Journal of JSCE, Vol. 3, 209-223, 2015 (Originally published in Journal of Society of Civil Engineers, Ser. B1, Vol. 69, No. 1, 71-82, 2013 in Japanese)

SCENARIO ANALYSIS FOR EVACUATION STRATEGIES FOR RESIDENTS IN BIG CITIES DURING LARGE-SCALE FLOODING

Toshitaka KATADA1, Noriyuki KUWASAWA2, Satoru SHIDA3 and Masaru KOJIMA4

1Member of JSCE, Professor, Research Center for Disaster Prevention in the Extended Metropolitan Area, Gunma University (1-5-1 Tenjin-cho, Kiryu-shi, Gunma 376-8515, Japan) E-mail: [email protected] 2Member of JSCE, Research Center for Disaster Prevention in the Extended Tokyo Metropolitan Area, Gunma University (IDA Co., Ltd.) (1-5-1 Tenjin-cho, Kiryu-shi, Gunma 376-8515, Japan) E-mail: [email protected] 3Member of JSCE, River Planning Division, Water and Disaster Management Bureau, Ministry of Land, Infrastructure, Transport and Tourism (1F, Joint Government Building (Chuo Godo Chosha) No. 3, 2-1-3 Kasumigaseki, Chiyoda-ku, Tokyo 100-8918, Japan) E-mail: [email protected] 4Member of JSCE, River Department, Kanto Regional Development Bureau, Ministry of Land, Infrastructure, Transport and Tourism (16F, Saitama New Urban Center Joint Government Building No. 2, 2-1 Shintoshin, Chuo-ku, Saitama-shi, Saitama 330-9724, Japan) E-mail: [email protected]

In major cities, large populations can hinder evacuation efforts in the event of a large-scale flooding. Therefore, for investigating evacuation strategies during large-scale flooding in such cities, population size and its impact must be considered. This study aims to examine the evacuation problems that arise when big cities are subjected to large-scale flood damage. In addition, it attempts to examine strategies that can over- come the evacuation problems. In this study, we developed a scenario simulator to elaborately simulate the flooding, evacuation of residents, and status of damage due to water exposure in the Edogawa Ward, which has a population of approximately 650,000. In addition, using a simulation that reflected the intended evacu- ation behaviors of the residents for a scenario in which the banks of the overflowed, we were able to identify the inherent evacuation problems of the city, including the damage caused by the presence of a large number of evacuees. Furthermore, we analyzed scenarios including measures to reduce the extent of damage. From this information, it became clear that countermeasures such as spatial and temporal disper- sion of evacuees and a reduction in evacuation demand are necessary.

Key Words: scenario analysis, evacuation strategies, big cities, large-scale flooding, simulation

1. INTRODUCTION tion and industries in such cities has also accelerated. Consequently, extensive damage by large-scale In recent years, the risk of flooding has grown be- flooding is now a real concern. Hence, development cause of the increased frequency of torrential down- of disaster-prevention measures targeting large-scale pours and stronger typhoons as a result of climate flood damage in big cities has become an urgent change associated with global warming. Therefore, concern. In particular, demand has risen for promot- the need for disaster-prevention measures targeting ing software-based countermeasures, including flood damage has also grown. In large cities, in par- evacuation strategies. ticular, the frequency of exposure to water has re- In April 2010, the Expert Panel on Large-Scale duced as a result of aggressive flood control Flood Disaster Countermeasures of the Central Dis- measures and facilities implemented in recent years. aster Prevention Council of the Cabinet Office made However, at the same time, concentration of popula- recommendations to reduce damage from large-

209 scale flooding in the Tokyo metropolitan area1). Ac- in some areas, the river banks did not experience cording to estimations by this Expert Panel, in the washout, and fortunately, large-scale external flood- case of flooding of the , which would ing did not occur. During the largest recorded flood- cause maximum damage to the Tokyo metropolitan ing of the Shonai River, the city of Nagoya issued area, an area of approximately 500 km2 would be an evacuation advisory that targeted over a million submerged in water and 2.3 million people would be residents living in 181 school districts within the exposed to flooding. The Expert Panel also indicat- city 25 minutes after the river had exceeded the wa- ed that evacuation policies must be examined based ter level, which was the standard for announcing an on the characteristics of the city to reduce the extent evacuation advisory. However, the number of resi- of damage. As highlighted in these recommenda- dents who evacuated to facilities specified by the tions and in a case involving the city of Nagoya, city was limited to approximately 5,000 people. presented in the next chapter, evacuation strategies Hence, based on the evacuation percentage, only a in large cities pose a difficult set of problems be- very limited evacuation could be performed2). If we cause a large population is affected. also consider the evacuees who fled to locations Based on an awareness of the abovementioned other than those that were designated as evacuation problems, the aim of this study is to understand the shelters, the actual number of evacuees was a little evacuation problems that arise with large-scale higher. However, if the Shonai River had burst its flooding in large cities, as well as to examine appro- banks at that time, it is not hard to imagine the priate countermeasures for such problems. First, to large-scale damage that would have occurred. How- specifically identify the problems associated with a ever, on the other hand, if most residents who were large number of evacuees present in cities with large warned to evacuate had actually evacuated, an un- populations, we developed a scenario simulator to precedented large-scale flood evacuation would simulate the evacuation of local residents necessitated have unfolded. In such a scenario, it is easy to imag- by flooding in the Edogawa Ward of Tokyo, which ine that the evacuation would have resulted in seri- has a population of over 650,000 people. We then ous confusion. Furthermore, if large-scale external used the developed simulator to analyze a scenario flooding had occurred, we cannot deny the possibil- that assumed a washout of the banks of the Arakawa ity that there would have been numerous casualties River. This information was then used to understand among the evacuees. the status of damage based on the behaviors of the As can be imagined from this case, the problem residents to examine measures to reduce damage, as of evacuating residents from large cities is a compli- well as to verify the effectiveness of such measures. cated concern, which, depending on the presence of Consequently, we were able to clarify the evacua- a large population to be evacuated, cannot be re- tion problems inherent to large cities where stereo- solved by merely promoting evacuation. Therefore, typical evacuations cannot be promoted or derived, to examine appropriate evacuation strategies for as well as the directionality of the strategies that large cities, rather than statically considering the should be adopted. regional population as the scale for a maintenance target or an evacuation target, we must also address problems that occur during evacuation, including 2. POLICY FOR EXAMINING FLOOD- evacuation behaviors using the way people think ING EVACUATION MEASURES FOR and act as dynamic elements that depend on the LARGE CITIES progression of the situation and the acquisition of information. (1) Evacuation problems during flooding of large cities (2) Policy for examining the evacuation of resi- When Typhoon No. 15 was located in the south- dents in the event of large-scale flooding of a ern ocean waters of Kyushu the day before it made large city landfall in eastern Japan and caused extensive dam- In this study, based on an awareness of the prob- age in September 2011, it fed damp air to the stag- lems described in the previous section, we devel- nated front line on and caused intermittent, oped a scenario simulator (hereinafter, simulator) to heavy downpours in the Gifu and Aichi prefectures. specifically determine the status of flood damage in At the Shonai River, the watershed for this region, a specific region to examine the problem of evacua- the torrential downpour caused the water level to tion in the event of large-scale flooding in a metro- rise rapidly, reaching a height of 6.23 m, thereby politan area. exceeding the previous peak caused by the Tokai Much of our nation’s simulation research target- torrential downpour in 2000 by approximately 50 ing the evacuation of residents focuses on confined cm. However, while excessive water was observed spaces such as underground complexes3) or targets

210 only a portion of the given region4). There are only teristics of the targeted region, the simulator few simulations that assume the evacuation of a elaborately models the elements that express the large population in a large city5), 6). However, local affected region such as its topography, road net- municipalities are examining the evacuation of resi- work, distribution of residents, and establishment dents. In studies that target large-scale flood damage and layout of disaster prevention facilities. in which an extensive area is submerged in water, at ● The response behavior of residents is an element the very least, the evacuation conditions of the entire that has a major impact on the extent of damage, area of each municipality must be considered. Fur- and therefore, the simulator reflects and models thermore, when a large city is affected, an examina- the residents’ level of awareness to examine ac- tion specifically considering the impact of popula- tual issues and strategies. tion size on the evacuation and devastation condi- ● To examine strategies that aim toward reducing tions is required. Other countries have also conduct- human suffering, the priority of measures with ed simulation research that pertains to evacuations different direct goals, such as improving infor- in the event of flood damage over an extensive area. mation transmission and providing evacuation This research can be classified into different types support, must be examined. Therefore, the simu- such as research to determine the time required for lator models human suffering, and uses the re- evacuations based on the population and size of the duction numbers as common evaluation indica- affected area, the attributes of its residents7), 8), predic- tors for different strategies. tions of time taken for evacuation using the number of evacuees, the OD, the network of roads9), 10), and micro-simulations that model the individual behavior 3. BUILDING THE SIMULATION MOD- of the evacuees and agent-based modeling11), 12). Of EL course, much of the evacuation responses that are examined in these research areas focus on efficiently In this study, to develop a simulator in accord- evacuating residents to areas that are not submerged ance with the policies mentioned in the previous by water, as typified by hurricane response in the chapter, we used comprehensive disaster scenario United States, and much of this research is not in simulators13), 14) that we had developed previously as accordance with the evacuation policies of Japan, a basic model. The basic model was configured us- which are based on municipality units. In addition, ing three different models, namely, an information as with this research, research that utilizes micro- transmission model to express the status of transmit- simulation models to determine individual evacua- ting disaster-related information, an evacuation be- tion conditions focuses on evacuation by car. In fact, havior model to express the evacuation conditions there is no simulation research that specifically and of the residents, and a flooding model to express the comprehensively models the conditions of a region hazard conditions. In this study, for the evacuation in the event of a flood, such as the conveyance of behavior model, reduction in evacuation speed in disaster-related information to residents, the behav- accordance with the extent of traffic jams and the ior of evacuees including walking, and the dynamic level of congestion caused by evacuees on foot was relationship with flooding. considered, and improvements were made to From an awareness of such problems and from achieve modeling with a more accurate representa- previous research, in our study, we considered the tion of the evacuation speed and the evacuation be- following policies as we developed a simulator and havior in a large city. conducted the examination. We targeted an entire large city and specifically aimed at modeling large- (1) Simulator configuration scale evacuation based on the intended behaviors of To model the evacuation conditions of residents its residents. in the event of large-scale flood damage, a simulator ● To identify the specific problems with residential configured from the computational models shown in evacuations, the simulator was designed to not Fig.1 was developed. These computational models only model hazards that pertain to flooding dis- can be largely classified as models that express haz- asters, but also to comprehensively model every ards related to flood disasters, models that express aspect, including evacuation behaviors and other social response in the event of a flooding disaster, social response behaviors. and models that estimate human suffering. The ● From information obtained by an examination computational model that expresses hazards com- that targets a large city, the simulation models prises a river model that expresses the change in the impact that the population scale has on dam- water level of the rivers and a floodplain model that age and evacuation behavior of residents. expresses the water depth and area in a flooded re- ● To conduct an examination based on the charac- gion within a dike area. On the other hand, the so-

211 Hazard Phenomenon Flood Region, flowing into a dike region due to a dike break or Water Depth, overflow. The dike break flow rate or overflow rate River Water Floodplain Flow Speed Model Level Model due to river flooding is calculated by the Honma 16) River Flood Region, Damage lateral overflow formula based on the calculated Water Water Depth, Occurrence mesh water level of the floodplain, corresponding to Level Flow Speed Model the water level in accordance with the river model, Disaster Key Information Evacuation Residential Transmission Informa- Behavior Computational and the relationship with the dike break height and Distribution Model tion Model Model embankment level height. Note that with this simu- Reference Social Response Information lator, the flooding conditions within the dike area Fig.1 Configuration of the simulation model. must be evaluated in conjunction with the evacua- tion behavior of the residents, and therefore a 10 m cial response model comprises an information detailed computational mesh is used. transmission model that expresses the conditions c) Information transmission model surrounding the transmission of disaster information The information transmission model expresses by the government to the residents in accordance conditions surrounding the transmission of warnings, with the hazard conditions, and an evacuation be- evacuation advisories, and other disaster-related in- havior model that expresses the decision by resi- formation to the residents through radio communi- dents to evacuate based on the acquisition of disas- cations for disaster prevention, mass media, and ter-related information and the conditions surround- other such means. Moreover, to model the condi- ing their movement to evacuation facilities. Fur- tions for which information is disseminated thermore, a damage occurrence model is used to throughout the region by word of mouth among res- estimate the scale of human suffering from the re- idents, the individual behavior of residents with re- sults output by the other two model groups. spect to conveying information is also considered. To implement this model, we used a disaster infor- (2) Overview of the simulation model mation transmission simulation model17) that we This section is a general overview of each compu- developed in a previous research. The disaster in- tational model presented in Fig.1. Note that each formation transmission simulation model consists of model basically utilizes existing technology, and parameters that handle the level of information therefore, the details are limited to the information transmission behaviors between residents. Moreover, presented by the documents listed in the references. it can model the conditions of transmission behavior a) River model according to measures such as the cohesiveness of The river model expresses temporal changes in the local community at the time of a disaster. This water levels and flow rates in river channels using model demonstrates information transmission in the the tidal level at the river mouth as the starting water event of a disaster by radio communications for dis- level and the hydrograph for the flow rate applied at aster prevention and by the media based on a model the upstream end as a boundary condition. Because of information transmission behavior of residents. the temporal changes in flooding conditions are ex- d) Evacuation behavior model pressed with this model, one-dimensional unsteady The evacuation behavior model primarily ex- flow calculation15) is used as the computational presses the conditions of movement of the residents method. In addition, with this model, points for cal- from their homes to their evacuation destinations via culating the flow rate and flow speed are placed in a roadways. The timing at which evacuations begin is transverse cross-section of the river channel to es- based on the timing at which such as evacuation tablish the river width and hydraulic radius, and cal- advisories are acquired as modeled by the infor- culations of the water level are performed using an mation transmission model. It includes the time intermediate point in this cross-section. If an area of spent on evacuation preparations. Furthermore, for convergent flow exists in the targeted region, the the evacuation destination and route, as a general flow rate at the downstream end of the area of con- rule, the evacuation facility closest to the resident’s vergent flow is calculated using a common water home and the shortest route to that facility are used. level calculation point for a downstream cross- Note that the modeling of this portion is described section of the branch streams and a common water in detail in the next chapter. level calculation point for the area of convergent During an evacuation, the position of an evacuee flow in the given river. is determined from the time that has lapsed since b) Floodplain model evacuation first started, as well as roadway data in The floodplain model is expressed by Cartesian accordance with the travel speed. The roads of a coordinate system-based two-dimensional unsteady targeted region are expressed as a network com- flow calculations15) of the behavior of floodwaters posed of links that express roads and nodes that ex-

212 press intersections. Each link has a changeable con- lated using the surface areas of the sidewalks and dition that expresses whether or not passage is pos- the number of evacuees in the given link. If 6 ≤ d, it sible at any given point in time. If a link that an was assumed that evacuees could not enter the given evacuee tries to enter is not passable, the unpassable link from a separate link. Furthermore, if the evacu- link is excluded from the network, and route search- ation took a long time, a reduction in speed due to es that use the resulting network are implemented to fatigue of the residents evacuating by walking is determine the appropriate detour. Moreover, the conceivable; therefore, the speed reduction due to flooding conditions expressed by the floodplain fatigue was considered using equation (3) 21). model are reflected every 10 seconds to update the r = 1 ⁄ (0.982 + e1.12t − 4) (3) road conditions, and the emergence of unpassable f locations in conjunction with the spread of the where rf is the speed coefficient due to fatigue and t flooded area and the detour behavior of evacuees in is the amount of time (hours) that has lapsed since accordance with such are also modeled14). the evacuation started. A reduction in speed in Both walking and the use of automobiles are flooded areas was also considered using equation modeled as evacuation measures. In the event of (4) 22). flood damage, evacuation is done by walking as a r = 1 – w ⁄ w (4) general rule, but for reasons such as vehicles being a w max valuable asset and movement by walking during a where rw is the speed coefficient due to flooding, w torrential downpour being difficult, evacuation by is the depth of flood water (m), and wmax is the depth car is widely used18). Travel speed is based on an of flood water that is the limit for walking (m, set to arbitrarily established speed. However, regarding 1 m for calculations during this study). While the the speed of vehicles, when an evacuee attempts to impact of flow speed should be considered with re- enter a link and the traffic volume of a link on spect to decreases in speed in accordance with which an evacuee is currently traveling changes, the flooding; but because it was difficult to model the change in speed is expressed in accordance with flow direction and the impact from shallow water traffic volume by updating the speed using Green- depths, in this study, only water depths that had a shields19) equation. more dominant impact were considered23). The above coefficients relating to speed decrease vc = vf · (1 – k ⁄ kj) (1) were considered, and the evacuation speed of pedes- where vc is the vehicle speed (km/h), vf is the free trian evacuees was expressed using equation (5). travel speed (km/h), k is the traffic density (cars/km), vw = rd · rf · rw · vs (5) and kj is the saturation density (cars/km, set to 140 for calculations for this study). Free travel speed is where vw is the pedestrian evacuation speed and vs is based primarily on the scale of the roads, and there- the initial value for pedestrian evacuation speed. fore, in this study, expressways were set to 30 km/h e) Damage occurrence model and all other roads were set to 20 km/h based on In the case of flood damage due to river flooding, surveys, such as a road traffic census of the targeted even if a resident is within the flooded area, the per- region. Moreover, traffic density was determined son will not necessarily become a victim. Therefore, using an inverse number of the spacing between the with this model, the flooded population was catego- front of the vehicle and the vehicle ahead. With this rized in three categories, as shown in Table 1, based model, vehicle lanes and traffic signals were omitted on the evacuation conditions of the residents, the to simplify the model, and priority was given to characteristics of their homes, and the flooding con- through-traffic at intersections. Because of evacua- ditions. This information was used as evaluation tion behaviors in large cities, congestion resulting indicators for the scale of human suffering. With from pedestrian evacuees occurs; therefore, reduc- this model, residents who were in a flooding situa- tion in the evacuation speed must also be considered. tion that made it difficult for them to walk were Therefore, the following equations were used with considered to be residents who required rescuing. reference to previous studies20) to model the reduc- To determine walking difficulty, an equation was tion in walking speed in accordance with the level used from Suga et al.23) that determined walking of congestion. difficulty based on body height and hydrodynamic force. rd = 1 (d < 1.5) rd = 1.3 – 0.2d (1.5 ≤ d < 6) (2) rd = 0.1 (6 ≤ d) where rd is the speed coefficient in accordance with the level of congestion, and d is the level of conges- tion (people/m2). The level of congestion is calcu-

213 Table 1 Classification of flooded population. Expressway Shelter Flooded Popu- National or Regional disaster Definition city road prevention center lation Other raod Edogawa Ward > r A person who stays in his or her own home with- Ichikawa City e iv Sidewalk width of 3.5 m R Person staying out evacuating, and who is not personally experi- a w a g o within a flooded encing flooding but whose home is within a d E Ichikawa City area flooded area. (The ground surface of the person’s Urayasu City home is flooding.) A person who has not evacuated or is in the Person in dan- process of evacuating, and while emergency ger within a rescue is not necessary, the person is located flooded area within flood waters.

A person who has not evacuated or is in the A Person requir- rak Katsushika Ward aw process of evacuating, is surrounded by flood a R ing emergency ive waters to the extent that walking is difficult, and r > rescue 021 an emergency rescue is necessary. km Koto Ward Fig.2 Region targeted by the study. Table 2 Calculation conditions. Classifi- References and Item Settings 4. ESTABLISHMENT OF THE BASIC cation thinking Assumed 200-year probable SCENARIO Design flood flooding flooding Starting Flooding AP + 2.2 m High water level In this chapter, we summarize the region targeted water level by the study and the calculation conditions that were Dike break Arakawa River Maximum dam- established (see Fig.2 and Table 2). areas left bank 7.0k age 651,733 people Population 303,245 house- Ward documents households (1) Establishing the calculation conditions holds Residents Walking speed a) Targeted region Walking based on sex and Statistical data speed In this study, the Edogawa Ward of Tokyo was age selected to examine evacuation strategies in the Layout 256 speakers Ward documents Loud- Audio range 280 m Ordinary value event of large-scale flooding. The Edogawa Ward is speakers Hearing rate 30 % Previous studies located in the southeastern part of Tokyo and is Number 10 cars Ward documents sandwiched between two major rivers, the Arakawa Publicity Speed 20 km/h Ordinary value River on the West and the Edogawa River on the cars Audio Range 80 m Ordinary value East. In addition, because the Edogawa Ward is lo- Hearing rate 30 % Ordinary value Phones cannot Communi- Communica- By word of cated at a position furthest downstream from both be used due to cation tion method mouth only congestion. rivers, if the embankments of the rivers collapse, the between Activity Not aggressively Survey of resi- residents geographical features of the area are such that the level communicated dents ward is at the risk of flooding regardless of where Sidewalk 2 m, some 3.5 m Ward documents the embankments collapse. Furthermore, because of width Roads 30 km/h on high- Ward documents Basic vehicle the effects of ground subsidence, 70% of the land ways, 20 km/h on speed area is a zero-meter region below the high-tide level, all other road and because there is no natural drainage, after flood- Shelters 106 shelter Ward documents Regional Evacuation ing has occurred, an extensive area will be covered disaster facilities 3 locations in flood waters for an extended period of time with prevention prolonged damage. Moreover, even within Tokyo, centers the Edogawa Ward is a large, prominent city with a population of approximately 650,000 people, and if maps as a reference. flooding should occur, there are concerns that a First, to model the residents, we prepared virtual large population will need to be evacuated. residents, based on the demographic composition as b) Conditions related to topography and popula- of January 2010 that was announced by the Edoga- tion wa Ward. This data included population statistics by First, a topographical map of the targeted region district, age, and gender. Next, we targeted detached was prepared with 10 m mesh precision using a la- homes and apartment buildings shown in residential ser profiler. Next, the distribution and shapes of maps by district, and randomly distributed virtual buildings were modeled using commercially availa- households for the number of households that exist- ble electronic residential maps. The number of sto- ed according to population statistics. We then dis- ries of each building was set using residential maps tributed the population by randomly allocating the and laser profiler data as a reference. Finally, net- residents prepared for each household premised on work data that expresses the roadways was prepared the conditions that detached homes contained only a using the road shapes contained in the residential single household.

214 Maximum Flood Water Depth Edogawa Ward Table 3 Conditional stages during flooding 0.0 m - 0.5 m or less Assumed Location of Dike Breakage presented with the survey. 0.5 m - 1.0 m or less No. Conditions 1.0 m - 2.0 m or less 1 Rain seems heavier than normal at home. 2.0 m - 3.0 m or less Kyuedogawa River > Heavy-rain warning or flooding warnings are Shinnakagawa River > 2 3.0 m - 4.0 m or less announced. Over 4.0 m Warnings continue for a long time, and the rain does 3 not stop falling. 4 An evacuation advisory is announced. Residents have become aware of conditions in which Arakawa River > 5 the embankment is likely to collapse. 021 6 Evacuation instructions are announced. km Notification has been received that the embankment 7 Fig.3 Assumed dike-breakage locations and maximum flood collapsed. water depth distribution. 8 Flooding reaches the area near a resident’s own home. 9 A resident’s own home begins to flood. c) Conditions related to hazards The river model was modeled based on the Ara- kawa River. For other rivers, topography data were residents were evacuated from that location to the utilized to model full capacity conditions. The as- next closest evacuation site. sumed flooding was set to the flooding levels of Regarding radio communications for disaster pre- September 1947 based on the design flood (1/200th vention, the 256 outdoor loudspeakers prepared by probability) of the Arakawa River. For the dike the ward were modeled. The rate at which infor- break, a 7.0 km location along the left bank of the mation was acquired from the outdoor loudspeakers Arakawa River was selected, which is a location was set to 30% based on a previous study24). Moreo- with a relatively large population that would be af- ver, the timing at which the evacuation advisory was fected by flooding within the targeted region. The issued and the timing at which evacuation instruc- width of the dike break at the river embankment was tions were issued were set as the point in time at set to 340 m based on the dike failure results along which the water level at the Iwabuchi site of the the Tone River caused by as a Arakawa River reached the water level determined reference. In addition, the timing of the dike break at for evacuation and when it reached a flooding risk the embankment was set to the point in time at level, respectively. Both timings were established by which a reference water level was reached with con- referencing the ward’s regional disaster prevention sideration of the current allowance height of the plan. Consequently, with the assumed flooding sce- embankment and maintenance conditions. Flooding nario, an evacuation advisory was issued 4 hours calculations based on these conditions provided re- and 44 minutes prior to the collapse of the embank- sults that indicated that a large portion of the region ment, and the evacuation instructions were issued 2 between the Arakawa River and the Shinnakagawa hours and 41 minutes prior to the collapse. River would be flooded (see Fig.3). d) Conditions pertaining to disaster-prevention (2) Intentions of residents regarding flooding facilities and disaster information evacuation and method for reflecting those in- The evacuation facilities were modeled by target- tentions ing shelters such as elementary and middle schools, In this section, we discuss the results of an and regional disaster-prevention centers (higher ele- awareness survey that was conducted to understand vation parks) designated in the regional disaster the intended behavior of the residents in the event of prevention plan for the Edogawa Ward. When con- a flood disaster and the method that was used to re- sidering shelters, we referenced usable floor space, flect that behavior in the evacuation behavior model. which excluded floors of buildings that were at risk a) Method for reflecting the intentions of the res- for flooding based on documentation from the idents and an overview of the survey Edogawa Ward, and the number of people that can In this study, we decided to consider the inten- be accommodated per shelter in an emergency evac- tions of the regional residents in three areas: (1) de- uation was set based on two people per 1.65 m2 (the termination of intent to evacuate, (2) timing at size of one tatami mat). In addition, because the area which evacuation began, and (3) behavior at the has considerable floor space in its regional disaster- time of evacuation. We expressed these intentions in prevention centers to accommodate over two million the evacuation behavior model. First, regarding the people, the number of people that can be accommo- determination of intent to evacuate, we presented dated in the centers was set as limitless. According the conditional stages during flooding, as shown in to the evacuation behavior model, if an evacuation Table 3, and identified the evacuation intentions of facility had exceeded its accommodation limit, the the residents at each stage. Furthermore, because the

215 Table 4 Overview of the resident awareness survey. 100 Survey period February 19–23, 2010 When subjected to persuasion to evacuate. Residents 20-years-old or older of the 80 Survey target (Includes people who Tokyo, Edogawa Ward would evacuate Survey method Internet survey 60 regardless of persuasion.) No. of responses 3,000 40 ● Basic attributes ● Awareness regarding flooding dis- 20 When not subjected to

Intent to Evacuate (%) persuasion to evacuate. Survey items asters Rateof of Determination ● Behavioral intentions regarding 0 No.1 No.2 No.3 No.4 No.5 No.6 No.7 No.8 No.9 flooding disasters than normal. Rain appears heavier announced. warnings are flooding Heavy rain or does not stop falling. long time, and the rain a for Warnings continue is announced. An evacuation advisory likely collapse. to You learn the dike is are announced. Evacuation instructions collapsed. dike that the Notified near your own home. Flooding reaches area to flood. Residents home begins impact of persuasion to evacuate is large25), we also decided to determine and understand the evacuation intentions when residents were subjected to persua- sion. Next, for the timing at which evacuation began, we decided to add the time required for preparations Fig.4 Rate of determination of intent to evacuate by conditions. after the determination of intent to evacuate was reached. Finally, regarding evacuation behavior, we b) Determination of intent to evacuate also decided to identify the evacuation methods and The percentages of people who decided to evacu- destinations. ate to a location outside of their homes at each of the The abovementioned behavioral intentions of the stages, No. 1–9, as shown in Table 3, which were residents in the event of a flooding disaster were obtained from a survey of the residents, and the per- determined using a survey targeting the residents of centages for a case in which the residents were also the Edogawa Ward. An overview of the survey is subjected to persuasion are shown in Fig.4. presented in Table 4. This survey was implemented As shown in the figure, if we first focus on data using an online survey service, and responses were for when the residents are not subjected to evacua- obtained from 3,000 residents of the Edogawa Ward. tion persuasion, a large increase is observed at the The survey asked questions regarding basic attrib- point in time that the residents learn about the evac- utes, including questions regarding the presence of uation advisory, but that percentage is limited to any family members who would have difficulty in 34%, which is approximately one-third of all the evacuating on their own and would require support. residents. Furthermore, the evacuation percentage To use the model to determine evacuation behav- increases dramatically to approximately 70% when iors that reflected the intentions of the residents, evacuation instructions are issued. We can also see responses obtained in advance through the survey that at the final point when the residents’ homes are were randomly selected for each household, and beginning to flood, the percentage of residents who behavior in line with the responses for each question have determined to evacuate reaches about 92%, was then applied to the given household. Rather meaning that approximately 8% of the people have than using the individual responses as is, another no intention to evacuate their own homes at any of conceivable method was to model behavior based the stages. on tallied results for each question. However, this In determining the intent to evacuate when sub- method was not adopted because it did not allow for jected to persuasion at each stage prior to the issuing expressing the correlation of the responses for each of the evacuation instructions, the percentage of res- question, and consistency in individual behavior idents who have determined to evacuate is about would be lost. Furthermore, differences were ob- 20% higher than when they are not subjected to served in the evacuation intentions between the res- evacuation persuasion. However, after the evacua- idential areas (six districts) and the number of sto- tion instructions are issued, that difference decreases. ries of the residence (single-family homes, single- These results show the importance of persuasion to story apartment buildings, two-story or higher sin- promote evacuation in the initial stages of a disaster. gle-family homes, two-story or higher apartment The timing of the occurrences in the simulation of buildings), and therefore, the responses were divid- each event corresponding to the conditions of Table ed into groups according to these two attributes. 3 is shown in Fig.5. To establish the timing of these Moreover, when selecting the response to be associ- events, we used the time difference between the ated with each household, the response was selected evacuation advisory and the embankment failure from the group for which the household matched the obtained as-is from the calculation results of the residential area and the number of stories of the res- abovementioned river model. Using the time differ- idence. ence, the timings when the flood warning and the river-related information would be transmitted were

216 Calculation Warning Evacuation River In- Evacuation Dike Time Require for 70 min Less than 10 min 6% are Started Information Advisory formation Instructions Failure Evacuation Preparations or more (No.1,2) (No.3) (No.4) (No.5) (No.6) (No.7) N=3,000 10% 10-20 min 60-70 min 21% 20% 01234567 (hour) 20-30 min 50-60 min 1% 9% Fig.5 Event occurrence timing. 40-50 min 4% 30-40 min 30% Evacuation Destination Evacuation Method set using a time of two hours prior to the issuance of Other 3% Location outside of Other Edogawa Ward 20% Bicycle 4% the evacuation advisory as the calculation start time. 2% Taxi 3% Area of higher Motorcycle 1% For this situation, conditions that had already oc- Nearby elevation 8% Automobile Ride in vehicle of elementary 15% curred prior to starting the simulation were assumed Home of a or middle Another home 1% relative or school as conditions No. 1 and No. 2. In addition, condition acquaintance Public 51% Walk 2% Nos. 8 and 9 occurred at different timings for each facility 73% Private facility 16% individual depending on the location and flooding 1% N=3,000 N=3,000 status of their homes. The timings shown in Fig.5 Fig.6 Time required for evacuation preparations and behav- actually show the start of information transmission ior at time of intention to evacuate. using the occurrence of an event as the trigger, and therefore, the exact timing of the occurrence of an evacuation was started was determined by adding event is the timing at which each resident receives the time required to prepare for evacuations once the information via transmission by radio communica- decision was made to evacuate. tions for disaster prevention, mass media, etc. d) Evacuation behavior Therefore, if a resident is unable to obtain infor- The tallied results for the survey items relating to mation, the relevant event does not occur. behavior at the time of evacuation are shown at the In the evacuation behavior model, the determina- bottom of Fig.6. tion of intent to evacuate is decided by residents First, the ward’s evacuation plan includes evacua- according to the conditions for receiving disaster tion to both regional disaster prevention centers information and evacuation persuasion. More spe- (higher elevation) as a basic evacuation destination cifically, if the information that was obtained was and shelters (elementary and middle schools) as information that indicated a situation in which the emergency evacuation destinations; however, incon- resident responded that he or she would decide to sistencies occurred between the ward’s evacuation evacuate, or was a later situation, the resident was plan and the evacuation destinations desired by the treated as having made the decision to evacuate. residents. More specifically, many respondents re- Furthermore, if information was obtained from an- plied that they would evacuate to a nearby elemen- other resident who had already decided to evacuate, tary or middle school, and when other public facili- this was viewed as having been subjected to evacua- ties were included, the percentage of respondents tion persuasion, and depending on the evacuation who intended to evacuate to a facility of the ward intention at the time that the resident was subjected reached 67%. On the other hand, respondents who to persuasion, the resident either decided at that expected to evacuate to a higher elevation location point in time to evacuate or would decide to evacu- was limited to 8%, which was a very limited per- ate, depending on the details of the information that centage. Twenty percent of the respondents imag- was received. ined broad-based evacuation to an area outside the c) Timing at which evacuation started ward. Responses pertaining to the time required for Next, when asked about their evacuation methods, preparations from the moment the decision was 73% of the respondents assumed that evacuation made to evacuate until the resident left his or her would be done by walking, which was characteristic home to evacuate are shown in the top portion of of the group. Evacuation using an automobile also Fig.6. According to these results, the number of included evacuating as a passenger in someone households that can complete preparations within 30 else’s car or using a taxi or motorcycle. The per- minutes from the time that the decision was made to centage of respondents falling into these categories evacuate was limited to 36%. Moreover, the re- was around 20%. When considered in terms of ab- sponse that accounted for the largest percentage of solute numbers, it is clear that the number of auto- responses indicated that a preparation time of 30–40 mobiles used for evacuation would be large. minutes would be required, and this was the case for In the evacuation behavior model, the evacuation one-third of all respondents. On the other hand, a destination and method for residents who have start- similar number of respondents indicated that they ed evacuation behavior were selected based on these would need one hour or longer to prepare. In the responses. To simplify the model, if evacuation was evacuation behavior model, the timing at which done to a public facility, a private facility, or a home

217 Presence of family member who Availability of evacuation support Table 5 Strategy scenarios. would have difficulty evacuating by family, etc. for persons having on his or her own power. difficulty evacuating. Symbol Strategy Scenario Modeling Method Implement strategies to support Support house- I don’t households that would have difficulty know holds who would A1 in evacuation and enable evacuations 13% have difficulty Can evacuate based on the condition of requiring because of evacuating. Not Present Present support from double preparation time. 90% 10% non-family members 4% Take measures to Improve information transmission so ensure that all that all residents receive the infor- A2 Can evacuate with households re- mation within one hour from the time family support Evacuation would 74% ceive information. that information transmission started. N=3,000be difficult in N=312 current state 9% Improve the awareness of residents so Improve evacua- that all residents decide to evacuate at Fig.7 Presence of family members who would have difficulty A3 evacuating on their own and availability of evacuation tion timing. least by the time they hear the evacua- tion advisory. support by family members, etc. Assume advance preparations during Improve evacua- normal conditions so that evacuation A4 tion preparation can be started within 30 minutes from of a relative or acquaintance within the ward, a con- time. the moment the decision is made to dition of evacuation to a shelter was assumed. In evacuate. addition, if a resident indicated that he or she would Prohibit danger- After information of a dike break has ous evacuations been obtained, prohibit the evacuation evacuate to an area outside the ward, such a case B1 by residents living of residents living on the third floor or was modeled using an evacuation to virtual facilities in upper stories. higher. distributed at the end points of the network of roads After the information of an evacuation that led to the outside of the ward. Furthermore, the advisory has been obtained, prohibit evacuation methods were selected as either walking Limit the number the evacuation of residents living on B2 of people targeted the third floor or higher. Evacuate all or by car. If the response indicated that evacuation for evacuation. residents living on the second floor or would occur by walking or by bicycle, the condition lower by the time the evacuation was modeled as evacuation by walking, and for all advisory has been issued. other cases, the condition was modeled as evacua- Guide evacuees to Based on evacuation to regional disas- regional disaster- ter-prevention centers, do not evacu- tion by car. C prevention cen- ate to shelters other than in an emer- e) Persons experiencing difficulty in evacuating ters. gency after flooding begins. on their own After an evacuation advisory has been Introduce roads The two graphs shown in Fig.7 are respective re- D1 issued, partition a part of the roads for for pedestrians. sponses to questions regarding (1) the presence of a use by pedestrians (see Fig.11). family member who would have difficulty evacuat- Decentralize Allow only those households with D2 evacuation desti- vulnerable family members to evacu- ing on his or her own power and (2) whether or not nations. ate to shelters. evacuation support could be provided for house- Complete the evacuation of house- holds having family members who would have dif- Promote evacua- holds with vulnerable family members D3 ficulty evacuating on their own. When we look at tion in advance. two hours prior to the evacuation these results collectively, it is clear that 10% of all advisory. households have family members who would have difficulty evacuating on their own, and at least about 10% of such households would have difficulty in obtaining support for evacuation by other family 5. SCENARIO ANALYSIS members. Therefore, evacuation on a household unit would be difficult. To analyze scenarios for examining measures to These results were reflected in the evacuation be- reduce damage in the event of large-scale flooding havior model, and a condition that included the in a large city, we first implemented simulations of presence of a person who would have difficulty scenarios that reproduced the intended behaviors of evacuating under his or her own power and for the residents, as understood from the survey (here- which evacuation would be difficult was modeled as inafter, the current status reproduction scenario). a state in which evacuation to a location outside of Using that scenario, we gained an understanding of their own home was not possible. Half of the re- the assumed-damage scale with the current condi- sponses of “I don’t know” regarding the availability tions and the factors thereof. Next, we implemented of support by family members and others for per- simulations of various scenarios that modeled dam- sons who would have difficulty evacuating were age-reduction strategies based on the factors that handled as not being capable of evacuating. generated the damage (see Table 5), and we then verified the results and identified new issues.

218 Composition of Residents People staying People in danger People requiring Number of people Composition of All Residents Requiring Emergency Rescue within a flooded area within a flooded area emergency rescue starting evacuation 7 70 63.5 65.2 N=651,733 N=37,158 Requiring 6 53.7 60 emergency rescue 48.5 49.4 49.3 37,158 people, 6% In process of 5 50 evacuation Completed evacuation 36,417 people, 98% 4 40 435,501 people 6.3 30 67% 3 Prior to starting 4.5 4.1 4.3 Not yet decide 2 3.7 3.9 4.1 3.8 20 to evacuate evacuation 3.5 2.4 133,148 people 741 people, 2% 1 10 In process of 20% 0.7 0.8 0.6 0.8 0.6 0.5 0.6 0.0 Have Who People of Number evacuation Flooded Population(10,000 people) 0 0 13,239 people Preparing for Current Support Measures to Improve Improve Implement people) (10,000 Evacuation Started Residents who have Status households ensure all evacuation evacuation all strategies. 2% evacuation Reproduction who have households timing (A3). preparation (A1-A4) 32,687 people difficulty evacuating 90 people, 12% difficulty receive time (A4). 5% evacuating information Did not receive (A1). (A2). Preparing advisory Fig.9 Flooded population and change in the number of people for 71 people, 10% evacuation starting evacuations according to strategy scenarios. 433 people, Prior to making Composition of Residents 58% decision to evacuate Requiring Emergency 147 people, 20% Rescue Prior to Starting case in which each scenario for each strategy was Evacuation N=741 conducted independently with respect to the current Fig.8 Results of simulation of current status reproduction. status reproduction scenario and for a case in which all the strategies were implemented collectively. The (1) Assumed damage with current conditions definition of each flooded population, such as the based on the behavioral intentions of the resi- number of people who stayed within the flooded dents area, is shown in the diagram in accordance with Figure 8 summarizes the conditions of each resi- Table 1. From these results, if we look at the change dent when 24 hours had passed since the beginning in the number of evacuees, when each strategy is of the current status reproduction scenario. The implemented independently, the evacuation percent- number of residents requiring emergency rescue age and the timing at which evacuation is started are would exceed 37,000, and it is clear that large scale improved, and as a result, more residents begin injury would result. It is also clear that 98% would evacuating compared to the current status reproduc- be injured while evacuating and would require tion scenario. When all the strategies are imple- emergency rescue. Regarding the contributing fac- mented collectively, it is clear that approximately tors, many residents decided to evacuate when the 650,000 people, a number that corresponds to the risk of flooding had increased after the evacuation entire population, begin to evacuate. However, when advisory (see Fig.4). Thus, it is clear that these resi- we focus on the flooded population, with the scenar- dents became engulfed in the flooding while they ios other than A2, it is clear that the introduction of were evacuating. Further, if we look at the composi- improvement measures has the opposite effect and tion of the residents requiring emergency rescue results in an increase in the number of people requir- prior to beginning evacuation, approximately 60% ing emergency rescue. In particular, with the current are in the middle of evacuation preparations, 20% status reproduction scenario, 45,000 residents stayed do not have any intention to evacuate outside their in the flooded region; however, when all of the own homes, and the rest are either residents who strategies were implemented, the number reduced to have not been able to receive any evacuation infor- zero. However, people requiring emergency rescue mation or residents who would have a difficult time due to serious injuries totaled 63,000, approximately evacuating. From these results, with the current sta- 1.7 times more than the level estimated with the cur- tus reproduction scenario, it is clear that injury is rent status reproduction. This result suggests that the caused by various factors, including a delay in the promotion of stereotypical evacuation behavior of determination of intent to evacuate. It is important all residents with current conditions has the risk of to note that many shelters ultimately reached an leading to an increase in injury. We also analyzed over-capacity state. the individual implementation of scenarios for each strategy, and found that while scenario A3 involved (2) Effects of improving evacuation behavior active evacuation at the evacuation advisory stage, Based on the problems identified from the current when the evacuation scale became very large, the status reproduction scenario, we determined the in- evacuation facilities became overcrowded, conges- jury reduction effects from improving the evacua- tion became a serious concern, and the amount of tion percentage and the timing at which evacuation injuries increased. In addition, scenarios A1 and A4 began. The assumed strategies are the scenarios A1– did not involve improvements in the timing at which A4, shown in Table 5. the decision to evacuate was made, and therefore, Figure 9 summarizes the number of people who the number of people evacuating under dangerous started evacuating and the flooded population for a conditions after the evacuation advisory increased,

219 People staying People in danger People requiring Number of people resulting in an increase in injuries. On the other within a flooded area within a flooded area emergency rescue starting evacuation 8 70 hand, by ensuring that information was thoroughly 65.2 *Implemented every strategy of A1, A2 and A4. 7 56.4 56.4 conveyed with scenario A2, the evacuation of resi- 6.3 53.6 60 6 dents who made the decision to evacuate at an early 50 5 4.3 4.3 stage was reliably started, and the evacuation timing 4.0 40 4 3.6 3.2 30 became faster. In addition, because evacuation in- 3 2.9 tentions were based on an awareness of the current 2 20 state, a significant increase in the number of evacu- 0.8 0.8 1 0.7 10 Number ofWho People Have 0.4 Flooded Population(10,000 people)

0.0 people)Started(10,000 Evacuation ees, as seen with scenario A3, did not occur, and as 0 0 If all residents Prohibit dangerous Limit the number Guide to regional a result, the amount of injuries decreased. have evacuated evacuations by of people targeted disaster- (Reference) residents living in for evacuation prevention centers From the above analysis, we were able to confirm upper stories (B1) (B2) (B2+C) the following problems that occur in conjunction Fig.10 Effects of limiting evacuees and guiding them to with an increase in the number of evacuees. regional disaster-prevention centers. Problem of risky evacuation behavior: If resi- dents who live in high-rise buildings stay in their Fig.10). However, when compared to the current own homes, and if their homes become flooded, status reproduction scenario, the reduction in the they require emergency rescue when they evacuate number of people requiring emergency rescue is their homes. minimal, and it is clear that a large amount of injury Problem of the capacity of evacuation facili- still occurs. Moreover, limiting the number of evac- ties: The capacity of shelters is expected to be ex- uees resulted in 40,000 people being surrounded by ceeded even with the current status reproduction flood waters, and therefore, separate measures that scenario, and is reached at an even earlier timing target these residents are required. when the number of evacuees increases. Problem of evacuee congestion: The increase in (4) Injury reduction effect by improving evacua- the number of evacuees causes serious vehicle and tion destinations pedestrian congestion, resulting in a slowdown of While 70% of the residents intend to evacuate to the evacuation time. a public facility within the ward, the total capacity Significantly, assuming a condition in which none of shelters is approximately 270,000 people. There- of the residents evacuate, the number of people re- fore, if a large-scale evacuation is implemented, it quiring emergency rescue decreases to approximate- would be unrealistic for residents to evacuate as in- ly 6,000 people, which is significantly lower than tended. On the other hand, regional disaster- the current status reproduction scenario. However, prevention centers have enough capacity to accom- the number of people who becomes surrounded by modate a total of over 2 million people. In the flood waters increases to 310,000, and therefore, ward’s evacuation plan, these centers are stipulated some response is required. as priority locations for the evacuation of residents. In the next analysis, we established a scenario (C) (3) Injury reduction effects from evacuation de- that used these regional disaster-prevention centers. mand reduction measures In this scenario, regional disaster-prevention centers Based on the results described in the previous were stipulated as the evacuation destination accord- section, we conducted simulations that assumed ing to the residential district. However, when infor- strategies to reduce risky evacuation behavior im- mation of a collapse of the embankment or flooding mediately before flooding and after flooding. More of the surrounding region was obtained, evacuation specifically, the simulations included prohibiting to the nearest evacuation location that included shel- evacuation by residents living on the third floor or ters was assumed as the emergency evacuation be- higher after a dike break (B1), evacuating all resi- havior. dents living on the second floor or lower by the time If we examine the results of adding the conditions the evacuation advisory was issued, and prohibiting of C to those of B2 (right side of Fig.10), we can the evacuation of residents living on the third floor confirm that injury is reduced by stipulating the ap- or higher once an evacuation advisory was issued propriate evacuation destination. However, since the (B2). These scenarios also assumed that the strate- decrease is approximately 4,000 people, the number gies for A1, A2, and A4 had already been imple- of people requiring rescue still remains over 30,000. mented. If we examine the simulation results for B1 and (5) Injury reduction effect from dispersing evac- B2 in Fig.10, it is clear that both scenarios signifi- uees cantly reduce injury when compared to the scenario While a reduction in the number of injuries was in which all residents were evacuated (left side of observed in improving the evacuation destination, a

220 People staying People in danger People requiring Time required to within a flooded area within a flooded area emergency rescue evacuate 140 5 121.7 *Implemented every strategy of A1, A2, A4, B2 and C. 4.3 4.3 4.3 120 4 97.4 3.2 82.8 3.2 100 3 64.4 80

2 60 1.2 (minutes) 0.9 40 1 0.7 Number of Passing 20 0.2 Pedestrians (people) 0.0 0.1 0.1 0 - 1,970 0 0

Without Introduction Decentralization Promotion of Average Time requiredto evacuate 1,971 - 7,587 people) Flooded Population (10,000 Dispersion of roads for of Evacuation Advance 7,588 - 19,008 (Reference) pedestrians Destinations Evacuation 19,009 - 46,539 (D1) (D2) (D3) 46,540 - 104,731 Regional Disaster Prevention Center Fig.12 Effects of dispersing evacuees. Pedestrian Exclusive Road Fig.11 Number of pedestrian evacuees passing on roads and areas where pedestrian-exclusive roads are introduced. nearby shelters, the evacuation time was reduced by another 20 minutes compared to D1. However, con- large number of evacuees were guided to a small gestion by evacuees heading to regional disaster- number of regional disaster-prevention centers. prevention centers was not adequately alleviated, Consequently, large-scale congestion occurred along and the injury reduction level was limited to approx- some roads particularly due to pedestrian evacuees imately 20,000 people. Finally, with strategy D3, (see Fig.11). Congestion occurred because of evac- the evacuation of vulnerable households was com- uees converging on specific roads in numbers that pleted at an early stage. Therefore, congestion at the temporarily exceeded capacity. Accordingly, to time of the evacuation advisory was alleviated, and eliminate the congestion, a strategy that spatially the number of people requiring emergency rescue and temporarily disperses evacuees is conceivable. was reduced to approximately 9,000 people, with To develop spatial dispersion strategies, first a sce- fewer experiencing injury compared to D2. Howev- nario was developed in which some of the roads er, the evacuation time of ordinary households head- were designated exclusively for pedestrian use after ing to regional disaster-prevention centers was not an evacuation advisory was issued (D1, see Fig.11 significantly improved. When the strategies of D1 to for the targeted roads), and a strategy was estab- D3 were implemented together, the number of peo- lished in which the evacuation destinations were ple requiring emergency rescue was reduced to zero. decentralized by selecting evacuation destinations, including shelters for vulnerable households (6) Summary of scenario analysis results (households with preschool-age children or elderly In this section, we summarize the issues and family members, approximately 24% of all house- strategies for reducing the number of flooding vic- holds) and households with family members who tims in large cities, as identified from the scenario would have difficulty in evacuating (D2), were es- analyses implemented in this study. tablished. With D1, it was assumed that the roads Improving the awareness of residents: The designated exclusively for pedestrian use were well awareness among residents regarding flooding dis- known, and in the search for pedestrian courses, the asters is low, and with the current awareness levels, roads were evaluated at half the normal cost. Next, many residents would engage in risky behavior in as a temporal dispersion strategy, a case in which the event of a flooding disaster, such as evacuating evacuation of vulnerable households was completed when their own homes are flooded. To reduce the two hours prior to the evacuation advisory (D3) was number of victims, civil authorities must understand established with the assumption that the prior evac- the flooding risks of cities and improve awareness to uation of vulnerable households was promoted. promote appropriate responses. Therefore, we must The flooded population and the average evacua- also promote disaster prevention education during tion time of pedestrian evacuees for conditions D1 normal periods. to D3 based on strategies B2 and C is summarized Reducing evacuation demand: Large-scale in Fig.12. From the figure, if we examine the results evacuations of cities result in problems such as con- for strategy D1, it is clear that the evacuation time gestion of evacuees and evacuation facilities ex- was reduced by approximately 40 minutes compared ceeding their capacities. Therefore, these evacua- to strategy C, which required approximately 2 hours tions also carry the risk of increasing the number of for evacuation; congestion was also alleviated. Fur- victims. Accordingly, measures to reduce the de- ther, a total of approximately 30,000 people requir- mand for evacuation are required, such as recom- ing emergency rescue was significantly reduced to mending that residents of high-rise buildings wait in approximately 1,000 people. Next, with strategy D2, their own homes. However, in zero-meter areas, because vulnerable households could evacuate to flood waters could remain for an extended period of

221 time, and therefore, secondary evacuation plans for 6. CONCLUSION use after human life has been secured must also be examined and implemented. This includes plans to In this study, our objective was to examine ap- promote reserve supplies in each home and rescue propriate evacuation strategies for large-scale flood plans in the event of a flood. damage in a large city. To achieve this, we devel- Dispersion of evacuees: If evacuation is initiated oped a simulator to model the spatial characteristics once an evacuation advisory is issued, the concern is of the region, including its topography and evacua- that congestion of evacuees could become serious. tion facilities, as well as the current status of the Therefore, an intrinsic issue is advancing with the residents of an area. This simulator was used to spe- temporal dispersion of evacuees by promoting pro- cifically model large-scale evacuations based on the active evacuation from an early stage. Achieving behavioral intentions of the residents using an entire this requires that information be aggressively pro- large city as the target. In addition, the simulator vided from an early stage using the latest technology was also capable of modeling conditions, such as for weather forecasts and water-level predictions. the disaster phenomenon, the transmission of infor- Appropriate evacuation destinations must also be mation, and the damage in the event of large-scale specified, and strategies to alleviate congestion must flooding; thus, we were able to use it as a tool to be implemented to spatially disperse evacuees. The examine evacuation measures in a large city. How- most serious concern arises when large numbers of ever, some issues remain that must be resolved to evacuees temporarily converge in small-scale evac- achieve a more realistic examination, such as con- uation facilities. Therefore, evacuation facilities that sidering the impact of inland flooding and flow can meet the evacuation demands of each region speed on the evacuation of residents. must be established. However, the establishment of In this study, we used the developed simulator to a small number of large-scale evacuation facilities conduct scenario analyses of the Edogawa Ward. comes with the risk of congestion becoming a seri- From the analyses, we could identify the evacuation ous problem, and therefore, caution must be exer- problems that are inherent in large cities, and found cised. Furthermore, if the establishment of evacua- that the promotion of stereotypical evacuation be- tion facilities is difficult, extensive evacuation that haviors temporarily creates a large number of evac- includes evacuation to higher elevations or areas uees, compounds the problems of congestion, and outside the region must also be examined, which causes evacuation facilities to exceed capacity, re- would require coordination with surrounding re- sulting in an increase in injuries. We also confirmed gions and measures to educate the residents. The that, to reduce injury in the event of a flooding dis- most effective measure for cities in low-lying areas aster in a large city, it is important that a wide range is to realize early-stage evacuation to areas outside of evacuation strategies be developed to improve the the flood zone by promoting both types of disper- awareness of the residents, reduce the evacuation sion measures. demand, and temporally and spatially disperse Transmitting information to all residents, and evacuees. supporting residents requiring relief in a disas- The results of this study are limited to indicating ter: To eliminate human suffering, it is essential to the framework for evacuation strategies in cities, support residents who do not receive evacuation and therefore, to devise practical evacuation plans information or would require relief assistance during for a specific region, examinations based on actual a disaster. It is particularly difficult for large cities conditions of a region must be implemented. Fur- to ensure that information is transmitted to every thermore, the appropriate implementation of an resident due to various problems such as the pres- evacuation plan requires the understanding of the ence of a large, diverse population, and the lack of residents in the region to be evacuated, and therefore, cohesiveness in communities. Conceivable activities to raise awareness are extremely important. measures to resolve this problem include diversify- ing the media sources used to transmit information ACKNOWLEDGEMENTS: We thank the cooper- and fostering independent efforts to obtain disaster- ation of the Edogawa Ward in the implementation of related information. Moreover, support for residents this study. Mr. Kyohei Hosoi of IDA Co., Ltd., co- requiring evacuation assistance is a basic problem, operated in the construction of the river model and which does not depend on regional characteristics. floodplain model, for which we are grateful. In cities with large populations, it is difficult for the government to be the sole source of support during REFERENCES disasters, and therefore, there is a demand to devel- 1) Central Disaster Prevention Council, the Expert Panel on Large-Scale Flood Disaster Countermeasures: Expert Panel op support organizations by engaging residents in Report on Large-Scale Flood Damage Measures, Flooding the region. in Tokyo - Measures that Should Be Taken To Reduce

222 Damage -, 2010. and disaster education, Doboku Gakkai Ronbunshuu D, 2) Central Disaster Management Council, the Expert Panel on Vol. 62, No. 3, pp. 250-261, 2006. Evacuating in a Disaster: Summary of Local Hearing Sur- 14) Kuwasawa, N., Katada, T., Oikawa, Y. and Kodama, M.: vey Regarding Typhoon No. 15 of 2011, 2012. Development of comprehensive scenario simulator in flood 3) Hori, M., Inukai, Y., Oguni, K. and Ichimura, T.: Study on time and its application to disaster education, Doboku developing simulation method for prediction of evacuation Gakkai Ronbunshuu D, Vol. 64, No. 3, pp. 354-366, 2008. processes after earthquake, Sociotechnica, Vol. 3, pp. 138- 15) Japan Society of Civil Engineers: Hydraulic Tallies [Fiscal 145, 2005. Year 1999 Edition], 1999. 4) Iida, S., Tachi, K., Takedomi, K., Kawamoto, K., Kaneki, 16) Ministry of Land, Infrastructure, Transport and Tourism, M., Hirakawa, R. and Tanioka, T.: Development of evac- Hokuriku Regional Development Bureau: Guidelines for uation analysis system for flood events and verification of Examining Flood Zones Along Torrential Rivers, 2003. applicability to emergency management, Advances in River 17) Katada, T., Oikawa, Y. and Tanaka, T.: Development of Engineering, Vol. 8, pp. 139-144, 2002. simulation model for evaluating the efficiency of disaster 5) Muraki, Y. and Kanoh, H.: Multiagent model for wide-area information dissemination, Journal of the Japan Society of disaster-evacuation simulations with local factors consid- Civil Engineers, No. 625/IV-44, pp. 1-13, 1999. ered, Journal of the Japanese Society for Artificial Intelli- 18) Takahashi, K. and Takahashi, Y.: Car Society and Flood gence, Vol. 22, No. 4F, pp. 416-424, 2007. Damage, Kyushu University Press, 1987. 6) Yukawa, S., Hatayama, M. and Tatano, H.: Observations 19) Greenshields, B.: A study of traffic capacity, Proc. of on simulation assessment of evacuation plans that assume Highway Research Board, Vol. 14, pp. 448-494, 1934. large-scale flood damage, Proceedings of the 73rd National 20) Central Disaster Management Council: 10th Expert Panel Convention of the Information Processing Society of Ja- on Evacuation Strategies for Inland Earthquakes Around pan, pp. 4.655-4.656, 2011. Tokyo, 2008. 7) Simonovic, S. P. and Ahmad, S.: Computer-based model 21) Foundation of River and Basin Integrated Communica- for flood evacuation emergency planning, Natural Haz- tions, Japan: Flooding Analysis and Residential Evacuation ards, Vol. 34, No. 1, pp. 25-51, 2005. Plan Including Embankment Collapse Strategies, 1993. 8) Kolen, B., Kok, M., Helsloot, I. and Maaskant, B.: Evacu- 22) Suetsugi, T.: Research to Improve the Accuracy and Appli- Aid: A probabilistic model to determine the expected loss cation of Flooding Analysis Techniques for Floodplain of life for different mass evacuation strategies during flood Management, Kyushu University Press, 1998. threats, Risk Analysis, Vol. 33, No. 7, pp. 1312-1333, 23) Suga, K., Uesaka, T., Yoshida, T., Hamaguchi, K. and 2012. Chen, Z.: Preliminary study on feasible safe evacuation in 9) El-Sergany, A. and Alam, S.: Trip distribution model for flood disaster, Annual Journal of Hydraulic Engineering, flood disaster evacuation operation, ITE Journal, Vol. 82, Japan Society of Civil Engineering, Vol. 39, pp. 879-882, pp. 42-47, 2012. 1995. 10) FLOODsite Consortium: Modeling the evacuation process 24) Disaster Social Engineering Lab, Gunma University: Field in case of flooding, FLOODsite Project Report, 2008. Survey Report on the September 2000 Tokai Torrential 11) Pel, A. J., Bliemer, M. C. J. and Hoogendoorn, S. P.: Downpour Disaster, 2001. EVAQ: A new analytical model for voluntary and manda- 25) Katada, T., Kodama, M., Asada, J., Oikawa, Y. and Ara- tory evacuation strategies on time-varying networks, Pro- hata, M.: Research on the situation dependence of deci- ceedings of IEEE Intelligent Transport Systems Confer- sion-making of inhabitants evacuation behavior in the To- ence, Beijing, China, 2008. kai Heavy Rainfall Disaster, Annual Journal of Hydraulic 12) Dawson, R., Peppe, R. and Wang, M.: An agent-based Engineering, Japan Society of Civil Engineering, Vol. 46, model for risk-based flood incident management, Natural pp. 319-324, 2002. Hazards, Vol. 59, Issue 1, pp. 167-189, 2011. 13) Katada, T. and Kuwasawa, N.: Development of tsunami (Received April 17, 2015) comprehensive scenario simulator for risk management

223