Progress in Disaster Science 23 (2024) 100357

Contents lists available at ScienceDirect

Progress in Disaster Science

journal homepage: [Link]/locate/pdisas

Integration of fuzzy AHP and explainable AI for effective coastal risk

management: A micro-scale risk analysis of tropical cyclones
Tanmoy Das a , Swapan Talukdar b , Shahfahad a , Mohd Waseem Naikoo c , Ishita Afreen Ahmed d ,
Atiqur Rahman a, * , Md Kamrul Islam e , Edris Alam f, g
a
Department of Geography, Faculty of Sciences, Jamia Millia Islamia, New Delhi 110025, India
b
Department of Geography, Asutosh College, Kolkata, West Bengal 700026, India
c
Department of Geography and Disaster Management, University of Kashmir, Hazratbal, Srinagar, Jammu & Kashmir 190006, India
d
Department of Civil Engineering, Indian Institute of Technology, New Delhi 110016, India
e
Department of Civil and Environmental Engineering College of Engineering, King Faisal University, AlAhsa 31982, Saudi Arabia
f
Faculty of Resilience, Rabdan Academy, Abu Dhabi 22401, United Arab Emirates
g
Department of Geography and Environmental Studies, University of Chittagong, Chittagong 4331, Bangladesh

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

Keywords: The east coast of India, especially the coastal region of Odisha, is highly threatened by tropical cyclones. This
Tropical cyclone risk study develops a detailed risk map for tropical cyclones in the coastal districts of Odisha at the micro level,
Multi-criteria decision-making focusing on the assessment of risk factors at the block level. Using a multi-criteria decision making (MCDM)
Explainable AI approach
approach, the study considers four primary risk components: Exposure, vulnerability, susceptibility, and miti-
XGBoost model
Coastal areas
gation options. The Explainable Artificial Intelligence (XAI) framework, which uses the XGBoost model in
Block level analysis conjunction with SHAP values, is applied to identify and elucidate the factors influencing risk levels in 69 blocks.
Results indicate that about 65% of the area is at high risk to tropical cyclone, especially in the northeastern and
central regions. In particular, 32 blocks are classified as high to very high-risk zones. The study shows a contrast
in risk levels, with blocks in the northeast and southeast at higher risk, while blocks in the southern regions such
as Ganjam and Puri and in the central parts of Kendrapara and Baleswar districts are at lower risk. The findings
from this study are crucial for local authorities to identify vulnerable areas and improve cyclone preparedness
and risk management strategies in Odisha.

1. Introduction India has ranked third in natural disasters over the last 20 years,
behind China and the USA [95]. Cyclones frequently hit the east coast of
Tropical cyclones are the most destructive weather events in coastal India, especially Odisha, causing significant loss and damage
regions, causing significant loss of life, destruction of property, agri- ([15,44,74]; [69]; [19]). In the past 109 years, the coastal regions of
cultural setbacks, and disruption of communications [14,78,83,88]. Odisha have been affected in 37% of all cyclone landfalls, making it the
Each year, an average of ninety tropical storms forms in different most vulnerable area in the country ([22]; [16,53]). The super cyclone
cyclone basins around the world [104,105]. Historical data from 1970 to of 1999 left around 350,000 people homeless and claimed almost
2010 recorded a total of 637 tropical cyclones worldwide [97]. Coastal 10,000 lives [55]. In recent years, Odisha has suffered severe destruction
regions around the world are regularly threatened and affected by these from major cyclones, including Yass in 2021, Fani in 2019, Hudhud in
cyclones [20]. Between 1970 and 2019, cyclones caused around 1.965 2014 and Phailin in 2013 [17,18,23,41,54,55]. In contrast to the 1999
trillion dollars in economic damage worldwide [101]. In the last two super cyclone in Odisha, there was an extensive evacuation campaign
hundred years, around 1.9 million deaths worldwide have been attrib- for Phailin and Hudhud. This was due to the introduction of a real-time
uted to tropical cyclones [33,86]. The increased vulnerability of coastal storm surge warning system and the dissemination of information. These
populations to these storms due to climate change and rising sea levels measures were introduced by the Indian National Center for Ocean In-
has become a growing problem. formation Services (INCOIS) under the Ministry of Earth Sciences,

* Corresponding author.
E-mail address: arahman2@[Link] (A. Rahman).

[Link]
Received 11 May 2024; Received in revised form 26 June 2024; Accepted 18 July 2024
Available online 22 July 2024
2590-0617/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license ([Link]
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Government of India [75]. located between 18◦ 58′22′′ N - 21◦ 43′45′′ N latitude and 84◦ 22′ E - 87◦ 28′
The integration of effective management and mitigation methods E longitude. This research area has an area of 22, 092 km2 and home to
offers the potential to significantly mitigate the harmful effects of over 10.1 million residents with a population density of 560 person/km2
tropical cyclones [8,49]. A thorough risk assessment is crucial for as per the Census of India, 2011. The study area comprises six coastal
developing effective mitigation measures [32,37]. The vulnerability of a districts of Odisha, selected based on administrative boundaries to
community depends on factors such as exposure, hazard and existing ensure comprehensive coverage of regions with varying degrees of
mitigation capacities [33]. Prioritization and effective allocation of re- cyclone exposure and vulnerability. These districts include both areas
sources based on the assessment of the importance of the parameters close to the coastline and those situated further inland with varying
helps in decision making and allocation of resources [10,70]. Assessing topographies. This deliberate selection aims to capture a broad spectrum
the importance of these parameters provides insights into potential risks of risk factors and to assess the spatial variation in cyclone risk across
and uncertainties and helps decision makers to make sound decisions different types of terrain and distances from the coast. The region is sub-
[76]. Both structural measures (e.g. cyclone shelters, health centers) and divided in 69 administrative blocks (Fig. 1, Supplementary Material 1).
non-structural efforts (e.g. cyclone warning systems, literacy rates, A “block” refers to an administrative unit within a district. It is an
policy implementation) are necessary to improve the resilience of intermediary level of administration between the district and the village.
communities and reduce the impact of tropical cyclones [33]. Assessing Blocks are also known as talukas, mandals, or tehsils in different states of
risks and understanding the importance of parameters can lead to a India. This area exhibits a tropical monsoon climate, experiencing an
reduction in disaster losses and better protection against future events average daily temperature of 28 ◦ C to 35 ◦ C and annual rainfall of 1400
[33]. In this context, our study uses the Fuzzy Analytical Hierarchy mm to 1700 mm. Additionally, Odisha is identified as part of the
Process (FAHP), a method that improves traditional decision-making Empowered Action Group (EAG) states, signalling challenging socio-
processes by allowing ambiguity and uncertainty through three economic circumstances. The demographic characteristics of state
different levels: Goal, Criterion and Factor. This approach facilitates a mainly feature a mix of population groups, including Scheduled Castes,
structured assessment of the risks associated with tropical cyclones by Scheduled Tribes, and various other communities. The literacy rate is
using expert opinions to prioritize and weight different risk parameters quite high in most of the blocks of Kendrapara, Jagatsinghpur, and Puri
([48]; [6]). The FAHP results, which reflect the risk assessments derived districts. Nevertheless, numerous districts within Ganjam and the
by the experts, serve as target variables for the subsequent imple- western regions of Balasore exhibit lower levels of literacy. The econ-
mentation of machine learning models. omy thrives on agriculture and fishing, leveraging the fertile plains of
To ensure the reliability and impartiality of these assessments, we the delta area. Moreover, tourism significantly contributes to the econ-
integrate Explainable Artificial Intelligence (XAI). Specifically, after the omies of Puri and Kendrapara districts, driven by religious, cultural, and
FAHP analysis, we use the XGBoost machine learning model to learn environmental allurements. Seasonal migration for employment,
from the data and predict the FAHP results. After training the model, we particularly to urban centres and other states, is common. The coastal
apply Local Interpretable Model-agnostic Explanations (LIME) and districts exhibit a rich tapestry of cultures, languages, and traditions,
SHapley Additive exPlanations (SHAP) within the XAI framework. These influenced by their historical and geographical context. This area was
tools clarify how the input variables affect the model predictions. They severely devasted by several super cyclone and the occurrence, strength,
provide transparency to the modeling results and allow stakeholders to and consequences of tropical cyclones are notably elevated because of its
understand why certain areas have been classified as high risk. This geographical positioning and specific topographical attributes. A record
integration of XAI is crucial for improving the credibility of model of past cyclone damages is provided in the supplementary section
predictions as it provides a clear, unbiased view of the factors influ- (Supplementary Material 3). Although the northern part of the Odisha’s
encing risk assessments [2,3,92]. Therefore, our study focuses on a coastline has distinctive funnel shape and this shape can channelize
detailed assessment of tropical cyclone risk in the coastal regions of cyclone, guiding them towards coast. Due to the high risk of tropical
Odisha at the block level. By analyzing various parameters at the block cyclones in this study area, there have been thousands of fatalities,
level, we aim to provide a comprehensive understanding of cyclone risk millions of displaced people, and billions of dollars in property damage.
and vulnerability within these administrative units using the FAHP.
Specifically, this study has four main objectives: 3. Database and method

• To assess and visualize tropical cyclone risk at block level in coastal 3.1. Database
districts of Odisha using FAHP to integrate expert opinion in quan-
tifying exposure, hazard, vulnerability and mitigation capacities. In this research, we employed diverse datasets encompassing 21
• To employ the XGBoost machine learning model to predict the out- criteria to assess tropical cyclone risk, with many of these criteria being
puts of the FAHP, ensuring the model’s predictions align with expert- dynamic and subject to regular updates. We utilized various data plat-
derived assessments. This step validates the reliability and accuracy forms, such as global, regional, and the Odisha state government’s
of the risk assessment process. official websites, for our data collection. Geospatial techniques were
• To apply XAI algorithms, specifically LIME and SHAP, to elucidate employed to produce the spatial data used in assessing these criteria.
how input variables influence the predictions of the XGBoost model. Table 1 provides further details about the data origins, duration, and
This objective aims to provide transparency and comprehensibility in intended uses.
the model outcomes, facilitating unbiased and informed decision-
making for cyclone risk management. 3.2. Method for preparing the risk map to the tropical cyclone
• To leverage the insights gained from the XAI-enhanced model to
inform and improve decision-making processes, enabling targeted In general, risk map is an instrument that allows the identification of
and effective risk management strategies that are responsive to the risk factors that may occur within a natural region. Risk map is used to
specific needs and vulnerabilities of the regions within Odisha. assist in identifying, prioritizing, and quantifying risks in a geographical
settings/region. But in particular, tropical cyclone risk map is a specific
2. Study area: coastal Odisha geographic visualization tool used to represent the likelihood and po-
tential impact of tropical cyclones in a particular region. In this research,
The current study was conducted focusing on the coastal regions of we have incorporated a range of both natural and human-influenced
Odisha, an eastern state in India boasting a coastline stretching 450 km, factors to evaluate the risk associated with tropical cyclones, employ-
abundant with natural riches. This region’s geographic boundaries are ing a multi-criteria assessment technique based on FAHP. The outcome

2
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Fig. 1. Location map of the study area.

of this method can be presented in a straightforward manner, facilitating significance of each parameter in the supplementary section (Supple-
the integration and consolidation of diverse criteria [48]. Numerous mentary Material 2). Twenty-one criteria layers were formed from four
equations for assessing tropical cyclone risk have been documented in risk components, each with a cell resolution of 30 m × 30 m maps un-
academic literature [34,74]. Using a thoroughly elaborated and exten- derwent classification employing the natural break statistical technique,
sive risk formula, incorporating all its elements, improves the accuracy which is a method that identifies natural groupings of data values. This
of risk evaluation. We considered the risk Eq. (1) for this study after technique seeks to minimize the variance within each class and maxi-
conducting a thorough review of the relevant literature [1,34,46]. The mize the variance between classes, resulting in more meaningful and
detailed methodology applied in this research is given in Fig. 7. representative categories based on inherent data pattern. The analysis
( ) was conducted using R-studio, QGIS, ArcGIS, and MS Excel. The
Vi*Ei*Hi
Ri = (1) following paragraphs elaborate on the criteria’s importance and map-
Mi
ping procedures.
where Ri indicates the risk, Vi refers to vulnerability, Hi is the hazard, Ei
3.2.2. Vulnerability mapping criterion
indicates exposure, and Mi is the mitigation capacity.
In the realm of risks and calamities, the term ‘vulnerability’ pertains
The details of risk assessment criteria have been discussed below:
to the conditions that render individuals, communities, assets, or sys-
tems prone to the impacts of hazards [96]. The susceptibility to tropical
3.2.1. Risk evaluation criteria
cyclones is influenced by various factors, which serve as parameters in
In the process of risk evaluation and examination, the pivotal stages
this research. Factors such as elevation, terrain incline, proximity to
involve the careful selection of suitable criteria and elements, organizing
cyclone paths and coastlines, and land use/land cover (LULC) have been
them into diverse categories, and aligning them with the various com-
identified as criteria for evaluating vulnerability.
ponents of risk. Decision criteria should be relevant to the study’s goals,
It is widely accepted that there is an inverse relationship between
easily understood, and amenable to statistical analysis. Based on the
altitude and susceptibility to cyclones. Higher elevations are less sus-
literature review and data evaluation, criteria were selected for this
ceptible to cyclonic impacts as hills and mountains act as natural
study and mapped to create spatial layers and also mentioned the

3
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Table 1 impact of cyclones, including storm surges, high winds, and heavy
Description of datasets used and distinct characteristics. rainfall. Using the Euclidean distance tool in ArcGIS, we calculated the
Data Data sources Period Outcome distances from the study area’s shoreline. This data was then converted
into raster layers to visualize the proximity of different regions to the
Landsat 8 United States 4 Landsat Coastal vegetation
Geological Survey images were & and use land coast. Areas closer to the shoreline are generally at a higher risk of
(USGS)Earth used (28th of cover experiencing severe cyclonic effects.
Explorer April 2022) Various types of land cover face susceptibility to tropical cyclones. In
Digital Elevation Earth Explorer 2011 Elevation, slope, the creation of the land cover map, depicted in Fig. 2e, four Landsat 8
Model (DEM) website of United storm surge
at 30 m spatial States Geological
images were employed. Initially, radiometric, geometric, and atmo-
resolution Survey (USGS) spheric corrections were conducted, alongside additional pre-processing
Demographic Census of India 2011 Density of stages and modifications, on the images [24]. Then, to distinguish be-
data population, tween the various types of land cover, Random Forest technique has
dependent
been applied, which is a machine learning algorithm based on the
population, female
population, literacy ensemble learning approach and utilizes decision trees as its base
rate, agricultural learners [66]. We categorized the land use and land cover (LULC) into
dependent seven groups, which included dense and sparse vegetation, shrubs,
population, cropland, built-up areas, water bodies, and bare land. In order to vali-
disabled
population, mud-
date the precision of the land cover map, we randomly selected around
wood-thatch 500 points from the high-resolution Google Earth imageries. The
houses, detailed methodology of accuracy assessment has been outlined in
Cyclone track International Best 1971–2021 Frequency of previous studies [24,89]. The result of accuracy assessment shows a
Track Archive for cyclone and
satisfactory accuracy of LULC classification with an overall accuracy of
Climate Stewardship distance from
(IBTrACS) cyclone track about 84.32% with respective kappa coefficient of 0.88. The accuracy
Health center, Official website of 2022, 2011 Nearness to assessment table has been provided in the supplementary section
Cyclone Odisha State cyclone shelter, (Supplementary Material 4).
shelter, & Government nearness to health
cyclone ([Link] & center, and cyclone
3.2.3. Exposure mapping criterion
warning [Link]/state-ha warning system
system zard) & Census of Exposure pertains to the presence of human resources, including
India individuals, infrastructure, housing, and production capacities, in areas
Road Bhukosh (Bhukosh. 2018 Distance to road prone to hazards. Evaluating exposure is crucial for gauging potential
[Link])
impacts and integrating risk mitigation into local development strate-
Rainfall (mm) Indian 1971–2021 Average annual
Meteorological precipitation map gies. Absence of exposure equates to absence of risk; areas with exposure
Organization (IMD) are highly vulnerable to risks, resulting in significant human and eco-
Wind speed (km/ National Oceanic & 1999–2019 Wind speed map nomic damages. Therefore, exposure plays an important role in deter-
h) Atmospheric mining the severity and potential consequences of any disaster [42,71].
Administration
In this research, we identified six criteria for evaluating exposure to
(NOAA)
tropical cyclones: density of population, proportion of female popula-
tion, proportion of disabled population, proportion of dependent pop-
barriers, reducing the cyclones’ destructive power [75,81]. Regions with ulation, housing structures, and population dependent on agriculture
higher elevation experience less damage compared to low-lying areas, (Fig. 3). We employed 2011 Census data for each criterion and produced
which are more prone to flooding and wind damage. Utilizing data from raster layers using ArcGIS software (version 10.8).
the Shuttle Radar Topography Mission (SRTM) DEM, with a spatial The destructive potential of any disaster is determined by how much
resolution of 30 m, we generated elevation maps to illustrate this people and their wealth are affected by a catastrophe after it occurs [62].
distinction (Fig. 2a). This data helps us understand the distribution of The clustering of individuals, particularly in urban settings, heightens
elevation across the study area and its potential impact on cyclone the likelihood of encountering catastrophic events. Coastal areas are
vulnerability. witnessing a rapid increase in population compared to other regions
Slope significantly affects the vulnerability of areas to tropical cy- globally, raising concerns about the susceptibility of coastal commu-
clones. Steeper slopes can mitigate the impact of cyclones by disrupting nities to cyclones. Variations in population density across different
wind patterns and reducing wind speed, whereas gentler slopes do not geographical areas play a significant role in determining exposure to
offer such protection. The slope was derived from the SRTM DEM data to tropical storms [64], with densely populated regions facing greater
create slope maps (Fig. 2b). This analysis helps identify regions where exposure compared to less populated ones. Vulnerable groups such as
steep slopes may provide a natural defence against cyclonic winds and children and the elderly (see Fig. 3b) are particularly susceptible to the
associated damage. impacts of tropical cyclones as they may have difficulty following and
Areas located near the historical cyclone tracks are more vulnerable implementing emergency protocols independently during evacuations
to cyclonic events. We used data from the International Best Track [30,37]. Their limited capacity to effectively manage the situation
Archive for Climate Stewardship (IBTrACS) to analyze cyclone tracks during the initial stages of a disaster evacuation renders them highly
from 1971 to 2021, revealing 53 spatial cyclone tracks of varying cat- prone to experiencing the adverse effects of tropical cyclones.
egories (1 to 5). Buffers were created around these tracks to determine Individuals with disabilities face a higher risk of injury or death
areas of heightened vulnerability based on their proximity to these during catastrophes [12,28]. Impairments and activity restrictions
historical paths. This analysis highlights regions that are more likely to hinder their ability to plan, evacuate, and recover from disasters, putting
experience cyclonic activity based on past occurrences. Utilizing Arc- them at a disadvantage [100]. Disabled individuals are disproportion-
GIS’s Euclidean distance tool, we computed the distances from the study ately affected by poverty, lack of social support, and exclusion [91],
area’s shoreline. Subsequently, the data underwent conversion into increasing their exposure to disasters. Many rely on assistance to evac-
raster layers. uate but may not receive help or early warnings during crises [74]. In
The nearness to the coastline is a crucial factor in determining India, women are especially vulnerable due to limited literacy, which
cyclone vulnerability. Coastal areas are directly exposed to the initial restricts their mobility, access to resources, and information during

4
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Fig. 2. Vulnerability assessment criteria.

Fig. 3. Exposure assessment criteria.

5
T. Das et al. Progress in Disaster Science 23 (2024) 100357

cyclones. and frequency [8]. In analyzing cyclone hazards, the study focused on
The type of material used in building homes significantly affects their factors such as the frequency of cyclones, wind speeds in km per hour,
ability to withstand disasters. Brick buildings are more resilient to ca- storm surges measured in meters, and the average annual precipitation
lamities compared to mud, wood, bamboo, or thatch. Regions with in millimetres
many homes made of the latter materials are more prone to damage Scientists estimate the time until the next tropical cyclone hits by
from tropical storms [58]. According to census data from 2011, dwell- anaylzing the average interval between cyclone occurrences. We
ings constructed from mud, wood, bamboo, and thatch showed a higher examined historical data, including paths and impacts to create a
susceptibility compared to alternative housing structures. Therefore, the cyclone frequency map for the region from 1971 to 2021. In the past
study considered the construction materials, such as wood, bamboo, and half-century, approximately 53 cyclones have made landfall in the area.
thatch. To gauge the occurrence rates of cyclones across various regions, we
Coastal regions face significant damage to agriculture due to tropical relied on both cyclone trajectory data and insights from local commu-
cyclones. The farming sector, including livestock, fishing, horticulture, nities. This data was then converted into spatial maps within a GIS
agroforestry, and crops, are all affected by direct damage from winds, framework.
rain, and floods due to storm surges. The fields may become unusable The extent of coastal inundation and flooding is influenced by the
due to saline water and sand brought in by high tides, and crop, fish, and magnitude of storm surges. In our study, we employed a combination of
animal output can be impacted indirectly by disease and infection [98]. a model based on historical data and SRTM DEM data to delineate storm
The severe impacts of the Super Cyclones of 1999 as well as Cyclone surge elevations (Fig. 4b). By examining storm surge records spanning
Phailin in 2013 on the livelihoods of farmers and agricultural production from 1971 to 2021, we identified the highest surge elevation of 7.5 m for
along India’s eastern coast were significant. The determination of de- a 51-year return period ([22]; Gopinath et al., 2023). A decay coefficient
pendency rates on agriculture relied on a five-category classification was derived from this elevation [106,107], which, when applied in
system derived from the 2011 Census of India. conjunction with the DEM data, facilitated the creation of the final storm
surge elevation map. The potential impact of a cyclone on human set-
3.2.4. Hazard mapping criterion tlements and the environment correlates directly with its maximum
A hazard refers to a procedure, phenomenon, or action that can cause wind velocity [35]. To generate a spatial representation illustrating the
human fatalities, injuries, property damage, societal disruption, eco- dispersion of cyclone wind velocities, information from past cyclones
nomic hardship, or environmental degradation [59]. Assessing tropical such as Cyclone Fani in 2019, Cyclone Hudhud in 2014, Cyclone Titli in
cyclone hazards involves examining historical records and current 2018, Cyclone Phailin in 2013, and Super Cyclone of 1999, was utilized.
research. The probability of risk is evaluated by considering variables Wind speed data sourced from the National Oceanic and Atmospheric
such as where and when hazardous events occur, as well as their severity Administration (NOAA) was utilized to ascertain wind velocities.

Fig. 4. Hazard assessment criteria.

6
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Converting the data from knots to kilometers, a wind speed map was For non-structural measures, we considered literacy levels as a proxy
then produced through spatial analysis, as depicted in Fig. 4c. Tropical for the effectiveness of education and awareness programs. Higher lit-
cyclones often bring heavy rainfall, but the study area lacked spatially eracy rates often correlate with better understanding and implementa-
distributed average rainfall data. To generate a map depicting rainfall tion of disaster preparedness measures. The presence of early warning
intensity, the monthly precipitation records (1971–2021) provided by systems, indicated by the cyclone alert systems, significantly enhances
the Indian Meteorological Department were utilized. Employing the community preparedness and response capabilities. Coastal greenery,
kriging technique within ArcGIS, data obtained from 19 rainfall stations such as mangroves and other coastal vegetation, plays a significant role
across Odisha were interpolated to construct an annual precipitation in reducing the impact of cyclones and represents an important non-
map for the entire state. Subsequently, the designated research zone was structural, nature-based solution that enhances the protective capacity
delineated from this map (Fig. 4d). of coastal areas [55,63]. Utilizing ArcGIS’s Euclidean distance tools, we
generated proximity layers for cyclone shelters and healthcare centres.
3.2.5. Mitigation capacity mapping criterion Coastal vegetation serves as a natural barrier, offering protection
Mitigation is the process of reducing the impact of a disastrous event, against water inundation during cyclones and mitigating their effects on
and it involves three basic categories of coastal hazards: protection, individuals, assets, and the ecosystem. Additionally, they help diminish
accommodation, and retreat. Mitigation capacity involves both struc- wind velocity and endure tremors [64]. The investigation utilized NDVI
tural (engineering plans & design such as cyclone shelters, health cen- examination of Landsat 8 imagery to recognize coastal vegetation.
tres, and transport networks) and non-structural (education, outreach Essential infrastructure, including highways, overpasses, drainage sys-
initiatives, enhancing awareness, and government policies) measures tems, and alternative routes, plays a vital role in emergency response,
([37]; [1]). Six criteria were employed to evaluate the extent of miti- rehabilitation efforts, and community resilience following disasters.
gation capacities. In terms of structural measures, we included prox- Relevant road network data were sourced from the Bhukosh website
imity to cyclone shelters, healthcare facilities, and distance from ([Link]), and spatial datasets for proximity to coastal
roadways. Cyclone shelters are engineered structures specifically vegetation and roadways were generated using the Euclidean distance
designed to protect communities during cyclones, making their prox- method in ArcGIS (refer to Fig. 5d).
imity crucial for community safety. The accessibility of healthcare fa- Enhancing literacy is essential for reducing the repercussions of cy-
cilities ensures the availability of medical support and emergency care, clones, as people with better education are inclined to adhere to evac-
which, while not a direct engineering plan, are integral components of uation guidelines and implement additional safety measures. Literacy is
structural resilience. Distance from roadways captures the accessibility crucial for disaster-prone areas, as it enables people to receive and
of transport networks essential for evacuation and the distribution of interpret information about mitigating disasters [77]. To generate the
relief materials, reflecting the importance of engineered road infra- raster layer for this research, we utilized literacy information obtained
structure in disaster management. from the 2011 Census. A warning system for cyclones has the potential

Fig. 5. Mitigation capacity criteria.

7
T. Das et al. Progress in Disaster Science 23 (2024) 100357

to minimize harm and loss of life through the timely alerting of strong
winds. For our investigation, we analysed information sourced from the
2011 Census, encompassing data on television, radio, mobile phones,
and internet accessibility at the district level, alongside data on warning
systems taken from website of Odisha State Government. We applied
weighted linear composite (WLC) approaches, with a concise overview
provided in supplementary material 5. The resultant classification
comprised three tiers: effective, moderate, and ineffective.

3.3. Modeling of coastal risk assessment

3.3.1. Normalization and ranking of parameters

In this study, we used the FAHP to assess the risks of tropical cyclones
by integrating expert opinion and empirical data. The FAHP method is
particularly effective in addressing the uncertainties and imprecision
associated with expert judgment [108,109]. First, we identified and
classified 21 sub-criteria under four main criteria: Vulnerability, Expo-
sure, Hazard, and Mitigation. These sub-criteria were classified using the
natural-break algorithm, which facilitated the normalization of the data
Fig. 6. Fuzzy triangular membership function.
to a uniform scale. This standardization was crucial to ensure compa-
rability between different data types [38,50]. Despite the normalization
process, the actual classified values were retained in our plots for better Table 3
interpretability. To rank the sub-criteria by importance, we categorized The membership function associated with the linguistic scale.
each criterion into five different groups and assigned values from 1 to 5,
Linguistic Crisp Triangular fuzzy Reciprocal triangular
with higher values indicating greater risk (Table 2). This approach variable numbers numbers fuzzy numbers
allowed us to convey the inherent importance of each parameter prior to
Equally strong 1 (1,1,1) (1, 1, 1)
normalization, providing a clear understanding of their relative impor-
Moderately 3 (2, 3, 4) (1/4, 1/3, 1/2)
tance in the context of tropical cyclone risk. strong
Strong 5 (4, 5, 6) (1/6, 1/5, 1/4)
3.3.2. Application of pairwise comparison and fuzzification Very Strong 7 (6, 7, 8) (1/8, 1/7, 1/6)
The core of the FAHP method is to perform pairwise comparisons Extremely 9 (9, 9, 9) (1/9, 1/9, 1/9)
Strong
between the criteria based on expert input. A linguistic scale is used for Intermediate 2 (1,2,3) (1/3, 1/2, 1)
these comparisons, which is then converted into triangular fuzzy 4 (3, 4, 5) (1/5, 1/4, 1/3)
numbers to capture the fuzziness of human judgment [82,90]. The 6 (5, 6, 7) (1/7, 1/6, 1/5)
fuzzification process is crucial in FAHP because it translates qualitative 8 (7, 8, 9) (1/9, 1/8, 1/7)
linguistic expressions into fuzzy membership functions, including
triangular, trapezoidal, and bell-shaped functions (Fig. 6). In this study, criterion by defuzzifying the fuzzy numbers (Table 4). In this step, the
triangular membership functions were specifically used to account for fuzzy weights derived from the pairwise comparison matrices were
the subjective nature of the ratings [5,85] (Table 3). converted into unique values required to synthesize the overall risk
assessment. We used the center of area (COA) method for defuzzifica-
3.3.3. Weight assignment and defuzzification tion, which was chosen for its straightforwardness and objectivity and
Following the fuzzification process, we assigned weights to each

Table 2
Ranking the sub-criteria’s relative importance in relation to the risk of tropical cyclones.
Component Criteria Ranking (based on risk)

Very low (1) Low (2) Moderate (3) High (4) Very high (5)

Elevation (m) > 537 101–536 41–100 6–40 <5

Slope (degree) 25.95–71.12 16.74–25.94 7.82–16.73 2.24–7.81 < 2.23
Vulnerability Nearness to Cyclone track (km) > 10 7–10 4–7 2–4 <2
Nearness to Coastline (km) > 73 48–73 29–48 13–29 < 13
LULC Water bodies, Sand bar Sparse veg., Shrub Dense Veg. Cropland Built up
Population Density (population/km2) 272–396 397–527 528–637 638–735 736–869
Dependent Population (%) 21.83–24.82 24.83–26.52 26.53–29.98 29.99–33.44 33.45–39.15
Female Population (%) 39.34–44.23 44.24–47.59 47.6–49.49 49.5–50.63 50.64–53.18
Exposure
Disable Population (%) 1.43–2.32 2.33–2.77 2.78–3.12 3.13–3.65 3.66–4.35
Mud, Wood, and Thatch Houses (%) 13.58–23.12 23.13–36.21 36.22–54.2 54.21–69.2 69.21–83.1
Agriculture Dependent Population (%) 38.48–53.87 53.88–63.72 63.73–70.47 70.48–75.48 75.49–82.57
Cyclone Frequency 0–1 1–3 3–5 5–8 8–11
Storm Surge (m) Unaffected Unaffected <3 3–6 6–10
Hazard
Wind Speed (km/h) 116–131 132–160 161–192 193–220 221–260
Average Annual Precipitation (mm) 1220 - 1347 1348 - 1478 1479 - 1601 1602–1742 1743 - 2096
Nearness to Cyclone Shelter (km) <2 2–4 4–7 7–10 > 10
Nearness to Health Centre (km) <2 2–4 4–7 7–10 > 10
Nearness to Coastal Vegetation (km) <5 5–15 15–30 30–50 > 50
Mitigation Capacity
Distance to Road (km) <1 1–2 2–3 3–4 >4
Cyclone Warning System – Effective Moderate Ineffective –
Literacy (%) 85.61–88.78 82.91–85.6 76.54–82.9 66.5–76.53 57.55–66.49

8
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Fig. 7. Flowchart outlining the methodology employed in the study.

corresponds to the approach described by Feizizadeh et al. [27]. The resulting risk index values were normalized on a scale of 0 to 1 and
detrended weights were then applied to the normalized criteria to classified into five risk categories: extremely high, high, medium, low,
calculate the overall risk score for each block in the study area. This and extremely low. This categorization process was facilitated by the
ensured that the weights accurately reflected the relative importance of utilization of Eq. (1) for risk evaluation.
each criterion as determined by expert judgment.
To check the consistency of the expert judgments, we calculated the 3.3.5. Validation of the models
consistency ratio (CR) for each pairwise comparison matrix. A CR value This study employed the Receiver Operating Characteristics (ROC)
of 0.1 or less was considered acceptable and indicates reliable judgments curve to authenticate the cyclone risk map. The ROC curve, depicted
[110,111]. This consistency check is crucial to ensure the robustness and graphically, illustrates the relationship between the true positive rate
credibility of the FAHP analysis. (TPR) and the false positive rate (FPR) across various potential diag-
nostic test thresholds. To determine how well the fit is, the area received
3.3.4. Creation of risk through the integration of all major criteria under the curve (AUC) calculation has been employed [72]. Further-
Overlay analysis was employed utilizing spatial layers and their more, feedback was collected from 112 individuals residing in the
associated weights to produce indices for hazard, vulnerability, expo- designated study region. The selection of individuals was based on their
sure, and mitigation capability. The risk index was derived through the knowledge and experience with cyclone impacts. The feedback focused
multiplication of the vulnerability, exposure, and hazard indices, then on verifying whether the respondents’ locations were truly experiencing
divided by the mitigation capacity index using raster calculator. The high or low cyclone risk as indicated by our model. They were asked

9
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Fig. 8. Spatial pattern of major four criteria for tropical cyclone risk model.

severe cyclone impacts were categorized as high-risk, while those with

minimal or no reported impacts were categorized as low-risk. Further,
comprehensive first-hand observations were conducted to precisely
ascertain the elevations of different areas depicted on the map and
confirm their susceptibility to tropical cyclones. By integrating input
from local residents and overlaying pertinent parameters, a total of 120
validation points were identified to create the inventory map. Using this
inventory map as a reference, data points were extracted from the
resulting cyclone vulnerability map at corresponding coordinates.
Following this, a curve analysis was performed using R Studio software
(Version 4.2.0).

3.4. Improvement in decision-making for managing coastal risk

assessment through the implementation of XAI

In our coastal risk assessment study, we employ an XGBoost model

enhanced by XAI techniques, in particular LIME and SHAP, to increase
the transparency and reliability of predictive models, which are crucial
for effective decision making in coastal risk management. Initially, risk
assessment is performed using FAHP, which compensates for potential
biases associated with expert weighting that may include subjective
judgments. Directly using an ML model for risk assessment is not
possible as there is no direct target variable for risk parameters, which
Fig. 9. Tropical cyclone risk model. makes it difficult to understand the influence of the parameters.
Therefore, we use FAHP to generate proxy target variables, which can
then be predicted with an ML model based on similar parameters used in
about their experience with cyclones, including the frequency and
FAHP modeling. By accurately predicting the FAHP results, we can
severity of impacts, as well as any observed damage to property and
understand the behaviour of the parameters without potential bias,
infrastructure. Finally, these feedbacks were converted into quantitative
which increases reliability and enables accurate management strategies.
values by categorizing their responses into high-risk or low-risk zones.
The first phase of our methodology involves the careful preparation
For instance, regions frequently reported by respondents as experiencing
of the data set that serves as the basis for the predictive models. The

10
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Fig. 10. The area under different levels of risk in every block.

generalizability of the model are evaluated using the validation set to

ensure that the model does not overfit the training data. The efficiency of
the model is measured using performance metrics such as accuracy,
precision, recall and the area under the Receiver Operating Character-
istic Curve (AUC-ROC). These metrics provide quantitative insight into
the model’s ability to apply learned patterns to new, unseen data, con-
firming its applicability in real-world scenarios.
After confirming the predictive accuracy of the model, LIME is
applied to improve the interpretability of the model by locally approx-
imating the global XGBoost model with simpler models to clarify the
logic behind certain predictions. This local interpretation is particularly
valuable in clarifying the reasons for categorizing certain areas as high
or low risk, which is helpful in developing targeted risk management
strategies. In addition, SHAP (SHapley Additive exPlanations) values are
used to quantify the impact of each predictor on the forecast results.
These values, derived from game theory, provide precise quantitative
Fig. 11. Validation of the risk map using the ROC curve. insights into the influence of each feature, and allow stakeholders to
identify and prioritize the most important factors in coastal risk
target variable derived from the FAHP results quantifies coastal risks assessment. By integrating XGBoost with LIME and SHAP, the approach
based on expert judgment of various risk factors. To simplify the not only improves the accuracy and efficiency of risk predictions, but
decision-making process and to be consistent with the binary nature of also builds stakeholder confidence and support for risk management
the XGBoost model, these risk values are divided into two classes: high strategies by making the decision-making process transparent and un-
risk and low risk, coded as binary values (1 for high risk and 0 for low derstandable. This integration effectively mitigates the potential biases
risk). The dataset comprises 21 predictors, including environmental, of FAHP and provides a more reliable and unbiased analysis that is
socio-economic, and infrastructural parameters, which represent a conducive to accurate coastal risk management.
comprehensive set of factors influencing coastal risk. The dataset is then
split into two subsets: 70% for the training set and 30% for the validation 4. Results
set to ensure robust learning and effective validation.
XGBoost, or Extreme Gradient Boosting, is used due to its advanced 4.1. Vulnerability assessment parameters
algorithmic capabilities to optimize computational speed and perfor-
mance across different datasets. The model uses an ensemble of decision In this research, we have selected five parameters (Fig. 2) and clas-
trees that are built sequentially, with each tree correcting the errors of its sified them into five categories to determine vulnerability. Based on our
predecessors through boosting. This iterative process gradually refines analysis (Fig. 2a & b), we have identified the Basudevpur, Tihidi, and
the accuracy of the model as each tree learns from the residuals of its Chandibali blocks of the Bhadrak district; most of the blocks of Ken-
predecessors. In the training phase, the XGBoost algorithm identifies and drapara and Kujang, Ersama, and Balikuda blocks of Jagatsinghpur
distinguishes high and low risk zones by recognizing complex patterns district are located at very low elevations and slopes. Consequently,
and correlations between predictors and tuning the decision trees to the these regions face significant susceptibility to tropical cyclones because
subtleties of the dataset. After training, the robustness and of their coastal geography, which is characterized by low elevation.

11
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Fig. 12. Determine the influence of different parameters on the risk of tropical cyclones in very high-risk blocks.

Fig. 13. Determine the influence of different parameters on the risk of tropical cyclone in low-risk blocks.

12
T. Das et al. Progress in Disaster Science 23 (2024) 100357

Table 4 requirements. This study found from Fig. 3d that Jaleswar and Bhograi
Weights assigned to criteria obtained from the pair-wise comparison matrices. blocks in the Baleshwar district, Kujang, and Tirtol blocks in the
Component Criteria Weight Jagatsinghpur district, and Kakatpur, Nimapada, Pipili, and Kanas
blocks in the Puri district have a higher proportion of the disabled
Vulnerability Elevation (m) 0.197
Slope (degree) 0.198 population. On the other hand, the disabled population is lowest in most
Nearness to Cyclone track (km) 0.19 of the blocks of Ganjam, Kendrapara, and western parts of the Bhadrak
Nearness to Coastline (km) 0.177 districts.
LULC 0.245 We also consider the proportion of houses made of mud, wood, and
Exposure Population Density (population/km2) 0.219
Dependent Population (%) 0.162
thatch (Fig. 3e), as well as the population that depends on agriculture
Female Population (%) 0.12 (Fig. 3f), to determine the exposure to tropical cyclones. A higher pro-
Disable Population (%) 0.107 portion of these parameters increases the exposure of any given region.
Mud, Wood, and Thatch Houses (%) 0.186 The northern regions of the surveyed zone, comprising the entirety of
Agriculture Dependent Population (%) 0.206
the Baleshwar, Bhadrak, and Kendrapara districts, exhibit a greater
Hazard Cyclone Frequency 0.15
Storm Surge (m) 0.45 prevalence of dwellings constructed from mud, wood, and thatch. This
Wind Speed (km/h) 0.3 suggests a relatively lower economic status within these districts. In
Average Annual Precipitation (mm) 0.1 contrast, most of the blocks of the Ganjam district have a low concen-
Mitigation capacity Nearness to Cyclone Shelter (km) 0.208 tration of such houses. In many blocks of the Baleshwar and Bhadrak
Nearness to Health Centre (km) 0.22
Nearness to Coastal Vegetation (km) 0.166
districts, as well as in the Surada, Sankhemundi, Digpahandi, Patrapur,
Distance to Road (km) 0.12 and Dharakote blocks of the Ganjam district, a significant portion of the
Cyclone Warning System 0.18 population relies on agriculture. Conversely, the majority of the Jagat-
Literacy (%) 0.106 singhpur district and the southern blocks of Ganjam district exhibit a
diminished dependency on agriculture among their populace.
In blocks where, previous cyclone tracks are more common, those
4.3. Hazard assessment parameters
areas are more vulnerable. Our study found that nearly all the blocks of
the Baleshwar district (Fig. 2c); Bant, and Basudevpur blocks of the
To assess the hazard related to tropical cyclones, we have chosen
Bhadrak district; the southern blocks (Derabish, Garadpur, Maha-
four parameters as mentioned in Fig. 4. The frequency of cyclones is very
kalpada, and Marsaghai) of Kendrapara districts; and Raghunathpur,
high in the northern blocks of the Baleshwar district, including Nilahiri,
Tritol, Kujang, and Biridi blocks of Jagatsinghpur district have the
Remuna, Basta, Baliapal, Jaleswar, and Bhograi, as well as in the Bau-
highest records of past cyclone tracks, making them particularly
devpur, Bhadrak, and Bant blocks of the Bhadrak district, and the
vulnerable. Certain land cover types exhibit greater vulnerability to
Derabish and Mahakalpada blocks of the Kendrapara district. In
tropical cyclone damage. This study involved dividing the land use and
contrast, most of the blocks in the Ganjam, Puri, Jagatsinghpur, and Puri
land cover map into seven distinct categories (Fig. 2e), considering the
districts have a low tropical cyclone frequency. We have classified the
fact that built-up areas and agricultural land are particularly vulnerable,
storm surge parameters (Fig. 4b) into four classes, and we have observed
while water bodies and sandbars are less vulnerable. Almost the entire
that most of the portions of the Ganjam and western parts of the
study area except for the mountainous portion of the Ganjam and
Baleshwar districts are unaffected due to high elevation. However, the
Baleshwar districts is highly vulnerable due to the higher concentration
Basta, Baliapal, and Bhograi regions within the Baleshwar district,
of built-up areas and cropland.
alongside the eastern and central areas of the Bhadrak district, along
with the entirety of Kendrapara, Puri, and Jagatsinghpur districts,
4.2. Exposure assessment parameters excluding the Biridi and Raghunathpur blocks, are classified as areas
highly susceptible to storm surges.
In this research on assessing exposure to tropical cyclones, we have In this research, it was found from Fig. 4c that all blocks of the
selected six parameters (Fig. 3), each of which has been classified into Kendrapara, Jagatsinghpur, Puri, and Ganjam districts are in a zone of
five categories. It appears that the population density is particularly high very high wind speeds. In contrast, all blocks of the Baleshwar district
(Fig. 3a) in the Basta, Baliapal, Bhograi, and Remuna blocks of the are in a low wind-speed zone. The annual average precipitation is
Baleshwar district, Derabish, Garadpur, and Marsaghai blocks of Ken- observed to be very high in the Baleshwar, Bhadrak, Kendrapara as well
drapara district, Biridi, Kujang, and Tritol blocks of Jagatsinghpur dis- as Jagatsinghpur districts, while it is very low in all blocks of the Ganjam
trict, Nimapada, Pipili, and Kakatpur blocks of Puri district, as well as district and the western blocks of the Puri district (Fig. 4d).
Sheraga, Kabisuryanagar, and Hinjilicut blocks of Ganjam district.
Therefore, these blocks have a higher exposure to tropical cyclones. On 4.4. Mitigation capacity assessment parameters
the other hand, the lowest density can be found in Mayurbhanj,
Keonjhar, Dhenkanal, and Nayagarh districts. The presence of a To evaluate the mitigation capacity in this study, we have selected
dependent population can increase a region’s vulnerability to natural six parameters (as shown in Fig. 5). The concentration of cyclone shel-
disasters. In this case, it appears from Fig. 3b that almost all blocks in the ters (Fig. 5a) is found to be very high in the Kendrapara district, the
Baleshwar district, except for Bhograi and Remuna, and nearly all blocks northern blocks of the Baleshwar districts, the central parts (central part
in the Ganjam district, except Kukudakhandi and Rangeilunda, have a of the coastal plain), the eastern blocks (Astaranga & Kakatpur), as well
higher concentration of dependent population. Meanwhile, Jagat- as the southern blocks of the Ganjam district. In contrast, the Jagat-
singhpur district and most parts of Puri district have a low concentration singhpur district has a very low number of cyclone shelters. The con-
of dependent population. centration of health centers (Fig. 5b) is observed to be very high
In the Bhadrak district, a higher concentration of female population throughout the entire study area, except for the Basudevpur block of the
(Fig. 3c) can be found in Basudevpur, Tihidi, Dhamnagar, and Bhandari Bhadrak district, Ersama block of Jagatsinghpur, Nimapada, Kanas, Gop,
Pokhari blocks. Similarly, in Kendrapara district, this can be seen in Krushnaprasad blocks of the Puri district, and some blocks of the Ganjam
Rajkanika, Ali, and Pattamundai blocks, as well as all eastern blocks and district.
Hinjilicut, Patrapur, and Digpahandi blocks in the western part of The road network concentration (Fig. 5d) in the entire study area is
Ganjam district. Additionally, disabled populations can increase a re- very good, although we have only considered national and state high-
gion’s vulnerability to hazards due to their specific needs and ways. A robust warning system is effective in reducing the immediate

13
T. Das et al. Progress in Disaster Science 23 (2024) 100357

impacts of tropical cyclones by facilitating communication and coordi- 4.5.3. Assessing tropical cyclone hazard
nation efforts between government agencies and the public. The Bha- We prepared the spatial pattern of the tropical cyclone hazard map
drak, Kendrapara, and Jagatsinghpur districts have a strong warning depicted in Fig. 8c based on four selected parameters. The hazard map
system, while the rest of the study area is covered by a moderate (Fig. 8c) illustrates that approximately 30% (6515 km2) of the total land
warning system (Fig. 5e). However, the Surada and Sanakhemundi area falls within the high-risk zones, predominantly encompassing the
blocks of the Ganjam district have an ineffective warning system. central regions of the Kendrapara (Derabish, Garadpur, Marsaghai,
Literacy rates can help reduce the impacts of disasters by increasing Mahakalpada, Rajkanika blocks), Jagatsinghpur (Tirtol, Kujang,
awareness, improving information dissemination, and promoting better Ersama, Balikuda blocks) and eastern parts of the Puri districts (Kakat-
decision-making. Almost all the blocks in Kendrapara, Jagatsinghpur, pur, Puri Sadar, Delanga, Gop, Brahmagiri blocks). The Basudevpur
and Puri districts, as well as Bant, Bhandari Pokhari, Dhamnagar, and block in Bhadrak and the northern blocks of Baleshwar districts are
Chandibali blocks in Bhadrak district, have a very high literacy rate. identified as highly hazard-prone areas. These regions exhibit significant
However, the entire Ganjam district and the Nilahiri, Oupada, Remuna, hazard intensity, attributed to factors such as heavy rainfall, low
Baliapal, and Jaleswar blocks of Baleshwar district comes under the very elevation, frequent cyclonic activity, and their proximity to coastal
low literacy zone (Fig. 5f). rivers and water bodies, rendering them vulnerable to severe storm
surges. Conversely, the southern (coastal areas bordering Andhra Pra-
4.5. Spatial pattern of tropical cyclone coastal risk model desh) and northeastern parts (areas along the border of West Bengal) of
the study area are predominantly categorized as low to very low haz-
4.5.1. Assessing vulnerability to tropical cyclone ardous zones. This designation encompasses most of the Ganjam district,
The entirety of this research area faces significant susceptibility to as well as the western blocks of Baleshwar (Oupada, Khaira, Soro,
tropical cyclones, particularly in the blocks adjacent to the shoreline, as Nilahiri, Similia, Remuna blocks), and the Bhadrak district (Bant,
illustrated in Fig. 8a. Conversely, the southwestern regions of the study Bhandari Pokhari blocks). Furthermore, the zones encompass the Biridi,
area exhibit reduced vulnerability, attributed to the presence of frag- Raghunathpur, and Jagatsinghpur blocks within the Jagatsinghpur
mented sections of the Eastern Ghats mountains. The study reveals that districts, constituting approximately 55% of the surveyed region. These
approximately 31.24% (6901.55 km2) and 34.43% (7606.28 km2) of regions were considered less vulnerable to hazards due to their reduced
the total area are categorized as very high and highly vulnerable zones, likelihood of storm surges, heavy rainfall, and strong wind velocities.
respectively. The majority of the blocks in Baleshwar, Bhadrak, Ken-
drapara, and Jagatsinghpur districts exhibit high to very high vulnera- 4.5.4. Assessing the mitigation capacity for tropical cyclone
bility to tropical cyclones, with exceptions in specific areas such as the Fig. 8d shows the spatial pattern of mitigation capacity which has
southern parts of the Nilahiri block and the northern parts of the Oupada been classified into five categories. Mitigation capacity indicates the
block in Baleshwar district, as well as the western and central parts of bounce-back capabilities of a particular region after any hazardous
Aali and the northern parts of Rajnagar blocks in Kendrapara district. event like a tropical cyclone. In this research, we found that 30%
The key regulating element for the high vulnerability in this region is the (6885.61 km2) area falls under high to very high mitigation capacity and
low elevation, flat slope, numerous cyclone tracks, closeness to the is located mainly in the coastal and central parts especially Rajkanika,
coastline, and calamitous LULC classes. Nearly 16.25% (3589.95 km2) Aali, Pattamundai, Mahakalpada blocks of Kendrapara district; Kujang,
area of this study area is covered by a moderate tropical cyclone Balikuda, Naugaon blocks of Jagatsinghpur district; Astaranga, Kakat-
vulnerability zone and its comprised western parts of Puri (Krushnap- pur, and Puri Sadar blocks of Puri district; and the eastern blocks of
rasad block) and Ganjam district (Jagannathprasad, Aska, Bellaguntha, Baleshwar district due to the presence of several cyclones shelter, health
Buguda blocks). Almost 4000 km2 (18%) are in this study area under a care center, a reliable cyclone warning system, a well-connected trans-
low to very low vulnerable zone which is mostly covered the northern portation network, as well as high literacy rate. On the other hand,
part of the Ganjam district. This phenomenon occurs due to the region’s almost 41% (9150.5 km2) area constitutes very-low and low mitigation
elevated altitude, distant position from the coast, limited frequency of capacity zone and it comprised nearly the entire Ganjam district except
cyclone paths, and lower density of urban development. for the southern blocks, eastern blocks of Puri, Bhadrak, and Baleshwar
districts. Although compared to other Indian coastal areas, the area
4.5.2. Assessing exposure to tropical cyclone relatively has overall good mitigation capabilities.
This research illustrates the spatial pattern of tropical cyclone
exposure, as depicted in Fig. 8b. The exposure map was created by 4.5.5. Assessment of tropical cyclone risk model
incorporating the weighting factors (referenced in Table 4) for the six Fig. 9 shows the spatial pattern of the tropical cyclone risk in the
parameters outlined in Fig. 4, which were then categorized into five study area. The resulting final risk map shows that the very high and
levels ranging from exceptionally high to exceptionally low. Approxi- high-risk zone covered 18.64% (4118 km2) and 19% (4200 km2)
mately 44% of the study area comprises zones categorized as high to respectively. These zones cover mainly the northeastern and south-
exceptionally high exposure levels. Most of the blocks in the north- eastern parts (area spanning from the Puri district to the border of
eastern part of the study area mainly in the Baleshwar (Jaleswar, Andhra Pradesh). The current study shows that the Jaleswar, Basta, and
Bhograi, Baliapal, Basta, Oupada, Soro, and Similia), followed by Bha- Baliapal blocks in the Balasore district; Basudevpur, Tihidi, Dhamnagar,
drak (Basudevpur, Tihidi, and Dhamnagar), Kendrapara (Rajnagar, and Banta blocks in the Bhadrak district; Rajnagar, Mahakalpada,
Mahakalpada, Rajkanika, and Derabish), and Ganjam (Sanakhemundi, Garadpur, and Derabish blocks in the Kendrapara district; Nimapada
Kabisuryanagar, Polasara, Jagannathprasad, Khallikote) district which and Kanas blocks in the Puri district, along with predominantly the
is located in southwestern parts are very highly exposed. The above- northeastern and western blocks of the Ganjam district, fall within the
mentioned region is highly susceptible because of its high population high to very high tropical cyclone risk zone. These regions are highly
density, high dependency rate, large percentage of women, high agri- vulnerable, exposed, and hazardous, with poor mitigating capacity. As a
cultural dependency rate, and poor quality of house construction. result, the risk of tropical cyclones in these areas is very high. The
Almost 31% area of this research region is covered by a moderate research area comprised nearly 42% (9150 km2) of the total area under
exposure zone and most of it lies in the northern and western blocks of the low to very low tropical cyclone risk zone. The southern areas of the
the Ganjam district, northern blocks of Puri, Aali, Marsaghai, Garadpur Ganjam and Puri districts, along with the central parts of the Balasore
blocks of Kendrapara districts, and two blocks of Bhadrak districts. and Kendrapara districts, are classified as low to very low risk zones due
to their low vulnerability, minimal exposure to hazards, and relatively
high level of mitigation capacity. We also determine the level of risk for

14
T. Das et al. Progress in Disaster Science 23 (2024) 100357

every block and it found from Fig. 10 that some blocks like Jaleswar, effective warning systems, reducing their vulnerability to tropical cy-
Baliapal, Basudevpur, Mahakalpada, Derabish, Nimapada, Kanas, and clones. Overall, our XAI approach, integrating the XGBoost model and
Jagannathprasad have >70% of these blocks area comes under High to SHAP values, allowed us to identify the critical parameters influencing
very high tropical cyclone risk zone. It also noticed that out of 69 blocks, coastal risk at the block scale. These insights can significantly contribute
22 blocks are under high-risk zone and some blocks are under low-risk to informed decision-making in coastal risk assessment and help prior-
zones such as Remuna, Nilahiri, Bahanga, Similia, Bhadrak, Aali, Bir- itize targeted interventions for risk reduction and disaster management
idi, Naugaon, Astaranga, Kakatpur, Brahmgiri, Chatrapur, Aska, and in vulnerable areas.
Chikiti. These blocks have 65 to 80% area under a low tropical cyclone
risk zone. 4.6.3. Management strategies
Based on our study’s findings and the application of Explainable
4.5.6. Validation of tropical cyclone risk model Artificial Intelligence (XAI) through the XGBoost model, we propose
The illustration in Fig. 11 displays the performance of the model specific management strategies for coastal risk assessment at the block
utilized in this research, using a prediction rate curve. The AUC (Area scale:
Under the Curve) of the prediction rate for the FAHP model, standing at
0.91, signifies that it accounts for 91.1% of the predictive accuracy of 1. Targeted Infrastructure Improvement: Our analysis revealed that
the applied model’s AUC. As AUC values range from 0 to 1, closer ex- the presence and quality of essential infrastructure significantly in-
amination of values nearing 1 reveals heightened accuracy, as high- fluence tropical cyclone risk. Therefore, a priority should be placed
lighted by Das and DSouza [23]. Typically, AUC values exceeding 0.80 on improving critical infrastructure in high-risk blocks. This includes
indicate the reliability of a model’s output. Thus, the AUC value of constructing more cyclone shelters and health centers, particularly in
91.1% obtained from the analysis underscores the effectiveness of our areas like Raghunathpur, Basudevpur, and Kanas, where these fac-
devised approach in evaluating cyclone risk. tors positively influenced risk. Upgrading existing infrastructure to
withstand cyclonic forces and ensuring its strategic distribution
4.6. Implementation of XAI through XGBoost model to improve decision- across vulnerable blocks will enhance the community’s resilience
making at block scale for managing coastal risk assessment during cyclone events.
2. Economic Diversification: Blocks with a high dependence on agri-
4.6.1. Implementation of XGBoost model culture, like Raghunathpur, Kanas, and Basudevpur, exhibit
The XGBoost model was then implemented as a classification prob- increased vulnerability to cyclone impacts. Encouraging economic
lem in Python. We successfully achieved a high accuracy of 0.88, indi- diversification and providing training for alternative income sources
cating the model’s excellent performance in predicting coastal risk can help reduce the reliance on vulnerable sectors and increase the
levels. However, accuracy alone does not provide insights into the un- community’s overall resilience. Initiatives that promote sustainable
derlying factors contributing to risk. livelihood options, like eco-tourism or other non-agricultural activ-
ities, can play a significant role in this regard.
4.6.2. Computation of SHAP in the form of XAI for improving decision 3. Enhancing Early Warning Systems: A robust and effective early
making warning system is critical for timely evacuation and preparedness.
To gain deeper insights into the model and the decision-making Prioritizing investments in advanced technologies and communica-
process, we extracted SHAP values from the XGBoost based LIME tion channels to disseminate warnings to the community in blocks
model. SHAP values helped us understand the importance of various with high influences from the cyclone warning system (e.g.,
parameters in influencing the tropical cyclone risk for each block. In our Raghunathpur and Basudevpur) can significantly reduce the loss of
analysis, we focused on six blocks from the very high-risk zone lives and assets during cyclone events.
(Raghunathpur, Basudevpur, Kanas, Kabisuryanagar, Nimapada, and 4. Improving Housing Standards: Blocks with a prevalence of poorly
Jagannathprasad) and six blocks from the low-risk zone (Remuna, constructed houses, like Raghunathpur, are at higher risk during
Chandibali, Biridi, Astaranga, Ganjam, and Kukudakhandi) depicted in cyclones. Implementing building codes and standards to improve the
Figs. 12 & 13 respectively. Our findings revealed that certain parameters structural integrity of houses can bolster resilience and reduce po-
had positive influences on the tropical cyclone risk, while others had tential damage. Housing improvement programs, especially for
negative influences. A positive influence meant that these parameters economically disadvantaged households, should be prioritized.
played a significant role in elevating the risk level for a specific block. 5. Protecting Coastal Ecosystems: Coastal ecosystems such as man-
Consequently, controlling these influential parameters could lead to risk groves and coral reefs play a vital role in buffering storm surges and
reduction in that block. For instance, in the Raghunathpur block, we protecting coastal communities. Blocks like Nimapada, Jagannath-
observed that the presence of a health center, poor-quality houses, an prasad, and Astaranga, located in the very high-risk zone, can benefit
agricultural-dependent population, and a cyclone warning system from restoration and conservation efforts to enhance their resilience
positively influenced the tropical cyclone risk. By improving the number against cyclones.
of health centers, constructing better houses, enhancing the warning 6. Community Capacity Building: Building strong community net-
system, and addressing agricultural dependence, we could potentially works and capacity for disaster preparedness and response is crucial.
reduce the risk in this block. Similarly, in the Basudevpur and Kanas Community-based training programs and awareness campaigns can
blocks, storm surges, agricultural-dependent populations, and cyclone empower residents in vulnerable blocks to take appropriate actions
warning systems exhibited high influences on tropical cyclone risk. In during cyclone events, reducing the overall impact.
Kanas block, however, cyclone shelters played a positive role in influ- 7. Integrating XAI in Decision-Making: The integration of XAI tech-
encing the risk, suggesting that increasing the number of cyclone shel- niques like SHAP values into risk assessments can provide policy-
ters might reduce the risk in these blocks (Fig. 12). makers and stakeholders with valuable insights into the most
Conversely, the blocks like Remuna, Chandibali, Biridi, Astaranga, influential parameters at the block scale. Encouraging the adoption
Ganjam, and Kukudakhandi were categorized under the very low-risk of XAI in other coastal risk assessment studies can lead to better-
zone due to factors such as sufficient cyclone shelters, health centers, informed decision-making and more targeted interventions for risk
relatively fewer poorly constructed houses, low dependence on agri- reduction.
culture, high literacy rates, and effective warning systems (Fig. 13).
Despite their proximity to the shoreline, these blocks displayed low-risk By implementing these management strategies, coastal communities
levels, primarily due to the presence of essential infrastructure and in high-risk blocks can enhance their resilience, reduce vulnerability,

15
T. Das et al. Progress in Disaster Science 23 (2024) 100357

and mitigate the potential impacts of tropical cyclones. However, it is resulting from climate change and rapid coastal population growth ne-
essential to engage stakeholders, local communities, and government cessitates the implementation of various measures to enhance commu-
agencies in collaborative efforts to ensure the successful implementation nity resilience. These measures include improving the early warning
of these strategies and the sustainable development of coastal regions. system by leveraging advanced technologies, fortifying critical infra-
structure, fostering community networks, encouraging sustainable
5. Discussion livelihoods, rehabilitating, and safeguarding coastal ecosystems, and
enforcing resilient community development policies [74,80]. By inte-
Coastal Odisha frequently experiences tropical cyclones, making it a grating these strategies, policymakers can effectively mitigate the im-
high-risk zone. However, limited research has focused on this region. In pacts of tropical cyclones and safeguard vulnerable coastal
this study, we assess vulnerability, exposure, hazard, and mitigation communities.
capacity separately and integrate these criteria to produce a tropical While our study offers valuable insights, it’s crucial to acknowledge
cyclone risk map at the block level. Unlike previous studies that used the its limitations. Our analysis relies on available data up to the time of the
Analytic Hierarchy Process (AHP) ([34]; [1,33]; [58,60]), we employed study, meaning that future changes such as climate variability and shifts
the Fuzzy Analytic Hierarchy Process (FAHP) for its flexibility and in land-use patterns could potentially alter the risk landscape. Moreover,
improved accuracy with linguistic variables and fuzzy numbers [51,93]. the lack of access to 2021 census data due to the COVID-19 pandemic led
Furthermore, we used Explainable Artificial Intelligence (XAI) to us to rely on census data from 2011, collected a decade earlier. Addi-
determine the importance of parameters at the block level, enabling tionally, the absence of data on tropical cyclone propagation speed
more informed decisions and effective disaster management. prevented its inclusion as a criterion in hazard mapping. In this research,
Our findings indicate that about 62% of the surveyed region is we utilized the Natural Break classification system, a statistical method
categorized as having high to very high vulnerability to tropical cy- that identifies natural groupings of data values to optimize the differ-
clones. This aligns with the findings of Sahoo and Bhaskaran [74], who entiation between categories. We also utilized a single classification
also identified high vulnerability in certain areas of Odisha. Addition- system (Natural Break) for all the different parameters. Every aspect of
ally, 43% of the overall area consists of blocks highly exposed to cy- the research criteria was impacted by both past events and present sit-
clones, predominantly clustered in the northeastern part of the region, uations. Looking ahead to potential changes in the climate, the
similar to the results of Mishra et al. [52], who identified highly exposed increasing sea levels become a pivotal consideration. However, the
areas in Jagatsinghpur and Puri districts. utilization of an outdated medium-resolution census dataset in our study
Our study innovatively distinguishes vulnerability and exposure diminishes overall confidence in the results. Additionally, the absence of
levels for tropical cyclone risk assessment, providing clearer insights comprehensive details regarding the progress of specific healthcare fa-
compared to prior studies that merged these factors [1,33,64]. Further cilities and cyclone shelters, along with irregular updates on pertinent
spatial analysis reveals that the southern blocks of Kendrapara, the data from the Odisha state government’s official website, implies the
entire Jagatsinghpur, and the eastern blocks of Puri district are situated potential omission of recent information. Furthermore, inherent un-
in highly hazardous zones, encompassing nearly 30% of the total area. certainties in modeling approaches, data quality, and underlying as-
This finding is slightly different from Boragapu et al. [9], who found that sumptions may impact the accuracy of the risk map. Additionally, our
the southern parts of Kendrapara were not entirely hazardous. Our study overlooks the dynamic nature of vulnerability and exposure,
analysis at the block level elucidates these differences. which can evolve over time due to socioeconomic changes and policy
Sahoo and Bhaskaran [75] used a multi-criteria decision-making interventions. This omission may restrict the long-term applicability of
approach to assess the susceptibility of Odisha’s coastal areas to tropical our risk assessment.
cyclones, incorporating hazard, exposure, vulnerability, and mitigation Our study paves the way for future research endeavors and encour-
capacity. Our study concurs with their conclusion that a significant ages scholars to delve deeper into enhancing tropical cyclone risk as-
portion of Odisha’s coastal region is at risk from tropical cyclones. sessments. Researchers may explore integrating advanced data analysis
Additionally, we determined the importance of each parameter at the techniques and machine learning algorithms to improve the precision of
block level, enhancing our understanding of local risk factors. The AUC risk predictions. Moreover, investigating the potential ramifications of
(Area Under the Curve) of the prediction rate for the FAHP model, climate change scenarios on cyclone risk could offer valuable insights
standing at 0.91, signifies that it accounts for 91.1% of the predictive into forthcoming vulnerability trends. There is also an opportunity to
accuracy of the applied model’s AUC. As AUC values range from 0 to 1, examine the influence of socio-cultural factors on shaping communities’
closer examination of values nearing 1 reveals heightened accuracy, as resilience to cyclone impacts. Additionally, research on the efficacy of
highlighted by Das and DSouza [23]. Typically, AUC values exceeding various risk reduction and mitigation strategies, particularly nature-
0.80 indicate the reliability of a model’s output [26,36]. Thus, the AUC based solutions like coastal ecosystem restoration, can contribute to
value of 91.1% obtained from the analysis underscores the effectiveness the development of resilient communities. Furthermore, the incorpora-
of our devised approach in evaluating cyclone risk. tion of XAI in risk assessments presents promising avenues. Exploring
A novel aspect of this study involves incorporating mitigation ca- alternative XAI methods and their applications in disaster risk man-
pacity into the risk evaluation process, which demonstrates the resil- agement can aid policymakers in better comprehending complex models
ience of the community [11,103]. This approach diverges from previous and instilling trust in the decision-making process.
studies by Sahoo and Bhaskaran [74] and Sarkhel et al. [80], which did Overall, we encourage researchers to adopt a multi-dimensional
not consider mitigation capacity and were conducted at the district approach, considering both natural and social factors, in conducting
level. By incorporating literacy rate as an additional parameter, our risk assessments. Collaborative efforts involving academia, government
study follows the framework of Hoque et al. [33] but provides a more agencies, and local communities can lead to more comprehensive and
comprehensive evaluation of risk. The elevated tropical cyclone risk effective strategies for disaster management and risk reduction, ulti-
observed in certain areas is largely attributed to factors such as limited mately enhancing the resilience of coastal regions facing tropical
availability of cyclone shelters and health centers, an ineffective warn- cyclones.
ing system, heavy reliance on agriculture, high population density, and
frequent cyclone occurrences. Key strategies for enhancing coastal risk 6. Conclusion
management involve improving the early warning system, bolstering
critical infrastructure, and promoting community resilience through In our investigation, we conducted a thorough examination of the
diverse livelihood options. risk posed by tropical cyclones and analysed the significance of various
Addressing the escalating frequency and intensity of cyclones factors at the block level within the coastal districts of Odisha. This

16
T. Das et al. Progress in Disaster Science 23 (2024) 100357

research marks the inaugural attempt in the area to apply remote editing, Funding acquisition.
sensing, GIS, and machine learning methodologies. Given Odisha’s
history of enduring the most cyclone landfalls nationwide, it stands as a
region of heightened susceptibility to tropical cyclones. Moreover, ac- Declaration of competing interest
cording to the 2011 census report, approximately 33% of Odisha’s total
population resides in these coastal districts, which are classified as The authors declare that they have no known competing financial
Empowered Action Group (EAG) states, indicating a less favourable interests or personal relationships that could have appeared to influence
socio-economic condition in the region. Hence, it’s essential to perform the work reported in this paper.
an in-depth risk evaluation and ascertain the importance of different
factors to pinpoint the area’s most susceptible, enabling the formulation Data availability
of suitable resilience tactics.
For the risk assessment, we employed 21 parameters categorized Data will be made available on request.
under four major criteria: exposure, vulnerability, hazard, and mitiga-
tion capacity. We also assessed the influence of these parameters on the Acknowledgement
heightened risk of tropical cyclones in specific blocks. The research
encompassed approximately 62% of areas highly susceptible to risks and This work was financially supported by the Deanship of Scientific
35% of zones prone to hazards. With the exception of the southwestern Research at the King Faisal University, Saudi Arabia (Grant: A370). The
region, the entire study area faces vulnerability to tropical cyclones. first author is indebted to UGC-India for availing doctoral fellowship
Particularly, blocks situated in the eastern and northern sectors of during this research work.
Ganjam, Bhadrak, and Baleshear districts, alongside portions of Jagat-
singhpur and Puri districts, are marked as high-risk areas. Appendix A. Supplementary data
Devastating natural calamities like tropical cyclones not only result
in casualties and economic losses but also hinder the social development Supplementary data to this article can be found online at [Link]
progress of the affected regions. They significantly impact the well-being org/10.1016/[Link].2024.100357.
of communities. Our holistic method offers valuable perspectives for
planners and local government officials to develop efficient coastal risk References
management plans that alleviate the impact of tropical cyclones in the
frequently impacted eastern coastal area of India. Additionally, these [1] Ali SA, Khatun R, Ahmad A, Ahmad SN. Assessment of cyclone vulnerability,
methods assist planners in developing location-specific mitigation hazard evaluation and mitigation capacity for analyzing cyclone risk using GIS
technique: a study on Sundarban biosphere reserve, India. Earth Syst Environ 2020;
measures, ultimately enhancing the resilience of coastal communities. 4(1):71–92.
The choice of standards, types of data, and measurement scales for [2] Al-Najjar HA, Pradhan B, Beydoun G, Sarkar R, Park HJ, Alamri A. A novel method
this evaluation may differ based on the particular geographic area. For using explainable artificial intelligence (XAI)-based Shapley additive explanations
for spatial landslide prediction using time-series SAR dataset. Gondw Res 2023;
instance, in Odisha, we used parameters such as population density, 123:107–24.
literacy rate, distance to cyclone shelters, and health center availability, [3] Alqadhi S, Mallick J, Alkahtani M, Ahmad I, Alqahtani D, Hang HT. Developing a
which are measured using various scales such as counts (e.g., number of hybrid deep learning model with explainable artificial intelligence (XAI) for
enhanced landslide susceptibility modeling and management. Nat Hazards 2023:
cyclone shelters), percentages (e.g., literacy rate), and distances (e.g., 1–29.
kilometers to nearest health center). These scales provide a quantitative [5] Baalousha HM, Younes A, Yassin MA, Fahs M. Comparison of the fuzzy analytic
basis for evaluating risk. hierarchy process (F-AHP) and fuzzy logic for flood exposure risk assessment in
arid regions. Hydrology 2023;10(7):136.
However, in other geographic regions, the relevant parameters and
[6] Bera S, Guru B, Chatterjee R, Shaw R. Geographic variation of resilience to
their measurement scales might differ. For example, in regions with landslide hazard: a household-based comparative studies in Kalimpong hilly
different socio-economic profiles or infrastructure setups, additional or region, India. Int J Disaster Risk Reduct 2020;46:101456.
[8] Bloemendaal N, Haigh ID, de Moel H, Muis S, Haarsma RJ, Aerts JC. Generation of
alternative parameters such as income levels, types of housing, and ac-
a global synthetic tropical cyclone hazard dataset using STORM. Sci data 2020;7
cess to emergency services might be necessary. Furthermore, the data (1):1–12.
types could vary, with some areas relying more on satellite imagery for [9] Boragapu R, Guhathakurta P, Sreejith OP. Tropical cyclone vulnerability
remote sensing, while others might have detailed local surveys assessment for India. Nat Hazards 2023;117(3):3123–43.
[10] Brennan MA, Flint CG. Uncovering the hidden dimensions of rural disaster
providing granular socio-economic data. The standards for assessing mitigation: capacity building through community emergency response teams.
vulnerability, exposure, and mitigation capacity may also need adjust- J Rural Soc Sci 2007;22(2):7.
ment. For instance, coastal areas with different climatic conditions or [11] Bubeck P, Pham TDM, Nguyen TNA, Hudson P. Disaster risk reduction on stage:
an empirical evaluation of community-based theatre as risk communication tool
demographic profiles might require tailored indices or thresholds to for coastal risk mitigation and ecosystem-based adaptation. Prog Disaster Sci
accurately reflect local conditions. These variations ensure that risk 2024;22:100323.
assessments are context-specific and relevant to the unique challenges [12] Chapman K, Norwood M, Shirota C, Palipana D, Kendall E. An undignified disaster
reality for Australians with disability. Aust Health Rev 2022;46(6):710–2.
faced by each area. This innovative method can be customized for [14] Chauhan A, Liou YA. Influence of major typhoons on ocean, atmosphere and air
evaluating risk and gauging the importance of factors in different coastal quality of Northwest Pacific during August 2020. Atmos Environ 2022;271:
environments affected by tropical storms. By recognizing and adapting 118923.
[15] Chauhan A, Kumar R, Singh RP. Coupling between land–ocean–atmosphere and
to these geographic variations, our approach offers a robust framework
pronounced changes in atmospheric/meteorological parameters associated with
for enhancing the resilience of coastal communities worldwide. the Hudhud cyclone of October 2014. Int J Environ Res Public Health 2018;15(12):
2759.
[16] Chauhan A, Singh R, Kumar R, Dash P. Change in land and ocean parameters along
CRediT authorship contribution statement the track of tropical cyclone Fani. In: IGARSS 2020–2020 IEEE international
geoscience and remote sensing symposium. IEEE; 2020. p. 5554–7.
Tanmoy Das: Writing – original draft, Software, Data curation, [17] Chauhan A, Singh RP, Dash P, Kumar R. Impact of tropical cyclone “Fani” on land,
ocean, atmospheric and meteorological parameters. Mar Pollut Bull 2021;162:
Conceptualization. Swapan Talukdar: Methodology, Formal analysis,
111844.
Data curation. Shahfahad: Writing – original draft, Methodology, [18] Chauhan A, Kumar R, Dash P, Singh RP. Dynamics of Amphan cyclone and
Investigation. Mohd Waseem Naikoo: Visualization, Formal analysis, associated changes in ocean, land meteorological and atmospheric parameters.
Data curation. Ishita Afreen Ahmed: Validation, Data curation. Atiqur MAUSAM 2021;72(1):215–28.
[19] Chauhan A, Kumar R, Liou YA, Singh RP. Effect of cyclones on atmospheric and
Rahman: Writing – review & editing, Supervision. Md Kamrul Islam: meteorological parameters. In: Asian atmospheric pollution. Elsevier; 2022.
Software, Resources, Data curation. Edris Alam: Writing – review & p. 521–47.

17
T. Das et al. Progress in Disaster Science 23 (2024) 100357

[20] Chettiyam Thodi MF, Gopinath G, Surendran UP, Prem P, Al-Ansari N, Mattar MA. [63] Pattnaik S. Multipurpose cyclone shelters and caste discrimination. Econ Pol Wkly
Using RS and GIS techniques to assess and monitor coastal changes of Coastal 2019;54(21):13.
Islands in the marine environment of a humid tropical region. Water 2023;15(21): [64] Pradhan B, Dikshit A, Lee S, Kim H. An explainable AI (XAI) model for landslide
3819. susceptibility modeling. Appl Soft Comput 2023;142:110324.
[22] Chittibabu P, Dube SK, Macnabb JB, Murty TS, Rao AD, Mohanty UC, et al. [66] Rahman MR, Shi ZH, Chongfa C. Soil erosion hazard evaluation—an integrated use
Mitigation of flooding and cyclone hazard in Orissa, India. Nat Hazards 2004;31: of remote sensing, GIS and statistical approaches with biophysical parameters
455–85. towards management strategies. Ecol Model 2009;220(13–14):1724–34.
[23] Das S, DSouza NM. Identifying the local factors of resilience during cyclone [69] Rao AD, Upadhaya P, Ali H, Pandey S, Warrier V. Coastal inundation due to
Hudhud and Phailin on the east coast of India. Ambio 2020;49(4):950–61. tropical cyclones along the east coast of India: an influence of climate change
[24] Das T, Naikoo MW, Talukdar S, Parvez A, Rahman A, Pal S, et al. Analysing process impact. Nat Hazards 2020;101(1):39–57.
and probability of built-up expansion using machine learning and fuzzy logic in [70] Ruangpan L, Vojinovic Z, Di Sabatino S, Leo LS, Capobianco V, Oen AM, et al.
english Bazar, West Bengal. Remote Sens (Basel) 2022;14(10):2349. Nature-based solutions for hydro-meteorological risk reduction: a state-of-the-art
[26] Fawcett T. An introduction to ROC analysis. Pattern Recogn Lett 2006;27(8): review of the research area. Nat Hazards Earth Syst Sci 2020;20(1):243–70.
861–74. [71] Safaeian M, Moses R, Ozguven EE, Dulebenets MA. An optimization-based risk
[27] Feizizadeh Bakhtiar, Roodposhti Majid Shadman, Jankowski Piotr, management framework with risk interdependence for effective disaster risk
Blaschke Thomas. A GIS-based extended fuzzy multi-criteria evaluation for reduction. Prog Disaster Sci 2024;21:100313.
landslide susceptibility mapping. Comput Geosci 2014. [Link] [72] Saha TK, Pal S, Talukdar S, Debanshi S, Khatun R, Singha P, et al. How far spatial
cageo.2014.08.001. resolution affects the ensemble machine learning based flood susceptibility
[28] Fuji Y, Sakaniwa R, Shirai K, Saito T, Ukawa S, Iso H, et al. The number of leisure- prediction in data sparse region. J Environ Manage 2021;297:113344.
time activities and risk of functional disability among Japanese older population: [74] Sahoo B, Bhaskaran PK. A comprehensive data set for tropical cyclone storm surge-
the JAGES cohort. Prev Med Rep 2022;26:101741. induced inundation for the east coast of India. Int J Climatol 2018;38(1):403–19.
[30] Geiger T, Gütschow J, Bresch DN, Emanuel K, Frieler K. Double benefit of limiting [75] Sahoo B, Bhaskaran PK. Multi-hazard risk assessment of coastal vulnerability from
global warming for tropical cyclone exposure. Nat Clim Change 2021;11(10): tropical cyclones–a GIS based approach for the Odisha coast. J Environ Manage
861–6. 2018;206:1166–78.
[32] Gopinath G. Critical coastal issues of Sagar Island, east coast of India. Environ [76] Sakizadeh M, Mohamed MM, Klammler H. Trend analysis and spatial prediction of
Monit Assess 2010;160:555–61. groundwater levels using time series forecasting and a novel spatio-temporal
[33] Hoque MAA, Phinn S, Roelfsema C, Childs I. Assessing tropical cyclone risks using method. Water Resour Manag 2019;33:1425–37.
geospatial techniques. Appl Geogr 2018;98:22–33. [77] Samarakkody A, Senanayake AC, Malalgoda C, Amaratunga D, Haigh R,
[34] Hoque MAA, Pradhan B, Ahmed N, Roy S. Tropical cyclone risk assessment using Liyanage C, et al. Inclusivity in online and distance disaster education: a review of
geospatial techniques for the eastern coastal region of Bangladesh. Sci Total educators’ views. Prog Disaster Sci 2023;20:100298.
Environ 2019;692:10–22. [78] Santos GDC. 2020 tropical cyclones in the Philippines: a review. Trop Cyclone Res
[35] Hoque MAA, Pradhan B, Ahmed N, Ahmed B, Alamri AM. Cyclone vulnerability Rev 2021;10(3):191–9.
assessment of the western coast of Bangladesh. Geomat Nat Haz Risk 2021;12(1): [80] Sarkhel P, Biswas D, Swain SS. A review of cyclone and its impact on the coastal
198–221. belts of Odisha. Int J Eng Res Technol 2019;8(5):759–62.
[36] Hosmer DW, Lemeshow S, Sturdivant RX. Applied logistic regression. Wiley; 2013. [81] Sattar MA, Biswas AAA, Islam MT, Hossain MA, Siddeqa M, Rahim MA, et al.
[37] Jaman T, Dharanirajan K, Sharma SS. Assessment of impact of cyclone hazard on Disaster vulnerability and mitigation of humanitarian issues in coastal Bangladesh:
social vulnerability of Bhadrak District of Odisha state during Phailin cyclone in local evidence and knowledge gaps. Prog Disaster Sci 2020;8:100138.
2013 and Titli cyclone in 2018 using multi-criteria analysis and geospatial [82] Senan CP, Ajin RS, Danumah JH, Costache R, Arabameri A, Rajaneesh A, et al.
techniques. Int J Disaster Risk Reduct 2021;53:101997. Flood vulnerability of a few areas in the foothills of the Western Ghats: a
[38] Khashei-Siuki A, Sharifan H. Comparison of AHP and FAHP methods in comparison of AHP and F-AHP models. Stoch Environ Res Risk Assess 2023;37(2):
determining suitable areas for drinking water harvesting in Birjand aquifer. Iran. 527–56.
Groundw Sustain Dev 2020;10:100328. [83] Shamsuzzoha M, Shaw R, Ahamed T. Machine learning system to assess rice crop
[41] Kumar A, Jeeva JC, Sarangi DN, Panda AK, Srivastava SK. Analysis of socio- change detection from satellite-derived RGVI due to tropical cyclones using remote
economic status of people in aspirational districts of Odisha for inclusive growth: sensing dataset. Remote Sens Appl Soc Environ 2024;35:101201.
socioeconomic status of aspirational districts of Odisha. J AgriSearch 2020;7(3): [85] Shekar PR, Mathew A. Assessing groundwater potential zones and artificial
158–62. recharge sites in the monsoon-fed Murredu river basin, India: an integrated
[42] Kumar S, Lal P, Kumar A. Turbulence of tropical cyclone ‘Fani’in the Bay of Bengal approach using GIS, AHP, and fuzzy-AHP. Groundw Sustain Dev 2023;23:100994.
and Indian subcontinent. Nat Hazards 2020;103(1):1613–22. [86] Shultz JM, Russell J, Espinel Z. Epidemiology of tropical cyclones: the dynamics of
[44] Kundu SN, Sahoo AK, Mohapatra S, Singh RP. Change analysis using IRS-P4 OCM disaster, disease, and development. Epidemiol Rev 2005;27(1):21–35.
data after the Orissa super cyclone. Int J Remote Sens 2001;22(7):1383–9. [88] Sobel AH, Camargo SJ, Hall TM, Lee CY, Tippett MK, Wing AA. Human influence
[46] Li K, Li GS. Risk assessment on storm surges in the coastal area of Guangdong on tropical cyclone intensity. Science 2016;353(6296):242–6.
Province. Nat Hazards 2013;68(2):1129–39. [89] Talukdar S, Singha P, Mahato S, Pal S, Liou YA, Rahman A. Land-use land-cover
[48] Lyu HM, Shen SL, Zhou A, Yang J. Risk assessment of mega-city infrastructures classification by machine learning classifiers for satellite observations—a review.
related to land subsidence using improved trapezoidal FAHP. Sci Total Environ Remote Sens (Basel) 2020;12(7):1135.
2020;717:135310. [90] Talukdar S, Naikoo MW, Mallick J, Praveen B, Sharma P, Islam ARMT, et al.
[49] Matthews T, Wilby RL, Murphy C. An emerging tropical cyclone–deadly heat Coupling geographic information system integrated fuzzy logic-analytical
compound hazard. Nat Clim Change 2019;9(8):602–6. hierarchy process with global and machine learning based sensitivity analysis for
[50] Meshram SG, Alvandi E, Singh VP, Meshram C. Comparison of AHP and fuzzy AHP agricultural suitability mapping. Agr Syst 2022;196:103343.
models for prioritization of watersheds. Soft Comput 2019;23:13615–25. [91] Tan R, Wang S, Jiang Y, Ishida K, Fuji MG. Nonlinear adaptive controller for omni-
[51] Meshram SG, Alvandi E, Singh VP, Meshram C. Comparison of AHP and fuzzy AHP directional walker: dynamic model improvement and experiment. ICIC Express Lett
models for prioritization of watersheds. Soft Comput 2019;23:13615–25. 2012;6(3):611–5.
[52] Mishra M, Pattnaik N, Dash MK, Kattel DB, Mishra SK. Analyzing trend of tropical [92] Teke A, Kavzoglu T. Explainable artificial intelligence empowered landslide
cyclone activity along Odisha Coast, India. 2020. susceptibility mapping using extreme gradient boosting (XGBoost). In: Advanced
[53] Mishra M, Kar D, Santos CAG, da Silva RM, Das PP. Assessment of impacts to the engineering days (AED). 6; 2023. p. 74–6.
sequence of the tropical cyclone Nisarga and monsoon events in shoreline changes [93] Tripathi AK, Agrawal S, Gupta RD. Comparison of GIS-based AHP and fuzzy AHP
and vegetation damage in the coastal zone of Maharashtra, India. Mar Pollut Bull methods for hospital site selection: a case study for Prayagraj City, India.
2022;174:113262. GeoJournal 2021:1–22.
[54] Mittal R, Tewari M, Radhakrishnan C, Ray P, Singh T, Nickerson AK. Response of [95] UNDRR. The global assessment report on disaster risk reduction 2019: risk and
tropical cyclone Phailin (2013) in the Bay of Bengal to climate perturbations. poverty in a changing climate. UNDRR; 2020. Retrieved from, [Link]
Climate Dynam 2019;53(3):2013–30. org/sites/default/files/2020-06/GAR2019_Online.pdf.
[55] Mohanty A, Dubey A, Singh RB. Policy and governance strategies for effective [96] UN-SPIDER. Report on space-based information for disaster management and
cyclone risk management in Odisha, India: a journey from 1999 super cyclone. In: emergency response. [Link]
Cyclonic disasters and resilience. Singapore: Springer; 2022. p. 155–84. mber-2020; 2020.
[58] Mondal M, Haldar S, Biswas A, Mandal S, Bhattacharya S, Paul S. Modeling [97] Weinkle J, Maue R, Pielke Jr R. Historical global tropical cyclone landfalls.
cyclone-induced multi-hazard risk assessment using analytical hierarchical J Climate 2012;25(13):4729–35.
processing and GIS for coastal West Bengal, India. Reg Stud Mar Sci 2021;44: [98] Wen S, Su B, Wang Y, Fischer T, Li X, Yin Y, et al. Economic sector loss from
101779. influential tropical cyclones and relationship to associated rainfall and wind speed
[59] Nott J. Palaeotempestology: the study of prehistoric tropical cyclones—a review in China. Global Planet Change 2018;169:224–33.
and implications for hazard assessment. Environ Int 2004;30(3):433–47. [100] World Health Organization. Guidance note on disability and emergency risk
[60] Np J, Gopinath G. Groundwater suitability zonation with synchronized GIS and management for health. Geneva. Switzerland. 2013.
MCDM approach for urban and peri-urban phreatic aquifer ensemble of southern [101] World Meteoroogical Organization. Climate change and water (WMO-No. 1225).
India. Urban Water J 2018;15(8):801–11. [Link]
[62] Parks RM, Anderson GB, Nethery RC, Navas-Acien A, Dominici F, 20global%20mean%20temperature%20for,temperature%20datasets%20(Figure
Kioumourtzoglou MA. Tropical cyclone exposure is associated with increased %201); 2020.
hospitalization rates in older adults. Nat Commun 2021;12(1):1–12.

18
T. Das et al. Progress in Disaster Science 23 (2024) 100357

[103] Zhao R, Fang C, Liu J, Zhang L. The evaluation and obstacle analysis of urban [108] Saaty TL. A scaling method for priorities in hierarchical structures. J. Mathematic.
resilience from the multidimensional perspective in Chinese cities. Sustain Cities Psychol. 1977;15(3):234–81.
Soc 2022;86:104160. [109] Mallick J, Singh RK, AlAwadh MA, Islam S, Khan RA, Qureshi MN. GIS-based
[104] Camargo SJ. Global and regional aspects of tropical cyclone activity in the CMIP5 landslide susceptibility evaluation using fuzzy-AHP multi-criteria decision-
models. J. Clim. 2013;26(24):9880–902. making techniques in the Abha Watershed, Saudi Arabia. Environ. Earth Sci.
[105] Ahmed R, Mohapatra M, Giri RK, Dwivedi S. An Evaluation of the Advanced 2018;77:1–25.
Dvorak Technique (9.0) for the tropical cyclones over the North Indian Ocean. [110] Kubler S, Robert J, Derigent W, Voisin A, Le Traon Y. A state-of the-art survey &
Tropic. Cyclone Res. Rev. 2021;10(4):201–8. testbed of fuzzy AHP (FAHP) applications. Expert systems with applications 2016;
[106] Haigh ID, Wahl T, Rohling EJ, Price RM, Pattiaratchi CB, Calafat FM, 65:398–422.
Dangendorf S. Timescales for detecting a significant acceleration in sea level rise. [111] Saaty TL. The analytic hierarchy process (AHP). The J. Operational Res. Soc.
Nat. Commun. 2014;5(1):3635. 1980;41(11):1073–6.
[107] Bhowmik SR, Kotal SD, Kalsi SR. An empirical model for predicting the decay of
tropical cyclone wind speed after landfall over the Indian region. J. Appl. Meteor.
Clim. 2005;44(1):179–85.

19