Beruflich Dokumente
Kultur Dokumente
Paper:
Tsunami Fragility
– A New Measure to Identify Tsunami Damage –
Shunichi Koshimura∗ , Yuichi Namegaya∗∗ , and Hideaki Yanagisawa∗∗∗
∗ Graduate School of Engineering, Tohoku University
Aoba 6-6-11-1104, Aramaki, Aoba-ku, Sendai 980-8579, Japan
E-mail: koshimura@tsunami2.civil.tohoku.ac.jp
∗∗ Geological Survey of Japan, National Institute of Advanced Industrial Science and Technology
Abstract. Tsunami fragility (fragility curve, or fragility Since the 2004 Sumatra-Andaman earthquake tsunami
function) is a new measure, we propose, for estimat- disaster, numerous efforts have been made to identify the
ing structural damage and fatalities due to tsunami at- tsunami damage mechanisms by widely deployed post-
tack, by integrating satellite remote sensing, field sur- tsunami survey teams reporting tsunami height, inunda-
vey, numerical modeling, and historical data analysis tion zone extent and damage [5–9]. These efforts have led
with geographic information system (GIS). Tsunami to new understandings of local aspects of tsunami inunda-
fragility is expressed as the structural damage prob- tion flow and damage mechanisms.
ability or fatality ratio related to hydrodynamic fea- However, their findings based on the inspection of lo-
tures of tsunami inundation flow, such as inundation cal aspects of tsunami damage make it difficult to identify
depth, current velocity and hydrodynamic force. It the “vulnerability” in a quantitative manner. The nature
expands the capability of estimating potential tsunami of vulnerability is associated with multitude of uncertain
damage in a quantitative manner. sources, such as hydrodynamic features of tsunami in-
undation flow, structural characteristics, population, land
use, and any other site conditions. To view tsunami vul-
Keywords: fragility curve, tsunami damage estimation, nerability comprehensively requires humongous amounts
remote sensing, numerical modeling, historical tsunamis of damage data, whereas post-tsunami survey rarely pro-
vides sufficient data because of limited survey time and
human resources. As Shuto (1993) concluded, degree of
1. Introduction damage may change with these uncertainties and a statis-
tical approach to these uncertainties should be taken.
In tsunami damage estimation efforts, several empiri- Herein, we propose Tsunami fragility (fragility curve
cal relationships between tsunami hazard and vulnerabil- or fragility function) as a new measure for estimating
ity have been used. Shuto (1993) proposed the tsunami tsunami damage. Tsunami fragility is defined as the struc-
intensity scale to discuss the structural damage based on tural damage probability or fatality ratio with particular
the empirical data from historical tsunamis in Japan, in regard to the hydrodynamic features of tsunami inunda-
terms of the damage and local tsunami height [1], and this tion flow, such as inundation depth, current velocity and
has been widely used in tsunami disaster assessment by hydrodynamic force. In principle, the development of
the Japanese government as a measure of tsunami dam- tsunami fragility requires that tsunami hazard informa-
age. When the local tsunami inundation depth exceeds tion and damage data should be used synergistically. We
2 m, for example, Shuto’s tsunami intensity scale sug- thus incorporate several approaches to constructing the
gests, complete destruction of wooden houses as shown tsunami fragility.
in Fig. 1. Izuka and Matsutomi (2000) suggested a struc- In order to obtain tsunami hazard information such as
tural destruction threshold related to the hydrodynamic inundation depth and current velocity, we performed a nu-
features of tsunami inundation flow throughout the field merical modeling of tsunami inundation with some model
surveys and laboratory experiments [2]. And some engi- validations, especially focusing on the 2004 Sumatra-
neering studies have proposed tsunami design forces on Andaman earthquake tsunami. In terms of the damage
structures based on the laboratory experiments [3, 4], but data including structural damage and fatalities, we use the
have not suggested the procedures for estimating struc- recent advances of remote sensing technologies expand-
tural damage. ing capabilities of detecting spatial extent of tsunami af-
Tsunami intensity 0 1 2 3 4 5
Tsunami height (m) 1 2 4 8 16 32
Damage : wooden house partly dam- completely destroyed
aged
Damage : masonry house withstand no data completely destroyed
Damage : reinforced concrete completely
building withstand no data destroyed
Damage : fishing boats damage 50 damage completely damaged
Fig. 1. Tsunami intensity scale and damage [1], modified from the original figure.
fected areas and damage on structures. In addition, some 1 Damage data acquisition : obtaining damage data
field data and historical documents are used to obtain the from satellite images, field surveys or historical doc-
tsunami fragility from historical events. uments, e.g. numbers of destroyed or survived struc-
Throughout the data integration and statistical analysis, tures with its spatial information.
we propose a concept and framework developing tsunami
fragility, and apply it in several approaches according to 2 Tsunami hazard estimation : estimating the hydro-
available data types from recent and historical events. dynamic features of tsunami by numerical modeling,
field surveys and from historical documents.
3 Data assimilation between the damage data and
2. Developing Tsunami Fragility – Methods tsunami hazard information : correlating the dam-
age data and the hydrodynamic features of tsunami
Fragility curves (or fragility functions) have conven- through the GIS analysis.
tionally been developed in performing seismic risk analy-
sis of structural systems to identify structural vulnerability 4 Calculating damage probability : determining the
against strong ground motion using damage data associ- damage probabilities by counting the number of
ated with historical earthquakes and spatial distribution of damaged or survived structures, for each range of hy-
observed or simulated seismic responses [10]. And they drodynamic features above.
have been implemented in estimating structural damage
against potential seismic risks in which various uncertain 5 Regression analysis : developing the fragility curves
sources such as seismic hazard, structural charachteris- by regression analysis of discrete sets of damage
tics, soil-structure interaction are involved [11, 12]. probability and hydrodynamic features of tsunami.
In earthquake engineering studies, the fragility curves In the sections that follow, we apply the above proce-
are defined by the following Eq. (1) or (2) ; dure in several approaches to construct tsunami fragility,
ln x − λ according to the data types such as satellite data, numeri-
PD (x) = Φ cal models, field surveys and historical documents.
ξ
x
1 (lnt − λ )2
= √ exp − dt (1)
−∞ 2πξ t 2ξ 2 3. Tsunami Fragility Determined from Satel-
lite Remote Sensing and Numerical Model-
x−µ ing
PD (x) = Φ
σ
x
Taking an advantage in satellite remote sensing, we
1 (t − µ )2 identify the spatial distribution of structural damage by
= √ exp − dt . (2)
−∞ 2πσ 2σ 2 tsunami. The highest spatial resolution of commercial op-
tical satellite imaging is up to 60-70 centimeters (Quick-
where PD (x) is the damage probability of structures, as a Bird owned by DigitalGlobe) or 1 meter (IKONOS op-
function of x or ln x such as the maximum ground accel- erated by GeoEye). Fig. 2 shows the result of visual
eration, velocity, and seismic intensity. Here, PD (x) is ex- interpretation [13] on structural damage by using a set
pressed by the standard lognormal or normal cumulative of pre and post-tsunami (the 2004 tsunami) satellite im-
distribution function Φ[(ln x − λ )/ξ ] or Φ[(x − µ )/σ ], ages (IKONOS) from Banda Aceh, Indonesia. Through
with two statistical parameters (λ , ξ ) or (µ , σ ), as the inspecting a set of pre and post-tsunami satellite images
mean of ln x (or x), and the standard deviation respec- visually or manually, presence of building roofs can be
tively. interpreted. The advantage of using high-resolution opti-
To develop tsunami fragility, we take a statistical ap- cal satellite images for damage interpretation is the capa-
proach synergistically using remote sensing, numerical bility of understanding structural damage visually and en-
model results, field surveys and historical documents, in ables us to comprehend its spatial extent in regional scale
five steps. where post-tsunami survey hardly get through because of
(a) ( )
Building/house damage (N=48910)
Destroyed (16474)
Survived (32436)
0 1 2 3 4 5
km
( )
(e)
0 50 100
m
Building/house damage
Inundation limit Destroyed Survived
Fig. 2. (a) Spatial distribution of structural damage interpreted from the pre and post-tsunami satellite images (IKONOS) [13].
Black dots indicate the interpreted structures as destroyed, and the gray dots as survived. The arrow points the expanded region
shown in the right panels (b) pre-tsunami and (c) post-tsunami satellite images and (d) interpreted damage. (e) The maximum
tsunami inundation depth obtained from the numerical model [14, 15].
Number of
the limitation of survey time and resources. However,
1200
Destroyed
note that no structural types were identified by the inter- Survived
pretation of satellite images. Also, the damage feature 1000
which can be identified from satellite images is only struc-
tural destruction or major structural failure which reveals 800
6.7
5.0
5.1
5.3
5.6
8.2
0.2
1.0 1.0
0 2 4 6 8 10 0.8 0.8
dmax
0.6 0.6
Fig. 4. A discrete set of damage probabilities and the me- 0.4 0.4
dian values of inundation depths that were compiled from
sample data. 0.2 0.2
0.0 0.0
6 0 2 4 6 8 0 2 4 6
Least-squares fit
5 1.0
4 0.8
)m( xamd
0.6
3
0.4
2
0.2
1
0.0
0 10 20 30 40
0
-3 -2 -1 0 1 2 3
F-1 Fig. 6. Tsunami fragility curves for structural destruction,
Fig. 5. An example of the plot on normal probability paper. in terms of (a) the maximum inundation depth, (b) the max-
imum current velocity and (c) the maximum hydrodynamic
force obtained from the numerical model. The solid lines
are the best-fitted curves of the plot (◦ : the distribution of
numerical model. And as a result of counting the number damage probabilities) with the parameters in Table 1.
of destroyed and survived structures within each inunda-
tion depth range (group), we obtain a relationship between
the damage probability and inundation depth, as a discrete mum hydrodynamic force F acting on a structure per unit
set of structural damage probabilities and tsunami inunda- width (kN/m). Here F is defined as the maximum drag
tion depths shown in Fig. 4. Then, we explore this rela- force per unit width of structures;
tionship with the form of fragility curve by performing the
1
regression analysis. F = CD ρ max{v2 d} × 10−3 . . . . . . . (3)
Taking an analogy of earthquake engineering studies 2
[10, 11, 17], we assume that the cumulative probability PD where CD is the drag coefficient (CD = 1.0 for simplic-
of damage occurrence is given as either Eq. (1) or (2) ity), ρ water density (= 1, 000 kg/m3 ), v current velocity
with two statistical parameters, (λ , ξ ) or (µ , σ ). Here, (m/s) and d inundation depth (m), and both of v and d are
the statistical parameters λ (or µ ) and ξ (or σ ) are ob- obtained at each time step of the tsunami inundation mod-
tained by plotting x (or ln x) and the inverse of Φ (Φ−1 ) eling. Note that the tsunami fragility with respect to the
on normal or lognormal probability paper, and conduct- inundation depth is given by the standardized normal dis-
ing the least-squares fitting of this plot, as shown in Fig. 5. tribution function with µ and σ , while those to the current
Hence, these parameters are obtained by taking the inter- velocity and hydrodynamic force are by the standardized
cept (= λ or µ ) and the angular coefficient (= ξ or σ ) in lognormal distribution functions with λ and ξ . The selec-
Eq. (1) or (2). tion of which curve is applied should be made by checking
Through regression analysis, the parameters are de- its fit to the datasets.
termined as shown in Table 1, to obtain the best fit of Fragility curves shown in Fig. 6 indicate the damage
fragility curve with respect to the maximum inundation probabilities of structural destruction equivalent to the hy-
depth (measured above the local ground level) dmax (m), drodynamic features of tsunami inundation flow. Houses
the maximum current velocity vmax (m/s) and the maxi- in Banda Aceh, for example, were especially vulnera-
ble when the local inundation depth exceeded 2 or 3 m,
0.6 0.6
overestimation in damage probabilities to the hydrody-
namic features of tsunami inundation flow. Further to be 0.4 0.4
+
$ !"
#
&
'
!
"
> @
+
Fig. 9. The result of visual interpretation of structural damage in Banda Aceh city and
the points of field measurements. See their original paper [19] for all the points and
areas investigated.
Table 2. Statistical parameters of tsunami fragility curves Table 3. The damage probabilities and measured tsunami
for fatality (Fig. 8). heights obtained by the visual inspection of satellite images
and field measurements. The area and points A to O are
Fragility curve µ σ R2 equivalent to Fig. 4.
(a) dmax (m) 3.92 1.15 0.62
(b) hmax (m) 5.37 0.76 0.72 Area hmax dmax Destroyed Pre-tsunami Damage
(m) (m) structures structures probability
A 7.0 5.4 40 41 0.98
B 7.1 5.0 51 57 0.89
aware of tsunami arrival have evacuated and tried to sur- C 7.6 4.7 32 45 0.71
vive. In other words, the fragility curve of Fig. 8 does not D 7.1 5.0 72 83 0.87
indicate the human’s survival possibility according to the E 8.2 4.9 46 57 0.81
F 6.6 3.7 51 60 0.85
local hydrodynamic features of tsunami inundation flow. G 4.6 2.5 2 31 0.06
Also, taking median to obtain the representative values of H 4.7 2.1 12 30 0.40
tsunami inundation depth according to each desa reflects I 12.0 – 72 75 0.96
higher variance of the plot compared with that of Fig. 6. J 9.7 7.9 60 61 0.98
K 6.1 3.3 51 56 0.91
For the above reasons, this fragility function should be in- L 6.6 3.8 36 50 0.72
terpreted as a macroscopic measure of tsunami impact, i.e. M 6.8 4.5 38 50 0.76
the occurrence of tsunami fatality significantly increase N 6.5 2.6 14 30 0.47
when the local inundation depth exceeds approximately O 4.8 – 6 31 0.19
2 m and the inundation height 4.5 (m), and almost impos-
sible to survive when the local inundation depth exceeds
1.0 1.0
6 m.
0.8 0.8
0.6 0.6
5. Tsunami Fragility from Satellite Remote
0.4 0.4
Sensing and Field Survey
0.2
0.2
Developing tsunami fragility may also be viewed using
0.0 0.0
satellite images and post-tsunami surveys. Namegaya and 0 2 4 6 8 0 2 4 6 8 10 12
Tsuji (2006) investigated the structural damage in Banda
Aceh city by the 2004 tsunami, using the visual inspec-
tion of QuickBird pre and post-tsunami satellite images Fig. 10. Tsunami fragility curves for structural destruction,
acquired on June 23, 2004 and December 28, 2004, in in terms of (a) the maximum inundation depth and (b) the
four areas of Banda Aceh city together with the mea- maximum inundation height, by using the visual inspection
surements of tsunami inundation depth and height [19]. of satellite images and field measurements. The dashed line
in (a) indicates the fragility curve from Fig. 6(a).
Fig. 4 is showing their result of visual interpretation of
structural damage in Banda Aceh city and the points of
tsunami measurements. The markers • and ◦ in the figure
denote survived and destroyed (washed-away) structures 6. Tsunami Fragility from Historical Data
interpreted from the satellite images focusing on the pres-
ence of their roofs. In constructing the tsunami fragility from historical
Using field survey results presents difficulties in cor- events, we incorporated the historical tsunami data on
relating the tsunami heights at all the points where the local tsunami damage and height. In Japan, the post-
structural damage was inspected. In this case, the dam- tsunami surveys would be conducted by many differ-
age probability is calculated by counting the number of ent organizations and individuals. After the 1896 Meiji
survived and destroyed structures within the dashed-line Sanriku earthquake tsunami, which caused approximately
circles of 100 m diameter (see Fig. 4) and the tsunami in- 22,000 casualties, the damage survey efforts were con-
undation depth or height is represented by the measured ducted by the central government [20] and an engineer
value at the center of each solid-line circle. Consequently, Soshin Yamana delegated by Iwate prefectural govern-
the relationships between the damage probabilities and ment (his report was published by Yamashita (1982) with
tsunami heights are obtained at 13 to 15 points in Banda comments and interpretations) [21]. Seismological re-
Aceh city (see Namegaya and Tsuji (2006) for details). searchers and Japan Meteorological Agency also con-
Table 3 is the result obtained at each area equivalent to ducted the survey after the 1933 Showa Sanriku earth-
Fig. 4. The tsunami fragility curves are determined as quake tsunami which caused approximately 3,000 casu-
Fig. 10, with statistical parameters of Table 4. Note that alties [22, 23].
two fragility curves in Fig. 10(a), although very similar, Hatori (1984) compiled the house damage data from
show differences in approaches of damage data compila- the historical documents of the 1896 Meiji Sanriku, 1933
tion (including number of samples and the inspected area) Showa Sanriku and the 1960 Chile tsunami events as
and methods to obtain tsunami hazard information. listed in Table 5. He defined the structural damage prob-
x µ σ R2 0.8 0.8
a + b/2
PD = . . . . . . . . . . . . . (4) 0.6 0.6
a+b+c PD PD
0.4 0.4 1896 Meiji
where a, b and c is the number of the houses catego- (Shuto)
0.2 1896 Meiji (Shuto) 0.2 Fragility
rized as destroyed/washed-away, moderate damage and Fragility Historical
Historical data data
only flooded, respectively. 0.0 0.0
Shuto (1987a, 1993) also compiled the documents and 0 5 10 15 20 0 2 4 6 8 10
Tsunami height (m) Inundation depth (m)
reports from the 1896 Meiij Sanriku tsunami (Table 5),
and determined the damage probability with four damage
Fig. 11. Historical tsunami data of Hatori (1984) and Shuto
categories ; (1987a, 1993), and tsunami fragility curves. (a) 1896 Meiji
a + b + c/2 Sanriku tsunami by Hatori, (b) 1933 Showa Sanriku tsunami
PD = . . . . . . . . . . . (5) by Hatori, (c) 1960 Chile tsunami by Hatori, (d) Three events
a+b+c+d
by Hatori, (e) and (f) 1896 Meiji Sanriku tsunami by Shuto.
where a, b, c and d is the number of the houses cate-
gorized as washed-away, completely destroyed, moderate
damage and only flooded, respectively. To increase the Table 6. Statistical parameters of tsunami fragility curves
reliability of data, he conducted the numerical modeling (Fig. 11). µ , σ for the regression of the historical data (solid
and the additional field survey to determine the reliability line) and µ , σ , µ , σ for upper and lower limits.
of the documents. Fragility curve µ σ R2 µ σ µ σ
Figure 11 plots historical data (damage probability ver- (a) 5.84 3.28 0.30 2.80 1.35 9.00 5.00
sus tsunami height and inundation depth) compiled by (b) 5.09 2.50 0.70 2.90 1.35 7.00 2.50
Hatori (1984) and Shuto (1987a, 1993), and the fragility (c) 4.66 1.09 0.80 N/A N/A N/A N/A
curves (solid lines) obtained by the least-squares fitting (d) 5.97 2.66 0.55 2.80 1.30 10.0 3.60
of Eq. (2). Since the historical data is highly dispersed, (e) 6.05 2.49 0.24 2.00 0.90 10.5 5.00
the dashed lines are also added to indicate the maximum (f) 5.49 1.26 0.36 N/A N/A N/A N/A
Name:
Hideki Yanagisawa
Address:
Aoba 6-6-11-1104, Aramaki, Aoba-Ku, Sendai 980-8579, Japan Affiliation:
Brief Career: Company Member, Tokyo Electric Power Ser-
2000-2002 JSPS Research Fellow, National Oceanic and Atmospheric vices Company Limited
Administration
2002-2005 Research Scientist, Disaster Reduction and Human Renovation
Institute
2005- Associate Professor, Tohoku University
Selected Publications: Address:
• S. Koshimura, T. Oie, H. Yanagisawa, and F. Imamura, “Developing
Higashi-Ueno 3-3-3, Taito-ku, Tokyo 110-0015, Japan
fragility functions for tsunami damage estimation using numerical model
and post-tsunami data from Banda Aceh, Indonesia,” Coastal Engineering Brief Career:
Journal, No.3, pp. 243-273, 2009. 2008-2009 Post-Doctoral Research Fellow, Graduate School of
• S. Koshimura, Y. Hayashi, K. Munemoto, and F. Imamura, “Effect of the Engineering, Tohoku University
Emperor seamounts on trans-oceanic propagation of the 2006 Kuril Island 2009- Company Member, Tokyo Electric Power Services Company
earthquake tsunami,” Geophysical Research letters, Vol.35, L02611, Limited
doi:10.1029/2007GL032129, 24, 2008. Selected Publications:
• S. Koshimura, T. Katada, H. O. Mofjeld, and Y. Kawata, “A method for • H. Yanagisawa, S. Koshimura, K. Goto, T. Miyagi, F. Imamura, A.
estimating casualties due to the tsunami inundation flow,” Natural Hazards, Ruangrassamee, and C. Tanavud, “Damage of mangrove forest by the
Vol.39, pp. 265-274, 2006. 2004 Indian Ocean tsunami at Pakarang Cape and Namkem, Thailand,”
Academic Societies & Scientific Organizations: Polish Journal of Environmental Studies, Vol.18, No.1, pp. 35-42, 2009.
• Japan Society of Civil Engineers (JSCE) • H. Yanagisawa, S. Koshimura, K. Goto, T. Miyagi, F. Imamura, A.
• Institute of Social Safety Science Ruangrassamee, and C. Tanavud, “The reduction effects of mangrove
• Japan Associaiton for Earthquake Engineering (JAEE) forest on a tsunami based on field surveys at Pakarang Cape, Thailand and
• Japan Society for Computational Engineering and Science (JSCES) numerical analysis, Estuarine,” Coastal and Shelf Science, Vol.81, pp.
• Americal Geophysical Union (AGU) 27-37, 2009.
Academic Societies & Scientific Organizations:
• Japan Society of Civil Engineers (JSCE)
• Japan Society for Mangroves