1 Introduction

Floods are one of the deadliest natural catastrophes occurring worldwide due to prolonged rainfall, but are aggravated by land-use changes, unscientific modifications of drainage channels, and unplanned development activities on flood plains and flood-prone areas (Deepak et al. 2020; Tehrany et al. 2015). Flood losses are generally categorized as direct or indirect as classified by Smith and Ward 1998. Direct losses are caused by physical contact of floodwater with individuals, property, or other items; whereas, indirect losses affect networks and social activities, resulting in losses such as traffic, trade, and public service interruptions (Nicholls et al. 2015), infectious disease outbreaks (Chadsuthi et al. 2021; Okaka and Odhiambo 2018), contamination of water resources (Ching et al. 2015; Sun et al. 2016; Yard et al. 2014), increased snakebite incidences (Ochoa et al. 2020), and psychological trauma (Crabtree 2013; Hajat et al. 2005; Paranjothy et al. 2011; Seyedin et al. 2017). Futhermore, delays and diverts in transportation networks cost money and loss of access to markets, affects employment prospects, health and education and social activities in far-flung places, which means that indirect losses are also incurred (Winter et al. 2016).

According to the Global Natural Disaster Assessment Report (2020), flood disasters with a frequency of 61.66% account for 40.92% of deaths, with 33.56% of the population affected, and 29.72% of direct economic losses worldwide for the year 2020. Out of the top ten countries, in terms of the number of people affected by flood disasters from 1900 to 2022, eight are from Asia (https://www.emdat.be/). Furthermore, in terms of the number of flood occurrences, out of the top ten countries (with 1,800 occurrences), seven countries are from Asia, which accounts for 1,339 occurrences (74.38%) (https://www.emdat.be/). The number of fatalities due to flooding are increasing in Asia (Franzke and Torelló i Sentelles 2020). Global warming and associated polar ice meltdown, glacier meltdown and sea-level rise are the major reasons for frequent flooding (Kumcu 2022; Swain et al. 2020a; Tabari 2020). Alike the rest of the world, flooding is one of the most prevalent natural hazards in India. India is one of the top ten countries (ranks third) most frequently affected by flooding (Liu et al. 2022). But what makes it different from the rest is the very high population density. According to EM-DAT (https://www.emdat.be/), the second highest number of deaths (30,115) was recorded in India after China due to riverine flooding between 1990 and 2022. In terms of total estimated damage during this time period, India ranks third (55.78 billion USD) after China and the United States (https://www.emdat.be/). Also, the total number of people affected by floods in India in the aforementioned period is 348,902,349, with China being at the top of the list (https://www.emdat.be/).

One of the recent catastrophic flood-affected areas in India is the narrow stretch of land, called Kerala, located between the Western Ghats and the Arabian Sea. Kerala is one of thestates in India having the highest population density (860 per km2), and is nestled in the foothills of the Western Ghats, an orographic edifice. This state has 44 medium-to-small rivers, all with a short traverse of an average of 100–150 km, flowing through all the physiographic zones, viz., lowland, midland, and highland, before debouching into the Arabian Sea (Sajinkumar et al. 2022). Because of the preponderance of the monsoonal climate, many parts of Kerala have been severely battered by floods, with the recent floods being more devastating (i.e., the 2018, 2019, 2020, and 2021 floods). The severe floods of 2018 and 2019 caused significant damage to infrastructure and property, and resulted in hundreds of deaths in Kerala (Hao et al. 2020, 2022; Hunt and Menon 2020; Mishra and Shah 2018; Vanama et al. 2021; Vishnu et al. 2019, 2020). What makes more difficult to estimate the accurate losses are: (i) the lack of a clear strategy, (ii) absence of a single, reliable method for estimating damage and costs, (iii) non-availability of multiple, but diversified, agencies in post-disaster reconstruction activities, and (iv) the longer resilience time (Donnini et al. 2017).

One of the pressing requirements in studying the flooding events is the assessment of vulnerability. Researchers generally uses AHP (Dandapat and Panda 2017; Deepak et al. 2020; Desalegn and Mulu 2020; Hoque et al. 2019; Hussain et al. 2021; Radmehr and Araghinejad 2015), analytical network process (Chukwuma et al. 2021), Bayesian belief network (Abebe et al. 2018), Weight of evidence-information value (Saha et al. 2021), frequency ratio (FR) (Saha et al. 2021; Sarkar and Mondal 2020), F-AHP (Duan et al. 2021), and support vector machine (Duan et al. 2021) for assessing flood vulnerability. Feloni et al. (2020) created flood vulnerability maps of the Attica region in Greece using the AHP and fuzzy-analytical hierarchy process (F-AHP) methods. They used physical vulnerability indicating factors such as digital elevation model (DEM), slope, critical aspect, horizontal overland flow distance, vertical distance of channel network, curvature index, SAGA wetness index, composite curve number, and daily-modified Fournier index. Ali et al. (2019) compared FR and AHP for demarcating the flood vulnerable zones in the Sundarbans region of India. They used only the physical vulnerability factors such as slope, elevation, topographic wetness index, land use land cover, amount of rainfall deviation, distance from river, and clay content in soil. Also, no researchers compared the AHP and F-AHP models to identify flood vulnerable zones using both physical-environmental and socio-economic indicators in any flood-prone area in the world. The majority of the existing literature focused solely on the areas' physical vulnerability factors (Desalegn and Mulu 2020; Feloni et al. 2020).

In any disaster study, the geographical ambience where the influencing parameters can be thoroughly studied could be a natural boundary like a drainage basin, whereas the implementation of management practices will be feasible within a political boundary. In India, the political set-up in the lowest rug is the local self governing (LSG) bodies. One of the worst-affected districts (which consist of a cluster of LSGs) is Kottayam (Vanama et al. 2021), located in the foothills of the Western Ghats, where the rivers Muvattupuzha, Meenachil, and Manimala were flooded during the recent years. Thus, this study aims to identify the flood vulnerable LSGs in the Kottayam district using both the AHP and F-AHP models, select the best one among these two models, and suggest suitable recommendations.

2 Materials and methods

2.1 Study area

Located beneath the Western Ghats, Kottayam district has longitudes of 76˚20′0' 'E and 77˚00′0'' E and latitudes of 9˚20′0'' N and 9˚55′0'' N and covers an area of about 2208 km2 (Fig. 1). Kottayam district experiences a tropical climate, with an average annual rainfall of 3130.33 mm (https://kottayam.nic.in/climate/). The eastern part of the district is composed of Precambrian metamorphic rocks that form steep terrain; the central part is a low plateau with Tertiary sediments and laterites; and the western part is a low plain covered by Quaternary fluvial or partly marine formations (Department of Mining and Geology 2016). Meenachil, Muvattupuzha, and Manimala are the district's major rivers (https://kottayam.nic.in/en/geography/), all originating from the Western Ghats and debouching into the Vembanad Lake. The Meenachil River has a length of 78 km, whereas the Muvattupuzha and Manimala rivers have lengths of 121 and 92 km, respectively. According to the data acquired from the National Remote Sensing Centre, 185.82 km2, 106.44 km2, and 88.96 km2 areas of the district were inundated by the floods that happened in the years 2018, 2019, and 2020, respectively. A total of 23 human fatalities have been recorded in Kottayam district during the 2018 monsoon season, while there were two during the 2019 monsoon season (Government of Kerala 2018a, 2018b, 2019a, b).

Fig. 1
figure 1

(Source: Elevation data of ASTER; Lithology data of Geological Survey of India)

Location map of the study area: a. South India, b. Kerala, c. elevation map with major rivers of the study area (Kottayam district), and d. Lithology map of the study area

Full size image

2.2 Data used

The modelling process involved the following five steps:

  1. i.

    Data for the 20 vulnerability indicators were gathered from various sources (Table 1). The thematic layers of these vulnerability indicators were created using ArcGIS 10.8 and ERDAS Imagine 9.2.

  2. ii.

    After creating the thematic layers of these indicators, the flood inundation data was gathered from the National Remote Sensing Centre (NRSC), India.

  3. iii.

    Validation locations were randomly selected within the flood-inundated areas.

  4. iv.

    The physical-environmental, socio-economic, and flood vulnerability maps were created using AHP and F-AHP methods. MS Excel and FisPro 3.8 were used to derive the weights of the AHP and F-AHP methods, respectively (Fig. 2).

  5. v.

    The vulnerability maps were validated utilizing the receiver operating characteristic (ROC) curve method. Validation of the results was done using IBM SPSS 26.0 .

Table 1 Inventory on the data used in this study
Full size table
Fig. 2
figure 2

Flowchart of the vulnerability modelling

Full size image

2.3 Methodology for vulnerability indicators

A total of 20 vulnerability indicators (nine physical-environmental variables and 11 socio-economic variables) have been considered. Physical-environmental vulnerability indicators such as slope angle, geomorphology, stream density, soil texture, land use/land cover (LULC), modified normalized difference water index (MNDWI), normalized difference vegetation index (NDVI), normalized difference built-up index (NDBI), water ratio index (WRI), and socio-economic vulnerability indicators such as total population, number of households, literacy rate, building roof type, building condition, household size, child population, access to information, building wall type, and marginalized populations (scheduled caste and scheduled tribe) have been utilized for the modelling.

2.3.1 Physical-environmental vulnerability indicators

Slope angle was computed from the ASTER GDEM using ArcGIS spatial (surface) analyst tools. The soil types of the study area were derived from the soil map published by the National Bureau of Soil Survey and Land Use Planning (NBSS&LUP) using ArcGIS tools. Land use and land cover (LULC) types of this area were derived from the Landsat 8 satellite images. The supervised maximum likelihood (cf. Joshi et al. 2022; Sisodia et al. 2014; Sunar Erbek et al. 2004) classification approach in the ERDAS Imagine software was utilized to classify the LULC types. Stream density was computed from the Survey of India (SoI) topographic map using ArcGIS spatial analyst (line density) tools. The geomorphic classes were initially derived from the Landsat 8 images through visual interpretation, but aided by field experience. This was verified with the SoI topographic map and high-resolution Google Earth images. This was done using ERDAS Imagine software. Water ratio index (WRI) was calculated using Eq. 1 (Shen and Li 2010) whereas modified normalized difference water index (MNDWI) was calculated using Eq. 2 (Xu 2006). Normalized difference built-up index (NDBI) was calculated using Eq. 3 (Zha et al. 2003) and normalized difference vegetation index (NDVI) was computed using Eq. 4 (Rouse et al. 1974).

The spectral indices such as WRI, MNDWI, NDBI, and NDVI were computed from the Landsat 8 images using ArcGIS spatial analyst (raster calculator) tool. The continuous data layers such as slope angle, stream density, WRI, MNDWI, NDBI, and NDVI were classified using the natural breaks method (cf. Ahmed 2015; Pradeep et al. 2022).

$${\text{WRI}} = \frac{{\left( {{\text{Green}} + {\text{Red}}} \right)}}{{\left( {{\text{NIR}} + {\text{MIR}}} \right)}}$$
(1)

where green, red, NIR and MIR stand for spectral reflectance of water in green, red, near-infrared and mid-infrared bands, respectively.

$${\text{MNDWI }} = \frac{{\left( {{\text{Green }} - {\text{ SWIR}}} \right)}}{{\left( {{\text{Green }} + {\text{ SWIR}}} \right)}}$$
(2)

where green, and SWIR stand for spectral reflectance of water in green, and short-wave infrared bands, respectively.

$${\text{NDBI }} = \frac{{\left( {{\text{SWIR}}1 - {\text{ NIR}}} \right)}}{{\left( {{\text{SWIR}}1 + {\text{NIR}}} \right)}}$$
(3)

where SWIR, and NIR stand for spectral reflectance in short-wave infrared and near-infrared bands, respectively.

$${\text{NDVI }} = \frac{{\left( {{\text{NIR}} - {\text{R}}} \right)}}{{\left( {{\text{NIR}} + {\text{R}}} \right)}}$$
(4)

where NIR and R stand for spectral reflectance in the near-infrared and red bands, respectively.

2.3.2 Socio-economic vulnerability indicators

All the eleven socio-economic vulnerability indicators were derived from the 2011 census data and were categorized into five classes through the natural breaks method (cf. Wubalem 2021) using ArcGIS tools.

2.4 AHP modeling

The AHP method, developed by Thomas L. Saaty (Saaty 1980) is used to make difficult problems into a hierarchy and find the best answer for the goal (Qazi and Abushammala 2020). The creation of the matrix for pair-wise comparisons, the computation of the eigen vector, the weighting coefficient (Tables 2 and 3), and the consistency ratio (Tables 4 and 5) are the major procedures involved with the AHP modelling (Akshaya et al. 2021; Amrutha et al. 2022; Nikhil et al. 2021; Thomas et al. 2021).

Table 2 Pairwise comparison matrix for Physical-environmental indicators
Full size table
Table 3 Pairwise comparison matrix for socio-economic vulnerability
Full size table
Table 4 Normalized matrix for physical-environmental vulnerability
Full size table
Table 5 Normalized matrix for socio-economic vulnerability
Full size table

Using Eqs. 5 and 6, the eigen vector (Vp), and the weighting coefficient (Cp) were determined (as in Akshaya et al. 2021; Nikhil et al. 2021; Thomas et al. 2021).

$$V_{P} = \sqrt[k]{{W1{\text{x}} \ldots Wk}}$$
(5)

where k = number of factors, and W = ratings.

$$C_{p} = \frac{Vp}{{Vp1 + \ldots Vpk}}$$
(6)

The priority vector [C], overall priority [D], and rational priority [E] were computed as in Danumah et al. (2016).

Equations 7, 8, and 9 (Akshaya et al. 2021; Nikhil et al. 2021; Thomas et al. 2021), were employed to compute the eigen value (λmax), consistency index (CI), and consistency ratio (CR).

$${\uplambda }_{\max } = \frac{\left[ E \right]}{k}$$
(7)
$${\text{CI = }}{\raise0.7ex\hbox{${\left( {\lambda_{\max } - k} \right)}$} \!\mathord{\left/ {\vphantom {{\left( {\lambda_{\max } - k} \right)} {\left( {k - 1} \right)}}}\right.\kern-\nulldelimiterspace} \!\lower0.7ex\hbox{${\left( {k - 1} \right)}$}}$$
(8)
$$CR = \frac{CI}{{RI}}$$
(9)

where RI = random index (Saaty 1980).

Saaty (1980) considers a CR of less than 0.1 acceptable. If the CR is larger than 0.1, repeat the analysis until the CR is acceptable. The CR (0.031) in this AHP modelling is found acceptable. As a result, the outcomes are reliable.

The weights derived by the AHP model for the physical-environmental vulnerability (PEV) and socio-economic vulnerability (SEV) indicators are depicted in Eqs. 10 and 11.

$$\begin{gathered} {\text{PEV }} = \left( {0.308 \times {\text{Slp}}.} \right) + \left( {0.223 \times {\text{Soil}}} \right) + \left( {0.157 \times {\text{LULC}}} \right) + \left( {0.108 \times {\text{SD}}} \right) + \left( {0.074 \times {\text{Geom}}.} \right) + \hfill \\ \left( {0.051 \times {\text{WRI}}} \right) + \left( {0.035 \times {\text{MNDWI}}} \right) + \left( {0.025 \times {\text{NDBI}}} \right) + \left( {0.018 \times {\text{NDVI}}} \right) \hfill \\ \end{gathered}$$
(10)
$$\begin{aligned} {\text{SEV = }} & \left( {0.187 \times {\text{TP}}} \right) + \left( {0.144 \times {\text{NH}}} \right) + \left( {0.082 \times {\text{BRT}}} \right) + \left( {0.074 \times {\text{BC}}} \right) + \\ & \left( {0.073 \times {\text{HHS}}} \right) + \left( {0.071 \times {\text{CP}}} \right) + \left( {0.070 \times {\text{MP}}1} \right) + \left( {0.067 \times {\text{MP}}2} \right) + \left( {0.064 \times {\text{ATI}}} \right) + \left( {0.060 \times {\text{BWT}}} \right) \\ \end{aligned}$$
(11)

2.5 F-AHP modeling

A combination of AHP and fuzzy logic theory was employed to weight the relevant factors in the F-AHP model (Eskandari and Miesel 2017). The F-AHP model overcomes the limitations of the AHP model by allowing decision-makers to assess their preferences within an acceptable interval (Afolayan et al. 2020). Buckley (1985) presented a method for comparing fuzzy ratios that has been adopted for the study. The key procedures involved are pair-wise comparison matrix creation (see Tables 6 and 7), geometric mean calculation (see Tables 8 and 9), determination of relative fuzzy weights, and computation of averaged and normalized relative weights (see Tables 10 and 11). The following are the steps in F-AHP modelling:

Table 6 Pair-wise comparisons of factors of physical-environmental vulnerability
Full size table
Table 7 Pair-wise comparisons of factors of socio-economic vulnerability
Full size table
Table 8 Geometric means of fuzzy comparison values: Physical-environmental vulnerability
Full size table
Table 9 Geometric means of fuzzy comparison values: socio-economic vulnerability
Full size table
Table 10 Relative fuzzy weights, averaged and normalized relative weights: Physical-environmental vulnerability
Full size table
Table 11 Relative fuzzy weights, averaged and normalized relative weights: Socio-economic vulnerability
Full size table

Step 1 Comparison of the indicators or alternatives.

The fuzzy triangular scale (2, 3, 4) will be used when factor 1 (P1) is less significant than factor 2 (P2). For the comparison matrix, the fuzzy triangle scale will be (1/4, 1/3, 1/2) (Ayhan 2013).

Equation 12 depicts the matrix.

$$\tilde{A}^{k} = \left[ {\begin{array}{*{20}c} {\tilde{d}_{11}^{k} } & {\tilde{d}_{12}^{k} } & {...} & {\tilde{d}_{1n}^{k} } \\ {\tilde{d}_{21}^{k} } & {...} & {...} & {\tilde{d}_{2n}^{k} } \\ {...} & {...} & {...} & {...} \\ {\tilde{d}_{n1}^{k} } & {\tilde{d}_{n2}^{k} } & {...} & {\tilde{d}_{mn}^{k} } \\ \end{array} } \right]$$
(12)

where\(\tilde{d}_{ij}^{k}\) by way of fuzzy triangular numbers, reflects the kth decision maker's preference for the ith factor over the jth factor (Ayhan 2013).

Step 2 \(\tilde{d}_{ij}\) was computed using Eq. 13, after averaging the preferences (\(\tilde{d}_{ij}^{k}\))

$$\tilde{d}_{ij} = \frac{{\mathop \sum \nolimits_{k = 1}^{K} \tilde{d}_{ij}^{k} }}{K}$$
(13)

Step 3 The matrix was modified applying Eq. 14.

$$\tilde{A} = \left[ {\begin{array}{*{20}c} {\tilde{d}_{11} } & \cdots & {\tilde{d}_{1n} } \\ \vdots & \ddots & \vdots \\ {\tilde{d}_{n1} } & \cdots & {\tilde{d}_{nn} } \\ \end{array} } \right]$$
(14)

Step 4 Computation of the geometric average using (Buckley 1985) Eq. 15

$$\tilde{r}_{i} = \left( {\mathop \prod \limits_{j = 1}^{n} \tilde{d}_{ij} } \right)^{1/n} ,^{{}} i = 1,2,....,n$$
(15)

where \({\widetilde{r}}_{i}\) = triangular values.

Step 5 Computation of fuzzy weight using the following three sub processes (5a, 5b, and 5c).

5a: Computation of vector summation of each \({\widetilde{r}}_{i}\)

5b: The fuzzy triangular number was substituted to convert it to an increasing order, after determining the (-1) power of the summation vector.

5c: Computation of fuzzy weights: Each \({\widetilde{r}}_{i}\) was multiplied with its reverse vector as in Eq. 16 to compute the weights

$$\begin{aligned} \tilde{w}_{i} = & & \tilde{r}_{i} \otimes (\tilde{r}_{1} \oplus \tilde{r}_{2} \oplus \ldots \oplus \tilde{r}_{n} )^{{ - 1}} \\ = (lw_{i} ,mw_{i} ,uw_{i} ) \\ \end{aligned}$$
(16)

Step 6 Eq. 17 was utilized for the de-fuzzification of fuzzy weights (Chou and Chang 2008)

$$M_{i} = \frac{{lw_{i} ,mw_{i} ,uw_{i} }}{3}$$
(17)

Step 7 Standardization of Mi using Eq. 18.

$$N_{i} = \frac{{M_{i} }}{{\sum\nolimits_{i = 1}^{n} {M_{i} } }}$$
(18)

The weights derived by the F-AHP model for the PEV and SEV indicators are depicted in Eqs. 19 and 20.

$$\begin{gathered} {\text{PEV }} = \left( {0.299 \times {\text{Slp}}.} \right) + \left( {0.223 \times {\text{Soil}}} \right) + \left( {0.159 \times {\text{LULC}}} \right) + \left( {0.111 \times {\text{SD}}} \right) + \left( {0.076 \times {\text{Geom}}.} \right) + \hfill \\ \left( {0.052 \times {\text{WRI}}} \right) + \left( {0.036 \times {\text{MNDWI}}} \right) + \left( {0.025 \times {\text{NDBI}}} \right) + \left( {0.018 \times {\text{NDVI}}} \right) \hfill \\ \end{gathered}$$
(19)
$$\begin{gathered} {\text{SEV }} = \left( {0.269 \times {\text{TP}}} \right) + \left( {0.208 \times {\text{NH}}} \right) + \left( {0.154 \times {\text{LR}}} \right) + \left( {0.112 \times {\text{BRT}}} \right) + \left( {0.081 \times {\text{BC}}} \right) + \left( {0.057 \times {\text{HHS}}} \right) \hfill \\ + \left( {0.041 \times {\text{CP}}} \right) + \left( {0.029 \times {\text{MP}}1} \right) + \left( {0.021 \times {\text{MP}}2} \right) + \left( {0.020 \times {\text{ATI}}} \right) + \left( {0.012 \times {\text{BWT}}} \right) \hfill \\ \end{gathered}$$
(20)

2.6 Validation of modelling results using the ROC method

ROC graphs are widely used to evaluate a classifier's performance. On the y-axis, sensitivity is plotted, while 1-specificity is plotted on the x-axis (Grimnes and Martinsen 2015). The AUC is a single scalar statistic that quantifies a binary classifier's total performance (Hanley and McNeil 1982). An AUC value of 0.7 to 0.8 is acceptable, 0.8 to 0.9 is excellent, and more than 0.9 is outstanding (Hosmer and Lemeshow 2000). The results of the study were validated using the flood inundation data provided by the NRSC for the years 2018, 2019, and 2020 (Fig. 3a, 3b, and 3c). In this research, a total of 100 locations within the flood-inundated areas have been randomly selected as a validation dataset (Fig. 4). The SPSS software was utilized for plotting the ROC graph and estimating the AUC values.

Fig. 3
figure 3

a. 2018 Flood inundated area; b. 2019 Flood inundated area; c. 2020 Flood inundated area

Full size image
Fig. 4
figure 4

Flood inundation data (2018–2020) and validation dataset

Full size image

3 Results

3.1 Vulnerability indicators

3.1.1 Slope angle

When the slope angle is less, the probability of flooding increases (Rahman et al. 2019). Runoff from rainfall accumulates and inundates areas with gentle slopes due to the low flow velocity in these areas (Lee and Kim 2021). The slope of the district ranges from 0 to 74.48° (Fig. 5a) and is categorized into five zones (0–4.67°, 4.68–9.93°, 9.94–17.52°, 17.53–27.75°, and 27.76–74.48°). The western portion of the district has gentle slopes, while the northeastern portion has steeper slopes.

Fig. 5
figure 5

a. Slope; b. Soil types; c. Land use/land cover types; d. Stream density

Full size image

3.1.2 Soil texture

The greater the infiltration rate of soil, the less likely a flood will occur (Islam et al. 2021). Impermeable formations, such as clay, enhance runoff rates, increasing the probability of flooding (Swain et al. 2020b). On the basis of soil texture, the soils are classified into clay, gravelly clay, gravelly loam, loam, and sand (Fig. 5b). The lower elevated parts of the district have predominantly clayey soil.

3.1.3 Land use/land cover (LULC)

Flooding is more likely in built-up areas because water cannot infiltrate and generate surface runoff (Islam et al. 2021). Increased forest and vegetation cover enhances water infiltration and reduces runoff depth, thus lowering the chance of flooding (Swain et al. 2020b). The land use/land cover (LULC) types comprise agricultural land, forest plantations, grassland, plantations (mixed vegetation), evergreen forest, built-up areas, water bodies, and wetlands (Fig. 5c). Agricultural land (paddy fields) and wetlands dominate the low-elevation part of the study area.

3.1.4 Stream density

Flooding is more likely in locations with lower stream densities because there aren't enough streams to drain out the surplus rainwater (Ajin et al. 2013; 2019). On the basis of the density of stream channels, the district is categorized into five zones (0–1.02 km/km2, 1.02–2.47 km/km2, 2.47–3.53 km/km2, 3.53–6.48 km/km2, and 6.48–10.87 km/km2). The lower elevation region of the study area has a stream density ranging from 0 to 1.02 km/sq.km (Fig. 5d).

3.1.5 Geomorphology

The geomorphic classes of the study area comprise the coastal plain, denudational hills, floodplain, islands, and plateau (Fig. 6a). The coastal plain and floodplains, located in the lower elevated region of the study area, are more susceptible to flooding due to flat topography, which favours inundation.

Fig. 6
figure 6

a. Geomorphology; b. Water ratio index; c. Modified normalized difference water index; d. Normalized difference built-up index

Full size image

3.1.6 Water ratio index (WRI)

Shen and Li (2010) suggest a WRI with a value greater than 1 for waterbodies. WRI values less than or close to 1 (especially values closer to and below zero) indicate that the soil moisture content is higher, making the area more vulnerable to flooding. The WRI of Kottayam district ranges from 0.30 to 1.45 and is grouped into five zones (0.30–0.54, 0.55–0.62, 0.63–0.78, 0.79–1.09, and 1.10–1.45) as depicted in Fig. 6b. The low elevated regions have a WRI value above 1.

3.1.7 Modified normalized difference water index (MNDWI)

The MNDWI (Xu 2006) suggests greater positive values for water bodies and smaller negative values for soil, vegetation, and built-up areas (Du et al. 2016). The higher MNDWI values represent areas with higher soil moisture content and are more vulnerable to flooding. The MNDWI of the study area ranges from -0.69 to 0.31 and is grouped into five zones (− 0.69 to − 0.24, − 0.23 to − 0.17, − 0.16 to − 0.11, − 0.10 to 0.05, and 0.06 to 0.31) as depicted in Fig. 6c. The low-elevated regions of the study area have positive MNDWI values and hence, more vulnerable to flooding.

3.1.8 Normalized difference built-up index (NDBI)

The positive NDBI values show built-up areas (Shahfahad et al. 2021). Urbanization increases the amount of impervious surface area in a location, slowing the hydrologic system's response time and thereby increasing the risk of flooding (Feng et al. 2021). The NDBI of the study area ranges from − 0.45 to 0.57 and is grouped into five zones (− 0.45 to − 0.23, − 0.22 to − 0.17, − 0.16 to − 0.10, − 0.09 to 0.00, and 0.01 to 0.57) as depicted in Fig. 6d. The low-elevated regions of the area have positive NDBI value and hence, induce flooding.

3.1.9 Normalized difference vegetation index (NDVI)

The NDVI ranges between − 1 and + 1 (Gessesse and Melesse 2019), and densely vegetated areas will be represented by positive values, and water and built-up areas, on the other hand, will be indicated by near-zero or negative values (Viana et al. 2019). The possibility of flooding is high in areas with lower NDVI values, as these values represent non-vegetated areas where the chance of surface runoff is greater. The thick vegetative cover can improve water infiltration and reduce runoff depth, lowering the risk of flooding (Swain et al. 2020b). The NDVI of the study area ranges from − 0.11 to 0.59 and is grouped into five zones (− 0.11 to 0.08, 0.09 to 0.25, 0.26 to 0.34, 0.35 to 0.41, and 0.42 to 0.59) as depicted in Fig. 7. The low-elevated regions of the study area have negative NDVI values.

Fig. 7
figure 7

Normalized Difference Vegetation Index (NDVI) of the study area. The highly urbanized Kottayam town is noteworthy in west-central region

Full size image

3.1.10 Total population

Areas with a larger population will be more exposed to a hazard and the evacuation process may be challenging, making them more vulnerable (Tascón-González et al. 2020). The population of the study area ranges between 7029 and 55,374 (Fig. 8a) and is divided into five zones (7029–14,339, 14,340–20,752, 20,753–27,977, 27,978–38,445, and 38,446–55,374).

Fig. 8
figure 8

a. Total population; b. No. of households; c. Literacy rate; d. Building roof type

Full size image

3.1.11 Number of households

As the rescue and evacuation processes will be much more complicated, the areas with a higher number of buildings will be more vulnerable than those with a lower number of buildings (Fernandez et al. 2016). The households in the study area range between 1706 and 14,366 (Fig. 8b) and are divided into five zones (1706–3556, 3557–4910, 4911–6952, 6953–9737, and 9738–14,366).

3.1.12 Literacy rate

A population's level and quality of education is a strong indicator of its vulnerability (Rasch 2016). Illiterates were thought to be more vulnerable because they lack the basic skills necessary to adapt to a risky situation (Deepak et al. 2020). The literacy rate of the study area ranges from 82.61 to 90.47% (Fig. 8c) and is categorized into five zones (82.61%, 82.62–87.66%, 87.67–88.63%, 88.64–89.61%, and 89.62–90.47%).

3.1.13 Building roof type

The roof of the buildings was categorized into different groups, viz., grass/thatch/bamboo/wood/mud, plastic/polythene, handmade tiles, machine-made tiles, burnt brick, stone/slate, G.I./metal/asbestos sheets, concrete, and others. Houses with concrete roofs will be more resistant to flooding. The majority of the houses in the study area have concrete roofs and hence, those houses were selected for the modelling. Houses other than concrete roofs will be more vulnerable to flooding. The percentage of buildings with concrete roofs ranges between 16.00 and 66.20% (Fig. 8d) and is categorized into five classes (16.00–27.90%, 27.91–36.90%, 36.91–45.47%, 45.48–55.60%, and 55.61–66.20%).

3.1.14 Building conditions

In the census data, the houses were categorized into three groups: good, liveable, and dilapidated. The dilapidated houses will be more prone to flooding. The percentage of dilapidated houses in the district ranges between 0.00 and 11.00% (Fig. 9a) and is grouped into five classes (0.00–1.40%, 1.41–3.20%, 3.21–5.00%, 5.01–7.20%, and 7.21–11.00%).

Fig. 9
figure 9

a. Building condition; b. Household size; c. Child population; d. Marginalized population 1

Full size image

3.1.15 Household size

The larger the household size, the greater the number of people impacted and the severity of the damage, and also, they have to share the resources. (Deepak et al. 2020). In this study, only houses with five or more residents were selected. The percentage of houses with five or more residents ranges between 10.13 and 16.60% (Fig. 9b), and is categorized into five classes (10.13–11.16%, 11.17–12.06%, 12.07–12.68%, 12.69–14.10%, and 14.11–16.60%).

3.1.16 Child population

Children are particularly vulnerable to flood hazards as finding higher ground is more difficult for them and thus, increasing the likelihood of drowning-related fatalities (Rasch 2016). The child population is categorized into five classes: 7.11–7.90%, 7.91–8.51%, 8.52–9.16%, 9.17–10.18%, and 10.19–13.25% (Fig. 9c).

3.1.17 Marginalized population 1

Scheduled caste (SC) and scheduled tribe (ST) are among India's most socio-economically disadvantaged groups. In this study, marginalized population 1 represents SC, whereas 2 denotes ST category. Areas with a higher number of SC residents are therefore more vulnerable to flooding. The percentage of the marginalized population 1 (SC) in the study area ranges from 0.76 to 22.31% (Fig. 9d) and is categorized into five: 0.76–3.74%, 3.75–5.63%, 5.64–8.33%, 8.34–11.92%, and 11.93–22.31%.

3.1.18 Marginalized population 2

As mentioned earlier, the ST population is a socio-economically weaker group in India. Hence, they are more vulnerable to flooding. The percentage of the marginalized population 2 (ST) in the study area ranges from 0.09 to 30.40% (Fig. 10a) and is categorized into five: 0.09–0.89%, 0.90–1.91%, 1.92–5.45%, 5.46–15.12%, and 15.13–30.40%.

Fig. 10
figure 10

a Marginalized population 2; b access to information; c building wall type

Full size image

3.1.19 Access to information

Vulnerability is influenced by one's capacity to access pertinent hazard information (Rasch 2016).Broadcasting of flood warnings through television and radio are frequent, and the community without access to television/radio will be more vulnerable to flooding. The percentage of people with access to information (television/radio) is categorized into five: 77.30–81.70%, 81.71–85.10%, 85.11–87.85%, 87.86–91.30%, and 91.31–97.30% (Fig. 10b).

3.1.20 Building wall type

The strength of a house's walls will determine how vulnerable it is to flood damage (Rasch 2016). The materials for the wall were grouped into grass/thatch/bamboo, plastic/polythene, mud/unburnt brick, wood, stone not packed with mortar, stone packed with mortar, G.I./metal/asbestos sheets, burnt brick, concrete, and others. Houses with walls made of burnt bricks and concrete are more resistant to flooding. The majority of the houses in the study area have walls made of burnt bricks and concrete and hence, those houses were selected for the modelling. The walls constructed using materials other than burnt bricks and concrete will be more vulnerable to flooding. The percentage of buildings with walls made of burnt bricks and concrete ranges between 0.50 and 90.80% (Fig. 10c) and is categorized into five classes (0.50–18.60%, 18.61–37.63%, 37.64–52.22%, 52.23–67.00%, and 67.01–90.80%).

3.2 Flood vulnerability

The flood vulnerability of Kottayam district is depicted in Figs. 11 and 12. The very high flood vulnerable zones are mainly confined to the western portion of the study area. This is due to the high to very high physical-environmental vulnerability and socio-economic vulnerability of this area. According to this study, a total of 12.29% and 11.81% of the district are categorized into very highly vulnerable zones by the AHP and F-AHP models, respectively (Table 12). The ROC curve analysis proved that both the models are effective, with an outstanding AUC value of 0.94 (Fig. 13). However, the AHP model (0.946) provided slightly better results than the F-AHP model (0.943). According to the vulnerability created using the AHP model, the panchayats (Hamlet level LSG), namely Neendoor, Vechoor, Aymanam, Arpookara, Chempu, Thalayolaparambu, Kumarakom, Thiruvarppu, Kaduthuruthy, Vazhappally, Kurichy, Udayanapuram, and Maravanthuruthu, and the municipality of Changanassery, are the very high flood vulnerable zones. From the study, it was found that a major portion of the very high flood vulnerable zone is agricultural land, followed by wetlands (Table 13).

Fig. 11
figure 11

Flood vulnerability: AHP method

Full size image
Fig. 12
figure 12

Flood vulnerability: F-AHP method

Full size image
Table 12 Area and percentage of flood vulnerable zones
Full size table
Fig. 13
figure 13

The ROC curves: flood vulnerability

Full size image
Table 13 Land use/land cover classes in each flood vulnerable zones
Full size table

3.3 Physical-environmental vulnerability (PEV)

The vulnerability maps for the physical-environmental indicators are depicted in Figs. 14 and 15. Major portions of the high and very highly vulnerable zones are confined to the western part of the study area. The area of each vulnerable zone is shown in Table 14. From the modelling, it is confirmed that slope, soil, LULC, stream density, and geomorphology are the major causative factors. The high and very high vulnerable areas have lower slopes, clayey soil, agricultural land (paddy fields), lowest stream density, and are coastal plains. A considerable percentage of the area of the high and very highly vulnerable zones has lower NDVI values, higher WRI, MNDWI, and NDBI values. Thus, it is proved beyond doubt that all the factors selected for the modelling are relevant. The AUC scores of 0.88 and 0.91 for the vulnerability maps created using the AHP and F-AHP methods confirm that the results are excellent and outstanding for these two models, respectively (Fig. 16). The AUC score of 0.91 proves that the F-AHP method has better prediction accuracy than the AHP method. According to the map created using the F-AHP method, the high and very high vulnerable zones together constitute 20.73% of the district.

Fig. 14
figure 14

Physical-environmental vulnerability: AHP method

Full size image
Fig. 15
figure 15

Physical-environmental vulnerability: F-AHP method

Full size image
Table 14 Area and percentage of vulnerable zones (Physical-environmental)
Full size table
Fig. 16
figure 16

The ROC curves: physical-environmental vulnerability

Full size image

3.4 Socio-economic vulnerability (SEV)

Figures 17 and 18 show the vulnerability maps for the socio-economic indicators. The ROC curve analysis (Fig. 19) proved that the F-AHP model (with an AUC score of 0.953) is better than the AHP model (with an AUC score of 0.92). Hence, the F-AHP model was selected as the best model. Both the models provided outstanding results. According to the vulnerability map created using the F-AHP model, panchayats viz. Kurichy, Vazhappally, Kaduthuruthy, Kanjirappally, Mundakkayam, Erumely and Changanasserry, and Erattupetta municipalities (middle level LSG) are very highly vulnerable. Most of the highly vulnerable LSGs are concentrated in the low-lying regions of the district. The highly and very highly vulnerable zones have lower literacy rates, lower access to information, a comparatively higher child population, larger household sizes, larger MP2, and a lower percentage of houses with concrete roofs. The low socio-economic conditions of the people living in the high and very high vulnerable areas impact egregiously on their lives and property as a result of a natural disaster. Even a frail disaster might lead to great loss and damage to society and its people.

Fig. 17
figure 17

Socio-economic vulnerability: AHP method

Full size image
Fig. 18
figure 18

Socio-economic vulnerability: F-AHP method

Full size image
Fig. 19
figure 19

The ROC curves: socio-economic vulnerability

Full size image

4 Discussion

The vulnerability studies, melding the PEV and SEV indicators, reveal that AHP and F-AHP as ideal methods. Through the use of the consistency measure, the AHP model improves the decision-making process (Olson 1988). During pairwise comparisons of factors, AHP allows for a certain level of inconsistency (Afolayan et al. 2020). Fuzzy logic, a method based on possibility theory (Thompson et al. 2012), can be highly useful in describing complex systems, especially those containing ambiguities and non-linearities (Chanal et al. 2021). The fuzzy logic method is advantageous since it is simple and easy to understand, and it does not require a large amount of data for training purposes. Many problems involving imprecise and uncertain data can be solved using the fuzzy logic method (Afolayan et al. 2020). F-AHP is an AHP method based on fuzzy logic theory (Putra et al. 2018). F-AHP will be used to address AHP's inability to handle imprecision and subjectivity in judgments (Carnero 2020; Liu et al. 2020). A few previous studies (cf. Ali et al. 2019; Feloni et al. 2020) compared the AHP versus FR models, and the AHP versus F-AHP models for flood vulnerability modelling. Moreover, the reliability of using such models depends on the factors that are selected for the analysis. The study conducted by Feloni et al. (2020) hasn’t considered the SEV indicators. Deepak et al. (2020) applied the AHP method and Random Forest models for demarcating the flood vulnerable zones of a municipality in the Ernakulam district (India). They considered both PEV and SEV indicators, but the study was limited to a smaller area. The present study found that the F-AHP model is more effective than the AHP model for the assessment of PEV and SEV, whereas the AHP model was found to be slightly better than the F-AHP for the flood vulnerability assessment. Akshaya et al. (2021), Khashei-Siuki et al. (2020), and Tripathi et al. (2021) also found that the F-AHP model is effective than the AHP model.

According to the PEV modelling, very high and high vulnerable zones have lower slopes, clayey soil, agricultural land, the lowest stream density, and are coastal plains. Lower NDVI, higher WRI, MNDWI, and NDBI values are found in a significant percentage in these zones. The clayey soil with poor infiltration capacity and lower slopes are ideal conditions for flooding. This observation concurs with the findings of Samanta et al. (2018), where they found that regions with poorly drained soil and lower slopes (0–5°) are the primary reasons for flooding. The low-lying, flood-prone areas of the Kottayam district are agricultural land (especially paddy fields) and coastal plains. These geomorphic/physiographic conditions usually facilitate flooding.

Based on the SEV modeling, highly and very highly vulnerable zones have lower literacy rates, lower access to information, a comparatively higher child population, larger household sizes, larger marginalized population 2, and a smaller percentage of dwellings with concrete roofs. Literacy can have a direct and indirect influence on disaster vulnerability (Hoffmann and Blecha 2020), which is defined as the ability to foresee, cope with, resist, and recover from natural hazards. Literacy can directly influence the knowledge, capabilities, skills, and perceptions acquired by individuals that allow them to effectively prepare for and deal with disaster shocks (Hoffmann and Blecha 2020). Indirectly, literacy makes individuals and households have indirect access to material, informational, and social resources, which can assist in reducing disaster vulnerability (Hoffmann and Blecha 2020). The findings of this study concur those of Sam et al. (2017), where they identified low literacy rates and weak housing structures are the primary causes for flood vulnerability. According to Peek (2008), children are psychologically vulnerable and may suffer from post-traumatic stress disorder during such disasters. They are also physically vulnerable to death, injury, illness, and abuse, and disasters often impair or delay their educational progress (Peek 2008). Moreover, the lack of financial resources of the marginalized populations to recover from disasters, are regarded as vulnerable to disasters (Fucile-Sanchez and Davlasheridze 2020).

4.1 Advantages of this or similar studies

The increasing frequency of flood occurrences due to heavy rainfall, and subsequent overflowing and discoursing of rivers in the study area led to heavy financial hardship and distress among individuals and governments. Usually, most countries have contingency funds to face any kind of disastrous situation. In the case of India, from 2006 onwards, both the center (Federal Government) and the states have established the National Disaster Response Fund (NDRF) and the State Disaster Response Fund (SDRF), respectively, as per the enactment of the Disaster Management Act 2005 by Parliament. These funds are only utilized as per the list of items and the norms of relief assistance that are included in its guidelines. As per the norms of the NDRF and SDRF, the government of India is paying 0.4 million INR for the death of a person due to flooding; ~ 0.1 million INR for a fully damaged house (in plain areas); and ~ 0.04 million INR/hectare for substantial loss of agricultural land (only to small and marginal farmers) (https://sdma.kerala.gov.in/sdrf-norms/). These assistance funds are significantly insufficient when compared to the actual value of the loss or damage to life and property. As the relief assistance was inadequate, the Kerala government increased the amount for some of the items. This additional fund is allocated from the Chief Ministers' Disaster Relief Fund (CMDRF). Hence, the exchequer is spending an enormous amount of money on flood damage. Thus, this sort of study will help in identifying areas vulnerable to flooding and taking precautionary measures so that the effects of flooding can be minimized. The huge amount of money saved through this can be utilized for implementing mitigation measures. Priority areas that require mitigation measures can also be identified through studies of this nature.

Akin to this, measures that are adopted in developed countries like the US, which do not provide any assistance to people for flood-related deaths, injuries, and damage (agriculture loss, infrastructure loss, and so on) from their budget, rather pay through flood insurance, and such options should also be explored by developing countries like India. For example, the Federal Emergency Management Agency (FEMA) manages the National Flood Insurance Program (NFIP) for the US, which is provided to the public by a network of more than 50 insurance firms (https://www.fema.gov/flood-insurance). Hence, implementation of disaster risk financing strategies is of utmost importance to reduce the burden of the government. The majority of the people in Kerala have a higher income than the national average, where the replacement cost of the average house would be more than 4–5 million INR (UNDP 2018). Implementation of a suitable disaster risk financing strategy is of utmost importance to society. Presently, the disaster risk financing schemes available in the state are mostly ex-post mechanisms such as budget reallocations and SDRF/NDRF, rather than ex-ante in nature, like insurance and catastrophic bonds (Government of Kerala 2019b). Hence, by implementing new flood insurance schemes with the support of private insurance companies, a huge amount of money spent from the budget can be saved and utilized for flood mitigation measures.

5 Conclusions

Before delving into the significance of this study, the major constrains are: this study utilized the 2011 census data, which is the only available valid data published by the Indian government. The next census data, scheduled for 2021, is overdue in lieu of the COVID-19 pandemic. Due to a lack of authentic data on the elderly population, this was not considered in this study. The data on the elderly population is very relevant for flood vulnerability assessment, as the elderly population is more vulnerable to flooding. Also, due to a lack of an adequate number of weather stations, the rainfall (the triggering factor) data was not considered in this modelling. Moreover, the gridded data from the India Meteorological Department (IMD) is available for free, but it has a lower resolution (~25 km).

And the following are the study's key findings:

  • The AHP and F-AHP methods classified 12.29% and 11.81%, respectively, of the study area as very highly vulnerable to floods.

  • Soil is the most important PEV indicator, followed by LULC, slope angle, stream density, and geomorphology.

  • A large portion of the high and very high flood vulnerability zones is agricultural land. As a result, the economic loss connected with agricultural land and crop destruction will be significant.

  • Lower literacy rates, lower access to information, a disproportionately higher child population, larger household size, a larger SC population, and a lower number of dwellings with concrete roofs characterize the high and very high flood vulnerable zones.

  • With an AUC value greater than 0.94, both the AHP and the F-AHP methods are outstanding in identifying flood-vulnerable zones.

  • The F-AHP method (AUC values: 0.919 and 0.953) outperforms the AHP method (AUC values: 0.883 and 0.92) for PEV and SEV modelling, while the AHP method (AUC value: 0.946) slightly outperforms the F-AHP method (AUC value: 0.943) for flood vulnerability modelling.

According to this study, both methods were found to be effective in mapping flood vulnerability and can be adapted to other locations with similar physiographical settings. The generated flood vulnerability maps will assist land-use planners in identifying critical locations in Kottayam district, allowing them to implement effective mitigation measures to prevent loss of life, infrastructure, and property, as well as avoid development in these areas.

Based on the findings of the present study, the following recommendations are made:

  • Implement a suitable disaster risk financing strategy and flood insurance schemes.

  • Conduct household survey-based studies to assess the willingness of communities in flood vulnerable and flood-affected areas to pay for flood insurance.

  • Install automated weather stations or automated rain gauges at the village/LSG level. This will enhance the monitoring of rainfall and provide more accurate weather predictions.

  • Clear debris, aquatic weed plants like water hyacinth (Eichhornia crassipes), and other obstructions from the stream channels and widen the stream channels so as to drain out the excess rain water.

  • Ban construction on the river course and floodplains, implement zoning of the areas based on the flood hazard level, and stop development activities that block the stream channels.

  • Make available high-resolution DEM (such as LiDAR data or drone images) to create high-resolution flood susceptibility maps for further micro-scale studies.