Current Issue

The 18 bottled water production sites (plants) studied are located in northern (Gyeonggi and Gangwon Provinces) and middle (Chungbuk Province) parts of Korea (Fig. 1). Each site has 1 to 6 production wells (a total of 63 wells) with some distant groundwater wells (total n=53) and nearby streams (total n=62, only a few meters away from the production wells). The depth of the production wells ranges from 67.4 to 291 m with a mean value of 183.1 m (Table 1). The wells mostly tap into fractured bedrock aquifers with a few meters of upper casing and bentonite grouting. The permitted production rate per well varies from 30 to 660 m³/day with a median value of 170 m³/day.

Fig. 1. 
Locations of bottled water production plants for this study in Korea and simplified geological map.

The geology of the production sites consists of plutonic rocks (mostly granite, biotite granite, granodiorite) (38.9%), metamorphic rocks (mostly banded gneiss, biotite gneiss, green schist) (50%), and limestones (11.1%). The overburden above the bedrock is approximately 10 m, composed of a few meters of upper soils and weathered rocks. The degree of fracture development in the bedrock aquifers varies at each site.

Table 2 shows the statistics of major parameters of the production wells (n=63). The natural groundwater levels (before pumping) ranged from 2.03 to 47.69 m below the ground surface with a mean value of 15.84 m. The maximum and minimum drawdown due to groundwater pumping were 139.47 and 1.32 m (mean=57.98 m), respectively. The hydraulic conductivity for the production wells varied (four orders of magnitude), and it ranged from 4.2×10^-6 to 3.0×10^-2 cm/sec with a median of 3.6×10^-5 cm/sec. The median value is within a typical range of fractured rocks (Freeze and Cherry, 1979) while the arithmetic mean (2.0×10^-3 cm/sec) is relatively higher than that for typical fractured rock aquifers.

The storage coefficients of the aquifers, where the production wells are tapped, are between 4.4×10^-6 and 1.96×10^-1, with a median value of 6.0×10^-3. This coefficient is higher for typical bedrock aquifers, indicating that these wells are generally highly fractured. The transmissivity of the production wells ranges from 0.69 to 984.96 m²/day (median= 4.64 m²/day), with specific capacity of 0.52-321.84 m³/day/m (median=4.45 m³/day/m). Considering the median values of specific capacity and drawdown, a volume of 289 m³ is pumped daily at each production well, which is relatively similar to the mean value of permitted production rate per well (227 m³/day; see Table 1).

This study focused on the water quality of production wells, groundwater wells and nearby streams. Therefore, in addition to the relevant geological and hydrogeological properties described above, we also collected water quality data including field measured parameters (water temperature, oxidation-reduction potential (ORP), dissolved oxygen (DO), electrical conductivity (EC), pH, chemical compositions, microbes, hazardous inorganic and organic materials, and aesthetic parameters), from the EIS reports for 18 bottled water production sites. We secured 630 water quality data sets for production wells (450 sets), surrounding groundwater (78 sets), and nearby streams (102 sets). It is noted that not all data sets have values for all of the above parameters.

In addition to the basic statistical analysis on the collected data and plots using chemical components levels, including Piper and Gibbs diagrams, we also conducted a principal component analysis (PCA) and cluster analysis (CA). For this, we excluded the data sets with not detected (n.d) values and finally included 12 quality components: Ca²⁺, Mg²⁺, Na⁺, K⁺, HCO₃^-, Cl^-, SO₄^2-, NO₃^-, KMnO₄ consumption, color, pH, and turbidity. Among the 630 data sets, only 115 data sets were selected for the analyses. We first tested the normal distribution of the data using the Shapiro-Wilk test at a 95% confidence level (Lee et al., 2009; Noshadi and Ghafourian, 2016). Furthermore, to avoid any prevailing dominance by parameters with large magnitudes and standardize the weight of each component, we conducted z-transformation on the analyzed parameters (mean=0, standard deviation=1).

PCA is a process to reduce the number of variables in data sets by computing the principal components and increasing interpretability with minimum information loss, using only the first few principal components and neglecting the rest (Jolliffe and Cadima, 2016). CA is a technique to group similar observations into a small number of clusters based on the observed values of several variables for each individual, and the grouping is represented by a dendrogram (Lee et al., 2009; Fabbrocino et al., 2019). In this study, the k-means clustering was applied to group the data sets. These multivariate analyses were conducted using the free statistical program Past 3.06 (Hammer et al., 2001).

Table 3 shows the statistical summary of field measured parameters and chemical compositions of water from production wells, surrounding groundwater wells and streams. The production well water temperatures ranged between 9.2 and 21.7℃ (mean=14.1℃); these values were similar to those of surrounding groundwater wells (range=8.2-19.5℃, mean=14.1℃; Fig. 2). Considering the groundwater temperatures (range=7.4-20.4℃, mean=14.2℃) of bedrock aquifers of Korea (Lee and Hahn, 2006), the water temperatures of production wells and surrounding groundwaters are considered to be normal and stable. The stream water temperatures, which were largely affected by atmospheric temperatures, were much more variable (range=0.7-24℃) than those of groundwater.

Fig. 2. 
Box plots of field measured parameters for production well, groundwater well, and stream water: (a) water temperature (℃), (b) oxidation-reduction potential (mV), (c) dissolved oxygen (mg/L), (d) electrical conductivity (μS/cm) and (e) pH (SU).

The ORP values of production wells, indicative of redox conditions, ranged from 89.5 to 496.9 mV (mean=230.8 mV); these were very similar to those of surrounding groundwaters (range=46.2-476.8 mV, mean=232 mV) (Fig. 2). However, the ORP of stream water was relatively higher (mean=267.3 mV). The ORP of the three water bodies was all indicative of their oxic conditions (see Fig. 2). These ORP values are consistent with those of DO. The DO ranges of production wells, groundwater wells and stream water were 4.6-9.6 mg/L (mean=7.8 mg/L), 4.3-10.5 mg/L (mean=8.1 mg/L), and 6.5-11.7 mg/L (mean=10.3 mg/), respectively. The high DO values of stream waters are expected because they are in direct contact with the atmospheric air. The high DO values of groundwater wells may be explained by rapid rainfall infiltration due to very shallow well depths. However, the very high DO values of production wells tapping the bedrock aquifers are quite strange. The very high values for both ORP and DO indicate a probable intrusion of nearby stream water into the production wells. This interpretation is also supported by very low and high water temperatures of production wells, which indicate that the groundwater of production wells is affected by the stream water with much varying water temperatures.

Mean electrical conductivities of the production wells, surrounding groundwaters, and stream waters were 176, 165, and 122 μS/cm, respectively. The high EC values of production well water and surrounding groundwaters indicate higher levels of dissolved solids, greater than that of the stream water. The pH values of the production well (mean=7.5) and stream water (mean=7.4) are similar but quite different from that of the surrounding groundwater (mean=7.1). This also indicates the production well water may be influenced by the stream water (see Fig. 2).

As shown in Table 3, concerning mean values for production wells, calcium (Ca²⁺) is the most prevailing cation (range=3.8-90.3 mg/L, mean=22.4 mg/L), followed by Na⁺, Mg²⁺, and K⁺ concentrations. The surrounding groundwater wells and stream water also show the same patterns. The anions showed the following order for production wells, HCO₃^-> SO₄^2->Cl^->NO₃^-; for surrounding groundwater wells, HCO₃^->NO₃^->Cl^->SO₄^2-; and for streams, HCO₃^-> Cl^->NO₃^->SO₄^2-. High levels of NO₃^- and Cl^- in the groundwater wells and stream waters indicate possible contamination by agricultural activities and domestic wastewater (Vengosh and Pankratov, 1998; Lee et al., 2017; Maimoona and Lee, 2019; Cellone et al., 2020).

The Pearson correlation coefficients among the major chemical constituents are presented in Table 4. In the case of the production well waters, Ca²⁺ showed high positive correlations with Mg²⁺ (0.68), K⁺ (0.63), HCO₃^- (0.65), and SO₄^2- (0.77). Because only two sites are located in the carbonated rock area, the overall water chemistry is considered to be less affected by calcite dissolution, while dolomite and gypsum dissolution may probably be based on the high correlations with Mg²⁺ and SO₄^2- (Nazzal et al., 2014). However, for the surrounding groundwater, Ca²⁺ shows high correlations with Mg²⁺ (0.77) and HCO₃^- (0.62), which indicates that the groundwater may be affected by some other factors as well as the water-rock interactions. The stream water showed an interesting correlation pattern. Ca²⁺ is highly correlated with Mg²⁺ (0.91), HCO₃^- (0.75), SO₄^2- (0.66) and Na⁺ is highly correlated with Cl^- (0.75). The latter correlation between Na⁺ and Cl^- was not found in the other two water bodies. Thus, we inferred that the stream water was largely influenced by domestic effluents.

Fig. 3 shows Piper diagrams of major components of the studied water bodies. Most of these (85.8%) are of the Ca²⁺(+Mg²⁺)-HCO₃^- type, a typical type of fresh shallow groundwater (Lee et al., 2013; Fenta et al., 2020). Small portions of the Na⁺-HCO₃^- (8.9%), Ca²⁺-Cl^- (2.9%), and Na⁺-Cl^- (2.4%, only stream water) types were also found. Considering that the study sites are quite far from the seas, the impact of seawater intrusion is not considered; however, the probable influence of domestic wastewater, ion exchange between Ca²⁺ and Na⁺, and/or dissolution from silicate minerals (albite and microcline) may be plausible (Maimoona and Lee, 2019; Patel et al., 2020; Sunkari et al., 2020).

Fig. 3. 
Piper diagram of major ion chemistry for production well, groundwater well, and stream water in the study area.

Fig. 4 shows the Gibbs diagrams of the production well water, surrounding groundwater, and stream water. For Na⁺ and Ca²⁺ ions, it is indicated that most of these (90.5%) were affected by rock weathering (water-rock interaction) while a small proportion (9.5%) was affected by precipitation (rainfall infiltration). In addition, regarding Cl^- and HCO₃^- ions, the majority of the waters (91.1%) were influenced by the water-rock interaction, while 8.9% were influenced by the dilution due to rainfall. Considering the anthropogenic source of Na⁺ and Cl^- from domestic effluents, it is reasonably inferred that the Gibbs diagram cannot appropriately explain the artificial origins of some constituents (Sajil Kumar, 2014; Zhang et al., 2018; Ryu et al., 2019).

Fig. 4. 
Gibbs diagram of production well, groundwater well, and stream water samples. (a) Total dissolved solids (TDS) versus Na⁺/(Na⁺ + Ca²⁺), (b) Total dissolved solids (TDS) versus Cl^-/(Cl^- + HCO₃^-).

The exploratory PCA has yielded four principal components whose eigenvalue is greater than 1 (Fig. 5). These four principal components explain 65.88% of the total variance of the data sets (25.09, 15.12, 14.78, and 10.89%, respectively). Therefore, we assumed that these four components were appropriate for data interpretation without significant information loss. Table 5 shows the loadings of each component on water quality parameters. PC1 has moderate to high positive loadings on Ca²⁺ (0.62), Mg²⁺ (0.65), HCO₃^- (0.63), and SO₄^2- (0.71). Therefore, this component is highly associated with water-rock interaction or weathering of silicate, carbonate, and/or sulfate minerals.

Fig. 5. 
Eigenvalues of principal components and cumulative % variance explained by the components. Four principal components were retained with the criteria of eigenvalue greater than 1. The symbol cross indicates the eigenvalue and bar means the cumulative % variance.

PC2 shows moderate loadings on Na⁺ (0.49), Cl^- (0.66), and NO₃^- (0.69). Thus, it appears to be highly associated with anthropogenic contamination, such as the arrival of domestic wastewater and effects of agricultural activities in upstream areas (Vengosh and Pankratov, 1998; Lee et al., 2017; Maimoona and Lee, 2019; Cellone et al., 2020). Surely these components may be provided by rainfall but the EIA reports confirm the existence of houses and dry fields upstream of some bottled water facilities. PC3 shows a negative loading on Na⁺ (-0.56) with moderate positive loadings for KMnO₄ consumption (0.68) and color (0.55). This component is related to organic contamination rather than the dissolution of silicate minerals (albite and microcline). PC4 has moderate positive loadings on color (0.44) and turbidity (0.67). Therefore, this component is closely associated with organic matter and fine particles suspended in water. High turbidity prevailed in the waters of carbonate rock areas.

From PCA, it is concluded that the water quality of the study areas is affected by weathering of silicate, carbonate, and sulfate minerals, and some anthropogenic pollution from domestic wastewater and agricultural activities.

Table 6 shows the mean water quality values of each cluster grouped by CA using k-means clustering. We predetermined three clusters (groups) based on the water chemistry and have examined the differences among them as well as how well they are classified by CA in terms of water chemistry. Contrary to our expectations, the clusters were not grouped according to the different water bodies. All water bodies are present in each cluster, indicating there are no significant differences in their water chemistry.

Of the 115 data sets, 8.7% are in Cluster 1, and most of these represent production well water in carbonate rock areas. Considering the high concentrations of Ca²⁺, Mg²⁺, HCO₃^-, and SO₄^2-, this cluster indicates the geogenic enrichment from weathering of relevant minerals. According to high values of Na⁺, Cl^-, NO₃^-, and turbidity, Cluster 2 (only 6.9%) is indicative of anthropogenic contamination with organic matter and agricultural fertilizers. Most of the water samples (84.3%) examined are classified into Cluster 3, which comprises many production well, surrounding groundwater, and stream water. This indicates that these water sources are not substantially different from each other in terms of their water chemistry. Furthermore, the water chemistry is the mid point between two end members, Clusters 1 and 2, representing geogenic and anthropogenic origins. Therefore, we inferred that the production wells are affected by anthropogenic contamination even though their water quality meets the Korean drinking water standards.

Fig. 6 shows some implications of anthropogenic effects on the water chemistry of production wells, surrounding groundwater wells, and stream water. Chloride (Cl^-) and nitrate (NO₃^-) are representative constituents indicating the human impact on water chemistry. The cumulative probability plots and their inflection points help determine the water affected by anthropogenic pollution (Park et al., 2005). The inflection points of Cl^- and NO₃^- are 15.84 and 16.21 mg/L, respectively. Using these values, we estimated 12.3% of waters examined were affected by the anthropogenic contamination with regard to Cl^- and NO₃^-.

Fig. 6. 
Cumulative frequency plots of (a) Cl and (b) NO₃ concentrations with inflection points and (c) scatter plot of Cl versus NO₃.

In addition to the water chemistry, we also examined contaminants of concern for production wells, as designated by the Korean Ministry of Environment (Fig. 7). The most frequently occurring contaminants are fluoride (F^-), arsenic (As), nitrate (NO₃^-), and some heavy metals. These are mostly of natural origins, except for high levels of nitrate. In some areas, anthropogenic organic contamination, such as dichloromethane and toluene, was found even though their values did not exceed the Korean drinking water standards. However, this indicates a lack of maintenance of production wells and fuel storage tanks at bottled water manufacturing sites.

Fig. 7. 
Histograms showing fluoride, arsenic, boron, nitrate-N, KMnO₄ consumption, Cu²⁺, color, pH, Zn²⁺, Cl^-, turbidity, SO₄^2-, Al³⁺ of the production well samples.