The municipality of Kaffrine is a locality located in the heart of Senegal, on the Dakar-Tambacounda axis, 258 km from Dakar. It is part of the department of the same name and is surrounded by the rural commune of Kahi ( Figure 1 ). It currently consists of nine (9) official neighboroods (Escale Est, Escale Ouest, Diamaguéne TP, Diamaguéne centre, Diamaguéne Ndiobéne, Pèye, Mbamba, Kaffrine II Nord and Kaffrine II Sud). Its surface area is 440 ha.

In 2023, the population of the municipality of Kaffrine is 57,307 according to the general census of the population and housing of Senegal of 2023 ( National Agency of Statistics and Demography (NASD), 2024 ). The economy is dominated by agriculture, livestock and forestry. It rains on average 800 mm per year; however, there is interannual variability. Temperatures are generally high, with significant variations. They fluctuate between 26 and 39˚C with an average of 29˚C.

The city is located in a vast lowland with flat relief. The contour lines determine a relief of slight depressions and mounds. Its relief constitutes to a certain extent, a difficulty for the installation of a sanitation network.

The main types of soils found in the municipality of Kaffrine are: sandy soils (Deck), clayey soils (Dior), silico-clayey soils, and lateritic soils which shelter quarries in certain places. These soils do not have the same sensitivity to the risk of flooding.

The hydrography is essentially made up of the Saloum Fossil Valley. As for groundwater, the city is irrigated by the Maestrichtian aquifer which is nearly three hundred (300) meters deep.

Hazard is a threatening event or a probability of occurrence in a region and during a given period of a phenomenon that could cause damage. It is a phenomenon resulting from factors or processes that escape, at least in part, human control ( Géoconfluences, 2019 ). It is a natural (atmospheric, hydrological, geological or geomorphological processes), technological or geopolitical phenomenon which can occur in a given space.

Hazard is a threatening phenomenon of natural and/or anthropogenic origin, likely to affect a given space, in particular by the nature and value of the exposed elements that this space supports (people, goods, activities, etc.). It is characterized by its nature, its identity, its probability of occurrence and its frequency when it can be estimated ( Gbeassor et al., 2006 ). According to the United Nations Disasters Relief Co-Ordinator ( UNDRO, 1979 ), hazard can be defined as a threatening event or a probability of occurrence in a region and during a given period of a phenomenon that could cause damage. Hazard is linked to the notion of chance. In the field of risk study, hazard is defined as the probability of occurrence of a phenomenon. It is often a physical, lithospheric or climatic process.

The notion of risk is very close to that of uncertainty. The notion of risk is therefore relative and depends on the way in which societies conceive their fragility in the face of perils ( Hangnon, 2009 ), refers to the possibility of a loss ( Fournier d’Albe, 1979 ). It follows from this that risk is the probability that a potentially dangerous phenomenon will occur, and which by its characteristics can cause damage and harm in a given space, at a given moment. The concept of risk includes two explanatory factors: on the one hand, the frequency and amplitude of events that can cause damage and on the other hand, the potential for damage or vulnerability that depends on the type, value and exposure of the elements affected by these hazards. It thus joins the usual definition proposed by MATE (2001) on the concept of natural risk which is the product of hazard and vulnerability: Risk = Hazard × Vulnerability. Vulnerability is defined as a degree of damage ( Veyret & Reghezza, 2005 ). In this study, we consider risk as the product of hazard and vulnerability.

The flood risk factors that are the subject of this study are as follows: slope, rainfall, soil type, altitude, building density, population density and level of coverage by rainwater drainage network.

Rainfall is a measure of the intensity of rain that brings water that can cause flooding. Given the unavailability of rainfall data by district and the small size of the study area (very low internal variability), we used the amounts of rainfall that fell in 2019, 2020 and 2021. In this way, we were able to measure the effect of the rain. The slope is one of the factors directly linked to the cause of flooding, because it facilitates the runoff of rainwater towards the lowlands. The type of soil plays a major role in the occurrence of floods ( Guelbeogo & Ouedraogo, 2022 ). Altitude is an important factor, because the speed of water circulation also depends on altitude ( Guelbeogo & Ouedraogo, 2022 ). Building density is a factor in flooding. Indeed, studies have shown that the characteristics of residential buildings can influence the occurrence or intensity of flooding ( Wilhelmi & Morss, 2013 ; Tanguy, 2012 ). Population density plays a fairly significant role in the occurrence of floods, because it leads to a reduction in free space, soil compaction, which can slow down the infiltration of rainwater, prevent or slow down the runoff of the latter. The level of coverage of the rainwater drainage network is a factor to be taken into account. Indeed, the failure and lack of coverage of this essential infrastructure can therefore be a factor aggravating the vulnerability of the population ( Pageon, 2008 ; Nicholls & Small, 2002 ).

Three types of data were collected. These are data on the natural physical environment (topography, rainfall, soils), urban data (population and level of coverage in the sanitation network), data on the relative importance of risk factors.

To collect the first data type, we used satellite images, Master Plan for the evacuation of rainwater for the municipality of Kaffrine and the data provided by the National Agency for Civil Aviation and Meteorology (NACAM). As for the second data type, they were collected from the NASD, the Regional Development Agency (RDA) and the National Sanitation Office (NSO). To collect the third data type, we conducted a survey of experts and stakeholders in the fight against flooding. A sample of 30 local authorities and experts was chosen to comment on the relative importance of flood risk factors. The survey consists of comparing two by two the different risk factors (hazard and vulnerability), by filling out a hazard and vulnerability matrix, presented in the form of a table. Each survey assigns a weight from one (1) to nine (9) according to the scale proposed by Saaty (1991) . The survey was conducted using a tablet equipped with the Kobocollect application. This application is reliable and easy to use.

To process and analyze physical environment data (altitude, slope), we followed the following steps. First, we made a digital terrain model (DTM) from 30 m resolution SRTM rasters provided by NASA. This model was validated by comparing the ground data with the DTM data. To do this, we used a sample of 43 ground control points spread out across the municipality and on which GPS surveys were made to measure altitude values. A correlation coefficient between the altitude values taken in the field and those obtained from the rasters was calculated. The result obtained is 0.773, closer to 1; which allowed us to validate the DTM.

The next step was to project a regular grid of 30 m × 30 m cells onto this raster to extract topographic data such as slope and altitude for each cell of the grid, by using zonal statistics method. The choice of cell size is based on the resolution of the raster used (30 m).

The topographic surveys allowed us to generate a linearization of the two existing networks using the Covadis software. The use of Google Earth Pro was necessary to obtain the topographic profiles of the area. The soil type of each neighborood was classified into categories according to its level of sensitivity to flooding using the Qgis software. Finally, for rainfall, the data was segmented into three years (2019, 2020 and 2021).

MCA is a method that was invented by the mathematician Thomas Saaty ( Saaty, 2000 ; Saaty, 2008 ). It is intended to help the decision-maker refine his decision-making process by examining the coherence and logic of his preferences. It is a method that can be used in the quantification of qualitative criteria, through weighting. It has already been applied in different fields with success ( Ramos et al., 2014 ; Corvin, et al., 2021 ; Kumar et al., 2010 ; Le Gallic et al., 2006 ). This method is able to identify and take into account the inconsistencies of decision-makers.

MCA is a rigorous method that includes a series of important steps: structuring the hierarchy, establishing priorities, and verifying the logical consistency of the analysis ( Saaty, 2008 ). It allows for measurement of the criteria of a given situation, based on the derivation of relative priorities from pairwise comparisons sharing a common attribute ( Saaty, 1994 ; Kendrick, 2007 ). Table 1 is used to calculate the weight of each flood criterion (factor).

In the hierarchical analysis process, the relative importance of component or criterion i with respect to component j (a_ij) is determined using the Saaty scale ( Saaty, 1991 ) and is assigned to the (i, j) th position of the pairwise comparison matrix. Automatically, the inverse of the number associated with the (j, i) th position can be calculated according to the following rule ( Chang et al., 2007 ):

$a_{ij}>0,a_{ji}=\frac{1}{aij},a_{ii}=1\forall i$ (1)

To calculate the weights of each factor, we first calculated the average of the fourteen (14) matrices called the raw value matrix proposed by the people surveyed to obtain a single matrix. This operation is valid for hazards and vulnerabilities. We then added the values of each cell for each column to obtain the total value of each column. Finally, to obtain the normalized matrix, we divided the value of each cell by the total of each column.

Once the raw matrix is filled, we add up each row. To obtain the weight of each factor, we divide each cell in the row sum column by the number of factors (hazard or vulnerability). The eigenvector (weight) indicates the order of priority or hierarchy of the characteristics studied. This result is important for the evaluation of the probability, since it will be used to indicate the relative importance of each operating criterion.

The eigenvector of the matrix can be found by the following formula:

$Wi={\left(\underset{j-1}{\overset{n}{{\displaystyle \prod }}}{a}_{ij}\right)}^{1/n}$ (2)

Furthermore, it must be normalized so that the sum of its elements is equal to unity. To do this, simply calculate the proportion of each element to the sum using the following equation:

$T=\left|W1/{\displaystyle \sum WiW2}/{\displaystyle \sum Wi\cdots Wn}/{\displaystyle \sum Wi}\right|$ (3)

Let T be the normalized eigenvector used to quantify and evaluate the importance of each criterion. The latter will be multiplied by the vector of weights of each factor to give the ratio of the weighted sums of the factors, the average of which gives the maximum eigenvalue (4).

$\lambda \max =T\cdot w$ (4)

where w is calculated by adding the columns of the comparison matrix.

In order to check the consistency of the response given by the respondents, Saaty (1977) proposes to calculate the CI (5). The next step is to calculate the consistency index (CI):

$CI=\frac{\left({\lambda }_{\max }-n\right)}{\left(n-1\right)}$ (5)

where, ${\lambda }_{\max }$: maximum eigenvalue of each factor in the matrix table and n the size of the matrix. The eigenvalue is the measure that will allow to evaluate the consistency or the quality of the solution obtained. To calculate the eigenvalue, we multiply each cell of the raw matrix by each value of the vector of the sum of the rows; which allows to obtain the vector of the sum of weighted values.

The following table shows the different values of the RI according to the number of factors.

The next step is to calculate the consistency ratio (CR) between the consistency index and the random consistency index using the following equation:

$CR=CI/RI$ (6)

where, CR is the ratio between CI and the random consistency index (RI). The RI index presented in ( Table 2 ), comes from a sample of 500 positive reciprocal matrices managed randomly, whose size reaches 11 by 11.

To validate the survey matrix, the CR is calculated (6). According to the work of Yurdakul and Tansel (2004) , the CR value must be less than 0.1 to conclude that the pairwise comparison judgments are consistent. On the other hand, if the CR value is greater than 0.1 the coefficients of the matrix are inconsistent and cannot be used for the analysis ( Wong & Li, 2007 ).

The data comes from different sources, spatial scales and different formats. It was therefore necessary to integrate them into a geographic information system. To do this, we first created a grid with cells measuring 30 m on each side to respect the 30 m resolution of the raster. This grid covers the entire extent of the municipality of Kaffrine.

We transferred the population data of the neighboroods to the cells of the generated grid, by multiplying the population density of the neighborood by the area of each cell of this neighborood. The values of the indicators of the different hazard factors (slope, soil, and altitude) were also transferred to the cells of the grid by superimposing the raster file (30 m) already corrected with the grid created and using the area statistics method. For the rainfall data, we discretized the statistical series into 3 classes representing each of the 3 years (2019, 2020 and 2021), then assigned the scores to each year according to the recorded rainfall intensity.

For the rainwater drainage network service level data, we first determined the centroids of each cell, then calculated the distance between these centroids and the nearest network line, using the distance matrix tool. Finally, the inverse of this distance was used to measure the level of rainwater drainage network coverage of each grid cell.

After entering the values of all the hazard and vulnerability factors for each grid cell, we weighted the values of each factor using the weights obtained from the multi-criteria analysis.

To map the indicators obtained, we created percentiles. The creation of percentiles consists of arranging the values of the hazard and vulnerability variables into five levels framed by the minimum and maximum of each variable and for each of the three years.

We used the Shuttle Radar Topography Mission (SRTM) data (30 m spatial resolution raster) provided by NASA to generate the Digital Terrain Model (DTM) (altitude map). To verify the validity of the results obtained, we took 43 ground control points, where altitudes were measured by a differential GPS. The altitudes obtained from the SRTM data were compared with those from the measurements taken at the ground control points, by calculating the Pearson correlation coefficient. The result obtained is 0.77, closer to 1; hence the validation of the DTM.

To validate the flood map, we took altitude measurements of 30 points well distributed in the study area. These points were projected onto the flood risk maps. The result shows that low-lying points are mostly located in high flood risk areas. This allowed us to validate our flood risk maps.

The results of the survey on the perception of experts and stakeholders on the relative importance of flood risk factors are contained in Table 3 and Table 4 .

Regarding the hazard factors, the results indicate a greater importance given to rainfall and slope/altitude ( Table 3 ). The water retention capacity of soils also has a significant weight, but lower than those of the previous factors. The factor with the lowest weight is the level of soil impermeability. Table 4 contains the results of the measurement of vulnerability factors. According to these results, the density of buildings has the greatest weight, followed by population density. The level of coverage by rainwater drainage network comes last.

To validate the survey matrix on the perception of the relative importance of the weights of flood risk factors, we calculated the consistency index (CI) and the consistency ratio (CR). The results are contained in Table 5 . They show that the R has a value lower than 0.1, which allows us to say that our consistency matrix is valid. Indeed, according to the Saaty method ( Saaty, 1977 ), when the value of R is lower than 0.1, we can conclude that the pairwise comparison judgments are consistent.

Knowing that we have only one rainfall station in the municipality of Kaffrine, the rainfall data for a single year do not experience any intra-municipal variation. Also, to have a variation in these data, we used the rainfall data for three years (2019, 2020, 2021). In addition, it should be noted that rainfall often changes from one year to the next. This allowed us to highlight the effect of rain on the occurrence of floods.

The additive combination of the hazard factor layers gives the maps below. These maps were developed for the years: 2019, 2020, 2021 ( Figure 2 ) to be able to follow the dynamics of the phenomenon over time. We can see that the hazard is very high over the entire northwest part, part of the south and the center of the municipality. The North-East, East, South-East and South-West parts are weakly exposed to flood hazard.

The results obtained also show that the hazard factor is almost the same in space for the years 2019, 2020 and 2022. The temporal dynamics do not seem to have an influence on the probability of flooding in the city of Kaffrine.

According to the results of our field investigations, the vulnerability factors did not change significantly during the three reference years (2019, 2020 and 2022). Therefore, we considered it more relevant to produce a single vulnerability map for these three years.

Figure 3 highlights the spatial distribution of the level of vulnerability to flooding in the city of Kaffrine. It can be seen, in general, that the level of vulnerability is higher in the north, while the south records low levels of vulnerability. The southern part of the city records the lowest risk levels. In addition, there are pockets of very low vulnerability in the center and enclaves of high vulnerability in the southwest.

Table 6 contains the percentage of the population exposed by risk level and by year. In 2019, 13.64% of the population was exposed to a very high-risk level and 15.28% to a high level of flooding. Almost the same proportions are observed in 2020 and 2021. A significant share of the population (nearly 29%) is exposed to high and high-risk levels.

As a reminder, this risk map results from the cross-referencing of the hazard map with the vulnerability map. By applying the percentile method, we distinguished 5 levels of flood risk: very high, high, moderate, low and very low. Figure 4 shows the spatial distribution of the flood risk level in the municipality of Kaffrine.

In general, we note that the flood risk level is higher in the northwest, particularly in the neighboroods of Mbamba, Peye, part of Diamagueune Centre. We also note pockets of very high flood risk level in the southwest, particularly in the neighboroods of Diamagueune centre and Escale ouest.

The areas that record the high flood risk level are found further northeast and north-central, particularly in the neighboroods of Bamba, Peye and in the northeast, particularly in the neighboroods of Kaffrine nord and Kaffrine sud. Moderate risk is recorded further north-east and in the centre, especially in the districts neighboroods of Kaffrine Nord and Kaffrine Sud, Escale Est, Escale Ouest.

Low risk areas mainly concern the neighboroods of Diamagueune Centre, Kaffrine Nord and Kaffrine, Escale Est, Escale Ouest. As for very low risk areas, they are found further south, in the center and center-west; they concern in particular the neighboroods of Diamagueune Ndiobéne, Diamagueune TP, Kaffrine Sud, Escale Est, Escale Ouest.