This study used spectroscopic methods to investigate the fate and dynamics of natural organic matter (NOM) as it traverses the treatment train at three water treatment plants (WTPs) in South Africa. The character, quantity, and removability of NOM at specific treatment stages was investigated by measuring changes in dissolved organic carbon (DOC) concentration, specific ultra-violet absorbance, UV absorbance, various spectroscopic indices, and maximum fluorescence intensity levels. A novel method of identifying and quantifying fluorescent fractions by combining synchronous fluorescence spectroscopy (SFS) and Gaussian peak fitting is presented. The dynamics of NOM removal were modeled using 2D-SFS correlation spectroscopy. Humic and fulvic substances dominated coastal plants and were the most amenable for removal by coagulation as shown by Hermanus WTP (plant H), which had a 42% DOC removal at the coagulation stage. Tyrosine-like, tryptophan-like and microbial humic-like substances were degraded or transformed concurrently at plant Flag Bushiole (FB) whereas, at plant H, fulvic-like matter was transformed first followed by tyrosine-like then humic-like matter. Through 2D-SFS, this study revealed that NOM transformation was varied as a consequence of NOM character, the type and dosage of treatment chemicals used, and WTPs operational parameters.
A novel method of identifying and quantifying fluorescent fractions by combining synchronous fluorescence spectroscopy and Gaussian peak fitting is presented.
The character of NOM at source and plant operational parameters play a role in the treatability of NOM.
Synergies of optical methods in tracking changes of NOM fractions due to water treatment processes.
Natural organic matter (NOM) is a complex mixture of organic compounds derived from decaying matter of terrestrial and or aquatic origin found in all natural waters. The source, climatic conditions and land use influence NOM concentrations and reactivity among different natural waters (Moyo et al. 2019). NOM is a precursor of disinfection byproducts (DBPs) such as haloacetic acids (HAAs) and trihalomethanes (THMs), which are carcinogenic. Moreover, NOM removal processes and disposal can be expensive, hence there is need for effective real-time monitoring and control techniques. This implies robust and sensitive analytical methods that can provide rapid information on NOM dynamics, reactivity and treatability need to be developed (Ncibi & Matilainen 2018).
Optical methods such as fluorescence excitation emission matrix (FEEM) spectroscopy with parallel factor (PARAFAC) analysis and ultraviolet-visible (UV-Vis) spectroscopy are widely used to characterize the origin, composition and reactivity of NOM. In particular, UV-Vis spectroscopy has been applied in tracing and tracking the origin, composition and reactivity of chromophoric dissolved organic matter (CDOM) in natural water (Zha et al. 2014). NOM-metal interactions and the generation of DBPs after chlorination have been tracked by the log-transformed absorbance (LnA) spectra, a variation of processed UV-Vis spectra (Li & Hur 2017). FEEM with PARAFAC analysis has been employed to fingerprint fluorescent dissolved organic matter (FDOM) fractions and sources of DOM (Ndiweni et al. 2019). PARAFAC analysis has proved advantageous over other chemometric techniques in analyzing fluorescence data because it can pick up and resolve minute variations in EEM datasets by extracting overlapping autonomous fluorophore groups (Hur & Cho 2012). However, complex mixtures such as NOM restrict the application of conventional fluorescence methods because of extensive spectral overlaps that reduce their sensitivity (Murphy et al. 2011).
To overcome spectral overlap and interference, techniques such as synchronous fluorescence spectroscopy (SFS) can be used. In SFS, individual fluorophores are decoupled from a pool of fluorescent NOM fractions in a sample by selecting an appropriate offset wavelength (Δλ) between excitation and emission wavelengths (Yu et al. 2013). This technique has been successfully used to study the seasonal variation of NOM in a Moorland water (Goslan 2003), and the influence of coagulation and ozonation on the character of effluent organic matter (Jeong et al. 2014). Further selectivity of SFS can be enhanced by combining it with two-dimensional correlation (2D-SFS) to resolve overlapping peaks. This is achieved by extrapolating the peaks over to the second dimension, and discriminating the chronological order of any subtle spectral changes brought about by external perturbations (Su et al. 2016).
Most studies have focused on the use of UV-Vis spectroscopy and fluorescent spectroscopy or SFS in detecting changes of NOM without investigating clear synergies among these methods (Gonçalves-araujo et al. 2016; Li & Hur 2017). Herein, we propose the assessment of NOM using complementary optical methods for tracking the transformation of NOM throughout the treatment trains of selected water treatment plants (WTPs) in South Africa. This study aimed at conducting a comparative assessment of the dynamics of NOM removal using combined optical techniques on samples collected after treatment stages. The objectives were: (1) to investigate the occurrence, distribution and fate of NOM fractions and subsequent transformations brought about by treatment processes; and (2) to determine the synergies of optical methods in tracking changes of NOM fractions due to water treatment processes.
MATERIALS AND METHODS
Water sampling was performed at three water treatment plants, namely: Flag Busheilo (FB) located inland, Hermanus (H) located on the south east coast, and Mthwalume (MT) located on the eastern coast (Figure A1). More information on the plants is found in the supplementary section (Tables A1 and A2). Triplicate samples were abstracted after each water treatment step using clean 1 L glass bottles with Teflon-lined screw caps. Potable multimeters were used onsite to measure conductivity, temperature, turbidity and pH. Ice boxes were used to transport the samples; upon arrival, a 0.45 μm GF/F filters were used to filter the samples thereafter stored at 4 °C and analysis was conducted within 48 h of sampling. The dissolved organic carbon (DOC) of all samples was determined using a total organic carbon analyser (TOC fusion, Teledyne Tekmar).
UV absorbance and fluorescence EEMs
Before analysis, samples were equilibrated to 25 °C then filtered through 0.45 μm GF/F filters. UV-Vis absorbance spectra, simulated synchronous scans at a wavelength offset of Δλ = 60 nm, and fluorescence EEMs were acquired using a fluorescence spectrometer (Aqualog, HORIBA, Jobin Yvon) in the wavelength range 200–800 nm. The excitation interval was set at 2 nm and the emission was recorded between 248.58–830.59 nm with an emission interval of 3.28 nm. The Raman water peak area was used to calibrate and correct the measured intensities and convert to Raman units (RU) from arbitrary units (AU). This was achieved by exciting the Raman water standard at an excitation wavelength of 350 nm and measuring the emission readings between 248.58 to 830 nm (Ndiweni et al. 2019).
Modeling techniques employed to track NOM composition and dynamics
Log-transformed absorbance spectra
The variable xijk represents the ith sample fluorescence intensity set at excitation and emission of k and j wavelength, respectively. The parameter aif varies with the quantity of the fth fluorophore contained in the ith sample (the score), and the emission and excitation spectrum of the fth fluorophore are denoted by the variables bjk and ckf, respectively (the loadings) (Ndiweni et al. 2019). The variable eijk denotes the residual variables of the model while the variable F denotes the number of components making up the fluorophore. Alternating least squares regression procedure was used to fit the model. The maximum fluorescence intensities (Fmax) were used to quantify the fluorophores after each treatment stage.
Two dimensional correlation spectroscopy
Two-dimensional correlation synchronous fluorescence spectroscopy was conducted to determine the rate of compositional variation of NOM due to the impact of treatment using the 2DShige software available from Kwansei-Gakuin University website, Japan. The order of treatment was the external perturbation. Briefly, Noda's rules state that only auto peaks are positive whereas cross peaks can either be positive or negative. A positive cross peak in the synchronous spectra with the wavelength pair λ1/λ2 means changes occurring at λ1 and λ2 are concurrent or synchronous, while a negative cross peak in the synchronous spectra with the wavelength pair of λ1/λ2 means changes occurring at λ1 and λ2 are inverse or asynchronous. Cross peaks can only be found in the asynchronous spectra. A positive cross peak of λ1/λ2 wavelength pair means the pace of change occurring at wavelength λ1 is faster than that at λ2 in the asynchronous spectra. A negative cross peak of λ1/λ2 wavelength pair within the asynchronous spectra implies that changes occurring at λ1 are slower than that at λ2.
RESULTS AND DISCUSSION
Characteristics of NOM components at source using synchronous scan and peak fitting
Gaussian fitting on SFS was processed to identify underlying bands making up the spectra (Figure 1). A near fit of the measured data and the Gaussian distribution bands was observed (R2 > 0.95). All 2D-SFS scans showed a prominent peak and three characteristic shoulders for all sampled water sources. Peak A was located within the wavelength range 260–314 nm, and ascribed to protein-like matter (PLM) (Hur et al. 2011). Peak B was responsible for the second shoulder in the wavelength range of 314–355 nm. This peak is mainly associated with microbial humic-like matter (MHLM) (Yu et al. 2013). The fulvic-like matter (FLM) component (Peak C) was the prominent peak for coastal plants (H and MT) and situated in the wave range 355–420 nm (Yu 2011). The humic-like matter (HLM) (Peak D and E) was in the wavelength range 420–500 nm, and was the third and not so prominent shoulder (Yu et al. 2013). Overall, the SFS scan identified a single peak and three shoulders, defined as PLM, MHLM, FLM, and HLM. The area under the SFS graph in a specific wavelength region has been used by most researches to enumerate the relative quantities of the fluorescent components in that specific wavelength region (e.g., Hur et al. 2011; Yu et al. 2013). This work reports on quantifying fluorescent components as proportional to the area of the resolved Gaussian curve in the respective fluorescence regions.
It was interesting to observe that peaks A and B for plant H (16.83 and 47.71 RU, respectively) doubled that of MT (6.40 and 19.05 RU, respectively), despite MT having higher DOC levels (12.52 mg/L C) and both these plants being located at the coasts. This finding proves the complexity of NOM. This could mean raw water from MT was less microbially impacted at the source than H (FI = 1.52 and FI = 1.38, respectively) (Table A2), suggesting the supply of DOC in MT was perhaps made up of biopolymers, which are non-fluorescent and not easily taken up by microorganisms (Kimura et al. 2018). Strikingly, peak C (FLM) was prominent in plant H (200.94 RU), about ten times more intense than MT and FB (23.01 and 30.59 RU, respectively). These results confirm the hypothesis that the brownish-yellowish colouration of feed water to plant H is characteristic of water laden with fulvic acids, as reported by Nkambule (2012). Of interest, peaks for HLM (Peaks D and E) was almost double those for coastal plants H and MT (59.98 RU and 43.93 RU, respectively) compared to inland plant FB (29.81 RU). This further confirms previous results that coastal plants contain feed water laden with humics (Nkambule 2012). Previous research done in South Africa indicated that water from plant H supports microbial growth and proliferation as shown by high BDOC (5 mg/L C), whereas for the other plants the mean value was 2 mg/L C (Nkambule 2012). This suggests microbial growth is supported by HLM because it acts as a substrate for their growth and proliferation, vis, high BDOC for plant H (Li et al. 2011).
Occurrence and enumeration of fluorescent dissolved organic matter fractions at source
The SOLO software, which is inbuilt into the Aqualog instrument, was used to process fluorescent data for PARAFAC using data from all the drinking water sources. This was performed so as to generate a large data pool and to come up with a universal model to cater for any variance among water sources. The aim was to determine the occurrence and quantity of fluorescent NOM fractions at source and to determine how such findings influence downstream processes with regards to NOM treatability. Figure 2 shows the loadings of the obtained spectra of the components. The identity of contributing fluorophores was ascertained by cross referencing using the OpenFluor database against those obtained globally (Murphy et al. 2014).
A split half criteria was used to validate and establish a four component model (Murphy et al. 2014). According to the OpenFluor database, components consisted of terrestrial HLM, FLM, and PLM and were denoted as C1, C2, C3 and C4, respectively. The Fmax values were used as a measure of the quantity and distribution of the contributing fluorophores. For all water sources, Fmax was higher for C1 and C2 than for C3 and for C4 (Table A2). This could mean C1 and C2 have high quantum turnover efficiency and low reactions to quenching effects than C3 and C4 (Baghoth Sharma & Amy 2010). Of note, Fmax values for C2 and C3 were higher for plant H. The feed water for plant H had a brownish-yellow colouration, mainly signaling the presence of FLM. This is in agreement with the SFS results in the previous section. The occurrence and distribution of Fmax of the tryptophan-like matter (C4) with origins from autochthonous microbial matter is important because previous research indicated that it is largely predominant in wastewater-impacted waters. The tracking of surface water impacted by wastewater is easily distinguished by the spectral signatures of tryptophan-like matter (Pifer & Fairey 2014). The presence of C4 therefore signals deliberate or opportune wastewater contamination.
Removal of bulk and specific NOM fractions at different stages of treatment
Dissolved organic carbon
Geographical location of the plants had a bearing on the levels of DOC available at source (Figure A1 and Figure 3). This implies catchment activities and bio-geochemical processes influence the quantity and quality of DOC found in each of the geographical regions (Ndiweni et al. 2019). The DOC concentration in the final water for all plants was below the limits laid out by the South African National Standards for water quality (SANS 241), which is ≤10 mg/L. The change of NOM quantity and quality as it traverses the treatment train is not adequately traced by DOC analysis alone. However, a reduction in DOC at each treatment stage is a good measure of process efficiency in removing NOM. Coagulation, usually coupled with flocculation, are the major processes attributed to the reduction of DOC in conventional WTPs. Similar treatment processes did not guarantee similar DOC removal efficiencies among the plants (Table A1 in Supplementary Material). This difference could be a function of coagulant type and dosage, and the character of NOM, or a combination of these (Table A1). Plant H uses alum as a coagulant, and had the highest DOC removal efficiency (38.6%). Previous research has shown that the DOC in waters laden with HLM, such as plant H, is easily removed by coagulation (Vasyukova et al. 2013).
There was an uncharacteristic incline in DOC content (5%) at the coagulation stage for FB. This could be the contribution of organic matter from the polymer coagulant used at that plant, which perhaps at the time and point of sampling was not fully mixed. Again there was an uncharacteristic spike in DOC concentration (58%) at the filtration stage for plant MT. This was attributed to clogged sand filters due for backwashing. Previous reports indicate that SUVA values below 2 L.mg−1.m−1 mean the type of NOM is non-humic, and SUVA values greater than 4 imply the greater component of NOM in the sample is humic in character. Research has shown that a 50% DOC removal is expected when the SUVA value is higher than 4 L.mg−1.m−1, and a 25% removal is expected when the SUVA value is below 2 L.mg−1.m−1 (Nkambule 2012). Hence, at the coagulation stage, more than 50% DOC removal was expected at plant H, and <25% from plants FB and MT. For plant H, this was close enough (42%), but for plants FB and MT the DOC removal at coagulation was - 5 and 31% removal, respectively. It must be noted that these removal guidelines apply at optimal conditions and are dependent on the type and dosage of the coagulants, hence the discrepancy from the guidelines by plants FB and MT (Ncibi & Matilainen 2018).
The analysis of NOM is not routine in South African WTPs because its removal in not prioritized. This work contributes to a new body of knowledge on NOM in the South African context, which can be classified together with DBPs as emerging pollutants of concern. Although the DOC concentrations in the treated water were below SANS 241, the critical DOC concentration necessary to offset the formation of DBPs is currently not known, therefore routine monitoring is recommended.
In general, all plants showed a progressive decline in the absorbance at 254 nm from the raw water through to finished final water (Figure 3). From these results it can be inferred that the aromatic composition of the water gradually decreased after every stage for all plants. However, there were anomalies at certain stages of the treatment train where an increase in UV absorbance was observed. For example, at the sedimentation stage for plants FB and MT (11 and 29%, respectively). Perhaps this was due to the accretion of suspended and dissolved UV absorbing microbial byproducts. The sedimentation basin can act as a batch bioreactor allowing the accumulation of bacteria (Cortina & Gonz 2016). The growth and proliferation of bacteria depends on the aerobic conditions, residence time, temperature and agitation in the basin (Cortina & Gonz 2016). Additionally, irregular cleaning of the sedimentation basin allows for the accumulation of bacteria.
Plant H had the highest UV254 absorbance (0.51 cm−1). This means, compared to other plants, the raw water source for plant H was laden with aromatic compounds. From these results, the order of NOM removal measured as UV removal was in the order H (80%) > MT (69%) > FB (32%).
Specific ultraviolet absorbance
Specific ultraviolet absorbance can give a measure of the transphilic, hydrophilic, and hydrophobic character of the NOM in the sample (Nkambule 2012). A SUVA value exceeding 4 L.mg−1.m−1 implies water is hydrophobic, a SUVA value in the range 2 to 4 L.mg−1.m−1 implies water is transphilic, and a SUVA value below 2 L.mg−1.m−1 indicates water of hydrophilic character. Raw water SUVA value for plant H was 7.1 L.mg−1.m−1, implying hydrophobic moieties dominate the NOM in this sample. The SUVA value of the final treated water from that plant was 2.1 L.mg−1.m−1 (Figure 3). These results are in agreement with a previous report that indicates NOM in final treated water in South Africa is transphilic in character (Nkambule 2012). There was no significant difference (p < 0.05) between the raw and finished waters in terms of SUVA change for FB and MT (0.45 and 0.2 unit change, respectively). However, there were significant variations in SUVA along the treatment train for these plants. For example, at the sedimentation stage, there was a 3.5 unit increase of SUVA for MT. This change from hydrophilic character to hydrophobic character could be due to the accumulation of microbial byproducts as discussed in the preceding section. Notably, the SUVA value at coagulation increased by 8 units at the coagulation for plant H, implying the hydrophobic character of NOM increased at this stage. This was unexpected because the hydrophobic NOM fraction has been found to be susceptible to removal by the coagulation process. Such an anomaly could be due to the type of coagulant, in this case alum. The agglomeration and aggregation of NOM as it forms flocs increases its molecular weight and the hydrophobic character (Hoffman et al. 2014).
Selective elimination of chromophoric dissolved organic matter fractions
Apart from a monotonic decrease in absorbance, UV-Vis spectra for drinking water are usually featureless. Chemometric methods such as differential spectra (DS) can be used to process UV-Vis data to show more insights on the activity of CDOM fractions (Yang et al. 2017). Therefore, DS can trace and track latent features that could reveal the dynamics of CDOM properties after critical stages of water treatment. It is noteworthy that DS of a particular treatment process followed the same trend for all plants, therefore respective treatment stages will be discussed together.
All plants showed a decreasing trend from low to high wavelengths at the coagulation stage. At wavelengths less than 290 nm, UV-Vis absorbance removal by coagulation was relatively constant, but at wavelengths greater than 290 nm all plants showed a rapid increase (Figure 4). This indicates that coagulation was effective in removing UV-Vis absorbing NOM. In addition, this could be because of the transformation occuring from DOC to particulate organic matter (POM) as NOM agglomerates into flocs (Yang et al. 2017). A correlation of 0.67 was established when considering the efficiency of coagulation as a unit process to remove total CDOM (tCDOM) with respect to UV254 removal. Such findings were expected because both parameters detect the fraction of NOM that fluoresces. Unexepected, however, was a strong correlation (R = 0.962) between total CDOM removal and DOC removal at the coagulation stage. This could be because DOC measurement is non selective to both fluorescent and non-fluorescent organic matter. Therefore for these plants, total CDOM removal can be used as a surrogate for DOC determination. Similar observations were reported by Wünsch et al. 2015. However, seasonal variations should be assessed to test the valididty of this finding.
In the sedimentation and filtration processes, the intensity of the absorbance showed an increasing trend. At these stages, water is retained for a long enough residence time for microorganisms to acclimatise with the environment and produce enzymes necessary to assimilate the available DOC in the water (Krzeminski et al. 2019). Therefore the increase in UV-Vis absorbance could potentially be due to the preference of microorganisms to assimilate low molecular weight NOM fractions from the water, yielding high molecular weight biomolecules as metabolic byproducts.
Typical for chlorinated water, a steady decline in absorbance over the scanned wavelength range after chlorination was observed. DS revealed a peak in absorbance in the wavelength region 260–270 nm at this stage (Figure 4). Lavonen et al. (2015) reported similar findings and deduced that the absorbance in this range is particularly reactive during disinfection with chlorine or chloroamine. Conjugated double bonds and activated aromatic rings of NOM are disrupted by chlorine at the disinfection stage, resulting in the production of smaller and hydrophilic organic substances (Krasner et al. 2013).
Tracking compositional variations of DOM due to treatment processes using 2D correlation spectra
Three auto-peaks with the wavelength pairs of 265/265; 295/295 and 335/335 nm were observed in the synchronous 2D correlation spectra for plant FB (Figure 5(a)). The results suggest changes that occurred at wavelengths 265, 295 and 335 nm were concurrent. The peaks at excitation wavelengths of 335, 295 and 265 nm are ascribed to microbial humic-like, tryptophan-like and tyrosine-like matter (Su et al. 2016). This implies these NOM fractions were transformed or degraded concurrently down the treatment train. The asynchronous spectra for FB showed positive cross peaks at 380/310, 380/265 and 280/310 nm wavelength pairs (Figure 5(b)), implying the order of change was in the order: 380 → 280 → 265 → 310 nm. The peaks centered at 380 and 310 nm are ascribed to microbial humic-like and tryptophan-like matter while peaks at 280 and 265 are ascribed to tyrosine-like matter (Su et al. 2016). For plant H, the synchronous spectra detected a single auto peak at 365/365 wavelength pair (Figure 5(c)), while the asynchronous spectra detected two positive cross peaks at the wavelength pairs of 365/265 and 450/365 (Figure 5(d)). According to Noda's rule (Su et al. 2016) this means that change at 365 nm is faster than that occurring at 265 nm but slower than that occurring at 450 nm. The peak at 365 nm was ascribed to fulvic-like matter, while peaks at 265 and 450 nm are mainly ascribed to tyrosine-like and humic-like matter, respectively. Therefore the transformation throughout the treatment plant was of the order: 450 → 365 → 265 nm. In the case of plant MT, three auto peaks of wavelength pairs 265/265, 300/300 and 365/365 nm, while two cross peaks of wavelength pairs of 300/265 and 365/280 nm were observed (Figure 5(e)). These results suggest that tyrosine-, tryptophan- and fulvic-like matter were transformed or degraded concurrently throughout the treatment process. Our previous study found tyrosine to be more susceptible to treatment than the other fractions (Moyo et al. 2020). This suggests NOM characteristics, treatment regimen and chemicals used play a role in the ease of removal of NOM fractions.
The asynchronous spectra detected three positive cross peaks at 480/355, 480/295 and 480/265 nm (Figure 5(f)). According to Noda's rule, spectral changes occurring at wavelength 480 are faster than those at 355, 295 and 265 nm. The peak at 480 nm is mainly ascribed to humic-like substances, while those at 355, 295 and 265 nm are ascribed to fulvic-like, tryptophan-like and tyrosine-like matter, respectively (Jarvis 2004).
Correlations relating NOM removal and spectroscopic parameters
A visual Spearman correlation matrix was generated using the XLSTAT statistical software (Figure 6). A correlation matrix gives a visual perspective of correlation data whilst at the same time returning the essential elements of statistical inference. Briefly, patterns are generated which identify both the sign and the intensity if correlations exist: positive correlations are identified by lines that slant from bottom left to top right whilst a negative correlation is depicted by lines which slant from bottom right to top left. The closer to each other the lines are, the closer the correlation is to zero. A mild correlation between the humification index (HIX) and UV254 removal was established (R = 0.796) (Figure A.2a). The HIX is associated condensed fluorescence organic matter, hence lower H:C ratios therefore serve as a perfect surrogate for humicity, suggesting its changes to be correlated to UV254 absorbance coefficient (Kamjunke et al. 2016). HIX for the raw water serves as an indicator of its susceptibility to removal by downstream processes. Raw water with a high HIX value corresponds with high quantities of humic substances. Humic substances are known to be easily removed by coagulation. Thus it is expected that higher HIX must translate to a larger NOM removal. Again the freshness index (β:α) correlated fairly with UV254 removal (R= 0.786) (Figure A.2b). The freshness index (ß:α), therefore changes in the ß:α ratio are expected to correlate with UV254 absorbance coefficient (Kamjunke et al. 2016). Conventional methods such as coagulation can easily remove NOM with low β:α values. A low β:α index indicates condensed and aged humic matter. The fluorescence index (FI) gives the ratio of terrestrially derived organic matter, usually lower values (e.g., degraded soil and plant organic matter) and microbially derived organic matter, usually higher values (e.g., leachate from algae and bacteria; extracellular release), therefore it is expected that the UV254 absorbance coefficient will be relative to the FI ratio (Hansen 2014). Previous research reports that terrestrially sourced NOM (FI>1.7) is less likely to be removed by conventional methods than microbially sourced NOM (FI<1.3) (Figure A.2c) (Yu et al. 2015). Specific ultraviolet absorbance (SUVA) is calculated as the quotient of UV254 absorbance coefficient and the DOC of the sample (Pifer & Fairey 2012). Therefore by definition it is expected that SUVA changes should correlate with UV254 absorbance. The UV254 reduction and SUVA gave a fair relationship (R= 0.745). Spectroscopic indices such as HIX and FI showed strong correlations that exceeded 0.9 with the Fmax values of C1 and C4. This could be because HLM and PLM contains aromatic groups. Aromatic compounds such as humic-like and protein-like NOM fractions tend to be condensed, inferring NOM fractions with a high level of humicity commensurate with HIX and FI spectroscopic ratios (Kamjunke et al. 2016).
Most conventional WTPs regard DOC and UV254 as the most appropriate surrogate measure of NOM quantity. However, measurement of DOC involves the use of expensive instruments and chemicals. There is therefore need for easily measurable parameters that give good correlations with DOC so as to act as a surrogate measure for DOC. The results indicate spectroscopic indices and DOC had a poor correlation. This was because measurement of UV254 and spectroscopic ratios rely on NOM moieties that fluoresce or absorb in the UV-Vis range, whereas DOC measurement is non-selective of the constituting organic matter (Baghoth Sharma & Amy 2010).
The work was motivated by the need to use optical methods to gain insightful understanding of the character of NOM at the three water plants in South Africa with the overarching aim of process optimization. The study revealed that NOM undergoes changes as the water is subjected to different treatment stages. Through 2D-SFS, it was observed that NOM transformation varied as a consequence of NOM character, the type and dosage of treatment chemicals used, and WTPs' operational parameters.
The key findings were:
Humic and fulvic substances dominated coastal plants and were the most amenable for removal by coagulation, as shown by plant H, which had a 42% DOC removal at the coagulation stage.
The character of NOM at source, coagulation chemicals used, and WTPs' operational parameters play a role in the treatability of NOM, as evidenced by the 2D-SFS data. Tyrosine-like, tryptophan-like and microbial humic-like substances were degraded or transformed concurrently at plant FB whereas at plant H, fulvic-like matter was transformed first followed by tyrosine-like then humic-like matter.
Because there was a poor correlation between spectroscopic indices and DOC, further research needs to focus on determining easily measurable parameters for use as NOM surrogates on a routine monitoring basis such as turbidity, conductivity and pH
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.