A comparative evaluation of GIS based flood susceptibility models: a case of Kopai river basin, Eastern India

Each year, floods are one of the most common natural disasters, wreaking havoc on ecosystems, human settlements, and individual lives all around the globe. Numerous factors, including flood intensity, amplitude, frequency, flow duration, changes in river course and geometry, etc., affect flooding's effectiveness. Shannon's entropy, frequency ratio, and weight of evidence have been used to identify flood susceptibility in the Kopai river basin. According to the results of these three models, the upper reach of the basin is comprised of non-flooded areas, while the lower part of the basin is home to extensive flooding. This makes the lower part of the basin more susceptible than the upper part. The final result of flood susceptibility zones is obtained using 11 thematic layers such as rainfall, soil moisture index, soil types, surface roughness,distance to the river, elevation, slope, drainage density, land use and land cover, normalised difference vegetation index, and normalized difference water index. Receiver operating characteristics curves have been produced using test data consisting of 370 data points to determine the validation of the models. The validation result demonstrates that the frequency ratio is 96.5% accurate. The weight of the evidence is 97.1% reliable, whereas Shannon entropy provides just 91.2% acuracy. Weight of evidence is therefore the best model for flood susceptibility zones identification for this Kopai river basin.


Introduction
Around the world, many natural and man-made disasters and catastrophes, such as floods, earthquakes, and landslides, are observed each and every year (Youssef et al. 2011;Du et al. 2013;Tehrany et al. 2014a).Recent years, however, have seen a sharp rise in this pattern as a result of several factors, including but not limited to: climate change, ecological degradation, rapid population growth, etc. (Caruso 2017).One of the most frequent and deadly of these is flooding, which annually claims thousands of lives and causes billions of dollars in damage in countries all over the world (Kowalzig 2008;Kourgialas and Karatzas 2011).Floods, on top of these other causes, also have a major negative impact on ecological systems.(Rahmati et al. 2015(Rahmati et al. , 2016;;Billa et al. 2006).This natural calamity is responsible for 31% of the global economy's destruction.In general, floods take place when considerable quantities of water surpass the normal limits, the riverbed becomes submerged, and this generates stagnant water during the fleeting season.
Asia is one of the places that is most susceptible to the repercussions of floods, which may be caused by a number of different natural and manmade reasons.Floods are responsible for a significant amount of human suffering in addition to the economic damage they inflict on a community.This is due to the fact that the real estate market is sensitive to the frequency and severity of flood catastrophes due to the destruction of infrastructure and the threat that it poses to people's lives (Balogun et al. 2021).In many countries, environmental and water management necessitate modelling of food susceptibility in order to provide early warning to the public about this danger and reduce the amount of harm that will occur (El-Haddad et al. 2021).The frequency of floods in 591 Page 2 of 18 India far exceeds that of any other natural calamity.Over 40 million hectares of land in this country are at risk of flooding, roughly one-eighth of the country's total land area (Gupta et al. 2003;Mohapatra and Singh 2003).The amount of land in India that is at risk of flooding has been steadily increasing, at a pace of 0.14 million hectares per year (Singh and Kumar 2013).Northern and northeastern states are more at risk of flooding due to the Ganga and Brahmaputra river basins.As reported by Kumar et al. (2005), many floods happen during the monsoon because of the seasonal and temporal fluctuations in rainfall patterns that cause rivers to discharge at high rates.Other reasons that contribute to the occurrence of floods include siltation of riverbeds, inadequate capacity of river banks to retain high flow, alteration of river course, failure of dams, poor drainage conditions in flood-prone regions, and glacial outbursts.About 30 million people in this country are impacted by floods annually, with over 1,500 lives lost.In 2003, Gupta and colleagues published a study proving this.India ranks second in terms of flooding, behind only Bangladesh, owing to its low elevation, narrow river beds, seasonal rain changes, large sediment flow, and other similar factors (Sarkar and Mondal 2020).As a result, identifying flood-prone areas is the primary responsibility of scientists in reducing human suffering and material destruction caused by floods.Numerous studies have been conducted by scientists to categorise and identify different types of floods and their immediate and long-term consequences.
The term "flood sensitivity mapping" is used to describe the process of evaluating and categorising the geographical distribution of floods, both those that now exist and those that might happen in a given location.As a result, tools for mapping areas at risk of flooding and gauging their sensitivity to flooding might be useful to decision-makers in charge of setting standards and regulations (Mirzaei et al. 2021).Recognizing flood-prone areas may help considerably in preventing or limiting flood damage, even when floods themselves cannot be averted (Sahoo and Sreeja 2017).In light of this, assessing the vulnerability of infrastructure to flooding is an important step toward disaster mitigation, although one that will be difficult owing to the complexity introduced by the participation of so many different conditioning variables.Scientists now believe that many comprehensive models for estimating flood risks have come to rely on remote sensing and geographic information system (GIS) techniques, due to the high degree of accuracy they provide (Pradhan 2009;Bates 2012;Wanders et al. 2014;Nikoo et al. 2016).The vast majority of the models that have been employed in the past were primarily concentrated on the hydrodynamic model, the hydrological model, multi-criteria decision analysis (MCDA), statistical models (SM), and machine learning (ML) approaches, all of which are included in the geographic information system (Singh and Kumar 2013;Elkhrachy 2015;Vojtek and Vojteková 2016;Rosser et al. 2017;Samanta et al. 2018;Tiryaki and Karaca 2018;Liuzzo et al. 2019;Santos et al. 2019;Tehrany et al. 2019a;Shahabi et al. 2020).The most commonly used models and techniques concerning flood susceptibility mapping include frequency ratio (FR) (Rahmati et al. 2015;Shafapour Tehrany et al. 2017;Samanta et al. 2018;Tehrany et al. 2019a;Sarkar and Mondal 2020), analytical hierarchy process (AHP) (Elkhrachy 2015;Dahri and Abida 2017;Das 2018;Rahman et al. 2019;Sepehri et al. 2020), shannon's entropy (Khosravi and Pourghasemi 2016;Haghizadeh et al. 2017), weights of evidence (WoE) (Tehrany et al. 2014b;Rahmati et al. 2015;Shafapour Tehrany et al. 2017;Costache 2019), artificial neural networks (ANN) (Kia et al. 2012;Ruslan et al. 2013;Elsafi 2014;Rahman et al. 2019;Kordrostami et al. 2020), fuzzy logic (Nandalal and Ratnayake 2011;Sahana and Patel 2019;Sepehri et al. 2020), support vector machines (Tehrany et al. 2014b(Tehrany et al. , 2015(Tehrany et al. , 2019b;;Nandi et al. 2017), biogeography based optimization and BAT algorithms (Ahmadlou et al. 2018;Wang et al. 2019), adaptive neuro-fuzzy inference system (ANFIS) (Wang et al. 2019;Vafakhah 2020), reduced error pruning trees (Khosravi et al. 2018), multivariate adaptive regression splines (Tien et al. 2019), maximum entropy (Haghizadeh and Rhamti 2017), random forest (Lee et al. 2017;Reza et al. 2020) etc.
This research on flood risk assessment has been carried out using the frequency ratio (FR), weights of evidence (WoE), and shannon's entropy (SE) models.According to the cited research, these statistical models are the most effective for both flood risk assessment and landslide vulnerability mapping.Several variables were analysed to determine how susceptible a certain area was to flooding: soil type; NDVI (normalised difference vegetation index); NDWI (normalised difference water index); distance to river; elevation; slope; drainage density; land use and land cover; rainfall; SMI (soil moisture index); and surface roughness.Several researchers have used these criteria (Table 1) chosen for the flood susceptibility map to identify flood-prone areas.Since it would be impossible to accurately predict flood-prone areas using just one model, we opted for a more robust trio of models in this investigation.This is confirmed by a large body of literature, which also reveals that each model has its own set of benefits and drawbacks that are unique to its implementation location.Our primary goal here is to create a one-of-a-kind flood risk map for the region under investigation and to find the bestfitting model among the top-tier models used.
So many machine learning and statistical models have been applied in various research related to basin floods.But the Kopai river basin of eastern India is tiny and is mainly recharged by rainfall water during the monsoon season.As a result, a devastating flood occurs from June to September.In this regard, a perfect plan is required to manage the loss of properties.No such work has yet been done to identify flood susceptibility zones using machine learning methods for small basins like Kopai.It is the first research where machine learning and advanced spatial statistical models have been applied to detect flood susceptibility zones.
Moreover, it is the first where a comparison has been found between Machine learning and advanced spatial statistical models.So, a critical research gap has been tried to fill in this research.If we look at the novelty of this research, it has already been discussed that Kopai is a very small but significant river basin in eastern India.Most people ignore its importance, but at the micro level, this river effectively impacts the economy and society.That's why this research may represent a novel purpose for improving flood management plans and policies after demarcating its flood susceptibility character.

Materials and methods
This section explains the methodology used, the data collected, and the theoretical underpinnings of the various flood susceptibility modelling strategies.We created databases, such a flood vulnerability map and flood conditioning variables, using the cleaned and reviewed data used in model construction and validation.Bivariate models (FR, SE, and WoE) were used to evaluate flood hazard, and validation strategies (ROC and AUC) were selected to test the models' accuracy.

Location of the study area
The basin of the Kopai River is considered to be one of the most significant river basins in Rarh Bengal.The Khajuri in Jharkhand is considered to be the river's headwaters, while the Babla river in Murshidabad is considered to be its confluence.The overall length of the river is around 176 kms, and the total area of the basin is approximately 1,533 square kilometres.The basin may be extended longitudinally from 87°13′00" East to 88°09′30" East and latitudinally from 23°26′18" North to 23°56′30" North.Figure 1 is an illustration of the geographic position of the Kopai river basin.

Flood inventory map
Historical floods give useful information that may be used as a foundation for the next step in the process, which is to create a map of flood inventory.Predicting future flooding requires reliable information from past and present floods.Not only did the Desk Research, but even the field research was done.An inventory of flood risk zones was first compiled using satellite images of the area, and this was checked during field verification.satellite images analysis was used to locate and identify all 274 flooded locations in the research area.These coordinates were found and confirmed by an experiment conducted in the field.The locations of 274 flooded regions are shown in Fig. 2. From this random

Frequency ratio model
The Frequency Ratio (FR) is a helpful tool for developing flood-prone maps because it is an observation-based model with a clear and straightforward explanation, high accuracy, and widespread usage (Javad et al. 2014;Rasyid et al. 2016;Thapa and Bhandari 2019;Sarkar and Mondal 2020).
The FR method is used to determine the causal relationship between the many variables that contribute to floods by analysing the observed correlations between the distribution of flood incidence sites and these factors (Samanta et al. 2018;Aprilia et al. 2021;Munir et al. 2022).The mathematical interpretation of the FR model is as follows: where, N pix(Fi) , represents the number of pixels in class i that contain flood, and N pix(Ni) represents the total number of pixels in class i.The measurable value of FR indicates the strength of the relationship between flood and the group of conditioning factors under consideration.For flood risk, (1) a score over 1 indicates a strong and beneficial relationship, while a score below 1 suggests a weak and potentially dangerous interaction and low flood risk.The Flood Susceptibility Index (FSI) was then calculated by summing the FR for each class using the provided formula equation.

Shannon's entropy model
In a framework, entropy is a mechanism used to prevent anomalies, inconsistencies, disruptions, and unpredictability.The entropy model, first proposed by Stephan Boltzmann to evaluate the degree of randomness in thermodynamic systems, was extended by Claude Shannon in 1948 and has subsequently been applied to the study of the spread of numerical variables (Shannon 1994).Utilizing the entropy idea, this method may also be used to flood vulnerability assessment (Lotfi and Fallahnejad 2010;Nascimento and Prudente 2018).Shannon's entropy model is based on the following equations: where FR depicts Frequency ratio values, and P ij represents the probability density.
(2) FSI = ∑ FR (3) H j and H jmax represent entropy values, I j depicts information coefficient, and W j refers to resultant weight values for each parameter.
Now the W j of each class was subsequently summarized using the given formula equation to determine the Flood Susceptibility Index (FSI):

Weight of evidence model
Using the Bayes theorem, it is a quantitative "data-driven" method for estimating future occurrence rates (Rahmati et al. 2015).Some researchers have been utilising this technique to create a flood risk assessment map recently (Tehrany et al. 2014b;Rahmati et al. 2015;Shafapour Tehrany et al. 2017;Costache 2019).Overlaying flood sites with each conditioning component and evaluating whether and to what extent the effective factor is responsible for prior floods allows one to compute the statistical correlation between flood locations and each conditioning factor (Wang et al. 2002;Weed 2005;Fan et al. 2011;Sharma 2012;Zhang and Agterberg 2018;Getachew and Meten 2021).Finding positive and negative weights is key to this model's foundation.According to this technique, the presence or absence of floods (B) in a given location is used to assign a positive or negative weight to each of the flood conditioning factors (A) involved.
where the probability is P, and the natural log is ln .B and B are both the existence and absence of the influencing variables.Furthermore,the presence of the flood is demarcated by A and absence by B .For an influencing element to have a positive weight (Wi +), it must be found in high concentrations in flooded regions, and this concentration indicates a positive correlation between the presence of the conditioning (4) factor and floods.And yet, the absence of the weighted elements that might have a detrimental impact (Wi-).
In this statistical equation, positive and negative values correspond to the positive and negative geographical associations represented by C.
The variance of the plus weights is described by S2(W +), while the variance of the minus weights is described by S2(W).The following formula can be used to estimate the range of values for the weights (Bonham-Carter 1994).
In place of an absolute weight is the studentized constant, which is defined as the ratio between contrasts and standard deviation.
After the final model weights have been obtained, FSI data may be used to create a flood susceptibility map.FSI values have been calculated using the following formula:

Flood susceptibility influencing factors
In order to construct the flood susceptibility map, a number of different types of maps (thematic maps) were produced.In order to accomplish this goal, factors such as soil type, the Normalized Difference Vegetation Index (NDVI), the Normalized Difference Water Index (NDWI), the distance to the river, elevation, slope, drainage density, land use and land cover, rainfall, the Soil Moisture Index (SMI), and surface roughness have been utilised in flood vulnerability assessments.The creation of the final product is dependent on many different factors, each of which is important in its own right.
1. Rainfall: It is one of the most important factors that contribute to the acceleration of flood events (Rahman et al. 2019).In the upper part of the basin, where yearly rainfall totals of more than 1,400 mm have been measured, intense downpours have been witnessed on occasion. (10) Page 7 of 18 591 On the other hand, the lowest rainfall is reported near the bottom end of the basin, which receives less than 1200 mm of precipitation annually (Fig. 3i).Because of this, the lower stretch of the river has been known to experience the most severe flooding, while having the greatest rainfall totals recorded in the upper reach.Therefore, the slope plays a significant part in the quickening of the stream.2. Elevation and Slope: Land elevation and slope govern river basin drainage, which affects flood vulnerability (Rahman et al. 2021).Flood risk is greater in a basin's bottom reach because runoff flows from highlands to lowlands.In the west of the basin, the maximum height is 110-160 m, while the lowest is 16-38 m (Fig. 3a).
The western section has a slope of 2.64-27.94%,whereas the eastern part, where the height is lowest, has a slope of 0.98 (Fig. 3b).

Land use and Land cover:
The pattern of land use for a region reveals the manner in which living people and natural processes utilise the land in that region (Ajin et al. 2013;Kaur et al. 2017).The creation of land use and land cover maps is an essential part of the flood risk assessment process.Vegetated regions are less susceptible to flooding events owing to the inverse correlation between the intensity of flooding and the amount of vegetation present (Mojaddadi et al. 2017).LULC has an immediate impact on When taken together, these fea-  buffer zones from the mainline, are shown in Fig. 3e.In other words, the further away from the river you are, the less likely it is that you will get flooded, and vice versa.6. Drainage Density: An increase in drainage density suggests an increase in surface runoff, which in turn increases the likelihood of floods.Evidence suggests that greater drainage density regions generate more surface runoff than lower drainage density areas (Kumar et al. 2007).Thus, drainage density, an important determinant in runoff generation, may affect the severity of flood danger (Ogden et al. 2011).In ArcGIS 10.5 (evaluation version), a drainage density map was created using a grid-based approach.As can be seen in Fig. 3d, the drainage density is highest in the top part of the basin and lowest in the bottom part.Although there is less drainage in the lower stretch, the slope makes it more prone to flooding.7. Normalized Difference Vegetation Index (NDVI): An increase in drainage density suggests an increase in surface runoff, which in turn increases the likelihood of floods.Evidence suggests that greater drainage density regions generate more surface runoff than lower drain-age density areas (Kumar et al. 2007).Thus, drainage density, an important determinant in runoff generation, may affect the severity of flood danger (Ogden et al. 2011).In ArcGIS 10.5 (evaluation version), a drainage density map was created using a grid-based approach.
As can be seen in Fig. 3d, the drainage density is highest in the top part of the basin and lowest in the bottom part.
Although there is less drainage in the lower stretch, the slope makes it more prone to flooding.
Because of the steepness of the slope, the upper course of the stream has been overgrown with vegetation, as shown by Fig. 3J; however, the plant cover in the lower length of the stream is more sparse.

Normalized Difference Water Index (NDWI):
The NDWI, or Normalized Difference Water Index, is a measure of the amount of liquid water detected in satel- lite photographs.It pinpoints all of the surface water bodies that have an effect on flooding time and frequency (Memon et al. 2015;Khalifeh Soltanian et al. 2019;Sivanpillai et al. 2020).Utilizing the raster calculator tool in ArcGIS 10.5 (evaluation version), NDWI was calculated from Landsat-8 satellite images.Following is the formula that has been used to determine NDWI.
NDWI values can vary between -1 to 1. Water bodies have NDWI values more than 0.5, whereas plants have values closer to 0 or even negative values, making the distinction between the two straightforward.The distribution of NDWI is shown graphically in Fig. 3K.9. Soil Moisture Index (SMI): The amount of moisture present in the soil is a significant factor in the control of the water cycle and the duration of the flooding period (Saha et al. 2018).Even if there is further rainfall when the soil already has a high moisture content, the soil will not be able to store any more water.As a direct consequence of this, overflow develops surrounding the land regions, which then transforms into flood (Ahlmer et al. 2018).Using the following algorithm in ArcGIS software, the soil moisture index (SMI) was calculated from the LANDSAT 8 satellite photos.
where LST Max refers to the highest temperature recorded on the land.LST Min is an abbreviation for "Minimum land Surface Temperature." Figure 3h shows that the river basin's lower reach has the highest level of soil moisture.This is the reason why the flood intensity is highest in the river basin's lower reach.
10. Surface Roughness: It is another component that may have a variety of effects on the vulnerability to flooding.The surface's roughness has a significant role in determining how susceptible it is to flooding, and vice versa.This parameter was taken using a digital elevation model, also known as a DEM, and the map was created with ArcGIS 10.5 software (evaluation version).
Figure 3c illustrates that the surface of the basin is rough in the upper reach of the basin, but in the lower reach of the basin, the surface is smooth.
The entire methodology from data collection to application of statistical model has been in detail in (Fig. 4).

Frequency ratio and flood susceptibility zones
The frequency ratio model has been conducted using eleven inputs.Table 3 displays the value of the frequency ratio for all parameters and their corresponding feature classes.In this case, the flood susceptibility increases as the frequency ratio increases, and vice versa.As a result, there is a robust positive correlation between FR levels and flood severity (Liuzzo et al. 2019).Results from the FR model are very sensitive to their choice of parameters (Figs.5a, 3a and d).The FR model shows that the upper Kopai river basin has a low to very low susceptible to flooding, whereas the lower basin has a significant risk of flooding (see Fig. 5a).Moreover, the FR model identifies drainage density and soil moisture index as critical influencing parameters for flood occurrence.Comparing Fig. 3h with Fig. 5a, it can be seen that both occur at the lower part of the river, where the basin accumulates an abundance of moisture and speeds up the risk of flooding.This model assigns an area of 362.403 and 191.184Km2 respectively to high and very high flood susceptible zones (Table 2).

Shannon's entropy and flood susceptibility zones
The susceptibility zones for flooding in the Kopai river basin have been extracted using another statistical approach, Shannon's entropy.In the part devoted to its methodology, this model's specific procedures are laid out. Figure 5b shows the output of Shannon's entropy.For example, in Shanno's model, high and very high flood susceptible zones encompass 522.757 and 314.266Km2 area, respectively, but in the FR model, these zones are substantially less (Table 2).Therefore, the outcome from this model is different from that of the FR model.

Weight of evidence and flood susceptibility zones
This model is identical to the FR model, as seen in Fig. 5c.A similar distribution of flood-prone areas is seen in both FR and WOE (Table 2).Based on this model's findings, the lower section of the Kopai river basin is more susceptible to flooding than the section closer to the river's source.
All three of these models are rather comparable; for all three, the highest value suggests a high risk of flooding, and vice-versa (Table 3).

Validation of models
Receiver operating curve (ROC) analysis was used to verify the accuracy of the aforementioned three models.In this study, the curve was used to graphically illustrate and quantitatively assess the prediction abilities of individual models.The area under the receiver operating characteristic curve is generated by plotting the sensitivity (y-axis) vs the 1-specificity (x-axis).In a quantitative way, the AUROC curve depicts the predictive power of a model and the accuracy of a map.The predictive accuracy value varies from 0 to 1 and may be broken down into five distinct categories.(Wang et al. 2011) viz.> 0.9 to 1 perfectly applicable, 0.8 to 0.9 highly applicable 0.7 to 0.8 moderately applicable 0.6 to 0.7 Poorly applicable 0.5 to 0.6 very poorly applicable.More than 300 sample points were obtained as test data for ROC sketching, and the test points were chosen to determine whether flood-susceptible areas are really flooded or not.In Fig. 6, we see three ROC curves for three different models, including the FR model, Shannon's entropy, and the weight of evidence.The area under the receiver operating characteristic (AUC) of the FR, Shannon's entropy, and WOE for the three models are 0.965, 0.912, and 0.971, respectively (Fig. 6).
That's because the FR model and WOE have been shown to be more accurate via validation, they should be considered the preferred statistical models for this area.If the output of flood susceptibility zones observed, it would be seen that Shannon's entropy offers more flooded regions than the other two models, but in actuality, the Kopai river basin only has flooded sections in the lower reach.The validation result of Shannon's entropy is comparatively low, as shown by the ROC curve based on field data, to provide a better explanation of flood susceptibility than the FR and WOE model.In this case, WOE offers the highest accuracy (97%) out of the three models.We thus accept the findings as a genuine model to determine the flood prone areas of this river basin.

Discussion
Finding out how susceptible a region is to flooding is the first and most important step in evaluating the risk of flooding.This is accomplished via flood susceptibility mapping.It is possible to identify locations that are prone to flooding and to put in place the required support measures in order to minimize losses caused by flooding.In our investigation and analysis, we employed a variety of geospatial datasets combined with machine learning and geographic information systems (GIS).The lower section of this river often floods during the monsoon season.While it is possible to take precautions against certain flood damage, this does not eliminate the risk completely.In order to lessen the destruction caused by floods, it is necessary to identify the main drivers of such occurrences and create a flood susceptibility map.However, floods are also affected by a wide variety of hydrological, geological, topographical, and morphological elements.Selecting relevant flood-affecting elements is a crucial step in flood susceptibility modelling since only some of these components are included in flood susceptibility models.
There has been no advanced algorithmic study of flood occurrences in this river basin.This research aims to fill this research gap by assessing the efficacy of three algorithms in forecasting flood-prone zones in this basin.
Flood modelling is a tricky task with many uncertainty.Machine learning algorithms can effectively deal with these unknowns if reliable historical flood inventory maps are available.To address this, we used aerial pictures and field verification to create a flood inventory map of the research  region.As long as reliable flood inventory maps from the past are accessible, machine learning algorithms can effectively deal with these unknowns.
Eleven key variables were considered in this assessment.Various factors contribute their own sets of probabilities to the occurrence of floods.Flooding is mostly determined by precipitation in plain areas.As the river basin is located in a tropical monsoon environment, the lower portion of the basin is often flooded due to heavy rainfall during the monsoon season.
In terms of ROC findings, the AUC for all three models was more than 0.90, showing that they successfully predicted flood susceptibility.
The WoE model has the highest area under the curve (0.971) of the three, beating out the FR and Shannon's entropy models.
For the purpose of flood risk management in the Kopai river basin, the provided models may prove to be an effective and new approach.

Conclusion
Based on the findings of three different statistical models, this research suggested different flood-prone zones.There are ten flood influencer factors that have been considered in order to run the three models.In order to produce a result that can be compared across models, the same parameters were used for each model.The findings of the three models are consistent with one another in the sense that in all of the models, the upper reach of the basin presented flood risk zones that were lower, while the lower reach gave flood risk zones that were greater.These five zones have been defined to demonstrate the severity of flood susceptibility; they are very high susceptible, high susceptible, moderately susceptible, low susceptible, and very low susceptible respectively.Very low susceptible zones are the least likely to be affected by flooding.
When it comes to the validation of the model, we can see that the frequency ratio and Shannon's entropy models are not as accurate as the weight of evidence model does.In other words, the weight of evidence model provides a higher level of precision.ROC has been plotted in this section to determine the curve and accuracy of the model.When compared to the other consecutive approaches, the WOE method demonstrates an accuracy of more than 97%.
The flood susceptibility map that was detailed in this study might be an invaluable resource for engineers, policymakers, planners, and administrative bodies in the Kopai river basin when it comes to the prevention of floods in that region.One further important distinction is that the approaches that were utilised in this research helped bridge the gap between the comprehension of catastrophic floods and the efficacy of their effects.It is recommended to use this strategy since it may help provide a better knowledge of the severity and frequency of floods, as well as the location of areas where people are at danger of being flooded.
Funding Authors receive no funding for this work.

Declarations
Conflicts of interests Authors have no conflict of interest to declare.

Fig. 1
Fig. 1 Location of the study area: a. India b.West Bengal c. Kopai river basin

Fig. 3
Fig. 3 Input data: a. Elevation b.Slope c. Surface roughness d.Drainage density e. Distance from river f.Landuse/landcover g.Soil h.Soil moisture index i.Rainfall j.NDVI k.NDWI

Fig. 4
Fig. 4 Methodology of the study

Fig. 5
Fig. 5 Flood susceptibility map based on (a) Frequency ratio (b) Shannon's entropy (c) Weight of evidence

Table 1
Literatures on selection of conditioning parameters for flood vulnerability assessment

Table 2
Area of flood susceptibility zones under different model