Safety Science 110 (2018) 192–203

Contents lists available at ScienceDirect

Safety Science
journal homepage: www.elsevier.com/locate/safety

A proposed semi-quantitative framework for comprehensive risk assessment T

of urban hazard installations considering rescue accessibility and evacuation
vulnerability
⁎
Qiansheng Zhaoa, Lee D. Hanb, , Nianxue Luoa
a
School of Geodesy and Geomatics, Wuhan University, 129 Luoyu Road, Wuhan, Hubei 430079, China
b
Department of Civil & Environmental Engineering, The University of Tennessee, 319 John Tickle Building, Knoxville, TN 37996, United States

A R T I C LE I N FO A B S T R A C T

Keywords: With rapid economic growth, previously less populated peripheral areas of major cities in China have now
Comprehensive risk assessment become densely populated. Many hazardous installations (e.g., chemical plants, tank farms, depots of explosive
Rescue accessibility materials, etc.) that used to be in these less populated areas are now found in the backyards of millions of citizens
Evacuation vulnerability going about their daily lives with minimal knowledge about the dangers these installations pose. A hazard
Hazard installation
mishap that may have been containable with no human loss just 30 years ago can now threaten the lives of
millions.
While numerous studies have assessed hazard-specific risks, few efforts have attempted to establish a com-
prehensive framework to assess multi-hazard scenarios. For this purpose, this study developed a framework for
spatial risk assessment of multiple hazardous installations in a metropolitan area. The framework includes three
quantitative models for the analysis of the inherent danger associated with, the accessibility for rescue opera-
tions at, and the potential effectiveness of mass evacuation from these hazardous installations. Based on this
framework and its quantitative models, a case study of the city of Beijing was conducted in this research.

1. Introduction installations at risk, and then help the urban planning managers and
emergency response managers to reduce the risk (Zhao and Chen, 2015;
In recent years, with rapid urbanization in China, the urban area is Zhao and Liu, 2017).
expanding rapidly, and many small country towns are becoming cities Assessing failure probabilities remains a problematic issue when it
(Xiang et al., 2011; Pan and Wei, 2015; 13th Five-Year Plan, 2015). comes to intervals of confidence (Pasman et al., 2017). In China, be-
Consequently, more hazardous installations that were beyond urban cause of a short period of industrialization, the country still has not
areas are located in urban areas and dense population districts (Huang accumulated enough event data and management experience. It is
et al., 2011). Many types of disasters are related to hazardous in- challenging to assess the possibility of disaster events occurring at ha-
stallations, but hazardous chemical leaks, gas explosions, and fire are zardous installations, because there are many factors such as manage-
the most typical manifestations of disasters (Zhou and Liu, 2012; Zhao ment level, worker level, and types of hazardous materials that need to
and Chen, 2015). The impact of these accidents often spreads to the be considered (Liu et al., 2007; Si et al., 2012; He et al., 2014; Wang,
adjacent urban areas, causing catastrophic effects and heavy casualties 2017). To simplify the assessment and make it practicable, in this study,
(Li, 2006; Georgiadou et al., 2010; Li et al., 2010a,b; Huang et al., 2011; we assume the possibility of the disaster events occurring of the ha-
Fan, 2014). Therefore, evaluating the potential hazards of major ha- zardous installations is 1, and we accordingly assess the risk of the
zardous installations in urban areas to adjacent urban areas is vital for consequences of a disaster. Some researchers have already used “the
public safety and environment protection (Zhao and Liu, 2017). The principle of maximum danger” and “the principle of probability sum-
ideal solution is to remove the hazardous installations by relocating the mation” as principles to calculate property losses and casualties. Fur-
factories or the warehouses that produce or deposit the hazardous thermore, they have used the Classification Standards of Major Hazard
materials, though this may be costly and time-consuming. A compro- Installations (exposure draft) as the standard to classify major dan-
mise solution is to assess the risk of the hazardous installations, find gerous chemical installations (Luo, 2013). In this study, we classified

⁎
Corresponding author.
E-mail addresses: [email protected] (Q. Zhao), [email protected] (L.D. Han), [email protected] (N. Luo).

https://doi.org/10.1016/j.ssci.2018.08.013
Received 5 December 2017; Received in revised form 16 August 2018; Accepted 16 August 2018
0925-7535/ © 2018 Elsevier Ltd. All rights reserved.
Q. Zhao et al. Safety Science 110 (2018) 192–203

factors related to hazardous installations into three categories. First, the Comprehensive risk
inherent risk of the installations, which depends on type and quantity of assessment framework
materials that they have. Second, rescue force accessibility around the (Section 2.1)
installations, which includes firefighting accessibility and medical ser-
vice accessibility. The more accessible the area around the installations
for rescue services, the less risky they are. Third, the evacuation vul-
nerability around the installations. The lower the evacuation vulner- Rescue force Evacuation
Intrinsic risk
accessibility vulnerability
ability, the less risky the installation is. How should these three parts of assessment
assessment assessment
the risk classification be individually assessed to give a comprehensive (Section 2.2)
(Section 2.3) (Section 2.4)
risk value of hazardous installations?.
Many researchers have worked on various aspects of this problem.
For example, Quantitative Risk Assessment (QRA), sometimes also
called Probabilistic Risk Analysis (PRA), is a typical risk-based ap- Comprehensive risk assessment
(Section 2.5)
proach and is well-known throughout the world (Zhao and Liu, 2017;
Zeng and Zio (2017)). Some researchers established a grid-based dif-
ferential risk model, which provides quantitative risk assessment and
Case study
calculation of individual risk, and explores its application in major
(Section 3)
hazard safety planning in urban areas (Wu et al. 2006; Weng et al.
2006; Zhao et al., 2013; De Silva et al., 2017). Wang et al. (2015)
provided a quantitative risk assessment means for identifying and Results and
classifying major hazard installations via an exemplary analysis of a discussion
small petrochemical plant. Wu and Hu (Wu and Hu, 2017) considered (Section 4)
the risk characteristics for a dense area of major port hazardous in- Fig. 1. the structural flow chart of this article.
stallations and established a quantitative risk assessment method, in-
cluding an assessment process, calculation model, core quantification
indexes and acceptable risk criteria. However, this research does not shows the structural flow chart of this article according to the logical
mention the surrounding evacuation and rescue force that can reduce process of the topic’s development.
the risk of the hazardous installations, we named this factors as com-
pensation factors. Risk factors are not only limited to population den- 2. Methodology
sity but, also, other factors like rescue service accessibility and eva-
cuation vulnerability can impact the spatial pattern of hazardous This section discusses the methodology of the comprehensive risk
installation risk. Concerning the evacuation in areas surrounding dis- assessment, including the framework for the comprehensive risk as-
asters, some studies have examined the optimization of evacuation sessment model, intrinsic risk assessment model, rescue accessibility
plans (Han, et al. 2005, 2006; Yuan and Han, 2010). Cova and Church assessment model, evacuation vulnerability assessment model, and the
(1997) proposed a method to assess evacuation vulnerability. Although comprehensive risk assessment model.
it is a quantitative method, it is beneficial for decision makers. Some
studies provide a relatively ideal solution for planning emergency 2.1. Framework of the comprehensive risk assessment model
traffic evacuation in open spaces, providing feasible evacuation routes
(Li et al., 2010b). There are many types of hazardous installations. Hazardous che-
In urban areas, the hazard caused by industrial installations to ad- mical leaks, major production safety accidents, gas explosions, mine
jacent areas is primarily a geospatial issue, which inevitably refers to collapses, major pollution, water, electrical, and other engineering
the spatial pattern and natural geographical condition of urban systems disasters are common manifestations of disaster. Among these, fire,
(Zhao and Chen, 2015). With the improvements of geospatial data in explosions, and poison gas leaks are the most common. Many directly or
temporal and spatial resolutions, risk calculation and mapping for risk indirectly cause other accidents. This study primarily discusses the
have grown significantly in volume (Zhao and Liu, 2017). In recent three forms of hazards mentioned previously.
years, the Geographic Information System (GIS) has become a vital Before beginning risk assessment, identification and classification of
component of the risk assessment framework, which has been suc- hazardous installations are needed. During the hazard risk assessment
cessfully demonstrated in regional risk assessment (Huang, et al., 2004; phase, there are two methods for risk assessment—the assessment of the
Shi, 2006; Delvosalle et al., 2006; Planas et al., 2006; Li et al., 2010, consequences and a probability-based risk assessment. The first
2014; Sebos et al., 2010; Herrero-Corral et al., 2012; Zhou and Liu, method, based on an assessment of the consequences, does not consider
2012; La Rosa and Martinico, 2013; Liu et al., 2013; Tena-Chollet et al., the probability of an event occurring (assume that the probability of an
2013; Meng, 2015; Thompson et al., 2015). event occurring is 1), directly related to the use of the physical damage
Although much research has focused on hazardous installation risk model. The second method, based on the probability of an accident, is
assessment, some significant limitations of these works are still ap- considered a risk assessment related to a physical damage model of the
parent. First, most of the risk assessment methods focus on one aspect of probability of an accident occurring. This study uses the first method,
the risk assessment process. Some research focusses on the social risk or for comprehensive disaster risk assessment in this study, as shown in
individual risk of the installation, or on evacuation routes planned the steps below:
around them. Some research focused on rescue access at installations,
and there is still no method of integrating all factors together to assess (1) Select the type of accident event occurring based on the type of
the risk of hazardous installations. Second, GIS technology is playing an hazardous materials;
increasingly crucial role in risk assessment, but there is still no frame- (2) Determine the physical damage model by accident type and cal-
work for risk data collection, risk modeling, and risk mapping. To ad- culate the spatial distribution of a dangerous dose of the hazardous
dress the limitations above, we introduced a semi-quantitative com- materials;
prehensive framework for hazardous installation risk assessment, (3) Calculate the individual risk through the spatial distribution of a
including intrinsic risk assessment, rescue force accessibility, and the dangerous dose, and calculate the social risk by population density;
evacuation vulnerability affecting risk around installations. Fig. 1 (4) Calculate the individual risk and social risk attenuation coefficients

193
Q. Zhao et al. Safety Science 110 (2018) 192–203

Data Model Result

Risk installations
distribution data
Intrinsic risk Intrinsic risk
assessment map

Risk zoning map

Assessment Model
Integrated Risk
Hospitals and Rescue Rescue
fire stations Accessibility Accessibility
distribution data Assessment map

Vulnerability
data: road and Vulnerability Vulnerability
population assessment map

Fig. 2. Framework of comprehensive risk assessment.

considering medical accessibility; Individual risk and social risk can describe the inherent risk of ha-
(5) Calculate individual risk and social risk attenuation coefficients zardous installations. Individual risk is the probability of death caused
considering firefighting accessibility; by accident for a person at a particular location with no protective
(6) Analyze the evacuation vulnerability of surrounding environments measures. Individual risk can be calculated by the level of the dan-
of the hazardous installations, and obtain the spatial distribution of gerous dose (thermal radiation, blast wave, and concentration, etc.) and
the vulnerability of the area to be evacuated; the mortality rate. Social risk describes the relationship between the
(7) Perform a comprehensive risk assessment considering inherent risk, probability of accidents and the number of accidents resulting in death
medical service access, firefighting access, and evacuation vulner- or injury, which can be calculated through the integration of individual
ability around the hazardous installations. risk and population density. In this study, individual risk and social risk
are used to present the inherent risk of the hazardous installations. The
The framework for comprehensive risk assessment includes three expression for individual risk for fires, explosions, and other dangerous
parts (Fig. 2): gas leak accidents is usually calculated using the following formula to
obtain individual risk value (Ji, et al., 2006):
(1) Data preparation: including risk installations data, rescue force
data, road network, and vulnerability data; d = C nt
(2) Models: including intrinsic risk assessment model, rescue accessi- P = A + B ln d
p−5
bility assessment model, evacuation vulnerability model, and a
comprehensive assessment model;
IR (P ) =
1
2π
∫ exp
−∞
( ) du
u2
2
(1)
(3) Results: collate the output of the models into a risk zoning map,
including the intrinsic risk map, rescue accessibility map, evacua- In the formula above:
tion vulnerability map, and comprehensive risk zoning map. C is the dangerous dose level (radiance, overpressure wave, gas
concentration and etc.);
n is the constant depending on the type of accident;
2.2. Intrinsic risk assessment t is the dangerous dose level’s effective period;
IR (P ) is individual risk;
Intrinsic risk of hazardous installations means the natural risk of the A , B are constants, obtained by animal experiments or historical
hazards; the physical damage model can calculate this. The inherent data;
risk is determined by the type and quantity of the hazardous materials. Social risk can be calculated by the following formula:
The research of the physical damage model is not the main purpose of
this study. Therefore, a brief introduction to common injury accidents E (N ) = ∬A IR (x, y) m (x, y) dxdy (2)
and the common disasters model is now presented. Fire, explosion, and
toxic gas leaks are common major accidents, often causing heavy ca- where E (N ) represents the expected fatality number and
sualties and substantial property losses. In this study, we take fires, m (x , y ) represents the population density of a location(x , y ) .
explosions and toxic gas leaks as examples, introduce the physical da- The Geographic Information System (GIS) provides an excellent
mage models and calculate the individual risk and social risk using a platform for quantitative risk analysis and showcases the results of as-
GIS platform. These three kinds of accident risk analysis are commonly sessments. Individual risk at any location (formula 1), calculated under
used in predictive models of injury (Ji, et al., 2006) and are shown in GIS, can be described by the following formulas:
Table1.

Table 1
General models of fires, explosions, and toxic gas leaks (Ji et al., 2006).
Accident type Fires Explosions Toxic gas leaks

General models Pool fire model; Jet flame model; Fireball model; Vapor cloud explosion model; Gas discharge model; Liquid relief model;
Radiation intensity attenuation model Limited vapor cloud explosion model; Adiabatic diffusion model;
Unrestricted vapor cloud explosion model; Gas diffusion model;
Boiling liquid expanding vapor explosion model Heavy gas diffusion model

194
Q. Zhao et al. Safety Science 110 (2018) 192–203

factors. After an incident occurs, rescue services are the main actors
seek to control a situation, prevent the spread of the accident, rescue
(x,y) personnel and organize mass evacuation. Timely and effective medical
treatment is essential for keeping the wounded out of danger, and
medical power promptly arrived at the scene to reduce casualties,
providing medical guidance for rescue operations. So the accessibility
of the rescue force is vital for the reduction of the risk. How to assess
the accessibility of rescue services around the hazardous installations?
In this study, the compensation factor to consider firefighting service
accessibility and medical rescue service accessibility.
2
2.3.1. Firefighting accessibility
According to China’s national conditions, fire brigades must arrive
1
within 15 min to the scene for rescue to be able to effectively prevent
further incidents (Chen and Guo, 1999). From the time when accidents
occurred to when the fire brigade arrived at the scene of the accident
encompassed 15 min: alarm 4 min, 1 min for ready out, 5 min transit,
Fig. 3. Individual risk calculation based on a grid. and firefighting activities occurred within 5 min. The distance from the
fire station to the scene is about 5 min’ drive at an average speed of
d = C n (x , y ) t 30 km per hour. In this study, we use the Firefighting Compensation
⎧
⎪ P = A + B ln d Coefficient (FCC) to describe firefighting accessibility around hazardous
⎪
⎪ p−5 installations. FCC is used to measure the compensation of firefighting
⎨
1
IR (x , y ) = 2π ∫ exp 2 du
−∞
( )u2
service to the risk of the hazardous installations. The smaller the FCC
⎪ n value, the more compensation the firefighting services have. The fol-
⎪ R (x , y ) = ∑ IRs (x , y ) lowing paragraph will descript how to calculate the FCC value.
⎪ (3)
⎩ s=1 The time for ready of firefighting service has two parts which are
the fire department response time and the equipping facilities time. The
IR (x , y ) represents the individual risk value of a single hazardous in-
fire department response time is divided into the police time (from
stallation at a location of (x , y ) ; IRs (x , y ) represents the individual risk
alarm to travel) and fire engine driving time. Fire response time is
value of number s hazardous installation at location of (x, y);
calculated as follows (Chen et al., 2006):
R (x , y ) represents the sum individual risk at location of (x, y) caused by
all hazard installations (Fig. 3). sf
tf = tf 1 +
Under a GIS environment, the process for the spatial intrinsic risk uf (4)
assessment of hazardous installations is:
(1) Data preparation where tf is fire response time(s); tf1 is the time for ready out(s), which is
Population density data: to ensure accurate calculation using the based on the regional average time; sf is the distance from the nearest
raster data model to represent population density, different precision fire station to the accident scene (m), and uf is the average drive speed
population density data can be resampled from the original raster data; (m/s).
Data of hazardous installations: including hazardous material type, This study considers the travel time of fire rescue forces as valid if
type of flammable hazards, the type and volume of explosives, toxic the area is within 5 min’ drive; or invalid if 5 min’ drive is insufficient.
material, and other substances by comparison with hazards associated Therefore, the fire rescue force cannot be considered to reduce the risk
with the type of the physical damage model; of a hazard if the drive time is more than 5 min. In such a situation, the
Accidental hazard probability: to determine the probability of an FCC is 1. When the time for the preparation of fire brigades and fire
accidental hazard occurring each year you can obtain relevant data engine driving speed are known, we can determine the route for the
through statistical data or methods. This study assumes the probability firefighting compensation region from sf (formula 5). The FCC value
of occurrence is 1 because this research is based on the consequences of according to drive time is shown in Table 2.
the accident risk analysis; sf = (t f −t f 1) uf (5)
(2) Meshing the regional area of assessment
Grids are calculated according to the definition of the size of the
required accuracy, and the entire region is divided into a unified grid 2.3.2. Medical rescue force accessibility
according to the necessary accuracy. Its center position locates the grid, Medical rescue compensation coefficient (MCC) is calculated with
and the coordinates of the positions are the output value of the grid. the qualified medical emergency rescue units which are located in re-
(3) Calculate individual risk gional risk compensation areas. The medical rescue compensation
Using the previous model (formal 3) to calculate the individual risk coefficient is determined by the emergency response time. The smaller
of each hazardous installation in the center of the value of each grid and the MCC value, the more compensation the medical rescue services
added them together; through the interpolated output corresponding to have. The following paragraph will descript how to calculate the MCC
the individual risk contours and individual risk level zoning map; value. Emergency response time is the time from when the hospital
(4) Calculate social risk
According to the administrative data and population density zoning Table 2
map, calculate the number of casualties in a given administrative re- The value of firefighting compensation coefficient (FCC).
gion.
Drive time (min) FCC value

2.3. Rescue accessibility assessment Drive time > 5 min 1.00

3 min < Drive time ⩽5 min 0.95
1 min < Drive time ⩽3 min 0.90
Risk compensation factors refer to the mitigation of consequence of Drive time ⩽1 min 0.85
unexpected events. This generally refers to ambulance and fire rescue

195
Q. Zhao et al. Safety Science 110 (2018) 192–203

Table 3 Vulnerability Level

Medical compensation coefficient value.
Drive time (min) Driving distance S (m) MCC value 1 2 3 4 5
Drive time > 10 min 1.00
5 min < Drive time ⩽10 min S2 0.90 1
Drive time ⩽5 min S1 0.80

Hazard Level
2
received a request to when rescue teams arrive at a scene, usually in-
cluding preparation time (from receipt of the “120” ambulance dispatch
instructions to going out) in the hospital and the driving time (ambu- 3
lance travel time) between the hospital and accident. According to the
rescue effect of the emergency response time, the area around the 4
hospital area is divided into two ranges, within 5 min’ reaction and
within 10 min’ reaction time of the compensation area.
The reaction time of the emergency medical units is calculated as 5
follows (Chen et al., 2006):
sm
tm = tm1 +
um (6) 1 2 3 4 5

where tm is the emergency response time (s); tm1 is the preparation time Fig. 5. Risk matrix.
(s); sm is the distance between hospital and accident scene (m); um is the
speed of the ambulance (m/s). Given the average speed um, the areas of 2.4. Evacuation vulnerability assessment
5 min and 10 min accessibility can be calculated. The corresponding
distances are S1 and S2, and the formula is: When assessing the accident area, it is hard to determine the loca-
tion of a scene before an accident occurs. Most studies on evacuation
Sm = (tm−tm1) um (7)
are about determining the evacuation scene but, in this situation, the
The medical compensation coefficient value is shown in Table 3. evacuation area is known. However, for an entire city, it is hard to
determine an accident scene, therefore, how do we assess the evacua-
tion capability of urban neighborhoods or evacuation vulnerability?
2.3.3. Integrated firefighting and the medical compensation coefficient This study combines the principal model of evacuation vulnerability
The integrated compensation coefficient (ICC) contains the fire- assessment Cova proposed in 1997 with our physical damage model of
fighting compensation and medical compensation coefficients, which hazardous risk assessment to determine the affected area (Cova and
can be calculated by the following formula: Church, 1997). The affected area is the evacuation unit which can be
ICC = FCC∗MCC (8) used to assess evacuation vulnerability. Accurately estimating area
clearance time for each evacuation area around hazardous installations
where ICC is the integrated compensation coefficient; FCC is fire- using a simulation model is not possible in this context because there
fighting compensation coefficient, and MCC is medical compensation are too many potential areas. However, another alternative is to de-
coefficient. After calculating the integrated compensation, the com- velop a static index estimate of evacuation difficulty. We refer to this
pensation risk (CR) can be calculated by the following formula: index as an evacuation vulnerability index (V). Evacuation difficulty is
defined as an estimate of the relative effort required to clear an area of
CR = ICC∗NR (9)
its population. “This concept embodies a composite of the potential for
where CR is compensation risk, and NR is intrinsic risk (individual risk congestion, accidents, and general difficulty in deploying response ve-
or social risk). hicles into the evacuation zone” (Cova and Church, 1997). There are

5 7 9
4 2 4 2 4 2

2 2 2
3
4 2 4 4 4 2 4 2
6 8
2 2
1 4 2 4 2
A
R

Fig. 4. Evacuation vulnerability at node A under the evacuation radius R (Left); Curve of evacuation difficulty at node A (Right).

196
Q. Zhao et al. Safety Science 110 (2018) 192–203

Population density Hazard installations distribution

Oil depot

s
Firecracker
warehouse

Fig. 6. Population density and hazard installations distribution (Processed).

Fig. 7. Individual risk.

numerous approaches to operationalizing a concept like evacuation evacuated out of radius R.

difficulty but, for our purposes, we will define it in Eq. (10) where P is A neighborhood’s evacuation vulnerability can also use related
the population involved, and C is a measure of the capacity of the exit problems in graph theory description: Given a network G and the cost of
choice set. each network edge strip, dividing the network into several small sizes in
a given sub-network, making the total costs of the sub-section of the
V= P/C (10) network connection minimum, evacuation vulnerability assessment can
be described as graph theory: given a graph G = (N , A) with costs cij
It is clear that this definition is a highly simplified view of what is (lanes), on its arcs and weights, ai (population), on its nodes, we choose
undoubtedly a complicated social, spatial-temporal process. It should to partition N into two node sets N1 and N 2 , and N = N 1 ∪ N 2 ,
be noted that this method is a comparison of starting scenarios, not N 1 ∩ N 2 = ∅, where N1 is a contiguous partition less than or equal to
predicted outcomes. In other words, the comparison is between initial size s that contains a root node i*, so as to maximize the total weight of
conditions, rather than an attempt to predict the outcome of an actual N1 relative to the total cost of the arcs cut between N1 and N2. This can
evacuation. be described as an integer programming problem:
Fig. 4 demonstrates how to calculate the evacuation vulnerability of Objective:
node A (node number 1) under the evacuation radius of R, assuming the
number of lane is 2, and people to be evacuated on every node is 4
(Fig. 4 Left), as a result, the value of P/C at the node A (Node 1) is 10
(Fig. 4 Right), because 10 is the maximum value of P/C when node A is

197
Q. Zhao et al. Safety Science 110 (2018) 192–203

ai is the weight of node i (population);

cij is the cost of arc ij (lanes);
Sxi ∗ xi is the distance between the root node to the nodes;
R is the radius;
i* is the root node, which should be calculated with vulnerability.
Algorithm steps:

(1) Building topology road network;

(2) Using road nodes to construct a Thiessen Polygon, and using the
Thiessen Polygon to divide the region and calculate the population
within every node;
(3) Using the hazard physical damage model to calculate the affected
area and radius for the evacuation area and radius;
(4) Repeat every node and do the same for spatial queries:
(5) Find the largest P/C in the limited area and assign the value to the
node;
(6) Assign the arc with the higher value of two nodes;
(7) Classify the arc by the value of P/C and get the evacuation vul-
Fig. 8. Firefighting compensation. nerability of the area.

∑ aixi
Maximize : i 2.5. Comprehensive risk assessment
∑ ∑ cijyij
i j (11) After intrinsic risk analysis, rescue force accessibility analysis and
evacuation vulnerability analysis, we need to integrate all the factors
Subject to:
into the final result risk assessment. In this study, we use risk matrix to
integrate all the factors (Fig. 5). We use “risk = hazard + vulnerability”
xi−yj ⩽ yij, ∀ i, j ∈ N
to integrate the two factors. To facilitate the calculation, we classify the
Sxi∗xi ⩽ R
individual risk after considering rescue force accessibility (hazard level)
xi∗ = 1
xi, yij, ∈{0, 1}, ∀ i, j ∈ N and evacuation vulnerability (vulnerability level) into five levels and,
(12)
then, by risk level to determine the final risk matrix (Fig. 5). The data
classification method can be an equal interval, quartile classification,
where
standard deviation classification, or natural breaks. In this study, the
most straightforward method of the equal interval is used, and it should
be noted that the degree of risk through the risk matrix table here
cannot be expressed as the absolute amount of risk, but only indicates

Fig. 9. Medical compensation.

198
Q. Zhao et al. Safety Science 110 (2018) 192–203

Fig. 10. Integrated risk compensation coefficient.

Fig. 11. Road network and Thiessen polygons.

the relative degree of risk. Conversely, this is a better method to assess data with a resolution of 30 m (Fig. 6, left). The hazardous installation
comprehensive risk. data includes 15 large oil depots and ten large firecracker warehouses
(Fig. 6, right).
We use an intrinsic risk assessment model to assess the individual
3. Case study
risk (Fig. 7). In this situation, with no rescue compensation, the intrinsic
individual risk relates to the hazard installations themselves.
In this section, we will use an area of Beijing as a case study to
practice the integrated risk assessment method proposed in the previous
sections. The data was processed because it is sensitive; thus, the in-
3.2. Rescue accessibility calculation
formation is not complete and accurate.
The entire rescue force and road data are substantial; thus, we chose
3.1. Intrinsic risk assessment a small region of the city. According to our previous method, we con-
structed a topology road network and used GIS tools to analyze the fire
The case uses two types of hazard installations, oil depots, and stations’ service area within a 1-min, 3-min, and 5-min drive (Fig. 8).
firecracker warehouses to calculate individual spatial risk and social These corresponded to the 1-min, 3-min, and 5-min compensation area.
risk. The data include population density and the location and quantity Fig. 9 shows the medical compensation area divided into a 5 min and
of hazardous installations. The population data is LandScan world-wide 10 min’ drive, and an invalid area. Fig. 10 shows the integrated

199
Q. Zhao et al. Safety Science 110 (2018) 192–203

Fig. 12. Evacuation vulnerability.

Fig. 13. Intrinsic individual risk.

compensation area, which we divided into five classes of compensation. has the greater corresponding P/C value. The evacuation path in a
dense area is relatively easy, and the corresponding P/C value is rela-
tively small.
3.3. Evacuation vulnerability calculation

Fig. 11(left) shows a part of the road network by lanes. Fig. 11 3.4. Comprehensive risk assessment
(right) shows a Thiessen polygon constructed by road nodes. The po-
pulation of network nodes can be extracted by every area of the In this case, we used the same area to compare the intrinsic risk and
polygon and population density. Fig. 12 shows the result of evacuation the risk considering rescue compensation. Fig. 13 shows intrinsic in-
vulnerability with a 1 km radius. The results according to P/C value is dividual risk mapping; the value is significantly related to the distance
divided into five levels—less than 150, 150–300, 300–450, 450–600, from hazard installations (the closer distance, the higher the value).
and greater than 600. The greater the value, the more difficulty in Fig. 14 shows individual risk mapping after considering the medical
evacuation, though the classification can be adjusted depending on the and fire compensation coefficient. Values here have significant re-
circumstances. Fig. 12 (left) represents the evacuation vulnerability in lationships with the distance from hazard installations and, although
the road network. Fig. 12 (right) represents the evacuation vulner- the risk has decreased significantly in the compensational area, some
ability in the Voronoi region. This figure assumes that the population local areas have a higher risk than those closer to hazard installations.
density is the same, the evacuation path is difficult in a sparse area, and This is because those positioned closer to the hazard installations have

200
Q. Zhao et al. Safety Science 110 (2018) 192–203

Fig. 14. Individual risk under rescue compensation.

Fig. 15. Risk mapping.

more rescue compensation. easier to evacuate people and it is more accessible to rescue services. In
Fig. 15 shows comprehensive risk mapping with evacuation vul- the low individual risk areas, the risk may be higher because this area is
nerability using a risk matrix method. It shows that, in the high in- beyond the rescue compensation area and has high evacuation vul-
dividual risk areas, risk has declined. This is because, in this area, it is nerability. For example, there is a noticeable high-risk area in the right

201
Q. Zhao et al. Safety Science 110 (2018) 192–203

middle; we found this region has fewer roads and is near a hazardous assessment methodology. China Saf. Sci. J. 16, 112–117.
installation. Although this qualitative approach cannot give the exact Delvosalle, C., Fievez, C., Pipart, A., Debray, B., 2006. ARAMIS project: a comprehensive
methodology for the identification of reference accident scenarios in process in-
value of risks in the surrounding area, the combination of the individual dustries. J. Hazard. Mater. 130, 200–219.
risk assessment results can help decision-makers to understand the risk De Silva, K., Gunasekera, M.Y., De Alwis, A.A.P., 2017. Derivation of a societal risk ac-
level around the hazard installations better. ceptance criterion for major accident hazard installations in Sri Lanka. Process Saf.
Environ. Prot. 111, 388–398.
Yuan, F., Han, L.D., 2010. A multi-objective optimization approach for evacuation
4. Conclusions planning. Procedia Eng. 3, 217–227.
Fan, B., 2014. Hybrid spatial data mining methods for site selection of emergency re-
sponse centers. Nat. Hazards 70, 643–656.
Quantitative and the real risk assessment of hazard installations is Georgiadou, P.S., Papazoglou, I.A., Kiranoudis, C.T., Markatos, N.C., 2010. Multi objec-
very difficult, because it is difficult to know the real probability of tive evolutionary emergency response optimization for major accidents. J. Hazard.
disaster occurring, and because the surrounding installations are very Mater. 178, 792–803.
Huang, B., Cheu, R.L., Liew, Y.S., 2004. Incorporating security in HAZMAT route plan-
complex in operation. In this study, we assume that the disasters have
ning using GIS and AHP. 83rd Transportation Research Board Annual Conference,
occurred and assess the sequence of these disasters. This is another Washington, DC.
method to tackle risk assessment problems. We proposed a semi- Han, L.D., Chin, S., Fang, Y., Hwang, H., 2005. Assessment of heavy vehicle impacts on
quantitative model for comprehensive risk assessment. We considered emergency evacuation operations. In: Proc. TRB 84th Annual Meeting.
Transportation Research Board, Washington, D.C.
the factors of intrinsic risk of the hazard installation itself, the rescue Han, L.D., Fang, Y., Chin, S.M., Hwang, H., 2006. Global optimization of emergency
accessibility for people and the evacuation vulnerability surrounding evacuation assignments. Interfaces. 36, 502–513.
installations. This is helpful for emergency response and urban plan- Huang, L., Wan, W., Li, F., Li, B., Yang, J., Bi, J., 2011. A two-scale system to identify
environmental risk of chemical industry clusters. J. Hazard. Mater. 186, 247–255.
ning. In our case study, we used this model to discover the most at-risk Herrero-Corral, G., Jappiot, M., Bouillon, C., Long-Fournel, M., 2012. Application of a
installations and surroundings. This will give good support for disaster geographical assessment method for the characterization of wildland–urban inter-
managers and urban planners. Decision makers will be able to reduce faces in the context of wildfire prevention: a case study in western Madrid. Appl.
Geogr. 35, 60–70.
risk from three aspects. First, they can relocate people and residents; He, Y., Wang, L.L., Duan, J.F., 2014. Research and application of hazard sources eva-
second, they can add more rescue forces near the installations to make luation measures for municipal construction projects. Ind. Saf. Environ. Protect. 40,
rescue more accessible; third, they can plan new roads to reduce eva- 59–61.
Ji, X.W., Weng, W.G., Shu, X.M., Shen, S.F., Fan, W.C., 2006. Research on the risk analysis
cuation vulnerability around hazard installations. The major contents of urban accidents. China Saf. Sci. J. 16, 119–124.
and results of this study are summarized below: Li, J., 2006. Assessment of explosion risk and its application to the city planning. Fire Sci.
Technol. 25, 540–542.
Liu, H., Tang, Y., Cheng, Y., 2007. Study on the classification of dangerous hazard eva-
(1) The study offers a framework for comprehensive risk assessment of
luation based on risk management method. China Saf. Sci. J. 17, 145–150.
hazard installations under a GIS environment, and discusses how to Li, F., Bi, J., Huang, L., Qu, C., Yang, J., Bu, Q., 2010a. Mapping human vulnerability to
use GIS and related technologies for urban hazard risk assessment. chemical accidents in the vicinity of chemical industry parks. J. Hazard. Mater. 179,
(2) Using a quantitative risk assessment model with urban hazard in- 500–506.
Li, Q., Chen, X., Chen, J., Tang, Q., 2010b. An evacuation risk assessment model for
stallations, we discussed the physical models of accidental injury, emergency traffic with consideration of urban hazard installations. Chin. Sci. Bull.
we studied industrial fire, explosions, and chemical gas diffusion, 2010 (55), 1000–1006.
and we offered a method for integrated intrinsic risk of the three La Rosa, D., Martinico, F., 2013. Assessment of hazards and risks for landscape protection
planning in Sicily. J. Environ. Manage. 127 (Supplement), S155–S167.
types of hazardous installations. Liu, R.Z., Borthwick, A.G.L., Lan, D.D., Zeng, W.H., 2013. Environmental risk mapping of
(3) After study the intrinsic risk of the hazardous installations the paper accidental pollution and its zonal prevention in a city. Process Saf. Environ. Prot. 91,
concerned the factors of risk compensation (fire protection and 397–404.
Luo, Z.M., 2013. Research on assessment and classification for major hazard installation
medical treatment). We also considered evacuation effectiveness of dangerous chemicals. Prog. Environ. Sci. Eng. 610–613, 904–909.
surrounding the installations, and integrated the intrinsic risk, Li, D., Zhang, C., Pizzol, L., Critto, A., Zhang, H., Lv, S., Marcomini, A., 2014. Regional
rescue accessibility, and evacuation vulnerability into a compre- risk assessment approaches to land planning for industrial polluted areas in China:
the Hulunbeier region case study. Environ. Int. 65, 16–32.
hensive risk assessment model.
Meng, Q., 2015. Spatial analysis of environment and population at risk of natural gas
fracking in the state of Pennsylvania, USA. Sci. Total Environ. 515–516, 198–206.
Acknowledgement Planas, E., Arnaldos, J., Silvetti, B., Vallée, A., Casal, J., 2006. A Risk Severity Index for
industrial plants and sites. J. Hazard. Mater. 130, 242–250.
Proposal of the 13th five-year program on national economic and social development of
This research was supported in part by the National Key Research the People’s Republic of China. In: The 18th Central Committee of the Communist
and Development Program of China (Grants No. #2017YFC1405304). Party of China, the Fifth Plenary Session. November 3, 2015, Beijing, China.
Some of the research for this paper was conducted as part of the efforts Pan, J., Wei, H., 2015. Blue Book of Cities in China: Annual Report on Urban
Development of China No. 8. Social Sciences Academic Press, Beijing, China.
of the Southeastern Transportation Center (STC) at the University of Pasman, H.J., Rogers, W.J., Sam, Mannan M., 2017. Risk assessment: what is it worth?
Tennessee. The primary sponsor for the STC is the United States Shall we just do away with it, or can it do a better job? Saf. Sci. 99, 140–155.
Department of Transportation through grant number DTRT13-G- Shi, P.J., 2006. Urban risk assessment research of major natural disasters in China. Adv.
Earth Sci. 21, 170–177.
UTC34. Sebos, I., Progiou, A., Symeonidis, P., Ziomas, I., 2010. Land-use planning in the vicinity
The China Scholarship Council also supported the research for one of major accident hazard installations in Greece. J. Hazard. Mater. 179, 901–910.
year study at the University of Tennessee and the National Natural Si, H., Ji, H., Zeng, X.H., 2012. Quantitative risk assessment model of hazardous che-
micals leakage and application. Saf. Sci. 50, 1452–1461.
Science Foundation of China (91224008). The authors also would like Cova, Thomas J., Church, Richard L., 1997. Modelling community evacuation vulner-
to thank the anonymous reviewers. ability using GIS. Int. J. Geogr. Inform. Sci. 11, 763–784.
Tena-Chollet, F., Tixier, J., Dusserre, G., Mangin, J.F., 2013. Development of a spatial risk
assessment tool for the transportation of hydrocarbons: methodology and im-
Appendix A. Supplementary material
plementation in a geographical information system. Environ. Modell. Software 46,
61–74.
Supplementary data associated with this article can be found, in the Thompson, M.P., Haas, J.R., Gilbertson-Day, J.W., Scott, J.H., Langowski, P., Bowne, E.,
Calkin, D.E., 2015. Development and application of a geospatial wildfire exposure
online version, at https://doi.org/10.1016/j.ssci.2018.08.013.
and risk calculation tool. Environ. Modell. Software 63, 61–72.
Wu, Z.Z., Duo, Y.Q., Wei, L.J., Guan, L., 2006. Quantitative area risk assessment method
References and its application in land use safety planning for major hazard installations. Eng. Sci.
8, 46–49.
Weng, T., Zhu, J.P., Ma, M.G., Liao, G.X., Wu, Z.Z., 2006. A Study on regional assessment
Chen, Y.Y., Guo, G.Q., 1999. Optimize the layout of city fire station. Fire Sci. Technol. 1, of risk of urban major hazard. Eng. Sci. 8, 80–89.
26–28. Wang, W.L., Dong, C.L., Dong, W.P., Yang, C.X., Ju, T.T., 2015. The design and im-
Chen, G.H., Zhang, J., Zhang, H., Yan, W.W., Chen, Q.G., 2006. Study on regional risk plementation of risk assessment model for hazard installations based on AHP-FCE

202
Q. Zhao et al. Safety Science 110 (2018) 192–203

method: a case study of Nansi Lake Basin. Ecol. Inf. 36, 162–171. Zhao, W.F., Wan, G.J., Zhang, G.W., Zhang, Y., 2013. Exemplary analysis on the quan-
Wang, Y.X., 2017. Research on Safety Capacity Evaluation Model of Chemical Industry titative risk assessment of major hazardous installations. J. Saf. Environ. 13,
Park Based on Quantitative Risk Analysis. Thesis of Master Degree. Chongqing 239–243.
University of Science and Technology. Zhao, M., Chen, Q., 2015. Risk-based optimization of emergency rescue facilities locations
Wu, J., Hu, Y.C., 2017. Quantitative risk assessment method for regional port major for large-scale environmental accidents to improve urban public safety. Nat. Hazards
hazard installations and its application. New York, IEEE. 2017 4th International 75, 163–189.
Conference on Transportation Information and Safety, 890–894. Zeng, Z., Zio, E., 2017. A classification-based framework for trustworthiness assessment of
Xiang, W., Stuber, R.M.B., Meng, X., 2011. Meeting critical challenges and striving for quantitative risk analysis. Saf. Sci. 99, 215–226.
urban sustainability in China. Landsc. Urban Plan. 100, 418–420. Zhao, M., Liu, X., 2017. Regional risk assessment for urban major hazards based on GIS
Zhou, Y., Liu, M., 2012. Risk assessment of major hazards and its application in urban geoprocessing to improve public safety. Saf. Sci. 97, 112–119.
planning: a case study. Risk Anal. 32, 566–577.

203