International Journal of Disaster Risk Reduction 37 (2019) 101162

Contents lists available at ScienceDirect

International Journal of Disaster Risk Reduction

journal homepage: www.elsevier.com/locate/ijdrr

Evaluating adaptation measures for reducing flood risk: A case study in the T
city of Colombo, Sri Lanka
D.J. Wagenaara,b,∗, R.J. Dahmc, F.L.M. Diermansea, W.P.S. Diasd, D.M.S.S. Dissanayakee,
H.P. Vajjaf, J.C. Gehrelsa, L.M. Bouwerg
a
Deltares, Delft, the Netherlands
b
Institute for Environmental Studies, VU University, Amsterdam, the Netherlands
c
Deltares USA Inc. Silver Spring, USA
d
University of Moratuwa, Moratuwa, Sri Lanka
e
CECB, Colombo, Sri Lanka
f
UNITAR-UNOSAT, Geneva, Switzerland
g
Climate Service Center Germany, Helmholtz-Zentrum Geesthacht, Hamburg, Germany

A R T I C LE I N FO A B S T R A C T

Keywords: Many cities around the world face frequent problems with flooding, which is expected to get worse due to
Flood risk anthropogenic climate change and further urbanization. To tackle these problems often local infrastructural
Adaptation measures adaptation measures are proposed. In this study a chain of state-of-the-art models is presented that can be used to
Damage-curves evaluate the benefits of such measures. Also, a method is presented to calculate the costs of not responding to a
Wetlands
changing environment that is slowly aggravating floods. These methods are applied to a case study in the city of
Colombo
Colombo in Sri Lanka. Colombo faces problems with floods that are expected to get worse by further wetland
reduction and climate change. Several local measures (infrastructural interventions) are proposed to tackle that
problem. This paper shows a method to quantify the expected reduction in future flood damages resulting from
the proposed measures, and compares the risk reduction to the proposed measure costs. This is done by creating
probabilistic inundation depth maps using a 1D2D hydrodynamic model. A detailed flood damage model and
socio-economic development scenarios are then applied to estimate damage with and without the measures. An
economic analysis is done to demonstrate the benefits of the measures, which can be used by decision makers.
Additionally, calculations are carried out of future flood risk increases when wetland reduction in Colombo
continues. In this case, the effect of stopping wetland encroachment is found to be larger than the effect of the
structural adaptation measures.

1. Introduction wetland encroachment are quantified. A probabilistic 1D2D hydro-

dynamic model is used to assess the effects of the hydraulic measures
Floods cause the largest portion of insured losses of all catastrophes and extra wetland encroachment. A detailed flood damage model is
around the world and this is expected to increase due to climate change subsequently used to assess the flood damage at different return per-
[1]. Investments in measures have often been inadequate [2]. This is iods. Five long-term socio-economic growth scenarios are applied to
because of several reasons: Postponement because of short-term eco- obtain the range of possible future expected flood damage, given the
nomic reasoning, no consensus on how to evaluate the return on in- different measures and urban encroachment developments. This is used
vestment for measures and fear of making irreversible decisions that as input for a cost-benefit analysis to assess the cost-effectiveness of
turn bad over time [1]. In this study a chain of state of the art models is measure packages and to calculate the costs (flood damages) of more
presented that can be used to evaluate the return on investments of such wetland encroachment.
measures. The individual techniques applied to this case study are well es-
This is applied to a case study in the city of Colombo in Sri Lanka, tablished. Probabilistic 1D2D hydrodynamic models have for example
where several measures are proposed to counter increasing flood risk. been applied in the Netherlands [3] and in the UK [4]. Over the last 15
In this study the effects of these proposed measures and efforts to stop years, flood damage models have advanced considerably in their level

∗
Corresponding author. Deltares, Delft, the Netherlands.
E-mail address: [email protected] (D.J. Wagenaar).

https://doi.org/10.1016/j.ijdrr.2019.101162
Received 30 December 2018; Received in revised form 3 April 2019; Accepted 17 April 2019
Available online 12 May 2019
2212-4209/ © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

Fig. 1. Colombo Metropolitan region and its major river and canals, the red points presenting the calibration locations shown in Fig. 3.

of detail as a result of faster computing and more available data. benefit analysis. Next, results of the analysis are presented, as well as
Especially for developed countries, standard software packages are now conclusions about the effectiveness of the measures under present and
available to calculate flood damage on a high resolution (e.g. Refs. future conditions; and the cost of inaction when wetland encroachment
[5,6]. The application of cost-benefit analyses for decision making on is not stopped.
public investments is common practice (e.g. Ref. [7]; also for flood risk
management [8–11].
2. Case study
The aim of this paper is to present a case study in which different
state of the art risk analysis techniques are combined to evaluate
Colombo is highly prone to flooding, and has experienced regular
adaptation measures. This is useful when detailed plans for specific
floods for the past 30 years, affecting over 1.2 million people annually
measures exist. It can, for example, be applied by organizations plan-
[16]. The recurrent floods in the Colombo metropolitan area are due to
ning to invest in a specific project. This is different from formulating an
a combination of factors including unauthorized constructions that
adaptation strategy, for which a broader view at possible future struc-
obstruct water flow, dumping waste in the drainage canals, backwater
tural and non-structural measures is required and therefore a much
build-up in the main canal system, lack of regular maintenance of the
larger number of scenarios/measures considered (see e.g. Ref. [12].
drainage system, and commercial development in wetland reservations
Some of the techniques applied in this study would require too much
[16]. These activities have reduced Colombo's capacity to cope with
calculation effort to be used for such broader purposes.
high intensity rainfall that has become more frequent and intense due
Earlier studies that looked at the entire flood risk chain such as Apel
to the impact of climate change.
et al. [13]; Chen et al. [14] or Metin et al. [15] do not include an
To counter this increasing flood risk a package with adaptation
economic analysis. Studies that included economic analysis such as
measures (structural interventions) to the water system are proposed
Kind [8] typically simplify the flood risk chain. Another paper looking
and efforts to stop further wetland encroachment are recommended. In
at the entire flood risk chain is Löwe et al. [34], but this paper is fo-
this case study the effects of these proposed measures and efforts to stop
cused on pluvial floods while this paper focuses on fluvial floods.
wetland encroachment are quantified.
This paper starts with a description of the case study area, followed
The Colombo metropolitan area is located in the western coastal
by an overview of the approach, and descriptions of the 1D2D hydro-
plains of Sri Lanka. It is the most populated region of the country with
dynamic model, the probabilistic model, the damage model and the cost
over 2.3 million people living in the study area. It encompasses the

2
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

Fig. 2. Overview of the risk calculation process.

business capital and largest city of Sri Lanka, Colombo. The study area 3. Data and methods
includes the Colombo basin and the Kolonnawa basin. These basins are
predominantly rainfed by the South-Western monsoon season from May 3.1. Overview
to September and during the inter-monsoon season from March to April
(sometimes extending to May) and October to November. The mean The study quantifies flood risk and compares the reduction in flood
annual rainfall of this region is approximately 2400 mm, and the inter- risk with the costs of adaptation measures. This process consists of a
monsoonal rainfall is responsible for almost half the annual precipita- hydrodynamic model, a statistical model, a damage model and a cost
tion. benefit analysis. The hydrodynamic model is used for multiple simu-
The Kelani River and Diyawannawa Oya are the main rivers in the lations of flood levels with different boundary conditions for sea water
study area. Diyawannawa Oya flows within the Colombo city areas and level, river discharge, local rainfall, and initial soil saturation condi-
then connects to the Kelani River. Kelani River is the second longest tions. A probabilistic model is then applied to develop inundation depth
river of Sri Lanka and has its outflow to the sea at the North boundary maps for different return periods based on the probabilities of the dif-
of the city of Colombo. Its flow varies between 800 and 1500 m³/s ferent boundary condition combinations. A detailed flood damage
during the wet season and 20–25 m³/s in the dry season, depending on model with exposure data for buildings and other objects and vulner-
the operation of three upstream reservoirs in the catchment. ability curves is then used to translate inundation depth maps into
Diyawannawa Oya is connected by a canal and tunnel system to drain damage maps for different return periods. These damage maps at dif-
the storm water to either the ocean or to the Kelani River. The water ferent return periods are used to calculate Expected Annual Damage
flow in the major canals is mainly governed by the hydraulic gradient, (EAD) maps, i.e. flood risk. Socio-economic scenarios are used to ex-
as about 80% of these canals have bed levels of about 1 m below mean trapolate the EAD into the future. Fig. 2 visualises this process.
ocean level. In total, the major canal system has 5 outfall locations; two The risk calculation process as described is carried out for three
governed by the Kelani River level (North Lock and Ambathale) and different flood event scenarios. These scenarios differ in whether or not
three towards the ocean (Mutwal Tunnel, Dehiwala canal and adaptation measures are implemented and in the degree of wetland
Wellawatta canal). When the Kelani River reaches the level of 1.5 m encroachment. Table 1 shows the three scenarios considered: no mea-
above mean ocean level, the North Lock gate is closed. By doing so, the sures and no more additional wetland encroachment (scenario 1); no
canal system loses approximately 30% of its total outfall capacity. measures and full wetland encroachment (scenario 3); and the struc-
Flooding can occur due to high rainfall intensities in the Colombo tural measures package (scenario 2). The difference between scenario 1
basin as well as high Kelani River discharges which then flood the and 2 is used to carry out the cost benefit analysis for the measures
Kolonnawa area. Flooding due to the first situation may be aggravated package. The difference between scenario 1 and 3 is used to calculate
during high(er) Kelani River levels and high or neap tide, both resulting the flooding costs when wetland encroachment is not stopped.
in decreased outflow capacity; or improved during low tide. Fig. 1 No calculation was carried out for the combination of wetland en-
shows the study area, its canal and river system and the major outfall croachment and measures, as the aim of the authorities is to stop
locations. wetland encroachment. The goal of scenario 3 is to calculate the ben-
Recent severe flooding of the Kolonnawa basin due to high Kelani efits of this policy, assuming the policy is carried out successfully.
River discharges occurred in 1998, 2016 and 2017, while in June 1992
and November 2010 torrential rain flooded Colombo with 494 mm in
19 h and 440 mm in 16 h respectively. In May 2010, a multi-day storm 3.2. Hydrodynamic model
event dropped 616 mm in 9 days on the city, though the daily rainfall
did not exceed 155 mm/day (approximately average annual maxima). A MIKE Flood model was applied in this study [18], consisting of a
These floods cost Colombo millions of dollars in economic losses due to hydrological module, a one-dimensional module for the major drainage
business interruption, in addition to severe damages afflicted on public canals and a two-dimensional module to represent flow and storage on
and private property [16]. In May 2016, operators were forced to close land. In the Kolonnawa basin area, the rainfall-runoff processes were
the North Lock gate due to the high Kelani River levels. Hence, the 10-
year rainfall event that occurred in that period (256 mm in a day) Table 1
needed to be stored in Colombo's canals and wetlands as neap tide Evaluated scenarios.
decreased the remaining outfall capacity even further. The heavy rains No more additional wetland Full wetland
have resulted in, for example, landslides in Kegalle and flooding in encroachment encroachment
Colombo, displacing more than 300,000 across the island with at least
No measures Scenario 1 Scenario 3
58 left dead and a further 130 missing [17].
Measures package Scenario 2 –

3
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

schematized using a Unit Hydrograph method since no observations for Nagalagama. This also affects the outflow at North Lock, as outflow
water level or discharges are available for this catchment. The Soil remained possible in the simulation, while this flow was in fact highly
Conservation Service-Curve Number (SCS-CN) method was applied to reduced due to the high Kelani River levels. This affects all water levels
estimate the surface runoff. The hydrology in the Colombo basin was in the simulations, resulting in an underestimation. The November
schematized by the NAM method, a lumped and conceptual catchment 2010 event was primarily driven by rainfall, showing acceptable model
runoff model. The rainfall recordings at Colombo Meteorological sta- performance even at the outfall locations near the ocean although using
tion were used for the entire study area as this is the only station with the Colombo Port tidal time series. The model performed reasonably
15-min time series. The output from the hydrological models (NAM and well for the May 2016 event; it underestimates the peak flood level in
SCS-CN), i.e. the discharge from each sub-catchment, was used as an the canal system by approximately 0.2 m; partly driven by the differ-
input for the one-dimensional hydrodynamic model. This model in- ence in observation and simulation at the Wellawatta outfall location.
cludes the primary drainage canals in the Colombo basin, reservoir/ After this model validation, the flood model was run using the
tank systems, and structures. In the Kolonnawa basin, secondary type stochastic variables, i.e. the synthetic boundary conditions and storm
canals which are draining to the Kelani River were also included. The events, derived for the probabilistic hazard assessment. The runtime of
Kelani River was schematized from the upstream Hanwella discharge a single simulation with these stochastic variables was approximately
monitoring station to the river mouth ending in the Indian Ocean. The 7 h, using an Intel ® Core ™ i7-6700 processor with 3.40 GHz CPU and
Indian Ocean tide level at Colombo Port gauge station was used as a 8 GB RAM. In the supplementary material there is also a validation
boundary condition at the outfall locations of Wellawatte, Dehiwala, where the flood of May 2016 was simulated and compared with the
Mutwal tunnel and the Kelani River mouth. The discharge at Hanwella observed flood extent. This shows a good match between the model and
gauge was applied as upstream inlet of the Kelani River. The one-di- the observations.
mensional part was connected to the two-dimensional hydrodynamic
flood model. A 1 × 1 m horizontal resolution lidar dataset is the basis 3.3. Probabilistic hazard assessment
for the elevation model. Sensitivity analysis was carried out with re-
spect to the grid extent and grid spacing to arrive at acceptable com- The main objective of the probabilistic hazard assessment is to de-
putational run times in view of the probabilistic analysis. This resulted rive maps of inundation depths for different return periods, which serve
in a 30 × 30 m grid, approx. 198.000 cells, and 76.000 active grid cells. as input for the damage model. The principal approach is to define the
Flood plain features such as bunds, embankments and elevated road range of potential (synthetic) events with known probabilities that may
spillways were assessed on their impact on the flood pattern and in- cause floods and then to subsequently [a] simulate these events with
cluded in the terrain model when deemed necessary. The existing land the hydrologic/hydrodynamic model to obtain the inundation depths in
use map was used to derive 7 categories of roughness coefficients. The the project area and [b] derive the probability of occurrence of each
resulting simulated maximum inundation depths are available at the event. Based on the combined information of [a] and [b], the prob-
same spatial grid extent; 30 × 30 m. This was the highest resolution for abilities of inundation depths in the area can be determined by using an
which the computation time remained feasible within the project. For appropriate probabilistic computation technique. This procedure is
the damage assessment these results were downscaled to a resolution of carried out for each scenario displayed in Table 1.
5 × 5 m. To define the synthetic events to be simulated in step [a], all re-
Due to the absence of discharge observations in the canal system, levant factors contributing to floods need to be taken into account.
separate calibration of the hydrological model and the hydrodynamic Floods in the Colombo Metropolitan Region can be caused by a com-
model is not possible. Additionally, no flood extent maps of historical bination of the following factors:
events in the Colombo basin are available. Water level data of flood
events are only available at a limited number of staff gauges in 1 High rainfall intensities in the Colombo Metropolitan Region;
Colombo's major canal system; however, these are infrequently mon- 2 High ocean water levels (tidal and storm surge effects taken to-
itored, and measurement locations have changed. Therefore, the joint gether), potentially causing backwater effects and a decrease in
hydrological and hydrodynamic model chain (i.e. the flood model) was drainage capacity;
simultaneously validated for its performance during three recent flood 3 High Kelani River discharges, potentially causing a decrease in
events: May 2010, November 2010 and May 2016. drainage capacity from the canal drainage system to the Kelani
During the 9-day period of 13–21 May 2010 pluvial flooding oc- River and also direct flooding from the Kelani river into the flood-
curred when in total 616 mm of rainfall was recorded. The daily value plains; and
never exceeded 155 mm/day nor did the hourly rainfall exceed 56 mm/ 4 Wet initial conditions in the soil and canal system, causing a de-
h; the latter two are average annual maxima. On 10–11 November crease in initial rainfall losses.
2010, a low-pressure system developed over Colombo overnight and in
a period of 16 h, a storm event of 440 mm was recorded, with hourly For each of these four factors a stochastic variable is defined in the
values up to 123 mm/h (50 year return period). This resulted in probabilistic hazard model. Statistical analyses of the stochastic vari-
widespread local pluvial flooding. The 5–16 May 2016 flood event is ables were carried out in order to determine the probabilities of oc-
characterized by heavy local rainfall of 257 mm/day and high Kelani currence of the simulated synthetic events (step [b] above). For rainfall,
River discharge resulting in fluvial flooding of the Kolonnawa basin. intensity-duration frequency (IDF-) curves were derived by fitting ex-
Due to this high Kelani River level the canal outfalls to the Kelani River treme value distributions to annual maximum and Peaks-Over-
were closed and no runoff in the Colombo basin could be discharged Threshold series for durations ranging from 15 min to 24 h. Synthetic
into the Kelani River. Combined with the heavy rainfall this resulted in rainfall hyetographs were derived from the IDF curves using the
pluvial flooding in the Colombo basin. ‘Chicago storm method” [19]. An area reduction factor of 0.9 was ap-
Only 4 staff gauges in the major canal system of Colombo have plied to account for the fact that rainfall is not equally extreme all over
(limited) recordings of all 3 events. In addition to these staff gauges, the the drainage area. More about this area reduction factor can be found in
Kelani River level is monitored hourly at Nagalagama gauge station. the supplementary material.
Fig. 3 shows the comparison between the observed and simulated water For the Kelani River discharge, an extreme value distribution was
levels at these gauge stations. It clearly shows a general under- derived from observed annual maxima at Hanwella gauge station. A
estimation of the flood levels during the May 2010 flood event. The representative unit hydrograph was subsequently derived from ob-
high discharge in the Kelani River was not recorded properly in the served hydrographs of the 35 events with the highest peak discharges.
available time series which affects the performance of the model at Additionally, the correlation between peak rainfall and peak discharge

4
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

Fig. 3. Comparison of observed (in red) and simulated (in blue) water levels for 3 flood events at five locations (shown in Fig. 1).

was modelled with a Gumbel-Hougaard copula-function [20]. Statistics simulated, 96 model simulations (2*4*6*2) would be required. How-
of peak ocean water levels were derived from time series at Colombo ever, 8 of these combinations do not cause flooding, i.e. when both the
Harbour station. This is the combined effect of the tidal level and the rainfall depth (20 mm) and river discharge (< =800 m3/s) are too low.
storm surge level. Theoretically it would have been better to separate These simulations are therefore not carried out, which leaves 88 model
these effects, but since storm surges at Colombo are small (a few simulations.
decimetres) the differences in results of the two approaches will be At each location in the study area, i.e. all grid cells of the two-di-
negligible. For initial conditions, it was found that the probability of mensional model, inundation depths are computed that correspond to a
relatively wet and relatively dry conditions were approximately equal set of return periods. For example, at each location the inundation
to 0.5. depth with an annual exceedance probability of 1/50 is computed. The
The probabilistic model computes the probability of exceedence of a combined results for all locations together form the 50-year flood map.
range of threshold inundation depths at each location in the project This map serves as input for the damage assessment. In this way, return
area. For this purpose, the probabilistic computation technique “nu- period inundation depth maps are made for 2, 5, 10, 20, 50 and 100
merical integration” is used. This method is for example applied in the year return periods.
statutory safety assessment procedure of flood defences in the More information about the probabilistic hazard assessment can be
Netherlands to determine exceedance probabilities of water levels [21]. found in the supplementary material. This includes also the derived
With this method, the set of potential events is discretized into an n- exceedance probabilities for the different stochastic variables.
dimensional computational grid, where n is the number of stochastic
variables (in this case, n = 4). The inundation depth at each location is 3.4. Flood damage assessment
evaluated with the hydrodynamic model for each n-dimensional grid
cell. Ideally, all possible combinations of realisations of the stochastic The flood damage model combines information on flood depth,
variables are simulated, but this requires an infinite number of simu- exposed objects and economic activities, and vulnerability functions
lations. The main challenge is therefore to find the right balance, i.e., to (see Fig. 4). The exposure maps contain elements at risk for 56 different
minimize the number of model simulations and at the same time not to damage categories identified by local stakeholders during several
exclude the events and processes that are relevant to high water levels. roundtable meetings. This includes combinations of all relevant
Table 2 shows the selected values of the stochastic variables for the building purposes, heights and building quality (normal or shanty
model simulations. If all the combinations of these realisations are building). Each damage category has a monetary maximum damage
value per element and a damage function assigned to it. A damage
function describes the relationship between inundation depth and da-
Table 2 mage fraction of the element at risk. The approach is described in a
Selected values of stochastic variables for the flood model simulations. mathematical form in equation (1) (Egorova et al., 2008).
Peak tide (m + MSL) Peak discharge 24 h rain (mm)a Initial conditions m n
(m3/s) Damage = ∑ si ∑ fij (dj) nij
i=1 j=1 (1)
0.4 300 20 wet
0.6 800 155 (T = 2) dry Equation (1) sums the damage for all damage categories m and all
1500 255 (T = 10) locations (grid cells) n. The damage for a single grid cell is a multi-
2500 370 (T = 25)
plication of the damage fraction fij (dj ) , the elements at risk nij and the
480 (T = 50)
635 (T = 100) value of risk of the different damage categories si . The damage fraction
originates from a damage function dependent on the inundation depth.
a
Values refer to 24-h point rainfall. The calculations are carried out on a 5 × 5 m resolution with the

5
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

Fig. 4. Visualization of the flood damage calculation procedure (also part of Fig. 2).

software Delft-FIAT [22]. methods to create purely data-driven damage functions taking into
This approach is considered the general procedure for flood damage account more variables than only the water depth (e.g. Refs. [28,29].
assessment [23]. The exposure data (elements at risk) is based on These methods could not be utilized in this study because they require a
building footprints for all buildings, land-use maps for agriculture, and large dataset of past flood damage records which is not available for the
coordinates of critical infrastructure objects. Vehicles, i.e. cars, trucks, study region. This is currently a common problem with these ap-
vans and motorcycles, are also included in this study. However, since proaches and they are therefore presently rarely utilized.
the location of a vehicle during a flood is unknown, the total number of In this study the indirect damages, business interruption and in-
vehicles in the city is equally distributed over the study area. The tangible damage are taken into account with a multiplication factor.
probability that someone drives their car to safety is included in the This multiplication factor is based on observations from previous floods
value at risk figure for the vehicles. This is estimated based on con- globally. Vilier [30] compared for a number of large-scale previous
versations with local car insurance companies [24]. flood events the observed business interruption and indirect damages as
The assessment of the values at risk and the subsequent develop- share of the total damage. From the data in Vilier [30] it can be con-
ment of the damage functions is further described in Dias et al. [24]; the cluded that indirect damage plus business interruption is usually
results of this assessment are shown in Fig. 5 and Table 3. The damage around 5–30% while some extreme events have a share higher than
functions and values at risk are based on household surveys, expert 100%. These extreme events are usually rare flood events that had a
consultations, bills of quantities and interviews with infrastructure large impact on high-tech industry, for instance the large floods in
agencies and insurance companies. The damage functions for vehicles Thailand of 2011 [30]. Most of the EAD calculated in this study origi-
were taken from Scawthorn et al. [6]. The damage functions only de- nates from frequent flood events and therefore the business interrup-
pend on the inundation depth. Flow velocity and flood duration are not tion/indirect damage share of the damage is expected to be low. Vilier
taken into account. This is a common assumption in flood damage [30] has not included intangible damage in his multiplication factor.
models [23]. Other influencing variables are already implied in simple Given the relatively small inundation depths and the relative flatness of
depth-damage curves [25] and are mostly important to model explicitly the terrain, mortalities within the Colombo metropolitan area are un-
in case of a lot of local variation or when transferring damage functions likely and affected people are the only intangible damage that is in-
that were developed for another location [26]. In this situation we also cluded for the cost-benefit analysis in this study. Based on these con-
expect that in general other hazard variables are less important for siderations the share of indirect losses is estimated to be between 10
damage on permanent buildings. For the temporary buildings, some and 50%, and therefore a multiplication factor of 1.1 and 1.5 has been
effects of such other hazard variables are implied in the damage curve. applied. The cost-benefit analysis is carried out with both values which
The damage function for the flood protection bunds is an exception; should be seen as a bandwidth of the result.
this only depends on the water level relative to mean ocean level at a
particular location: Nagalagama gauge station. The content damage for 3.5. Cost benefit analysis
semi-permanent (shanty) residential buildings is considered negligible.
At the time of the study 1 USD was equal to 150 Rs [24]. The purpose of a cost benefit analysis is to compare the cost of a
These damage functions and values at risk can be seen as im- measure with the reduction of future expected flood damages that this
provements compared to other studies (e.g. Ref. [27] since (i) actual measure is expected to result in. For this the damage at different return
building footprints and building costs were used rather than smearing periods and with different future scenario's is compared with and
them on land use areas; (ii) inundation depths up to 10 m were con- without measure. Our cost benefit analysis doesn't adjust for poverty,
sidered (although the flood depths did not exceed 6 m) – damage values this may lead to too little protection for people areas [31]. A more
at higher return periods will be underestimated if insufficient depths thorough analysis of this is provided in the discussion.
are used for the damage curves; and (iii) separate functions and values The first step in carrying out a cost benefit analysis is to calculate
were used for the structure and contents of the buildings – 3 categories the current Expected Annual Damage for the 4 scenarios (see Table 1).
were used for structure type and 7 for function (hence contents) type The EAD is calculated by combining the different damages at the dif-
(see Table 3; also [24]. ferent return periods. This is done with a convolution integral of the
The damage functions in this study are synthetic; there are also damage with respect of the exceedance probability.

6
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

Fig. 5. The damage functions for the categories with the largest damage shares. Left top, the building structure functions. Right top, the building content functions.
Right bottom, the temporary buildings function and left bottom, the vehicle functions [24].

Table 3
The 56 damage categories included in this study and their value at risk; for more information see Dias et al. [24].
Damage category Value at risk (kRs) Unit Damage category Value at risk (kRs) Unit

Residential - temporary - structure 6 m2 Schools - single storey - structure 30 m2

Residential - temporary - content 0 m2 Schools - single storey - content 6 m2
Residential - single storey - structure 30 m2 Schools - 2 storey - structure 60 m2
Residential - single storey - content 7.5 m2 Schools - 2 storey - content 12 m2
Residential - 2 storey - structure 60 m2 Hospitals - single storey - structure 30 m2
Residential - 2 storey - content 15 m2 Hospitals - single storey - content 25 m2
Warehouse - single storey - structure 30 m2 Hospitals - 2 storey - structure 60 m2
Warehouse - single storey - content 50 m2 Hospitals - 2 storey - content 50 m2
Warehouse - 2 storey - structure 60 m2 Telecom - access node 8350 #
Warehouse - 2 storey - content 100 m2 CEB Transformers 1000 #
Industrial - single storey - structure 30 m2 CEB Gantries 1000 #
Industrial - single storey - content 40 m2 Electricity grid substation 150,000 #
Industrial - 2 storey - structure 60 m2 Electricity primary substation 15,000 #
Industrial - 2 storey - content 80 m2 Telecom - exchange 300,000 #
Industrial - shanty - structure 6 m2 RDA roads 20 m
Industry - shanty - content 40 m2 NWSDB – meters & pipes 0.045 m2
Offices -shanty - structure 6 m2 CEB – electricity meters 0.03 m2
Offices - shanty - content 15 m2 Trucks 0.150 m2
Offices - single storey - structure 30 m2 Vans 0.175 m2
Offices - single storey - content 15 m2 Cars 0.175 m2
Offices - 2 storey - structure 60 m2 Trishaws 0.030 m2
Offices - 2 storey - content 30 m2 Motor cycles 0.025 m2
Small shops - shanty - structure 6 m2 CEB ring, radial & satellite substations 3000 m2
Small shops - shanty - content 40 m2 Cricket fields 0.041 m2
Small shops - single storey - structure 30 m2 Forest ecosystems 0.05 m2
Small shops - single storey - content 40 m2 Home garden 1.92 m2
Small shops - 2 storey - structure 60 m2 Agriculture Fields 2.95 m2
Small shops - 2 storey - content 80 m2 Rice 0.012 m2
Flood protection bunds 8545500 All bunds

7
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

pmax
The NPV in equation (3) (e.g. Ref. [32] is the Net Present Value of a
EAD = ∫ Damage (p) dp
measure. It is the present value of all future costs and benefits of the
0 (2)
measure. EADn, ref is the EAD in year n if the measure isn't carried in US
In equation (2) the EAD is in $ per year, p is the exceedance dollars. EADn, measure is the EAD in year n when the measure is carried
probability (1/y), pmax is the largest exceedance probability for which out in US dollars. Both EADn, ref and EADn, measure grow every year as the
damage is to be expected and damage (p) is the damage as function of city gets richer according to the different economic growth scenarios.
the exceedance probability. This function cannot be solved analytically cost0 are the initial costs of the measure. Mn are the maintenance cost in
and is represented by the different return periods for which the damage year n. In this case, n is the year in the calculation and r is the discount
is calculated. This integral can then be solved numerically. rate. When the NPV is zero, the discount r equals the IRR. This IRR is
The future EAD damage for the 3 different scenarios is projected to then applied to compare different projects with each other (a higher
account for future changes in economic values. This is important be- discount rate indicates a better investment). A high discount rate means
cause if the city becomes a wealthier in the future that would justify that future benefits of the measure have a low importance in the pre-
more expensive measures right now. For this purpose, 5 different sent. A high IRR therefore means that even if the future benefits have a
Shared Socioeconomic Pathway (SSP1-SSP5) or real economic growth low weight compared to the present measure cost the measure is still
projections for Sri Lanka are applied (O'Neill et al., 2017). In this case cost effective and it therefore is a good measure.
the real economic growth is applied per capita, so population growth is
neglected. This is due to the fact that limited space is available for new
buildings inside the study area, hence population growth may most 3.6. Wetland encroachment cost analysis
likely occur by replacing the current building stock with high rise
buildings and/or with increased wetland encroachment. High rise For the wetland encroachment scenario, we analyse the cost for a
buildings are not expected to be more at risk to floods than the current situation in which all current wetlands have disappeared. This would
buildings because the higher floors stay dry. Wetland encroachment is mean a reduction of 13.1 km2 of wetland area and 3.6 km2 of wetland
assumed to stop (except in scenario 3 in which it is specifically in- water bodies. The resulting flood damage for this scenario is then cal-
cluded). Therefore in scenario 1–2 the population growth will have no culated. A simplified damage model is developed to estimate the da-
impact on the flood risk, and the future EAD can be estimated only mage in the wetland areas that are not yet urbanized. A simplified
using the real economic growth per capita. Fig. 6 shows the 5 real approach is necessary because no exact knowledge is available re-
economic growth projections per capita based on O'Neill et al. (2017). garding the exposure in these future wetland areas to be encroached.
Assessment of a far-future EAD is less relevant than a near-future This simplified model applied damages per m2 urbanized area for dif-
EAD because of the time value of money. Therefore, the present value ferent inundation depths and is based on what the detailed damage
needs to be calculated for each future EAD before all future EAD can be model for Colombo calculates on average per m2 in the areas recently
summarized. To accomplish this, a discount rate is used: a variable that encroached. The maximum damage value in the simplified model is 55
determines how important the future is compared to today (a higher US dollar per m2 urbanized area. The damage function for the simpli-
discount rate means the future is less important). It is however chal- fied model is shown in Fig. 7.
lenging to determine this discount rate for Sri Lanka, because no Full wetland encroachment will only be reached at some future
standard discount rate is set by the Sri Lankan government. Therefore, date. The EAD increase due to the encroachment is considered linear
the discount rate is calculated in such a way that the costs and the between now and that future moment. To this end, the years 2030,
benefits are equal. This is called the Internal Rate of Return (IRR) - see 2050 and 2070 are taken as the moments that all wetlands have dis-
equation (3) (e.g. Ref. [32]. A higher internal rate of return means that appeared. The present value of the extra cost is calculated with three
the measure is more valuable. A cost-benefit analysis is not used for different discount rates: 5, 8 and 12%. This range of discount rates is
scenario 3, and only benefits are calculated; therefore we worked picked based on expert judgement and is supposed to cover a wide
without the IRR in that case. range. Together this provides an overview of the possible costs of in-
N N action on wetland encroachment. The calculation of present value fol-
EADn, ref − EADn, measure ⎛ Mn ⎞
NPV = ∑ − ⎜cost0 + ∑ lows the same approach as is used in equation (3).
(1 + r )n (1 + r )n ⎟ (3)
n=0 ⎝ n=0 ⎠

Fig. 6. Five real expected economic growth per ca-

pita projections, derived from O'Neill et al. [35]. The
names of the scenarios are: SSP1: Sustainability:
Taking the green road, SSP2: Middle of the road,
SSP3: Regional rivalry: A Rocky Road, SSP4: In-
equality; A road divided and SSP5: Fossil-fuelled
development: Taking the high road.

8
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

Fig. 7. Damage function for simplified damage model for encroached wetland areas. The maximum damage is 55 US dollar per m2 urbanized area.

3.7. Adaptation measures damage functions. The spatially distributed expected annual damage
for the study area is shown in Fig. 9. This study finds that the expected
The main causes for the flood challenges which the Metro Colombo annual flood damage for Metro Colombo is in the order of 13 M$/year.
faces are related to i) a lack of outfall capacity, ii) a lack of storage
capacity and iii) a reduced conveyance in the canal system. A set of 4.2. Impact of adaption measures
adaptation measures is developed to adapt to these challenges. This is
combined into one package. The package consists of 6 adaptation 4.2.1. Flood risk reduction: current situation
measures. The measures consist of widening of locks, a tunnel to dis- The intervention package causes a reduction in the EAD (see
charge water, a new pumping station and 3 diversions. All measures Table 5). They reduce the flood risk problem within the Colombo me-
focus solely on reducing flood risks in the Colombo metropolitan area tropolitan area by about 26%. The measures focus solely on the flood
and not on the Kolonnawa area which also has flood risk problems. The occurrence in the Colombo basin, and have little influence on the flood
Sri Lanka Land Reclamation and Development Cooperation (SLLRDC) hazards in the Kolonnawa basin. However, to investigate whether the
estimated the cost of the measures, see Table 4. measures justify the costs, it is important to look at the cost-benefit
analysis for all basins, and in this case at the IRR.
4. Results
4.2.2. Flood risk reduction: future situation
4.1. Flood hazard and risk The calculation of the IRR also includes the costs of the adaptation
measures and the future projected growth. The IRR is calculated twice
The flood model is simulated with all 88 combinations of stochastic with different assumptions about the indirect/business interruption and
variables. This resulted in 88 flood maps and a flood probability for intangible damage multiplication factor. The values reported in Table 6
each 30 × 30 m grid cell. Fig. 8 shows the resulting composite flood include a lower and a higher estimate, the lower estimate uses a 10%
map for these return periods. This analysis shows that every 10 years an multiplication factor and the upper estimate a 50% multiplication
area of 27 km2 is flooded within Metro Colombo. The flood damage factor.
corresponding to these inundation depths are calculated using the The IRR for the intervention package is between 6 and 10%. These

Table 4
Cost of the measures as estimated by the local government.
No. Measures Cost estimation (M USD)

1 North Lock Widening and Pumping Station 15

Installing six pumps with a capacity of 5 m3/s each which will operate once the Kelani river water level is higher than the Metro Colombo canal
water level.
2 New Mutwal Tunnel 36
Because of the poor outfall capacity of the existing Mutwal Tunnel, a new tunnel will be introduced with a 3 m diameter, 735 m length resulting
14.75 m3/s additional discharge capacity.
3 Kolonnawa Canal Diversion 20
Excess flood water of the Colombo catchment will be discharged via the Kolonnawa catchment during periods of lower water levels in the Kelani
River compared to Colombo catchment. This will increase the outfall capacity by 45 m3/s.
4 St. Sebastian South Pumping Station 10
Installing a new pumping station at St. Sebastian South canal with 10 m3/s capacity. This will be used once the inner canal water level is above
1 m msl.
5 Madiwela East Diversion 4.5
Installing four pumps with capacity of 5 m3/s each which will operate once the Kelani water level is higher than the Madiwela canal water level.
6 Madiwela South Diversion 33
A diversion of flood water coming from the three upper most catchments (12.75 km2) to another catchment nearby. This intervention will
reduce the runoff inflow to Colombo catchment by 50 m3/s.

9
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

Fig. 8. The flood extent at different return periods without any measures. Everything covered by a more frequent return period is also covered by any less frequent
return period.

IRRs are moderately high; it is up to decision makers to decide whether the number of combinations of the stochastic variables will determine
it is high enough for the investment. This depends on alternative in- the number of scenarios that are realistic.
vestment opportunities and their respective IRRs. The cost-benefit analysis did not take into account risk aversion,
income distribution and social welfare. Taking this into account would
4.2.3. Flood risk under increased future wetland encroachment ensure that measures are not mostly aimed at wealthier areas. Kind
The total cost (present value of all future flood risk) of inaction on et al. [31] showed that under some conditions these elements are im-
further wetland encroachment or the potential benefits of stopping portant to consider in a cost-benefit analysis. These conditions are the
wetland encroachment are shown in Table 7. These costs depend on the presence of high social vulnerability, incomplete damage compensa-
speed of the wetland encroachment - i.e. the year by which wetlands tion, and large income differences without a mechanism of redistribu-
have completely disappeared, the discount rate and the economic tion [31]. These conditions apply to some extent in Colombo and
growth projection. The total present value of the cost is between a therefore it may be important that future flood risk management studies
hundred million US dollar and three billion US dollar. Especially the in Colombo take this into account. Income data is required to take into
speed of the encroachment and the discount rate are very sensitive in account differences between income groups in the damage model and
this calculation. to derive correction factors. This income data was not available and
hence has not been taken into account in the present study.
5. Discussion Even though socio-economic changes and land-use changes (wet-
land encroachment) are taken into account, climate change is not. This
In the current study, only one measures package is considered. This will underestimate the future flood risk and therefore the benefits of the
package consists of multiple measures. Some of these measures could be measures are likely to be higher. However, it is expected that this will
more efficient than others, implying that there are possibly better only have a minor impact on the results. Due to the transient nature of
configurations of the measures package by only focusing on the most most climate change it is relatively far in the future, while the main
efficient measures. This is however difficult in practice because it would purpose of the adaptation measures is to solve a problem that is already
require many more time consuming hydrodynamic calculations for urgent today. Given the discount rates applied in the cost-benefit ana-
each individual measure. The set of 88 simulations which form the basis lysis, the short-term benefits weigh most heavily in the analysis and
of the flood return maps took over 600 h (25 days) of run time. This set therefore an underestimation of the flood risk further in the future will
of simulations is required for each measure, if assessed individually. likely only have a limited effect on the analysis. Climate change is a
With 6 measures and several potential combinations of measures, it is more important factor for areas that have less urgent problems, or when
therefore important to strike a balance between the detail of the cal- a lower discount rate is applied. Therefore, it might be useful to take
culations and the number of scenarios to be analysed. For the compu- climate change into account for similar studies in the future. This could
tation time especially, the resolution of the hydrodynamic model and be included in the probabilistic model by changing the probabilities of

10
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

Fig. 9. Flood risk map without any measures.

extreme conditions for future studies. Table 6

The current hydrodynamic model runs on a 30 × 30 m grid. A IRR based on the economic growth projection and the multiplication factor for
higher grid resolution could improve the results and typically leads to indirect/business interruption and intangible damage.
smaller flood extents [33]. In this study it wasn't feasible to calculate on Economic growth projection
a higher resolution grid because it would lead to too much computation
time. The effect of this relatively coarse grid is expected to be limited SSP1 SSP2 SSP3 SSP4 SSP5
because in this study mostly larger connected flood areas were expected
Measures package 7.8–9.5% 7.1–8.8% 6.0–7.7% 6.9–8.6% 8.4–10.1%
in the relatively flat area. In such cases a higher resolution would be less
important ( [33].
other studies that look at the entire flood risk chain such as Apel et al.
6. Conclusion [13] or Metin et al. [15] which do not include a detailed economic
analysis.
In this study we show how to combine several different models and A second analysis is performed to determine the benefits of stopping
techniques to determine the feasibility of investments for reducing wetland encroachment. This is done by calculating the present value of
flood risks. We integrated a state-of-the-art probabilistic model with all future expected flood damage with and without wetland encroach-
detailed hydrodynamic and flood damage models. This chain of models ment. This analysis shows that stopping wetland encroachment could
is used to calculate the reduction in expected annual damage to be save the Colombo metropolitan area between 100 million and 3 billion
expected if a set of measures to the hydraulic system is undertaken. An US dollars in future flood damages. This result is very sensitive to the
economic analysis is then added on top of this to calculate the Internal discount rate applied because most of this future flood damage would
Rate of Return (IRR) of the proposed measures. This combination of be far in the future. It is however clear that the effect of stopping
models is able to give valuable insights into the benefits of the measures wetland encroachment is much larger than the effect of the structural
and helps decision makers compare the investment in the measures adaptation measures.
with other possible investments. This approach goes a step further than

Table 5
The EAD with and without the measures (excluding multiplication factor for business interruption/indirect damage).
No measures (scenario 1) Intervention package (scenario 2)

Expected Annual Damage (EAD) [M$/y] 13.1 9.7

Reduction Expected Annual Damage (EAD) [M$/y] – −3.4
Reduction Expected Annual Damage (EAD) [%] – −25.9

11
D.J. Wagenaar, et al. International Journal of Disaster Risk Reduction 37 (2019) 101162

Table 7 [12] R.N. Jones, A. Patwardhan, S.J. Cohen, S. Dessai, A. Lammel, R.J. Lempert,
Present value (billion US$) of all extra future expected damages if wetland M.M. Qader Mirza, H. Von Storch, Foundations for decision making, in: C.B. Field,
encroachment continues. et al. (Ed.), Climate Change 2014: Impacts, Adaptation, and Vulnerability.
Contribution of Working Group II to the Fifth Assessment Report of the
Discount rate SSP1 SSP2 SSP3 SSP4 SSP5 Intergovernmental Panel on Climate Change, Cambridge University Press,
Cambridge, UK, and New York, NY, USA, 2014, pp. 195–228 (Chapter 2).
Full wetland encroachment reached in 2030 [13] H. Apel, A. Thieken, B. Merz, G. Bloschl, A probabilistic modelling system for as-
5% 2.41 1.93 1.33 1.80 2.99 sessing flood risks, Nat. Hazards 38 (2006) (2006) 79–100, https://doi.org/10.
1007/s11069-005-8603-7.
8% 1.00 0.83 0.65 0.81 1.17
[14] A. Chen, M. Hammond, S. Djordjevic, D. Butler, D. Khan, W. Veerbeek, From hazard
12% 0.44 0.39 0.33 0.38 0.49
to impact: flood damage assessment tools for mega cities, Nat. Hazards 82 (2016)
(2016) 857–890, https://doi.org/10.1007/s11069-016-2223-2.
Full wetland encroachment reached in 2050 [15] A.D. Metin, H. Apel, N.V. Dung, B. Guse, H. Kreibich, K. Schröter, S. Vorogushyn,
Discount rate SSP1 SSP2 SSP3 SSP4 SSP5 B. Merz, Effects of Changes along the Risk Chain on Flood Risk. EGU General
5% 1.95 1.53 0.97 1.39 2.49 Assembly 2017, (2017).
8% 0.70 0.57 0.41 0.53 0.84 [16] GFDRR, Transforming Colombo into a flood resilient Middle income country ca-
12% 0.26 0.22 0.18 0.21 0.29 pital, Proj. Highlights (4) (2013) 4 March 2013.
[17] GEOSRILANKA, Colombo Floods, (2016) https://geosrilanka.wordpress.com/
Full wetland encroachment reached in 2070 2016/05/18/colombo-floods-2016/.
Discount rate SSP1 SSP2 SSP3 SSP4 SSP5 [18] DHI, MIKE Flood. 1D-2D Modelling User Guide, (2017).
5% 1.54 1.19 0.71 1.07 2.00 [19] J.C. Keifer, H.H. Chu, Synthetic storm patterns for drainage design, J. Hydraul. Div.
8% 0.50 0.40 0.28 0.37 0.61 83 (1957).
12% 0.17 0.15 0.11 0.14 0.20 [20] P. Hougaard, A class of multivariate failure time distributions, Biometrika 73 (3)
(1986) 671–678 1986.
[21] C.P.M. Geerse, Probabilistic model to assess dike heights in part of The Netherlands,
International Symposium on Stochastic Hydraulics, Nijmegen the Netherlands,
Appendix A. Supplementary data 2005.
[22] K. Slager, A. Burzel, E. Bos, K. de Bruijn, D. Wagenaar, H. Winsemius, L. Bouwer,
M. van der Doef, User Manual Delft-FIAT Version 1, (2016) website: publicwiki.
Supplementary data to this article can be found online at https:// deltares.nl/display/DFIAT/Delft-FIAT+Home.
doi.org/10.1016/j.ijdrr.2019.101162. [23] B. Merz, H. Kreibich, R. Schwarze, A. Thieken, Review article “Assessment of
economic flood damage”, Nat. Hazards Earth Syst. Sci. 10 (2010) 1697–1724,
https://doi.org/10.5194/nhess-10-1697-2010.
References [24] P. Dias, N.M.S.I. Arambepola, K. Weerasinghe, K.D.N. Weerasinghe, D. Wagenaar,
L. Bouwer, H. Gehrels, Development of damage functions for flood risk assessment
[1] J.C.J.H. Aerts, W.J.W. Botzen, K. Emanuel, N. Lin, H. de Moel, Evaluating flood in the city of Colombo (Sri Lanka), Procedia Engineering 212 (2018) 332–339
resilience strategies for coastal megacities, Science 344 (6183) (2014) 473–475. https://dx.doi.org/10.1016/j.proeng.2018.01.043.
[2] R.J. Nicholls, S. Hanson, C. Herweijer, N. Patmore, S. Hallegatte, J. Corfee-Morlot, [25] D.J. Wagenaar, K.M. De Bruijn, L.M. Bouwer, H. De Moel, Uncertainty in flood
J. Château, R. Muir-Wood, Ranking Port Cities with High Exposure and damage estimates and its potential effect on investment decisions, Nat. Hazards
Vulnerability to Climate Extremes: Exposure Estimates Paris, France. Organisation Earth Syst. Sci. 16 (1) (2016) 1–14 2016.
for Economic and Co-operative Development (OECD), (2007), p. 62 (Environment [26] D. Wagenaar, S. Lüdtke, K. Schröter, L.M. Bouwer, H. Kreibich, Regional and
Directorate Working Papers, ENV/WKP(2007)1). temporal transferability of multivariable flood damage models, Water Resour. Res.
[3] F. Den Heijer, F.L.M. Diermanse, Towards risk-based assessment of flood defences 54 (5) (2018) 3688–3703 https://doi.org/10.1029/2017WR022233.
in The Netherlands: an operational framework, Comprehensive Flood Risk [27] Y. Budiyono, J. Aerts, J. Brinkman, M.A. Marfai, P. Ward, Flood risk assessment for
Management: Research for Policy and Practice, Proceedings of the Flood Risk 2012 delta mega-cities: a case study of Jakarta, Nat. Hazards 75 (2015) 389–413.
Conference in Rotterdam, November 2012, CRC Press, 9780415621441, 2012, p. [28] B. Merz, H. Kreibich, U. Lall, Multi-variate flood damage assessment: a tree-based
2012. data-mining approach, Nat. Hazards Earth Syst. Sci. 13 (2013) 53–64, https://doi.
[4] J.W. Hall, P.B. Sayers, R.J. Dawson, National-scale assessment of current and future org/10.5194/nhess-13-53-2013.
flood risk in England and Wales, Nat. Hazards 36 (2005) (2005) 147–164. [29] D. Wagenaar, J. de Jong, L.M. Bouwer, Multi-variable flood damage modelling with
[5] S.N. Jonkman, M. Bockarjova, M. Kok, P. Bernardini, Integrated hydrodynamic and limited data using supervised learning approaches, Nat. Hazards Earth Syst. Sci. 17
economic modelling of flood damage in The Netherlands, Ecol. Econ. 66 (2008) (2017) 1683–1696 2017 https://doi.org/10.5194/nhess-17-1683-2017.
77–90. [30] J.J.T.A. Vilier, Assessment of the Costs of Business Interruption Due to Large Scale
[6] C. Scawthorn, P. Flores, N. Blais, H. Seligson, E. Tate, S. Chang, E. Mifflin, Floods: A Comparison of Methods, Master Thesis Delft University of Technology –
W. Thomas, J. Murphy, C. Jones, M. Lawrence, HAZUS-MH flood loss estimation HKV, 2013.
methodology II, Damage and loss assessment, Natural Hazards Review 7 (2006) [31] J. Kind, W.J. Botzen, C.J.H. Aerts, Accounting for risk aversion, income distribution
72–81 2006. and social welfare in cost-benefit analysis for flood risk management, WIREs Clim
[7] T.M. Porter, Trust in Numbers: the Pursuit of Objectivity in Science and Public Life, Change 2016 (2016), https://doi.org/10.1002/wcc.446.
Princeton University Press, Princeton, NJ, 1995, p. 310. [32] M.A. Mian, Project Economics and Decision Analysis, second ed., PennWell Corp,
[8] J.M. Kind, Economically efficient flood protection standards for The Netherlands, J. 978-1593702083, 2011.
Flood Risk Manag. 7 (2) (2013) 103–117. [33] Y. Hsu, G. Prinsen, L. Bouaziz, Y. Lun, R. Dahm, An investigation of DEM resolution
[9] F. Messner, E. Penning-Rowsell, C. Green, V. Meyer, S. Tunstall, A. van der Veen, influence on flood inundation simulation, 12th International Conference of
Evaluating Flood Damages: Guidance and Recommendations on Principles and Hydroinformatics, HIC 2016, 2016.
Methods, Floodsite report T09-06-01 (2007). [34] R. Lowe, C. Urich, N. Domingo, O. Mark, A. Deletic, K. Arnbjerg-Nielsen,
[10] E.C. Penning-Rowsell, C. Johnson, S. Tunstall, The Benefits of Flood and Coastal Assessment of urban pluvial flood risk and efficiency of adaptation options through
Risk Management: A Manual of Assessment Techniques, Middlesex University Press, simulations – a new generation of urban planning tools, J. Hydrol. 550 (2016)
London, 2005. 335–367 July 2017.
[11] H. Van der Most, I. Tanczos, K.M. De Bruijn, D.J. Wagenaar, New, Risk-Based [35] O'Neill, et al., The roads ahead: narratives for shared socioeconomic pathways
standards for flood protection in The Netherlands, 6th International Conference on describing world futures in the 21st century, Glob. Environ. Chang. 42 (2017)
Flood Management (ICFM6), September 2014, Sao Paulo, Brazil. 2014, 2014. 169–180 January 2017, Pages.

12