You are on page 1of 10

Safety Science 134 (2021) 105079

Contents lists available at ScienceDirect

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

Internet of people enabled framework for evaluating performance loss and


resilience of urban critical infrastructures
Faxi Yuan a, *, Rui Liu a, *, Liang Mao b, Min Li c
a
M.E. Rinker, Sr. School of Construction Management, University of Florida, Gainesville, FL 32611, USA
b
Department of Geography, University of Florida, Gainesville, FL 32611, USA
c
Department of Anthropology and Sociology, Western Carolina University, Cullowhee, NC 28723, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Critical infrastructures (CIs) such as road networks play a critical role in transporting affected people to hospitals
Internet of People (IoP) and shelters during disasters. Timely evaluation of road networks’ performance loss and resilience enables
Natural disasters emergency management agencies (EMAs) to make quick and optimized decisions on resource allocation to
Road network analysis
critical road segments and bridges. However, real-time monitoring of road network conditions has been recog­
Performance loss
Resilience
nized as a challenge during major natural disasters because the affected regions are often inaccessible. This
Social media research introduces the Internet of People (IoP) enabled framework to assess road network’s performance loss
during disasters and proposes a performance loss rate to evaluate the road network resilience. The framework is
illustrated by a case study of Hurricane Florence in Wilmington, North Caroline, USA. Through semantic analysis
of road-related social media data, this study identifies the affected roads during the hurricane. To further
evaluate the road network resilience, this research utilizes road network analysis to generate a service zone for
each main hospital before and after the hurricane. The population-weighted travel time in each service zone are
compared before and after the hurricane to indicate the degree of performance loss. The findings show that our
IoP enabled framework can effectively support CI protection for building smart and resilient city environments.

1. Introduction qualitative evaluation results of CIs resilience mainly depend on the


survey data collected from subject matter experts (SMEs). The SMEs’
A resilient city should have the capacity to quickly respond to and feedbacks are specific to a single CI sector and reflect their subjectivity
recover from stressors, such as natural disasters (Cutter et al., 2008). and varying responses (Vugrin et al., 2010). As a result, the qualitative
Such capacity heavily replies on data-intensive response systems that method can hardly be applied across various CI sectors. On the other
can timely identify affected or damaged critical infrastructures (CIs) side, the quantitative methods for evaluating CIs resilience mostly used
during disasters. CIs include multiple interdependent physical systems the ratio or difference between CIs function after disruptions happened
such as roads, sewer systems, power grids, hospitals, and some others, and their full function before disruptions (Omer et al., 2009; Reed et al.,
which provide critical services in supporting community functioning 2009). Additionally, there are several studies concentrated on CIs in­
(Comes and Van de Walle, 2014). Failures of CIs can lead to severe terdependencies by system dynamics and network analysis (Guidotti
impacts on communities (Kim and Hastak, 2018), physical environment et al., 2016; Ouyang, 2014). These modelling methods depend heavily
(Lu et al., 2018) and public health (Patz et al., 2014). Hence, real-time on available data while the accessibility to the data of CIs and disaster is
evaluation of CIs’ performance loss and resilience can assist the emer­ limited and the data accuracy also remains insufficient (Labaka et al.,
gency management agencies (EMAs) to optimally deploy limited re­ 2016).
sources to the most needed areas, which benefits the protection of CIs Additionally, existing studies have indicated that the resilience
and establishment of resilient environments. evaluation methods vary according to various CIs sectors such as electric
Previous studies have applied both qualitative and quantitative power system, transport and water distribution system (Guidotti et al.,
methods for CIs resilience evaluation (e.g., Shen and Tang, 2015; Vugrin 2016; Hosseini and Barker, 2016). More methods and metrics for
et al., 2010). There are limitations in these methods. On one side, the assessing electric power system resilience have been proposed than that

* Corresponding authors.
E-mail addresses: faxi.yuan@ufl.edu (F. Yuan), liurui@ufl.edu (R. Liu), liangmao@ufl.edu (L. Mao), mli@wcu.edu (M. Li).

https://doi.org/10.1016/j.ssci.2020.105079
Received 26 March 2020; Received in revised form 24 June 2020; Accepted 25 October 2020
Available online 4 November 2020
0925-7535/© 2020 Elsevier Ltd. All rights reserved.
F. Yuan et al. Safety Science 134 (2021) 105079

for transport system (Shafieezadeh and Burden, 2014). Several studies for measuring road network resilience with social media data. The IoP
have utilized the closeness centrality of network and graph theory to framework considers the social media users as human sensors (Congosto
measure the transport system and supply chain resilience (Omer et al., et al., 2015) and collects the CIs-related data from the social media. As
2014; Soni et al., 2014). These methods compared the performance the concentration of this research is on the evaluation of systemic impact
index of transport system before and after disruptions occurred to on the road network, the technologies for data communication and
measure its resilience (Omer et al., 2014). Transport system such as road storage, the protocols for ensuring reliable and secure data communi­
network acts as a critical role to support affected people’s evacuation cation, and the design of user interface will be not be discussed with
activities which can be evaluated by the evacuation time. However, details in this paper. Utilizing the City of Wilmington, North Carolina,
evacuation activity was not considered in the mostly existing resilience one of the most affected regions in Hurricane Florence as the case study
evaluation methods for the road network. This research concentrates on area, this research validates the proposed IoP framework with the con­
resolving the data limitation issue and the involvement of evacuation centration on the performance loss evaluation of the road network. A
activity for evaluating road network resilience. comparative analysis of the population-weighted travel time in the
Recent studies have used remote sensing techniques such as satellite normal and hurricane periods is performed, and the variance in the
and UAVs for collecting CIs data during disasters (Hackl et al., 2018; population-weighted travel time is further used for assessing the per­
Yuan and Liu, 2018a). Remote sensing data for CIs is only accessible to formance loss of the road network in Wilmington NC. As a result, the
regional or federal stakeholders, while the affected communities and suggestion for evaluating the road network resilience is proposed by
local EMAs usually do not have access to this dataset (Fan and Mostafavi, using its performance loss rate in Hurricane Florence. The result can
2019). Additionally, remote sensing techniques cannot provide CIs data benefit the EMAs and urban planners in identifying the critical links in
at local level such as certain flooded roads or streets during hurricanes the road network and further building the resilient CIs. It also paves a
(Fan and Mostafavi, 2019), while this kind of information is particularly road for the future research on Internet of Things (IoT) enabled resilient
critical for the affected people and the EMAs to make evacuation stra­ CIs for smart city and resilient environment.
tegies during the crisis. Meanwhile, various studies have employed so­ The remainder of the paper is organized as follows. Section 2 pre­
cial media data to detect CIs disruptions during natural disasters (Chen sents related studies in critical infrastructure resilience and social media
et al., 2020; Fan and Mostafavi, 2019; Yuan and Liu, 2020; Petersen analysis for critical infrastructure resilience. Section 3 describes the
et al., 2017; Cervone et al., 2016; Congosto et al., 2015). Cervone et al. methods and materials for measuring road network resilience. Section 4
(2016) have employed social media data to identify the hot spots in provides the analysis results. Section 5 concludes this paper with dis­
2013 Boulder flood and further used remote sensing techniques to cussion and conclusions.
collect the damage data of the public transportation system. Fan and
Mostafavi (2019) used two specific CIs’ names in the city of Houston to 2. Related studies
filter the CIs-related Twitter data in Hurricane Harvey. Thereafter, they
mapped the frequency of CIs-related Twitter data over time to detect the 2.1. Critical infrastructure resilience
burst timeframes. Using the content similarity calculation, their research
built the semantic graph and further used the graph to identify the CIs, as a critical component of a community, include multiple
credible and critical CIs-related Twitter data to detect CIs’ disruptions in interdependent physical systems such as roads and power grids, which
Hurricane Harvey (Fan and Mostafavi, 2019). Hence, social media data provide critical services in supporting community functioning (Comes
indices its potential to support the awareness of disaster impacts on the and Van de Walle, 2014). The failures of CIs during natural disasters can
elements (e.g., roads and bridges) of CIs. bring severe influences on human systems such as their evacuation to
In this research, the authors follow the description of CI resilience in shelters and medical services (Kim and Hastak, 2018; Kadri et al., 2014).
Vugrin and Camphouse (2011), namely, the ability of CIs to “absorb, Recent literature and policies concentrating on critical infrastructure
adapt, and recover from the effects of a disruptive event while protection (CIP) have indicated a shift in emphasis from physical pro­
attempting to continue delivery of critical infrastructure services.” tection and asset hardening to that of resilience (Pursiainen, 2018;
Vugrin et al. (2010) have concluded two critical aspects for evaluating Vugrin and Camphouse, 2011). This shift was due to the acknowledge­
CIs resilience including systemic impact and total recovery effort. Sys­ ment that complete CIs protection can never be achieved due to CIs’
temic impact refers to changes in CIs performance while total recovery ageing process, varying disaster characters and the limited resources
effort means the quantity of resources used for recovery activities after (Croope and McNeil, 2011).
the disruption. This research is focused on the first aspect, systemic Various existing studies have concentrated on CIs resilience evalua­
impact, for evaluating CIs resilience. We take the road network as an tion methods and metrics (e.g., Vugrin and Camphouse, 2011). Firstly,
example of CIs sectors. Murdock et al. (2018) have measured road CIs resilience assessment methods mainly include the qualitative and
network performance loss by multiplying the number of affected people quantitative methods (Vugrin et al., 2010). Secondly, as CIs include
and delay in their travel time due to floods. Their measurement assumed various sectors such as transport, electricity, water and healthy, there
that people will travel to the same destinations during flood as that in were different methods for assessing the resilience for these different CIs
the normal period. The evacuation activity was not considered in Mur­ sectors (Panteli et al., 2017; Hosseini and Barker, 2016). This section
dock et al. (2018) while affected people need to evacuate to safe places introduces these two characters of the existing studies in CIs resilience
such as shelters in disasters. Through the identifications of affected evaluation.
people’s evacuation origins and destinations, this research will measure The qualitative method for evaluating CIs resilience needs the
the difference between evacuation time before and after disasters to resilience measurement index (RMI) as proposed in Panteli et al. (2017).
investigate road network resilience. For the EMAs, the awareness of The RMI include three level components. The first-level component in­
disaster impacts on road network, can help them to make suggestions for cludes preparedness, mitigation measures, response capabilities, and
the affected people’s selections of evacuation routes, but also benefit recovery mechanisms. Each first-level component has its second- and
them in deploying the limited resources to the critical links and roads on third-level components. In total, there are 10 second-level and 31 third-
the affected people’s evacuation routes. For the affected people during level components in the RMI. Users can collect the subjective evalua­
the crisis, the understanding of disaster impacts on road network can tions by subject matter experts (SMEs) for the 31 components in the RMI
help them make decisions on where to evacuate and which routes they to evaluate CIs resilience. Additionally, Fiksel (2003) has collected the
can use for their evacuations. responses of the SMEs to assess CIs three characters including absorptive
Based on the Internet of Things (IoT) architecture in Lin et al. (2017), capacity, adaptive capacity, and restorative capacity. The assessment
this research proposes the Internet of People (IoP) enabled framework results of these three capacities were further applied for evaluating CIs

2
F. Yuan et al. Safety Science 134 (2021) 105079

resilience. The qualitative methods depending on either the RMI or three (Hosseini and Barker, 2016; Shafieezadeh and Burden, 2014). Omer
capacities can be hardly applied consistently across different CI sectors et al. (2014) employed the difference of the closeness centrality of
due to the subjectivity of survey data and SMEs’ varying responses network before and after disaster events to measure CIs resilience. This
(Vugrin et al., 2010). is consistent with the quantitative method for evaluating CIs resilience
The quantitative evaluation methods for CIs resilience were also as in Reed et al. (2009). Youn et al. (2011) used the sum of reliability
popularly proposed and implemented in recent studies (Bruneau et al., (passive survival rate) and restoration (proactive survival rate) to
2003; Ouyang and Duenas-Osorio, 2014). Bruneau et al. (2003) have measure CIs system resilience. Soni et al. (2014) used the graph theory
quantified the expected degradation in CIs’ quality to measure their to establish a deterministic model to quantify the supply chain resil­
seismic resilience. Reed et al. (2009) have quantified the difference ience. Their method can capture the environment dynamics. During
between the capacity of CIs full functions and their capacity after di­ disruptions such as natural disasters, a critical function of transport
sasters approached. Omer et al. (2009) have employed the ratio of the system such as road network is to support affected people’s evacuations
telecommunication cable network’s delivery value after disasters to its which can be measured by the evacuation time. However, the quanti­
full-function delivery value to assess its resilience. Similarly, Henry and tative methods for measuring road network resilience either neglected
Ramirez-Marquez (2012) have applied the percentage of CIs’ recovered the travel time or failed to consider affected people’s evacuation activ­
delivery function since their disrupted state to measure their resilience. ities during disasters.
In addition, there are studies focused on the probability-based method
for measuring CIs resilience (Ouyang and Duenas-Osorio, 2014). Chang
and Shinozuka (2004) have measured CIs resilience as the probability 2.2. Social media analysis for critical infrastructure resilience
for situations where CIs’ robustness loss are lower than the maximum
acceptable loss of CIs’ performance and the time to CIs’ full recovery is Situation awareness was defined as ‘the perception of the elements in
less than the maximum acceptable disruption time. The CIs’ robustness the environment within a volume of time and space, the comprehension
loss and time to their full recovery were generated through a set of of their meaning, and the projection of their status in the near future’ in
Monte Carlo simulations. Furthermore, several studies have concen­ Endsley (1988). Within the disaster filed, situation awareness mainly
trated on CIs interdependencies to investigate CIs resilience. These refers to timely identification of disaster evolvements such as how the
studies have implemented system dynamics, agent-based modelling and floods and wildfires developed, and disaster impacts on infrastructures,
network analysis to explore CIs interdependencies (Guidotti et al., 2016; humans, economy, and environments, such as which roads and build­
Ouyang, 2014). For instance, Guidotti et al. (2016) have considered both ings were flooded and blocked in hurricanes and floods. Situation
the direct earthquake impacts on water distribution network and the awareness of timely disaster impacts on CIs can benefit emergency
indirect impacts from the damages on electric power network to model management agencies (EMAs) in delivering effective response strategies
CIs resilience. The critical challenges in implementing the above- (Yuan and Liu, 2020; Yuan et al., 2020; Yuan and Liu, 2019; Kryva­
mentioned quantitative methods for evaluating CIs resilience include sheyeu et al., 2016) and deploying the limited resources to the critical
limited availability and accessibility to data, and insufficient data ac­ nodes in the CIs network. However, the collection of real-time data of
curacy (Labaka et al., 2016). CIs’ functions during disasters is particularly challenging during major
As indicated in the second character of CIs resilience evaluation disaster events (Zhai et al., 2020; Yuan and Liu, 2018c; Yuan and Liu,
methods, different CIs sectors will need different variables for assessing 2018b). Traditional empirical data collection methods (e.g., field sur­
their performance loss and further resilience. Disaster/disruption impact vey) in crises, always takes a long time to reach victims (Spence et al.,
on CIs performances is a critical aspect for evaluating CIs resilience 2016), which can further delay the crisis responses.
(Shen and Tang, 2015). Vugrin et al. (2010) have summarized several Previous studies have investigated the implementation of remote
examples for different CIs sectors’ performance metrics. For instance, sensing techniques such as satellites and UAVs for collecting CIs data
they proposed to use the ‘number and efficiencies of cyber attacks’ to during disasters (Hackl et al., 2018; Adams et al., 2010). However, the
evaluate the performance of information technology, and to use the accessibility to the high-resolution satellite data is limited to regional or
‘average speed and cost of shipments’ to assess the performance of federal stakeholders, while the locally affected communities and EMAs
highway systems. Murdock et al. (2018) have used two different for­ cannot reach this dataset. Additionally, Fan and Mostafavi (2019) have
mulas to evaluate the flood impacts on power and transport systems. indicated that the CIs data from the satellites cannot provide the detailed
They applied the product of the number of affected people and the conditions of specific roads or bridges affected by floods or hurricanes,
outage duration to measure the flood impacts on power system, while while this information plays a critical role in supporting the EMAs and
utilized the product of the number of affected people and the difference affected people’s response activities such as the evacuation to shelters
between total travel time and typical time to reflect the disaster impacts and hospitals. Meanwhile, data collection for CIs (e.g., local roads) by
on transport system (Murdock et al., 2018). Hence, there is no general UAVs is also difficult due to the large affected areas in major natural
resilience assessment method which can fit in all CIs sectors. disaster events such as Hurricane Florence (Cervone et al., 2016).
In terms of simulation models on CIs resilience, more efforts have The advanced development of social media and data analytic
been spread to electric power system than the transport systems methods provides a solution to collect timely data of CIs’ functions in
disasters (e.g., Fan and Mostafavi, 2019). Social media users as human

Fig. 1. IoP Enabled Framework for Evaluating Road Network Resilience.

3
F. Yuan et al. Safety Science 134 (2021) 105079

sensors were found to post their feelings and observations of CIs’ func­
tions and conditions during a disaster (e.g., Cervone et al., 2016) and
normal periods (e.g., Congosto et al., 2015). These studies mainly
employed CIs’ names in the affected cities such as the Barker reservoir in
the City of Houston in Texas (Fan and Mostafavi, 2019) and Madrid
metro in Madrid (Congosto et al., 2015), to identify the CIs-related social
media data. On one side, there is very limited CIs’ names used for
filtering the CIs-related social media data in Fan and Mostafavi (2019)
and Congosto et al. (2015). On the other side, a systemic evolution for
the function loss or performance decrease of the CIs system/network in
the disaster period was not performed.
In summary, existing studies applying social media data for detecting
CIs’ failures and damages during disasters concentrate on CIs at the
individual level, while a systemic evaluation of the performance loss and
resilience of CIs system/network during disasters is not perfumed. The
timely evolution of the performance loss and resilience of CIs system/
network can help the EMAs to make optimal use of limited resources to
the broken CIs and further support the evacuation of affected commu­
nities to shelters and medical services during the crisis. Hence, this Fig. 2. The Hospitals and Road Network in Wilmington NC.
research proposes the Internet of People (IoP) enabled framework to
evaluate the performance loss and resilience of CIs system/network 3.3. Resilience evaluation
during the disasters.
This section focuses on the assessment of the performance loss of
3. Material and methods road networks in Wilmington NC. It mainly consists of four steps
including filter of road-related Twitter data, identification of affected
Aiming at evaluating CIs system/network resilience during natural and damaged roads in Wilmington NC, analysis of road network in terms
disasters such as Hurricane Florence, this research proposes the Internet of travel time from population centers to the main hospitals before and
of People (IoP) enabled framework as illustrated in Fig. 1. This research after the hurricane came to the city, and quantification of performance
uses the road network as an example of the CIs system. This framework loss of the road network to reflect the road network resilience in Wil­
has referred to the Internet of Things (IoT) architecture in Jia et al. mington NC. The following introduces each step.
(2019) and Lin et al. (2017). The performance loss of road network is
will be used for evaluating its resilience. Step 1. Selection of road-related social media data

3.1. Information collection This step employs a lexicon-based approach to filter the Twitter data
talking about the roads (road-related social media data) in Wilmington
Twitter users are treated as human sensors (Congosto et al., 2015) NC. Specifically, this research uses the road network shapefile released
and their posts on Twitter reflected their observed information in the by the North Carolina Department of Transportation (NCDOT) and ex­
hurricane. The perception layer of IoT architecture uses technologies tracts the road names appearing this shapefile. These road names are
such as RFID (radio-frequency identification) and Wireless Sensor Net­ further used to establish the road name lexicon. Compared with Fan and
works (Andrea et al., 2015; Zhao and Ge, 2013). The IoP framework in Mostafavi (2019), this research has expanded the number of keywords
this research replaces the physical sensors in the IoT architecture with (i.e., the road name lexicon) for filtering the CI-related social media
human sensors. Human sensors’ posts on Twitter were collected and data. Using this road name lexicon, this study implements the unsu­
further processed and filtered for resilience evaluation. Using Twitter pervised machine learning method to filter the road-related (i.e., CI-
API, this section collected the Twitter data posted in our case study area related) social media data posted during disasters. To implement the
during the crisis. unsupervised method, this section develops the algorithm to check if a
collected tweet contains the road names in the road name lexicon. If the
3.2. Data communication and storage tweet contains a road name in the road name lexicon, the algorithm will
label it as road related. Integrating the developed algorithm with Twitter
A comprehensive IoT architecture includes the network layer using API by Python, this research achieves the real-time and automatic
the technologies for data communication and storage such as 6LoWPAN annotation of road-related tweets. The road-related tweets will be
(Tan and Koo, 2014), Long-Term Evolution (LTE) (Lin et al., 2017), further manually annotated by researchers to decide if they contain the
Zigbee (Pandya and Champaneria, 2015) and cloud server (Jia et al., road damage information during disasters.
2019). The network layer of IoT architecture also involves with the
protocols such as IEEE 802.15.4 (Al-Fuqaha et al., 2015; Gan et al., Step 2. Identification of affected and damaged roads
2011), 6LoWPAN (Tan and Koo, 2014), ZigBee (Pandya and Champa­
neria, 2015) and Z-Wave (Tan and Koo, 2014). These protocols help to This research invites human annotators with disaster research ex­
promote reliable and secure communication in IoT architecture. As this periences to manually review the road-related social media data and the
research concentrates on the evaluation of performance loss and resil­ web links included in the data. The road names, types (e.g., street and
ience of the road network, the detailed technologies for data commu­ drive) and directions (e.g., from north to south), hurricane impacts on
nication and storage and protocols will not be introduced in this section. the roads such as flooded and blocked due to fallen trees or electric lines,
In the IoP framework of this research where the Twitter data was and their locations (with exact latitudes and longitudes) are recorded.
collected in 2018 (Hurricane Florence period), the main technologies for The affected roads are further integrated into the road network for the
data communication are 4G and Long-Term Evolution (LTE). Twitter road network analysis in the hurricane period.
users posted their feelings and observations through their mobile and
computer devices in Hurricane Florence. This research uses the local Step 3. Analysis of road network
database to store the collected Twitter data from human sensors.

4
F. Yuan et al. Safety Science 134 (2021) 105079

During hurricanes, many public buildings such as schools, hospitals


and senior centers are often used as temporary shelters in North Carolina
(ABC11 2018; North Carolina News, 2018; The News and Observer,
2018). As a result, this research selects the main hospitals as evacuees’
travel destinations and does not discriminate hospital characteristics
such as the number of beds and doctors. The hospitals are treated as
public buildings that can accommodate any evacuee in needs, not
necessarily patients. Four main hospitals in Wilmington NC are included
for analysis, namely Downtown Medical Center (D1), New Hanover
Regional Medical Center (D2), Wilmington Plasma Center (D3) and
Nunnelee Pediatric Specialty Clinic at Autumn Hall (D4), as shown in
Fig. 2. Based upon the road network and on-road speed limits (Fig. 2),
this research estimates people’s travel time to the nearest hospitals
before and during the hurricane. The population centers by block group
are considered as travel origins (O) (United States Census Bureau, 2010).
Based on the travel time of block groups, Voronoi diagrams are gener­
ated as the service zones of these four hospitals (Okabe et al., 2006).
That is, if the travel time from a population center (O) to a hospital (D) is
shorter than its travel time to all other hospitals, we define the block Fig. 3. The Collected Tweets Generated in Hurricane Florence.
group as a part of the service zone of hospital D.
Thereafter, this research calculates the shortest travel time tij from data during the hurricane in Wilmington NC, this research has not
each population center i to its closest hospital j before and after the considered the timely traffic data to calculate the travel costs.
hurricane. The calculation utilizes ArcGIS network analyst and imple­ In Eq. (3), the divergent issue cannot appear when calculating the
ments the maximum allowed travel speed (i.e., speed limitation value travel cost increase within each hospital’s service zone. Specifically,
from NCDOT). tij at the block group level are further aggregated to T’ Tj’ − Tj
service zone and the entire study area as a population-weighted sum, travel cost increase equals to the difference of Tjj and 1 (i.e., Tj ). Using
formulated in Eq. (1) and (2). Both equations are applied to the road the shortest travel time from each population center to hospitals, this
network before and after the hurricane to estimate performance loss of research defines the service zone for each hospital and further calculates
road networks due to hurricane. the population-weighted travel time within their service zones (Eq. (1)).
∑nj This indicates the values of travel cost before the hurricane are finite.
For the service zone of hospital j : Tj = ∑i=1nj
Pij tij
(1) After the hurricane came to the city, some roads and bridges were not
i=1 Pij available, which can result in the increase of travel time from population
∑4 ∑nj centers to corresponding hospitals. However, the values of Tj’ were not
i=1 Pij tij infinitely large. For instance, if all the accesses to hospital D1 were
(2)
j=1
For the entire study area : T = ∑4 ∑nj
j=1 i=1 Pij closed due to hurricane, the travel time from the population centers to
hospital D1 will be infinite. To resolve this issue and also consider the
where Tj is the population-weighted travel time within the service zone real situations in disasters, we propose to re-define the service zones of
of hospital j; nj is the total number of block groups in the service zone of all the hospitals after the hurricane. Therefore, if all the accesses to
hospital j (j = 1, 2, 3, and 4); Pij denotes the number of people in the hospital D1 were closed, we will eliminate the service zone of hospital
block group i within the service zone of hospital j; T is population- D1. The block groups within the service zone of hospital D1 before the
weighted travel time from population centers to the hospitals in the hurricane will be redistributed to the service zones of other hospitals (e.
entire city. g., D2, D3 and D4) using the shortest travel time. As a result, the value of
Tj’ can become larger compared with Tj while it cannot be infinitely
Step 4. Measuring the performance loss of road network large.

Following Gauthier et al. (2018), this paper uses the performance 4. Results
loss of the road network during disasters as an indicator of road network
resilience. Due to the impact of hurricane on roads, the service zone of 4.1. Social media data
each hospital may change before and after the hurricane, resulting in a
difference in people’s travel costs to the nearest hospital. The perfor­ Using Twitter API, the authors have collected the Twitter data posted
mance loss of road network is measured as the increase of people’s travel in the City of Wilmington, North Carolina (NC) starting at 11:00:00 am
cost, i.e., the difference between the shortest travel time from block on Sep 11, 2018, and ending at 11:59:59 pm on Sep 22, 2018 (US Eastern
groups to hospitals before and after the disaster, formulated as follows. Time). A spatial bounding box with latitude from 30.355591 to
Tj’ − Tj 40.638801, longitude from − 80.751429 to − 75.789148, was initially
For the service zone of hospitalj : FL Tj = (3) used for streaming Twitter data. Wilmington is one of the most affected
Tj
regions in the hurricane (USA Today, 2018) and therefore is selected as
T’ − T our case study region. Using the shapefile of Wilmington NC (data
For the entire study area : F L T = (4) source: New Hanover County), this research projects the collected geo-
T
located Twitter data into a map and filter the geo-located Twitter data
where j = 1, 2, 3, and 4, represents different hospitals; Tj and Tj’ are the posted in Wilmington NC.
population-weighted travel time within the service zone of hospital j This research collects 2,604,154 tweets within the pre-defined
before and after the disaster; FL Tj represents the performance loss of spatial bounding box. Using the shapefile of Wilmington NC, we
road network in the service zone of hospital j in terms of travel time; filtered the Tweets data posted in Wilmington NC (see nodes in Fig. 3).
FL Trepresents the performance loss of road network in the entire city of In total, 878 geo-located Tweets (with exact latitudes and longitudes)
Wilmington NC in terms of travel time. Due to the limitation of traffic were collected in Wilmington NC in Hurricane Florence.

5
F. Yuan et al. Safety Science 134 (2021) 105079

Table 1 including text and image refers to where the annotators get the infor­
Description of Affected Roads in Wilmington NC. mation of these affected roads in the road-related social media data.
Road names Data type Descriptions The road-related tweets account for 2.96% in the 878 geo-located
tweets collected in Wilmington NC. Hurricane Florence made landfall
Kerr Ave Text Kerr Ave after the hurricane was Flooded in for 2
days. in Wilmington on Sep 14, 2018 and dissipated on Sep 18, 2018. The
Burnett BLV Text Carolina Beach RD is impassable at the intersection temporal distribution of road-related tweets was illustrated in Fig. 4.
of Burnett BLV. Fig. 4 illustrates that road-related tweets were initially posted after
New Center Text Car was stalled on New Center DR East of College RD Hurricane Florence made landfall in Wilmington (Sep 14) and reached
DR due to flooded street.
Oleander DR Text High water was reported at the intersection of
the peak value on Sep 15. Before hurricane came to Wilmington (from
College RD and NC-132 and Oleander DR (US-76) in Sep 11 to Sep 13), there was no road-related tweet generated. When
Wilmington, NC. Hurricane Florence dissipated (Sep 18), there was no road-related tweet
Princess Place Text Princess Place DR is impassable between 17th ST and generated. After Sep 18, this research found four road-related tweets
DR 23rd ST.
posted from Sep 19, 2018 to Sep 22, 2018. Although the road-related
Whisper Creek Text Whisper Creek LN near Wade Park is blocked on both
LN ends by large trees down on road. tweets account for only 2.96% in the collected data, we have identi­
South College Text Flooding in the area of Oleander DR and South fied 13 affected roads such as flooded and blocked due to fallen trees,
RD College RD. through our manual annotations with the road-related data. However,
Carolina Text Carolina Beach RD is impassable at the intersection of the road-related tweets cannot reflect on when the affected roads were
Beach RD Burnett BLV due to downed power lines and trees.
Market ST Text High waters reported on Market ST (US-17 Business)
recovered. As some roads were reported as flooded even after Hurricane
between 11th and 21st ST. Downed power lines and Florence dissipated on Sep 18, 2018, such as Kerr Ave was reported as
trees were also reported. flooded on Sep 20, 2018, this research assumes that the affected roads
16th ST Text 16th ST has collapsed at the intersection with were not recovered in our study period from Sep 14, 2018 to Sep 22,
Greenfield ST in Wilmington, NC.
2018. Another limitation of Twitter data is that they cannot cover all the
NC-132 Text High water was reported at the intersection of
College RD and NC-132 and Oleander DR (US-76) in road conditions during disasters such as Hurricane Florence. Our future
Wilmington, NC. work will implement the IoT system for monitoring road conditions,
I40 East Image & That’s actually I40 (image). Wilmington is an island which will be introduced in the discussion section.
Text and you need a boat to get down I40 East.
US-421 Text US-421 washing out from the Northeast Cape Fear
Rive
4.3. Road network analysis
Others Image Posted with images without specific road names.

Before the hurricane (Fig. 5a), 18 block groups are included into the
4.2. Damaged roads description service zone of Downtown Medical Center (D1), while there are 24 block
groups for New Hanover Regional Medical Center (D2), 14 for Wil­
Using the road network shapefile of NCDOT, this study collects 518 mington Plasma Center (D3) and 14 for Nunnelee Pediatric Specialty
road names such as 6th Street and 5th Avenue in Wilmington NC. 18 Clinic at Autumn Hall (D4). Due the impact of hurricane on roads, the
road types are assigned with these 518 roads in Wilmington NC, travel time from many block groups to the hospitals were changed, and
including ALY, AVE, BLV, CIR, CT, DR, LN, LP, PKY, PL, PLZ, PT, RD, as a result, the service zone of each hospital changed accordingly
ROW, RUN, ST, WAY, and WLK. The descriptions of each road type are (Fig. 5b). For instance (Fig. 6), the failure of Market ST between 11th ST
presented in Table A1 in the Appendix (adjusted from Charlotte County and 21st ST has resulted in the major changes in the routes from pop­
Florida Government Portal). This research establishes the road name ulation centers to hospital D1. The travel route from the population
lexicon with the 518 road names of Wilmington NC. Each tweet con­ center to hospital D1 has shifted from the white route to the black route.
taining the road name appearing in the lexicon is labeled as a road- The collapse of 16th ST at the intersection with Greenfield ST has
related tweet. Utilizing the lexicon, we filter 26 road-related tweets. resulted in major changes in the routes from population centers to
Through manually review of the road-related tweets, this study iden­ hospital D2. The flood at the interception of New Center DR and College
tifies 13 affected or damaged roads by the flood due to Hurricane RD has blocked many routes from population centers to hospital D3.
Florence (see black sun cross in Fig. 5b). The affected or flooded roads In Fig. 5a and b, the four different colors including light grey, grey,
without specific names are labeled as ‘Others’ in Table 1. The de­ dark grey and black represent the service zones of hospital D1, D2, D3,
scriptions of these affected roads are illustrated in Table 1. Data type and D4. The rhombus points represent the locations of the population
centers in the block group, while the black sun cross points (see Fig. 5b)

Fig. 4. Temporal distribution of road-related tweets.

6
F. Yuan et al. Safety Science 134 (2021) 105079

Fig. 5. The Service Zones of Hospitals before (5a-left) and after (5b-right) Hurricane Florence in Wilmington NC.

Fig. 6. The Travel Scenario Samples in Service Zones of Hospital D1, D2, and D3.

7
F. Yuan et al. Safety Science 134 (2021) 105079

Table 2
Travel Time Before and After Hurricane Florence in Wilmington NC.
Service Number of block Number of block Served population Served population Population weighted Population weighted Population
zone groups before groups after before Hurricane after Hurricane travel time (min) before travel time (min) after weighted time
Hurricane Florence Hurricane Florence Florence Florence Hurricane Florence Hurricane Florence variance

D1 18 20 17,296 18,786 6.21 6.96 12.08%


D2 24 22 38,256 36,766 9.59 9.74 1.56%
D3 14 11 27,145 22,149 7.72 8.71 12.82%
D4 14 17 20,821 25,817 7.49 8.24 10.01%
Entire 70 70 103,518 103,518 8.11 8.64 6.54%
city

represent the affected roads as illustrated in Table 1. These service zones as a case study region, this study validates the IoP framework. This paper
were defined using the concept of the network-based Voronoi diagram. concentrates on the information collected from human sensors and the
If the travel time from a population center (O) to a hospital (D) is shorter resilience evaluation. Referring to the shapefile from NCDOT, we build
than its travel time to all other hospitals, this research assigns the block the road name lexicon and further apply this lexicon to filter the road-
group having this population center (O) with the service zone of the related (i.e., CI-related) social media data posted in Wilmington NC
hospital (D). For instance, within the service zone of hospital D1 during Hurricane Florence. The affected roads are identified through
(Fig. 5a), the travel time from the population centers within the light human annotations of the road-related social media data.
grey zone to hospital D1 is shorter than their travel time to other hos­ To conduct the road network analysis, this paper defines the four
pitals including D2, D3 and D4. This situation also happens within the main hospitals in Wilmington NC as the travel destinations and the
service zones of other hospitals. The affected or damaged roads were location of population centers by block group (United States Census
identified by the annotations of our two graduate students with disaster Bureau, 2010) as the travel origins. Using the concept of network-based
experiences with the road-related social media data. Voronoi diagram, this research defines the service zones of these four
main hospitals in Wilmington NC by considering the shortest travel time
4.4. Performance loss from the population centers to these hospitals with the road network in
Wilmington NC. To evaluate the road network performance, we intro­
Table 2 compares characteristics of hospital service zones before and duce the index of population-weighted travel time. Thereafter, this
after the hurricane. The served populations in the service zones of research computes the population-weighted travel time by service zones
hospital D1 and D4 have increased while that in the service zones of of these four hospitals and the City of Wilmington NC (see Eqs. (1) and
hospital D2 and D3 have decreased. The population-weighted travel time (2)). Considering the affected or flooded roads in Hurricane Florence,
in the service zones of these four hospitals have increased after the this study recalculates the population-weighted travel time from the
hurricane. In the service zone of hospital D1 (Downtown Medical Cen­ population centers to the hospitals. The new shortest travel time has
ter), the population-weighted travel time has increased by 12.08% been considered to re-allocate these block groups to the service zones of
during the hurricane period. In the service zone of hospital D2 (New the four hospitals in Wilmington NC (see Fig. 5a and b). The variance in
Hanover Regional Medical Center), the variance in the population- the population-weighted travel time from normal period to hurricane
weighted travel time is 1.56%, which is less than that in the service period is a 6.54% increase in the travel time per person, which is further
zone of hospital D1. The largest variance in the population-weighted applied for reflecting the performance loss in the road network of Wil­
travel time presents in the service zone of hospital D3 (Wilmington mington NC.
Plasma Center), which is 12.82%. The significant variance in the This research provides emergency management agencies (EMAs)
population-weighted travel time in the service zone hospital D3 was with a framework to evaluate the road network performance loss and
mainly caused by the floods at the intersection of New Center DR East resilience during the crisis such as Hurricane Florence. The under­
and the College RD. Hospital D3 is near this intersection and many routes standing of road network performance loss and resilience during the
from the population centers within its service zone to the hospital D3 disasters can benefit the EMAs to deliver on-target repair resources to
have changed. There is no affected or damaged road reported on social the critical roads whose failures can significantly reduce the road
media within the service zone of hospital D4 (Nunnelee Pediatric Spe­ network performance such as the floods at the intersection of New
cialty Clinic at Autumn Hall) which affects the routes from the popu­ Center DR East and the College RD seriously impacted the routes from
lation centers to the hospital D4. However, due to the travel time population centers to the hospital D3.
variances from three population centers to hospital D3, these three block Although many CIs resilience frameworks and metrics have been
groups were assigned with the service zone of hospital D4 after Hurri­ proposed with mathematic models, their applications are still limited
cane Florence came to the city. As a result, the population-weighted and lacking detailed prescriptions (Labaka et al., 2016). Mostly simu­
travel time in the service zones of hospital D4 has increased by 10.01%. lation models for CIs resilience are confronted with critical challenges
Considering the integrated road network in Wilmington NC, this including data limitation and insufficient data accuracy. Additionally,
research quantifies its performance loss by the population-weighted existing quantitative methods for measuring transport system resilience
travel time within the four service zones by Eq. (4). The results are mainly implement network attributes such as the closeness centrality
also illustrated in Table 2. The population-weighted travel time has seen and graph theory to reflect systemic impact on them (Omer et al., 2014;
an increase rate of 6.54% from the normal period to the hurricane Soni et al., 2014). The IoP framework not only involves social media
period. This result indicates the performance loss of the road network of data as a data source for exploring physical conditions of the road
Wilmington NC as 6.54% increase in the travel time per person after network, but also considers the evacuation activity for measuring road
Hurricane Florence came to the city. network performance during disasters. However, there is a limitation in
the current research which deserves further research efforts. There is no
5. Discussions and conclusion established threshold of the performance loss rate to evaluate the road
network resilience, where the performance loss rate refers to the in­
This research performs a case study of utilizing the quantitative crease rate in the population-weighted travel time. If the performance
method for evaluating road network resilience with its performance data loss rate reaches this threshold, we can say the road network is not
during real disturbance (i.e., Hurricane Florence). Using Wilmington NC resilient or at a medium or low resilience level. In this paper, we

8
F. Yuan et al. Safety Science 134 (2021) 105079

Table A1 performance loss evaluations of road networks in Wilmington NC.


Description of road types in Wilmington NC. Considering the limited number of affected roads reflected on social
Abbreviation Full name Descriptions media, this paper suggests the future research use physical sensors and
build the Internet of Things (IoT) enabled critical infrastructure infor­
ALY Alley a) A narrow passage; b) A short or narrow street; etc. *
AVE Avenue a) A wide or main thoroughfare; b) A thoroughfare mation system for evaluating the road network performance and resil­
running principally in a north–south direction and ience. The building of IoT enabled resilient critical infrastructure can
usually terminating at an east–west street. benefit critical infrastructure protection for smart cities and resilient
BLV Boulevard a) A wide, multi-lane arterial divided thoroughfare; b) environments. Future research can implement structural monitoring
A major thoroughfare running in a diagonal direction.
It must connect at least two sections and act as a
sensors for roads and bridges, wet road detection sensors and high-water
collector. detection sensors for monitoring road flooding, and traffic count sensors
CIR Circle A roadway containing a closed loop beginning and for monitoring congestion. As a result, more affected roads and detailed
ending at intersections with the same street, etc.* traffic conditions on the roads can be identified. Furthermore, the travel
CT Court a) A permanently closed streets; b) Dead end rights-of-
scenarios during hurricane period can involve more affected roads and
way under 1000 feet in length which run east and
west; etc. * real-time traffic conditions and further increase the reliability of the
DR Drive a) A curvilinear roadway of more than one thousand evaluation of road network performance loss.
feet (1000′ ) in length, generally designated by a name; In summary, this paper proposes and implements the IoP framework
b) Curving streets linger than 1000 feet; etc.* for evaluating road network performance loss and resilience during
LN Lane a) It could be short drives that begin and end in the
same street. (b) Circular or semicircular roads; etc.*
natural disasters. Through a comparative analysis of the population-
LP Loop a) could be short drives that begin and end in the same weighted travel time in the normal and hurricane periods, this study
street. (b) Circular or semicircular roads; etc.* quantifies the performance loss rate of the road network in Wilmington
PKY Parkway A special scenic route or park, drive abutting a park, NC in terms of 6.54% increase in the travel time per person. The IoP
greenway, or conservation area where zoning or
framework can be generalized to the evaluation of road network per­
topography would prohibit development on at least
one side of the roadway, etc.* formance loss and resilience in future disaster events across the US or
PL Place a) A short curvilinear or diagonal roadway less than other English-speaking countries. This research can benefit the EMAs in
one thousand feet (1000′ ) in length; b) A cul-de-sac or making optimal decisions to deploy limited resources to the critical links
permanent dead-end-road. (i.e., roads) on the evacuation routes. Future research is suggested on
PLZ Plaza None**
developing a performance loss rate threshold for evaluating resilience
PT Point None**
RD Road An identifiable route or path between two or more level and utilizing physical sensors (e.g., structural monitoring sensor
places, etc.* and high-water detection sensor) to monitor more roads in hurricanes
ROW Row None** and other natural disasters.
RUN Run None**
ST Street A street mainly facilitates public interaction rather
than transportation, etc.* Appendix A
WAY Way A minor street that changes direction or begins and
ends on the same thoroughfare, etc.* See Table A1.
WLK Walk None**

Note: References
*
See detailed descriptions from Charlotte County Florida Government Portal.
**
None: No official descriptions are found. ABC11, 2018. Florence: List of area shelters, emergency operations centers open.
Available at <https://abc11.com/emergency-shelters-wake-county-raleigh-durham-
cumberland/4211642/> (last accessed on Nov 14, 2019).
evaluate the performance loss of the road network in Wilmington NC as Adams, S., Friedland, C., Levitan, M., 2010. Unmanned aerial vehicle data acquisition for
a 6.54% increase in the travel time per person, while we cannot say the damage assessment in hurricane events. In: Proceedings of the 8th International
Workshop on Remote Sensing for Disaster Management, Tokyo, Japan.
road network is at high, medium, or low resilience level. Future research Alexander, L., Jiang, S., Murga, M., González, M.C., 2015. Origin–destination trips by
can conduct questionnaires among transportation and emergency purpose and time of day inferred from mobile phone data. Transp. Res. Part C:
management experts, and various case studies on the road network Emerg. Technol. 58, 240–250.
Al-Fuqaha, A., Guizani, M., Mohammadi, M., Aledhari, M., Ayyash, M., 2015. Internet of
performance analysis during disasters to define the threshold of per­ things: a survey on enabling technologies, protocols, and applications. IEEE
formance loss rate to evaluate the resilience level. Commun. Surv. Tutorials 17 (4), 2347–2376.
Another limitation of this research is the selected number of origins Andrea, I., Chrysostomou, C., Hadjichristofi, G., 2015. Internet of things: security
vulnerabilities and challenges. In: Proceedings of 2015 IEEE Symposium on
and destinations. This research employed population centers by block Computers and Communication (ISCC), Larnaca, Cyprus.
groups as travel origins. This cannot consider the evacuation dynamics Bruneau, M., Chang, S.E., Eguchi, R.T., Lee, G.C., O’Rourke, T.D., Reinhorn, A.M.,
for individuals during Hurricane Florence. Meanwhile, this paper Shinozuka, M., Tierney, K., Wallace, W.A., Von Winterfeldt, D., 2003. A framework
to quantitatively assess and enhance the seismic resilience of communities.
selected the four main hospitals as the travel destinations in the road Earthquake Spectra 19 (4), 733–752.
network analysis, while there can be other evacuation destinations such Cervone, G., Sava, E., Huang, Q., Schnebele, E., Harrison, J., Waters, N., 2016. Using
as temporary shelters and the current hospitals may not perform nor­ Twitter for tasking remote-sensing data collection and damage assessment: 2013
Boulder flood case study. Int. J. Remote Sens. 37 (1), 100–124.
mally in the hurricane period. Hence, future research can collect more
Chang, S.E., Shinozuka, M., 2004. Measuring improvements in the disaster resilience of
data on the locations providing medical and shelter services from the communities. Earthquake Spectra 20 (3), 739–755.
EMAs and hospitals. Additionally, existing studies have demonstrated Charlotte County Florida Government Portal, 2019. Street Type Definitions. available at
that using human mobility data (i.e., call detail records) from telecom­ <https://www.charlottecountyfl.gov/forms/streetnamesearch/lists/street%20type
%20definitions/allitems.aspx> (last accessed on Nov 14, 2019).
munications company (e.g., AT&T) for establishing the origin- Chen, Y., Wang, Q., Ji, W., 2020. Rapid assessment of disaster impacts on highways using
destination (O-D) matrix for humans’ evacuation activities is feasible social media. J. Manage. Eng. 36 (5) https://doi.org/10.1061/%28ASCE%
(Alexander et al., 2015). Each call detail record (CDR) data, includes the 29ME.1943-5479.0000836. In this issue.
Comes, T., Van de Walle, B., 2014. Measuring disaster resilience: The impact of hurricane
anonymous user ID, longitude, latitude, starting and end time of a phone sandy on critical infrastructure systems. In: Proceedings of ISCRAM 2014, University
call and SMS sending. Hence, future research can explore the CDR data Park, PA, USA.
to better define individuals’ travel origins and destination. The collec­ Congosto, M., Fuentes-Lorenzo, D., Sánchez, L., 2015. Microbloggers as sensors for public
transport breakdowns. IEEE Internet Comput. 19 (6), 18–25.
tion of more origins and destinations can increase the number of routes Croope, S.V., McNeil, S., 2011. Improving resilience of critical infrastructure systems
in the road network analysis, which can further increase the accuracy of postdisaster: recovery and mitigation. Transp. Res. Rec. 2234 (1), 3–13.

9
F. Yuan et al. Safety Science 134 (2021) 105079

Cutter, S.L., Barnes, L., Berry, M., Burton, C., Evans, E., Tate, E., Webb, J., 2008. A place- Patz, J.A., Grabow, M.L., Limaye, V.S., 2014. When it rains, it pours: future climate
based model for understanding community resilience to natural disasters. Global extremes and health. Ann. Global Health 80 (4), 332–344.
Environ. Change 18 (4), 598–606. Petersen, L., Fallou, L., Reilly, P., Serafinelli, E., 2017. Public expectations of social
Endsley, M.R., 1988. Design and evaluation for situation awareness enhancement. In: media use by critical infrastructure operators in crisis communication. In:
Proc., Human Factors Society annual meeting, Santa Monica, CA, USA. Proceedings of ISCRAM 2017, Albi, France.
Fan, C., Mostafavi, A., 2019. A graph-based method for social sensing of infrastructure Pursiainen, C., 2018. Critical infrastructure resilience: A Nordic model in the making?
disruptions in disasters. Comput.-Aided Civ. Infrastruct. Eng. https://doi.org/ Int. J. Disaster Risk Reduct. 27, 632–641.
10.1111/mice.12457. Reed, D.A., Kapur, K.C., Christie, R.D., 2009. Methodology for assessing the resilience of
Fiksel, J., 2003. Designing resilient, sustainable systems. Environ. Sci. Technol. 37 (23), networked infrastructure. IEEE Syst. J. 3 (2), 174–180.
5330–5339. Shafieezadeh, A., Burden, L.I., 2014. Scenario-based resilience assessment framework for
Gan, G., Lu, Z., Jiang, J., 2011. Internet of things security analysis. In: Proceedings of Int. critical infrastructure systems: Case study for seismic resilience of seaports. Reliab.
Conf. Internet Technol. Appl. (iTAP), Wuhan, China. Eng. Syst. Saf. 132, 207–219.
Gauthier, P., Furno, A., El Faouzi, N.E., 2018. Road network resilience: how to identify Shen, L., Tang, L., 2015. A resilience assessment framework for critical infrastructure
critical links subject to day-to-day disruptions. Transp. Res. Rec. 2672 (1), 54–65. systems. In: Proc., 2015 First international conference on reliability systems
Guidotti, R., Chmielewski, H., Unnikrishnan, V., Gardoni, P., McAllister, T., van de engineering (ICRSE), Beijing, China.
Lindt, J., 2016. Modeling the resilience of critical infrastructure: the role of network Soni, U., Jain, V., Kumar, S., 2014. Measuring supply chain resilience using a
dependencies. Sustain. Resilient Infrastruct. 1 (3–4), 153–168. deterministic modeling approach. Comput. Ind. Eng. 74, 11–25.
Hackl, J., Lam, J.C., Heitzler, M., Adey, B.T., Hurni, L., 2018. Estimating network related Spence, P., Lachlan, K., Rainear, A., 2016. Social media and crisis research: data
risks: a methodology and an application in the transport sector. Nat. Hazards Earth collection and directions. Comput. Hum. Behav. 54, 667–672.
Syst. Sci. 18 (8), 2273–2293. Tan, J., Koo, S.G., 2014. A survey of technologies in internet of things. In: Proceedings of
Henry, D., Ramirez-Marquez, J.E., 2012. Generic metrics and quantitative approaches for 2014 IEEE International Conference on Distributed Computing in Sensor Systems,
system resilience as a function of time. Reliab. Eng. Syst. Saf. 99, 114–122. Marina Del Rey, CA, USA.
Hosseini, S., Barker, K., 2016. Modeling infrastructure resilience using Bayesian The News & Observer, 2018. Coastal NC evacuees fleeing Hurricane Florence find shelter
networks: a case study of inland waterway ports. Comput. Ind. Eng. 93, 252–266. in Triangle schools. Available at <https://www.newsobserver.com/news/local/art
Jia, M., Komeily, A., Wang, Y., Srinivasan, R.S., 2019. Adopting Internet of Things for the icle218258835.html> (last accessed on Nov 14, 2019).
development of smart buildings: a review of enabling technologies and applications. United States Census Bureau, 2010. “Centers of Population.” available at <https://www.
Autom. Constr. 101, 111–126. census.gov/geographies/reference-files/time-series/geo/centers-population.html>
Kadri, F., Birregah, B., Châtelet, E., 2014. The impact of natural disasters on critical (last accessed on Nov 14, 2019).
infrastructures: a domino effect-based study. J. Homel. Secur. Emerg. Manage. 11 USA Today, 2018. Florence aftermath: 35 dead; flooding leaves North Carolina residents
(2), 217–241. seeking food, water. available at <https://www.usatoday.com/story/news/nation/
Kryvasheyeu, Y., Chen, H., Obradovich, N., Moro, E., Van Hentenryck, P., Fowler, J., 2018/09/18/florence-flooding-cuts-off-wilmington-food-water-distributed/1343
Cebrian, M., 2016. Rapid assessment of disaster damage using social media activity. 567002/> (last accessed on Nov 14, 2019).
Sci. Adv. 2 (3), e1500779. Vugrin, E.D., Camphouse, R.C., 2011. Infrastructure resilience assessment through
Kim, J., Hastak, M., 2018. Online human behaviors on social media during disaster control design. Int. J. Crit. Infrastruct. 7 (3), 243–260.
responses. J. NPS Center Homeland Defense Security 14, 7–8. Vugrin, E.D., Warren, D.E., Ehlen, M.A., Camphouse, R.C., 2010. A framework for
Labaka, L., Hernantes, J., Sarriegi, J.M., 2016. A holistic framework for building critical assessing the resilience of infrastructure and economic systems. Sustainable and
infrastructure resilience. Technol. Forecast. Soc. Chang. 103, 21–33. resilient critical infrastructure systems. Springer, Berlin, Heidelberg.
Lin, J., Yu, W., Zhang, N., Yang, X., Zhang, H., Zhao, W., 2017. A survey on internet of Youn, B.D., Hu, C., Wang, P., 2011. Resilience-driven system design of complex
things: architecture, enabling technologies, security and privacy, and applications. engineered systems. J. Mech. Des. 133 (10).
IEEE Internet Things J. 4 (5), 1125–1142. Yuan, F., Liu, R., 2018a. Integration of social media and Unmanned Aerial Vehicles
Lu, L., Wang, X., Ouyang, Y., Roningen, J., Myers, N., Calfas, G., 2018. Vulnerability of (UAVs) for rapid damage assessment in Hurricane Matthew. In: Proceedings of
interdependent urban infrastructure networks: Equilibrium after failure propagation Construction Research Congress 2018, New Orleans, LA, USA.
and cascading impacts. Comput.-Aided Civ. Infrastruct. Eng. 33 (4), 300–315. Yuan, F., Liu, R., 2019. Identifying Damage-related social media data during Hurricane
Murdock, H.J., De Bruijn, K.M., Gersonius, B., 2018. Assessment of critical infrastructure Matthew: a machine learning approach. In: Proceedings of Computing in Civil
resilience to flooding using a response curve approach. Sustainability 10 (10), 3470. Engineering 2019, Atlanta, GA, USA.
North Carolina Health News, 2018. State Stands Up Specialized Medical Shelters for Yuan, F., Liu, R., 2020. Mining social media data for rapid damage assessment during
Sicker Evacuees. Available at <https://www.northcarolinahealthnews.org/2018/ Hurricane Matthew: a feasibility study. J. Comput. Civil Eng. 34(3), 05020001,
09/18/state-stands-up-specialized-medical-shelters-for-sicker-evacuees/> (last 1–14.
accessed on Nov 14, 2019). Yuan, F., Li, M., Liu, R., 2020. Understanding the evolutions of public responses using
Okabe, A., Okunuki, K.I., Shiode, S., 2006. SANET: a toolbox for spatial analysis on a social media: Hurricane Matthew case study. Int. J. Disaster Risk Reduct. 51,
network. Geogr. Anal. 38 (1), 57–66. 101798.
Omer, M., Mostashari, A., Lindemann, U., 2014. Resilience analysis of soft infrastructure Yuan, F., Liu, R., 2018b. Feasibility study of using crowdsourcing to identify critical
systems. Procedia Comput. Sci. 28, 565–574. affected areas for rapid damage assessment: Hurricane Matthew case study. Int. J.
Omer, M., Nilchiani, R., Mostashari, A., 2009. Measuring the resilience of the trans- Disaster Risk Reduct. 28, 758–767.
oceanic telecommunication cable system. IEEE Syst. J. 3 (3), 295–303. Yuan, F., Liu, R., 2018c. Crowdsourcing for forensic disaster investigations: Hurricane
Ouyang, M., Duenas-Osorio, L., 2014. Multi-dimensional hurricane resilience assessment Harvey case study. Nat. Hazards 93 (3), 1529–1546. https://doi.org/10.1007/
of electric power systems. Struct. Saf. 48, 15–24. s11069-018-3366-0. In this issue.
Ouyang, M., 2014. Review on modeling and simulation of interdependent critical Zhai, W., Peng, Z.R., Yuan, F., 2020. Examine the effects of neighborhood equity on
infrastructure systems. Reliab. Eng. Syst. Saf. 121, 43–60. disaster situational awareness: Harness machine learning and geotagged Twitter
Pandya, H.B., Champaneria, T.A., 2015. Notice of retraction Internet of Things: Survey data. Int. J. Disaster Risk Reduct. 101611.
and case studies. In: Proceedings of 2015 international conference on electrical, Zhao, K., Ge, L., 2013. A survey on the internet of things security. In: Proceedings of 2013
electronics, signals, communication and optimization (EESCO), Visakhapatnam, Ninth international conference on computational intelligence and security, Emei
India. Moutain, Sichuan Province, China.
Panteli, M., Mancarella, P., Trakas, D.N., Kyriakides, E., Hatziargyriou, N.D., 2017.
Metrics and quantification of operational and infrastructure resilience in power
systems. IEEE Trans. Power Syst. 32 (6), 4732–4742.

10

You might also like