic realism is quite simpler : the modelling effort aims at

representing as weIl as possible the reality but the limitations of observations and

computational capacities restrict the "fitting", as long as the rnodels allows sorne

useful predictions it can be granted sorne scientific respectability. "AlI models are

wrong but sorne models are usefull" (Box, 1979).

The instrumentalists' views that aIl scientific theories of the past have proven to be

false to sorne extent, convey the idea that the current one will also turn out to be false

too. That does not prevent the enunciations of scientific statements but these are not

the reality. Instrumentalism in that sense is anti-realist especially since it also allows

that sorne statements about the nature ofthings may be subjective. Logically it

derives that the only justification for scientific theorising is empirical adequacy.

(Van Fraassen, 1980, Cartwright, 1983, Haack, 1994, Haack, 2002, Haack, 2003)

2.2 Environmental modelling equifinality

According to Ludwig von Bertalanffy, equifinality is the principle that in open

systems a given end state can be reached by many potential means. He prefers this

term, in contrast to "goal", in describing complex systems' similar or convergent

4
Literature review

behaviour. It emphasizes that the same end state may be achieved via many different

paths or trajectories. In closed systems, a direct cause-and-effect relationship exists

between the initial condition and the fmal state of the system. The idea of

equifinality suggests that similar results may be achieved with different initial

conditions and in many different ways. This phenomenon has also been referred to as

isotelesis (Von Bertalanffy, 1950).

The term equifinality has a long history in geomorphology, indicating that similar

landforms might arise as a result of quite different sets of processes and

histories.(Beven, 1996)

Landscape evolution models have recently been tested about their

equifinality.(Odoni, 2007)

But the most prolific advocate of the equifinality concept i:p. hydrological modelling

is most likely Keith Beven (Beven, 1996, , 1997, , 1998, Schulz et al., 1999, Zak and

Beven, 1999, Brazier et al., 2000, Beven and Freer, 2001, ,2006) . The origin of the

concept lies in purely empirical studies that have found many models giving good

fits to data. But for many modellers, the optimal model is still an aim that should be

pursue and the present "imperfections" of the model are a step toward the optimal

model(Beven, 2006).

The Generalised Likelihood Uncertainty Estimation (GLUE) methodology proposes

an evaluation frame to test the fitting and the uncertainty in a consistent

manner.(Beven and Binley, 1992). It intends to focus the attention on the fact that

there are many acceptable representations that cannot be easily rejected and should

be considered in assessing the uncertainty associated with predictions. The concept

5
Literature review

owes a lot to the Homberger-Speer-Young (HSY) sensitivity analysis of multiple

behavioural models (Homberger and Spear, 1981).

For any particular set. of observation, sorne of the acceptable or behavioural models

will be better in terms of one or more performance measures. But given the sources

of error in the modelling process, the behaviourals models cannot easily be rejected

as feasible representations of the system given the level of error in representing the

system. This can be viewed as a problem of decidability between feasible

descriptions of how the system is working.' (Beven, 2006)

To be able to represent different hypotheses about the processes of a hydrological

system, it is necessary to have sorne parametrisation of these processes. This usually

leads to a growth of complexity and multiplication of parameters quite often without

additional data collection to determine these additional parameters. Equifinality is

therefore rather inevitable because even in case of mathematically perfect model the

errors in the initial conditions, boundary conditions and output measurements leads

to a cloud of equally fitting solutions.

Environmental models are therefore mathematically ill-posed or ill-conditioned. The

information content available to define a modelling problem does not allow a single

or unambiguous mathematical solution to the identification problem.(Beck, 1987)

2.3 Flood damages

The best source regarding recent flood statistics is most likely the Dartmouth Flood

Observatory.(Brakenridge, 2010) The database holds the records of3703 floods

dating back to 1985.

A summary of the available data is presented in Table 1 : Flood damages since 1985
6
Literature review

Table 1 : Flood damages since 1985

Nbrof Sumof Sumof Sum of Damage


Years floods Dead Displaced (USD)
1985 69 3,034 5,831,709 5,882,203,782
1986 47 1,554 8,862,982 75,157,098,463
1987 45 3,218 1,452,016 2,667,700,000
1988 111 6,229 20,317,659 11,523,995,494
1989 112 9,838 8,620,018 1,673,442,200
1990 105 4,170 14,932,486 9,554,391,478
1991 122 150,169 17,804,633 82,285,582,000
1992 109 8,720 13,078,525 24,929,269,900
1993 99 7,453 35,102,006 20,759,890,046
1994 107 6,401 8,574,572 13,542,094,000
1995 115 7,625 48,575,631 45,860,002,000
1996 100 5,974 12,857,803 11,586,866,000
1997 159 10,361 6,475,540 11,140,926,000
1998 180 21,331 42,880,501 235,875,681,175
1999 100 33,792 57,106,721 55,222,677,688
2000 102 10,684 50,228,261 13,381,700,000
2001 172 5,624 37,993,069 13,217,061,635
2002 260 4,421 20,516,442 29,694,450,684
2003 297 4,554 21,667,574 7,564,443,015
2004 194 173,123 51,018,052 7,550,518,000
2005 171 10,140 19,367,962 82,249,650,000
2006 232 7,992 18,710,396 9,312,081,480
2007 243 12,506 36,013,671 22,265,410,000
2008 179 107,612 22,260,628 8,279,760,000
2009 158 3,897 8,804,644 387,000,000
2010 115 6,110 21,234,094 20,005,000
Grand Total 3703 626,532 610,287,595 801,583,900,041
Average 142 24,097 23,472,600 30,830,150,002

The damages cost of China and Pakistan floods in 2010, is probably not yet

reflected. UN officiaIs at the end of August 2010 report 14.5 millions displaced in

Pakistan (NydailyNews) and Chinese authorities are estimating the direct economic

losses at 3.9 billion USD (Xinhua, 2010).

The figures presented in this table coyer the last 25 years only. The deadliest flood in

history reached an estimated 3.7 millions fatalities in 1931 in China.(Wikipedia,

2010)
... "'~-"".""

7
Literature review

The 2007 flood in UK cost around 6 billions GBP with 13 casualties.(pitt, 2007)

These figures leave no doubt about the social priority that flood defence, flood

warning and flood mitigation may have.

2.4 Flood modelling

2.4.1 The Google Scholar metric

Searching Google Scholar about flood modelling faces two possible spelling for

modelling and a partial synonymy between flood and inundation.

Hydrologists seem to prefer flood while engineers may focus on inundation. The

recent Desktop review of 2D hydraulic modelling packages commissioned by the

UK Environmental Agency employs the words without much discrimination.

Inundation may be more urban, flood plain, flood extent may be more rural.

The number of hits excluding citation retumed by Google Scholar for the three

periods (before 1990, 1991 to 2000, after 2001) is presented in Table 2

Table 2 : Flood & inundation modelling - Google Scholar hits

inundation inundation inundation flood flood flood


Year modelling modeling model mode Il ing modeling model

<1990 8,320 8,890 8,700 18,300 18,700 23,600

1991-2000 9,790 9,760 9,430 17,300 17,400 26,300

2001~2010 17,000 17,000 17,300 17,300 17,400 24,200

The inundation model, model(l)ing has a fast growth during the present decade. The

flood model, model(l)ing is rather stagnating with even a slight dec1ine during the

8
Literature review

present decade but this may just reveal a change of vocabulary rather than a decline

in the number of publication related to the "unintended submersion with water".

2 .4.2 The hydrological connections

The flood modelling is a sub-division of a vast domain of hydrological modelling.

Hydrologie models are simplified, conceptual representations of a part of the

hydrologie cycle. They are primarily used for hydrologie prediction and for

understanding hydrologie processes. (Shaw et al., 2010).

The methods and concepts relating the movement of water segment the hydrological

modelling in : groundwater modelling, surface water modelling and composite

models.(Beven, 2004, Shaw et al., 2010).

Groundwater models represent groundwater flow systems, and are used by

hydrogeologists. Groundwater models are employed to si~ulate and predict aquifer

conditions. Surface water modelling are subdivided in rainfall-runoffmodel and

hydrological transport model. The composite model combine water movements

above and underground.

The flood modelling usually involve two type of modellings : the rainfall-runoff

model helps to derive a discharge from the rainfall (tbis particularly applies to

ungauged catchement) and the hydraulic model use the hydraulic laws to calculate

velocity and water level along the flow path.(O'Connell et al., 1970, Beven, 1993,

AI-Sabhan et al., 2003, Lai, 2005, Bastin et al., 2009, Sugiura et al., 2009, Shaw et

al., 2010).

9
,
"

::.
,
.':;

~
~"
=
""1
~

~
~
~
0
=
0
3
~

....
0

=-
~
Co System
""1
0
0'
f (randomness, space, time)
(JQ
r;"
~
3
0
Co Randomness?
~
(;;"
no yes

...... Spatial Variation?


0

Temporal Variation? r
&'
~
(t)

~
~'
Literature review

Chow presents a taxonomy of hydrological models based on randomness, space and

time. Figure 1 : Taxonomy ofhydrological models.

Revise perceptions The Perceptual Model: 3"


0
deciding on the pro cesses m
Pl
en
5"
Revise equations co
The Conceptual Model: 1»
"0
deciding on the equations '0
a
X
.:".~~.':"~'.
""
Debug code
3"
The Procedural Model: ~
getting the code to run on a computer 0"
::J

Revise parameter Model Calibration: ~


values getting values of parameters

Model Validation:
• ~ ~-:: 1.1 ~
good idea but difficult in practice

No

Figure 2 : A schematic outline of the different steps in the modelling process

In the modelling process outlined in Figure 2, Beven insists on the initial perceptual

model that he considers more complex that the conceptual model, but he views that

the reduction of complexity is often essential to reach mathematically supported

predictions.

2.4.3 The hydraulic models

The most important concepts of mathematical modelling of shallow water flows are
introduced first in one dimension (St-Venant equations), then in two dimensions
(shallow water equations). Note that the following sections are partly reproduced
from "Desktop review of 2D hydraulic modelling packages" (Néelz and Great, 2009)

2.4.3.1 1D modelling - The de St-Venant equations

The St-Venant equations can be expressed as follows:

dQ + dA = 0 Equation 1
dx dt

Il
Literature review

!:. dQ + !:.~ (Q2) + 0 dh - o(So - S.) = 0 Equation 2


A dt A dx A dx }

(i) (ii) (iii) (iv) (V)

where Equation (1) is referred to as the continuity or mass conservation

equation, and Equation (2) is the momentum conservation equation. In this,

Q is the flow discharge (Q = U.A where U is the cross-sectional averaged

velocity and A in the cross-section surface area), g is the acceleration of

gravity, h is the cross-sectional averaged water depth, So is the bed slope in

the longitudinal direction and SI is the friction slope (the slope of the energy

line).(Néelz and Great, 2009)

The various terms in the momentum conservation equation are as follows:

(i) local acce1eration term;

(ii) advective acceleration term;

(iii) pressure term;

(iv) bed slope term

(v) friction slope term

Here the momentum equation is expressed in conservative form. It is possible

to substitute UA for Q in Equations (1) and (2), expand Equation (2) and

simplify it using Equation (1) to yie1d the mathematically correct non

conservative form of the momentum equation. Use of the non-conservative

form may, however, lead to practical difficulties in its numerical

solution.(Néelz and Great, 2009, Sleigh and Goodwill, 2000)

The equations are referred as a set ofhyperbolic partial differential equations.

12
Literature review

A number of theoretical assumptions must be met for the St-Venant

equations to apply mainly:

• the bed slope is small

• the pressure is hydrostatic, that is, streamline curvature is small

• and vertical accelerations are negligible

• the effects of boundary friction and turbulence can be accounted for

by representations of channel conveyance derived for steady-state

flow.

The St Venant equations cannot be solved explicitly except by making some very
large assumptions which are unrealistic for most situations. Therefore numerical
techniques have to be used. Three groups of methods are usually used : fmite
difference, fmite elements and finite volumes.(Sleigh and Goodwill, 2000)

Finite difference (FD) methods rely on Taylor series expansions to express the value
taken by a variable (h, u, v and so on) at a given point, as ~ function of the values at
neighbouring points and of local derivatives ofincreasing orders. These Taylor
series are then combined to yield approximate expressions for the derivatives
involved in the shallow water equations, as a function of a fmite number of
neighbouring point values. The accuracy of the approximations can be controlled by
the order to which the Taylor series expansions are developed (the order of the so-
called truncation), which is also linked to the number of neighbouring points
involved. The implementation of finite difference methods is significantly more
straightforward on a structured grid, which is a computational grid that can
effectively be represented on a square matrix. This explains to some extent why their
popularity is currently in decay in the academic community (Alcrudo, 2004), as
unstructured grids lend themselves better to the modelling of environmental flows.
Software packages based on FD methods, however, are popular with a number of
UK consultants, due mainly to their compatibility with high resolution digital terrain
models and digital bathymetric models created from LiDAR and sonar
surveys.(Néelz and Great, 2009)

For the finite element methods, the solution space in divided into a number of
elements in 2D. In each element, the unknown variables are approximated by a linear
combination of piecewise linear functions called trial functions. There are as many
such functions as there are vertices defining the element, and each takes the value of
one at one vertex and the value of zero at all other vertices. A global function based

13
Literature review

on this approximation is substituted into the governing partial differential equations.


This equation is then integrated with weighting functions and the resulting error is
minimised to give coefficients for the trial functions that represent an approximate
solution (Bates et al., 2005).

Finite element methods benefit from a rigorous mathematical foundation (A1crudo,


2004) that allows a better understanding oftheir accuracy (Hervouet, 2007);
however, the technique has not been used as much as other approaches in
commercial software, perhaps because it is less accessible conceptually and produces
models that result in large run-times. Also, generating meshes can be time-
consuming when a suitable mesh generation tool is not available(Néelz and Great,
2009).

In the finite volume methods, space is divided into so-called fmite volume which are
2D (in this context) regions of any geometric shapes. The shallow water equations
(in conservative form) are integrated over each control volume to yield equations in
terms of fluxes through the control volume boundaries. Flux values across a given
boundary (ca1culated using interpolated variables) are used for both control volumes
separated by the boundary, resulting in the theoretically perfect mass and momentum
conservativeness of the approach (a flux into a finite volume through a boundary is
always equal to a flux out of a neighbouring one through the same boundary). In ID,
fmite volume methods are equivalent to fmite difference methods.(Néelz and Great,
2009)

One of the most significant advances in 1D models is the ability to link 1D and 2D
models (Syme, 1991, Evans et al., 2007). This can be applied in various ways:

• within a channel that one wishes to model partly in ID and partly in 2D;
• between a 1D drainage network model and a 2D surface flood model;
• between a 1Driver model and a 2D floodplain model;
• within a mainly 2D model where, for example, cul verts are modelled in ID
linking 2D cells between themselves.(Néelz and Great, 2009)

2.4.3.2 2D shallow water equations

The two-dimensional shallow water equations expressed in vector form are:


dU
dt
+ dF
dx
+ dG
dy
=H Equation 3

1
u= [f:l _
G-
[9
huv
h2
hv
+ hv 2
2
'

14
Literature review

Equation 4

where u and V are the depth-averaged velocities in the x and y directions,

respectively. Sox and Soy are the bed slopes in the x and y directions. The friction

slopes in the x and y directions can be expressed in a manner analogous to the 1D

formulation, as follows (assuming the use of Manning's n) (Néelz and Great, 2009):

n 2 u .Ju 2 + v 2 n 2 v .Ju 2 + v 2
SIx =- 4 and S,y =- 4 Equation 5
h'J h3

The viscosity is taken into account in more refme formulation but not presented here.

The contribution of the kinematic viscosity to the value of the viscosity coefficient E

is typically at least an order of magnitude smaller than the turbulent eddy viscosity

and for this reason is neglected. The apparent viscosity resulting from non-

uniformity in the horizontal velo city along the vertical direction is recognised as a

much more significant contributor to the value of E (Alcrudo, 2004). However, this

effect is poorly understood and is therefore neglected in most applications. The,

turbulent eddy diffusivity has been the object of more significant research (see Rodi

1980), but in the context of flood modelling it is generally not considered an

important parameter (Alcrudo, 2004). For overland flow conditions, it is unlikely

that the eddy viscosity will have a major effect on model predictions as friction will

dominate. It may however, for flow in and around structures, have a significant

effect upon local high-resolution predictions (Néelz and Great, 2009).

Similarly with the 1D formulation, it is possible to neglect the acceleration terms in

the 2D shallow-water equations (the terms involving u and v in U, F and G) to yield

15
Literature review

the 2D diffusion wave equations (Bradbrook et al., 2004). This is appropriate where

the flow is predominantly driven by local water surface slope and momentum effects

are less important, as is often the case in the context ofUK fluvial floodplains. Such

modelling approaches and recent practical applications are discussed in Hunter et

al.(2007).

An important mathematical property of the shallow water equations is that they are

non-linear (they do not satisfy the principle of superposition), in accordance with the

true non-linearity of the flow processes being modelled. One of the implications is

that shallow water flows are subject to shock waves, which are understood to be

discontinuous solutions of the shallow water equations (Toro and Toro, 2001).

Shocks on floodplains are mainly encountered in the form ofhydraulic jumps, that

is, transitions from supercritical to subcritical flows. These may be caused by local

changes in terrain topography (diminution of bottom slope, lateral expansion), or by

the effect of bottom friction.

2.4.3.3 1D versus 2D floodplain inundation modelling

The choice between a 1D and a 2D mode1 is relevant primarily in the context of river

floodplain modelling. The the ory of open channel flow in the form of ID St-Venant

equations is not applicable to urban flood flows where extreme non-uniformity and

spatial variability of flow patterns is common. Flows may happen in sequences of

fast moving shallow flows (possibly supercritical) and large still ponds, rather than

in the form of channe1s that are weIl defined over long distances. The significance of

storage and recirculation areas that clearly do not fit in a 1D description should not

be underestimated. Besides, urban flows rarely happen along routes that are clearly

identifiable in advance to allow the building of a model and running the simulations

16
Literature review

(unlike rivers). However, a case where ID modelling is as close as possible to being

appropriate can be found, for example, in Lhomme and al. (2006) (deep flooding in a

network of weIl defmed narrow streets). Similarly, in coastal flooding it is not

generally the case that floodplains may be reasonably represented as networks of

well-defined channels and therefore 1D floodplain modelling is rarely appropriate.

In river flooding contexts, however, ID (that is, ID models ofrivers with cross-

section extending over lateral floodplains) are appropriate for narrow floodplains,

typically where their width is not larger than three times the width of the main river channel.

The underlying assumption should be that the contribution of the floodplains to conveyance

can be quantified using recent advances in the estimation of compound channel conveyance.

An additional condition for such models to be valid is that the floodplains should not be

separated from the main channel by embankments, levees or any raised ground, where the

channel floodplain unit effectively behaves as a single channel. It is clear that ID river

models have limitations that can become significant in matiy practical applications.(Néelz

and Great, 2009) The flow is assumed to be unidirectional (generally happening in the

direction parallel to the main channel flow), and where this is not true (recirculation areas)

conveyance predictions can be severely overestimated. Situations where floodplain flow

"makes its own way" are frequent, but perhaps an even more significant issue is the fact that

ID cross-sections will offer a rather crude representation of floodplain storage capacity in

the case of large floodplains. A better balance between the correct representation of

floodplain conveyance and the correct representation of floodplain storage capacity can be

obtained through the use of ID+ models, where large "disconnected" floodplains are

modelled as storage reservoirs (white narrow floodplains can still be mode lied as part of

channel cross sections). This latter modelling approach has its own limitations: exchange

flows between the river and reservoirs and between the reservoirs are typically mode lied

using broad-crested weir equations (Evans et al., 2007), which are not always appropriate.

17
Literature review

Weir equations adapted for drowned (downstream controlled) flows are also used,

but the assumption that water levels are horizontal within each reservoir results in

incorrect water level predictions in the vicinity of reservoir boundaries, often causing

large errors in the predictions of exchange flows. These do not matter if the time

duration of the floodplain filling and draining is srnall cornpared to the duration of

the flood. Lastly, the size and location of floodplain storage cells and links between

thern are user-defined and therefore require sorne a priori understanding of flow

pathways in the floodplain which may result in circular reasoning within models.

The choice of a model type (ID, 2D- or 2D) for surface flow modelling is mostly

relevant in river flooding applications. 2D is the preferred choice in urban and in

coastal environments. 2D modelling of river floodplains can itselfbe divided into

two important classes of approaches, namely the one where only floodplains are

modelled in 2D (as part ofa combined ID/2D model) and the one where floodplain

flow and channel flow are modelled as part of the same 2D grid. The main advantage

of 2D modelling (over any other approach for floodplain modelling) are that local

variations of velo city and water levels and local changes in flow direction can be

represented (SYME, 2006). The approach also does not suffer from the limitations of

the 1D and 2D- approaches detailed in the previous paragraphs. It allows in principle

a better representation of floodplain conveyance, but a major limitation of combined

1D/2D models for river and floodplain systems is that the exchange processes

between the river and the floodplains are still modelled crudely (momentum transfer

is not modelled). A major drawback of 2D models is their computational cost. The

approach where the whole river and floodplain system is represented as part of a 2D

unstructured grid deserves special attention (see for example Horritt and Bates

18
Literature review

(2002)). This approach is not particularly common in UK practice, perhaps because

there is a long-established tradition of 1Driver modelling. Surveyed cross-sections

which are intended primarily for 1D models exist for a large proportion rivers in the

UK. Numerous existing ID models have been calibrated using measured data, and

ID Manning's n values are well-known for many rivers or river types. There is

therefore a c1ear incentive to make use of these data and knowledge by continuing to

build 1Driver models or to use existing ones. In addition, the grid resolution needed

to model a river in 2D is significantly fmer than what is typically applied on

floodplains, resulting in significantly increased computation times. These reasons

explain the current enthusiasm for combined 1D/2D modelling for river and

floodplain systems.(Néelz and Great, 2009)

2.4.4 2 D flood modelling packages

A recent desktop review, commissioned by the Environment Agency' s Science

Department and funded by the joint Environment Agency/Defra Flood and Coastal

Erosion Risk Management Research and Development Programme, has analysed 17

2d modelling packages currently used in UK.(Néelz and Great, 2009)

The vast majority use the shallow water equations and are commercially available.

Fewer (9) provides the possibility to integrate shock capturing. The possibility to

link 1D models exist for 7 of the packages reviewed.

The theory upon which 2D packages are based suggests that predictions using these

alternative approaches will differ where acceleration terms are significant.

19
Literature review

When simulation of inundation extent for dam break flooding is required, the ability

to model super and subcritical flows may become more crucial and shock wave

modelling may dominate the selection of packages.

TUFLOW, InfoWorks, Mike-Flood and JFLOW are presently the most commonly

used in UK but there was not fundamental cause to reject the other ones. Negotiation

of license, maintenance fees and training cost are open to negotiations ; 10 licences

would cost between ;(15,000 and ;(25,000; the annual maintenance costs are

typically between 10 and 20 per cent of licence costs. The recommended duration of

training courses is two to three days at a cost of around f2,000 for ten participants.

The fastest appears to be MIKE FLOD (DHI, 2010). The only one with a free

version in 2D is ISIS 2D (Halcrow, 2010a).

20
,1
.,'
"

0-3
~
0-
;;-
Vol

~
Name Physics Further information on Shock Developer Status Linkages 0
~
numerical scheme ca pturinÇ) ~
0
FINITE DIFFEREN CE SCHEMES 3
3
TU FLOW SWE Alternatinq Direct . 1mplicit No BMT-\J\J8M Commercial 0Nn 10 river and pipes solver 0
DIVAST SWE Alternatinq Direct . Implicit No Cardiff U niv . Re se arch As part of 1SIS 20 =
N
DIVAST-ND SWE Explicit 1YD- MacCormack Yes Cardiff U niv . Re se arch 0
ISIS2D SWE Altematina Direct . Imolicit No Halcrow Commercial Own 10 river solver "0
~
MIKE 21 SWE Alternatinq Direct. Implicit No OHI Commercial As part of M1KE FLOOD ~

Commercial ;J::"
MIKE FLOOD SWE MIKE21 No OHI Own 10 river (MIKE 11) an d urban drainage ~
(M IKE URBAN) solvers ~
~
<IJ
SIPSONlUIM SWE Alternatinq Direct . Implicit No U. of Exeter Research Own multiple linking element
SOBEK SWE Implicit - Staggered grid Yes OELTARES Commercial Own 10 river solver, vertical l ink :;
JFLOW Diffusive wave Explicit No J8A Internai ~
~
FINITE ELEMENT SCHEMES 1

'2
~,

TELEMAC 20 SWE No EDF Commercial ~


N

-
1

FINITE VOLUME SCHEMES ~


N
TELEMAC 20 SWE Tbc Yes EDF Commercial
=
Q.

MIKE 21 FM SWE Godunov based Yes OHI Commercial As part of MIKE FLOOD Ci
""'1
~
MIKE FLOOD SWE MIKE 21 FM Yes OHI Commercial OWn 10 river (MIKE 11) and urban drainage ~

(MIKE URBAN) solvers + MOUSE (7) :t"


N
Inf oWorks- RS SV\IE Roe's Riemann solver Yes Wal'ford Softw Commercial OWn 10 river solver <:>
<:>
InfoWorks-CS
HEMAT
SV\IE
SWE
Roe's Riemann solver
Roe's Riemann solver
Yes
Yes
Wal'ford Softw
Iran Wat. Res.
Commercial
Research
0Nn 10 urban dra inage solver ~ .. .
~
~

Cent. & Cardiff ~

~
BreZo SWE Explicit- R Riemann solver Yes U . of California Research
TRENT SWE Explicit- R Riemann solver Yes Nottinqham U . Research
(l)
OTHERS ""1
(l)
USFLooD-FP Norm. Flow in Explicit No U . of Bristol Research 10 kinematic wave treatment. Verticallink . .....
<:
(l)
x and y dir.
RFSM G ravity only Volume filling algorithm No HR-Wal'ford Internai Linked ta other components of national FRA ~
Flowoute Diffusive wave Ambiental Internai No technical information published.
Grid-2-Grid CEH No technical information published .
Floodflow Microdrainage No technical information published .
Literature review

Table 4 : Classification of inundation packages (Néelz and Great, 2009)

Method Description App licati on Typical Ou:puts Example


computation Models
times
10 Solution of the one- Design scale modell ng Minutes Water depth, cross-section Mike 11
dime1sional8t- which can be of the Jrder averaged velocity, and HEC-RAS
Venant equaticns . of 10sto 100s of km discharge at each cross- ISIS
depending on catchrnent section . InfoWorks
size . Inundation extent if RS
floodplains are part of 10
model , orthrough horiIontal
projection of water level .

10+ 10 plus a storege Design scale modell ng Minutes Asfor 10 models, pluswater Mike 11
cell approach t) the which can be of the )rder levels and inundation Extent HEC-RAS
simu lati on of of 1Os ta lODs of km in foodplain storage cells ISIS
floodJlain flow. depending on catchment 1nfoWorks
size, al sa has th e RS
potential for broa d scal e
application if used V\ith
sparse cross-sectior
data.
20- 20 rrin us the law of 8road scale modelling Hours Inundation extent LlSFLOOD-
conservation o! and applications Vllhere Water depths FP
morrentum forthe inertial effects are not JFLOW
floodJlain flow. important.
20 Sol uti on of th e tvvo- Design scale mode Il ng of Hours or days Inundation e~ent TUFLOW
dime1sional shallow t1e ordercf 10s ofkll. Water depths Mike 21
water equations. May have :he potential Depth-averaged velocijes TELEMAC
f)r use in broad scale SOBEK
modelling if applied "'vith InfoWorks-
very coars~ grids. LD

20+ 20 plus a solution for Predominently coastal Days Inundation extent TELEMAC
vertical velociti::s modelling 3pplications Water d epth s :::0
usin~ continuity only. where 3D '/elocity profiles 3D vel ocitie s
are important. Has also
been appli~d to reach
~l:al~ riv~r rrlULl~lIilly
problems i1 research
projects .
3D Solution of the three- Local predictions of Days Inundation extent CFX
dirne1sional t1 ree-d irnension al Water depths
Reynolds averaged velo city fields in main 3D velocities
Navier Stokes channels end floodp ains.
equa:ions.

2.5 Rainfall-runoffmodel

2.5.1 The Google Scholar metric

Searching Google Scholar about rainfall-runoff modelling faces two possible

spelling for modelling.

The number of hits excluding citation returned by Google Scholar for the three

periods (before 1990, 1991 to 2000, after 2001) is presented in Table 5

22
Literature review

Table 5 : Rainfall-runoff hits - Google Scholar

Nbr Rainfall- Rainfall- Rainfall-


Rainfall- Runoff Runoff Runoff
Year Runoff modelling modeling model
<1990 3,790 2,810 2,810 2,730
1991-2000 5,580 4,950 4,950 4,810
2001-2010 16,400 16,400 16,400 16,300

The present decades shows a 194% increase compare to the previous one (Rainfall-

runoffkeyword), the association with modelling increases from 74% prior to 1990 to

100% posterior to 2000.

2.5.2 The rationales of rainfall-runoff modelling

The main justification for rainfall-runoffmodelling is the limitations ofhydrological

measurements.(Beven, 2004). The available measurements need to be extrapolated

both in time and space particularly for the ungauged catchments (where

measurements are not available) and into the future (where measurements are not

possible) to explore the impact of foreseeable hydrological changes.

A significant proportion of rainfall-runoff models are used to formalize knowledge

about hydrological systems but the ultimate aim of prediction using models must be

to improve the decision-making about a hydrological problem (water resources

planning, flood protection, mitigation of contamination, licensing of abstraction ... )

(Beven, 2004).

2.5.3 The perceptual model

The perceptual model in hydrology undertook a dramatic paradigm shift with the

measurements of the Seine by Pierre Perault in 1668 to 1669 and ms calculations

proving that rainfall was sufficient to account for flow of the Seine. Before him, the

23
Literature review

opinions were that the stream were fed by underground channels whose

contributions out weighted the precipitations.(Linsley, 1967). Nevertheless during

the 17th and 18th centuries, numerous summit reservoirs to supply water to canals

crossing divides were constructed in Europe and England. There must have been

sorne shrewd estirnates of the runoff available in order to assure that these reservoirs

would meet the demand of the canals during protracted dry periods. There is no

evidence in the literature of the techniques used to make these estimates, and one

suspects that there was heavy dependence on the judgment of the planners.(Linsley,

1967). The perceptual model was their main reference. Keith Beven too insists on

the importance of the perceptual model as a summary of one' s perceptions of how

the catchments respond to rainfall under different conditions.(Beven, 2004) A

perceptual model is necessarily personaI. It depends on the training that a hydrologist

has had, what he read, the data and field sites he came in contact with. The

translation of the perceptual model into mathematical descriptions inevitably lead to

simplifications in sorne cases gross simplifications but often preserving sorne

capacity to make quantitative predictions. This process of deciding the equations to

represent the process leads to the conceptual model (see Figure 2 : A schematic

outline of the different steps in the modelling process)

The runoff generation processes involved in the perceptual models are iIIustrated in

Figure 3

24
Literature review

(a) Infiltration excess overland flow

~
~f
(b) Partial area infiltration excess overland flow

~
~f
(c) Saturation excess overland f10w

(d) Subsurface stormflow

(e) Perched subsurface stormflow

Figure 3 : A classification of process mechanisms in the response of hillslopes to rainfalls. (a) Infiltration
ex cess overland flow (Horton, 1933). (b) Partial area infiltration excess overland flow (Betson, 1964). (c)
Saturation excess overland (Cappus, 1960, Dunne and Black, 1970). (d) Subsurface stormflow (Hursh,
1936, Hewlett, 1961). (e) Perched saturation and throughflow (Weyman, 1970).

2.5.4 The conceptual model

At this point, the hypotheses and assumptions being made to simplify the

descriptions of the processes need to be made explicit. Keith Beven notes that in

25
Literature review

many articles and model users manuals, the goveming equations are presented

prominently but the underlying sirnplifying assumptions are not (Beven, 2004). This

however should be the starting point for the evaluation of a particular rnodel relative

to the perceptual model in mind.

2.5.5 The procedural model

The equations driving the conceptual model rnay not be resolved analytically within

the boundary conditions of the real system. This is often the case for partial

differential equations. A replacement of these equations by finite-difference or fmite-

volume equivalents is frequent. Sorne error/approxirnations are often introduced at

this points and great care should be paid to understand the impact of these

approximations.

2.5.6 The calibration

A variety of parameters are often involved relating soil properties and dimensions of

the catchments with the water discharge. Sorne are strongly time dependant, sorne

are constant during a given simulation. Some parameters like the ones related to

catchments size, drainage length and slope are easily available through GIS

techniques. The parameters relating to soil properties are often known at a very

different scale than the one being rnodelled (very limited soil samples representing

the whole catchments with very broad limit of confidence). The optimisation of the
r_ o- _ - _ . . . : - -

parameters is often done by various techniques of fitting a set of known outputs with

the available inputs in the rnodel. ,(Beven, 1993, Clarke, 1973, SCHUMANN, 1993,

Moradkhani and Sorooshian, 2008, Franks et al., 1997, Beven and Binley, 1992,

Beven, 2004, Beven, 2009).

26
Literature review

Parameters values determined by calibration are effectively valid only inside the

model structure used in calibration. Transferring these values to other model

structure or other catchments may generate serious errors.(Beven, 1993, Beven,

2006)

The concept of an optimum parameter set may be ill-founded in hydrological

modelling. The equifinality (cfr 2.2Environmental modelling equifmality) is most

likely inevitable

2.5.7 Validation jperformance evaluation

The validation process as understood by "classic" modellers consist at using a data

set with know results that was not used in the calibration (independent data) and

quantify the divergence between the modelled and observed output. Jens Christian

Refsgaard and Jesper Knudsen (Refsgaard and Knudsen, 1996) propose a testing

frame in 4 steps :

a) The split-sample test (SS) involves calibration of a model based on 3-5 years of

data and validation on another period of a appropriate length.

b) The differential split-sample test (DSS) involves calibration of a model based on

data before catchment change occurs, adjustment of model parameters to

characterize the change, and validation on the subsequent period.

j •• ".--:.. .. _ ...-..,.
c) In the proxy-basin test (PB) no direct calibration is allowed, but advantage may

be taken of information from other gauged catchments. Rence validation will

comprise identification of a gauged catchment deemed to be of a nature similar

to that of the validation catchment; initial calibration; transfer of model,

27
Literature review

including adjustment of parameters to reflect actual conditions within validation

catchment.

d) With the proxy-basin differential split-sample test (PBDSS), again no direct

calibration is allowed, but information from other catchments may be used.

Renee validation will comprise initial calibration on the other relevant

catchment, transfer of model to validation catchment, selection of two parameter

sets to represent the periods before and after the change, and subsequent

validations on both periods.

But these validation test are based on the idea of optimal model that is more or less

approached with the actual system modelled.(Beven, 2004, Beven, 2006, Anderson

et al., 2001)

The adoption of the equifinality proposition requires a much broader measure of

likelihood and uncertainty. The Generalised Likelihood Uncertainty Estimation

(GLUE) (Beven and Freer, 2001) achieves this objective by running many different

model runs with randomly choosen parameters sets. Each run is evaluated against

observed data by means of a likelihood measure. If a model is rejected it is given a

likelihood value of zero. The likelihood measure are then used to weight the

predictions of the retained models to calculate uncertainty estimates or prediction

limits for the simulation.

2.5.8 Rainfall-runoff model classification

There are many different ways of classifying hydrological models (see (Clarke,

1973, O'Connell, 1991, Singh, 1995). Rere after is a very basic classification adapted

from Beven (2004). The frrst taxon concems the spatial extension of the calculation

28
Literature review

unit: lumped or distributed modelling approach. Lumped models treat the catchment

as a single unit, with state variables that represent averages over the catchment area.

Distributed models make predictions that are distributed in space, with state

variables that represent local averages of storage, flow depths or hydraulic potential,

by discretizing the catchment into a large number of elements or grid squares and solving

the equations for the state variables associated with every element grid square.

Parameter values must also be specified for every element in a distributed model.

There is a general correspondence between lumped models and the 'explicit soil

moisture accounting' (ESMA) models ofO'Connell (1991), and betweendistributed

models and 'physically based' or process-based models. Even this correspondence is not

exact, however, since some distributed models use ESMA components 10 represent

different subcatchments or parts of the landscape as hydrologïcal response units while

even the most distributed models currently available must use average variables and

parameters at grid or element scales greater than the scare of variation of the processes.

They are consequently, in a sense, lumped conceptual models at the element scale

(Beven, 1989). There is also a range of models that do not make calculations for

every point in the catchment but for a distribution function of characteristics,

TOPMODEL, is a model of this type, but has the feature that the predictions can be

mapped back into space for comparison with any observations of the hydrological

response of the catchment. It could therefore be called, perhaps, a semi-distributed

model.(Beven, 2004)

A second consideration is whether to use a deterministic or stochastic model. Deter-

ministic models permit only one outcome from a simulation with one set of inputs and

parameter values. Stochastic models allow for some randomness or uncertainty in the

29
Literature review

possible outcomes due to uncertainty in input variables, boundary conditions or model

parameters. The vast majority of models used in rainfall-runoff modelling are used

in a detenninistic way, although again the distinction is not clear-cut since there are

examples of models which add a stochastic error model to the deterministic predic-

tions of the hydrological model and there are models that use a probability distribution

function of state variables but make predictions in a deterministic way. A working

rule is that if the model output variables are associated with some variance or other

measure of predictive dispersion the model çan he considered stochastic; if the output

values are single valued at any time step the mode! can he considered deterministic,

regardless of the nature of the underlying calculations.

There is one other modelling strategy based on :fuzzy methods that looks highly

promising for the future. In particular, fuzzy models would appear to offer the

potential for a more direct translation from the complexity.ofthe perceptual model

into a procedural model. Lohani and al. have recently compared rainfall-runoff

predictions based on Artificial Neural Network, fuzzy logic and linear transfer

functions on the upper Narmada basin (India). In their study, fuzzy models

outperformed the two other approaches.(Lohani et al., 2010).

2.5.9 ReFH package

2.5.9.1 Introduction

ReFH is a rainfall-runoffmodelling package for estimating design flood hydrographs

and analysis of observed events. It also includes reservoir routing methods for

estimating the attenuation of flood hydrographs by reservoirs. It is developed by

WHS HydroSolutions Ltd. The rationales for its use in the present study are

developed in para. 3.8 The choice of rainfall-runoff modelling package.


30
Literature review

The Revitalised Flood Hydrograph (ReFH) model is the new FEH rainfaII-runoff

method for UK design flood estimation. The methods used is based on the FEH

Supplementary Report No. 1 - The revitalised FSRlFEH rainfaII-runoffmethod.

The method was developed by The Centre for Ecology & Hydrology (CEH ) as a

direct replacement for the Flood Studies Report rainfall -runoff method, and the

software replaces the Micro-FSR software. This new lumped conceptual rainfall-

runoff model was calibrated using recent and large flood events and has added

emphasis on seasonal flooding, with the same underlying principles as the previous

method. It is an event based model.

The software is capable to :

1. Generate design flood hydrographs

• use a design rainfaII event of the required return period

• specify both soil moi sture and rainfall on a seasonal basis depending on the

degree of urbanisation within the catchment.

2. Analyse observed events

• visually inspect quality of the data for each event

• estimate model parameters using FEH CD-ROM catchment descriptors

• optimise model parameters through event analysis

o specify sub-daily event rainfall and runoff

o model antecedent soil moisture conditions using daily rainfaII and

evaporation

ouse regression fitting to select baseflow parameters on the hydrograph

for each event

31
Literature review

o optimise time-to-peak and maximum soil moisture content

o optionally optimise additional parameters

• use these parameters to generate design flood hydrographs at the site

3. Estimate the attenuation of flood hydrographs by reservoirs through reservoir

routing methods

4. Manage and view observed event data and modelled results

In addition to the software, the ReFH design model has been implemented as a freely

available spreadsheet version.

Note that the following sections are partly reproduced from "The revitalised

FSRlFEH rainfall-runoff method" (Kjeldsen, 2007)

2.5.9.2 Modelling concepts

Rainfall-runoffmodelling for design flood estimation has conventionally been based

on the modelling of individual events. At the most rudimentary level ail that is

required to reproduce the catchment-scale relationship between storm rainfall and the

corresponding stream flow response is :

• a volumetrie loss to account for hydrological processes such as evaporation,

soil moi sture storage, groundwater recharge and interception losses; and

.::,r... :,-•• -
• a time distribution model to represent the various dynamic modes of

catchment response.

However, the specification of the model developed to represent the rainfall-runoff

relationship is very much related to scale, both spatial and temporal. For instance, a

model relating the annual rainfall and runoff for a small homogeneous catchment

32
Literature review

may be very simple, while the relationship between hourly rainfall and runoff on a

large, heterogeneous catchment may be extremely complexe The Revitalised Flood

Hydrograph (ReFH) model has been developed for use in the revitalised FSRlFEH

rainfaIl-runoff method as a parameter-sparse hydrological model, representing the

major rainfall-runoff processes on a catchment scale.

In the following it is important to distinguish whether the ReFH model is used for

modelling an observed flood event or for generating a design flood event. When

modelling an observed event, the ReFH model is used as a deterministic mode)

trying to reproduce a flood event from historical series of observed rainfall and soil

moisture data. In contrast, a design event is a probabilistic estimate of a flood event

that will be exceeded on average once every T years, where T is the return period

(e.g. T= 100 years). When generating a design event, the input values of rainfall and

antecedent soi! moisture do not represent a particular historie event but are

generalised vaIues specified so that certain combinations will result in a flood event

of the required return periode

The ReFH model consists of three main components: a loss model converting total

rainfaIl into effective rainfall, a routing model and a baseflow model. The

connections between the three model components are shown in Figure 4 : ReFH

model concept diagram, together with the required input variables and model

parameters. In addition to the three main components, a soi! moisture accounting

model based on daily data is used to determine the state of the soil at the start of the

flood event based on long-term series of antecedent rainfall.

When simulating a flood event, the 10ss model is used to estimate the fraction of the

total rainfall volume turned into direct runoff. The direct runoff is then routed to the

33
Literature review

catchment outlet using the unit hydrograph convolution in the routing model and,

finally, the baseflow is added to the direct runoffto obtain total runoff. Each of the

three components, is explained further in the following sections.

Total ra infall

~ <~-~ Net rainfall


, ,

-..
--i--'"
,..-_....L.-_----,
•• -..~ ! .... ', .

Initial soli moisture ..... Loss Routing


Cini ......~ model model
{Cmax } (Tp )
~ Total f10w

Initial baseflow __ 1 Baseflow model 1


BFo _ _ _ _{_BR_,_B_L}_ _ ---'I 1

Figure 4 : ReFH model concept diagram

The loss model in ReFH is based on the Probability Distributed Model (PDM)

developed by Moore (1985) and widely used for a variety ofhydrological

applications in the UK. The PDM model is being used in a framework for a national

system for flood frequency estimation using continuous simulation modelling in the

UK (Lamb, 1999; Calver et al., 2005). Furthermore, the model has heen used in real-

time flood forecasting (Moore, 1999) and it has been used to investigate the impact

of climate change on runoff from small catchments in the UK (Prudhomme et al.,

2003).

Conceptually, the PDM assumes the catchment to consist of a number of individual

storage elements, each of a random soil moisture capacity C arising from a statistical

distribution. Assuming a uniform distribution of soil moisture capacities, if the

34
Literature review

storage elements are arranged in order from the highest (Cmax ) down to zero

capacity, the resulting PDM distribution of soil moisture capacity is shown in Figure

5 : Cumulative distribution of soil moi sture capacity . It is further assumed that the

storage elements interact such that the soil moisture is redistributed between stores

between rainfall events. Thus, at any time soil moisture is constant for all elements

of capacity greater than Ct and is at full capacity for elements of capacity smaller

than Ct as illustrated by the dotted area in Figure 5 : Cumulative distribution of soil

moisture capacity. During a storm, the depth ofwater in each storage element is

increased by rainfall (horinzontalline are a) and when rainfall exceeds the storage

capacity, direct runoff is generated. For the short duration of the storms under

consideration, the effects of evaporation and drainage out of the soils have not been

inc1uded.

EXeèSS

............. o
'-"'-~
"'-.. . . . .
~
--~.
.......
~.......
{I

..
............ ..".
~
.......

_.......
..--..
-
So~1 m(l~ s.1u re

-~. capaclty, C t
. ........
•• •
-
• •
C· ...--... ....... -..,..,.
• f ••
-_
-li---_.....
~
~

~,~ • •
~.. .
1 • • : - - :..

o GlJrnul a ~v.~ d~l~buUon of so~1 mo~stura capa'~ty

Figure 5 : Cumulative distribution of soil moisture capacity

Thus, a pulse of rain, Pt, on the soil gives 100% runoff from the area already at full

capacity and increases the moisture content in an other areas. The excess amount of

rainfall converted into direct runoff, qt, can be estimated through simple geometric

considerations as

35
Literature review

Equation 6

where the continuity equation Ct+M = Ct + Pt applies and Cilli is the soil moisture

content at the start of an event. The ratio q/P of rainfall transformed into direct

runoff is a measure of the percentage runoff, and Cmax is the only model parameter.

Once Ct exceeds Cmax, the model assumes that 100% of the rainfall is converted into

runoff. The loss model can be applied sequentially, where a loss is ca1culated

for each time step, or it can be applied to calculate a single loss of total rainfall

volume. In the revitalised FSRlFEH rainfall-runoff method, the former option has

been adopted, Le. a loss is calculated for each individual time step. As the soil

becomes increasingly wet during the storm, the loss decreases and the runoff rate

increases. The initial soil moi sture content (Cilli) is an important parameter when

applying the loss model, either for analysing an observed event or for simulating a

design flood event.

For the routing model, ReFH uses the unit hydrograph (UH) concept for routing the

net rainfall to the catchment outlet (direct runoff). A UH can be estimated directly

for each flood event through simultaneous analysis of the effective rainfall

hyetograph and the direct runoffhydrograph as described by Chow et al. (1988), for

example. The original FSRlFEH model adopted a standard triangular-shaped

instantaneous unit hydrograph (lUH) scaled to each catchment using the time-to-

peak (Tp) parameter, catchment area and the selected time step. The ReFH model

retains the concept of a standard IUH shape scaled to individual catchments, but

introduces a more flexible shape as shown in Figure 6 in the form of a kinked

triangle.

36
Literature review

UI it hyalrogra'ph

Uk ;;:; Unp,l IJ~

U,,;;;; 1 -'>Triangular IUH

Uk ;;:; D~ 1nfllrli-te T/3

l ime units
.:_______2_Tp_ _ - r~ ____•______-.~
.~------------ TB -----------------.~

Figure 6 : Shape of standard instantaneous unit hydr?graph adopted in ReFH

The kinked triangle is described by a time scaling parameter, Tp , and two

dimensionless parameters, Up and Uk , controlling the height and kink., respectively,

of the IUH. The parameter Uk is a multiplier applied to the ordinate Uc of a non-

kinked triangular IUH at 2Tp , i.e.

Equation 7

where TBt= 2 Tp /[h to ensure unit-area under the non-kinked triangular unit

hydrograph, illustrated by the broken line in Figure 6. Thus if Uk= 1 the IUH is a

simple triangle, but as Uk drops towards zero, the 'lost area' is transferred into the

IUH tail by extending the overall time base TB.

Equation 8

Attempts to relate the parameters controlling the height (Up ) and kink (Uk) of the

IUH to readily available catchment descriptors were unsuccessful. Instead, average

values of Up = 0.65 and Uk= 0.8 are recommended for use in the revitalised FSRlFEH

rainfall-runoffmethod. Because of the kink introduced at 2Tp the standard ReFH

IUH has a lower peak and a longer time base than the FSR IUH. To convert the

37
Literature review

dimensionless IUH in Figure 6 into the required units of m3 s-1 mm-1, a scaling factor

of AREA 1 (3.6 Tp) is applied, where AREA is in km2 and Tp is in hours. The IUH can

only be used directly when rainfall is given as a continuous function of time. If

rainfall is given as a sequence of depths in successive time steps I1t, the IUH must

first be converted to an equivalent I1t-hour UH. To transform the IUH in Figure 6into

a unit hydrograph of any given time step I1t, the ReFH model uses the S-curve

method as described in many standard hydrology textbooks such as Chow et al.

(1988). The S-curve method replaces the existing FEH approximation of adding half

the time step to the time-to- peak of the IUH. This approximation only works if I1t is

a small component of Tp and the unit hydrograph is not too skewed.

The baseflow mode} implemented in the ReFH model is based on the linear

reservoir concept, with a characteristic recession defined as an exponential decay.

The approach, discussed by Appleby (1974) allows the· separation of total flow in

baseflow and surface flow without knowing the rainfall input. It is based on the

contributing area concept, and assumes that the saturated area of the catchment that

produces surface runoff is the same area that also produces baseflow recharge, and

furthermore that the ratio of recharge to runoff, BR, is fixed. An unsaturated area

produces neither runoff nor recharge because rainfall is retained as soil moisture.

Rainfall that becomes recharge is assumed to pass through a linear storage (with a

lag value of BL ) before emerging into the same channel system that carries the

surface runoff. The baseflow hydrograph at the catchment outIet can be determined

by routing BR times the (as yet undefmed) surface flow at the outlet through the

groundwater store. The observed hydrograph at the outlet is then the sum of the

surface and baseflow hydrographs.


38
Literature review

The resulting baseflow model calculates the baseflow at successive time steps ~t

apart by linking the baseflow to the observed runoff and the estimated baseflow from

the previous time step as

where Qtis total observed flow at time t. For the case where the baseflow model is

being used to analyse an observed flood event, the constants kt, k2 and k3 are given

as :

k
1
= BR ( BL (1-k 3 )
k3 ) Equation 9
(l+BR) tH l+BR

k - BR (1_ BL (1-k 3 ))
Equation 10
2 - (1+BR) t1t 1+BR

k3 = exp ( - ;: (1 + BR) ) Equation Il

2.6 Satellite rainfall


...
~ '.

2.6.1 The rain gauges deficiencies

People have been measuring rainfall amounts for more than 2000 years, but a lot of

uncertainty remains regarding how much rain falls in remote areas of the globe.

(NASA,2010d).

Before TRMM's launch, the measurements of the global distribution of rainfall at the

Earth's surface had uncertainties of the order of50%.(NASA, 2010b)

39
Literature review

Traditionally rainfall has been monitored by open networks of raingauges at point

locations. However, such a sampling approach is ill-suited to a parameter which

seems to affect one percent or less of the surface of the Earth at any time (Barrett &

Martin, 1981), and which "typically occurs at only a very small fraction of the time

at any given location" (Theon, 1992). Furthermore, instantaneous precipitation

intensities range from 0 to > 125 mmJh, and characteristically vary over distances as

small as tens of metres, and over periods of minutes or even seconds, by orders of

magnitude of intensity. Meanwhile, because the gradients of rainfall intensity may

instead be very shaIlow, it is intrinsically difficult to establish the rain: no-rain

boundary with precision. It is also widely recognized that natural problems of

rainfall measurement by gauges have been exacerbated by problems which have

arisen in association with the gauges themselves: many different gauge types are in

everyday usage, the time intervals investigated by families of gauges differ, the

raingauges themselves affect the parameter they are designed to measure, and

deficiencies in rainfall station and data record management aIl adversely affect

the timeliness, availability and quality of rain gauge data.(Barrett, 1997)

The main limitations of recording gauges, due to the short intervals between

measurements, are that small rainfall totals may not be accurately recorded, some

rainfall may be missed during the emptying of storage vessels, and the delayed

delivery of water to the measuring container can introduce temporal errors (Linsley

et al., 1949, Sumner, 1988, Mc Anelly and Cotton, 1989, Robinson and Ward, 1990,

Horsfield, 2006, Beven et al., 2010) .

40
Literature review

Measurements from gauges are also subject to systematic errors due to shading and

sheltering, drops splashing in or out, condensation or evaporation of moisture, and

the occurrence of solid precipitation. Wind turbulence, initiated by local topography

and features but amplified if the gauge rim is above ground level, increases the wind

speed above the opening and carries raindrops downwind, causing rainfall totals to

be underestimated. The impact of this can be minimised by mounting the top of the

gauge level with the ground surface, building a turf wall around it, or placing it in a

pit surrounded by a mesh grid (Arlon and Meisner, 1987, Sumner, 1988, Horsfield,

2006, Beven et al., 2010).

Gauges provide good measurements of rainfall at a single point, but rainfall is highly

variable spatially and it is necessary to interpolate these values for most applications

in order to produce an estimate over an area. The accuracy of areal estimates is

dependent on gauge density and distribution, and on the ability of the available data

to represent patterns over a larger area; sparse point values may not effectively

monitor areas that experience pronounced spatial gradients, particularly in

convective regimes(Hildebrand et al., 1979, Arkin and Meisner, 1987, Sumner,

1988, Horsfield, 2006, Barrett and Beaumont, 1994). Point values are interpolated to

provide areal estimates for unsampled locations, but this can introduce smoothing,

which misses sorne fine-scale features (Hildebrand et al., 1979), and only perfectly

represents the actual situation where rainfall demonstrates spatial homogeneity (Oh

et al., 2002, Horsfield, 2006).

For the land area divided in grids of2.5° * 2.5 0


(= 3378 cells), 42.8 % have no

gauge, 41 % have only one, 10% have two to five gauges, 2.2% have ten and

more.(Bellerby, 2010). The density ofrain gauges is illustrated in Figure 7

41
Literature review

Reliability of recording is often compromised by the inadequate staffing and

protection of the gauges.(Gyawali, 2001).

Figure 7 : GHCN-Monthly Coverage Map for Precipitation (Burroughs, 2008).

2.6.2 The satellite rainfall concepts and instruments

Historically, satellite meteorology may be said to have been launched in April 1960

through the first Television and Infrared Observation Satellite (TIROS-l ).(Barrett,

1997). The foundations of satellite rainfall monitoring were laid by Lethbridge

(1967) who considered general relationships between cloud brightness and coldness

and rainfall, and by Barrett (1970) who prepared monthly rainfall maps for part of

the tropical Far East based on satellite cloud charts (nephanalyses), using climate

station data for calibration (Barrett and Martin, 1981). Since then ever-increasing

efforts have been made to exploit satellite visible, infrared and microwave data for

such purposes (Barrett and Beaumont, 1994). Visible and infrared data provide

information on cloud top characteristics which may be indicative of areas and rates

of rain falling from the bases of the clouds, but passive microwave radiation is

42
Literature review

scattered by, or emitted from the hydrometeors themselves embedded in the clouds.

A fundamental problem with satellite-based techniques relates to the frequencies of

the radiation data, these being characteristically rather low spatially (at best about 0.5

km in the visible and infrared, and as low as about 50 km in the passive microwave)

and infrequent temporally (ranging from every half an hour, or less, from

geostationary satellites to as low as once per day from polar orbiting satellite

systems).

Note that the following paragraphs are adapted from "Satellite rainfall climatology: a

review". (Kidd, 2001).

Perhaps the most important contribution that satellite observations can make to

climatological rainfall estimates is that their observations can be made quasi-

globally, a feature that was evident since the launch of the first meteorological

satellite.

Since the launch ofTIROS-l, polar-orbiting (or Low Earth Orbit) satellite sensors

have been a mainstay of global meteorological satellite observations. The current

polar-orbiting satellite series of the National Oceanographic and Atmospheric

Administration (NOAA) are direct descendants of the TIROS-1 satellite and are

characterized by dual satellite coverage of the globe twice daily at a basic resolution

of 1 km. These satellites constantly circle the Earth in an aImost north-south orbit,

passing close to both poles. The sun-synchronous nature of the orbit allows one

satellite to cross the Equator at 07 :30 h local time, with the other at 13 :40 h local

time: two of these satellites ensure that data for any region of the Earth are no more

than 6 hours old. The current NOAA series of satellites started in 1979 with the

43
Literature review

launch ofNOAA-6, with the newest satellite being NOAA-N launched on May 20,

2005(NOAA, 2010). Additional polar-orbiting visible/infrared (VislIR)

meteorological satellites are provided by China and Russia . (See

The primary instrument aboard the NOAA series of satellites is the Advanced Very

High Resolution Radiometer (AVHRR). This instrument senses radiation in the

visible and IR parts of the spectrum with a spatial resolution of 1 km, resulting in

many thousands of global measurements being collected daily. Complementing the

AVHRR : the Microwave Sounding Unit (MSU) and Stratospheric Sounding Unit

(SSU), and, more recently, the Advanced Microwave Sounding Dnit (AMSU), can

be used to help improve the satellite rainfall estimates. In addition, the use of the

High Resolution InfraRed Sounder (HIRS) can be used to detect the transparency of

clouds and, hence, their rainfall potential. However, the main limitation of the polar

orbiting satellites is the poor temporal sampling (about every 6 h) being insufficient

for sorne rneteorological applications. Geostationary meteorological satellites started

in 1974 with the Synchronous Meteorological Satellites (SMS). The geostationary

weather satellites are placed around the globe to provide continuous, repeat coverage

of the Earth. From an altitude of 35,400 km, each satellite can provide data from

about one third of the Earth' s surface, but due to image degradation towards the

edges it is necessary to have at least five geostationary satellites. Imagery can be

captured as frequently as every 30 seconds from the CUITent Geostationary Earth

Orbiting Satellite (GEOS) satellites, but more typically every 30 minutes. The

international cooperation between the participating nations is designed to ensure that

a minimum of 3-hourly data is provided, free of charge, to any other nation.

Although the resolution of the geostationary satellites is coarser than that of the

44
Literature review

polar-orbiting satellites (4 km in the IR at the sub-point), their ability for continuous

monitoring of the atmosphere is crucial for short-term monitoring and forecasting of

weather systems.

Whilst the sensors described above operate in the visible and IR regions of the

spectmm, another set of sensors have been developed for detecting microwave

radiation. Early experiments with passive microwave radiometry started in 1972 with

the Iaunch of the polar-orbiting Electrically Scanning Microwave Radiometer

(ESMR-5) operating at 19 GHz. This was followed by ESMR-6 in 1975, operating at

37 GHz. Results from these two instruments illustrated the benefits to be gained

from passive microwave instruments. The Scanning Multichannei Microwave

Radiometer (SMMR) was launched in 1978: this instrument collected data at five,

duai-polarized frequencies (6.6, 10.69, 18.0,21.0 and 37.0 GHz). Since 1987 the

US' s Defense MeteoroIogicai Satellite Program (DMSP) series of satellites have

carried the Special Sensor Microwave/lmager (SSMIl) providing dual polarized

observations at three frequencies (19.35, 37.0 and 85.5 GHz) and a vertically

polarized channel (22.235 GHz) with the resolution at the highest frequency being

13x15 km.

45
Literature review

Table 6 : Meteorological satellites

Low-Earth orbits Country of origin Sen or Co erage

NOAA USA Vis/IR 90 oN- 90oS


Meteor Rus ia Vis/IR 90 oN- 900S
FY-I China Vi /IR 90 oN- 900S
TRMM USA/Japan Vis/IR/MW /radar 40 oN-400S
DMSP USA Vi /IR/MW 89°N- 89°S

Geo tationary orbits Country of origin Sensors Sub-satellite point

....--.- ~--
Meteosat-7 Europe Vis/IR 0°
Meteosat-5 Europe Vi /IR 63°E
GOMS- I Russia Vis/IR 76°E
INSAT- ID lndia Vis/IR 74°E
INSAT-2E India Vis/IR 83°E
FY-2 China Vis/IR 105°E
GMS-5 Japan V~ /IR 1400E
GOES-IO USA V~ /IR 135°W
:r': • .- >
GOES-8 USA Vis/IR 75°\V

Despite the host of Vis/IR techniques developed, many showing great promise

especially over land, aIl have their limitations. The major drawback is that aIl the

techniques have to infer rainfall from the cloud top temperature or brightness. There

is sometimes little difference between the visible and IR signatures of precipitation

and non-precipitating clouds, often resulting in misclassification of rain areas.(Kidd,

2001).

In the meantime, it was established as early as the mid-1970s that the passive

microwave imagery could provide useful evaluation of rainfall over oceans, at least

on a climatological time and space scales (Rao and Abbott, 1976). The frequencies

0.81 cm (37 GHz), 1.43 cm (19.35 GHz) and 1.66 cm (18 GHz) were particularly

useful for delineating atmospheric liquid water content, precipitatable water and

rainfall intensity(Alishouse, 1983). The ability of the microwave to sense these

parameters is due to the capacity of the microwaves to penetrate clouds, precipitation

sized particles being the major source of attenuation at passive microwave

frequencies. The brightness temperatures measured by the satellites are dependent

46
Literature review

upon the radiation emerging from the Earth's surface and from the intervening

atmosphere between the Earth and the satellite (Allison et al., 1974). Early works by

(Adler et al., 1991, Weinman and Wilheit, 1981, Bellerby et al., 1998) all confirmed

that passive microwave radiometry had great potential for surface rain rate

estimation.

Despite the ability of satellites to monitor precipitation effectively, it must be noted

that absolute values are still uncertain: over sea areas there is a lack of adequate

validation data, surfaces in colder climatic iregions pose significant difficulties,

particularly to passive microwave rainfall (or precipitation) estimates, and the

frequency of observations from satellites leads to temporal samp-ling errors.

The various intercomparison exercises have all indicated that there are variations in

the amount and distributions of rainfall that individual algorithms retrieve. Assessing

which algorithm is 'best' is made more difficult by the Jact that, over regions where

greatest differences between the satellite products occur, generally are regions of

little of no surface validation data. Also, in common with existing disadvantages of

surface measurements, sorne error in the intercomparisons is due to the surface data

not being ofhigh enough quality for a direct comparison with satellite data.(Kidd et

al., 2003, Bellerby and Sun, 2005, Feidas, 2010)

2.6.3 The Tropical Rainfall Monitoring Mission

2.6.3.1 Concepts and instruments

The Tropical Rainfall Measuring Mission (TRMM) was launched in 1997,

augmenting the passive microwave capabilities . The TRMM satellite is in a low

Earth, non sun-synchronous orbit and carries sensors designed to enhance our

47
Literature review

understanding of rainfall. The TRMM Microwave Imager (TMI) was developed

from the SSMIl instrument, with the addition of a 10.7 GHz channel, with a best

resolution of 4 km at 85.5 GHz. This, together with the AVHRR derived Visible and

InfraRed Scanner (VIRS) constitutes the passive rainfall sensors(see Figure 8).

(Huffman et al., 2007, NASA, 2010c). The most significant instrument ofthis

satellite is the active microwave Precipitation Radar (PR) capable of providing

information on the horizontal and vertical distribution of rainfall at resolutions of 4

km and 250 m, respectively. It is the combination of these instruments that makes

the satellite quite unique. The co-Iocated, co-temporal sampling of the sensors

enables a better understanding of the rainfall processes to be achieved, thus leading

to better retrieval algorithms and rainfall estimates.(Kidd, 2001).

48
Literature review

Figure 8 : TRMM sensors chart

49
Literature review

2.6.3.2 TRMM data products

TRMM observed data are processed by NASA and NASDA, and distributed to

users. The definition of the TRMM products is presented in Table 7

Table 7: TRMM data level (NASDA, 2001)

Levet Definition
o Un essed instrument drt~ time ocdered, uali checked, no reœndanc .
1

The different products are presented in Table 8

Table 8 : TRMM Data products (NASDA, 2001)

Estimated Do
Sensor Processing Level Proœd Seme Unit"} Volume"'2
(Compressed)
149MB
PR IB2l Calibmled Received Power 1 orbit (16/day)
(6~70 MB)
149MB
IC2l Radar Refle ct ivity î orbit (16lday)
(4()-"SO MS)
Normalized R..J .... Swface 10MB
2A21 1 orbit (l61day)
Cross Section (s'J) (6-7 MB)
13MB
2A23 PR Qualihtive 1 orbit (16/day)
(6---7 MB)
241 :MS
2A25 Rain Profile 1 orbit (16/day)
(13-17 MSl
Monthly Statistics of Global Map (Monthly) 40MB
3A25
Rain Parameter (Grid: 5° x 5°, O.Y x O.S·) (26--27 MB)
Monthly Rain Rate Global Map (Monthly) 9.3 MB
3A26
8smg llStaJ:isticaJ Aletkod (Orid: 5° x 5°) \.~.. a..tf.LM9;'
14:MB
TM! IBll Brightness TemperZire 1 orbit (16/day)
(14 MB)
97:MB
2A12 Rain Profile 1 orbit (16/day)
(6.7-9 ME)
Global Map (Montbly) 53KB
3A11 Monthly Oceanic Ramfall
(Grid: 5" x S") (44 KB)
.r.,_-_':~-::.

92MB
Vm.S IBOl Radililce 1 orbit (16/day)
(90MB)
151 MB
COMB 2B3l Rain Profile 1 orbit (16/day)
(8MB)
Global Map (Monthly) 442KB
3B3l Monthly Rainfall
(Grid: 3° x 5°) (380.-410 KB)
TRMM &IR Global Map (S/day) 242KB
3B42
DailyRainfall Grid : 0.25 * 0.25 (1l~115 KBl
TRMM & Other Sources Global Map (Monthly) 242KB
3B43
Monthly Rainfall (Qid: 1° x 1") (242 KB)

50
Literature review

The data flow is presented in Table 9

Table 9 : TRMM data flow (NASDA, 2001)

LelJel 13

Le'/el2

1 ~IJ~I .~

,
PR
.
Fllonthly R:ain Rate
usln Il a Stalls.cal MetnoCi

Lo'/ol3
(Comhin::li P'OcllJct Iisino IR InfraRed
C PI G IDb:a1 P...eipitaüo n Ind IJ( .
-RM rv1 g Ot hY dGtG) SSMA SlleClalsensor
Micrawlve/lma ger
CAM6 Cllm:all! Assessment and
Monitoring SY!ium
G PCC G IDb III P recipi1 ëlton
Clmltology Center

2.6.3.3 3b42 rainfall product

The 3b42 rainfall estimates are sometimes referred as TMPA for TRMM

Multisatellite Precipitation Analysis but this is misleading as the TMPA includes the

3b43 product which is a monthlys estimate while the 3b42 is a 3 hourlyestimate.

51
Literature review

The 3b42 (Huffman et al., 2007) is a near real-time precipitation rate product at fine

time and space scales (3-hr, 0.25 degree X 0.25 degree latitude-longitude) over the

latitude band 50 degree N-S. This product makes use ofTRMM's highest quality

observations, along with high quality passive microwave-based rain estimates from

3-7 polar-orbiting satellites (e.g. AMSR, (2) SSMIIDSMP, (2) AMSUIPOES), and

all the geosynchronous IR sensors (Meteosat, GOES, GMS). The combined quasi-

global rain map at 3-hr resolution is produced by using TRMM to calibrate, or

adjust, the estimates from aIl the other satellites, and then combining aU the

estimates into the TMPA final products. The technique uses as much microwave

data as possible, and uses the geo-IR estimates to fill in gaps in the three-hour

analysis. The calibrations are computed using monthly accumulations of matched

data to ensure stability.

The primary merged microwave- infrared product is computed at the 3-hourly, 0.25 0

*0.25° latitude-longitude resolution. In common with the opep products, the

TMPA is designed to combine precipitation estimates from various satellite systems,

as weIl as land surface precipitation gauge analyses when possible, with the goal that

the final product will have a calibration traceable back to the single "best" satellite

estimate. In the present implementation, the calibration is based on TRMM

estimates. The TMPA is computed twice as part of the routine processing for

TRMM, first as an experimental best-effort real-time monitoring product about 9 h

after real time, and then as a post-real-time research-quality product about 10- 15

days after the end of each month. (For brevity, these will be referred to as the RT and

research products, respectively.) The fust bas been posted to the Web since February

52
Literature review

2002, while the second is available from January 1998, for a record that totals more

than 10 yr and continues to grow.

For the 3B42RT product (Huffman et al., 2009), files are produced every 3 hours on

synoptic observation hours (00 UTC, 03 UTC, ... , 21 UTC), these rainfall rates

correspond to the 3 hours centred on the nominal hour (ie: from 1:30 to 4:30 for the

03 UTC). The Version 6 TRMM product 3B42 is being computed with the TMPA

after real time, and constitutes the research-grade archive of TMPA estimates. The

version numbering for the TMPA-RT and official TRMM products are not related,

although both are currently numbered 6.

2.6.4 3b42 raiDfall rUDoff modelliDg

2.6.4.1 Google Scholar metric

The number of hits exc1uding citations returned by Google. Scholar for the various

combination of keyword with 3b42 is presented in Table 10

Table 10 : Google Scholar hits for 3b42 & hydrological models, flood, runoff

3b42 - 3b42
hydrological 3b42 3b42 flood
Year model runoff flood warning
<1990 0 1 2 0
1991-2000 2 0 1 0
2001-2005 19 5 11 5
2006 19 6 5 2
. .; ...
.,:..:-.,:- 2007 32 9 16 3
2008 48 21 26 10
2009 71 32 49 12
2010 35 22 30 8

53
Literature review

The surge of interest is linked to the growing historical depth of available data, the

excellent downloading options and the betler delineation of catchment offered by

world DEM such the SRTM(Jarvis et al., 2008) or the Aster GDEM (Jaxa, 2010) .

2.6.4.2 Results ofrainfall-runoffmodelling with 3B42 data

Large basins are often the focus of 3b42 rainfall-runoff modelling because they

present inaccessible areas and transboundary situation. The Mekong River is a case

where very advanced used of 3b42 data have been reported(Magome et al., 2008).

The study used 29 years of data to calibrate YHylBTOP model and obtain a volume

ration of 107% and a Nash-Sutcliffe coefficient of 85%.

Similarly in La Plata basin (Su et al., 2008), the 3B42 based runoff prediction had

an average Nash-Sutcliffe coefficient of 0.71 for the daily discharge.

In the Tapajos river basin (Collischonn et al., 2008b), the Nash-Sutcliffe coefficients

were in the bracket 0.68 to 0.99.

In China, the quality of the TRMM 3B42 rainfall data is validated with the gauged

rainfall data in a 3-year periode Forced by the gauged and TRMM 3B42 rainfall data,

two continuous hydrologic simulation cases are conducted to analyze the temporal

variability and the spatial distribution of the hydrologie process in Huaihe River

basin (32° Lat N) from 1998 to 2003 .The analyses show that the TRMM rainfall is

comparable to the gauged rainfall data in the hydrologic study. The simulated

streamflow hydrograph with the TRMM rainfall is also consistent with the observed

one at Bengbu station(Yang et al., 2009).

But further North in China, in the West Laohahe River basin at latitude of 41 0_

42.75°N, three-Iayer Variable Infiltration Capacity (VIC-3L) model cannot tolerate


54
Literature review

the nonphysical overestimation behaviour of 3B42RT through the hydrologic

integration proeesses, and as sueh the 3B42RT data have almost no hydrologie

utility, even at the monthly sca1e. In contrast, the 3B42 research data can produce

much better hydrologic predictions with reduced error propagation from input to

stream flow at both the daily and monthly sca1es.(Yong et al., 2010).

In Senegal, a comparison of several satellite rainfall estimates (SRFE) as input into

MIKESHE showed that as a raw input, 3b42 generated a streamflow with a

correlation (R2 ) = 0.41 with the observed',flow, while the best SRFE (CCD) had R2

= 0.68. After correction and recalibration the best correlation obtained was with

CPC-FEWS(R2 =0.86) , 3b42 ranked last in the 5 SRFEs tested.(STISEN et al.,

2010).

In the Ganges-Bhramaputra- Megna basin, MIKEBASIN was successfully setup

with a combination of gauges and TRMM data. (Nishat and Rahman, 2009).

Each of the studies mentioned above are using different rainfall-runoff models to

generate a discharge from a rainfall. Therefore, they may not be strictly representing

the performance of the 3B42 data (and a sharp difference seem to emerge between

the 3B42RT and 3b42 research(corrected on rain gauges monthly accumulation).

Most of the authors cited request some improvementlloca1 correction in the dataset

and a lot hope is placed in future Global Precipitation Measurement (GPM) to

resolve the present discrepancies.

55
Literature review

2.6.5 3b42 and flood warning system

Three cases of flood warning systems based on 3b42 have been identified : the

Global Flood Monitoring (Hossain et al., 2007), the Mekong Flood Forecasting

(Magome et al., 2008) and the EFLOOD package developed for the Evros

catchment, extending between Greece, Bulgaria and Turkey. (Fotopoulos et al.,

2010).

To offer a cost-effective solution to the ultimate challenge ofbuilding flood alert

systems for the data-sparse regions of the world, a modular-structured Global Flood

Monitoring (GFM) framework that incorporates satellite-based near real-time

rainfall flux into a cost-effective hydrological model for flood modelling quasi-

globally has been initiated by NASA. This framework includes four major

components: TRMM-based real-time precipitation, a global land surface database, a

distributed hydrological model, and an open-access web interface. Retrospective

simulations for 1998-2006 demonstrate that the GFM performs consistently at .

catchment levels. The interactive GFM website

(http://trmm.gsfc.nasa.gov/publications_dir/potential_flood_hydro.html) shows

close-up maps of the flood risks overlaid on topography/population or integrated

with the Google-Earth visualization too1. The global DEM is based on the

SRTM,(Jarvis et al., 2008) Global soil properties are extracted from the Digital Soil

of the World (F AO 2003). The MODIS land classification map is used as proxy of

land coyer/uses and a modified version of HEC-HMS is used to generate the runoff

from the 3b42RT dataset.(Hossain et al., 2007).

56
Literature review

In order to achieve effective flood risk forecasting for poorly gauged sub-basins in

the Mekong River Basin, the feasibility of using a currently available distributed

hydrological model and satellite-based precipitation datasets coupied with a simple

statistical approach was tested. A physically based distributed hydrological model,

the YHyMIBTOP model, was used to simulate at any grid for the whole Mekong

River Basin, including poorly gauged basins. Historical discharge data for the past

29 years were reconstructed and archived with validation using ground-based

observed hydro-meteorological data with other public domain datasets as input.

Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis

(TMPA) data were forced in the model. Simulated discharge by the YHyMIBTOP

model using satellite-based precipitation generally represents the trend of observed

discharge weIl and the possibility of real-time flood risk assessment was weIl

demonstrated.(Magome et al., 2008).

The most ambitious modelling using a automatic forcing of the model with 3b42RT

data is achieved by the Department of Water Resources and Environmental

Engineering, School of Civil Engineering, National Technical University of Athens,

Greece. (Fotopoulos et al., 2010). The system was developed for the Evros

catchment, extending between Greece, Bulgaria and Turkey. The main component is

called EFLOOD and is the core of the system. It retrieves the areal rainfall

estimation for a given time period and location from 3b42RT or 3B42 and computes

the expected runoff by means of a statistical rainfall-runoff model. The estimated

high discharge values are then compared to two pre-defined values that correspond

to the alarm level and to the maximum possible flow rate that can be routed through

the crosssections of Evros river respectively. If these two values are exceeded, then

57
Literature review

EFLOOD issues either an alarm warning or a flood event. During the period 2000-

2009, the system using 3B42 research had a successfui diagnostic of 87 %, faise

alarm 8%, missed flood 1%, false flood event 4%. The 3b42 RT (real time) was less

efficient : successful diagnostic of 77 %, false alarm 13%, missed flood 2%, false

flood event 80/0. But a locally recalibrated 3B42RT reached a better efficiency :

successful diagnostic of 95 0/0, faise alarm 2%, missed flood 2%, faise flood event

1%.

58
Methodology

3 Methodology

3.1 Research question and rationales

Is the error introduced by passive-microwave- and infrared-based satellite rainfall

data tolerable for sm aIl catchments flood plain modelling ?

This question arises as a substantial number of research are undertaken to propose

sorne hydrological rnodelling driven by re,m otely sensed rainfall.

Many of these studies stops at the estimation of a discharge based on a rainfall

information generated at catchment' s level. This usually where hydrologists stop.

The significance of a river discharge is manifold but one of the crucial consequence

of high flow is the inundation of the flood plain. The para 2.3 presents only a short

summary of flood damages but that should be sufficient to convince about the

requirements of flood defence and mitigation. These actions frorn the population

living in flood prone area require some spatial framework where the flood extent and

flood depth need to be as explicit as possible. This requires a modelling effort with a

proper treatment of the potential errors and uncertainty.

... '. '~-


"
The present study intends clearly to contribute to some engineering questions of data

reliability and the impact of error propagation in modelling flood plain. The

technologies involved in the acquisition of the data discussed here (space borne

radar, passive micro-wave & infrared captors and laser ranging) are in very rapid

development and each technological steps reveals huge potentials but also generate

risks of errors/ misrepresentation. The effort placed in this study is motivated by the

author's experimental tradition developed from an early age in a photo lab where the
59
Methodology

optimal combination of exposure and development duration had to be interpolated

between inadequate attempts (probe and automatic calibration were at their infancy

and unaffordable). The present focus is not to elucidate exhaustively the

performance and error of each instrument and its data processing but to look at a

final pragmatic result : a flood plain delineation processed with different data set.

The error generated by the passive-microwave- and infrared-based satellite rainfall

data will be identified by comparing the flood extent modelled based on stream

gauge data measurements, rain gauge data and passive-microwave- and infrared-

based satellite rainfall data. The modelling of the flood plain requires the

transformation of rainfall information into a river discharge which is done through a

rainfall-runoffmodel.(Beven, 2004). The link between the flood plain and the

satellite based estimation of rainfall is the backbone of the present study. Time and

word count constrains limit the investigation of connected· topics such as the

calibration of satellite based data, calibration of rainfall-runoff model, model

structure choice, estimation of errors and uncertainties, mobilisation of flood plain

model in emergency planning and management, and flood defence design but their

point of branching will be noted at the appropriate chapters.

Along the resolution of the central research question, some additional subsidiary

questions emerged while mobilising the various tools and software. One concem, the

adequacy of the ReFH method outside UK, the other is related to the use of LIDAR

data as the only source for river cross-sections.

60
Methodology

3.2 The flood plain of interest

The choice of the flood plain to be modelled resulted of practical considerations

emerging during the study rather than planned at its inception. A cruciallimiting

factor was to obtain a software package sufficiently recent, properly debugged and

well documented. These constraints result from the academic timeframe. The other

constraints is related to the access to satellite data, topographical data, river flow and

rain gauge data. A manageable flood plain South of Villers-sur-Semois (49.69° Lat

N, 5.56° Long E see Figure 9 : Location of Villers-sur-Semois) emerges as fulfilling


~;-.;.-.--~- ..
the following criteria:

a) Being of a size that the 2,500 cells of ISIS 3.3 free could modelled in 2

dimensions(Halcrow, 2010a). The size of the active area is 5.8 ha

b) located in the Semois basin for which river flow, rain gauge data, and LIDAR

data were available

c) Being located within the coverage of the TRMM satellite observation (Lat <

50° N (NASA, 2010b»

61
Methodology

Figure 9 : Location of Villers-sur-Semois

3.3 The reach of interest

The 1D module of ISIS free provides 250 nodes and withiil this limitation it was

possible to compute 103 sections at 50 meter distance downstream of the stream

gauge ofSt-Marie-sur-Semois (49.68495 °Lat N, 5.5634° Long E).

The mode lIed reach is 5,101 m long, cross two roads with bridges and does not have

any confluence with a tributary. The Figure 10 shows its location.

62
Methodology

Figure 10 : Location of the modelled reach

3.4 The catchment of interest

The catchment upstream of the stream gauge ofSt-Marie-sur-Semois was delineated

based on the 10 m DEM provided by the regional authority with the ArcHydro

utility (ESRI, 2009). The result returns 141.3 km2 , the official catchment size for the

gauging station is presented as 143 km2 • In the present study the 141.3 km2 value is

used.

3.5 The period studied

It is considered relevant to include the January 2003 flood in the study. At the time

of ordering the ground data, the plan included using the GSMap MVK+ which are

available for the 2003-2008 period. Therefore the data requested from the local

authorities included only November 2002 to December 2008. This is considered a

reasonable time span for an MSc dissertation. But the 3b42 data set is available from

January 1998 to present and the whole period could have been included in the study ~

The data processed here starts on 01/11/0201:00 AM and ends on 31/11/08 12:00

PM.

-.,-" --.:~ .. ~.

63
Methodology

3.6 The choice of passive-microwave- and infrared-based satellite

rainfall da ta

The initial intention was to compare several data set namely : CMORPH (Joyce et

al., 2004), TRMM 3B42 (NASA, 2008) and GSMapK+(USHIO et al., 2009) but the

time constraint made TRMM 3B42 the only possibility. The main advantages of

3b42 is the very comprehensive interface for data extraction: TOVAS(Kempler,

2010). The 3b42 product contains more inputs than the MW and IR results as it

incorporates also the observation from th~ space borne rain radar (see para 2.6.3.3).

The possibility of near real time access (latency of 6 to 9 hours) is an attractive

element for the 3b42RT product although this had no influence in the present study.

The data set used here is the 3b42 ver 6. Re se arch (NASA, 2008, NASDA, 2001).

3.7 The choice of lD-ZD flood plain modelljngosoftware

A large consensus supports the flood modelling based on an interaction between 1D

and 2D modelling.(Wallingford, 2010, DHI, 2010, Ernst et al., 2009, Pender, 2009,

Werner, 2004, Horritl and Bates, 2002, Syme, 1991). The Department of Geography

at the University of Hull has no license for any 1D-2D modelling software, ISIS free

was the only affordable option. The cost of the 4 days training in Leeds and London

was around 1,200 f despite a substantial rebate.

3.8 The choice of rainfall-runoff modelling package

The choice ofrainfall-runoffmodel is very open and the lack of experience and tight

deadline led to a choice based on the existing flood modelling recommendation in

UK where the ReFH method is a respected norm (Faulkner and Barber, 2009,

Kjeldsen, 2007). The parameterisation of the mode1looked rather simple at fust sight
64
Methodology

although this is the case for UK only where catchments descriptors are readily

available. The purchase of the software amounted to 294 f..

3.9 A brief introduction to the ReFH package operation &

parameters preparation

The rainfaIl-runoff concept is described in the heading 2.5.9. The parameterization

requires two types of catchment characteristics : the geometrical and the

hydrological.

The geometrical characteristics are :

1. (AREA) : Catchment drainage area derived from the Digital Terrain Model

(km2)

2. (AL TB AR) : Mean altitude of catchment (metres above sea level)

3. (ASPBAR) : Mean direction of aIl drainage path slopes (bearing in degrees)

4. (ASPVAR) : Invariability of slope directions (values approaching 1.0

indicate dominance of one direction)

5. (DPLBAR) : Mean of distances along drainage paths between each 50m grid

node and the outlet (km)

-.:_".- '.~

6. (DPSBAR) : Mean of aIl inter-nodal slopes, along drainage paths (mIkm)

7. (LDP) : Longest drainage path from source to outlet (km)

8. (URBEXT) : Extent ofurban and suburban land coyer

65
Methodology

AlI these parameters were derived from the 10m DEM provided by the regional

authorities.

The hydrological characteristics are :

1. (FARL): Index of flood attenuation due to reservoirs and lakes (1.0

indicates no attenuation)

2. (BFIHOST) : BaseFlow Index derived using the HOST (Hydrology Of Soil

Types) classification

3. (SPRHOST) : Standard Percentage Runoff derived using the HOST

classification

4. (PROPWET) : Proportion of time when soil moisture deficit was below 6mm

during 1961-90

5. (SAAR) Standard period (1961-1990) average annual rainfall (mm)

6. (SAAR4170) 1941-70 average annuaI rainfaiI (mm)

7. (RMED 1H) Median annual maximum 1-hour rainfall (mm)

8. (RMED1D) Median annuaI maximum 1-day rainfall (mm)

9. (RMED2D) Median annuaI maximum 2-day rainfall (mm)

RMEDIH, RMEDID, RMED2D have been calculated based on the 6 years hourly

data available.

The long term average rainfall (SAAR) have been extracted from FAO Loc_ Clim

database (Grieser et al., 2006).


66
Methodology

The BFI has been ca1culated by extracting the base flow from the 6 years data

available with the software ABSCAN.(Free Software Directory, 2006)

PROPWET has been estimated with a regression using the long term rainfall as

independent variable and the FEH-CD ROM data (WHS, 2010).

SPRHOST has been estimated with a multi-variate regression using as independent

variables: the hydrological properties of the European Soil Database (JRC, 2010),

the Corine Land Coyer 2000 database (Agency, 2000), DPLBAR, DPSBAR, BFI,

PROPWET and the SPRHOST data available from the HiFlows-UK v3.02 data set

(Environment Agency UK, 2010). Twenty four catchments with the same water

regime as the St-Marie-sur-Semois' catchment have been used to establish the multi-

variate regression. (R2=O.89) .

URBEXT was estimated based on a regression between the Corine Land Coyer 2000

database and the URBEXT data available from the HiFlows-UK_v3.02 data set.

The model is calibrated with rainfall-runoff data grouped in flood events. Each event

requires a synchronised set of rainfall-runoff observation. The timestep is variable

but aIl the events need to be presented with the same timestep.

The initial moisture is obtained from antecedent rainfall. For the evapotranspiration,

ReFH generate a sin curve based on an annual daily mean

During the calibration run, the following parameters are optimised :

FC : Field Capacity

RD : Routing depth

67
Methodology

FC scaling factor (SM)

CMax (mm): Maximum Soil Moisture Capacity

DK: Daily Soil Moisture Decay rate

Tp : Time to Peak

Up and Uk : height and kink of hydrograph

BL : Baseflow recession constant (lag)

BR : Baseflow recharge.

Once the model is calibrated, rainfall data and evapotranspiration annual daily mean

are introduced as input for the flood period and the antecedent period; the model

output a simulated discharge.

3.10 The rainfall data sources & processing

3.10.1 Rainfall Gauge

The typing bucket gauges used in the Semois basin are all equipped with processeur

and modem allowing the accumulation on hourI y basis and the transmission to a

central database (Water Control Data system for Hydrology and Water Management)

(RW, 2010) operated under the "Direction générale opérationnelle de la Mobilité et

des Voies hydrauliques". The houri y data were provided by "SPW - DGO. Direction

de la Gestion hydrologique intégrée" based on a request by email. Daily rainfall data

are available online at :

http://voies-hydrauligues.wallonie.be/opencms/opencms/fr/hydro/Archive/annuaires/index.html

68
Methodology

A total of 266,640 rain gauges records were used.

3.10.2 3b42 data set

The whole time span was downloaded pixel by pixel using the TRMM Online

Visualization and Analysis System (TOVAS). The URL

(http://disc2.nascom.nasa.gov/Giovanni/tovas/TRMM V6.3B42.shtml) allows a

spatial and temporal selection very convenient for a small area. The data can be

downloaded on Ascii, HDF, or binary format.

A total of 124,425 records were used.

3.10.3 Aggregation of rain gauges hourly data to 3 hourly data

3b42 data are provided as rain rate in mm!h. To transform them into accumulated

rain during the 3h covered, they were simply multiplied by 3. The rain gauge data

were summed during the same period (1.5 hour before, 1.5 hour after the nominal

time). Downscaling the 3b42 to a 1 hour timestep was considered too demanding for

the present study. It is noticing that the GSMap MVK+ product (USHIO et al., 2009)

has a one hour timestep and the latest version of CMorph (Joyce et al., 2004)
.r-.-\.W"_'" •

provides data with a 30 minutes timestep but only as rotating file set for the last 31

days. The archive are also with a 3 hours timestep.

3.10.4 Comparison between 3b42 pixel and individual rain gauges

Four rain gauge stations data were available with a 1 hour time step : Vresse, Sugny,

Bouillon, Bertrix. A fifth station was available with a Iday time step. Bouillon and

Fratin daily data are highly correlated (r=0.75) and it was decided to downscale the

Fratin data using the proportion of daily rainfall for each hourly observation from

Bouillon. As Fratin is the only rainfall station located within the catchment of

69
Methodology

interest (see Figure Il : Rain gauges localisation) and the inclusion of its data

improves the correlation between 3b42 and rain gauges, it was decided to base the

whole study on a data set including the downscaled data from Fratin station.

Rain gauges localisation

legend
• Raingauges
- SemorsR;ver
o Catchrrsntof St M.rie-sur-Semots
o Belgium boundaries

Figure Il : Rain gauges localisation

Based on the shortest distance between gauges and pixel centroid, the pairs in Table

Il are used for comparison.

70
Methodology

Table t t: 3b42 pixel paired with rain gauge

Pixel refno Rain Gauge

0 Sugny & Vresse

1 Bouillon

2 Bertrix

5 Fratin

The location of 3b42 pixel centroid and rain gauges is shown in Figure 12

A 3b42 pixel centroids & rain gauges location

o \t~. 1 2 3
• • • •

Frain.
Legend
6 5


• Rail Gauges
3M2 pixel centr oid
• •
D Semoi s Basin o 3.5
,! !!
7
I! !
14 Kilometres
Il

Figure 12 : 3b42 pixel centroid and rain gauges location

71
Methodology

3.10.5 Bias

The bias is the difference between the average of the whole time series of 3b42 data

and the rain gauges. It has been calculated for each pair 3b42-rain gauge. The

relative bias is the difference divided by the average of the rain gauge series.

3.10.6 Root Mean Square Error

The difference between each pair of3b42-rain gauge is squared for every timestep,

the square root of the mean square of these deviation is ca1culated for each pair of

3b42-rain gauge.

3.10.7 Probability of detection

The probability of detection (POD), also known as Hit Rate (HR), measures the

fraction of observed events that were correctly forecast. It has been evaluated for

each pair 3b42 - rain gauge.

3.10.8 False alarm Rate

The false alarm ratio gives the fraction of forecast events that were observed to be

non events.

3.10.9 Heidke Skill Score (HSS)

The HSS measures the fractional improvement of the forecast over the standard

forecast. It is normalized by the total range of possible improvement over the

standard, which me ans Heidke Skill scores can safely be compared on different

datasets. The range of the HSS is -00 to 1. Negative values indicate that the chance

forecast is better, 0 means no skill, and a perfect forecast obtains a HSS of. 1

(Online, 2009). The equation is as follows :

72
Methodology

HSS = _(h_I_"ts_+_co_~_re_c_t_n_e_g_at_iv_e_S_)_-_(_exp_e_c_t_ed_C_O~_/î_ec_t---:)ra;.=..;ndo.:..=..:..:.;;.nt
N - (exp ected correct }randont

where

l 1 [(hits + misses) (hits + false alarms) + -


~exPected correc tJra ndo nt =-
N
.. .
(correct negatlves + misses) ( correct negatlves + false alarms)_

3.10.10 Inverse distance weighing (IDW)

The weight of each source of rainfall data relative to the catchment area was

calculated by the following process =

1. Create 10m grid in the catchment area

2. Create the centroid point for each cell

3. Calculate the weight of each centroid point using the Shephard formula

-p
hi
Wi = n
Lj=l hj
-p Equation 12

Where p is where p is an arbitrary positive real number called the

power parameter (typically, p=2) and hi is the distance from the

scatter point to the interpolation point.

4. Average the weights relative to each source ofrainfall to obtain the weight

coefficient.

5. The weight coefficients multiply the rainfall data at each timestep.

6. The sum of the weighted rainfall provides the IDW rainfall in the catchment

area for each timestep.

73
Methodology

3.11 Source of discharge data

The stream gauges used in the Semois basin are aH equipped with processor and

modem allowing the averaging hourly basis and the transmission to a central

database (Water Control Data system for Hydrology and Water Management) (RW,

2010) operated under the "Direction générale opérationnelle de la Mobilité et des

Voies hydrauliques". The hourly discharge data were provided by "SPW - DGO.

Direction de la Gestion hydrologique intégrée" based on a request by email. Daily

discharge data are available online at :

http=//voies-hydrauliques.wallonie.be/opencms/opencms/fr/hydro/Archive/annuaires/index.html

The flow is recorded with a float-operated Thalimedes Shaft Encoder with integral

data logger designed for continuous, unatlended monitoring of water level. It

generates a measure every two minutes and calculates on site the hourly average

which is sent to the central database. An alert level can be'set to force a transmission

of data earlier than the routine interval. The precision is around 0.5 cm in water

depth (around 10 l/s).(Dierickx, 2010).

3.12 Source of evapotranspiration data

The evapotranspiration is retrieved from the GLDAS model. The data are available

on 3 hours 0.25 deg basis. The series was downscaled to 1 hour timestep by linear

interpolation. But the ReFH package returned an error while reading the evapo-

transpiration data and this could not be corrected. The option to simulate the ETP

based on the annual daily average was chosen.

74
Methodology

3.13 Preparation of cross sections

3.13.1 Correction of LIDAR data

The LIDAR data collected during 2000 to 2002 over the corridor of the main rivers

in Region Wallonne have recorded tirst and last pulse at a density of 8 pulses/m2

averaged to 1 elevation/m2 • The average vertical precision is Il cm. The last pulse is

expected to represent the altitude of the soil under the vegetation but many of the

trees around the river have echoed the signal rather than the river. This resulted in a

considerable distortion of the river profile'as illustrated in the Figure 13 to Figure 20.

As these LIDAR data were communicated very late in the course of the study (end of

July 2010), a rapid correction was deemed necessary. The small software Raster Edit

provided by the Departement of Geography at Hull University was used to

interpolate between points of consistent elevation but operating with the initial 1 m

DEM was too slow. The data were then aggregated to 4m using ArcGIS 9.3 and the

resample tool with bilinear interpolation option. The correction with RasterEdit have

been then tinalised on this 4 m DEM. It took 4 days ofwork to cover the 100 m

corridor along the river centralline on the 5 km reach. For the area outside the

corridor, patches of irrelevant elevation have been replaced by elevation originating

from the 10m DEM (Mokhadem, 2010) resampled to 4 m.

75
Methodology

Fi re 13 : Localisation of Ion

Figure 16: Cross-section before DEM correction

Figure 17 : Long section after DEM correction


- - - - - -
Figure 18 : Cross-section after DEM correction

'riaI_~ JI-----

w.--='-~
__ ro--- ..... _.. ~

-w. .. _ - -
rOllPoJ'i\n.'flND 'AeW - - - -

3)Vel;bDildtyQ:ltcn
J;1 QdsJ3)V~h.t1.mn~~Tet:IrIsnaœ

Figure 19: 3D view before DEM correction Figure 20: 3D view after DEM correction

3.13.2 Generation of cross sections with ISIS Mapper

A river centralline was elaborated based on the vector data of the topomap available

on the cartographie portal of the Region Wallonne, corrected visually based on the

Google Earth image. This centralline and the corrected 4 m DEM are used as input

in a ISIS Mapper tool to generate the cross-sections as presented in Figure 21

76
Methodology

er~ect I.,ayer lacis yje-N ~ngs !:ielp


; • ~H 1~ ;51 ~ 1~ e; ~ ~ ~o<; El 1\ . :!H 0 ~~ ~
···· !laD - 1 I m l . 4 ;m Mode.fingTooIBax
Matin Bridg" iiIJl 10 _ a
. . . 1-V' L.OlIO"-II
_ M.ninOlMd",,_Simpline.l è-IP' Sbep.'"
i:;:l 0 M.tinatMd em_SimpLine.l III tI' ISIS SpI

~œ MarieXS2 ~ :::::'ns
~ RNerCentrelile
t-# l OldRNtfS«OOns
;-t? Oraw RNer SecIj)ns
f-.# .rt
!,-~ EdIRNerSedioos
RNer Sednns from66 OAJ Fie
~ AssignRiterCerrtreÜ'le
~ ::;:,:~= on RNer cen~ U1e
~ nt~lateRrlerSe.ebaM
f-.:P' Ex1nIdZV.~ ..
'--# Save .. ISIS ED Ae
lm 2D _ a
:-tI' L• • d06l
$-(p loatiUOdeluode:$
":tP
Lo!.!t20UDd~Shapeflles
0-11' Cl'eale5S20Fies
- ~ Run SS20~lerfice
61 !-tP' .....rtllFAe
9..q. C,.... TUFLOW.,..
ffrÛ' Export TUFLOW"" To UF/UD Fe"""
~ lOAood ~ --.
:""tI' LoOl106l
:--t!' Lo.dTl.
'\

\
~# Lo"SilapeAe
L..# load Tabular csv Tex! fie. M Shepe fle
'-tP' EdiSll,pe,..
~ TriangulateShllpefJes
f-.q> ... TI.
~(""7'
; l "=''''-=
Ii)'':"'1~
-,=-I!l=(3'""11 !--# Md B.erylS5 R..... "'TIl
f=-- - - - - - 'I j-t? M6Tabu.rCSV Resulstonl
~ ChangeOispio!yedReJul
~ Generlleflooduap
'-V' ANna'eRe<U"
L#' R..:o<dAVl
!t 2D Aood ~
- lP' l ••dOEU
------ -------._---- \\
\
r-tl' lo, !l20Res:ib --""-_ 'i
~ GenerateFb:)dLlap .--....-----~
~ AniTla1eRes.ul:5
~ ReCGrdÂV1

Figure 21 : Generation of cross-sections in ISIS Mapper

3.14 Generation offlood events and antecedent rainfall sets

ReFH is an event based deterministic model, the separation of peak flow and

baseflow cannot be automatically carried out by the packages. Therefore, the five

largest peaks have been extracted from the data set available. This number is rather

arbitrary based on time constraints. The five peak flows appear to have created sorne

inundations in the modelled reach (see Figure 22 : Maximum stage 5 flood events) .

Once the peaks have been selected, the rising limb and recession limb were based on

the departure from the long term average flow and return to this long term average

flow. The period between the end of the recession limb to the beginning of the next

peak rising limb is considered an inter-flood period or antecedent rainfall. The

various periods are presented in Table 12.

77
Methodology

Table 12 : Flood and flood antecedent periods

Discharge m3/s

Period Start End Average Min Max Duration (days)


Pre FL1 01/11/200200:00 21/12/200207:00 3.59 0.86 19.12 50.29
Pre FL2 10/01/2003 10:00 30/12/2006 18:00 1.47 0.17 19.98 1450.33
Pre FL3 26/01/200708:00 01/12/200708:00 1.75 0.40 19.39 309.00
Pre FL4 15/12/200721:00 01/02/2008 02:00 2.86 0.79 Il.41 47.21
Pre FL5 13/02/200802:00 26/02/2008 18:00 1.41 1.21 1.94 13.67
Pre FL6 09/03/2008 02:00 30/11/200823:00 1.77 0.43 17.56 266.88
FL1 21/12/2002 08:00 10/01/200309:00 . 9.61 1.95 37.76 20.04
FL2 30/12/2006 19:00 26/01/200707:00 7.34 1.95 30.25 26.50
FL3 01/12/200709:00 15/12/2007 20:00 8.92 1.94 24.45 14.46
FL4 01/02/2008 03 :00 13/02/2008 01 :00 7.77 1.96 24.33 Il.92
FL5 26/02/2008 19:00 09/03/200801 :00 5.47 1.95 20.55 Il.25
General average
abs min max 1.87 0.17 37.76

The selection of the 5 large st peak as flood events is for practical purpose ; there

were peak flow of magnitude similar to the smallest ·flood flow during the inter flood

period. A more hydrological method would have been to simulate a variety of

discharge in the 1 D model and determine the flood threshold based on the presence

of a flow overtopping the banks at sorne point of the channel. This threshold may be

around 16 m 3/s. But the present work intends to focus on the flood plain extent and

minor floods are less significant for planning purpose.

78
6L

... -~ -.~~--

SlU<lt\<l poou S <l~hqs WnW!XBW : ZZ <lJn~!.!I

aL!Xl1 t.::------------
><20,002 l
x20_003 ( 'L ____ _
><20,004 î
1
f
f
x2O-'J05
-ZOJJ06 1 .----_:>-_:_-----------
:=:
-ZO,OO9
7
-ZO,010 L1
i :'~:~ ~

-~r>
-ZO'013 7
-ZO'014 7
-ZO'015 7
1
.20'016 , 7
-ZO=017 7
f -zo 016
7
f x2O'019
x2O'020
-zo'021 Î
1
"
x2O'022 1 '----------------- .._-- ...
x20'023 --,
-ZO'024 --,
-ZO'025 --,
-ZO=026 --,
x20027
x2O'028 -,
1 .20'029 --,
f -zo'OOO ,
x2O'031 Î
f X2O=032~
x2O_033 t----f
><20,034
x20_035
-ZO,036
-ZO,037
x2O,036
x20,039 "
x2O_040
-ZO,041
x2O_042
><20,043
x2O_044 .i
><20,045
x2O_046
.20,047
J
~
BRU

~ :=:
I~3
i x2tU>5) i~
f x20,054
x20,055
1 x20,055
.20,057
x20,056
i
x20_059
-ZO,06O
><20,061
-ZO,062
-ZO,063

o x20,066

x20,070
,~
1

-ZO,071

,0~
-ZO,072
x20,073
x20 074

i
.20=075
x2O_078
-ZO,077
x20_078
! -ZO,079

~ -ZO,06O
><20,061
,~ x20_082

~ -ZO,063
x20_084
,0~ .20,065
8 -ZO,066
><20,067
x20_088
x2O__

><20,090 -
-ZO,091
-ZO,092
><20,093
><20,094
-ZO,095

~~~I
:=::
><20=100 .
--_:~(~;%"':::::::-f::;;~:::::;:=~
---o f /.,
.
:-'--1' - r--,------'~ --------------------~~~~::::: :::::::.~~"'_
,
__t
~--------- -::: __ ==_ ..
!

Â3010PO-ql~W
Methodology

3.15 Calibration and validation set

Two ensembles of calibration and validation data have been tested. The first one

used the 4 smaller flood events (rain and discharge) for calibration (FL2 to FL5) and

tested the predictive capacity with the rainfalllinked to the FL 1 period. The second

used the FL 1, FL3 to FL5 flood events for calibration and FL2 rainfall for validation.

3.16 Running ISIS ID with the various flood events

The sections described at heading 3.13.2 are imported in ISIS ID, the two bridges

are modelled using approximate measurements based on available levels and basic

construction standards.

A Flow-Time boundary condition is used at the head providing a discharge

hydrograph based on 1 hour or 3hours timestep. A Normal/Critical Depth Boundary

(NCDBDY) unit is specified as a downstream boundary, this automatically generates

a flow-he ad relationship based on section data. (Ha1crow, 201 Ob)

A Manning N value of 0.030 for the channel and 0.015 for the flood plain is adopted

based on the similarity with the Trent River study (McGahey et al., 2006)

Six events are modelled : FLI and FL2 based on gauged flow, rain gauge based

modelled flow and 3b42 modelled flow.

The timestep is fixed at lOs; saving every 1200 s.

. .
..
;'::..~- '~ .:'.'

80
Methodology

3.17 Running the ISIS ID 2D with the various flood events

One major source of delay was that the version 3.4 of ISIS Free installed on a

QuadCore based computer with Windows 7 (the core computer for the present

study) did not process the link 1D-2D while the version 3.3 installed on a dualCore

with Windows XP did. A lot of time was wasted due to this bug.

The 2D domain, computational area and active area went through a succession of

reduction to fmally reach a do main that was computable with the limited number of

cell (2500).

The procedure of linking the 1D to the 2D model involves the creation of a link line

(z polyline) at the edge of the 2D domain, along the main channel.

The vertices of this link line are supplied with the 1D model nodes they should use

to pass the flow information to the 2D mode!.

The Water levellinking (type H) is used ; water levels from 1D nodes are taken by

the model, and imposed as a boundary condition on the 2D model. The 2D part of

the computational calculates the flow at these linked boundaries, and passes this

information back to the 1D part, where it is addedlremoved from the channel, to

ensure mass conservation. (Halcrow, 201 Ob).

The timestep of the 1D model is lOs, it is 1 s for the 2D model.

81
Results and discussion

4 Results& discussions

4.1 Comparisons of 3b42 data with rain gauges

4.1.1 Bias

Comparisons pixels by pixels with the five rain gauges in the Semois basin reveals a

considerable negative bias ranging from -0.11 to -0.18 mm /3 hours with an average

of -0.14. The relative bias (bias/mean) is in the bracket -27% to -39%.

The largest bias is observed between the pixel 1 and the rain gauge located in

Bouillon as seen in Table 13.

4.1.2 Root Mean Square Error

The RMSE is ranging between 1.64 and 1.91 with an average of 1.80. The largest

RMSE is observed at Bouillon (Table 13). The relative RMSE is really high: 426%

Table 13 : Individual gauges comparison

Rain Vresse Sugny Avg Bouillon Bertrix Fratin

gauges Sugny

Vresse

3b42 pix 0 0 0 1 2 5

RMSE 1.6942 1.6881 1.6453 1.9005 1.9065 1.8915

Bias -0.1384 -0.1140 -0.1262 -0.1856 -0.1339 -0.1106


:. ---.- _--r.·

Relative
bias -34% -29% -32% -39% -31% -27%

82
Results and discussion

4.1.3 Probability of detection

The probability of detection (POD) is the likelihood that the event would be forecast,

given that it occurred.(Ramirez-Beltran et al., 2007)

The POD is particularly low, ranging between Il and 15 % with an average of 13 %

4.1.4 False alarm Rate

The false-alarm rate is the proportion of forecast events that fails to

materialize.(Ramirez-Beltran et al., 2007)

The FAR is very low : 3 0/0

4.1.5 Heidke Ski Il Score (HSS)

The HSS measures the fractional improvement of the forecast over the standard

forecast. The range of the HSS is -00 to 1. Negative values indicate that the chance

forecast is better, 0 means no skill, and a perfect forecast obtains a HSS of. 1

(Online, 2009). The equation is presented in the heading 3.10.9.The HHS score of

the four 3b42 pixels ranges between 0.01 and 0.18 with an average of 0.14

4.1.6 Correlations

The correlation coefficient (r) 3b42 pixels with individual rain gauges ranges

between 0.37 and 0.46 with an average of 0.42. The highest correlation coefficient is

''''~''-'' , .... observed between the pixel 0 and the average of the gauges located in Vresse and

Sugny as presented in Table 14

'~".""'~.~ -.

83
Results and discussion

Table 14 : Correlations coefficients with individual rain gauges

Pixel Vresse Sugny Avg Vresse Sugny


0 0.45 0.43 0.46

Pixel Bouillon Bertrix Avg Bouillon Bertrix


1 0.42 0.40 0.43
Pixel Bertrix
2 0.40
Pixel Fratin
5 0.37

4.1.7 Individual gauges comparison performance review

The number of 3b42 pixel covering availcible gauges data is very low to draw

conclusion for the whole product. It is a small sample but it is the one operational for

the purpose of the present modelling exercise.

The performance of the 4 pixels of this study is rather poor ; comparing the 3b42

data at 0.5 0 resolution with a network of76 rain gauges in Greece, Feidas (Feidas,

2010) found that the 3b42 data had a low relative bias ( -4.2%), a much lower

relative RMSE (48.4%) and a much higher efficiency (0.76) and a correlation

coefficient of 0.88.

The difference are very like1y linked to the lower spatial resolution (the present study

used 0.25 0 resolution) and to the lower latitude (Greece is around 39 0 Lat N). The

present study is at the very extreme North of the 3b42 data domain.

A validation with rain gauges in Thailand (Chokngamwong and Chiu, 2008) shows

that around ISOLat N the 3b42 data most often overestimate the rain gauges except

for the Northem most region where altitude compounds with higher latitude and

mountainous c1imatology to produce a negative bias for the 3b42 data.

84
Results and discussion

The Northem most Ground Validation sites for the 3b42 data are located at 35° Nin

Nagoya (Japan) and Mesonet (Oklahoma USA) ; the difference of latitude

compounds with a large difference in continentality with the present study.(Hedge,

2010)

It seems relevant to note that a negative correlation exist in the present study

between the altitude of the rain gauges and the gauge to 3b42 correlation (r= -0.42).

This relation is also observed in the Laohahe basin, China around 41 ° lat N (Y ong et

al., 2010).

4.2 Comparisons of Inverse Distance Weighted rainfall for the

whole Semois watershed.

The Shephard IDW was calculated (see heading 3.10.10) and applied for the whole

Semois watershed for both the 3b42 data and the rain gauges. The correlation

coefficient between the 7 pixels and the 5 rain gauges is equal to 0.5. It is 20% better

than the average of individual rain gauges correlation coefficients. The relative bias

is -31 %. The improvement is related to the aggregative nature of the satellite

observations which have a better ability to represent areal rainfall than discrete rain

gauge observations.

4.3 Comparisons of Inverse Distance Weighted rainfall for the St

Marie-sur-Semois catchment.

The Shephard IDW was calculated (see 3.10.10) and applied for catchment upstream

of St Marie-sur-Semois gauging station for both the 3b42 data and the rain gauges.

Three rain gauges stations (Bertrix, Bouillon, Fratin) and three 3b42 pixels (3,4,5

see Figure 12) are evaluated for the correlation and bias. The correlation coefficient

85
Results and discussion

is equal to 0.39 and the relative bias to -34%. The reduction of scale and increase in

the rain gauges elevation seems to increase the error/divergence of the 3b42 data.

Seasonal effect will be discussed while discussing precipitations during flood events

(heading 4.4.)

4.4 Comparisons of rainfall during flood periods

As presented in the heading 3.14, five major flood events have been identified.

Table 15 : Five major flood events

Period Start End Duration (days)


FLI 21/12/200209:00 10/01/2003 09:00 20.0
FL2 30/12/200621:00 26/01/200709:00 26.5
FL3 01/12/200709:00 16/12/200700:00 14.6
FL4 01/02/200803:00 13/02/200803:00 12.0
FL5 26/02/2008 21 :00 09/03/2008 09:00 11.5

During these periods, the cumulated IDW rainfalls in ~t Marie-sur-Semois watershed

are as follows :

Table 16 : Accumulated rainfall du ring five flood events

Rain gauges IDW 3b42 IDW


Period (ace. mm) (ace. mm) Bias Relative bias
FL1 204.5 193.0 -11.5 -6%
FL2 261.3 76.5 -184.8 -71%
FL3 157.6 44.3 -113.4 -72%
FL4 83.7 31.2 -52.5 -63%
FL5 76.0 45.0 -30.9 -41%
Sum 783.1 390.1 -393.0 -50%

The relative bias has a wide bracket from -6% to -72%. The average relative bias

during flood periods is quite superior to the relative bias for the whole period of

observation. The fact that the five floods occurred during the winter leads to a

86
Results and discussion

degradation of the 3b42 performance (the algorithm does not differentiate weIl cold

with rain and cold without rain). This is also observed by F eidas in Greece (Feidas,

2010) who notes a -15% relative bias during the winter while during the other

seasons it is close to 0%. But Su (Su et al., 2008) in the La Plata basin (15° to 36°

Lat S) does not observe such a seasonal degradation.

Table 17 : Nbr of 3 hours period > 0 mm of acc. rainfall- Flood periods

Period Rain gauges IDW 3b42 IDW


FLI 109 16
FL2 123 10
FL3 74 5
FL4 43 7
FL5 49 9
Sum 398 47

The capacity to detect rain during the winter' s flood period seems really poor as
shown in Table 17 . The POD falls to 12% during the flood period while it is 14%
for the whole period of study.

4.5 Comparison of rainfall during inter-flood periods

As presented in the methodology the inter-flood periods are :

Table 18 : Inter-flood/high flow periods

Period Start End Duration (days)


Pre FLI 01/11/200203:00 21/12/2002 06:00 50.1
Pre FL2 10/01/2003 12:00 30/12/2006 18:00 1,450.2
Pre FL3 26/01/2007 12:00 01/12/200706:00 308.7
Pre FL4 16/12/200703:00 01/02/200800:00 46.9
Pre FL5 13/02/2008 06:00 26/02/2008 18:00 13.5

During these periods, the cumulated IDW rainfalls in St Marie-sur-Semois watershed

are as follows :

87
Results and discussion

Table 19 : Accumulated rainfall during five inter-flood events

Rain gauges
Period IDW 3b42 IDW Bias Relative bias
-
PreFLI 221.9 133.0 88.9 -40%
-
PreFL2 4,064.0 2,924.1 1,139.9 -28%
-
PreFL3 1,134.0 738.3 395.7 -35%
-
PreFL4 155.2 36.2 119.0 -77%
-
PreFL5 22.7 6.6 16.2 -71%
-
Sum 5,597.7 3,838.1 1,759.6 -31%

The relative bias is much lower than during the flood period. The winter effect is the

most available explanation.

The same applies to the POD (18%) as observed in Table 20.

Table 20 : Nbr of 3 hours period > 0 mm of ace. rainfall- inter flood periods

Period Rain gauges IDW 3b42 IDW


Pre FLI 236 13
Pre FL2 5238 742
Pre FL3 1217 173
Pre FL4 178 9
Pre FL5 30 2
Sum 961 172

4.6 Comparison ofmodelled flows

4.6.1 Introduction

As explained in the methodology, the revitalised FSRlFEH rainfall-runoff

method(Kjeldsen, 2007) was used with the adequate software package ReFH 1.0

sold by Wallingford HydroSolutions (WHS). The parameterisation of the model is

explained in the heading 3.9. Out of 5 events available, 4 were used for calibration 1

88
Results and discussion

was used for validation. Two combinations of events are reported here : combination

1 use the FL2 to FL5 events for calibration and FL 1 for validation; combination 2

use the FL 1 and FL3 to FL5 events for calibration and FL2 for validation.

Two comparisons are envisaged : a) the comparison of the modelled flow using the

rain gauges IDW with gauged flow measured at St-Marie . . sur-Semois at 1 hour

timestep; b) the comparison of the modelled flow using the 3b42 IDW at 3 hours

timestep with the 3 hours average of the gauged flow measured at St-Marie-sur-

Semois.

4.6.2 Total, Average, Peak, Minimum& Error

The descriptions of the flows is presented Table 21 and Table 22.

Table 21 : Flow descriptors - one hour timestep

One hour timestep (m3/s)

Standard
Period Source Total Average deviation Peak Minimum RMSE NSE
FLI Gauged 4538.8 10.3 9.3 37.8 2.0
Rain gauge Mod 3118.1 7.1 8.1 43.4 1.5 5.9 0.33
FL2 Gauged 4673.6 7.3 6.2 30.2 2.0
Rain gauge Mod 4027.9 6.3 7.8 43.5 1.2 5.5 0.23

89
Results and discussion

Table 22 : Flow descriptors 3hours timestep

Three hours
timestep m3/s

Standard
Period Source Total Average deviation Peak Minimum RMSE NSE

FLI Gauged 1,399.4 11.6 9.8 37.5 2.1


3b42
Mod 3,973.4 32.8 45.7 228.0 0.4 46 <0

FL2 Gauged 1,459 8 6.46 30.1 2.1


3b42
Mod 141 1 2.21 19.7 0.0 9 0.09

The sum of gauged flows differs between the two timesteps because the flow is

averaged on the 3 hours timestep and create a slightly different time period because

the departure and return to baseflow do not coincide exactly between the two

timesteps. The multiplication by 3 of the three hours sum is nearly equal to the sum

of the one hour timestep. The modelled flow based on rain gauges gives a result with

a reasonable error (relative RMSE = 66%) and providè a peak flow within a 50%

bracket ; this contrasts sharply with the 3b42 modelled flow which exhibits a relative

RMSE of258% and a peak flow within a 500 % bracket. The Nash Sutcliffe

coefficient of efficiency (NSE) differs sharply too between the two sources of

rainfall data in ratio of 1 to 25. There are few studies with which to compare the

present one regarding flow modelling with 3b42 data. One has been conducted in the

La Plata basin in South America, (Su et al., 2008), the timestep is daily and the

rainfall runoff model is the Variable Infiltration Capacity (VIC) semidistributed

hydrology model. The size of the basins observed ranges between : 62 k km2 and

1,100 k km2 , the NSE with model forced with 3b42 ranges between <0 and 0.71 for

daily flows and <0 and 0.8 for monthly flow. The weighted average are respectively

0.71 and 0.8. But there too, the peak flows are overestimated by the 3b42 product.

90
Results and discussion

4.6.3 Graphie presentation

FL1 - 20/12/2002 - 9/01/2003


50

45

40

35

30

..... 25
'"E
- RG Modelled Flow
20 - Gauged Flow
., -= n· "J'..,

15

10

Figure 23 : FLI rain gauges modelled

FL1 - 31/12/2002 ta 5/01/2003


250 ,-------------------------------------------------------------------

200 +-------------------------------------------------~----------------

150 +--------1~--------------------------------------~r_----_.--------

.....

- 3b42 Modelled Flow
100 +--------+-r--------------------------------------+-+-----~r_------
- Gauged Flow

50 +-------~_1,_++-------- . .----------------------+_--_+~~~~------

Figure 24 : FLI 3b42 modelled

91
Results and discussion

FL 2 - 29/12/2006 to 28/01/2007
50.0

45 .0

40.0

35.0

30.0

.... 25 .0
ë
- RG Modelled Flow
20.0 - GaugedFlow

15.0

10.0

5.0

Figure 25 : FL2 rain gauges modelled

FL 2 - 31/12/2006 to 22/01/2007
35.0

30.0

25.0

20.0

~
15 .0 - 3b42 Modelled Flow
- Gauged Flow

10.0

5.0

_....- ~ ... f'o.- • ...,;

0.0

()($~() ()çs~() ()çs~() ()($~() ()çs~() ()().~ ()çs~()


çf)ro ()()1 ()()1 ()()1 ()()1 ()()1 ()()1

\'\.'\.\'): \()'\.\'): \()'\.\'): \()'\.\'): \()'\.\'): \()'\.\'): \()'\.\'):


'\."'-' ()"? ()'b '\."? '\.'b '\."? '\.'b

Figure 26 : FL2 3b42 modelled

92
Results and discussion

Despite the high number of iterations allowed for the optimisation of the model, the

recession limb of the modelled flow appears faster than the "real" flow. A review of

ReFH methodology (Faulkner and Barber, 2009) notes that the model tends to

overshoot the peak flow and sharpen the recession. The present study notes that

when the model is calibrated using the 3b42 FLI event (the largest flow in the data

set) a slightly smaller flow (FL2) is predicted at a much lower peak value. It is the

only instance ofunderestimation of the peak flow.

4.7 Comparison of 10 modelling'results

4.7.1 Introduction

As presented in the methodology, the ISIS free software ver 3.3 and ver 3.4 were

used to model the flow depth and stage in a 5 km reach downstream for the gauging

station located at St-Marie-sur-Semois. The package uses the St Venant equations

(conservation of mass and momentum) to solve the relation between flow and water

levels. Three hydrographs are available for each flood periods studied = one

originating from the river gauging station at a 1 hour timestep, one from the flow

modelled with rain gauges at a Ihour timestep, one from the flow modelled with

3b42 data at a 3 hours timestep. These 6 flow streams generated 6 different runs for

the 1D model. One hundred three cross sections (30 m on each side of the central

line) based on a LIDAR DEM were used to compute the flow-stage data.

4.7.2 Maximum stage and overflow depth

The overflow depth is the difference between the flow stage and the elevation of the

lowest bank. It is computed from the maximum stage and analysed on the 103 cross-

sections. The results are presented in Table 23 : Maximum stages and overflow depth

93
Results and discussion

Table 23 : Maximum stages and overflow depth

Ratio
Ratio Rain 3b42
Rain Gauge Mod
Gauged Gauge 3B42 Mod/Gauged /Gauged
Period Descriptor Flow Mod Mod Flow Flow
FLI Average max stage (masl) 329.95 330.05 331.98 1.00 1.01
Average overflow depth (m) 0.26 0.32 2.12 1.24 8.19
Maximum overflow depth (m) 0.92 1.02 3.34 1.11 3.65
FL2 Average max stage (masl) 329.80 330.05 329.47 1.00 1.00
Average overflow depth (m) 0.17 0.32 0.04 1.89 0.26
Maximum overflow depth (m) 0.76 1.02 0.43 1.34 0.57

The divergence between the two modelled flows and the two events is substantial.

The errors between the rain gauge modelled flow and gauged flow are of higher

proportion than the error between the maximum overflow depth (14% vs Il % for

FL1, 44% vs 34 % for FL2). For 3B42 based maximum overflow depth too, the error

is of a lower proportion. That convergence of the two results is related to the

approximations in the resolution of the flow-depth equations. A larger sample may

be required to confmn this trend. The Figure 27, The maximum overflow depth for

4b42 is a misrepresentation of the reality because the cross-section were too narrow

to allow the spread of a such a large volume of water. The ISIS model builds "glass

walls" at the end of the cross-section to allow the continuation of the simulation.

94
Results and discussion

- .. _- .. _- ------- ... _-- -----------

-_::==:::::::-----

B
~I

nll8
LOO-œx
9Mrœx
mœx
ttQ-œx
oo-œx
~Mrœx
,oo-œx
OMl-œx
eœ-œx
IICO-œx
Lto-œx
Qto-œx
ga)-œx
HXl-œx
carœx
ltOœx
LOOCZX
œo-œx
6lOœx
----iœOœx
- - - - 1 a»œx
- - - - 1 œOœx
- - - - 1 sro-œx
- - - - 1 tm-œx
- - - - I t r o-œx
- - - - I I 2 O-œx
- - - - - 1 ,..-œx
- - - - - - 1 ozo-œx
----~ 6'O-œx
- - - - - i 9 . a-œx
- - - - - 1 LUrœx
' - - - - - - 1 9urœx
' - - - - - - - i s.a-œx
- - - - - - 1 .,o-œx
' - - - - - - - 1 uoœx
- - - - - - 1 ~.a-œx
- - - - - - 1 HO-œx
' - - - - - - - - 1 o,oœx
-------i~-œx
' - - - - - - - - I m-œx
------~~-œx
- - - - - - - - - I m-œx
-------~~-œx
-------~ ~-œx
---------I~-œx
1
--------I~-œx

~ii~~~~~à~aaaà~~~~~~~aa~~aaa~~~a~~~~~~~~~
(a.w)~_
- - - - - - - - 1 UIO-œx
-------~~-œx
1
Figure 27 : FLI Maximum stages Rain Gauged Modelled & Gauged Flow

95
.......... (mNl)

~~~~~~~~~~~~~~~~~~~~~g~~~~~~~~~~~~~~~~~~~~
x20_000 i . 1- . f L __. _- ":!.",= ___ .. !if- -1 •
x20_001
x20_002
x20JlO3
x20Jl04
x20_005
x'!UlO6
.2lU»7
x20_00s
x20JlOO
x20_010
x20_011
x20_012
x20_013
'1.20_01.
~ x2D_OUii
g x2O_016
](2CL011

! x20_018
><20_0'9

~ '1.20_020
x20_ 021

J x20_022
x20J123
,~ ><20_024

~
.20_025
.20_02&
,~ x20_CYO
x20_028
8 .20_029
i x20_030

~
x20_031
'1.20_032

,~ x20_033
x20_034
21 x20J135
~ x20_1Xl6

~ x20_031
x20_038
x20_039
x20_040
x20_041
'1.20_042
j
x20_043 - ~. _. 1
- - . ~
~ x20_044 g
g x20_ 045
x20Jl'8
,13
i;; .20_047
§
.~
~
i := : â
g
~
1 x203~1 i
~ ><20_052
~
',0!
x2fU)S3
x20_054
x20_055
j
~
x20_056 ;,
8 x2D_051

! x20Jl!i8
'1.20_058
~
f
;..
x20J11!O
'1.2(»-,)&1
x20_082
~ x20_083

~
x20Jl.'
x20JI65

~
x20_066
x20J)lJ7
x20JI68
x2(Ua
x20_070
x20_071
x20_0n
x20_073
x20_074
x20_015
x20_076

'~
x'2J)_On
x20_078
x20_079

.; x20_000
)(20_051
'1.20_082
,~ x20_083

,0~
x20_""'
x20_0115
x20_086
x2CU)87
8 x20_0Il8
x20_089
x20_090
x20_ 091
x20_092
x20_093
x20_094
'1.20_086
x20_096
x2(U»l
x20_098
x20_098
x20_ 100

-_Z==:::~:_::::r:::::.:~_ .
• 20_'01
x20_'02

MOI~ p\l~",~~ ~ A\'0l~ P\lIl\lPOIN Ztqr : S\l~lqS WnW!XHIN 11~ : 8Z \lJn~!~

uo!ssn~s!p pUB sHns~~


L6

MOI~ pa~m~~ ~ pallapow sa~m~~ U!BH - sa~Bls WnW!XBW - Z'~ : 6Z a.m~!~


~._- ------ ...... - - ------.~--

EIIrt.tbn(mAD)

~~~~~!î!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
x2!UlO1
I<2OJ'02
1<20_003
1<20_004
f 1<20_000

~
I<2OJ'OO
x2(UlO7

f~ 1<20_008
1<20_008
x20_0tO
ii x20_011

t x2O_012
x20_013
~ x20_014
K20_015
lt2O_018
x2O_011
x20_0t8
x20_019 .
I<2OJ'20
.20_021
1<20_022

r
~
1<20_023
I<2OJ'2.
1<20_025 1------1(-J -
x2OJ)26
x2O_027
1<20_028
1<20_029
1<20_030
dO_Olt
1<20_032
1<20_033
1<20_034
1<20_035 ---1.
I<2OJ>36
x2(U)37
1<20_038
f 1<20_008
~ .20_040

f x2(UJ41
x20_042 i
~ 1<20_043 -

~
x2(U)44
1<20_045 1
1 1<20_046 ,~
x203M7 §
ii 1<20_"""
~ 1<20_050

~
" x2CU)51
li .20_052
1<20_053
1<20_054
1<20_055
1<20_056
1<20_057
x20_058
x2O_059
x2(Ul60
x2fUl61
1<20_002
1<20_063
x2O_064
I<2OJ'65
1<20_066
x2fU)67
.
r._ .~_, ~ '.

1<20_066
x2(Ul6Q
1<20_070
x20_071
x20_0n
I<2OJ173
x20_014
x20_075
x20_076
x2fJ_071
x2D_078
x2(L079

:=: t1
1<20_062
x2(Ul83
I<2OJ'64
x20_065
1<20_066
x20J187

~ .20_088
x20J189
J x20_090
1<20_09'
1<20_092
x2tU)93
1<20_094
x20_095 ~

1<20_096
x20JJ97
1<20_098
I<2OJ>98
x2O_1oo
.:20_101
1<20_'02

uo!ssn~s!p plffi SnnS;;)(l


Results and discussion

zn.-oz><
lO~-OZX
ooa.-œx
~-œx

8OO-oz><
Leo-œx
8OO-œx
...-œx
...-œx
tm- Ill"
ZIlO-œx
Ulnex
oeo-œx
680-œx
ooo-œx
lWrœx
9OO-œx
... ...
...-ove
-
aiO- OZX
ZIlO-œx
LanJlX
080-'"
"",- ...
oro-...
UIrove
"",-ove
çro-œx
tro-œx
CiO- œ x
uo-œx
uo-ozx
O1D-ove
680-ove
lI9Il-œx
LSO- œx
1I9Il-'"
... ...
...,-œx
-
tlIO-œx
ZIlO-œx
LW- OZ:X
OIlO-œx
ISrœx
B!D-œx
LStrœx
!lOO-ove
çço-œx
..a-oz><
tÇO- OlX
zw-ove
19J- œx ~
ooo-œx
8M)- œ x
1
ltoO- OZX
m -ove
...,-ove
...,-œx
oo-œx
zlO-œx
'IO-œx
DIO- Ill"
6U)-œx
em-...
Lm- œx
900-'"
çm-œx
tœ-ove
tm- ...
lm- OZ x
.an,,"
oro-ove
OZO- OZX
BlIl-'"
lzn-...
• "fOZX
SlI)-OZX
tZXnJP
C~Xn2x
zzn-...
,m-œx
OlO-œx
OLO- OZX
8LO-OZX
lLO- œx
•• O- OZX
9lO- œx
••O- OZX
cm-œx
lLO- œx
uo-œx
OLO- œx
/lOO- ...
&(X)- ozx
LOO- œx
oro-ozx
ço:fœx
IOO- OZX
too- ...
zno-ove
u:xnJZx
ooo-ozx

Figure 30 : FL2 Maximum stages 3b42 modelled & Gauged Flow

98
Results and discussion

4.8 Comparison of the ID-2D modelling results

As ISIS is OpenID compliant package, its 1D and 2D mode1s are able to exchange

input-output during a synchronous run. The area modelled in 2 D had to be reduced

at a rectangle of81,000 m 2 due the 2500 cells limitation of the free version. The

village of Villers-sur-Semois is incIuded in the area as it is the critical asset in the

modelled reach. The area of the maximum flood extent are presented in Table 24

Table 24 : Flood maximum extent near Villers-sur-Semois (m 2)

3b42/
Gauged Rain gauge 3b42 Rain Gauge/ gauged
Period Flow modelled modelled GaugedFlow flow

FL1 40,003 31,752 53,900 0.79 1.35

FL2 36,946 36,946 18,473 1.00 0.50

The 3b42 case may be misrepresented as the extent nearly reached the Northern limit

of the available computing area. But it is a fact that the local topography and the

roads minimize the expansion of the flood plain developed by the 3b42 modelled

flow, the end result is a flood plain much doser to the other sources of rainfall.·

The Figure 31 and Figure 32 present the delineation of the flood extent on Google

Earth snapshot.

Animations and interactive map are posted online at :

http://web.me.com/albert.grela/3b42_flood~lainlWelcome.htm1

99
Results and discussion

-,- ".- .~ .. c::=] Max Flood Extent FI1 - 3b42 Modelled Flow
Max Flood Extent FI1 - Raln gauge<! modellad flow
[.=J Max Flood Extent FI1 - Gauge<! Flow
D 20 modelled area

Figure 31 : FLI flood extent

C:] Max Flood Extent FL2 - Gauge<! Flow


Max Flood Extent FL2 - 3b42 Modelled Flow
C=] Max Flood Extent FL2 - Raln Gauge ModeUe<! Flow
D 20 modelled aree

Figure 32 : FL2 flood extent

The Regional Authorities (Region Wallonne) have prepared an official mapping of

fluvial flood risk(Region Wallonne, 2010) which is compared with the present

modelling of the flood extent. The official definition of the high risk is :

100
Results and discussion

" an area with a flood return period equal or less than 25 years and a submersion of

at least 30 cm".

For the two flood events of the present study, the flood extent presented in Table 24

is re-delineated to correspond to a submersion by 30 cm ofwater. The results of

these extent with a depth >= 30 cm is plotted on the background of the official flood

risk map in Figure 33 and Figure 34

. '.1
21/12/02 to 10/01/03
FLl max sup 0.3 m depth : 31>42 mod
FL1 max sup 0.3 m depth : Rain Gauge mod
FU max sup 0.3 m depth : Gauged Aow mod

Figure 33 : FLI Flood extent with a depth of at least 30 cm and official flood risk map

101
Results and discussion

Figure 34 : FL2 Flood extent with a depth of at least 30 cm and official flood risk map

The match is rather good for FL 1 and a bit less for FL2, the two events are in 10
years return period range but FL 1 is the second largest discharge in a 28 years series
while FL2 is the 5th •

The Table 25 provides the ratio between the area classified as high risk and the
extent of the 30 cm depth during the two flood events: The convergence of the two
assessments (probabilistic from the Region Wallonne,event based for the present
study) converge remarkably weIl for the Gauged Flow and Rain Gauge Modelled.
The 3b42 based modelled exhibits a sharp divergence which is quite likely
underestimated for the reason expressed in para 4.7.2.

Table 25 : Ratio of FLI et FL2 flood extent with a depth superior to 30 cm and area c1assified as high risk

Reg Wal- High Risk area 24,728 m2


Gauged Raingauge 3b42
Flood extent Flow mode lied modelled
FLl>O.3 / RW HR 0.98 1.04 1.78
FL2>0.3 / RW HR 0.76 1.03 0.10
-'-.: ~ .--,,-

102
Conclusions

5 Conclusions

5.1 Flood plain modelling

The small size of the flood plain limits the extent of the conclusions but the fact that

it was possible to model it given the limited resources in expertise and software

license is a rather reassuring fact. The match with the official flood risk map (Region

Wallonne, 2010) is also a positive element to recommend this approach for assessing

risk with small chunk of area at a time. As ISIS Professional could be rented on

monthly basis, a large variety of cost options could be considered to prepare a flood

map. The annual estimated cost of flood damages reaching 30 billions USD

worldwide (Brakenridge, 2010) clearly suggests that savings on flood modelling

software may not be the most economically efficient. This also advocates that flood

modelling should gain sorne prominence in courses leading to "Environmental

modelling" graduation. Flood limits markers should be promoted and made available

online to maximise the possibility of model validation! verification.

5.2 ReFH adequacy outside UK

The model and optimization algorithm seem to perform weIl outside UK. The

hydrological properties of the European Soil Database (1RC, 2010), the Corine Land

Cover 2000 database (Agency, 2000) provide good correlations to propagate the data

collected in UK. A European rainfall-runoff model may have sorne relevance to

foster transboundaries collaboration. The experienced gained with the ReFH method

is probably worth considering but basin wide methods are probably more likely to

appeal to the pragmatic planner than national standards levelling the actual diversity

of hydrological conditions. The efforts made to provide rationales to the estimations

103
Conclusions

of discharge in ungauged catchments are a disincentive to establish instruments and

protocols to collect actual observations for specific catchments.

5.3 3b42 validity under the 50 ° Lat N

Without any doubt, this is the most disappointing part of the present study. The 3b42

data are clearly inappropriate for winter flood modelling close the 50° Lat N. Many

factors concur in this inadequacy : the precipitation radar does not coyer beyond the

38°, the ground validation station are aU in lower latitude and the ground elevation

does not seem perfectly included in the niin retrieval algorithme The low level of rain

event detection and the monthly adjustment based on rain-gauge tends to

overestimate the rain event reported by 3b42. This distorts the simulated hydrograph

very sharply (cfr Figure 24 ).

This was somehow to be expected as the literature review did not identify any

validation exercise at such latitude. Nevertheless the tRMM mission seems a

success confirmed by its extension and by the preparation of a constellation of

satellites based on a dual frequency rain radar: the Global Precipitation

Measurement Mission (GPM) is expected to provide rain estimation every 2 to 4

hours on a global coverage. The launch is schedule for July 2013.(NASA, 2010a)

5.4 LIDAR data processing

The LIDAR data are getting sorne popularity because the high spatial resolution and

their average elevation accuracy. The present study observed that the last pulse is far

to be always the soillevel. Trees along the river have often masked the river

bed/pound level. The time was too short to investigate the physical causes of this

error and the supplier of the data (Region Wallonne) had no explanation. The
-.~.":_ -.~ -

104
Conclusions

interpolation of levels between consistent ones was long and tedious but the small

software RasterEdit permitted a correction leading to a consistent generation of

cross-sections. This program is quite unique and highly relevant in the context of

LIDAR data set. It disserves sorne further development.

5.5 Data processing capa city

The rainfall data at 1 hour timestep were a considerable load (266,640 records) but

MS Excel 2007 absorbed it with a few limitations : automatic calculation had to be

disabled and logical tests had to be based 'o n numerical data only.

ISIS 1D performed the modelling of the 613 hours of FL2 at 10 sec timestep in a

very reasonable time : 235 seconds. This was performed on a computer with Intel

Quad Q9000 @ 2.00 GHz CPU with 4GB memory.

The 1D2D modelling of the same FL2 event took 24.5 minutes on a computer with

Intel Core Duo Mobile T2500 2.Ghz CPU with 2 GB memory.

The calibration for ReFH package took 12 minutes, running the model with the FL2

rain took 4 minutes, on the Quad computer described above.

These were reasonable processing time.

A Python script using the IDW object of ArcGIS 9.3 would have taken 9 days to

compute the whole period of rain data for the ID W of the Semois catchment.

105
Conclusions

5.6 Propagation of error and recommendation for local

observations

The 3b42 data do not provide any usable flood prediction tools at the location of the

present study. The rainfall-runoffmodel ReFH forced with rain gauges data, allows

sorne useful evaluation of the extent of a flood plain but many of the discharge

impact such as the average overflow depth are estimated with a bias of 56%. (see

Table 23 : Maximum stages and overflow depth). This may distort quite significantly

the design of flood defence structures. Th~ cost of flood damages and flood defences

should warrant a significant investment in hydrological data collection in all the

catchments where a flood mitigation investment is planned. The uncertainty of the

rainfall-runoffrelation particularly in ungauged catchments (Beven, 2009, Beven,

2006, Beven, 2004, Beven, 1996, Beven, 1993, Beven and Binley, 1992, Degré et

al., 2008, Faulkner and Barber, 2009) turns the infrastruc~e investment into a

casino's bet if carried out without consistent observation of the actual runoff.

The various attempt to recalibrate the 3b42 data ((Fotopoulos et al., 2010) or to use a

regional algorithm for satellite rainfall (STISEN et al., 2010) have been successful at

reducing the error and are like1y to be emulated in the near future.

5.7 Further work

The 1D2D mode1 of the present study could be used as a benchmark for a variety of

rainfall products such as CMorph, PERSIANN, GSMapMVK+. Recalibration of

3b42 data could be attempted although for the winter period the POD appears very

weak and is likely to distort the hydrograph beyond utility. Combination of satellite

borne data with ground radar have a very potential prospect as described by

106
Conclusions

Chandrasekar (Chandrasekar et al., 2008). Such attempt is possible at the same

location as it is covered by the rain radar installed at Wideumont (Kunstmaan, 2010).

More attention should be dedicated to the mass balance analysis for the ID2D model

and to the equifinaIity of the rainfall-runoffmodel.

107
Bibliography

6 Bibliography

ADLER, R. F., NEGRI, A. J. & HAKKARINEN, 1. M. 1991. Rain estimation from


combining geosychronous IR and low-orbit microwave data. Global and
Planetary Change, 4,87-92.
AGENCY, E. E. 2000. Corine Land Cover 2000 raster data - version 13 (02/2010) -
Datasets - EEA. :.
AL-SABHAN, W., MULLIGAN, M. & BLACKBURN, G. A. 2003. A real-time
hydrological model for flood prediction using GIS and the WWW.
Computers, Environment and Urban Systems, 27,9-32.
ALCRUDO, F. 2004. Mathematical modeIing techniques for flood propagation in
urban areas. IMPACT PROJECT technical report.
ALISHOUSE, J. 1983. Total precipitable water and rainfall determinations from the
SEASAT Scanning Multichannel Microwave Radiometer. J Geophys. Res,
88, 1929-1235.
ALLISON, L., RODGERS, E., WILHEIT, T. & FETT, R. 1974. Tropical cyclone
rainfall as measured by the Nimbus electrically scanning microwave
radiometer. Bulletin of the American Meteorological Society, 55, 1074-1089.
ANDERSON, M., BATES, P. & BATES, D. 2001. Model validation: perspectives in
hydrological science., Wiley Chichester.
ARKIN, P. & MEISNER, B. 1987. The relationship between large-scale convective
rainfall and cold cloud over the western hemisphere during 1982-84. Monthly
Weather Review, 115,51-74.
ARONICA, G., HANKIN, B. & BEVEN, K. 1998. Uncertainty and equifinality in
calibrating distributed roughness coefficients in a flood propagation model
with limited data. Advances in Water Resources, 22,349-365.
BARRETT, E. & BEAUMONT, M. 1994. Satellite rainfall monitoring: An
overview. Remote Sensing Reviews, Il,23-48.
BARRETT, E. & MARTIN, D. 1981. The use ofsatellite data in rainfall monitoring,
Academic Pr.
BARRETT, E. C. 1997. Satellite rainfall monitoring: recent progress and remaining
problems. IAHS PUBLICATION, 141-148.
BASTIN, G., MOENS, L. & DIERICKX, P. 2009. On-line river flow forecasting
with Hydromax: successes and challenges after twelve years of experience.
.- :"'----; ~~
15th IFAC Symposium on System Identification. Saint-Malo, France, July 6--
. ' ."
8,2009 .
BATES, P. D., LANE, S. N. & FERGUSON, R. 1. 2005. Computational jluid
dynamics: applications in environmental hydraulics, J. Wiley.
BECK, M. B. 1987. Water quality modeling: a review of the analysis of uncertainty.
Water Resources Research, 23, 1393-1442.
BELLERBY, T. 2010. Lecture 3 : Hydrometry. Module 16409 - GIS & Remote
Sensing. Hull; Departement of Geography - University of Hull.

108
Bibliography

BELLERBY, T. & SUN, J. 2005. Probabilistic and ensemble representations of the


uncertainty in an IRJmicrowave satellite precipitation product. Journal of
Hydrometeorology, 6, 1032-1044.
BELLERBY, T., TABERNER, M., WILMSHURST, A., BEAUMONT, M.,
BARRETT, E., SCOTT, J. & DURBIN, C. 1998. Retrieval of land and sea
brightness temperatures from mixedcoastal pixels in passive microwave data.
IEEE Transactions on Geoscience and Remote Sensing, 36, 1844-1851.
BETSON, R. 1964. What is watershed runoff. Journal ofGeophysical research, 69,
1541-1552.
BEVEN, K. 1989. Changing ideas in hydrology--The case ofphysically-based
models. Journal o/hydrology, 105, 157-172.
BEVEN, K. 1993. Prophecy, reality and uncertainty in distrlbuted hydrological
modelling. Advances in Water Resources, 16,41-51.
BEVEN, K. Year. Equifinality and Uncertainty in Geomorphological Modelling. In,
1996. John Wiley & Sons, 289-289.
BEVEN, K. 2004. Rainfall-runoffmodelling: the primer, John Wiley & Sons !nc.
BEVEN, K. 2006. A manifesto for the equifinality thesis. Journal ofHydrology,
320, 18-36.
BEVEN, K. & BINLEY, A. 1992. The future of distributed models: model
calibration and uncertainty prediction. Hydrological Processes, 6, 279-298.
BEVEN, K. & FREER, J. 2001. Equifinality, data assimilation, and uncertainty
estimation in mechanistic modelling of complex environmental systems
using the GLUE methodology. Journal ofHydrology, 249, 11-29.
BEVEN, K., SHAW, E., CHAPPELL, N. & LAMB, R. 2010. Hydrology in practice,
London, Taylor & Francis.
BEVEN, K. J. 2009. Environmental modelling: an uncertainfuture? : an
introduction to techniques for uncertainty estimation in environmental
prediction, London, Routledge.
BOX, G. E. P. 1979. Robustness in the strategy ofscientific model building,
WISCONSIN UNIV -MADISON MATHEMATICS RESEARCH CENTER.
BRADBROOK, K. F., LANE, S. N., WALLER, S. G. & BATES, P. D. 2004. Two
dimensional diffusion wave modelling of flood inundation using a simplified
channel representation. International Journal ofRiver Basin Management, 2,
211-223.
BRAKENRIDGE. 2010. Dartmouth Flood Observatory [Online]. Available:
http://floodobservatory.colorado.eduJ [Accessed].
BRAZIER, R. E., BEVEN, K. J., FREER, J. & ROWAN, J. S. 2000. Equifinality
and uncertainty in physically based soil erosion models: application of the
GLUE methodology to WEPP-the Water Erosion Prediction Project-for sites
in the UK and USA. Earth Surface Pro cesses and Landforms, 25, 825-845.
BURROUGHS, J. 2008. GHCN-Monthly - Data Coverage [Online]. Asheville Ne:
The National Environmental Satellite, Data, and Information Service
(NESDIS). Available: http://www.ncdc.noaa.gov/oa/climate/ghcn-
monthly/index.php [Accessed 18/05 2010}.
CAPPUS, P. 1960. Bassin expérimental d'Alrance. Etude des lois de l'écoulement.
Application au calcul et à la prévision des débits. La Houille Blanche, 493-
520.
CARTWRIGHT, N. 1983. How the laws ofphysics lie, Oxford University Press,
USA.

109
Bibliography

CASTREE, N., ROGERS, A. & SHERMAN, D. J. 2005. Questioning geography :


fundamental debates, Malden, Mass., Blackwell.
CHANDRASEKAR, V., BRINOI, V. N., RUTLEDOE, S. A., HOU, A., SMITH, E.,
JACKSON, G. S., GORGUCCI, E. & PETERSEN, W. A. 2008. Potential
Role Of Dual- Polarization Radar In The Validation Of Satellite Precipitation
Measurements: Rationale and Opportunities. Bulletin ofthe American
Meteorological Society, 89, 1127-1145.
CHOKNGAMWONG, R. & CHIU, L. S. 2008. Thailand Daily Rainfall and
Comparison with TRMM Products. Journal ofHydrometeorology, 9, 256-
266.
CLARKE, R. T. 1973. A review of some mathematical models used in hydrology,
with observations on their calibration and use. Journal ofHydrology, 19, 1-
20.
COLLISCHONN, B., COLLISCHONN, W. & TUCCI, C. 2008a. Daily hydrological
modeling in the Amazon basin using TRMM rainfall estimates. Journal of
Hydrology, 360,207-216.
COLLISCHONN, B., COLLISCHONN, W. & TUCCI, C. E. M. 2008b. Daily
hydrological modeling in the Amazon basin using TRMM rainfall estimates.
Journal ofHydrology, 360,207-216.
DEGRÉ, A., DAUTREBANDE, S., SOHIER, C. & DEBAUCHE, O. 2008.
Statistique des extrèmes dans les bassins faiblement jaugés: application diun
modèle global pluie-débit dans cinq bassins versants en région wallonne
(Belgique). Biotechnol. Agron. Soc. Environ, 12,379-391.
DH!. 2010. MIKE FLOOD - sprinting ahead ofthe competition [Online]. DH!.
Available:
http://dhigroup.com/News/2010/07/09/MIKEFLOODSprintingAheadOITheC
ompetition.aspx [Accessed 4/09/2010 2010].
DIERICKX, P. 31/08/20102010. RE: Specification ofSt Marie-sur-Semois stream
gauge. Type to GRELA, A.
DUNNE, T. & BLACK, R. 1970. An experimental investigation of runoff
production in permeable soils. Water Resources Research 6,478-490.
ENVIRONMENT AGENCY UK. 2010. HiFlows-UK [OnIine]. Available:
http://www.environment-agency.gov.uk/hiflows/91727 .aspx [Accessed 08/09
2010].
ERNST, J., CONINX, 1., DEWALS, B., DETREMBLEUR, S., ERPICUM, S.,
BACHUS, K. & PIROTTON, M. 2009. Social Flood Impacts in Urban
Areas. Integration of DetaHed Flow Modelling and Social Analysis. IAHR
2009 Congress - Water Engineeringfor a sustainable environment
edition:33. Vancouver date:9 - 14 August 2009: IAHR.
ESRI 2009. ARCGIS Hydro Data Model. 1.3 ed. Redlands, Calif: Environmental
Systems Researeh Institute, Ine.
EVANS, E. P., WICKS, J. M., WHITLOW, C. D. & RAMSBOTTOM, D. M. 2007.
The evolution of a river modelling system. Water Management, 160, 3-13.
FAULKNER, D. & BARBER, S. 2009. Performance of the Revitalised Flood
Hydrograph method. Journal ofFlood Risk Management, 2,254-261.
FEIDAS, H. 2010. Validation of satellite rainfall products over Greece. Theoretical
and Applied Climatology, 99, 193-216.

110
Bibliography

FOTOPOULOS, F., MAKROPOULOS, C. & MIMIKOU, M. A. 2010. Flood


forecasting in transboundary catchments using the Open Modeling Interface.
Environmental Modelling & Software, In Press, Corrected Proof.
FOUCAULT, M. 1976. Histoire de la sexualite, Paris, Gallimard.
FRANKS, S. W., BEVEN, K. J., QUINN, P. F. & WRIGHT, 1. R. 1997. On the
sensitivity of soil-vegetation-atmosphere transfer (SVAT) schemes:
equifinality and the problem of robust calibration. Agricultural and Forest
Meteorology, 86,63-75.
FREE SOFTWARE DlRECTORY. 2006. ABScan - Free Software Directory - Free
Software Foundation [Online]. Archive: http://www.thinknew.ca/analytics/.
Available: http://directory.fsf.org/projectlabscan/ [Accessed 08/09 2010J.
GRIESER, J., GOMMES, R. & BERNARDI, M. Year. New LocClim-the local
climate estimator ofFAO. In, 2006.
GYAWALI, D. 2001. Water in Nepal, Kathmandu, Himal Books.
HAACK, S. 1994. Evidence and inquiry: ,Towards reconstruction in epistemology,
Wiley-Blackwell.
HAACK, S. 2002. Defending Science within reason. Journal ofXiangtan Normal
University (Socim Science Edition).
HAACK, S. 2003. Truth,- Truths," Truth,-" and" Truths" in the Law. Harvard Journal
ofLaw & Public Policy, 26, 17-22.
HALCROW. 2010a. ISIS Professional overview [Online]. London: Halcrow.
A vailable: http://www.halcrow.comlisis/isisprofessional.asp [Accessed
14/03/2010 2010].
HALCROW. 2010b. ISIS Professional User Manual. [Accessed 11/09/2010].
HARRISON, D. L., SCOVELL, R. W. & KITCHEN, M. 2009. High-resolution
precipitation estimates for hydrological uses. Proceedings of the Institution of
Civil Engineers-Water Management, 162, 125":135.
HEDGE, M. 2010. Data Groupsfor Data Group Satèllite_CSI[Online]. NASA.
Available: http://mirador.gsfc.nasa.gov/cgi-
binimirador/presentNavigation.pl?tree=project&project=TRMM&dataGroup
=Subset&dataGroup=Satellite_ CSI&CGISESSID=a2c406da8e6b803bd6a02
39771755b5b [Accessed 24/082010].
HERVOUET, J. M. 2007. Hydrodynamics offree surfaceflows: modellingwith the
finite element method, John Wiley & Sons Inc.
HEWLETT, J. 1961. Soil moisture as a source ofbase flow from steep mountain
watersheds. Station Paper, 132.
HILDEBRAND, P., TOWERY, N. & SNELL, M. 1979. Measurement of convective
mean rainfall over small areas using high-density raingages and radar.
Journal ofApplied Meteorology, 18, 1316-1326.
HORNBERGER, G. M. & SPEAR, R. C. 1981. An approach to the preliminary
analysis of environmental systems. J, 7-18.
HORRITT, M. & BATES, P. 2002. Evaluation of ID and 2D numerical models for
predicting river flood inundation. Journal ofHydrology, 268, 87-99.
HORSFIELD, N. 2006. Development ofa Mass Balance Approach to Modelling
Cloud Lifecycles and Rainfall Using Satellite Observations. Doctor of
Philosophy, UNIVERSITY OF HULL.
HORTON, R. 1933. The role of infiltration in the hydrologic cycle. Trans. Am.
Geophys. Union, 14, 446-460.

111
Bibliography

HOSSAIN, #160, FAISAL, ANAGNOSTOU & N., E. 2004. Assessment of current


passive-microwave- and infrared-based satellite rainfall remote sensingfor
flood prediction, Washington, DC, ETATS-UNIS, American Geophysical
Union.
HOSSAIN, F., KATIYAR, N., HONG, Y. & WOLF, A. 2007. The emerging role of
satellite rainfall data in improving the hydro-poIitical situation of flood
monitoring in the under-developed regions of the world. Natural Hazards,
43, 199-210.
HUFFMAN, G., ADLER, R., BOLVIN, D., GU, G., NELKIN, E., BOWMAN, K.,
HONG, Y., STOCKER, E. & WOLFF, D. 2007. The TRMM multisatellite
precipitation analysis (TMPA): quasi-global, multiyear, combined-sensor
precipitation estimates at fme scales. Journal ofHydrometeorology, 8,38-55.
HUNTER, N. M., BATES, P. D., HORRITT, M. S. & WILSON, M. D. 2007.
Simple spatially-distributed models for predicting flood inundation: A
review. Geomorphology, 90,208-Z25.
HURSH, C. 1936. Storm water and absorption. Eos Trans. AGU, 17,301-302.
IFRCRCS. 2010. Red Cross Red Crescent - World Disasters Report 2009 [Online].
A vailable: http://www.ifrc.org/publicat/wdr2009/summaries.asp [Accessed].
JARVIS, A., H.L REUTER, NELSON, A. & GUEVARA, E. 2008. HoZe-filled
seamless SR TM data V4 [Online]. International Centre for Tropical
Agriculture (CIAT). [Accessed 12/01/20102010].
JAXA. 2010. Features ofASTER G-DEM[Online]. Jaxa. Available:
http://www.ersdac.or.jp/GDEMIE/2.html [Accessed 09/09/ 2010].
JOYCE, R., JANOWIAK, J., ARKIN, P. & XIE, P. 2004. CMORPH: A method that
produces global precipitation estimates from passive microwave and infrared
data at high spatial and temporal resolution. Journal ofHydrometeorology, 5,
487-503.
JRC. 2010. Google Earth KMZfilesfor European Soil Database [Online]. European
Commission - Joint Research Centre Institute for Environment and
Sustainability. Available:
http://eusoils.jrc.ec.europa.eu/ESDB_Archive/ESDBv3/googleearth/index.cf
m [Accessed 09/09/ 2010].
KEMPLER, S. J. 2010. Giovanni - TRMM Online Visualization and Analysis System
(TOVAS) [Online]. NASA. Available: http://gdatal.sci.gsfc.nasa.gov/daac-
bin/G3/guLcgi?instance_id=TRMM_3-Hourly [Accessed 24/08/20102010].
KIDD, C. 2001. Satellite rainfall climatology: a review. International Journal of
Climatology, 21, 1041-1066.
KIDD, C., KNIVETON, D., TODD, M. & BELLERBY, T. 2003. Satellite rainfall
estimation using combined passive microwave and infrared algorithms.
Journal ofHydrometeorology, 4, 1088-1104.
KJELDSEN, T., STEWART, E., PACKMAN, J., FOLWELL, S. & BAYLISS, A.
2006. Revitalisation of the FSRlFEH rainfall-runoff method. Technical
Report FD1913/TR, Department of the Environment, Food and Rural Affairs.
KJELDSEN, T. R. 2007. The revitalised FSRlFEH rainfall-runoff method. In:
HYDROLOGY, C. F. E. (ed.) Flood Estimation Handbook. Wallingford:
Centre for Ecology & Hydrology.
KJELDSEN, T. R. & JONES, D. A. 2009. A formaI statistical model for pooled
analysis of extreme floods. Hydrology Research, 40, 465-480.

112
Bibliography

KUNSTMAAN. 2010. Radar - IRM[Online]. Brussels: IRM Belgique. Available:


http://www.meteo.be/meteo/view/fr/123 884-Radar.html [Accessed 12/09/
2010].
KUSZMAUL, J. S., JOHNSON, E. G., EASSON, G., HOSSAIN, F. & ROBERT, G.
2008. Integration of NASA Global Precipitation Measurement Mission Data
into the SERVIR Flood Decision Support System for Mesoamerica.
LAI, Y. G. 2005. River and Watershed Modeling: Current Effort and Future
Direction. US-China Workshop on Advance Computational Modelling in
Hydroscience and Engineering. Oxford, Mississippi, USA.
LHOMME, J., BOUVIER, C., MIGNOT, E. & PAQUIER, A. 2006. One-
dimensional GIS-based model compared with a two-dimensional model in
urban floods simulation. Water science and technology, 54,83-91.
LINS LEY, R. 1967. The relation between rainfall and runoff:: Review paper.
Journalo/Hydrology, 5,297-311.
LINS LEY, R., KOHLER, M. & PAULHUS, J. 1949. Applied hydrology, McGraw-
Hill New York.
LORANI, A. K., GOEL, N. K. & BHATIA, K. K. S. 2010. Comparative study of
neural network, fuzzy logic and linear transfer function techniques in daily
rainfall-runoff modelling under different input domains - Lohani .. 2010 -
Hydrological Processes - Wiley OnIine Library.
MAGOME, J. U. N., FUKAMI, K., INOMATA, H. & TAKEUCHI, K. 2008.
FLOOD RISK FORECASTING FOR POORLy GAUGED BASINS IN
THE MEKONG RIVER BASIN USING A DISTRIBUTED
HYDROLOGICAL MODEL AND A SATELLITE DERIVED
PRECIPITATION DATASET. 6th Annual Mekong Flood Forum. Phnom
Pen: Mekong River Commission.
MC ANELLY, R. & COTTON, W. 1989. The precipitation life cycle ofmesoscale
convective complexes over the central United States.
MCGAHEY, C., SAMUELS, P. & KNIGHT, D. 2006. A practical approach to
estimating the flow capacity of rivers-application and analysis. Lisboa. River
Flow: Taylor & Francis.
MOKHADEM, A. 2010. ERRUISSOL [OnIine]. Namur. Available:
http://cartographie.wallonie.be/NewPortaiICarto/index.jsp?page=subMenuER
RUISSOL&node=32&snode=320 [Accessed 09/09/ 2010].
MORADKHANI, H. & SOROOSHIAN, S. 2008. General review ofrainfall-runoff
modeling: Model calibration, data assimilation, and uncertainty analysis.
Hydrological Modelling and the Water Cycle, 1-24.
MORIN, E. 1977. La méthode - La Nature de la Nature, Paris, Editions du Seuil.
NASA 2008. 3B42 algorithme
NASA. 201 Oa. GPM [Online]. Available: http://pmm.gsfc.nasa.govIPMM_gpm.htm
[Accessed 11/092010].
NASA. 2010b. NASA Facts: Science Issues - The Global Rain Gauge [OnIine].
Washington DC: National Aeronotic and Space Agency. Available:
http://trmm.gsfc.nasa.gov/overview_dir/rain_gauge.htmI[Accessed 03/04
2010].
NASA. 2010c. Tropical Rainfall Measuring Mission - Graduate school [Online].
Washington DC: National Aeronotic and Space Agency. Available:
http://trmm.gsfc.nasa.gov/overview_ dir/why-grad.html l Accessed 4/03
2010).

113
Bibliography

NASA. 201 Od. Tropical Rainfall Measuring Mission - University [Online].


Washington DC: National Aeronautics and Space Administration. A vailable:
http://trmm.gsfc.nasa.gov/overview_dir/why-univ.html [Accessed 3/04
2010].
NASDA 2001. TRMM Data Users Handbook. National Space Development Agency
of Japan.
NÉELZ, S. & GREAT, B. 2009. Desktop review of2D hydraulic modelling
packages, Bristol, Environment Agency.
NISHAT, B. & RAHMAN, S. M. 2009. Water Resources Modeling of the Ganges-
Brahmaputra-Meghna River Basins Using Satellite Remote SensingDatal.
JA WRA Journal of the American Water Resources Association, 45, 1313-
1327.
NOAA. 2010. NOAA Satellite Information Services Quickview - Office ofSatellite
Data Processing and Distribution [OnIine]. Available:
http://noaasis.noaa.govlNOAASIS/ml/quicklook.html [Accessed
24/08/2010].
NYDAILYNEWS. Pakistan floods expected to get worse, officiais say as up to 14
million people affected by disaster [OnIine]. A vailable:
http://www.nydailynews.com/news/worldl2010/08/13/2010-08-
13-pakistan_ floods_ expected_ to~et_worse_officiais_say_as_ up_ to_14_ mill
ion-people_ a.html [Accessed].
O'CONNELL, P. 1991. A historical perspective. Recent Advances in the Modeling of
Hydrologic Systems, D. Bowles and P. 0 'Connell (Editors). Kluwer
Academic Publisher, The Netherlands, 3-30.
O'CONNELL, P. E., NASH, J. E. & FARRELL, J. P. 1970. River flow forecasting
through conceptual models part II - The Brosna catchment at Ferbane.
JournalofHydrology, 10,317-329.
ODONI, N. A. 2007. Exploring equifinality in a landscape evolution model. PhD
Doctoral, PhD, UNIVERSITY OF SOUTHAMPTON.
OH, H., SOHN, B., SMITH, E., TURK, F., SEO, A. & CHUNG, H. 2002.
Validating infrared-based rainfall retrieval algorithms with l-minute spatially
dense raingage measurements over Korean peninsula. Meteorology and
Atmospheric Physics, 81,273-287.
ONLINE, S. 2009. Heidke Skill Score (HSS) [Online]. Available:
http://satreponIine.org/vesa/verif/www/english/msg/ver_ categ_forec/uos3/uo
s3_ko1.htm [Accessed 21/082010].
PENDER, S. N. E. A. G. 2009. Desktop review of2D hydraulic modelling packages.
In: AGENCY, E. (ed.).
PITT, M. 2007. The Pitt Review - Learning lessonsfrom the 2007 floods [Online].
London. A vailable:
http://archive.cabinetoffice.gov.uk/pittreview/thepittreview/interim_report.ht
ml [Accessed 1/09/20102010].
POPPER, K. R. 1963. Conjectures and refutations " the growth ofscientific
knowledge, London, Routledge & K. Paul.
RAMIREZ-BELTRAN, N., LAU, W., WINTER, A., CASTRO, J. & ESCALANTE,
N. 2007. Empirical probability models to predict precipitation levels over
Puerto Rico stations. Monthly Weather Review, 135,877-890.
RAO, M. & ABBOTT, W. 1976. Satellite-derived global oceanic rainfall atlas(1973
and 1974).

114
Bibliography

REFSGAARD, J. & KNUD SEN, J. 1996. Operational validation and


intercomparison of different types of hydrological models. Water Resources
Research, 32,2189-2202.
REGION WALLONNE. 2010. Aléa d'inondation [OnIine]. Available:
http://cartopro3.wallonie.be/alino/viewer.htm [Accessed 28/08/2010 2010].
RELIEFWEB. 20IOa. ReliejWeb» Document» GIEWS Country Brief China
(Mainland) 28-August-2010 [OnlineJ. Available:
http://www.reliefweb.intlrw/rwb.nsfldb900sid/MYAI-
88V5BZ?OpenDocument [Accessed 12/092010].
RELIEFWEB. 2010b. ReliejWeb» Pakistan: Floods - Jul2010 [Online]. Available:
http://www.reliefweb.intlrw/dbc.nsfldocl08?OpenForm&emid=FL-2010-
000141-PAK&rc=3 [Accessed 12/09 2010J.
ROBINSON, M. & WARD, R. 1990. Principles ofHydrology. McGraw-Hil1:
London.
RW. 2010. Hydrométrie [OnIine]. Namur~ Direction générale opérationnelle de la
Mobilité et des Voies hydrauliques. Available: http://voies-
hydrauliques. wallonie. be/opencms/opencms/fr/hydro/GestionIMesures/Hydr
ometrie/ [Accessed 11/09 2010J.
SAVENIJE, H. H. G. 200 L Equifinality, a blessing in disguise? Hydrological
Processes, 15,2835-2838.
SCHULZ, K., BEVEN, K. & HUWE, B. 1999. Equifmality and the problem of
robust calibration in nitrogen budget simulations. Soil Science Society of
America Journal, 63, 1934-1934.
SCHUMANN, A. H. 1993. Development of conceptual semi-distributed
hydrological models and estimation of their parameters with the aid of GIS 1

et estimation de leurs param"®tres"t


D"©veloppement de mod"®les hydrologiques conceptuels semidistribu"©s
l'aide d'Ùll syst"®me d'information
g"©ographique. Hydrological Sciences Journal, 38, 519 - 528.
SHAW, E., BEVEN, K., CHAPPELL, N. & LAMB, R. 2010. Hydrology in practice,
London, Spon Press.
SINGH, V. 1995. Computer models ofwatershed hydrology, Highland Ranch, CO.
SLEIGH, P. A. & GOODWILL, 1. M. 2000. The St Venant equations.
STISEN, #160, SIMON, SANDHOLT & INGE 2010. Evaluation ofremote-sensing-
based rainfall products through predictive capability in hydrologïcal runoff
modelling, Chichester, ROYAUME-UNI, Wiley.
SU, F., HONG, Y. & LETTENMAIER, D. 2008. Evaluation ofTRMM Multi-
Satellite Precipitation Analysis (TMPA) and Its Utility in Hydrologic
Prediction in the La Plata Basin. Journal ofHydrometeorology, 9, 622-640.
SUGIURA, T., FUKAMI, K. & INOMATA, H. Year. Development of Integrated
Flood Analysis System (IFAS) and Its Applications. In, 2009. 12-16.
SUMNER, G. 1988. Precipitation: process and analysis, Wiley.
SYME, B. 2006. 2D or not 2D? - An Australian perspective. 41st Defra Flood and
Coastal Risk Management Conference. York: DEFRA.
SYME, W. J. 1991. Dynamically Linked Two Dimensional/One-Dimensional
Hydrodynamic Modelling Programfor Rivers, Estuaries and Coastal Waters.
M.Eng. Sc, University of Queensland.
TORO, E. F. & TORO, E. 2001. Shock-capturing methodsfor free-surface shallow
flows, Wiley New York.

115
Bibliography

USHIO, T., SASASHIGE, K., KUBOTA, T., SHIGE, S., OKAMOTO, K.,
AONASHI, K., INOUE, T., TAKAHASHI, N., IGUCHI, T. & KACHI, M.
2009. A Kalman filter approach to the Global Satellite Mapping of
Precipitation (GSMaP) from combined passive microwave and infrared
radiometric data. Journal of the Meteorological Society ofJapan, 87, 137-
15l.
VALERIANO, O. C. S., KOlKE, T., YANG, D., NYUNT, C. T., VAN KHANH, D.
& CHAU, N. L. 2009. Flood simulation using different sources ofrainfall in
the Huong River, Vietnam/Simulation d'inondation à l'aide de différentes
sources d'information pluviométrique dans le bassin de la Rivière Huong,
Vietnam. Hydrological Sciences Journal, 54, 909-917.
VAN FRAASSEN, B. 1980. The scientific image, Oxford University Press, USA.
V AZQUEZ, R. F. & FEYEN, J. 2010. Rainfall-runoffmodelling of a rocky
catchment with limited data availability: Defining prediction limits. Journal
ofHydrology, 387, 128-140.
VON BERTALANFFY, L. 1950. The theory of open systems in physics and
biology. Science, 111,23-29.
WALLINGFORD, S. 2010. InfoWorks Rs. An integrated software solutionfor
simulatingflows in rivers, in channels and onfloodplains." [Online). Oxon:
MWH Soft Limited. A vailable:
http://www.wallingfordsoftware.comlproducts/infoworks_rs/ [Accessed
14/03/20102010].
WEINMAN, J. & WILHEIT, T. 1981. A survey of passive microwave and hybrid
remote sensing of precipitation. 1981. 3.
WERNER, M. 2004. A comparison of flood extent modelling approaches through
constraining uncertainties on gauge data. Hydrology and Earth System
Sciences, 8, 1141-1152.
WERNER, M. & DELFT, T. 2005. Spatial flood extent modelling: A performance-
based comparison. ProQuest Information & Learning, 300 N Zeeb Rd. PO
Box 1346 Ann Arbor MI 48106-1346 USA.
WEYMAN, D. 1970. Throughflow on hillslopes and its relation to the stream
hydrograph. Hydrological Sciences Journal, 15,25-33.
WHS. 2010. FEH-CD Rom 3 [Online]. Wallingford. Available:
http://www.hydrosolutions.co.ukIfehcd.html [Accessed 08/09 2010].
WIKIPEDIA. 2010.1931 Chinafloods - Wikipedia, thefree encyclopedia [Online].
A vailable: http://en.wikipedia.org/wiki/1931_China_ floods#cite_ note-pbsn-3
[Accessed 24/08/20102010].
XINHUA. 2010. Death tollfromfloods in NE China rises to 74; 71 still missing
[OnlineJ. Changchun. A vailable:
http://news.xinhuanet.comlenglish2010/china/2010-08/03/c_13428386.htm
[Accessed 24/08/2010 2010].
YANG, C., YU, Z., LIN, Z., HAO, Z., WANG, Z. & LI, M. 2009. Studyon
watershed hydrologic processes using TRMM satellite precipitation radar
products. Advances in Water Science, 20,461-466.
YONG, B., REN, L., HONG, Y., WANG, J., GOURLEY, J., JIANG, S., CHEN, X.
& WANG, W. 2010. Hydrologic evaluation of Multisatellite Precipitation
Analysis standard precipitation products in basins beyond its inclined latitude
band: A case study in Laohahe basin, China. Water Resources Research, 46,
W07542.

116
Bibliography

ZAK, S. K. & BEVEN, K. J. 1999. Equifinality, sensitivity and predictive


uncertainty in the estimation of criticalloads. Science of the total
environment, 236, 191-214.

117

You might also like