International Journal of Disaster Risk Reduction 20 (2016) 51–62

Contents lists available at ScienceDirect

International Journal of Disaster Risk Reduction

journal homepage: www.elsevier.com/locate/ijdrr

Multi-criteria location model of earthquake evacuation shelters to aid in

urban planning
crossmark
⁎
Jinghai Xua, , Xiaozhe Yina,b, Dingchao Chenc, Jiwen And, Gaozong Nied
a
College of Geomatics Engineering, Nanjing Tech University, 30 South Puzhu Road, Nanjing, China
b
Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, 11 Datun Road, Beijing, China
c
Jiangsu Institute of Urban Planning and Design, 88 Chaochangmeng Street, Nanjing, China
d
Institute of Geology, China Earthquake Administration, 1 Huayangjia, Beijing, China

A R T I C L E I N F O A BS T RAC T

Keywords:
Emergency facility location
Earthquake evacuation shelter
Geographic information system (GIS)
Multi-criteria modeling and optimization
Earthquake disaster mitigation
Earthquakes are serious natural disasters that can result in significant fatalities and economic loss. The building
of earthquake evacuation shelters is an effective way to reduce earthquake disaster risk and protect lives.
Current studies on facility location models commonly overlook multiple optimal criteria from an urban planning
perspective and are not suited to planning earthquake evacuation shelters. In this study, we describe seven
principles for locating earthquake evacuation shelters. Following these principles, we propose a multi-criteria
constraint location model that can be used to solve the location problem. We then present an iterative method to
solve the model. With the support of a geographic information system (GIS), the method is composed of three
steps: select candidate shelters, analyze the spatial coverage of candidate shelters and determine the shelter
locations. Finally, a case study is used to demonstrate the application of the multi-criteria model and the
corresponding solution method for its effectiveness in planning urban earthquake evacuation shelters. We
conclude that the evacuation shelter location model and solution method are effective and suitable to solve the
multi-criteria shelter location problem from an urban planning perspective.

1. Introduction
The majority of people worldwide now live in urban areas, and as
such cities are becoming more vulnerable to natural disasters, espe-
cially under the rapid development of the global economy [6].
Earthquakes are one such disaster that can cause severe damage to
cities. For example, the 2008 Wenchuan earthquake in China resulted
in 69,227 deaths, with 17,923 reported missing [34]. Similarly, 15,885
people died in the 2011 Tohoku earthquake in Japan. To mitigate the
damage caused by earthquakes, earthquake evacuation shelters have
been built in many cities. Since 1956, the Japanese government has
implemented special laws and standards to guide the building of
disaster evacuation shelters including the City Park Law, the Disaster
Resistance Plan and various design guides. In the great Hanshin
earthquake, more than 1000 city parks were used as earthquake
evacuation shelters; these areas provided valuable refuge for the
victims [19].
China has also experienced many serious earthquake disasters. The
great Tangshan earthquake in 1976 resulted in 242,769 deaths. After
that earthquake, the Chinese government began to realize the impor-
tance of urban planning in earthquake disaster mitigation. However,
the planning and building of earthquake shelters has been rather slow,
largely because of economic and technological reasons. In 2007 the
Chinese government released the National Standard for Urban
Planning on Earthquake Resistance and Hazards Prevention
(Standard No. GB 50413-2007). The standard states that earthquake
evacuation shelter planning is mandatory in China. In recent times,
China has experienced, on average, nearly one destructive earthquake a
year, for example, the 2008 Wenchuan earthquake, 2010 Yushu
earthquake, 2012 Yiliang earthquake, 2013 Minxian earthquake,
2014 Ludian earthquake and the 2014 Jinggu earthquake. These
events have fast-tracked the development of earthquake evacuation
shelter planning.
With the recent support for earthquake evacuation shelter plan-
ning, an earthquake evacuation location model from an urban planning
perspective is now required. The evacuation shelter location model is
essentially a facility location problem that has already been described
in a number of classical models: the p-median model, p-center model,
location set-covering problem (LSCP) and the maximal covering
location problem (MCLP). The p-median model supposes that the
facility number is determined and then finds the location of these
facilities to minimize the total (or average) service distance from all

⁎
Corresponding author.
E-mail address: xu_jing_hai@163.com (J. Xu).

http://dx.doi.org/10.1016/j.ijdrr.2016.10.009
Received 16 September 2016; Received in revised form 22 October 2016; Accepted 22 October 2016
Available online 26 October 2016
2212-4209/ © 2016 Elsevier Ltd. All rights reserved.
J. Xu et al.
demand points to service (facility) points [5]. The p-center model is
similar to the p-median model. It also supposes a fixed number of
facilities and finds the locations to minimize the maximal service
distance from facility points to demand points. The LSCP is a
mandatory covering model and its objective is to find the minimum
number of facilities to cover all demand points [15]. The MCLP model
aims to solve the problem by determining the facility number and then
finds the optimal location distribution to cover as many demand points
as possible.
These classical models are the foundation method with which to
describe and solve facility location problems. Many studies and
applications directly extend from them, for example Rahmaniani and
Shafia [26] extend the maximum covering facility location model by
considering its uncertainty and present a two-stage stochastic optimi-
zation location method. Jia et al. [21] extend the traditional model to
solve the location problem of a large-scale emergency medical service.
However, these models only consider a single optimal condition, that
is, either optimal service distance, minimum facility numbers or the
maximal coverage, which cannot be directly used for the location of
earthquake evacuation shelters.
Many studies tend establish a multi-criteria/multi-objective loca-
tion model to meet the requirements of more complex applications.
Nappi and Souza [24] present 10 criteria from qualitative and
quantitative perspectives for the location of temporary shelters. The
criteria provide a good reference for the building of a multi-criteria
location model in many application fields. Hadiguna et al. [14] study a
web-aided multi-criteria decision support system to assess public
facilities that can be used as earthquake and /or tsunami evacuation
centers. By considering multi-criteria and using a geographic informa-
tion system (GIS), Tsai and Yeh [30] study an optimal disaster
prevention evacuation location model. It provides the basis for local
governments to select evacuation shelters. Tamima and Chouinard [28]
develop an innovative multidisciplinary methodology, including road
conditions and distance between home and shelter, to estimate shelter
needs and evacuation scenarios when a major earthquake happens. Gu
and Wang [13] formalize the static and mobile facility location problem
as a bi-objective model and solve the model using a proposed
algorithm. This model expects to solve the static and mobile location
problem at the same time. Albacete et al. [1] conduct a similar study
that proposes a multi-criteria spatial analysis method in the home
searching process and maps suitable areas. Wood et al. [32] describe
multi-criteria decision analysis concept and the potential role of
geospatial modeling in solving problem of tsunami vertical-evacuation
siting. It put good foundation on further study of tsunami vertical-
evacuation shelters location.
Although these models contain multi-criteria, only a few are directly
suited to the location of earthquake evacuation shelters, especially from
the perspective of urban planning. Some criteria are important to
planning earthquake evacuation shelters, for example, the relationship
between population capacity and the service area (spatial coverage) of
shelters and selecting candidate sites (more criteria are listed in Section
2.2). However, few current studies simultaneously consider these
criteria.
Many multi-criteria models only provide a theoretical presentation
of the problem and are difficult to be solved and applied. The solution
to the problem of facility location is a difficult issue, and is generally
considered an NP (non-deterministic polynomial)-hard problem. The
use of heuristic algorithms is considered a good way to solve this
problem, including genetic algorithms, Lagrangian relaxation algo-
rithms, greedy algorithms, Tabu search algorithms and ant colony
algorithms [18]. These algorithms provide good solutions in many
applications; for example, Indriasari et al. [16] use three heuristics,
genetic algorithm (GA), tabu search (TS) and simulated annealing (SA)
to obtain optimization with the maximal service area problem. Babaie-
Kafaki et al. [2] propose modified hybridization of genetic algorithms
with neighborhood search based mete-heuristics to solve facility
52
International Journal of Disaster Risk Reduction 20 (2016) 51–62
location problems. They test the algorithm on large scale fuzzy bus
terminal location problems and get good results. Comber et al. [4]
describe a modified grouping genetic algorithm to solve the facility
location problem and present its application in the area of emergency
medical services. Saeidian et al. [27] compare the performance of GA
and bees algorithm (BA) in finding the optimum location of earthquake
temporary relief centers and conclude the GA is better than the BA in
simplicity and repeatability perspectives.
However, regarding the different multi-criteria restrictions in
planning earthquake evacuation shelters, these solutions may not
directly suitable to the present study. Furthermore, some of these
methods overlook the spatial distribution of facility sites and some of
them are too complex to realize in practical applications. In recent
years GIS has been used in the solving of location problems. GIS
possesses the unique advantage in dealing with a wide range of spatial
data analysis problems and decision-makers can always choose the
most suitable data modeling and analysis method in GIS to solve all
kinds of problems [29]. Some studies have achieved good results
addressing the facility location problem using the spatial functions of
GIS. Çetinkaya et al. [3] use GIS and fuzzy analytic hierarchy process
method in their study to determine the priority of all the potential
refugee camp sites and select the more suitable ones for Syrian
refugees. Li and Yeh [17] integrate genetic algorithms with GIS to
solve the problem of optimally selecting facility sites. Their study
obtained good results for optimal location searches under complex
situations. Marcelin et al. [22] adopt a p-median based modeling
framework together with GIS to explore where to site relief distribution
facilities after a potential hurricane happens in Leon County, Florida.
Esmaelian et al. [9] propose a multi-criteria spatial decision support
system that integrates GIS and a multi-criteria decision-making
method to identify evacuation shelters and emergency service loca-
tions. Elheishy et al. [8] describe a rough set approach combined with
GIS to solve evacuation shelters location problem. They present the
application of the method in Masoura, Egypt.
The location of earthquake evacuation shelters for urban planning
is a multi-criteria optimal location problem. This study is different
from previous research in that it provides detailed multi-criteria from
an urban planning perspective, and provides a corresponding simply
solution supported by GIS with a case study to demonstrate its
application.
The remainder of the paper contains the following sections: Section
2 presents a multi-criteria location model for earthquake evacuation
shelter location and Section 3 describes the GIS support solution to the
location model. In Section 4, we provide a case study to demonstrate
the application of the model and the related solution method. Finally,
in Section 5, we offer conclusions and directions for future research.
2. Model representation
2.1. Classification of earthquake evacuation shelters
Urban earthquake evacuation shelters are safe places that provide
evacuation protection and basic life support services after an earth-
quake. Various studies have classified evacuation shelters into different
categories. Dunn and Newton [7] classify them into three categories:
emergency shelter, fixed shelter and central shelter. Tsai and Yeh [30]
present a further three categories: emergency evacuation/refuge bases,
regional refuge shelters and city refuge shelters.
Based on the Chinese national standard GB 50413-2007, we classify
the earthquake evacuation shelters into three categories: temporary
shelter, fixed shelter and central shelter. Their characteristics and
planning requirements are listed in Table 1.
The first type of evacuation shelter is a temporary shelter. Because
it is only used for temporary evacuations, it is not necessary to find a
specific quantitative model for its location. Some guidance principles
will be effective regarding its location. Furthermore, the number of
J. Xu et al. International Journal of Disaster Risk Reduction 20 (2016) 51–62

Table 1
Planning and building requirements for earthquake evacuation shelters.

Type Duration of stay for Area requirement Effective area per Radius of service Travel time by Necessary facilities
refugees (hectare) population (m2) area (km) walking (minutes)

Temporary Temporary (generally less ≥ 0.1 ≥1 0.5 ≤ 10 No requirement

emergency shelter than 1 day)
Fixed shelter Short to medium stay 1–20 ≥2 2 ≤ 60 Basic living facilities
(generally less than 1
month)
Central shelter Long-term stay (generally ≥ 20, generally more ≥2 Referred to fixed Referred to fixed Basic, general and
less than 3 months) than 50 shelter shelter comprehensive living facilitates

central shelters in a city usually ranges from one to several because of
their large size. They are generally located in large vacant areas away
from the city center such as suburban areas. Because of the limited
candidate areas, the location of central shelters is not a complex matter
in China. However, the location of fixed shelters must meet multi-
objectives/criteria requirements including to build a minimal number
of shelters in a city, cover a maximal city territory range, service as
many people as possible and each shelter has a limited population
capacity. Thus, a multi-criteria constrained model is required to
achieve overall optimization for the shelter's location, which is the
objective of this study.
should try to maximize the spatial coverage of all shelters within the
planning region. The theoretical maximum coverage of all shelters is
equal to or even greater than the spatial coverage of the planning
region. However, this is not necessary because there are always some
areas with lower population density (or even no population) within the
planning region, for example lake and mountain regions.
P7: Population capacity principle. The population capacity of each
shelter is limited by the effective area of the shelter. Furthermore, the
population served by a shelter service area should be no more than its
capacity. Although this principle is obvious, it is sometimes deliberately
overlooked for the sake of simplifying the study and application [18]. In
this study, we will follow this principle.

2.2. The seven principles of planning earthquake evacuation shelters

2.3. Multi-criteria constraint location model

Looking at Liu [20], Nappi and Souza [24] and the Chinese national
The planning principles provide the foundation for the location of
standard GB 50413-2007, we summarize the principles governing the
earthquake evacuation shelters. We designed a multi-criteria constraint
location of earthquake evacuation shelters from an urban planning
location model to meet the requirements of these principles. The model
perspective.
is derived from a p-median model and location set covering the
P1: Safety principle. This is the most important principle. It means
problem model, and is formulated as Eqs. (1)–(8).
that candidate shelters should avoid any potential risk areas such as
n n
fault lines, liquefaction risks and chemical warehouses.
P2: Control of land use principle. In China, the government owns all
min ∑ ∑ t ij yj ,
i =1 j =1 (1)
the land—essentially there is no private land ownership. However, the
n n
land use right belongs to different companies, individuals or organiza-
tions. Thus, candidate shelters can only be selected from land parcels
max ∑ ∑ aixij ,
i =1 j =1 (2)
where their use rights are controlled by the government; this is denoted
by the land use type. Generally, candidate shelters will be selected from max∑ Aj ,
the following land: green spaces and educational land (e.g., city parks j=1 (3)
and schools). Another reason why these areas are appropriate is that
essential services (i.e., electric power, clean water) are needed in a subject to ∑ xij ≥ 1, i ∈ V ,
j ∈ Wi (4)
shelter, and parks and schools typically meet these requirements with
little or no rebuilding. n

P3: Nearest evacuation principle. When an earthquake occurs, the
shelters need be near to residential areas, and it should be easy for
residents to find the earthquake evacuation shelters. This principle also
means that the spatial distribution and number should be adapted to
suit the population density.
P4: Economic principle. The cost of building shelters is met by the
government. The spatial distribution and number of shelters should be
optimal to save government investments.
P5: Appropriate travel distance/time principle. People should only
have to travel a short distance/time to the shelters, and this can
improve the likelihood of a successful evacuation to shelters. However,
this also means that more shelters and a higher budget are required.
The travel distance/time from the demand point to the shelters should
be reasonable. Thus, an appropriate travel distance is required to
balance the budget and evacuation ratio. The Chinese national stan-
dard provides a clear rule that are the service distance of fixed
evacuation shelters should be no more than 2 km, which is about an
1-h walk.
P6: Maximum coverage principle. In this study, the maximum
coverage of a single shelter has already been determined by the Chinese
standard—no more than 2 km. Thus, the principle means that we
popj ≥ ∑ ai xij ,
i =1 (5)
dij ≤ tij ≤ rj , (6)
∑ Aj ≥ Aall ⋅δratio, and
j =1 (7)
xij , yj ∈ {0, 1}, j ∈ W , i ∈ V , (8)
where i is the index of demand points, j is the index of available shelters
and tij is the evacuation distance from demand point i to the shelter site
j. Furthermore, yj is the binary variable, equal to 1 if and only if a
shelter is located at site j and ai is the population of demand point i .
Additionally, xij is a binary variable, equal to 1 if and only if demand
point i is covered by site j, and Aj is the service area of site j. We use W
to represent the set of all available earthquake evacuation shelters, Wi is
the set of the shelters whose service area covers demand point i and V
indexes the set of demand points. The population capacity of site j is
denoted by popj , dij is the linear distance from demand pointi to the
shelter at site j and rj is the maximum service range of site j. Last, Aall is
the planning region, and δratio is the minimal threshold of coverage ratio
between the service area of all sites and the planning region. The

53
J. Xu et al.
Table 2
The objectives and relationships between the multi-criteria model and location
principles.
Equation Objectives
Eq. (1) Minimize the total distance between the population
that requires evacuation from homes/workplaces
and the shelters
Eq. (2) The shelters should cover as many people as possible
Eq. (3) Maximize the coverage of all shelters
Eq. (4) Each demand point is covered by at least one shelter
Eq. (5) Sum of the population of demand points covered by
site j should be no more than the population capacity
of this site
Eq. (6) The distance between the demand point and its
service point (shelter) should be no longer than the
service point's maximum service range
Eq. (7) The service area of the shelters should meet the
minimal coverage requirement; δratio may be
different in different planning regions, which is
controlled by a city's overall planning
Eq. (8) This is the auxiliary equation for Eqs. (1–7)
objectives of Eqs. (1)–(8) and their relationships with the location
principles are listed in Table 2.P1 and P2 are used to select candidate
shelters in which some will be formally selected and form W in Eq. (8).
3. GIS support solution to the multi-criteria model
Eqs. (1)–(8) represent the theoretical model to describe the
evacuation shelter location problem, and a method to solve this
problem is required to put it into practice. Heuristic algorithms are
widely used to provide optimal solutions for the facility location
problem [25]. However many solutions are complicated and suit for
special models. The Lagrangian heuristic algorithm is one of the most
commonly used heuristic algorithms. It improves the solvability of a
model by reducing its constraint conditions, absorbing difficult con-
straints into the objective functions and transforming the objective
functions into easier liner functions [12]. In this study, using a
Lagrangian heuristic algorithm and GIS, we provide a simply proximity
solution method for this model.
3.1. Overview of the solution
As shown in Fig. 1, the solution has three steps—the first is selecting
candidate sites for shelter location. The optimization of the spatial
distribution of the shelters is an important objective in the facilities’
location [23]. It is also one of objectives in this study (P3). Although
this objective may be easy to realize in urban planning (because the
locations are flexible and can be determined by the planners), we still
need to bear in mind that the locations of the shelters are also
constrained by P1 and P2. Thus, the first step is to find candidate
sites that are suitable for shelters. In this step, equations are built to
estimate the approximate number of candidate shelters and the priority
level of a site to be selected as a shelter.
In the second step, we analyze the spatial coverage (service area) of
each candidate shelter, one by one, with the support of GIS spatial
analysis function. The analysis of spatial coverage will not only
consider the influence of the road network but also the population
capacity of the candidate shelter. In the third step, the final shelters will
Step 1:Selecting the candidate shelters
• Estimate the approximate number
of candidate shelters
• Determine the priority of each
candidate shelter
Fig. 1. The three steps of the solution algorithm.
International Journal of Disaster Risk Reduction 20 (2016) 51–62
Related
principles
P3
P4
P6
P3
P7
P5
P6
P1 and P2
Fig. 2. Ideal distribution and service area of shelters.
be selected, determined by their total spatial coverage ratio in the
planning region and spatial distribution. Thus, the detailed process is
an iterative process, and is shown in Fig. 8.
3.2. Selecting the candidate shelters
3.2.1. Estimate the approximate number of candidate shelters
The number of candidate shelters is a rough input value for iterative
calculation. To simplify the estimation process, we make a number of
hypotheses for each candidate shelter (Fig. 2):
(1) Each candidate shelter has unlimited population capacity, which
means it can contain all the people within its coverage area.
(2) The total coverage ratio between the coverage of all shelters and
the planning area is 100%.
(3) Each candidate shelter in the planning region has an equal spatial
distribution.
Thus, the number of candidate shelters will be estimated as Eq. (9)
N0=Aall /2πR, (9)
where N0 is the number of candidate shelters, Aall is the area of
planning region, R is the maximum coverage of the candidate shelter,
which should be 2 km as based on the Chinese national standard.
The service area of these candidate sites will overlap and it will be
difficult for their service radiuses reach the maximum length for the
limited population capacity of the site. We adjust N0 with coefficient k
to estimate the number of candidate sites N , shown in Eq. (10). Based
on experience, we suggest k ranges from 3 to 6.
N = kN0 . (10)
3.2.2. Determine the priority of each candidate shelter
In accordance with P1 and P2, the actual location of the candidate
Step 2: Analysis the spatial coverage Step 3: Determination of
of single candidate shelter shelter location
• Union service area of each
• Building road network dataset candidate shelter
• Service area analysis • Calculate coverage ratio
54
J. Xu et al.
Table 3
Influencing factors of site priority.
Factor Explanation
Location condition This factor relates to P1. A site that has high priority
should be near to the population and far from risk
buildings and facilities such as gasoline stations,
chemical factories and fault lines.
Building (site) type and Sites that have a high priority should be leisure/
current usage entertainment green space or wide open spaces such
as parks, squares, schools and stadiums.
Reconstruction A priority site should be one that is easy to transform
into a shelter; for example, simple to access water and
electricity, with good vehicle access.
Effective area A site considered high priority should be a suitable
size (i.e., large enough). The effective area should
exclude lakes and steep mountains.
Anti-seismic quality of A building of high anti-seismic quality is considered a
building high-priority site.
shelters will not have an equal spatial distribution, as they are
determined by various constraints. Furthermore, in urban planning,
each candidate site must undergo a site visit by the planners to ensure
it is suitable for a shelter.
The candidate shelter can be an existing building within the
planning region (i.e., schools and parks) or planned buildings within
the region. Generally, there are many land parcels (or buildings on
these parcels) that are suitable for shelters; an evaluation criterion is
then required to determine the priority of different land parcels. A
comprehensive evaluation method is used in this study as shown in Eq.
(11), in which five factors are selected to evaluate the suitability of a
site. Thus, different candidate sites have different scores, and the site
with highest score is more likely to be selected as formal shelter.
5 shown in Fig. 4. According to P7, the population covered by the service
Q= ∑ αfi ,
i =1
where Q is the score of the candidate site, α is the weight of each factor
and fi (i = 1, 2…,5) is the single score of each factor of a candidate site.
The standard to evaluate these factors are listed in Table 3. The
planners evaluate the score of each factor for a candidate site, based on
their opinions.
3.3. Analysis of the spatial coverage of candidate shelters
Fig. 2 shows the ideal service area of earthquake evacuation
shelters. In reality, the service area will not be a circle, and the
maximum radiuses of the spatial coverage (service area) of different
shelter sites will differ. This will be influenced by two factors: the
population capacity of the candidate shelter and the travel distance
(time) from the demand point to the candidate shelter.
According to P2, the travel distance should be no more than 2 km.
Furthermore, the distance from the demand point to the candidate site
must be the actual distance via roads/footpaths, not the straight-line
distance as shown in Fig. 2. According to P7, the population within a
shelter's service area is considered to be the service population of that
particular shelter. That is, the service population should be no more
than the shelter's population capacity. Thus, the size of the service area
is also related to the population distribution in the planning region.
3.3.1. Building a road network dataset
GIS provides a spatial analysis tool to conduct an analysis of the
service area of a facility site. The road network is the basis with which
to use this tool, as shown in Fig. 3. We use ArcGIS software to build the
road network. There are two network models in ArcGIS: geometry
network and network dataset [35]. The first is widely used in the
representation and analysis of utility networks such as electronic and
drinking water networks. The second model is commonly used for road
International Journal of Disaster Risk Reduction 20 (2016) 51–62
network analysis. A network dataset that includes lines, points and
turns is well suited to service area analysis [10].
In China, Autodesk CAD is widely used in urban planning.
Therefore, in most cases, road data in a planning region is stored
using the AutoCAD format and cannot be used directly to build an
ArcGIS road network dataset. Thus, we had to transform the road
network data from the AutoCAD format. There are two key steps in this
process: (1) the ESRI shape file format is selected as an interchange-
able format to transform geometric information; and then we (2) set
the “road length” as the road network “impendence” attribute when
building the ArcGIS network dataset; therefore, the road length will be
used in the service area analysis.
3.3.2. Service area analysis of each candidate shelter
The service area of a candidate shelter is determined by its capacity,
actual road distance and population distribution. The capacity of a
shelter is calculated using Eq. (12):
⎧ Ae =λAall
⎨ ,
⎩ popcapacity =Ae / Ap (12)
where Ae refers to the effective refuge area of the shelter, Aall is the total
area of the candidate shelter and λ is the effective ratio, which means
not all of area in a shelter can be used. For example, a lake within an
evacuation park (shelter) cannot be used to house refugees.
Furthermore, λ is generally determined by previous urban planning,
popcapacity is the population capacity of the shelter and Ap is the per
capita refuge area (2 m2 in accord with the Chinese national standard).
The maximum service radius of a single site cannot be more than
the threshold rc (2 km). For example, the cumulative service length OA-
AB-BC cannot be more than rc (O is the proposed candidate site) as
area of the site should not exceed its capacity.
(11)
Therefore, population data is also important in the service analysis,
and are sourced from census data. The census generally only provides
the total population in a census tract. To calculate the population
covered by the service area of a candidate site, we suppose that within
the census tract, the distribution of the population is spatially equal.
Thus, we use Eq. (13) to calculate the population within the service
area:
Popsa=(Popct / Act ) Asa (13)
where Popsa is the population covered by the service area of a candidate
shelter, Asa is the area of the service area of the candidate shelter within
the census tract and Act is the area of the population census tract.
Furthermore, Popct is the total population of the census tract, as shown
in Fig. 4. When the covered population of a shelter service area is
greater than its capacity, we have to reduce its coverage to decrease the
service area within the census tract. Therefore, the covered population
will decrease until it is less than its population capacity.
With the support of the ArcGIS network dataset data model, ArcGIS
provides an ArcGIS service area analysis function in its extension
module. This function can generate the service area of a facility site
according to the input radius length of the service area.
Thus, we iteratively use the ArcGIS service area analysis function to
obtain the service area of a single candidate shelter. The analysis
process is shown in Fig. 5. In this process, the road distance from
demand points to the candidate site is iteratively decreased according
to the population capacity of the site and the population distribution.
The start value is the maximum service radius (rmax =2 km). If the
population ( popsa ) covered by the candidate site is more than its
capacity ( popcapacity ), then we will decrease the distance with an iterative
step (rstep ), that is, 20 m, until the capacity of the site is equal to or
greater than the population within its service area. Fig. 7a presents an
example of the service area of a site.
55
J. Xu et al. International Journal of Disaster Risk Reduction 20 (2016) 51–62

Fig. 3. Road network data in ArcGIS network dataset format.

3.4. Determination of shelter location

Fig. 4. Population tract and service area of candidate site.
Fig. 5 shows that the service area of each candidate site can be
analyzed successively according to its priority. However, when more
than two service areas are generated, we need to ensure the union of
the two and their optimization. Two GIS spatial operators will be used
in this process: union and contain. We suppose that Sai is the service
area of candidate site i and Saall is the union result of the service areas
of all candidate sites. The process can then be described by the c++
pseudo-code as shown in Fig. 6. The example in Fig. 7 explains the
process.
Therefore, the service areas of all the candidate sites can be
generated by the union and contain operators. The coverage ratio
between the candidate sites and the planning region can be calculated
using δ = Asa / Aall , where δ is the coverage ratio, Asa is the area of the
united service area of all candidate sites and Aall is the area of the
planning region.
We suppose that δt is the threshold of the coverage ratio, which is
determined by a city's general urban planning. If δ<δt , a new candidate
site (i+1) will be added to the location analysis until the threshold
condition is met; the location analysis will then be finished. The
complete analysis process is presented in Fig. 8.

Fig. 5. Flow chart of the service area analysis of a single candidate site.

Fig. 6. c++ pseudo-code of the union and optimization of service areas.

56
J. Xu et al.
Fig. 7. Example of the union and optimization of service areas: (a) Saall =Sa 0 ; (b) if Saall does not contain Sa1 then Saall =Saall uionSa1; (c) the result of the union operation; (d) if Saall
contains Sa2 , then delete candidate site 2 and join with Sa2 .
4. Case study
4.1. Background
The model and its corresponding solution method have been used
for the city of Yangzhou, for the planning of urban earthquake
evacuation shelters. Yangzhou is located at 32.3942°N, 119.4129°E
and is famous for its history and tourist attractions. In 2015
Yangzhou's GDP was 401.68 billion RMB, with a population of
approximately 4,459,760 (2010 census). With the rapid development
of its economy, Yangzhou is facing a high earthquake disaster risk. The
city has a large seismic fault (the Yangzhou–Tongling seismic fault)
running through it, and 59 earthquakes with a magnitude greater than
4.7 have been recorded. The most recent earthquake of such a
magnitude (4.9) occurred on July 20, 2012. A Yangzhou government
department, the Yangzhou Urban and Rural Construction Bureau
(YBRCB), have undertaken in recent times the planning and building
of earthquake evacuation shelters. Such efforts are an important way to
mitigate the effects of a serious earthquake.
In this study, the total area of the planning region of Yangzhou is
415.5 km2 . The thresholds used in this study are sourced from the
Seismic Prevention and Earthquake Disaster Mitigation Planning of
57
International Journal of Disaster Risk Reduction 20 (2016) 51–62
Yangzhou Central Region, which is published by the YBRCB. The
threshold δt coverage ratio between the service area of shelters and the
planning region is 60%. The value of δt is not large because much of the
area within the planning region is considered non-constructive land,
which means that there are no inhabitants and therefore no shelters are
required. As shown in Fig. 9, the land within the planning region is
categorized into three classes (non-constructive land is not included in
any of these categories).
4.2. Results
According to the model and methodology described in Sections 2
and 3, we first determine the candidate sites. The number of candidate
sites is estimated using Eq. (10). We set k=4.5 based on prior
experience, representing the mean range. Thus, the number of
candidate sites is 72. Each candidate site was then investigated (site
visit) and its priority evaluated in accordance with Eq. (11). The spatial
distribution and images are shown in Fig. 10.
The road network of central Yangzhou was built and is shown in
Fig. 10. We developed a program to realize the solution method using
Visual Studio 2010C# and ArcEngine 10.1 with regard to the metho-
dology described in Section 3 (Fig. 11). With the support of the
J. Xu et al.
Fig. 8. Process to determine the location of earthquake evacuation shelters.
program, 61 earthquake evacuation shelters were determined. The
distribution and service area of these shelters are shown in Fig. 11, and
detailed information regarding these shelters is shown in Table 4. Most
of the shelters are in the playgrounds of schools and parks, as these
sites provide existing utilities/facilities such as electricity and water.
Although some shelters are in school or parks yet to be built, all sites
are carefully investigated (via site visits) to ensure they are suitable for
an earthquake evacuation shelter.
The united service area of all shelters is 258.9 km2 and the total
coverage rate is 62%. The planning results have been reviewed by a
Fig. 9. Land use of planning region.
58
International Journal of Disaster Risk Reduction 20 (2016) 51–62
group of experts; their fields of expertise include urban planning,
earthquake risk mitigation, geography and urban management. Once
the results were accepted by the expert review group, the results were
released by the YBRCB. These will guide the construction of earthquake
evacuation shelters in Yangzhou in the future.
4.3. Discussion
The urban planning of earthquake evacuation shelters is an
important way to mitigate the damage caused by a significant earth-
quake, which is the responsibility of the government. Generally this
task is led by the urban and rural construction bureau of Chinese local
government. But the decision-makers of the government departments
are not good at the technical issues of the planning and the task will be
assigned to the related professional firms by the government.
However, comparing with traditional location problems, a multi-
criteria/multi-objective process brings with the evacuation shelters
location a number of challenges. An appropriate location model is
essential to solve the problem. In this study, seven principles from an
urban planning perspective have been proposed to describe the multi-
objectives of the location of earthquake evacuation shelters. We then
presented a multi-criteria location model to represent the location
problem. The model can be used by urban planning professionals and
related government decision-makers.
A simply solution method for this model is provided by using an
iterative solution algorithm. Three steps of the solution describe how to
use the multi-criteria location model. Since the solution method is
supported by the GIS, it can be programmed as a software package as
displayed in the Yangzhou case study, which makes the implement of
the model is convenient and easy. The location results of the case study
in Yangzhou meets the planning requirements, which demonstrate the
usefulness and effectiveness of the model and solution method.
The method used to solve the problem is an approximate algorithm
providing local optimization. Therefore, the service areas of the final
location of the evacuation shelters still have unnecessary overlaps, as
shown in Fig. 11. The operator in the final step of the solution (i.e., the
deletion of overlapping candidate sites) can be further optimized by
J. Xu et al. International Journal of Disaster Risk Reduction 20 (2016) 51–62

Fig. 10. Spatial distribution of the candidate sites.

considering more complex conditions.
Furthermore, the road network model can be enhanced by con-
sidering the effect of building collapse in an earthquake. The travel time
will become longer when buildings surrounding a road collapse.
Therefore, the travel distance (road length) becomes longer, and this
needs to be considered in a further study.
In this study, the population was equally distributed within a census
tract (see Fig. 12a). Eq. (13) is used to calculate the population covered
by the service area. However, this assumption may be inappropriate
because populations are not evenly distributed throughout census
tracts. Generally, people tend to live around main settlement points
(e.g., village or city center). Fig. 12b displays the real distribution of the
population in the whole census tract. The use of a kilometer-grid
format to determine the population looks promising to solve this
problem [31]. In contrast to traditional population data collected
within a census tract, this method provides the population distribution
in a kilometer grid or a high-resolution grid (i.e., 250-m grid), as shown
in Fig. 12c.

Fig. 11. Main interface of the earthquake evacuation location program.

59
J. Xu et al. International Journal of Disaster Risk Reduction 20 (2016) 51–62

Table 4 Table 4 (continued)

Detailed information regarding location results.
No. Population Site name Effective Radius of Existing
No. Population Site name Effective Radius of Existing capacity area of the maximum or
capacity area of the maximum or site (m2 ) service planned
site (m2 ) service planned range (m) site
range (m) site
III-4-3 4590 Sports fields of 13,770 1040 Existing
I-1-1 5310 Playground of 15,930 1920 Existing Jiangdu Sport
Pingshan Center
Experimental III-5-5 11,460 Planned 40,110 1540 Planned
Middle School Shuntian Park
I-2-6 4000 Playground of 12,000 1060 Planned III-5-2 5811 Playground of 20,340 1100 Existing
planned S02 Jiangdu Middle
High School School
I-2-7 3333 Playground of 10,000 860 Planned III-4-4 14,025 Jiangduyucai 28,050 1280 Existing
planned H16 Middle School
Primary School III-6-1 24,606 Planned 86,120 2000 Planned
I-2-1 11,667 Zuxi Park 35,000 1340 Existing Jiangping Park
I-4-1 17,325 Playground of 34,650 1260 Existing III-7-1 8000 Planned 228,000 1100 Planned
middle school Bingjiang green
affiliated to square
Yangzhou III-1-1 6000 Planned Hanji 18,000 2000 Planned
University Park
I-2-5 4000 Playground of 12,000 860 Planned V-1-1 81,400 Planned Xinbei 244,200 2000 Planned
planned Meiling lake Park
Primary School V-3-1 3000 Jiangdu Suma 9000 1700 Existing
I-2-3 4680 Playground of 14,040 880 Existing Middle School
Zuxi Middle IV-2-1 44,380 Yangzi jing Park 221,900 1120 Existing
School IV-6-1 14,250 Playground of 42,750 920 Existing
I-3-1 71,600 Songjiacheng 414,800 620 Existing Jinghai School
Park I-3-3 15,080 Planned Lotus 30,160 2000 Planned
I-3-2 2343 Yangzhou First 7030 880 Existing Pool Park
Middle School I-3-4 11,120 Meiling Middle 22,240 2000 Existing
I-4-4 5907 Yunhe Park 17,720 2000 Existing School
I-4-2 26,433 Qujiang Park 79300 1220 Existing 1-3-5 6540 Meiling Primary 13,080 2000 Existing
I-5-2 9870 Planned Wenfeng 19,740 1780 Planned School
Park I-7-1 20,000 Guifang Shahe 60,000 660 Existing
I-5-1 20,829 Planned 145,800 800 Planned Park
Sanwang Park I-7-2 16,980 Green land parcel 33,960 1200 Planned
I-5-3 7833 Planned Yuejing 23,500 620 Planned of east Jinghan
River Park River
II-6-3 5540 Planned park 19,390 760 Planned I-6-2 24,600 Green land parcel 73,800 2000 Planned
II-6-4 3429 Playground of 12,000 640 Planned on the
planned K02 intersection of
Middle School Zhuyuwang Road
II-6-2 11,629 Diehu Park 40,700 1340 Existing and a planned
II-2-3 6480 Planned street 12,960 1400 Planned road
park III-1-2 5640 Hangji Middle 11,280 2000 Existing
II-2-1 3429 Playground of 12,000 720 Planned School
planned III-1-3 12,690 Hangji sanxiao 25,380 1060 Existing
Jiangwang Primary School
Middle School III-3-6 24,100 Green land parcel 72,300 1620 Planned
IV-5-1 19,320 Planned Chahe 67,620 2000 Planned of west Jiangdu
Park Sewage Plant
IV-1-1 26,754 Yunxi Lake Park 93,640 2000 Existing III-5-3 30,000 Mangdao River 90,000 700 Existing
II-6-1 3140 Green square in 10,990 660 Existing Park
Hanjiang District II-5-1 11,940 Cuigan Middle 23,880 2000 Existing
II-5-3 17,194 Yingchao River 60,180 1680 Existing School
Park II-3-2 12,240 Weiyang Middle 24,480 3000 Existing
II-5-6 5786 Playground of 20,250 960 Existing School
Xinhua Middle II-3-3 3705 Weiyang Primary 7410 3000 Existing
School School
II-5-7 5557 Lianzheng 11,115 1000 Existing II-1-1 35,600 Yangzhou Middle 71,200 2000 Existing
Square School Western
II-5-2 8170 Laiquetai Square 28,595 1000 Existing Campus
II-4-1 18,789 Planned 65,760 2000 Planned IV-4-1 16,500 Guazhou Middle 33,000 2000 Existing
Yueliangwang School
Park II-1-2 48,118 Tiyu Park 240,590 2000 Existing
II-3-1 36,634 Sugan Park 128,220 2000 Existing II-4-2 20,000 Green land park 60,000 2000 Planned
I-6-1 42,743 Planned 249,600 2000 Planned near northern
Liaojiagou Park Runyang road
III-2-1 18,000 Xiancheng 36,000 1540 Existing
Middle School
III-3-2 7065 Playground of 14,130 860 Existing 5. Conclusion
Dujiang Middle
School
III-3-1 11,480 People Park 34,440 1,280 Existing The effects of a serious earthquake can be catastrophic, especially if
III-4-1 6300 Fudu Square 12,600 920 Existing occurring in or close to an urban area. The planning and building of
(continued on next page) earthquake evacuation shelters is an important measure to mitigate the

60
J. Xu et al.
Fig. 12. Spatial distribution of population data; the points represent the population distribution. The area surround by the grey boarder represent the census tracts and the area
surrounded by the red boarder is the service area of a shelter; a equal population distribution inside the census tract; b actual distribution according to settlements; c gridded
distribution (Adapted from Xu et al. [33]).
effects of natural disasters. However, such actions are made more
difficult by multi-criteria restrictions; thus, a multi-objective location
model is required to guide these type of projects. Classical location
models (i.e., p-center model, p- median model, LSCP and MCLP) are
single-criteria optimal models and cannot meet the requirements of
urban planning.
Seven principles have been introduced (i.e., safety, control of land
use, nearest evacuation point, economic constraints, appropriate travel
distance/time, maximum coverage and population capacity) to guide
the planning of earthquake evacuation shelters, and these are in
accordance with the Chinese national standard GB 50413-2007
(Standard for Urban Planning on Earthquake Resistance and
Hazards Prevention) and other studies. Under the constraint of these
principles, a multi-criteria location model for earthquake evacuation
shelters is proposed, extending the p-median model and location set-
covering problem model.
An iterative solution method to the multi-criteria model is explored
based on GIS. The solution method includes three main steps: selecting
candidate sites for shelter location, analyzing the spatial coverage of
single candidate shelters, and determining the final shelters according
to the spatial coverage ratio of the total service area of the shelters and
the planning region.
The solution method is simply and easy to be realized. Then a
corresponding location program was developed with ArcEngine and
Visual Studio 2010C# to achieve a solution to the problem, and the city
of Yangzhou was selected to demonstrate the application of the model
and location program. The number and spatial distribution of earth-
quake evacuation shelters in Yangzhou were analyzed and determined.
The planning results have been studied and accepted by an expert
panel and now have been adopted by a Yangzhou government depart-
ment (YBRCB).
We conclude that the proposed location model and corresponding
solution method are useful in the urban planning of earthquake
evacuation shelters. The model has the advantage of optimizing the
multiple location criteria from an urban planning perspective, in which
seven urban planning principles were considered. A GIS-supported
iterative method was explored to solve the model. Compared with other
solving methods, this method not only provides a solution to the model
but also considers the spatial distribution of the shelters (including
areas that can be selected as candidate sites) and road networks.
Furthermore, the method is easy to establish and operate.
Regarding future research, a more sophisticated location model and
solution method should be studied. Further criteria can be investigated
including the influence of earthquakes on the road network. It would
also be worthwhile to investigate a more reliable evacuation route from
demand points to shelters [11]. Population data is also important in the
selection of evacuation shelter location, as the urban population can
change rapidly throughout a 24-h period (e.g., daytime and nighttime
populations). An accurate population distribution model and data will
be very useful in a location model. In addition, a more sophisticated
location program should be investigated in the future.
61
International Journal of Disaster Risk Reduction 20 (2016) 51–62
Acknowledgments
This work was jointly supported by the Special Fund for Basic
Scientific Research Operations in Institute of Geology, China
Earthquake Administration (Grant No. IGCEA 1506), the Jiangsu
Surveying, Mapping and Geoinformation Science Research Project
(Grant No. JSCHKY201506), the Open Research Fund Program of
Shenzhen Key Laboratory of Spatial Smart Sensing and Services
(Shenzhen University) (Grant No. 201404), and the Scientific
Research Fund of Institute of Engineering Mechanics, China
Earthquake Administration (Grant No. 2016QJGJ16, Grant No.
2016QJGJ09). The authors would like to express their appreciation
to the editor and the referees for their comments and suggestions that
greatly improved the manuscript.
References
[1] X. Albacete, K. Pasanen, M. Kolehmainen, A GIS-based method for the selection of
the location of residence, Geo-Spat. Inf. Sci. 15 (1) (2012) 61–66.
[2] S. Babaie-Kafaki, R. Ghanbari, N. Mahdavi-Amiri, Hybridizations of genetic
algorithms and neighborhood search metaheuristics for fuzzy bus terminal location
problems, Appl. Soft Comput. 46 (2016) 220–229.
[3] C. Çetinkaya, E. Özceylan, M. Erbaş, M. Kabak, GIS-based fuzzy MCDA approach
for siting refugee camp: a case study for southeastern Turkey, Int. J. Disaster Risk
Reduct. 18 (2016) 218–231.
[4] A.J. Comber, S. Sasaki, H. Suzuki, C. Brunsdon, A modified grouping genetic
algorithm to select ambulance site locations, Int. J. Geogr. Inf. Sci. 25 (5) (2011)
807–823.
[5] R.L. Church, C.S. ReVelle, Theoretical and computational links between the p-
median location set-covering and the maximal covering location problem, Geogr.
Anal. 40 (1976) 406–415.
[6] K.C. Desouza, T.H. Flanery, Designing, planning, and managing resilient cities: a
conceptual framework, Cities 35 (2013) 89–99.
[7] C.E. Dunn, D. Newton, Optimal routes in GIS and emergency planning application,
Area 24 (3) (1992) 259–267.
[8] S.S. Elheishy, A.A. Saleh, A. Asem, A rough set and GIS based approach for
selecting suitable shelters during an evacuation process, J. Geogr. Inf. Syst. 5 (1)
(2013) 1–12.
[9] M. Esmaelian, M. Tavana, F.J.S. Arteaga, S. Mohammadi, A multicriteria spatial
decision support system for solving emergency service station location problems,
Int. J. Geogr. Inf. Sci. 29 (7) (2015) 1187–1213.
[10] ESRI, 〈http://resources.arcgis.com/en/help/main/10.2/index.html#/What_is_a_
network_dataset/004700000007000000/〉, 2016 (last access on 07.05.16).
[11] Z.X. Fang, X.L. Zong, Q.Q. Li, Q.P. Li, S.W. Xiong, Hierarchical multi-objective
evacuation routing in stadium using ant colony optimization approach, J. Transp.
Geogr. 19 (3) (2011) 443–451.
[12] G. Ghiani, L. Grandinetti, F. Guerriero, R. Musmanno, A Lagrangean heuristic for
the plant location problem with multiple facilities in the same site, Optim. Methods
Softw. 17 (6) (2002) 1059–1076.
[13] W. Gu, X. Wang, An interactive geospatial analysis platform for facility location
decision-making, Int. J. Numer. Anal. Model. 9 (2) (2012) 217–231.
[14] R.A. Hadiguna, I. Kamil, A. Delati, R. Reed, Implementing a web-based decision
support system for disaster logistics: a case study of an evacuation location
assessment for Indonesia, Int. J. Disaster Risk Reduct. 9 (2014) 38–47.
[15] S.L. Hakimi, Optimum locations of switching centers and the absolute centers and
medians of a graph, Oper. Res. 12 (3) (1964) 450–459.
[16] V. Indriasari, A.R. Mahmud, N. Ahmad, A.R.M. Shariff, Maximal service area
problem for optimal siting of emergency facilities, Int. J. Geogr. Inf. Sci. 24 (2)
(2010) 213–230.
[17] X. Li, A.G. Yeh, Integration of genetic algorithms and GIS for optimal location
search, Int. J. Geogr. Inf. Sci. 19 (5) (2005) 581–601.
[18] X.P. Li, Z.X. Zhao, X.Y. Zhu, T. Wyatt, Covering models and optimization
techniques for emergency response facility location and planning: a review, Math.
Methods Oper. Res. 74 (2011) 281–310.
J. Xu et al.
[19] B.Y. Liu, L.Y. Ye, J.Y. Su, The development and prospect of urban earthquake
disaster prevention plans, World Earthq. Eng. 24 (1) (2008) 68–72 (in Chinese).
[20] S.L. Liu, Urban emergency evacuation shelter location section and spatial layout:
using Nanjing as case study (Ph.D. thesis), Nanjing Normal University, Nanjing,
2012 (in Chinese).
[21] H.Z. Jia, F. Ordóñez, M. Dessouky, A modeling framework for facility location of
medical services for large-scale emergencies, IIE Trans. 39 (1) (2007) 41–55.
[22] J.M. Marcelin, M.W. Horner, E.E. Ozguven, How does accessibility to post disaster
relief compare between the aging and the general population? A spatial network
optimization analysis of hurricane relief facility locations, Int. J. Disaster Risk
Reduct. 15 (2016) 61–72.
[23] H. Miller, GIS and geometric representation in facility location problems, Int. J.
Geogr. Inf. Syst. 10 (7) (1996) 791–816.
[24] M.M.L. Nappi, J.C. Souza, Disaster management: hierarchical structuring criteria
for selection and location of temporary shelters, Nat. Hazards 75 (3) (2015)
2421–2436.
[25] H.H. Nasab, M. Tavana, M. Yousefi, A new heuristic algorithm for the planar
minimum covering circle problem, Prod. Manuf. Res. 2 (1) (2014) 142–155.
[26] R. Rahmaniani, M.A. Shafia, A study on maximum covering transportation network
design with facility location under uncertainty, J. Ind. Prod. Eng. 30 (2) (2013)
78–93.
[27] B. Saeidian, M.S. Mesgari, M. Ghodousi, Evaluation and comparison of Genetic
Algorithm and Bees Algorithm for location–allocation of earthquake relief centers,
62
International Journal of Disaster Risk Reduction 20 (2016) 51–62
Int. J. Disaster Risk Reduct. 15 (2016) 94–107.
[28] U. Tamima, L. Chouinard, Development of evacuation models for moderate seismic
zones: a case study of Montreal, Int. J. Disaster Risk Reduct. 16 (2016) 167–179.
[29] M. Tavana, W.R. Liu, P. Elmore, A practical taxonomy of methods and literature for
managing uncertain spatial data in geographic information systems, Measurement
81 (2016) 123–162.
[30] C.H. Tsai, Y.L. Yeh, The study of integrating geographic information with multi-
objective decision making on allocating the appropriate refuge shelters: using
Kengting National Park as an example, Nat. Hazards 82 (3) (2016) 2133–2147.
[31] P. Tenerelli, J.F. Gallego, D. Ehrlich, Population density modelling in support of
disaster risk assessment, Int. J. Disaster Risk Reduct. 13 (2015) 334–341.
[32] N. Wood, J. Jones, J. Schelling, M. Schmidtlein, Tsunami vertical-evacuation
planning in the U.S. Pacific Northwest as a geospatial, multi-criteria decision
problem, Int. J. Disaster Risk Reduct. 9 (2014) 68–83.
[33] J.H. Xu, J.W. An, G.Z. Nie, A quick earthquake disaster loss assessment method
supported by dasymetric data for emergency response in China, Nat. Hazards Earth
Syst. Sci. 16 (3) (2016) 885–899.
[34] J. Yang, J.H. Chen, H.L. Liu, J.C. Zheng, Comparison of two large earthquakes in
China: the 2008 Sichuan Wenchuan Earthquake and the 2013 Sichuan Lushan
Earthquake, Nat. Hazards 73 (2) (2014) 1127–1136.
[35] M. Zeiler, Modeling Our World: The ESRI Guide to Geodatabase Concepts, 2nd ed.,
Esri Press, California, USA, 2010.