Ahmadisharaf et al.

/ Journal of Hydrology (2016)

Spatial probabilistic multi-criteria decision making for assessment of flood management

alternatives

Ebrahim Ahmadisharafa, Alfred J. Kalyanapub,*, and Eun-Sung Chungb

38505-0001
2Assistant Professor, Department of Civil and Environmental Engineering, Tennessee Technological University, Cookeville, TN

38505-0001

* Corresponding Author. Tel.: +1 931 372 3561; Fax: 931-372-6239; Email address: [email protected]

3Assistant Professor, Department of Civil Engineering, Seoul National University of Science and Technology, Seoul 139-743,

Seoul, Republic of Korea

Abstract

Flood management alternatives are often evaluated on the basis of flood parameters such as depth and

velocity. As these parameters are uncertain, so is the evaluation of the alternatives. It is thus important to

incorporate the uncertainty of flood parameters into the decision making frameworks. This research

develops a spatial probabilistic multi-criteria decision making (SPMCDM) framework to demonstrate the

impact of the design rainfall uncertainty on evaluation of flood management alternatives. The framework

employs a probabilistic rainfall-runoff transformation model, a two-dimensional flood model and a spatial

MCDM technique. Thereby, the uncertainty of decision making can be determined alongside the best

alternative. A probability-based map is produced to show the discrete probability distribution function

(PDF) of selecting each competing alternative. Overall the best at each grid cell is the alternative with the

mode parameter of this PDF. This framework is demonstrated on the Swannanoa River watershed in

North Carolina, USA and its results are compared to those of deterministic approach. While the

deterministic framework fails to provide the uncertainty of selecting an alternative, the SPMCDM

framework showed that in overall, selection of flood management alternatives in the watershed is

“moderately uncertain”. Moreover, three comparison metrics, F fit measure, κ statistic, and Spearman

rank correlation coefficient (ρ), are computed to compare the results of these two approaches. An F fit

measure of 62.6%, κ statistic of 15.4% to 45.0%, and spatial mean ρ value of 0.48, imply a significant

1
 Ahmadisharaf et al. / Journal of Hydrology (2016)

difference in decision making by incorporating the design rainfall uncertainty through the presented

SPMCDM framework. The SPMCDM framework can help decision makers to understand the uncertainty

in selection of flood management alternatives.

Key Words: Spatial probabilistic multi-criteria decision making (SPMCDM), Probabilistic modeling,

Uncertainty analysis, Flood management, Design rainfall uncertainty.

1 Introduction

Flood management is complex and multifaceted, affected by different factors, involving various

stakeholders, competing alternatives and different tradeoffs (Levy et al., 2007; Hall and Solomatine,

2008; Schröter et al., 2014). Under these circumstances, multi-criteria decision making (MCDM) can

assist flood management by providing a systematic framework to deal with such complex problems.

Several MCDM techniques with different capacities can be identified based on the literature. There has

been a vast application of various MCDM techniques in different categories of flood management such as

flood risk mapping (Sinha et al., 2008; Meyer et al., 2009; Chen et al., 2011b; Zou et al., 2012), flood

hazard zoning (Fernandez and Lutz, 2010; Kourgialas and Karatzas, 2012; Stefanidis and Stathis, 2013;

Radmehr and Araghinejad, 2014; Papaioannou et al., 2015; Rahmati et al., 2015), flood risk assessment

(Lee et al., 2015; Malekian and Azarnivand, 2015), flood vulnerability analysis (Radmehr and

Araghinejad, 2015), site selection of flood mitigation measures (Ahmadisharaf et al., 2015b),

prioritization of flood mitigation strategies (Willette and Sharda, 1991; Bana E Costa et al., 2004; Levy,

2005; Chitsaz and Banihabib, 2015) and integrated assessment of long-term flood management scenarios

(Brouwer and van Ek, 2004). Main reason of applying MCDM for flood management is the inherent

complexity and multidisciplinarity, and the capability of MCDM techniques to structure such a

multifaceted problem into a simple quantifiable format.

Flood management should be considered a spatial problem because flood intensities and

characteristics vary with geographic location (Foudi et al., 2015). There has been a growing interest in

coupling GIS with MCDM techniques due to the capabilities of GIS in handling wide range of criteria

data from different sources (Chen et al., 2010). Conventional flood management decision making does

2
 Ahmadisharaf et al. / Journal of Hydrology (2016)

not account for the spatial variability of the evaluation criteria (Qi et al., 2013). Consequently, the

selected alternative might not be necessarily the best option for all locations within the region (Tkach and

Simonovic, 1997). In other words, while some areas may benefit from implementing an alternative, that

measure might aggravate the flood status in other locations. Using spatial MCDM (SMCDM) is more

desirable as it enables decision makers (DMs) to account for the spatial variability of flood characteristics,

namely, depth, velocity and duration. Considering the needs for spatial dimension of flood management

problems, this study uses Spatial Compromise Programming (SCP) (Tkach and Simonovic, 1997), to

evaluate a pool of flood management alternatives on a cell-by-cell basis.

Evaluation of flood management alternatives relies on flood parameters. The flood parameters are

produced by integrating hydrologic and hydraulic models. Both these models are associated with

uncertainty, which causes the prioritization of alternatives be highly risky if the model parameters are not

fixed. In hydrologic modeling, uncertainty in prediction of hydrograph arises from calibration\validation

data, model structure and parameters as well as input variables (Krzysztofowicz and Kelly, 2000;

Krzysztofowicz and Harr, 2001; McMillan et al., 2010). One of the primary inputs in the hydrologic

models is rainfall. The rainfall dataset is obtained directly from measurement or by statistically analyzing

the rainfall records. The latter is the commonly used approach to determine a design storm, in which

rainfall depth is assigned a return period by fitting a suitable probability distribution function (PDF). As

the inferred design storm through this statistical procedure is subject to uncertainty, so are the generated

hydrographs. These hydrographs feed to hydraulic models, which causes the uncertainty to be propagated

through these models. As a result, the produced flood parameters are associated with uncertainty

(Kalyanapu et al., 2012). In addition to this source of uncertainty, uncertainty in hydraulic modeling

might come from model structure and parameters, observed data, digital elevation model (DEM), land use

and soil data as well as choice of performance measures (Pappenberger et al., 2006; Smemoe et al., 2007;

Merwade et al., 2008; Aronica et al., 2012; Bhuyian et al., 2015). Probabilistic methods can be used to

incorporate this uncertainty into both hydrologic and hydraulic models (Di Baldassarre et al., 2010). One

concern about using probabilistic approach is the high computational time needed by hydraulic models (in

3
 Ahmadisharaf et al. / Journal of Hydrology (2016)

particular, multi-dimensional models) to perform the routing problem for multiple hydrographs. However,

recent advances in computational capabilities of flood models with tremendous speedup (e.g., Flood2D-

GPU by Kalyanapu et al. (2011)) can assist modelers in efficiently using probabilistic-based analyses.

Due to these advances, decision making needs to be improved by incorporating probabilistic frameworks.

The uncertainty in a MCDM may stem from selection of the criteria as well as criteria weights and

values (Hyde et al., 2003, 2004; Ascough et al., 2008; Chen et al., 2010; Ahmadisharaf et al., 2015a).

Most of the studies in the area of flood management have incorporated the uncertainty of criteria weights

through sensitivity analysis (Kang et al., 2013a) and considered the uncertainty of performance values by

using fuzzy methods (Lee et al., 2013; Kim and Chung, 2014; Kim et al., 2015). However, probabilistic

approach using all plausible performance values has not received any attention yet. Rational decision

making requires that the uncertainty of hydrologic predictions being quantified in terms of PDFs, which is

the most perfect uncertainty description method (Tung, 2011), subject to the available information and

knowledge (Krzysztofowicz, 1999). Probabilistic hydrologic predictions are more favorable as they are

scientifically more reliable, enabling rational decision making (Krzysztofowicz, 2001). Edjossan-Sossou

et al. (2014) stated that it is critical to adequately analyze uncertainty and examine its influences to

improve the decision making. Madani and Lund (2011) recommended use of more rigorous approaches to

inform the DM about the impacts of the uncertainty on prioritizing the alternatives. Mosadeghi et al.

(2013) highlighted the need for integrating simulation algorithms such as Monte Carlo (MC) method into

SMCDM in order to analyze the influences of uncertainty. Pappenberger et al. (2013) and Ronco et al.

(2014) emphasized that the uncertainties attributed to predicted flood risks must be clearly communicated

to the DMs. Nevertheless, the DMs are often poorly served with information about the impacts of

uncertainty on flood management decisions (Pappenberger and Beven, 2006; Rosner et al., 2014).

Therefore, the objective of this paper is to develop a spatial probabilistic MCDM (SPMCDM)

framework to prioritize the flood management alternatives considering the impact of the design rainfall

depth uncertainty. The unique aspect of this study is to present uncertainty level to the decisions on

prioritization of flood management alternatives by using the developed SPMCDM framework. This

4
 Ahmadisharaf et al. / Journal of Hydrology (2016)

framework is illustrated on the Swannanoa River watershed in North Carolina, USA. The remaining of

this paper is organized as follows: Section 2 describes the methodology that is used to develop the

SPMCDM framework; Section 3 introduces the case study to demonstrate the developed framework;

Section 4 discusses the case study results, including analysis of the impacts of design rainfall uncertainty

on decision making by comparison of deterministic and probabilistic frameworks; and Section 5 provides

a summary of the framework development and the study implications.

2 Methodology

The developed SPMCDM framework contains three modules as shown in Fig. 1: 1) Probabilistic

hydrologic modeling; 2) Hydraulic modeling; and 3) SMCDM. In the first module, a probabilistic

hydrologic model is employed to simulate rainfall-runoff transformation process. In the second module, a

flood model named Flood2D-GPU is applied to model subsequent flood. In the third module, SCP is used

to prioritize multiple flood management alternatives.

5
 Ahmadisharaf et al. / Journal of Hydrology (2016)

Fig. 1 SPMCDM framework schematic diagram developed in this study

2.1 Probabilistic Hydrologic Modeling Module

A probabilistic continuous semi-distributed conceptual hydrologic model is developed in GoldSim®

environment. GoldSim® is a dynamic simulation software with applications ranging from water resources

management to financial predictions, and provides a versatile user-friendly graphical user interface (GUI)

for probabilistic modeling. The model takes rainfall time series in tandem with the characteristics of river

cross sections and subwatersheds as input variables and generates the flow hydrograph at subwatersheds’

outlets and different locations of the river. It accounts for the spatial variability of rainfall, topography,

soil characteristics and land use by dividing the study watershed into multiple subwatersheds. The

Snyder’s method (Snyder, 1938), Soil Conservation Service (SCS) infiltration method (Natural Resources

Conservation Service (NRCS), 1986) and Muskingum technique are used to develop the unit hydrograph,

6
 Ahmadisharaf et al. / Journal of Hydrology (2016)

to simulate rainfall-runoff transformation process and to perform channel routing, respectively. In

addition to reaches and subwatersheds, reservoirs can be also included in the simulations. The model uses

level pool routing to route the flow through the reservoirs. Reservoir stage-storage-discharge table and

evapotranspiration rate must be entered by the user for reservoir computations. It is to be mentioned that

groundwater processes are not taken into account in the model. The deterministic model has been

successfully applied in previous studies such as Ahmadisharaf et al. (2015a) for hydrologic modeling and

Ahmadisharaf and Kalyanapu (2015) for channel routing.

The probabilistic hydrologic model is developed by extending a deterministic model to probabilistic.

For doing so, any desired input can be characterized through a PDF. The probabilistic model is capable of

accounting for multiple uncertain parameters. Furthermore, the correlation between the uncertain

parameters can be considered. However, in the present study, only design rainfall depth is considered as a

stochastic element. This is because rainfall depth was identified as the most influential parameter on the

hydrograph through sensitivity analysis. It is likewise in agreement with the general literature suggestion

(e.g., Merwade et al., 2008). A uniform PDF is chosen to characterize this element, in which lower and

upper confidence bounds estimated by Bonnin et al. (2004), are taken as minimum and maximum values.

Latin Hypercube Sampling (LHS) (McKay et al., 1979) is employed in this study for propagation of

uncertainty that is commonly preferred to the standard MC method as smaller sample size is required and

computational efficiency will be achieved accordingly (Helton and Davis, 2003; Hall et al., 2005;

Janssen, 2013). At each LHS realization, a hydrograph is generated by the probabilistic hydrologic model.

Resulting hydrographs feed to the second module.

2.2 Hydraulic Modeling Module

The hydraulic modeling module uses a two-dimensional (2D) physically-based flood model named

Flood2D-GPU. Developed in NVIDIA's CUDA programming environment, it is a 2D unsteady numerical

model that solves the Saint Venant equations using a first-order accurate upwind difference scheme to

generate flood depths and velocities (Kalyanapu et al., 2011). A staggered grid stencil is used to define

the computational domain with the water depth in the center of the cell, and horizontal and vertical

7
 Ahmadisharaf et al. / Journal of Hydrology (2016)

velocities on the cell edges (Kalyanapu et al., 2011). The model provides significantly reduced

computational time (by 80-88 times) comparing to the same flood model implemented in a CPU-based

environment (Kalyanapu et al., 2011). It is thus very advantageous for probabilistic analyses where high

computational cost is expected. The model has been successfully applied in flood hazard analysis

(Kalyanapu et al., 2012, 2013; Ahmadisharaf et al., 2013), flood damage estimation (Kalyanapu et al.,

2014) and investigation of the impact of land use/land cover change on floods (Yigzaw et al., 2013).

Required data for Flood2D-GPU are: 1) DEM for terrain representation; 2) Manning’s roughness; and

3) Flow hydrograph at the source location. The outputs include raw ASCII files of flow depth and

velocity in the x and y directions at various elapsed times during the simulation. These results are post-

processed using a geospatial toolbox within ArcGIS™ in order to determine maximum flood depth and

velocity. A code within MATLAB® is also implemented to determine duration of flooding. The generated

flood parameters feed to the third module, which is described in the next subsection.

2.3 SMCDM Module

This module integrates GIS with a distance-based MCDM technique, Compromise Programming (CP)

(Zeleny, 1973). This technique is utilized because it has a simple computational procedure. Furthermore,

it can directly use the original values of the decision criteria as it does not require the DM judgements to

determine the criteria values (i.e., it is not subjective). In CP, best alternative is the one with the closest

distance to the ‘utopia (or positive ideal solution)’, which provides the best value for each criterion. The

closest solutions to the utopia are named as compromise solutions. The distance from the ideal solution

for each alternative is measured by a distance metric, which is determined based on the following

equation:
n p
f + − fij 1
p| i
Lj = (∑ wi | )p (1)
fi + − fi −
i=1

where, Lj is the distance metric for alternative j, fi+ is the utopia for criterion i, fi- is the negative ideal

value for criterion i, fij is the value of criterion i for alternative j, wi is the weight of criterion i, n is number

of criteria and p is a distance parameter that varies from one to infinity (Zeleny, 1973). Parameter p

8
 Ahmadisharaf et al. / Journal of Hydrology (2016)

represents the importance of maximum deviation from the utopia. The selection of p depends on the type

of problem and desired solution (Tecle et al., 1998). In this study, p is selected to have the value of 2

(simple Euclidean distance) based on the recommendations by the literature (Simonovic, 1989; Pereira

and Duckstein, 2003; Baja et al., 2007; Chitsaz and Banihabib, 2015). Taking this value, each deviation

from the utopia is weighted in proportion to its magnitude.

Eq. (1) is used to determine the distance metric (L) in each grid cell. Thus, a spatial map of L is

produced for each alternative. Performing a cell-based comparison of all generated maps, the minimum

value of L is determined and the corresponding alternative is selected as the best in each grid cell. The

final output is a map showing the spatial variability of best alternative. The map is so-called best

alternative map (BAM).

To implement the SPMCDM, the abovementioned SMCDM procedure is repeated for each random

set of flood modeling results and a BAM is generated in each set. By statistical analysis of all BAMs, a

discrete PDF is produced at each grid cell, which shows the probability of the competing alternatives to

receive the first rank. The following equation is used to compute the probabilities in each grid cell

(Zarghami and Szidarovszky, 2011):

∑nin=1 bin
pj = (2)
n

in which, pj is the probability of alternative j to be best, n is the number of random samples, bin is 1.0

if the best alternative is j and is zero otherwise. Overall the best alternative can be ultimately selected as

any central tendency (i.e., mean, median, mode and so on) of this PDF (Zarghami and Szidarovszky,

2011). Here, mode is selected, which refers to the alternative with the highest pj (i.e., the alternative that

is selected in greater number of random samples). The greater the mode value, the less the prioritization

uncertainty. Additionally, the mode values can be assigned a qualitative uncertainty class in order to give

a more transparent translation of the uncertainty level. For doing so, the mode values are classified into

four different qualitative groups, which are presented in Table 1. Given the classification, a map showing

9
 Ahmadisharaf et al. / Journal of Hydrology (2016)

the spatial variability of qualitative uncertainty level of decision making can be produced. The SPMCDM

framework is demonstrated in the next subsection using a simple illustrative example.

Table 1 Qualitative uncertainty level classes used in this study

2.4 SPMCDM Illustrative Example

The example illustrates how to spatially select the best alternative by applying the SPMCDM framework.

Consider a hypothetical case study with size of 3x3 grid cells, in which four alternatives are evaluated,

including A, B, C and D. Two decision making frameworks are employed to identify the best alternative:

a) deterministic; and b) probabilistic.

2.4.1 Deterministic assessment of flood management alternatives

In the deterministic framework, criteria values are coupled with the criteria weights in the SCP

framework, and a BAM is produced as shown in

A
A A

D C C
D B B
Fig. 2.

A A A
D C C
D B B
Fig. 2 Deterministic best alternative map (BAM)

2.4.2 Probabilistic assessment of flood management alternatives:

In the probabilistic framework, it is assumed that five random samples are used for simplicity reasons.

Therefore, consider five BAMs are produced (one for each random sample). The five BAMs are

presented in

A A B A A A A A B D A B A A B
B C A B C B D C C D C D B C A
D B B D D B D B B D D B D C B

10
 Ahmadisharaf et al. / Journal of Hydrology (2016)

Sample 1 Sample 2 Sample 3 Sample 4 Sample 5

A A B A A A A A B D A B A A B
B C A B C B D C C D C D B C A
D B B D D B D B B D D B D C B
Sample 1 Sample 2 Sample 3 Sample 4 Sample 5
Fig. 3 BAMs using a probabilistic framework

From these BAMs, a map is produced for each alternative, in which the grid cells show the

probability that an alternative receives the first rank. Such a map is generated by repeating Eq. (2) for

each alternative in all the grid cells. For instance, the probability that alternative B receives the first rank

in grid cell (2,1) is calculated as:

1+1+0+0+1
pB = = 60%
5

Repeating these calculations for the four alternatives in all the grid cells, a map showing the

probability of receiving first rank is produced for each alternative. These maps are presented in

80 100 20 0 0 80 0 0 0 20 0 0
% % % % % % % % % % % %
0 0 40 60 0 20 0 100 20 40 0 20
% % % % % % % % % % % %
0 0 0 0 40 100 0 20 0 100 40 0
% % % % % % % % % % % %
Alternative A Alternative B Alternative C Alternative D
Fig. 4.

80 100 20 0 0 80 0 0 0 20 0 0
% % % % % % % % % % % %
0 0 40 60 0 20 0 100 20 40 0 20
% % % % % % % % % % % %
0 0 0 0 40 100 0 20 0 100 40 0
% % % % % % % % % % % %
Alternative A Alternative B Alternative C Alternative D
Fig. 4 Spatial variability of the probability of receiving first rank for the alternatives

Using these four maps, first, the mode value (i.e., highest probability) in each grid cell is determined.

Then, the corresponding alternative of this mode value in each grid cell is identified. This alternative is

the best option in each grid cell (i.e., the alternative with highest probability). It is similar to the final

output of the deterministic framework, but the uncertainty level is also provided by the probabilistic

11
 Ahmadisharaf et al. / Journal of Hydrology (2016)

framework. The overall probabilistic BAM with quantitative uncertainty level, which is the ultimate

output of the probabilistic decision making framework, is presented in

A A B A A B
80% 100% 80% U2 U1 U2
B C A B C A
60% 100% 40% U3 U1 U4
D B/D B D B/D B
100% 40% 100% U1 U4 U1
(a) (b)
Fig. 5 (a).

A A B A A B
80% 100% 80% U2 U1 U2
B C A B C A
60% 100% 40% U3 U1 U4
D B/D B D B/D B
100% 40% 100% U1 U4 U1
(a) (b)
Fig. 5 Overall probabilistic BAM with: a) quantitative uncertainty level; and b) qualitative uncertainty

level

Given the classification in Table 1, an overall BAM with qualitative uncertainty level can be also

generated. Such a map is visualized in

A A B A A B
80% 100% 80% U2 U1 U2
B C A B C A
60% 100% 40% U3 U1 U4
D B/D B D B/D B
100% 40% 100% U1 U4 U1
(a) (b)
Fig. 5 (b), which shows the spatial variability of the uncertainties in decision making for this

illustrative example. Comparing deterministic and probabilistic BAMs (

12
 Ahmadisharaf et al. / Journal of Hydrology (2016)

A A A
D C C
D B B
Fig. 2 vs.

A A B A A B
80% 100% 80% U2 U1 U2
B C A B C A
60% 100% 40% U3 U1 U4
D B/D B D B/D B
100% 40% 100% U1 U4 U1
(a) (b)
Fig. 5), capabilities of the SPMCDM framework can be obviously seen. First, in almost half of the

grid cells ([1,3], [2,1], [2,3] and [3,2]), a different alternative is selected. Selected alternatives in these

grid cells by applying the deterministic framework are A, D, C, and B, respectively. Second, while the

deterministic approach is unable to present the uncertainty of BAM, the SPMCDM frameworks provides

the DM with such valuable information. The uncertainty level in selection of these alternatives is 20%

(U4), 40% (U3), 20% (U4) and 40% (U3), respectively. Thus, in this example, overlooking the

uncertainty of criteria value can be misleading as it significantly affects the decision making results.

Furthermore, in grid cells with the same best alternative, deterministic framework fails to provide the

uncertainty level associated with implementing that alternative due to its inherent limitations.

The probabilistic BAMs enable the DM to understand the uncertainty in selection of an alternative.

The overall uncertainty level by selection of each alternative (which might be also desirable for the DM)

can be also estimated by spatially averaging the probability values of

A A B A A B
80% 100% 80% U2 U1 U2
B C A B C A
60% 100% 40% U3 U1 U4
D B/D B D B/D B
100% 40% 100% U1 U4 U1

13
 Ahmadisharaf et al. / Journal of Hydrology (2016)

(a) (b)
Fig. 5 (a). Doing so, spatial mean value of the probabilities is computed as 78% or overall uncertainty

level is moderately uncertain (U2 class). Thus, to select the best alternative, the DM not only has

exhaustive insight about the effectiveness of an alternative, but also understands the uncertainty of

selecting that alternative.

2.5 Comparison Metrics

To interpret the results in a spatial context, the deterministic and probabilistic BAMs are compared

within the inundation extent using three metrics, namely, F fit measure (Horritt and Bates, 2001), κ

statistic (Cohen, 1960), and Spearman rank correlation coefficient (ρ). These metrics have been

previously employed for relative comparison of rankings by two MCDM methods (e.g., Ahmadisharaf et

al., 2015a). In this study, they are employed for the same purpose. The measures can be computed by the

following expressions:
nj(dm∩pm)
F= (3)
nWet

P0 − PC
κ= (4)
1 − PC

6 ∑m
j=1 dj
ρ=1− (5)
m(m2 − 1)

where F is F fit measure, nj(dm∩dp) is the number of grid cells that alternative j is best in both

deterministic and probabilistic frameworks, nWet is total number of inundation grid cells, κ is coefficient of

agreement, P0 is the proportion of inundation grid cells that are in agreement in two different weight sets,

Pc is the proportion of inundation grid cells that are expected to be in agreement by chance, ρ is Spearman

rank correlation coefficient, m is the number of alternatives and dj is the difference between the ranks of

deterministic and probabilistic frameworks.

When F, κ or ρ takes the value of 1.0, it refers to a perfect agreement between two BAMs. In other

words, the lower these values the bigger the difference between the two BAMs and thus the higher the

significance of the uncertainties impact on the results. F fit measure represents the difference between two

14
 Ahmadisharaf et al. / Journal of Hydrology (2016)

BAMs by determining the number of grid cells that same alternative has been selected by both

deterministic and probabilistic frameworks. Thus, it represents the overall agreement of two BAMs. An F

value of zero indicates that there is no agreement between the two BAMs. The κ statistic indicates the

overall spatial variation of each alternative in BAM. Comparing two BAMs, κ statistic determines the

number of grid cells that both maps are in agreement about a specific alternative whether it is best or not.

A κ value of zero occurs when observed agreement is equal as chance agreement. The ρ measures the

difference between two sets of rankings (Kou et al., 2012). A ρ value of zero refers to complete

disagreement in the ranking by deterministic and probabilistic frameworks. All the three comparison

metrics provide a detailed representation of the variation in BAM.

3 Case Study

The presented framework is demonstrated using the Swannanoa River watershed located in Buncombe

County, the state of North Carolina, USA. The watershed, which is a part of the larger French Broad

River Basin, is located in western North Carolina Mountains, from Asheville to Montreat. The area is

selected due to its proximity to the south eastern coast of the US that exposes it to the potential path of

flood-causing hurricanes and tropical storms. There are developed areas in the watershed with City of

Asheville as the most urbanized area. Fig. 6 shows the study area including the computational domain and

cities as well as the US states and counties. The area has experienced several harmful flooding in the past,

including 1916, 1928, 1940, 1964, 1977 and 2004 events. The most severe flooding occurred in 2004

during hurricanes Francis and Ivan, and caused $54 millions damages to the structures, 11 fatalities as

well as disruption to the communities in the watershed (USACE, 2015). The most severe flooding

occurred in the eastern areas of the watershed, which mostly are not urbanized. This was substantially less

severe by the time it reached the densely populated region of the watershed, city of Asheville. While there

are warning systems in the watershed, no certified flood control reservoir or levee is located in the region.

For this study, the 33.3 km Swannanoa River reach is selected, which is bounded by an area of 173.1 km2,

upstream of the confluence of the Swannanoa River and French Broad River, including some parts of the

city of Asheville.

15
 Ahmadisharaf et al. / Journal of Hydrology (2016)

Fig. 6 Location of the Swannanoa River watershed along with the US states, counties and cities (Image

source: ArcGISTM)

4 Results and Discussion

4.1 Probabilistic Hydrologic Modeling

The probabilistic hydrologic model is applied to simulate rainfall-runoff transformation process. As noted

in the methodology section, the model was developed by extending a deterministic model to probabilistic.

The deterministic hydrologic model is first calibrated on the 1994 August flood event. As Swannanoa

River is an ungauged watershed, and there is only one flow gaging station on the river downstream. Thus,

it is impossible to directly use recorded flow data from an upstream gaging station in the flood modeling

tool. However, there are some rain gauge stations in the study watershed. Hourly rainfall time series taken

from National Climate Data Center (NCDC) are used as input in the GoldSim® model. Moreover, Web-

based Hydrograph Analysis Tool (WHAT) (Lim et al., 2005) is employed to generate hourly time series

of the baseflow. The GoldSim® model efficiency is tested by comparing the predicted hydrograph with

recorded data at Biltmore gaging station (USGS # 03451000) using four goodness-of-fit measures,

including the Nash-Sutcliffe Efficiency (NSE) (Nash and Sutcliffe, 1970), coefficient of determination

16
 Ahmadisharaf et al. / Journal of Hydrology (2016)

(R2), percent bias (PBIAS) and Root-Mean-Square Error (RMSE). CN, ratio of base time to time to peak

and lag time are chosen as the calibration parameters based on the model sensitivity analysis. Simulated

flow hydrograph by GoldSim® model is presented in Fig. 7 along with the observed hydrograph. The

results indicate that the model is in an excellent agreement with the observed flow in with NSE of 90.3%,

R2 of 90.6%, PBIAS of 5.5% and RMSE of 12.4 cms. Additionally, it is noticed that the model slightly

overpredicts the peak value by 0.2%, and slightly underpredicts time to peak and hydrograph volume by

1.9% and 8.4%, respectively. Thus, the remaining of the study is implemented by using this setting of the

GoldSim® model.

180
160 Observed
140
GoldSim
120
Flow (cms)

100
80
60
40
20
0
14-Aug 16-Aug 18-Aug 20-Aug 22-Aug 24-Aug
Time (hr)

Fig. 7 Observed and simulated hydrographs at Biltmore gaging station for August 1994 flood event

To implement the probabilistic model for 100-yr design flood in the study area, 100 LHS samples are

used. This number of random samples is settled because results of such a probabilistic model with a

single stochastic variable will not differ by more than 8% from the predicted value with 90% confidence

level based on Harr (1987). Design rainfall depth is assigned a uniform PDF with minimum and

17
 Ahmadisharaf et al. / Journal of Hydrology (2016)

maximum value of 177.8 to 217.7 mm whilst it has a deterministic value of 197.9 mm. This refers to

±10.1% error in design rainfall depth estimates. Antecedent moisture condition II (normal) is used as it is

typically suggested for the design events (NRCS, 1986). Executing the hydrologic model for the

watershed, which takes about 18 min, results in multiple stochastic flow hydrographs. Statistical analysis

of the given hydrographs shows that the peak value varies from 151.4 to 202.9 cms, while it has a

deterministic magnitude of 177.1 cms. It is to be mentioned that other hydrograph attributes such as time

to peak and duration have no variation in the 100 stochastic samples.

4.2 Hydraulic Modeling

Calibration of the flood model, which is discussed in detail by Ahmadisharaf et al. (2015a), revealed its

satisfactory performance in the study watershed. This process suggested an optimal Manning’s value of

0.05 to represent the surface roughness. In addition to this, randomly produced hydrographs from the first

module are used in tandem with a 23 m spatial resolution DEM generated from the National Map website

(http://nationalmap.gov/) to implement the flood model in the watershed. The spatial resolution is selected

based on Ahmadisharaf et al. (2013) suggestions for Flood2D-GPU application in the study area. A

courant number of 0.15 is used to maintain the model stability. Running the model, which takes about 3

days, results in multiple flood inundation maps for each flood parameter (depth, velocity and duration).

Statistical analysis of these maps indicates that there is 0.6 m increase in maximum flood depth, 0.02 m/s

increase in highest velocity, 8.5 hr increase in duration and 1.8 km2 increase in inundation extent. It is

noteworthy that these numbers are based on the comparison between deterministic simulation with the

most critical flood status (i.e., highest flood depth and velocity, longest flood duration and greatest

inundation extent) among all 100 LHS samples.

4.3 SMCDM Application

The MCDM-based framework is applied for the Swannanoa River watershed to assess a set of flood

management alternatives. Three hypothetical flood management alternatives are analyzed here, which are

three diversion channels along with the base-case model (i.e., without implementation of any flood

management alternative). Location of these mitigation options are presented in Fig. 8. For each

18
 Ahmadisharaf et al. / Journal of Hydrology (2016)

alternative, the DEM is adjusted by burning the grid cells with the depth of the diversion channels in

ArcGISTM. Three flood hazard parameters taken from Flood2D-GPU, including flood depth, velocity, and

duration, are used alongside costs of implementation as the decision criteria. The alternatives are ranked

based on equal weights for the four selection criteria (i.e., weight of 0.25). The weights are determined

subjectively by authors’ opinions and no attempt is made to elicit the stakeholders’ preferences as this

was outside the scope of the work. Two decision making frameworks are considered for assessing the

four alternatives, which are spatial selection of the best alternative by using a: a) deterministic framework;

and b) probabilistic framework using 100 LHS samples. It is to be noted that the results are presented

within the inundation extent. This is because of the fact that a large portion of the case study is unflooded

and therefore implementing the alternatives does not affect these locations. Consequently, the results of

such a comparison will introduce bias (Horritt, 2000). To avoid this type of biased findings, unflooded

area is disregarded and the comparison is conducted within the inundation extent. Moreover, the grid cells

within the channels are excluded from the analysis to avoid biased results. This is because of the fact that

the grid cells within the floodplain are the primary areas of concern for the DM and not those within the

channels.

19
 Ahmadisharaf et al. / Journal of Hydrology (2016)

Fig. 8 Location of the flood management alternatives

4.3.1 Deterministic Assessment of Flood management Alternatives

Applying the SMCDM module, a deterministic BAM is produced, which is visualized in

Fig. 9. A small portion of the inundation extent benefits from the base-case (13.3%) and diversion

channel (9.2%). On the other hand, diversion channel 2 provides benefits in more than half of the

inundation extent (58.4%). Scrutinizing the BAM in the populated regions (city of Asheville), which are

commonly the areas of concern, it can be seen that majority of the grid cells within the city of Asheville

20
 Ahmadisharaf et al. / Journal of Hydrology (2016)

benefits mostly (72.3%) from diversion channel 2. Hence, the overall decision based on the deterministic

assessment is to implement diversion channel 2. However, the uncertainty level by selecting this

alternative remains unclear due to the inherent limitations of the deterministic framework.

Fig. 9 Deterministic best alternative map (BAM)

4.3.2 Probabilistic Assessment of Flood Management Alternatives

Applying the SMCDM module, a BAM is produced for each random sample (total of 100 BAMs). After

statistical analysis of all BAMs, a discrete PDF is created in each grid cell. Four probability-based maps

(one for each alternative) are generated, which show the probability of an alternative to receive the first

rank at each grid cell. These maps are visualized in

(a)

21
Ahmadisharaf et al. / Journal of Hydrology (2016)

(b)

(c)

(d)

22
 Ahmadisharaf et al. / Journal of Hydrology (2016)

(e)

(f)

Fig. 10 (a-d). Using the mode value of the discrete PDFs in each grid cell, overall BAM is determined as

visualized in

(a)

23
Ahmadisharaf et al. / Journal of Hydrology (2016)

(b)

(c)

(d)

24
 Ahmadisharaf et al. / Journal of Hydrology (2016)

(e)

(f)

Fig. 10 (e). Almost no area within the inundation extent benefits from diversion channel 4 (0.9%) and a

small portion benefits from base-case (10.4%) and diversion channel 3 (15.7%). On the other hand,

diversion channel 2 provides the most benefit in a large portion of the inundation extent (73.0%).

Scrutinizing the BAM in the populated regions (city of Asheville), which are commonly the areas of

concern, it can be seen that the majority of these areas benefit from diversion channel 2. This includes the

majority of the grid cells in the city of Asheville (89.1%). Therefore, the overall decision based on the

probabilistic analysis is the implementation of diversion channel 2. Although this is the same as what

found earlier through the deterministic framework, probabilistic analysis refers to a better performance by

the selected alternative. Additionally, it is noteworthy that deterministic analysis overestimates the

performance of the three other alternatives in some cases. For instance, while the deterministic analysis

25
 Ahmadisharaf et al. / Journal of Hydrology (2016)

shows that diversion channel 4 provides the most benefit to 7.2% of the city of Asheville, probabilistic

assessment reveals that almost no part (1.5%) of this region benefits from this alternative.

In contrast to the deterministic approach, uncertainty level by selecting an alternative is also provided

by the probabilistic framework.

(a)

(b)

26
Ahmadisharaf et al. / Journal of Hydrology (2016)

(c)

(d)

(e)

27
 Ahmadisharaf et al. / Journal of Hydrology (2016)

(f)

Fig. 10 (f) shows the spatial variability of the uncertainty level of decision making following Table 1.

From statistical analysis of this map, the percentage of inundation extent with each uncertainty class is

presented in Fig. 11. In 92.8% of the grid cells, a lone definitive best alternative cannot be chosen. In

other words, only in 7.2% of the grid cells, an alternative receives the first rank in all 100 random

samples. Among the grid cells, 39.4% and 13.2% fall into U3 and U4 classes, respectively. In other

words, in more than half (52.6%) of the study area, the decision making is moderately to highly uncertain.

Overall the uncertainty level can be also determined by spatially averaging the corresponding

probabilities of the probabilistic BAM. As a result, spatial average mode value is 61.1%. That is, in

overall, selection of flood management alternatives in the watershed is “moderately uncertain” (i.e., U2

class). Thus, to select one of the feasible alternatives, the DM needs to be very cautious in prioritization

of these alternatives as the best alternative is moderately uncertain. Scrutinizing the BAM in the

populated regions (city of Asheville), the spatial average mode value is 63.0% in the city of Asheville.

That is, in overall, selection of flood management alternatives in the watershed is “moderately uncertain”

(or U2 class). The two decision making approaches are further compared in the next subsections.

28
Ahmadisharaf et al. / Journal of Hydrology (2016)

(a)

(b)

(c)

29
Ahmadisharaf et al. / Journal of Hydrology (2016)

(d)

(e)

(f)

30
 Ahmadisharaf et al. / Journal of Hydrology (2016)

Fig. 10 Outcomes of probabilistic decision making framework: (a) Base-case option probability; (b)

Diversion channel 2 probability; (c) Diversion channel 3 probability; (d) Diversion channel 4 probability;

(e) Overall probabilistic BAM; and (f) Uncertainty level of decision making

39.4%
40%
Percentage of Inundation
35%
30% 28.4%

25%
18.9%
Cells

20%
15% 13.2%

10%
5%
0%
U1 U2 U3 U4
Qualitative Uncertainty Class

Fig. 11 Percentage of different qualitative uncertainty classes in probabilistic BAM

4.3.3 Comparison of Deterministic and Probabilistic Frameworks

Fig. 12 shows the percentage of inundation cells that each alternative receives the first rank by using

deterministic and probabilistic frameworks. In contrast to the deterministic framework, which provides a

single definitive effectiveness level, the probabilistic framework provides a performance range of the

alternatives. It is noted that using the probabilistic framework leads to maximum difference of 77.9%,

55.4%, 75.6%, and 8.9%, in terms of the percentage of inundation cells occupied by the four alternatives,

respectively. The deterministic framework shows nearly the best performance of diversion channel 2,

while it shows approximately the worst performance of the other three alternatives. However, based on

Fig. 12, the effectiveness of the alternatives varies significantly by using the SPMCDM framework and

by incorporating the design rainfall depth uncertainty. A further interesting finding is that the

corresponding probability of the selected alternatives by the deterministic framework has a spatial

average mode value of 50.8%. This implies the fact that making decisions deterministically without

considering the uncertainty of criteria values can be very misleading. It is to be mentioned that the results

31
 Ahmadisharaf et al. / Journal of Hydrology (2016)

presented in Fig. 12 indicate the overall proportion of the grid cells in BAMs that each alternative

receives the first rank (e.g., 56.4% for diversion channel 2 in deterministic framework). These do not

account for the spatial variability of these grid cells. For instance, if an alternative is selected as best in

20% of the inundation cells in two different BAMs (i.e., two different random samples), it does not

essentially imply the both alternatives are equally effective. This is because these grid cells are not

necessarily in the same geographic location. To overcome this limitation in the representation of the

results, the differences in the BAMs are spatially explored through comparison metrics later in subsection

4.4.

97.5%
100% 91.8%
inundation cells covered

90% Probabilistic
80% Minimum
by an alternative

70% 60.5%
Percentage of

Probabilistic
60% 56.4% Maximum
50% Deterministic
40%
30%
20% 21.8% 16.7%
14.0%
10% 6.6% 7.8%
0% 1.3% 1.0% 0.3%
Base-Case Diversion Diversion Diversion
Channel 2 Channel 3 Channel 4
Alternative

Fig. 12 Percentage of inundation cells that each alternative receives first rank by using deterministic and

probabilistic frameworks

4.4 Comparison Metrics

As the ultimate comparison, three metrics are also used to further compare the results of decision making

by applying deterministic and probabilistic frameworks.

To compute the first agreement metric, F fit measure, deterministic and probabilistic BAMs are

spatially compared. Total of five cell-based comparisons are performed in ArcGIS™ and an output raster

is generated in each case. Each grid cell in this map is assigned a value of zero if it has same alternative as

best and 1.0 if it has a different alternative. The number of grid cells with same alternative (value of zero)

32
 Ahmadisharaf et al. / Journal of Hydrology (2016)

present the change in the BAM and refers to nj(dm∩pm) parameter in Eq. (3). Comparing the deterministic

scenario with probabilistic framework, an F value of 62.6% is found. This indicates 37.4% change in the

BAM by employment of the probabilistic framework. Bearing in mind the ±10.1% rainfall depth error,

37.4% change in BAM is considerable. The values of F fit measure also imply a remarkable change in the

BAM that addresses a potential uncertainty in the decision making by disregarding the design rainfall

uncertainty.

The κ measure is determined based on the Eq. (4) and then a relative accuracy matrix is built. For

each alternative, deterministic and probabilistic BAMs are compared. Each comparison includes the

determination of the following four quantities for each alternative (i.e. total of 16 quantifications): 1)

number of grid cells that both two frameworks accept an alternative as best; 2) number of grid cells that

first framework selects an alternative as best and the other one does not; 3) number of grid cells that

second framework selects an alternative as best and the other one does not; and 4) number of grid cells

that both two frameworks reject an alternative as best. Following Eq. (4), κ values are generated in each

case and summarized in Fig. 13. The κ values on the plot indicate the relative difference between

deterministic and probabilistic frameworks. Comparing the deterministic scenario with probabilistic, κ

value varies between 15.4% (base-case) to 45.0% (diversion channel 3). The small values of κ show that

the deterministic and probabilistic frameworks are not in consensus about selection of the alternatives and

decision making results vary greatly by employment of the developed probabilistic framework. Therefore,

the second comparison metric, κ likewise highlights the significance of incorporating design rainfall

uncertainty into the evaluation of flood management alternatives.

33
 Ahmadisharaf et al. / Journal of Hydrology (2016)

50% 45.0%
45%
40%
35% 32.2%
30%

κ (%)
25%
20% 15.4%
15%
8.0%
10%
5%
0%
Base-Case Diversion Diversion Diversion
Channel 2 Channel 3 Channel 4
Alternative

Fig. 13 κ values of the four alternatives between deterministic and probabilistic frameworks

The ρ is determined based on the Eq. (5). This is conducted by comparison of deterministic and

probabilistic BAMs. A map that shows spatial variability of ρ values is generated. A spatial mean ρ value

of 0.48 is computed, which refers to a large difference between the ranking by these two frameworks.

Also, the map is further statistically analyzed in order to determine the variation in ρ value and the values

are summarized in Fig. 14. The ρ values on the plot indicate the relative difference between the ranking

by deterministic and probabilistic frameworks. The frameworks are in perfect agreement in only 16.2% of

the inundation cells (i.e., ρ value of +1). In nearly one-third (28.9%) of the inundation cells, there is a

weak agreement (i.e., ρ value of less than +0.4) in the rankings by deterministic and probabilistic

frameworks. Moreover, there is a negative ρ value in 10.7% of the inundation cells, suggesting that the

rankings by these frameworks are inversely correlated. These numbers show that the deterministic and

probabilistic frameworks are not in an acceptable agreement. This also implies the significant impact of

design rainfall uncertainty on the decision making. Therefore, by neglecting the uncertainty of the design

rainfall depth, prioritization can be changed significantly.

34
 Ahmadisharaf et al. / Journal of Hydrology (2016)

29.4%
30%

Inundation Cells
25%

Percentage of
21.5%
20% 16.2%
14.3%
15%
10%
4.2% 3.9% 4.1%
5% 2.1% 1.4% 2.5%
0.5%
0%
-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
ρ

Fig. 14 Distribution of ρ values between deterministic and probabilistic frameworks

5 Summary and Conclusions

Flood management requires that all sources of uncertainty being addressed properly (Akter and

Simonovic, 2005). In this study, a SPMCDM framework was developed for assessment of flood

management alternatives. In contrast to the conventional deterministic decision making framework, this

study incorporates the uncertainty of design rainfall depth. The competing alternatives were evaluated by

using both deterministic and probabilistic SMCDM frameworks. Comparison of the results generated by

the two frameworks, highlighted the significant impact of design rainfall uncertainty on decision making.

While the deterministic framework fails to provide the uncertainty level associated with implementing an

alternative, the SPMCDM framework showed that in overall, selection of flood management alternatives

in the watershed is “moderately uncertain”. Thus, to select one of the feasible alternatives, the DM needs

to be cautious in prioritization of these alternatives as the best alternative is moderately uncertain. The

SPMCDM framework can help the DMs to understand the uncertainty associated with the selection of a

flood management option.

For many years, planners and DMs have employed deterministic frameworks in flood management.

While this approach is simple and understandable, it fails to adequately account for the uncertainty in

decision making. Overlooking the uncertainty of performance values might lead to incomplete

representation of the effectiveness of flood management alternatives and subsequent selection of

35
 Ahmadisharaf et al. / Journal of Hydrology (2016)

suboptimal alternatives (Meyer et al., 2009). As a result, millions of dollars might be wasted and people

life might be threatened. The uncertainty, which stems from different sources, is now receiving more

attention due to the impacts of nonstationary factors such as climate change and urbanization (Hutter and

Schanze, 2008). Hirsch (2011) suggested establishing a whole new decision making approach to account

for the nonstationarity. US Army Corps of Engineers’ (USACE) Flood Risk Management Program

(FRMP) and Federal Flood Risk Management Standard (FFRMS) needs the uncertainties attributed to

future changes being incorporated into the decision making by utilizing the best-available

hydrologic/hydraulic data and models that integrate current and future changes in flooding. With the

recent shift to a more sustainable flood management (SFM) approach, current decision making approach

needs to be updated to comply its key requirement (Kang et al., 2013b). In particular, in a long-term point

of view, which is a key element of SFM, uncertainty needs to be factored into the flood management

because the future is inherently uncertain (Edjossan-Sossou et al., 2014). There is thus an urgent need to

incorporate tools that are capable of uncertainty modeling, into the current decision making frameworks.

In this context, probabilistically-aided decision making frameworks are valuable due to the capabilities to

handle uncertainty in a scientifically reliable manner. There is a growing interest in modeling uncertainty

through probabilistic frameworks due to the inherent limitations of deterministic frameworks and

capabilities of probabilistic frameworks such as consideration of the interactions between uncertain

elements. Recent tremendous computational enhancement in flood models, enable DMs to utilize more

sophisticated tools in the decisions. While such probabilistic frameworks have been applied in

environmental and water resources problems, they have not been received any attention in the context of

flood management yet. The integrated framework advances the state of decision making, and introduces a

new method to explicitly account for uncertainties in decisions. It can therefore greatly assist DMs to

incorporate uncertainty into the flood management and make well-informed decisions through a

scientifically reliable manner. A better understanding of the uncertainty impacts can be used to develop

new strategies for protecting flood-prone areas from destructive flooding. This will guide future policy

36
 Ahmadisharaf et al. / Journal of Hydrology (2016)

and planning decisions, and promote more reliable flood management. The results will serve as an

example to floodplain managers and DMs working on complex floodplain systems.

Although the objectives of the present study are ambitious and attempt to begin a new approach in

flood management decisions, there are some limitations, which should be elaborated as follows in the

future.

1) Only the design rainfall depth uncertainty was incorporated in decision making framework. First,

given confidence limits on the design rainfall depth (which were the basis of uncertainty characterization

in our study) in Bonnin et al. (2004) have been estimated under the assumption that the data quality is

good. Uncertainties attributed to measurement errors and spatial interpolation are not included in those

estimations. Second, uncertainties arise from other sources were not taken into account in our study. It is

recommended that the impacts of the uncertainty of other parameters such as rainfall distribution, channel

and watershed characteristics as well as DEM on decision making are explored. In addition, all three

modeling tools of the SPMCDM framework have uncertainty. Not only intra-model, but also the selection

of the models itself. In the future, other hydrologic models such as HEC-HMS, TOPMODEL and etc.;

hydraulic models such as HEC-RAS, LISFLOOD-FP and etc.; and other MCDM techniques such as

TOPSIS, can be employed in order to verify and/or corroborate current study findings.

2) Design rainfall depth values that were used in this study, have been developed under the stationary

assumption (Bonnin et al., 2004). In order to consider this type of nonstationarity factors such as climate

change in decision making, the developed framework can be utilized. Moving forward to the future

research direction, the authors are currently working on applying the SPMCDM framework to make

decisions under climate change.

3) Design rainfall uncertainty was characterized through a uniform PDF. In general, the selection of

PDF itself can be a source of uncertainty (Alcamo and Bartnicki, 1987). This can be a source of

uncertainty in LHS application, which might not be negligible (Zagonjolli, 2007). Therefore, other PDFs

such as normal and triangular distributions are recommended to be tested and the decision making results

can be compared with current findings.

37
 Ahmadisharaf et al. / Journal of Hydrology (2016)

4) Only diversion channels were evaluated as flood management alternatives in present study. To

supplement the conclusions, other mitigation measures such as floodwalls and detention basins, should be

considered. By doing so, a better understanding of the significance of design rainfall uncertainty on

selection of flood management measures can be possible.

5) Criteria selection is critical and can affect the outcomes of decision making problem (Neves et al.,

2009; Chen et al., 2010). In this study, only three flood parameters, depth, velocity, and duration, along

with costs of implementation, were considered as decision criteria. Other flood parameters, including but

not limited to, arrival time and rate of rise can be added into the criteria set in order to investigate the

difference in decision making results.

6) Due to the important role of criteria weights in MCDM, weighting is determined by using

stakeholders and experts’ opinions (Yeh et al., 1999; Munda, 2006; Chen et al., 2011a). In addition to

this, sensitivity of the decision making outcomes to criteria weights should be analyzed to evaluate the

robustness of the selected alternative (Kaya and Kahraman, 2011). Nevertheless, these were out of the

scope of this paper and the weights were determined based on the authors’ preferences here in order to

show the general procedure. Consequently, these results should not be considered as the definitive and

final BAM. Weighing should be undertaken along with the specialists to incorporate their preferences and

to validate the results of this study and to address the related uncertainties. This can be performed by

using some interviews, questionnaires and workshops.

Acknowledgments

We are grateful to the financial support by the Center of Management, Utilization, and Protection of

Water Resources. We really appreciate Jason Lillywhite for his technical help in the development of the

hydrologic model within GoldSim® environment.

References

1. Ahmadisharaf, E., Bhuyian, M., Kalyanapu, A.J., 2013. Impact of spatial resolution on downstream

flood hazard due to dam break events using probabilistic flood modeling. 5th Dam Safety

Conference, Providence, RI. pp. 263-276.

38
 Ahmadisharaf et al. / Journal of Hydrology (2016)

2. Ahmadisharaf, E., Kalyanapu, A.J., 2015. Investigation of the impact of streamflow temporal

variation on dam overtopping risk: case study of a high-hazard dam. World Env. Water Resour.

Cong., Austin, TX. pp. 1050-1057.

3. Ahmadisharaf, E., Kalyanapu, A.J., Chung, E.S., 2015a. Evaluating the effects of inundation duration

and velocity on selection of flood management alternatives. Water Resour. Manage. 29(8), 2543-

2561.

4. Ahmadisharaf, E., Tajrishy, M., Alamdari, N., 2015b. Integrating flood hazard into site selection of

detention basins using spatial multi-criteria decision making. J. Environ. Plann. Manage., DOI:

10.1080/09640568.2015.1077104.

5. Akter, T., Simonovic, S.P., 2005. Aggregation of fuzzy views of a large number of stakeholders for

multi-objective flood management decision-making. J. Environ. Manage. 77(2), 133-143.

6. Alcamo, J., Bartnicki, J., 1987. A framework for error analysis of a long-range transport model with

emphasis on parameter uncertainty. Atmos. Environ. 21(10), 2121-2131.

7. Aronica, G.T., Franza, F., Bates, P.D., Neal, J.C., 2012. Probabilistic evaluation of flood hazard in

urban areas using Monte Carlo simulation. Hydrol. Processes 26(26), 3962-3972.

8. Ascough, J.C., Maier, H.R., Ravalico, J.K., Strudley, M.W., 2008. Future research challenges for

incorporation of uncertainty in environmental and ecological decision-making. Ecol. Modell. 219(3),

383-399.

9. Baja, S., Chapman, D.M., Dragovich, D., 2007. Spatial based compromise programming for multiple

criteria decision making in land use planning. Environ. Model. Assess. 12(3), 171-184.

10. Bana E Costa, C.A., Da Silva, P.A., Correia, F.N., 2004. Multicriteria evaluation of flood control

measures: The case of Ribeira do Livramento. Water Resour. Manage. 18(3), 263-283.

11. Bhuyian, M., Kalyanapu, A.J., Nardi, F., 2015. Approach to Digital Elevation Model correction by

improving channel conveyance. J. Hydrol. Eng. 20(5), 04014062.

39
 Ahmadisharaf et al. / Journal of Hydrology (2016)

12. Bonnin, G., Todd, D., Lin, B., Parzybok, T., Yekta, M., Riley. D., 2004. Precipitation frequency atlas

of the United States. National Oceanic and Atmospheric Administration (NOAA) Atlas 14.2, third

ed., Silver Springs, MD.

13. Brouwer, R., Van Ek, R., 2004. Integrated ecological, economic and social impact assessment of

alternative flood control policies in the Netherlands. Ecol. Econ. 50(1), 1-21.

14. Chen, H., Wood, M.D., Linstead, C., Maltby, E., 2011a. Uncertainty analysis in a GIS-based multi-

criteria analysis tool for river catchment management. J. Environ. Model. Softw. 26(4):395-405.

15. Chen, Y., Yu, J., Khan, S., 2010. Spatial sensitivity analysis of multi-criteria weights in GIS-based

land suitability evaluation. Environ. Model. Soft. 25(12), 1582-1591.

16. Chen, Y.R., Yeh, C.H., Yu, B., 2011b. Integrated application of the analytic hierarchy process and the

geographic information system for flood risk assessment and flood plain management in Taiwan. Nat.

Hazards 59(3), 1261-1276.

17. Chitsaz, N., Banihabib, M.E., 2015. Comparison of different multi criteria decision-making models in

prioritizing flood management alternatives. Water Resour. Manage. 29(8), 2503-2525.

18. Cohen, J., 1960. A coefficient of agreement for nominal scales. Educ. Psycholog. Meas. 20, 37-46.

19. Di Baldassarre, G., Schumann, G., Bates, P.D., Freer, J.E., Beven, K.J., 2010. Flood-plain mapping: a

critical discussion of deterministic and probabilistic approaches. Hydrol. Sci. J. 55(3), 364–376.

20. Edjossan-Sossou, A.M., Deck, O., Al Heib, M., Verdel, T., 2014. A decision-support methodology

for assessing the sustainability of natural risk management strategies in urban areas. Nat. Hazards

Earth Syst. Sci. 14(12), 3207-3230.

21. Fernandez, D.S., Lutz, M.A., 2010. Urban flood hazard zoning in Tucuman Province, Argentina,

using GIS and multicriteria decision analysis. Eng. Geol. 111(1), 90-98.

22. Foudi, S., Osés-Eraso, N., Tamayo, I., 2015. Integrated spatial flood risk assessment: The case of

Zaragoza. Land Use Policy 42, 278-292.

23. Hall, J., Solomatine, D., 2008. A framework for uncertainty analysis in flood risk management

decisions. Int. J. River Basin Manage. 6(2), 85-98.

40
 Ahmadisharaf et al. / Journal of Hydrology (2016)

24. Hall, J.W., Tarantola, S., Bates, P.D., Horritt, M.S., 2005. Distributed sensitivity analysis of flood

inundation model calibration. J. Hydraul. Eng. 131(2), 117-126.

25. Harr, M.E. 1987. Reliability-based design in civil engineering. McGraw–Hill, New York, NY.

26. Helton, J.C., Davis, F.J., 2003. Latin hypercube sampling and the propagation of uncertainty in

analyses of complex systems. Reliab. Eng. Syst. Saf. 81(1), 23-69.

27. Hirsch, R.M., 2011. A perspective on nonstationarity and water management. J. Am. Water Resour.

Assoc. 47(3), 436-446.

28. Horritt, M.S., 2000. Calibration of a two‐dimensional finite element flood flow model using satellite

radar imagery. Water Resour. Res. 36(11), 3279-3291.

29. Horritt, M.S., Bates, P.D., 2001. Effects of spatial resolution on a raster based model of flood flow. J.

Hydrol. 253(1), 239-249.

30. Hutter, G., Schanze, J., 2008. Learning how to deal with uncertainty of flood risk in long‐term

planning. Int. J. River Basin Manage. 6(2), 175-184.

31. Hyde, K., Maier, H.R., Colby, C., 2003. Incorporating uncertainty in the PROMETHEE MCDA

method. J. Multi‐Criteria Decis. Anal. 12(4‐5), 245-259.

32. Hyde, K.M., Maier, H.R., Colby, C.B., 2004. Reliability-based approach to multicriteria decision

analysis for water resources. J. Water Resour. Plann. Manage. 130(6), 429-438.

33. Janssen, H., 2013. Monte-Carlo based uncertainty analysis: Sampling efficiency and sampling

convergence. Reliab. Eng. Syst. Saf. 109, 123-132.

34. Kalyanapu, A.J., Hossain, A.A., Kim, J., Yigzaw, W., Hossain, F., Shum, C.K., 2013. Toward a

methodology to investigate the downstream flood hazards on the American River due to changes in

probable maximum flood due to effects of artificial reservoir size and land-use/land-cover patterns.

Earth Interact. 17(24), 1-24.

35. Kalyanapu, A.J., Judi, D.R., McPherson, T.N., Burian, S.J., 2012. Monte Carlo-based flood

modelling framework for estimating probability weighted flood risk. J. Flood Risk Manage. 5(1), 37-

48.

41
 Ahmadisharaf et al. / Journal of Hydrology (2016)

36. Kalyanapu, A.J., Judi, D.R., McPherson, T.N., Burian, S.J., 2015. Annualised risk analysis approach

to recommend appropriate level of flood control: application to Swannanoa river watershed. J. Flood

Risk Manage. 8(4), 368-385.

37. Kalyanapu, A.J., Shankar, S., Pardyjak, E.R., Judi, D.R., Burian, S.J., 2011. Assessment of GPU

computational enhancement to a 2D flood model. J. Environ. Model. Softw. 26(8), 1009-1016.

38. Kang, B.S., Lee, J.H., Chung, E.S., Kim, D.S., Kim, Y.D., 2013a. A sensitivity analysis approach of

multi-attribute decision making technique to rank flood mitigation projects. KSCE J. Civ. Eng. 17(6),

1529-1539.

39. Kang, M.G., Jeong, H.S., Lee, J.H., Kang, B.S., 2013b. Assessing national flood management using a

sustainable flood management framework. Water Policy 15, 418-434.

40. Kaya, T., Kahraman, C., 2011. An integrated fuzzy AHP–ELECTRE methodology for environmental

impact assessment. Expert Syst. Appl. 38(7), 8553-8562.

41. Kim, Y., Chung, E.S., 2014. An index-based robust decision making framework for watershed

management in a changing climate. Sci. Tot. Env. 473, 88-102.

42. Kim, Y., Chung, E.S., Jun, S.M., 2015. Iterative framework for robust reclaimed wastewater

allocation in a changing environment using multi-criteria decision making. Water Resour. Manage.

29(2), 295-311.

43. Kou, G., Lu, Y., Peng, Y., Shi, Y., 2012. Evaluation of classification algorithms using MCDM and

rank correlation. Int. J. Inf. Technol. Decis. Making 11(1), 197-225.

44. Kourgialas, N.N., Karatzas, G.P., 2012. Flood management and a GIS modeling method to assess

flood hazard areas-a case study. Hydrol. Sci. J. 56(2), 212-225.

45. Krzysztofowicz, R., 1999. Bayesian theory of probabilistic forecasting via deterministic hydrologic

model. Water Resour. Res. 35(9), 2739-2750.

46. Krzysztofowicz, R. 2001. The case for probabilistic forecasting in hydrology. J. Hydrol. 249(1), 2-9.

47. Krzysztofowicz, R., Herr, H.D., 2001. Hydrologic uncertainty processor for probabilistic river stage

forecasting: precipitation-dependent model. J. Hydrol. 249(1), 46-68.

42
 Ahmadisharaf et al. / Journal of Hydrology (2016)

48. Krzysztofowicz, R., Kelly, K.S., 2000. Hydrologic uncertainty processor for probabilistic river stage

forecasting. Water Resour. Res. 36(11), 3265-3277.

49. Lee, G.M., Jun, K.S., Chung, E.S., 2013. Integrated multi-criteria flood vulnerability approach using

Fuzzy TOPSIS and Delphi technique. Nat. Hazards Earth Sys. Sci. 13(5), 1293-1312.

50. Lee, G., Jun, K.S., Chung, E.S., 2015. Group decision-making approach for flood vulnerability

identification using the fuzzy VIKOR method. Nat. Hazards Earth Sys. Sci. 15(4), 863-874.

51. Levy, J.K., 2005. Multiple criteria decision making and decision support systems for flood risk

management. Stoch. Environ. Res. Risk Assess. 19(6), 438-447.

52. Levy, J.K., Hartmann, J., Li, K.W., An, Y., Asgary, A., 2007. Multi-Criteria Decision Support

Systems for flood hazard mitigation and emergency response in urban watersheds. J. Am. Water

Resour. Assoc. 43(2), 346-358.

53. Lim, K.J., Engel, B.A., Tang, Z., Choi, J., Kim, K.S., Muthukrishnan, S., Tripathy, D., 2005.

Automated web GIS based hydrograph analysis tool, WHAT. J. Am. Water Resour. Assoc. 41(6),

1407-1416.

54. Madani, K., Lund, J.R., 2011. A Monte-Carlo game theoretic approach for multi-criteria decision

making under uncertainty. Adv. Water Resour. 34(5), 607-616.

55. McKay, M.D., Beckman, R.J., Conover, W.J., 1979. Comparison of three methods for selecting

values of input variables in the analysis of output from a computer code. Technometrics 21(2), 239-

245.

56. Malekian, A., Azarnivand, A., 2015. Application of integrated Shannon’s Entropy and VIKOR

techniques in prioritization of flood risk in the Shemshak Watershed, Iran. Water Resour. Manage.

DOI: 10.1007/s11269-015-1169-6.

57. McMillan, H., Freer, J., Pappenberger, F., Krueger, T., Clark, M., 2010. Impacts of uncertain river

flow data on rainfall‐runoff model calibration and discharge predictions. Hydrol. Processes 24(10),

1270-1284.

43
 Ahmadisharaf et al. / Journal of Hydrology (2016)

58. Merwade, V., Olivera, F., Arabi, M., Edleman, S., 2008. Uncertainty in flood inundation mapping:

current issues and future directions. J. Hydrol. Eng. 13(7), 608-620.

59. Meyer, V., Scheuer, S., Haase, D., 2009. A multicriteria approach for flood risk mapping exemplified

at the Muddle River, Germany. Nat. Hazards 48, 17-39.

60. Mosadeghi, R., Warnken, J., Tomlinson, R., Mirfenderesk, H., 2013. Uncertainty analysis in the

application of multi-criteria decision-making methods in Australian strategic environmental

decisions. J. Environ. Plann. Manage. 56(8), 1097-1124.

61. Munda, G., 2006. Social multi-criteria evaluation for urban sustainability policies. Land Use Policy

23(1), 86–94.

62. NRCS, 1986. Urban hydrology for small watersheds. Conservation Engineering Division, NRCS, US

Department of Agriculture. Technical Release 55. Washington, DC.

63. Neves, L.P., Dias, L.C., Antunes, C.H., Martins, A.G., 2009. Structuring an MCDA model using

SSM: A case study in energy efficiency. Eur. J. Oper. Res. 199(3), 834-845.

64. Papaioannou, G., Vasiliades, L., Loukas, A., 2015. Multi-criteria analysis framework for potential

flood prone areas mapping. Water Resour. Manage. 29(2), 399-418.

65. Pappenberger, F., Beven, K.J., 2006. Ignorance is bliss: Or seven reasons not to use uncertainty

analysis. Water Resour. Res. 42(5), W05302.

66. Pappenberger, F., Matgen, P., Beven, K.J., Henry, J.B., Pfister, L., 2006. Influence of uncertain

boundary conditions and model structure on flood inundation predictions. Adv. Water Resour. 29(10),

1430-1449.

67. Pappenberger, F., Stephens, E., Thielen, J., Salamon, P., Demeritt, D., vanAndel, S.J., Watterhall F.,

Alfieri, L., 2013. Visualizing probabilistic flood forecast information: expert preferences and

perceptions of best practice in uncertainty communication. Hydrol. Processes 27(1), 132-146.

68. Pereira, J.M., Duckstein, L., 1993. A multiple criteria decision-making approach to GIS-based land

suitability evaluation. Int. J. Geog. Inf. Sci. 7(5), 407-424.

44
 Ahmadisharaf et al. / Journal of Hydrology (2016)

69. Qi, H., Qi, P., Altinakar, M.S., 2013. GIS-based spatial Monte Carlo analysis for integrated flood

management with two dimensional flood simulation. Water Resour. Manage. 27, 3631-3645.

70. Radmehr, A., Araghinejad, S., 2014. Developing strategies for urban flood management of Tehran city

using SMCDM and ANN. J. Comput. Civ. Eng. 28(6), 05014006.

71. Radmehr, A., Araghinejad, S., 2015. Flood vulnerability analysis by fuzzy spatial multi criteria decision

making. Water Resour. Manage. 29(12), 4427-4445.

72. Rahmati, O., Zeinivand, H., Besharat, M., 2015. Flood hazard zoning in Yasooj region, Iran, using

GIS and multi-criteria decision analysis. Geomatics Nat. Hazards Risk. DOI:

10.1080/19475705.2015.1045043.

73. Ronco, P., Gallina, V., Torresan, S., Zabeo, A., Semenzin, E., Critto, A., Marcomini, A., 2014. The

KULTURisk Regional Risk Assessment methodology for water-related natural hazards–Part 1:

Physical–environmental assessment. Hydrol. Earth Syst. Sci. 18(12), 5399-5414.

74. Rosner, A., Vogel, R.M., Kirshen, P.H., 2014. A risk‐based approach to flood management decisions

in a nonstationary world. Water Resour. Res. 50(3), 1928-1942.

75. Schröter, K., Kreibich, H., Vogel, K., Riggelsen, C., Scherbaum, F., Merz, B., 2014. How useful are

complex flood damage models? Water Resour. Res. 50(4), 3378-3395.

76. Simonovic, R.J., 1989. Application of water resources systems concept to the formulation of a water

master plan. Water Int. 14(1), 37–50.

77. Sinha, R., Bapalu, G.V., Singh, L.K., Rath, B., 2008. Flood risk analysis in the Kosi River Basin,

North Bihar using multi-parametric approach of Analytical Hierarchy Process (AHP). J. Indian Soc.

Remote Sens. 36(4), 335-349.

78. Smemoe, C.M., Nelson, E.J., Zundel, A.K., Miller, A.W., 2007. Demonstrating floodplain uncertainty

using flood probability maps. J. Am. Water Resour. Assoc. 43(2), 359-371.

79. Snyder, F.F., 1938. Synthetic unit‐graphs. Eos, Trans. Am. Geophys. Union 19(1), 447-454.

80. Stefanidis, S., Stathis, D., 2013. Assessment of flood hazard based on natural and anthropogenic

factors using analytic hierarchy process (AHP). Nat. Hazards 68(2), 569-585.

45
 Ahmadisharaf et al. / Journal of Hydrology (2016)

81. Tecle, A., Shrestha, B.P., Duckstein, L., 1998. A multiobjective decision support system for

multiresource forest management. Group Decis. Negotiation 7(1), 23-40.

82. Tkach, R.J., Simonovic R.J., 1997. A new approach to multi-criteria decision-making in water

resources. J. Geogr. Inf. Decis. Anal. 1, 25-44.

83. Tung, Y.K., 2011. Uncertainty and reliability analysis in water resources engineering. J. Contemp.

Water Res. Educ. 103(1), 13-21.

84. USACE, 2015. Swannanoa River Watershed Flood Risk Reduction Project. Available at:

http://www.lrn.usace.army.mil/Media/FactSheets/FactSheetArticleView/tabid/6992/Article/562061/s

wannanoa-river-watershed-flood-risk-reduction-project.aspx (Accessed June 23, 2015)

85. Willette, K., Sharda, R., 1991. Using the Analytic Hierarchy Process in water resources planning

selection of flood control projects. Socio-Econ. Plann. Sci. 25(2), 103-112.

86. Yeh, C.H., Willis, R.J., Deng, H., Pan, H., 1999. Task oriented weighting in multi-criteria analysis.

Eur. J. Oper. Res. 119(1), 130-146.

87. Yigzaw, W., Hossain, F., Kalyanapu, A.J., 2013. Comparison of PMP-driven probable maximum

floods with flood magnitudes due to increasingly urbanized catchment: The Case of American River

watershed. Earth Interact. 17(8), 1-15.

88. Zagonjolli, M., 2007. Dam break modelling, risk assessment and uncertainty analysis for flood

mitigation, Ph.D. Dissertation, Delft Univ. of Technol., Delft, Netherlands.

89. Zarghami, M., Szidarovszky, F., 2011. MCDA problems under uncertainty. in: Multicriteria analysis.

Springer-Verlag, Berlin, pp. 113-147.

90. Zeleny, M., 1973. Compromise Programming. in: Cochrane, J.L., Zeleny, M. (Eds.) Multiple criteria

decision making. University of South Carolina Press, Columbia, SC, pp. 262-301.

91. Zou, Q., Zhou, J., Zhou, C., Song, L., Guo, J., 2012. Comprehensive flood risk assessment based on

set pair analysis-variable fuzzy sets model and fuzzy AHP. Stoch. Environ. Res. Risk Assess. 27(2),

525-546.

46