Groundwater quality assessment: Well position differences
The results of the multivariate discriminant analysis (MDA) testing examining the differences among the six sampling sites are shown in Fig. 2. The results showed statistically significant differences between the six sampling sites (Wilks' lambda: 0.001; F = 26.2; p < 0.0000). All the variables reflecting water quality parameters were used in this model, of which pH, NO3− and Cl− had a statistically significant influence. It was visually apparent that the Z1 well was significantly different to all other wells, while the largest similarity was obtained between the B4 and B5 wells, as well as between the B1 and B2 wells. While wells B1, B2, B3, B4 and B5 were positioned along a line primarily exposed to primary groundwater flow, the more peripheral well, Z1, was likely influenced by secondary groundwater movement (Fig. 1d). The chemical composition of groundwater flows primary (main recharge) and secondary (additional recharge) with respect to the measured parameters in 2017 is shown in Table 1. The intensity of primary and secondary recharge flow influences the accumulation and washing out of anthropogenic pollutants, patterns for which have altered due to recent climate extremes (wet and dry season fluctuations). Well Z1 is in the vicinity of the city and exposed to urban influences, including solid waste dumps, household sullage and/or stagnant surface waters. These inputs may in some settings represent a greater pollution load compared to on-site (point) sources. Domestic waste in this region is often discharged into open ditches or channels and it may represent a significant source of groundwater contamination. Stagnant water pools may directly contaminate groundwater through infiltration or inundation of the sources when rainfall occurs. All of these impacts can contribute to the quality of recharge groundwater (secondary flow direction) outside the sanitary protection zones. Given that previous research has shown chemical pollution in cities and the surrounding areas impacts aquifer water quality (Erah et al. 2002; Ramesh and Elango 2012; Szymanski et al. 2018), it is of future importance to determine the concentration of harmful urban contaminants at the Ivanščak well field.
Descriptive statistics of groundwater physicochemical parameters at public supply wells (Z1, B1, B2, B3, B4, B5) is presented in Table 2. EC is primarily an indicator of environmental influences on groundwater quality, with anthropogenic inputs typically only having a minor contribution. The results indicated that mean EC values varied from 490 µS/cm (B3) to 576 µS/cm (Z1). Overall, groundwater EC values were approximately related to their position in the landscape, in terms of both their average values and their variability (minimum and maximum values). The comparative studies (Oyem et al 2014) of hydrochemical data showed that EC range of 825–1416 µS/cm. The mean values recorded for EC were 1108 µS/cm. Mondal et al (2008) indicates that EC varies from 605 to 5770 µS/cm in groundwater resources. Hence, compared to previous researches, our results showed the average electrical conductivity of groundwater. According to a previous analysis of groundwater quality (by the public supply company, “Koprivnica waters—public water supply”; https://www.kcvode.hr) at the study area, it was determined that certain indicators (e.g., nitrates or EC) were present at higher values in the secondary flow of recharge groundwater than in public supply wells (Table 1). This supports the hypothesis that well Z1 was recharged mainly via the secondary flow pathway.
The average pH values for the studied groundwaters ranged from 6.87 (B3) to 7.07 (Z1), making them suitable for human use (pH from 6.5 to 8.5) and within the limits allowed by the Croatian Drinking water standards (National Newspaper 125 (2013) and WHO recommendations (2011). Again, variations in the pH values were likely due to their relative position, being influenced by urbanization (Z1) and agricultural (B3) inputs. The minimum pH value of 6.54 (at B3) almost reached the lower permissible limit, indicating a deterioration in water quality compared to previous findings, which noted pH values typically > 7.0 (Nemčić-Jurec et al. 2017). Although pH is predominantly influenced by natural conditions, anthropogenic modifications may impact upon future projections. Indeed, previous research has demonstrated the anthropogenic influence on pH values at this study site (Duarte et al. 2013).
Permanganate oxidation (KMnO4) concentrations ranged from 0.43 mg/L (B5) to 0.50 mg/L (B2) and showed a similar trend across all wells with minimal spatial variation. The values were substantially lower than the allowable limits (National Newspaper 125 2013, WHO 2011), suggesting a low organic content in the soil and groundwater. The possible contamination identified by other researchers (Jurado et al. 2012; Kauffman and Chapelle 2010) was found to be either absent or insignificant at this site.
Previous studies into the groundwater quality at the Ivanščak well field identified the presence of elevated nitrate concentrations in the groundwater sampled from shallow wells (Nemčić-Jurec and Jazbec 2015). Our findings support this previous work, with measured average nitrate values varying from 28.2 mg/L (Z1) to 32.7 mg/L (B5). All groundwater samples had nitrate concentrations below the upper permissible limit for drinking water determined by national (National newspaper 125, 2013) or international legislation (WHO 2011). However, the maximum nitrate concentration (42.2 mg/l in the B1 well) exceeded recommended limit value of 25 mg/L (European Union (EU) Council Directive (1991) and was the highest nitrate concentration across all other areas in the county (Nemčić-Jurec et al. 2017). Considering that nitrate is predominantly of anthropogenic origin, the influence of the well position was expected to be a critical determinant for this elevated measurement. The position of the B1 well was strongly influenced by urbanization, industry and agriculture, which collectively contributed to the highest nitrate concentration (Fig. 1d). The variability in nitrate concentrations at B1 was likely a result of the well recharging from two different directions (primary and secondary). Indeed, it would appear that B1 was the well which was most exposed to the combined influence of both groundwater flow directions. Agriculture predominantly contributes to elevated nitrate in the primary water flow direction, with animal feces, as well as the application of manure or fertilizers, representing a significant diffuse source of nitrate contamination over a long period of time. Intensive livestock farming may have a significant impact on nitrate contamination, potentially contributing to a more widespread aquifer contamination at the study site. Urbanization and industry (solid waste dumps, household sullage, stagnant surface waters, stagnant water pools, domestic waste, etc.) affects water quality in the secondary water flow direction (depending on environmental factors) (Scanion et al. 2005; Rawat and Tripathi 2015). Finally, nitrate concentrations at B1 will also be determined by the antecedent ratio of water from both primary and secondary directions, coupled with topographical variations.
In the groundwater samples of the study area, the average chloride concentrations were 23.9 mg/L (Z1), 25.2 mg/L (B1), 19.4 mg/L (B2), 17.3 mg/L (B3), 17.0 mg/L (B4) and 20.0 mg/L (B5). The highest concentrations of chloride in the Z1 and B1 wells were likely related to their position. As outlined above, these wells were subject to pollution pathways from municipal (human sewage, roads, waste) and industrial waste or wastewater via the secondary flow direction (KPI 1 and P 12 piezometers; Table 1). The increase in chloride concentration (exceed maximal levels prescribed by WHO of 250 mg/l) due to discharge of municipal and industrial waste has been reported in earlier studies (Soomro et al. 2014; Trivedi et al. 2010; Eriksson et al. 2007; Liu et al. 2008). Conversely, the impact of agriculture may also contribute to elevated chloride levels via the primary flow direction. However, the measured chloride concentrations were far below regulatory limits, suggesting that anthropogenic impacts were not significant. Previous research (Negrel et al. 2003; Scanion, et al. 2005; Sakson and Brzezinska 2018) has identified that alluvial aquifers are particularly vulnerable to anthropogenically induced changes in major ions, such as chloride, especially in agricultural areas. Considering the alluvial nature of the Podravina region, chloride could easily be introduced into the groundwater, contributing to the variability of the pollutants in the aquifer. This may well be the reason for the large range between minimum and maximum chloride values in our study area (Table 2).
Risk assessment of groundwater contamination in public supply wells
The PCA results, including factor loading values and variance percentage, for the 5 wells are shown in Fig. 3. Eigenvalues provide a measure of the significance of individual factors, i.e., the factor with highest eigenvalue is the most important (Jacintha et al. 2017). The two primary factors extracted, Factors 1 and 2 (F1 and F2), explained 32.8% and 16.2% of the total variance, respectively. F1 correlated predominantly with nitrate concentrations suggesting that the groundwater quality was mainly controlled by this parameter. The high NO3− values highlight the potential role of manure or fertilizer application in agricultural areas or of urbanization (solid waste, sullage, surface water, broken sewage systems) upon aquifer contamination, a point which has been shown previously (Rawat and Tripathi 2015, Nemčić-Jurec et al. 2015). These observations suggest that anthropogenic sources contribute to the elevated NO3− concentrations in the study area. Low (negative) loadings of EC and chloride with F1, confirms that these parameters do not contribute to the variation in nitrate values. The second factor (F2) correlates predominantly with KMnO4 and pH values, indicating a second influence over the groundwater quality related to these parameters. As the degradation of organic material results in a pH decrease due to the formation of organic acids; F2 is may well be a degradation factor. Low (negative) loading of chloride and EC with F2, also suggests that their values were independent of pH or KMnO4. The low loading of EC and chloride would most likely be attributable to the natural solubility of aquifer minerals or natural sources of chloride.
The PCA results revealed that differences in groundwater quality were not restricted to any one or two measured parameters, indicative of both anthropogenic influences and natural mineralization. PCA results confirmed the descriptive statistics that nitrates and pH represent the greatest risk (as concentrations reach regulatory limits) for the deterioration of groundwater quality in public supply wells.
In order to cover all the geochemical processes affecting groundwater quality, groundwater monitoring was conducted for 10 years and the risk of contamination was assessed. The data collected from the study area and the results of current trends are shown in Fig. 4. Values used for the linear trend analysis represent mean concentrations measured monthly in all six wells for each of the water quality parameters.
The study area time series showed an increasing trend in nitrate, decreasing pH and permanganate oxidation, and steadily average concentrations for EC and chloride. Based on the theoretical assumptions and the temporal analysis as well as the variations and parameter trends (Hill and Neal 1997), the groundwater quality can be considered as predominantly natural (EC, pH, permanganate oxidation) and anthropogenic (nitrate, chloride) (Soomro et al. 2014; Nemčić-Jurec and Jazbec 2015; Nemčić-Jurec et al. 2013). An increasing trend of nitrate usage across the study area is a common response to increased human activities, resulting in human alteration of biogeochemical cycles, which has been demonstrated in other areas (Elisante and Muzuka 2017; Duarte et al. 2013). NO3− is very soluble in water and it is easily leached from soils into the groundwater. The elevated levels of nitrate in the groundwater may well have led to soil acidification. According to Elisante and Muzuka (2017), a significant negative correlation between the nitrate concentration and pH indicates the mineralization and nitrification of the organic nitrogen from animal manure. Our results show a negative correlation between pH (decreasing trend) and NO3− (increasing trend) over time. Given that the wells in the study area have been utilized for agricultural purposes for over 45 years, such results were largely expected. In terms of the study area, nitrate concentrations were lower (24 mg/L; KPI 3 and KPI 4 piezometers) from the primary recharge direction (west–northwest), when compared to average values at the public supply wells (sampling sites). Conversely, the groundwater from the secondary flow direction (southwest) nitrate concentrations were higher (39–51 mg/L; P 12 and KPI 1 piezometers) (Table 1). Consequently, the average nitrate concentration in the public supply wells was the result of mixing groundwaters from both directions. Considering that the secondary flow comes from the direction of Koprivnica city, the long-term effect of urbanization outside the sanitary protection zones is visually apparent (Fig. 1c).
The decreasing trend in permanganate oxidation (Fig. 4) during the study period may be the result of intensive agricultural activity, removal of crops and retention of soil organic matter and organic acids. The increasing use of mineral fertilizers instead of organic fertilizer (manure) at the study site, may also contribute to the reduction of organic matter in groundwater. Given that the concentrations of organic matter were low in all wells, even if there was an anthropogenic impact, they would appear to be insignificant toward the KMnO4 parameter.
Rainfall and sampling time influence on groundwater quality
Our results showed that a marked increase of nitrate, coupled with a decrease in pH and KMnO4 occurred in 2014 and 2015. In September 2014, the amount of rainfall was 38% (~ 1,300 mm/year) above the rainfall annual average in the study area (Crometeo 2014). It is conceivable that the more intense trends during this period were associated with the enhanced infiltration of contaminants (esp. nitrate) into the groundwater which consequently decreased pH values. Additional variability in pH values might well occur during perennial fluctuations in the climate. For example, it has been widely demonstrated that extremely low groundwater levels following dry periods promotes sulfide oxidation which increases the acidification of associated groundwaters (Knuttson 1994). Vidyalakshm et al. (2013) have shown that during heavy rainfall events, nutrients enter aquatic ecosystems more quickly, posing a higher risk for drinking water quality.
However, individual indicators behave differently during rainfall. The significantly decrease in KMnO4 (organic matter) after elevated rainfall may also be indicative of aquifer dilution processes. The influence of elevated rainfall and associated flooding was evident during 2014–2015, when KMnO4 concentrations were at their lowest. Previous research has shown that a decrease in organic matter concentrations could be accelerated during heavy rainfalls (Oyem et al. 2014; Sakson and Brzezinska 2018; Mondal et al. 2008).
Groundwater EC and chloride concentrations showed inconsistent trends over the study period (Fig. 4). Chloride concentrations in the groundwater suggested the presence of organic waste, which could have contributed to the variability observed in chloride concentrations during rainfall events. As in previous studies (Pu et al. 2011; Mondal et al. 2008), it is obvious that the low concentrations of chloride recorded during 2014 and 2015 were connected to the heavy rainfall, causing dilution of the groundwater during flooding.
Diffuse sources were the likely major contributor to nitrate in groundwaters (fertilizers from agriculture and urbanization), which were leached from the soil during periods of high rainfall. Intuitively, the behavior of parameters such as nitrate depends on their solubility and leaching rate during precipitation, coupled with the amount of pollutant which is stored in the soil within the aquifer catchment. Therefore, establishing direct links between rainfall and parameters is problematic, and as such, no correlation was observed in our data. Furthermore, studies into the connections between topography and climate upon groundwater fluxes have shown that groundwater configuration is controlled by geology and climate, which influence the residence time and location of subsurface contaminants (Condon and Maxwell 2015).
Investigations were also made into the time dependence (month and year of sampling) on the groundwater quality, analyzed by multiple regression analysis (MRA). The correlations among the groundwater quality and the predictor variable sampling time are shown in Table 3. The predictor variable had a significant influence on the parameters: electrical conductivity (β = 0.33) and chloride concentrations (β = 0.23). Based on the values of the beta coefficients and their levels of significance, it would appear that sampling time had a strong influence on the permanganate oxidation levels (β = 0.17), whereas the values for nitrate (β = 0.13) and pH (β = 0.11) were less dependent on sampling time. This supports the previous assertion that nitrate concentrations and pH value were more dependent on environmental factors such as pollution sources and land use, pollution distance and rainfall. MRA confirmed that the duration of monitoring was a very important factor in the assessment of environmental impacts (e.g., rainfall) on overall groundwater quality.
Water quality prediction
An evaluation of aquifer water quality is critical to ensure the provision of safe drinking water and has a direct impact on public health and the environment. Therefore, it is important to assess and predict the contamination of groundwater quality in the future using machine learning algorithms (artificial neural network, ANN) and similar methods. Some studies have investigated the performance of artificial intelligence techniques for predicting water quality components. For example, it was found that using a group method of data handling (GMDH) and support vector machine (SVM) model were more reliable in comparison with ANN (Haghiabi et al. 2018; Condon and Maxwell 2015). It is likely that similar models will be the subject of further research with the aim of predicting water quality across sites such as the one studied here.