Introduction

Flood is considered as the most devastating of events among other natural disasters and hazards. Floods occurred when overland flow surpasses and river channels inundate the land. Floods are considered as the most terrible climatic disaster in terms of loss of fauna and human resources (Youssef et al. 2011; Haghizadeh et al. 2017). Floods are cause of socio-economic transformations and consequences on environment and ecosystem. Floods can also cause damages to natural and man-made resources, agriculture, transportation, social infrastructures, cultural inheritance, economic feasibilities and many other aspects of human life and livelihood (Billa et al. 2006; Messner and Meyer 2006; Huang et al. 2008; Dang et al. 2011; Alvarado-Aguilar et al. 2012; Yu et al. 2013; Rahmati et al. 2016a, b). Thus, it is very essential to perform risk analysis, susceptibility mapping and vulnerability assessment of flood hazards as a pre-planned management aspect so that the extent of damage can be mitigated as much as possible (Feng and Wang 2011; Kia et al. 2012; Lee et al. 2012a; Esteves 2013; Markantonis et al. 2013; Schober et al. 2015; Khosravi et al. 2016; Rahmati et al. 2016a, b; Haghizadeh et al. 2017).

It is a quite difficult task to completely prevent the flood incidences but the devastating risk can be mitigated to some extent through proper risk and vulnerable zonation (Cloke and Pappenberger 2009). Therefore, flood risk mapping is considered as a very advantageous tool to detect the flood susceptibility areas (Tehrany et al. 2014; Rahmati et al. 2016a, b), as there are many factors behind flood occurring which differ with different geographical locations and geological settings. The results will become more weighted and acceptable if flood susceptibility mapping and hazard analysis are done through multi criteria instead of taking a single responsible factor (Booij 2005; Minea 2013; Xu et al. 2013; Poussin et al. 2014). In many flood vulnerable areas, disaster management is mainly focused on emergency response, disaster relief and rehabilitation activities. Several studies recommend that a paradigm shift is required from disaster relief and response to disaster risk and vulnerability reduction (Yodmani 2001). Such objective identification and detection of more to less vulnerable and risk areas are more crucial (Yang et al. 2015).

Recently, remote sensing (RS) and GIS technique have gained wide popularity in the field of vulnerability analysis (Haq et al. 2012; Moel et al. 2014). Remote sensing data help in acquiring sound information regarding land surface, topographic features, vegetation cover, climatic phenomena and many other relevant information of region. Concomitantly, GIS technique helps in preparing spatial database using RS data for flood vulnerable mapping. Remote sensing and GIS techniques were successfully applied in the field of flood vulnerability throughout the world, like urban flood hazard zoning using GIS and multi-criteria decision analysis in Tucuman Province of Argentina (Fernandez and Lutz 2010), ANN model for flood simulation using GIS in Johor River Basin of Malaysia (Kia et al. 2012), validation for predictive flood susceptibility mapping using GIS and frequency ratio model (Lee et al. 2012a), flood risk assessment of Kosi river basin, India, using remote sensing and GIS techniques (Mishra 2013), spatial prediction of flood vulnerable areas using GIS (Shafapour et al. 2013), flood vulnerability analysis and its verification (Tehrany et al. 2015a, b), GIS-based flood susceptibility assessment and its mapping in Iran (Khosravi et al. 2016), flood vulnerable mapping using GIS-based frequency ratio and weights-of-evidence models in Golastan Province of Iran (Rahmati et al. 2016a, b), flood susceptibility analysis through remote sensing and GIS-based frequency ratio (Samanta et al. 2018). There are many researches which had also revealed the capability of remote sensing and GIS in flood risk assessment.

Instead of using only spatial database, when different statistical techniques are involved along with GIS, the result become more accurate, logical and acceptable. Some of the mostly applied popular techniques along with GIS for vulnerable mapping are analytical hierarchy process (Chen et al. 2010; Ali and Ahmad 2018, 2019), frequency ratio model (Lee et al. 2012a; Tehrany et al. 2015a; Rahmati et al. 2016a, b), artificial neural network (Kia et al. 2012), logistic regression (Pradhan 2010), fuzzy logic (Pourghasemi et al. 2012), support vector machine (Tehrany et al. 2015a), decision tree (Tehrany et al. 2013), Shannon’s entropy model (Haghizadeh et al. 2017), etc. In the present study, AHP and FR model were used along with spatial datasets for flood vulnerability analysis. AHP was considered because it is a quite cost-effective, easy to understand and widely used technique (Naghadehi et al. 2009; Bunruamkaewa and Murayamaa 2011; Tavares et al. 2011; Zou et al. 2013; Eskandari et al. 2016; Guler and omralıoglu 2017; Ali and Ahmad 2018, 2019). AHP is based on the expert’s knowledge on decision making process. Consequently, Frequency Ratio is one of the important quantitative methods of bivariate statistical technique as well as accepted to apply in the study of natural hazards (Lee et al. 2012a; Mondal and Maiti 2013; Tehrany et al. 2015a, b; Khosravi et al. 2016; Rahmati et al. 2016a, b; Samanta et al. 2018).

The historical records evidenced that since 1990 to 2017, there were three severe and seven gentle flood incidences that hit the Sundarban region of India with loss of natural resources and thousand lives. Giving emphasize to this issue, effort was made to study about such aspect which help for recognising and predicting more vulnerable areas to flood. The main objective of present study was to establish and apply quantitative technique in combination with GIS for flood vulnerable mapping and estimate areas under risk at Sundarban region, West Bengal, India. Therefore, the study used multi-criteria decision tool and bivariate statistical technique with GIS to detect and spatial map of flood risk zones in Sundarban region. Different topographical, climatic and local factors were selected as input data layer in GIS environment and AHP & FR model was applied to estimate potential flood prone areas. This effort will very much helpful in impact assessment which can be beneficial to local government administrators, researchers and planners for predicting future flood risk zone and developing flood mitigation strategies.

Description of the study area

Sundarban is the World’s largest mangrove forest in delta region created by Ganges, Brahmaputra and Meghna rivers in the Bay of Bengal. The Sundarban region is spread over India and Bangladesh covering an area of around 25,000 km2, out of which 62% lies within Bangladesh and 38% in India (Ghosh et al. 2015). The Indian Sundarban covers about 9630 km2. Out of them 4264 km2 come under wetlands and dense mangrove forest while remaining 5366 km2 area is occupied by inhabited lands (Hazra et al. 2002). The study area is bounded by river Muriganga on the west and by rivers Harinbhahga and Raimangal on the east, while the Bay of Bengal located in the southern portion. It is the part of tide affected lower deltaic plain of Bay of Bengal. It lies between 21°40ʹ00″N to 22°35ʹ40″N and 88°01ʹ30″E to 89°05ʹ00ʹE (Fig. 1). Sundarban region has 19 blocks, out of which there are 13 blocks in South 24 Parganas and 6 blocks in North 24 Parganas districts of West Bengal, India.

Fig. 1
figure 1

Location map of Sundarban region, West Bengal, India

Full size image

The coastal location and geo-climatic characteristics of Sundarban region are mainly responsible for occurrences ofdisturbing natural hazards and calamities. The average annual temperature is about 35 °C which varies from 20 to 40 °C during winter to summer. The summer (pre-monsoon) extends from the middle of March to mid-June, and the winter (post-monsoon) from mid-November to February. Climatic phenomena with regular cyclonic depressions and storm with heavy rain are common from mid-March to mid-September. Monsoon is the main culprit in occurrence of heavy rainfall and upland flow. Average annual rainfall is about 1920 mm with about 11–25 mm mean deviation. Average humidity is about 82% and is more or less uniform throughout the year. The people residing in Sundarban region occupationally depend on the land, water and forest. Although agriculture and aquaculture are the main source of livelihood for the people living here, unavailability of fresh water in rivers and problems of brackishness in inland water sources make agriculture unfavourable and uncertain. Generally, the inhabitants are directly dependent on primary occupations, such as agriculture, fishing, wood and honey collection, etc. Due to geographical location, geo-physical characteristics and climatic factors, sundarban region is highly vulnerable to natural hazards. Cyclone and flood are two common natural hazards that hit concomitantly year after year. Different water-related threats viz. sea level rise, salinization of soil and water and flooding have made the Indian Sundarban a challenging place for human life (Sánchez-Triana et al. 2018).

The population of the Sundarban rose from nearly 1.2 million in 1951 to 4.4 million people in 2011 (Hazra et al. 2002; Sánchez-Triana et al. 2014). Out of more than hundred islands, 54 are currently inhabited with varying degrees (Danda 2010). Once the entire Sundarban was covered by dense mangrove forest, but recently much of its area has been converted into other land uses like aquaculture and agriculture (Ghosh et al. 2015). The embankment system that was planned to protect the Sundarban from the sea water and flood is not working well as study found that flood occurred due to storm surges during increasingly high tides (Kay et al. 2015). Mangrove forests are believed to provide more protection from storm events. Thus, population increase leads to local land use change, decrease in mangrove forest, increased aquaculture and human interference with natural ecosystem make this world heritage sites more vulnerable to natural hazards.

Materials and methods

Study design

Looking at these issues and hazards phenomena of Sundarban region, a study was planned to assess the flood-risk zones for future flood preparedness. The main objective of this study was to identify the flood vulnerable area in Sundarban region of India based on geospatial information and statistical model. Sundarban region was selected for this study, because it is the most flood vulnerable area in West Bengal and many general to severe flood incidences hit this area due to its coastal location and geo-climatic features. The plan and methodology designed to achieve such objective is presented in the following figure (Fig. 2). First of all a flood inventory map was prepared based on past flood data and field observation. Flood location shape files was created in GIS environment as training and testing points to correlate with flood conditioning factors and validate the result. Looking towards flood incidences, different types of topographical, climatic and local factors (considered as Decision factors) were considered for vulnerability assessment.

Fig. 2
figure 2

The plan and methodology used for flood vulnerable mapping and risk area estimation

Full size image

Analytic hierarchy process (AHP) was applied to measure the selected factor weight value (SFWV) of each considered factor and Frequency Ratio (FR) model was used to measure the degree of selected class weight value (SCWV) of these factors. The spatial data were converted into raster format from vector one and reclassified accordingly into different classes to put the class weight which was calculated using FR model. Based on SFWV and SCWV, the final flood vulnerable index (FVI) and flood vulnerability map were prepared. Finally, flood testing and training points were merged into FVI map and prediction accuracy was calculated.

Data source

The flood risk analysis and vulnerable mapping in Sundarban has been done using satellite images, digital elevation model (DEM), hydrological data and data from field survey. Satellite images include Landsat 8 (OLI & TIRS) and the SRTM DEM. Landsat 8 images of 26th November 2017 were collected from USGS Earth Explorer (https://earthexplorer.usgs.gov) which provides geometrically and radiometrically corrected satellite images of different time period throughout the globe in TIFF file format (.tif) and is in UTM projection by default. The data were selected with 0% cloud cover during download. The topographical data, i.e. SRTM (Shuttle Radar Topographic Mission) DEM was collected from Remote Pixel (http://remotepixel.ca) in geographic coordinate system-WGS84 by default. The annual rain gauge wise rainfall data of the study area were collected from Annual Flood Report, Irrigation and waterways Directorate, Govt. of West Bengal for the year 2012–2017. Survey of India (SoI) topographical sheet, Department of Disaster Management, the official portal of the Government of West Bengal (http://wbdmd.gov.in) and Google Earth Professional version were used for correction and preparation of base map of the study area and area of interest (AOI) was extracted by mask from satellite images. The major datasets, their sources and main purpose of uses are presented in the following table (Table 1).

Table 1 The sources from where different data were collected and their purpose of application
Full size table

Preparation of flood inventory map

In order to perform vulnerabilit analysis through AHP and FR model, it is required to prepare an inventory map of past floods. Flood inventory means the spatial distribution of flood locations in an area. Preparation of flood inventory map is a crucial part for flood vulnerability analysis or future prediction of flood. The future risk zonation of flood in an area is estimated by analysing the present and past flood events (Ohlmacher and Davis 2003). Hence, inventory map is essential for accurate vulnerable analysis of flood (Rahmati et al. 2016a, b). In the present study, the flood inventory map was prepared for Sundarban region using the annual flood report, Irrigation and waterways Directorate, Govt. of West Bengal, 2014 and personal field observations during 2017. The flood-affected locations were marked during field observations and mapped by utilising GPS points, satellite image (Landsat 8 OLI & TIRS), SRTM DEM data, SAS Planet (ArcGIS image) and Google Earth (Fig. 3).

Fig. 3
figure 3

Flood inventory map showing flood locations which used for training and testing

Full size image

The flood inventory map contains a total of 270 flood locations which were randomly divided into two data sets viz. 200 flood locations (75%) for training and 70 flood locations (25%) for testing or validation (Tunusluoglu et al. 2008; Pradhan 2010; Pradhan and Youssef 2010; Mondal and Maiti 2013; Pourtaghi and Pourghasemi 2015; Rahmati et al. 2016a, b). Finally, the flood inducing factors were incorporated with this flood inventory map to make out the importance of each factor in flood occurring.

Determination of flood inducing factors

The most vital stage of any risk assessment and zonation is to select appropriate factors or decision criteria to get suitable and best results. The natural hazards like flood, landslide, cyclone, etc. are largely depending on many factors for their occurring. Selection of effective factors is vital to make a flood hazard map in any catchment area (Kia et al. 2012). It is always considered as a complex task to select parameters consistently to produce flood susceptibility mapping (Tehrany et al. 2014). Therefore, to determine the most appropriate flood triggering factors, first of all a field survey was carried out. During the 30-day field visit in 2017, the most flood vulnerable areas were visited and personal views of local people were noted which were very useful for preparing the inventory map.

It is definite that rainfall is the most important factor among all others in creating flood in a region, but there are many other contributing factors also responsible with rainfall. The land elevation and slope play always a significant role along with rainfall in occurring flood. When rains fall on a catchment area, the characteristics of catchment like its shape, size, capacity of carrying water load, soil types, vegetation cover, etc. largely governed the amount of rainwater that reaches the waterways. Some rainfall is infiltrated by soil and vegetation, and the rest enters waterways as run-off. Thus, along with rainfall there are many flood causative factor. Looking at such aspects, previous works and local circumstances, seven important spatial factors including rainfall deviation, land elevation, slope angle, topographic wetness index, land use land cover, clay content in soil and distance from rivers were selected to prepare vulnerability map for sundarban region, India. The flood inducing factors were broadly categorised as climatic factors, topographic factors and local factors. As climatic phenomena always play significant role in creating flood. In the present study, rainfall deviation was taken as a main causative factor for spatial vulnerability analysis of flood.

Rainfall deviation: Heavy rainfall and flood occurrence are positively correlated. Havoc rainfall causes flash flood. For the study area, the intensity of rainfall increases from July to September and that is way these months are recorded as flooding months, i.e. monsoon session in India. Surface runoff due to heavy rainfall is very significant for estimating floods which are very rapid, flashy and of short duration (Pal and Samanta 2011). Here, the rainfall deviation was considered as a triggering factor for flood susceptibility analysis; because positive deviation of rainfall indicates sufficient of rainfall and possibility of flood concomitantly; negative deviation of rainfall indicates deficit of rainfall and possibility of drought. The average and recorded annual rainfall of Sundarban region was taken from 2009–2017 for calculating the deviation and spatial mapping. The rainfall deviation was calculated from average and recorded rainfall of each rain gauge station using the following equation;

$$Q = \left( { \frac{{\left( {L - Z} \right)*100}}{Z}} \right),$$

where Q is rainfall deviation, L is recorded rainfall, Z is average rainfall.

Based on calculated rainfall deviation data of five rain gauge stations, i.e. Uttarbag, Canning, Nimpith, Raidighi and Kachuberia, IDW (Inverse Distance Weight) interpolation technique was used in GIS environment to prepare the rainfall deviation map of the study area (Fig. 4).

Fig. 4
figure 4

Rainfall deviation

Full size image

Along with rainfall, topographic factors like land elevation, slope, aspect of slope, etc. also play significant role in csusing flood. Topographic data were extracted from DEM of the study area. The land elevation, slope angle and topographic wetness index were taken into consideration for the present study. Study area was extracted by mosaicking N21E087, N21E088, N21E089, N22E088 and N22E89 tiles of SRTM-DEM. By default DEM is in geographic coordinate system (D_WGS_1984 datum), but to extract the correct topographic features, it is required to convert in projected coordinate system (UTM_Zone_45 N). Thus, raster project system was used for the same.

Land elevation: The land elevation and degree of surface inclination are important factors which control the flood phenomena in a region. Elevations have always a vital role in preparing flood susceptibility maps (Rahmati et al. 2016b). Different elevations directly or indirectly influence the climatic characteristics in a certain region (Samanta et al. 2012). Different elevations make differentiations in receiving different amount of rainfall and temperature. Different rainfall and temperature caused varied soil forms and vegetation cover (Aniya 1985). Land elevation always has a very significant role in flood susceptibility mapping (Fernandez and Lutz 2010). Elevation map of Sundarban region was prepared in GIS environment (Fig. 5).

Fig. 5
figure 5

Land elevation

Full size image

Slope angle: There are many geological factors like local topography including slope, slope aspect, hill shade, etc. that also play important role in occurrence of flood in a region (Mishra 2013). As slope has a direct effect on surface runoff and propagates the process of infiltration, it is considered as a major factor in flood occurrence (Haghizadeh et al. 2017). Slope controls the surface runoff, the strength of water flow and process of soil erosion (Adiat et al. 2012) as well as vertical infiltration of water (Youssef et al. 2011). Same as elevation, the slope angle of the study area was also prepared and reclassified into different risk classes (Fig. 6).

Fig. 6
figure 6

slope of Sundarban region

Full size image

Topographic wetness index (TWI): The TWI, also called Compound Topographic Index (CTI) is a steady-state wetness index. TWI map defines the effect of topography on the amount of wet levels to form runoffs and it is considered as a suitable index in investigating flood potential of watersheds (Rahmati et al. 2016a, b; Haghizadeh et al. 2017). Topographic wetness index refers to spatial distribution of wetness and controls the overland flow of water (Samanta et al. 2018). In the present study, TWI was calculated using DEM through involving the upslope contributing area (a), slope raster and a pair of raster calculator. The following equation was run in raster calculator of GIS environment to generate the TWI map (Fig. 7):

$${\text{TWI}} = \frac{{{\text{Ln}}(\left( {{\text{``FLOWACC''}}*900} \right)}}{{{\text{Tan}}\left( {{\text{``SLOPE''}}} \right))}},$$

where FLOWACC is the flow accumulation raster which was created from flow direction raster and SLOPE is created by slope raster in degree.

Fig. 7
figure 7

Topographic wetness index

Full size image

Beside the climatic and topographic factors there are many local factors which are also responsible for occurring flood in a region. These local factors are varying with different geographical and geo-climatic region. Due to coastal location of Sundarban region, three local factors including LULC, clay content in soil and distance from river were also selected in flood vulnerability analysis.

Land use land cover (LULC): Land use categories play an important role in flooding because the hydrological processes including infiltration, evapotranspiration and run-off generation also governed by land use types (Rahmati et al. 2016a, b). Different type of land use controls the surface runoff and flood potential. Residential areas are mostly made by impermeable surfaces which increase the risk of run-off and inundation but vegetation covers are less prone to flooding because infiltration capability and vegetation density are positively correlated (Tehrany et al. 2013). Therefore, LULC is considered as a main local factor in flood vulnerable mapping in Sundarban region. LULC map was derived from Landsat 8 OLI & TIRS. Geometrically and radiometrically corrected Landsat 8 satellite data were collected from USGS Earth Explorer (http://earthexplorer.usgs.gov). Bands 7, 6, 4 (30 m spatial resolution) in RGB composite were used for FCC. Supervised classification was run in GIS environment based on created signature of detected surface objects (Fig. 8).

Fig. 8
figure 8

Land use land cover of Sundarban region

Full size image

Content of clay in soil: The presence of clay content in soil plays an important role in flood analysis. Soils comprising large proportion of sand and silt tend to form hard compacted surface, which generally lessens the rate of infiltration. The soil with high clay content and low organic matter is more easily compacted than with low clay content and high organic matter, which reduce the infiltration rate and increase runoff or overland flow (Rattan 1990). Giving emphasis on that point, clay content in soil was considered in the present study. Sample data from flood affected blocks of Sundarban were collected from Annual Flood Report, Irrigation and waterways Directorate, Govt. of West Bengal. Spatial data were created using Google Earth and GIS environment. For spatial mapping, interpolation was used (Fig. 9).

Fig. 9
figure 9

Content of clay in soil

Full size image

Distance from rivers: Those areas located nearer to river are more vulnerable to flood and its hazardous impacts are greater than areas located far. Flood generally takes place near the bank of river and inundates low-lying flood plain areas (Samanta et al. 2018). Certain distance from rivers has weighty impact on the flood and its extent (Glenn et al. 2012). Thus, distance from major rivers in Sundarban region was considered as one of the parameters in this study which was developed through proximity analysis in GIS environment (Fig. 10).

Fig. 10
figure 10

Distance from river

Full size image

Flood vulnerability analysis

In order to map flood vulnerability and spatial analysis, topographical, climatic and local factors were selected. Different criteria based spatial maps were created in both feature and raster format. The selected vector or feature layers were converted into raster layer (V2R) in order to put their rate and value of significance. The AHP and FR models were considered for the same.

AHP and the selected factor weight value (SFWV)

The main principle in analytic hierarchy process is to construct a matrix expressing the relative significance of a set of selected alternatives. It helps decision maker for taking suitable decision based on its importance and significance in certain phenomena. In AHP the problem can be recognised as how to derive weights, rankings or importance in a set of alternatives according to their value for occurring in some instances. It is the wide applicable technique of multi-criteria decision making (MCDM) problems. AHP technique was used in the present study to support the decision through giving comparison rank to the selected flood inducing factors by estimating the selected factor weight value (SFWV). To know the relative importance of different flood occurring factors in Sundarban region and to put their rank based on given preference, interview survey with local residents and field observation were accomplished. Then one single factor is compared with other factor in terms of whether it is very much more important, rather more important, as more important and so on down to very less important. Each of these factors was judged based on a numerical scale adopted from Saaty (Table 2).

Table 2 Preferred numerical scale to compare two flood inducing factors
Full size table

Based on the importance or value of the flood occurring factors, pair-wise comparison matrix was constructed. Out of total seven selected flood inducing factors, rank value was assigned to each of them on the scale ranging 1–9 based on their relative importance in this incidence. In matrix, the rank value of one alternative is reciprocal (i.e. 1/2 to 1/9) to its inverse comparison (Table 3). AHP was used to assemble flood causing factors hierarchically and to decide the selected factor weight value, i.e. eigenvector which corrected through measuring consistency ratio (CR) based on Saaty (1977, 1980) and Saaty and Vargas (2000).

Table 3 Flood inducing factors and their selected factor weight value (SFWV) in flood susceptibility mapping
Full size table

After relative rank is fitted to each factor, the factor weight values have to be calculated for classified sub-factors to judge the consistency in taking the scale of importance into consideration. Hence, the eigenvector was calculated by considering the following equation:

$$Ax \, = \, \lambda {\max}^{X} ,$$

where A is the comparison matrix of n number of criteria, x is the eigenvector of n size of criteria and λ is the Eigenvalue.

The eigenvalues help in measuring consistency in a set of pairwise rankings. For a consistent reciprocal matrix, the largest eigenvalue (λmax) is equal to the number of comparisons n. Hence, the Consistency Ratio (CR) is compulsory to calculate for the same. Saaty suggested that if the CR exceeds ‘0.1’, the set of decision is considered as ‘inconsistent’ and has to repeat again. Concomitantly, if CR is equal to ‘0’, it means the decision is perfectly consistent but the value between 0 and 0.1 is also considered as consistent (Saaty 1990). Consistency Ratio was calculated using the following equation:

$${\text{CR}} = \left( {\frac{{{\text{CI}} }}{\text{RI}}} \right),$$

where CR is the consistency ratio, CI is the consistency index and RI is the random index. However, RI was utilised from Saaty (1990) and CI was calculated by putting the values from above equations by using the following simple equation:

$${\text{CI}} = \left( {\frac{{\lambda_{\text{max} } - N }}{N - 1}} \right),$$

where λmax = average of x criteria and N is the total number of sub-factors.

AHP has the capability of application in diversified fields where many factors are responsible for occurrence of an incidence viz. assessment for ecotourism (Bunruamkaewa and Murayamaa 2011); for assessment of residential land use suitability (Patil et al. 2012); for industrial site selection (Rikalovic et al. 2014); in the selection of post-harvest technology (Montenegroa et al. 2014); in irrigation networks maintenance (Permana and Hadiani 2017); in selection of best underground mining technique (Naghadehi et al. 2009); for diseases transmission and risk mapping (Ali and Ahmad 2018, 2019) and many other studies of interest.

FR model and the selected class weight value (SCWV)

Frequency ratio is one of the important quantitative methods of bivariate analysis and well accepted to apply in flood susceptibility analysis. FR as a bivariate statistical analysis is based on the spatial correlation between dependent and independent factors. In the present study, the spatial correlation was calculated between flood training points as dependent factor and flood inducing determinants (topographic, climatic and local factors) as independent factors. Frequency ratio model was successfully applied in flood susceptibility and vulnerability analysis in different flood prone regions of the earth (Khosravi et al. 2016; Rahmati et al. 2016a, b; Samanta et al. 2018). Thus, in this study to calculate the FR for each class of all selected factors the following equation was used:

$${\text{FR}} = \frac{{\left( {P{\text{pix}}E / P{\text{pix}}T} \right) }}{{\left( {\varSigma {\text{pix}}E / \varSigma P{\text{pix}}T} \right)}},$$

where PpixE is the number of pixels containing flood points in class p, PpixT is the total number of pixels having in class p in the study area, ΣpixE is the total number of pixels containing flood points in class p and ΣPpixT is the total number of pixels in the in the study area.

If the outcome value of frequency ratio (FR) is more > 1.0, it indicates positive and strong relation between flood training points and the concerned class of the particular factor and high risk to flood, whereas a value of frequency ratio < 1.0 depicts negative relation and low significance to flood risk (Pradhan 2010; Lee et al. 2012a; Mondal and Maiti 2013; Rahmati et al. 2016a, b; Samanta et al. 2018). In the present study, the frequency ratio value for each class is considered as selected class weight value (SCWV). The flood vulnerable index (FVI) was also estimated in present study from AHP and FR model to show stretched value of flood vulnerability from high to low flood risk areas. Selected class weight value (SCWV) of each class of all selected factors and the selected factor weight value (SFWV) that has been chosen for flood occurring were taken into consideration to calculate the FVI using the following equation:

$${\text{FSI}} = \mathop \sum \limits_{n = 1}^{n} (w_{i} *{\text{FR}}),$$

where n is the total number of selected factors (n = 7 in present study), wi is the weight of factors (i.e. SFWV) and FR is the frequency ratio value of each class (i.e. SCWV).

Results and discussion

Flood vulnerability analysis is a planning and pre-hazards management approach that is required to reduce the risk. Due to coastal location, Sundarban region is always considered as a flood prone area. There are many evidences of devastating flood in this region with some interval and return period. This study highlighted the analysis of flood vulnerability based on decision making approach, i.e. analytic hierarchy process and bivariate statistical technique (frequency ratio). There are many floods inducing/conditioning independent factors which have significant role on flood assessment. Thus, statistical database for all seven selected conditioning factors (land elevation, slope, topographic wetness index, rainfall deviation, land use land cover, clay content in soil and distance from rivers) were prepared with respective sub-classes (Table 4). Their spatial relationship with flood vulnerability is discussed below along with measurement of accuracy.

Table 4 Class frequency ratio of flood inducing factors (SCWV) for flood vulnerability mapping
Full size table

Relationships among flood vulnerability and flood inducing factors

Effort was made to spatially correlate between flood vulnerable areas and flood inducing factors. Factor weight value and class weight value were considered in this regard. Factor weight value indicates the relative importance of each factor that was selected and calculated from AHP. Class weigh value indicates the relative importance of particular classes for each factor which offers important information for considering the role in flood creating. The frequency ratio of individual class for each factor is presented in Table 4. In climatic factor, rainfall always plays a significant role in flood vulnerability analysis. Rainfall deviation was taken to identify the probability to flood or drought, because deviation of rainfall is considered as best indicator for flood and drought (positive range indicates more rainfall than normal and negative range indicates less than normal rainfall). It was found from Table 4 and Fig. 4 that rainfall deviation between 11 and 15 mm has < 1 FR value, whereas 15–19 mm deviation has always > 1 FR value and these areas are more vulnerable to flood in comparison to low rainfall deviated areas.

The land elevation of Sundarban region is varied between < 2 and > 6 m. The spatial elevation map was prepared through extracting elevation raster value of digital elevation model which was classified into five classes using natural breaks interval. The elevation of core area covered by dense mangrove forest and Bay of Bengal was excluded from study of interest and only elevation of inhabited parts was considered. The FR values of land elevation between 2 and 5 m are > 1 which indicate positive relation with flood vulnerability. Slope angle of the study area varies from 0 to 34.15°. The calculated value of FR in slope gradient between 4.58° and 43.15° was found to be 4.91 which indicates this portion is very highly vulnerable to flood. On the other hand, FR value of low slope gradient was found < 1, which means this portion is very low susceptible to flood. The spatial database on topographic wetness index was reclassified into five classes (2.10–7.18, 7.19–9.61, 9.62–11.56, 11.57–13.92, 13.93–19.90). Higher TWI value denotes higher probabilities of flooding in a region. The highest FR value was calculated at 3.17 where the TWI varies between 13.93 and 19.90 and lowest FR value was 0.3271where TWI varies between 2.10 and 7.18.

Among different types of local factor, LULC plays a vital role in flood vulnerable analysis. Based on classification of satellite data, five land use land cover categories were considered, namely water bodies, water logged areas, vegetation, agriculture or crop land and settlements. Among these classification settlements agriculture was considered as much more vulnerable in comparison to other land cover types. The prominent land cover type in the study is agriculture including crop land and aquaculture which covered about 33% followed by vegetation, water logged areas, inland water bodies and settlement. The highest FR value was found as 3.74 and 0.75 for settlement and agricultural lands, respectively, and lowest FR value was found as 0.36 and 0.51 for water logged areas and vegetation, respectively. The factor weight value (SFWV) for LULC was calculated as 14% compared to other SFWV factor. Content of clay in soil was also considered as a weighted factor in this study. As soil with high clay content is more easily compacted than with low clay content which reduces the infiltration rate and increases runoff or overland flow and thatiss why the soil content has been given 10% factor weightage value. The findings show that high content of clay, i.e. 32.41–36.00% in soil has FR value > 1. The intensity of flood became higher in those areas near to the river bank and certainly less in those locations far away from the river and the risk was lesser than those nearer. The proximity analysis was done to create certain interval of distance from river. The result found that the distance from river to < 2, 4, 6, 8, and > 8 km has the FR value of 1.26, 1.21, 0.54, 0.56 and 0.48, respectively, which indicates that away from river having lesser FR value and least risk to flood.

Flood vulnerable mapping and risk area estimation

First of all, based on factor weight value and class weight value from AHP and FR analysis, final susceptible zones were created in GIS environment through overlay analysis. The SCWV of each sub-class of all selected factors as shown in Table 4 was used for the same. Then, flood vulnerability index (FVI) was calculated using summing the FR value of each selected flood inducing factor. The greater value of FVI means higher susceptibility to flood occurrence. Concurrently, lower value of FVI indicates less vulnerable to flood incidence. Basically, the FVI database was reclassified into five susceptible zones in order to know the spatial flood vulnerability zones of Sundarban region. The output zones were categorised into very high, high, medium, low and very low with covering area of 1273.88, 279.56, 1690.85, 588.65 and 1531.45 km2 respectively (Fig. 11 and Table 5). Some badly affected villages were also identified in the study area during field visit (Fig. 12). Areas along the river side of Gosaba, Namkhana, Patherpatima, Sandeshkali-II, Hingalganj, Kultali and Sagar were identified as very high to high vulnerable zones. On the other hand, western parts of Haroa, Canning-I, Canning-II, Jaynagar-I and Sagar were identified as low to very low, susceptible to flood. Very high to high flood vulnerability classes indicate that these areas would be more prone to fresh flood incidences for future.

Fig. 11
figure 11

Flood vulnerable map of Sundarban region retrieved using AHP and FR model

Full size image
Table 5 Spatial distribution of flood vulnerable classes and areas
Full size table
Fig. 12
figure 12

Showing some pictures during flood and after flood in Sundarban region, India. a The flood water overflowing on the road b the paddy field in Gosaba Block has submerged in water c the residential houses have broken along the canal side (d) and (e) sign of losses on human habitation: condition of residential households in Patherpatima block after few months of flood (f) A house from Sandeshkali-II after 6 months of flood

Full size image

Validation

It is really a crucial task in risk assessment and vulnerability analysis to measure the accuracy and validation of retrieved result. Accuracy assessment provides the validation of study and model applied. There are various techniques including area under curve (AUC), success rate (SR), prediction accuracy (PA), etc. for assessing the accuracy and validation (Chung and Fabbri 2003; Tien Bui et al. 2012; Mondal and Maiti 2013). In the present study PA and SR were considered to measure the accuracy through flood training and testing points which were calculated using simple equation;

$${\text{PA }} = \, \varSigma {\text{ap}}/\varSigma {\text{tp,}}$$
$${\text{SR}} = \varSigma {\text{sp}}/\varSigma {\text{tnp,}}$$

where, PA prediction accuracy, SR success rate, ap accurate testing points which were considered for flood vulnerability; tp total flood testing points, sp considered training points for success rate, tnp total training points.

Using the above equation, if the value of accuracy comes perfectly 1.0, it means completely accurate and the capability of model in prediction was done without considering any biasness, but the value > 0.75 is also considered as standard (Pradhan et al. 2010). In present study, prediction accuracy was calculated using 70 (25%) flood locations which were not taken during FR modelling and others 200 (70%) flood locations were used to calculate the success rate. For the same, class ranges from moderate to very high vulnerability were taken as potential flood zones which might be occurring in future. Success rate and prediction accuracy were calculated as 0.8450 and 0.8142 respectively (Table 6). Therefore, the prediction accuracy verified > 80% which validated the frequency ratio model that was used in flood vulnerability analysis for Sundarban region.

Table 6 Calculation of prediction accuracy and success rate for the flood susceptibility mapping
Full size table

Conclusion

The exercise of flood vulnerable mapping is done to define those areas which are under risk of flooding. Flood vulnerable map is an important aspect for planning suitable land use in flood prone areas. The main objective to design flood vulnerable map was to increase awareness of the possibility of flooding among the public, local authorities and other organisations. The study aimed to identify areas at risk of flooding which helps prioritise mitigation and response efforts. Geospatial technique and statistical method was used for the same. Based on selected flood inducing factors for analysing flood vulnerability, AHP was used to derive the maximum to minimum weightage rate of selected factors and FR model was applied to find the relationship between past flood incidences and possibility of future flood occurrence based on 200 flooded points. These flood point locations were shaped using datasets from historical flood records, satellite images and personal field survey. Random sampling method was used to select 75% input flood data for training to calculate frequency ratio of subclasses of each factor and 25% input flood data were used for validation. From the study analysis it was found that selected factor weight values of flood inducing factors were high for distance from river (0.3391), rainfall deviation (0.2144), land use land cover (0.1480) and content of clay in soil (0.1191) which indicated that these are most significant flood causing factors for Sundarban region. The SFWV of other contributing factors were 0.0850 for topographic wetness index, 0.0563 for slope angle and 0.0377 for land elevation which has also some role for flooding. The frequency ratio model evaluates the significance of each sub-class of individual factor which ranges from 0.08 to 4.91 indicates very low (< 1) to highly strong (> 1) significant in flood vulnerable mapping. These semi-quantitative and bivariate statistical methods are useful for supporting decision making for efficient flood management. The present study showed that climatic (rainfall) and local factors (distance from river, LULC and clay content in soil) have much more important contribution in comparison to topographic factor (land elevation and slope) because Sundarban region is flat coastal plain. The result of validation based on training and testing flood location points revealed that the prediction accuracy was 81.42% and success rate was 84.50% which may consider validating the FR model that applied in present study.

Flood vulnerable mapping is a difficult task to prepare and undertake at the community level because suitable numerical model is required for forecasting such extreme events. The main barrier for flood vulnerability analysis is lack of availability and accurate long-term extreme event data and suitable numerical model. Along with public awareness also create a barrier to implementation plan. But with developing and using suitable technique for vulnerability analysis, the risk area can be predicted and identified which has great strengths than other adaptation technique like flood proofing measures, flood emergency planning, facility of flood shelters and evacuation planning because vulnerability analysis is a flood preparedness plan. Therefore, different multi-criteria based numerical model could be applied globally to lessen the burden of flood hazard.