Content uploaded by Yan Yang
All content in this area was uploaded by Yan Yang on Oct 31, 2019
Content may be subject to copyright.
978-1-61284-848-8/11/$26.00 ©2011 IEEE
Estimating Seismic Losses of Schools Using
SELENA: the Case of Wenchuan Earthquake
Yan Yang1,2*, F. Benjamin Zhan1,2 and Lin Li1
1School of Resource and Environmental Science, Wuhan University, Wuhan, 430079, P.R. China,
2Texas Center for Geographic Information Science, Department of Geography,
Texas State University-San Marcos, San Marcos, TX 78666, US,
*Corresponding author, e-mail: firstname.lastname@example.org
Abstract—Estimation of the seismic risk of schools in developing
countries is an important but challenging task. In this paper, we
report an attempt to estimate the damage and losses of primary
schools during the Mw7.9 Wenchuan earthquake that ruptured
on 12 May 2008 using SELENA (SEismic Loss EstimatioN from a
logic tree Approach). In this study, the seismic hazard was
assessed with the probabilistic analysis procedures available in
SELENA. The capacity spectrum method (CSM) was applied to
determine the vulnerability function, which is a basic input for
estimating building damages and losses. The HAZUS
methodology was used for estimating human losses. This paper
reports preliminary results about the seismic risk of primary
schools in some affected areas of the Wenchuan earthquake. In
addition, it presents some findings about the usefulness of
SELENA for estimating damages and losses of schools using data
from a developing country like China.
Keywords-seismic loss estimation; SELENA; capacity spectrum
method (CSM); primary schools; Wenchuan Earthquake; seismic
The estimation of loss associated with schools during an
earthquake in developing countries is an important but
challenging task. Schools in many developing countries may
not be built to withstand strong ground shakes resulting from a
large earthquake. As a consequence, school children are
among the most vulnerable population groups when a large
earthquake strikes. This paper aims to estimate the seismic
damages and human losses associated with primary schools
during the Mw7.9 Wenchuan earthquake that occurred on 12
May 2008. We used SELENA (SEismic Loss EstimatioN from
a logic tree Approach) for the estimation.
SELENA is a tool for estimating risk and losses associated
with earthquakes. It was an open source software package,
developed by the International Center for Geohazards (ICG),
conducted by NORSAR (Norwegian Seismic Array) and its
collaborator of University of Alicante (in Spain). It can be
downloaded from the website of NORSAR:
http://selena.sourceforge.net/download.shtml. SELENA runs
in the MATLAB environment. It computes the degree of
building damages, the related socioeconomic losses (including
the quantity of casualties and economic losses) through a
spectrum displacement-based approach. SELENA provides
three analysis methods in the computation of ground shaking:
probabilistic, deterministic, and near-real-time. In this study,
the probabilistic analysis procedure was applied with the
shake maps provided by the U. S. Geological Survey (USGS).
In this study, county is the smallest available unit of area,
which is defined as the GEOUNIT or the minimum
geographical unit in SELENA. All the processes in the
calculations are based on the centroid of each county.
II. STUDY AREA AND ASSUMPTIONS
A. Study Area
Figure 1. Fifty studied counties in Sichuan, China; Green star is the
On 12 May, 2008, a violent earthquake struck in
Wenchuan, Sichuan, China. It was centered at 30.986°N,
103.364°E with Mw 7.9, and its focal depth was 19 km .
This disaster was the largest earthquake in China since the
1976 Tangshan earthquake, and it caused significant damages
and losses .
In this case, fifteen heavily affected counties (Fig. 1) in
Sichuan province, China, were studied. In 2007, the total study
area was approximately 3,147,000,000 m2, and the population
was 6,850,000. The total number of primary schools in the
study area was about 1,046 with 437,261 pupils in 2007. More
detailed statistical information about each county can be found
in TABLE I .
Supported by the China Scholarship Council (N0. 200910220)
To estimate the seismic risk with limited data, some
assumptions were made. All these assumptions were made
based on findings reported in the related literature (e.g.  and
). Given below is a summary of the assumptions used in the
xThe model building typologies in building inventory
were URM, RM1, RM2, C2, C3, and PC2 (see
Chapter 5 in ), excluding the C1 and S4, which are
other two possible building types in the study area
that were not used in the study due to the lack of data.
xThere was no High-Rise Building type, which is
defined by HAZUS  as over 8 floors (8+), in
xThere were no buildings with High-Code seismic
design in primary schools. The seismic code design is
based on the technical manual of HAZUS published
by Federal Emergency Management Agency (FEMA)
. According to the document of Sichuan
government, the average building area of each
student is assumed to be 3 m2.
xThere was no pupil in the schools at 2:00 AM. All
pupils were in the schools at 10:00 AM. Half of the
pupils were in the schools at 5:00 PM.
xAt 10:00 AM, the percentage of pupils in each
building type in each county for different occupancy
type is the same.
xThe percentages of each element in the three building
attributes (model building type, building height and
seismic design code) are given in TABLE II. The
numbers of 0.3, 0.3 and 0.4 in the parentheses in
TABLE II are the designed percentages of three
building attributes according to literature.
xThe elastic damping of each type of model building
in this study was respectively assumed as C2: 7%, C3:
10%, PC2: 7%, RM1: 10%, RM2: 7%.
III. DATA AND METHODS
In SELENA, the seismic risk estimation is conducted in
three steps: estimation of seismic hazard, building damage,
and seismic losses (including economic losses and human
losses). The general flowchart of the study on the seismic risk
analysis, using probabilistic analysis procedure without
economic losses, is described in Fig 2. Its content will be
sequentially described in this section.
Figure 2. Flowchart of SELENA using Probabilistic Analysis Procedure.
(revised from )
A. Seismic Hazard
An important provision of seismic risk and loss estimation
is the seismic ground motion (earthquake shaking represented
TABLE I. THE BASIC INFORMATION OF STUDY AREA
TABLE II. THE ASSUMED PERCENTAGES OF BUILDING INFORMATION
Model Building Type (0.3)
RM1 RM2 URM
Building Height (0.3)
Seismic Design Code (0.4)
Probabilistic AnalysisDeterministic Analysis Real-Time Analysis
(No. of Buildings)
(with HAZUS Method)
by its spectral characteristic and level). They are the basic data
for the computation of the physical damages to the affected
buildings, the consequent human losses, and economic losses
caused by the building collapse. The provision of seismic
ground motion is the first step of entire process of seismic risk
In SELENA, three analysis methods (Probabilistic
Analysis, Deterministic Analysis and Real-time Analysis) in
Fig. 2 can be used to estimate seismic ground motions .
xProbabilistic Analysis: Provision of spectral
accelerations, obtained from the probabilistic shaking
maps, with geographical coordinates of centroid in
xDeterministic Analysis: Definition of earthquake
scenarios based on historical events or user-specified
earthquakes and requirement of proper prediction
equations of ground motions.
xReal-time Analysis: Provision of real-time seismic
shaking amplitudes at the sites of seismic stations.
Probabilistic analysis, based on the shake maps, usually
considers all potential earthquake events and probability
distributions of expected damages and losses in a region and
the probable ground shakings generated by these epicenters. In
other words, probabilistic analysis represents the seismic risk
from an ensemble of earthquake scenarios. While
deterministic analysis focuses more on a historical or user-
defined earthquake scenario and predicts ground motion with
the specified local attenuation relation equation. However, it
needs to be mentioned that these two analysis procedures are
complementary to each other. The probabilistic analysis is also
suitable for an earthquake scenario with various probabilistic
elements. In this study, the seismic ground shaking, produced
by the Mw7.9 Wenchuan earthquake on 12 May 2008, were
applied in the probabilistic analysis to experimentally estimate
possible damages and human losses of this disaster.
The spectral ordinates Peak Ground Acceleration (PGA) at
0.01s, Spectral Acceleration (Sa) at 0.3s and Sa at 1.0s, used
as input data in the probabilistic analysis, are provided by the
Global Earthquake ShakeMaps of U.S. Geology Survey
Earthquake Hazard Program (TABLE III). These accelerations,
which do not include soil amplification, need to be
incorporated with a provision of seismic code and its soil
amplification factors to generate the elastic response spectrum,
whose damping factor ξ is 5%. Three seismic codes are
provided in SELENA: the US seismic code IBC-2006 ,
Eurocode 8 (Type1 and Type 2)  and Indian seismic
building code IS 1893. In this case, the IBC-2006 was
B. Building Damages
The estimation of building damage state is chiefly
dependent on building vulnerability (fragility), which is
strongly related to the seismic performance point and its
corresponding spectral displacement. A performance point of
a building type is the plotted point intersected by the capacity
curve and the spectral demand curve, which is decreased for
nonlinear effects. At present, many different methods are
available to identify the building performance point. In
SELENA, the Capacity Spectrum Method (CSM) is one of
such core methods and was adapted from the FEMA’s
software HAZUS-MH, which was tailored to the US situation
and difficult to be used in other countries.
The CSM was proposed by ATC-40 and described in
ATC-40 Chapter 2.4.1 . In this method, the relationship of
the capacity curve and the seismic demand of a building type
are required to be converted into that of the spectral
acceleration and spectral displacement (Sa-Sd) . As the
feature of the performance point is representing the damage
accumulation of an increased structural damping, the process
of computing performance point is iterative.
Once the performance point is found, the next step is to
assign damage probabilities P of the vulnerability (fragility)
function for each damage state ds. The function defined from
HAZUS  can be described as:
Where is the spectral displacement, ̅, is the median
of spectral displacement, is the threshold of damage state.
Φ is the standard normal cumulative distribution function,
is the standard deviation of the natural logarithm of spectral
displacement for damage state .
Further, to estimate the building damages, three impact
factors (the model building typology, the building height and
the level of seismic design code of each building) that can
influence the fragility function needed to be considered.
Because of the limited data that we have collected, the
estimation of Nc,d, the number of each building type (including
all three factors) in each geographical unit (i.e. county), were
calculated by the following equations based on (1) and
statistical data from :
Where i is the number of the counties, j is the number of
the building types (considered all the three factors), P is the
damage probability calculated from (1), PCb(j) is the total
percentage of the jth building type (considering all three
factors), Nc(i) is the total number of buildings in the ith county,
PCmbt(j) is the percentage of the first impact factor (model
building typology) in the jth building type (considering all
three factors), PCh(j) is the percentage of the second impact
factor (building height) in the jth building type, PCc(j) is the
Nc,d(i)(j)=P ∙Nc,b(i)(j) (2)
Nc,b (i)(j)= PCb(j) ∙ Nc(i) (3)
PCb(j)= PCmbt(j) + PCh(j)+ PCc(j) (4)
PCmbt(j) = 0.3 ∙(
) ∙ Pt(5)
PCh(j)= 0.3 ∙(
) ∙ Ph(6)
PCc(j) = 0.4 ∙(
) ∙ Pc(7)
percentage of the third impact factor (seismic design code) in
the jth building type. Pt (model building typology), Ph(building
height) and Pc (seismic design code) are respectively the
assumed percentages of each factor shown in TABLE II.
(model building typology), (building height), (seismic
design code) are respectively the total number of different
classes in an impact factor, for example, there are four
building types (C2L-low, C2M-low, C2L-mod, C2M-mod)
whose model building typology is C2, so the of C2 should
be 4. In addition, the sum of PCc,b(j) is 1.
C. Human Losses
The estimation of the number of casualties, including
injured pupils and fatalities, requires not only the statistic data
on pupils of each county but also the average number of pupils
staying in different building types in the schools and the
percentages of pupils respectively in and out of school
buildings at different time periods (2:00 AM, 10:00 AM, and
5:00 PM in this case) of a day.
There are two methodologies provided in SELENA to
compute human losses: the basic methodology  and the
HAZUS methodology . In this study, the later one was
applied using the following equations .
Where Pkilled is the probability of a pupil being killed, and
C and P
Dare the probabilities of the four types of
damage states (slight, moderate, extensive and complete). PH
and PI are the probabilities of the complete damage state
without collapse and with collapse. While PE,PF,PG,PJ and
PK represent the probabilities that pupils could be killed in the
corresponding damage state in (8). ENoccupants killed is the
expected number of pupils being killed, and Noccupants killed is the
number of pupils in a building at the time of an earthquake.
D. Input and Output Files
The input files for this study included five sections, which
are showed and described in TABLE III. More detail
information about the input files of SELENA program can be
found in the technical manual of SELENA 14.1  and 
The SELENA generated five output files in this study.
xdout1.txt: the damage probability of each building
type (considering all three factors) in each county.
xgmotionscen1.txt: the ordinates of ground shaking
incorporated with soil amplification and its related
xnobctout1.txt: the numbers of different building
damage levels (slight, moderate, extensive, complete)
in each building type in each county.
xhlbyinjur1.txt: the numbers of pupil casualties in
different injury states (low, median, heavy and death)
at different times (2:00 AM, 10:00 AM and 5:00 PM)
of the day in each county.
xtotalinjur1.txt: the cumulative numbers of pupil
casualties at different times (2:00 AM, 10:00 AM and
5:00 PM) of a day in each county.
IV. RESULTS AND DISCUSSION
The building damages and humans losses in primary
schools of each county were estimated in this study. The result
of building damages is represented in the numbers of buildings
in each different building type (a total of 22 types in this study)
considering of four different damage levels (slight, moderate,
extensive and complete) and is partly shown in Fig. 3. It can
be seen in Fig. 3 that the school buildings in Dujiangyan,
Pengzhou, Shifang, Mianzhu, Wenchuan, Anxian, Beichuan
and Qingchuan may have a higher probability to be
completely damaged than those in the other seven counties.
Figure 3. Estimated percentages of damaged buildings at diffierent damage
states in different counties.
Fig. 4, for example, illustrates the estimated number of
injured pupils in different damage states in each county. The
value at the end of each bar is the total number of injured
pupils in the county corresponding to that bar. The result
obviously illustrates that there are relatively more injured
pupils and higher possibilities of death in Dujiangyan,
Pengzhou, Shifang, Mianzhu, Anxian, Jiangyou and
Qingchuan, to which more attentions need to be paid in
seismic rescue operations.
The indicated information from both results of building
damages and human losses can be combined with other data,
such as geographical maps, demographics and socioeconomic
statistical data, in rescue missions after an earthquake. For
example, the results of building damages can be incorporated
with the economic values, which can be the cost for repair or
replacement of buildings in different material classes and
occupancy types in each of the damage state, to estimate the
economic losses. Incorporated with a local topographical map,
the estimated number of human losses can be a guide for
Pkilled = PA ∙ PE+ PB ∙ PF+ PC ∙ PG+ PD ∙ (PH ∙ PJ+ PI ∙ PK)(8)
ENoccupants killed = Noccupants∙Pkilled. (9)
0% 20% 40% 60% 80%
Percentage of different levels of building damage
Slight Moderate Extensive Complete None
rescue groups to decide a proper rescue sequence of the
However, the uncertainties existing in the input data and
the models in themselves are unavoidable at present. Though
SELENA provides the “logic tree” method to compensate for
the multiple situations of inputs, due to the limitation of data
TABLE III. THE INPUT FILES APPIED IN THIS STUDY AND THEIR DESCRIPTIONS
Input Files Sources & Descriptions
This file provides information about the shakecenter and the weight of each shakecenter for the logic tree.
(i = 1).txt
This file contains the ground shakings (PGA, Sa0.3 and Sa1.0) and the soil type in the centroid location of a county. The
shapefiles of PGA, Sa0.3 and Sa
1.0 were downloaded from the Globle ShakeMaps from the USGS website:
This file provides information about the soilcenter and the weight of each soilcenter for the logic tree.
(i = 1).txt
This file represents the soil type in areas around the centroid of a county. The Vs30 information was first downloaded
from the Custom Vs30 Mapping of the Global Vs30 Map Server on the USGS website:
http://earthquake.usgs.gov/hazards/apps/vs30/custom.php, and then converted to soil types based on the IBC-2006 code.
This is the IBC
-2006 code, applied in this case, for the soil classification scheme. The SELENA software
incorporates the Eurocode 8 (Type
s 1 and 2)  and the Indian standard IS1893 .
This file provides the capacity datasets as well as information about the fragility curves and the weights of these capacity
datasets and fragility curves for the logic tree.
The first column contains the name of each capacity curve provided in the capcurves folder. The Degradation Factor s (κ)
from the 4th to 6th columns are from TABLE 5.18 in HAZUS . The elastic damping of a model building type is based
on the recommendation of Newmark and Hall , and it is discussed in the Assumption section of this paper.
These files are in the
Each file indicates the capacity curve, represented by the spectral displacement (in
[m]) and the
spectral acceleration (in [m/s2
]), of each model building type. In this case, the capacity curves of the example
Bucharest provided by NORSAR were applied.
This file provides the values of the fragility curve parameters. In the calculation of the structure or nonstructural damage,
it needs to be combined with its corresponding capacity curves.
The file contains the total number of different
building types in each county. These can be computed by (2) –(7).
The pupil distribution in each county at disparate time periods and occupancy types.
file contains information about the percentage of pupils staying inside or outside school buildings,
different times during a day (
night (2:00 AM), day (10:00 AM), and commuting (5:00 PM)).
The collapse rate of each building type in every county. In this case, the collapserate.txt of the example Bucharest
provided by NORSAR was used in the analysis.
The file contains information about indoor casualty rate for slight damage. In this case, the indcasrates.txt of the example
Bucharest provided by NORSAR
Indoor casualty rate for moderate damage. In this case, the indcasratem.txt of the example Bucharest provided by
NORSAR was applied.
The file contains information about indoor casualty rate for extensive damage. In this case, the indcasratee.txt of the
example Bucharest provided by NORSAR was used.
The file contains information about i
ndoor casualty rate for complete damage (without collapse
). In this case, the
of the example Bucharest provided by NORSAR was adopted.
The file contains information about indoor casualty rate for complete damage (with collapse). In this case, the
indcasratecc.txt of the example Bucharest provided by NORSAR was applied.
The file contains information about outdoor casualty rate for moderate damage. In this case, the outcasratem.txt of the
example Bucharest provided by NORSAR was used.
The file contains information about outdoor casualty rate for extensive damage. In this case, the outcasratee.txt of the
example Bucharest provided by NORSAR was applied.
The file contains information about outdoor casualty rate for complete damage. In this case, the outcasratec.txt of the
example Bucharest provided by NORSAR was adopted in the analysis.
(i = 1, 2, …, 5)
The percentage of pupils in each building type in every county for different occupancy types. Notice, i = 1 means
occupancy type is RESIDENTIAL, While, 2 means COMMERCIAL; 3 means EDUCATIONAL; 4 means
INDUSTRIAL; 5 means HOTEL. In this case, it was assumed that the percentages of pupils in occmbtp3.txt are the same
in each building type in every county and those percentages in occmbtp(1/2/4/5) are 0.
provides header data for generating the damage output files.
This is an input file with information to make a choice among three selections that are used in the calculations or results:
(a) the method for performance point; (b) the format of damage result; (c) the methodology for estimating human losses.
In this study, the CSM method, the number of damaged buildings and the HAZUS methodology were respectively chosen
for (a), (b) and (c).
Figure 4. The number of injuries in different damage states in each county.
only considered a single situation of each input in this study.
More logic tree branches such as different building capacity
curves and fragility curves, which are different among
buildings and locations, can be incorporated into the analyses
based on the ground truth situations in the study area. In order
to better estimate the results in this study, an evaluation
against the ground truth data will also be included in future
To estimate the seismic risk in schools of developing
countries, the probabilistic analysis procedures and the CSM
methodology of SELENA were applied to determine the
seismic hazard and building vulnerability functions in the
study. The building inventory data, which was the most
difficult to collect, was obtained from a combination of the
general statistical data from the Sichuan State Information
Network  and the percentages of each impact factor of the
building vulnerability from the findings reported in the
literature. The analysis results indicate that high probabilities
of complete building damage state and high death rates would
occur in Dujiangyan County, Pengzhou County, Shifang
County, Mianzhu County, Anxian County and Qingchuan
SELENA, based on HAZUS, was designed for seismic risk
estimation that can be applied in different parts of the world.
In this case, we attempted to apply it to the Wenchuan
earthquake in China as an example of its applications in
developing countries. Though it also requires detailed building
information such as model building typologies, the heights of
each building type and their seismic design code, which are
not easy to collect, the software allows the flexibility to
provide an estimation of losses associated with primary
schools because parameters in SELENA can be modified
based on the requirements of different users.
The authors wish to acknowledge the support of China
Scholarship Council (No. 200910220). They also would like
to express appreciations to several colleagues who provided
valuable comments and offered their kind assistance.
 U.S. Geological Survey, “Magnitude 7.9 – Eastern Sichuan, Chian,”
Significate Earthquakes of Earthquake Hazards Program,
details, accessed at March 1, 2011.
 Z. Wang, “A preliminary report on the Great Wenchuan earthquake,”
Earthquake Engineering and Engineering Vibration, vol. 7(2), 2008, pp.
 Sichuan Bureau of Statistics, “Sichuan Statistical Yearbook 2008,”
Sichuan State Information Network, http://www.sc.stats.gov.cn/,
accessed at March 1, 2011.
 D. Wu, Y. Xiong, J. Cui, and Q. Luo, “The analysis of building damage
and reinforcement in Wenchuan County caused by the may 12, 2008
earthquake,” Journal of Seismological Research, vol. 33(2), 2010, pp.
216-220 (in Chinese).
 J. Li, X. Lu, X. Li, X. Ren, W. Liu, and Y. Tang, “Seismic damage of
reinforced concrete frame structures in Wenchuan Earthquake,”
Structural Engineers, vol. 24(3), 2008, pp. 9-11 (in Chinese).
 Federal Emergency Management Agency (FEMA), “Multi-hazard Loss
Estimation Methodology Earthquake Model HAZUS®MH MR4 Technical
accessed at September 29, 2010.
 S. Molina, D. H. Lang, C. D. Lindholm, and F. Lingvall, “User Manual
for the Earthquake Loss Estimation Tool: SELENA,”
http://www.norsar.no/c-144-SELENA-RISe.aspx, accessed at December
 International Code Council (ICC), “2006 International Building Code
(ibc-2006),” Technical report, United States, 2006, pp. 664.
 European Committee for Standardization (CEN), “Design for Structures
for Earthquake Resistance, Part1: General Rules, Seismic Actions and
Rules for Buildings,” Technical report, prEN 1998-1:200X, Eurocode 8,
Brussels, Belgium, 2002, pp. 224.
 Bureau of Indian Standards (BIS), “Indian Standard – Criteria for
Earthquake Resistant Design of Structures, Part 1- General Provisions
and Buildins (5th revision),” New Delhi, India, 2002, pp. 39.
 Applied Technology Council (ATC), “Seismic evaluation and retrofit of
concrete buildings,” Technical report, Redwood City, California, 1996,
 S. Molina, D. Lang, and C. D. Lindholm, “SELENA – An open-source
tool for seismic risk and loss assessment using a logic tree computation
procedure,” Computer & Geosceinces, vol. 36(3), 2010, pp. 257–269.
 Coburn, A. and Spence, R., Earthquake Protection, Chicester second ed.,
John Wiley & Sons Ltd., West Sussex, England, 2002, pp. 420.
 N. M. Newmark and W. J. Hall, “Earthquake spectra and design.”
Technical report, Earthquake Engineering Research Institute (EERI),
Oakland, CA: EERI, 1982.
0 1000 2000 3000
Number of Injuried Pupils in Different Damage States