Multivariate analysis of physicochemical parameters of water produced by a commercial atmospheric water generator (AWG) in an industrial area in South Africa

Water scarcity has become the world's third most serious threat, occurring when water usage is outstripping supply and restricted water resources are nearing or beyond sustainability limitations (UNICEF 2021). Water is a basic, essential, and vital resource for human survival, with more people consuming it than the sum of all foods combined (Badaró et al. 2021). The global population has expanded by 300% during the 20th and 21st centuries, while water consumption has climbed by 600% (Bagheri 2018). Furthermore, more than 2 billion individuals across the globe are unable to obtain safe drinking water, which is expected to rise with population growth and the effects of climate change (Moghimi et al. 2021).

Freshwater sources that are currently available, such as rivers, lakes, and groundwater, are suffering from man-made pollution and climatic water loss (Zhou et al. 2020). These contaminants from anthropogenic sources entering the water system, climate change and overdraft of pre-existing reservoirs are all putting a damper on the water supply, thus causing the scarcity of water (Bagheri 2018).

By 2025, about 48 countries will be facing water stress or scarcity (Bagheri 2018). In addition, by 2050, freshwater consumption will have increased by roughly a third (Moghimi et al. 2021). Furthermore, an estimated 3.6 billion people, or nearly half of the world's population, live in locations where water is potentially scarce for at least 1 month every year (Moghimi et al. 2021). Water shortage is a major danger to food security, a hindrance to economic growth, and exacerbates socio-political conflicts (Distefano & Kelly 2017).

This is clear evidence of the pressing and essential need for additional and new water sources, especially for drinking, to be provided to regions facing prolonged water shortages and specific short-term water crises (Kaplan et al. 2023). Therefore, it is vital to investigate innovative ways of water harvesting, supply, and storage to overcome expected challenges, such as overpopulation, drought, and floods.

Research has been done to see whether there are any new ways to exploit possible freshwater supplies, such as the purification of seawater or wastewater; many water purification technologies have been developed, including membrane filtration, filtering, multistage flash distillation, and the treatment of water through solar (Zhou et al. 2020). However, due to their reliance on sources of pure water, this advancement in technology can only be practicable in coastline regions, making it impossible for non-coastal regions (Zhou et al. 2020).

Generating water from the atmosphere has been considered as another alternative source for drinking water (Inbar et al. 2020; Kaplan et al. 2023). Moist air, which is available irrespective of geographic or hydrological circumstances, has been gaining popularity as a potential water supply (Zhou et al. 2020). The Earth's atmosphere holds an enormous and replenishable reservoir of water, with around 12,900 billion tons of fresh water (Li et al. 2018). Water, either in the form of liquid raindrops or vapour, is held in the Earth's atmosphere, which accounts for at least 10% of water bodies and supplies water in the amount of 50,000 km³. Furthermore, the cycle of water in nature ensures a long-term supply of water (Bagheri 2018).

As a result, an atmospheric water generator (AWG) can become a feasible option for meeting some of the world's freshwater needs (Bagheri 2018). The AWG systems can produce yields ranging from about 15 to over 10,000 l/day (Khalil et al. 2016; Ansari et al. 2022). The quality of the water produced by an AWG has been reported in several studies on inorganic characteristics (Beysens et al. 2017; Hong et al. 2019) and microbial parameters and contamination (Hu et al. 2018; Kaplan et al. 2023).

However, studies on water quality produced by a commercial AWG are limited. Therefore, it is essential to understand the physicochemical parameters and concentrations of potentially toxic elements (PTEs) in the produced atmospheric water and their distribution if present. Utilising multivariate statistical techniques is essential for effectively analysing extensive water quality datasets while minimising the loss of crucial information (Shrestha & Kazama 2007; Juahir et al. 2011; Samson & Elangovan 2017).

Multivariate statistical techniques like cluster analysis (CA), principal component analysis (PCA), factor analysis (FA), and discriminant analysis can effectively interpret complex data matrices, thus enabling a better understanding of water quality and other environmental systems by helping identify potential factors and sources. As a result, they serve as powerful tools for promptly addressing pollution problems (Reghunath et al. 2002; Simeonov et al. 2003, 2004; Ravikumar & Somashekar 2017).

Furthermore, multivariate statistical analysis has been utilized in various water quality assessments, such as that by Barakat et al. (2016), which conducted a study to assess surface water quality parameters, identify contamination sources, and evaluate their impact using correlation matrix, multivariate PCA, and CA techniques. Zhao et al. (2012) employed PCA to effectively assess water quality and identify sources of pollution in a lake. Furthermore, Nnorom et al. (2019) investigated sources of PTEs in springs, streams, boreholes, and hand-dug wells using multivariate statistical analysis.

Therefore, this study aimed to determine the quality of water produced by a commercial AWG located in the Ga-Rankuwa industrial area using multivariate statistical analysis. The objectives of the study are to assess whether the atmospheric water produced by an AWG complies with the water quality standards set by the South African Water Quality Guideline (SAWQG) and the World Health Organization (WHO) drinking water standard, to evaluate the physicochemical parameters and concentrations of PTEs in the produced atmospheric water and to identify potential factors (sources) of pollutants if present. To the best of the author's knowledge, this is the first study on the quality of water produced by a commercial AWG situated in an industrial area in Africa.

According to Raveesh et al. (2021), AWG technologies generate water from wet air, which is extensively and readily accessible in the atmosphere. The moisture in the atmosphere serves as a supply of fresh water that can be replenished, which can be potentially extracted regardless of geographical or hydrologic circumstances (Zhou et al. 2020).

For the air–water extraction process to take place in an AWG condenser unit, moist air is sucked into the AWG condenser, a compressor circulates the refrigerant through a condenser, and then the surrounding air is cooled by an evaporator coil (Figure 2). However, the controlled-speed fan pushes filtered air over the coil as warm air; this lowers the air temperature to its dew point, causing water to condense (Recor 2022). The water produced can be used to replenish fresh water and can be further purified and filtered to make it safe for human consumption (Raveesh et al. 2021).

Sampling of the atmospheric water was conducted over 12 months from November 2022 to October 2023 to account for all four seasons (summer, autumn, winter, and spring). A total of 12 water samples were systematically collected from the AWG water tank, with one sample obtained each month. Each sample was contained in a 500 ml polyethene bottle fitted with a secure lid to maintain sample integrity for subsequent analysis (Figure 3). The bottles were handled with the utmost sterility and caution to prevent external contamination during collection and transportation to the Aqua Air Africa (Pty) Ltd laboratory.

Each atmospheric water sample was accompanied by a blank sample containing 500 ml of distilled water in a tightly sealed bottle. The blank samples were treated and tested in the same way as the AWG water samples.

A water analysis was performed to determine the untreated AWG water quality and identify any potentially harmful elements. Once the water bottle samples arrived at the laboratory, chemical analysis of major elements (calcium, potassium sulphate, magnesium, chlorine free, chloride, and alkalinity) and trace elements (aluminum, copper, iron, manganese, nickel, silica, zinc, fluoride, nitrate, andammonia), was conducted using the HANNA H183399 Multiparameter Photometer (HANNA, South Africa).

Physical water parameters, pH, electrical conductivity (EC), and total dissolved solids (TDS) were measured using the pH/ conductivity/TDS bench meter (HANNA, South Africa). The turbidity was conducted using the HANNA H198703 Turbidimeter (HANNA, South Africa). The colour of the atmospheric water was determined using the HANNA Multiparameter Photometer. All the analyses were conducted according to the manufacturer's protocol (HANNA manual). The atmospheric water samples were further compared to the SAWQG standards and the WHO drinking water standard to assess if the water samples meet the criteria.

Multivariate statistical analysis was conducted using IBM^® SPSS^® Statistics 29 to examine various physicochemical parameters found in atmospheric water, including calcium, potassium, sulphate, magnesium, chlorine free, chloride, alkalinity, aluminium, copper, iron, manganese, nickel, silica, zinc, fluoride, nitrate, ammonia, pH, EC, TDS, turbidity, and colour of water. The multivariate statistical methods used in this study were univariate analysis (descriptive statistical analysis) and bivariate analysis (correlation analysis), CA, FA, which used the PCA extraction method.

Correlation analysis is a bivariate statistical technique utilized to quantify the association between two variables (Venkatramanan et al. 2013). CA is a useful classification tool that utilizes the Ward algorithmic method, which maximizes the variance between groups and minimizes it within the same group (Núñez-Alonso et al. 2019). It arranges the objects so that each object in the cluster matches the others based on a predefined criterion (Gulgundi & Shetty 2018). The hierarchical agglomerative clustering approach, represented by a dendrogram, was used in this study (McKenna 2003). The dendrogram provides a visual overview of the clustering process, illustrating the relationships between groups and their proximity while reducing the original data's dimensionality. In this study, CA was utilised to classify the water samples collected across different months according to their chemical composition.

FA is regularly utilized as a statistical method in studies related to hydrochemistry (Venkatramanan et al. 2013). FA is a statistical method utilized to determine the relationships among various observable quantitative variables and represent them in relation to a limited number of underlying factors (Suk & Lee 1999). In this study, FA was performed to identify the latent factors of the elements detected in the untreated AWG water through a rotated component matrix, describing the entire dataset (Hou et al. 2014; Guo et al. 2017).

Water chemistry is influenced by geological settings, climate, and human activities, and these varying factors interact in different environments, contributing to the complexity of complex water chemistry (Li et al. 2017; Towfiqul Islam et al. 2017; Wu et al. 2017). Table 1 presents the statistical summary of the physicochemical parameters that were used to examine the untreated AWG water quality in this study area.

The water quality analysis of atmospheric water was carried out to determine the different physicochemical parameters, namely, pH, EC, TDS, turbidity, ammonia, alkalinity, aluminum, calcium, chlorine free, colour of water, copper, chloride, fluoride, iron, manganese, magnesium, nickel, nitrate, potassium, silica, sulphate, and zinc. The suitability of atmospheric water for drinking purposes was evaluated based on the standards for drinking water set by the SAWQG (South African National Standard (SANS) 2015) and WHO drinking water standards (World Health Organization (WHO) 2022) (Table 1).

The pH of the atmospheric water ranged between 5.70 and 8.93, which corresponded with the SAWQG but exceeded the 8.5 pH limits for WHO drinking water in 1 out of 3 AWG water samples (Table 1). However, the average pH was 6.80, thus within both the SAWQG and WHO drinking water limits. These readings were similar to those recorded by Inbar et al. (2020), who found that the pH of atmospheric water ranges between 6.5 and 7.9. They also determined that pH values of atmospheric water can vary significantly across different sites due to the varying sources of ions (Inbar et al. 2020). Another study by Beysens (2018) observed that the pH level of dew water tends to stay relatively close to neutral (pH 7).

EC recordings ranged between 63.03 and 206.90 (μs/cm) with a mean value of 123 (μs/cm), which falls within the WHO drinking water limits but exceeded the SAWQG permissible limits of 170 μs/cm; these high levels were observed in 3 out of 12 AWG water samples (Table 1). It is worth noting that long-term use of water with high EC levels can lead to gastrointestinal irritation in humans (Ramesh & Elango 2012). As a result, filtration and purification treatment should be carried out before the water is consumed.

The TDS concentration in atmospheric water ranged between 30.92 and 103.40 mg/l, with an average of 60.95 mg/l. All the TDS concentrations were well below both the SAWQG and WHO drinking water permissible limits. The TDS readings, however, indicated that the atmospheric water within the study area was suitable for human consumption and irrigation (Rusydi 2018).

The atmospheric water turbidity concentrations were well below the SAWQG and WHO drinking water permissible limits, with ranges from 0.45 to 2.60 NTU and a mean of 1.28 NTU (Table 1). The colour of water concentration ranged between 0 and 107 with a mean of 46.50. This exceeded the SAWQG permissible limits in 9 out of 12 AWG water samples. Water colouration, according to Amfo-Otu et al. (2014), has no specific health effects; it is frequently associated with customer complaints and can cause aesthetic concerns. Therefore, the higher the levels of colouration in the atmospheric water, the more unappealing it is for some users to consume or use for domestic purposes; hence, filtration and purification treatment should be performed to remove colouration before consumption.

Ammonia concentration ranged from 1.42 to 92.20 mg/l, with a mean of 23.78 mg/l. This exceeded the SAWQG and WHO permissible limits in 11 out of 12 samples and, thus, may cause the water to be disqualified for drinking without proper post-treatment (filtration and purification treatment). However, ammonia in drinking water is not considered a direct health risk but rather aesthetic (SANS 2015). Studies conducted by Muselli et al. (2006), Inbar et al. (2021), and Kaplan et al. (2023) demonstrated that ammonia is prevalent in AWG water and dew water due to its common atmospheric origin. These may originate from agricultural, industrial, and vehicular sources. This implies that extracting gaseous ammonia is more effective from larger air volumes, as it passes through the condenser and into the AWG water, rather than just dissolving in water (Kaplan et al. 2023).

The following major ions were present at low concentrations in the atmospheric water: calcium, magnesium, and potassium. They were found to be compliant with both the SAWQG and WHO permissible limits (Table 1). Calcium ranged between 0 and 133 mg/l, with an average of 52.17 mg/l, while magnesium ranged between 5 and 40 mg/l, with an average of 10.67 mg/l, and potassium ranged between 0 and 1.30 mg/l, with an average of 0.58 mg/l. A deficiency of calcium and magnesium in drinking water can result in health issues such as tooth loss, rickets, and cardiac infarction (Inbar et al. 2020). The low concentrations were supported by Kaplan et al. (2023), who indicated that AWG water is typically characterized by a low concentration of minerals; therefore, mineral addition is important as a post-treatment.

The chloride was compliant with both the SAWQG and WHO permissible limits, ranging from 0 to 6.10 mg/l with a mean of 2.57 mg/l. It is noted that high levels of chloride in drinking water can lead to a salty taste and have a laxative effect on some individuals (Ramesh & Elango 2012). Both nitrate and sulphate were compliant with both the SAWQG and WHO permissible limits. Sulphate ranged from 17 to 74 mg/l with a mean of 37.42 mg/l, and nitrate ranged from 0 to 4.90 with a mean of 1.45 mg/l. A high concentration of nitrate in atmospheric water is not ideal, as this could result in gastric cancer, goiter, hypertension, and birth malformations (Ramesh & Elango 2012). The concentration of fluoride in the atmospheric water ranged from 0 to 0.38 mg/l and had a mean of 0.13 mg/l. All the samples fell within the desirable ranges as per SAWQG and WHO (Table 1). High levels of fluoride can lead to dental fluorosis, causing teeth discolouration and potential skeletal bone and kidney issues (Shaji et al. 2007). Furthermore, a deficiency of fluoride in drinking water can lead to tooth decay (Ramesh & Elango 2012).

Atmospheric water samples were monitored for potential toxic elements such as aluminium, nickel, copper, iron, manganese, silica, and zinc. Concentrations of aluminium, zinc, and copper were well below the permissible limits for both the SAWQG and WHO (Table 1). Other studies also revealed low copper levels in dew water which were consistently beneath drinking water standards by at least tenfold (Inbar et al. 2020). Dew water studies in an urban area near Paris, France, and in an urban, semiarid area in Chile detected median copper values of 0.02 and 0.04 mg/l, respectively, similar to this study (Beysens et al. 2017; Carvajal et al. 2018).

Nickel was the only metal that exceeded both the SAWQG and WHO permissible limits in 1 out of 12 samples. Nickel concentrations ranged from 0 to 0.22 mg/l, with a mean of 0.04 mg/l and a standard deviation of 0.06 mg/l (Table 1). Prolonged exposure to high levels of nickel in drinking water can be detrimental to health, potentially leading to lung cancer and intestinal erosion (Rosborg & Kozisek 2016). Thus, it is essential to further monitor the quantities of nickel in atmospheric water and conduct filtration and purification treatment before the atmospheric water is consumed. Studies conducted by Inbar et al. (2020) and Kaplan et al. (2023) also found high concentrations of nickel in atmospheric water that exceed the Israel and WHO drinking water standards. Both studies correlated the high nickel concentration with transportation or heavy traffic (Inbar et al. 2020; Kaplan et al. 2023).

The concentrations of iron, manganese, and silica were all below the permissible limits for both the SAWQG and WHO (Table 1). Iron ranged from 0 to 0.12 mg/l, with an average of 0.04 mg/l. Furthermore, manganese ranged between 0.01 and 0.30 mg/l, with an average of 0.18 mg/l. Silica ranged from 0.09 to 0.42 mg/l, with a mean of 0.025 mg/l. However, there are no guideline values to determine whether the limits were permissible or not.

Evaluating the correlation among different water quality variables provides valuable insights into the hydrochemical processes that influence chemical characteristics (Singh et al. 2017). Correlation analysis was performed using Pearson's correlation to describe the relationship between atmospheric water's physicochemical parameters, and the results are shown in Table 2. This study utilized Wang's (2018) classification method to analyze the correlation between atmospheric parameters. In this classification, r < 0.3 was considered of no correlation; 0.3 < r < 0.5 as ‘less correlation’; 0.5 < r < 0.8 as moderate correlation; and r > 0.8 as of high/strong correlation (Wang 2018).

Table 2

Correlation coefficient of the atmospheric water's physicochemical parameters

.	PH .	EC (⁠s/cm) .	TDS (mg/l) .	Turbidity (NTU) .	Ammonia .	Alkalinity .	Aluminium .	Calcium .	Chlorine free .	Colour of water .	Copper .	Chloride .	Fluoride .	Iron .	Manganese .	Magnesium .	Nickel .	Nitrate .	Potassium .	Silica .	Sulphate .	Zinc .
PH	1
EC (μs/cm)	0.079	1
TDS (mg/l)	0.091	0.997a	1
Turbidity (NTU)	0.335	0.541	0.521	1
Ammonia	− 0.367	0.250	0.229	−0.232	1
Alkalinity	−0.063	0.337	0.346	0.354	−0.194	1
Aluminium	0.397	−0.366	−0.350	−0.169	−0.489	−0.286	1
Calcium	−0.539	0.095	0.064	0.235	0.175	0.100	− 0.578b	1
Chlorine free	0.143	0.847a	0.842a	0.795a	−0.056	0.257	−0.288	0.208	1
Colour of water	0.286	0.830a	0.830a	0.821a	−0.219	0.478	−0.177	0.039	0.899a	1
Copper	−0.186	0.472	0.458	0.079	0.469	−0.459	0.095	0.126	0.409	0.185	1
Chloride	0.175	0.797a	0.790a	0.582b	−0.047	0.458	−0.183	0.078	0.814a	0.831a	0.300	1
Fluoride	0.279	−0.414	−0.432	0.020	0.209	−0.032	−0.044	0.163	−0.411	−0.362	−0.237	−0.330	1
Iron	−0.358	0.154	0.141	0.223	0.299	0.362	−0.095	0.215	0.199	0.194	0.273	0.433	0.137	1
Manganese	0.182	0.693b	0.699b	0.443	−0.079	0.146	0.001	−0.307	0.684b	0.781a	0.325	0.716a	− 0.631b	0.195	1
Magnesium	−0.168	−0.163	−0.170	−0.184	−0.043	−0.229	0.065	−0.200	−0.241	−0.030	−0.045	−0.196	−0.231	−0.040	0.379	1
Nickel	0.008	−0.026	−0.028	−0.036	−0.223	−0.138	0.113	−0.293	−0.088	0.163	−0.112	−0.085	−0.358	−0.201	0.511	0.953a	1
Nitrate	0.084	−0.491	−0.484	−0.320	−0.357	−0.331	0.861a	−0.373	−0.332	−0.388	0.149	−0.187	−0.083	0.100	−0.166	−0.117	−0.156	1
Potassium	−0.330	0.071	0.073	0.182	0.013	0.543	−0.537	0.224	0.219	0.215	−0.364	0.377	−0.117	0.545	0.214	0.018	−0.039	−0.280	1
Silica	−0.347	0.517	0.513	0.177	0.032	0.048	−0.119	0.161	0.425	0.518	0.455	0.441	− 0.580b	0.239	0.706b	0.616b	0.631b	−0.240	0.079	1
Sulphate	−0.002	0.914a	0.910a	0.390	0.542	0.073	−0.352	0.053	0.708a	0.637b	0.659b	0.626b	−0.325	0.227	0.653b	−0.047	0.006	−0.456	−0.021	0.518	1
Zinc	0.444	0.491	0.493	0.674b	0.130	−0.117	0.045	−0.235	0.645b	0.561	0.384	0.354	−0.081	0.122	0.526	−0.143	−0.042	−0.103	−0.063	0.042	0.589b	1

.	PH .	EC (⁠s/cm) .	TDS (mg/l) .	Turbidity (NTU) .	Ammonia .	Alkalinity .	Aluminium .	Calcium .	Chlorine free .	Colour of water .	Copper .	Chloride .	Fluoride .	Iron .	Manganese .	Magnesium .	Nickel .	Nitrate .	Potassium .	Silica .	Sulphate .	Zinc .
PH	1
EC (μs/cm)	0.079	1
TDS (mg/l)	0.091	0.997a	1
Turbidity (NTU)	0.335	0.541	0.521	1
Ammonia	− 0.367	0.250	0.229	−0.232	1
Alkalinity	−0.063	0.337	0.346	0.354	−0.194	1
Aluminium	0.397	−0.366	−0.350	−0.169	−0.489	−0.286	1
Calcium	−0.539	0.095	0.064	0.235	0.175	0.100	− 0.578b	1
Chlorine free	0.143	0.847a	0.842a	0.795a	−0.056	0.257	−0.288	0.208	1
Colour of water	0.286	0.830a	0.830a	0.821a	−0.219	0.478	−0.177	0.039	0.899a	1
Copper	−0.186	0.472	0.458	0.079	0.469	−0.459	0.095	0.126	0.409	0.185	1
Chloride	0.175	0.797a	0.790a	0.582b	−0.047	0.458	−0.183	0.078	0.814a	0.831a	0.300	1
Fluoride	0.279	−0.414	−0.432	0.020	0.209	−0.032	−0.044	0.163	−0.411	−0.362	−0.237	−0.330	1
Iron	−0.358	0.154	0.141	0.223	0.299	0.362	−0.095	0.215	0.199	0.194	0.273	0.433	0.137	1
Manganese	0.182	0.693b	0.699b	0.443	−0.079	0.146	0.001	−0.307	0.684b	0.781a	0.325	0.716a	− 0.631b	0.195	1
Magnesium	−0.168	−0.163	−0.170	−0.184	−0.043	−0.229	0.065	−0.200	−0.241	−0.030	−0.045	−0.196	−0.231	−0.040	0.379	1
Nickel	0.008	−0.026	−0.028	−0.036	−0.223	−0.138	0.113	−0.293	−0.088	0.163	−0.112	−0.085	−0.358	−0.201	0.511	0.953a	1
Nitrate	0.084	−0.491	−0.484	−0.320	−0.357	−0.331	0.861a	−0.373	−0.332	−0.388	0.149	−0.187	−0.083	0.100	−0.166	−0.117	−0.156	1
Potassium	−0.330	0.071	0.073	0.182	0.013	0.543	−0.537	0.224	0.219	0.215	−0.364	0.377	−0.117	0.545	0.214	0.018	−0.039	−0.280	1
Silica	−0.347	0.517	0.513	0.177	0.032	0.048	−0.119	0.161	0.425	0.518	0.455	0.441	− 0.580b	0.239	0.706b	0.616b	0.631b	−0.240	0.079	1
Sulphate	−0.002	0.914a	0.910a	0.390	0.542	0.073	−0.352	0.053	0.708a	0.637b	0.659b	0.626b	−0.325	0.227	0.653b	−0.047	0.006	−0.456	−0.021	0.518	1
Zinc	0.444	0.491	0.493	0.674b	0.130	−0.117	0.045	−0.235	0.645b	0.561	0.384	0.354	−0.081	0.122	0.526	−0.143	−0.042	−0.103	−0.063	0.042	0.589b	1

^aCorrelation is significant at the 0.01 level (2-tailed).

^bCorrelation is significant at the 0.05 level (2-tailed).

View Large

The results indicate a strong correlation (r = 0.997) between TDS and EC (Table 2). Both EC and TDS measure water salinity and are contributed by all soluble ions (Li et al. 2019). Both the EC and TDS displayed a strong correlation with sulphate, chlorine, the colour of water, and a moderate correlation with chloride and manganese (Table 2). Almost all trace elements had no significant correlation with TDS and EC except for manganese, which had a moderate correlation (r = 0.693 and r = 0.699, respectively).

TDS and EC measure the combined content of all inorganic and organic substances in a liquid in molecular, ionized, or micro-granular suspended form (Ezekwe et al. 2012). Elevated levels of dissolved constituents in water can indicate the presence of pollutants, reduced oxygen levels, and the existence of contaminant plumes. Therefore, the strong correlation between TDS and manganese, as well as EC and manganese, suggests that constant monitoring of manganese in the atmospheric water is vital, especially at elevated TDS or EC. According to Nnorom et al. (2019), highly correlated elements within a water body may exhibit similar hydrochemical characteristics.

Generally, strong positive correlations with correlation coefficients of 0.861 were observed between nitrate and aluminium and nickel and magnesium (r = 0.953). A moderate correlation was observed for zinc–turbidity, silica–manganese, sulphate–copper, silica–nickel, and zinc–sulphate, with correlation coefficients ranging from 0.589 to 0.706 (Table 2).

Venkatramanan et al. (2013) used CA as a method for combining groundwater wells into homogenous groups according to their water quality. Bhuiyan et al. (2011) also used CA to classify the surface, watercourse and Li et al. (2019) used CA to measure the similarity between water quality variables.

Cluster 1 consists of samples collected in November 2022, February 2023, March 2023, May 2023, June 2023, and September 2023. Cluster 2 are samples from December 2022 and October 2023. Cluster 3 consists of samples from April 2023, the only month without a corresponding month. Cluster 4 are samples from July 2023, August 2023, and January 2023. Therefore, the physicochemical parameters of atmospheric water collected during the months of each cluster exhibited the same water chemistry (pollution levels). Furthermore, according to the dendrogram, seasons do not play a role in the clusters as cluster 1 consisted of months that fall under all four seasons.

Tables 3 and 4 show results from FA conducted using 17 atmospheric water physicochemical parameters: chlorine, chloride, manganese, sulphate, fluoride, magnesium, nickel, silica, nitrate, aluminium, ammonia, alkalinity, copper, potassium, iron, calcium, and zinc. Six factors with eigenvalues ≥1 (red) were considered in displaying the variance in the data (Table 3). The eigenvalues for the factors were as follows: factor 1 had an eigenvalue of 4.926, factor 2 had an eigenvalue of 3.284, factor 3 had an eigenvalue of 2.537, factor 4 had an eigenvalue of 2.117, factor 5 had an eigenvalue of 1.395, and factor 6 an eigenvalue of 1.119. The six factors accounted for 90.461% of the total variance in the original data (Table 3). Factor 1 accounts for 24.995%, factor 2 accounts for 17.294%, factor 3 accounts for 15.426%, factor 4 accounts for 14.243%, factor 5 accounts for 10.284% and factor 6 accounts for 8.220% of the original variables.

The principal components (PCs) were rotated using the varimax method with Kaiser normalization (Table 4). Six factors were obtained for the atmospheric water parameters through FA performed on the PCs, which indicates that six main controlling factors influenced the quality of atmospheric water in the study area (Table 4).

The rotated component matrix (Table 4) was used to determine the loading of the variables under each factor. The factor loadings were classified as ‘strong’, ‘moderate’, and ‘weak’ according to the absolute loading values of >0.75, 0.75–0.50, and 0.50–0.30, respectively (Nnorom et al. 2019).

Factor 1 has a strong positive loading of chlorine, chloride, manganese, and sulphate; a moderate positive loading of silica, copper, and zinc; and a moderate negative loading of fluoride, indicating that it does not influence factor 1 sources due to it being too low (Table 4). The loadings in factor 1 indicate that this factor was multi-sourced from anthropogenic and natural activities such as agriculture and soil erosion processes (Bhuiyan et al. 2011) due to high loadings of chlorine, chloride, sulphate, and silica. Furthermore, anthropogenic activities such as steel and metal production, industrial combustion of lubricants, coal combustion, and vehicular road dust originate from the abrasion of vehicle parts (Taiwo et al. 2014; Das et al. 2015) due to moderate loadings of manganese, copper, and zinc.

Factor 2 has a strong positive loading of magnesium and nickel, a moderate positive loading of silica, a weak positive loading of manganese, and a weak negative loading of fluoride, indicating that it does not influence factor 2 sources due to it being too low (Table 4). Factor 2 loadings indicate that natural origins and anthropogenic activities influenced this factor. Magnesium and silica are due to soil erosion and leaching of rocks (Venkatramanan et al. 2013). Furthermore, nickel and manganese are the results of vehicle fuel combustion and traffic activities (Inbar et al. 2020). This is related to the findings of Heo et al. (2017), who discovered that the presence of nickel in the air is frequently connected to oil combustion, potentially originating from motorized vehicles.

Factor 3 has a strong negative loading of nitrate and aluminum, indicating that they do not influence factor 3 sources because their loadings are too low. Furthermore, factor 3 has a moderate positive loading of ammonia and a weak positive loading of sulphate and calcium. The loadings in factor 3 suggest that anthropogenic activities and natural origins influenced it. According to Muselli et al. (2006), Inbar et al. (2021), and Kaplan et al. (2023), ammonia and sulphate also originate from agricultural, industrial, and vehicular sources – moreover, Venkatramanan et al. (2013) state that calcium originates from soil erosion.

Factor 4 has a strong positive loading of copper, a weak positive loading of ammonia, a strong negative loading of alkalinity, and a moderate negative loading of alkalinity and potassium (Table 4). Therefore, alkalinity and potassium do not influence factor 4 sources because their loadings are too low. The loadings in factor 4 indicate that anthropogenic activities influenced this factor. The strong positive loading of copper indicates that steel and metal production, industrial combustion of lubricants, coal combustion, and vehicular road dust originating from the abrasion of vehicle parts were the main sources (Taiwo et al. 2014; Das et al. 2015). Furthermore, due to ammonia, other sources originate from agriculture and industrial emissions, according to Muselli et al. (2006), Inbar et al. (2021), and Kaplan et al. (2023).

Factor 5 has a strong positive loading of iron and a weak positive loading of ammonia and potassium (Table 4). The strong positive loading of iron indicates that this factor is mostly influenced by anthropogenic sources such as industrial and metallurgical processes, combustion of fossil fuels, and transport sources such as diesel wear, and tyre wear (Patel et al. 2012; Sanderson et al. 2014, 2016) influence this factor. The weak positive loading of potassium indicates that anthropogenic sources such as coal combustion and biomass burning were also sources of this factor (Yu et al. 2018). Additionally, due to ammonia, this factor was influenced by agriculture, and industrial emissions, according to Muselli et al. (2006), Inbar et al. (2021), and Kaplan et al. (2023).

Factor 6 has a moderate positive loading of zinc and a strong negative loading of calcium, indicating that it does not influence factor 6 sources due to its being too low (Table 4). The moderate positive loading of zinc indicates that this factor is mostly influenced by anthropogenic sources of vehicle abrasion (Heo et al. 2017; Inbar et al. 2020).

A quality assessment of atmospheric water that was produced by an AWG situated in an industrial area was conducted using multivariate statistical analysis. The comprehensive basic statistics results showed that most of the physicochemical parameters were well below both the SAWQG and WHO drinking water permissible limits. However, it is recommended that further monitoring should be given to pH, EC, the colour of water, ammonia, and nickel, as they were occasionally found to exceed both the SAWQG and WHO permissible limits for drinking water. The water produced by the AWG can potentially be used as a source of drinking water in heavily polluted areas, provided that specific contaminants, such as ammonia and nickel, are carefully monitored. It is also essential to conduct filtration and purification treatment before the atmospheric water is consumed.

The study used CA to group the 12 months that the atmospheric water samples were collected based on their water chemistry (pollution levels), helping evaluate the atmospheric water variation throughout the 12 months and four clusters were illustrated. Therefore, the physicochemical parameters of atmospheric water collected during each cluster (months) exhibited the same water chemistry (pollution levels). Furthermore, FA identified six pollution sources affecting the atmospheric water quality in the study area as anthropogenic sources (i.e., steel and metal production, vehicle fuel combustion, vehicle abrasion, traffic activities, agricultural, coal combustion, and biomass burning, and industrial emissions) and natural sources (soil erosion and leaching of rocks). The study has shown the effectiveness of using multivariate statistical analysis in monitoring atmospheric water quality, which could have a significant impact on atmospheric water management. It is recommended that multivariate statistical methods be integrated in future studies on pollution risk assessment of AWG-produced water environments in other South African regions.

The authors express their gratitude to the University of South Africa for the financial support provided to conduct this study. Additionally, Aqua Air Africa (Pty) Ltd is appreciated for granting permission to collect data on their premises.

The authors declare there is no conflict.

All relevant data are included in the paper or its Supplementary Information.

This study was funded by the University of South Africa Postgraduate Bursary Fund.

Material preparation, data collection, and analysis were performed by A.M. The conceptualisation of the study was done by A.M., L.L.S., M.M, and T.S. The first draft of the manuscript was written by A.M. writing-review, and editing was done by A.M., T.S.M., M. M., and L.L.S. All authors read and approved the final manuscript.