Abstract

The increase of fluorescent natural organic matter (fNOM) fractions during drinking water treatment might lead to an increased coagulant dose and filter clogging, and can be a precursor for disinfection by-products. Consequently, efficient fNOM removal is essential, for which characterisation of fNOM fractions is crucial. This study aims to develop a robust monitoring tool for assessing fNOM fractions across water treatment processes. To achieve this, water samples were collected from six South African water treatment plants (WTPs) during winter and summer, and two plants in Belgium during spring. The removal of fNOM was monitored by assessing fluorescence excitation–emission matrices datasets using parallel factor analysis. The removal of fNOM during summer for South African WTPs was in the range 69–85%, and decreased to 42–64% in winter. In Belgian WTPs, fNOM removal was in the range 74–78%. Principal component analysis revealed a positive correlation between total fluorescence and total organic carbon (TOC). However, TOC had an insignificant contribution to the factors affecting fNOM removal. Overall, the study demonstrated the appearance of fNOM in the final chlorinated water, indicating that fNOM requires a customised monitoring technique.

INTRODUCTION

Natural organic matter (NOM) in water sources reduces the performance of water treatment plants (WTPs), downgrades water quality, and increases operational costs (Sillanpaa et al. 2018). Fluorescent natural organic matter (fNOM) is largely made up of humic-like, fulvic-like, and protein-like components (Shao et al. 2014). Currently, parameters used for tracking recalcitrant fNOM are turbidity, total organic carbon (TOC), UV254, pH, and conductivity (Ndiweni et al. 2019). However, these monitoring tools do not have capabilities to reveal enough information to fast track the removal of problematic fNOM fractions. For instance, online turbidity measurements for raw water were proven to be unreliable compared to tryptophan-like (tryp-like) and humic-like fluorescence (FL) measurements (Sorensen et al. 2018).

A conventional WTP commonly uses different water treatment stages such as coagulation, sedimentation, sand filtration (SF), and chlorination. The effects of fNOM on water treatment include fouling of filtration media such as membranes and activated carbon (AC) (Shao et al. 2014). Moreover, during coagulation, flocculation, filtration, ozonation and chlorination, fNOM fractions can transform to form secondary products, which are more toxic and complex compared to the parent compounds (Brumer et al. 2019). Secondary product refers to compounds formed when ozone breaks down large molecules to small molecules that exhibit varying characteristics. Secondary product can also be used to refer to product formed when fNOM fractions react with chlorine disinfection by-products (DBPs), which are potentially formed (Chaukura et al. 2018). When traces of biodegradable fNOM are present in the chlorination stage, there is a possibility of bacterial regrowth and corrosion in the water distribution system (Baghoth & Amy 2012). These DBPs such as trihalomethanes and haloacetic acids stimulate cancer cells in humans; hence when optimising WTPs it is important to ensure that water is safe for consumption (Haarhoff et al. 2010).

Analysing unknown heterogeneous fNOM, some of which can occur at low relative concentrations or below detection limit, is a challenge (Liu et al. 2017). In order to identify fNOM fractions in WTPs, fluorescence excitation-emission matrices (FEEM) have been investigated to measure fluorescent intensities of fNOM in water samples (Murphy et al. 2018). The technique is user-friendly, highly selective, sensitive, and time- and cost-saving (Markechová et al. 2013). FEEM reveals the spectral properties of fNOM fractions and the data can be further processed by various applications such as fluorescence regional integration, and multi-way data analysis methods such as parallel factor (PARAFAC) analysis to detect the presence of different fNOM fractions and perform predictions of fNOM removal throughout the water treatment plant (Cuss et al. 2016; Sun et al. 2017). The relatively large datasets, resulting from fluorescence measurements, can further be analysed and categorised based on the most dominating functional substituents of fNOM fractions such as humic acid-like (HA-like), fulvic acid-like (FA-like), and protein-like materials (Baghoth & Amy 2012; Murphy et al. 2018). The PARAFAC model assists to interpret fNOM fractions denoted as components, as well as their scores, which are directly proportional to the concentration of fNOM fractions (Sun et al. 2017).

Fluorescence spectroscopy, and the use of PARAFAC, has proven to be reliable because of its high sensitivity and selectivity on most dominant fNOM fractions (Sun et al. 2017). Moreover, PARAFAC sensitivity increases with an increase in the size of the datasets. It is therefore suitable to use FEEM-PARAFAC in developing a model that can reveal the actual concentration of fluorophores provided the quantum yields or specific absorption coefficients are known (Sun et al. 2017). The interpretation of fNOM fractions in WTPs can be supported using protein compounds to humic compounds (P/H) ratio, which monitors the removal efficiency in the treatment stages (Arango et al. 2017). In previous research, this ratio revealed that in spring and summer the occurrence of HA-like material was higher compared to autumn, which could be due to observed high volume of leaves falling and flourishing algae (Arango et al. 2017). In this study, PARAFAC analysis was used to monitor the removal of fNOM in WTPs. This study is a follow-up to the previous study which focused on characterisation the properties of WTP source water. The presence of HA-like and protein-like compounds was monitored using P/H ratios. The objectives of the study were to: (1) determine the composition of the different fNOM fractions within the drinking water treatment process using FEEM datasets and PARAFAC model derived data; (2) use previous generated knowledge to monitor the removal of fNOM fractions throughout the WTPs, and (3) trace major fNOM fractions in the final stage for drinking water.

MATERIALS AND METHODS

Sampling

Water samples were collected during the winter (August) of 2017, and the summer (January and February) and spring (June) of 2018 from six WTPs in South Africa, namely Plettenberg Bay (A), in the Western Cape; Umzinto (B), in the KwaZulu Natal; Reitvlei (I) in Gauteng; Midvaal (C and D) in the North West, and Ebenezer (E), and Flag Bushielo (F), both in Limpopo. In Belgium, samples were collected during the spring (June) of 2018, and the selected WTPs (operated by De Watergroep) were De Gavers (G) in Harelbeke and Dikkebus (H) in Ieper. The treatment stages are as indicated in Table 1. The samples were analysed straight away to prevent NOM degradation over time. The following parameters were measured onsite in triplicate: pH, turbidity, and conductivity for South African and Belgium treatment plants.

Table 1

Monitoring parameters (turbidity, TOC, UV254 and PARAFAC outputs) removal from raw to the final water in plants in South Africa and Belgium

Turbidity removal from raw to the final water (%)
TOC removal from raw to the final water (%)
UV254 removal from raw to the final water (%)
PARAFAC components removed from raw water to final water (%)
WTPsMeasuredLiterature referenceMeasuredLiterature referenceMeasuredLiterature referenceTFFmax1Fmax2Fmax3Literature reference
Plant A: raw (surface water) – coagulation – settling – SF (sand filtration) – final 59.30  46.44 NOM 50% for raw water with 3.8 mg/L DOC (Callegari et al. 2017); TOC 47% for raw water with 5.6 mg/l (Matilainen et al. 2002); 48 .70 (Sillanpaa et al. 201876.67  72.25 75.38 100.00 9.12  
Plant B1: raw – coagulation – settling – SF – final   26.62  71.80 70.00 (Sillanpaa et al. 201859.44 47.59 56.03 89.43  
Plant F: raw (surface water) – coagulation – settling – SF – final 99.05  60.72 TOC 68% (Nissinen et al. 200181.82  89.22 86.62 100.00 80.04 Humic fractions 91% (Nissinen et al. 2001). TF 86% (HA-like, tyrosine-like, protein-like, tryp-like) (Markechová et al. 2013
Plant I: raw (surface water) – coagulation – DAF – final winter 75.66, summer 97.98  winter 0.51, summer 29.59  winter -, summer 38.10 36.10 (Sillanpaa et al. 201841.77 44.32 27.34 45.37 Humic substances 44% (Volk et al. 2005). 
Plant C and D: raw (surface water) – pre-ozonation – coagulation – DAF – post-ozonation – SF – final winter 99.40, summer 99.68      69.30 59.81 58.38 83.34 TF ∼50–70% (HA-like and protein-like), (Baghoth & Amy 2012). humic substances 65% (Sillanpaa et al. 2018
Plant E: raw (surface water) – aeration – coagulation – SF – final winter 89.67, summer 98.70  winter, summer 32.10    50.68 60.27 52.34 31.81  
Plant F: raw (surface water) – coagulation – settling – SF – final winter 74.74, summer 74.98  winter 0.29, summer  winter -, summer 33.33  49.94 61.61 46.90 56.28  
Plant G: raw (surface water) – nitrification pond – SF – ultra-filtration – AC filtration – final 94.33 Turbidity 95% for raw water with 5.5 FTU (Matilainen et al. 2002100.00 NOM 90% for raw water with TOC >10 mg/L (Volk et al. 2005). 90 90.00 (Sillanpaa et al. 201877.98 84.33 87.30   
Plant H: raw – coagulation – DAF – SF – AC – final 99.75  100.00  78.95  74.34 79.21 79.37 20.54  
Turbidity removal from raw to the final water (%)
TOC removal from raw to the final water (%)
UV254 removal from raw to the final water (%)
PARAFAC components removed from raw water to final water (%)
WTPsMeasuredLiterature referenceMeasuredLiterature referenceMeasuredLiterature referenceTFFmax1Fmax2Fmax3Literature reference
Plant A: raw (surface water) – coagulation – settling – SF (sand filtration) – final 59.30  46.44 NOM 50% for raw water with 3.8 mg/L DOC (Callegari et al. 2017); TOC 47% for raw water with 5.6 mg/l (Matilainen et al. 2002); 48 .70 (Sillanpaa et al. 201876.67  72.25 75.38 100.00 9.12  
Plant B1: raw – coagulation – settling – SF – final   26.62  71.80 70.00 (Sillanpaa et al. 201859.44 47.59 56.03 89.43  
Plant F: raw (surface water) – coagulation – settling – SF – final 99.05  60.72 TOC 68% (Nissinen et al. 200181.82  89.22 86.62 100.00 80.04 Humic fractions 91% (Nissinen et al. 2001). TF 86% (HA-like, tyrosine-like, protein-like, tryp-like) (Markechová et al. 2013
Plant I: raw (surface water) – coagulation – DAF – final winter 75.66, summer 97.98  winter 0.51, summer 29.59  winter -, summer 38.10 36.10 (Sillanpaa et al. 201841.77 44.32 27.34 45.37 Humic substances 44% (Volk et al. 2005). 
Plant C and D: raw (surface water) – pre-ozonation – coagulation – DAF – post-ozonation – SF – final winter 99.40, summer 99.68      69.30 59.81 58.38 83.34 TF ∼50–70% (HA-like and protein-like), (Baghoth & Amy 2012). humic substances 65% (Sillanpaa et al. 2018
Plant E: raw (surface water) – aeration – coagulation – SF – final winter 89.67, summer 98.70  winter, summer 32.10    50.68 60.27 52.34 31.81  
Plant F: raw (surface water) – coagulation – settling – SF – final winter 74.74, summer 74.98  winter 0.29, summer  winter -, summer 33.33  49.94 61.61 46.90 56.28  
Plant G: raw (surface water) – nitrification pond – SF – ultra-filtration – AC filtration – final 94.33 Turbidity 95% for raw water with 5.5 FTU (Matilainen et al. 2002100.00 NOM 90% for raw water with TOC >10 mg/L (Volk et al. 2005). 90 90.00 (Sillanpaa et al. 201877.98 84.33 87.30   
Plant H: raw – coagulation – DAF – SF – AC – final 99.75  100.00  78.95  74.34 79.21 79.37 20.54  

DOC, dissolved organic carbon; TF, total fluorescence; DAF, dissolved air flotation.

Further analysis of water samples

Total organic carbon was measured in triplicate using a Fusion TOC analyzer (Tekmar, Teledyne Instruments, USA) for South African samples, and using method 10129 and a TOC analyser (Hach, USA) for Belgian samples. To ensure they were representative of the organic matter in the WTPs, the samples were not treated further in the laboratory. For South African water samples, UV254 absorbance was measured concurrently with fluorescence measurements, while a UV-Vis spectrophotometer (Shimadzu, 1601, Japan) was used for Belgian samples. An Aqualog spectrofluorometer with built-in UV-Vis (HORIBA Jobin Yvon AquaLog, USA) was used to measure fNOM using Raman water as a blank in a quartz cuvette with 1 cm width light path length. The EEMs datasets for South African samples were obtained at 1 s integration time for excitation wavelengths 239.00–800.00 nm at 3.00 nm intervals, and emission wavelengths of 246.70–825.13 nm at an average interval of 4.54 nm. For Belgium samples, fluorescence spectra were obtained using a fluorimeter (Shimadzu, RF-5301, Japan) at 0.25 s integration time for excitation wavelengths 220.00–450.00 nm at 5 nm intervals, and emission wavelengths of 280.00–600.00 nm at 1.00 nm increments. The excitation and emission slits width were 5 nm for all measurements. The Raman scan was performed at excitation and emission slit width of 5 nm and a response time of 0.25 s using demineralised water at an excitation wavelength of 350.00 nm and an emission wavelength of 365.00–450.00 nm using 0.20 nm increments.

Method development for online monitoring of fluorescent natural organic matter

A modelling technique, PARAFAC, was used to mathematically separate fNOM fractions into a trilinear setup using an alternating least square algorithm. The PARAFAC model was developed in Matlab R2016a environment and using the DOMFlour tool and method (Cuss et al. 2016). The EEMs datasets correction was achieved by blank subtraction using the decomposition routines for the excitation–emission matrices (drEEM) toolbox to remove the influence of sample handling and instrument measurement instability. The measurements of UV absorbance were used to correct for the inner filter. Pre-processing of spectral corrected EEMs was limited to non-negative values and PARAFAC model running at no particular order. The fluorescence intensities of the samples were normalised using the area given by the Raman measurements and thereafter the units of sample fluorescence intensity were Raman units (RU) to ensure consistency (Sun et al. 2017). Robust PARAFAC models were developed by eliminating outliers and visually inspecting residuals to ensure they only consisted of noise. Prior PARAFAC model validation of the loadings of fNOM spectra was examined for accuracy and outliers removed. Inappropriate spectra were improved by adjusting iteration repetition and convergence from the default of 10−6 to a stricter 10−8 and 10−10. Thereafter, the split half analysis technique was performed to validate that the same PARAFAC model can be achieved through grouping differently EEMs datasets. The validated components were recorded in the drEEM toolbox (Markechová et al. 2013). The fNOM fractions were identified based on documented emission and excitation wavelength peaks. The HA-like fractions were located at emission and excitation wavelengths within the range 380–540 nm and 250–340 nm, FA-like range 380–540 nm and 200–250 nm, aromatic protein I range 280 nm–330 nm and 200 nm–250 nm, aromatic protein II range 330–380 nm and 200–250 nm, and soluble microbial products range 280–380 nm and 250–300 nm (Ndiweni et al. 2019). The P/H ratios (Equation (1)) were calculated using fluorescence intensities at specific wavelengths, whereby for protein-like compounds excitation at 275.00 nm and emission at 340.00 nm, and for humic material excitation at 350.00 nm and emission at 480.00 nm were used (Arango et al. 2017). 
formula
(1)
where is the excitation wavelength and is the emission wavelength.

RESULTS AND DISCUSSION

Quality parameters for water samples

The pH measurements for all the WTPs investigated in this study were consistently in the range 5.84–9.27, which is a huge variation, but for final disinfection stage the pH was within the SANS 241 acceptable level. Notably, the final stage for plant A had a high pH (9.27 ± 0.04). Although this pH is not harmful to human health, it causes aesthetic problems such as alkaline-like taste. The variation of turbidity and its removal do not indicate huge seasonal differences. In previous studies, turbidity removal was 50.00–90.00% and 95.00% for a four-stage conventional WTPs with raw water turbidity measured at 4.80 NTU and 5.50 NTU respectively (Matilainen et al. 2002; Callegari et al. 2017). In this study, the four-stage conventional WTPs' raw water turbidity was in the range 3.00–6.50 NTU and removal percentage was within the range 74.81–99.05% (Table 1). Turbidity substances facilitate transportation and protection and provide food for microbial species, and should, therefore, be kept to a minimum (Farrell et al. 2018).

In plant L, the nitrification process removes ammonium, which is an indicator of fecal pollution and the presence of nitrogen-based fertilizers (Westgate 2009). A reduction of 94.33% in the turbidity was observed for plants K and L, respectively (Table 1). However, the filtration stages were effective in reducing the turbidity to less than 1.00 NTU. The removal of the TOC for WTPs A to F from raw water to the final stages were within the range 21.00–73.00% during summer, whereas WTPs G, H, I, and J were within the range 13.00–24.00%. However, the removal of the TOC by WTPs G, H, I, and J during winter from raw water to the final stage was within the range 1.72–32.09%. Plant F TOC removal corresponded to previous studies (Nissinen et al. 2001; Callegari et al. 2017). The contamination of water sources originates mainly from wastewater treatment plants and informal settlements with poor waste management of sewage (Chaukura et al. 2018). During summer (January and February), South Africa experiences high levels of rainfall that generally keep dams and rivers full, potentially diluting TOC concentrations. The removals of TOC in Belgium WTPs using SF and ultra-filtration (UF) membranes were 52 and >98.00%, respectively. A previous study indicated that UF is more effective for the removal of large molecular weight NOM (Haarhoff et al. 2010). In another study, UF showed maximum efficiency of 90.00% (Prisciandaro & di Celso 2016). However, the efficiency of the membrane is compromised by fNOM fractions that cause fouling (Sillanpaa et al. 2018). The values recorded in South African winter (June and July) of 0.12–0.48 cm−1, in South African summer (January and February) of 0.05–0.13 cm−1, and in Belgian spring (May) of 0.01–0.04 cm−1 of UV254 for the final water indicate traces of light-absorbing aromatic compounds which are fNOM. It is deduced that there is a persistence of fNOM based UV254 measurements in the final disinfection stage. Although it reveals limited information, UV254 is currently a valuable tool to monitor NOM on-line.

A monitoring tool for fluorescence natural organic matter in WTPs

The EEM datasets were processed in PARAFAC analysis to calculate the values of the three validated major contributing components per plant at the maximum intensity, also referred to as relative concentrations. The wavelengths and the three-dimensional fluorescence spectra of the isolated major fNOM fractions in this study, classified as Fmax1, Fmax2 and Fmax3, are described in detail in a previous publication (Ndiweni et al. 2019). The maximum intensity (Fmax) during the South African winter (June and July) for Fmax1 raw water and final water was in the range 4.89–0.68 and 2.16–0.20 RU, respectively. During the South African summer (January and February) Fmax1 was in the range 1.62–0.05 and 0.60–0.02 RU in the raw water and final water, respectively. However, during the Belgian spring (May) Fmax1 was in the range 1.36–0.76 and 0.28–0.13 RU in the raw water and final water, respectively. The Fmax2 outcomes during winter were in the range 1.11–0.28 and 1.38–0.14 RU for raw water and final water, respectively, while during summer they were in the range 0.61–0.11 and 0.24 RU to less than the detection limit for raw water and final water, respectively. However, during spring Fmax2 was 0.86–0.47 and 0.11–0.10 RU for raw water and final water, respectively. The Fmax3 values during winter were in the range 0.54–0.23 and 0.66–0.08 RU for raw water and final water, respectively. In summer Fmax3 was in the range 0.57–0.02 and 0.37–0.01 RU for raw water and final water, respectively, while in spring it was in the range 0.46–0.10 and 0.31–0.03 RU for raw water and for final water, respectively. It is observed that this study is in agreement with a previous study in terms of sensitivity (Murphy et al. 2018). Samples with Fmax as low as 0.01 RU were detected in the final water.

Fmax1 mainly consist of fNOM fractions with large molecular weight and high hydrophobicity such as HA-like material, which is rich in terrestrial materials. HA-like material is dominant in water bodies due to decomposition of dead animals and plants (Shao et al. 2014; Ndiweni et al. 2019). Notably, Fmax2 consists of fNOM fractions that are a mixture of both high and low hydrophilicity and hydrophobicity as well as large and small molecular weight material. However, Fmax3 consists of protein-like substances, soluble microbial by-products (SMPs), which possess both low hydrophobicity and low aromaticity, and small molecular weight fractions (Zhu et al. 2017). Hence monitoring the removal of fNOM fractions throughout the WTPs (Figure 1) is important for fast-tracking changes in composition of fNOM fractions. This study tracked evolution of Fmax throughout the treatment process as a follow-up to a previous validated model for treatment plant water from sources such as rivers, dam and borehole (Ndiweni et al. 2019). Using PARAFAC analysis, a previous study isolated three major fNOM fractions from a diverse reservoir of heterogeneous polydisperse carbon-based species (Murphy et al. 2018). The fNOM total fluorescence (TF) removal during summer for South African WTPs ranged 69–85% and decreased to 42–64% in winter. In Belgian WTPs, fNOM removal ranged 74–78% (Table 1). A comparison of datasets with literature showed comparable fNOM removals (Table 1). In aquatic systems, photodegradation and biodegradation processes contribute to heterogeneous properties of fNOM (Hansen et al. 2016). For plants A–F, Fmax2 consisting of tryp-like and protein-like substances containing traces of SMPs and HA-like substances was efficiently removed by the filtration step (Figure 1(a)).

Figure 1

The removal of fluorescence natural organic matter in water treatment plants monitored using total florescence and maximum fluorescence. TF is total fluorescence divided by 10,000 to make data handling much easier. (a)–(i): Plants A to I, respectively. Aer: aeration; Coag: coagulation; Set: settling tank; Flot: flotation; Nitri: nitrification stage; Pondin: pond inlet; Pondf: pond outlet.

Figure 1

The removal of fluorescence natural organic matter in water treatment plants monitored using total florescence and maximum fluorescence. TF is total fluorescence divided by 10,000 to make data handling much easier. (a)–(i): Plants A to I, respectively. Aer: aeration; Coag: coagulation; Set: settling tank; Flot: flotation; Nitri: nitrification stage; Pondin: pond inlet; Pondf: pond outlet.

Furthermore, Fmax2 removal was possibly enhanced by variation in its composition such as aromaticity, molecular weight, and charge. Plant A and B Fmax1, which consists of HA-like materials, had removal range 58–79%, while Fmax2 had >98% removal. In the treatment process primary stages such as coagulation and flocculation target Fmax1 based on its characteristics such as high hydrophobicity, large molecular weight and its particle charge. In the succeeding treatment stages both Fmax2 and Fmax3 fNOM removals increase as they are targeted by final stages which are SF, AC filtration and UF (Markechová et al. 2013). The humic material removal was in agreement with previous studies for plant B and I (Nissinen et al. 2001; Volk et al. 2005). A relative increase of Fmax3 was observed in the settling and SF steps for plant B denoted as aromatic protein II with traces of SMPs and amino acids. This relative increase was also observed for plant B settling for TF, Fmax1 and Fmax2; plant F final stage for Fmax1; plant I coagulation; and plant C and D flotation and final stage for Fmax1 and Fmax3 (Figure 1). These process can promote carbon reduction and a shift in the functional groups of fNOM fractions (Khan et al. 2019). The Fmax3 for plant K which is aromatic protein I and II with traces of FA-like shows relative increase by 20.54% compared with Fmax1 and Fmax2 in the final stage.

Although the nitrification process for Plant H removes inorganic nitrogen, its removal efficiency for organic nitrogen is limited, and algae growth will therefore persist (Westgate 2009). As was mentioned in a previous study, for plant H the presence of algae that grow is highly possible due to the fertilizer distributed for agricultural reasons and run-off from the nearby land (Ndiweni et al. 2019). The removal of aromatic proteins decreased in the SF and AC filtration due to pore blockage and the possibility of a decrease in the adsorption capacity of the AC (Yang et al. 2018). Organic compounds such as tryp-like and protein-like substances are indicators of the significant contribution of wastewater effluent to the fNOM in water sources (Yang et al. 2018). For plant C and D, the removal of the hydrophobic acid containing traces of FA- and HA-like substances was 79.21%, which was mostly removed at the AC stage (Figure 1). The removal of SMPs with traces of aromatic protein was 79.37% and was mostly removed in the AC stage. Generally, the presence of aromatics continued to increase during the treatment process, possibly because the component consisting of proteins was not well defined; thus a small increase lies in the accuracy of the model and measurements. Based on the analysis of the results, the pressing issue for the development of the monitoring tool for discovered fNOM fractions is their appearance in the final water. This is an indication that WTPs are not operating optimally (Sillanpaa et al. 2018). The presence of fNOM fractions presents challenges such as the formation of secondary products due to accumulation of pollutants in the unit processes, which encourages biological activity within the treatment train (Brumer et al. 2019). The secondary products differ in chemical composition from their primary fNOM fractions, decreasing the removal efficiency and increasing toxicity (Brumer et al. 2019). The toxicity and side effects of traces of fNOM fractions remain unknown, hence the need to develop a monitoring tool that can rapidly track the occurrence of unknown fNOM fractions trace in the final treated water. When the characteristics of fNOM are known, it will be possible to optimise treatment stages, chemical dosage, and plant design for optimum fNOM fractions removal. Using recent technology, WTPs can improve their efficiency through atomisation, and predictions of the occurrence of fNOM fractions due to environmental activities and seasonal changes.

Correlation between measured and calculated parameters

To track the correlation the TF, Fmax, H/P ratio, fluorescence index, biological index, freshness index and site-measured parameter, principal component analysis (PCA) was employed. The indices were calculated for influent water as explained in a previous study (Box 1) (Ndiweni et al. 2019). The rotated component matrix illustrated main factors influencing the evolution in fNOM fractions throughout the treatment plants (Figure 2(a) and 2(b)). To ensure that the output of PCA is reliable a test was conducted using a Kaiser-Meyer-Olkin criterion, which had a value of 0.67, which is acceptable although the closer it is to 1.00 the more reliable it is (Ndiweni et al. 2019). The eigenvalues for principal component (PC) factors were 4.48, 3.83, and 1.50 for PC1, PC2, and PC3, respectively (Table 2). PC1 explained that the variance of 35.22% correlated to Fmax1, Fmax1 removal, Fmax2 removal, TF and TF removal. It is observed that the TF and TOC can be correlated; however, TOC had insignificant contribution in PC1 (Figure 2(a)). The main variables influencing the removal of fNOM are illustrated in Figure 2(b). PC1 is strongly correlated to Fmax information; therefore it can be used as a measure of fluorescence relative concentration. PC2 explains that 29.42% variation correlated to Fmax3, conductivity, P/H ratio, FI (fluorescence index), HIX (humification index), β:α (freshness index; β = recently produced NOM fractions, α = aged NOM fractions), and BIX (biological index); therefore it can be used as a measure of indicators. PC3 explains that 11.54% variation correlated to TF, Fmax2, UV254 be used as measure of UV254 related to Fmax2 fNOM fractions.

Box 1 Equations based on measured and calculated drinking water samples parameters
 
formula
(1)
 
formula
(2)
 
formula
(3)

Table 2

Principal component analysis for variables and their factors influencing fNOM removal throughout the treatment plant

VariablesFactors
123
TF −0.76a 0.10 0.53a 
TF removal% 0.97a 0.03 −0.14 
Fmax1 −0.83a −0.11 0.34 
Fmax3 0.17 0.67a 0.23 
Conductivity (μS/cm) 0.10 0.84a 0.32 
Fmax2 −0.20 0.10 0.73a 
Fmax1 removal% 0.92a 0.10 0.09 
Fmax2 removal% 0.91a −0.12 −0.17 
 UV254 −0.29 −.20 0.69a 
P/H ratio 0.00 0.88a −0.22 
FI 0.12 0.91a 0.04 
β:α −0.27 0.62a −0.33 
BIX −0.24 0.77a −0.33 
Eigenvalues 4.48 3.83 1.50 
% of explained variance 35.22 29.42 11.54 
VariablesFactors
123
TF −0.76a 0.10 0.53a 
TF removal% 0.97a 0.03 −0.14 
Fmax1 −0.83a −0.11 0.34 
Fmax3 0.17 0.67a 0.23 
Conductivity (μS/cm) 0.10 0.84a 0.32 
Fmax2 −0.20 0.10 0.73a 
Fmax1 removal% 0.92a 0.10 0.09 
Fmax2 removal% 0.91a −0.12 −0.17 
 UV254 −0.29 −.20 0.69a 
P/H ratio 0.00 0.88a −0.22 
FI 0.12 0.91a 0.04 
β:α −0.27 0.62a −0.33 
BIX −0.24 0.77a −0.33 
Eigenvalues 4.48 3.83 1.50 
% of explained variance 35.22 29.42 11.54 

aFactors with significant loadings.

Figure 2

Principal component analysis for drinking water treatment plants in (a) South Africa, and (b) Belgium.

Figure 2

Principal component analysis for drinking water treatment plants in (a) South Africa, and (b) Belgium.

It was observed that the overall TF intensity decreased with TOC from raw to final water for plants in South Africa (Figure A1, Supplementary Material). These results are in agreement with a previous study that showed a direct correlation between TF intensity and TOC (Baghoth & Amy 2012). However, the linearity of this relationship can be compromised mainly by high dissolved organic carbon concentration, and reduced intensity by species such as nitrates (Baghoth & Amy 2012). TOC represents the carbon content that is calculated as dissolved organic and particulate carbon, whereas TF denotes the intensity due to problematic targeted fNOM (unsaturated compounds). Using TOC, the carbon content was tracked throughout the WTPs. However, with TF this study revealed that there is a presence of persistent fNOM in final water. In the absence of sufficient residual disinfectant, the traces of fNOM in the final water may lead to microbial regrowth in the distribution system (Ndiweni et al. 2019). The TF data gives better information of the targeted problematic fNOM compared to the traditional measurement monitored, which is TOC. The PCA revealed that the TOC throughout the treatment plant had insignificant influence on the removal of fNOM fractions. Hence TF should be taken into consideration when evaluating the WTP efficiency and optimisation.

The protein to the humic ratio

In treatment processes the most dominating fNOM fractions are humic and protein material; their characteristics are as explained above. Therefore calculation of humic and protein ratio assists in development of fNOM fractions monitoring based on the most dominating components. The P/H ratio revealed that for South African WTPs there is an abundance of humic material compared to protein compounds (P/H ratio is <0.6) (Figure 3). However, for Belgian plants (P/H ≥ 1), presence of algae in the influent occurs at higher levels, which causes degradation of humified materials as the P/H ≥ 1 for plant G and for plant H is >0.6 (Figure 3) (Brumer et al. 2019). Furthermore, these findings demonstrated that: (1) the pool of fNOM enormously depends on pollutants in the raw water, and (2) the efficiency of WTPs is limited by the composition of fNOM. Previous studies indicate that protein-like compounds dominate due to the influence of algal boom, pollutants resulting from agricultural activities, and poorly treated wastewater effluent (Sorensen et al. 2018). Notably humic-like compounds are mostly removed in the coagulation, whereas humified material is not effectively removed prior to the SF or AC stages. However, they tend to compete for active sites with protein-like compounds (Zanacic et al. 2016). These results show that there is a significant reduction in the removal rate of the targeted protein-like compounds and the life span of the filtration medium is compromised. The higher P/H ratio values in the final water compared to raw water confirms that the traces of humified materials are higher than that of proteins compounds. This increase in P/H ratio may also be an indication of the formation of secondary fNOM fractions (Brumer et al. 2019). If the traces of humic compounds continue to increase in the final water, it may give rise to carcinogenic compounds which result when fNOM reacts with chlorine in the final water (Chaukura et al. 2018). The final water from plant G had higher humic materials in winter compared to summer. In comparison, plant C had water entering the WTP with high microbial activities processing humified materials in winter. However, both plants had higher humic compounds compared to the summer season. A previous study reported that during summer fNOM was not destroyed; instead there are spectral shifts which indicate transformation of the fNOM (Khan et al. 2019).

Figure 3

Tracking the presence of protein-like to the humic-like (P/H) ratio during the different treatment stages for drinking water treatment plants in South Africa and Belgium. In treatment stage abbreviations, the first letter refers to the plant; for explanation of the rest of the abbreviation, please see text and Figure 1 caption.

Figure 3

Tracking the presence of protein-like to the humic-like (P/H) ratio during the different treatment stages for drinking water treatment plants in South Africa and Belgium. In treatment stage abbreviations, the first letter refers to the plant; for explanation of the rest of the abbreviation, please see text and Figure 1 caption.

CONCLUSIONS

The removal of fNOM fractions during summer for South African WTPs was in the range 69–85% and decreased to 42–64% in winter. In Belgian WTPs, fNOM fraction removal ranged 74–78%. The removal or alteration of fNOM fractions is influenced by the chemical structure of fNOM, water treatment regimen, and seasonal changes. Although a low removal percentage does not reflect a high amount of fNOM fractions, the amount in the influent should be taken into consideration. Principal component analysis revealed that TF and TOC can be correlated: however, TOC had insignificant contribution to the factors affecting removal of fNOM. The P/H ratio revealed that for South African WTPs there is an abundance of humic material compared to protein compounds. However, for Belgian plants, it is likely that microbial activity occurs at higher levels, degrading humic materials. This PARAFAC model demonstrated there is appearance of fNOM fractions in the final chlorinated water. Hence further studies can predict amount, toxicity and side-effects of traces of fNOM factions in effluent.

SUPPLEMENTARY MATERIAL

The Supplementary Material for this paper is available online at https://dx.doi.org/10.2166/wst.2020.136

REFERENCES

REFERENCES
Arango
C. P.
,
Beaulieu
J. J.
,
Fritz
K. M.
,
Hill
B. H.
,
Elonen
C. M.
&
Pennino
M. J.
2017
Urban infrastructure influences dissolved organic matter quality and bacterial metabolism in an urban stream network
.
Freshwater Biology
62
(
11
),
1917
1928
.
https://doi.org/10.1111/fwb.13035
.
Baghoth
S. A.
&
Amy
G. L.
2012
Characterizing Natural Organic Matter in Drinking Water Treatment Processes and Trains
.
Thesis, Delft University of Technology
,
Netherlands
.
Brumer
A. M.
,
Vughs
D.
,
Siegers
W.
,
Bertelkamp
C.
,
Hofman-Caris
R.
&
Kolkman
A.
2019
Monitoring transformation product formation in the drinking water treatments rapid sand filtration and ozonation
.
Chemosphere
214
,
801
811
.
https://doi.org/10.1016/j.chemosphere.2018.09.140
.
Callegari
A.
,
Boguniewicz-Zablocka
J.
&
Capodaglio
A. G.
2017
Experimental application of an advanced separation process for NOM removal from surface drinking water supply
.
Separations
4
(
32
).
https://doi:10.3390/separations4040032
Chaukura
N.
,
Ndlangamandla
N. G.
,
Moyo
W.
,
Msagati
T. A. M.
,
Mamba
B. B.
&
Nkambule
T. T. I.
2018
Natural organic matter in aquatic systems – a South African perspective
.
Water SA
44
(
4
),
624
635
.
http://dx.doi.org/10.4314/wsa.v44i4.11
.
Cuss
C. W.
,
McConnell
S. M.
&
Guéguen
C.
2016
Combining parallel factor analysis and machine learning for the classification of dissolved organic matter according to source using fluorescence signatures
.
Chemosphere
155
,
283
291
.
https://doi.org/10.1016/j.chemosphere.2016.04.061
.
Farrell
C.
,
Hassard
F.
,
Jefferson
B.
,
Leziart
T.
,
Nocker
A.
&
Jarvis
P.
2018
Turbidity composition and the relationship with microbial attachment and UV inactivation efficacy
.
Science of the Total Environment
624
,
638
647
.
https://doi.org/10.1016/j.scitotenv.2017.12.173
.
Haarhoff
J.
,
Kubare
M.
,
Mamba
B.
,
Krause
R.
,
Nkambule
T.
,
Matsebula
B.
&
Menge
J.
2010
NOM characterization and removal at six Southern African water treatment plants
.
Drinking Water Engineering Science
3
,
53
61
.
https://doi.org/10.5194/dwes-3-53-2010
.
Hansen
A. M.
,
Kraus
T. E. C.
,
Pellerin
B. A.
,
Fleck
J. A.
,
Downing
B. D.
&
Bergamaschi
B. A.
2016
Optical properties of dissolved organic matter (DOM): effects of biological and photolytic degradation
.
Limnology and Oceanography
61
(
3
),
1015
1032
.
https://doi.org/10.1002/lno.10270
.
Khan
S. I.
,
Zamyadi
A.
,
Rao
N. R. H.
,
Li
X.
,
Stuetz
R. M.
&
Henderson
R. K.
2019
Fluorescence spectroscopic characterisation of algal organic matter: towards improved in situ fluorometer development
.
Environmental Science: Water Research & Technology
5
(
2
),
417
432
.
https://doi.org/10.1039/C8EW00731D
.
Liu
G.
,
Ya Zhang
Y.
,
Knibbe
W. J.
,
Feng
C.
,
Liu
W.
,
Medema
G.
&
Meer
W.
2017
Potential impacts of changing supply-water quality on drinking water distribution: a review
.
Water Research
16
,
135
148
.
Markechová
D.
,
Tomková
M.
&
Sádecká
J.
2013
Fluorescence excitation-emission matrix spectroscopy and parallel factor analysis in drinking water treatment : a review
.
Polish Journal of Environmental Studies
22
(
5
),
1289
1295
.
Matilainen
A.
,
Lindqvist
N.
,
Korhonen
S.
&
Tuhkanen
T.
2002
Removal of NOM in the different stages of the water treatment process
.
Environment International
28
(
6
),
457
465
.
Murphy
K. R.
,
Timko
S. A.
,
Gonsior
M.
,
Powers
L. C.
,
Wünsch
U. J.
&
Stedmon
C. A.
2018
Photochemistry illuminates ubiquitous organic matter fluorescence spectra
.
Environmental Science and Technology
52
(
19
),
11243
11250
.
https://doi.org/10.1021/acs.est.8b02648
.
Ndiweni
S. N.
,
Chys
M.
,
Chaukura
N.
,
Van Hulle
S. W. H.
&
Nkambule
T. T. I.
2019
Assessing the impact of environmental activities on natural organic matter in South Africa and Belgium
.
Environmental Technology
40
(
13
),
1756
1768
.
https://doi.org/10.1080/09593330.2019.1575920
.
Nissinen
T. K.
,
Miettinen
I. T.
,
Martikainen
P. J.
&
Vartiainen
T.
2001
Molecular size distribution of natural organic matter in raw and drinking waters
.
Chemosphere
45
,
865
873
.
Prisciandaro
M.
&
di Celso
G. M.
2016
On the removal of natural organic matter from superficial water by using UF and MF membranes
.
Desalination and Water Treatment
57
(
6
),
2481
2488
.
https://doi.org/10.1080/19443994.2015.1031184
.
Sillanpaa
M.
,
Ncibi
M. C.
,
Matilainen
A.
&
Vepsalainen
M.
2018
Removal of natural organic matter in drinking water treatment by coagulation: a comprehensive review
.
Chemosphere
190
,
54
71
.
https://doi.org/10.1016/j.chemosphere.2017.09.113
.
Sorensen
J. P. R.
,
Vivanco
A.
,
Ascott
M. J.
,
Gooddy
D. C.
,
Lapworth
D. J.
&
Read
D. S.
2018
Online fluorescence spectroscopy for the real-time evaluation of the microbial quality of drinking water
.
Water Research
137
,
301
309
.
https://doi.org/10.1016/j.watres.2018.03.001
.
Sun
H. Y.
,
Koal
P.
,
Gerl
G.
,
Schroll
R.
,
Joergensen
R. G.
&
Munch
J. C.
2017
Water-extractable organic matter and its fluorescence fractions in response to minimum tillage and organic farming in a Cambisol
.
Chemical and Biological Technologies in Agriculture
4
(
15
),
1
11
.
https://doi.org/10.1186/s40538-017-0097-5
.
Volk
C.
,
Kaplan
L. A.
,
Robinson
J.
,
Johnson
B.
,
Larry
W.
,
Zhu
H. W.
&
Lechevallier
M.
2005
Fluctuations of dissolved organic matter in river used for drinking water and impacts on conventional treatment plant performance
.
Environmental Science Technology
39
,
4258
4264
.
https://doi.org/10.1021/es040480 k
.
Westgate
P. J.
2009
Characterization of Proteins in Effluents from Three Wastewater Treatment Plants that Discharge to the Connecticut River
.
Thesis, University of Massachusetts Amherst
.
Yang
X.
,
Yu
X.
,
Cheng
J.
,
Zheng
R.
,
Wang
K.
&
Dai
Y.
2018
Impacts of land-use on surface waters at the watershed scale in southeastern China: insight from fluorescence excitation-emission matrix and PARAFAC
.
Science of the Total Environment
627
,
647
657
.
https://doi.org/10.1016/j.scitotenv.2018.01.279
.
Zanacic
E.
,
Stavrinides
J.
&
Mcmartin
D. W.
2016
Field-analysis of potable water quality and ozone efficiency in ozone- assisted biological filtration systems for surface water treatment
.
Water Research
104
,
397
407
.
https://doi.org/10.1016/j.watres.2016.08.043
.
Zhu
G.
,
Bian
Y.
,
Hursthouse
A. S.
,
Wan
P.
,
Szymanska
K.
,
Ma
J.
,
Wang
X.
&
Zhao
Z.
2017
Application of 3-D fluorescence: characterization of natural organic matter in natural water and water purification systems
.
Journal of Fluorescence
27
,
2069
2094
.
doi:10.1007/s10895-017-2146-7
.

Author notes

Present address: VEG-i-TEC, Ghent University, Campus Kortrijk, Graaf Karel De Goedelaan 5, B-8500 Kortrijk, Belgium

Supplementary data