10712_ten_Veldhuis_080410.

10712_ten_Veldhuis_080410.
Quantitative risk analysis of urban flooding
in lowland areas
ISBN: 978-94-6108-035-6
Layout and printed by: Gildeprint Drukkerijen - Enschede, the Netherlands
Quantitative risk analysis of urban flooding
in lowland areas
Proefschrift
Ter verkrijging van de graad van doctor
aan de Technische Universiteit Delft,
op gezag van de Rector Magnificus prof. K.C.A.M Luyben,
voorzitter van het College voor Promoties,
in het openbaar te verdedigen op
dinsdag 18 mei 2010 om 15.00 uur
door
Johanna Antonia Elisabeth TEN VELDHUIS
civiel ingenieur
geboren te Roosendaal
Dit proefschrift is goedgekeurd door de promotor:
Prof. dr. ir. F.H.L.R. Clemens
Samenstelling promotiecommissie:
Rector Magnificus
Prof. dr. ir. F.H.L.R. Clemens
Prof. dr. S.J. Tait
Prof. dr. F. van Knapen
Prof. dr. ir. L.C. Rietveld
Prof. dr. ir. J. Berlamont
Dr.ir. P.H.A.J.M. van Gelder
Dr. P. Le Gauffre
“Ουτος σοφωτατος εστι οστις εγνωκεν
οτι ουδενος αξιος εστι προς σοφιαν”
“Wisest is he who knows that he knows not”
Plato, Apology 23b,
cites Socrates
Summary
Over the last few decades, the interest in urban flood risk has been growing
steadily worldwide, as the frequency of flooding and the damage caused
by urban flood events have increased. Accelerated urbanisation has given
rise to increased building in flood-prone areas and expansion of impervious
areas, adding to the inflow into existing urban drainage systems and thus to
the probability of flooding. In addition, climate change predictions increase
concern over urban flood risk in cities around the world.
Analyses of urban flood risk require quantitative historical data on frequencies
and consequences of flooding events to quantify risk. Such data are scarce: data
collection takes place on an ad hoc basis and is usually restricted to severe events.
The resulting data deficiency renders quantitative assessment of urban flood
risks uncertain. The study reported in this thesis reviewed existing approaches
to quantitative flood risk analysis and evaluation of urban flooding guidelines.
It proceeded to explore historical data on flooding events from municipal call
centres in two cities in the Netherlands with the final aim to quantitatively
assess urban flood risk. Data from municipal call centres consist of texts
describing citizens’ observations of urban drainage problems. The texts provide
information about causes, locations and consequences of flooding events. Flood
risk analysis was applied according to a three steps approach: identification
of causes of flooding, followed by a quantification of flood probabilities and a
quantitative assessment of consequences of urban flooding.
Probabilistic fault tree analysis was applied to identify failure mechanisms of
urban flooding based on call text information about flooding causes. Flood
probabilities were quantified for each cause as well as contributions of individual
causes to the overall flood probability. Fault tree analysis results showed that
gully pot blockages stood out as the main cause of flooding; the contribution of
heavy rainfall to the overall probability of flooding was small compared to that
of blockages. This implies that component failures and human errors contribute
more to flood probability than sewer overloading by heavy rainfall.
Call information on flooding consequences was used to draw risk curves for a
range of flood damage classes: separate risk curves were drawn for consequences
associated with human health, damage to private property and damage related
to traffic disturbance. Risk curves for urban flooding depict flood damages on
the horizontal axis and their associated exceedance probabilities on the vertical
axis. The advantage of risk curves as opposed to a single expected value of
risk as a summary value is that risk curves show the contributions of small
and of large events to total risk. Call data per flooding event were used as a
measure of event severity, based on the finding of a strong correlation between
the amount of call data per event and rainfall volumes per event. The risk
curves showed that total flood risk was mainly constituted by small events.
Urban flood risk related to traffic disturbance was high compared to damage
to private properties. Flood risk related to human health was small, according
to call data information. Risk curves also showed that current flood protection
strategy is risk averse: it provides higher protection from flood events with
large consequences than from flood events with small consequences.
Flood waters that result form overloading of combined sewer systems are likely
to be contaminated. A screening-level microbial risk assessment was conducted
to estimate health risks to citizens associated with combined sewer flooding. The
assessment was based on analyses of samples from flooding events and samples
from combined sewers. The results indicated faecal contamination: faecal
indicator organism concentrations in samples from flood waters were similar to
those found in crude sewage under high flow conditions and Campylobacter was
detected in all samples. Annual risk values were calculated for low and for high
exposure scenarios: calculated annual infection risks vary from to 5x10-6 to 0.3;
the minimum value is for Cryptosporidium and a low exposure scenario, while
the maximum value is for Campylobacter based on high exposure scenario. The
results showed that health risk associated with flood waters from overloaded
combined sewers could be of the same order of magnitude as those associated
with swimming in surface waters exposed to combined sewer overflows.
Risk assessment results based on call data information showed that urban
flood risks in lowland areas are characterised by frequent flooding of roads and
occasional flooding of buildings. Whether or not such flooding is acceptable
for society depends on how flood risks can be compared to and balanced
against investments to prevent or reduce flood risk. Such risk-based decision
making requires risk outcomes that can be weighed against investments. To this
end, risk values from call data analysis were translated into monetary values
and into numbers of people affected by flooding. Translation into monetary
terms resulted in high flood risk associated with flooding of buildings and low
monetary flood risk associated with flooding of roads, cycle paths and footpaths.
When expressed in terms of the number of people affected by flooding, the
risk associated with flooding of buildings is low and the risk for flooding of
roads is high. This implies that for lowland areas, risk-based decisions using
monetarised values of flood damage put emphasis on flood risk for buildings.
Conversely, if risk-based decisions focused on numbers of people affected by
flooding, they would concentrate on damage associated with flooding of roads,
cycle and footpaths. In risk-based decision making for urban flood protection,
the choice what aspects to take into account and what level of flood protection
is considered acceptable is typically a political one.
The effectiveness of existing strategies to cope with urban flooding was
assessed based on a comparison of flood risk values associated with three
failure mechanisms of urban flooding and associated coping strategies: gully pot
blockage and cleaning, pipe blockage and cleaning and sewer overloading and
capacity increase. The results were expressed as quantitative flood risk values
in the form of a number of flooded locations per year per km sewer length, so
that the results of two cases could be compared. It was shown that cleaning
gully pots is a more efficient strategy to reduce flood risk than increasing sewer
pipe capacity or sewer cleaning frequencies, for the investigated cases. Based
on the same investment level, increasing gully pot frequencies was estimated
to result in about 10% decrease in flood risk values, whereas increasing sewer
cleaning or increasing sewer capacity resulted in less than 5% decrease.
Alternatively, flood risk can be reduced through reactive maintenance, by
realising short reaction times to calls on flood events in order to limit flooding
consequences. Call data showed that reactive handling is only efficient if the
quality of call information is sufficient to discriminate between calls indicating
large consequences that require immediate handling and those indicating small
or irrelevant consequences.
Finally, the resulting risk values were used to evaluate current frequencybased guidelines for urban flooding. The results pointed out a number of
shortcomings of frequency-based standards for urban flooding. First, all
potential causes of flooding, including hydraulic overloading and asset failures
should be taken into account to obtain a realistic flood risk estimate. Current
focus on hydrodynamic model simulations to evaluate urban flooding tends
to neglect the influence of asset failures. Second, flooding standards should
specify to what spatial scale they apply to ensure proper evaluation. In current
practice, flooding analysis is usually limited to a non-exceedance check of a
given standard, while system functioning under exceedance conditions is not
considered. If time-series are used to evaluate system functioning, including
exceedance conditions, spatial scale becomes important to be able to decide
whether what degree of exceedance is acceptable, by comparison to a generally
applicable standard. Third, standards should take flooding consequences into
account, because damage to society differs with various types and extents of
consequences. The advantage or risk-based standards is that, unlike frequencybased standards, they incorporate flooding probabilities and consequences and
that a risk-based approach looks at different kinds of failure mechanisms.
Quantification of urban flood risk based on historical series of call data is a
first step towards quantitative risk assessment and risk-based evaluation of
urban drainage systems. The advantage of call data is that they directly reflect
citizens’ experience with flooding; the disadvantage is that they represent only
a part of all flooding events. This thesis concludes with recommendations on
how to close existing knowledge gaps by improving existing call data collection
and storage. Suggestions are provided for additional data collection strategies
and methods for urban flooding in order to facilitate complete and reliable
quantitative urban flood risk analysis.
Samenvatting
De laatste tientallen jaren is de belangstelling voor stedelijke wateroverlastrisico’s wereldwijd gestadig gegroeid, doordat de frequentie van wateroverlast
en de schade veroorzaakt door wateroverlastgebeurtenissen zijn toegenomen.
Versnelde urbanisatie heeft geleid tot toenemende bebouwing in
overstromingsgevoelige gebieden en uitbreiding van verhard oppervlak, waardoor
de instroming naar bestaande rioleringssystemen is gestegen en daarmee de
kans op wateroverlast. Bovendien doen voorspellingen over klimaatverandering
de angst voor wateroverlastrisico’s in steden over de hele wereld toenemen.
Analyses van stedelijke wateroverlastrisico’s vereisen kwantitatieve historische
gegevens over frequenties en gevolgen van wateroverlastgebeurtenissen
om de risico’s te kunnen kwantificeren. Zulke gegevens zijn schaars:
gegevensverzameling vindt slechts op ad hoc basis plaats en blijft gewoonlijk
beperkt tot ernstige gebeurtenissen. Het hieruit volgend gebrek aan gegevens
maakt kwantitatieve berekeningen van wateroverlastrisico’s onzeker. De studie
waarvan dit proefschrift verslag is gestart met een evaluatie van bestaande
benaderingen voor kwantitatieve risicoanalyse en bestaande richtlijnen voor
stedelijke wateroverlast geëvalueerd. Vervolgens werden historische gegevens
over wateroverlastgebeurtenissen afkomstig van gemeentelijke meldpunten
in 2 Nederlandse steden geanalyseerd met het uiteindelijke doel om stedelijke
wateroverlastrisico’s te kwantificeren. Meldpuntgegevens bestaan uit teksten
die waarnemingen van burgers van problemen in het stedelijk watersysteem
beschrijven. De teksten bevatten informatie over oorzaken, locaties en gevolgen
van wateroverlastgebeurtenissen. Bij de risicoanalyse werd een drie-stappenbenadering gevolgd: identificeren van oorzaken van wateroverlast, gevolgd
door het kwantificeren van wateroverlastkansen en tenslotte van de gevolgen
van stedelijke wateroverlast.
Voor het identificeren van de faalmechanismen voor stedelijke wateroverlast
werd probabilistische foutenboomanalyse toegepast, op basis van informatie in
de meldingenteksten over oorzaken van wateroverlast. Wateroverlastkansen
werden gekwantificeerd voor elke oorzaak apart evenals voor de bijdrage
van elke oorzaak aan de totale kans op wateroverlast. Uit de resultaten van
de foutenboomanalyse bleken verstopte kolken de belangrijkste oorzaak
van wateroverlast; de bijdrage van hevige regenval aan de totale kans op
wateroverlast was klein in vergelijking met die van verstoppingen. Dit betekent
dat het falen van onderdelen en menselijke fouten meer bijdragen aan de kans
op wateroverlast dan het overbelast raken van riolen bij hevige regenval.
Meldinginformatie over wateroverlastgevolgen werd gebruikt om
risicografieken samen te stellen voor een serie schadeklassen van wateroverlast:
aparte risicografieken werden getekend voor gevolgen voor de publieke
gezondheid, voor schade aan private eigendommen en voor schade verbonden
aan verkeershinder. De risicografieken voor stedelijke wateroverlast
tonen een reeks toenemende wateroverlastschades met hun bijbehorende
overschrijdingskansen. Het voordeel van een risicografiek ten opzichte van een
samenvattende risico-uitkomst in de vorm van een gemiddelde is dat het aandeel
van kleine gebeurtenissen ten opzichte van grote gebeurtenissen in het totale
risico zichtbaar wordt. Het aantal meldingen per wateroverlastgebeurtenis
is gebruikt als een maat voor de ernst van de gebeurtenis, op basis van de
gevonden sterke correlatie tussen het aantal meldingen per gebeurtenis
en het neerslagvolume. De risicografieken lieten zien dat het totale
wateroverlastrisico met name werd bepaald door kleine gebeurtenissen.
Stedelijke wateroverlastrisico’s gerelateerd aan verkeershinder bleken groot
in vergelijking met wateroverlastrisico’s gerelateerd aan schade aan privaat
eigendom, op basis van het aantal meldingen. Wateroverlastrisico’s voor de
publieke gezondheid waren klein volgens de meldingeninformatie. Tenslotte
lieten de risicografieken zien dat de huidige beschermingsstrategie tegen
stedelijke wateroverlast risicomijdend is: de bescherming tegen gebeurtenissen
met grote gevolgen bleek hoger dan tegen gebeurtenissen met kleine gevolgen.
Water-op-straat dat een gevolg is van overbelasting van gemengde
riolen is waarschijnlijk besmet met ziekteverwekkers. Een verkennende
microbiologische risicoberekening is uitgevoerd om het gezondheidsrisico
voor burgers verbonden aan het overlopen van gemengde riolen te bepalen.
De berekening werd gebaseerd op analyses van monsters uit water-opstraatgebeurtenissen en uit gemengde riolen. De resultaten duidden op fecale
verontreiniging: de concentraties fecale indicatororganismen aangetroffen
in monsters uit water-op-straat zijn vergelijkbaar met de concentraties in
gemengde riolen onder hoge-afvoercondities en Campylobacter werd in alle
water-op-straatmonsters aangetroffen. Het jaarlijkse risico is bepaald voor een
hoog en een laag blootstellingsscenario: het berekende jaarlijkse infectierisico
varieerde van 5x10-6 to 0.3; de minimum waarde is voor Cryptosporidium en
een hoog blootstellingsscenario, de maximum waarde is voor Campylobacter
gebaseerd op een hoog blootstellingsscenario. De resultaten gaven aan dat
gezondheidsrisico’s verbonden aan water-op-straat afkomstig van overbelaste
gemengde riolen van dezelfde grootte-orde zijn als de risico’s verbonden aan
zwemmen in oppervlaktewater waarop riooloverstortwater uitkomt.
De berekende risico’s op basis van meldingeninformatie toonden aan dat
stedelijke wateroverlastrisico’s in laaggelegen gebieden worden gekenmerkt
door frequente wateroverlast op straat en sporadische wateroverlast in
gebouwen. Of dergelijke wateroverlast maatschappelijk acceptabel is hangt
af van de manier waarop wateroverlastrisico’s kunnen worden vergeleken
met en afgewogen tegen investeringen om wateroverlast te voorkomen of
te verminderen. Dergelijke risicogebaseerde besluitvorming vereist risicouitkomsten die kunnen worden afgewogen tegen investeringen. Hiertoe werden
de risico-uitkomsten uit de meldingenanalyse vertaald naar financiële termen en
naar aantallen mensen die wateroverlast ondervinden. Vertaling naar financiële
termen resulteerde in hoge risicowaarden voor wateroverlast in gebouwen en
lage risicowaarden voor wateroverlast op wegen, fiets- en voetpaden. Wanneer
risico werd uitgedrukt in aantallen mensen die wateroverlast ondervinden,
resulteerde dit in lage risicowaarden voor wateroverlast in gebouwen en hoge
risicowaarden voor wateroverlast op straat. Dit betekent dat voor laaggelegen
gebieden, risicogebaseerde besluiten die uitgaan van risico’s in financiële termen
de nadruk leggen op wateroverlast in gebouwen. Risicogebaseerde besluiten
op basis van het betrokken aantal mensen zouden zich juist concentreren
op wateroverlast op wegen, fiets- en voetpaden. De keuze welke aspecten in
beschouwing worden genomen bij risicogebaseerde besluitvorming en welk
risiconiveau aanvaardbaar is, is naar de aard een politiek besluit.
De effectiviteit van bestaande strategieën om wateroverlast aan te pakken
werden beoordeeld op basis van een vergelijking van de wateroverlastrisico’s
voor drie faalmechanismen en bijbehorende operationele maatregelen:
verstopping van kolken en kolken zuigen, verstopping van rioolleidingen en
rioolreiniging en overbelasting van riolering en capaciteitsuitbreiding. De
analyse werd gebaseerd op risicowaarden uitgedrukt in de vorm van het aantal
wateroverlastlocaties per jaar per kilometer rioolleiding, zodat de resultaten van
de twee studiegebieden konden worden vergeleken. Op basis van de uitkomsten
van meldingenanalyse bleek dat voor de onderzochte systemen het reinigen
van kolken een efficiëntere strategie is om wateroverlast te verminderen dan
het uitbreiden van rioleringscapaciteit of het reinigen van riolen. Bij hetzelfde
investeringsniveau leverde het verhogen van kolkreinigingsfrequenties naar
schatting 10% vermindering in het wateroverlastrisico, terwijl het verhogen
van de rioolreinigingsfrequentie of het uitbreiden van rioleringscapaciteit
minder dan 5% vermindering opleverde. In plaats van preventief, kan het
overstromingsrisico reactief worden aangepakt, door korte reactietijden voor
de meldingen na te streven teneinde de wateroverlastgevolgen te beperken. De
meldingengegevens toonden aan dat reactief handelen alleen efficiënt is als de
kwaliteit van meldinginformatie voldoende is om onderscheid te maken tussen
meldingen die grote wateroverlast betreffen en direct handelen vereisen en
meldingen die kleine wateroverlast of irrelevante problemen betreffen.
De risico-uitkomsten werden tenslotte gebruikt om bestaande richtlijnen die
zijn gebaseerd op wateroverlastfrequenties te evalueren. De resultaten brachten
een aantal beperkingen van frequentiegebaseerde normen voor wateroverlast
aan het licht. Ten eerste zouden alle faalmechanismen, inclusief overbelasting
en het falen van onderdelen in beschouwing moeten worden genomen om een
realistische schatting van het wateroverlastrisico te verkrijgen. De bestaande
nadruk op het gebruik van hydrodynamische modelsimulaties voor het evalueren
van wateroverlast hebben de neiging de invloed van het falen van onderdelen
te verwaarlozen. Ten tweede, zouden wateroverlastnormen specifiek moeten
aangeven voor welke ruimtelijke schaal ze gelden om tot een juiste evaluatie te
komen. In de huidige praktijk blijft wateroverlastanalyse doorgaans beperkt
tot een niet-overschrijdingscontrole van een gegeven norm. Het functioneren
van een systeem onder omstandigheden die de norm overschrijden wordt vaak
niet geanalyseerd. Als daarentegen tijdseries worden gebruikt om systemen te
evalueren, inclusief overschrijdingscondities, is de ruimtelijke schaal van belang
om te kunnen bepalen of overschrijding aanvaardbaar is door deze af te wegen
tegen een algemeen toepasbare norm. Ten derde zouden normen de gevolgen
van wateroverlast in beschouwing moeten nemen, omdat de maatschappelijke
schade afhangt van het type en de omvang van wateroverlastgevolgen.
Het voordeel van risicogebaseerde normen is dat zij, in tegenstelling tot
frequentiegebaseerde normen, zowel kansen als gevolgen in beschouwing
nemen en dat een risico-benadering verschillende faalmechanismen bekijkt.
Het kwantificeren van stedelijke wateroverlast op basis van historische
reeksen van meldinggegevens is een eerste stap op weg naar kwantitatieve
berekening van stedelijke wateroverlastrisico’s en risicogebaseerde evaluatie
van rioleringssystemen. Het voordeel van meldinggegevens is dat ze direct
weergeven hoe burgers wateroverlast ervaren; het nadeel is dat ze slechts
een deel van alle gebeurtenissen bestrijken. Dit proefschrift sluit af met
aanbevelingen hoe de bestaande kennishiaten te vullen door verbetering van
de bestaande verzameling en opslag van meldinggegevens. Daarnaast worden
suggesties gedaan voor aanvullende strategieën en methoden om gegevens
over wateroverlast te verzamelen, teneinde tot complete en betrouwbare
kwantitatieve risicoanalyse voor stedelijke wateroverlast te komen.
Contents
Summary
Samenvatting
7
13
1.
1.1.
1.2.
1.3.
1.4.
1.5.
Introduction and overview
Introduction
Urban flood risk literature
Flood event data
Thesis overview
References
23
26
35
38
40
2.1.
2.2.
2.3.
2.4.
2.5.
2.6.
Quantitative fault tree analysis for urban flooding
Introduction
Urban flood incident data
Quantitative fault tree model for urban flooding
Results of quantitative fault tree analysis for two
case studies
Discussion
References
3.1.
3.2.
3.3.
3.4.
3.5.
3.6.
Urban flood risk curves
Introduction
Classification of flooding consequences
Risk curves
Results and discussion
Conclusions
References
4.1.
Microbial risks associated with exposure to pathogens in
contaminated urban flood water
Introduction
2.
3.
4.
47
49
52
60
70
74
79
81
83
87
95
96
101
4.2.
4.3.
4.4.
4.5.
Materials and methods
Results and discussion
Conclusions en recommendations
References
103
110
117
119
5.1.
5.2.
5.3.
5.4.
5.5.
Quantification and acceptability of urban flood risk
Introduction
Quantification method of flood consequences
Results and discussion Conclusion
References
125
127
132
138
140
5.
6.
Risk-based urban flood management: improving
operational strategies
6.1.
Introduction
6.2.
Methods
6.3.
Results
6.4.
Conclusions
6.5.
References
7.
7.1.
7.2.
7.3.
7.4.
7.5.
7.6.
Discussion and recommendations
Contribution to answer research questions
Generalisation of results from case studies
Recommendations for further research
Recommendations for data collection for quantitative
urban flood risk analysis
Recommendations for analysis and handling of asset failures
References
Appendices
List of publications
Acknowledgements
Curriculum vitae
145
148
150
162
164
169
177
178
185
190
192
195
247
251
255
Chapter 1
Introduction and overview
Chapter 1
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
22
Introduction and overview
1.1
Introduction
Research context
Over the last few decades, the interest in urban flood risk has been growing
steadily worldwide, as the frequency of flooding and the damage caused by
urban flood events have increased (e.g. Ashley et al., 2005; Schreider et al.,
2005; Dutta et al., 2003). Accelerated urbanisation has given rise to increased
building in flood-prone areas and expansion of impervious areas, adding to
the inflow into existing urban drainage systems and thus to the probability of
flooding. In addition, climate change predictions increase concern over urban
flood risk in cities around the world (Wilby, 2007).
Urban flooding can be pluvial, fluvial or coastal flooding or a combination of
these. Urban pluvial flooding occurs as a result of rainfall-generated overland
flow ponding on the urban surface because it overwhelms urban underground
sewerage/drainage systems and surface watercourses by its high intensity or
is for some reason unable to enter drainage systems or water courses. Coastal
flooding is caused by high sea water levels and waves overtopping protection
structures; fluvial flooding is a result of overflowing of river banks. The focus
of this thesis is on urban pluvial flooding.
Protection from urban pluvial flooding is provided by urban drainage systems.
These are designed to function in accordance with prescribed flooding standards,
mostly defined in terms of maximum flooding frequencies. Standards are set by
local or regional authorities: cities, water boards or other governmental bodies
responsible of water policy. Some differentiate between occupational land
uses, like residential and commercial areas. By doing so, protection standards
implicitly seek to establish a trade-off between investment costs for flood
protection and expected damage from flooding: for higher expected damage,
stricter flooding standards apply.
23
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1
Problem
This trade-off is based on a qualitative assessment of expected flood damage; a
lack of quantitative historical data on flooding incidents prevents quantitative
assessment of urban pluvial flooding frequencies and damage. In the aftermath
of recent flooding in England and Wales (Ashley et al. 2005), Germany (Thieken
et al., 2005), USA (Hallegatte, 2008) and other areas, many more or less
quantitative flooding analyses have been conducted, dedicated to individual,
severe flood events. Flood risk analyses based on long data series comprising
series of flood events are rare, since data collection takes place on an ad hoc
basis and is usually restricted to severe events. This implies that compliance
with flooding standards is not checked based on structural collection of event
data to estimate return periods of flooding. Instead, compliance of systems with
flooding standards is usually checked by hydrodynamic model calculations
based on design storms with fixed return periods. Hydrodynamic models are
subject to uncertainties associated with external model input, especially rainfall
variability, errors in geometrical data and run-off catchment size, imperfect
functioning of sewer components and a lack of data for calibration (e.g. Rauch
et al., 2002; Pappenberger and Beven, 2006, Korving et al., 2009). As a result,
the outcomes of urban drainage models are to a great extent uncertain. The use
of flooding frequencies supplied by models to check compliance with flooding
standards may lead to unreliable conclusions and possibly to unnecessary
overdimensioning of drainage systems (Thorndahl et al., 2008).
In addition, hydrodynamic model calculations are sometimes used to prepare
flood risk maps, count the number of properties at risk of flooding and to estimate
flooding characteristics like flood depths and flow velocities. These outcomes
serve to quantify flood damage and to decide whether investments should be
made to reduce flood damage. Application of this approach for urban flooding
results in large uncertainties since overland flow models used to calculate flood
extents and flood depths are based on uncertain inputs from hydrodynamic
models and suffer from a lack of input and calibration data. Additionally,
establishment of depth-damage relations requires site-specific data on flood
24
Introduction and overview
damages that are seldom available. These uncertainties must be addressed in
order to improve quantitative assessment of urban flooding problems in order
to provide a better foundation on which to base decisions to reduce flood risk,
Research aim
Current methods for urban pluvial flooding analysis are based on hydrodynamic
models and damage assessments that are unreliable for various reasons, the
main being a lack of data on flood occurrences. Investment decisions for urban
flood protection require reliable estimation with known accuracy of flood
frequencies and damage, whether derived from a combination of models or
directly derived from event data. Data availability is central for either method
chosen. The general problem statement addressed in this thesis is:
What new insights can risk analysis based on historical series of flood occurrence data
provide with respect to characteristics of flood events in lowland areas?
The study reported in this thesis explores historical data on flooding incidents
from municipal call centres in two cities in the Netherlands with the final aim
to quantitatively assess urban pluvial flood risk. Municipal call centres receive
calls from citizens about urban drainage problems and register information
describing there observations. In the Netherlands, many municipalities have
a call register: 109 out of 190 municipalities according to a recent inquiry
(RIONED, 2007).
Flood risk is defined in this context as the product of flood probability and
associated consequences. The study specifically addresses flooding incidents in
lowland areas; these areas are characterised by a relatively high urban pluvial
flooding frequency and small associated incident damage compared to hilly
areas. As a result, questions as to what flooding standards to apply and how
to balance investments and effects of flooding on the urban environment are
different. Four research questions are derived from the main question to give
further direction to this study:
25
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1
1. What are the opportunities for using call centre data to identify causes that
contribute to urban pluvial flood risk and how can occurrences of these
causes be quantified?
Fault tree analysis is applied to identify possible causes of urban flooding and to
quantify the contribution of different causes based on call centre data.
2. What consequences of urban pluvial flooding should be taken into account
in a risk analysis and how can these be quantified?
Various kinds of consequences are compared, from potential microbial infection
to material damage and intangible consequences of flooding.
3. Can the results of quantitative urban pluvial flood risk analysis based
on historical data series from municipal call centres be used to support
decisions on how to efficiently improve flood protection?
4. Can risk-based standards for urban pluvial flooding provide a more
comprehensive basis to evaluate flood protection by urban drainage
systems than current frequency-based standards?
The obtained quantitative results are used to evaluate current flood protection
standards for two cities in the Netherlands and the efficiency of currently
applied solutions to prevent or alleviate urban pluvial flood risk.
1.2
Urban flood risk literature
Urban flooding guidelines, design criteria and methods to evaluate
compliance
Guidelines and design manuals for urban drainage systems have been developed
in the last decades, for instance by EU (CEN, 2008), the US Federal Highway
Administration (Brown et al., 2009) and in Australia (Pilgrim, 2001) that contain
prescriptions or recommendation for protection from pluvial flooding. Most
refer to maximum flooding frequencies and differentiate between occupational
26
Introduction and overview
functions, applying lower frequencies to more vulnerable or economically
valuable areas. This implies that flood protection levels are based on the concept
of risk, combining frequency and expected damage. While is frequencies are
defined quantitatively, the damage or vulnerability component is described
in a qualitative way. The European Guideline EN752 (CEN 2008) for drains
and sewer systems outside buildings contains the following requirement for
protection against flooding: “Flooding shall be limited to nationally or locally
prescribed frequencies taking into account the health and safety effects, costs,
extent to which any surface flooding can be controlled without causing damage
and whether it is likely to lead to flooding of basements.” The guideline also
states that “It is usually impracticable to avoid flooding from very severe storms.
A balance therefore has to be drawn between cost and the political choice of
the level of protection provided. The level of protection should be based on a
risk assessment of the impact of flooding to persons and property.” Directive
2007/60/EC on the assessment and management of flood risks (EU, 2007)
defines ‘flood risk’ as the combination of the probability of a flood event and of
the potential adverse consequences for human health, the environment, cultural
heritage and economic activity associated with a flood event. The directive
applies to coastal and river flooding; its application “may exclude floods from
sewerage systems”. It is interesting to note that both guidelines centre around
the concept of risk, which requires an analysis of flooding frequencies and
expected damage.
This thesis focuses on flooding problems in lowland areas based on case studies
in the Netherlands; the EN752 is the relevant guideline for this area. The EN752
guideline translates requirements for flood protection into two types of design
criteria, depending on the complexity of the design method applied. Simple
design methods are to be applied only to small schemes. Design criteria for
simple design methods are based on design storms that should not lead to sewer
system surcharge. Those for complex methods are based on hydrodynamic
model calculations for time-dependent design rainfall. The design storms
have a recommended return period of 1, 2, 5 or 10 years for rural, residential,
27
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1
commercial and city centre areas and underground railways and underpasses
respectively. The design criteria for complex design methods refer to flooding
frequencies that are to be calculated using a time-dependent rainfall input. The
recommended design criteria for flooding frequencies are 1 in 10, 20, 30 or 50
years for rural, residential, commercial and city areas and underground railway
and underpasses respectively.
In practice, design storms are traditionally used to evaluate the capacity of
small and large sewer systems, because calculations for time-dependent rainfall
series require long calculation times (Thorndahl et al., 2008). The European
guideline applies a 5 to 10 fold higher return period for flooding frequencies
based on time-dependent rainfall than for design storms leading to surcharge.
Thus, it implicitly assumes that the return period of surcharge based on design
storms corresponds with a 5 to 10 times longer return period of flooding. In
reality, this relation depends on specific urban drainage systems characteristics
like transport distances, invert levels, ground level variations, bottle neck
connections, storage capacity etc. When compliance with urban flooding
standards is based on design storm calculations, an unknown uncertainty is
introduced due to the unknown relation between return periods of design
storms and return periods of flooding. Additionally, surcharge and flooding
frequencies provide no information on flood damage so that separate analyses
should be conducted to assess expected damage.
In the Netherlands, local authorities decide upon the protection level against
flooding that sewer systems should provide. Commonly, the guideline for flood
protection is defined in terms of a maximum expected street flooding frequency
of once per year or once per two years (van Mameren and Clemens, 1997).
In many cases, the guideline does not specify to what area size this frequency
applies and how the guideline should be evaluated. In practice, flooding of
sewer systems is evaluated by hydrodynamic model calculations for a sewer
subcatchment or for a sewer system in a city as a whole. Thus, the area the
guideline is applied to depends on the boundaries of the available hydrodynamic
28
Introduction and overview
model. Design storms with return periods of one or two years are applied to
check system performance against the flooding guideline. Unlike the approach
in the European guideline, design storms are directly used to evaluate street
flooding and the return period of street flooding is assumed to be equal to the
return period of flooding as a result of time-dependent rainfall. To account for
uncertainty associated with the latter assumption, design storms of a higher
return period than the required return period of flooding are sometimes used or
surcharge to a certain level below ground level is taken as maximum acceptable
water level for a design storm instead of the rise of water levels up to ground
level.
Flood risk instead of flood frequencies
Required protection levels against flooding are mostly expressed in terms of a
maximum flooding frequency or, inversely, a minimum return period of flooding.
They do not provide sufficient information to support investment decisions for
flood reduction since they do not include flood damage. If investment costs are
to be balanced against the level of protection provided, the level of protection
should be based on a risk assessment of the impact of flooding to persons and
property (EN 752, 2008).
The word ‘risk’ is used and interpreted in many ways. A committee established
by the Society for Risk Analysis concluded after 4 years of deliberation that
no common definition for the word risk could be found and concluded that it
would be better to let each author define it in his own way explaining clearly
what way that is (Kaplan, 1997). A number of common concepts are generally
agreed upon in risk theory. The basic concept is that risk incorporates some
probability of unwanted events and consequences following that event. Kaplan
and Garrick (1981) in their article in the 1st issue of the journal of the Society of
Risk Analysis argue that the question “What is risk” really includes 3 questions:
1. What can go wrong?
2. How likely is it to happen?
3. If it does happen, what are the consequences?
29
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1
The answer to the first question can be considered as a scenario; e.g. overtopping
of a river dike and collapse of a river dike are different scenarios for flooding.
Several qualitative and quantitative methods are available to find scenarios for
unwanted events, see e.g. the fault tree handbook issued by NASA (Vesely et
al. 2002). The answer to the second question addresses uncertainty about the
occurrence of hazardous events or scenarios. The answer takes the form of a
frequency or probability. The answer to the 3rd question refers to a damage
index, resulting from an unwanted event.
The advantage of risk over frequency is that the concept of risk incorporates
both the frequency of events, mostly translated into a probability, and the
associated damage. By differentiating between occupational uses, current
flooding standards incorporate prioritisation according to the potential damage
of flooding to some extent. Risk-based standards do this in a more explicit and
quantitative way.
The philosophy of risk analysis is relatively recent; it was first applied in the
1960’s in the nuclear, aeronautic and chemical process sectors, where great
hazards and financial losses are involved upon occurrence of unwanted events
(Bernstein, 1996). Risk-based policies that regulate hazardous activities and
installations usually focus on potential casualties or fatalities. An example of a
quantitative risk measure is societal risk: the frequency of having an accident
e.g. in an industrial plant with at least a certain number of people being killed
simultaneously (e.g. VROM, 2005; HSE, 1989). Loss of life as a result of urban
flooding may occur in cities in developing regions of the world, e.g. in South/
South-East Asia (Mark et al, 2004). In most modern cities in the industrialized
part of the world urban flooding rarely causes casualties; flooding consequences
mainly consist of damage to properties and interruption of industrial and social
processes. (e.g. Apel et al., 2008).
Besides the danger of loss of life due to certain activities, risk can be expressed in
economic terms, in a cost-benefit analysis. This allows for an evaluation of costeffectiveness of mitigation measures and thus to optimise investments (Dutta et
30
Introduction and overview
al., 2003; Nussbaum, 2006; Olsen et al. 1998). For instance, the expected value
of economic damage is used as part of cost benefit analyses for flood prevention
measures in the UK and in The Netherlands (Jonkman et al., 2003). In both
approaches the benefits of a measure are determined by calculating the expected
value of the economic damage before and after implementation of the measure.
The difference between these two values is the benefit, which can be weighed
against the costs of the measures. Cost-benefit analysis has several important
drawbacks: translation of benefits of flood risk reduction into monetary terms
requires many assumptions that are subject to uncertainty and the translation
of all costs and benefits as a result of the investment to monetary values for the
year the investment is to be made, introduces additional uncertainty (Graham,
1981).
Urban flood modelling
In current urban drainage practice, hydrodynamic models are commonly
applied to check compliance with urban flooding standards, because local
data on flood events are unavailable or incomplete. The use of hydrodynamic
models has a drawback: because modelling results are subject to uncertainty,
they can be used for comparison between design options, but are usually too
inaccurate for quantitative evaluation of historical events. If model outcomes
are to be used in an absolute sense, to evaluate compliance with standards,
models should be calibrated and verified and uncertainty in model outcomes
must be quantified. Beven and Binley (1992), Pappenberger and Beven (2006),
Mannina et al. (2006), Thorndahl et al. (2008) and many others have drawn
attention to the importance of uncertainty analysis in hydrological and urban
drainage modelling and have demonstrated the impact of uncertainties on
model outcomes. Using a calibration of runoff volumes, Schaarup-Jensen et al.
(2005) showed a remarkable difference between an uncalibrated (using default
model values) and a calibrated urban drainage model, in predicted flooding
frequencies.
31
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1
Even if data are available for model calibration, uncertainties in model structure,
model parameter assumptions and inherent uncertainty in rainfall and run-off
characteristics lead to uncertainty in model outcomes. In addition, degradation
processes like sedimentation and pipe corrosion lead to development of further
discrepancies between the real system and theoretical model conditions.
Thorndahl et al. (2008) investigated different types of uncertainties in drainage
models, e.g. uncertainties in inputs (boundary conditions), parameters, model
structure, and conceptual uncertainties. They show that even for calibrated
models predicted values can deviate quite markedly from observed values. In
particular, the models tested perform somewhat poorly in predicting peaks and
tails of flow rates, peaks being of particular importance for correct prediction
of flooding. This implies that a comparison of model-predicted flooding
frequencies to flooding standards to check compliance may easily lead to
erroneous conclusions.
Information on frequencies of flooding from manholes is not sufficient to
perform a flood risk analysis: models should provide additional information
on flood depths and flooding characteristics in order to be able to asses flood
damage. Hydrodynamic sewer models can simulate the flow in pipe networks
and the rise of water levels at manholes up to ground level. Sewer models are
not adequate to simulate surface flooding (Schmitt et al., 2004), since they are
unable to simulate the transition from pressurised pipe flow to surface flooding.
In order to simulate flooding in a realistic manner, urban flood models need to
couple the underground and above-ground systems in what is referred to as the
dual drainage concept (Djordjevic et al., 2005). Mark et al. (2004) provide an
example of one-dimensional overland flow modelling for flooding simulation.
The greatest inaccuracy of this approach lies in the approximation of flooding
in streets by one-dimensional flow paths. One-dimensional models are seen as
a good approximation as long as the water remains within the street profile
and flow paths can be well identified. Still, the outcomes of one-dimensional
flow models are sensitive to the assumptions necessarily made to translate two-
32
Introduction and overview
dimensional flow processes into a one-dimensional flow. Two-dimensional flow
models are able to capture the reality of two-dimensional overland flow to a
greater extent. The downside of these models is a large data requirement, in
particular with respect to digital terrain information, and large computational
efforts. Even though more examples of coupled one-dimensional/twodimensional flow models have recently become available (e.g. Maksimovic et
al., 2009), validation of the two-dimensional models is hampered by a lack of
calibration data (Leandro et al., 2009). Where hydrodynamic sewer models
require data on in-sewer water levels and discharges, dual drainage models
additionally need calibration data from overland flooding events. Given
the infrequent occurrence of such events and practical difficulties to set up
monitoring of overland flow characteristics, such data are difficult to obtain.
No examples of calibrated dual drainage models have been found in the
literature. Due to large data requirements and computational efforts and the
lack of calibration data, it will take time before dual drainage models obtain
sufficient reliability of application in become common practice.
Urban flood damage modelling
Flood damage can be assessed based on relationships between flooding
characteristics and expected damage. This step in flood risk analysis is
indicated by either of the terms damage assessment or vulnerability analysis.
Flood damage modelling has been frequently applied in fluvial flood risk
analyses (e.g. Dutta et al., 2003; Thieken et al., 2005 and 2008, Meyer and
Messner, 2005). Most damage assessments focus on direct flood damages that
occur within the flooded areas. Indirect damage outside the flooded area is
often estimated as a fixed percentage of direct damage (Penning-Rowsell et al.,
2005). One feature most flood damage models have in common is that the direct
monetary flood damage is a function of the type of building and inundation
depth (Jonkman et al., 2003; Wind et al., 1999). Such depth-damage functions
are seen as the essential building blocks upon which flood damage analyses
are based, and they are internationally accepted as the standard approach to
assessing urban flood damage (Smith 1994). Thieken et al. (2005) showed that
33
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1
other flood characteristics, like flow velocity and flood water contamination, are
other important factors to explain flood damage. Potential damage to buildings
is usually estimated based on building values (e.g. Gersonius et al., 2008),
replacement values of buildings, e.g. derived from economic statistics, or more
detailed assessments of repair and replacement costs (Meyer and Messner,
2005). Merz and Thieken (2005, 2009- in press) and Apel et al. (2004, 2006,
2008 and 2009) conducted uncertainty analyses for flood risk quantification
associated with urban flooding. These studies refer to fluvial flooding in hilly
terrain, return periods of over 100 years and flood depths of several meters; they
investigate both theoretical scenarios of flooding and data from a large flood
event in Cologne in 2002. Their results show, among others, that uncertainty
in depth-damage functions dominates overall uncertainty for flood damages
with a return period below 10 years. In flood scenario assessments, the type of
modelled flood event influences which sources of uncertainty dominate. They
emphasise that a prerequisite for all applications of flood risk modelling is an
accurate calibration of the model system, which includes hydrodynamic and
damage modelling.
Flooding in lowland areas
The majority of studies in the field of flood risk analysis refer to severe flooding
such as fluvial flooding and pluvial flash floods, with flood depths of several
meters. This thesis focuses on pluvial flooding in lowland areas. In lowland areas
and flat terrains, pluvial floods rarely attain large flood depths; flood waters
spread over large areas, mostly resulting in flood depths of the order of tens of
centimetres. The associated flood damage is relatively small, as is illustrated by
the following two examples. A survey among households after the 2002 fluvial
floods in Germany showed that 1273 households specified monetary damage to
their residential building contents and 1079 specified building damage; associated
mean damages amounted to €16,335 and €42,093 per property, respectively
(Thieken et al., 2005). In September and October 1998, exceptionally heavy
rainfall occurred in the northern part of the Netherlands: more than 75 mm
in 24 hours. The average return period of this rainfall event was about 125
34
Introduction and overview
years (Jak and Kok, 2000). This event was classified as a national disaster
and fell under the Dutch Compensation Act and damage-experts investigated
all damage claims. According to their assessment, 1050 households suffered
flood damage; the average damage per residential building amounted to €2000
(1999 value) and 80% of the damages were below €2200. The damage in rural
areas was much higher: total damage to agricultural companies was estimated
at €330M; 85% of the total flood damage. This example illustrates that even
for a rare rainfall event, urban pluvial flood damage in lowlands remains small
compared to fluvial flooding. As a result of smaller damages per event, higher
flood event frequencies are generally accepted in lowland areas. In this thesis
an attempt is made to asses how the cumulative damage of flood events over the
lifetime of urban drainage systems compares between lowland areas with high
flooding frequencies and areas with lower flooding frequencies. A particular
difficulty in assessing flood risk in lowland areas is that intangible damage such
as traffic delay and inconvenience for pedestrians constitutes an important part
of total flood damage. These kinds of damage cannot easily be expressed in
monetary values which makes it difficult to assess total cumulative damage.
1.3
Flood event data
In this study data from two case studies in lowland areas were used; the cities of
Haarlem and Breda. Both represent medium-size cities of 147,000 and 170,000
inhabitants. One city is located in the western part of the Netherlands, in a
transition area between sand dunes and clay polders. The other is located in the
south of the Netherlands, on a transition between sandy soils and clay polders.
Ground levels vary mostly between 0m and 10m above Mean Sea Level, with
maximum ground level variations up to 20 meters. Both cities are primarily
served by gravity systems that are connected to a treatment plant by a pumping
station at the downstream end of the system. Figure 1.1 gives an overview of
construction periods of the urban drainage systems in both cities.
35
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32






R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1




















 

Figure 1.1 Constructed sewer lengths per 10-year-period for the urban drainage
systems of Breda and Haarlem.
Figure 1.2 shows the location of the two case studies in the Netherlands; table
1.1 presents a summary of urban drainage system characteristics for the two
cases.
Table 1.1 Characteristics of the urban drainage systems of Haarlem and Breda.
Urban drainage system characteristics
Number of inhabitants
Ground level variation
Storage in combined system below lowest overflow weir
Total length of gravity sewers
Total residential area
Total impervious area
Unit
m
m3
km
km2
km2
Haarlem
147,000
20
72,000
463
32
12
Breda
170,000
15
100,000
736
70
18
The primary data used in this thesis consist of data from municipal call centres
that register information on urban drainage problems observed by citizens. Call
 data are available for a period 10 years for Haarlem and 5 years for Breda.
Most calls refer to problems of flooding, ranging from local flooding on a
36


Introduction and overview
road or parking lot to flooding of entire streets and flooding inside residential
and commercial buildings. Since call texts describing citizens’ observations
provide information on time, location and characteristics of flooding, they
constitute a detailed series of flood event data. The advantage of these data is
that registration took place during or shortly after flood events which limits
distortion of the data by after-event actions and experiences. Since in lowland
areas flood frequencies are relatively high, time-series of historical flood event
data of 5 to 10 years are sufficient to obtain useful analysis results. In addition,
daily local rainfall measurements are available for both cases, which were
used to cross-check flood event data and rainfall characteristics. Call data and
rainfall data were used in various approaches of quantitative risk analysis.





Figure 1.2 Map of the Netherlands, Haarlem and Breda; locations of rain gauges in
Haarlem, H1 (Leiduin) and H2 (Schiphol) and in Breda, P1 (Prinsenbeek). Source:
Google maps, 2009
37

R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1
1.4
Thesis overview
This introduction chapter describes the context of urban flood management
and current research in the field of urban flooding, urban flood risk and flood
damage modelling. It identifies a number of shortcomings in existing approaches
and addresses some specifics of urban flooding in lowland areas. Based on these
observations, the main objective addressed in this study is formulated in the
form a central research problem and four research questions that guided the
research.
Chapter 2 presents a fault tree analysis for urban water infrastructure flooding.
It identifies possible mechanisms in urban water infrastructure that can lead to
flooding and their relative contributions to flood probability. While the focus in
urban flooding analysis is generally on fluvial flooding and flash floods caused
by heavy rain, this chapter compares the contribution of heavy rainfall to that
of other failure mechanisms for urban flooding.
In chapter 3 data from municipal call centres are explored to find out whether
they can be used to quantify urban flood risks associated with various possible
failure mechanisms. A data-driven approach based on historical data-series
of flooding events is used to quantify various types of consequences of urban
flooding. The results are presented in the form of risk curves that show the
probabilities of exceedance of a range of flooding consequences.
The primary function of urban drainage systems is to protect public health by
preventing contact with pathogens in wastewater. Chapter 4 investigates the
potential health risk to citizens of urban flood waters resulting from combined
sewer flooding, based on a screening-level microbial risk assessment.
Chapter 5 presents an attempt to translate tangible and intangible flooding
consequences into two types of common metrics in order to directly compare
their distribution to total flood risk. It compares the results for the two different
38
Introduction and overview
metrics and shows how the choice of metrics influences risk analysis outcomes,
hence the decisions based on these. The cumulative contribution to flood risk
of small flood events is compared to the contribution of rare, severe events to
address the question of whether severe events should get priority in flood risk
management over series of small events, or not.
Chapter 6 shows how flood risk analysis results can be used to evaluate
the efficiency of operational strategies and to identify efficient ways for
improvement. Three causes of flooding and associated flood management
strategies are compared and opportunities to enhance current strategies to
further reduce flood risk are highlighted.
Chapter 7 discusses the contribution of this research to current understanding
of urban flooding and urban flood management. Recommendations for further
studies are given as well in this chapter
Figure 1.3 shows how chapters interrelate.



















Figure 1.3 Relations between chapters in this thesis.
39
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1
1.5
References
Apel, H., Aronica, H., Kreibich, H., Thieken, A.H. (2009). Flood risk analyses – how
detailed do we need to be? Natural Hazards, 49, 79-98.
Apel, H., Merz, B., Thieken, A.H. (2008). Quantification of uncertainties in flood risk
assessments. International Journal of River Basin Management, 6(2), 149-162
Apel, H., Thieken, A.H., Merz, B., Blöschl, G. (2006) A probabilistic modelling system for
assessing flood risks. Natural Hazards, 38 (1-2), 79-100
Apel, H., Thieken, A.H., Merz, B., Blöschl, G. (2004). Flood risk assessment and associated
uncertainty. Natural Hazards and Earth System Science, 4 (2), pp. 295-308
Ashley, R.M., Balmforth, D.J., Saul, A.J. and Blanksby, J.D. (2005). Flooding in the future
- predicting climate change, risks and responses in urban areas. Water Science and
Technology. 52 (5): 265-273.
Ashley, R., Blanksby, J., Cashman, A. (2007). Building Knowledge for a Changing Climate.
Adaptable Urban Drainage - Addressing Change In Intensity, Occurrence And
Uncertainty of Stormwater (AUDACIOUS). Main report, draft 1.
Bernstein, P.L. (1998). Against the Gods: the remarkable story of risk. John Wiley and
Sons, New York, 1996.
Beven K.J. and Binley, A.M. (1992). The future of distributed models: model calibration
and uncertainty prediction. Hydrological Processes, 6, pp. 279–298.
Brown, S.A., Schall, J.D., Morris, J.L., Doherty, C.L., Stein, S.M., Warner, J.C. (2009).
Urban Drainage Design Manual Hydraulic Engineering Circular 22, Third Edition.
Publ. nr. FHWA-NHI-10-009. Federal Highway Administration, Washington, D.C.
CEN, 2008. Drain and sewer systems outside buildings. EN 752:2008: E. European
Committee for Standardisation, Brussels, Belgium.
Djordjević, S., Prodanović, D., Maksimović, C., Iveticć, M., Savić, D. (2005). SIPSON
simulation of interaction between pipe flow and surface overland flow in networks.
Water Sci. Technol. (52(5), 275-283.
Directive 2007/60/EC of the European parliament and of the council on the assessment and
management of flood risks, 23 october 2007.
Dutta, D., Herath, S., Musiake, K. (2003). A mathematical model for flood loss estimation.
Journal of Hydrology 277 (2003) 24–49.
Dutta, D., Herath, S., Wijesekera, S. (2003). Development and application of an urban
flood risk analysis system, Proceedings of the First International Conference of AsiaPacific Hydrology and Water Resources Association, Kyoto, Japan, pp. 821-826,
March 2003.
Gersonius, B., Zevenbergen, C., Puyan, N., Billah, M.M.M. (2008). Efficiency of private
flood proofing of new buildings - Adapted redevelopment of a floodplain in the
Netherlands. WIT Transactions on Ecology and the Environment, 118, pp. 247-259.
Graham, D.A. (1981). Cost-Benefit Analysis Under Uncertainty. The American Economic
Review, 71(4), 715-725.
Hallegatte S (2008) An adaptive regional input-output model and its application to the
assessment of the economic cost of Katrina. Risk Analysis 28(3), 779-799.
Health and Safety Executive (1989). Risk criteria for land-use planning in the vicinity of
major industrial hazards, HM Stationery Office.
Jak, M., Kok, M. (1999). A database of historical flood events in the Netherlands. In: Flood
Issues in Contemporary Water Management, 139-146. Proceedings of the NATO
Advanced Research Workshop on Coping with Flash Floods: Lessons Learned from
Experience, Malenovice, Czech Republic, May 16-21, 1999. Series: NATO Science
Partnership Sub-Series: 2: , Vol. 71. Marsalek, J.; Watt, W.E.; Zeman, E.; Sieker,
F. (Eds.).
40
Introduction and overview
Jonkman, S.N., van Gelder, P.H.A.J.M., Vrijling, J.K. (2003). An overview of quantitative
risk measures for loss of life and economic damage. Journal of hazardous materials,
A99, 1–30.
Kaplan, S. (1997). The words of risk analysis. Risk Analysis, 17(4), pp. 407-417
Kaplan, S., Garrick, B.J. (1981). On the quantitative definition of risk. Risk Analysis, 1(1),
pp. 11-27.
Korving, H., Van Noortwijk, J.M., Van Gelder, P.H.A.J.M., Clemens, F.H.L.R. (2009).
Risk-based design of sewer system rehabilitation. Structure and Infrastructure
Engineering, 5(3), 215 – 227.
Leandro, J., Djordjević, S., Chen, A.S., Savic, DA. (2009). A comparison of 1D/1D and
1D/2D coupled hydraulic models for urban flood simulation. Journal of Hydraulic
Engineering, 135(6), pp. 495-504.
Maksimović, C., Prodanović, D., Boonya-Aroonnet, S., Leitaõ, J.P., Djordjević, S., Allitt, R.
(2009). Overland flow and pathway analysis for modelling of urban pluvial flooding.
Journal of Hydraulic Research, 47(4), pp. 512-523.
Mannina, G., Freni, G., Viviani, G., Sægrov S., and Hafskjold, L.S. (2006.) Integrated
urban water modelling with uncertainty analysis, Water Science and Technology 54
(6-7), 379-386.
Mark, O., Weesakula, S., Apirumanekula, C., Boonya-Aroonnet, S., Djordjević, S. (2004).
Potential and limitations of 1D modelling of urban flooding. Journal of Hydrology
299 (2004) 284–299.
Merz, B., Thieken, A.H. (in press). Flood risk curves and uncertainty bounds. Natural
Hazards (2009, in press).
Merz, B., Thieken, A.H. (2005). Separating natural and epistemic uncertainty in flood
frequency analysis. Journal of Hydrology, 309(1-4), 114-132.
Merz, B., Kreibich, H., Thieken, A., Schmidtke, R. (2004). Estimation uncertainty of direct
monetary flood damage to buildings. Natural Hazards and Earth System Science, 4
(1), pp. 153-163
Meyer, V., Messner, F., (2005). National Flood Damage Evaluation Methods - A Review
of Applied Methods in England, the Netherlands, the Czech Republic and Germany.
UFZ-Diskussionspapiere 21/2005.
Natural Environment Research Council (NERC). Website: www.nerc.ac.uk.
Nussbaum, R. (2006). The cost and financing of flood damages. Houille Blanche, 6(2006),
55-60 Olsen, J.R., Beling, P.A., Lambert, J.H., Haimes, Y.Y.
Olsen, J.R., Beling, P.A., Lambert, J.H., Haimes, Y.Y. (1998). Input-output economic
evaluation of system of levees. Journal of Water Resources Planning and
Management. Volume 124(5), 237-245.
Pappenberger, F., Beven, K. J. (2006). Ignorance is bliss: Or seven reasons not to use
uncertainty analysis. Water Resour. Res., 42, W05302, doi:10.1029/2005WR004820.
Penning-Rowsell, E.C. E., Johnson, C., Tunstall, S., Tapsell, S., Morris, J., Chatterton, J.,
Coker, A., Green, C. (2005). The Benefits of Flood and Coastal Risk Management: A
Manual of Assessment Techniques. Enfield, Middlesex University Press.
Pilgrim, D. H (2001). Australian rainfall and runoff : a guide to flood estimation. Book VIII
Urban drainage systems. Institution of Engineers, Australia
Rauch, W., Bertrand-Krajewski, J.-L.,Krebs, P., Mark, O., Schilling, W., Schütze, M.,
Vanrolleghem, P.A. (2002). Deterministic modelling of integrated urban drainage
systems. Wat. Sci. & Tech., 45(3), 81-94.
RIONED Foundation, 2007. Analysis of pluvial flooding problems in the built environment
(in Dutch). Stichting RIONED, Ede, the Netherlands.
Schaarup-Jensen, K., Johansen, C., Thorndahl, S. (2005). Uncertainties related to extreme
event statistics of sewer system surcharge and overflow, 2005, Proceedings of 10th
International Conference on Urban Drainage, Copenhagen, Denmark August 21-26.
41
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 1
Schmit, T.G., Thomas, M., Ettrich, N. (2004). Analysis and modeling of flooding in urban
drainage systems. Journal of Hydrology 299 (2004) 300–311.
Schreider, S.Y., Smith, D. I., Jakeman, A. J. (2000). Climate change impacts on urban
flooding. Climate change, 47 (1-2).
Thieken, A., Müller, M., Kreibich, H., Merz, B. (2005). Flood damage and influencing
factors: New insights from the August 2002 flood in Germany. Water resources
research, 41(12), 1-16.
Thorndahl S., Willems, P. (2008). Probabilistic modelling of overflow, surcharge and flooding
in urban drainage using the first-order reliability method and parameterization of
local rain series. Water Research, 41, 455-466.
Thorndahl, S., Beven, K.J., Jensen, J.B., Schaarup-Jensen, K. (2008). Event based
uncertainty assessment in urban drainage modelling, applying the GLUE
methodology. Journal of Hydrology, 357, 421– 437.
Van Mameren, H., Clemens, F.H.L.R. (1997). Guidelines for hydrodynamic calculations
on urban drainage in the Netherlands: overview and principles. Water Science and
Technology, 36(8-9), 247-252.
Vesely, W., Dugan, J., Fragola, J., Minarick, J., Railsback, J. (2002). Fault Tree Handbook
with Aerospace Applications. Version 1.1. NASA Headquarters, Washington.
Uijt de Haag, P.A.M. and Ale, B.J.M. (2005). Guidelines for quantitative risk assessment.
Publication Series on Dangerous Substances (PGS 3), VROM, Den Haag.
Wilby, R.L. (2007). A Review of Climate Change Impacts on the Built Environment. Climate
change and cities, 33(1), 31-45.
42
Introduction and overview
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
43
Chapter 2
Quantitative fault tree analysis
for urban flooding
This chapter is based on an article that was published in Structure and Infrastructure
Engineering.
Veldhuis, J.A.E. ten, Clemens, F.H.L.R. and Gelder, P.H.A.J.M. van (2009).
Quantitative fault tree analysis for urban water infrastructure flooding. Structure
and Infrastructure Engineering, 99999:1, DOI: 10.1080/15732470902985876.
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
Context
The most common way to apply urban flood risk analysis is to determine
expected flood depths and flood extensions by means of some form of
hydrodynamic model calculations in order to assess the number and type
of properties at risk of flooding. This information is used to assess expected
damage and to decide whether flood reduction is required. This approach has
been developed for river and coastal flooding to quantify flood risk associated
with high water levels that leads to failure of dikes, river levees and other flood
protection structures. In the past decades a similar approach is applied to urban
pluvial flood risk analysis. The underlying assumption is that heavy rainfall
followed by overloading of urban drainage systems is the main cause of urban
pluvial flooding. Consequently it assumed that modelling the effects of system
overloading by heavy rainfall and quantifying associated flood risk, fully
captures urban pluvial flooding problems. Still, overloading by heavy rainfall
is only one of the possible failure mechanisms of urban drainage systems. This
chapter identifies other possible failure mechanisms of urban drainage systems
in a fault tree analysis and quantifies their contributions to overall flood
probability.
Abstract
Flooding in urban areas can be caused by heavy rainfall, improper planning or
component failures. Few studies have addressed quantitative contributions of
different causes to urban flood probability. In this chapter, probabilistic fault
tree analysis is applied to assess the probability of urban flooding as a result of
a range of causes. Causes are ranked according to their relative contributions.
To quantify the occurrence of flood incidents for individual causes, data
from municipal call centres were used, complemented with rainfall data and
hydrodynamic model simulations. Results showed that component failures
and human errors contribute more to flood probability than sewer overloading
by heavy rainfall. This applies not only to flooding in public areas but also
to flooding in buildings. Fault tree analysis has proved useful in identifying
relative contributions of failure mechanisms and providing quantitative data
for risk management.
Keywords: fault tree; flooding; risk; urban drainage
46
Quantitative fault tree analysis for urban flooding
2.1.
Introduction
Over the last few decades the interest in urban flood risk has been growing
steadily, as the frequency of flooding and the damage caused by urban flood
events have increased (Ashley et al., 2005). They state that accelerated
urbanisation has given rise to increased building in unsuitable areas and
expansion of impervious areas, both adding to the inflow into existing urban
drainage systems and thus to the probability of flooding. In addition, climate
change predictions increase concern over urban flood risk (Semadeni-Davies
et al., 2008). In the UK, the problem of urban flood risk has been addressed
in many studies. A baseline estimate of the current urban pluvial flood risk in
England and Wales concluded that the expected annual damage to residential
and commercial properties in urban areas amounts to ₤ 270 million (Ashley,
2006). Some 5000-7000 properties are flooded annually in England and Wales
by sewage (Ashley et al., 2005). No quantitative estimation studies of urban
flood risk in the Netherlands were found, not in general nor for specific cases.
Principal causes of flooding addressed in urban flood studies are heavy storm
events that lead to overloading of rivers and urban water infrastructures.
In addition, urban water systems are susceptible to component failures and
human errors. Analysis of call centre data from two municipalities of 10,000
to 170,000 inhabitants in the Netherlands has shown that hundreds of small
flood events occur each year in relation to these causes. Material damage to
private properties, local disturbance of urban traffic and nuisance for cyclists
and pedestrians are common consequences.
Quantification of flood risk requires data on flood incidents related to the
complete spectrum of potential causes. Additionally a methodology is needed to
quantify flood probabilities and consequences. A number of methods have been
developed in high-risk industries, such as the nuclear, aeronautic and chemical
industries, to quantify risk, including risk analysis methods and probabilistic
fault tree analysis (Kaplan and Garrick, 1981; Haimes, 1998; Vesely et al.,
47
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
1981 and 2002). Risk-based decision making in water resources matured as a
professional niche in the US in the 1980’s (Haimes, 1998). These methods have
been successfully applied in river flooding (Vrijling, 2001), but application to
urban drainage systems remains rare. In the UK, urban flood risk assessment
and management have received much attention recently and the approach
has been applied to several cases in the UK (FRMC, 2007). Probabilistic
techniques have had applications in urban drainage in research projects in
Denmark (Harremoes and Carstensen, 1994) and Belgium (Thorndahl and
Willems, 2008), amongst others.
Quantitative fault tree analysis is an example of a risk analysis technique that
effectively detects potential failure mechanisms and quantifies probabilities of
failure of complex systems based on failure data. A fault tree is a deductive
model that links a systems failure via reverse paths to all subsystems,
components, human errors etc. that can contribute to failure. It is very useful
to detect potential causes of flood events including both hydraulic overloading
and component failures. It quantifies both overall flood probability and
the relative contributions of individual causes of flooding based on their
probabilities of occurrence. The Fault Tree Handbook NUREG-0492 issued
by the US Nuclear Regulatory Commission in 1981 has been a leading technical
information source for fault tree analysis in the USA (Vesely et al.). In 2002
NASA issued a handbook for aerospace applications that contains additional
information on recent techniques (Vesely et al., 2002). Both handbooks also
provide a short overview of other approaches to the logical modelling of system
failure, e.g. failure mode and effect analysis and fault hazard analysis. Ang and
Tang provide a short introduction for applications in the field of structural
engineering (Ang and Tang, 1984).
In this chapter quantitative fault tree analysis is applied to urban flooding,
defined in this context as the occurrence of pools in an urban area. Quantitative
fault tree analysis is applied to the cases of two cities in the Netherlands, Haarlem
and Breda. These cities have urban drainage systems with a total length of 460
48
Quantitative fault tree analysis for urban flooding
and 1000 km that mainly consist of gravity sewers. Data from municipal call
centres, rain gauges and hydrodynamic model calculations are used to quantify
the probabilities of various causes of urban flooding.
Uncertainties in urban flood risk quantification are high due to a lack of
incident data registration for small incidents, that often pass unnoticed, and
low probabilities of large incidents so that long periods of data collection are
required to obtain sufficient data for risk quantification. Also attention tends to
focus on flood damage relief more than on data registration.
2.2.
Urban flood incident data
To quantify probabilities for fault tree events, data on flood incidences must be
collected. Potential sources of flood incident data are monitoring networks, call
centres, hydrodynamic models, fire brigade records and the media.
Monitoring networks in urban drainage systems can provide flood incident
information, if they have sufficient spatial density to detect all flood events
throughout urban areas. In practice, monitoring locations are limited to
pumping stations, overflow weirs and some additional points e.g. at special
constructions. This density is largely insufficient to register in detail all flood
incidents in an urban area.
Municipal call centres register call information on flood incidents. Incidents
that are sufficiently annoying to prompt citizens to make a call are recorded
in the call register. The network of callers is potentially very dense since every
citizen can be assumed to have access to a telephone. Still, calls do not give
complete coverage of flood incidents, because there is no guarantee that a call is
made for every event. It is on the other hand one of the best sources to provide
indication of events unacceptable to citizens.
49
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
Call data consist of a unique call number, date of the call, street name to indicate
the problem location and a telegram style text that describes what the caller has
said. In most cases a second text is added that describes the results of on-site
checking and actions undertaken to solve the problem. Call databases usually
contain categories that calls are assigned to and give an indication of the reason
a call was made. To be able to use call information for flood risk analysis these
categories are not specific enough and calls must be screened and classified
manually.
Data on flood events can also be derived indirectly from simulations of urban
drainage system behaviour under various rainfall conditions. One-dimensional
sewer models simulate flow through piped systems and can provide estimates
of flooding as a result of system overloading during heavy rainfall. Also pipe
blockages can be simulated, but flood estimates remain theoretical unless reallife data on occurrence of blockages are available to be used as input. The
description of inflow processes in these models is not sufficiently accurate
to provide estimates of flood incidents due to gully pot blockages, manifold
blockages and surface obstacles.
Overland flow models are developed and coupled with sewer models to support
quantification of expected consequences of flooding as a result of sewer overload
(e.g. Djordjevic et al., 2005).
Although hydrodynamic models can provide insight into expected flow paths
and flood frequencies, their use for probabilistic analysis is not straightforward.
Probabilistic analysis can be applied to rainfall data to compose design storms
with expected probabilities of occurrence that are fed into hydrodynamic
models. Expected rainfall probabilities must in some way be translated
into flood probabilities, which can be done for simple systems with linear
hydraulic behaviour, but becomes highly complicated for large, complex
systems. Alternatively probabilistic analysis can be applied to hydrodynamic
model results for long rainfall series of 10 or 25 years or more. This demands
long calculation times and a large amount of data storage and extensive data
50
Quantitative fault tree analysis for urban flooding
analysis. Additionally hydrodynamic models are subject to uncertainties and
tend to focus on hydraulic capacities of systems as designed or ‘as built’, having
difficulty with deviations caused by component failures. Some examples are
available where the vulnerability of model outcomes to component failures
and data uncertainties is assessed (Clemens, 2001) that show the complex
manipulations needed to obtain intended calculation results.
Other sources of flood incident information that have been investigated are
newspaper articles and on-line pages and fire brigade action records. The
Dutch Central Bureau of Statistics compiles yearly data on fire brigade actions
related to flooding. These data show that fire brigades in the Netherlands
assisted in between 2671 and 5540 cases of flooding yearly between 1994 and
2005. 80% of these cases concern flooding in buildings and 20% other than
buildings. Fire brigade records contain no information on the nature and cause
of flooding. Flooding in buildings for instance can be related to street flooding
or to burst drinking water mains inside buildings, high groundwater tables or
malfunctioning of rain pipes or in-house sewers. This lack of detail makes this
source of information unsuitable for fault tree analysis. News paper articles
often describe flood situations in detail, but newspaper reporting is selective:
calamitous events and events that in other ways disturb life in local communities
are likely to reach the newspapers, less striking events are not. Therefore this
information source has been discarded.
In this study, model simulations have been used to validate data from municipal
call centres by comparison of locations with frequent calls on flooding with flood
locations in simulation results for heavy rainfall conditions. In addition rainfall
data and calls have been compared directly for some logical checks: do calls on
flooding coincide with rain events and if not, is there a good explanation? Do
heavy rain events generate more calls than light events? Do calls that indicate
sewer overloading coincide with heavy rainfall events?
51
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
2.3.
Quantitative fault tree model for urban flooding
Definition of failure mechanisms
To explore what incidents can give rise to urban flooding a source-pathwayreceptor representation has been used to analyse urban water infrastructure
systems. Figure 2.1 shows a block diagram that represents the components of
such systems and their interconnections. Possible sources of water occurring
on urban surfaces are rainfall, river water that has flown over river banks,
drinking water e.g. from a burst pipe, groundwater that rises above ground
level and discharges e.g. from construction sites where groundwater abstraction
takes place. Under normal conditions, water on urban surfaces evaporates
or infiltrates or flows over the surface to an infiltration or storage facility or
a sewer system. Sewer systems transport water towards a treatment facility
or a pumping station. In case the hydraulic capacity of a pumping station or
treatment facility is insufficient to cope with the flow, water passes over a sewer
overflow to surface water. Surface water and groundwater are final receptors
in this system.
Flooding can occur when flow pathways are interrupted as a result of failing
system components. In branched systems interruption of a flow route leads
to flooding immediately or as soon as the storage capacity upstream of a
failed component is filled. In looped networks alternative flow routes are
available when one flow route gets blocked, which makes these networks less
vulnerability to component failures. Here the hierarchy of system elements is
important: failure of components in a main transport route is likely to cause
failure while failure in secondary routes can be compensated by alternative
routes. Pathway interruption also occurs due to errors during the design and
construction phase, e.g. when components are omitted, like gully pots that are
not connected to a sewer system.
Another mechanism that leads to urban flooding is system overload: when
water inflow exceeds the storage and transport capacity of one or more system
52
Quantitative fault tree analysis for urban flooding
elements. Normally urban drainage systems are designed to cope with weather
conditions up to a certain limit and overloads occur several times during a
system’s lifetime.
Figure 2.1. Block diagram for an urban drainage system. The diagram shows the system
components that, by their failure, can lead to the occurrence of water on urban areas.
Construction of fault tree model
The objective of fault tree analysis is to identify all possible failure mechanisms
that can lead to urban flooding in a systematic way. There are four basic
elements in the development of a fault tree: top event, basic events, AND gates
53
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
and OR gates (figure 2.2). The top event of a fault tree is the failure that is
subject of analysis, urban flooding in this case. Urban flooding is defined here
as the occurrence of a pool of water on the surface somewhere in an urban area
lasting long enough to be detected and cause disturbance. This includes the
appearance of water on the surface as a result of rainfall that is not properly
drained and of water that flows out of the drainage system onto the surface
due to a particular component failure. These failure mechanisms are analysed
in detail whereas the occurrence of pools on the urban surface due to failure
of other urban water systems: drinking water, groundwater or surface water,
are included in the fault tree, but not analysed in detail here. Basic events form
the most detailed level of a fault tree and stand for failures or conditions that
can be combined by AND or OR gates to create higher level states. The choice
of the basic level of a fault tree depends on the level of detail that is required
for a specific analysis. The AND gate links underlying events that must occur
simultaneously for the output condition to exist, while the OR gate generates
the output condition for any one of the underlying events.
Figure 2.2 Elements of a fault tree model
54
Quantitative fault tree analysis for urban flooding
In a systematic analysis seven failure mechanisms have been found that can give
rise to urban flooding, three of which are related to urban drainage systems:
1) Inflow route interruption: rainwater that falls on an urban surface
cannot flow away to a drainage facility and as a result forms pools on
the surface;
2) Depression filling: Rainwater that has fallen at an upstream location
flows over the surface to a downstream location where it cannot enter
a drainage facility but remains on the surface;
3) Sewer flooding: Water from the sewer system flows onto the surface
due to local system overload or downstream component failure;
4) Drinking water leakage: Drinking water flows onto the surface as a
result of a pipe burst or a leaking hydrant;
5) Groundwater flooding: groundwater table rises above ground level;
6) Surface water flooding: Surface water levels rise above bank levels or
overflow weir levels and surface water flows onto the surface directly
or via an urban drainage system;
7) External water discharge: An amount of water is discharged onto the
surface, e.g. extracted groundwater from a construction site or water
from a swimming pool that is replenished.
Figure 2.3 shows a fault tree for urban flooding for these 7 mechanisms. The
intermediate events form a first level in the tree; they in their turn result from
other events. Four events are included as undeveloped events since they will
not be analysed in detail. An “OR-gate” connects the top event to this first level
of events because occurrence of each individual event results in flooding.
55
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
Figure 2.3 Example of a fault tree model for urban flooding, first level. Three events are
to be developed deeper, to the level of basic events; four events remain undeveloped.
The ‘OR’ gate indicates that each individual intermediate event can lead to the top
event.
Inflow route interruption includes blockage of gutters, gully pots, gully pot
manifolds and high road verges that prevent water flow from a road surface
to adjacent green areas. Also absence of gutters, gully pots or manifolds
is included here. The second mechanism, depression filling is particularly
important in steep catchments where water rapidly runs down a slope and
fills up depressions at the bottom if no drainage facilities are available. When
facilities are available, flow pathways and potential failures become identical
to the inflow route interruption mechanism. Depression filling is different in
this respect that water that ends up in a depression comes largely from other,
upstream areas. The sewer flooding mechanism occurs when water reaches a
sewer system, but cannot enter because the system is full, or, in hydraulic terms,
the hydraulic gradient in the system is at or above ground level. This can be due
to system overload or to partial or complete blockage of components. Sewer
flooding also includes the mechanism where water has already entered a sewer
system and flows onto the surface due to rise of the pressure level above ground
level. A detailed fault tree for these failure mechanisms has been developed and
is available upon request.
56
Quantitative fault tree analysis for urban flooding
Quantitative fault tree analysis
Quantitative analysis of a fault tree provides the probabilities of occurrence
of basic events and of the top event. It also gives quantitative rankings of
contributions
of basic events to the top event. A failure probability model must



be chosen that suits the type of failure processes in the fault tree. In this analysis

the
occurrence of events is assumed to be a Poisson process, which implies that



the
probability that an event will occur in any specified short time period is



approximately
proportional to the length of the time period. The occurrences




of 
events in disjoint time periods are statistically independent. Under these



conditions, the number of occurrences x in some fixed period of time is a



Poisson
distributed variable:






    


           




 (2.1)


  

    









    





 
x

in a period
 of time
 t 



    

Where:
: probability
of
occurrences




   
λ
:
average rate of occurrence of events per time unit 
 



     

 


 



The
rate of occurrence λ is derived from failure data over a certain period of


time.
In a homogeneous Poisson process, the event occurrence rate λ is constant.




In 
a nonhomogenous Poisson process, λ is modelled as a function of time; this




model is useful to analyse trends, e.g. due to ageing processes. In this fault tree


 been assumed.
analysis a constant failure rate has


                  






Since
occurs
   occurrence of 1 or more events, the probability

   failure
   due
   
 to
 the









 
  be
 calculated

of 
failure can
from:









  


































(2.2)


 
     



Where: 
: probability
of one or more events
 





         
of no events
     : probability


              


              




chosen








The
time
period
tcan
be
atwill;
the
longer
t, the
higher
probability



 



the


 



 


so


 
of 
occurrence.
The
time
scale is preferably
chosen
as 
to fit the
frequency
of 

 

 
 
 

  














events. In the case of urban flooding flood events typically occur up to several



per 
month
and
the duration
of events
is in
the
order
of several
A 




 days.
 
times





           


 


 

  




 

 
57


         


          



 
 



 

 
 
 


 
















             


 

 


  
 
 

 

 


R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
time period of 1 week fits the event occurrence frequency and has been chosen
for the fault tree analysis of urban flooding.
This quantitative fault tree model is based on fixed probabilities of occurrence
of the basic events. The model can be developed further into a stochastic fault
tree model such as Reliability Block Diagrams or Dynamic fault trees in which
functional dependencies and fault-ordering is included. These extensions can
be subject for future study. The focus of this studyxx is primarily towards fault
tree modelling.
Independent events
Probabilistic fault tree analysis is more straightforward if successive events are
independent because probability distributions like the Poisson distributions are
only applicable on this condition. Successive flood events are independent if
the total urban drainage system has returned to its initial conditions between
two events. This includes all system components: pipes, basins, surfaces surface
infiltration capacity etc.
In practice usually insufficient data are available to check whether initial
conditions have been restored. A safe and practical assumption has been made
to separate independent events for this fault tree analysis. The main source
of urban flood water being rainfall, first a criterion has been defined for
independence of rain events. It is based on the length of the intermediate dry
period which must be sufficiently long to allow the drainage system to come
back to initial conditions. This period is typically in the order of 10 to 15 hours.
The intermediate period must not be longer than 24 hours because extremely
long events, in the order of several weeks, would result. This exceeds the
minimum return period of flood events and thus distorts probabilistic analysis.
Even though initial soil conditions may not have been entirely restored after 24
hours, the relative influence on system storage capacity is expected to be minor.
In addition it is assumed that blockages that give rise to flood incidents are
removed before the start of a new event, to assure independence of successive
58
Quantitative fault tree analysis for urban flooding
blockage events. Given that call data are used as data source for blockage
incidents, it is likely that problems are solved within a short time after calls are
made, since this is the main purpose of municipal call centres.
The identification of a criterion for the spatial independence of events is
less straightforward. Since hydraulic relationships control the flow patterns
throughout sewer systems, flood events at separate locations are likely to be
dependent. For this reason it is more convenient to evaluate the fault tree
model for an urban drainage system as a whole. In that case the fault tree model
provides probabilities of flood incidents on system level.
The number of flooded locations per event is used to quantify the consequences
of individual flood events and this information is combined with probabilities
to quantify flood risk. Flood risk, as defined in the European Flood Risk
Directive means the combination of the probability of a flood event and the
potential adverse consequences for human health, the environment, cultural
heritage and economic activity associated with a flood event (EU, 2007). Other
information on the extent of the flooding, if available, can be added to quantify
the consequences. There is no longer a need to define a criterion to separate
events at different locations, because consequences can be calculated on a
gradual scale.
59
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
2.4.
Results of quantitative fault tree analysis for two case
studies
Case studies characteristics and available data
The quantitative fault tree model has been applied to two case studies,
Prinsenbeek, a district in the city of Breda, and Haarlem. A municipal call
register, local rainfall measurements and a hydrodynamic sewer model are
available for both cases. Table 2.1 presents a summary of urban drainage system
characteristics for the two cases. Both are gravity systems that are connected
to a treatment plan by a pumping station at the downstream end of the system.
Table 2.1 Characteristics of the urban drainage systems of Prinsenbeek and Haarlem
Urban drainage system characteristics
Number of inhabitants
Ground level variation
Storage in combined system below lowest overflow weir
Maximum time needed to empty a full system storage
after rainfall: system storage/minimum capacity available
to pump rainwater
Total length of gravity sewer pipes (% combined)
Total residential area
Total impervious area (estimation in year)
- impervious area connected to combined system
- impervious area connected to separate system (% area
where 1st flush pumped to combined system)
Unit Prinsenbeek Haarlem
11,000 147,000
m
1
20
4700
72000
m3
hour
7.5
24
km
%
km2
km2
km2
km2
%
53.3
95
1.75
1.01
0.86
0.15
60
460
98
32
12.25
8.88
2.22
-
Call data are the most important data source to provide estimates of flood
incidents as a result of basic fault tree events. Call data are registered in both
cases by call centres that are part of the municipality. Call centre numbers are
made known to citizens through information brochures and occasional public
information campaigns. Calls are recorded in telegram style upon receipt; a text
reporting findings of on-site checking of the call is added within a few days, up
to a maximum of two weeks after the call. Call texts are analysed manually and
every call is assigned to a one of a list of classes that correspond with basic fault
tree events. A small number of call texts, about 1%, refer to more than 1 type
60
Quantitative fault tree analysis for urban flooding
of basic event; these calls are assigned to the various corresponding classes.
To check the reliability of call data, heavy rainfall incident frequencies derived
from call centre data are compared with those resulting from model simulations.
Also, frequent flood locations are compared. Every heavy rainfall incident that
results in flooding according to model simulations is reported by at least 1
call, in the call register. Most locations that suffer frequent flooding in model
simulations are reported in the call register as well. Only a number of locations
in Haarlem that in model simulations experience a high frequency of flooding
do not occur in the call register: these locations are situated in an industrial area
and are either not reported or the large impervious areas on private industrial
grounds are not well represented in the model so that in reality flood incidents
have a far lower frequency. Table 2.2 provides a summary of available call data
and rainfall data for the two cases studies.
Two different analyses have been conducted for the two case studies: for
Prinsenbeek, the sewer flooding failure mechanism has been analysed (figure
2.3, 2nd failure mechanism from left in fault tree) and for Haarlem the entire
fault tree has been analysed, except for depression filling because no data on
this mechanism are found in the call register.
61
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
Table 2.2 Data sources and characteristics case studies Prinsenbeek and
Haarlem
Municipal call registers
Period of call data
Prinsenbeek
31-07-2003 to
17-10- 2007
996
Total nr. of calls1 in urban-water call
category
Length of data series
1720 days
Rain gauges
Location of rain gauges (see also: figures 2.4 1 rain gauge in
and 2.5)
Prinsenbeek
Period of rainfall data
01-01-2002 to
31-10-2007
Time interval
5 minutes
Hydrodynamic sewer model
Simulated events:
Rainfall series
from local weather
station: 01/01/200231/10/2007
Haarlem
12-06-1997 to 02-112007
6361
3795 days
H1, H2, H3 in
Haarlem
H4: Leiduin - 3 km
SW of Haarlem
H5: Schiphol - 10 km
SE of Haarlem
H1, H2, H3: 17-062004 to 24-07-2005
H4: 01-01-1997 to
02-10-2007
H5: 01-01-1997 to
31-12-2007
H1, H2, H3: 2
minutes
H4, H5: day
Stationary rain: 40,
60, 70, 80, 90 l/s/ha
Design storms: T=1
year, T=2 years
(RIONED, 2004)
3 storms from data
series gauge H1
Correlation rain gauges Haarlem
Correlation between H4 and H5 (2003-2007)
Correlation betw. H1, H4 (18/11/04-23/07/05)
Correlation betw. H1, H5 (18/11/04-23/07/05)
0.635
0.81 (daily rainfall
from 8 to 8h for H1)
0.59 (daily rainfall
from 8 to 8h for H1)
Calls generated in weekend days are likely to be entered next working day: e.g. in 2004-2005
83 out of 104 Mondays hold complaints (80%), while 303 out of 521 working days hold
complaints (58%)
62
Quantitative fault tree analysis for urban flooding
Figure 2.4 Map of Prinsenbeek indicating the layout of the sewer system and the
location of the rain gauge P1.
63
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
Sewer flooding failure mechanism analysis for Prinsenbeek
The basic events for sewer flooding are sewer overloading by heavy rainfall,
pipe blockage and partial blockage or sedimentation of pipes and overflows
coinciding with rainfall. To analyse the contribution of these events, incidents
from call data are compared to flood incidents from a hydrodynamic model
simulation (Infoworks, Wallingford, version 8.5). The rainfall series that is
used as input for model simulation entirely overlaps the period of call data.
Incidents are counted for independent events; to this end the total rainfall
period is separated into independent rain events with dry periods of at least 10
hours in between. This results in 801 independent rain events. For each event,
the occurrence of flooding according to call data and to model simulation results
is compared and, if so, the number and locations of flood incidents. Figure 2.4
shows the lay-out of the case study area Prinsenbeek and the location of yhe
rain gauge.
In the call register 15 incidents of sewer flooding are found; model simulations
result in 4 flood incidents. These 4 incidents reflect cases of sewer overloading
during heavy rainfall and these are confirmed in textual information of calls
related to these incidents, e.g.: “Streets covered with water, water flowing
into our house”. The other 11 incidents in the call register are related to pipe
blockages, a wrong connection and a pump failure in a road tunnel. Call
information is not sufficiently detailed to discriminate between total or partial
pipe, valve or weir blockages. The frequency of sewer flooding is 0.07 per week
or 3.5 per year. The probability of this failure mechanism is 0.07/week or 0.9/
year. The relative contribution of blockage events to the sewer flooding failure
mechanism is 11 out of 15 (73%). The contribution of sewer overloading is
4 out of 15 (27%). The contribution of blockages is a conservatively biased
estimate, since not all potential blockages are reported in a call.
64
Quantitative fault tree analysis for urban flooding






Figure 2.5 Map of Haarlem that shows the location of rain gauges H1, H2 and H3
within the city area and the location of rain gauges H4 in Leiduin and H5 at Schiphol.
65


R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
Quantitative fault tree analysis for Haarlem
To find incident frequencies of all basic and undeveloped events in the fault
tree, every call in the Haarlem call register is screened and classified manually
for both causes and consequences of flooding. Cause classes correspond to
basic events and undeveloped events. Two “cause unknown” and “no problem
detected” classes are added for calls where call texts mention no clear cause
or indicate that no problem was found on-site. Consequence classes refer to
locations where flooding occurs, indicative of potential severity: flooding in
buildings, in basements, on public areas or in gardens and pastures. Figure 2.5
shows the lay-out of the case study area Haarlem and the location of the rain
gauges.
Daily rainfall data are available for the whole call data period and a period
of 1 dry day is used in this case to separate independent rain events. Calls
are assigned to independent rain events based on the date the call was made.
Incident frequencies are calculated for each basic event in the fault tree. The
fault tree model is used to calculate the top event probability for 4 scenarios
of flood consequences: flooding of streets, buildings, basements and gardens,
flooding in buildings only, flooding in basements only and flooding of streets
only. For each scenario individual contributions of basic events are quantified.
Table 2.3 gives 6 examples of basic events and their probabilities of occurrence.
In this case the inter-arrival time θ ≠1/λ, because the duration of events is not
negligible. Confidence intervals are calculated for incident frequencies and
probabilities based on uncertainties in the call data: 56% of call texts do not
explicitly mention occurrence of flooding. Inclusion of these calls in frequency
calculations gives a maximum estimate, whereas exclusion provides a minimum
estimate of flood incidents. Uncertainty also relates to calls that have been made
during dry periods. They represent 23% of the total number of calls. 48% of the
“dry event calls” can be explained because they report flood incidents for other
causes than rainfall, e.g. drinking water pipe bursts or a high groundwater
table. Detailed analysis shows that of the other 52%, some refer to a previous
66
Quantitative fault tree analysis for urban flooding
rain event whereas others seem to indicate that at the specific location rainfall
did occur. This is explained by spatial rainfall variation that the available data
from only two rain gauges for most of the analysed period cannot sufficiently
account for. The range between flood incident frequencies including and
excluding all dry-period-calls gives another bandwidth of uncertainty in flood
incident calculations.
Table 2.3 Six examples of basic events in the fault tree. The second column gives the
results for the event occurrence rate, the number of incidents associated with a basic
event divided by the number of weeks in the period of analysis (1997-2007). The third
column gives the probability of occurrence of basic events. 95% confidence intervals are
based on outcomes from different assumptions for incident analysis: in- or excluding
calls with no explicit consequence mentioned and in- or excluding calls during dry
periods.
Basic events in fault tree for urban
flooding
Period 1997-2007
Nr of incidents Basic event
Probability P
for basic event occurrence rate of at least one
[/10 years]
λ [week-1]
occurrence per
week [week-1]
Blocked or full gully pot
393 ± 209
0.72 ± 0.38
0.49 ± 0.17
Gully pot manifold blocked or broken 113 ± 66
0.21 ± 0.12
0.18 ± 0.09
No outflow available from a pool to a 60
± 10
0.11 ± 0.02
0.10 ± 0.02
rainwater facility
Sewer overloading
13
±1
0.02 ± 0.002 0.02 ± 0.002
Sewer pipe blocked
8
±4
0.01 ± 0.01
0.01 ± 0.01
Drinking water pipe burst
29
± 11
0.05 ± 0.03
0.05 ± 0.03
Gully pot blockages and gully pot manifolds cause the highest numbers of
flood incidents (table 2.3) and are subject to larger uncertainty than other basic
events. Sewer overloading incidents are reported with high certainty: in most
cases consequences are explicitly mentioned and few are reported during dry
periods.
The probability of flood incidents in buildings and basements is lower than
that of flooding in public areas (table 2.4). This is to be expected since in many
cases flood water flows over public areas before it runs into buildings. Flooding
67
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
of basements is mainly a result of high groundwater tables, for the case of
Haarlem. Blocked gully pots and gully pot manifolds, both component failures,
cause more flood incidents than sewer overloading by heavy rainfall, not only
for flooding in public areas, but also for flooding in buildings.
Table 2.4 Basic event incident numbers and probabilities in urban flooding fault tree
for 4 scenarios of flood consequences: (1)sum of all flood consequences, (2)flooding in
buildings only, (3)flooding in basements only, (4)flooding of public areas only. Incident
numbers of scenario 1 can be lower than sum of incidents of scenarios 2, 3 and 4 because
several types of consequences often occur simultaneously during a rain event.
Basic events in fault tree for urban
flooding, 4 flood consequence
scenarios
Period 1997-2007
Nr of basic
event
incidents
[/10years]
Prob. of at
least 1 occ.
per week
[week-1]
Nr of
basic event
incidents
[/10years]
Prob. of at
least 1 occ.
per week
[week-1]
Scenario
314
70
66
1
0.440
0.120
0.110
Scenario
45
6
12
2
0.080
0.011
0.022
14
8
46
37
Scenario
Blocked or full gully pot
17
Gully pot manifold blocked or broken 2
No outflow from a pool to a rainwater 2
facility
Sewer overloading
5
Sewer pipe blocked
0
Drinking water pipe burst
3
Groundwater table above ground level 46
0.025
0.015
0.066
0.066
3
0.031
0.004
0.004
1
0
1
1
Scenario
304
68
54
0.002
0.000
0.002
0.002
4
0.430
0.120
0.095
0.009
0
0.006
0.081
7
6
21
2
0.013
0.011
0.038
0.004
Blocked or full gully pot
Gully pot manifold blocked or broken
No outflow from a pool to a rainwater
facility
Sewer overloading
Sewer pipe blocked
Groundwater table above ground level
Drinking water pipe burst
Quantitative analysis: Monte Carlo simulation of fault tree
Mean basic event probabilities are used to calculate the top event probability
and rank the contributions of basic events. The quantitative analysis is based
on Monte Carlo simulation: the occurrences of basic events are simulated by
use of a random number generator. Each simulation that results in failure is
68
Quantitative fault tree analysis for urban flooding
stored, with the combination of basic events that caused the failure. A Monte
Carlo simulation for the case of Haarlem results in 7000 failures out of 10.000
simulations. The probability of the top event is 0.7 per week. Table 2.5 shows
the contribution of 5 basic events to the overall probability of failure.
Table 2.5 Results of 10.000 Monte Carlo simulations with the fault tree model for
Haarlem
Basic events
Contribution to total Contribution to
number of 7000 flood overall probability of
incidents
failure [%]
Blocked or full gully pot
Gully pot manifold blocked or broken
Not outflow available
Sewer overloading
Sewer pipe blocked
Drinking water pipe burst
5000
1770
1020
210
95
510
71
25
15
3
1
7
Sensitivity analysis for fault tree calculation
The sensitivity of the fault tree analysis to the probabilities of the basic events
is tested by changing the probabilities of the basic events between a lower and
an upper limit. Probability estimates based on call data are considered as a
minimum probability estimate since the likelihood of a false positive in the
register after cross-checking with rainfall data is small. Maximum estimates
are based on the number of basic events that could occur under unfavourable
conditions, with a minimum if maintenance and a maximum of human errors.
Estimates are made by expert judgment. For instance, the maximum expected
probability for gully pot blockage has been set equal to the probability of
occurrence of a rain event. The maximum estimate for no outflow has been set
equal to the average number of road reconstruction projects, assuming that all
of these result in some error that creates a no-outflow situation. The mistake is
assumed to be repaired after the first rain event.
69
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
Table 2.6 Results of the fault tree sensitivity analysis with minimum and maximum
probability estimates, for 10.000 Monte Carlo simulations
Basic events
Minimum estimate Maximum estimate
Total probability of failure
0.7
0.97
Contribution to overall probability of failure, minimum estimate [%]
Blocked or full gully pot
71
75
Gully pot manifold blocked or broken
25
44
Not outflow available
15
43
Sewer overloading
3
15
Sewer pipe blocked
1
22
Drinking water pipe burst
7
50
The probability of the top event rises to 0.97 when maximum estimated
occurrence probabilities are entered for all basic events (table 2.6). The
contribution of most individual basic events to the failure probability increases;
nevertheless gully pot blockages still contribute 75% to the top event probability.
The contribution of heavy rainfall events to the top event has increased from
5 to 15 %. The percentage contributions of the basic events do not add up to
100%, because basic events can contribute to the top event through various
combinations of basic events. The percentage indicates the ratio of the failures in
which the basic event is involved to the total number of failures. The pessimistic
maximum probability estimates result in many concurrences of basic events.
2.5.
Discussion
In this chapter a methodology is provided to conduct quantitative fault
tree analysis for urban water infrastructure systems and present results of
applications to two cases. To the authors’ knowledge, this is the first application
of probabilistic fault tree analysis to urban water infrastructure flooding. The
results show that component failures contribute significantly to urban flood
probability: gully pot blockage contributes 71%, gully pot manifold blockage
25% and pipe blockage 1% in a complete fault tree analysis for the case of
Haarlem. An analysis of only the mechanism of sewer flooding for the case of
70
Quantitative fault tree analysis for urban flooding
Prinsenbeek results in a frequency of 0.07 per week, where sewer blockage
contributes 73%. Nevertheless this type of failure mechanism receives only
minor attention in most flood risk studies that tend to focus on sewer overloading
by heavy rainfall which contributes only 3% to urban flood probability and
27% to sewer flooding in the presented cases. The results seem to justify further
extension of research and monitoring in this field.
The results presented are mainly based on call centre data and have a
conservative bias: only part of potential incidents is reported in calls. It is
expected that sewer overload incidents are largely covered, because their call
reports are confirmed in sewer model simulation results. The bias in incident
estimates for component failure and human errors is difficult to assess. A test
should be conducted in practice where urban areas are intensively monitored
during a number of rain events to capture all flood incidents and these should
be compared to the number of incidents that is reported to the call centre.
Fault tree analysis for urban flooding has been shown to provide useful data
for risk analysis and management: it reveals potential failure mechanisms and
quantifies failure probabilities and relative rankings of failure mechanism
contributions. These can be used to find and improve weaknesses in urban
water systems. A complete risk assessment requires two parameters: incident
probability and the severity associated with an incident (Haimes, 1998). This
chapter does not deal explicitly with incident severity, but some first insights
are given by comparing different flood consequence classes. We have shown
that the probability of flooding in buildings is lower than that of flooding in
public areas as may be expected since water often flows from public areas into
buildings. Flooding of basements is in the case Haarlem almost exclusively a
result of high groundwater tables and incidents are independent of rain events.
To appropriately quantify risk and justify risk reduction investments a good
severity metric must be available. Urban flood incidents involve intangible
consequences such as traffic delay and social distress and inconvenience. Much
information on this subject has been collected in research studies in the UK
71
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
(Penning-Rowsell et al., 2005) The next step in this study will be to evaluate
possibilities for a severity metric for urban flood consequences based on call
data and available references.
Risk management has traditionally been reactive where flood incidents caused
by blockages and human errors are concerned. Pipe blockages can be detected
by sewer pipe inspections, but inspection frequencies are generally too low,
in the order of once in 10 years, to undertake adequate preventive actions.
Other components, like gully pots and pumps, tend to have a fixed maintenance
frequency and failures are handled after they occur. The question whether a
proactive structured approach like fault tree analysis can actually reduce
incident frequencies compared to traditional approaches is yet unanswered.
Fault tree analysis provides insight into relative contributions of failure
mechanisms and can by that draw attention to failure mechanisms that were
previously overlooked or underestimated. If preventive maintenance to prevent
blockage or at least to prevent flooding caused by blockage can be effective is a
difficult question to answer, because the formation of blockages by sediments,
tree roots, objects dumped into sewers etc. is highly unpredictable.
Fault tree analysis is a methodology that can easily incorporate different
kinds of flood incident causes in the quantification of flood probability. Also
detection of weak points and unforeseen failure mechanisms is a strong feature
of this methodology. In this sense it complements information provided by
hydrodynamic model simulations of flooding: hydrodynamic models are well
capable of modelling expected flood frequencies as a result of heavy rainfall,
based on rainfall series. They can also, in combination with overland flow
models, simulate expected flow paths, if sufficient geographical information is
available. But modelling of flood causes related to blockages and errors and
quantification of associated flood probabilities requires complex manipulations
and can be done in more straightforward way in a fault tree.
This research has revealed opportunities for potential improvement in call data
registration to make data more suitable for risk analysis. Categories that are
currently used in call data registers primarily serve the purpose of efficient
72
Quantitative fault tree analysis for urban flooding
redirection of calls for handling by the relevant departments. If an additional
well-defined classification is created based on potential flood causes, and causes
of other incident types if desired, incidents reported in these classes could be
directly used as input for fault tree analysis. A consequence classification could
also be added to be able to derive probabilities of incidents of different severity.
Proper use of these classifications requires training of involved personnel at the
call centre or call handling departments. Alternatively, automatic classification
of calls based on call texts can be considered. First attempts have been to do this
for the case of Haarlem. Automatic classification is based on recurrent words or
word combinations in call texts and its potential accuracy depends on correct
and consistent use of words the texts. In both cases the benefit of improvements
relies on awareness of system users of the importance of accurate classification
and reporting.
To gain more insight in explanatory factors of flood incidents and their causes,
fault tree analysis can be applied to more cases to compare results for different
systems. Examples of potential explanatory factors for occurrence of pipe, gully
pot, gully pot manifold and pump blockages are system age, system component
types or materials, maintenance regime and population composition.
73
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 2
2.6.
References
Ang A.H.S. and Tang W.H. (1984). Probability concepts in Engineering Planning and
Design. Volume II: Decision, Risk and Reliability. John Wiley and Sons, Inc.
Ashley, R.M. (2006). The future for water and flood risk management in highly susceptible
urban areas. Lecture at EnviroWater 2006, Delft
Ashley, R.M., Balmforth, D.J., Saul, A.J. and Blanksby, J.D. (2005). Flooding in the future
- predicting climate change, risks and responses in urban areas. Water Science and
Technology, 52 (5), 265-273
Clemens, F.H.L.R. (2001). Hydrodynamic model in urban drainage: application and
calibration. PhD thesis, Delft University of Technology, Delft.
Djordjevic, S., Prodanovic, D., Maksimovic, C., Iventic, M., Savic, D. (2005). SIPSON –
Simulation of interaction between pipe flow and surface overland flow in networks.
Water Science and Technology, 52 (5), 275-283
EU, 2007. European Flood Directive. Directive 2007/60/EC on the assessment and
management of flood risks. Available from: http://ec.europa.eu/environment/water/
flood_risk
FRMC, Flood Risk Management Consortium (2007). Year Three Progress Report. http://
www.floodrisk.org.uk
Haimes, Y.Y. (1998). Risk Modeling, Assessment, and Management. New York: John
Wiley & Sons, Inc.
Harremoës, P., Carstensen, J. (1994). Deterministic versus stochastic interpretation of
continuously monitored sewer systems. European Water Pollution Control, 4 (5),
42-48.
Kaplan S., Garrick B.J. (1981). On the quantitative definition of risk. Journal of Risk
Analysis, 1(1), 11-27
Penning-Rowsell, E., Johnson, C., Tunstall, S., Tapsell, S., Morris, J., Chatterton, J.,
Green, C. (2005). The Benefits of Flood and Coasal Risk Management: A Manual
of Assessment Techniques (the Multicoloured Manual). Flood Hazard Research
Centre, Enfield
RIONED Foundation (2004). Leidraad Riolering, Module C2100, 17-20 (in Dutch). ISBN
978-90-73645-68-4. Stichting RIONED, Ede, the Netherlands.
Thorndahl S., Willems, P. (2008). Probabilistic modelling of overflow, surcharge and flooding
in urban drainage using the first-order reliability method and parameterization of
local rain series. Water Research, 41, 455-466
Semadeni-Davies, A., Hernebring, C., Svensson, G., Gustafsson, L. (2008). The impacts
of climate change and urbanisation on drainage in Helsingborg, Sweden: Suburban
stormwater. Journal of Hydrology, 350(1-2), 114-125
Vesely, W., Dugan, J., Fragola, J., Minarick, J., Railsback, J. (2002). Fault Tree Handbook
with Aerospace Applications. Version 1.1. NASA Headquarters, Washington.
Vesely W., Goldberg, F.F., Roberts, N.H., Haasl, D.F. (1981). Fault Tree Handbook.
NUREG-0492. US Nuclear Regulatory Commission, Washington.
Vrijling, J.K. (2001). Probabilistic design of water defense systems in the Netherlands.
Reliability Engineering and System Safety 74 (3): 337-344
74
Quantitative fault tree analysis for urban flooding
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
75
Chapter 3
Urban flood risk curves
This chapter is based on an article that was accepted for publication in Water
Science and Technology.
Veldhuis, J.A.E. ten, Clemens, F.H.L.R. (in press). Flood risk modelling based
on tangible and intangible urban flood damage quantification. Water Science and
Technology.
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
Context
In the previous chapter it was shown that overloading of urban drainage
systems by heavy rainfall is only one of the possible failure mechanisms that
can cause urban flooding. The contribution of this failure mechanism to
overall flood probability is small compared to other failure mechanisms. This
means that hydrodynamic modelling of sewer overloading by heavy rainfall
can provide only a partial picture of flood risk. The question remains what
the contribution of sewer overloading is to total flood risk, probabilities and
consequences, compared to other failure mechanisms. To answer this question,
a method must be found that incorporates all failure mechanisms to fully assess
flood risk. Failure mechanisms like blockage of gully pots and sewer pipes are
not well enough understood to predict their probabilities of occurrence and
associated flood consequences in a deterministic way. This means that datadriven modelling is the only way to quantify total urban flood risk. Such datadriven approach requires historical data-series of flooding events, including
information on their causes and consequences. This chapter explores data
from municipal call centres to find out whether the information they provide
can be used to quantify urban flood risks associated with all possible failure
mechanisms.
Abstract
The usual way to quantify flood damage is by application stage-damage
functions. Urban flood incidents in flat areas mostly result in intangible
damages like traffic disturbance and inconvenience for pedestrians caused by
pools at building entrances, on sidewalks and parking spaces. Stage-damage
functions are not well suited to quantify damage for these floods. This thesis
presents an alternative method to quantify flood damage that uses data from a
municipal call centre. The data cover a period of 10 years and contain detailed
information on consequences of urban flood incidents. Call data are linked to
individual flood incidents and then assigned to specific damage classes. The
results are used to draw risk curves for a range of flood incidents of increasing
damage severity. Risk curves for aggregated groups of damage classes show
78
Urban flood risk curves
that total flood risk related to traffic disturbance is larger than risk of damage
to private properties which in turn is larger than flood risk related to human
health. Risk curves for detailed damage classes show how distinctions can be
made between flood risks related to many types of occupational use in urban
areas. This information can be used to support prioritisation of actions for
flood risk reduction. Since call data directly convey how citizens are affected
by urban flood incidents, they provide valuable information that complements
flood risk analysis based on hydraulic models.
Keywords: flood risk; intangible damage; risk curve; urban drainage
3.1.
Introduction
Quantitative flood risk assessment consists of two steps: probability estimation
and flood damage quantification. Methods to quantify flood damage for severe
floods are usually based on stage-damage functions that quantify damage based
on inundation depth, flood duration and occupational land use (Thieken et al.,
2005). Such functions focus on damage to buildings and building contents,
which constitute the main part of total flood damage for severe floods (DEFRA,
2004). Such floods typically have a low probability of occurrence and affect
large areas at once.
Urban drainage systems in lowland areas are typically designed to cope with
rainfall events with return periods of two to five years (e.g. RIONED, 2004). As
a result, urban flood incidents occur at a regular basis. Many of these incidents
are characterised by small flood depths and small geographical extension. Stagedamage functions are not applicable to quantify damage for such small flood
depths (Merz et al., 2005; Apel et al., 2004; Dutta et al., 2003), because they
are generally developed for flood depths between 0 and 5 meters (e.g. Apel, in
press, Chang 2008 and Dutta, 2003) and uncertainty increases for applications
to smaller flood depths. Additionally, for many urban flood incidents, direct
damage forms a small if not negligible portion of flood consequences, where
79
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
intangible damage in the form of disruption of road traffic and inconvenience for
pedestrians caused by pools in front of shops, on parking lots and sidewalks is
more important. Indirect and intangible damages are more difficult to quantify
than direct damage. For convenience, indirect damage is sometimes quantified
as a fixed percentage of direct damage (FHRC, 2003), if indirect damage is
expected to be small compared to total damage. Previous studies have shown
that direct tangible damage cannot sufficiently describe flood consequences
and that intangible damage, particularly physical and mental health effects
should be included in the appraisal of flood risk alleviation schemes (Tapsell
et al., 2003). Few references are available on quantitative measures for these
indirect and intangible damages, compared to material damage and loss of
life. A method to translate intangible consequences into monetary values is by
assessing people’s willingness to pay (WTP) to prevent flood consequences.
This method was applied in the UK, where WTP to prevent health effects of
flooding was investigated in a series of questionnaires (DEFRA, 2004) which
resulted in an average sum per household. Fewtrell and Kay (2008) attempted
to quantify physical and mental health effects in terms of disability adjusted life
years (DALYs) based on interview results for people affected by floods. This
approach was also adopted by Lulani et al. (2008), who based their study on a
list of theoretical assumptions. Most methods that were proposed to quantify
intangible damage (e.g. Green et al., 1998 and Lekuthai et al., 2001, DEFRA,
2004) are based on indirect data, if any and include assumptions that are
difficult to verify.
In this chapter, municipal call data are used, that provide detailed information
on flood problems as encountered by citizens. The advantage of call data as
opposed to interviews and questionnaires is that the lag time between incident
occurrences and reporting of the consequences is very short. The purpose of
this study was to translate call information into quantitative values that can be
used for risk assessment. Risk is expressed in the form of a set of risk curves that
visualise risks expressed as exceedance probabilities for a range of consequence
severities.
80
Urban flood risk curves
3.2.
Classification of flooding consequences
Flood incident data
Call data consist of a unique call number, date of the call, street name to indicate
the problem location and a telegram style text that describes what the caller has
said. In most cases a second text is added that describes the results of on-site
checking and actions undertaken to solve the problem. Ten years of call data
on flood incidents in Haarlem and 5 years of call data for Breda were analysed
to quantify flood risk. Calls on urban drainage incidents were selected from the
database which resulted in dataset of 6361 and 7049 calls. Calls were assigned
to independent rain events which were defined by a separation of 24 or more
hours of dry weather.
Classification results are sensitive to class definition: the more narrowly a class
is defined, the lower the number of calls assigned to that class. Class boundaries
must be set in such a way that class sizes are more or less equal or, if class
sizes are different, these differences must be taken into account when results
are interpreted and compared between classes. This is more straightforward
for numerical values like ages in a population or flood damage figures than
for nominal values like call texts. For example, a class defined as “flooding at
bus stop” is by nature likely to receive fewer calls than “flooding on residential
road”, for two reasons: the total area of residential roads far exceeds that of bus
stops in any urban area and calls with short, undetailed call texts automatically
fall into classes of a more general definition, i.e. a class like “residential road” is
likely to get assigned many calls.
A consequence classification for urban flood-related calls has been developed for
this study, based on the primary functions of urban drainage systems (Davies
and Butler, 2004). This results in classes that can directly support evaluation
of urban drainage functioning and associated policy guidelines. Table 3.1
shows primary functions and consequence classes that have been used for call
consequences classification. For illustration, the numbers of calls in each class
81
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
for the case of Haarlem are added: the high number of calls in class “flooding
on residential/main street” compared to other classes shows how specificity of
class definition influences call numbers.
Calls are assigned to damage classes based on observations described in the call
texts: for instance, calls texts mentioning observations of toilet paper or excreta
are assigned to wastewater flooding; call texts indicating flooding of a shop or
a bus stop are assigned to the corresponding class. Classification results consist
of a matrix with individual flood incidents I1 to In in rows and damage classes in
columns. Table 3.2 shows a schematised example of the matrix.
Table 3.1 Primary functions of urban drainage systems and consequence classification
Primary functions
Consequence classes
Protection of human health:
physical harm or infection
Flooding with wastewater (toilet paper/bad smell/
excreta)
Nr. of calls
in class (city
of Haarlem)
20
Manhole lid removed
Protection of buildings and
Flooding in residential building
infrastructure against flooding: (house/flat/garage/shed)
damage to public and private
Flooding in commercial building
properties
(shop/restaurant/storage hall)
4
78
Prevention of road flooding:
traffic disruption
Flooding in tunnel(road/cycleway)
Flooding at bus stop/bus station/taxi stand
Flooding in shopping street/market place/commercial
centre
13
18
115
Flooding in front of entrance to shop/bar/restaurant/
library/hospital
55
Flooding on residential/main street
Flooding of sidewalk/cycle path
596
344
26
Table 3.2 Example of damage classes and call classification results
Damage classes
Flooding
Flooding in
incident nr. commercial building
Ii
Ij
Ik
82
0
0
1
Flooding in
residential
building
1
0
0
Flooding of
residential
road
20
5
12
Flooding in Flooding of
road tunnel wastewater
0
0
0
0
1
0
Urban flood risk curves
3.3.
Risk curves
Risk assessment studies often present the expected value of risk as a summary
value for a range of probabilities and consequences or they give a risk value for
a given scenario, e.g. a certain return period. Risk curves go one level deeper
and present risks for a range of probabilities and consequences (Kaplan and
Garrick, 1981). Risk curves for urban flooding depict flood damages on the
horizontal axis and their associated exceedance probabilities on the vertical
axis. Figure 3.1 gives an example of a risk curve, for a flood damage xi varying
from 0 to 100 on the horizontal axis and associated exceedance probabilities on
the vertical axis. The intersection of the curve with the vertical axis gives the
probability of any damage at all; the intersection with the horizontal axis gives
the maximum possible damage, with zero probability of exceedance. Values in
between are interpreted as probabilities of at least damage xi; this probability
increases or remains constant for decreasing damages. The staircase function
is the plotted result of a series of points representing damage for scenario i
and for each scenario. The staircase function can be regarded as a discrete
approximation of a continuous reality, represented by the smooth curve.
The area below the risk curve is a measure of total risk; the further risk curves
shift to the top-right-hand corner of the graph, the higher their associated total
risk. The advantage of risk curves compared to one value for expected risk is
that risk curves give insight into the contributions of small and large damages
to flood risk. If flood risk is mainly associated with small damage incidents, the
curve decreases steeply for small damages and more gently for high damages, as
is the case of the example in figure 3.1. If large damages mainly compose risk,
the curve is more or less flat for small damages and steeply decreases at large
damage values.
83
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
0.7
P robability of any damage at all
0.6
P robability of at leas t flood damage xi [-]
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
0.5
0.4
R is k of flooding, s taircas e function
R is k of flooding, s moothed line
0.3
0.2
Maximum poss ible damage
0.1
0
0
10
20
30
40
50
60
70
80
90
F lood damage x i [unit of damage]
Figure 3.1. Example of a risk curve (based on: Kaplan and Garrick, 1981): a
complementary cumulative distribution function (CCDF), i.e. the probability of
exceeding a given damage
Representation of risk in the form of risk curves requires availability of flood
incident data for a range of small to large flood damages. In this chapter call
classification results are used as a quantitative measure for intangible flood
damage, based on the assumption that the amount of calls per incident is
indicative of the number of affected citizens. This is confirmed by the correlation
between rainfall volume and numbers of flood-related calls per rainfall event:
a correlation coefficient of 0.76. This indicates that call numbers increase with
increasing rainfall volumes which are likely to induce more flooding (figure
3.2). The resulting curves for the number of calls per flood incident are similar
to FN-curves that show the probability of exceedance (F) as a function of
84
100
Urban flood risk curves
Nr of calls/event [-].
the number of fatalities (N) and are often used to quantify societal risk (e.g.
Bedford and Cooke, 2001).
300
S pearman's rank correlation
coefficient: 0.8
F requency of nr calls /event
250
F requency of rainfall vol/event
200
150
100
50
50
-50
0
0
50
-50
50
100
150
200
250
300
R ainfall volume/event [mm]
100
-100
Figure 3.2 Correlation between the number of calls per event and rainfall volume per
event. Frequencies of the number of data for call number per event and rainfall volumes
are displayed as well.
The probability of a certain damage, or amount of calls per incident, is derived
from the occurrence frequency of the damage. The occurrence of a given damage
is assumed to be a Poisson process. This implies that the probability that a
given damage will occur in any specified short time period is approximately
proportional to the length of the time period, that occurrences of evens in
disjoint time periods are statistically independent and that events do not occur
exactly simultaneously. Under these conditions, the number of occurrences x in
some fixed period of time is a Poisson distributed random variable:
85
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
    











  





 (3.1) 

   
   
    

 





 




   :
 of x occurrences

 of a given
 damage
 in a period of



probability
Where:














 




time

    

t 
 
  
  






:average rate of occurrence
of a given damage per time unit
λ    




 




  




 ofoccurrence
 
The
λ is derived
from call data
over 
a certain period
of 
time  


 
 
 
 rate











 

and
is assumed to be constant. The probability of any occurrence of a given





damage
then
calculated
from:
 
  can
be

 
     

  


  
 

   
 















 

 













































   
   













             (3.2)




  







      
 
       




 : probability




Where:
of at least one occurrence






 

       




         







of no occurrences
    : probability

 
         








                

The
time period
t 
be 
chosen
at will;
the longer
t, the higher
the














can







 







 







probability

 



















of occurrence. The time scale is preferably chosen so as to fit the frequency of





In our
dataset
flood incidents
occur
up toseveral
per
month
and   

events.
 
 
 

 times

 
 
















  of



 
  
 
 


the
duration
events
is 
in the
order 
of several
days. 
A time period
of 1 week




fits the incident occurrence frequency and has been chosen for this analysis.






The
results are used to plot risk curves for individual damage classes separately







 groups of damage, where calls over several classes are added
and for aggregated



           

plotted


 

up.
Risk
curves
are
in the form
of smooth
lines.       









            




 
 
 
 
 
 
 
 
 
 
 






















            










  
 
 

 
 

 
 










  








































             




 
 
 
 
 
 
 
 
 
 
 
 

 




          


          





86 

 
 
 





 





 

 
 


 













































             



           




 
 



 

 

 




 
 
 


  
 
  




























 






Urban flood risk curves
3.4.
Results and discussion
The municipal call classification results are given in table 3.3. Out of all classified
calls, 28% of the calls for Haarlem and 16% for Breda mention consequences
related to flooding. Flooding on streets is reported most often as a consequence.
This can be explained by the more general definition of this class as opposed to
e.g. flooding in front of entrance to building Therefore this class contains both
calls of real street-flooding and calls that due to a lack of detail in the call text
could not be assigned to more specific classes. This is a drawback of different
levels of detail in class definition that can only be avoided by generalising classes
which in its turn leads to a loss of information from detailed call texts.
Table 3.3 shows that detailed classification results in a number of sparse
consequence classes. In second instance, classes are lumped to a higher
aggregation level in order to obtain a more balanced classification dataset. The
classification results at the higher aggregation level are shown in table 3.3, as
totals in bold numbers.
87
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
Table 3.3 Call classification results for aggregated and for detailed flood consequence
classes, for the cases of Breda and Haarlem, for periods of 10 years and 5 years
Primary functions
Consequence classes
6.5
(nr)
28
9
37
141
7.6
34
1.9
16
0.9
173
26
74
423
13
18
117
9.7
1.5
4.1
3.4
1.4
3.4
0.7
1.0
6.5
63
26
63
309
22
17
4
1.2
0.9
0.2
Flooding in front of entrance to shop/bar/
library/hospital
55
3.1
43
2.3
Flooding in front of entrance to residential
building
65
3.6
66
3.6
Flooding on residential/main street
Flooding on cycle path
Flooding on sidewalk/footpath
Flooding on parking space
655
133
73
173
1302
1793
3563
1005
6361
36.5
7.4
4.1
9.7
Human health: physical Flooding with wastewater
harm or infection
Manhole lid removed
Total
Protection of buildings Flooding in residential building (house/
garage/shed)
and infrastructure:
damage to public and Flooding in commercial building (shop/
private properties
storage hall)
Flooding in basement
Water splashes onto building
Flooding of gardens/park
Total
Prevention of road
flooding: traffic
disruption
Nr. of calls/
Nr. of calls /
class:Haarlem class:Breda
Flooding in tunnel
Flooding at bus stop/taxi stand
Flooding in shopping street/place/
commercial centre
Total
Total number of calls relevant for flooding
No consequence mentioned
Consequence other than flooding
Total number of calls
(nr)
61
7
68
116
(%)
3.4
0.4
1229
23
25
70
1499
100% 1845
3035
2169
7049
(%)
1.5
0.5
66.6
1.3
1.4
3.8
100%
Figure 3.3 shows the classification results for the flood-related consequence
classes in percentages of the total number of flood-related calls. Flooding in
residential streets occurs most frequently in Haarlem and more dominantly so
in Breda. The 3 classes that relate to flooding in buildings represent 19% and
13% of the flood-related calls.
88
Urban flood risk curves
% of calls in clas s 70.00
60.00
50.00
Haarlem
B reda
40.00
30.00
20.00
10.00
w
as
te
wa
m
te
re
a
r
nh
si
de
ol
e
n
l
tia
co
m
l b id
m
ui
er
c ia ldin
g
l b
ui
ld
in
ba
g
bu
se
ild
m
in
e
g s p nt
la
ga
sh
rd
e n ing
s /p
ar
ks
tu
nn
co
bu e l
m
s
m
s
er
t
c ia op
l s
pu
tre
bl
ic
et
e
pr
nt
ra
iv
at
nc
e e
en
re
tra
si
de
nc
nt
e
ia
l s
tre
et
cy
cl
e pa
th
si
de
pa
w
rk
al
k
in
g sp
ac
e
0.00
C ons equence clas s es
Figure 3.3 Call classification results for flood-related consequence classes, for the cities
of Haarlem and Breda
Classification results at both aggregation levels are used to plot risk curves.
Figures 3.4 and 3.5 give 2 examples of risk curves for individual damage classes.
Flood consequence severity on the horizontal axis is expressed as amount of
calls per incident. The risk curves show that the maximum amount of calls
for flooding on streets is more than twice as high as for flooding in residential
buildings. The probability of at least 1 call is more than 3 times higher for
flooding on streets than flooding in residential buildings.
89
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
0.8
0.7
0.7
P robability of at leas t X calls
P robability of at leas t X calls
0.8
0.6
0.5
flooding on s treets ,
s moothed line
0.4
0.3
0.2
0.6
0.5
flooding in res idential buildings ,
s moothed line
0.4
0.3
0.2
0.1
0.1
0
0
0
5
10
15
20
25
30
0
2
4
6
8
10
F lood damage: amount of calls X per incident F lood damage: amount of calls X per incident
Figure 3.4. Risk curves (smoothed lines) and staircase functions for consequence class
‘flooding on streets’, based on call amounts per incident as a measure for consequence
severity
1
0.9
P robability of at leas t x calls
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
dis turbance of traffic
0.8
0.7
damage to private properties
0.6
0.5
threats to human health
0.4
0.3
0.2
0.1
0
1
3
5
7
9
11
13
15
17
19
21
23
25
27
Number of calls per incident
Figure 3.5. Risk curves (smoothed lines) and staircase functions for consequence class
‘flooding in residential buildings’, based on call amounts per incident as a measure for
consequence severity.
90
12
Urban flood risk curves
Risk curves for other consequence classes (see appendix 3) indicate that for
most consequence classes, maximum amount of calls per incident is below 5.
Maximum probabilities of at least 1 call per event vary from 0.009 per week for
lifted manholes to 0.13 per week for flooding on parking spaces. Risk curves
provide this information in a more accessible way than lists of numerical data.
The results indicate that most call texts are not detailed enough to be assigned
to detailed consequence classes and end up in a general class, here “flooding on
streets”. Still, detailed classification results help to identify which consequences
are mentioned more often than others. For instance flooding in residential
buildings is a detailed class that is mentioned up to 10 times per flood incident,
whereas flooding of tunnels is never mentioned more than once per incident in
our dataset.
Figure 3.6 shows 3 risk curves based on aggregated flood consequence classes.
The curves show that consequences for traffic are far more likely to be
mentioned by callers than damage to private properties and human health
consequences. The probability of at least 1 call on traffic consequences is almost
0.9 per week and the maximum amount of calls per incident is 28. Human
health consequences are mentioned in maximum 3 calls per incident; the
probability of at least 1 call per incident is 0.06 per week. Damage to private
properties generates a maximum of 12 calls per incident; the probability of at
least 1 call is almost 0.3 per week. Risk curves for aggregated consequences are
useful to quickly distinguish between higher and lower risks.
91
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
0.14
P robability of at leas t x calls [wk-1]
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
tunnel
0.12
cycle path
0.1
parking s pace
s idewalk/footpath
0.08
in front of entrance to commercial building
s hopping s treet
0.06
bus s top/taxi s tand
in front of entrance to res idential building
0.04
0.02
0
1
2
3
4
5
6
7
8
9
10
11
12
Number of calls per incident [-]
Figure 3.6 Risk curves for aggregated urban flood consequence classes
In figure 3.6, the risk curve for ‘disturbance of traffic’ lies furthest towards the
upper right corner of the graph, so the total associated risk is highest for this
curve. Total risk for damage to private properties is lower than for disturbance
of traffic and higher the risk of threats to human health. All risks are mainly
related to low-severity incidents in the sense that for most incidents only few
people report consequences.
This information is a useful input to check system performance for compliance
with policy guidelines. For instance if health protection is a priority, the lower
health risk compared to other risks as illustrated in figure 3.6 is in accordance
with policy guidelines. If prevention of traffic disturbance has a high priority,
the aggregated risk curve in figure 3.6 is a reason to consider the need for
improvements. Aggregated consequences give information about risks at the
level of primary functions of urban drainage systems. More detailed information
is required to decide whether the underlying types of consequences indeed
justify investments for improvement and if so, which actions are most effective.
92
Urban flood risk curves
For instance, figure 3.7 shows risk curves for detailed consequence classes
within the aggregated class ‘traffic disturbance’. Main contributions to the risk
of traffic disturbance are flooding of cycle paths and parking spaces, whereas
flooding of tunnels and bus stops contribute only little to traffic disturbance
risk. The more detailed the level of risk curves, the better investment needs can
be identified and motivated.
Uncertainty aspects
Flood risk estimations are subject to large uncertainties, whether based on
historical data, theoretical modelling or a combination of both (see e.g. Apel
et al., 2004 and Merz et al., 2004). Call data are a valuable source of historical
data on flood incidents that has been little researched so far. A source of
uncertainty particular for flood risk estimations based on these data is that call
data report only a portion of the actual flood incidents. It is unknown whether
reported incidents are representative nor what proportion they form of the total
amount of incidents. Also, call information can be subjective and comes from
non-experts whose information can be incorrect. This source of uncertainty is
greatly reduced when calls are checked on-site by technical experts or when
calls are handled by trained people using good protocols. On the other hand,
call data directly convey citizens’ experiences regarding adverse effects of
wastewater and flooding, which urban drainage systems are designed to protect
citizens from. Therefore call data are a useful source of information to prioritise
actions for flood risk reduction.
Risk curves can be made for data sources of historical flood incidents other
than call data as well. When data from various sources are available these can
be used to draw separate curves and compare these. Alternatively, historical
flood incident data can be combined by design a classification that fits the data
and can be used to draw risk curves. Combination of data sources does require
a careful assignment of data to individual incidents and consequence classes so
as to avoid double counting of consequences.
93
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
94
0
0,02
0,04
0,06
0,08
0,1
0,12
0,14
1
2
3
4
5
7
Number of calls per incident [-]
6
8
9
in front of entrance to residential building
bus stop/taxi stand
shopping street
in front of entrance to commercial building
sidewalk/footpath
parking space
cycle path
tunnel
Figure 3.7 Risk curves for detailed classes related to traffic disturbance
Probability of at least x calls [wk-1]
10
11
12
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
Urban flood risk curves
3.5.
Conclusions
A strong correlation was found between the amount of call data per incident and
rainfall volumes per incident. Based on this result, call data per incident were
used as a measure for incident severity. Risk curves were drawn that depict
flood risk for a range of flood incidents, from high-probability low-consequence
incidents to low-probability high consequence ones. Risk curves were plotted
for individual consequence classes and for aggregated consequence classes. The
risk curve for aggregated consequence classes showed that urban flood risk
related to traffic disturbance is high compared to damage to private properties.
Total flood risk related to human health is small. Examples of risk curves for
detailed consequence classes showed how distinctions can be made between
flood risks related to many types of occupational use in urban areas. This
information is useful to prioritise actions for flood risk reduction.
Since call data directly convey citizens’ experiences in urban flood incidents,
they give valuable information about the degree of protection that urban
drainage systems provide against adverse affects of wastewater and flooding.
Flood risk analysis based on hydraulic modelling and stage-damage functions do
not provide this type of information and mostly focus on severe, low probability
flood incidents. Call data complement these analyses in a valuable way.
95
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 3
3.6.
References
Apel, H., Thieken, A., Merz, B., Blöschl, G. (2004). Flood risk assessment and associated
uncertainty, Natural Hazards and Earth System Sciences, 4, 295-308
Bedford, T., Cooke, R.M. (2001). Probabilistic Risk Analysis: Foundations and Methods,
Cambridge University Press, New York, 2001.
Butler D., Davies, J.W. (2004). Urban Drainage. 2nd Edition. New York: E & FN Spon
Defra (2004). The appraisal of human related intangible impacts of flooding. Defra, Flood
management division, R&D Technical Report FD2005/TR, London, UK.
Dutta D, Herath S, Musiake K. (2003), A mathematical model for flood loss estimation,
Journal of Hydrology, 277(1–2), 24-49.
Fewtrell, L., Kay, D. (2008). An attempt to quantify the health impacts of flooding in the UK
using an urban case study. Public Health 122: pp. 446-451
Kaplan, S., Garrick, B.J. (1981). On the quantitative definition of risk. Risk Analysis, 1(1),
11-27.
Merz, B., Kreibich, H., Thieken, A., Schmidtke, R. (2004). Estimation uncertainty of direct
monetary flood damage to buildings. Natural Hazards and Earth System Sciences 4
(1), 153–163
Merz, B., Thieken, A. (2005), Separating natural and epistemic uncertainty in flood
frequency analysis, Journal of Hydrology, 309(1-4), 114-132.
Lulani, I., Steen van der, P., Vairavamoorthy, K. (2008). Analysis of the public health risks
of the urban water system in Accra by microbial risk assessment. WaterMill Working
Paper Series 2008, no. 8.
RIONED Foundation (2004). Leidraad Riolering, Module C2100, 17-20 (in Dutch). ISBN
978-90-73645-68-4. Stichting RIONED, Ede, the Netherlands.
Thieken, A.H., Muller, M., Kreibich, H., Merz, B. (2005). Flood damage and influencing
factors: New insights from the August 2002 flood in Germany. Water Resources
Research, 41(12), 1-16.
96
Urban flood risk curves
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
97
Chapter 4
Microbial risks of exposure to
contaminated urban flood water
This chapter is based on an article that was published in Water Research.
Veldhuis, J.A.E. ten, Clemens, F.H.L.R., Sterk, G., Berends, B.R. (in press).
Microbial risks associated with exposure to pathogens in contaminated urban flood
water, Water Research (2010), doi:10.1016/j.watres.2010.02.009
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
Context
The primary function of urban drainage systems is to protect public health by
preventing contact with pathogens in wastewater. Urban flood risk analyses
tend to focus on damage to flooded properties. If urban pluvial flooding
involves combined sewer systems, flood waters can be contaminated and pose
health risks to citizens. In this chapter it is investigated whether urban flood
waters can pose a health risk to citizens based on a screening-level microbial
risk assessment.
Abstract
Urban flood incidents induced by heavy rainfall in many cases entail flooding of
combined sewer systems. These flood waters are likely to be contaminated and
may pose potential health risks to citizens exposed to pathogens in these waters.
The purpose of this study was to evaluate the microbial risk associated with sewer
flooding incidents. Concentrations of Escherichia coli, intestinal enterococci and
Campylobacter were measured in samples from 3 sewer flooding incidents. The
results indicate faecal contamination: faecal indicator organism concentrations
were similar to those found in crude sewage under high flow conditions and
Campylobacter was detected in all samples. Due to infrequent occurrence of such
incidents only a small number of samples could be collected; additional data
were collected from controlled flooding experiments and analyses of samples
from combined sewers. The results were used for a screening-level quantitative
microbial risk assessment (QMRA). Calculated annual risks values vary from
5x10-6 for Cryptosporidium assuming a low exposure scenario to 0.03 for Giardia
assuming a high exposure scenario. The results of this screening-level risk
assessment justify further research and data collection to allow more reliable
quantitative assessment of health risks related to contaminated urban flood
waters.
Keywords: combined sewer, health risk assessment, urban flooding, wastewater
100
Microbial risks of exposure to contaminated urban flood water
4.1.
Introduction
The frequency of flooding and the damage caused by urban flood events have
increased over the past decades, mainly due to accelerated urbanisation (Ashley
et al. 2005). When urban flooding occurs in areas with combined sewer systems,
flood water is likely to be faecally contaminated and may pose health risks to
citizens exposed to pathogens in these waters. Faecal contamination of urban
flood waters was investigated after severe flooding in New Orleans following
Hurricanes Katrina and Rita (Sinigalliano et al., 2007) and after the Elbe floods
in Germany in 2002 (Abraham and Wenderoth, 2005). Elevated levels of faecal
indicator bacteria and microbial pathogens were found in floodwaters and in
sediments left in the urban environment after the flood. Faecal contamination
of floodwaters and subsequent contamination of drinking water sources have
been found for severe flood events in Bangladesh and Indonesia (Sirajul
Islam et al., 2007; Phanuwan et al., 2006). Physical and mental health effects
associated with severe floods have been studied by several authors (Fewtrell
and Kay, 2008; Ohl and Tapsell, 2000; Tapsell and Tunstall, 2003 and Tunstall
et al., 2006) based on interviews with people affected by floods. Lulani et al.
(2008) quantified combined health effects of flooding in terms of disabilityadjusted life years based on a list of assumptions.
Microbial health risks associated with faecally contaminated flood waters
are not only induced by severe, extensive flood events. Especially in lowland
areas flooding of combined sewers occurs almost yearly. For instance in the
Netherlands, a commonly applied design criterion for combined sewer systems
is a maximum flood frequency of once per year or per two years (RIONED,
2004). The reason why higher flood frequencies are accepted in lowland areas
is that expected damage of sewer flooding in flat areas is less than in sloping
areas: flood waters spread over larger areas, resulting in smaller flood depths
compared to sloping areas where flood waters concentrate in local depressions.
This implies that exposure of citizens to faecally contaminated flood waters may
occur on a regular basis. The spatial extent of these flood incidents is usually
101
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
small, flood waters covering a part of a street up to several streets (ten Veldhuis
and Clemens, 2009).
Occurrence of urban flood incidents is expected to increase in the future,
as climate change will induce more intense rainfall (e.g. Lenderink and van
Meijgaard, 2008) and ongoing urbanisation continues to increase inflow
to urban drainage systems. In addition, flooding caused by infrastructure
failures like pipe blockages is expected to occur more frequently in the future
as systems are ageing (ten Veldhuis et al., 2009). Increased flood frequencies
and growing population densities will increase health risks associated with
exposure to contaminated urban flood waters. Health risks associated with
combined sewer overflows (CSOs), which occur at a higher frequency than
flood incidents, have been investigated by Donovan et al. (2008) who find a
probability of contracting gastrointestinal illness from incidental ingestion of
water near CSOs ranging from 0.14 to nearly 0.70 over the course of a year
for visitors and recreators (e.g. swimmers), respectively, associated with the
presence of faecal pathogens indicated by the presence of faecal Streptococcus and
Enterococcus. Schets et al. (2008) investigated microbial quality of surface water
in canals and recreational lakes in Amsterdam that receive polluted water from
CSOs, raw sewage from houseboats and dog and bird faeces. The estimated
risk of infection with Cryptosporidium and Giardia per exposure event ranged
from 0.00002% to 0.007% and 0.03% to 0.2%, respectively, for occupational
divers professionally exposed to canal water. The effect of CSOs on surface
water quality has been investigated by Kay et al. (2008). They quantified faecal
indicator concentrations and export coefficients for catchments with different
land use and under specific climatic regimes. Urban areas are identified as one
of the key sources of faecal indicator organisms, with significantly higher values
occurring for high flow conditions, during or after rainfall. Curriero et al. (2001)
analysed the more general relationship between precipitation and waterborne
disease outbreaks for 548 reported outbreaks in the USA from 1948 through
1994. They found a statistically significant association between weather events
and disease; overflows from combined sewer systems are mentioned as one of
the potential sources of contamination.
102
Microbial risks of exposure to contaminated urban flood water
The purpose of this study was to conduct a screening-level quantitative
microbial risk assessment (QMRA) to evaluate the risk associated with
exposure of citizens to pathogens in flood waters resulting from combined
sewer flooding. Samples were collected and analysed for 3 sewer flooding
incidents and controlled flooding experiments were conducted to test survival
of pathogens in flood water. The results were used to conduct a screening-level
quantitative microbial risk assessment.
4.2.
Materials and methods
Experiments
Flooding incidents occur infrequently and often unpredictably in terms of
time and location; this makes sampling from flooding incidents a difficult task.
During a measurement campaign in the summer of 2007, several heavy rainfall
events occurred; one of those caused flooding at locations that were known to
flood regularly. During and shortly after a heavy rainfall event on 16 July 2007
water and sediments were sampled from 3 flooding incidents in the Hague, the
Netherlands. Rainfall lasted for more than 7 hours; the total rainfall volume
amounted to 25 mm. All 3 locations were served by combined sewers; streets
were partially flooded over a length of several hundred meters (figure 4.1).
Water samples were taken in duplicate during the flooding incidents; duplicate
sediment samples were taken at one location after flood waters had withdrawn.
The samples were cooled and analysed within 18 hours after sampling.
103
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
Figure 4.1 Flooding on 16 July 2007 at sampling site Scheveningen Boulevard II, the
Hague.
In addition, controlled urban flooding experiments were conducted to test
survival of microbial organisms in flood water. A metal ring (Ø 0.5 m) was
cemented to the street surface on a parking lot and the ring was filled with
wastewater from a nearby combined sewer. The wastewater was diluted with
non-chlorinated tap water to simulate dilution of wastewater with rainwater
during sewer flooding incidents. The dilution factor was chosen based on values
of E.coli and intestinal enterococci found in samples from the flood incidents
and values found in wastewater samples from the combined sewer system. The
controlled flooding experiments were carried out twice per day and on two
separate days, on 10 and 17 October 2007. A dilution factor of 1:20 was chosen
for the first experiment, under dry weather conditions; a factor 1:10 was chosen
for the second, when moderate rainfall had preceded the day of the experiment.
Samples were taken from the water in the ring every 10 minutes for a total
duration of 60 minutes, a typical timescale for urban pluvial flood events in
lowland areas. Samples of the undiluted wastewater were taken at t = 0 and 60
minutes.
104
Microbial risks of exposure to contaminated urban flood water
Figure 4.2 Map of Utrecht city centre; numbered arrows indicate 6 locations where
samples were taken from the combined sewer system. The controlled flooding
experiment was conducted on a parking lot near location 3.
A series of samples was taken from combined sewers during dry weather flow,
to collect data on concentrations of E.coli, intestinal enterococci, Cryptosporidium,
Giardia and Campylobacter in combined sewer water. E.coli and enterococci
concentrations were used to compare values in crude sewage to those in sewer
flood water to obtain a rough estimate of the dilution of sewage during flooding
incidents. Samples were taken from combined sewers in the city of Utrecht, the
105
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
Netherlands, where the controlled flooding experiments were also conducted,
at 6 locations (figure 4.2) and on two subsequent days. In addition, 23 samples
were taken from a combined sewer at 1 location throughout a day, between
7AM and 6PM, at time intervals of 30 minutes. This experiment was conducted
twice, on separate days. All samples were taken in duplicate; dilution series on
count plates were made in duplicate or triplicate. Table 4.1 gives an overview
of the experiments and analyses.
Table 4.1 Overview of experiments
Experiment
Sampling from urban
flooding incident, The
Hague, 16 July 2007, 3
locations
Purpose of experiments
Study concentrations in
water and sediment samples
from an urban flooding
situation
Sample analyses
4 water samples, 1 sediment
sample:
E.coli, intestinal enterococci,
Campylobacter
Controlled flooding
Study survival of microexperiments. Days: 10 and 17 organisms in urban flood
October 2007
water; duration: 60 minutes
4 samples:
E.coli, intestinal enterococci
Spatially distributed
sampling from combined
sewers: 6 locations
Days: 8 and 15 October
2007.
Temporally distributed
sampling from a combined
sewer: 1 location, 7 AM to 6
PM, time step 30 minutes.
Days: 3 and 22 October 2007
42 samples:
E.coli, intestinal enterococci,
Cryptosporidium, Giardia,
Campylobacter
Study concentrations of
microorganisms and 3 types
of pathogens in combined
sewer water under dry-flow
conditions
Study concentration range
of microorganisms over a
weekday
82 samples:
E.coli, intestinal enterococci
Analytical procedures
Samples from the flooding incidents, controlled flooding experiments and
from the combined sewers were analysed for E.coli and intestinal enterococci.
E.coli and intestinal enterococci were enumerated according to international
standards EN ISO 9308-3 (ISO, 1998a) and EN ISO 7899-1 (ISO, 1998b)).
Cryptosporidium, Giardia and Campylobacter were analysed in samples from 6
locations in the combined sewer system; analyses for Cryptosporidium and Giardia
were conducted according to EN ISO 15553 (ISO, 2006); Campylobacter was
determined as a most probable number according to EN ISO 17995 (ISO,
2005).
106
Microbial risks of exposure to contaminated urban flood water
Risk assessment
A screening-level quantitative microbial risk assessment was conducted
according to the approach described in WHO (2003) for recreational waters,
as the first step to identify where further data collection and quantitative
assessment may be most useful. The risk of infection with Cryptosporidium,
Giardia and Campylobacter was calculated for urban flood water, based on
concentration values found in combined sewer water, multiplied by a dilution
index and dose-response relations available in the literature (Haas et al., 1999,
Teunis et al., 1996, Schets et al., 2008).
Pathogen concentrations
Estimates of concentration values for Cryptosporidium, Giardia and Campylobacter
in flood water were based on arithmetic mean concentrations in samples from
the combined sewer system, since Cryptosporidium, Giardia and Campylobacter
were not analysed for the flooding incidents. The concentrations for the
combined systems were multiplied by a dilution factor that was chosen based
on values of E.coli and intestinal enterococci found in samples from the flooding
incidents and values found in samples from the combined sewer system. The
origin of dilution water during the flooding incidents was rainwater run-off that
flowed into the combined sewer system, mixed with sewage water, then flowed
onto the surface as the sewer became overloaded. Additionally, rainwater
that directly fell on the flooded location further diluted the flood water. The
resulting concentration values were used to determine the ingested pathogen
dose, which equals the pathogen concentration in flood water multiplied by the
individual ingested volume per exposure scenario.
Exposure scenarios
Two exposure scenarios were used to estimate infection risks for Cryptosporidium,
Giardia and Campylobacter: accidental ingestion of contaminated flood water
by a pedestrian splashed by passing traffic and accidental ingestion by a child
playing in the water. Ingestion volume for pedestrians was based on values
used for recreators, e.g. fishermen who have accidental contact with water:
107
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
10 ml per incident (Donovan et al., 2008). For children playing in the water,
the ingestion volume was based on that for swimmers (Schets et al., 2008),
assuming that children splash each other and crawl through the water: 30 ml
per incident. For each exposure scenario, infection risk was calculated for a
single exposure event.
Annual risk was then determined based on the assumption that a pedestrian
or a child experiences exposure events with an estimated exposure frequency,
according to:
PInf ,annual = 1 − (1 − PInf , single ) EF (4.1)
Where: PInf,Annual : annual infection risk
PInf,Single : single exposure infection risk
EF
: exposure frequency (exposures/year)
The exposure frequency depends on flood frequency and the presence of a
person at a flooded location. Both vary widely from one system to another
and between locations within a system. A range of exposure frequencies was
used to get an indication of annual risk, from 1 exposure in 10 years (exposure
frequency 0.1/year) to 1 exposure per year.
Dose response relationships
The risk of infection was estimated by using the exponential dose-response
model for Cryptosporidium and Giardia (Teunis et al., 1996, Teunis et al., 1997
and Ottoson et al., 2003):
PInf , Single = 1 − e − r µ 108
(4.2)
Microbial risks of exposure to contaminated urban flood water
Where: PInf,Single : single exposure risk of infection by a certain pathogen
r
: organism-specific constant: rCryptosporidium =0.0040 and rGiardia =0.0199
μ
: pathogen dose (ml)
The Beta Poisson dose-response model was used for Campylobacter (Medema
et al., 1996):
−α
PInf , Single
 µ
≈ 1 − 1 +   β
Provided β>>α
(4.3)
Where: : single exposure risk of infection by a certain pathogen
P*inf
μ
: pathogen dose (ml)
α, β : organism-specific constants; α=0.145, β=7.589.
Comparison with water quality standards
To get a further indication of the potential risk associated with urban flood
waters, concentrations of E.coli and intestinal enterococci in samples from
flood incidents were compared to water quality standards for bathing water
as defined by EU Bathing Water Directive 2006/7/EC (EU, 2006), outlined
in table 4.2. The guideline values refer to levels of risk based on exposure
conditions in large epidemiological studies. From Wiedenmann et al. (2006)
it can be inferred that the guideline value for excellent water quality of 200
intestinal enterococci/100ml corresponds with an attributable risk of 1 to 3%.
Although ingestion volumes for flood water are smaller and exposure times are
shorter than for recreational use of water, the EU Bathing Water Directive is
used to evaluate E.coli and intestinal enterococci values found in samples from
urban flood waters since no health-related standards for flood water exist.
109
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
Table 4.2 Bathing water classification values for inland waters, according to EU
Directive 2006/7/EC (EU, 2006).
Parameter
Excellent quality**
(cfu* 100 ml-1*)
E.coli
500
Intestinal enterococci 200
Good quality**
(cfu 100 ml-1)
1000
400
Sufficient quality***
(cfu 100 ml-1)
900
330
* cfu: colony forming units
** based upon a 95th-percentile evaluation
*** based upon a 90th-percentile evaluation
4.3.
Results and discussion
Flooding incidents
High numbers of E.coli and intestinal enterococci were found in samples of the
flood waters (table 4.3); values found in the sediment were 100 times higher
than in flood water. Campylobacter was detected in all samples. Enterococci
counts in water samples ranged from 5.0x104 to 3.7x105 cfu 100 ml-1, which is
slightly lower than concentration ranges found by Kay et al. (2008) in storm
sewage overflows during high-flow conditions in 12 study areas in the UK:
3.2x105 to 4.5x105 cfu 100 ml-1. The values could not be compared to values
from other flooding incidents, since in the references found samples were taken
weeks after the floods, from remnant flood waters in cellars or surface waters
affected by the floods (Abraham and Wenderoth, 2005; Phanuwan et al., 2006;
Sinigalliano et al., 2007). Analysis methods and parameters analysed in those
studies were also different.
110
Microbial risks of exposure to contaminated urban flood water
Table 4.3 E.coli and intestinal enterococci counts and presence/absence test results for
Campylobacter in samples from 3 urban flooding situations in The Hague on 16 July
2007.
Location
Sample type
Boulevard, I
Boulevard, II
Johan de Wittlaan
Valkenbosplein
Valkenbosplein
Flood water
Flood water
Flood water
Flood water
Sediment
7.0E+05
Campylobacter
Positive
Positive
Positive
Positive
Positive
Isolated sample Exp I
Isolated sample Exp II
6.0E+05
Enterococci count (CFU/100 ml)
Intestinal enterococci
cfu 100 ml-1
5.0x104
3.7x105
2.4x105
2.1x105
1.3x107
E.coli
cfu 100 ml-1
8.7x103
7.0x104
5.0x104
1.0 x105
1.08 x107
Exp I - 17/10/07 - surface
Exp I - 17/10/07 - bottom
5.0E+05
Exp II - 17/10/07 - bottom
4.0E+05
3.0E+05
2.0E+05
1.0E+05
0.0E+00
0
10
20
30
40
50
60
Time from start of experiment (minutes)
Figure 4.3 Enterococci counts in samples from two controlled flooding experiments
on 17 October 2007; a volume of wastewater that was used in the experiment was kept
separate and tested at the beginning and at the end of the experiment (isolated samples)
Controlled flooding experiments
Figure 4.3 shows intestinal enterococci values of two controlled flooding
experiments on 17 October 2007. During the first experiment, samples
were taken near the bottom of the flooded ring and near the water surface.
Enterococci values for these samples showed no difference between bottom
111
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
and surface sample values. Enterococci values found in the second experiment
were lower than in the first for no clear reason; the difference appeared to be
due to accidental variations in the sewer system where the samples were taken
from. Enterococci values did not show a downward trend with time over the
period of the experiment; the values varied only slightly and did not differ from
values in the volume of water that was kept in a closed, separate container. The
same was observed in the first experiment for enterococci values and in the first
and second experiments for E.coli values. This indicates that concentrations of
microorganisms in flood water did not change over the duration of the flood
incident, for incident durations up to 60 minutes. Abraham and Wenderoth
(2005) found high concentrations of pathogenic bacteria in flooded buildings
and on playgrounds days after the river Elbe floods in 2002 (Abraham and
Wenderoth, 2005). These results indicate that sewer flooding leads to the
presence of pathogens in the urban environment over prolonged periods of
time.
8
7
6
Log10CFU 100ml 1
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
5
4
3
2
1
0
03/10/07, E. Coli 08/10/07, E. Coli 15/10/07, E. Coli 22/10/07, E. Coli
03/10/07,
Enterococci
08/10/07,
Enterococci
15/10/07,
Enterococci
22/10/07,
Enterococci
Figure 4.4 Mean, 95% confidence intervals and range of log10 E.coli and enterococci
concentrations in samples from combined sewer systems on 4 sampling days.
112
Microbial risks of exposure to contaminated urban flood water
Sampling from combined sewer system
The variability in E.coli and intestinal enterococci values found (figure 4.4) was
low compared to values found by Kay et al. (2008) in samples of untreated
sewage in the UK, where values of faecal coliforms and enterococci vary by up
to a factor of 6. Mean values of enterococci were in the same order of magnitude
as those found by Kay et al. (2008) in crude sewage under base-flow conditions:
around 106 cfu 100 ml-1.
Intestinal enterococci values in combined sewer water were one order
of magnitude higher than values found in flood water (table 4.3), which
corresponds to a dilution factor of about 10 for flood water. For E.coli, values in
combined sewer water and in flood water varied almost 2 orders of magnitude.
Kay et al. (2008) found enterococci concentrations in untreated wastewater
and crude sewage 2 to 4 times higher under base-flow conditions compared to
high-flow conditions. A dilution factor of 10 was chosen in this study to obtain
an estimate for pathogen concentrations in flood water.
The results of pathogen analyses for 12 samples from combined sewers are
summarised in table 4.4. Cryptosporidium was found in 17% of the samples,
Giardia in 75% and Campylobacter in 25% of the samples. E.coli and intestinal
enterococci were present in 100% of these samples.
These values were of the same order of magnitude as those found by Schijven
et al. (1996) who analysed pathogen concentrations in crude wastewater
at 5 locations in the Netherlands. They reported average Cryptosporidium
concentrations of 17 oocysts/l and maximum concentration of 5.4x103 oocysts/l.
They found average Giardia concentrations of 200 cysts/l, with seasonal
variations from about 10 to 500 cysts/l and a maximum of 1.5x103 cysts/l. Few
studies report on Campylobacter in wastewater; the presence of Campylobacter
in Dutch surface waters influenced by sewage was confirmed by Schets et al.
(2008).
113
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
Table 4.4 Cryptosporidium, Giardia and Campylobacter in samples from combined sewers
in the city of Utrecht
Mean
(of positives)
Range
(min – max of positives)
No. of positive samples/
total samples
Cryptosporidium
(oocysts/l)
12
Giardia
(cysts/l)
5.8 x102
Campylobacter
(cfu l-1)
1.66x104
10-15
20 -1.7 x103
2.3x103-2.4 x104
2/12
9/12
3/12
Comparison with European bathing water quality guidelines
Values of intestinal enterococci and E.coli found in samples from urban flooding
situations are 1 to 3 orders of magnitude higher than values for good bathing
water quality according to the EU Directive 2006/7/EC. While ingestion
volumes and exposure frequencies for flood waters are lower, compared to
bathing water, pathogen concentrations are much higher. This means that
health risks of exposure to flood waters might rise above acceptable risk levels
that this directive is based on.
Risk assessment
A screening-level risk assessment was conducted based on values of
Cryptosporidium, Giardia and Campylobacter found in samples from combined
sewers and a dilution factor 10 for flood water. Table 4.5 summarises the values
used in the risk assessment calculations for each of the three pathogens. Table
4.6 shows single exposure and annual infection risks for 2 exposure scenarios.
These values give an indication of potential infection risks for urban flood
water. It is important to note that the development of a disease after infection
depends on a variety of factors specific to an individual’s immunity. Calculated
annual infection risks vary from to 5x10-6 to 0.3; the minimum value is for
Cryptosporidium based on 12 oocysts/l diluted by a factor 10, 10 ml ingestion
volume, exposure frequency once per 10 year and the maximum value for
Campylobactor based on 1.66x104 cfu l-1, diluted by a factor 10, 30 ml ingestion
volume and exposure frequency once per year.
114
Microbial risks of exposure to contaminated urban flood water
Infection risk values were available in the literature for exposure to pathogens
in surface waters affected by sewage discharges. Schets et al. (2008) found
infection risks per exposure event ranging from 6x10-7 for mean Cryptosporidium
concentrations in Amsterdam canal waters to 1.2 x10-2 for maximum Giardia
concentrations. Donovan et al. (2008) found annual risks of contracting
gastro-intestinal illness of 0.14 to nearly 0.70 for visitor and recreator scenarios
respectively, based on faecal Streptococcus and Enterococcus concentrations in
surface waters in the Lower Passaic River in New York. The results of the
screening-level risk assessment for urban flood waters showed that infection
risk values for urban flood water were in the same range as those found for
surface waters that receive sewage discharges.
Table 4.5 Summary of values used in risk assessment calculations for Cryptosporidium,
Giardia and Campylobacter, for 2 exposure scenarios. Annual risks correspond to
exposure frequencies of 0.1 (min) and 1 (max) per year
Microorganism Mean
Dilu-tion Ingestion
concentration factor
volume
adult-child
Cryptosporidium
Giardia
Campylobacter
12
5.8 x102
1.66x104
10
10
10
10-30
10-30
10-30
Exposure
frequency for
annual risk
Dose-response
relationship
0.1-1
0.1-1
0.1-1
Exponential
Exponential
Beta-Poisson
Table 4.6 Single exposure and annual infection risks for urban flooding situations,
for Cryptosporidium, Giardia and Campylobacter, for 2 exposure scenarios. Annual risks
correspond to exposure frequencies of 0.1 (min) and 1 (max) per year
Microorganism
Single exposure infection risk
Cryptosporidium
Giardia
Campylobacter
Annual infection risk (min-max)
Cryptosporidium
Giardia
Campylobacter
a
Pedestrian
Playing childa
5 x 10-5
1 x 10-2
2 x 10-1
1 x 10-4
3 x 10-2
3 x 10-1
5 x 10-6 - 5 x 10-5
1 x 10-3-1 x 10-2
2 x 10-2-2 x 10-1
1 x 10-5-1 x 10-4
3 x 10-3-3 x 10-2
3 x 10-2-3 x 10-1
In reality infection probabilities for a playing child are higher than the values calculated
here, due to the fact that the dose-response relations used are based on healthy adults
115
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
In lowland areas like the Netherlands, incidents of sewer flooding occur on a
regular, i.e. almost yearly, basis. Repeatedly, citizens are exposed to these flood
waters, as they walk or cycle through them. Abraham and Wenderoth (2005)
have drawn attention to health risks associated with faecally contaminated flood
waters during flood recovery and cleaning activities. In lowland areas, where
sewer flooding is a frequent phenomenon, exposure of people to contaminated
flood waters during daily life activities is at least as serious a reason for concern.
Given the regular occurrence of sewer flooding in lowland areas, it is especially
important that more data on pathogen concentrations in flood waters be
collected to make a more reliable health risk assessment. The need for more and
reliable data becomes more urgent as health risks associated with urban flood
incidents are expected to increase in the future, due to more intense rainfall
induced by climate change, ongoing urbanisation and increasing probability of
component failures in ageing systems (Ashley et al., 2005).
Health risks associated with combined sewer overflows to surface waters
receive much more attention than those related to urban flooding: the EU
Directive 2006/7/EC, EU Water Framework Directive and United States
Clean Water Act place requirements on regulators to manage sources of
microbial pollution for surface waters. The main reason is that recreational
use of contaminated surface waters is associated with higher ingestion volumes
thus a higher likelihood of exposure to pathogens compared to flood waters. On
the other hand, concentrations of pathogens in surface waters are lower (e.g.
Schets et al., 2008; Donovan et al., 2008) than those found in flood water from
overloaded combined sewers. Our study shows that the resulting health risk
could be of the same order of magnitude for both situations. Further studies are
needed to confirm this result; if they do, recommendations or guidelines to limit
exposure of citizens to flood waters in urban environments seem appropriate.
116
Microbial risks of exposure to contaminated urban flood water
4.4.
Conclusions and recommendations
Flood waters resulting from combined sewer flooding incidents are likely to
be contaminated and may pose potential health risks to citizens exposed to
pathogens in these waters. The aim of this study was to evaluate the microbial
risk associated with sewer flooding incidents. Concentrations of Escherichia
coli, intestinal enterococci and Campylobacter were measured in samples from
3 sewer flooding incidents. The results indicate faecal contamination: faecal
indicator organism concentrations were similar to those found in crude sewage
under high flow conditions and Campylobacter was detected in all samples. Due
to infrequent occurrence of such incidents only a small number of samples
could be collected; additional data were collected from controlled flooding
experiments and analyses of samples from combined sewers. The results were
used for a screening-level quantitative microbial risk assessment (QMRA).
Calculated annual risks values vary from 5x10-6 for Cryptosporidium assuming
a low exposure scenario to 0.03 for Giardia assuming a high exposure scenario.
The results of this screening-level risk assessment justify further research and
data collection to allow more reliable quantitative assessment of health risks
related to contaminated urban flood waters.
Collecting samples from flooding incidents is complicated by their
unpredictability. Registration of flood incidents by responsible organisations
will help to point out suitable locations for sampling. Many water authorities
have a call centre that receives calls from citizens who observe problems;
this information can be used to select locations that are repeatedly flooded.
Traditional monitoring by local sensors is not well fitted to collect information
on flooding incidents because spatial resolution is usually too low. Given a
flooding frequency of about once per year per city, instalment of permanent
sampling stations is not an option. A more efficient strategy could be to have
sampling teams stand-by when weather forecast predicts heavy storms and have
local representatives call out when flooding actually occurs. Samples must be
collected from a large geographical area or sample collection must be extended
117
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
over long periods of time to collect a sufficient amount of samples to be able to
draw reliable conclusions.
Exposure of citizens to waterborne pathogens is generally controlled by limiting
access to sites where pathogens are present, e.g. at wastewater treatment plants
and combined sewer overflows, or by reducing sources of pathogens to control
pathogen concentrations as is the case of surface waters for recreational use.
Pathogen concentrations in flood waters cannot be controlled by treatment
and exposure to flooded sites can hardly be avoided for flooding that occurs in
urban environments, on streets and pathways. The best way to control exposure
to pathogens in flood water is probably by raising awareness. If citizens are
aware of potential contamination of flood waters, they are more likely to avoid
ingestion of water and will keep their children away from flood pools.
118
Microbial risks of exposure to contaminated urban flood water
4.5.
References
Abraham, W. R., Wenderoth, D.F. (2005). Fate of facultative pathogenic microorganisms
during and after the flood of the Elbe and Mulde rivers in August 2002. Acta
hydrochim. hydrobiol. 33 (5), 449−454.
Ashley, R.M., Balmforth, D.J., Saul, A.J. and Blanksby, J.D. (2005). Flooding in the future
- predicting climate change, risks and responses in urban areas. Water Sci. Technol.
52(5), 265-273.
Curriero, F.C., Patz, J.A., Rose, J.B., Lele, S. (2001). The association
between extreme precipitation and waterborne disease outbreaks in
the United States, 1948-1994. Am. J. Publ. Health 91(8), 1194-1199
Donovan, E., Unice, K, Roberts, J.D., Harris, M. and Finley B. (2008). Risk
of gastrointestinal disease associated within exposure to pathogens in the
water of the lower passaic river. Appl. Environ. Microbiol. 74(4), 994-1003.
European Union (EU). (2006). Directive 2006/7/EC of the European Parliament and
of the Council concerning the management of bathing water quality and repealing
Directive 76/160/EEC. In: Off. J. Eur. Union 64L, 37-51.
Fewtrell, L., Kay, D. (2008). An attempt to quantify the health impacts of flooding
in the UK using an urban case study. Public Health 122: pp. 446-451
Haas, C.N., Rose, J.B., Gerba, C.P. (1999). Quantitative microbial risk assessment.
Wiley, New York. ISO (1998a). Water quality - Detection and enumeration of
Escherichia coli and coliform bacteria - Part 3: Miniaturized method (Most Probable
Number) for the detection and enumeration of E.coli in surface and wastewater. ISO
9308-3. International Organization for Standardization, Geneva, Switzerland.
ISO (1998b). Water quality - Detection and enumeration of intestinal enterococci - Part 1:
Miniaturized method (Most Probable Number) for surface and waste water. ISO
7899-1. International Organization for Standardization, Geneva, Switzerland.
ISO (2005). Water quality - Detection and enumeration of thermotolerant Campylobacter
species. ISO 17995. International Organization for Standardization, Geneva,
Switzerland.
ISO (2006). Water quality - Isolation and identification of Cryptosporidium oocysts and
Giardia cysts from water. ISO 15553. International Organization for Standardization,
Geneva, Switzerland.
Kay, D., Crowther, J., Stapleton, C.M., Wyer, M.D., Fewtrell, L., Edwards, A., Francis,
C.A., McDonald, A.T., Watkins, J., Wilkinson, J. (2008). Faecal indicator organism
concentrations in sewage and treated effluents. Water Res. 42, 442–454.
Kay, D., Bartram, J., Prüss, A., Ashbolt, N., Wyer, M.D., Fleisher, J.M., Fewtrell, L.,
Rogers, A., Rees, G. (2004). Derivation of numerical values for the World Health
Organization guidelines for recreational waters. Water Res. 38, 1296-1304.
Lenderink, G., van Meijgaard, E. (2008). Increase in hourly precipitation extremes
beyond expectations from temperature changes. Nature Geosc. 1, 511-514.
Lulani, I., Steen van der, P., Vairavamoorthy, K. (2008). Analysis of the public health risks
of the urban water system in Accra by microbial risk assessment. WaterMill Working
Paper Series 2008, no. 8.
Medema G., Teunis P., Havelaar A., Haas C. (1996). Assessment of the dose-response
relationship of Campylobacter jejuni. Int J Food Microbiol. 30(1–2), 101–12.
Ohl, , C. Tapsell, S. (2000). Flooding and human health. British Medical Journal. 321: pp.
1167-1168
119
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 4
Ottoson, J., Stenstrom, T.O. (2003). Faecal contamination of greywater and associated
microbial risks. Water Res. 37, 645–655.
Phanuwan, C., Takizawa, S., Oguma, K., Katayama, H., Yunika, A., and Ohgaki, S. (2006).
Monitoring of human enteric viruses and coliform bacteria in waters after urban
flood in Jakarta, Indonesia. Water Sci. Technol. 54(3), 203–210.
RIONED Foundation (2004). Leidraad Riolering, Module C2100, 17-20 (in Dutch). ISBN
978-90-73645-68-4. Stichting RIONED, Ede, the Netherlands.
Schets, F.M., Wijnen van, J.H., Schijven, J.F., Schoon, H., Roda Husman de, A.M.
(2008). Monitoring of waterborne pathogens in surface waters in Amsterdam, the
Netherlands, and the potential health risk associated with exposure to Cryptosporidium
and Giardia in these waters. Appl. Environ. Microbiol. 74(7), 2069-2078.
Schijven, J.F., Medem, G.J., Nijs de, A.C.M., Elzenga, J.G. (1996). Emission
and distribution of Cryptosporidium, Giardia and enteroviruses via domestic
wastewater (in Dutch). RIVM report nr. 289202014. Bilthoven, the Netherlands.
Sinigalliano, C.D., Gidley, M.L., Shibata, T., Whitman, D., Dixon, T.H., Laws, E.,
Hou, A., Bachoon, D., Brand, L., Amaral-Zettler, L., Gast, R.J., Steward, G.F.,
Nigro, O.D., Fujioka, R., Betancourt, W.Q., Vithanage, G., Mathews, J., Fleming,
L.E. and Solo-Gabriele, H.M. (2007). Impacts of Hurricanes Katrina and Rita on the
microbial landscape of the New Orleans area. PNAS, 104(21), 9029-9034.
Sirajul Islam, M., Brooks, A., Kabir, M.S., Jahid, I.K., Shafiqul Islam, M., Goswami, D.,
Nair, G.B., Larson, C., Yukiko, W., and Luby, S. (2007). Faecal contamination of
drinking water sources of Dhaka city during the 2004 flood in Bangladesh and use of
disinfectants for water treatment. data. J. Appl. Microbiol. 103, 80–87.
Tapsell, S., Tunstall, S. (2003). An examination of the health effects of flooding
in the United Kingdom. Journal of Meteorology 28. No. 183: pp. 341-349
Tunstall, S. Tapsell, S. Green, C. Floyd, P., George, C. (2006). The health effects
of flooding: social research results from England and Wales. Journal of Water and
Health. 4: pp. 365-380.
Teunis, P.F.M., Van der Heijden, O.G., Van der Giessen, J.W.B., Havelaar, A.H.
(1996). The dose-response relation in human volunteers for gastro-intestinal
pathogens. RIVM report nr. 284550002. Bilthoven, the Netherlands.
Teunis, P. F. M., Medema, G.J., Kruidenier, L., Havelaar, A.H. (1997). Assessment
of the risk of infection by Cryptosporidium or Giardia in drinking water from a surface
water source. Wat. Res., 31(6), 1333-1346,
Veldhuis ten, J.A.E., Clemens, F.H.L.R., van Gelder, P.H.A.J.M. (2009). Fault
tree analysis for urban water infrastructure. Structure and Infrastructure
Engineering,
URL:
http://dx.doi.org/10.1080/15732470902985876.
Veldhuis ten, J.A.E., Clemens, F.H.L.R. (2009). Flood risk modelling based on
tangible and intangible urban flood damage quantification. Proceedings of the 8th
International Conference of Urban Drainage Modelling, Tokyo, Japan.
Wiedenmann, A., Krüger, P., Dietz, K., López-Pila, J.M., Szewzyk, R., Botzenhart,
K. (2006). A randomized controlled trial assessing infectious disease risks from
bathing in fresh recreational waters in relation to the concentration of Escherichia
coli, intestinal enterococci, Clostridium perfringens, and somatic coliphages. Environ.
Health Perspect., 114, 228–236.
World Health Organisation (WHO) (2003). Guidelines for safe recreational water
environments: Vol. 1, Coastal and Fresh Waters. World Health Organisation,
Geneva, 219pp.
120
Microbial risks of exposure to contaminated urban flood water
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
121
Chapter 5
Quantification and acceptability of urban
flood risk
This chapter is based on an article that was presented at the “Road Map Towards
a Flood Resilient Urban Environment” conference in November 2009 and was
submitted for a special issue of the Journal of Flood Risk Management.
J.A.E. ten Veldhuis, F.H.L.R. Clemens (2010). How the choice of flood
damage metrics influences urban flood risk assessment.
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 5
Context
Previous chapters have shown that various causes contribute to urban flood
risk and that flood risks in lowland areas are characterised by frequent flooding
of roads and occasional flooding of buildings. The question is to what extent
flood risks are acceptable and how flood risks and investments to prevent
or reduce flood risk can be balanced to constitute a proper urban flood
management strategy. This chapter uses the results of flood risk quantification
in earlier chapters and translates there values into measures that can be used to
set priorities and to justify investments for flood risk reduction.
Abstract
This study presents a first attempt to quantify tangible and intangible flood
damage according to two different damage metrics: monetary values and
number of people affected by flooding. The data used are representative of
lowland flooding incidents with return periods up to 10 years. The results show
that monetarisation of damage prioritises damage to buildings compared to
roads, cycle paths and footpaths. When, on the other hand, damage is expressed
in terms of numbers of people affected by a flood, road flooding is the main
contributor to total flood damage. The results also show that the cumulative
damage of 10 years of successive flood events is almost equal to the damage of
a singular event with a T=125 years return period.
These quantitative risk outcomes provide a more comprehensive basis to
decide whether the current flood risk is acceptable compared to frequency
analyses based on design storms: differentiation between urban functions and
the use of different kinds of damage metrics to quantify flood risk provide the
opportunity to weigh tangible and intangible damages from an economic and
societal perspective.
Keywords
Flood risk, flood damage assessment, urban flooding
124
Quantification and acceptability of urban flood risk
5.1.
Introduction
Previous studies have shown that direct tangible damage cannot sufficiently
describe flood consequences and that intangible damage, particularly physical
and mental health effects should be included in the appraisal of flood risk
alleviation schemes (Tapsell and Tunstall, 2003). A proper aggregation of
quantified flood risk is key to support decision making and can be accomplished
by different flood damage metrics, monetary values being most commonly used.
This is understandable from a decision-making point of view, since monetary
values are most easily compared to capital investments. The question arises
whether decisions based on monetarised flood risk sufficiently account for all
types of urban flood damage, tangible as well as intangible, thus whether such
decisions result in proper flood protection.
In low-lying countries urban pluvial floods are characterised by small depths
and consequently small direct flood damage. For instance, in the Netherlands
direct pluvial flood damage rarely exceeds f5000/household (1998 value, van
der Bolt and Kok, 2000; net present value 2009 €3500, for an interest rate of
4% and translated into euros). As a result, the relative importance of intangible
damage like disturbance of traffic and inconvenience for pedestrians caused
by pools on parking lots and sidewalks increases. The situation of river floods
and flash floods is entirely different. Here, flooding spreads over large areas
and may lead to evacuation of people and complete disruption of communities.
Direct damage to buildings and infrastructure is large and cannot be compared
to the costs of traffic delay or inconvenience. The nature of intangible damage
is different as well: severe floods may cause psychological stress following
evacuation and insurance claim procedures. In lowland areas, pluvial flooding
does not lead to evacuation; damage to buildings, if any, consists of cleaning
costs and in some cases replacement of ground floor carpeting. Under these
conditions, the contribution of traffic delay and inconvenience becomes
important.
125
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 5
Current standards for urban pluvial flooding are usually based on flooding
frequencies and do not take flood damage into account explicitly. European
standards recommend a flooding frequency depending on occupation land use:
1 in 10, 20, 30 or 50 years for rural, residential, commercial and city areas
and underground railway and underpasses. Usually these flooding standards
are interpreted as maximum road flooding frequencies: hydrodynamic models
are used to check compliance with the standards and these calculate manhole
flooding. Implicit in this evaluation of flooding standards is the assumption
that most buildings are located above road level and that by protecting roads,
buildings are protected, too. In current practice, this assumption is not verified;
recent developments in 2D overland flow modelling should enable flooding
calculations at building level in the future.
Climate change predictions have triggered a debate urban among urban
drainage professionals in the Netherlands whether current standards should
be applied to roads and buildings alike or whether temporary flooding of roads
and public spaces can be accepted and only buildings should be protected. In
the light of this discussion flood damage estimation methods should be available
that adequately represent tangible and intangible damages associated with
flooding of buildings, roads and other infrastructure.
The aim of this chapter is to compare two types of metrics for urban pluvial flood
damage estimation incorporating tangible and intangible damage to buildings,
roads and other public spaces: monetary values based on stage-damage
functions and the number of people affected by flooding based on municipal
call centre statistics. The results are used to quantify urban pluvial flood risk for
a case study and to evaluate how the choice of metrics influences the outcomes
and, consequently, decisions to prioritise urban flood risk alleviation.
126
Quantification and acceptability of urban flood risk
5.2.
Quantification method of flood consequences
Data from call centres were classified according to damage classes. Table 5.1
gives a summary of primary functions and damage classes that were used for
call classification. For illustration, the numbers of calls in each class for the case
of Haarlem that were used in this study are added.
Table 5.1 Primary functions of urban drainage systems and damage classes used for
municipal call classification. The numbers of calls in each class are given for the case of
Haarlem city (calls totalled for rain events and dry events).
Primary functions
Protection of human health:
physical harm or infection
Protection of buildings and
infrastructure against flooding:
damage to public and private
properties
Prevention of road flooding:
traffic disruption
C1
C2
C3
C4
C5
C6
C7
C8
Damage classes
# of calls
Flooding with wastewater (toilet
20
paper/excreta)
Manhole lid removed
4
Flooding in residential building
78
(house/flat/garage/shed)
Flooding in commercial building
26
(shop/restaurant/storage hall)
Flooding on residential/main road
596
Flooding of sidewalk/cycle path
344
Flooding at bus stop/taxi stand/bus
18
or train station
Flooding in shopping street/
155
commercial centre
The assignment of classified calls to independent incidents results in a list of
incidents and numbers of calls per damage class per incident. These results
are translated into damage estimates per incident per damage class and total
damage estimates per damage class. Translations are based on a number of
assumptions with respect to the amount of damage and number of affected
people per call for different damage classes. Uncertainty is introduced through
the assumptions made for translation due to a lack of data. This uncertainty is
incorporated by assuming that each damage estimate has a uniform probability
distribution: it varies between a minimum and a maximum estimate and all
values in between have an equal probability:
127
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 5
 1
for α ≤ x ≤ β

f ( x) =  β − α
 0 for x < α or x > β

 
    

       




   distributed

Where: x : uniformly
variable
f (x)
β
α,
(5.1)




: probability density function of x
:
minimum and maximum boundaries of x

 

are calculated
 expected
 value
The
and the variance of a uniform distribution

as follows:

α + β
E ( X ) =    (5.2)








    2 

              
5.3





  






Where:
value of X
 : expected


E(X)

VAR(X)
:
variance
of
X



STD(X)
of X


: standarddeviation


        

Assumptions
for urban flood risk assessment metrics: stage damage curves

Stage-damage curves that are usually used in flood damage assessment are

based on information about depth, velocity and other characteristics of flood

waters.
If call texts are to be used as input for stage-damage curves, a flood

depth
must be derived from the call text. Call texts do not specify flood depths;

they
repeatedly mention that “water comes flowing into the house” or similar


statements.
Call texts indicate that floors and carpets are often wetted, yet

water depths are unlikely to exceed 10 cm: none of the calls mention high water

levels or high velocity flows. Since flood depths are small, only the low ranges

of stage-damage functions are applicable.


128







Quantification and acceptability of urban flood risk
In this study, stage-damage information from studies in Germany (Apel
et al., 2009) and the Netherlands (Gersonius et al., 2006) is used. As a first
approximation, a flood depth of 10 cm was assumed for all calls in classes
concerning flooding of buildings. Related damage according to stage-damage
functions varies from €10,000 to €30,000 for residential buildings. A minimum
of €1000 was assumed here to account for cleaning costs. None of the call texts
related to flooding of commercial buildings report damage to inventories, one
call mentions that customers tend to leave as water flows in. Since available
information does not suggest principle differences in costs, the same stagedamage functions were used for residential and commercial buildings. Yet for
commercial buildings a higher minimum of €2000 per flooded building was
assumed to account higher cleaning costs.
Assumptions for urban flood risk assessment metrics: costs of traffic delay
and annoyance
No references of stage-damage curves for traffic losses due to urban flooding
have been found. Traffic losses mainly relate to the costs of traffic delay, which
have been quantified in congestion cost studies. Most of these studies relate to
highways, few relate to traffic in urban areas. Bilbao-Ubillo (2008) quantified
congestion costs in urban areas at €12.50 per hour of delay. Based on traffic
counts for main roads in Haarlem (Haarlem, 2008) a minimum and a maximum
amount of vehicles were estimated for residential roads. A traffic delay of 5
minutes per vehicle was assumed for pools on residential roads, equal to a delay
of one cycle at traffic lights.
Flooding of cycle paths, sidewalks, bus stops etc. merely causes annoyance
to cyclists and pedestrians. A study in the UK (Defra, 2004) quantified the
willingness-to-pay to avoid health impacts associated with flooding. Health
impacts included physical and psychological effects of homes being flooded.
Although these effects refer to more serious flooding situations, the willingnessto-pay (WTP) value from this study was taken as an upper boundary: €220.
The lower boundary was set at €0.
129
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 5
Assumptions for translation of call data into monetary damage are summarised
in table 5.2, for all classes.
Table 5.2 Assumptions damage metrics for flood risk assessment
Damage classes
C1
C2
C3
C4
C5
C6
C7
Monetary damage
Remarks
Min(€) Max(€)
Flooding with wastewater
0
220
Max: WTP to prevent health
effects of flooding
Manhole lid removed
0
220
Idem C1
Flooding in residential building
1000 30000 Min: cleaning costs only; max:
flood depth 10 cm, medium
building value
Flooding in commercial building
2000 30000 Idem C3; min cleaning costs for
larger building surface
Flooding on residential/main road
10
700
10-700 vehicles; 5min delay/
vehicle; €12.5/hr
Flooding of foot//cycle path
0
220
Idem C1
Flooding at bus stop/taxi/train station
0
220
Idem C1
Assumptions for urban flood risk assessment metrics: affected people
Table 5.3 summarises assumptions used in this study for the numbers of affected
people per call in every damage class. Assumptions for car and cycle traffic were
based on figures from the yearly statistics report for the city of Haarlem, year
2007 (Haarlem, 2008). Other assumptions are based on oral communications.
Table 5.3 Assumptions for number of affected people per call in damage class
Damage classes
C1
C2
C3
C4
C5
C6
C7
# affected people
Remarks
Min Max
Flooding with wastewater
10
100 10-100 pedestrians or cyclists
cycle or footpath
Manhole lid removed
5
500 5-500 cyclists or cars on road or
cycle path*
Flooding in residential building
2
5 Size of household
Flooding in commercial building
2
10 Owner, personnel and customers
Flooding on residential/main road
30
500 30-500 vehicles per 15 min.*
Flooding of foot/cycle path
5
115 5-115 cyclists per 15 minutes*
Flooding at bus stop/taxi/train station 10
20 10-20 travellers waiting at bus
stop/station
*Source: Haarlem, 2008. Yearly statistics 2007
130
Quantification and acceptability of urban flood risk
Acceptability of flood risk
Based on the quantified risk outcomes, the acceptability of flood risk was
assessed. The acceptability of flooding and the need for investments to reduce
flood risk can be based on societal consideration or it can be viewed as an
economic decision problem or a combination of these. Economic cost-benefit
analyses offer the advantage of a direct comparison between costs and benefits
in monetary terms; the disadvantage of translating all parameters into monetary
values is the amount of uncertainty that is introduced through assumptions that
have to be made for translation. Additionally, the damage schematization made
by this translation does not necessarily reflect public perception of the potential
loss.
Societal risk is usually expressed in the form of an FN-curve that displays the
probability of exceedance of the number (N) of deaths or casualties. A limit line
can be drawn in a graph depicting an FN-curve to define maximum acceptable
risk. Such limit lines can be described by the following formula (e.g. Jonkman
et al., 2003):
1 − FN ( x) <
C
xn
(5.1)
Where n is the steepness of the limit line and C the constant that determines
the position of the limit line. A limit line with a steepness of n = 1 is called
risk neutral; a line with steepness n = 2 is called risk averse (Vrijling and van
Gelder, 1997). Similarly, an FD-curve displays the probability of exceedance
as a function of the economic damage, D (Jonkman et al., 2003).
The type of flood events investigated in this thesis cause some tangible and a lot
of intangible damage. The results can be depicted as an exceedance curve of the
number of calls (C) per event, an FC-curve. The expected value of flood risk
equals the area under the FN-curve (Vrijling and van Gelder, 1997). Similarly,
the expected value of flood risk in terms of the number of calls per event equals
the area under the FC-curve.
131
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 5
5.3.
Results and discussion
The results of flood damage quantification for the case of Haarlem are
summarised in table 5.4. The results show that total flood damage over the
period 1997 to 2007 amounts to between 153 kEUR and 1688 kEUR and that
18,000 to 296,000 people are affected by flooding. Table 5.5. shows expected
values and variances of damage in each consequence class, calculated according
to formulas 5.2 and 5.3.
Table 5.4 Total urban flood damage for damage classes C1 to C7 for the city of
Haarlem, period 1997-2007; based on assumptions for damage quantification according
to monetary values and numbers of affected people.
Damage classes
C1
C2
C3
C4
C5
C6
C7
132
Flooding with wastewater
Manhole lid removed
Flooding in residential building
Flooding in commercial building
Flooding on residential/main road
Flooding of sidewalk/foot/cycle path
Flooding at bus stop/taxi/train station
Total
monetary monetary affected affected
damage
damage
people
people
*1000 EUR *1000 EUR
*1000
*1000
min %
0
0
0
0
98 64
50 33
5
3
0
0
0
0
153 100
max % min % max %
2
0
0
0
1
0
1
0
0
0
2
1
980 58
0
0
1
0
250 15
0
0
0
0
349 21 15 83 250 85
87
5
2 11 41 14
19
1
1
6
2
1
1688 100 18 100 296 100
Quantification and acceptability of urban flood risk
Table 5.5 Total urban flood damage for damage classes C1 to C7 for the city of Haarlem,
period 1997-2007; expected values and variance
Damage classes
monetary
monetary affected affected
damage
damage people
people
*1000 EUR *1000 EUR
*1000
*1000
E(X)
C1
C2
C3
C4
C5
C6
C7
Flooding with wastewater
Manhole lid removed
Flooding in residential building
Flooding in commercial building
Flooding on residential/main road
Flooding of sidewalk/foot/cycle path
Flooding at bus stop/taxi/train station
Total
1
0
539
150
177
44
10
921
STD(X)
%
0.5
0.2
255
58
99
25
5.5
E(X) STD(X)
0
1
0
0
132
22
1
156
0.2
0.4
0.08
0.06
68
11
0.3
The results show that flooding of buildings contributes most to flood damage
expressed in monetary values, whereas road flooding affects the largest number
of people. In other words: flooding incidents that affect many people do not
cause large monetary damage. This outcome was obtained for one of the two
case studies, the city of Haarlem. The results presented in chapter 4 show that
ratios between consequences classes related to building flooding and those
related to street flooding are similar for the 2 case studies, Haarlem and Breda.
Therefore, the results shown in table 5.4 are likely to be representative of
flooding incidents with return periods of less than 10 years in medium size cities
in lowland areas. The question to what extent results of the two case studies can
be generalised to other cities in lowland areas is discussed in chapter 7.
Figure 5.1 gives a graphical presentation of the data in table 5.5. It shows
that monetary damage to residential building (class C3) is significantly larger
than monetary damage to commercial buildings (C4) and monetary damage
due to flooding of roads (C5), of sidewalks and cycle paths (C6) and of bus
stops (C7). Monetary damage to commercial buildings is of the same order of
magnitude as monetary damage due to flooding of roads. This is a result of a
low incidence of flooding of commercial buildings associated with large damage
133
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
per incident and of a high incidence of road flooding with small damage per
incident. The number of people affected by road flooding is larger than for all
other classes. The expected values of numbers of people affected for classes C1
to C4 and C7 are less than 1% of the expected value of the number of people
affected for class 5. Figure 5.1 shows that even if damage estimates are subject
to large uncertainty as a result of assumptions underlying cost calculations,
discrepancies between damages in most classes are significant.
900














Monetary damage
800
200
180
Nr of affected people

600

140






120
500
100
400
80
300
60
200
40
100
0
Nr of affected people (*1000)
160
700
Monetary damage (kEUR) .
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 5
20


C1
0
C2
C3
C4
C5
C6
C7
Figure 5.1 Total urban flood damage for damage classes C1 to C7 for the city of
Haarlem, period 1997-2007. Data points show mean values of monetary damage and
number of affected people per class, error bars show standard deviations from the mean.
Figure 5.2 is based on the same results as table 5.4; instead of total values at city
level, the minimum and maximum values in figure 5.2 are expressed in terms
of damage per km sewer length per year. Threats to human health caused by
wastewater flooding and uplifted manholes are almost negligible, as a result
of low occurrence and low damage values. Monetarised damage to buildings
134
Quantification and acceptability of urban flood risk
exceeds other kinds of monetarised damage, yet the number people affected by
building flooding is low. Flooding of roads, cycle paths and foot paths results in
low monetary damage, yet affects large numbers of people.
Nr of affected people/km sewer length/year (-)
City: Haarlem
-60
-40
-20
0
20
C1, wastewater
40
60
Nr of affected people min
Nr of affected people, max
C2, manhole lid
Monetary damage min. estimate
Monetary damage max. estimate
C3, res. building
C4, comm. building
C5, res/main roads
C6, foot/cycle path
C7, bus/train stop
250
200
150
100
50
Monetary damage/km sewer length/year (EUR) .
0
-50
-100
-150
-200
-250
Figure 5.2 Monetary flood damage in EUR per km sewer length per year and number
of people affected by flooding per km sewer length per year for damage classes C1 to
C7, case of Haarlem
Acceptability of flood risk
Based on the quantified risk outcomes presented in table 5.4 and figure 5.2 FCcurves were drawn for urban flood risk expressed in terms of numbers of calls.
The resulting FC-curve is shown in figure 5.3. The expected value of flood risk
equals the area under the FN-curve (Vrijling and van Gelder, 1997). Similarly,
the expected value of flood risk in terms of the number of calls per event equals
the area under the FC-curve in figure 5.3 An example of a limit line for damage
to properties is drawn in figure 5.3, for n=1 and C=10-2, according to formula
5.1. This example shows that for the chosen risk neutral limit line, the risk of
damage to properties is acceptable for events with more than 7 calls, that are
135
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
below the limt line and is unacceptable for small events, with 1 to 7 calls above
the limit line. The result signifies that the risk level resulting from current flood
protection strategy is risk-averse, protection from larger events is higher than
required according to a risk neutral approach whereas protection from small
events is low compared to a risk neutral approach.
Annual probability of events with at least N calls/km sewer
length
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 5
1




0,1



0,01




0,001





















































disturbance of traffic
damage to private properties
threats to human health
Limit line private properties
0,0001
0,00001
1
10
N: Number of calls per event
Figure 5.3 Exceedance curves of the number of calls per event, for 3 types of flood
damage, for the case of Haarlem. A limit line for damage to properties, for n=1, is drawn
as an example.
Economic evaluation of risk acceptability is usually based on monetarised values
of flood risk. The acceptable economic risk can be defined as a fixed maximum
expected flood damage or can be the outcome of an economic optimisation of
flood damage versus investment costs for flood protection.
For comparison, the cumulative costs of building flooding as a result of small
flood events, as calculated in this study, is compared to the costs of building
136
100
Quantification and acceptability of urban flood risk
flooding as a result of a singular rare event. The cumulative costs for small
events are derived from table 5.4; rare event damage data are derived from Van
der Bolt and Kok (2000). Their data concern a pluvial flood event in 1998 with
an estimated return period of 125 years. This event was classified as a national
disaster and fell under the Dutch Compensation Act. Table 5.6 presents a
summary of the cumulative costs of successive events over a 10 year period
versus the costs of the T=125 years event.
This table shows that the cumulative monetary damage to buildings per
affected person over a period of 10 years is of the same order of magnitude as
the damage per person for a T=125 years event. Damage per affected person is
based on the expected value of damage estimates and estimates of the number
of affected people.
While the severe event damage was considered eligible for compensation by the
national government, cumulative damage is not compensated; the responsibility
is left with private owners to seek insurance against pluvial flood damage.
Table 5.6 Cumulative flood damage to buildings and roads for 10 years of successive
events versus singular event damage to buildings for a rare event
Flooding of buildings
Monetary Number Monetary costs/ Monetary costs/
costs of people affected person affected person/
(*1000 €) affected
(€)
year(€)
Flooding of buildings (tangible damage)
Expected value of cumulative
689
490
costs of small events, 10 years
Costs per household of T=125
3.11
24002
years event
Flooding of roads, cycle paths etc. (intangible damage)
Expected value of cumulative
230 155,000
costs of small events, 10 years
Sewer tax (partially spent on flood protection)
Cumulative sewer taxes, 10 years
68,0003 147,000
1400
134
1360
55
1.5
0.14
450
45
2009 value, based on1999 value €2000 and interest rate 4%
1
1050 houses, average household size 2.3 (CBS)
2
3
Average sewer tax 1997-2007: €90/year; 76,000 households
137
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 5
This outcome confirms a risk-averse attitude: small accidents are more easily
accepted than one single rare accident with large consequences, even though
the expected damage is similar in both cases (Vrijling, 2001). The results also
show that for people affected by flooding of buildings, the yearly damage is
likely to exceed the amount of yearly sewer tax paid.
In an economic evaluation, the question is whether more efficient flood
protection could be achieved by investments to reduce flood risk and if so,
whether it is more efficient to reduce the probability or the consequences
component of flood risk. Given the uncertainties in the current study, the
outcome of such evaluations is inevitably uncertain. Appendix 1 illustrates the
effect of call data uncertainty on potential decisions for flood risk reduction. A
comprehensive evaluation of investments versus reduction of flood risk requires
additional knowledge on the costs and effects of maintenance strategies, for
gully pot cleaning, sewer cleaning, repair of manifolds etc. that can be obtained
from experiments, preferably on real-world scale.
5.4.
Conclusion
This study is a first attempt to gain insight into different kinds of flood damage
and to find quantitative measures for comparison of direct damage and
indirect, intangible damage. Flood quantification studies tend to be based on
monetarisation of damage, which leads to a prioritisation of tangible damage
to buildings over intangible damage associated with flooding of roads, cycle
and footpaths. Application of different kinds of damage metrics provides the
opportunity to weigh tangible and intangible damages in various ways and to
evaluate flood damage in a more balanced way.
The results show that flood protection for the investigated case is risk-averse:
protection from small events is low compared to larger events. The results also
show that the number of people affected by tangible damage is small compared
138
Quantification and acceptability of urban flood risk
to those affected by intangible damage. Based on the available data it cannot
be concluded whether the current protection level is an economic optimum: the
effect of investments to reduce flood risk, especially those related to increased
maintenance, are too uncertain. The final question to be answered is whether
the distribution of damage over small and large events and over tangible and
intangible damage correctly reflects a safety level that is considered acceptable
by society. This is in essence a political question, because costs and benefits
of flood protection contain aspects that may be valued differently by different
stakeholders.
If flooding standards and investments prioritisation are to be based on a risk
approach, data and model predictions must be able to discriminate between
different kinds of flood damage and flooding causes to support policy
development and decision making. Recent development of two-dimensional
overland flow models can help to make a distinction between flooding
consequences related to roads and buildings. Data on asset failures are essential
to quantify their contribution to flood risk. Call data, complemented with other
flood incident observations can help to provide data on asset failures and can
be used for flood model calibration and verification.
139
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 5
5.5.
References
Apel, H., Aronica, G.T., Kreibich, H., Thieken, A.H. (2009). Flood risk analyses—how
detailed do we need to be? Natural Hazards 49, 79–98
Bilbao-Ubillos, J. (2008). The costs of urban congestion: Estimation of welfare losses
arising from congestion on cross-town link roads. Transportation Research Part A
42 (2008), 1098–1108
CBS, Central Statistics Bureau (s.d.). http://www.statline.cbs.nl; visited 25.11.2009.
Defra (2004). The appraisal of human related intangible impacts of flooding. Defra, Flood
management division, R&D Technical Report FD2005/TR, London, UK.
Gemeente Haarlem (2008). Jaarstatistiek Haarlem 2007 (in Dutch). Yearly statistics for
Haarlem, 2007. Haarlem, the Netherlands.
Gersonius, B., Zevenbergen, C., Puyan, N., Billah, M.M.M.. (2006). Efficiency of private
flood proofing of new buildings – Adapted redevelopment of a floodplain in the
Netherlands. Proceedings of FRIAR conference, UK, 2006.
Tapsell, S. and Tunstall, S. (2003). An examination of the health effects of flooding in the
United Kingdom. Journal of Meteorology, 28(238), 341-349.
Thieken, A.H., Muller, M., Kreibich, H., Merz, B. (2005). Flood damage and influencing
factors: New insights from the August 2002 flood in Germany. Water Resources
Research, 41(12), 1-16.
Van der Bolt, F.J.E. and Kok, M. (2000). Hoogwaternormering regionale watersystemen.
Schademodellering. (Standardisation of flooding in regional water systems. Damage
modelling). HKV Lijn in Water and Alterra, the Netherlands.
Vrijling, J.K. (2001). Probabilistic design of water defense systems in the Netherlands.
Reliability Engineering and System Safety, 74(2001), 337-344.
Vrijling, J.K., van Gelder, P.H.A.J.M. (1997). Societal risk and the concept of risk aversion.
In: C. Guedes Soares (Ed.), Advances in Safety and Reliability, vol. 1, Lissabon,
1997, pp. 45–52.
140
Quantification and acceptability of urban flood risk
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
141
Chapter 6
Risk-based urban flood management:
improving operational strategies
This chapter is based on an article that was presented at the LESAM conference in
Miami, November 2009.
J.A.E. ten Veldhuis, J. Dirksen and F.H.L.R. Clemens, Evaluation of operational
strategies to control sewer flooding based on failure data. In: Proceedings of
the 3rd International Leading Edge Conference on Strategic Asset Management
(LESAM), AWWA, Miami, 2009.
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
Context
As previous chapters have shown, various causes contribute to urban flood
risk, thus various kinds of actions can be undertaken to reduce flood risk. In
this chapter three causes of flooding are compared, evaluates associated flood
management strategies and specifies efficient ways to enhance current strategies
to further reduce flood risk.
Abstract
Data from call centres at two municipalities were analysed in order to
quantify flooding frequencies and associated flood risks for three main failure
mechanisms causing urban flooding. The aim was to find out whether current
operational strategies are efficient for flood prevention and if directions for
improvement could be found. The results show that quantified flood risk for
the two cases is well above the standard which is defined in sewer management
plans. The analysis pointed out that gully pot blockages are the main cause
of flooding. Reactive handling of calls, as is currently applied, is inefficient
if all calls are reacted upon since a small portion of all calls report serious
consequences like flooding in buildings or wastewater flooding. Preventive
cleaning of sewer pipes proves to be an efficient strategy to reduce flooding
due to sewer blockages as flood risk associated with sewer blockages is lower
in case of higher cleaning sewer frequencies. Sewer blockages often have
serious consequences, thus preventive handling is to be preferred to reactive
cleaning. According to the results of this analysis, reduction of flooding sewer
overloading is not of primary concern, because serious consequences for this
failure mechanism are rare compared to other failure mechanisms.
Three flood reduction strategies are compared with respect to their efficiency
in flood risk management in a fictitious decision making example. The results
show that increasing gully pot blockages frequency is a more efficient strategy
to reduce flood risk than increasing sewer cleaning frequency or increasing
sewer pipe capacity. Keywords: asset management, flooding, urban drainage
144
Risk-based urban flood management: improving operational strategies
6.1.
Introduction
In recent years, increased media attention for urban flood incidents and
uncertainties in climate change predictions, have inspired discussions
among urban drainage managers about the need for investments in
sewer systems to improve urban flood prevention. Research in the area
of civil structures like dams, dikes and water supply systems (Tuhovčák,
2007) has shown how risk analysis can support design and operational
decisions, in particular those that involve uncertainties. These may include
uncertainties about future developments like climate change as well as
uncertainties about the functioning and condition of drainage systems.
In chapter 2, quantitative fault tree analysis was applied to urban flooding in
order to detect and quantify causes of urban flooding. It was shown that the
contribution of component failures to flood incident frequency was larger than
that of sewer overloading by heavy rainfall. Typically, component failures in
urban drainage systems are hard to detect and inspection techniques are costly,
since most of the system is underground. As a result, inspection frequencies are
usually low and urban drainage operators often resort to reactive maintenance
to solve failures. The objective of this chapter is to evaluate operational
strategies for prevention of sewer flooding based on a risk assessment, in order
to find out whether currently applied strategies are efficient and how they can
be improved. Current strategies largely build upon many years of practical
experience supported by few quantitative data. This study uses failure data
related to sewer flooding incidents to quantify flood risks.
Operational strategies fall into two main categories (Bedford and Cooke,
2001): corrective and preventive strategies. Corrective strategies aim to repair
a defect, fault or failure after it has occurred; preventive strategies form part of
regular servicing. Table 6.1 summarises types of strategies and scheduling of
operational activities.
145
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
Table 6.1 Strategies and scheduling of operational activities (from: Bedford and Cooke,
2001)
Scheduling of activities
Calendar-based
Corrective
-
Condition-based
Upon observation of
degradation; functionality still
in place
If suitable opportunity
presents itself and degradation
has been observed
When component is in a state
that disables the system;
usually immediately after
failure
Opportunity-based
Emergency
Preventive
Fixed cycles of operational
activities
If suitable opportunity
presents itself, while no
degradation has been observed
-
Corrective strategies are applicable when failures can be detected rapidly
and do not have immediate disastrous consequences. They consist of repair
actions in response to detected failures. Corrective strategies require condition
monitoring and inspection to identify the point at which repair is needed.
Preventive strategies consist of maintenance activities based on a fixed schedule
or following opportunities. Operators decide upon what strategy to prefer based
on efficiency in terms of time, energy and costs. In urban drainage practice such
decisions are usually made implicitly, without explicit quantification of time,
energy or costs of strategy implementation versus prevented consequences.
This chapter focuses on three sewer failure mechanisms that are main
contributors to sewer flood risk: sewer overloading, sewer pipe blockage and
gully pot blockage. Common strategies to avoid failure according to these
mechanisms are briefly summarised for the situation in the Netherlands.
Sewer overloading is dealt with by defining a design standard for flooding
frequency, usually once per year or per 2 years (RIONED, 2004). Compliance
with this standard is checked by mostly unvalidated model calculations conducted
in the design stage. Calculations are repeated approximately every 10 years. If
according to these calculations sewer flooding frequency exceeds the design
standard, an improvement measure is designed and implemented following a
146
Risk-based urban flood management: improving operational strategies
preventive approach. If model results are not trusted or if insufficient budget
is available, improvements are postponed or cancelled. Besides the preventive
approach, complaints from citizens about flooding may form a reason to react
and implement structural improvements.
Sewer blockage is tackled in two ways: following inspection and upon citizens’
complaints. Sewer inspection is complicated and expensive compared to other
infrastructure, because it must be done with special equipment that can enter
the sewers, typically a camera mounted on a robot vehicle, which in addition
requires previous sewer cleaning. As a result, sewer inspection frequencies
are usually low, of the order of once every 10 years. When blockages occur
in the period between inspections and lead to flooding, these are resolved
only if citizens complain about the flooding. Since most sewer systems in the
Netherlands are looped networks, pipe blockage normally leads to flooding
in main transport routes and where local transport capacity is critical. Gully
pots are usually cleaned once a year; vulnerable locations like market places
and shopping streets are often cleaned 2 or 4 times yearly. In addition, gully
pots are cleaned upon complaints, usually within a maximum period of 1 or
2 weeks after the complaint was made. These strategies have developed over
many years of practical experience and in the Netherlands there is a common
agreement among sewer managers that this is an efficient way to cope with
failure mechanisms. This is reflected in corresponding recommendations laid
down in the Dutch Sewer Guidelines (RIONED, 2007). The aim of this
chapter is to find out whether failure data confirm this common agreement
about the efficiency of current strategies and if analysis of failure data can point
out directions for improvement.
147
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
6.2.
Methods
Urban flood incident data
Data on urban flood incidents were obtained from municipal call centres that
register information from citizens’ calls about observed flood problems and
ensuing information from technical staff after on-site investigation. Sewer
inspection data were not used, since data sets were small and inspection data
have proved to be unreliable (Dirksen et al., 2007). Call data from the cities of
Haarlem and Breda were analysed to detect characteristics of failure processes
for the three failure mechanisms described in the introduction of this chapter.
Table 6.2 summarises characteristics of the sewer systems and maintenance
regimes for the two cases.
Table 6.2 Summary of data for the cities of Haarlem and Breda: sewer system
characteristics, maintenance regime
Data case study
Number of inhabitants
Length of sewer system (% combined)
Total surface connected to sewer system
Total number of gully pots
Maximum ground level variation
Maintenance regime
Gully pot cleaning
Sewer cleaning
Haarlem
147000
460 km (98%)
1110 ha
42500
20 m
Breda
170000
740 km (65%)
1800 ha
80000
10 m
1x/year + upon calls
1x/year + upon calls
62km/yr (13% of total 65km/yr (6% of total
sewer length)
length)
Relative contributions of the failure mechanisms to flooding frequency
were quantified as well as their expected consequences. Consequences were
quantified in terms of the number of calls per failure mechanism per flooding
incident. Most calls refer to only 1 location, so that the number of calls per
incident equals the number of reported flooded locations per incident for
95% of all incidents. Call data were verified by checking consistence of call
information with respect to rainfall data and hydrodynamic model calculation
results (see chapter 2).
148
Risk-based urban flood management: improving operational strategies
R1
Probabilistic risk analysis
R2
Occurrence of flooding was evaluated in terms of flooding frequencies and
R3
flood risk related to various consequences: flooding in buildings, wastewater
R4
flooding and flooding in general, including the former two and flooding of
R5
streets, sidewalks, gardens etc. Flooding frequencies were drawn from incident
R6
occurrences over the period of available data. Flood risk was quantified by
R7
multiplication of incident occurrence probability by average number of locations
R8
per incident. The average number of locations per incident was assumed to be
R9
equal to the average number of calls per incident; this generalisation holds for
R10
95% of all incidents.
R11
         




(6.1) 

R12
 
  


    








R13



Where: R
: risk of flooding in amount of flood locations in



R14
 period oftime

t
 
       
R15
      


 
: probability of flooding in period of time t


R16
 
:
Average
consequence
of
flooding
incidents




R17
expressed as the number of locations per incident:



R18

total number of calls divided by total number of


R19

flooding incidents


R20





R21
Risk-based decision making for flood risk reduction



R22
Risk-based urban flood management uses outcomes of urban flood risk analysis



to support decisions for flood reduction. Quantitative risk analysis results for
R23



the case of Breda were used to demonstrate how quantitative risk values based
R24



on
call
data
analysis
can
support
decisions
for
urban
flood
risk
reduction.
Three



R25

possible
actions
to
reduce
flood
risk
were
compared:
increasing
sewer
capacity


R26



to reduce sewer overloading, increased sewer cleaning frequency to reduce
R27



sewer blockage and increasing gully pot cleaning to reduce blockage. To account
R28



for the effect that call data represent only a part of the total number of flood

 R29

incidents true an estimate is made of the percentage of citizens that is expected

 R30

to make a call to a municipal call centre out of the total number of citizens



R31



R32





149


















R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
who observe unsatisfactory urban drainage conditions. Based on Wiechen et al.
(2002) and Devereux and Weisbrod (2006) the expected percentage of citizens
who make a call was estimated between 2% and 30%.
The effect of uncertainties in call data on flood risk estimates is discussed in
detail in a sensitivity analysis in Appendix 1. The effects of flood risk reduction
actions were estimated based on expert judgment, since insufficient data were
available to quantify the effect of these actions.
6.3.
Results
Tables 6.3 and 6.4 give the results of call data analysis for the 3 failure mechanisms
‘gully pot blockage’, ‘sewer pipe blockage’ and ‘sewer overloading’, for the cases
of Haarlem and Breda. A distinction is made between the classification results
for rain events and dry events and between various groups of consequences.
Comparison between failure mechanisms
Tables 6.3 and 6.4 show that calls which explicitly report flooding-related
consequences make up 25% of all calls for Haarlem and 38% for Breda. A
small portion of these calls report flooding in buildings or flooding with
wastewater. The results for flooding in buildings and flooding with wastewater
were analysed separately, because these are severe consequences compared to
flooding of streets and parks. Flooding of streets never causes traffic disruption
or damage according to the call texts, probably because both case study areas
are more or less flat.
For both cases gully pot blockages are reported far more often than the other
two failure mechanisms. The amount of calls per incident is also highest for
gully pot blockages, indicating that more locations per incident are affected.
This applies for all flooding-related calls together as well as for calls on flooding
in buildings and calls on wastewater flooding separately. Sewer overloading
rarely leads to flooding in buildings or flooding with wastewater. The same
applies for sewer blockage in Haarlem; in Breda blocked sewers are a frequent
150
Risk-based urban flood management: improving operational strategies
cause of flooding in buildings. In Haarlem blocked sewers are the main cause of
wastewater flooding Calls that report wastewater flooding caused by gully pot
blockage mostly refer to erroneous connections to gully pot mains which results
in wastewater flooding. Some calls were misclassified and refer to blockage of
house connections instead of gully pots.
The amount of flood-related calls during dry incidents is lower than during rain
incidents, except for flooding with wastewater which occurs more or less as
often during dry and rain incidents. Detailed investigation of call texts shows
that flood-related calls during dry incidents often refer to rainfall on previous
days. Reference to previous days is especially common on Mondays, since call
centres are closed during the weekend. Other dry incident calls do not refer to
particular incidents; these calls usually report minor flooding.
Table 6.3 Results call data analysis Haarlem, 3 failure mechanisms for sewer flooding.
Call data for Haarlem cover a period of 10 years; in this period 566 independent rain
incidents occurred and 566 dry incidents following each rain incident.
Haarlem
# of incid # of calls
flooding-related
Failure mechanisms
consequence classes
Rain incidents
Gully pot blockage
202
897
Blocked sewer pipe
6
6
Sewer overloading
10
15
TOTAL
218
918
Dry incidents
Gully pot blockage
Blocked sewer pipe
Sewer overloading
TOTAL
111
5
2
118
178
5
2*
185
# of incid. # of calls # of incid # of calls
flooding in buildings
flooding with
wastewater
55
0
5
60
7
0
1
8
110
0
6
116
8
0
1*
9
2
3
0
5
2
3
0
5
3
5
0
8
3
5
0
8
*Calls refer to rainfall on previous days; 1 call was misclassified: should have been ‘Illegal discharge’
151
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
Table 6.4 Results call data analysis Breda, 3 failure mechanisms for sewer flooding. Call
data for Breda cover a period of 5 years; in this period 251 independent rain incidents
occurred and 251 dry incidents following each rain incident.
Breda
Failure mechanisms
# of incid. # of calls
flooding-related conseq.
classes
# of incid. # of calls
flooding in buildings
# of
# of
incid. calls
flooding with
wastewater
Rain incidents
Gully pot blockage
137
978
40
66
5
5
Blocked sewer pipe
28
36
14
14
2
2
Sewer overloading
18
25
4
6
2
2
TOTAL
183
1039
58
86
9
9
Dry incidents
Gully pot blockage
108
265
22
22
6
7
1
1
Blocked sewer pipe
24
28
11
12 *
3
3 **
0
0
Sewer overloading
7
7 **
TOTAL
139
300
36
37
7
8
*some of the calls were misclassified; they refer to blocked house connections instead of blocked main sewers
**calls refer to rainfall on previous days or problems that occur during rainfall in general; for 1 call the
cause is not entirely clear
Comparison between cases
To allow for comparison between the two cases, the results in tables 6.3 and 6.4
were divided by the total sewer length and the total length of the measurement
period for each case. This results in incident frequencies per 100 km sewer length
per year for the 3 failure mechanisms. Figure 6.1 shows incident frequencies for
Haarlem and Breda per 100 km of sewer length and per year, for rain incidents.
The graph shows that incident frequencies of gully pot blockages are similar for
the cases of Haarlem and Breda: 4.2 and 3.9 per 100 km sewer length per year,
for all flood-related consequences. Gully pot blockages cause about 1 incident
of flooding in buildings per 100km per year for both cases. The frequency of
flooding with wastewater is low: below 0.2 per 100km per year for both cases,
for each of the flooding mechanisms.
152
Risk-based urban flood management: improving operational strategies
Incident frequency of sewer pipe blockages is approximately 8 times higher
for Breda compared to Haarlem, for all flood related consequences. The same
applies to dry incidents (results not shown here). A possible explanation is that
sewer cleaning frequency in Haarlem is twice as high as in Breda (see table
6.2). In addition, a recent evaluation report of urban drainage management in
Breda (Gemeente Breda, 2008) mentions that in 2004 and 2005 many sewers
were cleaned that hadn’t been cleaned for a long time. This was not reflected
in a reduction of the amount of ‘sewer blockage’ calls for 2006 and 2007, which
may indicate remaining backlog in maintenance work. Ages of sewer pipes
cannot account for the difference in blockage frequency; the distribution of
pipe lengths over pipe ages is similar for both cities.

 


























Figure
6.1 Comparison of the number of incidents per kilometre sewer length per year

for between Haarlem and Breda for 3 different selections of flood consequence classes,
for rain incidents
153
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
Incident frequency of sewer overloading is three times higher for Breda
compared to Haarlem. A possible explanation is that older parts of the system
in Breda were designed according to a lower design standard and that system
capacity was not adjusted at a later stage. Recent hydrodynamic calculations
for 4 subcatchments in Breda have indeed shown that system capacity in 3
of these areas does not comply with the design standard (Gemeente Breda,
2008). Other areas will be evaluated in the coming years. Also, the frequency
of occurrence of rainfall incidents in Breda could have been higher over the
study period compared to Haarlem. This could not be confirmed, since only
daily rainfall data were available for Haarlem and sewer overloading is mainly
influenced by peak intensities over short durations.
As mentioned earlier, detailed investigation of call texts for dry incidents shows
that many of these calls in fact refer to previous rain incidents or do not refer to
a particular event. This implies that most calls for dry incidents do not report
additional incidents, thus that probabilities calculated for rain incidents are
representative of total probabilities of flooding, as reported by citizens.
Probabilities of occurrence of incidents in various classes were quantified
following equation 3, as well as average consequences per incident in terms of the
number of reported locations per incident. These values were used to quantify
flood risk, according to equation 1. Table 6.5 gives the results of probabilities
and quantified risk for flooding-related consequences. The accumulated risk of
flooding incidents for 3 failure mechanisms is 0.19 locations/km sewer length/
year for Haarlem and 0.29 locations/km/year for Breda, for rain incidents
and for all flood-related consequences. The accumulated risk of flooding in
buildings is less than 10% of risk for all flood-related consequences. In both
cases, gully pot blockages contribute most to flood risk. These quantified risk
values can be used in decision making in order to decide whether flooding risks
should be reduced and what failure mechanism should be handled with priority
for risk reduction.
154
Risk-based urban flood management: improving operational strategies
Table 6.5 Summary of flooding risks for case studies of Haarlem and Breda, for rain
incidents: probabilities of flooding incidents and average risk per failure mechanism per
year. All values are calculated per year and per kilometre of sewer length.
Prob. of incid.
(km-1.year-1)
Haarlem
Gully pot blockage
Blocked sewer pipe
Sewer overloading
Total
Breda
Gully pot blockage
Blocked sewer pipe
Sewer overloading
Total
Flood risk
Prob. of incid.
(locations. km-1. (km-1.year-1)
yr-1)
Flood risk
(locations.km-1.
yr-1)
flooding-related consequences
Flooding in buildings
0.041
0.001
0.002
0.180
0.001
0.003
0.19
0.010
0.000
0.001
0.020
0.000
0.001
0.024
0.039
0.008
0.005
0.280
0.010
0.007
0.29
0.010
0.004
0.001
0.019
0.004
0.002
0.025
Evaluation of operational strategies
- Gully pot cleaning
The results show that handling of gully pot blockages should be a priority in
sewer management, since these are the main cause of flooding in general as
well as for flooding in buildings. At present, investments in preventive cleaning
constitute 15% of the total maintenance budget in both municipalities; 5% of
the total budget is spent on reactive handling upon gully pot calls. The results
in tables 6.3 and 6.4 show that reactive handling upon calls is not an efficient
strategy, because only 3% of all gully-pot-calls report serious consequences, i.e.
flooding in buildings or flooding with wastewater. Nevertheless it is current
practice in many municipalities to conduct investigation or direct cleaning
actions on-site upon every call. Much efficiency can be gained in handling of
gully pot blockages by reacting only to those calls that indeed have serious
consequences. This selection can be made at the call centre, by obtaining
additional information from callers, e.g. based on a number of standard
questions.
155
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
The blockage process of gully pots largely unknown so that occurrence of
blockages remains unpredictable, which complicates preventive handling.
Since most municipalities in the Netherlands apply similar regimes of gully
pot cleaning, no reference is available to compare the effect of higher or lower
preventive gully pot cleaning frequencies. The costs of planned gully pot
cleaning are low: about €3 to €6 per gully pot compared to €100 to €200
per reactive action. On the other hand, preventive cleaning involves all gully
pots, whereas reactive cleaning according to current strategies applies to less
than 1% of all gully pots yearly. Therefore, two options should be investigated
for their potential for cost reduction: experimenting with selective handling to
reduce reactive cleaning costs and optimizing preventive cleaning frequencies.
- Sewer pipe blockage
The difference in sewer blockage probability and associated risk of flooding
between Breda and Haarlem indicates that increasing preventive sewer
cleaning frequency can be an efficient strategy to reduce flooding induced by
sewer blockage. Preventive handling is a more desirable strategy than reactive
handling, since in the case of Breda half of the sewer blockages have serious
consequences, i.e. flooded buildings and wastewater flooding.
- Sewer overloading
The cities of Breda and Haarlem established standards for sewer flooding
induced by sewer overloading in their strategic plans: a maximum flooding
frequency of once per 2 years. In Breda a lower standard of once per year
applies to some areas. The standards do not specify to what geographical area
they apply: single location, street, sewer catchment of the entire city. The risk of
flooding caused by sewer overloading is about 1 location per year for Haarlem
and 5 locations per year for Breda. If the standard applies to the city as a whole
it is not satisfied; if it applies to a district or subcatchment it is easily satisfied.
The risk of flooding by sewer overloading is low compared to other failure
mechanisms; probability is low and few calls report serious consequences, i.e.
flooding inside buildings or flooding with wastewater. The costs of prevention
156
Risk-based urban flood management: improving operational strategies
can be high, if pipe dimensions have to be increased. In those cases, prevention
of blockages is a more efficient strategy to reduce flood risk. Prevention of
flooding by sewer overloading should only be considered in cases of serious
consequences or if prevention can be achieved by low-cost measures like
increasing the heights of doorsteps at building entrances.
Risk-based decision making for flood risk reduction
The urban drainage policy plan for the city of Breda states the following
maximum acceptable flooding frequencies for roads: once or twice per year for
residential areas, once per two years for commercial areas and the city centre
(Gemeente Breda, 2008). Flooding of buildings is not explicitly distinguished
from flooding of roads; protection levels of buildings therefore depend on the
relation of their building level to street level: building levels above street level
are likely to experience less flooding, those below street level more frequent
flooding than roads. This aspect is not addressed in the urban drainage policy
plan.
Table 6.6 summarises the results of call data analysis for the case of Breda, for
flooding of roads and of buildings separately. The contribution of the three
most important causes of flooding was also quantified. This table shows that
flooding frequencies exceed maximum values prescribed in the policy plan and
indicate a need for flood reduction.
Table 6.6 Outcome of call data analysis: flood risk in nr of calls/km sewer length/year,
city of Breda, period 2003-2007, total sewer system length 740km.
Flooded
Locations/km/yr
Total all causes
Sewer overloading
Sewer blockage
Gully blockage
Total
Roads
Buildings
0.3
0.003
0.003
0.2
0.206
0.03
0.002
0.004
0.02
0.026
157
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
Under the assumption that calls represent 2% to 30% of all real flood occurrences
(Wiechen et al., 2002; Devereux and Weisbrod, 2006), the uncertainty range
in real flood risk in terms of the number of calls per km sewer length per year
is summarised in table 6.7.
Table 6.7 Uncertainty range of quantified flood risk in nr of calls/km sewer length/
year, city of Breda, under the assumption that calls represent 2% to 30% of real flood
occurrences.
Flooded
Locations/km/yr
Roads
#calls
Sewer overloading
Sewer blockage
Gully blockage
Total
0.003
0.003
0.200
0.206
Min real
occurr
0.01
0.01
0.67
0.69
Max real
occur
0.15
0.15
10.00
10.30
Buildings
# calls
0.002
0.004
0.020
0.026
Min real
occurr
0.007
0.013
0.067
0.087
Max real
occurr
0.10
0.20
1.00
1.30
If, based on these results it is decided that flood risk should be reduced, various
actions can be taken to address these flooding causes. Table 6.8 summarises
actions that can be undertaken to reduce flood risk for three individual causes
of flooding: sewer overloading, sewer blockage and gully pot blockage. Due to
a lack of data on the effect of actions, especially of maintenance related actions,
the estimated effect of each action was based on expert judgment.
Table 6.8 Actions to reduce flood risk, for each of the three analysed flooding causes.
Costs were estimated based on investment and maintenance costs for 2 case studies;
effect was estimated based on expert judgment
Flooding cause Action to reduce
Estimated cost Estimated effect: flood
associated flood risk M€/km/year
risk reduction outcome
(locations/km/yr)
Sewer overloading Enlarge sewer pipe:
Reduction by 16.67% of
sewer overloading-related
events
Sewer blockage Increase cleaning
0.05
Reduction by 14% of sewer
frequency
blockage-related events
Gully blockage Increase cleaning
0.05
Reduction by 10% of gully
frequency
pot blockage-related events
* based on €1000/m sewer length replacement, 40 years amortization, interest rate 0.04
158
0.05*
Risk-based urban flood management: improving operational strategies
Sewer overloading is reduced by implementation of a structural measure:
enlargement of a sewer pipe. Blockages are handled by increasing maintenance
frequencies. Three measures of similar yearly investment cost are used for
comparison. The following assumptions were made with respect to the effects
of measures in relation to investment costs (table 6.9).
Table 6.9 Assumptions underlying estimates of the costs and effects of measures to
reduce flood risk
Flood reduction
Enlargement of
sewer pipe to reduce
flooding due to sewer
overloading
Cost assumptions
1 location at a time: 1000 m pipe
enlargement by replacement with
larger diameter;
Investment cost: €1,000,000 or
€50,000 per year;
Effect assumptions
Reduction of 1 flooded location
per year (where capacity is
enlarged) out of average 6
flooded locations per year:
reduction 1/6 or 16.67%.
Increase sewer
cleaning frequency
Yearly costs of sewer cleaning
are €180,000.
Increase cleaning costs with
€50,000/yr: cleaning frequency
increases by 28%.
Comparison of 2 cases with
different cleaning frequencies
shows that 2 times higher
cleaning frequency corresponds
with half the number of calls/year
(50% reduction). It is assumed
that 28% increase of frequency
results in 14% reduction in the
number of calls/year
Increase gully pot
cleaning frequency
Yearly costs of gully pot cleaning
are €150,000. Increase cleaning
costs with €50,000/yr: cleaning
frequency increases by 33%.
No data are available to estimate
the effect of increased gully
pot cleaning. The expected
bandwidth of reduction induced
by 33% frequency increase is
0-33%. It is assumed that 33%
increase in cleaning frequency
leads to 10% reduction in the
number of calls.
159
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
Table 6.10 Uncertainty range of quantified flood risk in nr of locations/km sewer
length/year, city of Breda, as a result of 3 different flood reduction measures, for road
flooding and for building flooding.
Locations/km/yr
Road flooding
Sewer overloading
Sewer blockage
Gully blockage
Total
Locations/km/yr
Building flooding
Sewer overloading
Sewer blockage
Gully blockage
Total
Enlarge sewer pipe
Min occur Max occurr
0.008
0.125
0.010
0.150
0.667
10.000
0.685
10.275
Enlarge sewer pipe
Min occur Max occurr
0.006
0.083
0.013
0.200
0.067
1.000
0.086
1.283
Increase sewer
Increase gully pot
cleaning frequency
cleaning frequency
Min occur Max occurr Min occur Max occurr
0.010
0.150
0.010
0.150
0.009
0.129
0.010
0.150
0.667
10.000
0.600
9.000
0.685
10.279
0.620
9.300
Increase sewer
Increase gully pot
cleaning frequency
cleaning frequency
Min occur Max occurr Min occur Max occurr
0.007
0.100
0.007
0.100
0.011
0.172
0.013
0.200
0.067
1.000
0.060
0.900
0.085
1.272
0.080
1.200
The relation between actions and reduction of call numbers is summarized in
table 6.10. Comparison of the results in table 6.10 with those in table 6.7 shows
that increasing gully pot cleaning frequency is most effective of the 3 strategies
to reduce flood risk. Sewer pipe enlargement and increasing sewer cleaning
frequency have only marginal effect on total flood risk. This follows from the
small number of calls, thus flooded locations, related to sewer overloading and
sewer blockage compared to gully pot blockage.
Table 6.11 summarises investment costs and minimum and maximum flood risk
estimates in terms of the number of flooded locations per year for the current
situation and after execution of each of the three flood reduction measures.
Figure 6.2 gives a graphical representation of the data in table 6.11. It shows
that for the same investment level, increasing gully pot maintenance is the most
effective measure to reduce flood risk. The effect of increased gully pot cleaning
frequency is about 10 times higher than that of enlarging sewer pipe capacity or
increasing sewer cleaning frequency. Uncertainty in flood risk results derived
from call data does not influence this conclusion. It only influences absolute
values of quantitative flood risk outcomes.
160
Risk-based urban flood management: improving operational strategies
Table 6.11 Summary of yearly investment costs and resulting flood risk in terms of
the number of flooded locations/km sewer length/year, for 3 flood reduction measures.
Uncertainty margins are based on the estimated representation of flood-related calls
compared the real number of flooded locations
Effect of investments;
Do
Enlarge
Increase
nr. of flooded locations/km/yr nothing sewer pipe sewer cleaning
frequency
Investment
Road flooding
Min (calls represent 30% of
real occurrences)
Max (calls represent 2%
of real occurrences)
Building flooding
Min (calls represent 30% of
real occurrences)
Max (calls represent 2%
of real occurrences)
Increase gully
pot cleaning
frequency
€0/yr
€50,000/yr €50,000/yr
€50,000/yr
0.687
0.685
0.685
0.620
10.300
10.275
10.279
9.300
0.087
0.086
0.085
0.080
1.300
1.283
1.272
1.200
㄀㈀⸀ ⸀ ㄀ 䴀愀砀㨀 㔀 砀 挀愀氀氀猀
䴀椀渀㨀 ㄀ ⼀㌀砀挀愀氀氀猀
䤀渀瘀攀猀 琀洀攀渀琀 䴀갠
⸀ 㠀
㠀⸀ ⸀ 㘀
㘀⸀ ⸀ 㐀
㐀⸀ 夀 攀愀 爀氀礀 椀渀瘀攀猀 琀洀攀渀琀 ⠀䴀 䔀 唀 刀 ⤀
一爀 漀昀 挀愀 氀氀猀 ⼀欀洀 猀 攀眀攀爀 氀攀渀最琀栀⼀礀爀
㄀ ⸀ ⸀ ㈀
㈀⸀ ⸀ ⸀ 䐀漀 渀漀琀栀椀渀最 䔀 渀氀愀爀最攀 䤀渀挀爀攀愀猀 攀 䤀渀挀爀攀愀猀 攀 䐀漀 渀漀琀栀椀渀最 䔀 渀氀愀爀最攀 䤀渀挀爀攀愀猀 攀 䤀渀挀爀攀愀猀 攀
猀 攀眀攀爀 瀀椀瀀攀 猀 攀眀攀爀
最甀氀氀礀 瀀漀琀
猀 攀眀攀爀 瀀椀瀀攀 猀 攀眀攀爀
最甀氀氀礀 瀀漀琀
挀氀攀愀渀椀渀最 挀氀攀愀渀椀渀最
挀氀攀愀渀椀渀最 挀氀攀愀渀椀渀最
昀爀攀焀甀攀渀挀礀 昀爀攀焀甀攀渀挀礀
昀爀攀焀甀攀渀挀礀 昀爀攀焀甀攀渀挀礀
刀 漀愀搀猀
䈀 甀椀氀搀椀渀最猀
Figure 6.2 Yearly investment costs and resulting flood risk in terms of the number of
flooded locations/km sewer length/year, for 3 flood reduction measures.
161
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
6.4.
Conclusions
Data from call centres at two municipalities reporting problems related to urban
drainage were analysed in order to quantify flooding frequencies and associated
flood risks for three main failure mechanisms. The results were used to evaluate
current operational strategies for prevention of flooding. The aim was to find
out whether current operational strategies based on practical experience are
efficient and if directions for improvement could be found. Quantified flood
risk for the 2 cases is 0.19 flooded locations per km sewer length per year and
0.29 locations per km per year. This is well above the standard defined as a
flooding frequency of once per year. The analysis pointed out that gully pot
blockages are the main cause of flooding. The efficiency of current gully pot
cleaning strategy can be increased by limiting reactive handling to those calls
that report serious consequences, which is a small portion of all calls. Also
optimisation of preventive cleaning frequencies can reduce costs. Preventive
cleaning of sewer pipes proves to be an efficient strategy to reduce flooding
due to sewer blockages as flood risk associated with sewer blockages is lower
in case of higher cleaning sewer frequencies. Sewer blockages often have
serious consequences, thus preventive handling is to be preferred to reactive
cleaning. According to the results of this analysis, reduction of flooding sewer
overloading is not of primary concern, because serious consequences for this
failure mechanism are rare compared to other failure mechanisms.
It was shown that based on call data analysis effective strategies flood risk
reduction can be identified. Currently, information about the effect of flood
reduction measures is lacking to adequately assess the effect of actions for
flood risk reduction. Based on the availability it could be shown that increasing
gully pot blockage is the most efficient action to reduce flood risk, given
data uncertainty. If differences between cause incidences are large, as in the
presented case study, call data are sufficient to decide how flood risk can be most
efficiently reduced. If differences are small, call data do not provide sufficient
accuracy to distinguish between causes. Additional data must be collected to
162
Risk-based urban flood management: improving operational strategies
assess flood risk more accurately and to estimate the effect of flood reduction
measures. The effect of structural measures can be estimated based on model
simulations if a reliable hydrodynamic model is available. Little information is
currently available to estimate the effect of different maintenance frequencies;
experiments with varying maintenance frequencies and methods should be
conducted to obtain insights into the effect varying maintenance strategies and
to support relating decisions in urban flood risk management.
163
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 6
6.5.
References
Bedford, T., Cooke, R. (2001). Probabilistic risk analysis: foundations and methods.
Cambridge University Press, Cambridge
Devereux, P.J., Weisbrod, B.A. (2006). Does “satisfaction” with local public services
affect complaints (voice) and geographic mobility (exit)? Public Finance Review, 34(2),
123-147.
Dirksen, J., Goldina, A., ten Veldhuis, J.A.E.Clemens, F.H.L.R. (2007). The role of
uncertainty in urban drainage decisions: uncertainty in inspection data and their
impact on rehabilitation decisions. In: Proc. of 2nd Leading Edge Conference on
Strategic Asset Management, Lisbon, Portugal
Gemeente Breda (2008). Verbreed GRP Breda 2009-2013 (in Dutch). Extended municipal
sewer management plan 2009-2013. Gemeente Breda, the Netherlands.
Gemeente Haarlem (2008). GRP Haarlem 2007-2011 (in Dutch). Municipal sewer
management plan 2007-2011. Gemeente Haarlem, the Netherlands.
RIONED Foundation (2004). Leidraad Riolering, Module C2100, 17-20 (in Dutch). ISBN
978-90-73645-68-4. Stichting RIONED, Ede, the Netherlands.
RIONED Foundation (2007). Leidraad Riolering, Module D1100, 43-70 (in Dutch). ISBN
978-90-73645-68-4. Stichting RIONED, Ede, the Netherlands.
Tuhovčák, L., Ručka, J., 2007. Hazard identification and risk analysis of water supply
systems. In: Proc. of 2nd Leading Edge Conference on Strategic Asset Management,
Lisbon, Portugal
Vesely, W., Dugan, J., Fragola, J., Minarick, J., Railsback, J. (2002). Fault Tree Handbook
with Aerospace Applications. Version 1.1. NASA Headquarters, Washington.
Wiechen van, C.M.A.G., Franssen E.A.M., de Jong, R.G., Lebret, E. (2002). Aircraft noise
exposure from Schiphol Airport: a relation with complainants. Noise and Health,
5(17), 23-34.
164
Risk-based urban flood management: improving operational strategies
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
165
Chapter 7
Discussion and recommendations
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
Historical series of data from municipal call centres show that urban flooding in
lowland areas occurs frequently, up to hundreds of times per year (ten Veldhuis
et al., 2009). Research in the field of urban flooding tends to concentrate on
flooding caused by rare, heavy rainfall events: hydrological studies are dedicated
to extreme rainfall characteristics and the effects of climate change (Ntegeka
and Willems, 2008) and modelling studies develop routines to simulate system
overloading by heavy rainfall and overland flow patterns (Maksimovic et
al., 2009; Djorjevic et al., 2005). Since heavy rainfall events typically occur
at low frequencies, of the order once per several years, they cannot account
for the high frequency of occurrence of urban flooding in lowland areas. The
question is what causes these high-frequency events, what consequences they
have and what the distribution of causes and consequences is over the series
of events? Risk analysis addresses causes and consequences of events and
associated probabilities of occurrence. Hence the general problem statement
for this thesis: what new insights can risk analysis based on historical series of
flood occurrence data provide with respect to characteristics of flood events in
lowland areas?
This chapter addresses the contributions of the research reported in this thesis
to respond to the four research questions that in the introduction chapter were
derived from the general problem statement. In this thesis data from municipal
call centres that describe observations by citizens of urban flood events are
used. The advantage of this type of data is that it is flexible to accommodate
descriptions of many kinds of flooding characteristics; the disadvantage is that
the information characteristics change with observation qualities of individual
citizens and their readiness and ability to provide details. The uncertainty
aspects of the use of call data in flood risk analysis and how data uncertainty
influences the validity of results and conclusions drawn is addressed in appendix
1. Recommendations for further study are given at the end of this chapter,
as well as directions for practical application of risk analysis in urban flood
management.
168
Discussion and recommendations
7.1.
Contribution to answer research questions
1. What causes contribute to urban pluvial flood risk and how can these
causes be quantified?
Fault tree analysis is applied to identify causes of urban flooding and to quantify
the contribution of different causes to overall flood probability. The results of
this study show that gully pot blockages are the main cause of flooding. They
contribute 71% to the overall probability of flooding in the investigated case
study. Other causes are, in decreasing order of contribution magnitude:
− blocked gully pot manifolds;
− areas not connected to urban drainage systems;
− high groundwater tables, drinking water pipe bursts;
− sewer overloading and blocked sewer pipes.
This result shows that asset failures are a more important cause of urban
flooding than overloading of urban drainage systems due to heavy rainfall. The
same applies to road flooding and flooding of buildings. In a study for the UK,
Arthur et al. (2009) obtained a similar result for the city of Edinburgh, showing
that more than 75% of sewer-related flood events were due to blockages, while
16% was due to hydraulic overloading. Renard and Volte (2009) conducted a
study of flood observation data for Grand-Lyon and found that 43% of flood
events was due to blockages of inflow devices like gully pots, 27% was related
to problems in sewer pipes and 19% concerned infiltration facilities, for the
period 1988-2005. Caradot et al. (submitted) state that, in a study for the city
of Mulhouse, 400 to 600 interventions were made yearly since 1993 to solve
flooding problems. Of these interventions, 37% of flooding problems was
due gully pot blockages, 27% was caused by improper behaviour of building
constructors and 27% was due to improper behaviour of citizens. A summary
of results from the case studies is given in table 7.1.
169
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
Table 7.1 Contributions of failure mechanisms in urban drainage systems to urban
flooding. Only percentage values for mechanisms that appear in all studies are shown.
Failure mechanism
Results case Results case study UK Results case studies
study the
(Arthur et al., 2009)
France (Renard and
Netherlands
Volte, 2009; Caradot et
al., submitted)
Haarlem
Edinburgh
Lyon
Mulhouse
Gully pot blockage
71%
54%, 37%
Blockages
75%
Hydraulic overloading 3%
16%
Problems in sewer pipes 1%
27%
These results point out the important role of asset failures as a cause of urban
flooding. This implies that risk analysis based on hydrodynamic modelling of
design storm and rainfall series provides an incomplete picture of urban flood
risk, since it only addresses flooding caused by overloading due to heavy rainfall.
Presently, many urban flooding studies ignore the effect of asset failures on
flooding. If asset failures are not taken into account in flood risk analysis, flood
frequencies and flood risk are likely to be underestimated. Additionally, flood
risk reduction measures that are chosen and designed based on these results
are likely to be ineffective, as a part of flooding problems remains unaddressed.
Flood risk analysis should include all potential causes of flooding to obtain
flood risk estimates representative of reality and to properly determine what
type of flood risk reduction measures are most effective.
2. What consequences of urban pluvial flooding should be taken into account
in a risk analysis and how can these be quantified?
Various kinds of urban flooding consequences are compared in this thesis,
including material damage and intangible consequences of flooding and
potential microbial infection due to exposure of citizens to contaminated flood
waters.
In chapter 4 of this thesis, tangible and intangible damages of urban pluvial
flooding are investigated for two lowland case studies. Tangible damages
include flooding of residential and commercial buildings, intangible damages
170
Discussion and recommendations
refer to traffic delay caused by road flooding and inconveniences to road users,
especially pedestrians and cyclists. It is shown that lowland areas are frequently
affected by flooding and that the frequency of occurrence of intangible damages
is higher than that of tangible damage.
A first attempt is made to translate both tangible and intangible damage into
common quantitative measures in order to be able to directly compare their
contributions to total flood damage. Two types of quantitative measures are
compared: monetary values and the number of people affected by flooding. The
results show that even though the frequency of occurrence of tangible damage
is lower, the monetary damage associated with tangible damage is much higher
than that for intangible damage. On the other hand, the number of people
affected by tangible damage is far smaller than the number of people affected
by intangible damage: over a period of 10 years, intangible damage affects up
to hundreds of times as many people as tangible damage.
The cumulative monetary damage to buildings from small flood events over a
period of 10 years, is estimated at about 10% of the damage to buildings during
the 1998 pluvial flood event in the Netherlands, with an estimated return period
of 125 years. This result illustrates that the cumulative damage of small flood
events over a period of 125 years is likely to be of the same order or magnitude
as the singular-event damage of a 125 year return period. This result shows that,
while flood risk analyses tend to focus on severe events and tangible damages
(e.g. Jonkman et al., 2003; Dutta et al., 2003; Apel et al., 2006; Thieken et al.,
2005), damage of small flood events should be taken into account to obtain a
complete and representative flood risk estimate for urban catchments.
The results of this thesis also show that large numbers of people are affected
yearly by intangible flood damage. Translation of this damage into costs
to citizens and society is not straightforward. In a study by Defra and the
Environment Agency in the UK (Defra, 2004) willingness-to-pay (WTP)
was used to quantify human-related intangible impacts of flooding. This
study involved 1510 face-to-face interviews and focused on willingness to
pay to prevent physical and psychological health effects of flooding of private
property. Similar studies could be conducted to develop methods for translation
171
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
of intangible flood damage into values that can be used in decision making and
policy related to urban flood risk.
Studies, e.g. in New Orleans, Dhaka and Jakarta (Sinigalliano et al., 2007;
Sirajul Islam et al., 2007 and Phanuwan et al.) demonstrated elevated
concentrations of microbial contaminants in flood waters and sludge after
severe flooding events. In chapter 3 of this thesis, a screening level microbial
risk analysis for urban pluvial flooding shows that flooding of combined sewer
systems produces health risks of the same order of magnitude as those associated
with swimming in recreational waters affected by combined sewer overflows.
This result indicates that small urban flooding events, with frequencies of
occurrence of up to several times per year, pose non-negligible health risks to
citizens, due to their high frequency of occurrence.
3. Can the results of quantitative urban pluvial flood risk analysis based
on historical data series from municipal call centres be used to support
decisions on how to effectively improve flood protection?
Call data provide details on causes of flooding events, consequences of flooding
and locations affected by flooding. 92% to 95% of the calls analysed in this thesis
contain information on flooding causes; 32% to 52% of the calls contain details
on flooding consequences; all calls include address details. This thesis shows
that call data analysis enable identification of flooding causes and quantification
of their contributions to flood risk. Call data also enable to distinguish between
contributions of flood causes to different types of consequences, such as
flooding of buildings, roads and tunnels. Thus, it supports selection of most
effective measures to improve flood protection. In chapter 5 it is shown that for
the cases analysed in this thesis, flood protection is most effectively improved
by prevention of gully pot blockages and sewer pipe blockages. Additionally,
it is shown that sewer pipe blockages can effectively be reduced by increasing
sewer cleaning frequency based on a comparison between cleaning frequencies
and flooding induced by sewer blockage for two cases. Available data do not
provide sufficient information to conclude whether increasing the frequency of
172
Discussion and recommendations
routine gully pot cleaning is effective to prevent gully pot blockages. Insight
obtained from call data analysis does suggest that reactive handling of gully
pot blockages can be made more efficient if actions are prioritised according
to the severity of observed consequences. The same holds for prioritisation of
investments for flood prevention: building flooding and tunnel flooding have
more disruptive consequence than flooding of residential streets and these
locations should therefore get priority in preventive action.
To summarise, call data analysis is useful to support decisions by setting
priorities based on observed consequences and predicting what type of flood
prevention strategy is likely to be most effective based on contributions of
flooding causes. The reliability of such decisions depends on the reliability of
call data. The influence of call data uncertainty on flood risk analysis outcomes
and related decisions is discussed in appendix 1.
4. Can risk-based standards for urban pluvial flooding provide a better basis
to evaluate urban drainage systems than current frequency-based standards
and guidelines?
Most current flooding standards and guidelines (e.g. CEN, 2008; RIONED,
2004) are expressed in terms of flooding frequency with no or limited reference
to flooding consequences. In addition, standards often do not make explicit
whether they are applicable at city level, or at the level of individual locations
or sewer subcatchments. For instance, if a few locations in a city suffer from
high flooding frequencies, this means that the system does not comply with
flooding standards at city level; yet all other individual locations inside the
city experience lower flooding frequencies and individually do comply with
standards.
In chapter 6 of this thesis, urban flood risk is quantified in terms of the number of
flooded locations per km sewer length per year. The number of flooded locations
is specified for various types of consequences: health-related consequences,
flooding of buildings and flooding of roads. In table 7.2 the results based on call
data analysis for 2 case studies are compared to flooding guidelines as defined
in policy plans for each case study. The urban drainage policy plans for Breda
173
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
and Haarlem (Gemeente Breda, 2008; Gemeente Haarlem, 2008) state that
those parts of the system that are covered by hydrodynamic models (circa 25%
for Breda and 75% for Haarlem) were evaluated based on design storms with
a return period of 2 years (RIONED, 2004; van Mameren and Clemens, 1997;
van Luijtelaar and Rebergen, 1997). Remaining areas are to be evaluated in
the future. Evaluation took place per subcatchment area, i.e. area connected to
a main pumping station; the size of subcatchment varies from about 100 ha to
1000 ha of semi- and impervious connected to urban drainage systems. When
model simulations indicated locations prone to flooding, these were studied in
further detail and solutions were designed and included in future investment
programs. Complaints were looked at to identify additional problem locations;
these were likewise studied in further detail.
The comparison shows that the investigated systems are far from complying
with flooding guidelines at city level: flooding frequencies vary from 9 to 33
flood events per year. Evaluation of frequency-based standards and guidelines
is often based on design storms, which implies that hydraulic overloading is
the sole failure mechanism taken into consideration. The results show that for
hydraulic overloading only, the investigated systems do not comply with the
standards at city level, unless flooding consequences are limited to building
flooding. If guidelines are evaluated per kilometer sewer length, guidelines are
easily complied with, both for street flooding and flooding of buildings, for all
failure mechanisms together.
174
Discussion and recommendations
Table 7.2 Policy guidelines for urban flooding the Netherlands and outcomes of
quantitative flood risk analysis based on historical flood incident data.
Policy guideline
Return period of design storm that urban
drainage system can cope with
Spatial scale undefined
Breda
Residential areas:
T=1 year
frequency: 1.0/year
Commercial areas:
T=2 years
frequency: 0.5/year
Haarlem
All areas:
T=2 years
frequency: 0.5/yr
Flooding frequency analysis
Based on historical flood incident data
City level (year-1)
Per km sewer length
(year-1)
Breda Haarlem
Breda Haarlem
All flooding consequences
33
25
0.045
0.055
Street flooding
31
22
0.042
0.047
Building flooding
16
9
0.021
0.020
Flooding due to hydraulic overloading only
Breda Haarlem
Breda Haarlem
All flooding consequences
3.8
1.0
0.005
0.002
Street flooding
2.5
0.6
0.003
0.001
Building flooding
0.8
0.5
0.001
0.001
These results point out a number of shortcomings of frequency-based standards
and evaluation for urban flooding. First, all potential causes of flooding,
including hydraulic overloading and asset failures should be taken into account
to obtain a realistic flood risk estimate. Second, flooding standards should
specify to what spatial scale they apply to ensure proper evaluation. If the
applicable spatial scale is not specified, the scale for application can be chosen
at will and outcomes of different evaluation can no longer be compared. Third,
standards should take flooding consequences into account, because damage to
society differs with various types of consequences. For instance, the results in
table 7.2 show that street flooding frequencies for roads are about two times
higher than flooding frequencies of buildings. This is a direct consequence of the
general construction level of buildings, which is about 15 cm above street level
in the Netherlands. If maximum flooding frequencies defined in the standards
are interpreted as road flooding frequencies, a safety margin for flooding of
175
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
buildings follows as an automatic result. Standards cannot allow for such
ambiguity of interpretation; their conditions of application should be explicitly
defined. It must be clear whether standards apply to all areas and occupational
functions alike, or whether specific functions need higher protection levels than
others to reflect differences in expected flood damage.
Unlike frequency-based standards, risk-based standards incorporate flooding
consequences and they take all known failure mechanisms into account instead
of focusing on one failure mechanism. This study showed how quantitative flood
risk values can be obtained from time-series of flood event data, in the form of
a number of flooded locations per year per km sewer length. This result was
further specified to flooded buildings per year per km and flooded roads per
year per km. The results were also used for quantification in terms of monetary
values and the number of people affected by flooding, per year, per km sewer
length, based on a number of assumptions with respect to the monetary damage
and number of people affected per flooded location.
Risk values, whether expressed in terms of flooded locations, number of people
affected or monetary damage per km sewer length per year can be used as a
starting point to develop risk-based standards. The setting of standards is in
essence a political decision that is informed by knowledge of current flood risk
and required investments to obtain flood risk levels in the future. Risk-based
standards in terms of monetary risk values have the advantage of providing
a direct investment cost versus damage costs comparison. The advantage of
number of flooded locations per km per year, specific for buildings, roads,
economical and societal functions is that flood risk can be directly derived from
flood occurrence data, without the need for translations based on uncertain
assumptions, as is the case for translation into monetary values.
Priority setting between different urban functions takes place either way: or
by differentiating protection levels between urban functions or by assuming
different monetary values associated with flooding of locations occupied by
different urban functions. The first makes priority setting an explicit part of the
political decision process, the latter makes it part of the risk assessment process.
176
Discussion and recommendations
7.2.
Generalisation of results from case studies
This research was based on an analysis of historical flood event data from 2 case
studies in the Netherlands. The case studies are representative of conditions in
densely populated, lowland areas in developed countries: small ground level
gradients, high groundwater tables, high building density, urban drainage
mainly provided by combined sewer systems. Studies based on case studies in
France and the UK (Caradot et al., submitted; Arthur et al., 2009) found that
urban flooding occurs at frequencies of hundreds of times per year in these case
studies as well and that flooding is mainly caused by asset failures. Since these
case studies are not situated in lowland areas, these conclusions are likely to be
true in general, in lowland and in hilly areas.
Another outcome for the Netherlands’ case studies was that the cumulative
risk of small pluvial flood events over a period 10 years is of the same order of
magnitude as the risk associated with a 100-years return period pluvial flood
event. Studies that quantify flood risks associated with small, high-frequency
events are rare and no references have been found that describe the cumulative
effect of these events. Frequencies of flooding are similar for the Dutch case
studies and those in France and the UK; the question is whether the amount
of damage associated with high-frequency events and with rare events is also
of the same order of magnitude. High-frequency events are characterised by
small flood depths; damage consists of intangible damage and small tangible
damage. The amount of damage mainly depends on population density, the
type of urban functions affected and lay-out of streets and buildings. Areas
with similar population density and urban lay-out are likely to experience
similar high-frequency flood damage. Areas with smaller spatial building
density and elevated building constructions are likely to experience less highfrequency damage. Damage associated with rare events depends on urban layout and ground level gradients: during these events, urban drainage systems
get overloaded and water flows mainly over the surface towards depressions.
If gradients are steep, flood depths in depressions rise rapidly and associated
damage to urban functions located in these depressions is likely to be high
177
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
compared to that in lowland areas. In that case, tangible damage for rare events
may exceed the cumulative damage due to high-frequency events.
7.3.
Recommendations for further research
The method applied in this thesis to quantify urban flood risk based on historical
flood event data is only a first step towards a unified approach for quantitative
urban flood risk assessment and risk-based evaluation of urban drainage
systems. Several knowledge gaps exist that impede a proper quantification of
urban pluvial risk and that need further attention.
1. Knowledge of flooding causes:
Flood risk analysis includes an analysis of all potential failure mechanisms leading
to urban flooding. In chapter 2 of this thesis it is shown that the contribution of
asset failures to flood risk is large compared to the contribution of overloading
due to heavy rainfall events. Blockage of inflow devices (especially gully pots)
is the most frequent cause of flooding, for flooding of buildings and of roads.
The discrepancy between the contribution of this cause of flooding and others
is large enough to draw this conclusion given data uncertainty.
Other causes of flooding, like heavy rainfall and pipe blockage, have lower
frequencies of occurrence that differ less from one another. More extensive data
collection and analysis is required to properly quantify contributions of these
causes to urban flood risk and to asses what causes lead to severest damage.
Insufficient data are available presently to assess how contributions of flooding
causes depend on system characteristics and maintenance activities. As more
data on flooding causes become available for various urban drainage systems,
it will be possible to analyse these relationships. Understanding of blockage
processes will enable prediction of blockage occurrence. This knowledge
supports development of efficient maintenance strategies, preventive handling
of assets and improved design of assets to reduce their failures sensitivity. The
importance of such knowledge is growing as the failure potential of assets is
178
Discussion and recommendations
expected to increase in the future due to ageing of urban drainage systems,
especially in western countries.
2. Knowledge of flooding consequences:
In this thesis an attempt is made to quantify consequences of urban flooding
and to translate tangible and intangible consequences into two kinds of common
measures: monetary values and numbers of affected people. Many assumptions
have to be made for such translation, due to a lack of information on relations
between various consequences and the chosen common measures. Assumptions
relate to the amount of damage to buildings and building contents as a result of
uncertain flood characteristics and uncertain property values; to the amount of
damage due to traffic delay and to the amount of intangible damage related to
inconveniences for road-users as roads and parking lots are flooded.
Uncertainty in direct damage to properties can be reduced by collecting data
on costs of flood events, e.g. from insurance reports, in combination with data
on flood characteristics. Indirect costs like traffic delay and inconvenience are
difficult to translate into monetary terms; such translations inevitably result
in uncertain outcomes, because traffic densities and velocities in urban areas
are difficult to predict (Liu et al., 2006) and the costs of traffic delay for urban
traffic are difficult to estimate (Bilbao-Ubillos, 2008). Information on stress
and inconvenience can be obtained via interviews with affected people or
through call centres by asking specific questions to callers. Willingness-to-pay
is a possible way to obtain monetary assessments of intangible damage due to
traffic delay and inconvenience; this method was applied in the UK to assess
intangible health effects of flooding (Defra, 2004).
The question is whether translation into common measures, monetary or other,
is desirable given the large variety of consequences and the uncertainties
involved in translation. Instead of translating consequences into common
measures, the numbers of events and affected locations per consequence
category can be directly used to quantify flood risk, as shown in this thesis.
Alternatively, a common measure can be applied per consequence category, for
179
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
instance: monetary values for damage to buildings and building contents, time
loss for flooding of roads, number of lives threatened for flooding of emergency
routes, number of people affected by inconvenience for flooding of sidewalks
and parking lots.
Further research is needed to find out whether translation of flood risk into
common measures is feasible and if so, what common measure should be
chosen and what uncertainty as a result of this translation can be accepted. If
uncertainties involved in translation in common measures are unacceptable,
common measures per consequence category are an alternative option. This
leaves the question of how to integrate or compare risks associated with
different consequence categories to decision makers.
3. Knowledge of efficiency of flood risk reduction measures:
Risk-based standards form a basis to assess the performance of urban drainage
systems and the need for flood risk reduction. The efficiency of alternative flood
reduction measures is to be assessed by comparing the costs of measures with
the benefits of reduced flood risk. Cost-benefit analysis is a possible method to
do this. Even though it offers the advantage of a direct comparison between
costs and benefits in monetary terms, it has several important drawbacks:
translation of benefits of flood risk reduction into monetary terms requires
many assumptions that are subject to uncertainty and the translation of all
costs and benefits as a result of the investment to monetary values for the year
the investment is to be made, introduces additional uncertainty. Finally, the
damage schematization made by this translation does not necessarily reflect
public perception of the potential loss.
Further research is needed to develop a method to quantify the benefits of
flood reduction measures that properly incorporates tangible and intangible
consequences, that accumulates benefits over the application time of reduction
measures and compares accumulated benefits with investment costs.
180
Discussion and recommendations
In addition, there is a lack of knowledge on the effect of flood reduction
measures on flood risk, thus a lack of knowledge to assess the benefits of flood
reduction investments. The effect of investments to increase system capacity
can be assessed using hydrodynamic models to a certain extent. There are some
questions to be answered as to how to properly use these models to quantify
flood risk:
− What model accuracy is required to be properly assess flood occurrence;
− What combination of underground sewer model and surface flow model
can provide sufficient accuracy to assess flood depths and flood extent;
− What accuracy is required to be able to quantify differences between flood
reduction measures;
− What rainfall series is to be used to assess future benefits of flood reduction
measures?
The effects of changes in maintenance strategies are largely unknown. Since
the development of blockages is difficult to predict, field experiments should be
conducted to determine the effect of variations in maintenance frequencies and
methods on the occurrence of flooding associated with blockage and to assess
the effect of combinations of preventive and reactive maintenance strategies on
flood risk.
4. Knowledge on acceptability of flood risk
Once methods become available to quantify flood risk, the outcomes can be used
to evaluate urban drainage system performance and to decide upon the need
for flood risk reduction. Contrary to frequency-based analysis, risk analysis
enables to evaluate the acceptability of flooding in view of the consequences
(Vrijling, 2001). Such evaluation is based on a comparison to some standard
or guideline that represents acceptable flood risk. An important question to be
addressed in the definition of risk-based standards is what level of flood risk is
acceptable in relation to the level of investment required for flood protection.
There are no absolute answers as to what flooding consequences are acceptable
and how much investment can be borne by society to prevent more severe
181
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
consequences. Answers to these questions are the outcome of societal preferences,
political and management discussions. Societal and economic developments
can give rise to changes in the desired protection level; for instance, higher
economic values at risk of flooding can lead to higher protection levels, lower
willingness to pay for flood protection due to poor economic conditions can
lead to lower protection levels.
The results of this study demonstrate that the cumulative direct, monetary
damage associated with small, frequent flooding events over the lifetime of the
investigated systems (50-100 years) is of the same order of magnitude is that
associated with rare, severe events. This implies that for the cases investigated
in this study, investments for urban flood protection provide an equal balance
between protection from small flooding events and severe events, for direct,
tangible damage. This balance is not the deliberate outcome of a chosen flood
protection strategy, as flood risk was never quantified at the time the strategy
was established. It is worthwhile to investigate whether the present situation is
considered acceptable by society: residents, property owners and politicians.
The numbers of calls received yearly at municipalities suggests this is not the
case.
Possibly, flood protection can be improved by shifting the balance towards one
side or another, without changing flood protection investments: for instance
by increasing protection from small events while decreasing protection from
severe events. This option is relevant in the light of climate change: if climate
change will give rise to a higher frequency of occurrence of severe events, flood
protection can be kept stable by increasing system capacity to bring back flood
risk associated with severe events to present levels or by increasing cleaning
frequencies, while maintaining current system capacity.
Shifting the balance means that citizens will be affected by flooding differently:
damage associated with small events comes frequently and is borne by individual
citizens and partly covered by insurance companies where it concerns damage to
building contents. Damage of severe events is rare and is borne by individuals,
sometimes covered by insurance and sometimes compensated by regional
182
Discussion and recommendations
or national government. Severe events tend to affect large areas at once and
are more likely to cause societal and economic disruption than small events.
Shifting the balance towards better protection from small events means people
and properties will be affected by flooding less frequently, thus will have to
recover from flood damage less frequently. Even if total damage over systems’
and people’s lifetimes remains unchanged, it may preferably to have to recover
rarely from severe damage than frequently from small damage. On the other
hand, the possibility of societal or economic disruption due to severe events
may be a reason to prefer flood risk reduction associated with these events.
Severe urban flood events are almost invariably caused by heavy rainfall
(not including river and sea flooding that are outside the scope of this thesis).
Blockages usually have a local effect; they are the main cause of small flood
events, yet are unlikely to cause severe flood events. This implies that a reduction
of flood risk associated with small events requires investments in intensified
cleaning of gullies and gully pots to prevent blockage, whereas risk reduction
associated with rare events requires investments to increase transport capacity
or to protect properties from flooding.
Further research is needed to find out what aspects of flood risk should be
taken into account to assess acceptability of flooding. Once the aspects that
influence flooding acceptability are known and can be assessed quantitatively,
risk-based standards can be developed. The risk level in the standard represents
the acceptability of flooding; the aspects that are to be taken into account to
assess acceptability are to be based on knowledge of flood risk characteristics;
the choice of the level of acceptable flood risk is the outcome of a political
decision process.
5. Knowledge to support risk-based decisions to manage urban flooding
As stated earlier, risk-based evaluation provides a more complete and realistic
picture of flooding problems than evaluation based only on frequencies. A riskbased approach offers additional advantages in decision support: If flood risk is
too high compared to the standard, flood risk can be reduced in two ways: by
183
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
reducing flood frequency and by reducing flood consequences. This approach
opens up additional options for system improvement: a decision maker can
decide to give priority to reduce flood frequencies of all main road tunnels or
to reduce flood consequences by increasing protection of flood-prone buildings
by raising pavement levels and door-sills. Given that intangible damage affects
many people, this can be a reason to prioritise investments to reduce flood risk
associated with this type of damage, even if associated monetary damage is
small.
If required investments for flood protection exceed the amount of damage that
can be prevented, a reactive approach to flood risk can be a viable alternative.
Especially if investments prevent flood damage that affects only a few people,
as is the case for flooding of buildings due to small flood events, the question is
whether investments are justifiable if the costs are to be borne by society as a
whole. This broadens the spectrum from flood risk prevention to raising flood
resilience, i.e. the ability of a physical and socio-economic system to recover
from flooding (de Bruijn, 2004). Possible reactive actions to promote resilience
include compensation of flood damage to individuals and offering the possibility
of insurance, which shifts the choice of reactive action on flood risk from water
authorities to individual owners of flood-prone property.
Risk-based urban flood management comprises finding a balance between
protection from frequent and rare flood events, between preventive and reactive
actions, between protection of different regions and occupational functions,
between tangible and intangible damage. Development of a unified approach
for quantitative urban flood risk assessment and risk-based evaluation of urban
drainage systems begins with the definition of a common method to assess and
evaluate flood risk. The following components of this method particularly need
to be clearly defined:
- how to assess flood frequencies and consequences: how to quantify
flood frequencies, including definition of individual flood events, what
consequence categories to include, how to quantify consequences;
184
Discussion and recommendations
- how to quantify flood risk: as a probability distribution of flood consequences
or as the expected value of flood consequences; in terms of damage per km
sewer length per year, per consequence category or alternative terms that
allow for a comparison between systems of different sizes and characteristics.
- how to evaluate flood risk: to compare expected value or maximum value or
90th or 95th percentile value of flood risk to the risk level defined in the flood
risk standard.
The risk level established in flood risk standards or guidelines is a political
decision, yet the way it is defined is preferably prescribed at a central, i.e.
national or supranational, level to enable comparison between systems in
different regions.
7.4 Recommendations for data collection for quantitative urban
flood risk analysis
The implementation of methods for quantitative urban flood risk analysis
can be started by setting up data collection and storage of urban flood event
characteristics. Figure 7.1 schematizes data about urban flooding, data
characteristics and relations. In this thesis, call data and rainfall data are used to
quantify urban flood risk. Rainfall data are used to define independent rainfall
events and to look for relation between rainfall volumes and numbers of calls.
Call data are used to identify causes and consequences of flood events and
their frequencies of occurrence. In figure 7.1, the option of including additional
measurements is indicated; data characteristics are given only for call data and
rainfall data.
185
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
Table Water level measurements
Urban flood measurement:
- flooding-related call
- water level sensor
measurement
Link: Time,
Location
Tables Other measurements
Link: Time
Table Call data
Table Call data:
- date, time
- location
- cause class(es)*
- consequence class(es)
- comment
Link: cause class index nr.
Table Cause classes:
- cause class description
- cause class number
Link: consequence class index nr.
Table Consequence classes:
- consequence class
description
- consequence class number
- flood depth
- flood extent
- flood duration
Table Rainfall data
Table Rainfall data:
- date, time
- rainfall volume
Link:
Date, time
Link: Date, time;
criterion separation
independent rain events
Table Rain events/dry
periods:
- start date, time
- end date, time
- code rainfall (1)/dry(0)
- event number

 Figure 7.1 Schematic presentation of data collection for quantitative urban flood risk
analysis
Rainfall data for flood risk analysis
In this thesis, daily rainfall measurements are used; as a result, relations between
rainfall intensities and numbers of calls cannot be investigated. Urban drainage
systems are especially sensitive to short-duration peak rainfall, therefore
rainfall data are preferably collected at short time intervals: 5 to 10 minutes. If
rainfall data are available, the effect of peak rainfall intensities on flood risk can
be examined. Additionally, the interval of rainfall data influences the definition
of independent rainfall events. In this thesis, a 24-hour dry period is used as a
criterion to separate independent events. The use of daily rainfall data leads
to rainfall events with long durations of up to 36 days, whereas in reality 24-
186
Discussion and recommendations
hour periods are likely to have occurred in between, yet not coinciding with
daily rainfall measurement period. If rainfall data are available at shorter
time intervals, the definition of rainfall events and dry periods can be made to
represent reality more accurately.
Call data for flood risk analysis
Many municipalities collect call data; 109 out of 190 municipalities that took
part in a questionnaire survey in the Netherlands (RIONED, 2007). Yet few
use it to analyse the condition of their infrastructure, e.g. in a risk analysis. The
reason is that call centres mainly aim at routing calls to a relevant department,
where it is to be handled efficiently. Call centres are not oriented towards
collecting information for an analysis of problem causes and consequences; thus
valuable information is wasted. This situation can be improved by structuring
the way call information is entered into a call database.
Current call databases usually have a time field, address field, main category
and subcategory fields and one or more open text fields for comments on flood
event characteristics. Main and subcategory fields are defined so as to facilitate
call handling.
The address and comments fields in call databases used in this thesis are open
text fields that contain a large diversity of texts. This complicates structured
storage and analysis of the data. For instance, street names are spelled
in different ways and comments vary from a few words to an almost literal
transcription of call conversation. To prevent the need for manual call-to-call
processing, text fields in call databases should be pre-structured as much as
possible. For instance, street names are to be picked from a pre-defined list to
prevent different spellings of the same street name. Similarly, other location
details such as “at the corner of”, “at the entrance of” should be pre-defined and
put in a different data field, separate from street name. Comment fields should
be used only for information that is not essential for further analysis or that is
too rare to be included in a predefined structure.
If call data are to be used for risk analysis, cause and consequence classes are to
be added to the current categories for call handling. The more detailed classes
187
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
are defined, the more information they can contain and the smaller the need for
additional open text comments.
The list of cause classes used in this thesis is based on fault tree analysis; all basic
events that appear in one or more call texts are included. The list of consequence
classes is composed pragmatically, based on consequence descriptions found
in the call texts from 2 databases. The lists of cause and consequence classes
preferably contain a high level of detail from the beginning in order to avoid
the need for adding new classes to the database at a large stage. Adding new
classes leads to inconsistency in the database and discontinuity in time series,
which can only be avoided by reclassifying all calls according to the new class
definition, a large time investment that is to be avoided. The link between calls
and classes can made via index numbers, or directly in the same database. The
advantage of using index numbers is that the database remains more concise
and that class descriptions can be adjusted without having to make changes in
the database.
Reliability and accuracy of call data
Even if cause and consequence classes are predefined, the reliability of call
classification at the call remains subject to uncertainty. Call centre employees
handle a wide variety of problems of which they have no specialised knowledge,
such as outfall of traffic lights, damage to street furniture and flooding.
Additionally, call centre teams tend to change rapidly and as a result the effect
of training employees on technical issues quickly erodes. As a result, causes and
consequences of flood events are sometimes identified erroneously. A possible
solution to this problem is to have technical personnel check every call on-site
and enter cause and consequence classes accordingly, as is done for the two
cases used in this thesis. In one of the cases, illustrations of urban drainage
problem causes and consequences were made available at the call centre to
support proper identification. Training and instruction, preferably in the field,
is helpful and should be repeated at regular intervals. Even though training
and on-site checking are time-consuming procedures, it is more efficient than
manual classification afterwards or losing the information altogether.
188
Discussion and recommendations
Additional flood event measurements
Call data cover only a part of all flood events; to obtain a more complete register
of flood events, additional data collection is needed.
Water level sensors can provide additional information on flooding: especially
locations where flooding occurs and flood depth. Advantages of water level
sensors are that they provide objective measurements and that they can provide
continuous measurements of flood characteristics throughout a flood event. A
disadvantage is that a sensor provides data of a single location: therefore sensors
are useful to measure flooding at locations that are known to be flood-prone.
The spatial density of a sensor network for blockage detection must be high
to be able to detect all flooding due to blockage. It is estimated in this thesis
that call data represent about 20% of true urban drainage problems, including
failures of gully pots, pipes, pumps, infiltration facilities etc (see appendix
1). To obtain near 100% coverage for asset failure detection, sensors should
be placed in every gully pot, i.e. about every 50 meters. Such dense sensor
networks are not a feasible option, economically nor practically. To obtain
higher sensor cost-efficiency, the number of sensors can be reduced to cover
only vulnerable locations, such as entrances to hospitals and emergency centres,
shopping streets, tunnels and residential areas at or below street level. Sensors
can provide little information about causes and consequences of flooding.
Configurations of two or more sensors can be used to identify the direction of
flooding, e.g. by placing sensors inside a sewer, gully pot connection and in a
gully pot, and to identify flooding consequences by placing sensors at various
locations in a street profile, including sidewalks and gardens.
189
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
Health-risks of sewer flooding
Data collection on a larger scale is required to quantify health risk associated
with combined sewer flooding with greater certainty. Collecting samples from
flooding incidents is complicated by their unpredictability. Registration of flood
incidents by responsible organisations will help to point out suitable locations
for sampling; call data can be especially helpful in this respect. Sampling teams
should be stand-by when weather forecast predicts heavy storms to reach
flooded locations at rapidly as possible for sampling. Local representatives can
be asked to call out as soon as they observe flooding or cameras or sensors can
be installed to observe flooding at locations known to be flood-prone. Samples
must be collected from a large geographical area or sample collection must be
extended over long periods of time to collect a sufficient amount of samples
to be able to draw reliable conclusions. Sampling from flooded locations is
complicated by possible variations in pathogen concentrations in time and
space due to ongoing rainfall and exchange between sludge and standing
water. ISO guideline 5667, parts 1, 10 and 12 provide guidance on the design
of sampling programmes and sampling of wastewaters and bottom sediments.
More research is needed to define sampling programmes, necessary sampling
frequencies and applicable techniques for health risk assessment of urban flood
waters.
7.5 Recommendations for analysis and handling of asset failures
Asset failures prove to be an important cause of flooding. Data collection of
asset failures and their consequences is essential in order to be able to assess
their effect on urban drainage system performance, to adequately handle asset
failures and to predict future failure likelihood. Data should be used in statistical
analyses to estimate probabilities of occurrence of asset failures and in risk
analyses to assess the effects of failures on flood risk. In addition, failure data
can be used as input in hydrodynamic model simulations to predict how failures
affect hydraulic processes, to identify locations that are vulnerable to flooding
190
Discussion and recommendations
as a result of asset failures and to evaluate the effect of improved handling of
asset failures. Data collection on asset failures in urban drainage systems in
relation to flooding problems should focus on gully pot blockages (including
blockage and breakage of gully pot connections), sewer pipe degradation
leading to root intrusion, sedimentation and partial or full pipe blockage and
on blockage of rainwater infiltration facilities. The role of pump failures was
investigated by Korving et al. (2006); pump failures lead to increased combined
sewer overflows, yet their influence on the occurrence of flooding is limited.
Currently, asset failures are handled by a combination of preventive, routine
cleaning activities and reactive handling of problems. The results of this study
suggest that improvements in the efficiency of asset management could be
made, by shifting the balance further towards preventive handling of failures.
Reactive handling of gully pot blockages and pump failures is expensive
compared to preventive handling, either because travel times between
individual cleaning actions are long compared to routine cleaning (gully pot
blockage) or because repair actions take more time than routine maintenance
(pump failures). Analysis results showed that higher sewer cleaning frequency
leads to fewer pipe blockages, which suggests that cleaning frequency can be
optimised further.
Since few data are available to assess the effectiveness of varying cleaning
frequencies for sewer pipes, pumps and gully pots, experiments with varying
cleaning frequencies should be conducted to test effectiveness with respect to
flood prevention. Besides that, changes in the layout or the design of inflow
devices could be investigated to prevent blockages. Gully pots often have a dual
function of run-off water conveyance and sand trap. Anti-odour screens added
in most type applied in the Netherlands. This combination of conveyance,
sand trap and screen leads to high susceptibility to blockage. Sand traps aim
to prevent blockage of pipes and damage to pumps; it is worth investigating
whether this function can be accommodated in a different way.
191
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Chapter 7
7.6.
References
Apel, H., Thieken, A.H., Merz, B., Blöschl, G. (2006) A probabilistic modelling system for
assessing flood risks. Natural Hazards, 38 (1-2), 79-100
Arthur S., Crow, H., Pedezert, L., Karikas, N. (2009). The holistic prioritization of proactive
sewer maintenance. Water Science and Technology, 59(7), 1385-1396.
Bilbao-Ubillos, J. (2008). The costs of urban congestion: Estimation of welfare losses
arising from congestion on cross-town link roads. Transportation Research Part A,
42 (2008), 1098–1108.
Caradot, H., Granger, D., Rostaing, C. Cherqui, F. Chocat, B. (submitted for NOVATECH
2010). Risk assessment of overflowing sewage systems: methodological contributions
of two case studies (Lyon and Mulhouse) (in French).
CEN, 2008. Drain and sewer systems outside buildings. EN 752:2008: E. European
Committee for Standardisation, Brussels, Belgium.
De Bruijn, K.M. (2004). Resilience and flood risk management. Water Policy, 6 (1), pp.
53-65
Defra (2004). The appraisal of human related intangible impacts of flooding. Defra, Flood
management division, R&D Technical Report FD2005/TR, London, UK.
Djordjević, S., Prodanović, D., Maksimović, C., Iveticć, M., Savić, D. (2005). SIPSON
simulation of interaction between pipe flow and surface overland flow in networks.
Water Sci. Technol. (52(5), 275-283.Dutta D, Herath S, Musiake K. (2003), A
mathematical model for flood loss estimation, Journal of Hydrology, 277(1–2), 24-49.
Gemeente Breda (2008). Verbreed GRP Breda 2009-2013 (in Dutch). Extended municipal
sewer management plan 2009-2013. Gemeente Breda, the Netherlands.
Gemeente Haarlem (2008). GRP Haarlem 2007-2011 (in Dutch). Municipal sewer
management plan 2007-2011. Gemeente Haarlem, the Netherlands.
ISO (2006). Water quality - Sampling - Part 1: Guidance on the design of sampling
programmes and sampling techniques. ISO 5667. International Organization for
Standardization, Geneva, Switzerland.
ISO (2006). Water quality - Sampling - Part 10: Guidance on sampling of wastewater. ISO
5667. International Organization for Standardization, Geneva, Switzerland.
ISO (2006). Water quality - Sampling - Part 12: Guidance on sampling of bottom sediments.
ISO 5667. International Organization for Standardization, Geneva, Switzerland.
Jonkman, S.N., van Gelder, P.H.A.J.M., Vrijling, J.K. (2003). An overview of quantitative
risk measures for loss of life and economic damage. Journal of hazardous materials,
A99, 1–30.
Korving, J., Clemens, F.H.L.R., van Noortwijk, J.M. (2006). Statistical modeling of the
serviceability of sewage pumps. Journal of Hydraulic Engineering, 132(10), 10761085.
Liu, H., van Duylen, H.J., van Lint, J.W.C., Salamons, M. (2006). Urban arterial travel
time prediction with state-space neural networks and kalman filters. Transportation
Research Record, 1968, 99-108.
Maksimović, C., Prodanović, D., Boonya-Aroonnet, S., Leitaõ, J.P., Djordjević, S., Allitt, R.
(2009). Overland flow and pathway analysis for modelling of urban pluvial flooding.
Journal of Hydraulic Research, 47(4), pp. 512-523.
Ntegeka, V., Willems, P. (2008). Trends and multidecadal oscillations in rainfall extremes,
based on a more than 100-year time series of 10 min rainfall intensities at Uccle,
Belgium. Water Resources Research, vol. 44, art no. w07402.
Renard F., Volte E., 2009, Study of pluvial flood events for the urban drainage system of
Grand-Lyon (in French), TSM juillet août 2009.
192
Discussion and recommendations
RIONED Foundation (2004). Leidraad Riolering, Module C2100, 17-20 (in Dutch). ISBN
978-90-73645-68-4. Stichting RIONED, Ede, the Netherlands.
RIONED Foundation (2007). Inquiry of flood problems in the built environment (in
Dutch). RIONED Foundation, Ede, the Netherlands.
Tapsell, S. Tunstall, S. (2003). An examination of the health effects of flooding in the United
Kingdom. Journal of Meteorology, 28(238), 341-349.
Veldhuis, JAE ten, Clemens, FHLR & Gelder, PHAJM van (2009). Fault tree analysis for
urban flooding. Water science and technology, 59(8), 1621-1629.
Thieken, A.H., Muller, M., Kreibich, H., Merz, B. (2005). Flood damage and influencing
factors: New insights from the August 2002 flood in Germany. Water Resources
Research, 41(12), 1-16.
Van Luijtelaar, H., Rebergen, E.W. (1997). Guidelines for hydrodynamic calculations on
urban drainage in the Netherlands: backgrounds and examples. Water Science and
Technology, 36(8-9), 253-258.
Van Mameren, H., Clemens, F.H.L.R. (1997). Guidelines for hydrodynamic calculations
on urban drainage in the Netherlands: overview and principles. Water Science and
Technology, 36(8-9), 247-252.
Vrijling, J.K. (2001). Probabilistic design of water defense systems in the Netherlands.
Reliability Engineering and System Safety, 74, 337-344.
193
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 1
Sensitivity analysis
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 1
This appendix specifically addresses uncertainty in the historical data series used
in this study and analysis results: how does data uncertainty influence the validity
of results and conclusions and can the conclusions be generalised to other lowland
areas?
196
Sensitivity analysis
Influence of uncertainty in call data on quantitative urban pluvial flood risk
results
Flood risk estimations are subject to large uncertainties, whether based on
a combination of physical models or on a data-driven approach. Physical
modelling approaches suffer from a lack of input and calibration data, model
structure and parameter uncertainties and inherent uncertainties in natural
phenomena like rainfall and run-off processes. Data-driven approaches suffer
from data uncertainty, uncertainty due to phenomena that are not represented
by available historical data and inherent uncertainties in natural processes.
A particular source of uncertainty for flood risk estimations based on call
centre data is that call data represent a sample of the total number of flood
occurrences, while the constitution of the sample cannot be controlled. It is
unknown whether the characteristics of reported incidents are representative of
the total collection of incidents nor how the number of report incidents relates
to the total number of incidents. Also, call information can be subjective and
comes from non-experts whose information can be incorrect. The latter source
of uncertainty is greatly reduced when calls are checked on-site by technical
experts as was the case of the data used in this study. The advantage of call
data is that they directly convey citizens’ experiences regarding adverse effects
of flooding. Hence, call data indicate the acceptability of flooding problems to
citizens.
To provide an estimate of the relation between the number of incidents
reported in call data and the total number of incidents, a full coverage of a
series of incidents would be needed for comparison to incidents reported in
calls. Since flooding incidents, especially those associated with blockages are
unpredictable, full registration of all flooding incidents would require a dense
observation network in time and space. To set up a dense sensor network of 1
or 2 sensors per km of sewer length or a continuous video or radar registration
just to validate data is not a feasible option.
197
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 1
References from customer research and complaint behaviour research provide
an estimate of the percentage of people that expresses dissatisfaction out of
the total number of people that is dissatisfied. Wiechen et al. (2002) compared
characteristics of complainants and non-complainants about aircraft noise in
the area around Schiphol Airport. They found that above a noise level of 55
dB, 2% to 7% of the total inhabitants in the noise exposed areas ever made a
complaint to the responsible agency. Out of the group of people who expressed
high annoyance by aircraft noise in a questionnaire, 19% had voiced their
complaints to the responsible agency.
Devereux and Weisbrod (2006) investigated satisfaction levels with public
services in Chicago, based on a telephone survey among 658 respondents. The
respondents were asked for their satisfaction levels about garbage collection,
street condition, police service and the quality of parks. Their results show that
3% to 9% of the respondents per category voiced a complaint. Of the group of
respondents who are very or somewhat dissatisfied about garbage, streets or
police, 23 to 26% voiced their most important complaint in this category. Of
the other respondents, who expressed some degree of satisfaction, 5% to 11%
voiced a complaint. Few people complained about parks: 3% of dissatisfied and
less than 1% of satisfied respondents.
Phau and Baird (2008) investigated complaint behaviour among Australian
consumers related to random service and purchase actions. They found that
50% of respondents will complain when they are dissatisfied with a product or
service. Kau and Loh surveyed complaint behaviour of mobile phone purchasers
in Singapore; 35% of respondents voiced a complaint to their mobile phone
provider.
198
Sensitivity analysis
The reasons for consumers to complain have found to be diverse: dissatisfaction
is an important though not always sufficient reason to complain. Other
influencing factors are the expected benefit of complaining and time and energy
spent in the complaint process.
Based on these aspects, an estimate is made of the percentage of citizens that
is expected to make a call to a municipal call centre out of the total number of
citizens who observe unsatisfactory urban drainage conditions. The examples of
complaints about aircraft noise and public services are closest to the situation of
complaints about unsatisfactory urban drainage conditions. Thus, the expected
percentage of citizens who make a call is between 2% and 30%.
Higher percentages were found for customer complaints after direct purchase
of goods or services; these situations are characterized by a higher direct
personal involvement or investment of customers and a higher interest in
obtaining a positive outcome. These percentages are therefore considered less
representative for complaints about urban drainage conditions.
Additionally, risk assessment for urban flooding is preferably based on the
real number of flood occurrences. This may include occurrences that were
not observed by citizens, e.g. during the night. This is similar to the lowest
percentage of 2% for noise complaints, where the 98% non-complainers
includes citizens who experience noise and decide not to complain and citizens
who are inside a noise range but do not experience noise.
Besides the size of the sample represented by citizens’ calls from the real
number of occurrences, the distribution of calls over cause and consequence
classes forms an additional source of uncertainty. The question is whether all
causes and consequences are equally represented in the sample. Representation
of calls in cause classes depends on how easily a cause can be recognized by
lay-people during the flood incident or by specialists who come to investigate
the call afterwards. An evaluation of original classes, assigned to calls at the call
centre upon reception of the call, compared to reassigned classes by a specialist
based on the call text and information added after on-site inspection shows that
199
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 1
gully pot blockage are easily recognized, whereas blocked sewers and gully
pot manifolds are difficult to recognize as the cause of flooding. This implies
that there may be hidden calls referring to these classes that were erroneously
labeled in other classes. The same goes for detailed consequence classes like
flooding at bus stops and flooding in front of shops. It is likely that many calls
in the “flooding of residential road” class could be assigned to more detailed
classes if more information were available. This problem was overcome by
focusing on aggregated classes: flooding in buildings, flooding on roads and
health-related flooding consequences. These classes are easily distinguishable
even for lay-people.
In this study risks were quantified by multiplying probability and consequences
of flooding events. The probability is quantified per class of causes or
consequences, based on the number of events in which a particular class is
mentioned; consequences are quantified based on the assumption that each call
represents flooding at one location. Analysis results showed that this is true for
95% of all reported incidents. This implies that missed calls have the following
effect on quantified risk:
− cause class entirely missed for an incident: cause class probability
underestimated
− consequence class entirely missed for an incident: consequence class
probability underestimated
− locations missed for an incident: magnitude of consequences (number of
locations) underestimated
Likelihood of missing calls depends on the abundance of calls per class,
visibility of specific cause and consequence classes and felt urgency of citizens
to respond. Gully pots en heavy rainfall are likely to be overrepresented in call
data compared to gully pot connection and sewer pipe blockages since these
causes are more easy to recognise. For the same reason, flooding of buildings
and roads is more likely to be reported than health-related consequences.
200
Sensitivity analysis
For quantitative flood risk estimates this implies that the probability of gully
pot blockages and heavy rainfall is likely to be correctly estimated, while
the probability of other causes of flooding is likely to be underestimated.
The probabilities of road flooding and building are also likely to be correctly
estimated. The magnitude of consequences likely to be underestimated; it is
more sensitive to uncertainty because every flood event that goes unreported
directly influences consequences magnitude. Contrarily, probabilities depend
on only one report per class per event; they are not influenced if incidents
within the same event are missed.
Decision problem: need for urban flood reduction for the case of Breda
The influence of data uncertainty on quantitative risk analysis results and the
consequences for decisions based on these results is investigated by analyzing
a typical decision problem for a case study of flood risk management. Acquired
insights are used to assess the impact of uncertainty in call data on flood risk
analysis and related decisions in general.
The urban drainage policy plan for the city of Breda states the following
maximum acceptable flooding frequencies for roads: once or twice per year for
residential areas, once per two years for commercial areas and the city centre
(#Breda, 2007). Flooding of buildings is not explicitly distinguished from
flooding of roads; protection levels of buildings therefore depend on the relation
of their building level to street level: building levels above street level are likely
to experience less flooding, those below street level more frequent flooding than
roads. This aspect is not addressed in the urban drainage policy plan.
Call data analysis for the city of Breda has shown that flooding frequencies
exceed these maximum prescribed values and indicate a need for flood
reduction. Table A.1 summarises the results of call data analysis, distinguishing
between flooding of roads and buildings. The contribution of the three most
important causes of flooding is also quantified.
201
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 1
Table A.1 Outcome of call data analysis: flood risk in nr of calls/km sewer length/year,
city of Breda, period 2003-2007
Flooded
Locations/km/yr
Total all causes
Sewer overloading
Sewer blockage
Gully blockage
Total
Roads Buildings
0.3
0.003
0.003
0.2
0.206
0.03
0.002
0.004
0.02
0.026
Under the assumption that calls represent 2% to 30% of all real flood
occurrences, the uncertainty range in real flood risk in terms of the number of
calls per km sewer length per year is summarized in table A.2.
Table A.2 Uncertainty range of quantified flood risk in nr of calls/km sewer length/
year, city of Breda, under the assumption that calls represent 2% to 30% of real flood
occurrences.
Flooded
Locations/km/yr
Roads
#calls
Sewer overloading
Sewer blockage
Gully blockage
Total
0.003
0.003
0.200
0.206
Min real Max real
occurr
occur
0.01
0.15
0.01
0.15
0.67
10.00
0.69
10.30
Buildings
# calls
0.002
0.004
0.020
0.026
Min real Max real
occurr
occurr
0.007
0.10
0.013
0.20
0.067
1.00
0.087
1.30
If, based on these results it is decided that flood risk should be reduced, various
actions can be taken to address these flooding causes. Table A.3 summarises
actions that can be undertaken to reduce flood risk for three individual causes
of flooding: sewer overloading, sewer blockage and gully pot blockage.
202
Sensitivity analysis
Table A.3 Actions to reduce flood risk, for each of the three analysed flooding causes.
Costs are estimated based on investment and maintenance costs for 2 case studies; effect
is estimated based on expert judgment
Flooding cause Action to reduce
Estimated
Estimated effect: flood risk
associated flood risk cost
reduction outcome
M€/km/year (locations/km/yr)
Sewer overloading Enlarge sewer pipe:
0.05*
Sewer blockage Increase cleaning
frequency
Gully blockage Increase cleaning
frequency
0.05
0.05
Reduction by 16.67% of sewer
overloading-related events
Reduction by 14% of sewer
blockage-related events
Reduction by 10% of gully pot
blockage-related events
* based on €1000/m sewer length replacement, 40 years amortization, interest rate 0.04
Sewer overloading is reduced by implementation of a structural measure,
enlargement of a sewer pipe. Blockages are reduced by increasing maintenance
frequencies. Three measures of similar yearly investment cost are used for
comparison. The effect of each of the measures is estimated based on expert
judgment. The following assumptions are made with respect to the effects of
measures in relation to investment costs:
− Enlargement of sewer pipe to reduce flooding due to sewer overloading:
1 location at a time: 1000 m pipe enlargement by replacement with larger
diameter; Investment cost: €1,000,000 or €50,000 per year; Effect:
reduction of 1 flooded location per year (where capacity is enlarged) out of
average 6 flooded locations per year: reduction 1/6 or 16.67%.
− Increase sewer cleaning frequency: yearly costs of sewer cleaning are
€180,000. Increase cleaning costs with €50,000/yr: cleaning frequency
increases by 28%. Effect: comparison of 2 cases with different cleaning
frequencies shows that 2 times higher cleaning frequency corresponds
with half the number of calls/year (50% reduction). It is assumed that 28%
increase of frequency results in 14% reduction in the number of calls/year
− Increase gully pot cleaning frequency: yearly costs of gully pot cleaning
are €150,000. Increase cleaning costs with €50,000/yr: cleaning frequency
increases by 33%. Effect: no data are available to estimate the effect of
203
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 1
increased gully pot cleaning. The expected bandwidth of reduction induced
by 33% frequency increase is 0-33%. It is assumed that 33% increase in
cleaning frequency leads to 10% reduction in the number of calls.
Table A.4 Uncertainty range of quantified flood risk in nr of locations/km sewer length/
year, city of Breda, as a result of 3 different flood reduction measures, for road flooding
and for building flooding.
Locations/km/yr
Road flooding
Sewer overloading
Sewer blockage
Gully blockage
Total
Locations/km/yr
Building flooding
Sewer overloading
Sewer blockage
Gully blockage
Total
Enlarge sewer pipe
Min occur Max occurr
0.008
0.125
0.010
0.150
0.667
10.000
0.685
10.275
Enlarge sewer pipe
Min occur Max occurr
0.006
0.083
0.013
0.200
0.067
1.000
0.086
1.283
Increase sewer
Increase gully pot
cleaning frequency
cleaning frequency
Min occur Max occurr Min occur Max occurr
0.010
0.150
0.010
0.150
0.009
0.129
0.010
0.150
0.667
10.000
0.600
9.000
0.685
10.279
0.620
9.300
Increase sewer
Increase gully pot
cleaning frequency
cleaning frequency
Min occur Max occurr Min occur Max occurr
0.007
0.100
0.007
0.100
0.011
0.172
0.013
0.200
0.067
1.000
0.060
0.900
0.085
1.272
0.080
1.200
The relation between actions and reduction of call numbers is summarized in
table A.4. Comparison of the results in table A.4 with those in table A.2 shows
that increasing gully pot cleaning frequency is most effective of the 3 strategies
to reduce flood risk. Sewer pipe enlargement and increasing sewer cleaning
frequency have only marginal effect on total flood risk. This follows from the
small number of calls, thus flooded locations, related to sewer overloading and
sewer blockage compared to gully pot blockage.
Table A.5 summarises investment costs and minimum and maximum flood risk
estimates in terms of the number of flooded locations per year for the current
situation and after execution of each of the three flood reduction measures.
Figure A.1 gives a graphical representation of the data in table A.5. It shows
that for the same investment level, increasing gully pot maintenance is the most
204
Sensitivity analysis
effective measure to reduce flood risk. The effect of increased gully pot cleaning
frequency is about 10 times higher than that of enlarging sewer pipe capacity or
increasing sewer cleaning frequency. Uncertainty in flood risk results derived
from call data does not influence this conclusion. It only influences absolute
values of quantitative flood risk outcomes.
Table A.5 Summary of yearly investment costs and resulting flood risk in terms of
the number of flooded locations/km sewer length/year, for 3 flood reduction measures.
Uncertainty margins are based on the estimated representation of flood-related calls
compared the real number of flooded locations
Effect of investments;
Do nothing Enlarge
Increase
Increase gully
nr. of flooded locations/km/yr
sewer pipe sewer cleaning pot cleaning
frequency
frequency
Investment
Road flooding
Min (calls represent 30% of
real occurrences)
Max (calls represent 2%
of real occurrences)
Building flooding
Min (calls represent 30% of
real occurrences)
Max (calls represent 2%
of real occurrences)
€0/yr
€50,000/yr €50,000/yr
€50,000/yr
0.687
0.685
0.685
0.620
10.300
10.275
10.279
9.300
0.087
0.086
0.085
0.080
1.300
1.283
1.272
1.200
Sensitivity of decisions to data uncertainty and data need for risk-based
decisions on urban flooding
1. Identify most vulnerable components in sewer system with respect to
causing flooding. Vulnerable components with respect to flooding are
those components that are most likely to fail and contribute most to flood
risk. Call data have shown to provide sufficient accuracy to identify gully
pots as the most vulnerable component in sewer systems, with respect to
flooding. Gully pot blockages stand out against other causes to such an
extent that uncertainties in call data do not influence this conclusion. In
order to distinguish between component vulnerabilities that differ less
205
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 1
conspicuously from others, more accurate and more complete data sets
are needed. In particular the relative contributions of sewer blockages and
heavy rainfall should be supported by additional data, since this distinction
cannot be made by call data based on above-ground observations and ex
post analysis by experts.
2. Identify most vulnerable locations to flooding in catchment.
The vulnerability of locations to flooding can be interpreted in various
ways: locations that suffer flooding most frequently, those that suffer most
severe consequences or those that raise most protest from citizens. The first
two aspects are summarised in quantitative flood risk assessment, the latter
is revealed in call texts and letters and petitions to local authorities. Flood
risk assessments are typically aiming to be objective; citizens’ protests are
subjective. The use of call data for flood risk analysis implies introduction
of a degree of subjectivity into quantitative risk assessment outcomes. This
effect is diminished by the large number of call data: the call database
shows that the maximum number of calls per street represents 1% of the
total number of calls. This indicates that the data are not susceptible to bias
introduced by excessive calling of one or a few individuals and that call
data are sufficiently representative to identify most vulnerable locations in
a catchment.
3. Evaluate urban drainage systems respect to urban flooding standards.
Urban flooding standards mostly define a maximum flooding frequency
or a maximum surcharge frequency; some distinguish between different
occupational land uses. Call data analysis results in an estimate of flooding
frequencies and of flood risk; they provide a sufficient level of detail to
distinguish between occupational land uses and even between road types
and buildings uses. Call data provide a better basis to check compliance
with standards than hydrodynamic model simulations or singular-eventbased evaluation, because they include a wider range of flooding causes and
consequences. The main drawback of call data for risk quantification is that
they represent a sample of unknown size of real flooding incidents. This
means that quantitative flood risk based on call data always underestimate
206
Sensitivity analysis
the true flood risk, while the degree of underestimation is unknown. Still, it
provides a risk estimate that is closer to reality than model simulations that
focus on heavy rainfall events and do not include asset failures as a cause
of flooding.
4. Decision to prioritise locations for investments to reduce flood risk.
This decision problem is similar to decision 2, if prioritisation takes
place according to flood risk. If other aspects are taken into account,
like possibilities to combine investments with other maintenance or
construction activities in order to gain efficiency, additional data regarding
these respects is needed.
5. Decision in what flood reduction measure to invest, for a certain location
or area, in order to most efficiently reduce flood risk. Call data analysis can
identify the main causes of flooding for a particular location. Besides this,
information about the effect of flood reduction measures is needed. The
effect of structural measures can be estimated based on model simulations;
little information is available to estimate the effect of different maintenance
frequencies. If differences between cause incidences are large, call data
are sufficient to decide how flood risk can be most efficiently reduced.
If differences are small, call data do not provide sufficient accuracy to
distinguish between causes. Additional data must be collected to assess
flood risk more accurately and to estimate the effect of flood reduction
measures, especially varying maintenance frequencies.
Most decision problems require data collection in addition to call data to provide
more accurate risk assessments and to allow distinctions between options that
differ little. Ideally, data would provide a full sample of flood occurrences,
including cause and consequence details. This would require a high temporal
and spatial resolution of data collection. A dense sensor network, e.g. one that
is constituted of sensors in gully pots and house connections could provide
such information. The installation and operational costs of such of network
are high and the reliability depends on the quality of the sensors, data transfer,
storage and analysis. Alternatively, satellite images can provide high-resolution
207
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 1
spatial data, yet the temporal resolution is low, typically weekly or monthly
data collection. Another drawback is that satellite images are disturbed by
clouds, while satellite radar images are not well fit for interpretation of flooded
surfaces, especially at the level of detail require for the urban scale.
Since the aim of urban flood protection is to protect citizens and their
possessions from the harmful effects of flooding, citizens’ observations are a
valuable source of information to be used in flood risk analysis. The use of
call centres to register citizens’ observations and complaints is widely spread
among authorities; public, e.g. municipalities, as well as private, e.g. water
companies. The quality of call data can be enhanced in several ways to improve
the reliability of flood risk analyses. Additionally call data can be complemented
with data from other sources.
Call data have several advantages over other types of flood data, like data
from water level sensors and ex post interviews with people affected by floods.
Sensors have the advantage of providing more objective measurements; yet to
collect details on flooding causes and consequences, a combination of sensors
would be needed which results in an expensive monitoring set-up. Ex-post
interview have the drawback of collecting information with a certain delay,
which inevitably result in information loss, since interviewed people may have
forgotten details of not have paid attention to certain information details.
208
Sensitivity analysis
References
Wiechen van, C.M.A.G., Franssen E.A.M., de Jong, R.G., Lebret, E. (2002). Aircraft noise
exposure from Schiphol Airport: a relation with complainants. Noise and Health,
5(17), 23-34.
Devereux, P.J., Weisbrod, B.A. (2006). Does “satisfaction” with local public services
affect complaints (voice) and geographic mobility (exit)? Public Finance Review,
34(2), 123-147.
Phau, I., Baird, M. (2008). Complainers versus non-complainers retaliatory responses
towards service dissatisfactions. Marketing Intelligence and Planning, 26 (6), 587604.
Forbes, S.J. (2008). The effect of service quality and expectations on customer complaints.
Journal of industrial economics, LVI(1), 190-213.
209
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2
Automatic classification of call data for
quantitative urban flood risk analysis
Appendix 2
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
212
Automatic classification of call data for quantitative urban flood risk analysis
Introduction
In the past, comprehensive flood risk analyses have been limited by a lack
of data or a lack of knowledge of the complex interactions between rainfall
conditions and flooding consequences (Apel et al., 2006). Although it is widely
recognised that hydrodynamic models are indispensable tools for successful
flood management, the development of such models is to date limited by a
lack of spatially distributed evaluation data (Werner et al, 2005; Schumann
et al., 2008). Monitoring networks in urban drainage systems can provide
the required information, if they have sufficient spatial density to detect all
flood events throughout urban areas. In practice, monitoring locations are
limited to pumping stations, overflow weirs and some additional points e.g. at
special constructions. This density is largely insufficient to register in detail all
flood incidents in an urban area. Additionally, monitoring networks in urban
drainage suffer from data loss due to sensor failure, communication failures etc.
(Dirksen et al., 2009).
Municipal call centres register information on urban drainage problems observed
by citizens. The network of callers is potentially dense since every citizen can
be assumed to have access to a telephone. Though calls do not necessarily give
complete coverage of flood incidents, because there is no guarantee that a call is
made for every event, it is one of the best sources to provide indication of events
unacceptable to citizens: citizens make calls to point out abnormal situations
that they citizens want to see solved.
Calls related to urban drainage cover a variety of details on problem causes
and consequences that traditional monitoring and modelling find difficult to
address, such as details on in-house flooding and maintenance-related problems
like pipe blockages. The drawback of this type of data is their unstructured
nature: call texts vary in the level of detail and type of information provided,
depending on what is provided by the caller and how much of that is reported
at the call centre. To be able to use call information in flood risk analysis, calls
must be screened and classified to obtain consistent output that can be used in
quantitative analysis.
213
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2
Even though many municipalities have a call register, 109 out of 190
municipalities that took part in a recent inquiry in the Netherlands (RIONED,
2007), few use it to analyse the occurrence of problems in their infrastructure.
One reason is that manual classification of calls is time-consuming due to the
large numbers of calls: hundreds or thousands per year per municipality. Yet
call data have proven to provide valuable information to detect causes and
consequences of urban flooding that cannot be provided by other types of
monitoring data (Arthur et al., 2009, ten Veldhuis et al., 2009).
This chapter examines the possibility of automatic call classification based on
call texts for the purpose of urban drainage system analysis and quantitative
risk assessment. To this end, some well-known classification routines are tested
by application to two call databases containing about 6300 calls each.
Automatic classification of municipal calls may be compared call routing
where a call is routed to a destination based on words or grammar fragments
in call texts (Garfield and Wermter, 2006; Gorin et al., 1997). The task of call
classification differs from call routing for helpdesk applications where routing
is preferably based on a minimum of information, e.g. only the first caller’s
utterance. Call classification for application in risk assessment tries to retrieve
as much information as possible from a call. Municipal calls typically contain
natural spoken language (Gorin et al., 1997) that comes from one or two
sources: call centre employees write down in telegram style what callers have
actually said and in part of the databases technical employees enter text on how
they handled calls. The information content of both texts differs and the second
text may even contradict what was stated in the first, because a problem was
found to be different from the one described upon on-site investigation.
This article is structured as follows: first, principles of classification pattern
recognition are discussed in brief, followed by a description of the datasets that
are used for automatic classification experiments. After that, the set-up is given
of some initial automatic classification experiments that have been conducted.
214
Automatic classification of call data for quantitative urban flood risk analysis
The results are outlined, followed by a discussion and some additional notes and
observations. Finally, conclusions derived from the outcome of our experiments
are presented.
Materials and methods
Pattern recognition
Call classification is a special case of pattern recognition, a research field that
aims to assign observations to classes based on observations’ characteristics,
expressed as a number of features.
There are numerous books and other texts that provide a good introduction
to the field of pattern recognition (e.g. Duda et al. (2001), Jain et al. (2000)
and Bishop (2006)), while various more dedicated texts concerned with text
categorization are also available (e.g. Sebastiani, 2002). Here we only provide a
brief sketch of some of the essentials that should enable the reader to understand
the illustrations given and the basic experiments carried out in this work.
One of the main questions in pattern recognition studies is how, and to what
extent, one can decide on the class label of a new object, based on a comparison
of object characteristics with those of objects with known labels? The basic
idea is that particular, typically statistical patterns in the characteristics of an
object provide weak or strong clues about the true class label of this object. E.g.
a relatively high number of occurrences of the words “yellow” and “submarine”
lowers the probability that a text is about Elvis. The initial collection of
observations for labelled objects is used to try to find general patterns and
relations that can subsequently be used to predict the label of new objects.
In pattern recognition language a label predictor is referred to as a classifier,
the act as classification, the overall error made in this prediction is called the
classification or generalization error (in a way this is the probability that a
new object will be labelled with the wrong label), while finding the patterns
215
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2
and relations in the training data is called learning or training. In addition, the
different labels are called classes and the characteristics chosen to describe the
objects are features. One important step in pattern recognition can accordingly
be stated as devising features based on which successful classification can be
performed. Another is the choice of the actual classifier that is to be trained
using these features and the associated class labels. There is an immense amount
of literature on various types of classification approaches and procedures. We
discuss two simple schemes, first nearest neighbour (1NN) classification and
nearest mean classification (NMC), that should give a good initial impression
of how such classification could be performed.
Having measured N features for every object -- this could for example be word
counts of “submarine”, “yellow”, “haze”, “purple” or any other word that might
help us to distinguish different classes from each other -- we can represent
every object as a vector in an N-dimensional space (the section on word counts,
which can be found below, details our particular choice of features). Now,
the 1NN classifier operates in this vector space and labels new objects with
the same label as the object that is nearest to the new one and for which one
knows the label. Nearest is in terms of the distance between the feature vectors
in the vector space. The idea behind 1NN is simple and intuitive: the nearer
features are to each other, the more similar the original objects probably are,
and therefore chances are high that their labels are also the same. NMC, on
the other hand, relies on more global statistics, but is no more complicated
than 1NN. In the classifier training phase, one determines the mean of every
class, i.e., the average feature vector for every category is computed, which
again is an N-dimensional vector. In the classification phase, every new object
is assigned to the class mean that is closest, i.e., it gets the label belonging to
that class.
216
Automatic classification of call data for quantitative urban flood risk analysis
A final concept that needs to be introduced is the learning curve. Learning
curves plot the generalization error with varying training set size or feature set
size. The first type of curves investigates how much there is to gain from adding
more and more data to the training set and can be used to decide whether it is
worth the effort to collect more labeled data. The latter type of curves provides
insight into how a classifier behaves under a varying number of features for a
particular fixed number of training objects. As it turns out, adding more and
more features, and hence more and more information about the individual
objects, does not necessarily mean that classification performance will improve.
This maybe counterintuitive behavior of classification schemes is often coined
the curse of dimensionality (Duda et al., 2001; Jain et al., 2000; Bishop, 2006).
Available datasets for classification
Two call databases were available for this study, including all calls related to
urban drainage for 2 municipalities in the Netherlands: Haarlem and Breda.
The datasets consist of 6991 and 6361 calls respectively over a period of 5 and
10 years (Table 1).
Table A2.1 Summary of data for two cities with available datasets: sewer system
characteristics, call data in municipal call register
Data case study
Number of inhabitants
Length of sewer system (% combined)
Total surface connected to sewer system
Total number of gully pots
Period of call data
Total number of calls on urban drainage
Length of data series
Haarlem
147000
460 km (98%)
1110 ha
42500
12-06-1997 to
02-11-2007
6359
3788 days
Breda
170000
740 km (65%)
1800 ha
80000
31-01-2003 to
23-10-2007
6980
1726 days
217
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2
Table 2 gives some examples of call texts from the call datasets. The examples
illustrate how the type of information and details vary between call texts.
Features are selected from these call texts to be used for automatic classification.
Table A2.2 Example of call texts
Date
2-5-2002
Call text
On the Karel Doormanlaan near the apartment complex Spaarnhoven,
much water remains on the street after a storm. Elderly people have trouble
entering the building. Can this be solved? Action: 10/05, Gully pot cleaned.
25-10-2005 At the busstop on the Zuiderzeelaan and the busstop to the west 2 or 3 gully
pots are blocked. The busstop is flooded. Action: 2 gully pots cleaned and
flushed
15-5-2007 This caller on the Veenbergstraat nr 20 has problems with moisture under his
residence. There are also rats in the residence. She thinks it has to do with
the bad condition of the sewer in the street; the street is full of pits and holes.
Please contact caller and take a look in this street. Action: Solved by owner.
22-5-2007 Flooding of bicycle tunnel. 14-06-07 situation ok, problem solved
Definition of classes
We used sets of manually classified data from a quantitative flood risk analysis
study (ten Veldhuis et al., 2009). Class definitions were defined based on a fault
tree analysis; this resulted in six classes that correspond with potential causes
of urban flooding (table 3). The calls were manually classified by technical
specialists based on the information in the call texts. The manually classified
datasets provide the training and test sets for the development of an automatic
classification procedure.
218
Automatic classification of call data for quantitative urban flood risk analysis
Table A2.3 Class definitions used in manual classification and manual classification
results
Class definition
1
2
3
4
5
6
# entries/class
Breda
Blocked inflow process (gutters, gully pots, manifolds) 1767
Sewer overloading by heavy rainfall
20
Blocked sewer pipe or pump
222
Blocked or broken house connection
131
Problem related to other urban water system
47
components: groundwater/surfacewater/drinking water
Not relevant
1301
# entries/class
Haarlem
2455
12
32
61
124
493
Selection of features
The basic features employed in this work are based on individual words in
the call texts, a typical choice in text classification. More complex word
combinations and grammatical constructions were not used.
To start with, call texts are split into separate words. This gives 216231 separate
words spread over 8544 vocabulary units, i.e., unique words. In order to reduce
the size of the database, all words that occur only once have been removed.
This reduces the number of different words to 4489. Words of only 1 or 2
characters have been removed as well since most of these are words with low
information content like the Dutch definite article “de”. This results in a list of
4378 words that are used to compile an initial dataset of word count features in
the following way. Every single call text, of the total of 6359 and 6980 call texts,
is represented by a 4378-dimensional feature vector in which every dimension,
every feature, corresponds to the number of times a particular word, from the
4378 words, occurs in the call text. This feature set size is very large, which
implies that a very large number of training records is needed for training and
calculation times for classifier training and testing are long. Additionally, a
high-dimensional feature space may result in the earlier-mentioned curse of
dimensionality. Therefore, latent semantic analysis was applied to reduce the
initial number of features before starting the experiments. Latent semantic
analysis is a multivariate analysis technique that is very similar to well-known
219
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2
principal component analysis and it selects and combines features based on
their (relative) importance (Manning et al., 2008). The result is a reduced list
of 1024 features that are ranked according to decreasing importance.
Classifiers for automatic call classification experiments
Three classifiers were tested to give a first idea of the applicability of
automatic classification of municipal calls. Two of them have been introduced
above, i.e., the nearest mean classifier (NMC) and the first-nearest neighbor
(1NN) classifier. A third classical and well-known classifier we used in our
experiments is Fisher’s linear discriminant (FLD), also referred to as linear
discriminant analasis (LDA) (Bishop, 2006, Duda et al., 2001, Webb, 2002).
These classifiers were chosen because of their straightforward structure and
associated short calculation times, which facilitates our experiments. Moreover,
results obtained employing these relatively straightforward classifiers, which
can be seen as representatives from different parts of the classifier spectrum
(Mansilla and Ho, 2004)., should give an indication of their potential use of
pattern recognition in automatic call classification.
Experimental set-up: Learning curves
For practical application of automatic call classification, classifiers are to be
trained anew for each new call center dataset. The natural way to proceed is
to provide a training set from the dataset for which calls have been classified
manually. The smaller the size of the dataset that is needed for training, the fewer
calls need to be classified manually and the less time-consuming application to
new datasets will be. This in turn enhances the usefulness of automatic call
classification for practical applications. Dataset size depends on the number
of features needed for classification and on the number of records needed for
classifier training. Classifier performance for different dataset sizes is tested in
learning curve experiments. Three experiments have been conducted with the
available datasets of 6359 and 6980 call records and 1024 features.
220
Automatic classification of call data for quantitative urban flood risk analysis
- Learning curve for feature set size
A learning curve of classification error as a function of the number of features
shows how many features are needed to obtain a minimum classification error.
As features set size increases, more information is available for classification;
more features also require a higher number of training records
- Learning curve for training set size
A learning curve of classification error as a function of the number of training
objects provides information to determine the required dataset size, for a given
number of features, to obtain sufficient accuracy of the classification results.
In practical applications, sufficient accuracy depends on the sensitivity of
applications to classification errors. For new datasets the required dataset size
determines the number of records that is to be trained manually.
Application of automatic call classification results for quantitative fault tree
analysis
The applicability of automatic call classification results in quantitative risk
analysis is tested in a quantitative fault tree analysis for urban flooding. Figure
1 shows the fault tree model that was used, including four failure mechanisms
that can give rise to urban flooding. Three failure mechanisms are related to
urban drainage systems and are represented as basic events in the tree: blockage
of inflow processes, e.g. blockage of gully pots gully pot connections, hydraulic
overloading as a result of heavy rainfall and blockage of sewer pipes and pumps.
Problems in other water systems are not analysed in detail; failure related to
these water systems are lumped into an undeveloped event, represented by a
diamond symbol instead of a circle. More information on the construction of
the fault tree can be found in (Sebastiani, 2002). The fault tree analysis results
obtained based on automatically classified calls are compared to the results
based on manual classification of the calls.
221
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2








 


 


 







Figure A2.1. Fault tree model for urban flooding used to test sensitivity of quantitative
fault tree analysis results to errors in automatic classification results.
Results
Learning curve number of features
Figure 2 shows learning curves for the Breda and the Haarlem datasets, for
the three classifiers LDA, NMC and 1NN, for increasing feature set size. The
classification error rate, i.e. the rate of wrongly classified calls out of the total
number of calls, has a clear minimum for LDA and NMC as a results of the
counter-intuitive effect of increasing error-rate with increasing feature set size.
The error-rate in the 1NN curve is almost insensitive to the size of the feature
set; error rates for all feature set sizes are above the minimum errors for LDA
and NCM. The optimum number of features, based on these learning curves,
is 200 for LDA and 300 for NMC. The plots also show that the minimum error
rate for Breda is higher than for Haarlem. This will be explained later in this
chapter.


222

Automatic classification of call data for quantitative urban flood risk analysis








         





         




         

         
























         


         

Figure A2.2 Learning curves for feature number increasing from 100, with steps of 100
to 1000, for the Breda and Haarlem datasets. 50% of the dataset is used for training and
50% is used for testing. Boxplots are based on 10 repetitions of the training and testing
procedures of the classifiers.
Learning curve training set size
Learning curves for increasing training set size were created by successively
using 10% up to 90% of the dataset for training and the other 90% down to
10% of the dataset for testing. Feature set sizes of 200 for LDA, 300 for NMC
and 300 for 1NN were applied, based on the learning curves for feature set size.
223

R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32


















        

        

 




        






        






R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2


        




        

Figure A2.3 Learning curves for increasing fractions 0.1 to 0.9 of the dataset used
for training; the remainder of the dataset is used for testing. Boxplots are based on 10
repetitions of the training and testing procedures of the classifiers.
Figure 3 shows how error rates decrease with increasing training set size; the
uncertainty in error rate increases as a result of smaller test sets as training
set sizes grow. The lowest mean error rate for the Breda dataset is 0.18 and is
obtained applying LDA, when 90% of the dataset is used for training, i.e. 6282
training records. The lowest mean error rate for the Haarlem dataset is 0.13
and is obtained applying LDA, when 60% or more of the dataset is used for
training or at least 3815 training records. The lower error rate for the Haarlem
dataset is explained by the presence of one large class that contains 77% of the
call records. This implies that if all records were erroneously assigned to this
largest class, the error rate would be 0.23. The Breda dataset is more balanced;
the largest class contains 51% of the call records, corresponding with an error
rate of 0.49 if all records were assigned to this class.
224
Automatic classification of call data for quantitative urban flood risk analysis
In order study the nature of the classification errors in more detail, class
confusion matrices were created that show the results for all classes for both
the true (manually classified) labels and the labels assigned through automatic
classification. Table 4 shows the confusion matrix for LDA, for the Breda
dataset; correctly labelled records are on the matrix diagonal, erroneously
labelled records are off-diagonal.
Table A2.4 Class confusion matrix for the results of LDA, for 200 features and 50% of
the dataset used for training and for testing. Classification error rate: 0.20
True
labels
1
2
3
4
5
6
Assigned labels
1
2
3
Inflow process blocked
1503 19
30
Overloading by heavy rainfall 7
7
0
Blocked sewer pipe or pump 36
1
108
House connection problem
12
6
11
Other water system problem 5
0
0
Not relevant
132 5
55
Sum assigned
1695 38
204
4
27
1
9
79
2
20
138
5
2
0
2
0
29
12
45
6
186
5
66
23
11
1077
1368
Sum
True
1767
20
222
131
47
1301
3488
Σcorrect
/Σtrue
0.85
0.35
0.49
0.60
0.62
0.83
0.80
The matrix shows that the classifier has special difficulty in distinguishing
records for the smallest class, class 2, which has the lowest correct/true ratio
of 0.35. This is probably due to the lower number of available records for
training in this class. Surprisingly, class 5, which also has a small class size, has
a correct/true ratio of 0.62, higher than the ratio for the larger classes 3 and 4.
The confusion matrix for NMC (not shown here) has a correct/true ratio above
0.5 for all classes except class 2. For the Haarlem dataset, LDA gives a low ratio
for class 2 of 0.08, while NMC gives a ratio of 0.58. Class 5 scores are good for
LDA and NMC for both datasets, which implies that class 5 is easy to recognise
for these classifiers. Classification results for class 1 are robust: correct/true
ratios are above 0.85 for LDA and NMC for both datasets. This is a result of
the large size of class 1 compared to other classes.
225
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2
Sensitivity of fault tree analysis results to errors in automatic call
classification
Probabilities of occurrence of events in the fault tree were calculated based
on manual and automatic call classification results for the events in the tree.
Automatic classification results for LDA, 200 features are used. Probabilities
are derived from the number of calls in each class, divided by the number
of independent flooding events. Quantitative fault tree analysis is based on
Monte Carlo simulation: the occurrences of basic and undeveloped events are
simulated with the use of a random number generator. Each simulation that
results in failure is stored, with the combination of causes that led to flooding.
A Monte Carlo simulation for the case of Breda with manually classified calls
results in a probability of flooding of 0.68 per event per 100 km sewer length. A
Monte Carlo simulation with automatically classified calls results in a probability
of flooding of 0.66/event/100km. Tables 5 and 6 show the contributions of the
basic events to the overall probability of flooding for the 2 simulations. The
results show that errors in the automatic classification procedure have only
limited influence on the outcomes of the fault tree calculations. The overall
probability of flooding remains approximately the same: the contribution of the
main failure mechanism, blockage of inflow processes is 92% for both manual
and automatic classification results. The contribution of the smallest failure
mechanism, overloading, changes by 1%, from 2% for manual classification to
3% for automatic classification.
Table A2/5. Results of 10,000 Monte Carlo simulations with the fault tree model for
Breda, manual classification
Flood causes
Inflow process blocked
Overloading by heavy rainfall
Blocked sewer pipe or pump
Other water system problem
226
Contribution to overall probability of failure
9212 out of 10,000 (92%)
156 out of 10,000 (2%)
1654 out of 10,000 (17%)
344 out of 10,000 (3%)
Automatic classification of call data for quantitative urban flood risk analysis
Table A2.6. Results of 10,000 Monte Carlo simulations with the fault tree model for
Haarlem, automatic classification
Flood causes
Inflow process blocked
Overloading by heavy rainfall
Blocked sewer pipe or pump
Other water system problem
Contribution to overall probability of failure
9153 out of 10,000 (92%)
314 out of 10,000 (3%)
1572 out of 10,000 (16%)
339 out of 10,000 (3%)
Discussion
Learning curves for varying feature set sizes show that LDC and NMC suffer
from the “curse of dimensionality”: minimum error rates are obtained for
feature set sizes of 200 and 300 and error rates rise rapidly for larger feature set
size. 1NN is less sensitive to feature set size and error rates vary only little with
varying feature set sizes.
Error rates decrease as the training set grows, up to half the total dataset; larger
training set sizes give only limited improvement of the error rate. This implies
that a training set size of about 3000 records is needed for application of the LDA
and NMC classification schemes to new call datasets. The 1NN classification
scheme performs poorly for this classification task: it results in high error rates
compared to LDA and NMC. In this case, error rates decrease more slowly
with increasing training set size and have not yet reached a minimum when
90% of the dataset is used for training. Potentially, the addition of more records
could bring the performance of 1NN to the level of LDA or NMC, but in the
current situation one of the latter classifiers is clearly to be preferred over 1NN.
Confusion matrices for LDA and NMC show that small classes are most
sensitive to classification errors. This implies that classification accuracy for
these classes could improve if data sets with larger numbers of calls in these
classes were available. In practical applications, call numbers increase with
time as a call centre stays in operation. As data set size grows, larger training
and test sets become available and classifiers can be retrained to obtain higher
227
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2
accuracy. It is more efficient, and possibly equally effective, to purposefully
acquiring examples of the smaller classes only in order to improve their
accuracy. Obviously, overall performance improvements might be obtained
by choosing yet another classification technique from the large number of
approaches that have already been proposed in the literature (see [referenties
naar PR en ML literatuur]). What is potentially more powerful is to develop
classifiers and construct features that are more dedicated to handling municipal
call data as the integration of the correct prior information should generally be
beneficial. Nonetheless, the power and potential of the presented methods and
their variations should be apparent from the initial study we offered.
Minimum error rates of 0.18 and 0.13 are obtained for the datasets of Breda
and Haarlem, for the LDA classification scheme. Application of classification
results in quantitative fault tree analysis shows that error rates of 0.18 and
0.13 for Breda and Haarlem do not distort the outcomes of the analysis: the
ranking of failure mechanisms and their contributions to the overall probability
of flooding change by at most 1%.
For other applications in risk assessment absolute probabilities of occurrence
of individual classes may be needed; in that case error rates of more than 30%,
as obtained in the presented applications for small classes, are likely to be
unacceptable. For such applications, larger data set sizes for smaller classes
are required or alternative, more elaborate classification schemes could be
explored to obtain lower error rates. The same is true if calls are used to identify
vulnerable locations for flooding, for specific failure mechanisms. In that case,
correct labelling of individual calls is important which is more sensitive to
classification errors than the total number of calls per class.
Instead of training a classifier for anew for each individual call database, the
trained classifier of one database can be directly applied for classification of
a new database. If the classifier has good portability from one database to
another, it will provide acceptable classification results for the new database.
This means no new classifier needs to be trained to classify new databases. This
228
Automatic classification of call data for quantitative urban flood risk analysis
offers opportunities for broad application of automatic call classification: once
a classifier with good portability is found and trained, many databases can be
trained with the same classifier. Whether classifiers with good portability can
be found and trained is a topic for further research.
Conclusion
The results of this study show that simple automatic classification schemes like
LDA and NMC can classify call datasets with error rates below 0.2. Classifiers
perform better for large class sizes than for small classes, probably due to the
larger number of available training objects. The presence of one large class in the
Haarlem dataset, containing 77% of the call records results in a low error rate of
0.13; for the Breda dataset with a more balance distribution of calls over classes,
an error rate of 0.18 can be obtained. Application of automatically classified
datasets in quantitative fault tree analysis shows that obtained classification
accuracy is sufficient to correctly rank failure mechanisms according to their
contributions to the overall probability.
Acknowledgement
The authors want to thank Breda and Haarlem municipalities for making
available the data in their call centre database. We also would like to thank
the people that develop and maintain the Matlab pattern recognition toolbox
PRTools (prtools.org; van der Heijden, F. and Duin, R. and De Ridder, D. and
Tax, DMJ (2004). Classification, parameter estimation, and state estimation:
an engineering approach using MATLAB. John Wiley & Sons Inc), especially
Dr. R.P.W. Duin.
229
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 2
References
Arthur S., Crow, H., Pedezert, L., Karikas, N. (2009). The holistic prioritization of proactive
sewer maintenance. Water Science and Technology, 59(7), 1385-1396.
Bishop, C.M. (2006). Pattern recognition and machine learning. Springer, 2006
Breiman, L., Freidman, J.H., Olshen, R.A., Stone, J.S., Classification and regression trees,
Wadsworth, 1984
Duda, R. O., Hart, P. E., Stork, D. G. (2001). Pattern Classification (2nd ed.), John Wiley
and Sons, 2001
Dirksen, J., Veldhuis, J.A.E. ten, Schilperoort, R. P. S. (2009). Fault tree analysis for dataloss in long-term monitoring networks. Water Science and Technology, 60(4), 909915
doi: 10.2166/wst.2009.427
S. Garfield and S. Wermter, “Call Classification using recurrent neural Networks, SVMs
and finite State Automata”, Knowledge and Information Systems 9(2), 2006. pp. 131-156.
A.L. Gorin, G. Riccardi, J.H. Wright. “How may I help you ?”, Speech Communication 23,
1997. pp. 113-127.
A.K. Jain, R.P.W. Duin, J. Mao. (2000). Statistical pattern recognition: A review. IEEE
Transactions on pattern analysis and machine intelligence, 4-37
Manning, C. D., Raghavan, P., Schütze, H. (2008). Introduction to information retrieval.
Cambridge University Press, 2008.
Mansilla, E., Ho, T.K (2004). On classifier domains of competence. In: Proceedings of the
17th International Conference on Pattern Recognition, pp. 136–139 (2004).
RIONED Foundation. “Inquiry on flood problems in the built environment”, RIONED
Foundation, 2007 (in Dutch)
Sebastiani, F. (2002). Machine learning in automated text categorization. ACM computing
surveys (CSUR, 34(1), 1-47.
Ten Veldhuis, J.A.E., Clemens, F.H. L. R. and van Gelder, P.H. A. J. M.(2009).
Quantitative fault tree analysis for urban water infrastructure flooding’, Structure
and Infrastructure Engineering. doi: 10.1080/1573247090298587
Webb, A. Statistical Pattern Recognition, John Wiley & Sons, New York, 2002
230
Automatic classification of call data for quantitative urban flood risk analysis
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
231
Appendix 3
Risk curves for urban pluvial flooding
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 3
This appendix presents risk curves for all consequences of urban pluvial flooding
used in the analysis of data from the municipal call centre of the city of Haarlem,
over the period 1997 to 2007.
234
Risk curves for urban pluvial flooding
Introduction
Risk assessment studies often present the expected value of risk as a summary
value for a range of probabilities and consequences or they give a risk value for
a given scenario, e.g. a certain return period. Risk curves go one level deeper
and present risks for a range of probabilities and consequences (Kaplan and
Garrick, 1981). Risk curves for urban flooding depict flood damages on the
horizontal axis and their associated exceedance probabilities on the vertical
axis. Figure A3.1 gives an example of a risk curve, for a flood damage xi varying
from 0 to 100 on the horizontal axis and associated exceedance probabilities on
the vertical axis.

 

















 



Figure A3.1. Example of a risk curve (based on: Kaplan and Garrick, 1981): a
complementary cumulative distribution function (CCDF), i.e. the probability of
exceeding a given damage
235
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 3
Risk curves for urban flooding depict flood damages on the horizontal axis and
their associated exceedance probabilities on the vertical axis. The intersection
of the curve with the vertical axis gives the probability of any damage at all;
the intersection with the horizontal axis gives the maximum possible damage,
with zero probability of exceedance. Values in between are interpreted as
probabilities of at least damage xi; this probability increases or remains constant
for decreasing damages. The staircase function is the plotted result of a series
of points representing damage for scenario i and the exceedance probability
for each scenario. The staircase function can be regarded as a discrete
approximation of a continuous reality, represented by the smooth curve. The
area below the risk curve is a measure of total risk; the further risk curves shift
to the top-right-hand corner of the graph, the higher their associated total risk.
The advantage of risk curves compared to one value for expected risk is that
risk curves give insight into the contributions of small and large damages to
flood risk. If flood risk is mainly associated with small damage incidents, the
curve decreases steeply for small damages and more gently for high damages,
as is the case of the example in figure A3.1. If large damages mainly compose
risk, the curve is more or less flat for small damages and steeply decreases at
large damage values.
Preparation of call data to construct risk curves
Table A3.1 summarises results of call classification for consequence classes of
urban pluvial flooding, for the case of Haarlem. Sixteen consequence classes
are distinguished, based on information in the call texts. Classified calls are
subsequently assigned to independent rainfall events, as described in chapter 2
of this thesis. This results in a matrix of events and consequence classes; each
cell in the matrix gives the number of calls received per event per consequence
class. For each consequence class, the incidence of numbers of calls per event is
determined. The result is illustrated in table A3.2, where X is the number of call
per event per consequence class. A small number of calls per consequence class
per event means that the amount of associated flood damage is small. Table
A3.2 shows that this is the case of most events: call incidence 1 per event per
236









































 
































 

 
 


 


 

 
 




 










Risk curves
for 
urban pluvial flooding












 


              



















             



            

















 

 

Call

 

 


 

class (X=1)
occurs most
frequently.
incidence
of more
than 10 per
event





























































(X>=10)
occurs
only
for 3
consequence
classes.




 


















 
 

 



 

 

























The
results
in
table
A3.2
are
used
to
calculate
probabilities
of
occurrence
of






consequence classes. The occurrence of events is assumed to be a Poisson








 

in 

process, 
which
implies that the
probability that
an event
will 
occur
any  






































 


is


 






 
 

 

specified
short
time
period
approximately
proportional
to the length
of the















             
time period. The occurrences of events in disjoint time periods are statistically



             


independent.
Under these 
conditions, the
of occurrences
x in
some 






 


number
 

 
 





































fixed period
of 
time
is a
Poisson
distributed
variable:





 
   













 


 
 
  
 
 
 
 
 
 
 


  


             









   
 




(A3.1) 





 








 

: probability
of time t


Where:
 
of xoccurrences

 in a period

  

 
 




λ
: average rate of occurrence of events per time unit











 









 



Since failure
occurs due to the occurrence of 1 or more events, the probability










can
 be
 calculated
  





 failure

      from:

of








                 








 










 
  









(A3.2)

  
















     




of one or more events


 


:probability

Where:
 

  : probability

of no events


 

The time period t can be chosen at will; the longer t, the higher the probability
of occurrence. The time scale is preferably chosen so as to fit the frequency of
events. In the case of urban flooding flood events typically occur up to several
times per month and the duration of events is in the order of several days. A
time period of 1 week fits the event occurrence frequency and has been chosen
for the construction of risk curves.
237





 

R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 3
Table A3.1 Call classification results for aggregated and for detailed flood consequence
classes, for the cases of Haarlem, for a period of 10 years
Primary functions Consequence classes
Human health:
physical harm or
infection
Flooding with wastewater
Manhole lid removed
Flooding in residential building (house/garage/shed)
Flooding in commercial building (shop/storage hall)
Flooding in basement
Water splashes onto building
Flooding of gardens/park
Flooding in tunnel
Flooding at bus stop/taxi stand
Flooding in shopping street/place/commercial centre
Flooding in front of entrance to shop/bar/library/hospital
Flooding in front of entrance to residential building
Flooding on residential/main street
Flooding on cycle path
Flooding on sidewalk/footpath
Flooding on parking space
Total number of calls relevant for flooding
No consequence mentioned
Consequence other than flooding
Total number of calls
Protection of
buildings and
infrastructure:
damage to public
and private
properties
Prevention of road
flooding: traffic
disruption
238
Nr. of calls/
class:Haarlem
(nr) (%)
61
3.4
7
0.4
116
6.5
34
1.9
173
9.7
26
1.5
74
4.1
13
0.7
18
1.0
117
6.5
55
3.1
65
3.6
655 36.5
133
7.4
73
4.1
173
9.7
1793 100%
3563
1005
6361
Risk curves for urban pluvial flooding
Table A3.2 Incidence of events with X calls per class, for consequence classes E0, E101
to E116. Call incidence above 0 is shaded in grey.
X: number of events with X (for X 1 to 30) calls per class; E101: Flooding in residential
building; E102: Flooding in commercial building; E103: Flooding in basements; E104:
Flooding on streets; E105: Flooding of tunnel; E106: Flooding on cycle path; E107:
Flooding on footpath; E108: Flooding on parking space; E109: Flooding at bus stop;
E110: Flooding with wastewater; E111: Manhole lifted due to flooding; E112: Flooding
of green areas (parks/gardens).
X E0 E101 E102 E103 E104 E105 E106 E107 E108 E109 E110 E111 E112
1
100
38
15
33
75
9
42
37
46
17
25
4
27
2
61
6
5
12
35
0
11
4
7
0
2
1
5
3
41
1
0
3
19
0
7
0
8
0
1
0
2
4
22
1
1
1
12
0
1
2
5
0
0
0
1
5
27
0
0
2
6
0
2
0
3
0
0
0
0
6
17
1
0
1
4
0
0
0
2
0
0
0
2
7
18
2
0
3
2
0
1
0
1
0
0
0
0
8
9
0
0
0
1
0
0
0
1
0
0
0
0
9
8
1
0
0
6
0
0
0
0
0
0
0
0
10
7
1
0
0
0
0
0
0
0
0
0
0
0
11
6
0
0
1
1
0
0
0
0
0
0
0
0
12
1
0
0
1
2
0
0
0
0
0
0
0
0
13
3
0
0
0
0
0
0
0
0
0
0
0
0
14
6
0
0
0
0
0
0
0
0
0
0
0
0
15
3
0
0
0
1
0
0
0
0
0
0
0
0
16
2
0
0
0
1
0
0
0
0
0
0
0
0
17
2
0
0
0
0
0
0
0
0
0
0
0
0
18
2
0
0
0
1
0
0
0
0
0
0
0
0
19
0
0
0
0
0
0
0
0
0
0
0
0
0
20
4
0
0
0
1
0
0
0
0
0
0
0
0
21
1
0
0
0
0
0
0
0
0
0
0
0
0
22
3
0
0
0
0
0
0
0
0
0
0
0
0
23
1
0
0
0
0
0
0
0
0
0
0
0
0
24
2
0
0
0
0
0
0
0
0
0
0
0
0
25
2
0
0
0
0
0
0
0
0
0
0
0
0
26
1
0
0
0
1
0
0
0
0
0
0
0
0
27
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
28
1
0
0
0
1
0
0
29
1
0
0
0
0
0
0
0
0
0
0
0
0
30
0
0
0
0
0
0
0
0
0
0
0
0
0
>30
17
0
0
0
0
0
0
0
0
0
0
0
0
239
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Risk curves for consequence classes of urban pluvial flooding

Figures A3.2 and A3.3 give 2 examples of risk curves for individual damage

classes. Flood consequence severity on the horizontal axis is expressed as

amount of calls per incident. The risk curves show that the maximum amount


 
 
 
of calls 
for flooding
on 
streets
is more
than twice
as highas
for flooding
in 
             
residential buildings. The probability of at least 1 call is more than 3 times

higher for flooding on streets than flooding in residential buildings.





R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 3



















Figure A3.2. Risk curves (smoothed lines) and staircase functions for consequence class
          
‘flooding
on streets’,
based on
call amounts
per
incident as
a measure


 
 

forconsequence
  

severity
240
Risk curves for urban pluvial flooding






















          
Figure A3.3. Risk curves (smoothed lines) and staircase functions for consequence
            
class ‘flooding in residential buildings’, based on call amounts per incident as a measure

 for consequence severity.

           

Risk graphs for other consequence classes (figures A3.4 to A3.7) show that for

most
consequence
classes,
the 
maximum
number of
calls
per incident
is below



 
 

  
5. Probabilities of at least 1 call per event vary from 0.009 per week for lifted

 manholes to 0.13 per week for flooding on parking spaces. Most risk curves

 decrease steeply for increasing numbers of calls per event, indicating that flood
 risk for most consequence classes is associated with small events.



241



R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32

R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 3









































 
Figure
A3.4.
Risk
curves
for
consequence classes
related
to 
flooding
in buildings,
based

on call amounts per incident as a measure for consequence severity.


242



Risk curves for urban pluvial flooding











































Figure A3.5. Risk curves for consequence classes related to flooding of streets, based

on call amounts per incident as a measure for consequence severity.
243


R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32


R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 3

































Figure A3.6. Risk curves for consequence classes related to flooding with wastewater,



green



of
 

lifted
manholes,
flooding of
spaces
and
flooding
in front
entrances
to residential
           
buildings, based on call amounts per incident as a measure for consequence severity.

244



Risk curves for urban pluvial flooding


 































Figure
A3.7.
Risk
curves
consequence
classes
relatedto
flooding
in
front
of




for



 
shopping
streets,



entrances
to commercial
facilities,
flooding in
flooding
of 
tunnels
and
              
flooding at bus stops, taxi stands and bus and train stations, based on call amounts per

 incident as a measure for consequence severity.

245



R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Appendix 3
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
246
List of publications
List of publications
International peer-reviewed journals
ten Veldhuis, J.A.E., Clemens, F.H.L.R., Sterk, G., Berends, B.R. (in
press). Microbial risks associated with exposure to pathogens in
contaminated urban flood water, Water Research (2010), doi:10.1016/j.
watres.2010.02.009
Veldhuis, J.A.E. ten, Clemens, F.H.L.R. (in press). Flood risk modelling based
on tangible and intangible urban flood damage quantification. Water
Science and Technology (2010).
Dirksen, J., Veldhuis, J.A.E. ten, Schilperoort, R. P. S. (2008). Fault tree
analysis for data-loss in long-term monitoring networks. Water Science
and Technology, 60(4), 909-915. doi: 10.2166/wst.2009.427
Veldhuis, JAE ten, Clemens, FHLR & Gelder, PHAJM van (2008). Fault
tree analysis for urban flooding. Water Science and Technology, 59(8),
1621-1629.
doi: 10.2166/wst.2009.171
Ten Veldhuis, J.A.E., Clemens, F.H. L. R. and van Gelder, P.H. A.
J. M.(2009). Quantitative fault tree analysis for urban water
infrastructure flooding’, Structure and Infrastructure Engineering. doi:
10.1080/15732470902985876
Book contributions
Veldhuis, J.A.E. ten: Contributions to chapter 6, Urban Flood Risk Assessment.
In: C. Zevenbergen, A. Cashman, N. Evelpidou, E. Pasche, S. Garvin
and R. Ashley, Urban Flood Management, CRC Press/Balkema – Taylor
& Francis Group, London, 2011 (in press)
Dirksen, J, Goldina, A, Veldhuis, JAE ten & Clemens, FHLR (2007).
The role of uncertainty in urban drainage decisions: uncertainty in
inspection data and their impact on rehabitation decisions. In: Strategic
Asset Management of Water Supply and Wastewater Infrastructures.
Editor(s): Helena Alegre, Maria do Ceu Almeida
Publication Date: 15 Sep 2009 • ISBN: 9781843391869
247
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
List of publications
Editorship
Meulen, M van der, Veldhuis, JAE ten & Schilperoort, RPS (Eds.). (2007).
5th International Conference on Sewer Processes and Networks. Delft:
TU Delft. (TUD)
Non peer-reviewed journals
Veldhuis, JAE ten & Clemens, FHLR (2007). Uncertainty in risk analysis of
urban pluvial flooding: a case study. Water Practice and Technology,
4(1),
doi: 10.2166/WPT.2009.018
Conference Proceedings
Veldhuis, J.A.E. ten, Harder, R.C., Loog, M. (accepted for oral presentation).
Automatic classification of municipal call data for quantitative urban
drainage system analysis. First European Congress of the IAHR. IAHR,
Edinburgh, May 2010.
Veldhuis, J.A.E. ten, Clemens, Dirksen, J., Clemens, F.H.L.R. (2009).
Evaluation of operational strategies to control sewer flooding. In s.n.
(Ed.), 3rd Leading-edge conference on strategic asset management –
LESAM2009 (pp. 1-8). AWWA, Miami, November 2009.
Veldhuis, J.A.E. ten, Clemens, F.H.L.R. (2009). Flood risk modelling based on
tangible and intangible urban flood damage quantification. In s.n., 8th
International Conference on Urban Drainage Modelling (1-8). IWA,
Tokyo, September 2009.
Dirksen, J, Veldhuis, JAE ten & Schilperoort, RPS (2009). Fault tree analysis
for data-loss in long-term monitoring networks. In s.n. (Ed.), 10th IWA
conference on instrumentation, control and automation (pp. 1-8). IWA,
Cairns, June 2009.
Veldhuis, JAE ten, Clemens, FHLR & Gelder, PHAJM van (2008). Fault
tree analysis for urban flooding. In s.n. (Ed.), ICUD08 (pp. 1-10). IWA,
Edinburgh, September 2008.
248
List of publications
Sterk, G., Veldhuis, JAE ten, Clemens, FHLR & Berends, BR (2008).
Microbial risk assessment for urban pluvial flooding. In s.n. (Ed.), 11
ICUD08 (pp. 1-10). IWA, Edinburgh, September 2008.
Dirksen, J, Goldina, A, Veldhuis, JAE ten & Clemens, FHLR (2009).
Probabilistic modeling of sewer deterioration using inspection data.
In s.n. (Ed.), IWA 2nd leading edge conference on strategic asset
management – LESAM (pp. 1-8). IWA, Lisbon, October 2008.
Veldhuis, JAE ten, Clemens, FHLR (2007). Uncertainty in risk analysis of
urban pluvial flooding: a case study. In R Schilperoort & M van der
Meulen (Eds.), Sewer processes and networks (pp. 41-50). IWA, Delft,
August 2007.
National publications
Veldhuis, J.A.E. ten (2009). Wat modellen niet vertellen over wateroverlast. In:
Water binnen gemeentegrenzen. Handvatten voor de practicus. Uitgave
van Land+Water. Editor: Bas Keijts. Koninklijke BDU Uitgevers B.V.
Man, H de, Leenen, I & Veldhuis, JAE ten (2009). Water in de stad: een risico
voor de volksgezondheids. H2O: tijdschrift voor watervoorziening en
waterbeheer, 9(33), 54-69.
Ven, FHM van de, Veldhuis, JAE ten, FHLR (2009). Betekent een ander
klimaat grotere riolen? H2O: tijdschrift voor watervoorziening en
waterbeheer, 16/17, 28-28.
Man, H de, Clemens, FHLR, Veldhuis, JAE ten, Moens, M & Klootwijk, M
(2008). Onderzoek naar de nauwkeurigheid van debietmetingen in deels
gevulde leidingen. Rioleringswetenschap en -techniek, 8(31), 33-46.
Veldhuis, JAE ten, Rooij, NF de & Moens, M (2006). Gemeente Breda
zet stedelijke wateropgave op de kaart. H2O: tijdschrift voor
watervoorziening en waterbeheer, 18-21.
249
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
List of publications
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
250
Acknowledgements
Acknowledgements
This thesis is the result of three and a half years of research investigating
problems of urban flooding in lowland areas. The main reason for undertaking
this work is that I felt that the existing approach of urban flooding analysis that
strongly focuses on hydrodynamic modeling was not addressing urban flooding
problems appropriately. According to my experience as an urban drainage
manager in Breda, models were unable to simulate reality: they showed flooding
locations that had no record of flooding and they missed locations where people
had to protect themselves from recurrent flooding by installing barriers in front
and back doors. An important reason was that we lacked monitoring data to
calibrate the hydrodynamic models, as is the case of many cities throughout
the world. Even if cities have a monitoring network installed, it takes a lot
of energy to obtain reliable data and large data gaps are more of a rule than
an exception. That is why I felt alternative methods should be developed to
obtain information on urban drainage system functioning. Not to replace
hydrodynamic models, but to complement data from monitoring networks.
Preferably this complementary information should come from data sources that
are widely available: data that is collected anyhow and can be put to use for
analysis. Call centre data proved to be such a data source and analyzing these
data has been an enriching experience and great fun, too, at times.
This thesis could not have been completed without the help of people that
I have had the pleasure to work with over the past few years. After I left
university I have been in doubt whether or not I should pursue a PhD degree.
To do this while working on a full time job is not evident and going back to
university as PhD student never seemed a realistic option. Until the option
of a university job combining research and education popped up. François,
I’m grateful to you for suggesting this opportunity to me and for guiding me
into the scientific world. My colleagues in urban drainage research provided
a stimulating working environment and were a great support in busy times.
Rémy, I have good memories of the time we organized the SPN-conference
251
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Acknowledgements
together; Ivo, thanks for upgrading my knowledge of hydraulics; Jojanneke,
you were always ready to read through my texts critically. My work load was
much relieved by Robin Harder, who helped me for several years to develop
accessible call databases that were a prerequisite to be able to make sense of the
data. I enjoyed working with Boyd Berends and Frans van Knapen at IRAS,
who introduced me into the particulars of microbiological risk assessment.
Gerdien Sterk’s graduation work provided the data that allowed me to add this
component to my research study.
This research would not have been possible without the help of representatives
from the cities of Haarlem and Breda who provided their call data registers and
hydrodynamic model databases. I want to thank Timo Nierop and Marjolijn
Hoobroeckx who provided a lot of data and information for Haarlem, Mathijs
Vromans, Martijn Klootwijk and Michel Moens who supplied the data that
I needed from the call center and of the urban drainage system, while Peter
Verwijmeren supplied detailed series of rainfall data from his weather station.
Throughout the first years of my research I received valuable feedback from a
committee of scientists and practitioners. Hans Korving, Hugo Gastkemper,
Aad Oomens, Egbert Baart, Pieter van Gelder, Wil Thissen, the variety of
your backgrounds always gave rise to interesting discussions and provided
me with new food for thought. I’m grateful to Chris Zevenbergen for his
ever enthusiastic support and stimulating advice: our meetings at IHE and
for COST C22 have invariably given my work a boost of fresh energy. I
also want to mention the Flood Resilience Group from Unesco-IHE: Berry,
William, Sebastian, the atmosphere in your group has been inspiring. I owe
much to Richard Ashley whose independent thinking and meticulous reading
and comments have stimulated me to get the better out of my abilities. I also
thank you for introducing me into the COST C22 group, where I met a lot of
interesting people. The quality of my first articles greatly benefited from the
critical scrutiny by Marcel Donze whose characteristic comments and advice I
can never forget.
252
Acknowledgements
Finally, I want to thank my family for supporting me through sometimes
stormy times. Daniëlle, who would have thought we would spend this time at
the university together; it has been good to have a sister nearby. Jack and José,
I’m grateful to you for your help and understanding. Mom and Aad, you know
me better than anyone; I’m very happy we are sharing this experience together.
Cas, you are an unrivalled support and I’m sure we will find a new target for
discovery soon.
253
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Acknowledgements
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
254
Curriculum vitae
Curriculum vitae
Marie-claire (Johanna Antonia Elisabeth) ten Veldhuis was born on October
15, 1969 in Roosendaal, the Netherlands. She followed her secondary education
at St.-Oelbertgymnasium, from 1982 to 1988. In 1988 she started her study of
Civil Engineering at Delft University of Technology, from which she graduated
in 1995. Her MSc-graduation focused on an analysis of water quality problems
in a pond in Amsterdam that was fed by water from a separate storm water
system. After graduation she went to Belo Horizonte in Brasil to work on a
research project studying flooding problems in low-income areas of the city.
Upon her return in 1996 she started to work at IWACO BV as a water systems
consultant. She worked on various water quality and water management studies
and was sent abroad several times to work on a European project in Israel,
Jordan and the Palestinian Territories. In 2002 she moved to the city of Breda
to become manager of urban water and sewerage, which she did in cooperation
with a team of engineers and operators. Building on this experience she started
PhD research at Delft University in 2006. She combined this research with a
position as assistant professor, teaching master courses on urban drainage. Her
present research interests are data acquisition by monitoring and data mining
of existing data sources for analysis and operation of urban drainage systems.
255
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
Curriculum vitae
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
R17
R18
R19
R20
R21
R22
R23
R24
R25
R26
R27
R28
R29
R30
R31
R32
256
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement