The water quality data, including 232 sampling locations across the country were obtained from Senegal’s water planning department (“Direction de la Gestion et la Planification des Resources en Eau”). All water samples were collected in 250 and 500 ml polyethene bottles from 2013 to 2019. The chemical analyses, which included electrical conductivity (EC, measured in the field with a multi-parameter probe, HANNA, HI 9828), total dissolved solids (TDS) derived from EC (TDS = 0.64 * EC), Cl⁻, SO 4 2 , NO 3 − , F⁻, HCO 3 − , Ca²⁺, Mg²⁺, Na⁺ and K⁺ were conducted by the hydrochemistry laboratory of the Geology Department of Dakar University. The ion chromatography method with AQUION-DIONEX equipment was used for the chemical analyses. The exchange processes were carried out through AS14 A-AERS 500 columns for anions and CS12 A-CERS 500 for cations. The carbonate species were analyzed by titrimetry using a 0.05 N sulphuric acid solution. From more than 600 samples, ionic balances (IB) were used as cross-validation for chemical analysis. Samples with IB > 5% were removed from the database and 380 samples were retained for this present study.

Descriptive statistics, correlation matrix and multivariate statistical models (PCA and HCA), together with classical hydrochemistry data visualization (e.g., Piper and Schoeller diagrams) were used to identify the potential factors influencing the water chemistry and to group the samples according to their water quality characteristics. The correlation matrix illustrates the relationship between all pairs of chemicals and is an effective tool for summarizing and visualizing patterns in large datasets. On one hand, PCA is typically used to locate the directions of maximum variance in high dimensional data in order to project them onto a reduced dimensional subspace while preserving most of the information. It is used in this study to identify likely factors that influence water chemistry based on chemical loads onto the principal components. All chemical parameters were included except for TDS which was derived from EC values. 214 water samples (we removed missing values) were considered for the PCA and only components with eigenvalues ≥ 1 were considered significant following Kaiser criterion [35]. HCA, on the other hand, is a useful method for objectively arranging a large dataset into clusters based on a set of characteristics. Euclidean distances were used as a metric of dissimilarity between sets of observations, and Ward’s criterion for the linkage rule [36]. All statistical analyses and data visualizations were carried out in Python using the scikit-learn machine learning and seaborn packages.

All data sets used in this present study were compiled in a GIS environment by establishing a database geographically referenced to the UTM-WSG84 projection. The spatial distribution of groundwater quality parameters was computed using the inverse distance weighted (IDW) method with ArcGIS software’s spatial analyst modules. IDW is a spatial interpolation technique that estimates values between measurements [37]. The prediction maps of TDS, EC, cations, and anions are classified according to [38] standards for drinking water. These raster data are then used to generate the groundwater quality index (GWQI) map, which represents the composite influence of different water quality parameters on overall water quality for drinking purposes.

The GWQI computation has been accomplished through two steps: 1) weighting assignment of each criterion by AHP, 2) quality rating scale and demarcation of GWQI. More details about the computation of the water quality index can be found in [31] [39] [40] [41].

The adequacy of the groundwater for irrigation purposes was evaluated using indicators including sodium percentage (Na%), sodium adsorption ratio (SAR), magnesium hazard (MH) and residual sodium carbonate (RSC) which are listed in Table 2. The rating was then based on standard value intervals following [46] [47] [48] [49].

The statistical summary of the groundwater quality parameters in comparison with WHO standards are presented in Table 3. The electrical conductivity (EC) values of groundwater samples ranged from 19 to 6480 µs·cm⁻¹, with an average of 1764 (median of 1190 µs/cm) which is slightly above the WHO standard (1500 µs·cm⁻¹). The high standard deviation value (1478 µs·cm⁻¹) revealed different water types ranging from fresh to saline water, suggesting different geochemical processes occurring in the aquifer. Cations and anions had average values below the WHO standards, except Na⁺ and Cl⁻ which mean values were well above (317 and 360 mg·l⁻¹, respectively). The ionic dominance patterns were in the order of Na⁺ > Ca²⁺ > Mg²⁺ > K⁺ for cations and Cl⁻ > HCO 3 − > SO 4 2 − > NO 3 − > F⁻ for anions.

The correlation matrix (Figure 2) showed that EC has a strong positive correlation with Cl⁻ (0.96), Na⁺ (0.94) and K⁺ (0.79), and a moderate correlation with F⁻ (0.60) and HCO 3 − (0.56) reflecting the significant contribution of these elements to the acquisition of groundwater mineralization such as 1/rock-water interaction, 2/ion exchanges, 3/salinization through actual or past marine intrusion in the coastal area and 4/freshening and mixing through leakage from upper aquifers (e.g., Paleocene and Eocene aquifers) and groundwater recharge in the unconfined area [16] [17] [19]. The Cl⁻ gave a strong positive correlation with

Na⁺ (0.91) suggesting the same origin from either halite dissolution or salinization processes (e.g., seawater intrusion) as commonly reported in groundwater coastal areas [50] [51] [52] [53]. Since equal amounts of Na⁺ and Cl⁻ are mixed in groundwater as a result of halite dissolution, the ratio will be close to one. The Na⁺/Cl⁻ ratios for the study area ranged from 0.33 to 28.75, with a mean value of 2.35, suggesting the possibility of additional mechanisms releasing sodium into the groundwater. Indeed, if the ratio is less than 1 and close to the marine ratio value (0.86) it may be due to seawater intrusion [54] [55]. Instead, a ratio greater than 1 suggests that sodium sources are from ion exchange reactions [51] [56] or Silicate weathering [57]. Furthermore, the Simpson ratio (Cl⁻/ HCO 3 − ) can also be used to classify the impacts of seawater intrusion on groundwater mineralization [58]. Generally, groundwaters that have not been contaminated by seawater intrusion have a Simpson ratio of less than 0.5. In this study, 27% of samples had a value greater than 0.5, which confirmed marine processes’ occurrence in groundwater mineralization. Reference [59] suggested a paleowater origin of saline waters in the west part of the aquifer instead of actual seawater intrusion induced by over-exploitation. Na⁺ showed a poor correlation with Mg²⁺ (1.57) and non-significant (at p = 0.05) with Ca²⁺, due probably to the substitution of Mg²⁺ or Ca²⁺ in the groundwater with Na⁺ in the aquifer reservoir which corroborated the cation exchange processes [17] [50] [60]. The significant correlation found between Ca²⁺ and Mg²⁺ (0.86) suggested the presence of magnesium calcareous minerals in the aquifer, such as dolomite (CaMg(CO₃)₂), and their dissolution into groundwater. The positive high correlation between F⁻ and Na⁺ (0.75), HCO 3 − (0.72) and K⁺ (0.70), and negative poor correlation between Ca²⁺ (−0.30) indicated leaching of fluoride-bearing minerals such as apatite [59]. There was no significant correlation between NO 3 − and the other parameters, suggesting another source such as anthropogenic activities nearby the sampling site or leaking from upper aquifers as evidenced in the western and northern part of the study area by [14] [17]. Thus, rock-water interaction, residual/actual seawater water intrusion, cation exchange and localized mixing and pollution resulted in the hydrogeochemical evolution of groundwater quality in the Maastrichtian aquifer.

Results from PCA are presented in Table 4, which summarized the determined Principal Components (PC) loadings, their eigenvalues and the percentage of the variance contributed to each PC. Three components with an eigenvalue greater than 1 were sufficient to explain 81.5% of the variance in the correlation matrix as follows: 47.9% by PC 1, 22.9% by PC 2, and 10.6% by PC 3. The PCA loading circle plot and the contribution of each variable to the three principal components are presented in Figure 3. PC 1 was strongly loaded by Na⁺, K⁺, EC and Cl⁻, and by F⁻ and HCO 3 − to a lesser extent (Figure 3(b)), contributing to salinization processes mainly controlled by rock-water interaction (dissolution of halite and fluorite, silicate weathering) but also seawater intrusion and cation exchange reactions. PC 2 was loaded by the couple Ca²⁺/Mg²⁺, which comes primarily from dolomite dissolution (Figure 4(c)). PC 3 accounted for NO 3 − and SO 4 2 − derived from localized pollution around boreholes and gypsum (CaSO₄, 2H₂O) dissolution as evidenced by [14] [59] (Figure 4(d)).

Due to the variability in ionic contents and involved hydrogeochemical processes deciphered with the correlation matrix and PCA, HCA was applied to identify the similarity between water samples.

The dendrogram distinguished four clusters (C1, C2, C3 and C4), with C3-C4 linked at higher distances (Figure 4(a)). Table 5 summarizes details on each cluster relative to the mean value of chemical parameters and the main water types based on the Piper (Figure 4(b)) and Schoeller (Figure 4(c)) diagrams. The first cluster, C1 (N = 148) was characterized by freshwater (average EC value of 594 µs/cm) with mean values of all ions below WHO standards and a mixed-HCO₃ water type. This cluster was mainly influenced by freshening, cations exchange reactions and mixing processes. Instead, C2 (N = 3) was mainly saline (Na-Cl) water type with higher mineralization compared to other clusters (EC = 5953 µs/cm) and ionic contents well above WHO standards except for F⁻ (0.6 mg·l⁻¹) and HCO 3 − (203.3 mg·l⁻¹). C3 (N = 17) exhibited brackish water type (Na-Cl) but with lower ionic contents than C2 (EC = 3247 µs/cm), except for HCO 3 − , F⁻ and NO 3 − which were higher compared to other clusters. C2 and C3 are mainly influenced by salinization processes and potentially by localized anthropogenic activities around boreholes. Finally, C4 (N = 46) presented similarities with C3, depicted by brackish water type (Na-mixed) with lower mineralization contents than C2 and C3.

The spatial distribution of TDS values, cations, and anions concentration, classified in terms of WHO standards are shown in Figure 5 and Figure 6. The TDS prediction map (Figure 5(a)) indicated an increasing gradient of mineralization from the East characterized by freshwater (<500 mg·l⁻¹) towards a meridian band located in the centre part of the area where a significant rise in mineral contents is observed. Within this brackish central band, TDS is higher than the

WHO standards, ranging from 1.5 to 3.8 g·l⁻¹. Reference [59] suggested that the high mineralization observed in this area is the result of past marine intrusion during the Eocene and Holocene periods but also rock interaction with paleowaters characterized by low transit time. As indicated by [16], apart from the Diass Horst, South and southeast borders, the most recent recharge episodes took place during the Holocene humid periods (between 12,000 and 3000 years BP). In addition, radiocarbon dating (¹⁴C) has shown that the Maastrichtian aquifer contains paleowaters in the brackish band that are over 125,000 years old [16] [59]. Towards the West, on the other side of the brackish band, TDS

decreases again to reach values <500 mg·l⁻¹ in the Diass Horst which constitutes the recharge zone of the aquifer [17]. The spatial distribution of Ca²⁺ (Figure 5(b)) and Mg²⁺ (Figure 5(c)) concentrations revealed that most of the study area has values less than 75 mg·l⁻¹, except the northern part of the study area, which is higher than the WHO standards. This high concentration in the North, especially for Ca²⁺ (241 mg·l⁻¹), would come from the reverse cation exchange process with Na⁺, induced by the occurrence of the Guier dome (anticlinal structure of Lake Guier associated with faults) which allows mixing with the upper aquifer [61] [62]. Regarding Na⁺ (Figure 5(d)) and K⁺ (Figure 5(e)) concentrations, their spatial variations were relatively similar to TDS, especially for Na⁺ whose concentration is higher compared to K⁺. From a few tens of milligrams in the East, the concentration gradually increases towards the West to reach, at the level of the brackish band; values higher than 12 and 200 mg·l⁻¹ for K⁺ and Na⁺ corresponding to the WHO threshold. F⁻ (Figure 6(a)) and Cl⁻ (Figure 6(d)) prediction maps were also similar to those of TDS and Na⁺ with a high concentration greater than the WHO standard in the brackish band. According to [59], the high fluoride content does not come from residual seawater, as may be the case for chloride and Sodium. They would come from the interaction between groundwater and rock from fluoride minerals associated with phosphate sediments (e.g., apatite fluoride) as also evidenced in other studies [63] [64] [65]. The prediction map of HCO 3 − showed that most of the study area outside of the brackish band has values below WHO standard (<300 mg·l⁻¹). Sulfate contents were lowest in the eastern and south-eastern parts of the country, where the concentrations were less than 100 mg·l⁻¹ (Figure 6(c)). While in the central and northern parts of the country, a higher concentration was observed exceeding 600 mg·l⁻¹ in some zones due to the dissolution of gypsum minerals. Nitrate contents were also low, less than 5 mg·l⁻¹ in most samples (Figure 6(e)). The only exceptions are a few samples collected in the central and the North-West part, where concentrations above 100 mg·l⁻¹ have been recorded. However, in this part of the country, the Maastrichtian is confined and covered with layers thicker than 200 m, therefore pollution from agriculture is not likely to occur. This nitrate would originate from the leaking and mixing of diverse types of water percolating from the upper aquifers, which are vulnerable to pollution.

Analytical hierarchical processes (AHP) have been used to compute suitable weights of the ten physicochemical parameters using a pairwise comparison matrix (Table 6). Fluoride and nitrate were given the highest weights, followed by EC, SO 4 2 − , and Cl⁻. Instead, HCO 3 − has given the lowest weight.

The spatial distribution of GWQI revealed that 20% of the study area falls in unfit water corresponding primarily to the brackish band, characterized by high TDS, F⁻, Cl⁻, Na⁺ and K⁺ contents (Figure 7). Poor to extremely poor water represents 26% of the study area and is located on both sides of the brackish band. Conversely, excellent, and good water types represent 25% and 29% of the study area situated in the East, southeast, and far Western parts of the country. Thus, the Maastrichtian aquifer has an essential reserve of high water quality standards, representing more than 54% of its total surface area in Senegal.

SAR, %Na, MH, and RCS indices were used to ascertain the suitability of the Maastrichtian aquifer for agriculture usage. For irrigated water, sodium hazard is a valuable tool to evaluate suitability. A significant increase in SAR at the root zone may influence soil permeability and cause the formation of progressively permeable layers, resulting in water logging [66] [67]. Furthermore, high Na⁺ concentrations in irrigated water may cause absorption by clay particles, which is regulated by Ca²⁺ and Mg²⁺ ion exchange. In this present study, SAR varied between 0.09 and 115 (Figure 8(a)) and Na% ranged from 2.09 to 99.56 (Figure 8(b)). The suitable water types for irrigation based on these indices were observed in the East, Southeast and the western part of the country. Conversely, unsuitable, and doubtful water for irrigation was observed in the brackish band and its vicinity. Other, essential indices for irrigation water ratings are MH and RCS [68]. An excess of magnesium ions in groundwater degrades soil quality by increasing alkalinity, which may reduce agriculture productivity [26]. MH values of the study area ranged from 5.7% to 98% (Figure 8(c)) and RSC varied between −21 and 12.4 meq·l⁻ (Figure 8(d)). Their spatial variation showed that most of the aquifer has suitable water for irrigation except in some parts in the brackish band and the South, West, and Southeast Regions to a lesser extent.