Introduction

Flood is one of the most frequent, widespread, and disastrous natural hazards in Tunisia (Abida and Ellouze 2008). Flash floods happen very suddenly and are difficult to forecast. Flood events generally cause damage to agricultural crops and property, roads and railways, and may even result in the loss of human lives. Millán (2014) affirmed that flash flood hazard is mainly related to the size of the concerned catchment and its geomorphologic characteristics. According to Taubenbock et al. (2011), meteorological conditions and urbanization rates are arguably the most significant driving forces of flood hazard. In arid areas, floods are generally caused by storms of high intensity and are often of relatively limited extent. In these areas, the type of slope and the nature of soil may enhance the potential of rapid surface runoff during intense rainfall. The activation of surface runoff can also be increased by land use modification, urbanization, and fire-induced alteration. Moreover, the dynamics of the socio-economic system and the expansion of urban areas, especially in flood-prone areas, may significantly contribute to raise the damages from flooding events (Elmer et al. 2012).

Gabes Catchment, located in Southern Tunisia on the Mediterranean Sea, is confronted to the impacts entrained by flash floods. The basin is characterized by sparse vegetation, steep slopes, poor soil development, large impermeable areas, high urbanization, and an ever growing population. Besides, rainfall in the region is characterized by its erratic distribution, ranging from extended drought periods to extreme rainfall events resulting in flash floods. In the past, Gabes Region was hit by many flash floods, the most significant of which is the flood of 1962, which resulted in 50 deaths, 7000 persons without shelter and important material losses (Fehri 2014). In the main meteorological station of Gabes City, the analysis of daily precipitation data from September 1983 to December 2014 revealed that intense events represent approximately 23% of the total number of rainy days. The ‘extreme’ and ‘very extreme’ events, estimated by percentile indices, correspond to 49.5 and 60.8 mm, respectively. Lately, an extreme rainfall event hit the Gabes Region in June 2014, causing human deaths and major material losses. This resulted in the stagnation of storm water in the numerous low zones of the study area, endangering thereby human health and causing disastrous environmental impacts.

Various methods, used to assess flood susceptibility, are available in the literature. In the context of lacking knowledge about flood phenomena, expertise is required to provide analyses for decision and hazard management purposes using multi-disciplinary qualitative and quantitative approaches. The former is based on expert knowledge to afford a relative indication of hazard while the latter offers quantitative results, based on calculated data and/or modeling. However, these approaches are complementary. Enormous progress has been made in the development of susceptibility mapping and hazard zoning, whereby much of this progress is based on the extensive use of geographic information system (GIS), survey data and remote sensing techniques (Kachouri et al. 2014). GIS environment presents effective tools for handling, integrating, and visualizing diverse spatial data sets (Dixon 2005). The extension of hazard areas is an important consideration in the prediction of flood hazard and the management of natural resources.

One of the most important tools used in hazard mapping is the combination of an analytical hierarchy process (AHP) method (Saaty 1980) with a GIS platform. This process has received significant consideration among multi-disciplinary decision makers and has demonstrated its value in various studies related to natural hazard evaluation, including soil erosion hazard mapping (Kachouri et al. 2014), flood hazard (Fernandez and Lutz 2010), and landslide susceptibility mapping (Feizizadeh et al. 2013). The AHP represents a powerful tool for the analysis of complex decision problems based on an approach of multi-criteria evaluation, generally involving incommensurable data or factors. The integration of AHP into a GIS environment was shown to be efficient in the development of automated methods for quantifying the spatial variability of flood hazard and the associated problems (Pourghasemi et al. 2012).

The application of AHP method has obtained considerable attention and value in various studies related to natural hazard assessment (Dambatta et al. 2009). However, the uncertainties that are associated with AHP techniques may significantly impact the results. The uncertainties are mainly due to incomplete and inaccurate data which are combined into flooded susceptibility values and parameters used in the combination rules. A strong correlation is presented between data uncertainty and parameters uncertainty, since model parameters are obtained directly from measured data, or indirect calibration (Ascough et al. 2008). The large number of parameters and the heterogeneity of data sources make the uncertainty of results difficult to quantify. Thus, AHP technique should be exhaustively evaluated to ensure its robustness under a wide range of possible conditions. The incorporation of probabilistic uncertainty into the AHP technique is considered to quantify the sensitivity and uncertainty of the proposed model. The beta-PERT distribution has been widely used for modeling expert’s judgments and providing a close fit to normal distributions with little demand for data (Coates and Rahimifard 2009; Lake et al. 2010). This technique uses the most likely, minimum, and maximum values of expert estimates to generate a probability distribution that provides a possibility of measuring the level of confidence in AHP decision (Chen et al. 2011).

The objective of this study is to examine the magnitude and the extent of flood hazard in Gabes Basin using GIS and AHP techniques. However, this latter is based on expert opinions and thus may be subjected to cognitive limitations with uncertainty and subjectivity (Pourghasemi et al. 2012). Therefore, Monte Carlo simulation-aided analytic hierarchy (MC–AHP) is used to quantify the sensitivity and minimize uncertainty of AHP model. This technique has been already applied to landslide susceptibility mapping (LSM), soil erosion hazard, and pollution control (Cao et al. 2016). To the authors’ knowledge, the evaluation of flood hazard remained limited to sensitivity analysis and this the first time that the MC-AHP technique is applied to flood susceptibility mapping (Fernandez and Lutz 2010). The results obtained by the MC–AHP approach were validated based on their confrontation with remote sensing data for the extreme rainfall event that hit the region in June 2014.

This paper is organized as follows. The next chapter (2) presents the study area and describes the methods: AHP, MC–AHP and the validation exercise. Our results are reported and discussed in chapter 3. A conclusion closes the paper in chapter 4.

Materials and methods

Study area and data analysis

Gabes Watershed, located in Southern Tunisia on the Mediterranean Sea, covers a drainage area of 95 km2. It is characterized by mild slopes and altitudes varying from 0.3 m in the northeast to 235 m in the southwest (Fig. 1). This area is subjected to the influences of both warm and humid air masses coming from the desert and the Mediterranean Sea, respectively. Rainfall is characterized by its shortage, irregularity and erratic distribution, leading to dry and intensive rainy periods, with extreme flood events. Average rainfall is approximately 185 mm/year. Average monthly temperatures vary from 12 °C in January to 29 °C in August, with an annual average of 21 °C. This area is influenced by occasionally strong winds of variable directions. The downstream part of the basin is characterized by its development in terms of housing, industrial, and agricultural activities. Both natural and anthropogenic factors contributed to the damages and impacts of floods.

Fig. 1
figure 1

Location map and digital elevation model of Gabes Catchment

Full size image

Six evaluation factors were considered in order to develop a flood susceptibility map of the area (Fernandez and Lutz 2010; Wang et al. 2011; Elsheikh et al. 2015; Dahri et al. 2016; Danumah et al. 2016). They include litho-facies, slope gradient, land use, elevation, rainfall, and drainage density. The analysis starts with the digital elevation model (DEM), created from both real elevation data and SRTM 30 m data. Real elevation data were derived from the urban management plan. The obtained elevation data were organized as grid data, corresponding to the 1:30,000-scale. The determination of the absolute accuracy of SRTM data is based on the computation of the standard deviation statistic for the elevation differences between the SRTM data and a reference dataset (GPS point measurements) (Gorokhovich and Voustianiouk 2006). The performance of SRTM data involved firstly the conversion of SRTM raster data set into xyz data and then the comparison with observed data in the same location. The results obtained show that a mean difference in elevation between the two data sets is 1.1 m for 110 point measurements. The coefficient of correlation between the two data sets was 0.95. These results indicate a strong positive correlation between the two data sets. Further analysis of the SRTM surface involved a comparison between contours generated from SRTM and the contour maps at a 1:25,000-scale published by the Office of Topography and Cartography (OTC), Tunisia. The results obtained show a reasonably good superposition of the two contours derived from the two disparate sources. Further, the superposition of stream networks generated from the two different origins revealed that the two surfaces are generally close. The obtained DEM was adopted to obtain digital thematic layers. Digital slope and stream network layers were also determined on the basis of DEM data. D8 flow direction is the method used to determine the flow direction for each cell according to the steepest descent to one of its 8 neighboring cells (O’Callaghan and Mark 1984). The flow accumulation raster is used for stream delineation. The threshold value for stream network identification depends mostly on climate, basin physical characteristics as well as the spatial resolution of DEM used (Yamamoto 2009). In this study, 1% of the maximum flow accumulation is considered as the default stream threshold value (Girish et al. 2013), computed to be 1055 for Gabes Watershed. This would imply that all cells with flow accumulation value above the threshold are considered to be part of the stream network. A drainage density map was produced from stream network data using the Kernel Density Algorithm. Drainage density is expressed as the total length of the stream network per unit area. The litho-facies basin map resulted from digitalization of the geological maps (1:50,000) as vector files and then rasterized.

Land use data were obtained by classifying Landsat TM 8 images (Mertikas and Zervakis 2001). The images were classified using a maximum likelihood classification algorithm in a per-pixel classification approach (Sharma et al. 2013). The image classification was run using an appropriate signature file (Fig. 2). Five land use types were identified, including (1) urban area, (2) oasis, (3) forest, (4) low vegetation, and (5) bare soil. This method showed fine details not achieved by photo-interpretation (Mertikas and Zervakis 2001). The overall accuracy and the Kappa coefficient were found to be 93.5% and 0.92, respectively, which clearly shows the accuracy of the classification method (Table 1). The spatial distribution of rainfall data is used to characterize the climatic factor. The thematic layer of rainfall spatial distribution was produced using kriging interpolation (Krige 1951) of data gathered by 6 meteorological stations to create a continuous raster rainfall data within the catchment and its surroundings.

Fig. 2
figure 2

Land use of Gabes Catchment

Full size image
Table 1 Kappa statistic and confusion matrix
Full size table

According to Cova (1999), flood hazard assessment is based on the highest rainfall situation. Indeed, extreme events are generally used in flood mapping. Rainfall factor in this study was derived from the 2014 extreme event. The maximum registered rainfall over a period of almost 6 h varied between 65 and 110 mm all over Gabes Catchment, showing an erratic spatial rainfall distribution. Downtown Gabes, located downstream, is the area that acquired the maximum of precipitation (from 107 to 110 mm). Gumbel distribution was applied to estimate the recurrence of the annual maximum rainfall recorded during 1 day (R × 1d) over the period extending from 1983 to 2014 (Gumbel 1958; Fehri and Yadh 2016). Table 2 shows that the 2014 extreme rainfall is characterized by a 50 year return period. This is in agreement with the findings of Bourges (1974) who reported that the majority of extreme rainfall events that caused severe damage and human losses in Gabes Region have a return period of 50 years. Finally, all data were projected using the Universal Transverse Mercator (UTM) projection system and arranged as raster maps with a resolution of 20 m for further analysis.

Table 2 Statistics of extreme daily rainfall (mm) using Gumbel statistical distribution
Full size table

Standardization

Standardization technique is used to translate various inputs of a decision problem to a common scale, to allow comparison, and to overcome the incommensurability of data (Azizur Rahman et al. 2012; Eastman 1997). The standardization process consists in the transformation of the input raster cell values into the [0, 1] range. In this study, the fuzzy module is applied to translate all factors to a common scale [0, 1]. A fuzzy set is an important part of fuzzy logic model (Bellman and Zadeh 1970; Zadeh 1983). The fuzzy sets design is defined by membership functions and rule bases. The membership is the approach that defines the shapes of the fuzzy sets. Fuzzy logic defines the interval between 0 and 1 in order to indicate the various states of truth (Zadeh 1988). This process facilitates the combination of various raster layers regardless of their original measurement scales (Gorsevski et al. 2012). The fuzzy function is selected in such a way that highly suitable raster cells, in terms of achieving the analysis objective, reach high standardized values and less suitable cells are associated with low values (Azizur Rahman et al. 2012).

Analytical hierarchy process (AHP)

Factor weights are assigned to each criterion based on its importance relative to each of the other factors using Saaty’s Method (Saaty 1980). AHP is applied to help decision makers make pair-wise comparisons between the factors (Table 3), through an importance scale varying from 1 to 9. The method is weighed for each criterion (W i ) by taking the eigenvector corresponding to the largest eigenvalue of the matrix, and then normalizing the sum of the components to unity (Eq. 1).

$$\sum\limits_{i = 1}^{n} {W_{i} } = 1$$
(1)
Table 3 Scales for pair-wise AHP comparisons (Saaty and Vargas 1991)
Full size table

The creation of AHP model is based on the following steps:

  • Step 1: Define the problem and structure its hierarchy which includes a main goal, factors and alternatives.

  • Step 2: Construct the original basic data which is the pair-wise comparison matrix, PCM, of n factors, established on the basis of Saaty’s scaling ratios as defined in Eq. 2. PCM is a matrix with elements a ij . The matrix has also the property of reciprocity (Eq. 3). The construction of pair-wise comparison matrix (PCMs) is based on the surveys and interviews of experts’ opinions (eight experts).

    $${\text{PCM}} = \left[ {a_{ij} } \right] , i , j = 1, 2, 3 \ldots n$$
    (2)
    $$a_{ij} = 1/a_{ji}$$
    (3)

  • Step 3: Use the pair-wise comparison matrix obtained by experts’ opinions for each non-diagonal element to generate a corresponding beta distribution (Fazar 1959). Three inputs data (i.e., minimum, maximum, and most likely) are used to fit a beta-PERT distribution. This approach makes the modeling of expert opinions in decision making processes an ideal technique. The PERT formula estimates the average values (mean) and standard deviations (stdev) of non-diagonal elements as defined in Eqs. 4 and  5. These parameters are necessary to calculate shape factors (α and β) from the corresponding beta-PERT distributions (Eqs. 6 and 7).

    $${\text{mean}} = \frac{{{ \hbox{min} } + 4\,{\text{modal}} + \hbox{max} }}{p}$$
    (4)
    $${\text{stdev}} = \frac{\hbox{max} - \hbox{min} }{p}$$
    (5)
    $$\alpha = \left( {\frac{{{\text{mean}} - { \hbox{min} }}}{{{ \hbox{max} } - { \hbox{min} }}}} \right).\left( {\frac{{({\text{mean}} - { \hbox{min} }).(\hbox{max} - {\text{mean}})}}{{{\text{stdev}}^{2} }} - 1} \right)$$
    (6)
    $$\beta = \left( {\frac{{\hbox{max} - {\text{mean}}}}{{{\text{mean}} - \hbox{min} }}} \right).\alpha$$
    (7)

  • Step 4: Stochastically produce the beta-PERT distributions and generate the elements of matrix as defined in Eq. 8.

    $$a_{ij} = \hbox{min} + {\text{beta}}(\alpha ,\beta ).(\hbox{max} - \hbox{min} )$$
    (8)

  • Step 5: Normalize the matrix obtained in step 4. The normalized elements are named bij which formed a new matrix Mn. The normalization step is defined as shown in Eq. 9

    $$b_{ij} = a_{ij} /\sum\limits_{i = 1}^{n} {a_{ij} }$$
    (9)

    Factor’s weights are calculated as presented in Eq. 10. The results of Factor’s weights are presented in Table 4.

    $$W_{i} = \frac{{\sum\nolimits_{j = 1}^{n} {b_{ij} } }}{{\sum\nolimits_{i = 1}^{n} {\sum\nolimits_{j = 1}^{n} {b_{ij} } } }},\,i, \, j = 1, \, 2, \, 3 \ldots \, n$$
    (10)

    Equation 11 represents the relationship between the maximum eigenvalue (λ max) and its corresponding eigenvector (W) (Chen et al. 2010).

    $$M_{n} W = \lambda_{\hbox{max} } W$$
    (11)

  • Step 6: Compute the consistency ratio (CR), which reflects the degree of consistency of judgements and determines the quality of the comparison (Saaty 1977). The average random consistency index (RI) is provided in Saaty (1980) (Table 4). The consistency index (CI) results from Eq. 12. CR index is calculated using Eq. 13. CR varies from 0 to 1. A CR value exceeding 0.1 is a sign of inconsistency, and a revision of the preference matrix is recommended. A CR of the order of 0.10 or less represents a reasonable level of consistency.

    $${\text{CI}} = \frac{{(\lambda_{\hbox{max} } - n)}}{n - 1}$$
    (12)
    $${\text{CR}} = \frac{\text{CI}}{\text{RI}}$$
    (13)
Table 4 Random index RI for different n values
Full size table

Monte Carlo AHP (MC–AHP)

To improve the performance of the proposed model, an innovative Monte Carlo simulation-aided analytic hierarchy process (MC–AHP) is used. This is an important approach to quantify the sensitivity and the uncertainty when weights are assigned based on subjective expert opinion, or personal preference (Gomez-Delgado and Tarantola 2006). This approach is related to probability distributions of all non-diagonal elements of the pair-wise comparison matrix (PCM). In this approach, the same steps and constraints of AHP are used with introduction of beta-random (betarnd function). The betarnd is a function in MATLAB that generates beta distributed numbers within the interval of [0, 1] using to shape factors α and β. The Eq. 8 used in AHP method is replaced by Eq. 14.

$$a_{ij} = \hbox{min} + {\text{betarnd}}(\alpha ,\beta ).(\hbox{max} - \hbox{min} )$$
(14)

The repetition of the same steps (step (4) to step (6)) of AHP, with Eq. 14 instead of Eq. 8, is considered with a number of replications (raying from 100 to 10.000). The relative weights for factors are obtained (Table 5) and plotted as probability density functions rather than as point values (Jing et al. 2013).

Table 5 Factors weights from AHP and MC–AHP approaches
Full size table

Weighted linear combination (WLC)

Weight values derived from AHP and MC–AHP were assigned to layer factors. Weighted linear combination (WLC) is considered as a decision rule to derive composite maps using GIS platform (Malczewski 1999, 2006). WLC is a concept which aggregates maps by applying a standardized score to each class of factors and a weight to the factors themselves. Factor weights are used to determine the hazard condition (R) as defined by Eq. 15.

$${\text{R}} = \sum\nolimits_{{}} {W_{i} \times X_{i}}$$
(15)

W i is the mean weight values of the factor i, and X i is the potential rating of the factor i.

Validation

Satellite data were analyzed to identify flood-prone areas (Zhang et al. 2002). A Landsat TM 8 image, captured in June 3, 2014, was used to validate the AHP flood hazard delineation for selected sample points. All bands images were orthorectified, georeferenced to the UTM projection 32 and converted to digital numbers (Knight and Kvaran 2004). Atmospheric correction was also performed to obtain the surface reflectance of all raw images (Devadas et al. 2012). The Landsat image was corrected using ENVI software (Knight and Rvaran 2014; Williams 2008; ITT Visual Information Solutions 2009). Inundated areas were classified based on thresholds of NDWI (Normalized Difference Water Index) and NDVI (Normalized Difference Vegetation Index), considered to be sufficient to accurately map and monitor the presence of water in natural and artificial ponds. This methodology was tested and verified in arid and semiarid regions (Gond et al. 2004).

Results and discussion

Both GIS and AHP techniques are used to identify flood hazard areas. The AHP method is generally used to describe hazard evaluation (Erkut and Moran (1991). However, for large-scale applications, it is important to incorporate stochastic simulations to better reflect real conditions. This takes into account uncertainty in a real application, which shall increase the confidence of a decision maker in the final results. Monte Carlo simulation, known to be particularly useful in stochastic modeling, was considered. MC–AHP method is based on the fact that input variables are presented as statistical distributions that are derived by best fitting the data collected. Numerous iterations were performed in order to quantify results and identify the optimum solution. In the following, results of both AHP and MC–AHP are compared based on an error quantification analysis. MC–AHP results were validated using the methodology of Gond et al. (2004). Obtained flooded areas were also verified by survey data collected after a particular flood event.

The obtained flood hazard map of the study area (Fig. 3) was subdivided into five classes, ranging from ‘very low’ to ‘very high’ hazard. The areas labeled as ‘very high hazard’ are strongly influenced by the impervious areas in the urban sector. Impermeable litho-facies and high drainage density tend to increase the flood hazard. The map shows that the northeastern part of Gabes Catchment has the highest flood hazard, with a flooded area of 11.9% of the total area (Table 6). 20.5% of the total area is shown to be under a high level of flood hazard. The areas with very low to low flood hazard are generally located in the southwestern part of the study basin and represent 21.7% of the basin. Table 5 shows that 45.9% of the total area is found to be under a moderate flood hazard. The flood hazard influence is increased mainly downstream of the catchment, characterized by high urbanization rates. Moreover, anarchic urbanization in flood-prone areas and the lack or the limited capacity of the storm water collection system are considered as important factors contributing to flood risk.

Fig. 3
figure 3

Final flood susceptibility map derived by AHP approach

Full size image
Table 6 Subdivision of the different flood risk classes within Gabes Catchment
Full size table

The region of ‘very high’ flood hazard is also located along the coast, which represents the zone that receives the highest amount of rainfall and the most important extreme events. This was demonstrated by Ellouze et al. (2009) who indicated that rainfall increased from the west (continent) to the east (coast line) in southeastern Tunisia regions. Indeed, a high precipitation over a short period of time is the most important factor responsible for triggering floods (Wang et al. 2011). Finally, the urban area located downstream of Gabes Catchment is characterized by mild slopes, inducing thereby water stagnation.

The analytical hierarchy process (AHP), as a method of factors weighting, suffers from sensitivity and uncertainty, as reported by (Banuelas and Antony 2004). The Monte Carlo simulation-aided analytical hierarchy process (MC–AHP) approach is expected to quantify the sensitivity and uncertainty of AHP model. MC–AHP integrated the beta-PERT distribution, which was shown to closely resemble the realistic probability distribution with little demand of data. A PERT normal distribution was considered because the triangular distribution is limited in its ability to model real-world estimates (Jing et al. 2013). The proposed approach addresses the uncertainty resulting from insufficient information and subjectivity judgements in group decision making problems. For all MC–AHP simulations, the consistency ratio was shown to be less than 0.1 (Fig. 4). The results obtained after 10.000 replications show that MC–AHP presented similar results if sorted into five classes (Fig. 5).

Fig. 4
figure 4

Consistency ratio of MC–AHP (10.000 Monte Carlo replications)

Full size image
Fig. 5
figure 5

Flood susceptibility map derived by the MC–AHP method

Full size image

The probability density distributions of factor weights obtained by the kernel-smoothing method are plotted in Fig. 6. Compared to other probability density distributions, land use probability function has the most important weights (Fig. 6). Regardless of experts’ judgments, the land use factor increased flood hazard especially downstream. The high flood hazard areas are also attributed to mild topography and natural depressions or low points (called Sebkhas) inducing water stagnation. Figure 6 also shows similar flood hazard contributions of both ‘rainfall’ and ‘litho-facies’ factors. Finally, elevation contributed with the minimum weight to flood hazard. The results obtained by AHP and MC–AHP were compared. The beta-PERT distribution adopted in the proposed approach significantly reduced standard deviations of the overall scores. Other indices such as confidence interval with a confidence level of 95% were derived for both AHP and MC–AHP as shown in Table 7. In the AHP approach, the overall scores had large variation intervals (Fig. 7). Unlike the traditional AHP method, MC–AHP approach tends to concentrate the weight around the mean with a small confidence interval (Fig. 7). The MC–AHP model mainly reduced the standard deviations of weight factors. This method minimizes the error produced by subjectivity judgments of the AHP model.

Fig. 6
figure 6

Probability density estimates of factor weights

Full size image
Table 7 Statistics for overall scores from the AHP and MC–AHP techniques
Full size table
Fig. 7
figure 7

Confidence interval of 95% plot of the overall scores for the different factors

Full size image

The results obtained by the MC–AHP approach were also compared to the Landsat analysis of June 2014 flood event (Fig. 8). The examined flooded areas in 2014 show that 226 observed flood samples are used to validate the accuracy of the results obtained by MC–AHP approach. The total observed inundation area is estimated to be 62 ha (0.65% of total basin area). Two hundred and seven from a total of 226 observed flooding zones are located in a high to very high susceptibility zone. Flood samples are mainly concentrated downstream of the catchment. 95.5% of observed flooding areas are especially located in Gabes City surrounding the streams and anarchic urbanization (Fig. 8). The lack or the limited capacity of the storm water collection system also contributed to water stagnation. Indeed, Gabes City lacks an appropriate drainage system in order to afford a significant level of conveyance, avoiding thereby the multiplication of flooding zones. Storm water is simply conveyed by gravity via roads and streets to the Mediterranean Sea.

Fig. 8
figure 8

Land-sat analysis of June 2014 flood event in Gabes Catchment

Full size image

Conclusion

Analytical hierarchy processes (AHP), with and without Monte Carlo simulation, were applied, and flood susceptibility maps of Gabes Catchment were developed. The analysis considered six variables, including elevation, slope, land use, drainage density, litho-facies, and rainfall. This latter was derived from the spatial distribution of the June 2014 extreme rainfall event, characterized by a return period of 50 years since the concept of hazard is especially associated with extreme events.

The results obtained by both AHP and MC–AHP approaches are similar for five hazard classes. The beta-PERT distribution adopted in the MC–AHP approach significantly reduced standard deviations of overall scores. MC–AHP modified the original factor weights and produced limited confidence intervals. The analysis showed that reducing the error of the input weights results in improving the flood map accuracy. The obtained MC–AHP results show that 33.5% of the basin area is characterized by a high to a very high flooding hazard. Land use was shown to be the most important factor contributing to flood hazard.

To validate MC–AHP model, Landsat analysis of the 2014 flood event was used. The examined flooded areas proved that among 226 observed flood samples, 209 zones are characterized by a high to a very high susceptibility hazard. 96.3% of observed flooding areas are located in Gabes City surrounding the streams and anarchic urbanization.