An evaluation of groundwater vulnerability assessment methods in a rapidly urbanizing city: evidence from Dakar, Senegal

In rapidly growing cities in the tropics, unregulated urban development presents a major risk to groundwater quality. Here, we assess the vulnerability of an unconfined aquifer of Quaternary sands in the Thiaroye area of Dakar (Senegal) to contamination using four GIS-based indices (DRASTIC, DRASTIC_N, SINTACS, SI). Our correlation of assessed vulnerability to observed impact is semi-quantitative, relating observed groundwater quality, based on nitrate concentrations and tryptophan-like fluorescence to vulnerability degrees (i.e. coincidence rates). We show that considerably more of the Thiaroye area has a “very high vulnerability” according to SI (36%) relative to DRASTIC (5%) and SINTACS (9%); “high vulnerability” is estimated using DRASTIC_N (100%), DRASTIC (66%) and SINTACS (69%). Single-parameter sensitivity tests show that groundwater depth, soil, topography, land use and redox parameters strongly influence assessments of groundwater vulnerability. Correlation with observed nitrate concentrations reveals aquifer vulnerability is better represented by SI (coincidence rates of 56%) relative to DRASTIC_N (43%), SINTACS (38%) and DRASTIC (34%). The underestimation of groundwater vulnerability in Dakar using DRASTIC, DRASTIC_N and SINTACS is attributed to their reliance on an assumed capacity of the unsaturated zone to attenuate surface or near-surface contaminant loading, which in the low-income (Thiaroye) area of Dakar is thin and affords limited protection. The inclusion of a land-use parameter in SI improves the characterization of groundwater vulnerability in this low-income, rapidly urbanizing area of Dakar.


Introduction
Groundwater resources in drylands are often the only perennial source of freshwater to meet domestic, agricultural and industrial demands. Tropical drylands are often characterized by limited, seasonal surface water availability that is a function of their climate. The greater vulnerability of surface water to pollution is also an obstacle to its sustainable use. Use of shallow groundwater for drinking and other domestic purposes is an especially common feature of many urban low-income communities (Howard et al. 2003;Gaye and Tindimugaya 2019). Urban groundwater use includes not only utility withdrawals but also private self-supply for residential, commercial, industrial and agricultural uses (Foster et al. 2018). Urban self-supply of water and direct use of local wells by low-income households is often of vital importance to a large proportion of a city's inhabitants (Grönwall et al. 2010).
Vulnerability assessments are commonly conducted in areas where water resources are stressed due to This article is part of a Topical Collection in Environmental Earth Sciences on Groundwater quality and contamination and the application of GIS, guest edited by Narsimha Adimalla and Hui Qian.
anthropogenic activities (Singh et al. 2015). Groundwater vulnerability assessment studies have, to date, shown urban areas to be increasingly prone to groundwater contamination (NRC 1993;Alam et al. 2012;Singh et al. 2015;Kumar et al. 2015). Three primary methods exist by which groundwater vulnerability is commonly assessed: (1) a subjective overlay and index method based on the rating of individual hydrogeological factors (Kumar et al. 2016); (2) processbased mathematical models that are data intensive (Neshat et al. 2014); and (3) statistical models that describe the contamination potential for a specified geographical region using the available data in the regions of interest (NRC 1993;Kumar et al. 2016).
The subjective overlay and index method is the simplest and most widely used method in sub-Saharan Africa where the availability of hydrogeological data is commonly limited (Oke and Fourie 2017). Among the different overlay and index methods, DRASTIC is one of the most popular due to its performance and ease of use (Barbulescu 2020). In the last two decades, DRAS-TIC has been employed in several studies to assess groundwater vulnerability in Asia (Rahman 2008;Babiker et al. 2005;Yin et al. 2011;Muhammad et al. 2014;Ghosh et al. 2015;Shrestha et al. 2017), Africa (Ojuri and Bankole 2013;Jourda et al. 2007;Saidi et al. 2009;Hamza et al. 2008Hamza et al. , 2010Neh et al. 2015;Oke 2020) and the Americas (Klug 2009;Agyemang and Beauty 2017). DRASTIC does not, however, explicitly consider spatial variability in anthropogenic factors that may be of critical importance in rapidly urbanizing, low-income areas (Singh et al. 2015). Modifications to DRASTIC have been developed to explicitly represent groundwater vulnerability to anthropogenic pollution from surface/sub-surface sources. SINTACS (Civita et De Maio 1997), DRASTIC_N (Voutchkova et al. 2020) and the Susceptibility Index or SI (Riberio et al. 2000) are examples of these, which have demonstrated efficacy in assessing groundwater vulnerability in many environments (Batista 2004;Frances et al. 2002;Stigter et al. 2006;Hamza et al. 2008;Ake et al. 2011;Afonso et al. 2016;Ribeiro et al. 2017;Batchi et al. 2017;Si et al. 2017;Marjuanto et al. 2019;Voutchkova et al. 2020).
In Dakar (Senegal, Fig. 1), the piped water supply is provided by: (1) surface water via a pipeline from Lake Guiers 250 km away representing 50% of the drinking water supply; and (2) groundwater from local aquifers that include deep Maastrichtian sands, Paleocene limestones and infrabasaltic/ northern coastal Quaternary sands. Despite this conjunctive use of groundwater surface water, the city faces chronic shortages in drinking water. In densely populated, lowincome areas of Thiaroye in Dakar for example, self-supply of water via dug wells as well as hand pumps is common to adapt to limitations in access to the piped water supply. Use of urban groundwater is, nonetheless, compromised by declining groundwater quality, primarily as a consequence of effluent from on-site sanitation systems (Cissé-Faye et al. 2019;Diaw et al. 2020). Due to the importance of urban groundwater in the provision of safe water in pursuit of UN Sustainable Development Goal 6 (access to safe water for all by 2030), effective tools to identify areas at risk of groundwater contamination, especially when monitoring is limited, are vital.
Extensive nitrate pollution in the unconfined aquifer of Quaternary sands in Dakar (Thiaroye) has been revealed by previous studies (e.g. Tandia 2000;Cissé-Faye 2001Diédhiou et al. 2012;Diaw et al. 2020). The high concentrations of nitrate have prompted earlier assessments of the vulnerability of shallow groundwater to contamination from its urban environment by Cissé-Faye (2001) and Madioune et al. (2005 and using DRASTIC (Aller et al. 1987) and GOD (Foster 1987) assessment methods. These studies provided an overview of the areas of vulnerability of the Thiaroye groundwater to nitrate pollution at the regional scale. These methods did not, however, consider directly the impact of human activities on groundwater quality. Further, assessed vulnerabilities were not explicitly reconciled to observed groundwater contamination. This study evaluates the ability of DRASTIC, DRASTIC_N, SINTACS and SI models to assess the vulnerability of an unconfined aquifer comprising well-sorted Quaternary sands to contamination in a rapidly urbanizing environment. Here, we: (1) assess urban groundwater vulnerability using DRAS-TIC, DRASTIC_N, SINTACS and SI; and (2) evaluate the respective performance of these models to predict urban groundwater contamination. The ultimate goal is to inform pragmatic assessments of groundwater vulnerability in urban areas of sub-Saharan Africa where there is commonly strong dependence on the use of on-site sanitation.

Study area of Dakar (Senegal)
Dakar is a typical fast-growing metropolitan city in Sub-Saharan Africa. Its population increased five-fold from ~ 583,000 inhabitants in 1971 to 3.1 million inhabitants in 2013. The population in the suburban areas (e.g. Thiaroye, Pikine, Guediawaye, Parcelles assainies, Keur Massar and Rufisque districts) is estimated at 1,500,000 with a mean population density of 9335 inhabitants per km 2 . The study area is characterized by great speed of urbanization that started after the Sahelian drought that began in the 70 s (Fig. 2a) and led to rapid expansion of peri-urban and unplanned settlements (Sow 2009).
The study focuses on low-income communities served by on-site sanitation (septic tanks) within the Thiaroye watershed in the Dakar region. It is located at the collar of the peninsula with an area of 60.4 km 2 (Fig. 2b). The relief is materialized by depressed dunes where three hydrogeological units are identified: the lakes area to the north, Niayes area or closed dune depressions to the west and dune area which covers most of the basin. The climate is semiarid with mean annual precipitation ranging between 450 and 500 mm, which primarily falls during the rainy season between July and October; annual temperatures vary between 21 and 29 °C .
The study area is underlain by a Quaternary sand aquifer system which extends along the Senegal northern coastal zone from Dakar. The aquifer system itself is underlain by Eocene marl to clay formations which outcrop in the south.
Quaternary deposits constitute the aquifer reservoir and comprise mainly unconsolidated clayey sands, coarse sands, eolian sands which form the Ogolian dunes in the coastal zone. The aquifer is unconfined throughout the study area. Recharge is thought to occur primarily by the direct infiltration of rainfall (i.e. diffuse recharge) but also via contributions from other sources such as wastewater and irrigation waters that contaminate the groundwater Diouf et al. 2012;Cissé-Faye et al. 2019).

Materials and methods
To assess groundwater vulnerability to pollution at a watershed scale in Dakar, four indices (DRASTIC, DRASTIC_N, SINTACS and SI) were applied. These models estimate the vulnerability of groundwater to contamination and are expressed in the form of vulnerability maps (Kumar et al. 2016).

DRASTIC method
DRASTIC is an empirical model that estimates groundwater vulnerability in aquifer systems based on in-situ hydrogeological information (Aller et al. 1987). It is widely used to assess the intrinsic vulnerability of groundwater to a wide range of potential pollutants (Al-Abadi et al. 2017). DRASTIC assesses groundwater vulnerability to contaminants generated by human activities. It considers geological, hydrological and hydrogeological characteristics but not those of human activities (Hamza et al. 2010). The acronym DRASTIC designates the parameters included in the method: groundwater depth (D), net recharge (R), aquifer lithology (A), soil type (S), topography (T), impact of the unsaturated zone (I) and hydraulic conductivity (C). The seven parameters pertain to hydrogeological factors that influence contaminant transport and attenuation processes. The degree of influence of each parameter is quantified by a numerical value known as a 'parametric weight' of between 1 and 5; Each parameter is listed in classes associated with ranges from 1 to 10 (Table 1). Smaller scores denote a lower vulnerability to contamination. A numerical value referred to as the DRASTIC vulnerability index (DI) is computed by aggregating the products of the ranges by the weights of the corresponding parameters in Eq. 1: where DI is the vulnerability index, D, R, A, S, T, I and C comprise the seven parameters of the DRASTIC method and the subscript "w" refers to the weight of the parameter (1) DI = D w × D r + R w × R r + A w × A r + S w × S r + T w × T r + I w × I r + C w × C r and "r" to its associated rating. There are seven classes, each corresponding to different degrees of vulnerability (Table 2).

SINTACS method
SINTACS (Civita et De Maio 1997) is the Italian version of the DRASTIC model. It takes into consideration the same parameters with different weights and dimensions. The acronym SINTACS stands for the initials of the following seven factors: Groundwater depth (S = soggiacenza), Net recharge (I = infiltrazione), Impact of the unsaturated zone (N = effetto di autoepurazione del non-saturo), Type of soil (T = tipologia della copertura), Lithology of the aquifer (A = caratteristiche idrogeologiche dell'acquifero), Hydraulic conductivity (C = conducibilità dell'acquifero) and Topography (S = l'acclività della superficie topografica). Unlike  DRASTIC, SINTACS allows for weighting factors to vary spatially. Five possible scenarios are distinguished: • "Normal impact" scenario for unconsolidated sediment aquifers where water table is not deep (< 10 m below ground). The areas related to this scenario correspond to stable regions, in terms of land use, with or without cultivated land, low use of pesticides, fertilizers and irrigation and widely dispersed urban areas; • "Severe Impact" scenario which corresponds to the same aquifer type subject to intensive land use where there is considerable use of pesticides, fertilizers and irrigation or where there exist, dense industrial and urban settlements, liquid and solid waste deposits; • "Significant Drainage from a Surface System" scenario for areas with high infiltration to the aquifer from a surface water system; • "Very karstified terrain" scenario; • "Fractured terrain" scenario.
The dominant scenario of the hydrogeological conditions in the study area is "Severe Impact". SINTACS is calculated similarly to DRASTIC. Weights assigned to the different parameters in the different SINTACS versions as well as vulnerability classes are presented in Tables 3  and 4.

Susceptibility index (SI) method
SI is a simplified version of the DRASTIC method and developed in Portugal by Ribeiro (2000). It is used to assess vertical specific vulnerability to anthropogenic pollution (Hamza et al. 2010). The term specific vulnerability is used to define the vulnerability of aquifer by a particular contaminant or group of contaminants. The method also considers properties of potential contaminants and their relationship to the different components of intrinsic vulnerability. It employs five parameters. The first four are identical to that used in the DRASTIC method: groundwater depth (D), net recharge (R), aquifer lithology (A) and topography (T). The fifth parameter (OS) is the translation of human activities through land use. Land cover classes have ranges from 0 to 100 (Table 5). The CORINNE Land Cover classification (European Community 1993) has been used to classify the land use type. The susceptibility index is calculated by Eq. 2: The SI method presents four vulnerability degrees according to the indices values obtained (Tables 6 and 7).
(2) SI = D w × D r + R w × R r + A w × A r + T w × T r + OS w × OS r    (2020) that considers nitratespecific groundwater vulnerability through the inclusion of a parameter representing the redox potential of the aquifer. This approach recognizes explicitly that the vulnerability of groundwater to nitrate contamination is influenced by the potential for denitrification to take place (Hansen et al. 2016), converting nitrogen in nitrate to a chemically reduced form (e.g. aqueous NO 2 − , or NH 4 + or gaseous N 2 or NH 3 ). The new N-parameter infers redox conditions in terms of a sampling depth above or below the interface between oxic and anoxic conditions in the saturated zone. DRASTIC_N is computed according to Eq. 3: where N R and N W are the rating and the weight for the additional parameter N. According to Voutchkova et al. (2020), a high nitrate pollution potential (rating 10) is assigned where oxic conditions restricting denitrification occur. A medium nitrate pollution potential (rating 5) is assigned to sampling depths that reside within ± 5 m of the interface between oxic and anoxic conditions; a low nitrate pollution potential (rating 1) is assigned to sampling greater than 5 m below the oxic/anoxic interface.

Single-parameter sensitivity analysis of each vulnerability assessment
A single-parameter sensitivity analysis was used to assess the influence of each of the different model parameters on the vulnerability measure. In this analysis, the actual or effective weight of each parameter is compared with the (3) DRASTIC N = DRASTIC + N W N R assigned or theoretical weight. The effective weight of a parameter is calculated using Eq. 4: where W is the effective weight of the parameter in one polygon, Pr and Pw are respectively the weight and the range of this parameter and V the total vulnerability index.

Input data
Assessment of groundwater vulnerability requires the use of reliable data. Bibliographic research and field measurements were used to acquire information on hydrogeological characteristics in the study area (Fig. 3). Table 8 shows the input data used. All amassed data were processed in ArcGIS (v. 10.2.2) for cartographic analyses and ERDAS Imagine for satellite image processing. The model builder tool in Arc-GIS enabled computation of raw data to obtain the various vulnerability maps according to a methodological approach described in the diagram below (Fig. 4).

Groundwater depth
Depth to groundwater from the surface determines the thickness of the unsaturated zone through which infiltrating water must pass before reaching the water table in the Thiaroye aquifer. This parameter therefore influences the degree, extent, attenuation and degradation processes of the pollutant. Data were acquired by 33 field measurements made in April 2017. This dry season period is characterized by the deepest groundwater levels. The point data collected were then spatialized by interpolation using the inverse weighted distance (IDW). The results closest to reality were obtained using high power squared. Recorded errors vary from − 1 to 1 m with a RMSE of 0.5 m.

Net recharge
Net recharge is the amount of water infiltrated per unit area that reaches the water table (Aller et al. 1987). This parameter is involved in the vertical transport of pollutant from the soil surface to the capillary fringe. It also controls the amount of water available for dispersion and dilution of solutes in the unsaturated zone. Recharge assessment in the study area has been extensively studied. Many approaches such as basic water balance calculations (Martin 1970;Cissé-Faye 2001), groundwater modelling  (Table 9). The net recharge assessment (4) W = Pr.Pw V .100

Aquifer lithology
This parameter refers to the geological characteristics of the aquifer. The data were extracted from Cissé-Faye (2001) and Madioune et al. (2011), which relied on stratigraphic logs,

Soil type
Soil type data from Maignien (1959) were used to determine soil typology. The description of the different units was based on granulometric and textural soil analyses. Soil texture data were obtained from the FAO World Soil Harmonized Database.

Topography
The study area topography is obtained from the digital elevation model covering the study area. The model was downloaded from the USGS website. The slope (%) was then calculated using the spatial analysis tool in ARCGIS 10.2.2.

Impact on the vadose zone
The vadose zone (unsaturated zone) is the unsaturated part of an aquifer above the water table. It is an important variable in the estimation of vulnerability, because it influences the residence time of pollutants and hence the attenuation probability. Unsaturated zone thematic map was based on the sandy facies predominating over clayey sands (Madioune et al. 2011). Unsaturated zone facies are therefore identified according to their degree of confinement.

Hydraulic conductivity
The hydraulic conductivity thematic map was derived from point data obtained by pumping tests (Geohydraulique 1972). This information was then converted into raster data by digital interpolation using the IDW interpolation.

Landuse
To determine landuse classes, a Landsat 8 OLI/TIRS image (scene 205-50) from April 03, 2017 was acquired and uploaded to the USGS database. Supervised classification was then performed with the ERDAS Imagine (v. 2014) software to obtain a landuse matrix file. The employed method of classification is Corine Land Cover (European Community 1993). Seven land use classes were observed: built, coastal dunes, surface water, vegetation, wetland, vegetable garden and landfill.

Redox state of the aquifer (N)
The nitrate reduction capacity incor porated in DRASTIC_N was estimated from a redox water type characterisation and depth-dependent nitrate reduction capacity outlined in Voutchkova et al. (2020). Analysis of sediment logs and groundwater chemical speciation (O 2 , NO 3 − , Fe 2+ and SO 4 2− ), which control and reflect denitrification capacity, was done to assess the position of the redox interface between oxic and anoxic conditions. The predominance of sandy facies and chemically oxidised species in solution (i.e. NO 3 − , dissolved O 2 and SO 4 2− ) confirm low denitrification capacity of the sampled shallow aquifer; a maximum rating of 10, reflecting a high nitrate pollution potential was therefore assigned to this parameter.

Groundwater quality data to test vulnerability assessments
To test the outcomes of aquifer vulnerability assessments, groundwater quality data that specifically relate to human activity were employed and included nitrate and fluorescent natural organic matter, tryptophan-like fluorescence (TLF) which describes fluorescence occurring from a range of compounds within the excitation-emission wavelengths associated with the fluorescence peak of the amino acid tryptophan (Baker et al. 2002). TLF data, expressed in ppb was recently used as an indicator of faecal pollution in the Thiaroye area of Dakar (Sorensen et al. 2020). These elements were chosen as specific contaminants to conduct a correlative analysis with the vulnerability index, on the one hand and to assess the groundwater reduction capacity, on the other hand. Nitrate concentrations were monitored from 2016 to 2020 in 40 groundwater samples whereas TLF were observed during 2018 from 73 groundwater samples. For the reduction capacity assessment of the aquifer, sulphate and dissolved oxygen were monitored from 2018 to 2020 on 40 samples whereas total iron data were available for 28 samples collected in 2020 (Table 10).

Vulnerability assessment
Four vulnerability maps applying DRASTIC, DRASTIC_N, SINTACS and SI models were generated (Fig. 5). Computed vulnerability indices are subdivided into three classes (moderate, high and very high). The "Very high vulnerability" class is more prevalent applying SI (36%) compared to DRASTIC and SINTACS (5% and 9%, respectively). In contrast, the "High vulnerability" class is more commonly computed in DRASTIC_N, DRASTIC and SINTACS (100%, 65% and 69% respectively) relative to SI (45%). DRASTIC vulnerability indices range from 129 to 188. "Moderate vulnerability" occurs in 30% of the study area (17.5 km 2 ) and corresponds to a depression occupied by lakes to the north-east, the zone of the Niayes de Pikine to the west and part of the southern zone at Thiaroye (Fig. 5a). These areas are characterized by a lower recharge rate and a generally sandy reservoir. "High vulnerability" is computed for 65% of the study area (39.7 km 2 ) and primarily located in the western agglomeration zone as well as, in part, to the south. The relatively shallow depth to groundwater, sandy soil texture and high recharge rates explain the computation of this "high vulnerability". "Very high vulnerability" is computed for only a small fraction (5%) of the study area, specifically located to the west in the agglomeration zone in the counties of Pikine and Thiaroye. It is also found to the south and southeast of the study area (South Thiaroye, Mbao, Yeumbeul and Keur Massar). A single parameter sensitivity analysis (Table 11) reveals that depth to groundwater, soil type and topography are the primary factors that influence the class of vulnerability index with actual weight (29%, 11% and 6%) greater than their respective default weight (22%, 9% and 4%). The net recharge, the characteristics of the aquifer and the impact of the unsaturated zone have comparatively less influence on the computed vulnerability index with average effective weights slightly lower than their default weights. As for the hydraulic conductivity, its influence is lower with an effective weight of 5% clearly lower than its default weight (13%).
For DRASTIC_N, the vulnerability indices range from 159 to 238. The "High vulnerability" class was computed throughout the study area (Fig. 5d). Variations in the index show that highest values occur in the agglomeration areas of Thiaroye, Diamaguene, Mbao and Keur Massar; lowest values are observed in the northern lake area. The sensitivity analysis shows DRASTIC_N to be influenced by depth, soil, topography and redox state of the aquifer. The other parameters (aquifer characteristics, impact of the vadose zone and hydraulic conductivity) are of lower sensitivity.

Evaluation of vulnerability analyses
Comparison of the vulnerability maps shows substantial differences in assessed vulnerability. Indeed, the "Very high" degree of vulnerability is greatest in the SI method (36%) relative to DRASTIC and SINTACS. A "High" vulnerability class is predominantly assigned using DRASTIC_N, DRASTIC and SINTACS methods. Table 12 highlights the degree of vulnerability of the different methods according to the occupancy rate. The semi-quantitative analysis of the correlation between these concentrations and the vulnerability categories are presented in box-plots (Fig. 6). The relationships between assessed groundwater vulnerability and observed groundwater quality range considerable (Figs. 6 and 8). We consider comparisons for nitrate first and then TLF.
For nitrate concentrations, DRASTIC shows 88% of samples (n = 35 with NO 3 − from 69 to 599 mg/L) are correlated with the "High" vulnerability. The first quartile varies from 69 to 224 mg/L whereas the interquartile range is 224 to 448 mg/L. Observed nitrates are less represented in "Moderate" (n = 4 ranged from 101 to 460 mg/L) and "Very high" (n = 1; 210 mg/L) vulnerability category. For DRASTIC_N, only the "High" vulnerability class is present. Nevertheless, the distribution of nitrate concentrations differs between minimum (179 to 209 mg/L) and maximum (210 to 240 mg/L) "High" vulnerability index values. Among the 43 samples, only four (101, 105, 315 and 459 mg/L) are correlated to 179-209 vulnerability index. Most of the samples coincide with "High" vulnerability index comprised between 210-240 where nitrates concentration are ranged from 70 to 599 mg/L (n = 39, median = 325 mg/L). SINTACS presents a variability of nitrate concentrations similar to DRASTIC. The "High" vulnerability category involves 90% (n = 35) of the samples ranged from 69 to 599 mg/L with the first quartile from 69 to 213 mg/L. and the interquartile ranged from 213 to 452 mg/L. The "Moderate" and "Very high" categories are less common with n = 1; 315 mg/L and n = 3 ranged from 209 to 401 mg/L respectively. For SI, the "High" vulnerability category concerns 35% of the samples (n = 14) with concentrations between 101 to 567 mg/L. The first quartile are ranged from 101 to 179 mg/L and the interquartile from 179 to 401 mg/L. The "Very high" vulnerability is found in 65% of the samples (n = 26) with concentrations between 69 and 599 mg/L. The first quartile are between 69 and 222 mg/L and the interquartile range between 222 and 449 mg/L.
To validate the vulnerability maps, a spatial correlation was made with observed nitrate concentrations. The nitrate concentrations map was obtained using average discrete values estimated to vary from 25 to 710 mg/L. Those values were spatialized by IDW interpolation (errors are within -13 and 10 mg/L; RMSE = 5 mg/L) and classified by their degree of contamination based on Stigter et al. (2006) (Fig. 7). The nitrate map thus Fig. 6 Box-plots of observed NO 3 − /TLF and vulnerability classes, representing the median, mean, 50th and 75th percentiles as vertical boxes and symbols as values vertical distribution obtained shows that the WHO guideline value (WHO 2004) for drinking-water quality (50 mg/L) is largely exceeded throughout the study area. Highest nitrate concentrations (> 300 mg/L) are observed throughout much of the study area.

Discussion
Comparison of observed nitrate concentrations with four vulnerability maps reveals substantial differences. To facilitate validation of these assessment methods, a new set of maps was created by subtracting the assessed vulnerability class from observed nitrate contamination class as per Stigter et al. (2006). This process argues that when the difference between the classes is in the order of minus one, zero or one (meaning that the vulnerability class is one class higher, equal or lower than the nitrate contamination class), the vulnerability assessment is considered correct. If a difference in the order of two or three classes is obtained, the vulnerability assessment is considered overestimated or underestimated respectively. Finally, when the difference is higher or lower by four or five, the vulnerability assessment is extremely overestimated or extremely underestimated, respectively. The analysis of the maps obtained, coupled with the coincidence rate determination (Fig. 8) shows a better vulnerability/contamination correspondence given by the SI model (56%). For the DRASTIC, SINTACS and DRASTIC_N methods, the correspondence rate is much lower (34, 38 and 43% respectively). Mapped correlations of aquifer vulnerability to observed nitrate concentrations in Fig. 9 also show the degree to which the four vulnerability assessments correctly estimate and either underestimate or overestimate the contamination risk.
Generally, the assessments by DRASTIC and SINTACS provide similar results of aquifer vulnerability. Both methods use the same parameters notwithstanding the higher weightings employed by SINTACS. Both underestimate vulnerability by a similar magnitude (coincidence rate). This underestimation occurs when the "Moderate to High" vulnerability and "High" vulnerability classes coincide with "Extremely high" levels of nitrate contamination. It may be due to the fact that these methods only assess intrinsic vulnerability without taking into account pollution risk derived from the vulnerability of the aquifer and pollutant load (Foster 1987). Nevertheless, good correspondence is established between vulnerability and contamination classes with a coincidence rate of 25% for the DRASTIC method and 28% for SIN-TACS. The inclusion of the new redox state parameter (N) in DRASTIC_N improves the vulnerability assessment, despite the low sensitivity of some parameters such as the impact of the unsaturated zone and the hydraulic conductivity. Notwithstanding this improvement, coincidence rates remain significantly lower than the SI method (see below), despite the application of the maximum rate (10) for redox state condition.
In the SI method, explicit consideration of land cover with high values attributed to the urbanized zone (LU = 75) and both soil and impact of the vadose zone parameters remove are the reasons for the good vulnerability estimation. In very small proportions (1%), an overestimation of vulnerability is established for the SI and SINTACS methods. In the discontinuous built-up areas located to the east of the watershed, the risk of pollution is low but the parameters of vulnerability are high. The assignment of high rating to land use (LU = 70) for SI as well as high weight to intrinsic recharge parameters and soil type for SINTACS lead to this overestimation. It is, however, preferable to an underestimation, in the sense that it involves the safe side of uncertainty. In other words, if vulnerability assessment were to be used by planners or decision-makers, negative consequences of uncertainty associated to underestimation would be avoided (Stigter et al. 2006).
Beyond these findings, the applicability of the DRASTIC and SINTACS methods to the study area is questionable. These methods take into account certain intrinsic parameters that in most cases are constrained in the study area. This is the case for I "Impact of the vadose zone" and S "soils" parameters. Indeed, in the urbanized zone with relatively shallow groundwater depth (less than 2 m), a vast network of on-site sanitation facilities, primarily septic tanks exist above the Thiaroye aquifer (Cissé-Faye et al. 2019). Recent work has shown densities of on-site sanitation vary between 1 and 70 septic tanks and pit latrines per hectare, with an  Stigter et al. (2006) average density of 20 per hectare . This autonomous management of domestic wastewater suggests the use of soil as a filter bed which, because of its attenuation capacity, contributes to the reduction of pollution risks. This zone is characterized by the presence of septic tanks that are mostly leaking, latrines and sumps often built at depth (1 to 2 m). The wastewater is either injected directly into the groundwater or percolates through a very thin unsaturated zone to reach the capillary fringe. This faecal loading is expected to play an important role in groundwater pollution as long as the ability of the unsaturated zone to attenuate contamination is limited. This is well observed in both DRASTIC and SINTACS methods integrating these parameters where, in the urbanized zone, vulnerability is underestimated.

Conclusions
Groundwater vulnerability to nitrate contamination in a low-income area of Dakar (Senegal) was assessed at the catchment scale (~ 60 km 2 ) using DRASTIC, DRASTIC_N, SINTACS and SI overlay and index methods. Vulnerability indices obtained from these models identify 3 classes of vulnerability: moderate, high and very high. Prevalence of the "Very high" degree of vulnerability is greater in the SI method (36%) relative to DRASTIC and SINTACS (5 and 9% respectively). Intrinsic vulnerability models (DRASTIC and SINTACS) present similarities in their assessment of vulnerability as they employ a common set of parameters but consider neither the nature of the pollutants nor factors managing the specific vulnerability such as land use. Single-parameter sensitivity tests show that groundwater depth, soil and topography have the greatest influence over the vulnerability rating. The shallowness of the water table and low relief of Dakar are considered to have the greatest influence on the vulnerability of groundwater to contamination. An improved correspondence is observed between assessed contamination by nitrate using the SI method (coincidence rate of 56%) compared to DRASTIC, SINTACS and DRASTIC_N with lower coincidence rates of 34, 38 and 43%, respectively. In the Thiaroye area of Dakar, representation of groundwater vulnerability using the SI method is improved by the explicit inclusion of land use. Our evidence from Dakar suggests that in urban areas of tropical Africa with shallow water tables, the SI method may prove to provide a more robust representation of groundwater vulnerability. Such assessments may prove invaluable for planning where groundwater quality monitoring data are absent.