Nineteen brands of bottled mineral waters were used for this study out of which seven are local brands and twelve mineral waters are imported brands. We bought those bottled in local supermarkets and independent food stores throughout Djibouti (Table 1). The water samples were collected between January and March 2019. As indicated on their labels, all the sample bottled were

validated for one year from the production date as per the Djibouti Ministry of Health certification. All bottled water were in polyethylene terephthalate (PET) containers with plastic screw caps. The capacity of these bottled varies from 0.5 to 2 liters. The data collected on the chemical characteristics of the main brands of water marketed in Djibouti comes directly from the labels glued to the bottled available on the market (eight important parameters appear most often i.e. Ca²⁺, Mg²⁺, Na⁺, K⁺, Cl⁻, NO 3 − , HCO 3 − and SO 4 2 − ).

Mineral water composition was studied by means of correlation analysis (CA), principal component analysis (PCA) and hierarchical cluster analysis (HCA) using Statistical Package for the Social Sciences (SPSS version 20) and SPAD. Application of multivariate analysis to complex data sets has attracted high scientific interest in recent years and they are now used in a wide range of application. Several multivariate statistical based studies have been performed to analyze bottled water utilizing different physico-chemical parameters [9] [15] [16].

Correlation analysis was applied to describe the degree of relation between two water chemistry parameters. Correlation is the ratio of the covariance of two variables to the product of their standard deviations [25]. The resulting correlation coefficient is a unitless number that ranges between −1.0 and +1.0. A value of −1.0 represents a perfect inverse relationship between the two variables, whereas +1.0 occurs when the two variables react in exactly the same way as their values change. A correlation coefficient of zero suggests that the two variables are independent of each other.

Principal component analysis was used to examine the relationships among the variables. In fact, this method helps to reduce the number of variables in the data matrix, to select the most discriminating parameters and to investigate to the overall variation of data. This method can be performed only if correlation among variables in the data matrix occurs [26]. Rotation of principal components was carried out using the varimax normalized method and only factors with eigenvalues greater than one were taken into consideration [27].

Hierarchical cluster analysis is a combination of techniques to classify large data into clusters on the basis of similarities or dissimilarities. Thus, resulting groups are similar to each other but distinct from other groups. In this study, HCA was used for searching the natural grouping among bottled water from different brands. The bottled water brands were classified according to their major ion composition. The data were standardized and the Euclidean distance was used as similarity measurement. The Ward’s method was used to obtain hierarchical associations. The result of the HCA is presented as a dendrogram with the numbers of groups selected based on visual examination of the dendrogram.

The chemical composition of nineteen bottled mineral waters marketed in Djibouti is summarized in Table 2. However, it is interesting to control the quality and consistency of the analysis carried out in major ions of bottled water. Calculation of the ionic balance error is essential to be able to present and to interpret the results of analyzes displayed on the labels of these bottled [5].

Water has a net neutral charge and therefore the positive charge contributed by the cations must equal the negative charge contributed by the anions, Equation (1):

( ∑ cations ) = ( ∑ anions ) (1)

*L.D: Limit of Detection.

Of the 19 physico-chemical variables (consisting of major ions, minor ions, trace elements, and physical parameters) in the compiled database, eight variables (Ca²⁺, Mg²⁺, Na⁺, K⁺, Cl⁻, NO 3 − , HCO 3 − and SO 4 2 − ) occur most often and were thus utilized by the multivariate pattern recognition methods. Analysis of chemical and physical properties of these bottled waters was carried out by official laboratories that have been certified by the Djiboutian government and accuracy and precision of the laboratory results were not questioned in this study.

However, the chemical composition of the nineteen bottled mineral waters has made the subject of a first verification by Ionic Balance Error (IBE). According to Freeze and Cherry, the percentage ionic balance error is calculated as Equation (2) [28]:

IBE ( % ) = ( ( ∑ cations − ∑ anions ) / ( ∑ cations + ∑ anions ) ) × 100 (2)

The ionic balance method allows to control the quality of the analysis carried out on the major elements. This validation process is a prerequisite for a reliable presentation of the results of analysis and their interpretation. The IBE can be positive or negative. A positive IBE indicates that the water sample has a higher concentration of cations than anions. Conversely, a negative IBE indicates that anions are more abundant.

Table 3 represents the IBE calculated on each of the mineral water samples.

Applying this check technique, on data of 19 brands waters collected, exhibits that 36.8% of data (7 out of 19 measurements) exceeded the acceptance limit (±10%) of ionic balance error percent. This indicates that around one third of the samples have measuring or sampling errors, which means that the resultant data quality is insufficient for drawing reliable conclusions about water quality. Deviations from electroneutrality come from random or systematic error in one or more constituents or an incomplete analysis that neglects some significant constituent [29]. Seven bottled waters mineral (Bio, Iljano, Biyo Safi, Tayse-Le, Vitall, Bahçepinar, Hamidiye) presented an ionic balance error above +/− 10% are excluded from the analysis database. Therefore, 12 bottled waters mineral out of the 19 collected were selected for statistical analysis (Table 4).

The maximum, minimum, standard deviation, mean and coefficient of variation (CV %) values of the 9 chemical parameters of 12 bottled water brands selected out of 19 bottled marketed in Djibouti were determined and presented in Table 4. Comparison of these values with those by World Health Organization [30] and the United State Environment Protection Agency [31] are also shown in

Table 5. The chemical composition of natural water depends on many factors which includes chemistry of atmospheric precipitation, mineralogy of the rocks encountered along the flow path, residence time of the surface or groundwater, topography and climate [32] [33]. Hence, each bottled water brands has its own physical characteristics and chemical properties that are defined by a unique combination of these factors.

As can be seen in Table 4, the coefficient of variation of major constituents bottled water varies from 61.32% to 167.96% for ion concentration. Sulfate has the biggest coefficient of variation (167.96%), followed by calcium (140.64%), magnesium (115.87%), nitrate (99.48%), bicarbonate (90.52%), chloride (80.68%), potassium (69.17%) and sodium (61.32%). The values coefficient of variation showed that there a great variability between the water in terms of richness in chemical elements with the exception of pH which has a low coefficient of variation (2.57%). In this study, two local bottled mineral waters (Palmaraie and Okar) have similar chemical composition (Table 4). This can be explained by the fact that the water comes exclusively from the same groundwater. Güler et al. pointed out the samples with similar chemical characteristics often have similar hydrologic histories, similar recharge areas, infiltration pathways, and flow paths in terms of climate, mineralogy, and residence time [32]. On the other hand, we see in Table 4 that the major ions in all brands demonstrate a great variations in composition which could be attributed to natural environment from which the water is taken (geological setting, climate, etc, ...), source water composition and type of treatment/purification technique(s) applied during the production [32]. Comparisons of the values chemical composition of the twelve brands water selected with those set by the United State Environment Protection Agency [31] and World Health Organization [30] international standards are also shown in Table 5.

WHO: World Health Organization; USEPA: United State Environmental Protection Agency.

pH is a measure of the concentration of hydrogen ions and hydroxyl ions in water. The limit of pH value for drinking water is specified as 6.5 - 8.5 (Table 5). The pH of all brands was around neutral ranging from 7 to 7.7 with a mean of 7.35. All brands were within recommended prescribed by WHO and USEPA (Table 5). The pH has no direct impact on the consumer [34].

The Highest concentration of HCO₃ is found in Hépar brands (383.7 mg/l), while the lowest recorded in Super Gulf was 14 mg/l with a mean 162.97 mg/l. The source of bicarbonate is attributed to natural processes such as dissolution of carbonate mineral in the presence of CO₂ in soil [35]. In the present study none of the local brands had HCO₃ level exceeding the above standards, whereas five on the imported brands (Hépar, Contrex, Evian, Thonon and Courmayeur) have HCO₃ levels that exceeded the standard guideline recommendations. It has been reported that, drinking of mineral water rich in bicarbonates prevent or improve type 2 diabetes in humans [36].

Concentration of Cl ranges from 0.3 - 40 mg/l with a mean of 19.44 mg/l. Most of the water brands contain lower amounts of Cl. None of the values exceeded the WHO and USEPA permissibles limits. An excess concentration of chloride (>250 mg/l) causes water to taste salty.

Concentration of NO₃ in the studied water bottles varied from <0.3 to 8 mg/l with a mean of 2.80 mg/l. All the water brands studied have NO₃ levels falling short of WHO and USEPA limits. Nitrate is the most common chemical contaminant in the world’s groundwater and aquifers. It can be converted in the stomach infants to nitrite (NO₂), which can then oxidize hemoglobin to methemoglobin, making it difficult to transport oxygen around the body [37] [38].

Table 4 shows that the SO₄ concentration varied between 2.74 to 1530 mg/l with a mean of 349.31 mg/l. However, three of the imported brands (Contrex, Courmayeur and Hépar) have sulfate levels that exceeded the recommendations of WHO and USEPA standards.

Concentration of Ca varied from 8 to 576 mg/l with a mean of 155.95 mg/l. The minimum and maximum Mg concentration were 2.4 and 119 mg/l with a mean 29.24 mg/l. However, three brands (Contrex, Courmayeur and Hépar) had Ca levels that exceeded the WHO and USEPA standards.

All the water brands have Mg levels well within the international standards limits expect two of the imported brands (Contrex and Hépar). They exceed 50 mg/l level recommended by WHO standards.

Regarding the concentration of Na ranging from 0.6 to 14.2 mg/l with a mean of 6.60 mg/l, they were very well within the recommended WHO and USEPA standard limits for drinking water.

As can be seen from Table 4, the potassium level varied < 0.2 to 6 mg/l with a mean of 2.09 mg/l. None of the values exceeded the maximum limit of 12 mg/l set by WHO).

Water hardness is caused by minerals, such as magnesium and calcium. The total water hardness, as Ca²⁺ and Mg²⁺ ions, was calculated according to Equation (3) [39].

Total Hardness = 2.5 [ Ca 2 + ] + 4.1 [ Mg 2 + ] (3)

where [Ca²⁺] and [Mg²⁺] are the concentrations of calcium and magnesium (in mg/l) displayed on the bottled labels, while 2.5 and 4.1 are their molar mass ratios per 100 g of CaCO₃.

Bottled water brands for this study were classified based on total hardness (TH) as shown in Table 6. General guidelines for classification of water are: 0 to 60 mg/L as calcium carbonate is classified as soft; 61 to 120 mg/L as moderately hard; 121 to 180 mg/L as hard; and more than 180 mg/L as very hard [40]. According to this classification, water hardness of the bottled water ranged from 62.8 to 1860.4 mg/l CaCO₃ with a mean value of 509.8 mg/l (Table 6). Seven of the brands with hardness between 62.8 and 96 mg/l are considered moderately hard and five brands with hardness between 295.6 and 1860.4 mg/l are considered very hard. All the 12 brands bottled are qualified by hard water category. Hardness of water which is due to the presence of calcium and magnesium salts in water, does contribute towards total calcium and magnesium human dietary needs, which has a beneficial effect on bone structure [40]. Water hardness has no known adverse effects. However, some evidence indicates its role in heart disease [41]. Moreover, epidemiological studies suggest that consumption of Mg may reduce the frequency of sudden death and Ca may help prevent osteoporosis in humans [42].

The WHO limits hardness for drinking water between 100 - 500 mg/l. It may be seen from Table 6 that the total water hardness of all the three water brands (Hépar, Courmayeur and Contrex) is above than the limits established by WHO/USEPA. They can be crucial in the prevention of magnesium deficiency,

as well as contribute significantly to the daily calcium intake for persons with calcium deficient diets. However, the use of very hard water for infants or pregnant women is not recommended.

In this study, the water type is defined by all ionic constituents that contribute at least 25% to the total anionic or cationic composition of water (Table 7). The most frequently observed water type is Mg-Ca-Cl-HCO₃ and Ca-Mg-HCO₃. All the studied water brands are dominated by either Ca or Mg except two brands (Courmayeur and Mont Blanc), which are dominated by Ca-SO₄ and Ca-HCO₃-SO₄.

Bottled water is classified into eight major water types according to the Piper’s diagram (Figure 1).

In order to perform a comparison between different bottled water types, main components (Na, K, Ca, Mg, Cl, SO₄ and HCO₃) of the 12 bottled waters selected among 19 are plotted on the Piper diagram [43]. The diagram displays the relative concentrations of the major cations and anions on two separate trilinear plots, together with a central diamond plot where points from two trilinear plots are projected. As shown in Figure 1, most of the water brands are Ca, Mg, HCO₃, SO₄ and Cl type water.

The Pearson’s correlation coefficient between major ions in bottled water brands (n = 12) are presented in Table 8. The variables having coefficient value (r) > 0.5 are considered significant. Strong positive correlations between SO₄ and Ca (r = 0.99), SO₄ and Mg (r = 0.91), Ca-Mg (r = 0.89), HCO₃ and Mg (r = 0.70), HCO₃ and Ca (r = 0.63), HCO₃ and NO₃ (r = 0.56), HCO₃ and SO₄ (r = 0.55), Mg and Na (r = 0.50) and K and Na (r = 0.51). No significant correlations were obtained for Cl and NO₃ with the others variables, and this corresponds well to their chemical composition which was low in all the bottled water brands.

PCA of the water quality variables extracts two components with eigenvalue > 1.0 according to the method of the percentage of inertia. This method resulted in two principal components (PC1 and PC2), which account ~68.67% of the total variance in the dataset (Table 9). Figure 2 shows results of the PCA analysis for 12 brands of bottled waters. Analysing the projections of the variables on the factorial plane by plotting PC1 against PC2 (Figure 2) and respective PC contribution (Table 9), it is concluded that the first principal component (PC1) accounted for 45.85% of the total variance and is characterized by high positive contribution of Mg, Ca, SO₄ and HCO₃ (with contribution 95%, 92%, 91% and 83%, respectively). This component appears to be clearly dependent on geological composition of the substrate, being located mostly in association with carbonate rocks [9] [18]. This PC1 may be considered as an index of water hardness which Ca and HCO₃ are the major dissolved species in limestone aquifers, while the presence of Mg is attributed to either magnesian calcite or dolomite [44].

The second Principal Component (PC2) represents 22.82% of the total variance within the data and is characterized by negative contribution in K, Na and Cl (with contribution 75%, 87% and 52%, respectively). This component represents dissolution of the evaporite rocks.

HCA was used to find the natural grouping of the 12 bottled waters selected out of the nineteen waters brands marketed in Djibouti in accordance with chemical composition parameters. The result of the HCA is presented in the form of a dendrogram (Figure 3). Based on eight parameters, the HCA classified 12 bottled waters brands in four groups. The first group is composed of three brands (Contrex, Hépar and Courmayeur), the second grouped four brands (Evian, Hayat, Mont Blanc and Thonon), the third contained one brand (Volvic)

and the fourth groups comprised four brands (Alain, Okar, Palmaraie and Super Gulf).Two local bottled (Okar and Palmaraie) are of similar ionic profile as bottled water imported (Al ain) according to the amount of 8 major ions. The characteristic parameter of the first group is Cl (ion related to the salinity of the water) and the forth group is SO₄, Ca and Mg (result the dissolution of the carbonate rocks). Or the second and third groups have no characteristic parameter. The results of HCA coincide with those obtained from PCA.