Development of a New Generation of Flood Inundation Maps—A Case Study of the Coastal City of Tainan, Taiwan

water

Article Development of a New Generation of Flood Inundation Maps—A Case Study of the Coastal City of Tainan, Taiwan

Dong-Jiing Doong 1, Weicheng Lo 1,*, Zoran Vojinovic 2,*, Wei-Lin Lee 1 and Shin-Ping Lee 3 1 Department of Hydraulic and Ocean Engineering, National Cheng Kung University, Tainan 70101, Taiwan; [email protected] (D.-J.D.); [email protected] (W.-L.L.) 2 UNESCO-IHE, Westvest 7, 2611 AX Delft, The Netherlands 3 Disaster Prevention Research Center, National Cheng Kung University, Tainan 70101, Taiwan; [email protected] * Correspondence: [email protected] (W.L.); [email protected] (Z.V.)

Academic Editor: Zoran Vojinovic Received: 22 September 2016; Accepted: 31 October 2016; Published: 8 November 2016 Abstract: Flood risk management has become a growing priority for city managers and disaster risk prevention agencies worldwide. Correspondingly, large investments are made towards data collection, archiving and analysis and technologies such as geographic information systems (GIS) and remote sensing play important role in this regard. GIS technologies offer valuable means for delineation of flood plains, zoning of areas that need protection from floods and identification of plans for development and scoping of various kinds of flood protection measures. Flood inundation maps (FIMs) are particularly useful in planning flood disaster risk responses. The purpose of the present paper is to describe efforts in developing new generation of FIMs at the city scale and to demonstrate effectiveness of such maps in the case of the coastal city of Tainan, Taiwan. In the present work, besides pluvial floods, the storm surge influence is also considered. The 1D/2D coupled model SOBEK was used for flood simulations. Different indicators such as Probability of Detection (POD) and Scale of Accuracy (SA) were applied in the calibration and validation stages of the work and their corresponding values were found to be up to 88.1% and 68.0%, respectively. From the overall analysis, it came up that land elevation, tidal phase, and storm surge are the three dominant factors that influence flooding in Tainan. A large number of model simulations were carried out in order to produce FIMs which were then effectively applied in the stakeholder engagement process.

Keywords: ﬂood inundation map (FIM); 1D/2D SOBEK model; storm surge; stakeholder engagement

1. Introduction The relentless migration of people from rural to urban areas and increase in development activities are making considerable pressure on water management systems, and especially on the management of emergencies and disasters. In view of the threats of climate change, natural disasters such as floods are likely to strengthen this trend in the coming years [1–3]. Correspondingly, flooding in urban areas has become a growing priority for city managers and disaster risk prevention agencies, and those communities that are situated in the coastal zone are further threatened by storm surges and rising sea levels [4]. Flood prediction, prevention, and mitigation are essential. Structural measures play important roles, which should primarily focus on the protection of human health and safety, valuable goods and property. In this regard, many researchers and practitioners are placing their efforts on multi-objective optimization techniques for evaluation and comparison of different measures [5–8]. In many situations, effective mitigation of flood hazards requires a change of paradigm from structural defensive actions against floods to understanding and

Water 2016, 8, 521; doi:10.3390/w8110521 www.mdpi.com/journal/water Water 2016, 8, 521 2 of 20 modeling different causes of flood risk [9–11] and correspondingly developing a range of structural and non-structural measures that can minimize that risk. In addition to the more traditional structural measures, flood warning systems have been widely developed and applied [12–17]. However, the establishment of an operational flood warning system is often demanding and expensive task (e.g., establishment of a real-time monitoring network), and effective real-time simulation of floods poses a great challenge [18–22]. Therefore, it is of the utmost importance to provide timely and sufficiently accurate real-time flood information to decision makers and the public. A flood warning system produces information every few hours to one day by coupling models and observations with rainfall forecasts [23]. This approach is sufficient for urgent responses at the flooding stage. Another strategy is to produce Flood Inundation Maps (FIMs). The main purpose of FIMs is depict the potential depth and extent of flooded zones in relation to scenarios of the possible initiation of floods coupled with their consequences in the light of different control and mitigation actions. Such maps have been widely researched and used [24–28]. Although FIMs may have a certain degree of uncertainty due to data collection and modeling technologies applied [29], such maps offer valuable means to enhance our understanding of local flood risks. FIMs can provide such information that is required by municipal authorities to more effectively inform citizens and adopt appropriate flood management strategies in advance. The warning time provided by a FIM is usually three days or longer, which is sufficient enough to evacuate citizens within areas suffering high flooding potentials or take preventive actions. In some instances, such maps are used for flood insurance purposes [30,31]. In developing flood maps it is necessary to generate scenarios of the possible initiation of floods and evaluate their consequences in the light of different flood mitigation strategies. With numerical models, it is possible to explore the generation and propagation of floods around urban features such as road or pavement curbs which can play a significant role in diverting the shallow flows through urban areas. Numerous hydraulic models have been developed and applied for this purpose [18,32–35]. However, head losses owing to flow over or round such features are difficult to accommodate and a significant body of literature suggests that great care and caution should be applied in data processing and modeling techniques [36–43]. Data collection plays an important role for constructing a FIM. The lack of adequate flood observations during flooding events represents a great challenge in producing reliable flood models [44]. Development of a FIM needs geographical and hydrological data as well as historical records, operating rules of sluices and reservoirs and reports of regulation projects, as shown in Table1.

Table 1. List of data required for development of a ﬂood inundation map.

Geography Hydrology Others • DEM • Rainfall • Historical ﬂooded extent and depth • Aerial image • Discharge • Operating rules • River/drainage cross section • Water level • Regulation reports • Reservoir • Tide level • Hydraulic structures • Wave • Sewer system • Coastal dyke • Sea bathymetry • Land use condition

Countries such as USA, Japan, UK, Germany, Bangladesh and the Netherlands have developed their own flood inundation map libraries which are used for different purposes [20,24,32,45–48]. The first generation of FIMs was developed in Taiwan in 1999. The coarse DEM (200 × 200 m) was used and the sewer system was not included in the simulation. In 2007, the second generation FIMs were produced. The model simulations included sewer system and storm surge influences. The DEM resolution of 40 × 40 m was used. The present work describes the development of the Water 2016, 8, 521 3 of 20 third generation of FIMs in Taiwan. The DEM resolution of 5 × 5 m is used and numerous hydraulic structures (e.g., sewer and drainage systems, detention basin, and pumping stations) are included in the numericalWater 2016, 8 models., 521 The influence from storm surge and wave overtopping is particularly addressed3 of 20 in the numerical modeling work. Correspondingly, due to the complexity of the system and the local sewer and drainage systems, detention basin, and pumping stations) are included in the numerical conditionsmodels. a largeThe influence number offrom scenarios storm surge was modeled.and wave The overtopping present paperis particularly describes addressed the development in the of this thirdnumerical generation modeling of FIMs, work. and Correspondingly, discusses the due causes to the of flooding complexity in theof the coastal system city and of Tainan.the local conditions a large number of scenarios was modeled. The present paper describes the development 2. Theof this Study third Area generation of FIMs, and discusses the causes of flooding in the coastal city of Tainan. Taiwan is located in a subtropical zone in one of the main typhoon tracks. The average 2. The Study Area annual precipitation in Taiwan is 2500 mm, reaching 3000 to 5000 mm in mountainous regions. Most of theTaiwan precipitation is located in is a concentrated subtropical zone in thein one typhoon of the main and monsoontyphoon tracks. season The during average the annual summer. Taiwanprecipitation has experienced in Taiwan more is 2500 than mm, 350 typhoonsreaching 3000 and to one 5000 thousand mm in mountainous rainfall events regions. in the Most past of 100 the years. The maximumprecipitation hourly is concentrated precipitation in the reached typhoon 300 and mm, monsoon and the season maximum during one-day the summer. precipitation Taiwan has reached experienced more than 350 typhoons and one thousand rainfall events in the past 100 years. The 1748 mm, which is 93.4% of the world record [49]. The Water Resources Agency (WRA) of Taiwan maximum hourly precipitation reached 300 mm, and the maximum one‐day precipitation reached reported1748 that mm, approximately which is 93.4% 3000of the buildings world record are damaged[49]. The Water by floods Resources annually, Agency with (WRA) an associated of Taiwan loss of approximatelyreported that 400 approximately million USD, 3000 which buildings is approximately are damaged by 4.6 floods times annually, more than with the an lossassociated caused loss by fire damageof approximately in Taiwan. In 400 addition million to USD, extreme which precipitation is approximately during 4.6 typhoons,times more the than river the characteristicsloss caused by and topographyfire damage of Taiwan in Taiwan. are In important addition to factors extreme that precipitation lead to flooding. during typhoons, The rivers the in river Taiwan characteristics are short and haveand steep topography slopes that of Taiwan exceed are 1/100 important in upstream factors that reaches lead to and flooding. 1/200–1/500 The rivers in in downstream Taiwan are short reaches. Concentratedand have steep rainfall slopes in short that exceed and steep 1/100 river in upstream basins reaches generates and rapid 1/200–1/500 flow increases in downstream and flow reaches. peaks. TainanConcentrated is the rainfall oldest in city short located and steep in southwestern river basins generates Taiwan. rapid It is flow bordered increases by and the Taiwanflow peaks. Strait to the west,Tainan as shown is the in oldest Figure city1. Inlocated 2016, in Tainan southwestern City had Taiwan. a population It is bordered of 1.9 by million the Taiwan in its Strait 37 districts. to the west, as shown in Figure 1. In 2016,2 Tainan City had a population of 1.9 million in its 37 districts. 2 The total area of Tainan is 2200 km2 , with an average population density of 860 residents/km2 . The total area of Tainan is 2200 km , with an average2 population density of 860 residents/km . More Morethan than 80% 80% of of the the population population lives lives in ina 495 a 495 km km2 urbanurban area area (23% (23% of the of thecity) city) located located in southwestern in southwestern 2 TainanTainan near near the coast.the coast. The The population population density density in in the the Tainan urban urban area area is is 4500 4500 residents/km residents/km2. .

Figure 1. Location, river distribution and topography of Tainan City. Figure 1. Location, river distribution and topography of Tainan City. Six rivers run through Tainan: the Bajhang River, Jishui River, Jiangjun River, Tsengwen River, SixYanshui rivers River run and through Erren Tainan:River (from the north Bajhang to south). River, Midwestern Jishui River, Tainan Jiangjun is an River,alluvial Tsengwen plain of the River, YanshuiYanshui River River and and Erren Tsengwen River (from River, northwith a tofew south). hills and Midwestern mountains distributed Tainan is in an the alluvial east. The plain only of the

Water 2016, 8, 521 4 of 20

Yanshui River and Tsengwen River, with a few hills and mountains distributed in the east. The only river that runs through the urban area of Tainan is the Yanshui River, with a length of 41.3 km and a watershed area of 340 km2, as shown in Table2. The coastline of Tainan City is 63.7 km. Water 2016, 8, 521 4 of 20 Table 2. List of rivers run through Tainan City. river that runs through the urban area of Tainan is the Yanshui River, with a length of 41.3 km and a watershedRiver area of 340 km Length2, as shown (km) in TableWatershed 2. The coastline (km2) of TainanSlope City Mean is 63.7 Discharge km. (cms) Bajhang River 81 474 1/42 27 Table 2. List of rivers run through Tainan City. Jishui River 65 379 1/118 15 Jiangjun RiverRiver Length 24 (km) Watershed 158(km2) Slope 1/800Mean Discharge (cms) 5 TsengwenBajhang River River 13881 1177474 1/42 1/20027 75 Yanshui RiverJishui River 4165 340379 1/118 1/295 15 10 Erren RiverJiangjun River 6324 350158 1/800 1/786 5 8 Tsengwen River 138 1177 1/200 75 Yanshui River 41 340 1/295 10 Tainan has frequentlyErren River flooded63 in the past, see350 Figure 2.1/786 The floods that8 occurred in Tainan City were generally induced by multiple factors, including rainfall which exceeds the drainage system capacity,Tainan a poor has condition frequently of theflooded drainage in the system,past, see floodFigure defense 2. The floods failure, that rising occurred sea in level Tainan due City to high tideswere and/or generally storm induced surges, by and multiple urbanization. factors, including According rainfall to which historical exceeds records, the drainage several system typhoons, suchcapacity, as Morakot a poor in 2009 condition and Kongreyof the drainage in 2013, system, have flood hit Taiwan. defense These failure, typhoons rising sea were level accompanieddue to high by tides and/or storm surges, and urbanization. According to historical records, several typhoons, such abundant rainfall that caused serious damage, especially in Tainan. More than 300 projects have been as Morakot in 2009 and Kongrey in 2013, have hit Taiwan. These typhoons were accompanied by implementedabundant rainfall in Tainan that City caused to serious prevent damage, flooding especially over the in Tainan. past 10 More years, than but 300 these projects projects have been have not solvedimplemented the problem. in Tainan Tainan City City to is prevent selected flooding as the studyover the area past not 10 onlyyears, because but these of projects its significant have not urban developmentsolved the but problem. also because Tainan City rainfall is selected is the as dominant the study factor area not that only influences because of flooding. its significant urban Adevelopment real-time floodbut also warning because rainfall system is is the particularly dominant factor needed that influences for Tainan flooding. to enable more effective and immediateA real‐time responses. flood warning However, system real-time is particularly model needed simulations for Tainan are to computationally enable more effective demanding and and timeimmediate consuming responses. and However, as such real they‐time make model the simulations use of numerical are computationally models difficult demanding for practicaland real-timetime application.consuming and In as this such regard, they make the pre-produced the use of numerical flood inundationmodels difficult maps for offerpractical the real possibility‐time of overcomingapplication. issues In thatthis areregard, associated the pre with‐produced the use flood of numerical inundation models maps in offer real-time the possibility flood forecasting of overcoming issues that are associated with the use of numerical models in real‐time flood forecasting and decision-making. and decision‐making.

(a) (b)

Figure 2. Photos of flooding in Tainan City. (a) flooding on the street; (b) flooding in the underpass. Figure 2. Photos of flooding in Tainan City. (a) flooding on the street; (b) flooding in the underpass. 3. Development of Flood Inundation Maps 3. Development of Flood Inundation Maps 3.1. Methodology 3.1. Methodology In the present work, one‐dimensional (1D) hydraulic flow model was used to simulate flows in Indrainage the present channels. work, This one-dimensional model was then coupled (1D) hydraulic with the two flow‐dimensional model was (2D) used model to simulate to simulate flows in drainageflows channels. along the floodplain. This model Physically was then‐based coupled computational with the two-dimensionalmodel SOBEK which (2D) was model developed to simulate by flowsDeltares along the (https://www.deltares.nl/en/) floodplain. Physically-based was used computational in the present model work. SOBEK SOBEK whichhas been was successfully developed by Deltaresapplied (https://www.deltares.nl/en/ in various flood forecasting, drainage) was used system in the optimization, present work. irrigation SOBEK system has control, been successfully sewer appliedoverflow in various design, river flood morphology, forecasting, salt drainage intrusion and system surface optimization, water quality studies irrigation [44,50–54]. system control, sewer overflowSOBEK design,is a coupled river 1D morphology,‐2D hydrodynamic salt intrusion model. The and 1D surface St. Venant water equations quality are studies solved [ 44for,50 a –54]. series of cross‐sections of the main channel and the overbank perpendicular to the main channel. The water level is spatially interpolated in the 1D computational grid points and imported into the 2D

Water 2016, 8, 521 5 of 20

SOBEK is a coupled 1D-2D hydrodynamic model. The 1D St. Venant equations are solved for a series of cross-sections of the main channel and the overbank perpendicular to the main channel. The water level is spatially interpolated in the 1D computational grid points and imported into the 2D flood model based on a finite difference, staggered grid solution. The flow is modeled by solving the continuity equation and the momentum equations in the x- and y-directions as follows: (uh) ∂ (vh) ∂h + + = 0 (1) ∂x ∂y ∂t √ ∂u ∂u ∂u ∂h n2u u2 + v2 + u + v + g + = 0 (2) ∂t ∂x ∂y ∂x h4/3 √ ∂v ∂v ∂v ∂h n2v u2 + v2 + u + v + g + = 0 (3) ∂t ∂x ∂y ∂y h4/3 where y is the depth of the channel from the reference level, x is the longitudinal distance along the channel, t is time, h is the water head elevation from the reference level, u is the flow velocity in the x-direction, v is the flow velocity in the y-direction, and n is the Manning coefficient. Carrivick [49] used a SOBEK 2D hydrodynamic model to reconstruct the characteristics of high-magnitude outburst floods in the proglacial area of Kverkfjöll, Iceland, in anastomosing networks of simultaneously inundated channels. The country-wide SOBEK 1D hydrodynamic model (abbreviated LSM) was developed to model the surface water distribution as part of the Delta model, which is a set of models used to analyze the decisions related to the long-term fresh water supply and flood risk management in the Netherlands [51,52]. Linde et al. [55] used the SOBEK model to simulate low-probability flood peak events in the Rhine basin. Vanderkimpen et al. [50] compared the flood modeling using MIKE FLOOD with SOBEK. However the SOBEK model has been improved in recent years. Wei et al. [54] employed the SOBEK channel flow (CF) module and the rainfall-runoff (RR) module in river channel simulations to estimate river flow risks under climate change in the Tsengwen River basin, Tainan, Taiwan. Wu et al. [53] integrated the SOBEK model for flood simulations, a landslide model, and a coastal model to assess disaster impacts comprehensively according to extreme rainfall scenarios under climate change. The first step for constructing a FIM is the data collection. This study entirely collected hydrological and geomorphologic data as well as all necessary records and regulation reports in Tainan City. Upon the instantiation of the model, the next step involved hydrologic analyses, such as a frequency analysis, and establishing of rainfall patterns that can be used as inputs into the flood modeling work. Hydrologic analyses were conducted for river level, discharge and storm surge estimation. The flood model was calibrated by adjusting model parameters to match historical flood events. The model was then validated to determine the accuracy of flood simulation in relation to independent flood events. The last step in the development of FIMs was simulation of flood extents and flood depths using the calibrated and validated model.

3.2. Data Used For construction of the FIM for Tainan City, large amounts of different kinds of data were collected. They are shown below.

3.2.1. Digital Elevation Model (DEM) Gridded arrays of elevations with 40 × 40 m spatial resolutions in hillside ﬁelds and mountainous areas (elevation > 100 m) and with 5 × 5 m resolutions in low-lying areas (elevation < 100 m) were collected. The DEM used in this study is generated by Light Detection and Ranging (LiDAR) with 10–15 cm accuracy in height, provided by Ministry of Interior of Taiwan. The DEM-based topography of Tainan City is also shown in Figure1. Water 2016, 8, 521 6 of 20

3.2.2.Water River 2016 Cross-Sectional, 8, 521 Shape 6 of 20 Water 2016, 8, 521 6 of 20 3.2.2.Cross-sectional River Cross‐ shapeSectional varies Shape with position in the river and discharge. The elevation variations in the cross sections of six rivers in Tainan were collected. The interval of the cross-sectional shape data is 3.2.2.Cross River‐sectional Cross‐Sectional shape varies Shape with position in the river and discharge. The elevation variations in 300 m in non-urban areas and 20 m in urban areas. the crossCross sections‐sectional of six shape rivers varies in Tainan with position were collected. in the river The and interval discharge. of the Thecross elevation‐sectional variations shape data in isthe 300 cross m in sections non‐urban of six areas rivers and in 20Tainan m in wereurban collected. areas. The interval of the cross‐sectional shape data 3.2.3. Regional Drainage System (Open Channel) is 300 m in non‐urban areas and 20 m in urban areas. 3.2.3.Open-channel Regional Drainage rainwater System drains (Open are Channel) extremely common in the low-lying countryside of Tainan City,3.2.3. as shownOpen Regional‐channel in Drainage Figures rainwater3 System and drains4. (Open The are drainChannel) extremely is a common type of sewerin the low system.‐lying countryside The regional of Tainan drainage systemCity, generally Openas shown‐channel consists in Figures rainwater of 3 a and secondary drains 4. The are drain drainageextremely is a type systemcommon of sewer with in system. the a low network The‐lying regional countryside of small drainage drains of systemTainan attached. The smallgenerallyCity, as drains shown consists transport in ofFigures a secondary rainwater 3 and 4. drainage The to drain a primarysystem is a type with drainage of a sewernetwork system. system of small The composed drains regional attached. ofdrainage main The system drains small that servedrainsgenerally large transport areas. consists The rainwater of main a secondary drains to a primary are drainage connected drainage system to systemwith rivers. a network composed There are of smallof 662 main rainwater drains drains attached. that drains, serve The as largesmall shown in Figureareas.drains4. Generally, The transport main thedrains rainwater design are connected returnto a primary period to rivers. drainage for aThere drain system are is 662 25 composed years. rainwater The of drains, main shapes drains as and shown that elevations in serve Figure large of 4. drain Generally, the design return period for a drain is 25 years. The shapes and elevations of drain cross‐ cross-sectionsareas. The were main collecteddrains are forconnected the development to rivers. There of FIMs. are 662 rainwater drains, as shown in Figure 4. sectionsGenerally, were the collected design return for the period development for a drain of FIMs. is 25 years. The shapes and elevations of drain cross‐ sections were collected for the development of FIMs.

(a) (b)

Figure 3. General form( aof) regional drainage channel in Tainan City. (a) a( bsmall) regional drainage Figure 3. General form of regional drainage channel in Tainan City. (a) a small regional drainage channelFigure 3.with General width form 5 m; of(b )regional a large regional drainage drainage channel channel in Tainan with City. width (a) 20 a smallm. regional drainage channel with width 5 m; (b) a large regional drainage channel with width 20 m. channel with width 5 m; (b) a large regional drainage channel with width 20 m.

Figure 4. Network of river, drainage and hydraulic structures in Tainan City. Figure 4. Network of river, drainage and hydraulic structures in Tainan City. Figure 4. Network of river, drainage and hydraulic structures in Tainan City.

Water 2016, 8, 521 7 of 20 Water 2016, 8, 521 7 of 20

3.2.4.3.2.4. Rainwater Sewer (Underground) TheThe rainwaterrainwater sewer (i.e., drainage) system is comprised of a large underground pipe network. ItIt collects and transportstransports rainwaterrainwater runoffrunoff toto rivers.rivers. Manholes are installedinstalled wherever therethere isis aa change ofof gradientgradient oror alignment. The completion rate of the rainwater sewer system in Taiwan isis higher than 60%,60%, andand thethe network isis mainly located inin urban areas (Figure(Figure5 5).). The The design design return return period period for for a a sewer sewer systemsystem inin TaiwanTaiwan isis approximatelyapproximately 2–52–5 yearsyears accordingaccording toto thethe developmentdevelopment conditions of the area, andand thethe timetime ofof concentrationconcentration of thethe systemsystem rangesranges from 3030 minmin toto 22 hh [[56].56]. The locations,locations, lengths, shapesshapes andand elevationselevations ofof sewerssewers werewere collected.collected. There are 43354335 undergroundunderground sewersewer pipespipes andand 83458345 manholesmanholes inin TainanTainan City.City.

Figure 5. Sewer system in North District of Tainan City. Figure 5. Sewer system in North District of Tainan City.

3.2.5.3.2.5. Satellite and Aerial ImagesImages SatelliteSatellite imagesimages from from Formosa-II Formosa taken‐II taken by optical by optical sensor (resolutionsensor (resolution 2.0 m) and 2.0 aerial m) orthoimages and aerial takenorthoimages by LIDAR taken (resolution by LIDAR (resolution 0.25 m) with 0.25 a m) scale with of a scale 1:5000 of were1:5000 used were to used divide to divide or adjust or adjust the sub-watersheds,the sub‐watersheds, which which were were deﬁned defined according according to rainwater to rainwater drainage drainage or sewer or sewer systems. systems. This study This collectedstudy collected more than more 400 than images 400 images in Tainan in Tainan City. City.

3.2.6.3.2.6. Land Use LandLand useuse conditionsconditions shownshown inin Figure6 6 were were collected collected to to identify identify the the land land roughness roughness and and derive derive thethe ﬂowflow behavior.behavior. In Tainan City,City, moremore thanthan 45%45% ofof thethe landland isis usedused forfor agriculture.agriculture. OnlyOnly 35%35% ofof TainanTainan City City comprises comprises villages villages or or urban urban areas. areas.

Water 2016, 8, 521 8 of 20 Water 2016, 8, 521 8 of 20

Figure 6. Land use in Tainan City. Figure 6. Land use in Tainan City. 3.2.7. Hydraulic Structures and Instruments 3.2.7. Hydraulic Structures and Instruments Hydraulic structures such as sluice gates, dykes, reservoirs and detention basins, and hydraulic Hydraulicinstruments structures such as pumping such as stations sluice are gates, also dykes, included reservoirs in the flood and model. detention Bridges basins, that cross and rivers hydraulic instrumentswere also such included. as pumping For the purposes stations areof the also present included study, in the the data flood concerning model. Bridges hydraulic that structures cross rivers wereand also instruments included. Forin Tainan the purposes City were of also the presentcollected. study, The data the items data concerninginclude elevations hydraulic of 81 structures river and instrumentsdykes, 1206 sluice in Tainan gates, City 955 werebridges, also 38 collected. ponds and The detention data items basins, include 13 reservoirs elevations and 380 of 81 pumping river dykes, 1206 sluicestations gates, (fix and 955 mobile), bridges, including 38 ponds fixed and detentionand mobile basins, pump 13machines. reservoirs The and amount 380 pumping of hydraulic stations (fix andstructures mobile), and including instruments fixed in six and rivers mobile is shown pump in machines. Table 3. The amount of hydraulic structures and instruments in six rivers is shown in Table3. Table 3. Amount of hydraulic structures in six rivers of Tainan City.

Table 3. AmountCross of hydraulicDetention structuresSluice in six riversPumping of Tainan City. River Name Manhole Bridge Section Basin Gate Station Bajhang River Cross2084 Detention7 Sluice25 Pumping17 351 120 River Name Manhole Bridge Jishui River Section1866 Basin5 Gate59 Station7 472 71 Jiangjun River 2549 0 33 6 282 30 BajhangTsengwen River River 20841991 76 2516 178 351320 36 120 JishuiYanshui River River 18661842 512 5926 77 2333 472 36 71 JiangjunErren River River 25491662 03 3360 63 282676 8 30 Tsengwentotal River 199111,994 633 16219 848 4434 320 301 36 Yanshui River 1842 12 26 7 2333 36 Erren River 1662 3 60 3 676 8 3.2.8. Rainfall Records Total 11,994 33 219 48 4434 301 The locations, elevations, record lengths and data contents from 99 rainfall stations were collected in Tainan City and the nearby cities of Chiayi and Kaohsiung. The present study used 37 3.2.8. Rainfall Records stations that have recorded hourly rainfall for more than 20 years. The locations, elevations, record lengths and data contents from 99 rainfall stations were collected in Tainan3.2.9. CityRiver and Water the Level nearby and cities Discharge of Chiayi and Kaohsiung. The present study used 37 stations that have recordedHourly hourly data from rainfall 45 river for level more stations than 20 and years. 12 river discharge stations were collected. The river level and discharge data are mainly used for calibration of the 1D hydrodynamic model. 3.2.9. River Water Level and Discharge Hourly data from 45 river level stations and 12 river discharge stations were collected. The river level and discharge data are mainly used for calibration of the 1D hydrodynamic model. Water 2016, 8, 521 9 of 20

3.2.10. Seaside Data Hourly sea level data from 3 tidal stations along the Tainan coast were collected. Tidal data are used for a storm surge analysis and as the downstream boundary of the river in the 1D hydrodynamic simulation. In addition, data were collected from a marine data buoy located at a depth of 20 m and 1 km off the Tainan coast. Wave data (height, period, and direction) and 200 m resolution bathymetry data are used for the run-up analysis together, especially during the typhoon period.

3.2.11. Historical Flood Events Data from historical flood events, including flood depths and flood extents, over the past 10 years were collected for model calibration and validation purposes. These data mostly come from the reports from local villages and local stakeholders.

3.3. Simulation and Veriﬁcation

3.3.1. Rainfall Analysis River runoff is the driver of flood simulations, and it is assumed to be mainly generated by precipitation. The occurrence and quantity of runoff are dependent on the characteristics of the rainfall event, i.e., the intensity, duration and distribution. The objective of the rainfall analysis was to obtain the amount of rainfall for the corresponding duration, the return periods and the hyetographs for each station. Annual maximum rainfall depths for durations of 6, 12, 24 and 48 h were obtained using historic hourly data from 37 rainfall stations in Tainan with records of longer than 20 years. Six statistical distributions, including Normal, Log-Normal, Extreme Value Type I, Person Type III and Log-Person Type III, were used to fit the rainfall data. Two goodness-of-fit tests (Chi-square and Kolmogorov–Smirnov tests) are used to evaluate the best fitting model. The rainfall depths for recurrence intervals of 2, 5, 10, 25, 50, 100, 200 and 500 years are derived for the corresponding durations. Table4 shows the results for the Tainan rainfall station (Station No. 467410). The results of the frequency analysis were used to develop the relationship between rainfall intensity (or depth), duration, and frequency (or return period) at all sites and to create IDF (Intensity–Duration–Frequency) curves which used to estimate rainfall intensity according to assumed duration and return period. Horner’s equation is used to fit the IDF curves in this study.

Table 4. Results of frequency analysis on rainfall and storm surge and calibration of the coefﬁcients of Horner formula at Tainan Station (Station No. 467410).

Return Period (Years) Items Duration (h) 2 5 10 25 50 100 200 500 6 131 179 207 240 262 283 302 326 12 177 237 273 314 343 370 396 428 Rainfall (mm) 24 203 289 347 419 473 526 579 650 48 229 329 395 479 541 602 664 745 Storm surge (m) 1.45 1.50 1.57 1.67 1.74 1.80 1.87 1.96 a 1143 1131 1062 959 888 827 776 721 Horner formula b 39.6 31.1 21.9 9.5 0.7 −7.1 −14.0 −21.8 coefﬁcient c 0.66 0.61 0.58 0.54 0.51 0.49 0.47 0.44

A design hyetograph is then required to obtain the temporal distribution of runoff according to total rainfall depth. Kimura et al. (2014) [57] incorporated a modified ranking method to create a design hyetograph. In this study, the alternating block model [58] was used to derive duration-specific and return period-specific hyetographs from IDF curves. Figure7 shows the designed storm hyetograph at Tainan Station for 2-, 10-, 50- and 100-year return periods and a 12-h duration. Water 2016, 8, 521 10 of 20 Water 2016, 8, 521 10 of 20

40 30 T=2 years T=10 years

(%) 20 (%) 20 10 百分比百分比百分比 Percentage (%) Percentage Percentage (%) 0 0 0 3 6 9 12 0 3 6 9 12 Time(hr)Time (hr) Time(hr)Time (hr) 30 30 T=50 years T=100 years 20 20 (%) (%)

10 10 百分比百分比 Percentage (%)

Percentage (%) Percentage 0 0 0 3 6 9 12 0 3 6 9 12 Time(hr)Time (hr) Time(hr)Time (hr)

Figure 7. TheThe design design hyetograph for 2 2-,‐, 10 10-,‐, 50 ‐ -andand 100 100-year‐year return periods and 12-h12‐h duration.

3.3.2. Storm Surge Analysis A storm surge is usually characterized as an abnormal rise in in water level level generated generated by by a a storm, storm, over and above the predicted astronomical tide. The total water level is the superimposition of the tide, stormstorm surgesurge and and wave wave run-up. run‐up. The The water water level level rises rises when when a typhoon a typhoon approaches, approaches, especially especially since sincethe waves the waves are large are large enough enough to generate to generate significant significant run-up. run When‐up. When the total the watertotal water level level is higher is higher than thanthe sea the dyke sea dyke elevation, elevation, overtopping overtopping occurs. occurs. In this In study,this study, the High the High Water Water of Ordinary of Ordinary Spring Spring Tide (HWOST)Tide (HWOST) is used is as used the base as the tide base height. tide The height. maximum The stormmaximum surges storm derived surges from 105derived typhoons from from 105 1980typhoons to 2013 from are 1980 obtained to 2013 by numericalare obtained simulation by numerical using thesimulation MIKE 21 using program. the MIKE The same 21 program. distributions The sameand goodness-of-fit distributions and used goodness in the frequency‐of‐fit used analysis in the frequency were employed analysis to were test employed the best fitting to test model the best of fittingstorm surge.model Theof storm heights surge. of storm The heights surge onof storm the Tainan surge coast on the were Tainan estimated coast were for return estimated periods for return of 2, 5, 10,periods 25, 50, of 100, 2, 5, 200 10, and25, 50, 500 100, years, 200 as and shown 500 years, in Table as 4shown. in Table 4.

3.3.3. Model Setup In the present work, modules of R-RR‐R (Rainfall–Runoff), 1DFLOW (Rural and Urban) and 2D Overland Flow in SOBEK Advanced Version 2.13.002 were used. The work used R R-R‐R in areas with elevations higherhigher than than 100 100 m, m, which which are are defined defined as highlandas highland areas. areas. Runoff Runoff routing routing to downstream to downstream areas (elevationareas (elevation < 100 < m) 100 was m) performed was performed using using 1DFLOW. 1DFLOW. When When the runoff the runoff rates exceed rates exceed the capacity the capacity of the ofriver the channel, river channel, the 2D Overlandthe 2D Overland Flow module Flow takes module over takes computations over computations and simulates and the simulates propagation the propagationalong the floodplain. along the floodplain. The DEMDEM withwith resolution resolution of of 40 40× ×40 40 m m was was used used for for uniform uniform land land use areas,use areas, and aand 10 × a 1010 m× DEM10 m wasDEM used was for used multi-use for multi areas‐use in areas highland in highland parts. Additionally, parts. Additionally, a 5 × 5 ma DEM5 × 5 wasm DEM used was to construct used to constructthe computational the computational grids in urban grids areas. in urban areas. The parameters of the SCS Runoff Curve Number Method were set in the R R-R‐R modules for each watershed. The cross cross-sectional‐sectional shapes of rivers, regional drainage systems and properties of sewers and hydrologic structures described in Section 3.3 were usedused asas inputsinputs intointo thethe hydraulichydraulic model.model. Simulations were carried out separately in each watershed due to computational limitations. limitations. There are numerous regional drainage channels in in Tainan City, in addition to underground sewer systems. Geomorphologically, Geomorphologically, these these different different elements elements can can also also be be considered considered as asriver river systems. systems. A Adrainage drainage basin basin receives receives runoff runoff for for a aspecific specific topographic topographic region. region. The The basins basins vary vary in in size size from from a a few km2 toto several hundreds of km 2.. ArcGIS ArcGIS software software was was used used to to divide the the drainage basins based on the DEM characteristics. The drainage basins were corrected according to satellite and aerial images when the auto‐division results were biased. Figure 8 shows the watershed partitions (or delineations) in Tainan City.

Water 2016, 8, 521 11 of 20 when the auto-division results were biased. Figure8 shows the watershed partitions (or delineations) in Tainan City. Water 2016, 8, 521 11 of 20

Figure 8. Partition (or delineation) of sub‐watersheds. Figure 8. Partition (or delineation) of sub-watersheds. 3.3.4. Calibration and Validation 3.3.4. Calibration and Validation Records from historical flood events were used to calibrate and validate the model. The two Records from historical flood events were used to calibrate and validate the model. The two indicators used to assess the model performance based on the agreement between the model‐ indicators used to assess the model performance based on the agreement between the model-simulated simulated flood extent and observations are the Probability of Detection (POD) and Scale of Accuracy flood extent and observations are the Probability of Detection (POD) and Scale of Accuracy (SA) indicators: (SA) indicators: POD % Ac (4) POD (%) = (4) A f A SA( )%= c (5) SA % (5) A f + A0 − Ac where A0 is the simulation flood extent, Af is the historical flood extent, and Ac is the overlap between where A0 is the simulation flood extent, Af is the historical flood extent, and Ac is the overlap between simulated and recorded flood extents. The optimal simulation performance is achieved when both simulated and recorded flood extents. The optimal simulation performance is achieved when both POD and SA approach 100%. The denominator of SA presents the incorrect simulation is considered POD and SA approach 100%. The denominator of SA presents the incorrect simulation is considered in the indicator that can be estimated by 100%‐SA. Flood simulation is asked by Water Resources in the indicator that can be estimated by 100%-SA. Flood simulation is asked by Water Resources Agency of Taiwan to have POD and SA values higher than 60%. To correct the flooded depth, the Agency of Taiwan to have POD and SA values higher than 60%. To correct the flooded depth, the bias bias associated with simulated and observed peak flood depths should be lower than 0.2 m. associatedFour with historical simulated flood and events observed caused peak by floodtyphoons depths were should used befor lower model than calibration, 0.2 m. and two additionalFour historical events (one flood typhoon events causedand one by rainstorm) typhoons were were used used for for model model validation. calibration, Manning’s and two additionalroughness events coefficient (one was typhoon the calibration and one parameter rainstorm) in the were model used calibration for model work. validation. This was Manning’sdone via roughnessthe trial‐and coefficient‐error procedure. was the calibration The values parameter of Manning in coefficients the model calibrationare found to work. be 0.025 This for was sewer done viasystems the trial-and-error and between procedure. 0.027 and 0.032 The valuesfor drainage of Manning systems. coefficients The model arecalibration found to results be 0.025 shows for that sewer systemsthe POD and is between higher than 0.027 70% and (maximum 0.032 for drainage88.1% for systems.Typhoon The Fanapi model flood calibration simulation) results and the shows SA is that thehigher POD than is higher 60%, thansee Table 70% 5. (maximum 88.1% for Typhoon Fanapi flood simulation) and the SA is higher than 60%, see Table5.

Water 2016, 8, 521 12 of 20 Water 2016, 8, 521 12 of 20

TableTable 5. 5. ResultsResults from from the the model model calibration calibration and and verification veriﬁcation work. work.

ModeledModeled ObservedObserved OverlapOverlap Area Area Task EventEvent WatershedWatershed FloodFlood Area Area FloodFlood Area Area 2 SASA (%) (%) POD POD (%)(%) 2 2 (km(km) 2) (km(km) 2) (km(km) 2) TyphoonTyphoon Morakot Morakot JiangjunJiangjun watershed watershed 37.137.1 40.940.9 29.329.3 60.160.1 71.6 (2008)(2008) SanyehSanyeh drainage drainage system system 10.510.5 13.213.2 9.19.1 62.462.4 69.0 TyphoonTyphoon Fanapi Fanapi CalibrationCalibration JiangjunJiangjun watershed watershed 7.77.7 6.26.2 5.45.4 64.364.3 88.1 (2010)(2010) Typhoon Kongrey Typhoon Kongrey Sanyeh drainage system 8.9 9.3 7.3 66.8 78.5 (2013) Sanyeh drainage system 8.9 9.3 7.3 66.8 78.5 (2013) Typhoon Kongrey Typhoon Kongrey Jiangjun watershed 9.9 10.4 7.8 62.3 74.9 Veriﬁcation (2013) Jiangjun watershed 9.9 10.4 7.8 62.3 74.9 Verification (2013) 0812 Rainstorm Sanyeh drainage system 6.5 6.2 5.1 68.0 82.9 0812 Rainstorm Sanyeh drainage system 6.5 6.2 5.1 68.0 82.9

TheThe flood flood events events induced by TyphoonTyphoon Kongrey Kongrey (August (August 2013) 2013) and and the the rainstorm rainstorm that that occurred occurred on on12 August12 August 2014, 2014, referred referred to as to rainstorm as rainstorm 0812 henceforth,0812 henceforth, were used were for used validation. for validation. Typhoon Typhoon Kongrey Kongreycaused serious caused floodingserious flooding in the Jiangjun in the Jiangjun watershed, watershed, and rainstorm and rainstorm 0812 triggered 0812 triggered flooding flooding in the inSanyeh the Sanyeh drainage drainage system. system. The rainfall The rainfall totals during totals Typhoonduring Typhoon Kongrey Kongrey at three stationsat three fromstations upstream from upstreamto downstream to downstream on the Jiangjun on the RiverJiangjun are River shown are in shown Figure 9in with Figure corresponding 9 with corresponding tidal elevation. tidal elevation.The maximum The maximum three-day three rainfall‐day depth rainfall is 636 depth mm, is and 636 themm, maximum and the hourlymaximum rainfall hourly intensity rainfall is intensity100 mm/h. is 100 The mm/h. simulated The simulated flood extent flood and extent depth and during depth Typhoon during Typhoon Kongrey areKongrey shown are in shown Figure in10 . FigureFigure 10.11 showsFigure another11 shows validation another validation event in the event Sanyeh in the drainage Sanyeh system. drainage The system. flood The conditions flood conditionsdetermined determined by field investigations by field investigations are shown in are the shown figures. in The the maximumfigures. The flood maximum depths are flood 2.41 depths m and are4.4 2.41 m in m the and events, 4.4 m respectively. in the events, Table respectively.5 shows the Table performance 5 shows of the validation. performance The results of validation. suggest The that resultsthe POD suggest is 75% that and the SA POD is 62% is 75% in the and Jiangjun SA is 62% watershed in the Jiangjun flood simulation watershed during flood simulation Typhoon Kongrey. during TyphoonAdditionally, Kongrey. the POD Additionally, and SA were the 83% POD and and 68%, SA respectively, were 83% and for the 68%, flood respectively, simulation for in the the Sanyeh flood simulationdrainage system in the during Sanyeh rainstorm drainage 0812. system This during quantitative rainstorm assessment 0812. This is satisfactory, quantitative however, assessment we will is satisfactory,try to improve however, further we it in will the try subsequent to improve work. further it in the subsequent work.

FigureFigure 9. 9. RainfallRainfall histogram histogram at at three three stations stations (Xinying, (Xinying, Wangye Wangye Temple, Temple, and and Xiaying) Xiaying) in in Jiangjun Jiangjun watershedwatershed and and the the tidal tidal curve curve downstream downstream during during Typhoon Typhoon Kongrey in August 2013.

Water 2016, 8, 521 13 of 20 Water 2016, 8, 521 13 of 20 Water 2016, 8, 521 13 of 20

Figure 10. Validation of flood simulation for Jiangjun watershed after Typhoon Kongrey in August 2013. Figure 10. Validation of ﬂood simulation for Jiangjun watershed after Typhoon Kongrey in August 2013. Figure 10. Validation of flood simulation for Jiangjun watershed after Typhoon Kongrey in August 2013.

Figure 11. Validation of flood simulation for Sanyeh drainage system after rainstorm occurred on 12 Figure 11. Validation of flood simulation for Sanyeh drainage system after rainstorm occurred on 12 FigureAugust 11. Validation 2014. of flood simulation for Sanyeh drainage system after rainstorm occurred on August 2014. 12 August 2014. There are at least three potential causes of bias in flood simulations. Firstly, the historical flood surveysThere do are not at fully least reflect three thepotential actual causes conditions. of bias The in floodweather simulations. during flood Firstly, events the washistorical not always flood Theresurveyspermitting are do at thenot least use fully of three satellitereflect potential the and actual aerial causes conditions. imaging. of bias Additionally, The in weather flood simulations. surveys during implementedflood Firstly, events was theafter historicalnot peak always flood flood surveyspermittingtimes do may not thenot fully usehave reflect of recorded satellite the and actualthe aerialcorrect conditions. imaging. extents Additionally,and The depths, weather and surveys during dictations implemented flood from events local after residents was peak not flood are always permittingtimessubjective, may the usenotdiverse have of satellite and recorded have and highthe aerial correct degrees imaging. extents of uncertainty. Additionally,and depths, Secondly, and surveys dictations the implementedsewer from system local residents data after have peak are a flood subjective, diverse and have high degrees of uncertainty. Secondly, the sewer system data have a times may not have recorded the correct extents and depths, and dictations from local residents are subjective, diverse and have high degrees of uncertainty. Secondly, the sewer system data have a degree of uncertainty. Furthermore, rapid changes in the developments and land use represent another Water 2016, 8, 521 14 of 20 cause of uncertainty. Thirdly, there is a degree of uncertainty due to simplified model physics. Due to the abovementioned uncertainties, flood model simulations for which POD or SA values were higher than 60% were considered acceptable.

3.4. Flood Inundation Maps FIMs are designed to represent the possible ﬂood conditions under various rainfall scenarios. Two types of rainfall scenarios were assumed: one with topography-based cumulative rainfall (CR) and the other with periodic rainfall (PR). Topography-based cumulative rainfall assigns rainfall amounts for various rainfall durations, but the amounts depend on topography. The rainfall amount in highland areas is 2.1 times higher than in the low-lying areas. The type of periodic rainfall is based on rainfalls of several return periods for various durations. Thirty-four scenarios consisting of quantitative and periodic rainfalls were assumed. To consider the effect of the sea, another eight scenarios with simultaneous rainfall and storm surge events were assumed for various return periods and 24-h durations. In total, 42 scenarios were used to produce the FIMs in the present work. These scenarios are given in Table6.

Table 6. Forty-two scenarios are assumed for producing FIMs.

Duration (h) Type Scenario Code 6 12 24 Cumulative Rainfall 150, 250, 350 200, 300, 400 200, 350, 500, 650 CR-D-D (ex. QR-6-150; QR-24-350) (mm) Periodic rainfall 2, 5, 10, 25, 50, 100, 200, 500 PR-D-T (ex. PR-6-100) (return year) Storm surge 2, 5, 10, 25, 50, 100, 200, 500 PR-D-TS (ex. PR-6-100S) (return year)

FIMs are designed to be exported as high-resolution, low-compression digital images such as TIFF or JPG formats. An FIM is printed on an A0 size (841 × 1189 mm) poster. Figure 12 shows the FIMs for a cumulative rainfall of 400 mm and durations of 6, 12, and 24 h. The maximum flood depth for a rainfall of 400 mm in 6 h (scenario code CR-6-400) is higher than 3 m in Annan District (between the Tsengwen and Yensui Rivers). The total flooded area of scenario CR-6-400 is 1.73 times that of CR-24-400. However, the flood induced by scenario CR-6-400 is only 1.26 times that of CR-6-100. Therefore, the rainfall intensity is a more influential factor than the total rainfall amount. Furthermore, the tide level was found to have an important influence on flood simulations, especially in the low-lying areas (elevation < 2 m). When the sea surface reaches the high tide phase, flooding in the low-lying areas of Tainan City is significant, especially in the old Tainan City areas, regardless of the rainfall intensity or depth. When the rainfall intensity increases (for example, in CR-6-200, CR-12-450, and CR-24-650), the flooding increases from elevation 2 m to 5 m in some areas. Figure 13 shows the FIMs with 100-year rainfall and durations of 6, 12, and 24 h. These rainfall scenarios with different return periods are similar to those of cumulative rainfall, but they are presented differently to satisfy different requirements. Flooding began to occur in the coastal area (elevation < 1 m) for low-intensity rainfall events such as PR-12-2 or PR-24-5. The flooding is due to low-lying land in the coastal area. When the rainfall intensity increases, such as in scenarios PR-6-2 and PR-12-10, the flood extent expands due to the influence of low-lying areas. However, when the rainfall intensity increases, such as in scenarios PR-6-10, PR-12-50, and PR-24-100, the flood extent reaches elevation > 10 m in some areas because the river discharge capability decreases due to tidal chokage. Figure 14 shows the results of various scenarios. The flood extents of a 100-year rainfall event for durations of 6, 12 and 24 h (scenario codes PR-6-100, PR-12-100 and PR-24-100) are overlapped, as is the influence of the storm surge (PR-24-100S). The most extreme scenario (100-year rainfall for 24 h) triggered severe floods, especially in the southern Tainan region (Figure 13c); however, the storm surge considerably affected low-lying areas in the western part of Tainan City. Water 2016, 8, 521 15 of 20 Water 2016, 8, 521 15 of 20

FigureFigure 12. 12. ProductionsProductions of offlood ﬂood inundation inundation map map for for Tainan Tainan City City under under conditions: conditions: (a ()a cumulative) cumulative rainfall rainfall 400 400 mm mm for for 6 6h h(CR (CR-6-400);‐6‐400); (b ()b cumulative) cumulative rainfall rainfall 400 400 mm mm for 12for h (CR 12 h‐12 (CR-12-400);‐400); and (c and) cumulative (c) cumulative rainfall rainfall 400 mm 400 for mm 24 h for (CR 24‐ h24 (CR-24-400).‐400).

Water 2016, 8, 521 16 of 20 Water 2016, 8, 521 16 of 20

FigureFigure 13. 13. ProductionsProductions of offlood ﬂood inundation inundation map map for for Tainan Tainan City City under under conditions: conditions: (a) ( a100) 100-year‐year rainfall rainfall for for 6 h 6 h(PR (PR-6-100);‐6‐100); (b) ( b100) 100-year‐year rainfall rainfall for for 12 12h (PR h (PR-12-100);‐12‐100); and (c)and 100‐ (yearc) 100-year rainfall rainfallfor 24 h for(PR 24‐24 h‐100). (PR-24-100).

Water 2016, 8, 521 17 of 20 Water 2016, 8, 521 17 of 20

Figure 14. Overlap of the flooded area under various scenarios. Figure 14. Overlap of the flooded area under various scenarios. 3.5. Non‐Technical Improvement 3.5. Non-Technical Improvement In additional to technical development, non‐technical improvement on the FIM is possible, i.e., In additional to technical development, non-technical improvement on the FIM is possible, the stakeholder engagement. Two stakeholder meetings were held in the developing process. One i.e.,was the held stakeholder during the engagement. model simulation Two stage, stakeholder and the meetingsother was wereheld when held the in theFIMs developing were produced. process. OneThe was participants held during of both the modelmeetings simulation included FIM stage, producers and the (the other Water was Resources held when Agency the FIMs of the were produced.central Thegovernment), participants groups of both that meetings provided included FIM development FIM producers and technical (the Water support Resources (university Agency of theprofessors), central government), first‐line managers groups (water that provided resource FIMbureaus development in local city and governments), technical support and public (university users professors),(local residents). first-line The managers objective (water of the resource first meeting bureaus was in to local decrease city governments),the uncertainty andin records public of users (localhistorical residents). flood Theevents objective based on of inputs the first from meeting local residents was to decreaseand their the memories/perceptions uncertainty in records and of historicalpersonal flood records. events The basedpurpose on of inputs the second from meeting local residents was to educate and theirthe first memories/perceptions‐line managers and local and personalresidents records. regarding The the purpose conditions, of the results second and meeting uncertainty was toof educate FIMs. Additionally, the first-line that managers meeting and localprovided residents information regarding on the web conditions, search systems, results and and users uncertainty provided of feedback FIMs. Additionally,to improve the that FIMs. meeting provided information on web search systems, and users provided feedback to improve the FIMs. 4. Conclusions 4. ConclusionsThe present paper describes the development of third‐generation FIMs at the city scale. Unlike numerous studies that have focused on an urban area or a village scale, this study simulates a large The present paper describes the development of third-generation FIMs at the city scale. city with a population of 1.8 million encompassing an area of 2200 km2 and consisting of six large Unlike numerous studies that have focused on an urban area or a village scale, this study simulates a watersheds and 3162 sub‐watersheds. Substantial amounts of data were collected, with particular large city with a population of 1.8 million encompassing an area of 2200 km2 and consisting of six large efforts devoted on measuring artificial open channel regional drainage systems. The 1D/2D coupled watershedsmodel SOBEK and 3162 was sub-watersheds.used for flood simulations. Substantial The amounts indicator of POD data (Probability were collected, of Detection) with particular was effortsfound devoted to be up on measuringto 88.1% for artificial flood simulation open channel in Typhoon regional Fanapi. drainage Another systems. indicator The 1D/2DSA (Scale coupled of modelAccuracy) SOBEK was was found used to for be flood up to simulations. 68% in 0812 rainstorm The indicator flood POD simulation. (Probability Values of above Detection) this threshold was found to bewere up totreated 88.1% as for satisfactory, flood simulation which was in Typhoon in accordance Fanapi. with Another the standard indicator defined SA (Scale by the of Accuracy)Water wasResources found to Agency be up toof 68%Taiwan. in 0812The inundation rainstorm map flood database simulation. is able Values to generate above approximate this threshold results were treatedquickly as satisfactory,based on a forecasted which was cumulative in accordance rainfall with that the can standard be obtained defined few days by thein advance Water Resources of the Agencyhazardous of Taiwan. event. The This inundation capability enables map database a quick is(second able to scale) generate view approximate of potential flooding results quicklyin support based on aof forecasted initial decision cumulative making rainfall activities. that canThe besame obtained maps fewcan daysalso be in advanceused for ofland the hazardoususe planning, event. Thisdevelopment capability enables analysis, a insurance quick (second applications, scale) view and retreatment of potential strategy flooding preparation. in support A of flood initial warning decision makingsystem activities. that combines The same real‐time maps observations can also be and used numerical for land models use planning, is undoubtedly development necessary analysis, in order to produce accurate forecasts. However, the associated computations may require considerable insurance applications, and retreatment strategy preparation. A flood warning system that combines time (hours) and efforts, which makes the use of numerical models very challenging. Hence, the use real-time observations and numerical models is undoubtedly necessary in order to produce accurate of FIMs plays an important role in flood risk management applications. forecasts. However, the associated computations may require considerable time (hours) and efforts, which makes the use of numerical models very challenging. Hence, the use of FIMs plays an important role in flood risk management applications. Water 2016, 8, 521 18 of 20

In total, forty-two scenarios were used in the development of FIMs in the Tainan City. In the present work it was found that the rainfall intensity affects flood conditions more than the accumulative rainfall depth. The tidal phase is another important factor associated with the flooding scale, in the low-lying areas in Tainan City. Areas with elevation less than 1 m are frequently flooded during the high tide phase no matter how high is the scale of the rainfall intensity and depth. Storm surges are the third key factor that influences flooding in the coastal area of Tainan City. The study revealed that the extreme scenario (100-year rainfall for 24 h) would cause severe flooding in the western Tainan. It was found that one-third of the entire Tainan City would be flooded under this extreme scenario. The third-generation FIMs presented in this paper are based on various scenarios of high-intensity rainfall data. The third-generation of FIMs represents an improvement from the previous generation of FIMs and as such it will supersede the previous maps.

Acknowledgments: The study was jointly supported by the Ministry of Science and Technology of Taiwan (MOST 104-2923-I-006-001-MY3) and the European Union Seventh Framework Programme (FP7/2007-2013) PEARL (Preparing for Extreme And Rare events in coastal regions) (Grant No. 603663). Author Contributions: Dong-Jiing Doong, Weicheng Lo, Zoran Vojinovic, Wei-Lin Lee and Shin-Ping Lee contributed equally in the work and in production of the present manuscript. Conﬂicts of Interest: The authors declare no conﬂict of interest.

References

1. Price, R.K.; Vojinovic, Z. Urban flood disaster management. Urban Water J. 2008, 5, 259–276. [CrossRef] 2. Mynett, A.E.; Vojinovic, Z. Hydroinformatics in multi-colours—Part red: Urban flood and disaster management. J. Hydroinform. 2009, 11, 166–180. [CrossRef] 3. Vojinovic, Z.; van Teeffelen, J. An Integrated Stormwater Management Approach for Small Islands in Tropical Climates. Urban Water J. 2007, 4, 211–231. [CrossRef] 4. Aerts, J.; Botzen, W.; Bowman, M.; Dircke, P.; Ward, P. Climate Adaptation and Flood Risk in Coastal Cities; Routledge: London, UK, 2013. 5. Vojinovic, Z.; Sanchez, A. Optimising sewer system rehabilitation strategies between flooding, overflow emissions and investment costs. In Proceedings of the 11th International Conference on Urban Drainage, Edinburgh, UK, 31 August–5 September 2008. 6. Barreto, W.; Vojinovic, Z.; Price, R.; Solomatine, D. Multiobjective Evolutionary Approach to Rehabilitation of Urban Drainage Systems. J. Water Res. Plan. Manag. 2010, 136, 547–554. [CrossRef] 7. Vojinovic, Z.; Solomatine, D.; Price, R.K. Dynamic least-cost optimization of wastewater system remedial works requirements. Water Sci. Technol. 2006, 54, 467–475. [CrossRef][PubMed] 8. Vojinovic, Z.; Sahlu, S.; Torres, A.S.; Seyoum, S.D.; Anvarifar, F.; Matungulu, H.; Barreto, W.; Savic, D.; Kapelan, Z. Multi-objective rehabilitation of urban drainage systems under uncertainties. J. Hydroinform. 2014, 16, 1044–1061. [CrossRef] 9. Vojinovic, Z.; Hammond, M.; Golub, D.; Hirunsalee, S.; Weesakul, S.; Meesuk, V.; Medina, N.; Sanchez, A.; Kumara, S.; Abbott, M. Holistic approach to flood risk assessment in areas with cultural heritage: A practical application in Ayutthaya, Thailand. Nat. Hazards 2016, 81, 589–616. [CrossRef] 10. Sathish Kumar, D.; Arya, D.S.; Vojinovic, Z. Modeling of urban growth dynamics and its impact on surface runoff characteristics. Comput. Environ. Urban Syst. 2013, 41, 124–135. [CrossRef] 11. Sanchez, A.; Medina, N.; Vojinovic, Z.; Price, R. An integrated cellular automata evolutionary-based approach for evaluating future scenarios and the expansion of urban drainage networks. J. Hydroinform. 2014, 16, 319–340. [CrossRef] 12. Krzhizhanovskaya, V.V.; Shirshov, G.S.; Melnikova, N.B.; Belleman, R.G.; Rusadi, F.I.; Broekhuijsen, B.J.; Gouldby, B.P.; Lhomme, J.; Balis, B.; Bubak, M.; et al. Flood early warning system: Design, implementation and computational modules. Procedia Comput. Sci. 2011, 4, 106–115. [CrossRef] 13. Krzysztofowicz, R. Recent advances associated with flood forecast and warning systems. Rev. Geophys. 1995, 33, 1139–1147. [CrossRef] 14. De Roo, A.P.J.; Gouweleeuw, B.; Thielen, J.; Bartholmes, J.; Bongioannini-Cerlini, P.; Todini, E.; Bates, P.D.; Horritt, M.; Hunter, N.; Beven, K.; et al. Development of a European flood forecasting system. Int. J. River Basin Manag. 2003, 1, 49–59. [CrossRef] Water 2016, 8, 521 19 of 20

15. Doong, D.J.; Chuang, L.Z.-H.; Wu, L.C.; Fan, Y.M.; Kao, C.C.; Wang, J.H. Development of an operational coastal flooding early warning system. Nat. Hazards Earth Syst. Sci. 2012, 12, 379–390. [CrossRef] 16. Jubach, R.; Tokar, A.S. International severe weather and flash flood hazard early warning systems—Leveraging coordination, cooperation, and partnerships through a hydrometeorological project in Southern Africa. Water 2016, 8, 258. [CrossRef] 17. Keoduangsine, S.; Robert, G. An appropriate flood warning system in the context of developing countries. Int. J. Innov. Manag. Technol. 2012, 3, 213–216. 18. Horritt, M.S.; Bates, P.D. Evaluation of 1D and 2D numerical models for predicting river flood inundation. J. Hydrol. 2002, 268, 87–99. [CrossRef] 19. Hunter, N.M.; Bates, P.D.; Horritt, M.S.; Wilson, M.D. Simple spatially-distributed models for predicting flood inundation: A review. Geomorphology 2007, 90, 208–225. [CrossRef] 20. Merz, B.; Thieken, A.H.; Gocht, M. Flood risk mapping at the local scale: Concepts and challenges. In Flood Risk Management in Europe; Begum, S., Stive, M.J.F., Hall, J.W., Eds.; Springer: Dordrecht, The Netherlands, 2007; pp. 231–251. 21. Neal, J.; Villanueva, I.; Wright, N.; Willis, T.; Fewtrell, T.; Bates, P.D. How much physical complexity is needed to model flood inundation? Hydrol. Process. 2012, 26, 2264–2282. [CrossRef] 22. Panayiotis, D.; Aristoteles, T.; Oikonomou, A.; Pagana, V.; Koukouvinos, A.; Mamassis, N.; Koutsoyiannis, D.; Efstratiadis, A. Comparative evaluation of 1D and quasi-2D hydraulic models based on benchmark and real-world applications for uncertainty assessment in flood mapping. J. Hydrol. 2016, 534, 478–492. 23. Pan, T.Y.; Chang, L.Y.; Lai, J.S.; Chang, H.K.; Lee, C.S.; Tan, Y.C. Coupling typhoon rainfall forecasting with overland-flow modeling for early warning of inundation. Nat. Hazards 2014, 70, 1763–1793. [CrossRef] 24. Islam, A.S.; Bala, S.K.; Haque, A. Flood inundation map of Bangladesh using Modis surface reflectance data. J. Flood Risk Manag. 2010, 3, 210–222. [CrossRef] 25. Porter, J.; Demeritt, D. Flood-risk management, mapping, and planning: The institutional politics of decision support in England. Environ. Plan. A 2012, 44, 2359–2378. [CrossRef] 26. Gilles, D.; Young, N.; Schroeder, H.; Piotrowski, J.; Chang, Y.J. Inundation mapping initiatives of the Iowa flood center: Statewide coverage and detailed urban flooding analysis. Water 2012, 4, 85–106. [CrossRef] 27. Alfieri, L.; Salamon, P.; Bianchi, A.; Neal, J.; Bates, P.D.; Feyen, L. Advances in pan-European flood hazard mapping. Hydrol. Process. 2014, 28, 4067–4077. [CrossRef] 28. Bhatt, C.M.; Rao, G.S.; Diwakar, P.G.; Dadhwal, V.K. Development of flood inundation extent libraries over a range of potential flood levels: A practical framework for quick flood response. Geomat. Nat. Hazards Risk 2016, 1–18. [CrossRef] 29. Jung, Y.; Merwade, V.; Kim, S.; Kang, N.; Kim, Y.; Lee, K.; Kim, G.; Kim, H.S. Sensitivity of subjective decisions in the GLUE methodology for quantifying the uncertainty in the flood inundation map for Seymour reach in Indiana, USA. Water 2014, 6, 2104–2126. [CrossRef] 30. Tariq, M.A.U.R.; Hoes, O.A.C.; van de Giesen, N.C. Development of a risk-based framework to integrate flood insurance. J. Flood Risk Manag. 2014, 7, 291–307. [CrossRef] 31. Guzzetti, F.; Pasquinelli, A. Flood risk management: The role of geo-information in the insurance industry. GEOmedia 2015, 19, 20–24. 32. Schumann, G.; Bates, P.D.; Horritt, M.S.; Matgen, P.; Pappenberger, F. Progress in integration of remote sensing–derived flood extent and stage data and hydraulic models. Rev. Geophys. 2009, 47, GR4001. [CrossRef] 33. Bates, P.D.; de Roo, A.P.J. A simple raster-based model for flood inundation simulation. J. Hydrol. 2000, 236, 54–77. [CrossRef] 34. Tayefi, V.; Lane, S.N.; Hardy, R.J.; Yu, D. A comparison of one- and two-dimensional approaches to modelling flood inundation over complex upland floodplains. Hydrol. Process. 2007, 21, 3190–3202. [CrossRef] 35. Chen, J.; Hill, A.A.; Urbano, L.D. A GIS-based model for urban flood inundation. J. Hydrol. 2009, 373, 184–192. [CrossRef] 36. Vojinovic, Z.; Tutulic, D. On the use of 1D and coupled 1D-2D approaches for assessment of flood damages in urban areas. Urban Water J. 2009, 6, 183–199. [CrossRef] 37. Vojinovic, Z.; Bonillo, B.; Chitranjan, K.; Price, R. Modelling flow transitions at street junctions with 1D and 2D models. In Proceedings of the 7th International Conference on Hydroinformatics, Acropolis, Nice, France, 4–8 September 2006. Water 2016, 8, 521 20 of 20

38. Vojinovic, Z.; Seyoum, S.; Salum, M.H.; Price, R.K.; Fikri, A.K.; Abebe, Y. Modelling floods in urban areas and representation of buildings with amethod based on adjusted conveyance and storage characteristics. J. Hydroinform. 2012, 15, 1150–1168. [CrossRef] 39. Vojinovic, Z.; Seyoum, S.D.; Mwalwaka, J.M.; Price, R.K. Effects of model schematisation, geometry and parameter values on urban flood modelling. Water Sci. Technol. 2011, 63, 462–467. [CrossRef][PubMed] 40. Meesuk, V.; Vojinovic, Z.; Mynett, A.E.; Abdullah, A.F. Urban flood modelling combining top-view LiDAR data with ground-view SfM observations. Adv. Water Resour. 2015, 75, 105–117. [CrossRef] 41. Abdullah, A.; Vojinovic, Z.; Price, R.K.; Aziz, N.A.A. A methodology for processing raw LiDAR data to support urban flood modelling framework. J. Hydroinform. 2012, 14, 75–92. [CrossRef] 42. Abdullah, A.; Vojinovic, Z.; Price, R.K.; Aziz, N.A.A. Improved methodology for processing raw LiDAR data to support urban flood modelling—Accounting for elevated roads and bridges. J. Hydroinform. 2012, 14, 253–269. [CrossRef] 43. Abdullah, A.; Rahman, A.; Vojinovic, Z. LiDAR filtering algorithms for urban flood application: Review on current algorithms and filters test. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2009, 38, 30–36. 44. Carrivick, J.L. Application of 2D hydrodynamic modelling to high-magnitude outburst floods: An example from Kverkfjoll, Iceland. J. Hydrol. 2006, 321, 187–199. [CrossRef] 45. De Moel, H.; Aerts, J.C.; Koomen, E. Development of flood exposure in the Netherlands during the 20th and 21st century. Glob. Environ. Chang. 2011, 21, 620–627. [CrossRef] 46. Lowe, A.S. The federal emergency management agency’s multi-hazard flood map modernization and the national map. Photogramm. Eng. Remote Sens. 2003, 69, 1133–1135. [CrossRef] 47. Chen, A.S.; Hsu, M.H.; Tsng, W.H.; Huang, C.J.; Yeh, S.H.; Lien, W.Y. Establishing the database of inundation potential in Taiwan. Nat. Hazards 2006, 37, 107–132. [CrossRef] 48. Moel, H.D.; Alphen, J.V.; Aerts, J.C.J.H. Flood maps in Europe–methods, availability and use. Nat. Hazards Earth Syst. Sci. 2009, 9, 289–301. [CrossRef] 49. Fang, X.; Kuo, Y.H.; Wang, A. The impacts of Taiwan topography on the predictability of Typhoon Morakot’s record-breaking rainfall: A high-resolution ensemble simulation. Weather Forecast. 2011, 26, 613–633. [CrossRef] 50. Vanderkimpen, P.; Melger, E.; Peeters, P. Flood modeling for risk evaluation—A MIKE FLOOD vs. SOBEK 1D2D benchmark study. In Flood Risk Management: Research and Practice; Samuels, P., Huntington, S., Allsop, W., Harrop, J., Eds.; CRC Press: London, UK, 2009; pp. 77–84. 51. Prinsen, G.F.; Becker, B.P.J. Application of SOBEK hydraulic surface water models in the Netherlands hydrological modelling instrument. Irrig. Drain. 2011, 60, 35–41. [CrossRef] 52. Prinsen, G.; Weiland, F.S.; Ruijgh, E. The delta model for fresh water policy analysis in the Netherlands. Water Resour. Manag. 2015, 29, 645–661. [CrossRef] 53. Wu, T.Y.; Li, H.C.; Wei, S.P.; Chen, W.B.; Chen, Y.M.; Su, Y.F.; Liu, J.J.; Shih, H.J. A comprehensive disaster impact assessment of extreme rainfall events under climate change: A case study in Zheng-wen river basin, Taiwan. Environ. Earth Sci. 2016, 75, 597. [CrossRef] 54. Wei, H.P.; Yeh, K.C.; Liou, J.J.; Chen, Y.M.; Cheng, C.T. Estimating the risk of river flow under climate change in the Tsengwen river basin. Water 2016, 8, 81. [CrossRef] 55. Linde, A.H.T.; Aerts, J.C.J.H.; Bakker, A.M.R.; Kwadijk, J.C.J. Simulating low probability peak discharges for the Rhine basin using resampled climate modeling data. Water. Resour. Res. 2010, 46, 1–19. [CrossRef] 56. Lee, C.S.; Ho, H.Y.; Lee, K.T.; Wang, Y.C.; Guo, W.D.; Chen, Y.C.; Hsiao, L.F.; Chen, C.H.; Chiang, C.C.; Yang, M.J.; et al. Assessment of sewer flooding model based on ensemble quantitative precipitation forecast. J. Hydrol. 2013, 506, 101–113. [CrossRef] 57. Kimura, N.; Tai, A.; Chiang, S.; Wei, H.P.; Su, Y.F.; Cheng, C.T.; Kitoh, A. Hydrological flood simulation using a design hyetograph created from extreme weather data of a high-resolution atmospheric general circulation model. Water 2014, 6, 345–366. [CrossRef] 58. Chow, V.T. Applied Hydrology; McGraw Hill: New York, NY, USA, 1988.

© 2016 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/).