**Contents**


## **About the Editor**

**Tiago Miguel Ferreira** holds a PhD in Civil Engineering. His research domains include the evaluation and mitigation of urban risks—including seismic, fire, and flood—and the structural safety assessment and retrofit of ancient stone masonry buildings. He is currently a Junior Researcher at the Institute for Sustainability and Innovation in Structural Engineering (ISISE) of the University of Minho, invited Assistant Professor at the Department of Civil Engineering of the University of Coimbra, and Editor-in-Chief of *Conservar Patrim´onio*, a SCOPUS and Web-of-Science indexed journal dedicated to heritage studies.

## *Editorial* **Recent Advances in the Assessment of Flood Risk in Urban Areas**

#### **Tiago Miguel Ferreira**

ISISE, Institute of Science and Innovation for Bio-Sustainability (IB-S), Department of Civil Engineering, University of Minho, 4800-058 Guimarães, Portugal; tmferreira@civil.uminho.pt

Received: 23 June 2020; Accepted: 25 June 2020; Published: 29 June 2020

The adverse effects of flood disasters in urban areas have been increasing in severity and extent over the past years. The amounts of losses resulting from these events are also increasing exponentially, particularly in highly urbanised urban areas, where the effects of intensive land use and climate change are particularly extreme [1]. All this despite our scientific knowledge, technical competence and computational capacity to develop highly sophisticated and accurate forecasting and simulation models being higher than ever, as is our capacity to map and analyse flood-related data.

In order to tackle this global issue, it is fundamental to keep on promoting and developing fundamental and applied research that allows the better targeting of interventions to improve resilience, reduce vulnerability and enhance recovery, as well as assisting decision-makers in delivering more effective flood risk-reduction policies. The present Special Issue of Water aims to contribute to this goal by providing a space in which to share and discuss recent studies and state-of-the-art methodologies focused on the assessment and mitigation of flood risk in urban areas. It includes nine high-quality research articles authored by eminent scholars from India, Italy, Korea, Portugal, Romania, Singapore, Spain, Taiwan, Thailand and Vietnam, who had the tremendous generosity to join me in this project. The range of topics covered by these nine studies is extraordinarily vast, reflecting the complexity of the current challenges associated with the topic.

Lee et al. [2] present an interesting discussion on the role played by drainage facilities in urban flooding. Based on the analysis of a series of past flooding events, the authors propose a resilience index to be used to diagnose the status of urban drainage systems. Such an indicator, easily calculated using real-time rainfall data, allows for the identification of intervention needs and for the evaluation of the impact of structural and non-structural interventions aimed at reducing urban inundation—thus providing a proper means of supporting more efficient decision-making processes, fostering urban resilience.

Afifi et al. [3] propose a GIS-based high-resolution flood loss and risk assessment model to analyse the impacts of flooding events in residential urban areas. For such, different flood loss, hazard and vulnerability indicators—such as water depth, elevation, distance to first responders and population density—are combined using an Analytic Hierarchy Process (AHP). After being comprehensively presented and discussed, the model is used to generate a series of hazard, vulnerability and flood risk maps for the city of Tainan in Taiwan. Among other interesting particulars, the authors provide a critical discussion on the effects (measured in terms of estimated losses) of different spatial model resolutions.

Park and Lee [4] present a study aimed at contributing to the reduction and minimisation of flood damage in the case of heavy rainfall in urban areas, by increasing urban spatial efficiency through the grading of flood risk. To this end, the authors suggest an urban flood risk assessment model able to account for different urban planning elements, namely the land use and building characteristics. The capacities and the potential of the model are discussed through its application to Changwon City in Korea. The role played by land-use planning measures to prevent flooding, as well as the possibility of classifying flood risk areas according to the land use of the districts and of evaluating intervention priorities by selecting high-risk areas, are some of the aspects discussed by the authors.

Littidej and Buasri [5] analyse the way in which the Digital Elevation Model (DEM) and Flood Risk Susceptibility (FRS) prediction models are impacted by urban growth in Muaeng District, Nakhon Ratchasima (Thailand). The scope of this study comprises three main objectives: to optimise the Cellular Automata (CA) model for predicting the expansion of built-up sites, to model a linear regression method for deriving the transition of the digital elevation model (DEM), and to apply a Geographic Weighted Regression (GWR) for analysing the risk of the stativity of flood areas in the province. Among other noteworthy outcomes, this study shows that the CA model can accurately predict the expansion of built-up areas using land-use data updated at 2-year intervals.

Mihu-Pintilie et al. [6] develop a method for flood vulnerability assessment under real (average discharge) and mathematical (calculated discharge) hydrological data based on HEC-RAS, high-density LiDAR data, and 2D hydraulic modelling. The authors compute four 2D streamflow hydraulic scenarios in order to test the flood mitigation capacity of hydro-technical constructions located downstream on the Bistri¸ta River, North-East Romania. With this application, the authors show that multi-scenario results obtained from 2D hydraulic modelling can be applied to obtain flood hazard parameters (such as the flood depth, flood extent, flood velocity, or water surface elevation), which can be subsequently used in recognising and responding to flood threats at the local level.

Bernardini and Quagliarini [7] address the important issue of individuals' evacuation in urban flood scenarios by providing general and unified modelling approaches to estimate evacuation speed variations depending on an individual's excitement (walking or running), floodwater depths and individuals' features (age, gender, height and their average speed on dry surfaces). Speed data from previous experiments are organized using linear regression models. Finally, the authors discuss the possible implementation of these models to simulate evacuees' motion in floodwaters (considering different confidence degree levels) and to assess the community's flood risk and the effectiveness of risk-reduction strategies.

Dhara et al. [8] discuss the suitability of using different sources of satellite imagery for capturing the extent of flood inundation in urban areas. In this proof-of-concept study, the authors show that machine-learning algorithms—namely the Support Vector Machine Regression (SVR) technique—can be efficiently used to integrate data from three different satellite sensors (Landsat, MODIS and Sentinel-2) to derive one single integrated value. Because it relies on the use of freely available data, this approach constitutes an economical and efficient alternative, mainly when field data are not readily available, which is particularly frequent in regions with difficult access or with high rates of urban growth. The city of Can Tho (Vietnam) is used to prove the concept and to illustrate the potential of the approach.

Ferreira and Santos [9] discuss the application of an integrated flood risk assessment approach, which combines flood hazard and building vulnerability indicators to identify and classify risk and to narrow intervention priorities. The Historic City Centre of Guimarães, in Portugal, is exploited by the authors to illustrate the application of the methodology. After modelling the flood hazard using the hydrologic–hydraulic method and evaluating the flood vulnerability of the buildings by resorting to a simplified vulnerability assessment method, they provide a comprehensive analysis of the outputs in both an individual and integrated manner. Finally, they also use a risk-matrix approach to aggregate these hazard and vulnerability outputs and categorise the buildings into different qualitative levels of risk.

Arosio et al. [10] analyse the direct and indirect impacts of rainfall flooding in Mexico City. To do so, they resort to a graph-based methodology in which the exposed elements are organized as nodes on a graph that is used to propagate impacts from directly affected nodes to other nodes across network links. The authors discuss how the impacts are propagated along different orders of the impact chain for increasing return periods, also comparing the risk curves between direct and indirect impacts. Finally, they highlight the extent to which the reduction in the demand for services from consumers and the loss of services from suppliers are respectively contributing to the final indirect impacts, as well as how different impact mitigation measures can be formulated from these results.

I would like to express my sincere gratitude to the authors, who have kindly shared their scientific knowledge and experience through their contributions; to the peer reviewers, who have contributed to enhancing the quality of the articles published in this Special Issue; and to the managing editors of the journal, who have supported and promoted this project. Last but not least, I would like to leave a word of appreciation to the authors who, unfortunately, saw their manuscripts declined as a result of the high-quality standards adopted in the selection and review of the received articles.

**Funding:** This research was funded by the Portuguese Foundation for Science and Technology (FCT) through the postdoctoral grant SFRH/BPD/122598/2016.

**Conflicts of Interest:** The author declares no conflict of interest.

#### **References**


© 2020 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## **Real-Time Integrated Operation for Urban Streams with Centralized and Decentralized Reservoirs to Improve System Resilience**

#### **Eui Hoon Lee <sup>1</sup> , Young Hwan Choi <sup>2</sup> and Joong Hoon Kim 3,\***


Received: 4 December 2018; Accepted: 26 December 2018; Published: 2 January 2019

**Abstract:** Recently, the number of extreme rainfall events has increased because of climate change. The ever-widening impervious area in urban watersheds also continuously augments runoff volume. Most measures to prevent urban inundation are structural, such as the construction, rehabilitation, and replacement of urban drainage facilities. Because structural measures require time and money, nonstructural measures are also required for the efficient prevention of urban inundation. Current operations in Korea focus on the individual operation of urban drainage facilities while neglecting the status of effluent streams. A study on urban drainage facilities that considers the status of urban streams is necessary to improve the operation of drainage facilities in urban areas. A revised resilience index is suggested to evaluate measures. For the historical rainfall event in 2010, the system resilience for current and integrated operations was 0.199 and 0.238, respectively. For the 2011 event, the system resilience for current and integrated operations was 0.064 and 0.235, respectively. The integrated operation exhibited good performance for the 2010 and 2011 events. Based on the results of this study, an operation as a nonstructural measure for the total management of urban areas is proposed. The revised resilience index could support decision-making processes for flood-management plans.

**Keywords:** integrated operation; urban stream; urban drainage facility; revised resilience index

#### **1. Introduction**

As the discharge in urban areas has increased because of urbanization, various measures to prevent inundation have been considered. Structural measures focus on expanding the capacity of urban drainage facilities through construction, rehabilitation, and replacement. The construction of additional drainage facilities can be classified as structural because it requires money and time. Nonstructural measures, such as the operation of urban drainage facilities (centralized and decentralized reservoirs) and flooding alerts, are required to maximize the efficient management of these structural measures. In general, the status of urban streams is not considered because urban drainage facilities in Korea are individually operated to prevent urban inundation. The management of urban drainage systems should include all components, such as urban streams, pump stations, and detention reservoirs; however, previous studies did not consider these components. The previous studies focused on the individual real-time control (RTC) and cooperative operation of urban drainage facilities in a single drainage area. The related studies of RTC in urban drainage can be classified into categories of individual operation and cooperative operation in a drainage area, as shown in Table 1.

**Table 1.** Classification of real-time control (RTC) in urban drainage systems.


Most of the studies related to individual operation have been conducted according to the unit time of operation. Individual operation has been suggested for single urban drainage facilities using various techniques [1–7]. Additionally, individual operation considering structural and other nonstructural measures has been applied to the design and forecasting field [8–11]. The cooperative operation of a single drainage area has been conducted for reducing the flooding of a city by applying the cooperation of the sewer system and the pump facility in the same drainage area. Cooperative operation with optimization, including single and multi-objective functions, has been applied to urban drainage systems [12–14]. Additional approaches for cooperative operation using the urban inundation model and the concept of resilience have been introduced [15–18].

These operations can be categorized as individual and cooperative operation of urban drainage facilities in a single drainage area. However, the integrated operation of urban drainage facilities has not yet been examined while considering the status of urban streams. Previous literature on RTC in urban drainage systems focused on the prevention of urban inundation. These approaches (i.e., individual operation and cooperative operation in a drainage area) can reduce the damage that urban inundation causes. However, it is difficult for these approaches to respond under extreme and localized torrential rainfall conditions, and they do not consider sustainable urban development in cities. Therefore, this study proposes a new operation approach that considers the overflow in urban streams and the inundation of inland drainage areas simultaneously as well as the improvement of system resilience. Therefore, it is suggested that this integrated operation could reduce inundation and improve resilience in target watersheds. The term "integrated operation" originated from a research project (Development of advanced techniques in combined inland-river systems) that was supported by the Ministry of Land, Infrastructure, and Transport of the Korean government.

In the 2000s, new approaches that considered various models and components were suggested for more sustainable and resilient cities [19–22]. These studies are based on the resilience of various factors in urban systems and have led to studies on system resilience in civil engineering, many of which have been performed in the field of water resources engineering. Ever since the concept of resilience was applied to water resources, many studies on the resilience of urban drainage systems have been conducted. A system resilience with conceptual framework and global analysis approach has been proposed for urban drainage systems [23–27]. The resilience index in these studies was calculated by comparing normal and extreme rainfall events.

System resilience is the ability of preparation for, reaction to, and recoverability against a system failure that is generated as a malfunction of system components (e.g., the pump and gate in pump stations), and the resilience index indicates system functionality. Therefore, to maintain high system functionality, the system needs to consider system resilience. For example, in urban drainage systems, if 80 out of 100 units of water can be discharged and there are 20 units of flooding volume, the system functionality is 80%. The resilience in urban drainage systems is based on the quantification of the system functionality at each time point. The resilience in urban drainage systems has been proposed in terms of, and quantified by, the rainfall amount, target area, and flooding volume [10].

One resilience index consists of the unit area, the rainfall amount considering the time of concentration, and the flooding volume [16]. Another resilience index consists of the watershed area, the total rainfall amount, and the flooding volume [10]. A drawback of these two indices is that it is difficult to highlight the differences between each value. In the previous indices, the denominator was very large compared with the flooding volume per minute in the numerator. This was because the denominator was multiplied by the rainfall amount considering the time of concentration (or the total rainfall amount) and the unit area (or the watershed area). The fundamental reason for overestimation in these two indices originates from the rainfall amount in the denominator. The duration for calculating the rainfall amount at each time point is longer than the duration of the flooding volume at each time point. If the flooding volume occurs at 60 min and the time of concentration is 30 min, the rainfall amount is calculated as the sum of 30 min to 60 min (or 1 min to 60 min). The calculation of the rainfall amount needs to be improved.

The revised resilience index can better reflect the status of urban drainage systems. It is calculated using real-time rainfall data and is easy to apply. Various conditions, such as flooding in times of no rainfall and no flooding during rainfall events, can now be considered. A revised resilience index is needed to overcome the disadvantages of the previous indices.

#### **2. Materials and Methodologies**

#### *2.1. Integrated Operation Considering Urban Streams*

The operation of pump stations should be integrated considering the levels of monitoring nodes and urban streams within the target watershed. RTC in this study means the operation/control of urban drainage facilities considering real-time rainfall data per minute. This approach was applied to each drainage area. Monitoring nodes were selected using the method described by Lee et al. [16], and integrated operation was based on the level of urban streams. When the stream level is high, the freeboard of levees on one or both sides may not be enough to prevent flood risk. In this case, pump stations on the stream do not operate drainage pumps as early as expected; however, they reserve a certain amount of water as the operating level in the centralized reservoirs (CRs) of the pump station increases. The CR in urban drainage systems is located in the downstream of the network in a single drainage area. It receives all inflow from the network, and the inflow in the CR is discharged by drainage pumps in pump stations.

The objective of integrated operation in urban areas is to ensure rapid and safe drainage and to secure additional storage capacity. In addition, it can prevent the backwater effect that is caused by high water levels in the CR and reduce the risk of overflow in urban streams. Moreover, this operation system allows for the effective management of urban streams when the system does not have sufficient capacity. This study looks at the development of an integrated operation system considering sustainable urban management, which applies the revised system resilience index and focuses on Korean urban areas. In Korea, it is difficult to expand the width of streams or increase the height of levees because the tops of levees are used as roads. The integrated operation can balance the spare capacity between urban drainage facilities and urban streams. A schematic of the integrated operation is shown in Figure 1. The proposed approach consists of three schemes: (1) integrated pump operation, (2) the revised system resilience index, and (3) the selection of monitoring nodes.

**Figure 1.** A schematic of integrated operation in urban areas. LWL, low water level; HWL, high water level.

#### *2.2. Integrated Pump Operation*

The operation of pump stations in Korea is determined by the level of the CR. The initial CR operating level (*Oi*) is calculated as the sum of the required depth, screen head loss (*Hs*), and freeboard for mechanical operation (*Fm*). *Vr* is the value of the product of the initial pump capacity and pump preparation time (*Tp*). The value of *Hs* varies from 0.1 m to 0.3 m, and the value of *Fm* varies from 0.0 m to 0.2 m. *Hs* is determined by the presence or absence of a CR in pump stations, and *Fm* by the total pump capacity. Both should be determined to prevent the cavitation of drainage pumps. These three factors are added to the bottom of the CR (*Bcr*) to give the initial operating level of drainage pumps (*Oi*). Other operating levels, except for the initial operating level, are based on the calculation of the required depth. Early operating levels in each pump station are calculated using the method of Lee et al. [16]. Equation (1) is used to calculate the initial operating level of the drainage pumps in the CR.

$$\dot{O}\_i = \frac{P\_{\text{i}} \times T\_p}{4V\_{\text{r}}A\_{\text{l}}} + H\_{\text{s}} + F\_{\text{ill}} + B\_{\text{cr}} \tag{1}$$

where *Oi* is the initial operating level of the drainage pumps in the CR, *Pi* is the capacity of the initial pump, *Tp* is the preparation time of the initial pump, *Vr* is the required volume in the CR, *Al* is the average area at each elevation in the CR, *Hs* is the screen head loss, *Fm* is the freeboard for mechanical operation, and *Bcr* is the bed elevation of the CR. Figure 2 shows the determination of operating levels for drainage pumps in CRs.

**Figure 2.** Determination of operating levels for drainage pumps in the centralized reservoir (CR).

As the local government in Korea has already determined the operating order for all pump stations, the new operation in the CR has the same operating order as the current one. Pump operation should be determined by operating level and order, considering the cavitation of drainage pumps. When the level of a monitoring node reaches a determined level for a new operation, drainage pumps in the CR are operated early. The operation of the decentralized reservoir (DR) is applied to the integrated operation if there is a DR in the drainage area. The DR's operation is also based on the level of the monitoring nodes. Pumps in the DR are operated until the level of the monitoring node reaches the operating level of the CR. Lee et al. [16] suggested the cooperative operation of the CR and DR, which means the combined operation of the CR and the DR. In this study, we propose an integrated operation for urban streams with all drainage facilities, including the CR and the DR.

#### *2.3. Revised Resilience Index*

Resilience is the ability to recover from a failure because of internal and external shocks to a community system. It involves the ability to recover from flooding or a malfunction of drainage facilities in urban drainage systems. Recently, the concept of resilience has been used in the evaluation of various systems, including future strategies by governments, academia, and enterprises. Todini [28] applied the resilience index to the design of a looped water distribution network using a heuristic approach. A framework to quantitatively assess the improvement in a community's resilience to seismic hazards was developed by Bruneau et al. [29]. Godshalk [30] recommended constructing resilient cities to withstand the threats of natural hazards and terrorism.

Structural and functional resilience indices in urban drainage systems were suggested in [25,26]. These consist of the total flood volume, the total inflow into the system, the mean duration of nodal flooding (computed for all flooded nodes in the system), and the maximum nodal flood duration or total elapsed time (simulation time). The resilience index in the two studies is the computed design functional resilience index for the design of urban drainage systems. It is not based on real rainfall but on rainfall scenarios. In addition, its application is very complicated because it requires the estimation of the areal reducing factor. Lee et al. [10] introduced the resilience index for urban drainage systems, which is based on the flooding volume per minute, the area of the target watershed, and the total rainfall amount. In this index, the flooding volume has a low impact if the area of the target watershed and the total rainfall amount are large. All values of the resilience index in the previous study are close to 1; therefore, it is difficult to see a difference between current and new measures. In this study, a revised resilience index is introduced to overcome this difficulty.

The main difference between the previous and revised resilience indices lies in the calculation of the denominator in the performance evaluation function. In the previous resilience index, the total rainfall amount is applied to the denominator, while in the revised resilience index, the rainfall amount at each minute is applied to the denominator. The value of the denominator in the previous resilience index is larger than that in the revised resilience index. In addition, these values are classified into various cases because the denominator cannot be calculated when the rainfall amount is zero in the revised resilience index. The revised resilience is calculated as 1 when the rainfall amount and flooding volume are zero. The revised resilience is calculated as 0 when the rainfall amount is zero and the flooding volume is not zero. The rainfall amounts in both resilience indices are different, even though the performance evaluation function is calculated every minute in both. The total amount of rainfall in the watershed is used in the previous resilience index, while the current rainfall amount in the watershed is used in the revised one. The resilience index comprises the performance evaluation function, which is shown in Equation (2):

$$\mu(T)\_t = \max\left(0, 1 - \frac{F\_t}{R\_t \times A}\right) \tag{2}$$

where *u(T)t* is the performance at time *t*, *Ft* (in m3) is the flooding volume at time *t*, *Rt* is the rainfall amount (in m) at time *t*, and *A* is the area of the target watershed (in m2). If the value of the performance evaluation is 1, then there is no flooding. Table 2 categorizes the value of utility performance according to the numerator and denominator of Equation (2). In the revised index, the duration of failure is calculated in the equation for resilience because the performance evaluation function is estimated at each minute and the equation for resilience is based on the value of the performance evaluation function.


**Table 2.** The value of the performance evaluation according to the numerator and denominator.

The sum of the numerators at each time point gives the total flooding volume in the target watershed, and the sum of the denominators at each time point gives the total rainfall amount. The resilience in the target watershed is calculated by the value of the performance every minute. The equation for resilience is shown in Equation (3).

$$R\_s = \frac{1}{T} \int\_0^T u(T)\_t dt\tag{3}$$

where *Rs* is the resilience in the target watershed, and *T* is the entire duration of the rainfall event. The revised resilience index can be calculated for and applied to all watersheds.

#### *2.4. Selection of Monitoring Nodes*

Monitoring nodes are points where inundation first occurs within an urban drainage system [10,16]. Three synthetic rainfall durations were selected on the basis of the time of concentration (tc) (e.g., tc, 2tc, and 3tc).

Time of concentration was used as a variable to allow for a sufficient duration of rainfall to be simulated considering the total amount of rainfall in the drainage system. Additional durations, up to three times as long as the time of concentration, were selected to simulate longer-duration rainfall events. For example, if the time of concentration is 30 min, the three durations are 30, 60, and 90 min. Initially, 1 mm of synthetic rainfall data were produced. This was then increased in 1-mm increments until the first flooding event occurred. The process for finding the first flooding nodes is shown in Figure 3.

**Figure 3.** The process for finding the first flooding nodes.

A report detailing the design of discharge in streams in Korea is available to the public [31]. Most streams in Korea have several monitoring nodes under bridges. Monitoring nodes for integrated operation can be selected from among these. Various discharges for each frequency can be applied to the selected simulation model, and the freeboards for each monitoring node can be checked. The monitoring node with the smallest value of freeboard is then selected for use in the integrated operation. The schematic of the monitoring nodes in streams for integrated operation is shown in Figure 4.

In Figure 4, among the four candidates, candidate B can be selected as the monitoring node because the freeboard at candidate B is smaller than the required freeboard. The candidate with the largest difference is selected if the freeboard at several candidates is smaller than the required freeboard. In Korea, monitoring candidates are generally located on bridges, and the required freeboard at each station of streams is determined by the design frequency. The required freeboard of the stream is high if the design frequency of the stream is high. For example, the required freeboard of the stream is 0.6 m

for a 100-year frequency, and the required freeboard is 0.8 m for a 200-year frequency. The design frequency of the downstream is larger than that of the upstream if the design frequencies of the upstream and downstream are different.

**Figure 4.** A schematic of the monitoring nodes in streams for integrated operation.

#### *2.5. Model Formulation*

This study consisted of five parts, as shown in Figure 5. First, synthetic rainfall data were generated for the selection of monitoring nodes. Second, monitoring nodes in each drainage area were selected to operate drainage pumps in the CR. Third, a monitoring node in an urban stream was selected for integrated operation in each CR. Fourth, the integrated operation of pump stations was conducted using the level of a monitoring node in an urban stream. Finally, a revised resilience index was suggested and applied to urban drainage systems in a target watershed.

**Figure 5.** The flowchart for this study.

A storm water management model (SWMM) was used for the rainfall runoff simulation of the target watershed [32]. Evaporation and losses can be simulated in the SWMM. In Korea, the evaporation data are generally provided per month and not per minute. Monthly averages in evaporation can be applied to the SWMM, which is only used during dry periods. For infiltration among losses, Horton's infiltration equation was used. Each parameter was changed during calibration. Other losses were considered using the percentage of impervious area at each sub-catchment.

#### *2.6. Study Area*

The Dorim stream was selected as the target watershed, and is shown in Figure 6. It has a total area of 41.93 km2. Furthermore, there are 11 pump stations (Mullae, Dorim2, Daerim3, Daerim2, Guro1, Guro2, Guro3, Guro4, Sinlim1, Sinlim2, and Sinlim5), two detention reservoirs (Daerim and Gwanak), and two branches (the Daebang and Bongchun streams) along the stream.

**Figure 6.** Description of the target area (Imagery © 2017 Centre National d'Etudes Spatiales/Astrium, DigitalGlobe, DigitalGlobe, Map data © SK telecom).

The characteristics of Dorim stream consist of river length, levee length, mean width of basin, shape factor, and slope. Among them, the mean width of basin is the value of area (A) divided by the river length (L) and the shape factor is the value of area (A) divided by the square of river length (L). Table 3 shows the characteristics of the Dorim stream [31].



A rainfall station is located in the center of the target watershed. Table 4 provides information on the pump stations in the target watershed, including CR capacity, drainage pump capacity, drainage area, high water level (HWL), and low water level (LWL) [33–44].


**Table 4.** Information on pump stations in the target watershed.

The total drainage area of the 11 pump stations is 884 ha (8.84 km2), which is approximately 21% of the total drainage area of the Dorim stream. The Daerim detention reservoir was constructed in 2009 and the Gwanak detention reservoir in 2016. They are described in Table 5 [45,46].


**Table 5.** Description of the detention reservoirs in the target watershed.

Inundations occurred at the Dorim stream in 2010 and 2011. The total annual precipitation in 2010 was 2.075 mm, and the total amount of rainfall was 253 mm when the flooding occurred on 21 September 2010. The total annual precipitation in 2011 was 2.014 mm, and the total amount of rainfall was 378 mm when the flooding occurred on 27 July 2011 [47]. These historical rainfall events have a frequency of approximately 100 years, which is greater than the design frequency of the urban drainage facilities [47,48].

The new model of the sewer network in the target watershed is based on the geographic information system data for urban drainage areas and streams supplied by the Seoul Metropolitan Government. Figure 7 shows the digital elevation model, hill shade, slope, and aspect of flow direction in the target watershed [45].

**Figure 7.** Information on the target watershed: (**a**) Digital elevation model (DEM), (**b**) Hill shade, (**c**) Slope, and (**d**) Aspect of flow direction.

As shown in Figure 8, the sewer network in the target watershed consists of 4137 sub-catchments, 4544 nodes, and 4710 links.

**Figure 8.** The sewer network in the target watershed.

#### *2.7. Generation of Synthetic Rainfall Data*

For each drainage area and urban stream, synthetic rainfall data were generated using the Huff distribution [49], and monitoring nodes were selected by means of a rainfall runoff simulation. The synthetic rainfall data were used for the selection of monitoring nodes, and the historical rainfall

data were selected to verify the effect of the integrated operation. The Huff distribution was used because its third quartile is appropriate for the design and operation of drainage facilities in Korea [50]. The design of Korean urban drainage facilities is based on this distribution. Equation (4) shows the regression equation for the third quartile of the Huff distribution in Seoul [51].

$$y = 37.835x^6 - 106.21x^5 + 105.18x^4 - 44.549x^3 + 9.1084x^2 - 0.3603x + 0.0005\tag{4}$$

where *y* is the ratio of cumulative rainfall, and *x* is the ratio of the total rainfall duration. The process of generating synthetic rainfall data from the Huff distribution consists of three steps. First, a cumulative distribution is generated using the regression equation of the Huff distribution. Second, the separated ratio is produced by the cumulative distribution. Third, the rainfall amount and duration are applied to the separated ratio [51]. This process is illustrated in Figure 9.

**Figure 9.** Generation of synthetic rainfall data by the Huff distribution.

Moreover, the sewer network in the target watershed was manually calibrated using the 2010 and 2011 rainfall events. After the calibration of each sub-watershed in each CR, the total runoff was calibrated downstream of each drainage area. The peak discharge, peak time, total runoff volume, and minimization of the root mean square error were focused on the calibration of the model. The percentage of impervious area, width, and percentage of slope in each sub-catchment were used to calibrate the model. In addition, the maximum infiltration rate, minimum infiltration rate, and decay constant in Horton's infiltration equation were used to calibrate the model.

#### **3. Application and Results**

#### *3.1. Selecting Monitoring Nodes in Each Drainage Area and Stream*

The selection of monitoring nodes is required for the integrated operation of drainage facilities, such as a CR and a DR. This is because a CR's operation, or a cooperative operation between a CR and a DR, is based on the level of monitoring nodes. In this study, monitoring nodes were selected using two methods, based on the first flooding node and the maximum flooding node.

The first flooding node generally occurs between branch conduits, rather than between main conduits. This makes it difficult to use for the integrated operation of drainage facilities. A section is categorized as the main conduit based on the product of the runoff coefficient (C) and drainage area (A). If this is greater than 0.12 km2 (CA ≥ 0.12 km2), then it is the main conduit. If it is smaller than 0.12 km<sup>2</sup> (CA < 0.12 km2), then it is a branch [16]. The first flooding node is selected using the results of rainfall runoff simulations, with synthetic rainfall events generated by the Huff distribution. The amount of synthetic rainfall is increased from 1 mm in 1 mm increments, and this is applied to the runoff model until the first flooding event occurs. The first flooding nodes in each drainage area are shown in Table 6.

**Table 6.** The results of the first flooding nodes for the operation of drainage facilities in each drainage area.


The rainfall amount required to cause the first flood of the drainage system in the Daerim3 pump station is higher than that in the others. This means that the drainage system in the Daerim3 pump station is relatively strong against initial flooding. Conversely, the drainage systems in the Sinlim1, Sinlim2, and Sinlim5 pump stations are relatively weak against initial flooding. Some drainage areas have different first flooding nodes for different rainfall durations. The drainage areas in Guro1, Dorim2, and Daerim3 show different first flooding nodes at 90 min because some conduits in these areas have reverse gradients, causing different initial flooding patterns. However, the Guro2, Guro3, Guro4, Guro4, Sinlim1, Sinlim2, Sinlim5, Mullae, and Daerim2 pump stations all demonstrated the same first flooding node for all durations. When a drainage area had different first flooding nodes at 90 min, the node that appeared in the largest number of results was selected.

The maximum and first flooding nodes were selected as the monitoring nodes for the integrated operation in each drainage area. To select the maximum flooding nodes, historical rainfall events were used, rather than synthetic ones, as simulations using synthetic rainfall events produce various maximum flooding nodes, making the selection of monitoring nodes difficult. Rainfall data from 23 September 2010 and 27 July 2011, when historical flooding occurred in the target watershed, were used to identify the maximum flooding nodes in each drainage area. The maximum flooding nodes in each drainage area are shown in Table 7.

**Table 7.** The results of maximum flooding nodes for the operation of drainage facilities in each drainage area.


Several maximum flooding nodes demonstrated a greater flooding volume than those in other drainage areas, namely Sinlim1, Sinlim2, Sinlim5, Mullae, and Dorim2. This is because of the capacity shortage of conduits and backwater effects produced by the level of the CR. A single node is selected as a monitoring node if the first flooding node is the same as the maximum flooding node. Two nodes are selected as monitoring nodes if the two are different. In the real-time integrated operation, Guro1, Guro3, Guro4, Sinlim1, Sinlim5, Mullae, Daerim2, and Daerim3 each have two monitoring nodes, while Guro2, Sinlim2, and Dorim2 have one. If the depths of the two monitoring nodes differ, they are converted into a dimensionless parameter. The depth of the monitoring node is converted to 1.0D if it is 1.5 m, and the level is converted to 0.6D if it is 0.9 m.

In Korea, monitoring nodes in urban streams are constructed under bridges. There are six bridges across the Dorim stream: the Dorim, Guro1, Sindaebang, Gwanakdorim, Sinlim3, and Seoul National University Bridges. All streams in Korea have a designed freeboard, and this is 0.6 m for the Dorim stream. The monitoring candidates in the Dorim stream are shown in Figure 10.

**Figure 10.** The monitoring candidates for the Dorim stream (Imagery © 2017 Centre National d'Etudes Spatiales/Astrium, DigitalGlobe, DigitalGlobe, Map data © SK telecom).

Rainfall events with various frequencies (30, 50, 80, 100, and 200 years) and with a duration based on the time of concentration in the Dorim stream (360 min) were applied to select the monitoring nodes. The water level in the Dorim stream, height of the bank, and freeboard for events with 30-, 50-, 80-, 100-, and 200-year frequencies are shown in Table 8.




**Table 8.** *Cont.*

For the 30-year frequency, all monitoring candidates satisfy the designed freeboard of the Dorim stream. For the 80- and 100-year frequencies, the right bank at Sindaebang Bridge lacks a freeboard. Overflow occurs here with a 100-year frequency. The other monitoring candidates satisfy the designed freeboard of the Dorim stream. However, the banks of various monitoring candidates, such as the left bank at the Dorim Bridge, both banks at the Sindaebang Bridge, and the right bank at the Sinlim3 Bridge, also lack a freeboard. Overflow occurs at the right bank at the Sindaebang Bridge, and the overflow from a 100-year frequency event is the same as that for an 80-year frequency event. Both banks at the Dorim, Sindaebang, and Gwanakdorim Bridges, and the right bank at the Sinlim3 Bridge, also lack a freeboard. Overflow occurs at the Dorim and Sindaebang Bridges. The results in Table 8 shows that overflows at Sindaebang Bridge occur with 80-, 100-, and 200-year frequencies. This section of the Dorim stream is most vulnerable to overflow.

#### *3.2. Results of the Rainfall Runoff Simulation*

Historical rainfall events in 2010 and 2011, when flooding occurred in the Dorim stream, were selected for the rainfall runoff simulation, in which the current and integrated approaches for operating drainage facilities, including CR and DR operations, were applied to the target watershed. The integrated operation, which includes the use of the CR, was applied to each pump station in the Dorim stream. The preparation time was between 5 and 30 min. The calculation was used to determine the initial operating level in the CR. This was applied to the Daerim3 pump station as follows. The product of the initial pump discharge (233 m3/min) and the preparation time (30 min) were divided by 4. The required volume in the CR was 1711 m<sup>3</sup> and the average area at each elevation in the CR was 11,400 m2. The required depth was calculated by dividing the required volume in the CR by the average area at each elevation [16]. Table 9 shows the operating levels of the drainage pumps in each drainage area of the Dorim stream.

These data are shown for both the integrated and current operations. The results, shown in Figure 11a, indicate that the integrated operation produces a lower flooding volume than the current operation (which produces 2,905,874 m3).


**Table 9.** The current and new operations of the CR for each drainage area of the Dorim stream.

**Figure 11.** The results of current and integrated operations: (**a**) 2010; (**b**) 2011.

Overall, the integrated operation demonstrated good results, although they varied slightly according to the operating levels. When the level of the monitoring node was 0.2D, the maximum flooding volume was 2,743,103 m3, and the minimum flooding volume was 2,741,478 m3 with a level of 0.3D. Figure 11b shows the results of the current and integrated operations for 2011. The results in Figure 11b also show that the integrated operation produced a lower flooding volume than the current operation (3,312,733 m3) for all levels of the monitoring node. The integrated operation once more showed good results, although they still slightly differed from one other according to the operating levels. The maximum flooding volume was 2,951,973 m<sup>3</sup> when the level of the monitoring node was 0.1D, and the minimum flooding volume was 2,944,196 m<sup>3</sup> when the level was 0.9D. The results of

Figure 11 show that the integrated operation was steadily better than the current operation at all levels of the monitoring nodes.

In Figure 12, the results of flooding volume over time using the integrated operation and current operation are compared. The results in 2010 and 2011 are shown in Figure 12a,b, respectively. The results in 2010 and 2011 show that the integrated operation had less flood volume per minute than the current operation.

**Figure 12.** Rainfall and flooding volume over time: (**a**) 2010; (**b**) 2011.

Moreover, this study applied the system resilience to verify the ability to prepare for and recover from the malfunction (failure) of drainage facilities and inundation (system degradation) of drainage systems. The proposed resilience index was applied to the current and integrated operations for the 2010 and 2011 events. The results of the two operations were compared for both years, when the level of the monitoring node at the beginning of the integrated operation (including the CR operation) was 0.8D. The results of system resilience for the current and integrated operations are shown in Table 10.


**Table 10.** The results of system resilience for current and integrated operations.

For the 2010 event, the system resilience of the current operation was 0.199, whereas that of the integrated operation was 0.238. This means that the integrated operation increases the system resilience of the urban drainage system in the target watershed by 0.039. The value of system resilience ranges from 0 to 1. A high value of system resilience in urban drainage systems means a more resilient drainage system. The integrated operation makes the urban drainage system in the target watershed resilient to failure (flooding).

For the 2011 event, the system resilience of the current operation was 0.064, whereas that of the integrated operation was 0.235. The system resilience increment between the current and integrated operations was thus 0.171. The system resilience of the current operation for the 2011 event was lower than that for 2010 because the total flooding volume of the 2011 event was larger than that in 2010 and system failure occurred frequently in 2011, as the flooding volume was widely distributed. The system resilience of the target watershed in Table 9 was calculated as low because the value of performance was calculated as zero if flooding occurred when there was no rainfall.

#### **4. Conclusions**


**Author Contributions:** E.H.L. carried out the survey of previous studies. E.H.L. and Y.H.C. wrote the original manuscript. E.H.L. conducted the simulations. Y.H.C. revised the original manuscript. E.H.L., Y.H.C., and J.H.K. conceived the original idea of the proposed method.

**Funding:** This research was funded by the National Research Foundation (NRF) of Korea in the Korean government (MISP) (No. 2016R1A2A1A05005306).

**Acknowledgments:** This work was supported by a grant from The National Research Foundation (NRF) of Korea in the Korean government (MSIP) (No. 2016R1A2A1A05005306).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Residential Flood Loss Assessment and Risk Mapping from High-Resolution Simulation**

**Zulfahmi Afifi 1, Hone-Jay Chu 2,\* , Yen-Lien Kuo <sup>3</sup> , Yung-Chia Hsu 4, Hock-Kiet Wong <sup>5</sup> and Muhammad Zeeshan Ali <sup>2</sup>**


Received: 12 March 2019; Accepted: 8 April 2019; Published: 10 April 2019

**Abstract:** Since the patterns of residential buildings in the urban area are small-sized and dispersed, this study proposes a high-resolution flood loss and risk assessment model to analyze the direct loss and risk impacts caused by floods. The flood inundation simulation with a fine digital elevation model (DEM) provides detailed estimations of flood-inundated areas and their corresponding inundation depths during the 2016 Typhoon Megi and 2017 Typhoon Haitang. The flood loss assessment identifies the impacts of both events on residential areas. The depth-damage table from surveys in the impacted area was applied. Results indicated that the flood simulation with the depth-damage table is an effective way to assess the direct loss of a flood disaster. The study also showed the effects of spatial resolution on the residential loss. The results indicated that the low-resolution model easily caused the estimated error of loss in dispersed residential areas when compared with the high-resolution model. The analytic hierarchy process (AHP), as a multi-criteria decision-making method, was used to identify the weight factor for each vulnerability factor. The flood-vulnerable area was mapped using natural and social vulnerability factors, such as high-resolution DEM, distance to river, distance to fire station, and population density. Eventually, the flood risk map was derived from the vulnerability and flood hazard maps to present the risk level of the flood disaster in the residential areas.

**Keywords:** flood; 3Di; loss assessment; analytic hierarchy process (AHP); risk map

#### **1. Introduction**

Floods are some of the most catastrophic natural disasters that include severe economic impacts, especially if floods happen in big cities around the world. In Taiwan, floods in the last 25 years have resulted in losses of \$518 million USD. The flood losses are approximately 4.6 times the losses caused by fire damage [1]. Floods result in the exposure of many properties in residential areas to standing water. Inundation may cause structural damages, such as on wall linings, and property damage, including electronic devices [2]. Flood damages and losses can be broadly divided into two types: direct losses and indirect losses. Direct flood losses are those caused by the physical or structural impact of a flood event while indirect losses are not the direct economic losses of a flood. In the majority of damage estimation studies, losses are restricted to direct losses for each element [3]. The evaluation of flood loss is related to the use of damage functions from potential asset damaging [4]. However, flood loss

is highly correlated to flood water depth. The depth-damage curve or table was created via a field survey of direct loss in a flood event in the flood loss model [5]. The loss assessment includes asset exposure, susceptibility to suffer damage, and damaging potential [6]. Moreover, flood risk was usually defined as the product of hazard e.g., probability of occurrence and vulnerability or estimated cost of foreseeable damage [7]. Risk assessment was generally measured as the expected loss degree of risk as a consequence of a hazardous event [8].

Flood loss and risk assessment have been determined through flood hazard data, i.e., spatial distribution of flood area, exceedance probability, and asset vulnerability data [6]. However, most flood information is only based on point discharge measurements at discrete locations without spatiotemporal information of flood inundation [9]. For the lack of sufficient depth measurements, we offer a flood loss and risk assessment that integrates 2D hydraulic model results and high-resolution digital elevation model (DEM) data, especially during the typhoon events. Considering the pattern of residential area in the urban area, the water depth of each residential area is more appropriate based on the high-resolution inundation simulation. A simulation-based flood loss and risk assessment can provide an effective way for the spatial estimation and visualization of flood loss and risk in potential flooding areas. Eventually, the impact of estimated loss or risk using a geographic information system (GIS) is practical for flood loss assessment and disaster management [4,5,10].

This study proposes a GIS-based high-resolution flood loss and risk assessment model to analyze the direct loss and risk impacts caused by a flood in a residential urban area. The model with fine grids provides a high-resolution estimation of a flood-inundated area and its corresponding loss and risk. The flood simulated water depths based on typhoon events during the 2016 Typhoon Megi and 2017 Typhoon Haitang were used in this research. The flood loss assessment identifies the impact of a simulated flood water depth in the residential area. The highest water depths in the simulation of the typhoon event were used to estimate the maximum loss in residential buildings. The depth-damage table of the impacted area was applied in our survey for flood loss assessment. In addition, risk is defined as the product of hazard e.g., high-resolution water depth and flood vulnerability. The flood-vulnerable area was mapped using natural and social vulnerability factors, such as distance from the river, distance to the fire station, elevation, and people density. The analytic hierarchy process (AHP) was used to acquire the weight factor of each vulnerability factor. A flood risk map was derived from the vulnerability and flood hazard maps to show the risk of the flood disaster.

#### **2. Datasets and Methods**

Figure 1 describes the process of each element in different colors. Each component represents the proposed material and method used to derive the results. The black and blue colors indicate the materials used in the beginning of the study. The current land use map was used to analyze the impact of the flood disasters on the residential area. The high-resolution (1 m resolution) DEM was used in this study. Water depth data were generated from the 3Di model [11]. The DEM, distance to river, distance to fire station, and population density were used as natural and social vulnerability factors. The green color indicates the process of deriving the flood loss and hazard map. The orange color indicates the process of generating the vulnerability map. The blue color indicates the process of deriving the risk map as the final step. A GIS model was also used to make the flood loss and risk process automatic. This model combined loss and risk functions and processes developed using Python. The proposed model enhances the applications with the purpose of flood loss and risk assessment by combining functions and data related to flood loss, hazard, and vulnerability (Figure 1).

**Figure 1.** Flowchart of the GIS-based flood loss and risk assessment.

#### *2.1. Flood Model*

The 3Di model was developed by Professor Stelling from TU Delft, Netherlands in 2010 [11]. The 3Di model combines four numerical methods including the subgrid method, bottom friction based on the concepts of roughness depth, the staggered-grid finite-volume method for shallow water equations with rapidly varying flows, and the quad-tree technique [12]. The 3Di model with sub-grid and quad-tree methods can handle a large number of computational grids with high-resolution topographic data [13]. The 3Di numerical method utilizes a finite volume. The method works by solving the 2D shallow water equations and combining the data through the quad-tree method. Such process increases the density of computational cells in important areas, such as urban areas, levees, and roads. It also ensures that the flow field will perform smoothly and reduces errors affected by the terrains. 3Di can efficiently perform under high-resolution terrain data. The integrated 3Di model consists of two main components including 2D and 1D-flow components. In the model simulation, 3Di uses 1D and 2D equations in terms of terrain conditions.

Compared with other inundation models, the most special part is the technique of the simulation cell in the 3Di model, such as the sub-grid and quad-tree methods. The sub-grid method works before starting a simulation, and it provides a series of temporary tables that contain the water volume–level relationship of each simulation cell and link during the simulation when a water level is being solved. Accordingly, we determine the volume of water that has changed between two adjacent cells. The quad-tree method is suitable for simulating high-resolution terrain data or a large simulation area. It has been widely used in inundation simulation or hydraulic calculation. This method can increase the flexibility of simulation cells and is efficient when used for simulation.

In this study, the high-resolution simulation model was based on the high-resolution DEM. The distributed rainfall data used in the 3Di flood simulation data were the Quantitative Precipitation Estimation and Segregation Using Multiple Sensors (QPESUMS). QPESUMS is the rainfall time series developed by the Central Weather Bureau, the Soil and Water Conservation Bureau, and the Water Resources Agency in 2002. This time series integrates radar, satellite, rain gauge station, and lightning observation data and is combined with the QPESUMS development. The size of QPESUMS rainfall data is 1.3 km × 1.3 km, and it collects rainfall data every 10 minutes. Observed water level data from two stations were used to validate the flood model data result. The gauging stations of the Wangliao Bridge and the Yanzi Bridge are located in the Yongkang District. The water level data will be updated every 10 min. Moreover, the simulated water levels and depths from the 3Di are the vector data in the form of quad-tree data of the water level in the study area. These data were then converted into raster data for the subsequent loss and risk assessment. The flood raster data were converted to different resolutions to determine the effects of differences in the resolution. The flood losses in using different spatial resolutions (1, 5, and 10 m) were compared in this study.

#### *2.2. Loss Assessment*

The flood loss assessment was calculated based on a direct assessment method of the impact of flooding on residential buildings. This assessment used the maximum water depth data of the flood simulation model from the 3Di model during the flood events and the depth-damage table from the field survey.

The depth-damage table was generated via the field survey in the flood area in the Rende District using the synthetic method [14]. Fourteen local residents joined the loss survey. The survey was conducted using a questionnaire given to residents in the flood disaster-prone areas. The table was generated from the survey results of the flooding areas, depths, and the losses in the residential area including furniture, clothes, electronic devices, and domestic appliances [14]. The survey data implied that no costs were incurred by the community in the disaster-prone area for the flood recovery work. The building recovery work and cleaning process were performed by the homeowners themselves and assisted by the government, such as firefighters and military personnel.

#### *2.3. Risk Assessment*

A flood risk map is useful for increasing the awareness of local communities, local authorities, and other organizations on flood risks. The determination of flood-risk areas is normally accomplished through a subjective method using vulnerability data and the water depth of flood, with the concept of risk itself interpreted in various ways. With a combination of the data hazard and vulnerability, the flood risk map was generated using automatic procedures. The flood risk was generated by multiplying the flood hazard map with the vulnerability map [15,16].

$$
\mathbb{R} = \mathbb{H} \times \mathbb{V},
\tag{1}
$$

where R is the risk map, H is the flood hazard map, and V is the flood vulnerability map. The maximum water depth of each typhoon event was determined from the flood simulation. Furthermore, the average water depth from two events were applied for the flood hazard map. In this study, the five-level risk map was generated for flood risk assessment.

#### *2.4. Vulnerability Assessment*

Flood vulnerability mapping is the process of determining the degree of susceptibility of a given place to flooding. Vulnerability is the most crucial component of risk since it determines whether exposure to a hazard constitutes a risk that may actually result in a disaster. Flood vulnerability can be obtained from the weighted product model. This flood vulnerability map will be combined with available vulnerability data in the study area.

$$\mathbf{V} = \Pi\_{\mathbf{i}=1}^{\mathbf{n}} \mathbf{V}\_{\mathbf{i}}^{\mathbf{w}\_{\mathbf{i}}} \,\tag{2}$$

where wi is the weight of the factor i (i.e., DEM, distance to river, distance to fire station, and population density) from the AHP. Vi is the vulnerability map for factor i and n is the number of factors. The final result is a flood vulnerability map that has information on vulnerability levels ranging from very low to extreme levels.

The vulnerability factor data in this study include natural and social factors, such as the DEM, distance to river, distance to fire station, and population density. In Tainan, low-lying areas have a high vulnerability to the dangers of floods. The DEM data were used as one of the vulnerability factors in this study. Flood happens when rainwater exceeds the capacity of waterways or rivers. Distance to

the river is also an important factor in the analysis of vulnerability to flooding. River line area data were also included to calculate vulnerability in terms of the distance of an area to river flow. Areas that are close to the river will tend to have a higher value of vulnerability than other areas located far away from the river. Distance to the river is determined through GIS. Moreover, fire stations in Tainan are considered the first agency that can handle emergencies during a disaster, including floods. A region that has a distance far from a fire station will have a higher vulnerability when a flood occurs than an area located close to the fire station. This vulnerability is interpreted as the first response to an emergency situation to prevent casualties and dangers that can cause losses due to flooding. People can acquire help from firefighters and military services. Thus, fire stations are critical facilities when considering the social vulnerability index [17]. In addition, population density is an important factor for vulnerability that must be considered. The Ministry of Interior Affairs provides the village-based population data. The polygon data are converted to a raster dataset through GIS.

#### *2.5. AHP for Vulnerability Weight*

The flood disaster vulnerability map was created by processing some of the available vulnerability factor data. A multi-criterion decision-making (MCDM) method was used to determine the importance of vulnerability factors from the vulnerability factor data. One of the MCDM methods used was the AHP. AHP is a semi-quantitative decision-making value judgment approach that fulfills the objectives of decision makers [18]. In this study, the AHP model [19] was used. The AHP continued until the obtained consistency ratio was less than 0.1.

Twenty participants were selected for the AHP survey. Among them, 40% were local students from the International Master Program on Natural Hazard Mitigation and Management Program, 50% were hydraulic engineering students, and 10% were social science students. The backgrounds of the participants were acquired to examine their various perspectives.

#### *2.6. Study Area*

The study area is Tainan City in Taiwan, which covers an area of 137 km2. The locations of the study area (blue polygon) and residential area (orange polygon) are shown in Figure 2. The river (Erren River) moves through the middle of the city. Typhoons, e.g., the 2016 Typhoon Megi and 2017 Typhoon Haitang, were accompanied by abundant rainfall (352 and 430 mm with 24-hour accumulated rainfall) that caused serious damages in Tainan.

**Figure 2.** Study area and residential area.

#### **3. Result and Discussion**

#### *3.1. Flood Hazard Map, Depth-Damage Table, and Vulnerability Weight by AHP*

A flood hazard map was created on the basis of the data of the flood simulation model that was processed with an automatic GIS model. The flood simulation data were obtained from the research generated from the 3Di model. Water depth data from two flood events were used in this study (the 2016 Typhoon Megi and the 2017 Typhoon Haitang). The average water depths from both typhoon events were used to generate the flood hazard maps. Water level data from two stations known as the Wangliao Bridge and the Yanzi Bridge were used to validate the flood model data result. Figure 3 shows the comparison between flood simulations and observation of the water level in the two typhoon events. The red color represents the observation of the water level and the blue one represents the simulation of the water level. The results showed a good fitting between the simulated and observed water levels. Figure 4 shows the maximum flood water depth of each typhoon for the flood hazard map. The intensity level of the blue color indicates water depth. The dark blue indicates the highest water level in the study area. The automatic model was developed based on the flood data simulation caused by the typhoon disaster. In the 2016 Typhoon Megi and 2017 Typhoon Haitang, the results showed that the maximum flood water level ranged from 0.25 m to 2 m in the entire study area.

Table 1 shows the depth-damage table generated from the field survey in the flood area in the Rende District. No damage losses occurred if the water depth was less than 0.15 m. The flood loss increased with water depth in the depth-damage table. The damage per area is at maximum if the water depth is higher than 2 m. Figure 5 shows the vulnerability factor data such as (a) DEM, (b) distance to river, (c) distance to fire station, and (d) population density. In Figure 5a, the elevation of the study area is between −0.5 m and 40 m. Figure 5b shows that the residential area is close to rivers. In Figure 5c, the fire station is uniformly distributed except in the southeast area. Figure 5d shows that the population density is less than three persons per 25 m<sup>2</sup> in most areas but is crowded with 10 persons to 59 persons per 25 m<sup>2</sup> in the area. The urban population is concentrated in the core area of the northern part. The vulnerability factor is an important element for minimizing the impact of floods in the future.

Table 2 shows the vulnerability weight value from the AHP. The results of the AHP method is the weight value used to determine the level of importance of the vulnerability factor. The two most important factors are the population density (weight = 0.310) and distance to river (weight = 0.271). Vulnerability weights contribute in the evaluation of flood risk and are sensitive due to their specific local factors [20]. AHP modeling can identify vulnerability on a local scale but is difficult to impeccably quantify because it depends on the quality of the indicators and survey data.


**Table 1.** Depth-damage table for the residential area from the field survey.

**Table 2.** Vulnerability weight values from the AHP.


**Figure 3.** Comparison between flood simulation (blue color) and observation (red color) from the water level station at the (**a**) Yanzi Bridge during Typhoon Megi, (**b**) the Wangliao Bridge during Typhoon Megi, (**c**) the Yanzi Bridge during Typhoon Haitan, and (**d**) the Wangliao Bridge during Typhoon Haitang (unit: m).

(**a**) **Figure 4.** *Cont.*

**Figure 4.** Flood hazard map: maximum flooding water depth during (**a**) 2016 Typhoon Megi and (**b**) 2017 Typhoon Haitang (unit: m).

**Figure 5.** *Cont.*

**Figure 5.** Vulnerability factors for (**a**) DEM (unit: m), (**b**) distance to river (unit: m), (**c**) distance to fire station (unit: m), and (**d**) population density (unit: persons/25 m2).

#### *3.2. Flood Loss Comparison Using Di*ff*erent Resolutions*

In this subsection, the effects of different spatial model resolutions are discussed. Based on the loss comparisons with different spatial resolutions in Table 3, the estimated flood losses and flooding area at the fine resolutions of 1 m and 5 m resolutions were similar but were different from the 10 m resolution. Table 4 shows the comparison in flood loss at resolutions of 1 m to 5 m and 10 m. The loss result from the 10-m resolution contains a large difference among the cases. We explored the sensitivity of the flood loss estimates with the spatial representation in the flood loss model. The estimated loss at the 1 m or 5 m resolution is more appropriate due to the dispersed and small-sized residential buildings in Tainan City. Since the pattern of the residential area is dispersed, the high-resolution flood model offers relatively detailed loss assessment in the flood areas. This loss value will increase or decrease depending on the used scale of the study area. The case evaluated the sensitivity of the spatial resolution. In summary, the loss differences reached 7.8% and 11.2% during two events when compared with the 1-m and 10-m resolutions. Spatial resolutions used in the assessment had a direct impact on the potential flood heights [21]. However, the flood loss evaluation was often restricted due to the lack of dense observation data [22]. The adoption of a high-resolution flood simulation approach represents small-scale structural elements and small topographic variations [23]. The detailed simulation information is relevant for assessing the flood loss and risk in an urban area.

For flood management, the loss and risk are identified effectively based on simulation approaches [5,8]. The simulation model is applied to determine the maximum water depths during typhoon events without sufficient flood depth measurements. Since the patterns of residential areas in the study area are distributed, inundation depths in a residential area can be better identified using the high -resolution simulation model. However, the benefits of using the complex building representation and high-resolution flood inundation model are uncertain because of the lack of sufficient data for model calibration and validation [23].


**Table 3.** Flood losses from different spatial resolutions.


**Table 4.** Flood loss comparison oflmresolution.

#### *3.3. Vulnerability and Risk Maps*

A vulnerability map was derived from four vulnerability factors. Figure 6 shows the vulnerability results classified into five categories. Every vulnerability range is represented by a unique color. Level 1 (dark green color) indicates the areas with a very low vulnerability. Level 2 (light green) indicates the areas with low vulnerability, and level 3 (yellow color) indicates the areas with a moderate level of loss. Levels 4 and 5 (orange and red colors) indicate areas with high and very high vulnerabilities, respectively. The vulnerability map shows that the areas with the highest vulnerability are located in the south and center line because they are close to the river and are composed of lowland with high population density.

Figure 7 shows the risk map of the study area. The risk level is highly dependent on flood hazard and vulnerability. The flood risk is classified into the five levels of colors. Level 1 (dark green color) indicates the areas with very low risk. Level 2 (light green) indicates the areas with low risk, and level 3 (yellow color) indicates the areas with a moderate level of loss. Levels 4 and 5 (orange and red colors) indicate the areas with high and very high risks, respectively. In the risk map, most areas have no risk of flooding even though some have high vulnerability. The result also showed that the most areas at risk are located in the South District near the river area. The areas at risk of flooding are located around the river and have low elevation. The Tainan City Government can use the risk map as a reference to reduce the level of vulnerability in the areas around the river. Such a goal can be achieved by adding fire station units and reducing population density around the river. Reducing the value of this vulnerability factor will also have an impact for decreasing the risk of flood loss in the future.

The risk map analysis indicated that most areas (883,526 m2) have a very low risk level, as denoted by the dark green color in Figure 7. A total of 269,192 m2 areas are located in the low-risk area, as indicated by the light green color. Only 8099 m<sup>2</sup> areas are located in the moderate level of risk, as represented by the yellow color. In the flood risk maps, the southwestern parts of the district near the river were most vulnerable to floods. The risk value was calculated based on the water depth from the hazard map and the value of the vulnerability map. Most areas are not at risk for floods but some have high vulnerability. The government can reduce the risk value and reduce the social vulnerability value through population reduction and adding fire station in areas with high water depth value. Figure 8 identifies all residential areas that have risk values and very low risk level. Risk value is affected by the vulnerability factors and the hazard map in the risk-generating process. In Figure 8, no residential polygon that has a high level risk or a very high level risk was determined. Therefore, the residential polygon in the study area faces a low potential risk against flood disaster loss in the future. The government needs to pay attention to the areas, even those with less than level 3 of risk.

**Figure 6.** Vulnerability map in five levels from low (green color) to high (red color).

**Figure 7.** Risk map in five levels from low (green color) to high (red color).

**Figure 8.** Risk map in a residential area: (**a**) risk in the entire study area, (**b**) details in the Rende District, and (**c**) details in the South District.

Flood risk is a product of the probability of the flood event occurrence and its consequence to society and the environment [24–26]. The 24-h accumulated rainfall for Typhoon Megi or Haitang exceeds 350 mm over the 200-year return period. The risk map over the 200-year return period was generated in this study. The risk maps with various flood probability distributions would be recommended in the development of flood loss and risk assessment. However, the uncertainty in flood loss and risk assessment can lead to significant over-estimations or under-estimations [27]. The main uncertainties lie in the scarcity of data, the assessment model, and the human dimension e.g., the uncertainties within DEM accuracy, parameterization of the hydraulic model, house types, and field surveys [27,28]. The further study will address the impact of uncertainty in flood loss and risk assessment. In addition, the effects of the urban imperviousness [29], land development, and climate change can be quantified in hazard mitigation planning.

#### **4. Conclusions**

This study offers a GIS-based flood loss and risk assessment model to analyze the direct loss and risk impacts caused by floods in a residential urban area with small-sized residential buildings. Since the pattern of a residential area in the study area is dispersed, the water depths of a residential area are easier to identify using the high-resolution model. This study integrates high-resolution inundation simulation to generate fine-resolution flood hazard and risk maps. The high-resolution model provides detailed information of water depth in each pixel inside the residential area. The effect of the simulation resolution on flood loss was identified. The spatial resolution of simulation greatly affects the results of loss value. The assessment model shows different resolution estimations of a flood-inundated area and its corresponding loss. Moreover, vulnerability factors used in this study were prepared based on the importance of each of the vulnerability value factors in the study area. The AHP identified the weights of the vulnerability factors. The risk map is the result of combining flood hazard and vulnerability maps. Spatial flood risk maps show that the southwestern parts of the district near the river are most vulnerable to floods.

Future studies will consider flood risk under various return periods in order to identify the expected annual damage for each return period. Data collection of filled survey and social vulnerability factors, such as household income, and uncertainty analysis must also be added to develop a reliable loss and risk estimator.

**Author Contributions:** Z.A. contributed to model development, paper writing, and reporting of the results. H.-J.C. contributed to the study design, the model concept, data preparation, and the writing of the manuscript. Y.-L.K. and Y.-C.H. contributed to the improvement of the draft manuscript. H.K.W. contributed to the 3Di model. M.Z.A. contributed to paper processing. All authors have seen and approved the final version.

**Funding:** MOST 107-2119-M-006-024 and 107-2622-M-006-001-CC2 funded the APC.

**Acknowledgments:** The students from the International Master Program on Natural Hazards Mitigation and Management, NCKU, supported this work. The authors thank reviewers for valuable comments and suggestions.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **The Development and Application of the Urban Flood Risk Assessment Model for Reflecting upon Urban Planning Elements**

#### **Kiyong Park <sup>1</sup> and Man-Hyung Lee 2,\***


Received: 15 March 2019; Accepted: 25 April 2019; Published: 1 May 2019

**Abstract:** As a city develops and expands, it is likely confronted with a variety of environmental problems. Although the impact of climate change on people has continuously increased in the past, great numbers of natural disasters in urban areas have become varied in terms of form. Among these urban disasters, urban flooding is the most frequent type, and this study focuses on urban flooding. In cities, the population and major facilities are concentrated, and to examine flooding issues in these urban areas, different levels of flooding risk are classified on 100 m × 100 m geographic grids to maximize the spatial efficiency during the flooding events and to minimize the following flooding damage. In this analysis, vulnerability and exposure tests are adopted to analyze urban flooding risks. The first method is based on land-use planning, and the building-to-land ratio. Using fuzzy approaches, the tests focus on risks. However, the latter method using the HEC-Ras model examines factors such as topology and precipitation volume. By mapping the classification of land-use and flooding, the risk of urban flooding is evaluated by grade-scales: green, yellow, orange, and red zones. There are two key findings and theoretical contributions of this study. First, the areas with a high flood risk are mainly restricted to central commercial areas where the main urban functions are concentrated. Additionally, the development density and urbanization are relatively high in these areas, in addition to the old center of urban areas. In the case of Changwon City, Euichang-gu and Seongsan-gu have increased the flood risk because of the high property value of commercial areas and high building density in these regions. Thus, land-use planning of these districts should be designed to reflect upon the different levels of flood risks, in addition to the preparation of anti-disaster facilities to mitigate flood damages in high flood risk areas. Urban flood risk analysis for individual land use districts would facilitate urban planners and managers to prioritize the areas with a high flood risk and to prepare responding preventive measures for more efficient flood management.

**Keywords:** climate change; urban flood risk; flood damage; urban disaster; land use

#### **1. Introduction**

The degree and scale of flood hazards have massively increased with the changing climate in the last decades. The larger-scale flash floods than in the past have brought fast-moving and rapid-rising water with force, resulting in tremendous life and property losses, as well as social disruption worldwide [1].

Floods are natural processes in river systems [2,3]. However, humans have occupied and urbanized floodplains for their urbanization attractiveness due to their planar morphology and water availability [4,5]. The latter has translated into the growth of flood risk zones for human settlements and infrastructure due to a greater concentration of people and structures [6,7]. Recent urban growth has not taken the space that rivers require to temporarily store flows during floods into consideration [4]. Global efforts have focused more on implementing flood control infrastructure, such as dikes, dams, and channelization, but despite these efforts, modern cities still remain vulnerable to flood risk [8].

Furthermore, the potential for flood casualties and damages is also increasing in many regions due to the social and economic development, which implies pressure on land-use, e.g., through urbanization. Flood hazard is expected to increase in frequency and severity, through the impacts of global change on climate, resulting in severe weather in the form of heavy rains and river discharge conditions [9].

More cities are becoming hotspots for risk and disaster [10], mainly as a result of rapid urbanization, population growth, and the impacts of climate change [11].

Throughout the world, cities have been affected by the increasing impacts of floods. In the period 1998–2008, more than 2900 events were registered [12]. Recent estimates indicate that urban zones exposed to flooding will increase 2.7 times by the year 2030 [13]. Globally, the previewed scenario demonstrates an increase in the frequency and magnitude of floods due to the changes in precipitation patterns resulting from climate change and accelerated urban expansion [14–17]. It is estimated that by 2050, 70% of the world's population will be concentrated in urban areas [13,15,18].

Countermeasures to urban flooding should be considered in long-term perspectives because the impacts of climate change are unpredictable and complex [19].

The development of appropriate flood risk management strategies for flooding should be considered in long-term perspectives (e.g., expected future rainfall amounts, although climate change impacts can be unpredictable and complex) and should focus on increasing an area's resilience to flooding. Recently, as rainfall has become more concentrated over short periods of time, substantial amounts of damage due to pluvial flooding have occurred in urban areas. This includes damage of social infrastructures, as well as losses of human life and properties. Conventionally, storm water management has been focused on drainage systems via underground pipes. However, these conventional approaches (structural measures) have had problems in many cases because they were designed based on historical events. This makes it difficult to deal with extreme rainfall events that exceed the designated capacity. The frequency of extreme rainfall events is expected to increase with projected climate change, which may cause conventional stormwater management systems to be exceeded [20].

While information on the distribution of flood hazard at a national to global scale is extremely valuable, knowledge about how it is distributed with reference to the population and built assets is critical to water resource managers, city planners, and policy-makers [21].

In the past, natural disasters were accepted as an unavoidable calamity, and the emphasis was placed on how to cope with the aftermath of the disaster, rather than how to prevent it. However, the prevention measure shall shift its focus to preemptive prevention from a holistic point of view considering the large scale, unexpectedness, and complexity of the disaster. This study suggests the development and application of the urban flood risk assessment model to reflect upon urban planning elements (land use, building characteristics). In this respect, measuring in terms of urban planning is the most effective preemptive measure to reduce damage from natural disasters, and it focuses on the idea that the development shall be inhibited in the area that is vulnerable to flood damage and carries a high risk from the stage of land utilization planning in order to minimize the damage. In addition, an urban utilization plan may be an important policy measure in achieving the goal of keeping the city safe by restricting development in areas vulnerable to natural disaster. It is important to note that the impact of urban land use on disaster damages may escalate when unplanned and thoughtless developments prevail without the construction of appropriate infrastructure facilities and that this may become a reality while damages are currently increasing in urban areas due to reckless land use. Reducing the area for urban land use would naturally reduce the resilience of disaster

damages. However, it would be difficult to expect such a reduction in the area for urban land use as there has not been any reduction recently. Therefore, the purpose of this study is to analyze the flood risk by concentrating on land use and building characteristics reflecting on urban planning elements. The flood risk is classified to maximize the spatial efficiency during the flooding events and to minimize the following flooding damage. The paper concludes by stressing the necessity of land use planning measures to prevent flooding. It is possible to classify flood risk areas according to the use of districts and administrative districts, evaluate priorities, and enable efficient management by selecting areas with a high flood risk.

#### **2. Methods**

In Sections 2.1 and 2.2, the theoretical perspectives are presented. Following this, Section 2.3 explores the relationship between urban space and flooding. In Section 2.4, the overall process of this study is described. Furthermore, in Section 2.5, the evaluation item of the analysis is presented. In Section 2.6, the fuzzy classification is introduced. Finally, in Section 2.7, the background of the selection of the analysis area is given.

#### *2.1. Vulnerability and Exposure of Urban Flood*

The risk of disaster is discontinuous and local. The only way to reduce the risk is to reduce the vulnerability within the system. The risk that places stress on the system is purely natural and is caused by variations outside of the system, and there is nothing that people can do to reduce the risk itself. However, human beings can only reduce the vulnerability to natural disasters by changing social systems or social infrastructure.

The concept of vulnerability defined by the IPCC is based on the view that combines vulnerability as the result of external stress and vulnerability as an internal state of the system. Vulnerability is the state of being easily affected by adverse impacts of climate change, including climate variations and extreme events, or the degree of inability to cope with it, and is the function of the characteristics, scale, and speed of climate variations to which a system is exposed and the sensitivity and susceptibility of the system [22,23]. In this respect, vulnerability should be regarded as potential exposure to damage rather than an estimation of damage due to a sudden change in climate or stress based on probability.

In the 8th IHP Strategic Plan of UNESCO, the specific concept of vulnerability was defined as in Figure 1, based on the disaster characteristics-damage relationship, by focusing on the idea that there is a difference in the scale of damage due to vulnerability, even for natural disasters of the same intensity. As explained above, the conceptual definition of vulnerability may be slightly different. However, it can be summarized that when the impact of natural disasters is great, the vulnerability of a system is considered high if it has a small capacity to cope with it. On the contrary, the capacity to cope with disaster can be high, even when the impact of natural disasters is high. This system can have the opportunity for development while coping with disasters appropriately. If the system has a small capacity to cope and the natural disaster has a small impact, the system may still have residual risks. If the impact is small and the system has a great capacity to cope, the system may promote sustainable development.

Degree of exposure has been defined as the person, property, system, or other components in dangerous areas affected by a potential loss and can be measured by the number of people and assets in the area. To quantitatively estimate the risk involved with the hazard, the exposure can be combined with specific vulnerabilities of the elements exposed to a specific risk.

**Figure 1.** Relationship between the damage caused by hydrological extremes and the severity of the event [24].

#### *2.2. Risk of Urban Flood*

The risk can be defined as the possibility that the loss will be greater than what was generally expected. In other words, risk is a concept of probability based on possibility, but it can be re-defined as the difference between expectation and reality considering the amount of loss [25].

According to ISO 31000, risk is defined as the "effect of uncertainty toward an object" or "the combination of the probability that an event would occur and the outcome of the event." The probability that an event will occur is related to the source of the disaster and its properties, and the outcome is related to vulnerability, which influences the scale of damage and the capacity to reduce damage.

There are still many cases where the meaning of risk is not clearly defined and is used confusedly to mean the degree of risk, hazardousness, or vulnerability. In this study, the components of risk considered are riskiness, vulnerability, and exposure, and also include the ability to adapt to risks.

#### *2.3. Relationship between Urban Space and Flooding*

Urban spaces carry a high risk of massive and complex human and property damages as they are quite vulnerable to disasters due to the concentration and densification of the population and facilities, the increased interdependence of various urban facilities and activities, the development of lowlands and slopes, and the development of underground space [26].

According to the fifth report of the IPCC, the artificial activity of human beings accounts for an absolute ratio of 95% of the cause of climate change. Paradoxically, however, human beings are also the ones that are most affected by diversified and escalated damages from various disasters caused by climate change. The city is a space where such human activities are concentrated, and the city, as well as the environment that surrounds people, influence and are influenced by climate change. Recent studies on disasters show that the impact of a natural disaster is caused by the interaction between natural phenomena and people [27]. Eventually, the spaces that explain this phenomenon gather and form a city.

Urbanization and climate change have direct and indirect impacts on disasters or cause such disasters within urban spaces. Urbanization and climate change, which have great relevance across all areas of the urban space, are considered to be important paradigms related to natural disasters. If urbanization was the global phenomenon that drew the attention of the whole world in the 1900s, the international issue of the 2000s is climate change. The most common opinion among experts is that the influence of climate change will appear more prominently in society along with the acceleration of urbanization [28]. They argue that climate change is increasing natural disaster damage around the world, and the trend of urbanization is escalating damages [22].

The urbanization rate around the world expanded from 23.8% in 1950 to 50% in 2010. Urbanization is a population-concentrating phenomenon that appears after the concentration of various facilities and functions, and it is accompanied by various environmental problems [29]. The KOSIS (Korean Statistical Information Service) data also predicted that the urbanization rate will reach 60% or over as of 2030. Although the increase rate of urbanization has decreased after exceeding 50% in 2010, the rate of urbanization itself is expected to increase continuously. Assuming that the current environment remains the same, it is clear that the damage resulting from natural disasters will increase [30].

Urbanization is concentrating the limited urban space with artificial factors such as population, industries, and facilities, and increasing urban spaces vulnerable to natural disasters through indiscreet land use. In particular, unpredictable natural disasters due to climate change not only increase the risk even further in urban spaces with high vulnerability and urban spaces with less recovering capabilities but also form a vicious cycle of causing damage from natural disasters.

#### *2.4. Methods of Risk Analysis*

The basic ideas and concepts are shown above. The vulnerability analysis was carried out based on urban space characteristics and building characteristics, including non-structural characteristics, by setting 100 m × 100 m geographic grids as the evaluation units alongside urban flooding analysis based on environmental factors such as topographical characteristics and rainfall characteristics. The urban flood risk was assessed and analyzed based on these results and using the risk mapping technique. Concepts and processes important to their construction are described herein and in Figure 2.


**Figure 2.** Analysis framework of urban flood risk.

#### *2.5. Evaluation Item*

Objective indicators and values were calculated to analyze the risk of flooding. Each indicator was selected based on the following standards.

The land price for each use district provided by the Korea Appraisal Board is considered an important indicator of flood damage because property damage represents the major loss in the case of a flood [31]. As the Ministry of Construction and Transportation established long-term water resource use planning in 2001, local characteristics of water resources were carefully examined to identify investment priority in watershed planning and development. As a result, Potential Flood Damage (PFD) was estimated, for which the property value plays an important role [32]. In the research of Han [33], the correlation between various indicators of flood vulnerability and the damage costs for flooding from 1971 and 2000 in different watersheds was analyzed. As a result, the property density had a significant correlation with the damage cost of flooding, which was a result attributed to the estimation method of flood damage cost largely reflecting the land price [33].

The underground area index is directly related to the flood reference system to prevent building submergence within the flood-water disaster prevention criterion for buildings [34]. In a study of Jang [35], the vulnerability of buildings due to flooding was found to be closely related to the presence and size of underground space. The top 10 items with the greatest damage were located in places where there were significant portions of underground space, while the bottom 10 items had no underground space. Therefore, the size of underground space was important for determining the vulnerability of buildings. Additionally, it is considered that rainwater adversely affects the building structures by moisture, warping, cracks, and corrosion when rainwater fills the underground space or penetrates through the wall [35].

The floor area ratio is a concept of disaster that collectively refers to the case where urban spaces lose their function and damages are multiplied due to a high density. Because of the high utilization of urban space, complex factors such as underground space and humans instigate unpredictable large-scaled disasters. In the United States, the Community Rating System (CRS), which was established in 1990 to operate the flood insurance system, has been employed to evaluate the building height issued by the Federal Emergency Management Agency (FEMA) for 19 items. Since the damage is expected to vary depending on the area and density of the building, the indicators are selected by considering urban functional damage [31].

A decline of building shows that the incidence of accidents is frequent as the durability reaches its limit. Moreover, the development-centered paradigm of urban planning has escalated an event to a series of failures, resulting in a disaster. Therefore, urban disasters have a high potential to result in a collective paralysis of urban functions [32].

The material of buildings is a robustness-related index. Unanwa et al. (2000) stated that property values of the building and exterior wall types were set. In the case of building interiors, it has been shown that many damages occur during flooding due to the heavy use of wood [36].

#### *2.6. Fuzzy Classification*

Fuzzy set theory has been developed and extensively applied since 1965 [37]. It was designed to supplement the interpretation of linguistic or measured uncertainties for real-world random phenomena. These uncertainties could originate with non-statistical characteristics in nature that refer to the absence of sharp boundaries in information. However, the main source of uncertainties involved in a large-scale complex decision-making process may be properly described via fuzzy membership functions [38].

Fuzzy classification is, alongside neural networks [39] and probabilistic approaches [40], a very powerful soft classifier. As an expert system for classification [41], it takes into account uncertainty in sensor measurements, parameter variations due to limited sensor calibration, vague (linguistic) class descriptions, and class mixtures due to a limited resolution. Fuzzy classification consists of an

n-dimensional tuple of membership degrees, which describes the degree of class assignment μ of the considered object obj to the n considered classes.

$$f\_{\text{class},obj} = \begin{bmatrix} \mu\_{\text{class}\_1}(obj) \ \mu\_{\text{class\\_2}}(obj), \dots \ \mu\_{\text{class\\_n}}(obj) \end{bmatrix} \tag{1}$$

Crisp classification would only provide information on which membership degree is the highest, whereas this tuple contains all information about the overall reliability, stability, and class mixture. Fuzzy classification requires a complete fuzzy system, consisting of the fuzzification of input variables, resulting in fuzzy sets, fuzzy logic combinations of these fuzzy sets, and defuzzification of the fuzzy classification result to get the common crisp classification for map production. Fuzzy logic is a multi-valued logic quantifying uncertain statements. The basic idea is to replace the two Boolean logical statements "true" and "false" by the continuous range from 0 to 1, where 0 means "false" and one means "true", and all values between 0 and 1 represent a transition between true and false. Avoiding arbitrary sharp thresholds, fuzzy logic is able to approximate real-world complexity much better than the simplifying Boolean systems do. Fuzzy logic can model imprecise human thinking and can represent linguistic rules. Hence, fuzzy classification systems are well-suited to handle most sources of vagueness in remote sensing information extraction. The mentioned parameter and model uncertainties are considered by fuzzy sets, which are defined by membership functions. Fuzzy systems consist of three main steps, including fuzzification and the combination of fuzzy sets [42].

In the flood damage risk classification analysis, it is very difficult to quantify and compare the flood damage in commercial and residential areas when the same area is submerged. In other words, it is ambiguous to express the level of flood damage as numerical values. Therefore, to overcome the linguistic ambiguity for decision-making in previous studies and to analyze complex relationships between different indicators and indices, the fuzzy logic method is adopted to perform a more objective analysis of flood risk by deriving quantitative and accurate indicators.

#### *2.7. Analysis Area*

The distribution of flood damage in Korea analyzed by the Korea Ministry of Land, Infrastructure, and Transport is shown in Figure 3 [37], and Changwon City is selected as the target area due to its advantages of data construction because it includes areas with both high and low flood damage [31].

**Figure 3.** Distribution of national flood damage in Korea [43].

Changwon City has become one of the first successful administrative integration models in the nation as it integrated Changwon, Masan, and Jinhae cities in July 2010, and it is becoming the first growth base of the Southeastern Greater Economic Zone in Korea. The administrative district consists of the five districts of Masan Happo-gu, Masan Hoewon-gu, Seongsan-gu, Euichang-gu, and Jinhae-gu, 351 legal smaller districts, and 62 administrative smaller districts. The administrative area of Changwon City is 746.58 km2 [44].

As a result of examining the land use situation from 1975 to 2007, using the land cover map of Changwon City, the urbanization rate greatly increased from 3.95% in the 1990s to 13.61% in recent years. In addition, a countermeasure to flood damage in Changwon city is needed because the urbanization rate is planned to increase from 14.13% in 2020 to 15.72% in 2025, according to the step-by-step development plan of Changwon City Basic Plan 2025 [44]. In response, the risk according to flood damage was analyzed in Changwon City.

#### **3. Results and Discussion**

#### *3.1. Analysis of Vulnerability*

To select the indicators capable of assessing the flood vulnerability to climate change, indicators related to land use and building in the non-structural aspect that were proposed in previous research were selected, and the final indicators were selected through content validity analysis. To calculate the quotient, the fuzzy methodology was used to derive the position function, and the fuzzy inference rules were established and implemented to analyze the vulnerability of the districts of Changwon, as seen in Figure 4. As all the spaces, buildings, and facilities in the city are basically the subject of disaster prevention, the characteristics of the areas were analyzed in 100 m × 100 m geographic grids, which maintains the livelihood of the citizens and the urban function.

**Figure 4.** Flood damage scheme: (**a**) land price—floor area ratio, (**b**) land price—underground area, (**c**) land price—decline of building, (**d**) land price—material of building, (**e**) floor area ratio—underground area, (**f**) floor area ratio—decline of building, (**g**) floor area ratio—material of building, (**h**) underground area—decline of building, (**i**) underground area—material of building, and (**j**) decline of building—material of building.

There were a total of 177,193 cases of the officially assessed land prices for each lot in Changwon city, and the data were constructed using the individual officially assessed land prices provided by the Korea Appraisal Board. There were a total of 66,269 cases of the floor area ratio, 15,021 cases of underground area, 97,346 construction declines of buildings, and 69,932 cases of building materials according to the building register. The map of Changwon City was divided into 77,737 cells of 100 m × 100 m geographic grids, and maximum and minimum values, as well as the average value, were standardized for the officially assessed land price, floor area ratio, underground area, decline of building, and building material indices, and fuzzy analysis was conducted to evaluate the vulnerability. As a result of the fuzzy analysis, the index with the highest vulnerability to flooding was land

price, followed by underground area, floor area ratio, decline of building, and material of building. The implication of the result is that the land price and the underground space are direct indicators of property damage and physical damage in the case of flooding, whereas the floor area ratio is directly related to the only corresponding floor and indirectly related to other floors for its inconvenience in urban functional aspects. Furthermore, it is shown that decline of building and material of building are robustness-related indexes, which is less important than other indicators.

A higher fuzzy score indicates a higher degree of flood risk (greater flood damage), and a lower fuzzy score indicates a lower degree of flood risk (no damage). The fuzzy values for 100 m × 100 m geographic grids were derived from fuzzy analysis through standardized values of each indicator for officially assessed land price, floor area ratio, underground area, decline of building, and building materials, listed in Table 1.


**Table 1.** Fuzzy values for 100 m × 100 m geographic grids.

Based on the fuzzy analysis value of each indicator, 100 m × 100 m geographic grids were prioritized for flood damages. When the same area was flooded, the area with the most flood damage in a social and economic sense, which is the area with high vulnerability, was analyzed as the area with the highest development density in Changwon centering on the central commercial area. In addition, the areas that played key functions in the city in the past as the old town center, although currently declining, showed high vulnerability.

Once the vulnerability grade for land use has been determined based on standardization of the fuzzy score, it is possible to classify based on the level of risk. The standard color was determined by designating the area with the highest vulnerability as the red zone, the area with high vulnerability as the orange zone, the area with intermediate vulnerability as the yellow zone, and the area with low vulnerability as the green zone. If the resulting values are applied to the map of Changwon, it is possible to derive the land use classification map, as shown in Figure 5.

**Figure 5.** Urban flood vulnerability analysis.

#### *3.2. Analysis of Exposure*

The analysis using HEC-Ras is a typical flood analysis model that conducts a simulation by constructing the river crossing data, the center line of the river, the river bank line, and and using the flood amount and the flood level as the boundary conditions. This analysis process is shown in Figure 6.

**Figure 6.** Flood depth and flooded area analysis using HEC-Ras.

The urban flood map prepared by using the one-dimensional model HEC-Ras based on the basic environmental factors such as rainfall data and topographical data developed by the Ministry of Interior and Safety was used in this study [45]. The urban flood map is shown in Figure 7 below. The urban flooding area evaluation confirmed that the flood depth was high in the northern area adjacent to the river and reservoir and also high in the areas adjacent to Changwon City Hall, which play key roles in Changwon City and are also considered as the town center.

**Figure 7.** Urban flood exposure analysis.

#### *3.3. Analysis of Risk*

Figure 8 displays the graded urban flood risk by overlapping the non-structural factors, which are the criteria for socioeconomic damages, derived based on the study by Kron [46] and Brooks et al. [47], with the flood simulation analysis map. This map analyzed the flood depth based on the land use classification map, formed of vulnerability analysis data and data on disaster characteristics (rainfall, topography), including the degree of exposure.

The analysis of urban flood risk confirmed that the red zone was mainly found in the central commercial area, where various infrastructures, including public institutions that play key functions in the city, are located. This appears to be related to the development considering the economy, convenience, and efficiency of the city. The orange zone shows a fan-shaped distribution centering on the red zone, which also demonstrates the tendency of branching out from major functions and socio-economic parts of the city, and by use area, they were classified into distribution commercial areas, including general commercial areas and quasi-residential areas. The distribution of the yellow zone centered on the residential area confirms that it is distributed in the area that was formed a long time ago. It is considered that this is closely related to the aging of buildings. The green zone was mainly distributed in green areas where the environmental values or the environment rather than the value of development land use need to be preserved.

**Figure 8.** Urban flood risk analysis.

#### *3.4. Overall Analysis Results*

The results of the risk analysis by district are shown in Table 2. The area with the highest urban flood risk and the highest percentage of red zone in urban flood risk was Euichang-gu (13.07%), followed by Seongsan-gu (4.07%), Masan Hoewon-gu (2.68%), Masan Happo-gu (1.87%), and Jinhae-gu (1.78%), in respective order. The orange zone was also distributed more frequently in the window (14.78%) and in Seongsan-gu (12.37%) than in other areas. The distribution of the yellow zone was most prominent in Euichang-gu (3.96%), followed by Seongsan-gu (2.10%) and Masan Happo-gu (1.39%), respectively. The green zone, which is the safe area with the lowest urban flood risk, was most distributed in Masan Happo-gu (92.21%), Jinhae-gu (86.99%), and Masan Hoewon-gu (85.89%).


**Table 2.** Risk for each administrative district.

Masan Happo-gu has been active as the general commercial area since being developed a long time ago centering on Masan Port and served as the transportation hub with railroads in the past, although it is now an abandoned railroad site. The neighborhood commercial area and semi-residential area were formed around the district, and the urban flood risk in these areas appears to be high. The flood risk in the industrial areas is relatively low as they were recently formed sporadically in mountainous areas. Considering the results in terms of building characteristics, a high fuzzy value for construction year across the whole use area indicates that the city was formed long ago and that the aging of buildings is becoming serious. Additionally, the basement is distributed more in the industrial area than other use areas.

In Masan Hoewon-gu, a general commercial area is mainly formed around Masan Station and its neighborhood, with commercial areas, semi-residential areas, and general residential areas around it. The industrial area was extensively formed in the southeastern area in the past and the flood risk is more or less high as the vulnerability of the building materials is high. According to building characteristics, the construction year was higher than other indicators in the same way as Masan Happo-gu, indicating that the buildings are aged. In particular, the officially assessed land price of the neighboring commercial area is much higher than other indicators, and this suggested that there is a great demand for daily necessities and services from the residents of the neighborhood residential area and the area appeared to have been specialized for that purpose.

Seongsan-gu is a comparatively new city area, where the commercial area and residential area have been recently developed, centering on public institutions including the city hall, and the fuzzy value for the materials is low as the buildings are not aged and many of them were constructed recently. The officially assessed land price and floor area ratio are high throughout the use district, particularly in the general commercial area, central commercial area, and semi-residential area, indicating that it is an area with high asset value. The fuzzy value is high, particularly as the highly dense central commercial area is activated, and also as many semi-residential areas incorporate commercial facilities among the residential areas. The reason why the fuzzy value is low in the green area is that it has a low asset value and it is quite unlikely that development of the area would occur in this region.

Euichang-gu is located near Seongsan-gu. Like Seongsan-gu, the commercial areas and residential areas have recently been developed around public institutions, including the provincial government buildings, and the fuzzy value of the city's land price is considerably higher than other indicators, indicating that the asset value of the region is high. The green area demonstrated the lowest fuzzy value like other areas, indicating the lowest vulnerability.

Jinhae-gu has a low asset value and the fuzzy value of the construction year was high throughout the whole use district, indicating that the city has existed for a long time and the buildings are aged compared with other areas. In the Jinhae area, semi-residential areas and general commercial areas showed more or less a higher vulnerability, and fuzzy values of industrial areas and green zones were low.

By use district, the ratio of red zones of commercial areas, including central commercial areas and general commercial areas, was higher, as shown in Table 3, and the risk declines in the order of residential areas > industrial areas > green areas.

These results imply that the precautionary measures for flooded areas should primarily focus on the central commercial areas among the use district of the urban areas. In addition, the green zones have the least risk, as shown in the analysis results, and the potential to reduce or minimize the flood risk and may demonstrate more flexibility than the structural measures in various aspects considering the uncertainty of climate change. Moreover, the green zones are more effective than aged structural systems (sewage, storage facilities) in the long term, and not only reduce the rainwater runoff rate, but also reduce water pollution, which makes them one of the most useful resources, while the structural (sewer) system generates pollution and increases water pollution. Therefore, the results confirmed the fact that the flood risk could be reduced by arranging green zones appropriately.


**Table 3.** Risk for each use district.

In order to apply measures in terms of urban planning to mitigate urban flood damages, it is necessary to establish an objective and reasonable method of setting dangerous areas, such as disaster prevention areas, and this study emphasizes that measures for disaster prevention shall be established by dividing the results of the urban flood risk analysis into 100 m × 100 m geographic grids and primarily considering the areas with a relatively high ratio of red zones.

#### **4. Conclusions**

The development and expansion of the city have brought about a variety of environmental problems, and the disaster in the city is gradually becoming larger and diversified due to the influence of climate change. Under this circumstance, the importance of urban disaster prevention has been emphasized, and this study focused on flooding, which accounts for a significant portion of disasters related to climate change. This study intended to contribute to the reduction and minimization of flood damage in the case of heavy rainfall in urban areas, which are concentrated with population and major facilities, by increasing the urban spatial efficiency through the grading of flood risk. This study developed and applied a flood risk assessment model to Changwon City and obtained the results described below.

First, as an assessment model, an objective and scientific evaluation model which reflects urban spatial concepts and characteristics of buildings was developed by applying an urban flood risk assessment model. Additionally, the non-structural characteristics were identified and the land use and vulnerability of building units were analyzed to minimize damage from natural disasters as a long-term measure. This study intended to construct an urban spatial model to primarily apply to less vulnerable and risky areas and suggested a new paradigm for the integrated study of 'environment + city' rather than individual planning or the study of environmental planning and urban planning. The developed model provides urban planners with a flood risk map when developing flood risk management strategies to ensure that a strong decision is made on flood management options such as land use, which represents a major challenge for flood risk, and disaster management challenges faced by climate change. Therefore, the ability of urban planners to make such decisions is important in reducing the social, economic, and physical (infrastructure) impact of flood damage.

Second, a set of indicators with non-structural characteristics that are related to land use and building characteristics were derived to assess vulnerability. The priority for each 100 m × 100 m geographic grid for flood damage was determined based on the fuzzy analytical values for each indicator: officially assessed land price, floor area ratio, underground area, decline of building, and material of building. When the same area was flooded, the area with the largest flood damage in a social and economic sense, which is the area with high vulnerability, was the area with the highest development density, including the central commercial area. The areas that had key functions in the city in the past, as the old town center, although currently declining, demonstrated high vulnerability.

Third, an urban flooding map was constructed by using the HEC-Ras model based on the basic environmental factors, such as rainfall data and topographic data, which were developed by the Ministry of the Interior and Safety. The results of urban flooding area evaluation confirmed that the flood depth of the northern area adjacent to the river and reservoir is high. The flood depth was also high in the areas adjacent to Changwon City Hall, which play key roles in Changwon City, the target area, and which are also considered as the town center.

Fourth, the flood risk was analyzed by overlapping the results of vulnerability analysis and exposure analysis. As a result of analyzing the risk of urban flooding by four grades: green, yellow, orange, and red zones, it was confirmed that red zones were formed centering on the central commercial areas, where a variety of infrastructures, including public institutions with key functions in the city, were developed. This is related to the development of the economic, convenience, and efficient aspects of the city. The orange zone has a fan-shaped distribution centering on the red zone, which also shows the tendency of branching out from major functions and socio-economic parts of the city, and by using districts, they were classified into distribution commercial areas, including general commercial areas and semi-residential areas. The distribution of the yellow zone centered on the residential area, confirming that it is distributed in the area that was formed a long time ago. It is deemed that this is closely related to the aging of the buildings. The green zone was mainly distributed in green areas, where the environmental values or the environment rather than the value of development land use need to be preserved.

Lastly, the analysis of the urban flood risk in each administrative district showed that Euichang-gu (13.07%) had the highest urban flood risk with the highest percentage of red zone in Changwon, followed by Seongsan-gu (4.07%), Masan Hoewon-gu (2.68%), Masan Happo-gu (1.87%), and Jinhae-gu (1.78%), in respective order. This was attributed to the generally high property values and building density in the commercial areas of Seongsan-gu and Euichang-gu, which represent the central region for new town development. In other words, the land use plan should be established by allocating the green zone with the lowest urban flood risk and other use districts appropriately and disaster prevention facilities and space facilities that can reduce flooding should be allocated appropriately to prepare measures primarily for the areas with high risk. However, the difference in the analysis results of the five districts, which differed from each other, indicates that the degree of damage from disasters may vary according to the local environment and characteristics, newly developed town areas, the decline of the old town center and geographical location.

**Author Contributions:** K.P. (Kiyong Park) conceptualized the research, performed the formal analysis, and wrote the first draft of the paper. M.H.L. (Man-Hyung Lee) provided feedback on the research approach, and reviewed the first draft of the paper. All authors revised the paper and agreed on the final version of the paper.

**Acknowledgments:** This paper was financially supported by Ministry of the Interior and Safety as "Human resource development Project in Disaster management".

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
