Damage Estimation using Shock Zones: A case study of Amphan tropical cyclone

The tropical cyclone affects millions of people living in the coastal regions. The changing climate has led to an increased intensity and frequency of cyclones, therefore increasing the damage caused to people, the environment, and property. The Bay of Bengal is most prone to tropical cyclones, which affects Bangladesh and the eastern coastal region of India due to geographical proximity. Hence, a comprehensive understanding of the inundation damage and intensity caused becomes essential to focus the relief efforts on the affected districts. This study identied the shock zone and assessed the inundation associated damage caused by recent cyclone Amphan in the area of Bangladesh and West Bengal in India. The shock zonation was based on the track of cyclones, cyclone wind speed zones, elevation, wind impact potentiality, and agricultural population area. The identication of the affected area was done using integrated Landsat and SAR data, and economic damage cost was assessed using the Asian Development Bank’s (ADB) Unit price approach. The total people affected due to inundation are 2.4 million in India and 1.4 million in Bangladesh and the damage totaled up to 5.4 million USD. The results of this study can be used by concerned authorities to identify the shock zones and be used for rapid assessment of the damages.


Introduction
Tropical cyclones are storms that cause extensive damage to property, disruption of transport and communication networks, loss of human and animal lives, and environmental degradation (Dube et Mallick et al., 2017). Further, these coastal regions are highly vulnerable due to large population density, high poverty rates, and the presence of temporary infrastructure. According to Paul and Dutt (2010), more than 1 million people were killed by cyclonic disasters since 1877 in coastal Bangladesh. Further, sea-level rise due to global warming will intensify the impacts of tropical cyclones on people's lives and livelihood across the coastal districts of both India and Bangladesh (Karim and Mimura, 2008 On 20th May 2020, the tropical cyclone 'Amphan' hit the coast of India and Bangladesh, accompanied by severe storm surges and rainfall (wind speeds up to 195kmph or 121mph). The cyclone caused causalities, killing around 88 people and leaving thousands homeless in India and Bangladesh (Aljazeera, 2020). The cyclone struck at a time when the region had already been ailing with the impact of the COVID-19 pandemic. In such a situation, the relief and recovery measures get further complicated. Therefore, nding the risk zones and estimating the damage is essential to provide an idea about the loss of property, agricultural and livestock, and various primary livelihoods. Some news reports and government organizations published estimated damage for a particular area (Sud and Rajaram, 2020) or speci c aspects. Detailed reports on risks and overall damages in the entire cyclonic affected coastal and adjacent districts were not available. This study develops a spatial framework that includes cyclone shock zones and damage and output loss intensity. UN-SPIDER recommended damage estimation practice and unit cost methods were combined to estimates output loss for the entire ood inundated areas caused by cyclone Amphan. This study seeks to analyse the situation of the areas majorly affected by recent inundation and ooding caused by the Amphan cyclone. Firstly, the study assesses the categories of Amphan shock zones to identify potentially exposed areas, rather than following the common risk zonation approach. Secondly, developing a spatial damages assessment framework to account for the economic cost of inundation and ooding on the crop, livestock, and housing units. In this study, the maps produced by risk assessment would be very helpful to identify the spread and intensity of disaster to create the most effective disaster mitigation plan in this area.

Data
This study uses socio-economic, disaster, and climate-related data, administrative GIS layers, and satellite data to estimate the cyclone severity and damage intensity. Socio-economic data was used to estimate human exposure to the disaster and the household crisis. Climate data depicts the last 48 hr. update of the cyclone event i.e., its track, intensity, and area of in uence. The historical disaster records were used to assess the current loss and to compute area-wise disaster damage intensity. Further, GIS layers and remote sensing data served the local to regional level damage and loss information of crop, forest, property, and human life. The database and its preliminary preparation process are illustrated in Table 1. • History of previous cyclone track data is valuable to understand the regional risk from the cyclone events • Spatial estimated population dataset is useful to estimate the current population under threats • The reliability of the population spatial dataset is checked using 2011 as a reference year.
• Total low-income rural population data is extracted from the global human exposure dataset 2015.
• The 2010 man-animal proportion is used to compute the livestock population in 2020.
• Built-up grid data is used to validate the household unit for the year 2014.
• WorldPop population grid data is used to estimation the inundation impact on human life

Methods
The study was carried out in four major steps. First, shock zones were de ned using the cyclone characteristics and sociodemographic situation shock zones. Second, using remote sensing and GIS tools LULC and ood-affected areas were demarcated. Third, the impacts of inundation were estimated using LULC, inundation area, population, and poverty situation.
Finally, cyclone shock zones and their associated cost are estimated to understand the association between cyclone intensity and damage. Detail work ow of these steps is illustrated in Figure 1 The overall accuracy of the LULC data was computed to be 91% and the Kappa coe cient was 89%.
Inundation change analysis performed in the GEE Sentinel-1C-band GRD imagery with VV, VH polarization, and 'DESCENDING' pass direction for the dates 4th May 2020 and 22nd May 2020 was used for pre and post-inundation situation (Uddin et al., 2019).
In the GEE, all the data is pre-processed (i.e., noise removal, radiometric correction, and terrain correction, and nally, backscatter scattering to decibel conversion). The intensity of change per pixel was estimated by dividing after ood mosaic with before ood mosaic. A binary ood layer was prepared using a threshold of 1.25, where values above 1.25 were assigned a score of 1 and all other pixels are assigned a score of 0 . quantity vary signi cantly between rural and urban areas. The damage assessment for the infrastructure and forestry sectors was not attempted due to inconsistencies in the media report and the absence of reliable spatial data.
This study followed two steps to estimate the loss value which is in a comparable form. Firstly, inundation affected cropland, livestock, and housing loss area was estimated using the 5km x 5km grid. Secondly, inundated cropland and the built-up area within each grid was multiplied with the unit price of the crop, livestock, and housing. The values are converted into USD using country-speci c exchange rate. Details steps are illustrated in Table 2.The total cyclone damage and loss amount are estimated for each cyclone shock zone and district.

3.3Amphan and poverty severity
Grid wise population data and ood-affected areas were intersected to estimate population exposure to inundation. Also, low income rural and urban population from the global exposure dataset 2014 were used to estimate the district-wise poor population share. Next, the association between district-level poverty rate and damage intensity was computed to understand the cyclone induced poverty. Further, food accessibility and total risk information were combined to assess the severity of poverty. This estimation identi es the areas which require urgent relief and the approximate amount of monetary support needed.

3.4Amphan cyclone severity zone
Cyclone wind speed zones, elevation, wind impact, distance from cropland and vegetation, business loss, and agricultural population area was used to prepare Amphan severity zones. All these variables were normalized such that higher values indicate high severity and lower value indicate low severity. These zones are then intersected with the damage data to understand the association between cyclone shocks intensity and damage intensity.
Four zones were extracted using GeoSOM Ward's criterion-based 'Contiguity-constrained hierarchical clustering' process, available in the SPAWNN toolkit algorithm (Hagenauer and Helbich, 2016). The spatial clustering approach is a two-step process to de ne the Best Matching Unit (BMU). Firstly, based on the input dataset i.e., wind speed, distance from the cyclone track, elevation, and proportion of rural area closest neurons are identi ed. Secondly, a distance criterion is applied to the identi ed closest neurons.
This study used radius = 1 to reduce the spatial dependencies in BMU. Each input variable and its BMU structure are presented in a hexagonal space. This spatial clustering is based on self-organising neural network techniques which are useful for its information extraction capability from the large spatial dataset. occur each year. The wind speed ranges from 75km/hr to 260km/hr, with half of the cyclones falling in the super cyclonic storm or extremely severe cyclonic storm range. The overall losses depend upon the intensity of the cyclone and the geographic region where it hits (Table 3).   In Bangladesh, more than 10% of GDP came from the agriculture, livestock, sheries, and forestry sectors in the nancial year 2019. This sector manages to feed more than 40% of the labour force. More than 83% of households in the country live in rural areas and are dependent on agricultural and allied sectors. Agriculture is still the dominant activity in the country with 8 Mha net cropped area (54%). Paddy is the major crop (77%), followed by pulses (2.8%), oilseed (2.78%), spices and condiments (2.53%), and vegetable (1.27%).
Cyclone Amphan severely affected the agricultural sector in India and Bangladesh. Severe rainfall caused heavy crop damage. A total of 0.11 Mha and 0.06 Mha cropland was severely affected due to inundation in the districts of West Bengal, and Bangladesh respectively, which was under kharif crops ( Figure 4). Mostly rice, pulses, vegetation, and mango fruit were affected in the districts of West Bengal, whereas largely paddy cultivation was affected in Gopalganj (4.7 million USD), Jassore (4.4 million) and Jhenaidah ( Figure 5).

Housing damage
Cyclone Amphan caused extensive damage to the housing units. Both kutcha and semi-pucca houses were affected due to ooding and associated cyclonic wind. The assessment considered household units that were submerged underwater for around 3-4 days after the cyclone, as spatial information on affected kutcha and pucca structures is unavailable. Therefore, this study assumes that both kutcha and pucca structures were partially damaged due to heavy rainfall and storm surge. Almost 0. Parganas (6%) (Figure 8). The share of the poor population in Bangladesh is quite high as compare to India, therefore Amphan damage intensity has made a severe crisis in Bangladesh.

Cyclone shock zones
The shock zones de ne the varied vulnerability (Figure 8). The districts were clustered into four shock zones using the variableswind speed, distance from the cyclone track, elevation, and proportion of the rural area. The highest shock zone lies along the Loading [MathJax]/jax/output/CommonHTML/jax.js cyclone track, experienced the highest wind speed, was prone to storm surge due to low elevation, and had a greater proportion of rural areas (Table 5).
In India, the wind speed experienced was highest for Zone I, followed by Zone II, Zone III, and lastly Zone IV. The wind speed showed a declining trend from Southeast to North West. Zone I in Bangladesh faced the greatest wind speed, followed by Zone III, Zone IV, and Zone II respectively. Wind speed showed a decline from west to east and from south to north in Bangladesh.
The cyclone travelled from south-west to north-east. Hence, the north-west and south-east area of interest lie farthest from the cyclone track. In India, Zone I and II and in Bangladesh Zone IV and Zone I were closest to the path of the cyclone.
The lower elevation areas have the highest chance of being affected due to cyclone, experiencing ood and storm surges. In India, for elevation Zone I lies at the lowest elevation, followed by zone II and zone III respectively. In Bangladesh, Zone I had the lowest elevation followed closely by zone 3 and zone II.
Rural tracts have greater vulnerability owing to the damage caused to agricultural land, livestock and temporary infrastructures such as mud and thatched houses. Bangladesh is predominantly rural, hence more vulnerable as compared to India

Discussion And Conclusion
Spatial risk zonation is commonly used to identify the exposed area to certain disasters. In general, historical data and large heterogeneous actors are incorporated into the risk assessment models. Based on the exposed area measurement approach this study demarcates four shock zones based on elevation, distance from cyclone track, wind speed, and rural area share based on GEOSOM clustering. These zones explain the degree of initial shock that Amphan created over India and Bangladesh. Extracted shock zones are useful to understand the potential exposed area and damage intensity. This study used these zones to estimate the inundation and ood-related damage caused by the cyclone. Estimation suggested that the cyclone intensity and inundation damage was not mutually inclusive. The northern and eastern parts of the study area did not face severe wind and were quite far from the eye of the cyclone, but inundation damage was still high. Physical setting and rainfall intensity caused huge damage within the 200km radius. A sudden weakening of the translation speed of Amphan caused an increase in rainfall and inundation damage in Bangladesh, as also noticed in Kossain's (2018) ndings. These generalized climatic risk micro-zonation and ranking mechanisms can induce event-speci c bias to future management plans.
The spatial damage assessment framework integrated UN-SPIDER SAR-based ood damage estimation and Asian Development Bank's (ADB) unit price approach to assess the economic loss. All these types of damage and loss estimation framework largely underestimate the actual cost. However, the current spatial approach is more useful and cost-effective to identify the worst affected location and damage intensity more clearly. Only the inundation associated with economic loss was captured. News reports and govt. organizations published estimated total damage, but a detailed report on Amphan inundation impact was not available for the entire cyclonic affected coastal and adjacent districts.According to government sources the total estimated damages of Amphan is 132 million USD in Bangladesh and PCD Global estimates show 12.4 billion USD in India, whereas our estimation suggests only ood and inundation caused total damage of 5.4 billion USD. Flood and inundation caused 31 % of total damage in India, 7% in Bangladesh. If we considered the global average ood impact, up to average of 5.5% people will be expected to fall below the poverty line in Bangladesh and West Bengal.
This framework is more reliable and t for ood damage assessment but can be usedto estimate damage and loss for all types of natural disasters if the required information is available. However, the current approach has several limitations such as Grid level assessment (5km x 5 Km), partial ood damage assessment, which leads to gross generalization due to unit cost approach, and classi cation error. First, in general, the grid-based estimation is sometimes unable to provide high spatial accuracy, which can also increase the underestimation error. However, large coverage areas and different sources of information can be easily Loading [MathJax]/jax/output/CommonHTML/jax.js integrated using this approach. Second, partial estimation of inundation damage is captured through the above methodology as data on types of standing crops was not available. Therefore, district-level gross estimates were apportioned to the grid level.
Similarly, damaged kutcha and pucca household locations were not available, even the degree of damage information was not available. Hence, partial damage cost is uniformly applied to estimate the property damage. Third, the unit cost approach introduces large generalization. Even though the previous estimates largely follow this approach, no other standard universal tools are available. Fourth, some LULC classi cation errors and ood detection errors are unavoidable, although these are minimized using rigorous accuracy checks and post-classi cation data standardization.
Even though the spatial damage estimation has always been a subject to underestimation as it only captures the material cost, hence underestimating the actual economic cost. The spatial approach is still useful as it is suitable to assess the pattern and degree of damage and identi es the worst affected location. This spatial damage estimate framework is indispensable to improve the relief and support mechanism. The damage estimation is essential to channelize the resources and funds to the affected people and proper locations. This method is quite cost-effective and e cient as large-scale spatial datasets as tools and techniques are readily available. Remote sensing and GIS technology can be used to for a rapid estimation of the aftereffect of any disaster event possible. This research adds to the research by developing a spatial direct damage estimation framework using freely available spatial dataset.

Declarations
Ethical statement We con rm that any aspect of the work covered in this manuscript that has involved human patients has been conducted with the ethical approval of all relevant bodies and that such approvals are acknowledged within the manuscript.
Declared by Authors-