Flood risk analysis using fuzzy models

H.K. Nandalal and U.R. Ratnayake

Department of Civil Engineering, University of Peradeniya, Peradeniya, Sri Lanka

Correspondence Abstract
H.K. Nandalal, Department of Civil
Engineering, University of Peradeniya,
Risk is a combination of the factors that determine vulnerability and exposure
Peradeniya, KY 20400, Sri Lanka potential for people to a hazard. The computation of flood extents and the
Email: hemalin@[Link] identification of vulnerable elements help to determine high-risk zones due to
floods in advance, which helps to take mitigatory measures effectively and
DOI:10.1111/j.1753-318X.2011.01097.x efficiently. This study examines how effectively the risk with respect to floods can
be assessed using a fuzzy approach taking the frequently flooding Kalu-Ganga
Key words River basin in Sri Lanka as the study area. Flood extent for a 100-year return period
Flood modelling; flood risk; fuzzy models; rainfall was determined using Hydrologic Engineering Center’s Hydrologic Mod-
hazard; vulnerability.
elling System (HEC-HMS)- and HEC-River Analysis System-based models. Flood
extent and mean flood depth were taken as hazard indicators while population
density and dependency ratio were used as vulnerability indicators. Based on these
indicators, flood risk was determined for the lowest administrative divisions
within the inundated area using conventional risk assessment approaches. A
methodology was proposed and applied to assess risk assuming the above
indicators as fuzzy variables. Comparison of the results obtained from the two
approaches indicates the proposed fuzzy-based method, which takes uncertainty in
the determination of hazard, vulnerability and risk levels into account, as
providing more accurate results.

modelling methodology that may facilitate an easier transi-

Introduction tion between humans and computers for decision making.
Flood hazard, vulnerability and risk assessment has become Fuzzy methods can also provide a useful tool to handle
a useful tool in modern flood risk management. However, a imprecise and uncertain information. This methodology has
common methodology, to identify and measure the risk and undergone several developments and is currently widely
vulnerability to disasters to determine disaster-risk manage- used in various areas, for example in controllers of washing
ment and disaster-relief priorities, is yet to be developed. machines or video cameras, optical processors, traffic con-
There are several flood risk assessments based on a trol, etc. This approach has been proven to be very useful in
conventional crisp approach using integers or numbers as medical diagnosis (Lascio et al., 2002), information technol-
hazard factors and vulnerability factors, which determine ogy (Lee, 1996), reliability analysis (Sadiq et al., 2004) and in
the level (low, medium or high) of hazard or vulnerability many other applications (Lawry, 2001), where reported data
(Anselmo et al., 1996; Messner and Meyer, 2006; Tingsan- are either qualitative or decision making is performed based
chali and Karim, 2005, 2010; Buchele et al., 2006). Birkmann on expert opinions.
(2007) outlined several approaches that use such factors to The application of fuzzy logic is becoming popular in the
measure the risk and vulnerability to hazards of natural field of water resources management. Esogbue et al. (1992)
origin. Tingsanchali and Karim (2010) attempted to find the modelled flood damage reduction by an appropriate inte-
optimum values that can be assigned to those factors. gration of structural and nonstructural measures with the
However, the level of hazard, vulnerability and risk are objective of optimizing the flood damage reduction due to
difficult to express precisely by a crisp value due to their recurrent floods via fuzzy sets methodologies. Sadiq and
inherent vague nature. Husain (2005) developed a fuzzy-based methodology for
Fuzzy logic emerged as a more general form of logic that estimating the aggregative risk of various scenarios of
can handle uncertainty. Since the pioneering work of Zadeh drilling waste discharge in the marine environment. Chen
(1996) as cited in Ross (2004), fuzzy logic has been used as a and Chang (2010) used fuzzy sets for incorporating


c 2011 The Authors J Flood Risk Management 4 (2011) 128–139
c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
Flood risk analysis 129

objective and subjective uncertainties to address water and the river flows from a height of about 2250 m MSL
resources redistribution alternatives in a transboundary (Figure 1).
channel reservoir system. However, the application of fuzzy Rain that falls on the upper catchment of the river flows
models to evaluate risk in water-related hazards is scarce. through a length of about 36 km and from an elevation of
This study presents an indicator-based approach for flood about 2250 m MSL to 14 m MSL at Ratnapura town. There
risk analysis using fuzzy logic that adequately captures the onwards, the Kalu-Ganga River travels through a compara-
vagueness inherent in such approaches and compares its tively flat terrain of 14 m MSL to sea level through a distance
results with the conventional crisp approach used in flood of about 70 km, collecting a large amount of water flowing
risk analysis. from the other tributaries on the way, making the lower
The Kalu-Ganga River basin in Sri Lanka was selected as reaches of the river prone to frequent floods. For this study,
the study area as flooding in this river basin is the most daily rainfall data at 14 rain gauging stations and daily
frequent natural hazard in the country. Because of its discharge data at three discharge gauging stations (Table 1)
geographical location, the Kalu-Ganga River basin receives displayed in Figure 2 were used.
a large amount of rainfall during monsoon seasons. This
causes flooding almost every year in the Kalutara and
Ratnapura districts. As stated by Churchill and Hutchinson Table 1 Rainfall and discharge gauging stations used
(1984), relief and aid programmes have been essentially the No. Name Longitude Latitude
only collective adjustment to flooding in Sri Lanka, which is
Rainfall gauging stations 1 Horana 80.07 6.75
still in practice. Increase in the human population and 2 Raigama 80.18 6.77
migration to cities have made the flood risk management 3 Horagoda 80.25 6.50
an inevitable step that should be taken to reduce the 4 Morapitiya 80.27 6.53
vulnerability of the human population to frequently occur- 5 Galathura 80.28 6.70
ring floods in Sri Lanka. 6 Gonapanigala 80.33 6.57
7 Pussella 80.35 6.80
8 Kudawa 80.42 6.43
Study area 9 Galaboda 80.47 6.70
10 Lellopitiya 80.50 6.68
The Kalu-Ganga River catchment covers 2658 km2 and is 11 Hapugastenna 80.52 6.72
dominated by forest, residential and agricultural cropland 12 Wellandura 80.57 6.53
land-use types. It experiences an average annual rainfall of 13 Alupolla 80.58 6.72
4000 mm, which varies from 2800 mm in the lower reaches 14 Ratnapura 80.40 6.60
to 5300 mm in higher elevations. Geographically, the catch- Discharge gauging stations 1 Ratnapura 80.40 6.68
2 Ellegawa 80.22 6.73
ment lies between 6.321N and 6.901N latitude and 79.901E
3 Putupaula 80.07 6.61
and 80.751E longitude as per the WGS84 coordinate system

Figure 1 The Kalu-Ganga catchment and the area under investigation.

J Flood Risk Management 4 (2011) 128–139 c 2011 The Authors

c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
130 Nandalal and Ratnayake

Figure 2 Hydrological gauging stations.

There is evidence that this basin has been a habitat of to a rainfall of a 100-year return period and the use of a
humans for as long as 28 000 years. At present, the total flood simulation model to develop the inundation area.
population that occupies the river basin is about 1 036 000
Hydrologic model
and around 12% of this population lives in flood-prone
areas. The population density of the study area is 312 per-
The hydrological model used is Hydrologic Engineering
sons/km2. This river basin attracts population further due to
Center’s Hydrologic Modelling System (HEC-HMS) devel-
fair weather conditions, availability of water, fertile agricul-
oped by US Army corps of Engineers [United States Army
tural land and gem mining industry, making more popula-
Corps of Engineers, Hydrologic Engineering Centre (USACE-
tions vulnerable to frequent flood hazards.
HEC), 2009a]. The HEC-HMS has many well-known and
The Kalu-Ganga River passes through two administrative
well-applicable hydrologic methods to simulate the rainfall–
districts, Ratnapura and Kalutara, as shown in Figure 2. This
runoff process in river basins. The application of HEC-HMS
paper focuses on analysing floods in the upper reaches of the
for modelling watersheds is extensive and some of the recent
Kalu-Ganga River in the Ratnapura District. In Sri Lanka,
applications are Pingel et al. (2005) and Chu and Steinman
administrative divisions within a district are named as Divi-
(2009). Dharmasena (1997) has modelled selected rivers in Sri
sional Secretariat (DS) divisions, while as the lowest adminis-
Lanka including the Kalu-Ganga River and concluded that
trative division, a DS division is divided into several Grama
even linear black box models including HEC1 provide a
Niladhari (GN) divisions. In the study, six DS divisions in the
satisfactory tool for the simulation of flows in river basins in
Ratnapura district namely, Kiriella, Kuruwita, Ratnapura,
the wet zone of Sri Lanka. The HEC-HMS model was initially
Pelmadulla, Elapatha and Ayagama, were considered and the
calibrated and then verified for the basin. At the calibration
GN divisions were considered for a unit-wise risk assessment.
stage, some model parameters were obtained from basin and
stream characteristics. Others were estimated based on a trial
Methodology and error procedure using a set of observed rainfall and runoff
The study presents the identification of inundation areas for data. The basin model needed for the HEC-HMS model was
a 100-year return period rainfall, followed by the assessment developed using Geospatial Hydrologic Modelling Extension,
of flood-risk using two different approaches. In the first HEC-GeoHMS (USACE-HEC, 2003), in the ArcView plat-
approach, hazard and vulnerability elements were assumed form. Runoff data for rainfall of a 100-year return period
to vary in a crisp manner while the second treated these required for the inundation development were computed
elements to vary in a fuzzy manner. using this calibrated and verified model.
Hydrodynamic model
Development of flood inundation area
Development of inundation area includes the use of a To obtain the extent of flood inundation, HEC-GeoRAS
rainfall–runoff model to determine the peak discharge due (USACE-HEC, 2009b) and River Analysis System (HEC-RAS)


c 2011 The Authors J Flood Risk Management 4 (2011) 128–139
c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
Flood risk analysis 131

(USACE-HEC, 2010) software were used. The HEC-GeoRAS commonly used terms to describe the potential threats that
is an ArcGIS extension specifically designed to process geos- natural disasters such as floods pose to human life, environ-
patial data for use with the HEC-RAS. It allows users to pre- ment and infrastructure.
and postprocess the data for the HEC-RAS.
The HEC-RAS is a one dimensional (1D) model that is
Conventional approach of risk assessment
very popular among the researchers and managers of flood-
plains as a flood inundation model (Tate et al., 2002; Knebl Flood-hazard assessment. Hazard as defined in United
et al., 2005; Pappenberger et al., 2005). Horritt and Bates Nations International Strategy for Disaster Reduction
(2002) compared 1D and 2D model codes (HEC-RAS, (UNISDR) (2009) is a dangerous phenomenon, substance,
LISFLOOD-FP and TELEMAC-2D) in an optimization human activity or condition that may cause loss of life,
framework, and reported that in some cases, 1D models injury or other health impacts, property damage, loss of
such as HEC-RAS are very effective in predicting flood livelihoods and services, social and economic disruption, or
extent. environmental damage.
The ‘Steady Flow Water Surface Profile’ component of the Flood-hazard assessment is the estimation of the overall
HEC-RAS, which is designed for applying in flood plain adverse effects of flooding for a particular area. It depends
management and flood insurance studies, was utilized in the on many parameters such as depth of flooding, duration of
study by modelling the Kalu-Ganga River from Ratnapura flooding, flood wave velocity and rate of rise of water level.
to Ellagawa. The main river was divided into five reaches One or more parameters can be considered in the hazard
within this stretch. The modelled stretch includes 12 tribu- assessment depending on the characteristics of the study
taries. The implementation of the HEC-RAS requires inputs area and floods (Tingsanchali and Karim, 2005). Consider-
that come from three basic categories of data: Geometric ing the characteristics of the study area, two major
data, Basin characteristics and Flow data. The requisite parameters, namely depth of flooding and percentage area
geometric data include stream centrelines and cross-section of flooding, were considered for the assessment of hazard of
cut lines and these are prepared using the HEC-GeoRAS the land units considered.
graphical user interface. A hazard index, HI, was introduced to represent the
Manning’s friction coefficient n represents basin charac- degree of hazard corresponding to different flood depths.
teristics. The Manning’s n value is highly variable and As recommended in past studies (e.g. Chowdhury and
depends on a number of factors including surface rough- Karim, 1997), four hazard categories were used and each
ness, vegetation, channel irregularities, channel alignment, category was represented by a hazard index. To devise a
scour and deposition, obstructions, etc. Reference is made to scale for HI, flooding areas were divided into four depth
Chow (1959), together with the land-use information of the categories based on three critical flood depths: 0.6, 1.0 and
basin, to attribute the selected part of the basin with the 3.5 m. Selection of the critical depths was based on the
initial values of Manning’s n. guidelines in the flood-plain development manual US
Discharge and normal water level values represented the Bureau of Reclamation (USBR) (1988). Based on these three
upstream and downstream boundary conditions. Having critical values of flood depth (D), hazards were classified as
completed the set up of the system with the requisite model low (D  0.6 m), medium (0.6 m o D  1.0 m), high
parameters and variables, calibration and verification were (1.0 m o D  3.50 m) and very high (3.50 m o D). As
performed. The initially used Manning’s n values were recommended by Tingsanchali and Karim (2005), a linear
varied to yield the rating curve at Ellagawa gauging station. scale was used for HI to represent different hazard categories
Subsequently, the inundation extent was obtained using the as presented in Table 2.
HEC-RAS model for a flood expected to occur due to a 100- For the practical application of the predicted results, the
year return period rainfall. hazard was estimated for a land unit and represented by a
The HEC-RAS model simulation results were exported to number: the hazard factor (HF). The HF for a land unit was
the HEC-GeoRAS for necessary processing and visualization taken to represent hazards due to flood depth (HFD) and
of flood extents. The results of this simulation were then
checked against flood depths obtained from a field survey
for a flood of the scale of a 100-year return period. Table 2 Hazard index for depth of flooding
Depth (D) of flooding (m) Hazard category Hazard index (HI)
Assessment of risk
0 o D  0.6 Low 1
0.6 o D  1.0 Medium 2
The concepts of hazard, vulnerability and risk have been
1.0 o D  3.5 High 3
used extensively in various disciplines with a different
3.5 o D Very high 4
meaning for facing hazardous events. They are the most

J Flood Risk Management 4 (2011) 128–139 c 2011 The Authors

c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
132 Nandalal and Ratnayake

inundated area (HFA). A land unit here is the lowest potentially damaging natural phenomena. The vulnerability is
administrative division, while a piece of land of expressed on a scale from 0 (no damage) to 1 (total damage).
200 m  200 m is the computational grid cell. The DEM of The vulnerability factor (VF) of each land unit was assessed
200 m  200 m grid cells was generated from the 1:10 000 using the parameters, population density and dependency
contour maps (contour interval of 5 m). A land unit is ratio of the land unit. Population density is the most
several times larger than a computational grid cell and, indicative factor of the social vulnerability of a community,
therefore, flooding of more than one depth category may which is calculated by dividing the total population by the
occur in the same land unit. Exposure of a land unit to the area of the land unit. The total dependency ratio described in
flood hazard is taken as 1, assuming that all land units are United Nation (UN) (2009) is used as a parameter to assess
equally exposed to it. the vulnerability of each land unit, with slight modifications
The hazard factor HFD for the flood depth of each land suitable for the local conditions. In this study, the ratio used
unit was computed based on the fraction area under each is the number of persons of age under 201 persons aged 55
depth category and the corresponding hazard index using or over per 100 persons of the total population. Thus, the
the following equation: vulnerability factors with respect to population density (VFP)
and total dependency ratio (VFA) are
!,
X
Nd X
Nd
HFD ðiÞ ¼ Aði; jÞHIi ðjÞ Aði; jÞ ð1Þ Total population
VFP ¼ ð5Þ
j¼1 j¼1 Land area

Number of persons under age 20 þ number of persons aged 55 or over

VFA ¼  100 ð6Þ
Total population

where i is the land unit identification number and j As each of the above vulnerability factors were in different
represents the depth category; Nd is the total number of scales, they were standardized as an index using the following
depth categories; Hij(j) is the hazard index for the area under equation:
depth category j in land unit i; and A(i, j) is the area under
depth category j in land unit i. VFK ðiÞ
VFKs ðiÞ ¼ ðK ¼ P or AÞ ð7Þ
The hazard factor HFA for flood area of each land unit was ðVFK Þmax
computed as the percentage area under flood irrespective of
depth using the following equation: where VFKS ðiÞ is the standardized vulnerability factor of the
land unit i, VFK(i) is the original hazard factor for land unit
Area under flood in land unit i and (VFK)max is the maximum VF in the range.
HFA ðiÞ ¼  100 ð2Þ
Total area of land unit i UN/ISDR (2004) stressed the need for population data
disaggregated by age and also UN (2009) indicated the
As each of the above hazard factors were measured on a importance of taking the total dependency ratio as a
different scale, they were standardized as an index using the vulnerability indicator. Therefore, in this research, both the
following equation: population density and the total dependency ratio are
considered equally important in the assessment of vulner-
HFK ðiÞ ability in the Kalu-Ganga River basin. Thus, assigning equal
HFks ðiÞ ¼ ðK ¼ D or AÞ ð3Þ
ðHFK Þmax weight to both of these parameters, the average value, VF(i),
was taken as the vulnerability factor of the land unit i as given
where HFKs ðiÞ is the standardized hazard factor of the land
in the following equation:
unit i, HFK(i) is the original hazard factor for the land unit i
and (HFK)max is the maximum HF in the range. Finally, VFðiÞ ¼ ðVFPs ðiÞ þ VFAs ðiÞÞ=2 ð8Þ
assigning the same weight for both of these factors, the
average value was taken as the hazard factor of the land unit
as in the following equation: Flood risk assessment. In general, risk is a concept that
incorporates the concepts of hazard and vulnerability. It is
HFðiÞ ¼ ðHFDs ðiÞ þ HFAs ðiÞÞ=2 ð4Þ customary to express risk (R) as a functional relationship of
hazard and vulnerability. The magnitude of risk for a land
Flood-vulnerability assessment. Vulnerability is a measure unit was estimated by a risk factor, RF(i), which was
of the intrinsic susceptibility of an element at risk exposed to computed as the product of the hazard factor and the


c 2011 The Authors J Flood Risk Management 4 (2011) 128–139
c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
Flood risk analysis 133

vulnerability factor as in the following equation: categories, having a certain degree of membership in each
category (27% low as well as 68% medium vulnerable).
RFðiÞ ¼ HFðiÞ  VFðiÞ ð9Þ
Fuzzy inference. The fuzzified variables are related to each
other with a knowledge-based rule system. A statement
Fuzzy approach of risk assessment about the resulting variable has to be made for all possible
combinations of the categories of all variables. Suppose the
Quantifying vulnerability, hazard and risk is difficult as model consists of two variables: population density and
many elements (e.g. flood depth, population density, etc.) flood depth. If each of these variables is defined in two
may contribute towards these factors in complex ways. states (i.e. ‘low’ and ‘high’), then four rules are required to
Normally, knowledge of these elements is vague. Besides, describe the resulting output variable (say, risk index). The
vulnerability, hazard and risk may have nonlinear relation- rules describing the system can be:
ships with these elements and the weights to be attached to
these elements are less known (Kavi Kumar and Viswa- Rule 1: If the population density is low and flood depth is
nathan, 2006). Therefore, hazard and vulnerability aspects low, then the risk is low.
were considered as fuzzy in the assessment of risk due to Rule 2: If the population density is low and flood depth is
floods and the results were compared with the crisp conven- high, then the risk is medium.
tional approach described previously. Rule 3: If the population density is high and flood depth is
low, then the risk is medium.
Fuzzy inference system Rule 4: If the population density is high and flood depth is
high, then the risk is high.
Fuzzy set theory is useful in translating linguistic state-
ments such as ‘high’ or ‘low’ into numerical values. These rules transform the input variables into an output
As described by Zadeh (1996), it is a formal attempt to variable. The output variable, which is the risk in this work,
capture, represent and work with objects with unclear also has to be defined by membership functions, usually low,
or ambiguous boundaries. The use of fuzzy systems con- medium and high risk, as in Figure 3. The number of rules
sists of three steps: fuzzification, fuzzy inference and to be considered will depend on the number of membership
defuzzification. function considered for the input and output variables.
There are three common methods of deductive inference
Fuzzification. This involves the translation of propositions for fuzzy systems based on such linguistic rules: (1) Mamd-
into quantitative values using membership functions. hani systems, (2) Sugeno models and (3) Tsukamoto models
For instance consider the proposition: ‘if population (Ross, 2004). The first inference method, the Mamdhani
density is high, the vulnerability is high’. In binary logic, system (Mamdhani and Assilian, 1975 as cited in Ross,
levels such as ‘high’ and ‘low’ are assigned sharp 2004), is the most common method used in practice and it
boundaries, whereas in fuzzy logic, it is possible to assign was used as the fuzzy inference system in the study.
nonsharp (or fuzzy) boundaries. As shown in Figure 3, the
Defuzzification. This final step of the fuzzy inference
membership functions defined in a fuzzy model describe
system is necessary for combining the results of each rule
the ‘degree of belief ’ of a particular value of a variable.
into a single result. Although there are several mathematical
As shown in Figure 3, a population density of, say,
approaches available for defuzzification, the most com-
36 persons/ha need not be assigned to either a ‘low’ or a
monly used approach, the ‘centre of gravity’ method, was
‘medium’ vulnerable category, but can be a member of both
used in the analysis.
Thus, the procedure adopted involved the following
steps: (a) translation of input indicators into fuzzy numbers
using the fuzzification procedure, (b) application of fuzzy
inference to relate the input indicators and output index
and (c) use of defuzzification to translate the fuzzy index
into a crisp value, which is taken as the risk index of the
land unit.
The flood area, flood depth, population density and
dependency ratio were defined as fuzzy variables using the
methods described and then used to calculate a fuzzy
measure of risk. The fuzzy toolbox in the MATLAB software
Figure 3 Example of a membership function. was used to implement the fuzzy calculations.

J Flood Risk Management 4 (2011) 128–139 c 2011 The Authors

c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
134 Nandalal and Ratnayake

Results and discussion stations and the last row represents the average value for all
three rainfall events.
Flood flow simulation The Nash Sutcliffe efficiency coefficient indicates the high
efficiency of the developed event-based model. Visual in-
River flow generation spection of Figure 4 further indicates the suitability of the
model in the prediction of floods.
A frequency analysis was performed using the Gumbel
distribution for all fourteen rainfall gauging stations using
rainfall data for 23 years from 1987 to 2009. Then rainfalls of Development of the inundation area
a 100-year return period at these gauging stations were
obtained from the distribution. These values were averaged A digital terrain model for the study area was created using
using the Thiessen polygon method to compute the aerial topographic maps of scale 1:10 000 and spot levels. Then the
average rainfall. This rainfall was used in the calibrated river network between Ratnapura and Ellagawa was gener-
HEC-HMS-based model for the basin to generate river ated using the HEC-GeoRAS. After the introduction of
discharges that are required for the hydrodynamic model to geometrical information such as riverbanks, cross sections
obtain flood inundation areas and depths. and land-use data required for the determination of surface
roughness, etc., on to the river network, they were exported
to the HEC-RAS hydrodynamic model.
Calibration and verification of the hydrological The hydrodynamic model, HEC-RAS, was calibrated
model using the rating curve at the Ellegawa gauging station. The
Manning coefficient n for the river and the land area was
The hydrological model, HEC-HMS, was calibrated for the adjusted by trial and error at the calibration stage and its
entire Kalu-Ganga River basin using one rainfall event and value was found to be in the range of 0.030–0.055. Subse-
verified with three other rainfall events (Nandalal and quently, using the HEC-RAS model, the inundation extents
Ratnayake, 2010). Rainfall data at 14 rainfall gauging with the corresponding depths for a flood expected to occur
stations and runoff data at three river gauging stations were due to a rainfall of return period 100 years were developed.
used at this step. Figure 4 shows the comparison of the Figure 5 depicts the inundation area along the Kalu-Ganga
computed and observed discharge hydrographs of a rainfall River between Ratnapura and Ellagawa.
event in June 1994 at Ellegawa gauging station during the
calibration.
In general, the agreement between the computed and the
Table 3 Nash Sutcliffe model efficiency coefficient for observed gauging
observed runoff hydrographs for both model calibration and
points for the model
verification is acceptable. The hydrographs were compared
using the Nash Sutcliffe model efficiency coefficient (Nash Event Ratnapura Ellagawa Putupaula
and Sutcliffe, 1970) calculated for computed discharge and 1993 May 0.920920 0.942308 0.700043
observed discharge at the three gauging stations, Ratnapura, 1993 October 0.932305 0.942336 0.803982
Ellagawa and Putupaula. Table 3 shows the Nash Sutcliffe 1994 May 0.912886 0.959965 0.957709
0.924131 0.922629 0.860085
model efficiency coefficient yielded for these three gauging

Figure 4 Comparison of hydrographs at calibration.


c 2011 The Authors J Flood Risk Management 4 (2011) 128–139
c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
Flood risk analysis 135

Figure 5 Flood inundation area along Kalu-Ganga River.

Figure 6 Flood risk map: conventional approach.

numbers in the figures are the identification numbers given

Risk assessment for the GN divisions within the Ratnapura district by the
administration. These GN divisions were categorized into
Conventional approach in risk assessment five risk zones based on scores of five equal intervals,
as recommended by Tingsanchali and Karim (2005). The
Risk due to flooding was computed at the GN division level. risk areas were named as a very low-risk zone for
The inundation areas and depths obtained from the hydro- 0.001 o RF  0.2, a low-risk zone for 0.2 o RF  0.4, a
dynamic model provide exposure that is needed for the medium-risk zone for 0.4 o RF  0.6, a high-risk zone
calculation of risk. for 0.6 o RF  0.8 and a very high-risk zone for 0.8 o
The computed risk factors (RF) varied in the range RF  1.0. According to this analysis, 33 GN divisions were
0.004–1.0 among the considered GN divisions. Sixty-seven rated as very low risk, 26 GN divisions as low risk, 7 GN
GN divisions were identified within the 100-year return divisions as medium risk and 1 GN division (Ratnapura, no.
period floodplain area as shown in Figures 6 and 10. The 284) as very high-risk divisions (Table 5). No GN divisions

J Flood Risk Management 4 (2011) 128–139 c 2011 The Authors

c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
136 Nandalal and Ratnayake

Figure 7 Membership functions used to fuzzify the hazard factors.

Figure 8 Membership functions used to fuzzify the vulnerability factors.

turned out to be in the high-risk zone. Figure 6 presents the

risk levels of the 67 GN divisions.

Fuzzy approach in risk assessment

At this step, the flood area, flood depth, population density

and population vulnerability were treated as fuzzy variables.
Fuzzy membership functions for the hazard and vulnerabil-
ity factors were selected to reflect the crisp values used in the Figure 9 Fuzzy membership functions for risk.
conventional method to enable a comparison between these
two methods. The selection of the form of the membership If (Hazard Area is high) and (Hazard Depth is Low) and
function is subjective and depends on the context. Triangu- (Dependency Ratio is medium) and (Population Density
lar, trapezoidal, type-Z, type-S and bell-shaped functions are is Low) then (Risk is medium).
the most commonly used in engineering applications. The If (Hazard Area is very high) and (Hazard Depth is Low)
membership functions used to fuzzify the hazard and and (Dependency Ratio is low) and (Population Density
vulnerability factors are as shown in Figures 7 and 8, is Low) then (Risk is medium).
respectively. As indicated in the figures: four membership Risk is also treated as a fuzzy variable. The membership
functions (two triangular and two trapezoidal) were selected functions assumed for the risks are given in Figure 9. As
to fuzzify the flood area; two membership functions (type-Z shown in the figure, five membership functions were used
and type-S) were selected to fuzzify flood depth; three for the risk.
membership functions (one triangular and two trapezoidal) The estimated risk factors (RF) based on the fuzzy logic
were selected to fuzzify the dependency ratio; and three approach varied in the range 0.159–0.92 among the GN
membership functions (type-Z, bell and type-S) were divisions. According to the analysis, the numbers of GN
selected to fuzzify the population density. These member- divisions rated as very low risk, low risk, medium risk, high
ship functions were selected to express best representation of risk and very high risk are as in Table 4. Figure 10 depicts the
each factor after running several trials. Therefore, at the spatial distribution of the risks at the GN division level.
fuzzy inference stage, 72 rules were set to capture the risk
value. Some of the rules set are displayed below. Comparison of conventional and fuzzy approaches
If (Hazard Area is low) and (Hazard Depth is Low) and in risk assessment
(Dependency Ratio is low) and (Population Density is
Medium) then (Risk is low). Treatment of hazard, vulnerability and risk as fuzzy variables
If (Hazard Area is low) and (Hazard Depth is High) and changes the risk level of the GN divisions. The identification
(Dependency Ratio is low) and (Population Density is of Mudduwa/East (no. 277), Godigamuwa (no. 283), Mud-
Low) then (Risk is very low). duwa (no. 276), Ratnapura/west (no. 286) and Ratnapura


c 2011 The Authors J Flood Risk Management 4 (2011) 128–139
c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
Flood risk analysis 137

Table 4 Number of GN divisions in different risk categories from the conventional approach and the fuzzy approach
Very low risk Low risk Medium risk High risk Very high risk
Conventional approach 33 26 7 0 1
Fuzzy approach 8 12 29 13 5

GN, Grama Niladhari.

Figure 10 Flood risk map: fuzzy approach.

(no. 284) GN divisions as very high-risk zones was found to A comparison of the results obtained from two ap-
be in agreement with the historical information. These GN proaches was made based on the actual flood relief expenses
divisions, which are having high inundation areas, are also incurred during a flood event that occurred in June 2008
the most densely populated. According to the conventional obtained from the records of the Ratnapura Divisional Unit
crisp approach, only one GN division (Ratnapura) was rated of the Department of Social Services. This flood was in the
as very high risk. The number of GN divisions rated in each calibre of a 100-year return period flood. Table 5 presents
category using both approaches is compared in Table 4. As the actual relief expenses per unit area and the risk levels
the table shows, the conventional method seems to be resulted from the crisp and fuzzy approaches for a few GN
underestimating the risk due to floods in these areas. divisions. As it shows, the Ratnapura GN division is
The factors used to quantify the hazard and vulnerability categorized as a very high-risk division by both methods.
were measured in different units. Therefore, to combine The risk levels of the GN divisions obtained by the fuzzy
them in the crisp approach, it was necessary to standardize approach were found to be decreasing with the relief
the hazard factors and vulnerability factors divided by their expenses. However, the results obtained by the crisp ap-
maximum value in the range. For example, the depth hazard proach do not show such a consistent change in the risk level
factor, which ranges from 0 to 17 m for all land units, has to with the expenses. These results confirm the superiority of
be divided by the maximum value of 17 m, to bring all into the fuzzy approach over the crisp approach.
the range 0–1. Similarly, area hazard factors that ranged
from 0 to 6 466 030 m2 were also divided by the largest area
of 6 466 030 m2. This enabled the conversion of both hazard Conclusions
factors to a value within the range 0–1. This standardization
results in a relative hazard or vulnerability among all the A flood risk analysis of a flood-prone area immensely
land units instead of their real values in the crisp approach. supports the decision makers in taking correct decision at
The fuzzy approach does not have this setback as the risk the right time. It helps in all phases of a flood-related
derived is independent of the units of the factors considered disaster, i.e. pre, post and during the disaster. In Sri Lanka,
to indicate the hazard and vulnerability of the land units. during a disaster, all relief activities are carried at the GN
This is a major benefit of the fuzzy approach. division level and, therefore, flood risk analysis considering

J Flood Risk Management 4 (2011) 128–139 c 2011 The Authors

c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
138 Nandalal and Ratnayake

Table 5 Flood relief expenses for June 2008 flood and the risk levels obtained by the crisp and fuzzy approaches for a few GN divisions
Risk criteria

GN division Relief expense/ha (LKR) Crisp Fuzzy

Ratnapura (no. 284) Rs. 8085.00 Very high risk Very high risk
Godigamuwa (no. 283) Rs. 5108.00 Medium risk Very high risk
Muwagama (no. 278) Rs. 4511.00 Low risk High risk
Pallegedara (no. 73) Rs. 2547.00 Medium risk High risk
Angammana (no. 280) Rs. 2004.00 Very low risk Medium risk
Pahala Hakamuva (no. 203) Rs. 1260.00 Low risk Medium risk
Mada Baddara (no. 124) Rs. 505.00 Very low risk Low risk
Withangagama (no. 135) Rs. 43.00 Very low risk Very low risk

GN, Grama Niladhari.

the land units as GN divisions is the most suitable. The References

HEC-HMS- and HEC-RAS-based models were found to be
Anselmo V., Galeati G., Palmieri S., Rossi U. & Todini E. Flood
very suitable in the derivation of flood inundation area in
risk assessment using an integrated hydrological and hydraulic
the Kalu-Ganga River basin.
modelling approach: a case study. J Hydrol 1996, 175, (4),
Among the different hydraulic parameters that determine
533–554.
hazard due to a flood, the flood depth and inundation are Birkmann J. Risk and vulnerability indicators at different scales:
found to be two major parameters that can be used to applicability, usefulness and policy implications. Environ
determine hazard level sufficiently. Hazards 2007, 7, (1), 20–31.
The vulnerability of an area always depends on the Buchele B., Kreibich H., Kron A., Thieken A., Ihringer J., Oberle
population of that area and also the age-wise distribution P., Merz B. & Nestmann F. Flood-risk mapping: contributions
of the population. The vulnerability factors, the population towards an enhanced assessment of extreme events and
density and the ratio of the dependent population to the associated risks. Nat Hazards Earth Syst Sci 2006, 6, (4),
total population, were found to be effective in the determi- 485–503.
nation of risk in the area. Chen H. & Chang N. Using fuzzy operators to address the
The conventional method-based flood risk levels did not complexity in decision making of water resources
capture the variability of the hazard and vulnerability as redistribution in two neighbouring river basins. Adv Water
expected. This is mainly due to the standardization process Resour 2010, 33, (6), 652–666.
that had to be carried out at all levels to equate to one scale. Chow V.T. Open channel hydraulics. New York: McGraw-Hill
The fuzzy logic-based approach has captured the levels of Book Co, 1959.
indicator parameters, hazard and vulnerability factors effec- Chowdhury J.U. & Karim M.F. A risk based zoning of storm surge
tively and resulted in a fair risk distribution. The fuzzy prone area of the Ganges tidal plain. UNCRD Proc Ser 1997,
approach does not need standardization as the input and 17, (2), 171–185.
output membership functions are always in the scale of 0–1 Chu X. & Steinman A. Event and continuous modeling with
irrespective of the variation of the factors considered. The HEC_HMS. J Irrig Drainage Eng 2009, 135, (1), 119–124.
results obtained by the fuzzy approach matched better than Churchill R.R. & Hutchinson D.M. Flood hazard in Ratnapura,
the crisp approach when compared with the flood relief Sri Lanka: individual attitudes vs collective action. Geoforum
expenses incurred at GN divisions for a recent flood. This 1984, 15, (4), 517–524.
strengthens the success of the application of fuzzy approach Dharmasena G.T. Application of mathematical models for flood
in risk assessment. forecasting in Sri Lanka. In: G.H. Leavesley, H.F. Lins, F.
The presented approach is found to be very much Nobilis, R.S. Parker, V.R. Schneider & F.H.M. van de Ven, eds.
applicable in risk assessment as it grasps the fuzziness of Destructive Water: Water-Caused Natural Disasters, their
hazard, vulnerability and risk by taking uncertainty into Abatment and Control. Wallingford, UK: IAHS Publication
account when their different categories are defined. no. 239, 1997, 225–235.
Esogbue A.O., Theologidu M. & Guo K. On the application of
fuzzy sets theory to the optimal flood control problem arising
Acknowledgements in water resources systems. Fuzzy Sets Syst 1992, 48, (1),
155–172.
The authors acknowledge the University Grant Commission Horritt M.S. & Bates P.D. Evaluation of 1D and 2D numerical
of Sri Lanka for providing the necessary funds to carry out models for predicting river flood inundation. J Hydrol 2002,
this research. 268, (1–4), 87–99.


c 2011 The Authors J Flood Risk Management 4 (2011) 128–139
c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management 
Flood risk analysis 139

Kavi Kumar K.S. & Viswanathan B. Vulnerability to globalization Tate E.C., Maidment D.R., Olivera F. & Anderson D.J. Creating a
in India – Relative rankings of States Using Fuzzy Models. terrain model for floodplain mapping. J Hydrol Eng 2002, 7,
Research Paper No. 2006/40, United Nations University, (2), 100–108.
World Institute for Development Economic Research Tingsanchali T. & Karim M.F. Flood hazard and risk analysis in
(UNU-WIDER), 2006. the southwest region of Bangladesh. Hydrol Process 2005, 19,
Knebl M.R., Yang Z.L., Hutchison K. & Maidment D.R. Regional (10), 2055–2069.
scale flood modeling using NEXRAD rainfall, GIS, and HEC- Tingsanchali T. & Karim M.F. Flood-hazard assessment and
HMS/RAS: a case study for the San Antonio River Basin Summer risk-based zoning of a tropical flood plain: case study of the
2002 storm event. J Environ Manag 2005, 75, (4), 325–336. Yom River, Thailand. Hydrol Sci J 2010, 55, (2), 145–161.
Lascio L.D., Gisolfi A., Albunia A., Galardi G. & Moschi F. United Nations (UN). World Population Ageing. United Nations,
A fuzzy-based methodology for the analysis of diabetic New York, 2009. Available at [Link]
neuropathy. Fuzzy Sets Syst 2002, 129, (2), 203–228. population/publications/WPA2009/WPA2009_
Lawry J. A methodology for computing with words. Int J [Link] (accessed 26 May 2010).
Approximate Reasoning 2001, 28, (2–3), 51–89. United Nations International Strategy for Disaster Reduction
Lee H.M. Applying fuzzy set theory to evaluate the rate of (UNISDR). UNISDR Terminology on Disaster Risk
aggregative risk in software development. Fuzzy Sets Syst 1996, Reduction. 2009. Available at [Link]
79, (3), 323–336. terminology/[Link] (accessed 23 March
Messner F. & Meyer V. Flood damage, vulnerability and risk 2010).
perception – challenges for flood damage research. In: UN/ISDR. Living with risk. A global review of disaster reduction
J. Schanze, E. Zeman & J. Marsalek, eds. Flood Risk initiatives. 2004 version. Geneva: UN Publications, United
Management–Hazards, Vulnerability and Mitigation Measures. Nations International Strategy for Disaster Reduction, 2004.
Dordrecht, The Netherlands: Springer, 2006, 149–167. United States Army Corps of Engineers, Hydrologic Engineering
Nandalal H.K. & Ratnayake U.R. Event based modelling of a water- Centre (USACE-HEC). Geospatial Hydrologic Modeling
shed using HEC-HMS. J Inst Eng, Sri Lanka 2010, 43, (2), 28–37. Extension: HEC-GeoHMS. User’s manual, Version 1.1, Davis,
Nash J.E. & Sutcliffe J.V. River flow forecasting through California, 2003.
conceptual models, part-1 – a discussion of principles. United States Army Corps of Engineers, Hydrologic Engineering
J Hydrol 1970, 10, (3), 282–290. Centre (USACE-HEC). Hydrological Modeling System HEC-
Pappenberger F., Beven K., Horritt M. & Blazkova S. Uncertainty HMS. User’s manual, Version 3.4, Davis, California, 2009a.
in the calibration of effective roughness parameters in United States Army Corps of Engineers, Hydrologic Engineering
HEC-RAS using inundation and downstream level Centre (USACE-HEC). HEC-GeoRAS: GIS tools for support
observations. J Hydrol 2005, 302, (1–4), 46–69. of HEC-RAS using ArcGIS. User’s manual, Version 4.2, Davis,
Pingel N.P.E., Jones C. & Ford D. Estimating forecast lead time. California, 2009b.
Nat Hazards Rev 2005, 6, (2), 60–66. United States Army Corps of Engineers, Hydrologic Engineering
Ross T.J. Fuzzy logic with engineering applications, 2nd edn, UK: Centre (USACE-HEC). HEC-RAS River Analysis System.
John Wiley & Sons Ltd, 2004. User’s manual, Version 4.1, Davis, California, 2010.
Sadiq R. & Husain T. A fuzzy-based methodology for an US Bureau of Reclamation (USBR). Downstream Hazard
aggregative environmental risk assessment: a case study of Classification Guidelines. ACER Technical Memorandum
drilling waste. Environ Model Software 2005, 20, (1), 33–46. No. 11, Assistant Commissioner-Engineering and Research,
Sadiq R., Rajani B. & Kleiner Y. A fuzzy-based method to evaluate Denver, Colorado, 1988.
soil corrosivity for prediction of water main deterioration. Zadeh L.A. Fuzzy logic computing with words. IEEE
ASCE J Infrastruct Syst 2004, 10, (4), 149–156. Transactions. Fuzzy Syst 1996, 4, (2), 103–111.

J Flood Risk Management 4 (2011) 128–139 c 2011 The Authors

c 2011 The Chartered Institution of Water and Environmental Management
Journal of Flood Risk Management