Introduction

A flood or inundation can be generally defined as an overflow of water that submerges land that is usually dry. It occurs rapidly, generally within 1 h of rainfall, and sometimes accompanied by landslides, mudflows, bridge collapse, damage to buildings, and fatalities (Hapuarachchi et al. 2011).

Flooding is considered as the most frequent natural disasters during the last decades (Rahmati et al. 2016a, b). The frequency of such natural catastrophe is growing each year (Ushiyama et al. 2017), and results from several factors including climate change, urbanization, population growth, the partial or total cover of watercourse by construction of roads or other structures, and unsuitably dimensioned rainwater drainage channel (Green et al. 2000). Floods impact both socioeconomic and environmental attributes of the affected area (Meyer et al. 2013). Tunis has suffered floods regulary since 1969, caused by land use changes and climate variability. The most violent one was in 2003 and in 2017 whose immediate impacts included loss of human life, the destruction of infrastructures, and damage to property. The areas of Ettadhamen, Ettahrir, Ariana, and Ksar Saïd regions which are located in the Tunis were the worst hint. To date, the infrastructure damage due to flooding causes the intermittent supply of water and poor service delivery of wastewater treatment, transport, electricity, education system, communication network, and health care. Reduction in purchasing power in the floodplains can leave communities economically vulnerable.

Studies to improve stormwater sanitation and flood protection have already been carried out since 1966, including the DHU’s Flood Protection Study launched by DHU and the Japan Cooperation Agency on 1994. Unfortunately, many developments proposed in these studies have not been fully implemented and some projects have negative impacts related to undersizing rainwater channels, or underestimation of hydraulic parameters (DHU 2015).

The 2003 floods showed the limits of the infrastructure and its inability to protect the population against the socioeconomic damage and against the environmental impact. Therefore, it is necessary to assess the flood risk by adopting the appropriate tools in the study area.

Flood risk assessment has been studied by several authors such as Arrighi et al. 2018, Benson et al. 2016, Ortiz et al. 2016, and Shankman and Keim 2016. In the past decades, numerous techniques and methods have been tested in order to simulate flooding and runoff. An important parameter the “Manning roughness coefficient” was integrated in these methods, which comprises a flow resistant factor and a function of land cover in flooded areas (De Roo 1999). A segmentation and classification approach of IKONOS-2 imagery (Van der Sande et al. 2003) aims to produce detailed land cover maps for flood damage assessment as well as detailed maps of roughness coefficients for flood simulation models. The fuzzy mathematical models are introduced in the field order to improve probability estimation and evaluation of flood risk with incomplete data sets (Qiong et al. 2012; Pourghasemi et al. 2012). However, these different techniques and methods lack comparison between parameters in order to obtain the most accurate results and give weights close to reality.

In this study, the aim was to identify flood-susceptible areas in order to find solutions for the assessment and management of floods and to determine which region deserves the most engagement in the development of risk diminution measures or mitigation procedures. Flood mapping and risk analysis of different regions need the use of multiple factors (Poussin et al. 2014). As a result, multicriteria decision analysis (MCDA) methods have been successfully used in various researches and have been recognized as an important tool for analyzing complex decision problems. MCDA integrate a large number of criteria such as technical, environmental, and socioeconomic factors (Liu et al. 2003) to achieve a perfect decision (Ghanbarpour et al. 2013). This process is established using the existing conditions as a reference to identify flood hazards areas (Yang et al. 2011). Coupled to the Geographical Information System (GIS), the MCDA was used for mapping inundation zones. Indeed, the GIS techniques facilitate treatment and analysis of the spatial data and make easy the visualization, interpretation, and evaluation of the MCDA results (Wang et al. 2011). Eight spatial parameters have been used in this study: Land use-Land cover, drainage density, rainfall, lithology, elevation, slope, soil and groundwater level.

Analytical hierarchy process (AHP) has been developed by Saaty in 1980 and it is considered a mathematical technique of multi-criteria decision-making (Matori et al. 2014). AHP technique has been applied to flood susceptibility mapping and has proved its effectiveness, especially in groundwater mapping (Kaliraj et al. 2014; Razandi et al. 2015), landslide susceptibility mapping (Althuwaynee et al. 2014), and in chemical fields (Hanratty and Joseph 1992; Pirdashti et al. 2009).

This technique evaluates the importance of factors, according to weight values. After, the weight determination, the Flood Hazard Index (FHI) can be computed and used to assess the rate of the flooding probability (Elkhrachy 2015). Hazard Index Sensitivity (FHIS) is identified to examine the sensitivity pairwise comparison method against the changes in criteria weights. Consequently, we noticed that the FHIS is efficient in situations as where uncertainty exists in the definition of the importance of each factor (Yahaya 2008).

The new approach proposed in this study is applied for the first time in the region. It allows us to compare between the parameters, and test and classify them according to their intervention in the production of the flood in the studied zone. This proposed approach is applicable in various zones. The scientific results which are derived from this work can assist planners, engineers, and governments to perform proper actions in order to prevent and mitigate the flood occurrence in the future.

Material and methods

Study area

The study area is located in the North-East of Tunisia (Fig. 1a), North Africa, between 10° 1´ 0´´ E to 10° 20´ 30´´ E longitude and from 36° 45´ 0´´ N to 36° 58´ 0´´ N latitude, overlooking the Mediterranean Sea in the East. It extends from Morneg in the South to the Djebel Ammar (328 m) and Ennahli (201 m) anticlines in the North, and it is bounded by the Djebel Sidi Salah and Djebel Ain El Krima anticlines in the West (Fig. 1b). The surface which is concerned by this study is about 524.4 km2 including two sebkhas, that of Ariana (50 km2) and Séjoumi (30 km2) and it even integrates the Lake of Tunis (40 km2). Ariana, Soukra, and Manouba plains, surrounds these humid fields. The Tunis sector is characterized by a semi-arid climate (Emberger 1955) with a hot winter of an annual mean precipitation of 469.5 mm, recorded at five meteorological stations from 1985 to 2015 (DGRE 2015, CRDA 2015). The mean monthly rainfall distribution has shown that the rainiest period extends from October to March, with a maximum mean monthly value in December (77 mm). The driest months of the year are June, July, and August with a minimum recorded in July (3.2 mm). The mean of monthly temperature reaches its maximum (34.5 °C) in August and its minimum (7.7 °C) in January. Monthly inter-annual wind variability ranges from 20 to 27.8 km/h (INM 2015). The mean annual precipitation for the period 1985–2015 is about 469.5 mm, which corresponds to the mean annual rainfall water volume of 246.3 × 106 m3. Potential evapotranspiration has been computed using the Thornthwaite method (Thornthwaite and Mather 1957). The real evapotranspiration (ETR) is about 1097.76 mm and the potential evapotranspiration (ETP) is about 41.9 mm/years.

Fig. 1
figure 1

a Location of Tunisia. b Location of the study area

Full size image

Hydrographical network covering the study area is composed of permanent and impermanent streams. There are three major rivers discharge to the Essijoumi Sebkha (Gueriana, Bardo, El Meleh) and two rivers (Guereb and Roriche) discharge to the Tunis Lake. The geological formations outcropping in the study area are composed of Triassic, Cretaceous, Paleogene, Neogene, and Quaternary series. The Triassic is exposed at jebel Ammar and Jebel Mergueb. It is composed mainly of gypsum, gypsum clay, marl, and dolomite (Boutib 1998). The high structures (Djebel Ammar 328 m, Djebel Ennahli 201 m) are covered essentially by limestone or marl-limestone alternation from Jurassic to Eocene ages (Pini 1971; Pini and Kchouk 1971). The hills are composed mainly of sandstone, clay, and sand of Olig-Pliocene (Bujakka et al. 1971). The continental Mio-Pliocene series lay unconformably on the Oligocene deposits (Ben Ayed et al. 1979; Boutib 1998). The Miocene is characterized by sandstone, sand, and sandy clays (Lajmi 1968). The Pliocene strata expose mainly at Djebel Ain El Krima and Ariana-Belvédére where they are represented by fluvial-deltaic sediments (Boutib 1998).

Finally, the Quaternary deposits cover the greater part of the study area, usually composed of silt, conglomerate, clay, silt, and sandy limestone accumulation (Oueslati 1994). The phreatic aquifer of the sector is mainly formed during the Plio-Quaternary. Rainwater directly infiltrates into the soil, recharging the aquifer (DHU 2015). In the major part of the area, the water table depth is about 6 m while it is relatively deep (greater than 10 m) in a small area around the hills and Djebels. The transmissivity varies between 3.10−3 and 5.10−3 m2/s.

In the arid and semi-arid regions, especially in the rapidly expanding towns and cities, intense urbanization has led to a high demand of groundwater resources (Farah et al. 1997; Murthy 2000) (Table 1). Consequently, the groundwater quality has been degradated (Fritch et al. 2000). There are various sources of groundwater pollution such as discharge of untreated sewage in water courses. It is the most important source of pollution of both surface and groundwater (Naik et al. 2008).

Table 1 Resources and exploitation of ground water (DGRE 2015)
Full size table

The Tunis region has been undergoing an intense population growth. It increased by 814,853 thousand people between 1994 and 2014 (INS 2015) (Table 2). Results increased the demand for housing and new residences (Givisiez and Oliveira 2009), urban expansion, and industrial centralization (Chabbi and Abid 2008). Fast, visible, and critical changes in land use and settlement have been taking place at a remarkable, reshaping the landscape and affecting the environment (Weng 2001). Waterproofing rate of the soil (IMP) varies between 10 and 90% (DHU 2015). More than half of the Tunis sector watershed has been altered by human activities. A natural flow of rainwater is seriously prevented, because the rain that would normally seep into the soil has to flow across construction surfaces and through storm-water pipes, increasing the volume and velocity of runoff (Tucci et al. 1995) causing environmental risks, such as flood hazard.

Table 2 Population growing (INS 2015)
Full size table

Methodology

The methodology adopted in the current study is summarized in the diagram of the Fig. 2. The main goal of this study is to define flood-susceptible areas by using the GIS environment (Forte et al. 2006), SRTM (Shuttle Radar Topography Mission), and Remote sensing data (Brivio et al. 2010). The second objective is to calculate Flood Hazard Index (FHI) by incorporating multiple criteria decision-making (MCDM) (Drobne and Lisec 2009) and analytical hierarchy process (AHP) (Generino et al. 2014).

Fig. 2
figure 2

Conceptual model based on multi-criteria decision analysis (MCDA) and analytical hierarchy process (AHP) for flood susceptibility mapping by GIS

Full size image

Initially, various data sources were used and introduced in ArcGIS including satellite images, time series observations, spatial data, and geological and topographic maps. However, a selected set of parameters were integrated into the present study according to their significant role in flood area selection. Only eight thematic layers were approved for the inundation zoning including drainage density, land use/land cover, rainfall, slope, soil type, elevation, lithology, and groundwater level.

The slope map of the study area was deduced directly from DEM (digital elevation model) data extracted by SRTM global 1 arcsecond (30 m). The slope percentage values were estimated by the spatial analysis tool in Arc-info-GIS 9.3. SRTM was further utilized to develop the elevation map of the area using ArcGIS software. The drainage density network was created using the line density analysis Tool in ArcGIS.

The piezometric map of the year 2015 was estimated by the GIS model kriging tool. It allows, as a result, the groundwater level surface to be defined at ungauged locations from local information. The annual mean rainfall was calculated using monthly rainfall data collected from the General Department of Water Resources (DGRE) of Tunisia in five rain gauges for 30 years from 1985 to 2015. Subsequently, an annual mean rainfall map was generated and saved as a layer in ArcGIS platform. Either the soil map taken from the General Department of Water Resources was georeferenced and digitized as a layer in ArcGIS. Lithological map was produced by assembling the four geological maps of Tunis, La Goulette, La Marsa, and Ariana sectors on a 1/50000 scale. Afterwards, land use/land cover map was obtained by the treatment of satellite images from Landsat 8 Sensor on a scale of 1/500000 using ENVI model and converted into raster format (30 m resolution) legible by ArcGIS.

The factors used to assess the flood-susceptible areas do not have the same degree of influence. Evaluating the effect of each factor alone does not provide the complete required picture. The integration of all factors together was necessary. Indeed, the influence of factors on each other is determined using a schematic sketch according to Shaban et al. (2001). This method allows to determine the weight of each factor on all other factors. One (1) point is assigned to a major effect and half point (0.5) to a minor effect. Next, the rate factor is calculated as the summation of the points attributed to the effects of the factor (Shaban et al. 2006).

To conduct a comprehensive assessment of the influence of each factor on flood generation in the study area, a weight value equivalent to its relative importance was assigned to each factor. This weight was determined by pairwise comparison process proposed by Saaty in 1980.

The first stage of the AHP technique consists to define of the decision problem. Preparing a pairwise comparison matrix constitutes the second stage of the technique (Yaralıoğlu 2004). An n*n dimensional pairwise comparison matrix of the conditioning factors is prepared at this stage. The importance value scale proposed by Saaty (1980) is respected (Table 3). In this stage, every factor was given an arithmetic value between 1 and 9, depending on its significance compared to the other factor which is coupled to it. An arithmetic value of 1 means that both factors are equally significant. However, a value of 9 indicates that a row factor is much more significant than the corresponding column factor which it has been compared to Saaty 1980.

Table 3 Saaty scale (1980)
Full size table

The proposed methodology suggests a pairwise comparison, using an 8 × 8 matrix. Afterwards, a numerical index named consistency ratio (CR) was used to examine the consistency pairwise comparison matrix (Table 4).

Table 4 Random inconsistency indices
Full size table

In Table 5, the criteria selected are sorted in a hierarchical manner. The values of each row characterize the importance between two parameters. The first row of the table illustrates the importance of land use-land cover in regard to the other parameters which are placed in the columns. For example, land use-land cover is significantly more important from the groundwater level, and therefore assigned the value 9. Land use-land cover has been considered as the most important parameter since flooded areas are often located in urban zones. These weights were normalized according the method of the AHP. The model was integrated into a GIS system in order to realize a flood hazard map (Stefanidis and Stathis 2012).

Table 5 A matrix of pairwise comparisons of eight criteria for AHP process
Full size table

Afterwards, the consistency ratio (CR) was determined. If CR ratio is superior to 0.1 the judgments may be too inconsistent to be reliable. However, when CR is equal to 0, the judgments seem to be absolutely consistent (Elkhrachy 2015). To calculate the value of (CR), the consistency index (CI) is divided by the random index (RI) Eq. (1). The RI is dependent on the number of criteria (Kazakis et al. 2015).

$$ \mathrm{CR}=\mathrm{CI}/\mathrm{RI} $$
(1)

Where CI is the consistency index, and RI is the random inconsistency indices.

$$ \mathrm{CI}=\left(\uplambda -\mathrm{n}\right)/\left(\mathrm{n}\hbox{--} 1\right) $$
n:

number of factors and λ: average value of the consistency vector.

In the present case of study, the eight factors used for hazard area mapping are examined separately. We also divided each factor into a number of classes and gave a weight for every class. The maximum and minimum weights for each class of every factor vary respectively from 1 to 10. Next, the weights assigned to different classes of the individual criterion and their normalized weights are calculated. The class weights used are related to the effect of each class to the flood hazard. All factors use the same classification; however, they do not contribute to the same amplitude (Stefanidis and Stathis 2012).

First, the information collected from different data sources is processed in the GIS. Then, the FHI is calculated by the following expression Eq. (2) to evaluate the flooding probability rate (Elkhrachy 2015):

$$ \mathrm{FHI}=\sum \limits_{\mathrm{i}=1}^{\mathrm{n}}\mathrm{Wi}\ \mathrm{Ri} $$
(2)

Where Ri is the rating of the factors, Wi is the weight of each factors, and n correspond to the number of the parameters.

Therefore, Eq. (2) is written in the next form:

$$ \mathrm{FHI}={\mathrm{W}}_{\mathrm{LULC}}\ {\mathrm{R}}_{\mathrm{LULC}}+{\mathrm{W}}_{\mathrm{EL}}\ {\mathrm{R}}_{\mathrm{EL}}+{\mathrm{W}}_{\mathrm{LI}}\ {\mathrm{R}}_{\mathrm{LI}}+{\mathrm{W}}_{\mathrm{R}\mathrm{F}}\ {\mathrm{R}}_{\mathrm{R}\mathrm{F}}+{\mathrm{W}}_{\mathrm{DD}}{\mathrm{R}}_{\mathrm{DD}}+{\mathrm{W}}_{\mathrm{SP}}\ {\mathrm{R}}_{\mathrm{SP}}+{\mathrm{W}}_{\mathrm{SL}}{\mathrm{R}}_{\mathrm{SL}}+{\mathrm{W}}_{\mathrm{GL}}{\mathrm{R}}_{\mathrm{GL}} $$

At the end, single parameter sensitivity analysis is determined to assess the impact of each parameter on the vulnerability index. This test of sensitivity is produced because the numerical values attributed to these parameters are arbitrary (Zghibi et al. 2016). The initial arbitrary weights by AHP are replaced with effective weights (Kazakis et al. 2015; Stefanidis and Stathis 2012), and calculated according to the next equation Eq. (3):

$$ \mathrm{W}=\left[\left({\mathrm{P}}_{\mathrm{r}}\ast {\mathrm{P}}_{\mathrm{w}}\right)/\mathrm{V}\right]\ast 100 $$
(3)

Where W is the effective weight of each parameter, Pr is the parameter’s rating, Pw the parameter’s weight, and V is the aggregated value of the applied index.

The effective weights are then employed to calculate the revised Flood Hazard Index (FHIS) by the sensitivity analysis method. For the evaluation of flood risk-prone areas, the parameters’ classes of each factor selected should be adjusted to the local characteristics of each study area.

Results and discussion

Multi-influencing factors of flood Hazard zones

Slope map

Topography has a main effect on flood production and redistribution. The slope percent is an indicator surface to define the flood sensitivity (Youssef et al. 2011). An area with a low slope has a higher risk because it is easily to be submerged by flood. On the other hand, a surface with a high slope is less exposed to flood risk, because these floods can be rapidly drained to down slope (Chen et al. 2015). Therefore, this factor plays an important role in determining the amount of surface runoff and infiltration, and thus affecting flood sensibility (Rahmati et al. 2016a, b). Based on the slope, the study area can be divided into five slope classes (Fig. 3a). This weight of each class (betwen 1 at 10) is according to its intervention in the generation of the flood. The weight 10 is attributed to the first category (0 to 2%), composed by the nearly flat terrain highly exposed to the inundation. This range represents about 54.4% of the study area.

Fig. 3
figure 3

a Slope. b Elevation

Full size image

Elevation map

The altitude has significant impact on the propagation of flooding in the study area. In addition, this parameter has an important role in the control of the followed direction movement and in the depth of the flood. The resulting map was grouped into five classes showed in Fig. 3b.

Rainfall map

Averages of annual rainfall were calculated according to monthly rainfall data from five rain gauges. We used the data for 30 years between 1985 and 2015, where the mean annual precipitation is around 469.5 mm. Subsequently, a mean annual rainfall map was generated by employing interpolation techniques of kriging used by ArcGIS software (Fig. 4a). Precipitation scale was divided into 4 classes from 440 to 520 mm. The first class represents values under 460 mm (6.62%), the second one represents values between 460 and 480 mm (77.27%), the third one represents values between 480 and 500 mm (12.5%), and the fourth one represents values over 500 mm (3.59%). A weight of 2 is proposed to characterize the lowest precipitation interval and a weight of 10 is attributed to the highest level range.

Fig. 4
figure 4

a Rainfall. b Lithology

Full size image

Lithological map

The fourth significant parameter used to define the flood zone is lithology (Svoboda 1991). This factor is related to the permeability and the capacity of water storage that vary according to rock types (Oikonomidis et al. 2015). When the rock is impermeable, the rainfall water infiltration is less important, which favors extension of runoff surfaces and amplify flood event effects (Kourgialas and Karatzas 2011). The lithology map of the study area was produced by an analysis in a GIS environment of the geological maps of Tunis sector that were prepared at the 1:50000 scale by the Topography and Cartography Office (OTC 1980) in Tunisia. The study area was covered by various types of lithological formations extending from the Triassic to the Quaternary (Bujakka et al. 1971; Maamri and Ben Salem 2000) (Fig. 4b).

Land use/land cover map

Land use/land cover constitute the principal factor to identify areas, which are susceptible to be submerged by flooding (Nuissl et al. 2009). Land use influences infiltration rate. For example, forest and vegetation favor water infiltration (Kourgialas and Karatzas 2011). On the other hand, urban area, which is mostly made from impermeable surfaces and bare lands, increase the storm runoff (Tehrany et al. 2014). These land use-land cover classes are determined from the satellite images. Therefore, we obtained the five classes of the land cover, such as urban areas, forest, irrigated land, crop land, and bare soil. A considerable surface of the studied area is urbanized (51, 5%) (Fig. 5a).

Fig. 5
figure 5

a Land use-land cover. b Soil

Full size image

Soil map

In the current study, the thematic soil type map was displayed in a GIS layer that ranked soils based on their textures and structures. Effectively, the texture and structure of each soil type can influence significantly their permeability and consequently their capacity of water storage. Thus, we assigned the lowest weight value (2) to sand class and the highest one (10) to the soil type corresponding to urban areas (Senanayake et al. 2016) (Fig. 5b).

Groundwater level map

In fact, during floods, a rapid change in groundwater outflow was observed due to the new infiltrated water during heavy rainfall. This water flow coming in a brief time range can obviously increase the groundwater level and affects the soil infiltration capacity (Troch et al. 1993). In this way, it increases the hydraulic pressure in the aquifer and in areas where groundwater is very near to surface (Upton and Jackson 2011). The depth of the groundwater level was measured all over the piezometric network and the spatial interpolation of these piezometric values for different time scale allows a mapmaking of groundwater level surface to be constructed using GIS software. The GIS is one of the most appropriate softwares for contouring groundwater level data, which is generally a non-stationary variable (Gundogdu and Guney 2007). The depth of the water table was prepared by considering the highest elevations of the static water levels (2015). The resulting map was classified into five classes and areas underlined by shallow depth (0–2 m depth) were considered as the most vulnerable to the risk of flooding (Fig. 6a). Nevertheless, the water table depth in the Tunis sector is lower than this value and varies between 0 and 32 m above sea level, decreasing gradually from the West to the East. According to the water-table depth, 70% of the Tunis region is characterized by a very high to moderate flood susceptibility. The less susceptible sites (> 8 m) with ratings 1 and 2 represent nearly 30% of the total area. However, the most sensitive sites with ratings 8 and 10 are located in the center part and near the coast.

Fig. 6
figure 6

a Groundwater level. b Drainage density

Full size image

Drainage density map

Drainage density is an important element of flood control measures (Ogden et al. 2011), which expresses the length of rivers per unit of area (Km/km2) (Magesh et al. 2012). The drainage density map was produced employing the line density analysis tool in the Arc GIS environment. After that, a reclassification into four classes (from low to high), according to the class boundaries of Table 6, has been conducted. High drainage density (1.5 to 1.9 Km) was recorded in the relief of the study area. On the other hand, the low drainage density (0–0.5 km) is recorded on the plains (Fig. 6b). The raster drainage network density file was assigned a weight of 6% (0.06) in the calculation of the flood susceptibility final map and according to AHP methodology.

Table 6 Classes of the factors and according weights
Full size table

Interrelationship and pairwise comparisons of the factors

Multi-influencing factors of flood-susceptible areas

The eight influencing factors, established in the preceding section, were used to estimate the flood sensitive areas and generate the corresponding map, such as lithology, slope, elevation, land use-land cover, drainage density, soil type, rainfall, and groundwater level. The dangerous zones cannot be estimated by considering the effect of each factor separately, since all factors do not have the same degree of influence in the hazardous areas (Kazakis et al. 2015). The factors weights were given by applying the relation determined by Shaban et al. (2001), which take into account the influence of each factor on all other factors. A continuous line between two factors indicates that one factor has a major and principal effect on the other pointed by the arrow. A discontinuous line between two factors indicates that one factor has a minor and indirect effect on the factor pointed by the arrow (Fig. 7).

Fig. 7
figure 7

Interrelationship between that multi-influencing factor concerning the flood-susceptible areas

Full size image

To quantify the two types of effects, one (1) point is attributed to a major effect and half a point (0.5) to a minor one (Shaban et al. 2006). Next, the rate factor is calculated as the summation of the points attributed to the effects of the factor (Shaban et al. 2006). According to this relation, the rate weight of the land use-land cover factor is significantly higher than the groundwater level.

Relative weights of criteria

A pairwise comparison is shown in Table 5, using an 8 × 8 matrix, where diagonal elements are equal to 1. The first row of the table illustrates the importance of land use/land cover in regard to the other parameters which are placed in the columns. Since our basin containing urban areas, land use/land cover has a higher influence in flood occurrence compared to the other factors. Land use/land cover is significantly more important than groundwater level, and therefore assigned the value 9. On the first hand, elevation and lithology are considered less important than land use/land cover. On the other hand, they are considered more important than rainfall, drainage density, slope, and soil-type factors. Although, all these parameters were prioritized in other previous studies.

Determined consistency ratio (CR)

RI is dependent on the number of criteria (Matori et al. 2014) (Table 4). In this study, the criteria are eight and the result RI = 1.41.

Finally, the consistency ratio has been calculated CR = 0.013 (Eq. 4), since CR value is inferior than 0.1 and the weights consistency are accepted.

$$ \mathrm{CR}=\mathrm{CI}/\mathrm{RI} $$
(4)
CI:

Consistency index and RI = Random inconsistency indices.

CI:

(λ -n) / (n – 1)

n:

number of factors and λ: average value of the consistency vector.

λ:

8,13

CI:

0,019

CR:

0,013 < 0,1

Classes of the factors and according weights

The contribution of each class to the hazardous areas expressed by a percentage is demonstrated in the last column of Table 6. The percentage of each class was computed as the ratio of the total factor weight to the grand total. Based on the results, the weight of the urban area is significantly higher than the other class of land use-land cover factor (bare soil, irrigated land, crop land, and forest).

Flood Hazard Index (FHI)

The FHI is found to evaluate the rate of flooding probability (Elkhrachy 2015), which was calculated by the following equation:

$$ \mathrm{FHI}=0.23\ {\mathrm{R}}_{\mathrm{LULC}}+0.18\ {\mathrm{R}}_{\mathrm{EL}}+0.18\ {\mathrm{R}}_{\mathrm{LI}}+0.15\ {\mathrm{R}}_{\mathrm{R}\mathrm{F}}+0.10{\mathrm{R}}_{\mathrm{DD}}+0.08\ {\mathrm{R}}_{\mathrm{SP}}+0.05\ {\mathrm{R}}_{\mathrm{SL}}+0.03\ {\mathrm{R}}_{\mathrm{GL}} $$

Flood susceptibility map

The eight maps that were developed after the weighting method and AHP were combined and superposed in the (GIS) environment. The flood susceptibility map supports the identification of inundation sensitive areas, while records of historical flood events verify the accuracy of the methodology.

The final map of the flood sensitivity area was constructed and classified into five major classes with flood potentiality from very low to very high. Respectively, we find that 2.85% of the study area represents the very low class, 7.5% of the area represents the low class, 17.95% of the area represents the moderate class, 20.63% of the area represents the high class, and 51.06% of the area represents the very high class (Fig. 8). It appears that areas of very high potentiality occupy a surface of 270.37 km2, which covers the largest part of the study region. Historically, this region has suffered from several floods (on 1989, the Gulf of Tunis, 2003 the Tunis sector, October 2007 the area of Sabbalet Ben Ammar of the region of Ariana and in 2017, the Manouba region).

Fig. 8
figure 8

Flood susceptibility map

Full size image

Obviously, we can observe the coincidence between the zones of the highest flooding risk and the urban areas. This index accordingly received a higher weight. However, this effect is especially critical to inundation and produces relatively a significant impact on the resultant flood susceptibility map. It is also interesting to notice that three criteria, land use/land cover, elevation, and lithology (Figs. 5a, 3b, and 4b), received high weights which can be obviously observed, particularly in the Northeast, south-east, and central area. However, the influence of other criteria (rainfall, drainage density, slope, soil, and groundwater level) are not visible. These criteria are assigned to the smallest weights and do not have a significant impact on the inundation phenomenon. Therefore, these results provide a baseline information, which needs to be considered during flood management.

Sensitivity analysis FHIS

Single parameter sensitivity was developed to evaluate the impact of each parameter on the flood hazard index. This sensitivity test type is produced because the numerical values attributed to these parameters are essentially arbitrary (Al-Adamat et al. 2003; Zghibi et al. 2016). In terms of sensitivity analysis, the initial arbitrary weights given by the theoretical method (AHP) are replaced with “effective weights” (Kazakis et al. 2015) and calculated according to the following equation:

$$ \mathrm{W}=\left[\left({\mathrm{P}}_{\mathrm{r}}\ast {\mathrm{P}}_{\mathrm{w}}\right)/\mathrm{V}\right]\ast 100 $$
(5)

Where:

W:

the effective weight of each parameter, Pr: the parameter’s rating, Pw: the parameter’s weight and V: the aggregated value of the calculated index.

$$ \mathrm{FHIS}=0.3\ {\mathrm{R}}_{\mathrm{LULC}}+0.21\ {\mathrm{R}}_{\mathrm{EL}}+0.18\ {\mathrm{R}}_{\mathrm{LI}}+0.14\ {\mathrm{R}}_{\mathrm{R}\mathrm{F}}+0.07\ {\mathrm{R}}_{\mathrm{DD}}+0.04\ {\mathrm{R}}_{\mathrm{SP}}+0.05\ {\mathrm{R}}_{\mathrm{SL}}+0.02\ {\mathrm{R}}_{\mathrm{GL}} $$

The effective weights are then used to calculate the revised Flood Hazard Index of the sensitivity analysis (FHIS) (Akbari and Rahimi Shahrbabaki 2011). The FHIS analyze the same parameters signed previously in studied and classified by the FHI, but with different weights (the average effective weight of the sensitivity analysis). FHIS estimated in a range of different values of the criteria and its comparison with the FHI shows the dependence and sensitivity of flood on the parameters selected. For the assessment of flood hazard areas, the parameters classes of each factors selected should be adjusted to the local characteristics of each study area (Kazakis et al. 2015).

Table 7 shows the effective weight presented by the sensitivity analysis and the theoretical weight attributed by the AHP method. The sensitivity analysis concluded that land use-land cover, elevation, rainfall, and lithology have a bigger influence in the studied region. The effective weight of the rainfall, drainage density, slope, and groundwater level (14%, 7%, 4%, and 2% respectively) represented low effective weight compared to their theoretical weight (15%, 10%, 8%, and 3% respectively).

Table 7 Statistics of the effective weights-sensitivity analysis
Full size table

Conclusion

Inundation is natural risk that impacts the populations and structures within the affected areas. The research proposes an efficient methodology to precise delineate the flood-susceptible areas. The spatial multicriteria decision analysis (MCDA) method was used in Tunisia to evaluate the effect of several factors, causing flood phenomenon generation, on the flood hazard occurences and the vulnerability of the flooding areas to each factor. This method analyzes spatially eight parameters such as land use/land cover, elevation, slope, lithology, drainage density, rainfall, soil, and groundwater level. The impact of each parameter on all other parameters was determined by applying the technique proposed by Shaban et al. (2001). Next, the MCDA approach is coupled with AHP techniques and GIS. GIS enhances here the visualization capability and increases the assessment efficiency, the acquisition, storage, retrieval, manipulation, and data analysis to develop information that can support decisions would be made in a shorter time. To estimate flood hazard, the influenced factors were fixed and their corresponding weights in accordance with flood formation were estimated, using analytic hierarchy process and pairwise comparison. This new approach does not require any training stage and it is easy to apply it. Moreover, it is applied to any regions. The main advantage of the method is the property given by the statement “depending on expert knowledge.” However, this is also the main disadvantage. Expert subjectivity, particularly in pairwise comparisons, constitutes the main drawback of the AHP.

This study demonstrates that the land-use/land-cover factor is the main parameter of flooding. The highest weight was assigned to the land use-land cover parameter and the lowest weight to the groundwater level. This clarify that most floods occur due to human intervention to extensive urbanization and the natural torrential environment.

Flood susceptibility map in terms of natural and anthropogenic factors was constructed and classified into five major classes with flood potentiality from very low to very high. Respectively, we find 2.85% (very low), 7.5% (low), 17.95% (moderate), 20.63% (high), and 51.06% (very high). After that, FHI was coupled with a sensitivity analysis process to have a revised index FHIS. This index evaluates the impact of each criterion on the method, which leads to a better understanding of each parameter on the flood susceptibility map.

Finally, the produced maps from the current study can be used as a guide for flood management in the Tunis region. The knowledge about high-risk zones is important for local authorities to manage inundation properly in order to organize the implementation of the appropriate projects for flood protection.