The potential application of fluorescence spectroscopy for monitoring of organic matter concentration and character at four water treatment facilities was investigated. Results are presented showing impacts on natural organic matter (NOM) due to intake location on the same water body and from individual unit processes including ozonation, granular-activated carbon filtration, and coagulation/flocculation. For validation and comparison of fluorescence methods, organic matter was quantified and characterized using liquid chromatography-organic carbon detection (LC-OCD). Principal component analysis (PCA) and parallel factors analysis were used for dimensionality reduction and to represent individual organic components observed through fluorescence excitation-emission matrices. Fluorescence results generally agreed with LC-OCD characterization, indicating that complete treatment reduced organic concentrations and preferential removal of humic-like material was associated with coagulation/flocculation. PCA results indicated higher concentrations of humic-like material at the Island water treatment plant intake that was not well reduced by inline polyaluminum chloride coagulation and direct filtration. Through fluorescence spectroscopy, ozonation increased Rayleigh scattering, which is correlated to small colloidal/particulate concentrations. Full-scale results from four water treatment plants presented demonstrate that fluorescence methods can characterize NOM, providing similar identification of trends to LC-OCD, with possible online application and use in real-time water treatment process control.

INTRODUCTION

The monitoring of natural organic matter (NOM) has proved to be of great importance in drinking water treatment, for example playing a role in disinfection byproduct formation and membrane fouling (Peiris et al. 2009). For both surface and ground waters, NOM consists of a wide array of compounds including humic acids, carbohydrates, and proteins. This diversity is problematic when considering removal techniques since there is evidence that both functional groups and molecular weight distributions of compounds are integral to determining treatability (Her et al. 2003). Two promising techniques that have been shown to provide information regarding multiple NOM fractions include liquid chromatography-organic carbon detection (LC-OCD) (Huber et al. 2011) and multivariate analysis of fluorescence excitation-emission matrices (FEEM) (Peiris et al. 2010; Bieroza et al. 2011).

Fluorescence spectra represent fluorescence intensity values collected at multiple iterations of excitation and emission wavelengths (Her et al. 2003). The position (excitation-emission region) of a given peak can be correlated with specific compound groupings and the maximum intensity of the peak correlated with concentration (Bieroza et al. 2011). To further extract spectral information, multivariate analysis techniques have been used to track changes in NOM composition. Common techniques applied include two-way principal component analysis (PCA) and multi-way parallel factor analysis (PARAFAC) (Peiris et al. 2010). Fluorescence-based measures offer an opportunity to implement online monitoring of NOM fractions, which could better aid treatment process control and understanding operational impacts at water treatment facilities. Current online NOM measures typically only provide estimates of total concentrations, such as total organic carbon (TOC) or ultraviolet absorbance. Characterization techniques including LC-OCD or resin fractionation are complex, expensive, and are not implementable for online monitoring. Recent studies have investigated this application using PARAFAC analysis (Sanchez et al. 2014). Baghoth et al. (2011) tracked changes to PARAFAC component results, LC-OCD fractions, and other NOM measures resulting from several treatment processes including, but not limited to, rapid sand filtration, ozonation, biological activated carbon filtration, and coagulation. They demonstrated correlations of LC-OCD fractions and PARAFAC components and also identified unique process impacts on individual fractions, indicating the possible use of fluorescence spectroscopy for online monitoring. Sanchez et al. (2014) conducted a long-term study demonstrating the suitability of fluorescence/PARAFAC for coagulant selection based on optimizing organic concentration reduction.

The purpose of this study was to investigate the application of fluorescence spectral analysis for water treatment monitoring by tracking spatial and treatment impacts on NOM at four water treatment plants (WTPs) (F.J. Horgan, Island, R.C. Harris, and R.L. Clark). The overall goal is to establish the validity and understand limitations of fluorescence methods for water quality monitoring. The data and results in this work are intended to complement and extend other recent full-scale studies that have tracked changes in NOM due to treatment processes using PARAFAC analysis. The work presented is novel for simultaneous collection of samples at four WTPs, application and comparison of PCA and PARAFAC analysis, and employing a measure of Rayleigh scattering that has been linked to colloidal/particulate material (Peiris et al. 2010).

METHODS

Water treatment plant characteristics

A summary of the intake depth and length, as well as treatment processes at each of the four WTPs, is shown in Table 1.

Table 1

Intake locations and treatment processes

Treatment facilityIntake depth (m)/distance from shore (m)CoagulationCoagulantFilter media
Horgan WTP 18/2900 Inline PACl in winter, alum (above 10 °C) in addition to polyelectrolyte 1.5–2.2 m GAC above 0.25 m sand 
Harris WTP 15/2200
15/2100 
Hydraulic flocculation, sedimentation Alum 0.25–0.30 m GAC or anthracite over 0.25 m sand 
Island WTP 83/4850
83/4650
83/4700 
Inline PACl 0.91 m anthracite over 0.25 m sand 
Clark WTP 11/1600 Mechanical flocculation, sedimentation Alum 0.46 m anthracite over 0.31 m sand 
Treatment facilityIntake depth (m)/distance from shore (m)CoagulationCoagulantFilter media
Horgan WTP 18/2900 Inline PACl in winter, alum (above 10 °C) in addition to polyelectrolyte 1.5–2.2 m GAC above 0.25 m sand 
Harris WTP 15/2200
15/2100 
Hydraulic flocculation, sedimentation Alum 0.25–0.30 m GAC or anthracite over 0.25 m sand 
Island WTP 83/4850
83/4650
83/4700 
Inline PACl 0.91 m anthracite over 0.25 m sand 
Clark WTP 11/1600 Mechanical flocculation, sedimentation Alum 0.46 m anthracite over 0.31 m sand 

Alum: aluminum sulfate; PACl: polyaluminum chloride; GAC: granular-activated carbon; Inline: addition of coagulant directly prior to media filtration.

Sampling

There were two distinct sampling phases to this study. Phase I consisted of an assessment of NOM spatial changes at the intake of all four plants as well as due to overall treatment. Phase II focused on the Horgan WTP and examined samples before and after major unit processes such as ozonation and activated carbon filtration. During Phase II six sampling points at the Horgan WTP included raw water, post-ozonation, settled water (post-coagulation/settling), post-granular-activated carbon (GAC) filtration (new GAC and older GAC), and treated water (post-chlorination) as it entered the distribution system. New GAC was replaced in the summer of 2012 (June–July), at the end of the Phase I study. The older GAC varied in age (2+ years), but was assumed to be fully exhausted in terms of adsorption potential. In total, 140 samples were collected during Phase I between April and July on a roughly biweekly basis. During Phase II, 138 samples were collected every 2 weeks from September to July. LC-OCD samples were less frequent, collected once per month.

Total organic carbon

Organic carbon measurements were based on the wet oxidation method as described in Standard Method 5310 D (APHA 2005) using an O.I. Analytical Aurora 1030 organic carbon analyzer (College Station, TX, USA).

Liquid chromatography-organic carbon detection

LC-OCD analyses were conducted using the method described by Huber et al. (2011). LC-OCD allows for simultaneous quantification of several organic matter fractions. It consists of a DOC analyzer preceded by chromatographic separation on the basis of molecule size. The method allows for quantification of five molecule weight ranges labeled biopolymers, humics, building blocks, low molecular weight (LMW) acids, and LMW neutrals. Further description of these classes can be found in Huber et al. (2011). Samples were passed through a 0.45 μm filter to remove any particulates prior to analyses. A weak cation exchange column (250 × 20 mm, Toso, Japan) provided chromatographic separation of the NOM fractions. The mobile phase used was a phosphate buffer (MLE, Dresden, Germany) exposed to UV at a flow rate of 1.1 mL/min. Calibration was performed on-site using potassium hydrogen phthalate. Data acquisition and processing was conducted using a customized software program (ChromCALC, DOC-LABOR, Karlsruhe, Germany).

Fluorescence spectra collection and analysis

FEEM were collected using a Perkin Elmer Luminescence Spectrometer LS50B (Waltham, MA, USA). Collection of intensity values occurred at 10 nm excitation wavelength increments and 1 nm emission wavelength increments. Excitation-emission ranges were 250–380 and 300–600 nm, respectively. Scan rate was set to 600 nm/min, slit width was set to 10 nm, and photomultiplier tube voltage was set to 775 V. The instrument settings were determined based on ranges used in previous studies (Bieroza et al. 2011), protocols that increase resolution (Peiris et al. 2009), and in-house testing to optimize FEEM collection. UV-grade polymethylmetacrylate (PMMA) cuvettes with four optical windows were used (VWR, Mississauga, Canada). It has been shown that while PMMA cuvettes reduce the intensity of excitation wavelengths below 285 nm, this approach is acceptable for the purposes of distinguishing NOM elements using fluorescence (Peiris et al. 2010). To account for scattering interference, blank subtraction was performed using spectra collected of Milli-Q© water using the same instrument settings (Her et al. 2003). Inner filtering effects were not taken into account due to the low organic content (Stedmon & Bro 2008) of all water samples (TOC < 2.61 mg/L; UV absorbance at 254 nm < 0.03 cm−1).

PCA was applied and scattering volumes calculated using in-house scripts written in the R language (Ihaka & Gentleman 1996). PARAFAC was carried out using the n-way Toolbox in MATLAB V7.12.0 (Andersson & Bro 2000). More detailed description of the process for interpretation of the multivariate results can be found in Peiris et al. (2010) and Andersen & Bro (2003).

RESULTS AND DISCUSSION

Fluorescence spectra

The fluorescence spectra for each of the four water intakes exhibited similar characteristics. Two main peaks a and b, along with a pronounced shoulder peak c, were identified from the spectral plots of fluorescence intensities (examples shown in Figure S1 of the Supplementary material, available online at http://www.iwaponline.com/ws/015/013.pdf). Comparison with published work showed peak a to represent humic-acid-type matter (Ex/Em: 270/450 nm) whereas peak b was consistent with fulvic-acid-type material common to fresh waters (Ex/Em: 340/440 nm) (Chen et al. 2003; Murphy et al. 2008). Peaks of high intensity were observed in both first- and second-order Rayleigh scattering regions (Ex/Em: 300–380/300–380 nm and Ex/Em: 250–300/500–600 nm), which are related to colloidal and particulate concentrations (Peiris et al. 2010). Of note is Rayleigh scattering which occurs from molecules and particles that are much smaller than the incident wavelength of light (Jonasz & Fournier 2007), therefore, in the context of this work << 200 nm. The extended shoulder peak c, between emission wavelengths of 350 and 400 nm is believed to result from protein-like material (Chen et al. 2003). The excitation/emission location of peaks a and b were highly consistent (5 nm emission; equal excitation) among raw waters of the four intakes. The conformity of peak locations provided evidence of similar organic character.

Analysis of fluorescence spectra

By applying both PCA and PARAFAC as fluorescence spectra analysis tools, a direct comparison can be made between methods. Rayleigh scattering peaks were removed within a 20 nm area prior to analysis in an effort to increase the representation of dissolved organic matter components using PCA or PARAFAC. The volume contained by Rayleigh peaks (regional integration) was calculated as a separate measure.

Multivariate analysis was applied independently to several data subsets. For Phase I, two data sets were analyzed separately: pI1 – raw water (68 samples) and pI2 – raw and treated water (140 samples). For Phase II, samples collected at all sampling points were analyzed as one data set (pII1: 138 samples). Organic fractions represented by principal components (PCs) and PARAFAC components for all data subsets were observed to be similar. The first 4 PCs (each representing over 2% of explained variance), cumulatively explained 92.2%, 93.0%, and 82.8% of the variance in pI1, pI2, and pII1 sample sets, respectively.

The physical meaning of each principal component was determined through analysis of loading plots presented in Figure S2 of the Supplementary material (available online at http://www.iwaponline.com/ws/015/013.pdf). The loading plot for PC1 displayed high values in areas representing humic-like substances. PC2 loading values indicated negative representation of humic- and fulvic-like components and positive representation of protein-like material. The loading plot for PC3 showed pronounced negative representation of protein-like components and PC4 represented a component possibly representative of polyaromatic hydrocarbons (PAH) (Ex/Em: 340/400 nm) (Murphy et al. 2008). PCA applied to all data subsets resulted in similar compounds represented by PC1–PC3. Representation of PAH through PC4 was unique to the Phase I data sets, possibly indicating PAH were only present in Phase I sample(s). PCA produces relatively vague and unsupervised organic class representations which can be difficult to interpret and are not consistent across differing data sets. To account for rotational freedom of PCs, scores were adjusted for positive representation of the major component.

Optimal PARAFAC components were determined through both an observation of sum of squared errors, core consistency, and split-half analysis (Andersen & Bro 2003). An optimized PARAFAC model including two components under no constraints was determined for each data set. Component 1 (C1) represented humic- and fulvic-like compounds whereas component 2 (C2) represented protein-like material through the observed tryptophan peak (Ex/Em: 280/360) (Figure S2 in Supplementary material). Components identified from each data set conformed well.

Phase I: analysis of raw water spatial and temporal differences

Raw water TOC results associated with the intakes at all four water treatment facilities varied between 1.85 and 2.61 mg/L. Periods of increasing or decreasing trends in TOC levels were generally experienced simultaneously by all four sources. A repeated measure ANOVA (analysis of variance) was used to determine significant differences among the four source waters while accounting for temporal changes in water quality. Post hoc testing was carried out using a pairwise t-test when ANOVA results indicated significant differences. The raw water TOC from the Island WTP intake was significantly lower than the other three treatment plants (95% confidence). The maximum absolute difference (between the Island WTP and Harris WTP) was 0.13 mg/L TOC and considered to be low.

Differences in organic character were first examined using LC-OCD. Concentrations of all fractions were observed to be significantly equal, including overall DOC, at a 95% confidence level. Scores from PCA and PARAFAC analysis of data set pI1 (Figure 1) show the intake water quality for each plant to be highly variable, possibly indicating higher sensitivity to changes in organic matter concentrations. Humic-like material represented by PC1 and C1 was found to be higher at the Island WTP intake when compared to all other sources based on paired t-tests (p < 0.02). Paired t-tests indicated that the PC3 score was only significantly different between the Harris WTP and the Island WTP (p = 0.03). No significant differences were found using C2 scores. Scattering scores from all sources showed no differences in overall trends, although some unique differences were seen on certain sampling days (Figure 1).

Figure 1

Raw water PC1/C1 (humic-like) and PC3/C2 (protein-like) scores and Rayleigh scattering total area vs. time. Data have been smoothed using local polynomial regression fitting to provide better representation of trends.

Figure 1

Raw water PC1/C1 (humic-like) and PC3/C2 (protein-like) scores and Rayleigh scattering total area vs. time. Data have been smoothed using local polynomial regression fitting to provide better representation of trends.

Phase I: analysis of treatment impacts

The impact of overall treatment processes at each of the four plants was illustrated by calculating the difference between treated and raw water values (difference = raw – treated). Positive difference values indicated that treatment reduced the concentration of the NOM fraction. ANOVA was carried out on the calculated differences from all four plants to observe significant differences in treatment impacts. Statistical analysis did not include data collected beyond 19 June due to process and equipment changes at the Horgan WTP. At this time, ozonation commenced and the GAC filter media was replaced, which increased TOC removal by approximately 0.3 mg/L.

The Horgan, Harris, and Clark WTPs reduced TOC by similar amounts (0.16–0.21 mg/L), while the Island plant had a statistically significantly smaller effect (0.04 mg/L TOC reduction) (p < 0.04). During Phase I, the Island plant was unique in using PACl rather than alum as a coagulant. Furthermore, intake water quality monitoring indicated higher representation of PC1 and C1 components at the Island plant. Although inconclusive without more focused studies, possibly organics represented by the fluorescence components are not readily treated by coagulation or PACl. The observed TOC reduction (∼8–11%) at all facilities except the Island WTP (2% reduction) was comparable to treatment performance reported in the literature for plants receiving lake waters with low organic content (2–4 mg/L TOC) (Volk et al. 2000; Uyak & Toroz 2007).

Resulting changes of NOM fractions due to treatment, as analyzed by LC-OCD, are shown in Table 2. Of note is the lack of significant differences between raw and treated water due to treatment from the Island WTP, indicating the ineffectiveness of inline coagulation using PACl for organic matter concentration reduction on this water intake. However, the Island WTP is the exception and the findings are supported by previous studies which indicate that coagulation most significantly affects humic substances (Baghoth et al. 2011) and biopolymers (Volk et al. 2000).

Table 2

Significant changes in LC-OCD fractions between raw and treated water

PlantDOCBiopolymersHumicsBuilding blocksLMW neutralsLMW acids
Horgan WTP Yes (−0.11 mg/L) No Yes (−0.06 mg/L) Yes (−0.05 mg/L) No No 
Island WTP No No No No No No 
Harris WTP Yes (−0.17 mg/L) No Yes (−0.13 mg/L) No No No 
Clark WTP Yes (−0.24 mg/L) Yes (−0.07 mg/L) Yes (−0.09 mg/L) Yes (−0.09 mg/L) No No 
PlantDOCBiopolymersHumicsBuilding blocksLMW neutralsLMW acids
Horgan WTP Yes (−0.11 mg/L) No Yes (−0.06 mg/L) Yes (−0.05 mg/L) No No 
Island WTP No No No No No No 
Harris WTP Yes (−0.17 mg/L) No Yes (−0.13 mg/L) No No No 
Clark WTP Yes (−0.24 mg/L) Yes (−0.07 mg/L) Yes (−0.09 mg/L) Yes (−0.09 mg/L) No No 

Yes = significantly different (95% confidence); No = significantly similar (95% confidence).

Differences between raw and treated water scores from PCA and PARAFAC were also calculated (Figure 2). It is important to note that score values reported from these multivariate analysis methods were not calibrated to external standards. As such they should be taken in the context of values within the same data set. PC1 and C1 scores from the Horgan WTP showed that humic-like substances were reduced throughout the entire study period. Harris WTP PC1 scores were found to be significantly reduced by treatment (95% confidence); however, C1 was not affected. All other treatment plants were observed to have insignificant effects on PC1 or C1, as determined through t-tests. Fluorescence results show differing trends from LC-OCD, with greatest humic removal observed at the Horgan WTP and no humic removal at the Clark WTP. These results indicate unique organic class representation between these methods. This is not unexpected since LC-OCD separates organic compound classes based on molecular size and hydrophobicity, while fluorescence relies on responses from organic fluorophores which typically possess aromatic functional groups.

Figure 2

Average difference in PC and PARAFAC component scores and scattering volume due to treatment. Note: Vertical bars represent one standard deviation; n = 15 – 16.

Figure 2

Average difference in PC and PARAFAC component scores and scattering volume due to treatment. Note: Vertical bars represent one standard deviation; n = 15 – 16.

In terms of protein-like material (PC3/C2), no significant effects were observed with respect to PC3 at any of the facilities. In contrast, a marked decrease in C2 score was associated with treatment at the Horgan WTP and a lesser, although significant, decrease was also noted at the Harris WTP. These two WTPs are distinct because of their use of GAC in the filter beds. Furthermore, Horgan WTP GAC filter layers are deeper (1.5–2.2 m), where greater protein- and humic-like removal was observed, as opposed to 0.25–0.30 m at the Harris WTP. As with humic-like results, disagreement between LC-OCD and protein-like fluorescence was observed, confirming unique organic class representation between the two methods. Furthermore, this disagreement implies differing impacts of treatment on aromatic amino acids, through fluorescence, and large hydrophilic biopolymers, which includes both polysaccharides and proteins, represented through LC-OCD. Disagreement of protein-like trends identified by PC3 and C2 imply distinct NOM constituents represented through PCA and PARAFAC. Scattering volumes, which have been identified to provide representation of colloidal concentrations (Peiris et al. 2010), were only significantly affected at the Harris WTP.

Phase II: impact of specific unit processes

To facilitate an in-depth assessment of specific treatment processes and identify the resulting impact of process changes at the Horgan WTP, additional sampling of TOC, LC-OCD, and FEEM was conducted. An outline of Phase II sampling points at the Horgan WTP is given in the methods section. Sampling at the Clark WTP examined the performance of conventional unit processes: raw water, settled water, filtered water, and treated water.

In terms of TOC, significant reductions were associated with ozonation (0.08 mg/L), new vs. old GAC (0.29 and 0.28 mg/L), and from overall treatment (0.22 mg/L) at the Horgan WTP. Results from LC-OCD indicated few significant changes resulting from any given treatment process (95% confidence) at the Horgan WTP. Ozonation significantly increased biopolymers (20 μg/L) and reduced humic substances (60 μg/L). Overall, treatment significantly decreased bio-polymer concentrations (30 μg/L), humic substances (111 μg/L), and building blocks (52 μg/L) which are thought to be breakdown products of humic substances (Huber et al. 2011). There was a high degree of variability in treatment effects on LC-OCD NOM fractions, indicating treatment impacts on these components were inconsistent over the study period.

Analysis using fluorescence-based measures showed similar high variability to LC-OCD results, particularly PARAFAC results. For the Horgan WTP (Figure 3), ozonation showed significant reduction of PC1/PC2 (humic-like) removal and increased scattering. Preferential impact of ozonation of humic-like fluorescent components was also observed by Baghoth et al. (2011). The observation of increased scattering volumes, which are correlated with concentrations of colloids or molecules much smaller than the excitation wavelength, as a result of ozonation possibly results from the breakdown of high molecular weight organics into smaller fragments (Świetlik & Sikorska 2004). Coagulation/settling significantly reduced C1 (humic-like) and C2 (protein-like) scores even though PC1/PC2/PC3 was not affected. Baghoth et al. (2011) and Sanchez et al. (2014) indicated preferential removal of humic-like substances by coagulation. No significant difference between C1 and C2 removals from coagulation were observed in this study. The older GAC was found to significantly reduce PC1 and C1 (humic-like) scores, scattering, and C2 (protein-like) material. In contrast, new GAC was only found to significantly affect PC1 and scattering components. Biofiltration is utilized at the Horgan WTP and removal of protein-like material exclusive to older GAC is possibly due to a more established biological community in the older filters compared to media replaced at the very beginning of Phase II. Removal of protein-like material through biofiltration has been noted in other studies (Peldszus et al. 2011). New GAC treatment impacts on C1 and C2 were much more variable when compared to older GAC and offered no improved treatment performance (Figure 3). It is proposed that the variability in new GAC resulted from decreased adsorptive capacity with time, although no significant trend was evident based on C1/C2 score vs. time (data not shown). PC1/PC2 scores and scattering were significantly reduced as a result of overall treatment (raw – treated), while other parameters were not significantly affected.

Figure 3

Average difference in PCA/PARAFAC scores due to unit processes at the Horgan WTP, n = 16 – 18. Error bars represent one standard deviation.

Figure 3

Average difference in PCA/PARAFAC scores due to unit processes at the Horgan WTP, n = 16 – 18. Error bars represent one standard deviation.

CONCLUSIONS

Analysis of raw water organic composition at four WTPs demonstrated strong similarities despite differences in spatial and temporal characteristics. While LC-OCD results did not show significant differences, fluorescence-based measures (PCA and PARAFAC) indicated higher humic concentrations at the Island WTP which had an intake that was both deeper and further from shore. Overall treatment assessments from Phase I exhibited lower NOM removal from the Island WTP indicating inline coagulation and dual media filtration was not effective for organic reduction at this intake, although scattering was reduced.

When considering individual unit processes (Phase II), few significant changes to NOM character were observed and the variability of treatment performance was high as noted by established (LC-OCD and TOC) and fluorescence-based measures. Pre-ozonation caused an increase in scattering, while coagulation/flocculation reduced humic and protein-like components. Overall, GAC was observed to play the largest role in humic-like and particulate removals at the Horgan WTP. In comparison to new GAC, performance of older GAC was more stable and reduced protein-like components more effectively. Treatment performance analyzed using PARAFAC component scores appeared to be more variable when compared to PCA either indicating increased noise or increased sensitivity; however C1 and C2 provided a less ambiguous representation of specific NOM constituents.

In summary, this work compared fluorescence-based measures as a rapid NOM characterization technique for low organic content source waters. When compared to LC-OCD, fluorescence measures identified similar effects on generalized organic fractions due to treatment and demonstrated increased sensitivity to differences in raw water character. Although the degree of variability in results is high, the similarity in trends to LC-OCD indicates the potential suitability of fluorescence as an online and rapid NOM characterization method. Furthermore, inclusion of scattering changes associated with particulates and colloids is not possible when using LC-OCD or TOC. Although implementation of fluorescence for online water treatment monitoring is promising, this study has identified several issues with variability in performance assessment. Long-term online studies, where samples are collected at much shorter time intervals, are needed as a next step to understanding the application of this method in full-scale water treatment monitoring.

ACKNOWLEDGEMENTS

This work was funded in part by the Canadian Water Network and the Natural Sciences and Engineering Research Council of Canada (NSERC) Chair in Drinking Water Research at the University of Toronto. We would like to thank David Scott (Toronto Water) for his help with this work as well as the staff at F.J. Horgan, R.L. Clark, Island, and R.C. Harris water treatment plants. We are grateful to Dr Monica Tudorancea and Dr Sigrid Peldzsus (University of Waterloo) for performing LC-OCD analyses.

REFERENCES

REFERENCES
Andersen
C. M.
Bro
R.
2003
Practical aspects of PARAFAC modeling of fluorescence excitation-emission data
.
Journal of Chemometrics
17
(
4
),
200
215
.
Andersson
C. A.
Bro
R.
2000
The N-way Toolbox for MATLAB
.
Chemometrics and Intelligent Laboratory Systems
52
(
1
),
1
4
.
APHA
2005
Standard Methods for the Examination of Water and Wastewater
, 21st edn.
American Public Health Association (APHA)/American Water Works Association/Water Environment Federation
,
Washington, DC
,
USA
.
Chen
W.
Westerhoff
P.
Leenheer
J. A.
Booksh
K.
2003
Fluorescence excitation-emission matrix regional integration to quantify spectra for dissolved organic matter
.
Environmental Science & Technology
37
(
24
),
5701
5710
.
Ihaka
R.
Gentleman
R.
1996
R: A language for data analysis and graphics
.
Journal of Computation and Graphical Statistics
5
(
3
),
299
314
.
Jonasz
M.
Fournier
G.
2007
Light Scattering by Particles in Water: Theoretical and Experimental Foundations
.
Academic Press
,
London
.
Peiris
B. R. H.
Budman
H.
Moresoli
C.
Legge
R. L.
2009
Acquiring reproducible fluorescence spectra of dissolved organic matter at very low concentrations
.
Water Science & Technology
60
(
6
),
1385
1392
.
Sanchez
N. P.
Skeriotis
A. T.
Miller
C. M.
2014
A PARAFAC-based long-term assessment of DOM in a multi-coagulant drinking water treatment scheme
.
Environmental Science & Technology
48
(
3
),
1582
1591
.

Supplementary data