Content uploaded by Tadashi Nakasu

Author content

All content in this area was uploaded by Tadashi Nakasu on Sep 11, 2019

Content may be subject to copyright.

HUMAN VULNERABILITY INDEX FOR EVALUATING

TSUNAMI EVACUATION CAPABILITY OF

COMMUNITIES

Yozo GOTO1 and Tadashi NAKASU2

1 Member of JAEE, Senior Engineering Adviser, Kaihatsu-Toranomon Consultant Co., Ltd,

Tokyo, Japan, gotoyozo@mti.biglobe.ne.jp

2 Member of JAEE, Academic Researcher, Chulalongkom University,

Bangkok, Thailand, Tadashi.N@chula.ac.th

ABSTRACT: The rate of fatalities caused by tsunamis vary from community to

community depending on geographical and socio-psychological features peculiar to each.

If the relationship between fatalities rate and geographical and socio-psychological features

can be quantitatively formulated, this can be a concrete means for evaluating a community’s

vulnerability with regard to evacuation (hereafter, evacuation vulnerability) and developing

effective measures that can reduce loss of human life. Therefore, the authors of this paper

proposed to apply a Human Vulnerability Index (HVI), defined as fatality rate divided by

rate of incidence of washed-out buildings, to evaluate the evacuation vulnerability of

municipalities. Using reliable public databases, the authors analyzed the HVIs of twenty

municipalities that were heavily damaged by the tsunami of the 2011 Great East Japan

Earthquake. Then they applied a multiple-regression analysis using the following four

factors as explanatory variables: 1) time allowance for evacuation; 2) preparedness; 3) road

serviceability; and 4) warning effect. They thus extracted a reliable formula (R=0.904),

which enabled them to quantify the effects of these factors on the HVI. Future tasks are to

generalize the formula through application to other tsunami disasters and to establish a

numerical evaluation of geographical and socio-psychological features to enable estimation

of the tsunami evacuation capability of a municipality and the effect of tsunami

countermeasures before a tsunami occurs.

Key Words: Tsunami evacuation, Human Vulnerability Index, The Great East Japan

Earthquake, Geographical feature, Socio-psychological feature

1. INTRODUCTION

1.1 Objectives of this study

Many people lost their lives because they could not effectively evacuate from the tsunami of the 2011

Great East Japan Earthquake (GEJE). People’s vulnerability to a tsunami, i.e., their lack of ability to

escape from areas that are impacted by a tsunami, must essentially be correlated with the geographical

and socio-psychological features of each area. Therefore, if it becomes possible to establish a numerical

- 1 -

Journal of Japan Association for Earthquake Engineering, Vol. 18, No. 6, 2018

model of such a correlation and then measure the vulnerability of the tsunami-impacted area, it will

advance knowledge regarding tsunami-induced human loss and make it possible to evaluate the area's

effort to improve its geographical and socio-psychological features. Moreover, generalization of the

model to make it applicable to other areas and other tsunami disasters will produce a tool for measuring

areal vulnerability to future tsunamis and enable municipalities to prioritize the order of their

countermeasures. Hence, the authors have proposed a Human Vulnerability Index (HVI), which is

defined as the fatality rate divided by the rate of incidence of buildings damaged, in order to measure

the probability of human loss of an area. Furthermore, they have analyzed the correlation between HVI

and geographical and socio-psychological features.

The objective of this paper is to verify the hypothesis that HVI is correlated with major geographical

and socio-psychological features by establishing a numerical correlation model, and to demonstrate that

factors of tsunami evacuation vulnerability can be macro-assessed by this model.

1.2 Literature review

Studies examining features of earthquakes and vulnerability of buildings using the relationship between

number of fatalities and number of collapsed buildings have a history of one hundred years. Imamura1)

used the ratio of the number of damaged buildings to the number of fatalities to denote the features of

an earthquake and presented his findings in a paper for Earthquake Prevention Study Committee Report

No.77 published in 1913. Coburn, et al.2) defined the ratio of the number of fatalities to the number of

residents in collapsed buildings as a Lethality Ratio, and analyzed the causes of deaths. Ohta, et al.3)

improved Kawasumi’s equation, D = 0.01H1.3 (D: number of fatalities, H: number of collapsed

buildings), and proposed a regression equation for D using data from 35 earthquakes from 1872 to 1978.

Miyano and Ro4), using data from 1950 and later earthquakes in which most of the human damage was

caused by collapsed buildings, proposed a relational expression among hypocentral distance, human

damage and building damage. Murakami5) analyzed Lethality Ratios of single-family houses,

multifamily houses and non-wooden multifamily houses by applying multiple regression analyses to

zone-by-zone numbers of collapsed buildings and fatalities in Ashiya due to the 1995 Great Kobe

Earthquake. Matsuda6) defined the number of destroyed houses (collapsed by shaking, burning and being

washed away by tsunami) per death as an HD value, and applied this to characterize the features of each

earthquake disaster since the Meiji era. Moroi and Takemura7) defined the D/S value as the number of

fatalities divided by the number of occupants of collapsed buildings, and applied this to analysis of the

features of damage due to the 1995 Kobe Earthquake. Takemura8) analyzed the ratios of the number of

destroyed houses (collapsed by shaking, burning, being washed away by tsunami, being buried, and so

on) divided by the number of fatalities from major earthquakes since the 1872 Hamada Earthquake, and

reported that inland near-field earthquakes caused about one fatality per ten collapsed buildings.

The above studies mainly analyzed the relationship between the shaking of earthquakes and number

of fatalities. For the study on tsunami disasters, Kuwasawa, et al.9) surveyed the reactions of people in

Owase (one of the tsunami risk-cities in Mie Prefecture) at the time of the 2004 Kii Peninsula Southeast

Offshore Earthquake. This earthquake did not cause a destructive tsunami, but the shaking prompted

some people to evacuate from the coast. Therefore, Kuwasawa, et al. focused on the consciousness of

evacuation and proposed a decision-making model for tsunami evacuation through a multiple regression

analysis using five explanatory variables: earthquake intensity, tsunami risk awareness, prejudice of

normalization, risk degree of dwellings and preparedness.

After the GEJE, Suzuki and Hayashi10) tried to derive a relationship between human damage and

tsunami hazard of the coastal area. They discussed for each local municipality (hereafter, LM) the

correlation between fatalities rate, tsunami intensity, geographical features, population exposure,

disaster awareness of people and assumed tsunami height. Koyama, et al.11), using 500-meter mesh

population data, extracted age and gender distributions of both daytime population and night-time

population in the inundated area and in the building washed-out area. They pointed out that the fatality

rate rose according to aging of population in LMs and that the missing person rate rose where building-

washed-out areas spread over most of the inundated area. Tanishita and Asada12) analyzed the fatalities

rates of 59 regions of Minami-sanriku, Miyagi Prefecture, using the tsunami inundation depth,

- 2 -

experience from 1960 Chilean Tsunami, viewability of the sea and distance to high land as explanatory

variables. They reported that the fatalities rates were high in areas that the 1960 Chilean Tsunami had

not inundated and the sea could not be seen.

Goto13) proposed a Victim Index, defined as number of fatalities divided by number of damaged

buildings, and applied this to six villages in Yamada, Iwate Prefecture. He reported that the rate of

tsunami-experienced persons in the village, cognition rate of tsunami warnings, degree of traffic jams,

rates of participation in tsunami drills, rate of people who helped others and length of evacuation routes

could be related to the Index of each village. Nakasu, et al.14) 15) proposed an HVI, defined as fatalities

rate divided by collapsed-house rate, as a longitudinal analysis tool. And they applied their HVI to the

major municipalities in the Sanriku ria coast of Japan based on their records of damage due to the 1896

Meiji Sanriku Tsunami, the 1933 Showa Sanriku Tsunami and the 2011 Great East Japan Earthquake

and Tsunami (GEJET) disaster, and they analyzed the historical shift of vulnerabilities.

On another front, the City Bureau, Ministry of Land, Infrastructure, Transport and Tourism of Japan

(MLIT) interviewed 10,603 refugees from the coastal LMs of six prefectures from Aomori to Chiba and

asked about their emergency actions16). The City Bureau analyzed the effects of geographical and social

features by grouping the LMs into four zones: urban area close to hills; farming-fishing village close to

hills; urban area in the plain; and farming-fishing village in the plain. Additionally, effects of traffic

jams, risk of human and car mixed evacuations and the effect of steep roads in each LM were analyzed

using trip data of evacuees17).

1.3 Characteristics and Meanings of this Study

This study is intended to:

(1) redefine Nakasu’s HVI by adding the rate of people at home at the time of an earthquake;

(2) measure HVIs of twenty municipalities in Iwate, Miyagi and Fukushima Prefectures numerically

using data of the FSC archive (introduced in 2.2.3 and Appendix 1);

(3) verify HVI as an index for measuring tsunami evacuation vulnerability of LMs; and

(4) present a prediction equation for HVI through a multiple regression analysis that uses four

characteristic values of geographical and socio-psychological features as explanatory variables.

Expansibility of this study is as follows. If HVI can be generalized to make it adaptable to areas

other than the GEJE region, it will be possible to foresee factors causing human damage in risk areas of

coming tsunamis, and to numerically evaluate a way to improve evacuation vulnerability.

The uniqueness of this study is the development of a numerical model of actual evacuation

vulnerabilities of LMs suffered by the large tsunami of GEJE using site-specific geographical and socio-

psychological features. The preceding study by Kuwasawa, et al.9) modeled individual decision-making

for evacuation using data of an earthquake not accompanied by a destructive tsunami. The study by

Suzuki and Hayashi10) analyzed damage caused by GEJE with respect to each LM, but only discussed

qualitatively the effects of tsunami height, exposed population and assumed scenario tsunami height.

On the other hand, the tsunami evacuation simulation using a multi-agent model can be utilized for

disaster education as it can show evacuation action through animation. However, because the effects of

geographical and socio-psychological features are included in the simulation, factors affecting

evacuation vulnerability cannot be explicitly analyzed.

1.4 Structure of this paper

The second chapter describes the analyzed areas, data and their usages. The third chapter describes the

formulation of HVI and the calculation procedure, and verifies the hypothesis that HVI, i.e., the quotient

of risk and hazard, can measure vulnerability. The fourth chapter verifies the hypothesis that HVI can

be defined by geographical and socio-psychological features through multiple regression analysis on the

relationship between HVIs and those features. It also validates HVI through sensitivity analysis on the

effects of geographical and socio-psychological features to the number of fatalities using the regression

equation of HVI. The fifth chapter discusses the availability of some additional explanatory variables

- 3 -

and the method of quantification of the geographical and socio-psychological features when HVI is

utilized as a prediction tool. The sixth chapter lists conclusions of this study.

The contents of this paper originated from the author’s papers for the 2016 JAEE annual conference

and the 16th WCEE18). However, the formulation of HVI has been revised and discussions added.

2. STUDIED AREAS AND USED DATA

2.1 Analyzed Municipalities

Fig. 1 shows LMs that experienced 78 or more fatalities (including missing persons) due to GEJE,

excluding the neighborhood of the Fukushima nuclear power plants. Of these, 20 LMs, from A to T,

were selected for this analysis. The LM labeled U in the figure suffered 188 fatalities, but was left out

of the analysis because: 1) only a small part of the boundary is facing the sea; 2) almost all of the

inundated area was covered by factories and warehouses; 3) 35% of the fatalities were daytime visitors

Drawing from digital map data of Geospatial Information Authority of Japan,

and Japan municipality boundary data of ESRI Japan

Fig. 1 Municipalities to be analyzed

U

Iwate Pref.

Miyagi Pref.

Fukushima Pref.

50km

A

B

C

D

E

F

G

H

I

J

K

L

M

V

N

O

W P

Q

R

X

S

T

Tokyo

Sendai

- 4 -

from other LMs; and 4) the rate of damaged residential houses was low: 4.7% (the rate for the 20 LMs

was 70.1% on average). Additionally, LMs V and W were not included in the target because of the large

deviation in age of the limited samples. And LM X was left out because data of people who had

undertaken tsunami risk preparedness was estimated unnaturally and was extremely small (these

numerical values can be found in footnotes 2) and 3) of Appendix 2).

Another approach would have been to split each LM into areas, such as beach-by-beach or bay-by-

bay, so that features of the areas could be analyzed more clearly. However, such splitting would

unavoidably reduce the number of available data to less than thirty, and there would have been a loss of

significance of the statistical analysis. Therefore, this study did not employ such splitting of LMs.

2.2 Data

2.2.1 Number of fatalities

To obtain the number of fatalities, data from the local governments of Iwate19), Miyagi20) and

Fukushima21) Prefectures were utilized. The number of fatalities is defined as the number of dead

persons found in the LM area not including indirect deaths, and missing persons counted by the LM.

Dead visitors from other LMs were included in the number of fatalities because, in order to analyze

evacuation vulnerability of an LM in the daytime, the number of fatalities including visitors should be

used. 16 LMs published the number of fatalities of their own citizens. The numbers only differed by

1.9% in total from the number of fatalities defined above and the standard deviation of the difference

was 7.8%.

2.2.2 Population and number of houses

The 2010 Census was used to obtain for populations and numbers of houses. As the census does not

include the number of houses of LMs I, J, K and R, these numbers were estimated by the method denoted

in footnote*3 of Table 2.

2.2.3 Number of damaged houses and inundation depth

Data of the Fukkou-Shien-Chousa archive22) (hereafter, the FSC archive) was used to obtain the number

of damaged houses and inundation depth data. The FSC archive, outlined in Appendix 1, is a GIS

database compiling data from interview surveys conducted by the City Bureau of MLIT.

15) All the

damaged house statistics including inundation depth are in one folder for each LM of the archive. In this

study, the number of houses that were washed out by the tsunami was used as the number of damaged

houses. For the inundation depth, the average of inundation depths evaluated at the locations of washed-

out houses was applied.

There might be some biased counting in the data having fuzzy meaning because many people had

to participate to compile the data for different prefectures and there are many LMs in them. Therefore,

cumulative addition curves of numbers of washed-out houses were calculated for each LM taking

building area as a parameter. The curves were averaged with respect to three prefectures: Iwate (six

Fig. 3 Cumulative addition curves of washed-

out houses vs. building area > 30m2

Building area (

m

2)

Miyagi

Fukushima

Iwate

Rate of cumulative addition

(number in area 500m2 as 1)

Fig. 2 Cumulative addition curves of washed-

out houses vs. building area

Building area (

m

2)

Miyagi Fukushima

Iwate

Rate of cumulative addition

(number in area 500m2 as 1)

- 5 -

LMs), Miyagi (eleven LMs) and Fukushima (three LMs). Then, the three cumulative additional curves

were drawn, as seen in Fig. 2, taking building area as the horizontal axis and rate of cumulative addition

as the vertical axis. Non-negligible discrepancy is seen among the three curves. The counting rules

concerning attached small houses, such as barns and garages, seemed to differ among the three

prefectures. Therefore, houses of less than 30m2 in building area were omitted from the cumulative

addition and the three curves were re-drawn as shown on Fig. 3. By omitting small houses, the curves

of Iwate and Miyagi are closely aligned. Although the curve of Fukushima does not align well, this

study decided to use the number of washed-out houses greater than 30m2 in the building area, taking

into account the small number of target LMs in Fukushima.

2.2.4 Sufferers for analysis, rate of people at home and evacuation trip data

Attributes of people who suffered and their evacuation trips were retrieved from the individual

evacuation method folder of the FSC archive. The archive compiled individual attributes, evacuation

trips and answers to questionnaires covering 10,603 sufferers in the coastal LMs of six prefectures, from

Aomori to Chiba. However, in order to focus on the behavior of the persons who might die if they would

not evacuate, this study selected the data of people who were at home at the time of the earthquake or

returned home before the tsunami arrived, and concurrently whose houses had completely collapsed due

to the tsunami. Hereafter, this study denotes these persons as sufferers for analysis.

The rate of people at home was defined as the number of sufferers for analysis divided by the total

number of people whose houses were completely destroyed. The values are listed in Table 2 and are

consistent with the values of a preceding study23). It should be noted here that the individual attribute

data of the FSC archive defined “completely collapsed houses” to include “washed-out houses”.

Evacuation routes and elapsed time for this study were extracted from the trip data of the FSC

archive. However, as this was based on interview surveys and not on instrumental observation, it should

be recognized that the accuracy is somewhat limited.

2.2.5 Data correction with age

Distribution of age data on individuals in the FSC archive deviated from that of the population.

Therefore, age distribution was compared with the census of small segments24) which had an average of

137 segments per LM, and correction coefficients with age were defined extracting a group of small

segments which covered tsunami inundation area.

Then, the data on individuals in the FSC archive were weighted by the correction coefficient so that

the age data distribution came close to that of the census when summing up the data for each LM (for

details, see Goto’s past paper25)). Table 1 shows the average and standard deviation of the correction

coefficients, which were divided into three age ranges: 20-49, 50-69 and over 70. It might have been

desirable to divide them into six ranges by adding a sex range. However, such detailed division was not

applied because there were too few data.

2.2.6 Discussion on use of survivors' data

While it is more favorable to use data including fatalities for the analysis of evacuation vulnerability, it

was impossible to get such data with the same accuracy as for survivors. Therefore, this study used only

the data of survivors, as mentioned in 2.2.4.

The ratio of fatalities to sufferers for analysis (defined at 2.2.4), however, needs to be checked. The

total number of sufferers for analysis was estimated as 127,000. This was based on the rate of people at

home (75%) and the total number of completely destroyed houses in the object region (70,800 houses)

with 2.4 people living in one house (2.66 persons per a household in Iwate, Miyagi and Fukushima

according to the 2010 Census, with vacant houses estimated to be 10%). The total number of fatalities

Table 1 Average and standard deviation of correction coefficients with age of twenty LMs

Age 20~49 50~69 Over 70

Average 1.209 0.9 1.044

Standard deviation 0.357 0.137 0.185

- 6 -

in the subject region was 16,900. Mikami26) reported that 80% of the fatalities were estimated to have

been at home or in the process of evacuation. Therefore, the number of fatalities among sufferers for

analysis was 13,000, that is, 80% of 16,900. Consequently, the ratio of fatalities to the sufferers for

analysis is estimated at around 10%.

If the data concerning fatalities could be added, some aspects of evacuation vulnerability might be

emphasized; for example, evacuation roads could be longer, the rate of preparation of emergency carry-

out bags would be lower, and so on. Consequently, analysis without data of fatalities might lack

sharpness. However, it is not such a simple on-off phenomenon such that evacuation roads longer than

a certain length automatically result in evacuation failure. Success or failure of tsunami evacuation

depends on many widely dispersed factors. As the data of fatalities were 10% of that of sufferers for

analysis, the authors considered the analysis that lacked data of fatalities to not cause a fatal error.

This study might be misunderstood as having developed an evaluation method of the number of

fatalities by using survivor data. However, the main subjects of this study are a regression analysis of

HVI and its validation. Although evaluation of the number of fatalities is introduced in Chapter 5 of this

paper, it is aimed at validating the availability of the regression HVI. Therefore, the validity of using

data from survivors in the regression analysis for HVI should be discussed, and this study considered it

possible, as mentioned above.

3. FORMULATION OF HVI

3.1 Definition of HVI

This study defined HVI as an index for measuring the evacuation vulnerability of people in an area. It

is hypothesized that the number of fatalities (= risk) could be computed by multiplying the exposed

population by HVI (= vulnerability) and by inundation depth of a tsunami (= hazard), as described by

Eq. (1). For simplification, hazard is expressed by inundation depth, although the flow velocity may

affect it.

Number of fatalities Exposed population × HVI × Inundation depth (1)

HVI is formulated by Eq. (2).

Number of fatalities caused by a tsunami 1

Population in area Rate of people at home

HVI = × 100 (2)

Number of washed-out houses

Number of residential houses in area

Both the number of fatalities and the number of washed-out houses monotonically increase with

increased depth of tsunami inundation. However, taking their ratio, HVI becomes independent of

inundation depth, as shown by Eq. (2). In addition, the denominator and numerator of Eq. (2) are divided

by the number of residential houses and the population in the object area, respectively, enabling HVI to

be non-dimensional. Residential houses and population in a common area have to be counted. Therefore,

this study used each municipality as the common area, considering data accessibility and versatility.

Exceptionally, Ishinomaki was divided into rias coast area and flatland area, and Sendai was divided

into administrative wards because of its broadness.

The definition of completely collapsed houses includes washed-out houses. It might have been worth

considering applying the number of completely collapsed houses instead of the number of washed-out

houses in Eq. (2) in order to fit the definition of "sufferers for analysis" mentioned in subsection 2.2.4.

However, as the "number of washed-out houses" of Eq. (2) is used only to evaluate hazard, it is not

necessary to use the same definition as in 2.2.4. Therefore, this study used the number of washed-out

houses for the reason mentioned in section 3.3.

Multiplication of 100 in Eq. (2) is for improved readability of the HVI value.

×

- 7 -

3.2 HVI of each LM (Local Municipality)

The HVIs of LMs from A to T of Fig. 1 are listed in Table 2, together with their calculation parameters.

The calculated HVIs are scattered from 3 to 33. Chapters 4 and 5 will verify that this scattering means

the difference among evacuation vulnerabilities of the LMs.

3.3 Verification of independence of HVI of inundation depth

HVI and inundation depth must be independent of each other in order to verify the hypothesis that HVI

represents vulnerability. The right end column of Table 2 lists the average inundation depths evaluated

at the locations of washed-out houses in each LM. Fig. 4 is a plot of the HVI of each LM with the

averaged inundation depth on the horizontal axis. The dotted line in the figure is the linear regression

Table 2 Parameters and HVI

LM Fatalities

*1 Population*2 Residential

houses*3

Rate of people

at home*4

Washed-out

houses*5 HVI Inundation

depth*6

A 514 59,430 25,010 0.718 1,453 20.7 5.48 (m)

B 752 18,617 7,950 0.729 1,990 22.1 5.40

C 1229 15,276 6,130 0.658 2,989 25.1 7.98

D 1040 39,574 18,420 0.724 2,303 29.0 7.45

E 419 40,737 16,580 0.698 2,397 10.2 7.06

F 1,763 23,300 8,550 0.803 4,210 19.1 11.04

G 1,326 73,489 25,670 0.741 5,817 10.7 7.23

H 812 17,429 5,540 0.752 3,836 8.9 10.71

I 850 10,051 3,450 0.762 2,268 16.9 13.41

J 1,106 23,611 8,105 0.881 3,974 10.8 8.19

K 2,597 137,215 56,765 0.823 4,163 31.4 4.54

L 1,086 42,903 15,450 0.835 2,862 16.4 3.80

M 78 20,416 6,650 0.895 829 3.4 4.54

N 345 132,306 70,640 0.561 1,407 23.3 4.28

O 950 73,134 25,820 0.623 1,888 28.5 4.68

P 270 34,845 11,520 0.819 1,089 10.0 3.37

Q 698 16,704 5,310 0.756 2,061 14.2 5.99

R 99 8,224 3,068 0.827 440 10.1 7.58

S 636 70,878 25,050 0.631 1,083 32.9 5.28

T 330 342,249 147,740 0.733 914 21.3 3.31

*1 Number of dead and missing persons, except related deaths, found in each LM. Iwate, Miyagi and

Fukushima Prefectures compiled and published them on the web.

*2 Extracted from 2010 Census data.

*3 Extracted from 2010 Census data; however, the numbers for I and R were not published and for J

and K, only sum of J + K was published. Therefore, the numbers for I, R and K were evaluated

using the rate of number of residential houses to that of population of neighboring LMs. J was

evaluated by subtracting K from the sum.

*4 Rate of people who were in their houses at the time of the earthquake or returned to their houses

before the tsunami, and whose houses completely collapsed. Extracted from the FSC archive.

*5 Number of washed-out houses with building areas greater than 30m2. Extracted from the FSC

archive.

*6 Average of inundation depths evaluated at the locations of washed-out house. Extracted from the

FSC archive.

- 8 -

line, which indicates a somewhat decreasing HVI with increased inundation depth. The correlation

coefficient of HVI and the average inundation depth was −0.184. However, as the significance level p

was 0.437, correlation between HVI and inundation depth was denied from a statistical point of view.

As an alternative, other HVIs were calculated using the number of completely collapsed houses

including washed-out houses, instead of the number of washed-out houses only, and analyzed

correlation to the average inundation depths evaluated at locations of completely collapsed houses. The

correlation coefficient was 0.123 and p was 0.605. Therefore, the independence of HVI would be

somewhat improved if the number of completely collapsed houses were used. However, as Takemura8)

pointed out, the definition of completely collapsed house has altered historically and the recent

administrative definition seemed to depart from the structural definition to some extent. In addition,

considering utilization of overseas data, such a clear definition as washed-out house should be applied.

4. MULTIPLE REGRESSION ANALYSIS ON HVI AND VALIDATION

4.1 Explanatory variables

Many multiple regression analyses on HVIs were executed, taking many different combinations of

geographical and socio-psychological features as explanatory variables, as shown in Appendix 2. The

following four explanatory variables were extracted as the best combination. The clarity of geographical

and socio-psychological meanings of the variables were emphasized in the extraction:

Allowance period A: Tsunami arrival time / Distance to a safe place

Preparedness P: Rate of people who had prepared emergency carry-out bags

Road serviceability R: Rate of car-using evacuees × Car speed

Warning effect We: Tsunami warning height × Cognition rate

4.1.1 Allowance period A

Allowance period A is defined as tsunami arrival time divided by distance to a safe place. Tsunami

arrival time means the time for the tsunami to arrive at each LM on the coast after the earthquake and is

evaluated as 35 minutes for the rias coast area and 53 - 65 minutes for the flatland area, as shown in

Table 3, by referring to previous studies17), 23). Distance to a safe place is the average moving distance of

persons in an LM who evacuated their completely collapsed houses, without detour, to high land or

inland non-inundation area or vertical evacuation facilities. The individual moving distance was

calculated from the individual's evacuation trip data of the FSC archive, using the same method as Goto’s

preceding study25).

When analyzing the average moving distances, it should be noted that in many LMs the number of

car-using evacuees was roughly equal to that of walking evacuees. Therefore, the following equivalent

Fig. 4 HVI vs. Average inundation depth

HVI

HVI = 22.0 - 0.569D

p value: 0.437

Ave ra

g

e inundation de

p

th

(

m

)

- 9 -

evacuation distance, which converted the evacuation distance of a car to that of a pedestrian, was used

(Table 3).

ed(i) = W(i) × rw(i) + D(i) × rc(i) × 0.244 (3)

where, ed(i): Equivalent evacuation distance of LM i

W(i): Average walking evacuation distances of LM i (average of the distances from house

to a safe place)

rw(i): Rate of walking evacuees of LM i

D(i): Average car-driving-evacuation distances of LM i (average of the distances from

their house to a safe place)

rc(i): Rate of car-using evacuees of LM i

0.244 = ΣW(i) / ΣD(i): Conversion factor of car-driving-evacuation distance into walking

evacuation distance (average of ratios of all target LMs)

Fig. 5 shows the correlation between HVI and allowance period A. The plots are scattered considerably,

but HVI tended to increase when the allowance period decreased.

4.1.2 Preparedness P

Preparedness P is defined as the rate of sufferers for analysis who had prepared emergency carry-out

bags beforehand, and was extracted from personal interview data of the FSC archive (refer Table 3). The

Table 3 HVI and element parameters for explanatory variables

LM HVI

Tsunami

arrival

time

Equivalent

evacuation

distance

Car

velocity

Rate of car-

using-

evacuees

Number o f

data ed, rc

*1

Tsunami

warning

height

Cognition

rate of

warning

Number o f

data cr

*2

t (minute) ed (m) v (km/h) rc h(m) cr

A 20.7 35 184 8.6 0.504 114 3 0.513 184

B 22.1 35 157 6.4 0.439 35*3 3 0.366 167

C 25.1 35 247 6.4 0.453 90 3 0.368 151

D 29.0 35 175 6.1 0.253 134 3 0.505 209

E 10.2 35 126 8.4 0.586 181 3 0.422 257

F 19.1 35 166 11.1 0.445 145 3 0.566 308

G 10.7 35 247 7.1 0.503 357 6 0.620 490

H 8.9 35 186 7.6 0.551 179 6 0.702 275

I 16.9 35 159 6.2 0.387 94 6 0.589 120

J 10.8 35 135 8.8 0.465 156 6 0.563 236

K 31.4 60 433 6.9 0.487 460 6 0.576 789

L 16.4 60 478 7.7 0.516 68 6 0.430 151

M 3.4 65 166 9.5 0.736 42 6 0.675 57

N

23.3 65 510 13 0.956 31 6 0.501 57

O 28.5 65 501 8.9 0.601 88 6 0.535 155

P 10.0 65 551 18.5 0.703 73 6 0.644 149

Q 14.2 60 425 17.9 0.922 96 6 0.437 134

R 10.1 65 293 13.2 0.721 43 3 0.520 74

S 32.9 65 348 14.1 0.925 97 3 0.301 193

T 21.3 53 190 8.1 0.552 78 3 0.320 110

*1

N

umber of evacuees who evacuated their houses to a safe place without detou

r

, and whose houses completely

collapsed.

*2 Number of sufferers-for-analysis, namely, people who were at their houses at the time of the earthquake or

returned to their houses before the tsunami, and whose houses completely collapsed.

*3 A part of the data is missing.

- 10 -

Table 4 Explanatory variables and Regressed HVI

LM HVI

Allowance

period Preparedness Road

serviceability

Warning

effect Number of

data

*1

Regression

HVI

*2

A

= t / ed

(

minute/m

)

P R = v × rc

(

km/h

)

We = h × c

r

(

m

)

A 20.7 0.190 0.498 4.34 1.539 184 14.9

B 22.1 0.222 0.406 2.81 1.098 167 20.7

C 25.1 0.142 0.429 2.90 1.104 151 32.1

D 29.0 0.200 0.342 1.54 1.515 209 30.1

E 10.2 0.278 0.480 4.92 1.266 257 10.3

F 19.1 0.211 0.369 4.94 1.698 308 15.5

G 10.7 0.141 0.403 3.57 3.720 490 18.2

H 8.9 0.188 0.469 4.19 4.212 275 10.3

I 16.9 0.220 0.376 2.40 3.534 120 14.2

J 10.8 0.259 0.361 4.09 3.378 236 10.0

K 31.4 0.139 0.418 3.36 3.456 789 19.1

L 16.4 0.126 0.434 3.97 2.580 151 21.9

M 3.4 0.392 0.410 6.99 4.050 57 4.2

N

23.3 0.127 0.272 12.43 3.006 57 18.5

O 28.5 0.130 0.289 5.35 3.210 155 23.9

P 10.0 0.118 0.412 13.01 3.864 149 12.4

Q 14.2 0.141 0.351 16.51 2.622 134 12.5

R 10.1 0.221 0.414 9.52 1.560 74 10.4

S 32.9 0.187 0.133 13.04 0.903 193 37.0

T 21.3 0.279 0.225 4.47 0.960 110 23.1

*1 Number of evacuees who evacuated directly to a safe place without detour and whose

houses completely collapsed.

*2 Regression HVI is discussed in section 4.2.

correlation between HVI and preparedness P is shown in Fig. 6. To prepare an emergency carry-out bag

beforehand is evidence of consciousness that emergency evacuation might be needed. Considerable

correlation was observed as the significance level p was 0.019.

The rates of beforehand-executing of "talking about evacuation method, communication tool,

designated place, and so on among family", "checking tsunami hazard map" and "participating in

tsunami evacuation drill arranged by the community" were also extracted and were analyzed to

determine their correlations with HVI. However, the p values were 0.545, 0.474 and 0.286, respectively,

and no significant correlations with HVI were seen.

4.1.3 Road serviceability R

Ideally, road serviceability R should be defined by the road traffic capacity, namely, multiplication of

the available number of cars and their velocity, in each LM area. However, such data could not be

obtained easily. Therefore, data on the actual performance of the tsunami evacuation was used. The

better the R, the greater the actual number of car-using evacuees and the higher the car velocity are

assumed to be. The car velocity v and the rate of car-using evacuees rc were extracted from the

evacuation trip data of the FSC archive, and their multiplication is applied as R (Table 3).

Fig. 7 shows the relationship between R and HVI. As there are outliers N, Q and S and significance

level p is 0.653, no correlation is seen. However, if the outliers are skipped, p becomes 0.039 and

correlation is improved, as shown by the red dotted line. As LMs N, Q and S are flatland and low

population areas, cars could be driven at their natural velocity. Therefore, the HVIs of these LMs must

have been affected strongly by other factors. For example, in LM S, preparedness P is 13% (the average

of 20 LMs was 37%), and the evacuation rate is 54% (the average was 77%). Hence, many people in

LM S could have lacked wariness over tsunamis, and lost their lives without attempting evacuation.

- 11 -

4.1.4 Warning effect We

The first announcement of a large-tsunami warning forecast tsunami heights as three meters for the

Iwate and Fukushima coasts and six meters for the Miyagi coast. Multiplication of announced tsunami

height h and cognition rate of warning cr is defined as Warning effect We. Fig. 8 shows the relationship

between We and HVI. Although a certain level of fluctuation is seen, such correlation as significance

level p being 0.045 is confirmed.

4.2 Correlation among explanatory variables, and comprehensive influence of geographical features

Correlation coefficients of the four explanatory variables are listed in Table 5. While it is desirable for

the correlation coefficients to be low, preparedness P was to some extent related to road serviceability

R and warning effect We . Correlation between P and the five element variables v, rc, h, cr and ed were

analyzed and the results are listed in Table 6. P indicates a negative correlation to the rate of car-using

evacuees rc. Therefore, many of the people who prepared emergency carry-out bags beforehand seemed

to have intended to evacuate on foot. P indicates a positive correlation to cognition rate of warning cr.

Persons who prepared for evacuation were maintaining a wariness over tsunamis and attention to

tsunami warnings. When the sensitivity analysis on the explanatory variables is conducted, these

correlations should be considered.

On the other hand, geographical features could have a strong influence. Therefore, a dummy variable

that defined the rias coast region (north of Oshika peninsula) as 0 and the flat coast region (southwest

of Oshika peninsula) as 1, was introduced and its correlation with the element parameters (Table 3) and

the explanatory variables (Table 4) was analyzed.

The results are shown in Table 7. Geographical features are distinctly correlated to tsunami arrival

time t and evacuation distance ed, but does not affect allowance period A because the correlations are

compensated for through calculation of the ratio between t and ed. Correlation to the rate of car-using

evacuees rc is clear. The number of car-using evacuees rc might have increased because the evacuation

Fig. 5 HVI vs. Allowance period A

Allowance period

A

(min./

m

)

H

V

I

HVI = 28.4 - 51.7

A

p value: 0.062

HVI = 36.7 - 49.2

P

p value: 0.019

Fig. 6 HVI vs. Preparedness

P

HVI

Preparedness P

HVI = 24.55 - 1.487

R

p value : 0.039

Fig. 7 HVI vs. Road serviceability

R

Road serviceability R (km/h)

HVI HVI = 19.58 - 0.211

R

p value : 0.653

Fig. 8 HVI vs. Warning effec

t

We

HVI = 26.29 - 3.258

I

e

p

value: 0.045

Warning effect We (m)

HVI

- 12 -

distance was long and many roads were easy to drive on in the flat region. However, in conclusion, the

geographical features do not correlate with HVI, as p is 0.649 and the correlation coefficient is 0.109.

The reasons are inferred as follows. LMs M, R and T are located in the flat coast region, but hills are

near the coast, which shortens evacuation distances. While, in LMs K and O, HVIs are pushed up due

to traffic jams lowering car velocities.

4.3 Formulation of regression equation and result of multiple regression analysis

A regression equation is formulated as shown Eq. (4). To prevent HVI from being negative, a monomial

of exponential terms is applied.

HVI = eα × Aβ × Pγ × RΔ × Weε (4)

Taking the natural logarithm of both sides of Eq. (4), Eq. (5) is obtained.

ln(HVI) = α + β×ln(A) + γ×ln(P) + Δ×ln(R) + ε×ln(We) (5)

Applying the explanatory variables listed in Table 4, regression coefficients α, β, γ, Δ and ε for Eq.

(5) were evaluated by linear multiple regression analysis (IBM SPSS Statistics was used), and the

following regression equation for HVI was obtained.

Regression HVI = 2.886 × A-1.117 × P-0.852 × R-0.423 × We-0.441 (6)

The evaluated regression coefficients, their significance probabilities p and standardized coefficients are

listed in Table 8. As each p value is less than 0.02, good fit of the analysis is confirmed. The standardized

coefficient of β was maximum and that of Δ was second. This means that allowance period A is most

effective on ln(HVI) and road serviceability R is second. VIF, Variance Inflation Factor, is the value that

increases if correlation between the explanatory variables becomes higher, and for a VIF of more than

10, the multiple regression analysis becomes unstable because of multicollinearity. As the VIF of each

explanatory variable is less than 10, the solution of the multiple regression analysis is confirmed to be

stable27).

Table 9 lists evaluation values of the regression equation, with a multiple correlation coefficient R

of 0.904 and an adjusted determination coefficient R2 of 0.768 was achieved. The values of the

regression HVIs are calculated from the Eq. (6) and are listed in the right end column of Table 4.

Additionally, the plot of the relationship between HVI and the regression HVI is shown in Fig. 9. A

trend of similarity is recognized from the plots, even though the standard error is 4.66 and the deviation

from the one-to-one line is 12 at maximum.

Table 5 Correlation factors between explanatory

variables

A P R

We

Allowance period : A 1 0.040 -0.214 -0.105

Preparedness : P 0.040 1 -0.380 0.238

Road Serviceability : R -0.214 -0.380 1 0.058

Warning effect : We -0.105 0.238 0.058 1

Table 6 Correlation facto

r

s of preparedness

P to element variables

P

Car velocity : v -0.287

Rate of car-using-evacuees : rc -0.415

Height of warned tsunami : h 0.087

Cognition rate of warning : cr 0.454

Evacuation distance : ed 0.250

Table 7 Correlation factors between geographical aspect vs. element and explanatory variables

t

ed A=t / ed

P

v

rc

R=v×rc

h

cr We=h×cr HVI

Rias or flat 0.981 0.744 －0.143 －0.447 0.563 0.690 0.638 0.302 －0.123 0.138 0.109

- 13 -

4.4 Validation of HVI through sensitivity analysis

In order to estimate the number of fatalities using the regression HVI, Eq. (2) is deformed to Eq. (7).

The regression HVI of Table 4 is applied to Eq. (7), and the results are listed in Table 10. The deviation

of the estimated number from the actual number is 0.5%.

Risk Exposed population

Number of fatalities = Population × Rate of people at home

× Regression HVI × (7)

Vulnerability Hazard

In addition, one of the four explanatory variables was multiplied by the rate of increase of standard

deviation 1 σ, and regression HVI was calculated keeping the other three variables at their original

values. Then, the number of fatalities was evaluated from Eq. (7), and is listed in rows (1) - (4) of Table

11. If the same cost were required to increase each explanatory variable by the rate of 1 σ, increase of

allowance period A would be the most effective, as already presumed through the comparison of the

standardized coefficients of Table 8. If all explanatory variables were increased by the rate of 1 σ, the

number of fatalities would decrease by 41%, as shown in row (5) of Table 11.

Table 8 Regression coefficients and evaluation values

α β γ Δ ε

Regression coefficient 1.060 －1.117 －0.852 －0.423 －0.441

Significance probability p 0.017 0.000 0.003 0.002 0.004

Standardized coefficient ----- －0.655 －0.465 －0.477 －0.424

VIF 1.102 1.387 1.252 1.287

Table 9 Evaluation value of

r

egression equation

Correlation coefficient R 0.904

Determination coefficient R2 0.817

Adjusted R2 0.768

Standard deviation σ 0.270

Significance probability p 0.000

N

umber of washed-out houses 1

Number of residential houses 100

×

Fig. 9 HVI vs. Regression HVI

Regression HVI

HVI

- 14 -

As shown in Table 7, a positive correlation exists between preparedness P and warning cognition

rate cr, which is the element variable of Warning effect We. Therefore, the number of deaths should be

calculated when P and cr are increased together. The result is shown in row (6) of Table 11 and proved

that the increase of P + cr is as effective as the increase of allowance period A.

Negative correlation existed between preparedness P and the rate of car-using evacuees rc, which

was the element variable of road serviceability R. Therefore, the case in which P and cr are increased

and rc is decreased by the rate of 1 σ was analyzed. The calculated result is shown in row (7) of Table

11. The decrease of rc has the effect of compensating the increases of P and cr.

For the actual evacuation during the GEJET, P and rc in the LMs studied were in negative correlation.

However, if improvements of roads for evacuation, optimization of car usage and sophistication of

disaster education were simultaneously implemented, P and rc would not be in negative correlation and

their synergistic effect would be expected to develop.

5. DISCUSSION

5.1 Selection of explanatory variables and their combination

The hypotheses that HVI is independent of inundation depth and can be an index for evaluating

evacuation vulnerability according to geographical and socio-psychological features are verified

through the analyses described in section 3.3 and chapter 4. An inevitable dispersion exists in the data

from the interview surveys. Therefore, it can be emphasized that a set of explanatory variables that

achieves such high-accuracy regression as correlation coefficient 0.904 was developed. However, this

set of explanatory variables does not comply with the sufficient condition, so additional explanatory

variables are discussed.

5.1.1 Earthquake intensity

The rates of people of the LMs who immediately thought that a tsunami would come because of the

large shaking of GEJE were 49.8% on average, with a standard deviation of 0.173. However, the

correlation coefficient between this rate and HVI was as low as −0.188, and multiple regression analysis,

including this rate as one of the explanatory variables, did not improve regression accuracy. The JMA

seismic intensities of LMs studied were from 5-upper to 6-upper, and the durations of the shaking were

similarly long. Therefore, the earthquake intensity that had the effect of prompting people to evacuate

did not differ much among the LMs and the differences among the HVIs of the LMs in this study were

estimated to be determined by factors other than earthquake intensity.

However, if HVI is applied to a tsunami disaster that is accompanied by an earthquake of different

intensity, the effect of earthquake intensity should obviously be added to the multiple regression analysis

as one of the explanatory variables.

Table 11 Sensitivity analysis on explanatory variables

Modified explanatory variable Number of fatalities

(1) Allowance period A +1σ 12,080

(2) Road serviceability R +1σ 13,540

(3) Preparedness P +1σ 14,090

(4) Warning effect We +1σ 14,220

(5) All explanatory variables +1σ 6,890

(6) P +1σ and warning cognition rate cr +1σ 12,900

(7) P +1σ, warning cognition rate cr +1σ

and rate of car-using evacuees rc –1σ 15,130

Table 10 Estimated number of

fatalities vs. actual number

N

umber of

fatalities

Actual number 16,900

Estimated number

by Eq. (7) 16,819

- 15 -

5.1.2 Aging rate

A high damage rate of aged people is a well-known feature11), 12). Therefore, HVI is presumed to be high

in the area of high aging rate. However, the correlation coefficient between the aging rate, which is

defined by the ratio of the population over 70 years old to that over 20 years old, and HVI is as low as

−0.094. The accuracy of the multiple regression analysis, which was added to the aging rate as one of

the explanatory variables, was not improved. Here, the average of the aging rate of 20 LMs was 0.284

and the standard deviation was 0.0256. Therefore, the variation of the aging rates was generally low.

Moreover, LM I and LM J, which had relatively higher aging rates, were on the rias coast and had

experienced tsunamis frequently. And as reported by the interview survey of Goto and others22), the

experience seemed to have transmitted from one generation to the next. In such LMs, awareness of

tsunamis was shared among the people and there were fewer fatalities. Thus, the HVI might not become

high, even if the aging rate is high.

5.1.3 Visibility of sea

Tanishita12), 28) reported a tendency in which the number of fatalities increased in areas where there was

no direct view of the sea coast. This study could not take into account the visibility of the sea coast

because the minimum resolution of the area was the size of each LM. If a finer areal resolution were

applied, the visibility of the sea coast would be evaluated by utilizing 3-dimensional GIS and such, and

its effect could be studied by adding this as one of the explanatory variables.

5.1.4 Coastal levee

The structural effect of coastal levees is automatically reflected in the hazard term of Eq. (7), because

the presence or absence of coastal levees affects the inundation depth and rate of washed-out houses. As

for the spiritual effect, people in the area without effective coastal levees might have heightened

wariness over tsunamis and evacuate quickly. These effects are reflected in preparedness P.

On another front, taking coastal levees as one of the explanatory variables, it could be possible to

analyze the following effects: The presence of coastal levees might reassure people living near-by and

induce them to stay in their houses, and they might conceal the tsunami and cause people to delay their

evacuation; conversely, it might delay the tsunami inundation. However, these effects of coastal levees

could not be analyzed in this study, because no data that covered the location and height of all coast

levees of 20 LMs were found.

5.2 Evaluations of key parameters to analyze the evacuation vulnerability of a tsunami-anticipated

area using HVI

In order to estimate HVIs for tsunami-anticipated areas and to evaluate evacuation vulnerability, or to

analyze the factors that affect the number of fatalities using Eq. (7), the number of washed-out houses

and the values of the four explanatory variables must be established in advance.

5.2.1 Number of washed-out houses

The number of washed-out houses can be calculated through the following steps:

(1) Obtain the inundation depth distribution from the results of tsunami inundation simulations that the

central or prefecture governments provide to the LMs, and

(2) Calculate the number of washed-out houses using a fragility curve. The fragility curve indicates the

relationship between inundation depth and rate of washed-out houses, as proposed by Koshimura,

et al.29) and others.

However, death by tsunami can happen even if the number of washed-out houses is zero. For these

cases, it is necessary to develop another type of explanatory variable that expresses the tsunami hazard.

This is a future research issue.

5.2.2 Allowance period A (tsunami arrival time/ evacuation distance)

The tsunami arrival time can be estimated from tsunami simulation conducted by central or prefecture

governments. The evacuation distance can be calculated by GIS-aided search of roads that connect

- 16 -

starting points with safe places. The former is people's locations when the earthquake strikes, like home,

and the latter is the outer area of the assumed inundation area or the vertical evacuation facility.

5.2.3 Preparedness P (rate of persons who had prepared emergency carry-out-bags beforehand)

Data from recent existing surveys can be utilized. In LMs that have not completed such a survey, the

data can be collected by questionnaire.

Evaluating the preparation rate of emergency carry-out bags helps to evaluate the level of people's

self-directive risk awareness. It should be noted that disaster education such as only urging people to

prepare emergency carry-out bags does not directly reduce the number of fatalities.

5.2.4 Road serviceability R (car velocity x rate of car-using evacuees)

R is obtained by multiplying car velocity by the rate of car-using evacuees. There are several ways to

estimate car velocity: analyzing the road network capacity against car use demand, using a traffic flow

simulator to analyze the effect of traffic jams, measuring car velocity at a car use evacuation drill, and

so on. The rate of car-using evacuees must be a conceivable rate, not the target rate of a disaster

prevention plan.

5.2.5 Warning effect We (announced tsunami height x cognition rate of the warning)

Warning effect, We, is the equivalent value of the product of announced tsunami height and cognition

rate of warning. In the case of GEJET, it was reported that many people in the area where 3 meters was

announced as the forecast tsunami height by the first warning received this as a sign of safety and some

of them missed the timing of evacuation30). Therefore, 6-meter or 3-meter warning might be a criterion

for people to think about evacuation. However, as the effect of the warning to push people to evacuate

varies with past tsunami experience and announcement history of forecast height in the area, the

effective tsunami height for announcement should be evaluated considering these historical factors.

Cognition rate should not be estimated through analogy of GEJET, because emergency alert emails

and other new IT tools to be issued by governmental agencies to disaster forecast areas have been

introduced. The effect of such tools should be checked on such occasions as disaster drills.

6. CONCLUSIONS

(1) The Human Vulnerability Index (HVI) introduced by this study was verified to be independent of

tsunami height, and to be an index that can express evacuation vulnerability of a studied area being

evaluated by geographical and socio-psychological features.

(2) A multiple regression analysis, which set HVI as a target variable and used four explanatory variables,

allowance period A, preparedness P, road serviceability R and warning effect We, extracted a

regression formula that achieved a multi-correlation coefficient of 0.904 and an adjusted coefficient

of determination of 0.768. The number of fatalities calculated using the regression HVI deviated

from the actual number by only 0.5%.

(3) Sensitivity analysis of the four explanatory variables concerning the number of fatalities indicated

allowance period A as the most effective and road serviceability R as second, if the same variation

rate was applied to one of the four variables. Preparedness P was closely related to wariness over

tsunamis and hence linked with the cognition rate of warnings cr, which is the element variable of

We. Therefore, if these two factors are combined, the effectiveness will be almost the same as that

of A.

(4) Factors indicated by the sensitivities of these explanatory variables have been qualitatively reported

by previous surveys. Nevertheless, by modeling the effects of the variables using HVI, sensitive

factors become clear and the efficiency of countermeasures by improving the factors can be

numerically evaluated.

(5) Issues for the future are to test HVI for other tsunami disasters and improve the reliability of the

regression formula, and to establish a comprehensive evaluation method for explanatory variables

in order to apply HVI to other areas subject to tsunami hazard.

- 17 -

ACKNOWLEDGMENT

The City Bureau of Ministry of Land, Infrastructure, Transportation and Tourism of Japan and the

Center for Spatial Information Science of the University of Tokyo collected significant data regarding

people's evacuation from the Great East Japan Earthquake tsunami and uploaded the data as the Fukkou

Shien Chosa archive. The authors would like to express much thanks and respect for their contributions.

Many appreciations to their effective discussions are given to Professor Hitomi Murakami of

Yamaguchi University, Professor Masayoshi Tanishita of Chuo University, and the members of the

JAEE research committee “Evacuation Research Committee 2012-2016”. Associate Professor Maki

Koyama of Gifu University is appreciated for her kind advice concerning data exploration. The authors

are also grateful to Dr. Raya Muttarak from the International Institute for Applied Systems Analysis

(IIASA) for her kind notification of the importance of HVI’s potential uses.

APPENDIX 1 Outline of Fukkou Shien Chosa archive (FSC archive)

The FSC archive is a GIS database uploaded by the Center for Spatial Information Science of the

University of Tokyo. The original data for the archive is the outcome of the “Survey for Reconstruction

of Damaged Cities suffered by the East Japan Great Tsunami, 2011 (Fukkou Shien Chosa in Japanese)”

conducted by the City Bureau of Ministry of Land, Infrastructure, Transportation and Tourism of Japan.

In line with the contents of the survey report16), the FSC archive web page21), and the paper by Sekimoto,

et al.31), the outline of the archive is described below.

1. The database lists individual persons and business offices in 62 local municipalities (hereafter LMs),

on the coast from Aomori to Chiba Prefectures. The numbers of samples are 10,603 individuals and 985

business offices. In the survey of the individuals, investigators visited shelters, temporary houses and

partially damaged houses, and interviewed people who were affected by the tsunami. The survey term

was from September to December of 2011.

2. The sample rate of the individuals was 1.5% - 3% of over 20 age population in the inundation area of

LMs. The minimum number of samples in an LM was 20, and in reverse, if number exceeded 500, the

sample rate was gradually decreased and the number of samples was limited to around 1,500 at most. In

addition, the affected area was divided into two zones, houses completely collapsed and houses partially

damaged and inundated, and the number of samples was allocated to be proportional to the population

of each area.

For the people in the area where houses had completely collapsed, those in temporary houses were

interviewed, and for the people in the area where houses were partially damaged and inundated, those

in their own houses were interviewed.

The sex and age distribution of samples were targeted to be similar with those of the population in

each LM. However, the actual distribution of samples somewhat deviated from the targets in the LMs

studied. There were fewer 20-39 year old males and females and 40-59 year old males. Conversely,

there were more over 60 year old males and females, and 40-59 year old females.

3. The archive consists of open data and semi-open data, and governmental or research users can access

the latter through a registration procedure. However, for downloading the data, the users are required to

promise to take measures to protect personal information and privacy in future publication of their study

results.

4. The archive contains a GIS data definition document and each LM's past reconstruction plan as well

as each LM's database of many kinds of damage of the 2011 Great East Japan Earthquake. Namely,

inundation area, inundation depth, inundation trace, damage overview, building damage, public

infrastructure damage (river, coast, steep slope, erosion control facility, windbreak storm surge forest,

road, port, sewer, park, and green space), lifeline damage (water and gas), public service damage (bus,

- 18 -

hospital and welfare), cultural asset and educational facility damage, and sufferer and their evacuation

manner data are compiled in the database.

5. Outline of data used in this study:

(1) Inundation area data: Polygon data.

(2) Damaged building data: Polygon data of all buildings in the inundation area with attributes such as

floor area, structure, usage, classification of damage, year build*, residential house or non-residential

house*, adequacy for the use as evacuation points, and inundation depth at building location. (*

means data of some MLs are missing).

(3) Evacuation trip data of individual sufferers: Polyline data of individual evacuation trips with

attributes such as staying time, start time, arrival time, movement method, purpose of trip, trigger of

evacuation start, and tsunami visibility.

(4) Refuge place data of individual sufferers: Point data of refuge place with attributes such as name and

type.

(5) Sufferer data of each administrative area: Polygon data of administrative areas in each municipality

with attributes such as population before disaster, number of households before disaster, number of

deaths, number of missing, number of deceased visitors, and number of decreased households.

However, it should be noted that these data were collected before the end of June of 2011 and a

considerable number of items are missing.

(6) Interview data of individual sufferers: At the top of the evacuation action sub-folder, a table of

interviews of individual sufferers is uploaded. The contents are:

(a) whereabouts of the interviewee at the time of the earthquake,

(b) number of stories of the building where the interviewee was in at the time of the earthquake,

(c) whereabouts of the interviewee's family at the time of the earthquake,

(d) anticipated tsunami coming just after the earthquake or not,

(e) damage of the place where the interviewee was located,

(f) actions the interviewee took after the earthquake,

(g) heard the large tsunami warning or not, heard the height of tsunami forecast or not,

(h) source of the heard warning, impression of warning upon hearing it,

(i) heard evacuation alert from LM or not,

(j) most beneficial source of information in the period from the earthquake to the day's sunset,

(k) intended doing evacuation before arrival of tsunami or not,

(l) evacuation place after the earthquake until sunset of the day, type of evacuation place,

(m) movement method, movement purpose,

(n) trigger for decision to start evacuation,

(o) reason for using a car for evacuation,

(p) problem of road for evacuation,

(q) problem of first evacuation place,

(r) watched hazard map beforehand or not,

(s) saw a board or a sign or a marking that indicated the direction and the place of evacuation, or not,

(t) made preparations for evacuation, such as securing furniture, preparing emergency carry-out bag,

talking with family about tsunami emergency, pre-confirming evacuation place and road, checking

tsunami hazard map, participating in community evacuation drill and so on, or not,

(u) knew the location of the designated place or building for evacuation near the place that was at the

time of the earthquake or not,

(v) was able to go there or not,

(w) sex, age, job of interviewee, number of families living in the same house,

(x) saw the tsunami after the earthquake or not,

(y) damage to interviewee’s house by tsunami or by shaking of the earthquake,

(z) injury to interviewee and his/her family by tsunami or by shaking of the earthquake.

- 19 -

APPENDIX 2 Major trials for selecting explanatory variables (tried about 5 times of this table)

Start

time of

evacua-

tion

Tsunami

arrival

time:t

Evacua-

tion

dist ance

(here

after, ED)

Walking

ED

Car-

using

ED

Equiva-

lent

ED:ed

4)

Allow-

ance

period:

t/ed

5)

Rate of

car-

using-

evacuees:

cr

Car

velocity:

v

Road

service-

ability:

v × cr

6)

Rate of

over 70

age

persons

Tsunami

warning

height :

h

7)

Rate of

antici-

pating

tsunami:

R

h

×

RCogni-

tion rate

of

warning:

cr

Warning

effec t

h × cr

8)

Checked

the

hazard

map or

not

Partici-

pated

evacua-

tion drill

or not

Prepared-

ness

9)

Rate of

returnin g

home +

detour-

ing

0.681 0.357 0.057 0.007 0.056

0.684 0.315 0.168 0.758 0.013 0.063

0.681 0.311 0.069 0.984 0.013 0.066

0.704 0.351 0.338 0.365 0.006 0.047

0.716 0.325 0.327 0.397 0.507 0.019 0.047

0.684 0.316 0.142 0.011 0.754 0.079

0.681 0.311 0.084 0.022 0.111 0.940

0.704 0.351 0.338 0.365 0.006 0.047

0.694 0. 334 0.051 0 .449 0.008 0.026

0.719 0.369 0.036 0.213 0.014 0.069

0.666 0.332 0.029 0.010 0.043

0.717 0. 375 0.027 0 .176 0.038 0.029

0.760 0.456 0.026 0.025 0.036 0.016

0.766 0.470 0.040 0.055 0.086 0.009

0.782 0.501 0.025 0.008 0.063 0.005

0.660 0.323 0.109 0.371 0.007

0.814 0.565 0.008 0.004 0.036 0.004

0.730 0.440 0.060 0.006 0.009

0.819 0.544 0.076 0.004 0.575 0.037 0.004

0.825 0.557 0.099 0.004 0.032 0. 408 0.007

1

0) 0.814 0.532 0.971 0.044 0.007 0.059 0.005

0.817 0.539 0.013 0.665 0.083 0.051 0.005

0.760 0.456 0.026 0.025 0.036 0.016

0.797 0.530 0.015 0.004 0.068 0.013

0.644 0.248 0.415 0.049 0.872 0.052

0.730 0.440 0.060 0.006 0.009

0.744 0.426 0.003 0.007 0.176 0.061

0.650 0.306 0.009 0.025 0.122

0.793 0.523 0.003 0.003 0.127 0.031

0.748 0.471 0.008 0.003 0.012

0.744 0.425 0.012 0.038 0.544 0.041

0.804 0.545 0.002 0.003 0.087 0.009

0.748 0.433 0.023 0.015 0.461 0.041

0.842 0.596 0.040 0.011 0.021 0.126 0.003

0.842 0.596 0.139 0.002 0.023 0.125 0.003

0.810 0.523 0.108 0.005 0.062 0.605 0.007

0.805 0.513 0.077 0.005 0.080 0.010 0.950

0.748 0.433 0.011 0.051 0.534 0.048

0.739 0.456 0.011 0.021 0.037

0.799 0.534 0.042 0.008 0.082 0.009

0.831 0.571 0.036 0.023 0.036 0.162 0.006

0.803 0.568 0.013 0.020 0.006

0.822 0.577 0.010 0.065 0.280 0.006

0.838 0.603 0.078 0.010 0.509 0.004

0.831 0.619 0.031 0.008 0.003

0.850 0.631 0.024 0.027 0.258 0.003

0.861 0.656 0.057 0.004 0.148 0.009

0.836 0.599 0.009 0.009 0.004 0.558

0.851 0.673 0.010 0.007 0.000

0.867 0.686 0.006 0.012 0.218 0.000

0.869 0. 691 0.006 0.010 0.189 0.000

0.860 0.669 0.010 0.009 0.380 0.001

0.823 0. 617 0.001 0.002 11) 0.001

0.837 0.621 0.001 0.015 0.064 0.050

0.895 0.747 0.000 0.003 0.002 0.001

0.897 0.753 0.000 0.001 0.008 0.000

0.874 0.700 0.000 0.002 0.034 0.018

0.797 0.566 0.002 0.006 0.131

0.904 0. 768 0.000 0.002 0.004 0.003

0.904 0.752 0.000 0.004 0.886 0.006 0.004

0.904 0. 752 0.000 0.003 0.843 0.009 0.027

1

2) 0.827 0.599 0.001 12) 0.151 0.017 0.044

Sign ific an ce le ve l p of e xp lana to ry va ria ble

(Multip le regres sion a nalysis was ap plied to the s et of va riables of which ce lls a re fille d b y p va lues ).

R

Corre-

lation

coefficient

R2

Adjusted

determi-

nation

coefficien t

1) An alyze d

using the data

of 19 local

mun icipa litie s

(here after LMs)

of Iwate and

Miyagi.

Regression

equation of

po lyn omial

terms was

ap plied .

Th e fo llowin g 2)

and 3), same

type of

po lyn omial

terms was

ap plied .

2) An alyze d

using the data

of 17 LMs of

Iwate and

Miyagi.

3) 20 LMs of

Iwate, Miyagi

and Fukushima.

Analyzed using

the data of 20

LMs o f I wat e,

Miyagi and

Fukus hima.

Monomial of

exponential

terms was

ap plied .

1) The 19 LMs are Miyako, Yamada, Otsuchi, Kamaishi, Ofunato, Rikuzentakata, Kesennuma, Minamisanriku, Onagawa, Ishinomaki (rias coast area),

Ishinomaki (flatland area), Higashimatsushima, Shichigahama, Miyagino-ku of Sendai, Wakabayashi-ku of Sendai, Natori, Iwanuma, Watari and Yamamoto.

2) Miyagino-ku and Iwanuma were omitted because of their large deviation of age distribution from Census. (Data of under 49 age of Miyagino-ku was 19%,

whereas Census in the area was 57%. Data of over 70 age of Iwanuma was 5%, whereas Census was 18%.)

3) Shinchi, Minamisoma and Iwaki of Fukushima were added, whereas Soma of Fukushima was excluded because its preparation rate of emergency carry-out-

bag was unnaturally low. (Preparation rate of emergency carry-out bag of Soma was 5.1%, while the rate of 20 LMs was 37.2% in average.)

4) Equivalent ED = Average walking ED × Rate of walking evacuee+Average car ED × Rate of car-using evacuee×0.244. (Here, ED is the distance of road

from evacuee's home to the boundary of inundation area.)

5) Allowance period = Tsunami arrival time/Equivalent ED

6) Road serviceability = Rate of car-using evacuees × Car velocity (The better the road, the higher the car velocity and the higher the rate of car users.)

7) Tsunami forecast height broadcast by the first tsunami warning (Iwate 3m, Miyagi 6m, Fukushima 3m)

8) Warning effect = Tsunami forecast height × Cognition rate of tsunami warning

9) Preparedness= Rate of persons who prepared emergency take-out bag beforehand

10) Hereafter，HVI was modified by applying the rate of persons in their houses, as denoted by Eq. (2).

11) Hereafter, when calculating the preparedness rate, the denominator was replaced by the number of persons whose houses were completely collapsed to the

number of persons who were at home or returned home before the tsunami’s arrival and concurrently whose houses were completely collapsed.

12) Road serviceability was replaced by a dummy variable that defined the rias coast region (north of Oshika peninsula) as 0 and the flat coast region (southwest

of Oshika peninsula) as 1．

- 20 -

REFERENCES

1) Imamura, A.: Disaster Prevention Committee Report 77., Disaster Prevention Committee, 1913. (in Japanese)

2) Coburn, A. and Spence, R.: Earthquake Protection, 2nd Edition, Chichester, John Wiley & Sons Ltd., 2002.

3) Ohta, Y., Goto, N. and Ohashi, H.: An Empirical Construction of Equations for Estimating Number of Victims

at an Earthquake, Zisin, Vol. 36, No.6, pp. 463-466, 1983. (in Japanese)

4) Miyano, M. and Lu, H. G.: The Effect of Fault Distance to the Relationship between Human Casualties and

Housing Damage due to Earthquakes, Journal of Japan Society for Natural Disaster Science, No.13-3, pp.

287-296, 1995. (in Japanese)

5) Murakami, H.: Chances of Occupant Survival and SAR Operation in the Buildings Collapsed by the 1995

Great Hanshin Earthquake, Japan, Proceedings of 11th World Conference on Earthquake Engineering, Paper

No. 852, 1996.

6) Matsuda, I.: Changes of Human Losses and Deaths due to the 1995 Hyogo-ken Nanbu Earthquake,

Comprehensive Urban Studies, Vol. 61 pp.155-166, 1996. (in Japanese)

7) Moroi, T. and Takemura, M.: Comparison of Fatality Risk between the 1891 Nobi Earthquake, the 1948 Fukui

Earthquake and the 1995 Hyogoken-Nanbu Earthquake, Zisin, Vol. 52, pp.189-197, 1999. (in Japanese)

8) Takemura, M.: Earthquake and Disaster Prevention ~from vibration to aseismic design~, Chukoshinsyo,

Chuokoron-Shinsha Inc., 2008. (in Japanese)

9) Kuwasawa, N., Kanai, M., Hosoi, K. and Katada, T.: Study on the effect of disaster education considering

decision making evacuation from a tsunami, Infrastructure Planning Review, JSCE, Vol.23, pp.345-354，

2006. (in Japanese)

10) Suzuki, S. and Hayashi, H.: A Preliminary Critical Causal Analysis on the Mortality Caused by the 2011 East

Japan Earthquake Tsunami, Journal of Social Safety Science, No.15, pp.179-188, 2011. (in Japanese)

11) Koyama, M., Ishii, N., Furukawa, A., Kiyono, J. and Yoshimura, A.: Municipality’s Mortality Rate according

to Inundation Level and Age Classes on the 2011 Great East Japan Earthquake, Journal of JSCE A1 (Structure

and Earthquake Engineering), Vol.69, No.4, I_161-I_170, 2013. (in Japanese)

12) Tanishita, M. and Asada, T.: Factors Associated with District Victim Rate by Tohoku Earthquake Tsunami

in Minamisanriku Town, Journal of JSCE A1 (Structure and Earthquake engineering). Vol.70, No.4, I_66-

I_70, 2014. (in Japanese)

13) Goto, Y.: Index to Evaluate Tsunami Evacuation Potential and its Validation at Yamada, Iwate Prefecture,

Journal of Disaster Research, Vol.9, No.7, pp.730-742, 2014.

14) Nakasu, T., Ono Y. and Pothisiri W.: Forensic Investigation of the 2011 Great East Japan Earthquake and

Tsunami disaster: a case study of Rikuzentakata, Disaster Prevention and Management: An

International Journal, Vol. 26, Issue 3, DOI: 10.1108/DPM-10-2016-0213, 2017.

15) Nakasu, T., Ono Y. and Pothisiri W.: Why did Rikuzentakata have a high death toll in the 2011 Great East

Japan Earthquake and Tsunami disaster?: Finding the devastating disaster’s root causes, International Journal

of Disaster Risk Reduction, Vol.27, pp.21-36, 2017.

16) Ministry of Land, Infrastructure, Transport and Tourism: Summary of Tsunami Affected Urban Area

Reconstruction Method Investigation Survey from the Great East Japan Earthquake

http://www.mlit.go.jp/toshi/toshi-hukkou-arkaibu.html. (in Japanese, last accessed in February, 2015)

17) City Bureau of Ministry of Land, Infrastructure, Transport and Tourism: About the Arrangement of Escape

Routes, Evacuation facilities and Evacuation Guidance assuming Tsunami Evacuation (Third Edition) 2013,

http://www.mlit.go.jp/common/000233464.pdf. (in Japanese, last accessed in June, 2016)

18) Goto, Y. and Nakasu, T.: Human Vulnerability Index for Evaluating Tsunami Evacuation Capability of

Communities, Proceedings of 16th World Conference on Earthquake Engineering, Paper No.1669, 2017.

19) Iwate Prefecture: Iwate Disaster Management Portal - Human Damage / Building Damage Status due to

Tohoku Region Pacific Offshore Earthquake Summary as of July 31,

http://www2.pref.iwate.jp/~bousai/index.html. (in Japanese, last accessed in September, 2015)

20) Miyagi Prefecture: Damage caused by the Great East Japan Earthquake as of July 31, 2015

http://www.pref.miyagi.jp/uploaded/attachment/321498.pdf. (in Japanese, last accessed in September, 2015)

21) Fukushima Headquarter for Disaster Countermeasures: Damage Status due to the 2011 Tohoku Region

Pacific Offshore Earthquake (1666th Report) as of October 31,

https://www.pref.fukushima.lg.jp/uploaded/life/241248_559150_misc.pdf. (in Japanese, last accessed in

October, 2016)

22) City Bureau of Ministry of Land, Infrastructure, Transport and Tourism of Japan and the Center for Spatial

Information Science of the University of Tokyo: Fukkou Shien Chosa archive.

http://fukkou.csis.u-tokyo.ac.jp/default/about. (in Japanese, last accessed in March, 2014)

- 21 -

23) Goto, Y., Ikeda, H., Ichiko, T., Ogawa, Y., Kitaura, M., Sato, S., Suzuki, H., Tanaka, T., Nakamura, M.,

Mikami, T., Murakami, H., Yanagihara, S. and Yamamoto, K.: The Joint Survey Group about the Tsunami

Evacuation of the Great East Japan Earthquake and Field Study by the Sub-group charged with Yamada and

Ishinomaki – Analysis on Data Characteristics – , Journal of JAEE, Vol. 15 (2015) No. 5, p. 5_118-5_143,

2015. (in Japanese)

24) Statistics Bureau of Ministry of Internal Affairs and Communications: Statistics Viewed on the Map

(Statistical GIS)

http://e-stat.go.jp/SG2/eStatGIS/page/download.html. (in Japanese, last accessed in October, 2016)

25) Goto, Y.: Impacts of Returning Home or Dropping In Just Before or During Evacuation from the 2011 East

Japan Great Earthquake Tsunami, Journal of JAEE, Vol. 16 (2016) No.10, p.10_86-10_104, 2016. (in

Japanese)

26) Mikami, T.: The Survey Analysis about the Victims by the Tsunami under the Great East Japan Earthquake

– Yamada-cho and Ishinomaki City –, Journal of JSCE A1(Structure and Earthquake engineering), Vol.70,

No.4, I_908-I_915, 2014. (in Japanese)

27) Tanaka, Y. and Tarumi T.: Statistical Analysis Handbook - Multivariate analysis, ISBN 4-320-01492-8,

Kyoritsu Shuppan Co., Ltd., pp.48-49, 1995. (in Japanese)

28) Tanishita, M.: Factors associated with district tsunami victim rate by age class in Ishinomaki city 2011 Great

East Japan earthquake and tsunami, ResearchGate, DOI: 10.13140/RG.2.1.1493.8002/1, 2016.

https://www.researchgate.net/publication/299752194. (in Japanese, last accessed in November, 2016)

29) Koshimura, S. and Gokon, H.: Structural Vulnerability and Tsunami Fragility Curves from the 2011 Tohoku

Earthquake Tsunami Disaster, Journal of JSCE B2 (Coastal engineering), Vol.68, No.2, I_336-I_340, 2012.

(in Japanese)

30) Japan Meteorological Agency: On the Direction of Improvement of the Tsunami Warning based on the

Damage of the Tsunami caused by the Tohoku Region Pacific Offshore Earthquake, September 12, 2011,

http://www.jma.go.jp/jma/press/1109/12a/torimatome.pdf. (in Japanese, last accessed in October, 2016)

31) Sekimoto, Y., Nishizawa, A., Yamada, H., Shibasaki, R., Kumagai, J., Kashiyama, T., Sagara, T., Kayama,

Y. and Ootomo, S.: Data Mobilization by Digital Archiving of the Great East Japan Earthquake Survey,

Theory and Applications of GIS, GIS Association of Japan, Vol.21, No.2, pp.1-9, 2013.

(Original Japanese Paper Published: May, 2017)

(English Version Submitted: June, 4, 2018)

(English Version Accepted: August 22, 2018)

- 22 -