Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Vol.:(0123456789)
Natural Hazards (2021) 107:2389–2417https://doi.org/10.1007/s11069-020-04429-3
1 3
ORIGINAL PAPER
Can we detect trends in natural disaster management with artificial intelligence? A review of modeling practices
Ling Tan1 · Ji Guo2,3 · Selvarajah Mohanarajah4 · Kun Zhou5
Received: 31 July 2020 / Accepted: 11 November 2020 / Published online: 22 November 2020 © Springer Nature B.V. 2020
AbstractThere has been an unsettling rise in the intensity and frequency of natural disasters due to climate change and anthropogenic activities. Artificial intelligence (AI) models have shown remarkable success and superiority to handle huge and nonlinear data owing to their higher accuracy and efficiency, making them perfect tools for disaster monitoring and manage-ment. Accordingly, natural disaster management (NDM) with the usage of AI models has received increasing attention in recent years, but there has been no systematic review so far. This paper presents a systematic review on how AI models are applied in different NDM stages based on 278 studies retrieved from Elsevier Science, Springer LINK and Web of Science. The review: (1) enables increased visibility into various disaster types in different NDM stages from the methodological and content perspective, (2) obtains many general results including the practicality and gaps of extant studies and (3) provides several recom-mendations to develop innovative AI models and improve the quality of modeling. Overall, a comprehensive assessment and evaluation for the reviewed studies are performed, which tracked all stages of NDM research with the applications of AI models.
Keywords Natural disaster management · Artificial intelligence · Stage analysis
* Ling Tan [email protected]
Ji Guo [email protected]
Selvarajah Mohanarajah [email protected]
Kun Zhou [email protected]
1 School of Applied Meteorology, Nanjing University of Information Science and Technology, Nanjing 210044, China
2 School of Economics and Management, Shanghai Maritime University, Shanghai 201306, China3 Collaborative Innovation Center On Climate and Meteorological Disasters, Nanjing University
of Information Science and Technology, Nanjing 210044, Jiangsu, China4 School of Computer Science, Mathematics and Computer Science, University of North Carolina
At Pembroke, Pembroke, NC 28372, USA5 School of Management Science and Engineering, Nanjing University of Information Science &
Technology, Nanjing 210044, China
2390 Natural Hazards (2021) 107:2389–2417
1 3
1 Introduction
According to the Emergency Events Database (EM-DAT), there were 348 natural disasters on average between 2008 and 2017. Associated with these disasters were 67,572 fatali-ties, 198.8 million injured or homeless people, and $166.7 billion economic losses per year (EM-DAT 2019). The intensity and frequency of natural disasters such as floods, storms, droughts, heat waves and wildfires have gradually increased, and the range affected has also continued to expand due to climate change and anthropogenic activities (Kalantari et al. 2019). Consequently, the influence caused by these disasters has created serious chal-lenges to the sustainable development of the economy as well as the safety of human life and property.
In recent years, with the increasing availability of large databases and more powerful computing power, artificial intelligence (AI) models continue to be used to process and interpret data at a more reliable rate (Dwivedi et al. 2019). Meanwhile, its application in natural disaster management (NDM) has become increasingly widespread. In order to reduce the impact of natural disasters effectively, AI models have been gradually applied in different fields of NDM due to its ease of use, high speed operation and acceptable accu-racy (Rajaee. 2019), such as retrieving and integrating disaster information (Budiharto. 2015), searching and rescuing victims (Liu et al. 2016) and conducting post-disaster loss analysis (Bai et al. 2018).
Although several empirical studies have been conducted on NDM with the usage of these AI models, the results have not been systematically compared and analyzed. To the best of our knowledge, this study is the first attempt toward the overall analysis concerning AI for NDM. Three contributions are made through this paper. First, an interdisciplinary review was conducted to present a clear picture for using AI models to carry out NDM research on various disaster types in different stages. Second, trends and the practicality of extant studies were examined through critically observing and investigating gaps. Third, the challenges were summarized, and potential future directions were anticipated, which can provide basic guidance for both disaster managers and academics.
The remainder of the paper is organized as follows: Sect. 2 analyzes the concept of AI and characteristics of NDM. In Sect. 3, selection process and statistical analysis of studies included in the paper are described. Section 4 reviews common AI models to deal with NDM in different stages for various disaster types. On this basis, challenges and possible directions are depicted in Sect. 5.
2 Related analysis of artificial intelligence (AI)and natural disaster management (NDM)
2.1 Artificial intelligence (AI)
AI is a comprehensive subject that integrates computer science, mathematics, philosophy, neuroscience, psychology, cybernetics, linguistics and other professional fields. Despite the fact that AI has received considerable attention from the government and practition-ers, there is no widely acknowledged definition. For example, as AI can perceive its envi-ronment in order to take actions that maximize the chance of success, Poole et al. (2010) defined AI as “computational agents that act intelligently”. Russell et al. (2016) defined
2391Natural Hazards (2021) 107:2389–2417
1 3
AI as a system of cognitive functions related to human attributes, such as learning things, making speeches, solving problems and then can sense and take actions in the environment to achieve its goals or maximize performance parameters entity. With the development of big data and Internet of things technologies, Kaplan et al. (2019) defined AI as the abil-ity of a system to interpret external data correctly, learn from such data and achieve spe-cific goals through flexible adaptation. Although definitions vary among scholars, the basic meaning is the same, that is, the machines perform specific roles and tasks to enhance and substitute manual tasks and activities.
2.2 Stages of natural disaster management (NDM)
Due to the increasing impact of natural disasters, many countries have to take a series of disaster management actions before, during and after disasters, especially those that have experienced severe disasters or are vulnerable to natural disasters. According to the occur-rence, development and evolution of disasters, NDM is usually divided into different stages including disaster preparedness (period before the disaster), disaster response (period dur-ing and after the disaster) and disaster recovery (a long period after the disaster) (Wex et al. 2014). In the stage of disaster preparedness, the main measures are to formulate emergency plans, carry out emergency drills and improve capabilities including early warning and monitoring, the resilience and recovery abilities (Ahmadi et al. 2015; Mashi et al. 2019). In the stage of disaster response, the main purpose is to maintain the stability of the socio-economic system in disaster situations, save people’s lives, and reduce the economic losses (Caunhye et al. 2012; Wu et al. 2018). The key measures taken at this stage include the implementation of early stage plans, the analysis of disaster environments, the deployment of emergency rescue teams and material facilities, emergency material transportation and personnel evacuation (Anaya-Arenas et al. 2014; Poblet et al. 2018). In the stage of disaster recovery, the main purpose is to shorten the time of recovery, and effective measures can be taken such as the formulation of recovery and reconstruction plans, the allocation of
Emergency plans
Emergency drill
Early warning and monitoring
Disaster-resistance and recovery capabilities
Carry out plans
Disaster preparedness Disaster response
Environmental analysis
Emergency rescue teams and material facilities
Emergency supplies
Evacuation of disaster victims
Disaster recovery
Deploy internal and external resources
Data analysis
Infrastructure repair
Recovery and reconstruction plans
Disaster assistance
Natural Disaster management
Fig. 1 Stage analysis of NDM
2392 Natural Hazards (2021) 107:2389–2417
1 3
internal and external resources, post-disaster data analysis, infrastructure restoration, and disaster assistance (Sahebjamnia et al. 2015). The process of NDM is shown in Fig. 1.
3 Analysis of articles included in the study
3.1 Identification and selection of articles
The articles included in this paper were identified and selected through a four step pro-cess. First, search terms were used to identify all related articles in English. The search terms defined in the identification process were ′′artificial intelligence and disaster′′ or ′′AI and disaster′′. Articles are available online as of March 20, 2020. Databases for arti-cles are Elsevier Science, Springer LINK and Web of Science. In the process of identi-fication, a total of 1267 articles were obtained. Second, 1,267 articles went through the process of removing duplicates. Articles were obtained from different databases, and thus existed duplicates inevitably. From this process, 112 articles were excluded, 1155 articles remained. Third, the title, keywords and abstract of remaining articles have been screened to determine whether it is related to the research question, and review articles have also been removed. Finally, the full texts of the remaining articles were reviewed. From the identification and selection process, 278 articles were included. The specific process is shown in Fig. 2.
3.2 Statistical analysis of included articles
The developments, knowledge accumulation and maturity of a research field can be measured by its literature publications, which is of great significance to evaluate the
N=1,272
Searching databases with search terms Removing duplicates
N=1,155
Screening title,keys and abstract
N=436
Checking full text
N=278
Fig. 2 The process of article identification and selection
Fig. 3 Publications of AI models used in the field of NDM
2393Natural Hazards (2021) 107:2389–2417
1 3
development trend and dynamic evolution. First, this paper described the literature output and the changing trend of NDM with AI models (see Fig. 3). In general, publications have shown an upward trend, which means the topic has clearly been noted by scholars. Espe-cially since 2013, publications have increased rapidly.
Second, this paper generated commonly used AI models (see Table1) and showed their yearly application (see Fig. 4), which was designed to communicate to the readers that AI models increased in popularity among NDM. In general, the ten most commonly used applications are artificial neural network (ANN) (Chau et al. 2005), support vector
Table 1 Summary of statistics in methodology, disaster type and research stage
Models (Top 10) Proportion (%) Disaster type (Top 10) Proportion (%)
Artificial neural network (ANN) 19.46 Natural disaster 29.08Support vector machine (SVM) 11.92 Flood 26.95Fuzzy logic (FL) 7.79 Landslide 13.48Regress algorithm (RA) 6.81 Earthquake 9.57Genetic algorithms (GA) 6.08 Rainfall 4.26Random forest (RF) 5.6 Drought 4.26Robotics 5.11 Fire 3.9Bayesian (Bayes) 3.65 Storm 3.9Extreme learning machine (ELM) 3.16 Collapse 1.42Decision tree (DT) 2.68 Debris flow 1.42Wavelet analysis (WA) 2.43 Research stage ProportionExpert system (ES) 2.19 Disaster preparedness 63.73Particle swarm optimization (PSO) 1.95 Disaster response 28.17Search algorithm (SA) 1.95 Disaster recovery 8.1
Fig. 4 Number of major AI models applied per year
2394 Natural Hazards (2021) 107:2389–2417
1 3
machine (SVM) (Jiao et al. 2016), fuzzy logic (FL) (Rodríguez et al. 2011), regress algo-rithm (RA) (Ragettli et al. 2017), genetic algorithm (GA) (Zhou et al. 2019), random forest (RF) (Wang et al. 2015), robotics (Geng et al. 2019), bays (Cheng et al. 2016), extreme learning machine (ELM) (Li et al. 2017) and decision tree (DT) (Choi et al. 2018). Among the retrieved articles, the proportion of the ten most commonly used applications reached 72.26%. Other AI models, such as wavelet analysis (WA) (Seo et al. 2015), expert system (ES) (Miller-Hooks et al. 2007), particle swarm optimization (PSO) (Tayfur et al. 2018) and search algorithm (SA) (Perumal et al. 2017), etc., are also used in NDM research. From the perspective of the application of AI models annually, Bays and ES are the earlier applied AI models, ANN, SVM, FL, RA and GA are more frequently used AI models in the field of NDM.
Third, this paper summarized the proportion of disaster type and research stage. From the statistical results, the ten most studied disaster types are natural disaster, flood, land-slide, earthquake, rainfall, drought, fire, storm, collapse and debris flow, accounting for 98.23% of retrieved articles. Other disasters are ice, rock burst and tsunami, accounting for 1.77%. From the stage analysis of NDM, the application proportion of disaster prepar-edness is the highest, reaching 65.92%, far higher than 26.59% of disaster response and 7.49% of disaster recovery.
Fourth, this paper provided journals with AI models applied in the field of NDM with ≥ 3 articles (see Fig. 5). From 1986 to March 2020, a total of 278 articles were identi-fied, involving 133 journals. Among them, a total of 20 journals with ≥ 3 articles published, accounting for 49.28% of total publications. Journal of Hydrology with a major position in output (19 articles), accounting for about 7.28% of total publications. Water and Remote Sensing followed closely behind. In addition, scholars have also published articles in jour-nals such as Safety Science, Information Sciences, Ecological Indicators and Journal of
Fig.5 Journals published on NDM with AI models
2395Natural Hazards (2021) 107:2389–2417
1 3
Cleaner Production, etc. From the statistical results of the journals, it can be seen that there are no relatively concentrated journals so far, and the publications of research results are relatively scattered.
Finally, this paper provides articles with the top 10 citations per year (see Table2). Arti-cles ranked in the top 10 in annual citations are all after 2010, including 1 in 2011, 3 in 2015, 3 in 2017, 2 in 2018 and 1 in 2019 as of March 20, 2020. From the statistical results of the annual high citations, the more influential articles in this field are ongoing. Through the analysis of these important articles, the foundational knowledge of applying AI models for NDM can be revealed.
4 Applications of AI models in NDM
What are the preferred AI models used in different NDM stages? How about the research progress of different disaster types? In the following sections, this paper will summarize the application of AI models in different stages of NDM with numerous disaster types, and investigate practicality, gaps, trends of the extant studies.
4.1 Disaster preparedness
Disaster preparation is the stage most studied by scholars. The applied methods include ES, ANN, FL, SVM, adaptive neuro-fuzzy interface system (ANFIS), etc. The research focuses principally include disaster risk assessment, disaster early warning and monitoring, concerning flood, earthquake, landslide, debris flow, collapse and other different disaster types.
4.1.1 Natural disaster
From the perspective of natural disasters, the available studies can be divided into five aspects: risk assessment, website application, emergency resource demand prediction and emergency warning system construction. The research methods mainly concentrate on sin-gle methods. Table 3 provides research progress of scholars on natural disasters in the stage of disaster preparedness.
Risk assessment. For example, Pozdnoukhov et al. (2009) used data-driven feature selection, SVM and ANN produced spatial maps of hazard-related parameters from point observations and available auxiliary information. As the cloud and snow recognition tech-nologies for multispectral satellite images have played an important role in resource inves-tigation, natural disaster monitoring and environmental pollution, Xia et al. (2019) pro-posed an improved in-depth residual network with multi-dimensional input for cloud and snow recognition.
Website application. NDM websites play an important role in aiding people through various disasters. However, it is seldom applied to practical applications due to the com-plexity of those websites. AI models provide new ideas for dealing with the huge flow of information. For example, Chou et al. (2014) developed an ontology-based evaluation tool to evaluate the utility of NDM websites. Zheng et al. (2019) proposed a method to achieve both high-precision and high-recall in collecting event-related tweets based on active learn-ing and multiple-instance learning.
2396 Natural Hazards (2021) 107:2389–2417
1 3
Tabl
e 2
Top
10
pape
rs c
ited
per y
ear o
n N
DM
with
AI m
odel
s
Aut
hor
Year
Title
Ann
ual C
itatio
ns
Kho
srav
i et a
l20
18A
com
para
tive
asse
ssm
ent o
f dec
isio
n tre
es a
lgor
ithm
s for
flas
h flo
od su
scep
tibili
ty m
odel
ing
at H
araz
wat
ersh
ed,
north
ern
Iran
43
Stum
pf e
t al
2011
Obj
ect-o
rient
ed m
appi
ng o
f lan
dslid
es u
sing
Ran
dom
For
ests
42.3
Goe
tz e
t al
2015
Eval
uatin
g m
achi
ne le
arni
ng a
nd st
atist
ical
pre
dict
ion
tech
niqu
es fo
r lan
dslid
e su
scep
tibili
ty m
odel
ing
31.6
7H
ong
et a
l20
18Fl
ood
susc
eptib
ility
ass
essm
ent i
n H
engf
eng
area
cou
plin
g ad
aptiv
e ne
uro-
fuzz
y in
fere
nce
syste
m w
ith g
enet
ic
algo
rithm
and
diff
eren
tial e
volu
tion
30.3
3
Wan
g et
al
2015
Floo
d ha
zard
risk
ass
essm
ent m
odel
bas
ed o
n ra
ndom
fore
st28
.17
Che
n et
al
2017
Land
slid
e sp
atia
l mod
elin
g: In
trodu
cing
new
ens
embl
es o
f AN
N, M
axEn
t, an
d SV
M m
achi
ne le
arni
ng te
chni
ques
27.7
5Sh
irzad
i et a
l20
17Sh
allo
w la
ndsl
ide
susc
eptib
ility
ass
essm
ent u
sing
a n
ovel
hyb
rid in
telli
genc
e ap
proa
ch25
.25
Cha
pi e
t al
2017
A n
ovel
hyb
rid a
rtific
ial i
ntel
ligen
ce a
ppro
ach
for fl
ood
susc
eptib
ility
ass
essm
ent
25Ja
afar
i et a
l20
19H
ybrid
arti
ficia
l int
ellig
ence
mod
els b
ased
on
a ne
uro-
fuzz
y sy
stem
and
met
aheu
ristic
opt
imiz
atio
n al
gorit
hms
for s
patia
l pre
dict
ion
of w
ildfir
e pr
obab
ility
29
Seo
et a
l20
15D
aily
wat
er le
vel f
orec
astin
g us
ing
wav
elet
dec
ompo
sitio
n an
d ar
tifici
al in
telli
genc
e te
chni
ques
23
2397Natural Hazards (2021) 107:2389–2417
1 3
Emergency resource demand prediction. For example, Dawson et al. (2000) presented a method for emergency resource demand prediction using case-based reasoning corre-sponding to the characteristics of emergency resource demand prediction.
Emergency warning system construction. For example, Kang et al. (2016) introduced the idea of designing an emergency warning system based on deep learning. The proposed system is identified suitable for existing infrastructure, such as closed-circuit televisions and other monitoring equipment. Experimental results show that, in most cases, the system will instantly detect emergencies, such as traffic accidents and natural disasters.
4.1.2 Flood
From the perspective of flood management, the available studies can be divided into six aspects: flood risk assessment, water level prediction, streamflow forecasting, rainfall-runoff process modeling, flood frequency analysis and flood forecasting. Table 4 presents research progress on floods in the stage of disaster preparedness, including research objec-tives, research methods and applications.
Flood risk assessment. Studies can be divided into two types due to research methods, e.g., single method and hybrid method. Hybrid methods merge different methods in order to improve the overall results of the assessment regarding accuracy, time and range (Hal-legatte 2008). The first type is a single method. For instance, Wang et al. (2015) adopted RF and SVM to evaluate the risk of regional floods. Ragettli et al. (2017) used DT, clas-sification and regression trees to model the flash floods in ungauged mountain catchments. Darabi et al. (2019) applied based algorithm rule-set production and quick unbiased effi-cient statistical tree to generate the flood risk graph. The second type is hybrid methods. For instance, Chau et al. (2005) adopted ANN and ANFIS based on GA for flood disaster prediction. Zhou et al. (2019) proposed a fuzzy inference system based on cyclic adap-tive network with GA and least squares estimator embedded to make multi-step flood prediction.
Water level prediction. According to the research methods, studies can be divided into two types. The first type is a single method. For instance, Jiao et al. (2016) used emergency empirical mode decomposition, radial basis function natural networks and SVM to forecast hydrological data. Tayfur et al. (2018) applied ANN, GA, ant colony optimization, and PSO methods to predict flood hydrology. The second type is hybrid methods. For instance, Seo et al. (2015) adopted the wavelet-based artificial network and wavelet-based adaptive neuro-fuzzy information system to predict the daily water level of the reservoir.
Table 3 Research progress of natural disaster in the stage of disaster preparedness
Disaster type Objective Applications or examples
Natural disaster Risk assessment SVM + ANN (Pozdnoukhov et al. 2009); Multidimensional deep residual network t (Xia et al. 2019)
Website application Ontology (Chou et al. 2011); Ontol-ogy (Chou et al. 2014); Active learning (Zheng et al. 2019)
Emergency resource demand prediction Case-based reasoning (Liu et al. 2012)Emergency warning system Deep learnin (Kang et al. 2016)
2398 Natural Hazards (2021) 107:2389–2417
1 3
Tabl
e 4
Res
earc
h pr
ogre
ss o
f floo
d in
the
stag
e of
dis
aste
r pre
pare
dnes
s
Dis
aste
r typ
eO
bjec
tive
Met
hodo
logy
App
licat
ions
or e
xam
ples
Floo
dR
isk
asse
ssm
ent
Sing
le m
etho
dsR
F +
SVM
Wan
g et
al.
(201
5); A
NN
Fle
min
g et
al.
(201
5); G
enet
ic p
rogr
amm
ing
Hu
(201
6); D
T +
Cla
s-si
ficat
ion
and
regr
essi
on tr
ees R
aget
tli e
t al.
(201
7); P
oiss
on re
gres
sion
+ R
F Sa
dler
et a
l. (2
018)
; DT
Kho
srav
i et a
l. (2
018)
; Gen
etic
alg
orith
m ru
le-s
et p
rodu
ctio
n (G
AR
P) +
Qui
ck u
nbia
sed
effici
ent s
tatis
ti-ca
l tre
e D
arab
i et a
l. (2
019)
; Nai
ve b
ayes
Kho
srav
i et a
l. (2
019)
; Sup
port
vect
or re
gres
sion
Wu
et a
l. (2
019)
; Nai
ve b
ayes
+ A
ltern
atin
g D
T +
RF
Che
n t a
l. (2
020)
Hyb
rid m
etho
ds(G
A-A
NN
) + A
NFI
S C
hau
et a
l. (2
005)
; GA
-AN
N W
u et
al.
(200
6); B
aggi
ng- L
ogist
ic m
odel
tree
Cha
pi
et a
l. (2
017)
; AN
FIS
+ G
A +
Diff
eren
tial e
volu
tion
Hon
g et
al.
(201
8); (
Redu
ced
erro
r pru
ning
tree
s)-
Bag
ging
-Ran
dom
subs
pace
Che
n et
al.
(201
9); P
SO-E
LM B
ui e
t al.
(201
9); A
NFI
S W
ang
et a
l. (2
019)
; Re
curr
ent-A
NFI
S Zh
ou e
t al.
(201
9)W
ater
leve
l pre
dict
ion
Sing
le m
etho
dsEn
sem
ble
empi
rical
mod
e de
com
posi
tion +
Rad
ial b
asis
func
tion
neur
al n
etw
orks
+ SV
M Ji
ao e
t al.
(201
6); G
A +
Diff
eren
tial e
volu
tion +
PSO
+ H
arm
ony
sear
ch P
erum
al e
t al.
(201
7); L
ong
shor
t ter
m
mem
ory
netw
ork
Lian
g et
al.(
2018
); A
NN
+ G
A +
Ant
col
ony
optim
izat
ion +
PSO
Tay
fur e
t al.
(201
8)H
ybrid
met
hods
AN
FIS
Baz
arts
eren
et a
l. (2
003)
; (W
avel
et-A
NN
) + (W
avel
et-A
NFI
S) S
eo e
t al.
(201
5)St
ream
flow
fore
casti
ngSi
ngle
met
hods
AN
N +
Bay
s Chi
ang
et a
l. (2
018)
; Wav
elet
tran
sfor
mat
ion +
Mul
tigen
e ge
netic
pro
gram
min
g + A
NN
Had
i et
al.
(201
8); G
ene-
expr
essi
on p
rogr
amm
ing
Bag
atur
et a
l. (2
018)
Hyb
rid m
etho
dsA
NN
+ A
NFI
S +
Wav
elet
neu
ral n
etw
orks
+ (A
NFI
S-W
avel
et) B
adrz
adeh
et a
l. (2
015)
Rai
nfal
l-run
off p
roce
ss m
odel
ing
Sing
le m
etho
dsD
T +
AN
N D
awso
n et
al.
(200
0); A
dapt
ive
boos
ting
algo
rithm
+ P
SO L
iu e
t al.
(201
4)H
ybrid
met
hods
AN
N +
AN
FIS
Pram
anik
et a
l. (2
009)
Floo
d fr
eque
ncy
anal
ysis
Sing
le m
etho
dsA
NN
+ S
uppo
rt ve
ctor
regr
essi
on G
izaw
et a
l. (2
016)
2399Natural Hazards (2021) 107:2389–2417
1 3
Streamflow forecasting. Studies in this theme also take the form of single methods and hybrid methods. The first type is a single method. For instance, Hadi et al. (2018) used wavelet transformation, multigene genetic programming, and ANN to conduct monthly streamflow forecasting. The second type is hybrid methods. For instance, Badrzadeh et al. (2015) adopted ANN, ANFIS, wavelet neural networks, and hybrid ANFIS with multi res-olution analysis using wavelets for river flow forecasting at Casino station on Richmond River, which is highly prone to flooding.
Rainfall-runoff process modeling. Studies can be divided into single methods and hybrid methods. The first type is a single method. For instance, Dawson et al. (2000) adopted DT and ANN to simulate river flows in flood-prone basins using actual hydrological data. Liu et al. (2014) applied an adaptive boosting algorithm and PSO to simulate single process-based rainfall-runoff process. The second type is hybrid methods. For instance, Pramanik et al. (2009) adopted ANN and ANFIS to estimate the flow in the downstream section of the river using flow data from the upstream position.
Flood frequency analysis. Currently, there are a few studies on flood frequency analysis. For example, Gizaw et al. (2016) applied ANN and SVR to estimate regional flood quan-tiles for two study areas, one with 26 catchments located in southeastern British, Colum-bia, and another with 23 catchments located in southern Ontario, Canada.
4.1.3 Landslide
Studies on landslide disasters focus on reliability assessment. Research results usually com-bine single methods with hybrid methods. The first type is a single method. For instance, Kojima et al. (2006) presented an inverse analysis of unobserved trigger factors for slope failures and landslides, based on structural equation modeling. Steger et al. (2016) applied two statistical classifiers (logistic regression and generalized additive model) and two machine learning techniques (RF and SVM) Studies on landslide disaster focus on the reli-ability assessment. Research results usually combine single methods with hybrid methods. The second type is hybrid methods. For instance, Sdao et al. (2013) applied neuro-fuzzy network to conduct a comprehensive assessment of landslide sensitivity. Chen et al. (2017) used three popular machine learning models, i.e. maximum entropy, SVM, ANN and their sets to carry out spatial simulation of landslide sensitivity. Table 5 lists research progress of scholars on landslides in the stage of disaster preparation.
4.1.4 Earthquake
Studies of earthquake management started early, and the main content focused on risk assessment, presenting the form of single methods and hybrid methods. The first type is a single method. For instance, Miyasato et al. (1986) used ES to predict the possible risk level of buildings due to earthquake hazards. Hanna et al. (2007) put forward an alterna-tive general expression network model to predict the liquefaction condition in soil depos-its, which contributes to the research in making accurate predictions of earthquake places. Huang et al. (2018) a developed a method to identify the arrival time delay by using the revolutionary natural network and deep learning technologies, and then determined the source location of the micro-seismic events in the underground mine. The second type is hybrid methods. For instance, Chen et al. (2012) developed an evolutionary SVM infer-ence system that integrated two AI models, e.g., SVM and fast messy GA to evaluate the seismic resistance of school buildings. Thomas et al. (2016) proposed randomized ANFIS
2400 Natural Hazards (2021) 107:2389–2417
1 3
Table 5 Research progress of landslide in the stage of disaster preparedness
Disaster type Objective Methodology Applications or examples
Landslide Risk assessment Single methods Structural equation modeling Kojima et al. (2006); ANN Cani-ani et al. (2008); Fuzzy inference system + ANN Vahidnia et al. (2010); ELM Cao et al. (2012); Logistic regression + General-ized additive model + RF + SVM Steger et al. (2016); Alternating DT Shirzadi et al. (2019)
Hybrid methods Neuro-fuzzy Network Sdao et al. (2013); SVM + RF + Boot-strap aggregated classification trees (bundling) with penal-ized discriminant analysis Goetz et al. (2015); Maximum entropy + SVM + ANN + ( ANN-SVM, ANN-Maximum entropy, ANN-Maximum entropy-SVM, SVM-Maximum entropy) Chen et al. (2017)
Table 6 Research progress of earthquake in the stage of disaster preparedness
Disaster type Objective Methodology Applications or examples
Earthquake Risk assessment Single methods ES Miyasato et al. (1986); Bayes Peizhuang et al. (1986); Regres-sion neural network Hanna et al. (2007); SVM + Fast messy genetic algorithms Cheng et al. (2014); ES Ikram et al. (2014); Pattern recognition neural network + Recurrent neural network + RF + Linear program-ming boost ensemble Asim et al. (2017); Convolutional neural network Huang et al. (2018)
Hybrid methods Evolutionary SVM inference system Chen et al. (2012); Smart firefly algorithm + Least squares SVM Chou et al. (2016); Ran-domized ANFIS Thomas et al. (2016)
2401Natural Hazards (2021) 107:2389–2417
1 3
Tabl
e 7
Res
earc
h pr
ogre
ss o
f oth
er d
isas
ters
in th
e st
age
of d
isas
ter p
repa
redn
ess
Dis
aste
r typ
eO
bjec
tive
Met
hodo
logy
App
licat
ions
or e
xam
ples
Rai
nfal
lR
isk
asse
ssm
ent
Sing
le m
etho
dsSV
M N
ayak
et a
l. (2
013)
; SV
M N
ayak
et a
l. (2
017)
; DT
+ ba
ggin
g + R
F +
Boo
sting
Cho
i et a
l. (2
018)
; A
NN
+ SV
M +
RF
Sula
iman
et a
l. (2
018)
Fire
Ris
k as
sess
men
tSi
ngle
met
hods
Wire
less
sens
or n
etw
orks
+ A
rtific
ial b
ee c
olon
y al
gorit
hm +
Ant
col
ony
optim
izat
ion
Kum
ar e
t al.
(201
6);
Logi
stic
regr
essi
on +
RF
+ B
oosti
ng c
lass
ifica
tion
trees
Mits
opou
los e
t al.
(201
7); A
NN
+ SV
M S
ayad
et
al.
(201
9)H
ybrid
met
hods
AN
FIS
+ G
A +
PSO
+ S
huffl
ed fr
og le
apin
g al
gorit
hm +
Impe
rialis
t com
petit
ive
algo
rithm
Jaaf
ari e
t al.
(201
9)D
roug
htR
isk
asse
ssm
ent
Hyb
rid m
etho
dsM
ultip
le li
near
regr
essi
on +
Rad
ial b
asis
func
tion
neur
al n
etw
orks
+ L
east
squa
res s
uppo
rt ve
ctor
regr
essi
on
Sadr
i et a
l. (2
012)
; Mul
tilay
er p
erce
ptro
n A
NN
+ A
NFI
S +
SVM
+ A
utor
egre
ssiv
e in
tegr
ated
mov
ing
aver
-ag
e Ja
lalk
amal
i et a
l. (2
015)
; AN
N +
AN
FIS
+ SV
M M
okht
arza
d et
al.
(201
7); (
Wav
elet
- Aut
o re
gres
sive
in
tegr
ated
mov
ing
aver
age
mod
el- A
NN
) + (W
avel
et-A
NFI
S) S
oh e
t al.
(201
8); A
NN
Ach
our e
t al.
(202
0)C
olla
pse
Ris
k as
sess
men
tSi
ngle
met
hods
Bay
s + K
NN
Che
ng e
t al.
(201
6); A
NN
+ G
A +
SVM
Sum
an e
t al.
(201
6); G
ravi
tatio
nal s
earc
h al
go-
rithm
+ R
F +
SVM
+ N
aive
Bay
s Lin
et a
l. (2
018)
Hyb
rid m
etho
dsEv
olut
iona
ry ri
sk p
refe
renc
e fu
zzy
SVM
infe
renc
e m
odel
Che
ng e
t al.
(201
2); F
irefly
Alg
orith
m +
the
Leas
t Sq
uare
s SV
M H
oang
et a
l. (2
016)
Deb
ris fl
owR
isk
asse
ssm
ent
Sing
le m
etho
dsA
dapt
ive
boos
ting
mac
hine
lear
ning
alg
orith
m P
ai e
t al.
(201
4)W
arni
ng sy
stem
con
struc
tion
Hyb
rid m
etho
dsSh
ared
nea
r nei
ghbo
rs n
eura
l net
wor
k C
hang
et a
l. (2
007)
Ice
Ris
k as
sess
men
tSi
ngle
met
hods
Mul
tivar
iabl
e tim
e se
ries m
odel
Li e
t al.
(201
4)H
ybrid
met
hods
Gra
y co
rrel
atio
n an
alys
is +
ELM
bas
ed o
n th
e ad
aptiv
e w
hale
opt
imiz
atio
n al
gorit
hm im
prov
ed b
y ch
aotic
si
ne c
osin
e op
erat
or W
ang
et a
l. (2
019)
Stor
mR
isk
asse
ssm
ent
Sing
le m
etho
dsA
NN
+ R
F H
art e
t al.
(201
9); D
eep
neur
al n
etw
ork
Kim
et a
l. (2
019)
Rock
bur
stR
isk
asse
ssm
ent
Sing
le m
etho
dsG
A +
ELM
Li e
t al.
(201
7)
2402 Natural Hazards (2021) 107:2389–2417
1 3
to predict ground motion parameters related to seismic signals. Table 6 provides research progress for scholars on earthquakes in the stage of disaster preparation.
4.1.5 Other disasters
Other disasters include rainfall, fire, drought, collapse, debris flow, storm, ice and rock burst. The research content primarily centers on risk assessment. Table7 presents research progress for scholars on other disasters in the stage of disaster preparedness.
Rainfall. The main research methods are single methods. For example, Nayak et al. (2013) developed an algorithm based on SVM, to predict extreme rainfall with a lead time of 6–48 h in Mumbai, using mesoscale (20–200 km) and synoptic scale (200–2000 km) weather patterns. Choi et al. (2018) developed prediction models of heavy rain damage using machine learning such as DT, bagging, RF, and boosting based on big data for the Seoul Capital Area in the Republic of Korea.
Fire. Research results usually combine single methods with hybrid methods. The first type is a single method. For instance, Kumar et al. (2016) combine artificial bee colony algorithm and ant colony optimization algorithm to solve a non-deterministic polynomial hard and finite problem of wireless sensor networks in designing the framework for forest fire detection and monitoring. Mitsopoulos et al. (2017) attempted to determine the role of fire suppression tactics and behavior, weather, topography and landscape features on differ-ent fire sizes based on logistic regression analysis, RF and boosting classification trees. The second type is hybrid methods. For instance, Jaafari et al. (2019) provided a new compara-tive analysis of four hybrid artificial intelligence models for the spatially explicit prediction of wildfire probabilities.
Drought. The main research methods are hybrid methods. For instance, Jalalkamali et al. (2015) proposed to predict drought using and comparing the multilayer perceptron ANN, ANFIS, SVM model, and the autoregressive integrated moving average multivari-ate time series. Soh et al. (2018) adopted wavelet- auto regressive integrated moving aver-age model- ANN model and the latest wavelet-ANFIS model to forecast the standardized precipitation evapotranspiration index at the Langat River Basin for different time ranges (1 month, 3 months and 6 months).
Collapse. Studies can be divided into single methods and hybrid methods. The first type is a single method. For instance, Cheng et al. (2016) proposed a novel approach for slope collapse assessment integrated the Bayesian framework and the KNN density estimation technique. The second type is hybrid methods. For instance, Hoang et al. (2016) proposed a hybrid AI for slope stability assessment based on firefly algorithm and the least squares support vector classification.
Debris flow. Studies can be divided into single methods and hybrid methods. The research contents include risk assessment and warning system construction. For example, Chang et al. (2007) proposed the shared near neighbors neural network to build debris flow early warning system. Pai et al. (2014) used the unique strengths of the adaboost and rough set theory in classification and rule generation to analyze debris flow.
Ice. Studies can be divided into single methods and hybrid methods. For example, Li et al. (2014) presented a model based on a multivariable time series to predict the icing load of a transmission line. Wang et al. (2019) adopted gray correlation analysis and a new ELM based on the adaptive whale optimization algorithm improved by chaotic sine cosine operator to forecast the ice coating damage to power line in southern China.
2403Natural Hazards (2021) 107:2389–2417
1 3
Storm and Rock burst. The main research methods are single methods. For example, Hart et al. (2019) tested the ability of machine learning techniques, e.g., ANN and RF, to predict the individual trees within a forest most at risk of damage in storms. Li et al. (2017) proposed the rock burst prediction model according to GA and ELM.
4.1.6 Remarks
Studies in the stage of disaster preparedness are relatively rich. Research contents involve risk assessment, disaster prediction, warning system construction, etc. Besides, scholars have developed various AI models for different research topics. However, the analysis in Sect. 4.1 still reveals some gaps in this field.
The neglected role of climate change adaptation. Global warming, especially the changes of extreme climate values accompanied by global warming, will further increase the frequency and intensity of natural disasters (Kalantari et al. 2019). Therefore, incor-porating climate change factors into the studies in the stage of disaster preparedness can not only analyze the formation mechanism of disasters more clearly, but also simulate the trajectory of disasters more accurately, which is very important for risk assessment and disaster prediction.
The neglected role of resilience against disasters. At present, studies of risk assessment are usually based on the information of assessment region, such as topography, historical data and environmental information. However, resilience of the affected area is rarely con-sidered. In general, resilience refers to the ability or capacity to absorb or cushion against disaster damage or loss (Rose and Liao 2005). It can help the affected area reduce the prob-ability of structural or system failure, and quickly return to normal (Xie et al., 2018). Resil-ience emanates both from internal motivation and the stimulus of private or public policy decisions (Rose 2004). For example, the optimization of investment decisions before the disaster has an important impact on the improvement of transportation networks, informa-tion networks and other infrastructure. Taking these factors into account enables a more accurate assessment of disaster risk.
The neglected role of disaster preparedness, planning and management. Disasters have the characteristics of sudden. If proper measures are taken in advance, the effects can be greatly reduced. How to prepare response plans and rescue materials in advance to reduce the negative impact of disasters according to the different disaster types? Disaster prepar-edness, planning, and management and their impact on disaster response have not yet been mentioned in the existing literature.
4.2 Disaster response
In the stage of disaster response, the main research contents include disaster environment analysis, emergency rescue and personnel evacuation. Applied methods include robotics, ANN, FL, SVM, etc. Flood, earthquake, landslide, rainfall and other disaster types are involved in this stage.
4.2.1 Natural disaster
Studies of natural disasters in the stage of disaster response can be divided into five aspects: search and rescue, emergency response decision, emergency evaluation, envi-ronmental analysis, website application and emergency materials transportation. Table 8
2404 Natural Hazards (2021) 107:2389–2417
1 3
Tabl
e 8
Res
earc
h pr
ogre
ss o
f nat
ural
dis
aste
r in
the
stag
e of
dis
aste
r res
pons
e
Dis
aste
r typ
eO
bjec
tive
Met
hodo
logy
App
licat
ions
or e
xam
ples
Nat
ural
dis
aste
rSe
arch
and
resc
ueSi
ngle
met
hods
Robo
tics B
litch
et a
l. (1
996)
; Rob
otic
s Cha
tterje
e et
al.
(200
5); R
obot
ics S
hiro
ma
et a
l. (2
005)
; Ro
botic
s Virk
et a
l. (2
008)
; Rob
otic
s + H
RL
Liu
et a
l. (2
016)
; Rob
otic
s Pen
g et
al.
(201
8); R
obot
-ic
s + A
ttent
ion-
base
d C
omm
unic
atio
n ne
ural
net
wor
k G
eng
et a
l. (2
019)
Emer
genc
y re
spon
se d
ecis
ion
Sing
le m
etho
dsIn
tegr
atin
g hi
erar
chic
al ta
sk n
etw
ork
Tang
et a
l. (2
015)
; Rob
otic
s Ram
chur
n et
al.
(201
6); M
ulti-
agen
t mar
kov
deci
sion
pro
cess
Ram
chur
n et
al.
(201
6)H
ybrid
met
hods
Ont
olog
y-su
ppor
ted
case
-bas
ed re
ason
ing
Am
aile
f et a
l. (2
013)
; Fuz
zy e
vide
ntia
l dec
isio
n-m
akin
g tri
al a
nd e
valu
atio
n la
bora
tory
met
hod
Han
et a
l. (2
018)
Emer
genc
y ev
acua
tion
Sing
le m
etho
dsA
gent
-bas
ed m
odel
Man
ley
et a
l. (2
016)
; GA
+ N
N +
FL
Shar
ma
et a
l. (2
018)
; Age
nt-b
ased
dis
cret
e-ev
ent s
imul
atio
n m
odel
ing
fram
ewor
k N
a et
al.
(201
9)En
viro
nmen
tal a
naly
sis
Sing
le m
etho
dsRo
botic
s + B
ackp
ropa
gatio
n ne
ural
net
wor
k B
udih
arto
(201
5); R
obot
ics +
Mac
hine
lear
ning
Ofli
et
al.
(201
6); R
obot
ics +
Res
idua
l neu
ral n
etw
ork
Zhao
et a
l. (2
018)
Web
site
app
licat
ion
Sing
le m
etho
dsA
utom
ated
kno
wle
dge
extra
ctio
n ag
ent G
oh e
t al.
(200
5); o
ntol
ogy
Li e
t al.
(201
4); M
achi
ne le
arn-
ing
Avv
enut
i et a
l. (2
016)
; Nat
ural
lang
uage
pro
cess
ing
Ala
m e
t al.
(201
9)Em
erge
ncy
mat
eria
ls tr
ansp
orta
tion
Sing
le m
etho
dsA
nt c
olon
y op
timiz
atio
n Li
u et
al.
(201
7)
2405Natural Hazards (2021) 107:2389–2417
1 3
provides research progress of natural disaster in disaster response stage, including research objectives, research methods and application.
Search and rescue. Studies of disaster search and rescue mainly adopt single methods such as robotics technology. For example, Blitch et al. (1996) discussed key issues in the application of robotic systems to urban search and rescue activities and discussed ongoing development of a knowledge-based system for efficient management of automated search assets. Chatterjee et al. (2005) described rescue robot features in the context of search and rescue effort coordination. Geng et al. (2019) presented a novel approach, the attention-based communication neural network, to simulate the cooperation strategies automatically in multi-robot exploration problems.
Emergency response decision. Studies can be divided into single methods and hybrid methods. The first type is a single method. For instance, Ramchurn et al. (2016) devel-oped an algorithm, based on a multi-agent Marko decision process representation of the task allocation problem and then integrated the algorithm into a planning agent to solve the task allocation problem posed by emergency response planning. The second type is hybrid methods. For instance, Han et al. (2018) proposed a novel enhanced fuzzy eviden-tial decision-making trial and evaluation laboratory method to improve the performance of emergency systems.
Emergency evacuation. The main research methods are single methods. For example, Manley et al. (2016) presented an agent-based model called exitus which is capable of determining the extent to which collective behavior, and overall evacuation time of pas-senger groups is affected by changes in the built environment for large, complex structures. Sharma et al. (2018) combined GA with neural networks and FL to explore how intelligent agents can learn and adapt their behavior during an evacuation.
Environmental analysis. The main research methods are single methods. For example, Ofli et al. (2016) proposed a hybrid crowdsourcing and real-time machine learning solu-tion to rapidly process large volumes of aerial data for disaster response in a time-sensitive manner. Zhao et al. (2018) selected the unmanned aerial vehicles remote sensing system as the platform to acquire image data, and studied the problem of image recognition based on residual neural network.
Website application. For example, Alam et al. (2019) revealed that various AI tech-niques from natural language processing and computer vision fields can exploit comple-mentary information generated during disaster events.
Emergency materials transportation. For example, Liu et al. (2017) proposed an emer-gency materials scheduling model in the case of material demand and vehicle amount con-tinual alteration based on dynamic programming and ant colony optimization.
4.2.2 Flood
Studies of floods in the stage of disaster response mainly include two aspects: flood spread-ing, drainage and inbound areas detection. Table 9 presents research progress on floods in the stage of disaster response.
Flood spreading and drainage. The reasonable control of floods can improve the supply of groundwater, make the soil more fertile and increase the nutrients in the soil. Research methods include both single and hybrid methods. For example, Hsu et al. (2013) applied a historical and an optimized ANFIS respectively to stimulate the real-time operation crite-ria of various pumping machines for controlling floods. Rahimi et al. (2014) used analytic
2406 Natural Hazards (2021) 107:2389–2417
1 3
Tabl
e 9
Res
earc
h pr
ogre
ss o
f floo
d in
the
stag
e of
dis
aste
r res
pons
e
Dis
aste
r typ
eO
bjec
tive
Met
hodo
logy
App
licat
ions
or e
xam
ples
Floo
dFl
ood
spre
adin
g an
d dr
aina
geSi
ngle
met
hods
Ana
lytic
hie
rarc
hy p
roce
ss +
GA
Rah
imi e
t al.
(201
4)H
ybrid
met
hods
AN
FIS
Hsu
et a
l. (2
013)
Inun
date
d ar
eas d
etec
tion
Sing
le m
etho
dsRo
botic
s + R
F Fe
ng e
t al.
(201
5); D
eep
lear
ning
Fen
g et
al.
(201
8)H
ybrid
met
hods
SVM
+ re
gula
rized
ker
nel F
ishe
r’s d
iscr
imin
ant a
naly
sis m
achi
ne
lear
ning
Irel
and
et a
l. (2
015)
2407Natural Hazards (2021) 107:2389–2417
1 3
hierarchy process of multi-criteria decision analysis and GA to find the most suitable area location for flood spreading operation in the Gareh Bygone Plain of Iran.
Inundated areas detection. For example, Feng et al. (2015) used unmanned aerial vehi-cles to provide high-resolution data for accurate detection of inundated areas under com-plex urban landscapes, and RF classifier to extract flooded areas in the spectral-textural feature space. Feng et al. (2018) applied deep learning approaches on user generated texts and photos to retrieve high quality eyewitnesses of rainfall and flooding events.
4.2.3 Other disasters
Other disasters include earthquakes, fires and storms. The research content mainly focuses on risk assessment. Research methods of these disaster types focus on single methods. Table10 provides research progress for scholars on other disasters in the stage of disaster response.
Earthquake. Research contents mainly include two aspects, e.g., website application and emergency evacuation. For example, Gelernter et al. (2013) used natural language pro-cessing methods to identify references to streets and addresses, buildings and urban spaces consisted of Twitter messages sent immediately following the February 2011 earthquake in Christchurch, New Zealand. Song et al.(2017) collected big and heterogeneous data of 1.6million users over 3 years on earthquakes that have occurred in Japan over 4 years, including news report data, transportation network data, and then built an intelligent sys-tem based on a deep learning architecture for understanding and predicting human evacua-tion behavior and mobility following different types of natural disasters.
Fire. Research contents mainly include two aspects, e.g., studies of emergency evacua-tion and emergency behavior. For example, Miller-Hooks et al. (2007) proposed a concept for ES that, through the use of sensor technology, can permit real-time assessment of the extent of blast and fire damage to a building, and can be used to aid the rescue workers and evacuees in rescue efforts and safe egress. Tissera et al. (2012) used cellular automata for modeling the dynamics of fire and smoke propagation to investigate behavioral dynam-ics for pedestrians in an emergency evacuation. Innocente et al. (2019) demonstrated the feasibility and potential of employing swarm intelligence and swarm robotics to fight fires autonomously, with a focus on the self-coordination mechanisms for the desired firefight-ing behavior to emerge.
Table 10 Research progress of other disasters in the stage of disaster response
Disaster type Objective Methodology Applications or examples
Earthquake Website application Single methods Natural language processing Gelernter et al. (2013)
Emergency evacuation Single methods Deep learning Song et al. (2017)Fire Emergency evacuation Single methods Expert decision support system Miller-Hooks
et al. (2007); Cellular automata Tissera et al. (2012)
Emergency behavior Single methods Swarm intelligence + Swarm robotics Innocente et al. (2019)
Storm Environmental analysis Single methods Robotics Murphy et al. (2008)
2408 Natural Hazards (2021) 107:2389–2417
1 3
Storm. The main research content is environmental analysis. On Oct. 24, 2005, Hur-ricane Wilma, a category 5 storm, made landfall at Cape Romano, Florida. The Center for Robot-Assisted Search and Rescue at the University of South Florida for the first time used unmanned sea surface vehicles for emergency response and established their suitability of disaster management by detecting damage to seawalls and piers, locating submerged debris (moorings and handrails), and determining safe lanes for sea navigation.
4.2.4 Remarks
According to the analysis of Sect. 4.2, we further discussed possible research deficiencies and gaps to provide suggestions for future studies in the stage of disaster response.
The neglected role of rescue resources allocation. To a great extent, the rescue capabil-ity in disaster response depends on the efficiency of rescue resources and facilities. Due to the dynamic and uncertainty of the disaster response environment, the supply and demand of resources are often unbalanced. In this case, the design of resource sharing mechanisms between adjacent facilities needs further study. In addition, the types, severity, location and spread of disasters are often different, so it is often necessary to transfer transportation modes with different types and capacities. Existing studies pay less attention to this issue.
The ability of management to make emergency decisions in disaster response needs to be further improved. The emergency response decision of the management department needs to consider many stakeholders under the conditions of time pressure, risk and uncer-tainty. It can be seen from past experience that poor response to disaster management will increase the adverse consequences (Rolland et al. 2010). Emergency decision-making often depends on the accumulation of previous knowledge and experience, but responding to emergencies is greatly limited to relevant experience. In addition, many studies have prob-lems obtaining data, and it is often difficult to overcome. It is necessary to design AI mod-els in the absence of data or whenever the data are not available. These problems restrict the development of emergency decision-making during the stage of disaster response to some degree, and further breakthroughs are needed.
4.3 Disaster recovery
There are relatively few studies at the stage of disaster recovery. The disaster types are nat-ural disasters, floods, earthquakes and tsunamis. The main research content is the analysis of disaster data, such as damage evaluation. Table 11 provides research progress of various disaster types in disaster recovery stage, including research objectives, research methods and applications.
4.3.1 Natural disaster
Research contents mainly include four aspects: damage evaluation, website application, post-disaster reconstruction and assistance.
Damage evaluation. AI models are mainly used to evaluate the severity of disasters or draw disaster damage maps. For example, Rodríguez et al. (2011) developed a fuzzy rule-based decision support system based on FL, and compared it with multiple linear regres-sion, linear discriminant analysis, classification trees and SVM to assess disaster sever-ity. Kou et al. (2014) proposed an efficient disaster assessment ES, which integrates FL, survey questionnaire, Delphi method and multi-criteria decision-making methods, and the
2409Natural Hazards (2021) 107:2389–2417
1 3
Tabl
e 11
Re
sear
ch p
rogr
ess i
n th
e st
age
of d
isas
ter r
ecov
ery
Dis
aste
r typ
eO
bjec
tive
Met
hodo
logy
App
licat
ions
or e
xam
ples
Nat
ural
dis
aste
rD
amag
e ev
alua
tion
Sing
le m
etho
dsFu
zzy
met
hodo
logy
+ M
ultip
le li
near
regr
essi
on +
Lin
ear d
iscr
imin
ant a
naly
sis +
Cla
ssifi
catio
n tre
es +
SVM
Ro
dríg
uez
et a
l. (2
011)
; FL
+ E
S K
ou e
t al.
(201
4); V
isua
l-Bag
-of-W
ords
+ H
istog
ram
of g
radi
ent o
rient
a-tio
ns +
Gab
or w
avel
ets +
SVM
+ R
F +
Ada
boos
t Vet
rivel
et a
l. (2
016)
; U-n
et c
onvo
lutio
nal n
etw
ork
(Bai
et
al.
(201
8)W
ebsi
te a
pplic
atio
nSi
ngle
met
hods
Mac
hine
lear
ning
Bai
et a
l. (2
016)
; Dee
p le
arni
ng Y
ang
et a
l. (2
019)
; Nat
ural
Lan
guag
e Pr
oces
sing
Sha
n et
al.
(201
9); C
onvo
lutio
n ne
ural
net
wor
k + B
i-Dire
ctio
nal l
ong
shor
t ter
m m
emor
y A
hmad
et a
l. (2
020)
Post-
disa
ster r
econ
struc
tion
Sing
le m
etho
dsPr
elim
inar
y te
mpo
rosp
atia
l age
nt-b
ased
mod
el +
Gam
e-th
eore
tical
app
roac
h N
ejat
et a
l. (2
012)
Dis
aste
r ass
istan
ceSi
ngle
met
hods
ES S
ingh
(200
7)Fl
ood
Dam
age
eval
uatio
nSi
ngle
met
hods
SVM
Hua
ng e
t al.
(201
5); T
rans
fer l
earn
ing +
Las
so re
gres
sion
Jian
g et
al.
(201
8)Ea
rthqu
ake
Dam
age
eval
uatio
nSi
ngle
met
hods
Robo
tics +
SVM
Xu
et a
l. (2
018)
Hyb
rid m
etho
dsD
ecis
ion
tree-
exp
ert s
yste
m C
heem
a (2
007)
; Hyb
rid n
euro
-fuz
zy sy
stem
bas
ed o
n a
spec
ial t
hree
-laye
r fe
edfo
rwar
d ar
tifici
al n
eura
l net
wor
k an
d fu
zzy
rule
bas
es C
arre
no e
t al.
(201
0); M
achi
ne le
arni
ng F
rank
et
al.
(201
0)Ts
unam
iD
amag
e ev
alua
tion
Sing
le m
etho
dsD
eep
lear
ning
Sub
lime
et a
l. (2
019)
2410 Natural Hazards (2021) 107:2389–2417
1 3
experimental results show that the proposed ES is not only accurate, but also has strong adaptability to different environments.
Website application. The main purpose of this topic is to extract public sentiment and loss data through information published by the public on social media. For example, Yang et al. (2019) introduced the deep learning method to extract fine-grained public emotional information from Chinese social media big data to assist in disaster analysis. Shan et al. (2019) developed a disaster damage assessment model based on social media texts.
Post-disaster reconstruction and assistance. Singh (2007) develop a prototype ES for monitoring and evaluating food aid by international disaster relief organizations. Nejat et al. (2012) developed a preliminary temporospatial agent-based model combined with game-theoretical approach account for homeowners’ dynamic interactions with their neigh-bors in housing re-establishment in the context of post-disaster recovery.
4.3.2 Other disasters
The main research content of floods, earthquakes and tsunamis is damage evaluation. For example, Huang et al. (2018) established an SVM model suitable for flood evaluation to achieve comprehensive assessment of multidimensional disaster indicators in one-dimen-sional continuous space. Jiang et al. (2018) presented a new approach to extract urban waterlogging depths from video images based on transfer learning and lasso regression. Carreno et al. (2010) used hybrid neuro-fuzzy system based on a special three-layer feed-forward ANN and fuzzy rule to provide decisions about safety, habitability, and reparabil-ity of the buildings during the emergency response phase. Xu et al. (2018) used SVM and unmanned aerial vehicles for earthquake damage mapping. Sublime et al. (2019) presented a state-of-the-art deep learning approach to evaluate both the extent and the severity of the damages applied to satellite images taken from Tohoku tsunami of 2011.
4.3.3 Remarks
According to the analysis of Sect. 4.3, we further discussed possible research deficiencies and gaps to provide suggestions for future studies at the stage of disaster recovery.
The accuracy of disaster loss assessment needs to be improved. Technologies such as remote sensing images and drones have become useful tools for mapping damage mapping in most areas of the world. The mission of recognizing objects in an image can be briefly summarized in four components, they are classification, detection, localization, and seman-tic segmentation (Pi et al.2020). The first challenge to this approach is robustly processing a large amount of data to identify and map objects of interest in real-time, and there is often a trade-off between accuracy and speed (Pi et al.2020). Therefore, although the use of AI models to map and estimate the impact of disasters has made great progress, image resolution, information quality and evaluation speed still need to be improved.
The neglected role of the effectiveness emergency management evaluation. Based on the gradually improved organizational system and a more scientific concept of disaster prevention and control, the effectiveness of disaster emergency management continues to be improved. However, each disaster will bring new problems, affecting the effectiveness of the original emergency management system and plan. How to grasp the law of disas-ter emergency management more accurately and comprehensively, and test the effect of emergency treatment, so as to formulate more effective policies for making the most of the emergency management effectiveness, seems to be important and urgent.
2411Natural Hazards (2021) 107:2389–2417
1 3
5 Conclusion and outlook
In the previous sections, the concept of AI and the stage analysis of NDM were given, types of AI methods applied in various stages of NDM were collected addressing their specified issues. This conclusive section attempts to summarize identified difficulties and future research directions.
5.1 Current NDM challenges with AI models
5.1.1 Disaster preparedness
The research content in the stage of disaster preparedness is primarily concentrated on risk assessment. The data used are mainly based on historical data or collected data through smart sensors. Through historical data, AI models can be used for disaster risk analysis and regional sensitivity assessment (Khosravi et al. 2018). The information of smart sensors mainly comes from remote sensing, monitoring instruments and social media. It can be used to regularly create and update disaster sensitive maps and enrich information sources about disaster situations (Stumpf et al. 2011; Chou et al. 2011).
However, due to the lack of theory and algorithm strategy to represent heterogeneous data, the application requirements of multi-source data pose challenges to data integration and analysis (Nweke et al. 2019). Judging from the current research progress, studies on data integration and extraction are mainly concentrated in the data generated by remote sensing, which is not consistent with the reality of the increased multi-source data. In fact, each data source has important information that cannot be acquired by other sources, and the lack of consideration between different data sources can lead to information redundancy or loss. Therefore, how to identify and evaluate the complementarity of multi-source data, overcome the heterogeneity of data format, time range and semantics in order to reduce the redundancy or loss of data and obtain valuable information is an important challenge for current studies in the stage of disaster preparedness.
5.1.2 Disaster response
How to make adequate emergency decisions in the disaster environment is the focus of the disaster response stage. First, disaster managers or management departments need to use AI models to support time-sensitive decision making, such as the application of rescue robots for search and rescue work, the application of Markov decision-making and case-based reasoning methods to assemble response plans. (Blitch et al. 1996; Amailef et al. 2013; Ramchurn et al. 2016). Second, the disaster victims also need to make emergency response decisions to evacuate from the disaster area (Sharma et al. 2018).
However, decision-making in a disaster environment often demands consideration of the actions of other participants to make decisions alone or jointly. For example, in the emer-gency evacuation of disaster victims, by observing their behavior, it can be found that the emergency response decision-making process will be influenced by the interaction with other subjects (Tissera et al. 2012). Besides, how to use cooperative multi-robot teams in search and rescue environments are also a challenging research field. Therefore, in the pro-cess of disaster emergency decision-making, it is essential to investigate and practice how
2412 Natural Hazards (2021) 107:2389–2417
1 3
to combine hierarchical reinforcement learning, analytic hierarchy process and other meth-ods to clarify the probable relationship between objectives and different path choices, so as to help decision makers find the best task sequence.
5.1.3 Disaster recovery
Studies mainly emphasize on post-disaster data analysis at the stage of disaster recovery. The analysis and summarization of post-disaster data can convey the impact of disasters, and can capture the complex and dynamic relationships between different subjects and their performance, which is essential for enhancing the efficiency of disaster management (Jiang et al. 2018; Fan et al. 2019).
At present, the main content of post-disaster data analysis is to carry out disaster loss assessment in order to give special attention. For example, extremely vulnerable sectors and areas should be fully considered before disasters so that targeted protection measures can be taken. Simultaneously, relevant policy support can also be formulated to those sec-tors and areas so as to reduce economic losses during post-disaster recovery and recon-struction. A commonly used method is to adopt remote sensing images for automatic map-ping after a disaster, and then use machine learning technology to instantly classify or label the damage with ES (Kou et al. 2014; Xu et al. 2018). However, remote sensing mapping may have different characteristics in terms of sensor field, image proportion and scene complexity, and information extraction is also elusive, which restricts the accuracy of loss assessment to some extent. In addition, how to verify the effect of emergency response through data analysis after a disaster, capture and solve the problem of low efficiency of disaster management, such as unoptimized resource allocation, so as to optimize the disas-ter management process and upgrade the efficiency of disaster emergency management, is also a difficulty that needs to be solved in the stage of disaster recovery.
5.2 Possible directions and opportunities
First, the impact of disasters is huge, so every decision in the process of disaster manage-ment is vital. Disaster managers or departments need to have a sufficient understanding of the process of producing predicted results or determining decision plans, so as to direct the following disaster prevention work with confidence. Up to now, there has always been a distrust of AI, which is chiefly due to the fact that its computation process is in the form of a black box, leading to the lack of explanatory results, interactivity and operability with users (Tizhoosh et al. 2018). At present, how to strengthen the interpretability of AI, vali-date AI model predictions and conduct retrospective analysis by comparing prospective AI predictions with actual situations, should be further explored in the future.
Second, in order to further improve the level of NDM, AI models also need to be con-tinuously innovative. The era of big data has increased the types of data not previously used for analysis, such as data from social media, which has advanced the application of AI models in NDM. Multi-source data analysis based on data fusion is an important part of gathering complete disaster data. There is a strong need to further investigate the synergy of AI and Big Data in NDM. Besides, the large amount of scenario data extracted dur-ing the disaster process often carries noise, rumors and fake information, which seriously hinders the precise extraction of disaster data (Stieglitz et al. 2018). Therefore, for obtain-ing complete and reliable information in the context of big data, it is vital to continue to
2413Natural Hazards (2021) 107:2389–2417
1 3
develop AI models, integrate multi-source heterogeneous information, along with carry out disaster features extraction, classify and identify.
Third, studies of various disaster types are presented in the form of single methods and hybrid methods. To better the quality of prediction, there appears an ever-increasing trend in constructing hybrid methods. What are the differences between the research results of single methods and hybrid methods in terms of accuracy, robustness, computation cost and speed? Can hybrid methods be used to solve the problem that single methods cannot address at present? This deserves enough attention.
Fourth, the advanced performance of AI is intelligent machines. They are deep inte-gration of various technologies such as smart sensing, big data analysis applications, machine learning technologies. Although rescue robots currently have the ability to carry out disaster search and rescue work individually or in teams, they do not have autonomous intelligent behaviors and can only complete single-mode operations according to the set procedures. Therefore, how to make breakthroughs in crucial technologies, for example, autonomous learning and autonomous behavior, and finally create rescue robots with human-like behavior ability is also an important direction of future NDM research.
Acknowledgements This research was supported by: National Social and Scientific Fund Program (16ZDA047, 18ZDA052, 17BGL142); The Natural Science Foundation of China (91546117, 71373131).
Compliance with ethical standards
Conflict of interests The authors declare that they have no conflict of interest.
References
Ahmadi M, Seifi A, Tootooni B (2015) A humanitarian logistics model for disaster relief operation consid-ering network failure and standard relief time: a case study on San Francisco district. Transp Res Pt e-Logist Transp Rev 75:145–163
Alam F, Ofli F, Imran M (2019) Descriptive and visual summaries of disaster events using artificial intel-ligence techniques: case studies of Hurricanes Harvey, Irma, and Maria. Behav Inf Technol 1:1–31
Amailef K, Lu J (2013) Ontology-supported case-based reasoning approach for intelligent m-Government emergency response services. Decis Support Syst 55(1):79–97
Anaya-Arenas AM, Renaud J, Ruiz A (2014) Relief distribution networks: a systematic review. Ann Oper Res 223(1):53–79
Badrzadeh H, Sarukkalige R, Jayawardena AW (2015) Hourly runoff forecasting for flood risk management: Application of various computational intelligence models. J Hydrol 529:1633–1643
Bai Y, Mas E, Koshimura S (2018) Towards operational satellite-based damage-mapping using U-Net con-volutional network: a case study of 2011 Tohoku Earthquake-Tsunami. Remote Sens 10(10):1626
Blitch JG (1996) Artificial intelligence technologies for robot assisted urban search and rescue. Expert Syst Appl 11(2):109–124
Budiharto W (2015) Intelligent surveillance robot with obstacle avoidance capabilities using neural net-work. Comput Intell Neurosci 3:1–5
Carreno ML, Cardona OD, Barbat AH (2010) Computational tool for post-earthquake evaluation of damage in buildings. Earthq Spectra 26(1):63–86
Caunhye AM, Nie X, Pokharel S (2012) Optimization models in emergency logistics: a literature review. Socio-Econ Plan Sci 46(1):4–13
Chang FJ, Tseng KY, Chaves P (2007) Shared near neighbours neural network model: a debris flow warning system. Hydrol Process 21(14):1968–1976
Chatterjee R, Matsuno F (2005) Robot description ontology and disaster scene description ontology: analy-sis of necessity and scope in rescue infrastructure context. Adv Robot 19(8):839–859
Chau KW, Wu CL, Li YS (2005) Comparison of several flood forecasting models in Yangtze river. J Hydrol Eng 10(6):485–491
2414 Natural Hazards (2021) 107:2389–2417
1 3
Chen CS, Cheng MY, Wu YW (2012) Seismic assessment of school buildings in Taiwan using the evolu-tionary support vector machine inference system. Expert Syst Appl 39(4):4102–4110
Chen W, Pourghasemi HR, Kornejady A, Zhang N (2017) Landslide spatial modeling: Introducing new ensembles of ANN, MaxEnt, and SVM machine learning techniques. Geoderma 305:314–327
Cheng MY, Hoang ND (2016) Slope collapse prediction using bayesian framework with K-Nearest neighbor density estimation: case study in Taiwan. J Comput Civil Eng 30(1):04014116
Choi C, Kim J, Kim J, Kim D, Bae Y, Kim HS (2018) Development of heavy rain damage prediction model using machine learning based on big data. Adv Meteorol. https ://doi.org/10.1155/2018/50249 30
Chou CH, Zahedi FM, Zhao H (2011) Ontology for developing web sites for natural disaster management: methodology and implementation. Ieee Transactions on Systems Man and Cybernetics Part a-Sys-tems and Humans 41(1):50–62
Chou CH, Zahedi FM, Zhao H (2014) Ontology-based evaluation of natural disaster management websites: a multistakeholder perspective. MIS Q 38(4):997–1016
Darabi H, Choubin B, Rahmati O, Haghighi AT, Pradhan B, Klove B (2019) Urban flood risk mapping using the GARP and QUEST models: a comparative study of machine learning techniques. J Hydrol 569:142–154
Dawson CW, Brown MR, Wilby RL (2000) Inductive learning approaches to rainfall-runoff modelling. Int J Neural Syst 10(1):43–57
Dwivedi YK, Hughes L, Ismagilova E, Aarts G, Coombs C, Crick T, Duan Y, Dwivedi R, Edwards J, Eirug A, Galanos V, Ilavarasan PV, Janssen M, Jones P, Kar AK, Kizgin H, Kronemann B, Lal B, Wil-liams MD (2019) Artificial intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy. Int J Inf Manage. https ://doi.org/10.1016/j.ijinf omgt.2019.08.002
EM-DAT (2019) Natural disasters 2018. The OFDA/CRED International Disaster Database. Universite´ Catholique de Louvain, Brussels
Fan C, Mostafavi A (2019) Metanetwork framework for performance analysis of disaster management sys-tem-of-Systems. IEEE Syst J 14:1265–1276
Feng Q, Liu J, Gong J (2015) Urban Flood mapping based on unmanned aerial vehicle remote sensing and random forest classifier-a case of Yuyao. China Water 7(4):1437–1455
Feng Y, Sester M (2018) Extraction of pluvial flood relevant volunteered geographic information (VGI) by deep learning from user generated texts and photos. ISPRS Int Geo-Inf 7(2):39
Gelernter J, Balaji S (2013) An algorithm for local geoparsing of microtext. Geoinformatica 17(4):635–667Geng M, Xu K, Zhou X, Ding B, Wang H, Zhang L (2019) Learning to cooperate via an attention-based
communication neural network in decentralized multi-robot exploration dagger. Entropy 21:294Gizaw MS, Gan TY (2016) Regional flood frequency analysis using support vector regression under histori-
cal and future climate. J Hydrol 538:387–398Hadi SJ, Tombul M (2018) Monthly streamflow forecasting using continuous wavelet and multi-gene
genetic programming combination. J Hydrol 561:674–687Hallegatte S (2008) An adaptive regional input-output model and its application to the assessment of the
economic cost of Katrina. Risk Anal 28(3):779–799Han Y, Deng Y (2018) An enhanced fuzzy evidential DEMATEL method with its application to identify
critical success factors. Soft Comput 22(15):5073–5090Hanna AM, Ural D, Saygili G (2007) Neural network model for liquefaction potential in soil deposits using
Turkey and Taiwan earthquake data. Soil Dyn Earthq Eng 27(6):521–540Hart E, Sim K, Kamimura K, Meredieu C, Guyon D, Gardiner B (2019) Use of machine learning techniques
to model wind damage to forests. Agric For Meteorol 265:16–29Hoang ND, Pham AD (2016) Hybrid artificial intelligence approach based on metaheuristic and machine
learning for slope stability assessment: a multinational data analysis. Expert Syst Appl 46:60–68Hsu NS, Huang CL, Wei CC (2013) Intelligent real-time operation of a pumping station for an urban drain-
age system. J Hydrol 489:85–97Huang L, Li J, Hao H, Li X (2018) Micro-seismic event detection and location in underground mines
by using convolutional neural networks (CNN) and deep learning. Tunn Undergr Space Technol 81:265–276
Huang Z, Zhou J, Song L, Lu Y, Zhang Y (2010) Flood disaster loss comprehensive evaluation model based on optimization support vector machine. Expert Syst Appl 37(5):3810–3814
Innocente MS, Grasso P (2019) Self-organising swarms of firefighting drones: Harnessing the power of col-lective intelligence in decentralised multi-robot systems. J Comput Sci 34:80–101
Jaafari A, Zenner EK, Panahi M, Shahabi H (2019) Hybrid artificial intelligence models based on a neuro-fuzzy system and metaheuristic optimization algorithms for spatial prediction of wildfire probability. Agric For Meteorol 266–267:198–207
2415Natural Hazards (2021) 107:2389–2417
1 3
Jalalkamali A, Moradi M, Moradi N (2015) Application of several artificial intelligence models and ARI-MAX model for forecasting drought using the Standardized Precipitation Index. Int J Environ Sci Technol 12(4):1201–1210
Jiang J, Liu J, Qin CZ, Wang D (2018) Extraction of urban waterlogging depth from video images using transfer learning. Water 10(10):1485
Jiao G, Guo T, Ding Y (2016) A new hybrid forecasting approach applied to hydrological data: a case study on precipitation in northwestern China. Water 8(9):367
Kalantari Z, Ferreira CSS, Koutsouris AJ, Ahlmer AK, Cerdà A, Destouni G (2019) Assessing flood prob-ability for transportation infrastructure based on catchment characteristics, sediment connectivity and remotely sensed soil moisture. Sci Total Environ 661:393–406
Kang B, Choo H (2016) A deep-learning-based emergency alert system. ICT Express 2(2):67–70Kaplan AM, Haenlein M (2019) Siri, Siri, in my hand: Who’s the fairest in the land? On the interpretations,
illustrations, and implications of artificial intelligence. Bus Horiz 62(1):15–25Khosravi K, Pham BT, Chapi K, Shirzadi A, Shahabi H, Revhaug I, Prakash I, Bui DT (2018) A compara-
tive assessment of decision trees algorithms for flash flood susceptibility modeling at Haraz water-shed, northern Iran. Sci Total Environ 627:744–755
Kojima H, Obayashi S (2006) An inverse analysis of unobserved trigger factor for slope stability evaluation. Comput Geosci 32(8):1069–1078
Kou G, Ergu D, Shi Y (2014) An integrated expert system for fast disaster assessment. Comput Oper Res 42:95–107
Kumar R, Kumar D (2016) Hybrid swarm intelligence energy efficient clustered routing algorithm for wire-less sensor networks. J Sens 3:1–19
Li P, Zhao N, Zhou D, Cao M, Li J, Shi X (2014) Multivariable time series prediction for the icing process on overhead power transmission line. Sci World J. https ://doi.org/10.1155/2014/25681 5
Li TZ, Li YX, Xl Y (2017) Rock burst prediction based on genetic algorithms and extreme learning machine. J Cent South Univ 24(9):2105–2113
Liu J, Xie K (2017) Emergency materials transportation model in disasters based on dynamic programming and ant colony optimization. Kybernetes 46(4):656–671
Liu S, Xu J, Zhao J, Xie X, Zhang W (2014) Efficiency enhancement of a process-based rainfall-runoff model using a new modified AdaBoost.RT technique. Appl Soft Comput 23:521–529
Liu W, Hu G, Li J (2012) Emergency resources demand prediction using case-based reasoning. Saf Sci 50(3):530–534
Liu Y, Nejat G (2016) Multirobot cooperative learning for semiautonomous control in urban search and res-cue applications. J Field Robot 33(4):512–536
Manley M, Kim YS, Christensen K, Chen A (2016) Airport emergency evacuation planning: an agent-based simulation study of dirty bomb scenarios. Ieee Transactions on Systems Man Cybernetics-Systems 46(10):1390–1403
Mashi SA, Oghenejabor OD, Inkani AI (2019) Disaster risks and management policies and practices in Nigeria: a critical appraisal of the national emergency management agency act. J Disaster Risk Reduction 33:253–265
Miller-Hooks E, Krauthammer T (2007) An intelligent evacuation, rescue and recovery concept. Fire Tech-nol 43(2):107–122
Mitsopoulos I, Mallinis G (2017) A data-driven approach to assess large fire size generation in Greece. Nat Hazards 88(3):1591–1607
Miyasato GH, Dong W, Levitt RE, Boissonnade AC (1986) Implementation of a knowledge based seismic risk evaluation system on microcomputers. Artif Intell Eng 1(1):29–35
Murphy RR, Steimle E, Griffin C, Cullins C, Hall M, Pratt K (2008) Cooperative use of unmanned sea sur-face and micro aerial vehicles at Hurricane Wilma. J Field Robot 25(3):164–180
Nayak MA, Ghosh S (2013) Prediction of extreme rainfall event using weather pattern recognition and sup-port vector machine classifier. Theor Appl Climatol 114(3–4):583–603
Nejat A, Damnjanovic I (2012) Agent-based modeling of behavioral housing recovery following disasters. Comput Aided Civil Infrastruct Eng 27(10):748–763
Nweke HF, Teh YW, Mujtaba G, Al-Garadi MA (2019) Data fusion and multiple classifier systems for human activity detection and health monitoring: review and open research directions. Inf Fusion 46:147–170
Ofli F, Meier P, Imran M, Castillo C, Tuia D, Rey N, Briant J, Millet P, Reinhard F, Parkan M, Joost S (2016) Combining human computing and machine learning to make sense of big (aerial) data for dis-aster response. Big Data 4(1):47–59
Pai PF, Li LL, Hung WZ, Lin KP (2014) Using adaboost and rough set theory for predicting debris flow disaster. Water Resour Manag 28(4):1143–1155
2416 Natural Hazards (2021) 107:2389–2417
1 3
Perumal M, Tayfur G, Rao CM, Gurarslan G (2017) Evaluation of a physically based quasi-linear and a con-ceptually based nonlinear Muskingum methods. J Hydrol 546:437–449
Pi Y, Nath ND, Behzadan AH (2020) Convolutional neural networks for object detection in aerial imagery for disaster response and recovery. Adv Eng Inform 43:101009
Poblet M, García-Cuesta E, Casanovas P (2018) Crowdsourcing roles, methods and tools for data-intensive disaster management. Inf Syst Front 20(6):1363–1379
Poole DL, Mackworth AK (2010) Artificial intelligence: Foundations of computational agents. Cambridge University Press, Cambridge, UK
Pozdnoukhov A, Foresti L, Kanevski M (2009) Data-driven topo-climatic mapping with machine learning methods. Nat Hazards 50(3):497–518
Pramanik N, Panda RK (2009) Application of neural network and adaptive neuro-fuzzy inference systems for river flow prediction. Hydrol Sci J-J Sci Hydrol 54(2):247–260
Ragettli S, Zhou J, Wang H, Liu C, Guo L (2017) Modeling flash floods in ungauged mountain catchments of China: a decision tree learning approach for parameter regionalization. J Hydrol 555:330–346
Rahimi S, Roodposhti MS, Abbaspour RA (2014) Using combined AHP-genetic algorithm in artificial groundwater recharge site selection of Gareh Bygone Plain. Iran Environ Earth Sci 72(6):1979–1992
Rajaee T, Ebrahimi H, Nourani V (2019) A review of the artificial intelligence methods in groundwater level modeling. J Hydrol 572:336–351
Ramchurn SD, Wu F, Jiang W, Fischer JE, Reece S, Roberts S, Rodden T, Greenhalgh C, Jennings NR (2016) Human-agent collaboration for disaster response. Auton Agents Multi-Agent Syst 30(1):82–111
Rodriguez JT, Vitoriano B, Montero J, Kecman V (2011) A disaster-severity assessment DSS comparative analysis. OR Spectrum 33(3):451–479
Rolland E, Patterson RA, Ward K, Dodin B (2010) Decision support for disaster management. Oper Manag Res 3:68–79
Rose A (2004) Defining and measuring economic resilience to disasters. Disaster Prev Manag 13(4):307–314
Rose A, Liao SY (2005) Modeling regional economic resilience to disasters: a computable general equilib-rium analysis of water service disruptions. J Reg Sci 45(1):75–112
Russell SJ, Norvig P (2016) Artificial intelligence: a modern approach. Pearson Education Limited, Malaysia
Sahebjamnia N, Torabi SA, Mansouri SA (2015) Integrated business continuity and disaster recovery plan-ning: towards organizational resilience. Eur J Oper Res 242(1):261–273
Sdao F, Lioi DS, Pascale S, Caniani D, Mancini IM (2013) Landslide susceptibility assessment by using a neuro-fuzzy model: a case study in the Rupestrian heritage rich area of Matera. Nat Hazards Earth Syst Sci 13(2):395–407
Seo Y, Kim S, Kisi O, Singh VP (2015) Daily water level forecasting using wavelet decomposition and arti-ficial intelligence techniques. J Hydrol 520:224–243
Shan S, Zhao F, Wei Y, Liu M (2019) Disaster management 2.0: A real-time disaster damage assess-ment model based on mobile social media data—A case study of Weibo (Chinese Twitter). Saf Sci 115:393–413
Sharma S, Ogunlana K, Scribner D, Grynovicki J (2018) Modeling human behavior during emergency evac-uation using intelligent agents: a multi-agent simulation approach. Inf Syst Front 20(4):741–757
Singh N (2007) Expert system prototype of food aid distribution. Asia Pac J Clin Nutr 16:116–121Soh YW, Koo CH, Huang YF, Fung KF (2018) Application of artificial intelligence models for the predic-
tion of standardized precipitation evapotranspiration index (SPEI) at Langat River Basin, Malaysia. Comput Electron Agric 144:164–173
Song X, Shibasaki R, Yuan NJ, Xie X, Li T, Adachi R (2017) DeepMob: learning deep knowledge of human emergency behavior and mobility from big and heterogeneous data. ACM Trans Inf Syst 35(4):1–19
Steger S, Brenning A, Bell R, Petschko H, Glade T (2016) Exploring discrepancies between quantitative validation results and the geomorphic plausibility of statistical landslide susceptibility maps. Geo-morphology 262:8–23
Stieglitz S, Mirbabaie M, Ross B, Neuberger C (2018) Social media analytics – challenges in topic discov-ery, data collection, and data preparation. Int J Inf Manage 39:156–168
Stumpf A, Kerle N (2011) Object-oriented mapping of landslides using Random Forests. Remote Sens Environ 115(10):2564–2577
Sublime J, Kalinicheva E (2019) Automatic post-disaster damage mapping using deep-learning techniques for change detection: case study of the Tohoku Tsunami. Remote Sens 11(9):1123
Thomas S, Pillai GN, Pal K, Jagtap P (2016) Prediction of ground motion parameters using randomized ANFIS (RANFIS). Appl Soft Comput 40:624–634
2417Natural Hazards (2021) 107:2389–2417
1 3
Tissera PC, Printista AM, Luque E (2012) A hybrid simulation model to test behaviour designs in an emer-gency evacuation. Procedia Comput Sci 9:266–275
Tizhoosh HR, Pantanowitz L (2018) Artificial intelligence and digital pathology: challenges and opportuni-ties. J Pathol Inform 9:38
Wang W, Peng W, Tong L, Tan X, Xin T (2019) Study on sustainable development of power transmission system under ice disaster based on a new security early warning model. J Clean Prod 228:175–184
Wang Z, Lai C, Chen X, Yang B, Zhao S, Bai X (2015) Flood hazard risk assessment model based on ran-dom forest. J Hydrol 527:1130–1141
Tayfur G, Singh VP, Moramarco T, Barbetta S (2018) Flood hydrograph prediction using machine learning methods. Water 10(8):968
Wex F, Schryen G, Feuerriegel S, Neumann D (2014) Emergency response in natural disaster management: Allocation and scheduling of rescue units. Eur J Oper Res 235(3):697–708
Wu D, Cui Y (2018) Disaster early warning and damage assessment analysis using social media data and geo-location information. Decis Support Syst 111:48–59
Xia M, Liu W, Shi B, Weng L, Liu J (2019) Cloud/snow recognition for multispectral satellite imagery based on a multidimensional deep residual network. Int J Remote Sens 40(1):156–170
Xie W, Rose A, Li S, He J, Li N, Ali T (2018) Dynamic economic resilience and economic recovery from disasters: A quantitative assessment. Risk Anal 38(6):1306–1318
Xu Z, Wu L, Zhang Z (2018) Use of active learning for earthquake damage mapping from UAV photogram-metric point clouds. Int J Remote Sens 39(15–16):5568–5595
Yang T, Xie J, Li G, Mou N, Li Z, Tian C, Zhao J (2019) Social media big data mining and spatio-temporal analysis on public emotions for disaster mitigation. ISPRS Int Geo-Inf 8(1):29
Zhao K, He T, Wu S, Wang S, Dai B, Yang Q, Lei Y (2018) Application research of image recognition tech-nology based on CNN in image location of environmental monitoring UAV. EURASIP J Image Video Process. https ://doi.org/10.1186/s1364 0-018-0391-6
Zheng X, Sun A (2019) Collecting event-related tweets from twitter stream. J Assoc Inf Sci Tech 70(2):176–186
Zhou Y, Guo S, Chang FJ (2019) Explore an evolutionary recurrent ANFIS for modelling multi-step-ahead flood forecasts. J Hydrol 570:343–355
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.