Page 1
APPLICATIONS OF PRINCIPAL COMPONENT ANALYSIS OF
FLUORESCENCE EXCITATION-EMISSION MATRICES FOR
CHARACTERIZATION OF NATURAL ORGANIC MATTER
IN WATER TREATMENT
by
Nicolás Miguel Peleato
A thesis submitted in conformity with the requirements
for the degree of Master of Applied Science
Graduate Department of Civil Engineering
University of Toronto
© Copyright by Nicolás Miguel Peleato 2013
Page 2
ii
APPLICATIONS OF PRINCIPAL COMPONENT ANALYSIS OF FLUORESCENCE
EXCITATION-EMISSION MATRICES FOR CHARACTERIZATION OF NATURAL
ORGANIC MATTER IN WATER TREATMENT
Nicolás Miguel Peleato
Master of Applied Science, 2013
Graduate Department of Civil Engineering
University of Toronto
ABSTRACT
Quantification of natural organic matter (NOM) in water is limited by the complex and
varied nature of compounds found in natural waters. Current characterization techniques, which
identify and quantify fractions of NOM, are often expensive and time consuming suggesting the
need for rapid and accurate characterization methods. In this work, principal component analysis
of fluorescence excitation-emission matrices (FEEM-PCA) was investigated as a NOM
characterization technique. Through the use of jar tests and disinfection by-product formation
tests, FEEM-PCA was shown to be a good surrogate for disinfection by-product precursors.
FEEM-PCA was also applied in order to characterize differences in humic-like, protein-like, and
Rayleigh scattering between multiple source waters and due to differing treatment processes. A
decrease in Rayleigh scattering influence was observed for a deep lake intake, and multiple
processes were found to significantly affect humic-like substances, protein-like, and Rayleigh
scattering fractions.
Page 3
iii
AKNOWLEDGEMENTS
This work was funded in part by the Natural Sciences and Engineering Research Council
of Canada (NSERC) Char in Drinking Water Research, as well as through an Ontario Graduate
Scholarship.
I would like to thank my supervisor, Professor Robert C. Andrews, for his guidance,
expertise, and support of my research. My thanks also go out to Professor Susan Andrews for
her assistance with analytical chemistry and troubleshooting chromatography issues. Assistance
from Jennifer Lee, Jim Wang, John Armour, Sabrina Diemert, Dana Zheng, Liz Taylor-
Edmonds, and Alex Balmus was also greatly appreciated.
This work would not have been possible without invaluable support from the
Peterborough Utilities Commission (PUC) and Toronto Water with special thanks to Kevan
Light, René Gagnon, and David Scott.
Finally, my appreciation goes out to my parents, family, friends, and Aidan for their love
and continual support.
Page 4
iv
TABLE OF CONTENTS
ABSTRACT ............................................................................................................................... ii
AKNOWLEDGEMENTS ......................................................................................................... iii
TABLE OF CONTENTS .......................................................................................................... iv
LIST OF TABLES .................................................................................................................. viii
LIST OF FIGURES .................................................................................................................... x
NOMENCLATURE ................................................................................................................. xii
1 Introduction ........................................................................................................................... 1
1.1 Background .................................................................................................................. 1
1.2 Objectives ..................................................................................................................... 2
2 Literature Review .................................................................................................................. 3
2.1 Disinfection by-products .............................................................................................. 3
2.1.1 DBP formation theory ........................................................................................... 3
2.1.2 Modelling DBP formation .................................................................................... 4
2.2 Measurement techniques for natural organic matter .................................................... 7
2.2.1 Liquid chromatography – organic carbon detection ............................................. 8
2.2.2 Fluorescence excitation-emission matrices ........................................................... 8
2.3 Multivariate analysis of fluorescence spectra ............................................................ 11
2.4 Principal component analysis ..................................................................................... 12
2.4.1 Mathematical definition and derivation .............................................................. 12
2.5 Application to natural waters ..................................................................................... 14
3 Materials and methods ......................................................................................................... 15
3.1 Experimental protocols .............................................................................................. 15
3.1.1 Jar tests and DBP formation ............................................................................... 15
3.1.2 Peterborough pilot plant ...................................................................................... 17
3.2 Analytical methods .................................................................................................... 19
Page 5
v
3.2.1 Quality control .................................................................................................... 19
3.2.2 Total and dissolved organic carbon .................................................................... 19
3.2.3 Ultraviolet absorbance ........................................................................................ 20
3.2.4 Liquid chromatography – organic carbon detection ........................................... 22
3.2.5 Chlorine residuals ............................................................................................... 23
3.2.6 pH measurements ................................................................................................ 24
3.2.7 Fluorescence excitation-emission matrices ......................................................... 24
3.2.8 Principal component analysis ............................................................................. 24
3.2.9 Trihalomethanes .................................................................................................. 25
3.2.10 Haloacetic acids .................................................................................................. 29
4 Disinfection By-Product Modelling using Principal Component Analysis of Fluorescence
Excitation-Emission Matrices ............................................................................................. 36
4.1 Introduction ................................................................................................................ 36
4.2 Method overview ....................................................................................................... 38
4.2.1 Source waters ...................................................................................................... 38
4.2.2 Jar tests ................................................................................................................ 38
4.2.3 Disinfection by-product formation and analysis ................................................. 39
4.2.4 DOC, TOC, and UVA measurements ................................................................. 39
4.2.5 Fluorescence spectra collection .......................................................................... 40
4.2.6 Fluorescence data analysis .................................................................................. 40
4.3 Results and discussion ............................................................................................... 41
4.3.1 Conventional parameter results from jar tests .................................................... 41
4.3.2 Fluorescence results from jar tests ...................................................................... 43
4.3.3 Relationships between NOM indicators ............................................................. 47
4.3.4 Disinfection by-product formation prediction .................................................... 48
4.4 Conclusions ................................................................................................................ 54
5 Contributions of spatial, temporal, and treatment impacts on natural organic matter
character using principal component analysis of fluorescence spectra ............................... 55
5.1 Introduction ................................................................................................................ 55
5.2 Method overview ....................................................................................................... 56
Page 6
vi
5.2.1 Water treatment plant characteristics .................................................................. 56
5.2.2 Total organic carbon ........................................................................................... 57
5.2.3 Liquid chromatography – organic carbon detection ........................................... 59
5.2.4 Fluorescence spectra collection .......................................................................... 59
5.2.5 Statistical analysis ............................................................................................... 61
5.3 Results and discussion ............................................................................................... 61
5.3.1 Analysis of raw water ......................................................................................... 61
5.3.2 Analysis of treatment efficiency ......................................................................... 66
5.4 Conclusions ................................................................................................................ 70
6 Pilot-scale coagulation optimization including NOM characterization with principal
component analysis of fluorescence spectra ........................................................................ 71
6.1 Introduction ................................................................................................................ 71
6.2 Methods ...................................................................................................................... 72
6.2.1 Pilot plant ............................................................................................................ 72
6.2.2 Fluorescence excitation-emission matrices ......................................................... 73
6.2.3 Principal component analysis ............................................................................. 74
6.2.4 Statistical assessment .......................................................................................... 74
6.2.5 Tukey’s method .................................................................................................. 75
6.3 Results and discussion ............................................................................................... 76
6.4 Performance evaluation using fluorescence spectra .................................................. 78
6.4.1 Spectral features of Peterborough water ............................................................. 78
6.4.2 Results of principal component analysis and model selection ............................ 80
6.4.3 Performance of coagulation on NOM fraction removal ..................................... 81
6.4.4 Performance of filtration on NOM fraction removal .......................................... 83
6.4.5 Performance of FEEM-PCA scores related to common NOM indicators .......... 84
6.5 Conclusions ................................................................................................................ 86
7 Summary, Conclusions and Recommendations .................................................................. 87
7.1 Summary .................................................................................................................... 87
7.2 Conclusions ................................................................................................................ 87
Page 7
vii
7.3 Recommendations ...................................................................................................... 88
REFERENCES ......................................................................................................................... 89
8 Appendices .......................................................................................................................... 98
8.1 Raw Data .................................................................................................................... 98
8.1.1 Disinfection by-product modelling (Section 4) .................................................. 98
8.1.2 Toronto NOM project data tables ..................................................................... 108
8.1.3 Peterborough pilot coagulant optimization data tables ..................................... 115
8.2 Calculations .............................................................................................................. 118
8.2.1 Analysis of variance .......................................................................................... 118
8.2.2 Tukey’s method ................................................................................................ 118
8.2.3 Statistics on water quality differences at Peterborough Pilot ........................... 119
8.3 Process flow diagrams of City of Toronto water treatment plants .......................... 121
8.4 Program code ........................................................................................................... 126
8.4.1 Real RLS files (Python 2.6) .............................................................................. 126
8.4.2 Manipulate for PCA input (Python 2.6) ............................................................ 128
8.4.3 Deconvolute loadings (Python 2.6) ................................................................... 130
8.5 Increased THM formation in PACl treated water .................................................... 131
8.5.1 Background ....................................................................................................... 131
8.5.2 Method: pH adjusted DBP Formation Tests ..................................................... 131
8.5.3 Results and Discussion ..................................................................................... 134
8.5.4 Controls and Blanks .......................................................................................... 136
8.5.5 Water quality ..................................................................................................... 137
8.5.6 Comparison of pilot and jar test results ............................................................ 138
8.5.7 DBP Formation Results .................................................................................... 139
8.5.8 Conclusion ........................................................................................................ 144
Page 8
viii
LIST OF TABLES
Table 2.1: Selected existing DBP models ...................................................................................... 6
Table 2.2: Commonly reported excitation/emission location of NOM fluorophores .................... 9
Table 3.1: Jar test reagents ........................................................................................................... 15
Table 3.2: Jar test method outline ................................................................................................ 16
Table 3.3: Total and dissolved organic carbon instrument conditions ........................................ 20
Table 3.4: Accuracy and method detection limit of TOC analysis .............................................. 20
Table 3.5: Description of LC-OCD fractions (Huber et al., 2011) .............................................. 23
Table 3.6: THM analysis instrument conditions .......................................................................... 25
Table 3.7: Accuracy and method detection limit of THM analysis ............................................. 26
Table 3.8: Trihalomethane analysis method outline .................................................................... 28
Table 3.9: Stock HAA specie concentrations .............................................................................. 30
Table 3.10: HAA analysis instrument conditions ........................................................................ 30
Table 3.11: Calibration range and check standard target concentrations for HAAs ................... 31
Table 3.12: Quality control standard deviation and HAA method detection limits .................... 32
Table 3.13: HAA analysis method outline ................................................................................... 33
Table 4.1: Source water characteristicsa ...................................................................................... 38
Table 4.2: Excitation/emission peak location for each water source ........................................... 43
Table 4.3: Variance explained by each PC for each sample set .................................................. 45
Table 4.4: Linear relationship strength (R2) between FEEM-PCA and traditional NOM
indicators .................................................................................................................... 48
Table 4.5: Results of non-linear regression for various NOM indicators and total THM
concentrations ............................................................................................................ 53
Table 4.6: Results of non-linear regression for various NOM indicators and total HAA
concentrations ............................................................................................................ 53
Table 5.1: Intake descriptions (CTC Source Protection Committee, 2012) ................................ 57
Table 5.2: Summary of treatment processes ................................................................................ 57
Table 5.3: Excitation/emission location of peaks for each water source ..................................... 64
Table 5.4: Variance explained by PCA model ............................................................................. 64
Page 9
ix
Table 5.5: Significant changes in LC-OCD fractions between raw and treated water ................ 68
Table 6.1: Peterborough raw water quality yearly ranges ........................................................... 72
Table 6.2: Variance (%) explained by PCs for Models 1 and 2 .................................................. 80
Table 6.3: Strength of linear correlation between FEEM-PCA and common NOM indicators .. 85
Table 8.1: PC 1 Scores ................................................................................................................. 99
Table 8.2: PC 2 Scores ................................................................................................................. 99
Table 8.3: TOC .......................................................................................................................... 100
Table 8.4: UVA .......................................................................................................................... 100
Table 8.5: SUVA ....................................................................................................................... 101
Table 8.6: THMs ........................................................................................................................ 102
Table 8.7: HAAs ........................................................................................................................ 103
Table 8.8: Raw water LC-OCD results ...................................................................................... 108
Table 8.9: Treated water LC-OCD results ................................................................................. 109
Table 8.10: Treated water PC scores ......................................................................................... 110
Table 8.11: Raw water PC scores .............................................................................................. 111
Table 8.12: Raw water TOC ...................................................................................................... 113
Table 8.13: Treated water TOC ................................................................................................. 114
Table 8.14: FEEM-PCA scores .................................................................................................. 115
Table 8.15: Statistical comparison of pilot and full-scale treatment trains: water quality
parameters ................................................................................................................ 116
Table 8.16: Statistical comparison of pilot and full-scale treatment trains: chlorine residual and
disinfection by-products ........................................................................................... 117
Table 8.17: Jar Test Water Quality T-Test Results .................................................................... 135
Table 8.18: Pilot Plant Water Quality T-Test Results ............................................................... 137
Table 8.19: T-Tests for Normalized TTHM Jar Test Results .................................................... 141
Table 8.20: T-Tests for Normalized TTHM Pilot Plant Results ................................................ 142
Table 8.21: T-Tests for Normalized THAA Jar Test Results .................................................... 144
Table 8.22: T-Tests for Normalized THAA Pilot Plant Results ................................................ 144
Page 10
x
LIST OF FIGURES
Figure 2.1: Example freshwater FEEM with humic acid (α), fulvic acid (β), and protein-like (δ)
peaks. (a) top-down view, (b) 3D view. (Peiris et al., 2010). ................................... 10
Figure 3.1: Process flow diagram of Peterborough water treatment plant .................................. 18
Figure 3.2: Example TOC calibration .......................................................................................... 21
Figure 3.3: TOC quality control chart (July to Sep., 2012) ......................................................... 21
Figure 3.4: LC-OCD chromatogram of surface water (Huber et al., 2011). ............................... 22
Figure 3.5: Example calibration chart for THM species (July 2012) .......................................... 26
Figure 3.6: Trichloromethane (TCM) quality control - low concentration target: 10 ug/L ......... 27
Figure 3.7: Trichloromethane (TCM) quality control - high concentration target: 60 ug/L ........ 27
Figure 3.8: Example HAA calibration curves (July 2012) .......................................................... 31
Figure 3.9: Example HAA quality control chart for DCAA ........................................................ 32
Figure 4.1: DOC, UVA, and SUVA results for all waters over the range of coagulant doses .... 42
Figure 4.2: Example fluorescence spectra (Otonabee River raw water)...................................... 43
Figure 4.3: Fluorescence spectra of Ottawa River raw water ...................................................... 44
Figure 4.4: Fluorescence spectra of Lake Simcoe raw water ...................................................... 44
Figure 4.5: Flourescence specctra of Lake Ontario raw water .................................................... 44
Figure 4.6: Loading plots for three principal components of the ‘all samples’ model from sample
set XA. ........................................................................................................................ 46
Figure 4.7: Scores from all sample PCA model (XA) .................................................................. 47
Figure 4.8: Total THMs concentration vs. alum dose. ................................................................ 50
Figure 4.9: Total HAAs concentration vs. alum dose ................................................................... 50
Figure 4.10: Measured values vs. predicted values of THMs from non-linear regressions of
source-specific models ............................................................................................... 52
Figure 5.1: Toronto WTP intake locations. ................................................................................. 58
Figure 5.2: Raw water LC-OCD results ...................................................................................... 62
Figure 5.3: Example raw water spectra from all four water treatment plants .............................. 63
Figure 5.4: Loading plots for first 3 principal component ........................................................... 65
Figure 5.5: Averages and MSD for raw water FEEM-PCA scores ............................................. 66
Page 11
xi
Figure 5.6: Comparison of treatment performance for all plants: TOC ...................................... 67
Figure 5.7: Comparison of treatment performance for all plants: LC-OCD ................................ 68
Figure 5.8: Average difference in PC score due to treatment ...................................................... 69
Figure 6.1: Simplified process flow diagram of pilot and full-scale treatment train ................... 72
Figure 6.2: Example FEEM illustrating scattering regions (FORS and SORS) .......................... 74
Figure 6.3: TOC and UVA of pilot plant settled water ................................................................ 77
Figure 6.4 Average monthly total loss of head for pilot plant filter ............................................ 78
Figure 6.5: Raw water FEEM – 2D view .................................................................................... 79
Figure 6.6: Raw water FEEM - 3D view ..................................................................................... 79
Figure 6.7: Raw and treated water samples (Model 1), PC 1 loading plot ................................. 82
Figure 6.8: Raw and treated water samples (Model 1), PC 2 loading plot .................................. 82
Figure 6.9: Treated water samples (Model 2), PC 1 loading plot ................................................ 82
Figure 6.10: Treated water samples (Model 2), PC 2 loading plot .............................................. 83
Figure 6.11: Impact of coagulation on PC scores. Averages and MDL (bars) for each treatment
train (n = 9 – 10). ....................................................................................................... 83
Figure 6.12: Impact of filtration on PC scores. Averages and MDL (bars) for each treatment train
(n = 9 – 10). ................................................................................................................ 84
Figure 8.1: THM speciation graphs, chlorine = 2.5 mg/L ......................................................... 104
Figure 8.2: THM speciation graphs, chlorine = 3.5 mg/L ......................................................... 105
Figure 8.3: HAA speciation graphs, chlorine = 2.5 mg/L ......................................................... 106
Figure 8.4: HAA speciation graphs, chlorine = 3.5 mg/L ......................................................... 107
Figure 8.5: F.J Horgan process flow diagram ............................................................................ 122
Figure 8.6: Island process flow diagram .................................................................................... 123
Figure 8.7: R.C. Harris process flow diagram ........................................................................... 124
Figure 8.8: R.L. Clark process flow diagram ............................................................................. 125
Figure 8.9: UVA vs. TOC for Jar Test Results: Full Range ...................................................... 135
Figure 8.10: UVA vs. TOC for jar test results: filtered water samples ..................................... 136
Figure 8.11: pH vs. Total THM Concentration for Jar Test Samples ........................................ 140
Figure 8.12: pH vs Total THM Concentration for Pilot Plant Samples ..................................... 141
Figure 8.13: pH vs. total HAA concentration jar tests ............................................................... 142
Figure 8.14: pH vs. total HAA concentration pilot samples ...................................................... 143
Page 12
xii
NOMENCLATURE
% Percent
oC Degrees Celsius
γ Fraction of integral fluorescence emission
ε Molar absorptivity
λ Wavelength (nm)
μ Average
φ Quantum yield
General form of principal component
A Absorbance
a.u. Arbitrary units
Al2(SO4)3 Aluminum sulphate
Alum Aluminum sulphate
ANN Artificial neural networks
ANOVA Analysis of variance
b Path length
BCAA Bromochloroacetic acid
BDCAA Bromodichloroacetic acid
BDCM Bromodichloromethane
Br- Bromide
c Concentration
C Covariance matrix
CDBM Chlorodibromomethane
DBAA Dibromochloroacetic acid
DBPs Disinfection by-products
DCAA Dichloroacetic acid
DOC Dissolved organic carbon (mg/L)
DPD N,N-diethyl-p-phenylenediamine
F Fluorescence intensity
FEEM Fluorescence excitation-emission matrix
Page 13
xiii
FEEM-PCA Principal component analysis of fluorescence excitation-emission matrices
FORS First order Rayleigh scattering
GAC Granular activated carbon
GC-ECD Gas chromatography with electron capture detection
H2SO4 Sulfuric acid
HAA5 Sum of five haloacetic acids: Monochloroacetic acid (MCAA),
Dichloroacetic acid (DCAA), Trichloroacetic acid (TCAA),
Monobromoacetic acid (MBAA) and Dibromoacetic acid (DBAA)
HAA9 Sum of nine haloacetic acids: Monochloroacetic acid (MCAA),
Dichloroacetic acid (DCAA), Trichloroacetic acid (TCAA),
Monobromoacetic acid (MBAA), Dibromoacetic acid (DBAA),
Bromochloroacetic acid (BCAA), Bromodichloroacetic acid (BDCAA),
ChlorodibromoaceticaAcid (CDBAA), and Tribromoacetic acid (TBAA)
HAAs Haloacetic acids
HI 705 Hyper+Ion 705 PACl
I Intensity
KHP Potassium hydrogen phthalate
L Litre(s)
LC-OCD Liquid chromatography with organic carbon detection
LMW Low molecular weight
MAE Mean absolute error
MBAA Monobromoacetic acid
MCAA Monochloroacetic acid
MDL Method detection limit
mg/L Milligrams per litre
Milli-Q® Purified water from Millipore Corporation equipment
mL Millilitre(s)
MSD Minimum significance difference
MSE Mean square error
MTBE Methyl-tert-butyl ether
MX Mutagen-x
μL Microlitre(s)
Page 14
xiv
µg/L Microgram(s) per litre
n Number of samples
NaOCl Sodium hypochlorite
nm Nanometer
NOM Natural organic matter
OCD Organic carbon detection
OND Organic nitrogen detection
p Loading matrix
PACl Polyaluminum chloride
PARAFAC Parallel factor analysis
PC Principal component
PCA Principal component analysis
PFD Process flow diagram
pH Hydrogen ion concentration
PLS Partial least squares
R2 Pearson’s correlation coefficient
SCADA Supervisory control and data acquisition
SD Standard deviation
SORS Second order Rayleigh scattering
SUVA Specific ultraviolet absorbance
T Temperature
t Time
t Scores matrix
TBM Tribromomethane
TBAA Tribromoacetic acid
TCAA Trichloroacetic acid
TCM Trichloromethane
THMs Trihalomethanes
THM4 Sum of four trihalomethanes: Tribromomethan (TBM), Trichloromethane
(TCM), Chlorodibromomethane (CDBM), and Bromodichloromethane
(BDCM)
TOC Total organic carbon, mg/L
Page 15
xv
TOX Total organic halides
USEPA United States Environmental Protection Agency
UV Ultraviolet
UVA Ultraviolet absorbance
V Volume
WTP Water treatment plant
Page 16
1
1 Introduction
1.1 Background
Organic matter in naturally occurring waters is comprised of a complex heterogeneous
mixture of humic acid, fulvic acid, proteins, carbohydrates, and other organic compound classes
(Her et al., 2003). The monitoring of natural organic matter (NOM) has proven to be of great
importance in the water treatment field. NOM has been shown to be related to disinfection by-
product formation, negatively affects water treatment processes such as membrane filtration
(Peiris et al., 2009), and promotes biological growth in distribution systems (Baghoth et al.,
2011). The diversity of compounds within the definition of NOM is problematic when
considering removal techniques (Pifer and Fairey, 2012). There is evidence that both functional
groups and molecular weight distributions of compounds considered to be components of NOM
are integral to determining treatability (Matilainen et al., 2011).
Quantification of NOM has traditionally been through gross measurement parameters
such as total and dissolved organic carbon (TOC/DOC). These parameters represent all
oxidizable organic content in the water and do not provide information on specific compound
classes within the definition of NOM. Given the evidence that size, functionality, and structure
are important to treatability and reactivity, NOM quantification techniques with the ability to
identify specific compound classes are of interest for a variety of applications in the water
treatment field (Barrett et al., 2000).
NOM characterization using fluorescence spectra has been gaining traction as a
promising method (Zepp et al., 2004; Bieroza et al., 2011). This type of analysis is completed
through the collection of three dimensional fluorescence excitation-emission matrices (FEEM),
which represent fluorescence intensity at various excitation-emission pairs. In comparison to
other NOM characterization techniques, including liquid chromatography - organic carbon
detection (LC-OCD), FEEM analysis is fast and provides consistent results with high sensitivity
(Peiris et al., 2008). Traditional FEEM analysis involves identifying the location and magnitude
of peaks of fluorescence intensity. The position (excitation-emission region) of the peak can be
correlated with a specific compound and the maximum intensity of the peak with concentration
(Bieroza et al., 2011). This method, referred to as ‘peak picking,’ is limited in that it neglects to
capture the heterogeneity of NOM fractions (Peiris et al., 2010). The fluorescence intensity at
specific excitation/emission pairs is influenced by several variables including concentration,
quantum yield, and absorptivity. Molecular structure has a noted effect on some of these factors
Page 17
2
which can be common among different molecules. By ‘peak picking’ the nature of overlapping
structural properties between distinct compounds and many of the potentially relevant
chromophores are neglected (Persson and Wedborg, 2001).
To further extract information and, in particular, changes in NOM composition between
several samples, multivariate analysis techniques can be employed. Common advanced analysis
techniques involve using two-way principal component analysis (PCA) and multi-way parallel
factor (PARAFAC) analysis (Peiris et al., 2010). These multivariate techniques reduce the
dimensionality of the data set (Persson and Wedborg, 2001), simplifying analysis, while
incorporating the entire fluorescence spectrum (Bieroza et al., 2011).
1.2 Objectives
Specific research objectives for this study can be summarized as follows:
1. Investigate the applicability of principal component analysis of fluorescence
excitation-emission matrices (FEEM-PCA) for characterization of NOM.
2. Compare FEEM-PCA as an indicator of disinfection by-product (DBP) precursor
material to traditional NOM measures.
3. Investigate spatial and temporal changes in NOM characteristics between raw waters
for treatment facilities in a given watershed using FEEM-PCA.
4. Determine changes to NOM fractions due to various treatment processes with FEEM-
PCA and other characterization techniques.
Page 18
3
2 Literature Review
2.1 Disinfection by-products
Major DBPs classes produced from chlorination or natural waters include
trihalomethanes (THMs), haloacetic acids (HAAs), haloacetonitriles, halopicrin, cyanogen
chloride and bromide, chloral hydrate, and mutagen-x (MX) (Oxenford, 1997). Despite the
identification of hundreds of distinct DBP types, 30 - 60% of the total organic halides (TOX)
formed have been accounted for (Singer, 1994; Krasner et al., 2006). Control and regulations for
water utilities using chlorine typically revolve around concentrations of trihalomethanes (THMs)
and a subset of haloacetic acids (HAAs) (Beggs et al., 2009), which occur at the highest
concentrations (Hua and Reckhow, 2008). The regulation of only certain classes addresses the
impracticality of measuring all DBPs and provides a surrogate for other un-identified compounds
(Hrudey, 2009).
Coinciding with the discovery of THMs in drinking water in the 1970s, the USEPA
released findings of an association between consumption of chlorinated water and increased
cancer risk. This spurred the classification of chloroform as a carcinogen and ultimately the
regulation of THMs in drinking water by the USEPA (Singer, 1994). Furthermore, it opened up
a field of necessary research into the potential health effects of chlorination DBPs on humans.
There is some indication of potential carcinogenicity and adverse health effects from chlorinated
DBP exposure (McBean et al., 2008; USEPA, 1999), although current evidence has not linked
realistic human consumption levels of regulated DBPs in drinking water, THMs and HAA5, and
elevated human health risk (Hrudey, 2009; Wang et al., 2007).
2.1.1 DBP formation theory
Disinfection by-products are formed through the oxidation of NOM present in natural
source waters (Oxenford, 1997; Sadiq and Rodriguez, 2004). NOM is comprised of a complex
heterogeneous mixture of humic acid, fulvic acid, proteins, carbohydrates, and other organic
compound classes (Her et al., 2003). Each of these classes has unique reactivity with
disinfectants to form DBPs (Barrett et al., 2000), although multiple studies have identified
humic-like substances to be the largest contributor to DBP formation (Childress et al., 1999;
Liang and Singer, 2003).
Page 19
4
In addition to precursor concentrations and reactivity, several other factors affect the rate
and total DBP formation including pH, temperature, reaction time, chlorine dose, and bromide
ion concentration (Chen and Westerhoff, 2010). Increased reaction times generally lead to
greater DBP formation and consumption of additional chlorine residual (Rodriguez and Sérodes,
1999). The effect of pH is specific to DPB specie; THM formation potential increases with
increasing pH, while formation potential for HAAs and most other DBP classes are reduced with
increasing pH (Sadiq and Rodriguez, 2004). Increasing chlorine dose as well as water
temperature are also associated with increasing DBP formation potential on the basis of reagent
availability and reaction kinetics (Gallard and U. von Gunten, 2002).
2.1.2 Modelling DBP formation
Control of DBP formation is typically achieved through the reduction of precursor
material (Singer, 1994; Oxenford, 1996). While it is possible to remove DBPs once they have
formed, it is more cost efficient to limit the production through NOM reduction (Sadiq and
Rodriguez, 2004). Predictive DBP models have been successfully linked to NOM estimation
parameters which estimate the amount of DBP precursor material, including total organic carbon
(TOC), dissolved organic carbon (DOC), UV-absorbance (UVA) at 254 nm, and specific UV-
absorbance (SUVA) (Chen and Westerhoff, 2010; Chowdhury et al., 2009). Results from DBP
modelling studies show conflicting results regarding which NOM surrogate is most appropriate
for DBP formation prediction. Some studies indicate that SUVA (DOC divided by UVA at 254
nm) has been shown to correlate best with DBP formation (Kitis et al., 2001; Barrett et al.,
2000). In contrast, Uyak and Toroz (2007) found UVA more closely correlated with DBP
formation potential than DOC or SUVA while other authors have reported greater prediction
strength using DOC (Sohn et al., 2004). Based on the EPA database of reported DBP
concentrations, Sohn et al. (2004) presented DOC based models had a significantly greater
predictability when compared to using UVA.
Chowdhury et al. (2009) provided a comprehensive list of reported DBP models and their
strength (as R2). This list has been adapted and is presented as Table 2.1 to illustrate historically
successful model types and parameters used for DBP modelling. Of particular note is the lack of
NOM fraction information used in previous models. Although linear regression techniques have
been applied, it has been shown that empirical power based models best describe DBP formation
(Amy et al., 1987; Sohn et al., 2004). A generalized form of a power function for THM
prediction is shown as equation 2.1.
Page 20
5
where, NOM is the concentration of NOM, Br- is concentration of bromide, T is
temperature, Cl2 is chlorine dose, and t is time. k, a, b, c, d, e, and f are constants
determined through regression. Coefficients can be found through several methods,
although forward stepwise regression has typically been used (Amy et al., 1998).
Due to the complexity of compound types within NOM and their unique reactivity in
DBP formation (Barrett et al., 2000; Chen et al., 2008) it is postulated that methods which can
characterize NOM fractions will improve the accuracy of DBP formation models. Although
significant work has been done on relating specific constituents of NOM with DBP formation
potential (Minear and Amy, 1996; Hua and Reckhow, 2007), little work has focused on including
fractionation information into predictive DBP models (Chowdhury et al., 2009). Furthermore,
NOM characterisation has commonly relied on chemical fractionation using adsorption resins
(Matilainen et al., 2011) or physical fractionation using size exclusion columns (Huber et al.,
2011; Pifer and Fairey, 2012). These NOM fractionation techniques are limited for widespread
application due to high instrument and sample costs, long analysis time, and required personnel
expertise.
Hua and Reckhow (2007) performed a comprehensive study on DBP formation potential
from NOM fractions obtained using XAD resins. Through total organic halides (TOX)
measurements, the authors found that hydrophobic, high molecular weight compounds produced
a greater amount of unidentifiable organohalides. Furthermore, they concluded that THM
precursors were generally less hydrophobic than those for HAAs. Yang et al. (2008) used
regional integration of fluorescence excitation-emission matrices (FEEMs) for determining DBP
formation of NOM fraction extracts. The results from regional integration performed worse for
prediction of chloroform and nitrile formation in comparison to SUVA. Pifer and Fairey (2012)
presented increased correlation of chloroform formation with maximum intensities from parallel
factor (PARAFAC) analysis of FEEMs coupled with SUVA in comparison to SUVA by itself.
(2.1)
Page 21
6
Table 2.1: Selected existing DBP models
Source Model R2
Rodriguez et al. (2000) a) THMs = 0.044(DOC)1.030
(t)0.262
(pH)1.149
(D)0.277
(T)0.968
b) THMs = 1.392(DOC)1.092
(pH)0.531
(T)0.255
a) 0.90
b) 0.34
Serodes et al. (2003) a) log(HAAs) = 2.72+0.653(TOC)+0.458(D)+0.295(t)
b) log(HAAs) = 1.33+2.612(TOC)+0.102(D)0.255(T)+0.102(t)
c) HAAs = -8.202+4.869(TOC)+1.053(D)+0.364(t)
d) THMs = 16.9+16.0(TOC)+3.319(D)-1.135(T)+1.139(t)
e) log(THMs) = -0.101+0.335THMo+3.914(TOC)+0.117(t)
f) THMs – 21.2+2.447(D)+0.499(t)
a) 0.89
b) 0.80
c) 0.92
d) 0.78
e) 0.89
f) 0.56
Sohn et al. (2004) a) THMs = 10-1.385
(DOC)1.098
(D)0.152
(Br)0.068
(T)0.609
(pH)1.601
(t)0.263
b) THMs = 0.42(UVA)0.482
(D)0.339
(Br-)0.023
(T)0.617
(pH)1.601
(t)0.261
c) THMs = 0.283(DOC*UVA)0.421
(D)0.145
(Br-)0.041
(T)0.614
(pH)1.606
(t)0.261
d) THMs = 3.296(DOC)0.801
(D)0.261
(Br)0.223
(t)0.264
e) THMs = 75.7(UVA)0.593
(D)0.332
(Br)0.0603
(t)0.264
f) THMs = 23.9(DOC*UVA)0.403
(D)0.225
(Br-)0.414
(t)0.264
g) THMs = (THM @ pH 7.5, T = 20oC)*1.156
(pH-7.5)1.0263
(T-20)
h) HAA6 = 9.98(DOC)0.935
(D)0.443
(Br)-0.031
(T)0.387
(pH)-0.655
(t)0.178
i) HAA6 = 171.4(UVA)0.584
(D)0.398
(Br)-0.091
(T)0.396
(pH)-0.645
(t)0.178
j) HAA6 = 101.2(DOC*UVA)0.452
(D)0.194
(Br)-0.0698
(T)0.346
(pH)-0.6235
(t)0.18
k) HAA6 = 5.228(DOC)0.585
(D)0.565
(Br)-0.031
(t)0.153
l) HAA6 = 63.7(UVA)0.419
(D)0.640
(Br)-0.066
(t)0.161
m) HAA6 = 30.7(DOC*UVA)0.302
(D)0.541
(Br)-0.012
(t)0.161
n) HAA6 = (HAA6 @ pH 7.5, T = 20oC)*0.932
(pH-7.5)1.02
(T-20)
a) 0.90
b) 0.70
c) 0.81
d) 0.87
e) 0.90
f) 0.92
g) 0.92
h) 0.87
i) 0.80
j) 0.85
k) 0.92
l) 0.92
m) 0.94
n) 0.85
Uyak et al. (2005) THMs = 0.0707(TOC+3.2)1.314
(pH-4.0)1.496
(D-2.5)-0.197
(T+10)0.724
0.98
Hong et al. (2007) THMs = 10-1.375
t0.258
(D/DOC)0.194
pH1.695
T0.507
(Br-)0.218
BDCM = 10-3.201
t0.297
pH1.695
T0.507
(Br-)0.218
TCM = 10-0.748
t0.210
(D/DOC)0.221
pH1.374
T0.532
(Br-)-0.184
0.87
0.87
0.86
Note: DBP concentrations from all models reported in µg/L. THM: trihalomethanes, HAA: haloacetic acid, D: chlorine dose
(mg/L), DOC: dissolved organic carbon, TOC: total organic carbon, UVA: ultraviolet absorbance, Br-: bromide concentration
(mg/L), pH: hydrogen ion concentration, t: time, T: temperature (oC)
Page 22
7
2.2 Measurement techniques for natural organic matter
In both surface and ground waters, NOM consists of a wide array of compounds
including humic substances, carbohydrates, carboxylic acids, phenols, amino acids, and proteins
(Thurman, 1985; Her et al., 2003). Due to the complex variety of natural organic matter,
quantification is accomplished through the use of surrogates. Typical surrogate quantification
measures of NOM, such as TOC/DOC, only provide gross measurements (Gone et al., 2009).
Both TOC and DOC are measures of total mass of organic carbon, typically obtained through
complete oxidation (Matilainen et al., 2011). This means that TOC/DOC, while providing
information on the total concentration of organic material offers no information on NOM
character and reactivity.
Through a variety of methods it has been established that humic substances make up the
majority of organic carbon (40-60% of the dissolved fraction) in natural waters (Sohn et al.,
2007). Humic substances are a broad group which represent a heterogeneous mixture of high
molecular weight aromatic and aliphatic organic compounds characterized by functional groups
such as carboxyl, phenolic and/or enolic OH, alcoholic OH, and quinolic CO (Mobed et al.,
1996). In most natural freshwater sources, the second most abundant organic carbon fraction is
bio-polymers, which have molecular weights of 10 kDA or greater. This fraction tends not to
have a response to UVA indicating the lack of unsaturated structures however, does contain
organic nitrogen indicating the presence of polysaccharides, proteins, and amino acids (Huber et
al., 2011).
Another commonly used NOM surrogate is UVA, which has been shown to account for
specific structures and certain functional groups (Sadiq and Rodriguez, 2004). NOM has been
found to absorb light over a wide range of wavelengths, while inorganic constituents of natural
waters do not absorb above wavelengths of approximately 230 nm. The majority of NOM
chromophores contain aromatic functional groups such as phenols and various aromatic acids,
which absorb light in the UV region (< 400 nm), and are principally associated with humic
substances (Korshin et al., 1996). For the majority of applications in water treatment, UVA is
measured at a wavelength of 254 nm, which is thought to be most selective to aromatic
chromophores. Studies have shown that the aromaticity of DOC strongly defines its reactivity
with oxidants, such as chlorine, making UVA a theoretically ideal DBP precursor surrogate
(Weishaar et al., 2003).
Page 23
8
In an effort to address the non-selectiveness of TOC/DOC, a measure of SUVA was
developed which incorporates a measure of aromaticity using UVA. SUVA is defined as the
ratio of UVA at 254 nm to DOC (Edzwald and Tobiason, 1999). As with UVA, it has been
thought that SUVA would perform well as a DBP precursor surrogate since it incorporates a
measure of aromaticity. However, Weishaar et al. (2003) showed that water samples with
similar SUVA values displayed a wide range of THM formation potentials. They concluded that
while SUVA was a good indicator of the DOC character, however did not provide enough
source-specific reactivity information for DBP formation prediction.
Two analysis techniques which have been shown to provide information on multiple
NOM fractions include liquid chromatography-organic carbon detection (LC-OCD) (Huber et
al., 2011) and multivariate analysis of fluorescence excitation-emission matrices (FEEM) (Zepp
et al., 2004; Bieroza et al, 2011).
2.2.1 Liquid chromatography – organic carbon detection
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 and into hydrophilic and hydrophobic fractions (Huber and Frimmel, 1992). Hydrophilic
NOM is then divided into five subcategories: biopolymers, humic substances, building blocks of
humic substances, low molecular weight (LMW) acids and LMW neutrals. A UV detector
provides information regarding the specific UV absorbance (SUVA) and qualitative levels of
inorganic colloids for each previously identified category. Finally, an organic nitrogen detector is
used to provide additional information regarding the nature of specific fractions, such as the
fraction of bio-polymers considered to be proteins (Huber et al, 2011).
2.2.2 Fluorescence excitation-emission matrices
Analysis of fluorescence spectra is completed through collection of fluorescence
excitation-emission matrices (FEEM). These matrices represent fluorescence intensity at various
excitation-emission pairs. When compared to other NOM characterization techniques, including
liquid chromatograph organic carbon detection (LC-OCD), FEEM analysis is fast and provides
consistent results with high sensitivity (Peiris et al., 2008). Furthermore, fluorescence measures
are more selective and less susceptible to interferences when compared to NOM characterization
using UVA (Persson and Wedborg, 2001).
Page 24
9
FEEM analysis is based on the realization that several components of NOM fractions
exhibit unique fluorescence signatures that allow them to be distinguished from each other in a
fluorescence spectrum (Coble et al., 1990; Zepp et al., 2004). Through FEEMs, typically humic
acid, fulvic acid, and aromatic proteins (tryptophan or tyrosine-like) compounds are identifiable
(Bieroza et al., 2011). There is evidence that humic substances originating from either terrestrial
or anthropogenic sources may be differentiated through FEEMs (McKnight et al., 2001). NOM
compounds typically identifiable through fluorescence and their reported unique excitation-
emission signatures are presented in Table 2.2. An example FEEM with humic acid fulvic acid,
protein-like, as well as first and second order Rayleigh scattering (FORS/SORS) peaks labelled,
is shown as Figure 2.1 (Peiris et al., 2010).
Table 2.2: Commonly reported excitation/emission location of NOM fluorophores
Compound group Excitation/Emission
wavelength (nm)
Description and source Author
Protein-like 275/<300 Amino acids, free or
bound in proteins
Hudson et al., 2007
275/310 Tyrosine-like Coble et al., 1996
275/340 Tryptophan-like Coble et al., 1996
Humic-like 260/380-460
Terrestial humic-like.
Peak A: humic acid
Coble et al., 1996
350/420-480 Terrestial humic-like. Peak
C: fulvic acid
Coble et al., 1996
300-370/400-500 Humic-like Murphy et al., 2008
2.2.2.1 Theory
Fluorescence occurs when chromophores are excited and then relax to a ground state
through the radiation of photons. Due to non-radiative decay, and other processes some of the
energy adsorbed is lost resulting in the emitted photon to not equal the excitation energy (Stokes’
shift) (Hudson et al., 2007). Fluorescence intensity (photons emitted) is dependent on several
factors including absorptivity, quantum yield, and concentration of the chromophore, as well as
cell path length (Roch, 1997; Persson and Wedborg, 2001).
Page 25
10
Figure 2.1: Example freshwater FEEM with humic acid (α), fulvic acid (β), and protein-like (δ)
peaks. (a) top-down view, (b) 3D view. (Peiris et al., 2010).
The energy absorbed by a chromophore subjected to excitation photons can be described
by Beer’s law:
where, A is the absorbance, ε is the molar absorptivity of a specific chromophore, b is the
path length, and c is the concentration of chromophores.
The absorbance (A) is what contributes to the excitation of the chromophore. In complex
mixtures, such as NOM in water, multiple chromophores exist with distinct properties expressed
through the molar absorptivity (ε). Therefore, total absorbance of the sample is the sum of the
molar absorptivity and concentration of each chromophore multiplied by the path length (Persson
and Wedborg, 2001). After excitation at a specific wavelength, λi, the intensity of emission
through relaxation to the ground state is described by the quantum yield (φ) and the fraction of
the integral fluorescence emission and a specific wavelength (γ(λi)) (Roch, 1997). Quantum
yield is defined as the number of quanta emitted divided by the total quanta absorbed (Persson
and Wedborg, 2001), in effect the efficiency of fluorescence for a specific compound.
Therefore, for a given intensity of excitation light, Io(λi), the fluorescence of a chromophore, F,
can be expressed as follows (Roch, 1997):
(2.2)
(2.3)
Page 26
11
For multiple chromophores in solution, as with Beer’s law, the total fluorescence is
additive (Persson and Wedborg, 2001). Factors such as absorptivity and quantum yield are
highly dependent on the structure, weight, and functional groups associated with a compound
(Baghoth et al., 2011). Therefore, the broad peaks observed in FEEMs of natural waters are
explained by the overlapping structural properties of NOM chromophores (Persson and
Wedborg, 2001).
2.3 Multivariate analysis of fluorescence spectra
Historically analysis of FEEMs has been completed through identifying location and
maximum intensity of fluorescence peaks. The position (excitation-emission region) of the
peaks are correlated with specific compound groups such as humic and fulvic acids, while the
maximum intensity is correlated with concentration (Her et al., 2003). As discussed in section
2.2.2.1, the fluorescence intensity is highly dependent on the molecular structure of the
chromophores, which can be similar between several distinct molecules. The method of tracking
concentration through the maximum intensity of fluorescence peaks, referred to as ‘peak
picking,’ does not capture the nature of overlapping structural properties between diverse
compounds and many of the potentially relevant chromophores are neglected (Persson and
Wedborg, 2001; Peiris et al., 2010).
To further extract information and, in particular, changes in NOM composition between
several samples, multivariate analysis techniques can be employed. Multivariate analysis
methods such as PCA are able to reduce dimensionality of data sets by modelling variance
between samples. This allows for the incorporation of entire fluorescence spectrum in analysis,
creating a more complete representation of the heterogeneity of NOM (Bieroza et al., 2011).
Furthermore, information related to the fundamental properties of PCA can garner information
about the sample set being analyzed. PCs represent maximum variance in the sample set and,
therefore, describe the spectral areas which most significantly change between samples.
Several multivariate analysis techniques have been successfully applied to FEEM
analysis including PCA, partial least squares (PLS), parallel factor analysis (PARAFAC),
artificial neural networks (ANNs) and others (Persson and Wedborg, 2001; Stedmon et al.,
2003). Recent publications related to freshwaters have generally employed either PCA or
PARAFAC (Her et al., 2003; Peiris et al. 2010). Each of these analysis techniques vary in their
assumptions and algorithms for data decomposition. For example, PCA requires each PC to be
Page 27
12
mutually independent with maximized variance representation (Joliffe, 1986). While there is
some evidence that PARAFAC produces variables more representative of individual NOM
compounds, it is not capable of analyzing negative values (Bieroza et al., 2011) and cannot
incorporate light scattering regions due to their non-linear response (Bahram et al., 2006).
Three types of scattering have been identified in typical FEEMs including Rayleigh,
Tyndall and Raman scattering, with Rayleigh scattering typically being the most pronounced
(Zepp et al. 2004). Although it eliminates the possibility of analyzing changes to scattering,
which have been associated with particulates (Peiris et al., 2010), modelling efficiency has been
shown to improve for PCA through the subtraction of scattering regions (Bahram et al., 2006).
2.4 Principal component analysis
PCA is a multivariate data analysis technique that is used to approximate a large data
matrix through observed patterns (Wold et al., 1987). Approximation of the data patterns is
achieved by obtaining new, mutually independent, variables that are mathematically represented
by linear combinations of the original variables. In generalized terms, equation 2.4, the original
data matrix X is decomposed into the product of two small matrices, ti and pi which are referred
to as the scores matrix and the loading matrix, respectively. The residuals generated from the
approximation are represented in another matrix E. For multiple samples, q, in the matrix X, the
product and residuals are summed (Peiris et al., 2010).
∑
2.4.1 Mathematical definition and derivation
Given a matrix of n random variables, X, one could describe patterns of the n variances
through all of the possible ⁄ covariances. For a small n, this would be realistically
feasible, however in applications where n is very large this method of describing variance
between random variables would be cumbersome. In essence, principal component analysis
attempts to describe these variances through a reduced set of variables, called principal
components (Joliffe, 1986).
Principal components are linear functions of the original set of variables (Wold et al.,
1987), X, therefore generally can be defined as , where θT is a vector of n constants,
(2.4)
Page 28
13
transposed. The first PC ( ) describes the most variance in X. As part of PCA, individual PCs
are defined as mutually independent. Therefore, the second PC ( ) must also describe a
maximum amount of variance within X, while being uncorrelated with (Joliffe, 1986).
The PCs are calculated through finding the eigenvectors of the covariance matrix of X,
referred to as C. For any arbitrary i’th PC,
∑
Where, θi is the i`th eigenvector of C with associated eigenvalue, λi. A constraint is typically
added where θTθ = 1 so that by definition the variance of αi is equal to the i`th eigenvalue, λi.
As discussed, each subsequent PC maximizes the amount of variance represented. The
variance represented by a PC can be defined as follows with respect to the covariance matrix:
A common approach to finding the PCs is to use Lagrange multipliers (Joliffe, 1986).
For a Langrage multiplier, λ, equation 2.7 is maximized through differentiation (equation 2.8)
with respect to θ1 in order to find the first principal component:
[
]
Therefore, λ is the eigenvalue of C associated with the eigenvector, θ1. This leads to the
conclusion that in order to maximize variance represented by the PC, the eigenvalue has to be
maximized. In general, the i’th largest eigenvalue of the covariance matrix is associated with the
i’th PC. For PCs greater than 1, mutual independence must also be considered. Joliffe shows a
proof on how correlation between eigenvectors can be accounted for using the approach
discussed above (Joliffe, 1987).
(2.5)
(2.6)
(2.7)
(2.8)
Page 29
14
2.5 Application to natural waters
Significant work with FEEM analysis for NOM characterization in marine and
freshwaters has been completed over the last couple of decades. Coble et al., 1990 analyzed
dissolved organic matter (DOM) in sea waters using FEEMs. They were successful in
identifying three distinct fluorophores and tracking changes in concentration with depth based on
maximum peak intensities. To-date, at least four NOM groups related to distinct FEEM peaks
have been identified to represent protein and humic substances (Zepp et al., 2004). Baker et al.,
(2002) used fluorescence spectrophotometry to identify the presence of pollution in catchments.
They reported that protein-like fluorescence correlated well with wastewater influence and were
able to estimate the impact on natural waters using this technique. Hudson et al. (2007) also
illustrated the use of fluorescence to detect anthropogenic sources of pollution in natural waters.
Peiris et al. (2011) has shown the application of FEEMs for prediction of ultrafiltration
membrane fouling events. Furthermore, membrane fouling for both ultrafiltration and
nanofiltration has been linked predominantly to the protein-like fraction (Peiris et al., 2010).
In light of these varied applications of FEEM analysis for NOM characterization, it is of
interest to further explore and refine the use of fluorescence spectra in the drinking water field.
The incorporation of multivariate analysis techniques for handling high-dimensionality of data
has significant promise for increasing analysis sensitivity and selectivity.
Page 30
15
3 Materials and methods
3.1 Experimental protocols
3.1.1 Jar tests and DBP formation
Jar tests were conducted to mimic conventional water treatment processes including
coagulation, flocculation, sedimentation, and filtration. Two coagulant types were used in this
study: alum (General Chemical, Parsipanny, NJ), and Hyper+Ion 705 (HI705) polyaluminum
chloride (PACl) (General Chemical, Parsipanny, NJ). The jar tests were conducted using a PB-
700 Standard Jar Tester paddle stirrer with six square, acrylic 2-L jars (Phipps & Bird,
Richmond, VA). The protocol followed was adapted from the USEPA’s Enhanced Coagulation
Guidance Manual (USEPA, 1999). Following addition of the coagulant into each jar,
coagulation was simulated through rapid mix (100 RPM) for 1.5 minutes and flocculation by
reducing the mixing speed to 30 RPM for 15 minutes. The water was then allowed to stand (no
mixing) for 30 minutes to simulate sedimentation. Vacuum filtration was performed on the
settled water using 1.2 micron glass microfibre filters (Whatman 934-AH, Florham Park, NJ) to
simulate media filtration (Wassink, 2011). For specific analyses including DOC and LC-OCD
the water was filtered through 0.45 micron membrane filters (Pall Corporation Supor®-450, Port
Washington, NY) to ensure that particulates (> 0.45µm in size) had been removed. Following
filtration, chlorination was simulated by dosing sodium hypochlorite (12% Cl2, BioShop Canada,
Inc., Burlington, ON) and allowing a 24 hour reaction time in a 250 mL sealed bottle, without
headspace.
Reagents used in the jar tests are summarized in Table 3.1 and Table 3.2 provides a
method outline for the jar tests.
Table 3.1: Jar test reagents
Reagent Source
Aluminum sulphate (Al3(SO4)318H2O General Chemical (Parsipanny, NJ), 48.5%
Polyaluminum chloride (Hyper+Ion 705 PACl) General Chemical (Parsipanny, NJ), 100%
Sodium hypochlorite (NaOCl) BioShop Canada Inc. (Burlington, ON), 12%
Page 31
16
Table 3.2: Jar test method outline
Raw water stored in refrigerator (4oC)
Allow for raw water to reach room temperature before testing (23±2oC)
Chlorine Solution Preparation
1. Fill a 100 mL volumetric flask with Milli-Q water and 1000 µL of 12% NaOCl stock
solution. Invert flask 5 times to ensure mixing.
2. Transfer to 125 mL amber bottle, cap with Teflon lined cap and store at 4oC Yields a 600
mg/L Chlorine Solution
3. Concentration of free chlorine in solution is checked using the DPD method
Jar Test
1. Take a 100 mL sample of raw water and measure pH. From this take duplicate 40 mL
sample for UV254 and TOC analysis
2. Fill six jars with 2 L of raw water using a 1 L graduated cylinder.
3. Add required volume of coagulant (and acid determined from previous step). Add acid
first and mix well prior to adding coagulant.
4. Using the jar testing apparatus, stir at 100 rpm for 90 seconds (rapid mix)
5. Reduce speed to 30 rpm for 15 minutes (flocculation)
6. Turn of stirrer, raise the paddles out of the sample and allow 30 minutes for settling.
7. Collect a 500 mL sample of the supernatant from each jar, filter using a 1.5 µm glass
fibre filter and measure pH.
8. Take duplicate 40 mL sample from the 500 mL for UV254 and TOC analysis
DBP Formation Test:
1. From the filtered supernatant, transfer just under 250 mL into a 250 mL bottle.
2. After adding chlorine, fill up the remainder of the headspace with Milli-Q water. Cap
with Teflon-lined screw caps and set out at room temperature (23 ± 2oC)
3. After 24 hours, measure free and total chlorine fractions using DPD method.
4. Quench chlorine with 0.050 g of ascorbic acid
5. Collect two 25 mL samples for DBP analysis in sealed vials with no headspace
Page 32
17
3.1.2 Peterborough pilot plant
The Peterborough pilot plant was designed and operated to mimic the operation of the
full-scale water treatment plant located on the Otonabee River in the City of Peterborough,
Ontario. The full-scale plant was built in 1922 with expansions in 1952, 1965, and 1995. The
pilot plant, which is comprised of two identical treatment trains, was put online in 2011 to collect
data for process optimization work. To date, the pilot plant has been used principally to compare
the effects on water quality from the use of polyaluminum chloride (PACl) as a coagulant
compared to the currently used aluminum sulphate (alum).
A detailed process flow diagram is presented as Figure 3.1. Both the full-scale and pilot-
scale trains draw water from the same intake in the Otonabee River. When the temperature of
the raw water falls below 12oC, a free residual of approximately 0.05 mg/L chlorine is applied at
the intake for zebra mussel control. Following intake, the coagulant is added during a flash mix.
Flocculation is allowed for in multiple large tanks followed by sedimentation using parallel plate
settlers. Settled water then flows through granular dual media filters. The filter effluent is
combined prior to entering a chlorine contact tank for primary disinfection. After the initial dose
of chlorine and a hydraulic retention to allow for sufficient contact time, additional chlorine is
added for secondary disinfection. Hydrofluosilic acid and sodium silicate are also added at this
point for fluorination and corrosion control, respectively. Sodium silicate addition is intended to
elevate the pH, which is typically depressed through alum addition, for corrosion prevention
prior to entering the distribution system.
3.1.2.1 Operational scheme
The pilot plant was being used to observe the possible cost savings and process benefits
of using Hyper+Ion 705 (HI705) polyaluminum chloride (PACl) as a coagulant as opposed to
alum. For comparison of these two coagulants, the operation scheme was to match settled water
TOC for both pilot treatment trains as well as the full-scale plant. The full-scale plant coagulant
dose was selected by the WTP staff based on raw water quality. Pilot-scale doses of both alum
and PACl are selected based on monthly jar tests to match full-scale settled water TOC.
TOC measurements were collected every two minutes by one online analyzer (General
Electric Sievers 900 On-Line), which alternated between four sampling points: two pilot plant
trains, full-scale, and raw water. The analyzer cycled between the sampling points
approximately every 2.5 hours. The manufacturer reported accuracy of this machine was ±2% of
the TOC value (General Electric, 2005).
Page 33
18
Figure 3.1: Process flow diagram of Peterborough water treatment plant
Page 34
19
3.2 Analytical methods
3.2.1 Quality control
Quality control measures were used in TOC/DOC, UVA, THM, and HAA analyses.
These included calibration, setting quality control limits, and periodic quality checks. Details
pertaining to each analysis method are presented in their specific method descriptions. Quality
control limits were set based on Standard Method 1020 (APHA, 2005). Limits were set based on
the standard deviation of 8 quality control standards as follows:
Control limit was set to µ ± 3 SD
Warning limit was set to µ ± 2 SD
where, µ was the average response or concentration from 8 quality control standards, and SD
was the standard deviation.
If any of the following conditions were satisfied, the instrument would be recalibrated:
2 consecutive measurements outside the control limit
3 of 4 consecutive measurements outside the warning limit
Method detection limits were also calculated for each analysis method. The MDL was
determined through multiplication of the standard deviation from 8 quality control standards and
the critical value of the Student’s t distribution at a 99% confidence level.
3.2.2 Total and dissolved organic carbon
For jar tests, both total and dissolved organic carbon was analyzed using an O-I
Corporation model 1030 TOC Analyzer (College Station, Texas). Pilot plant TOC methods are
described in section 3.1.2. Organic carbon measurements were based on the wet oxidation
method as described in Standard Method 5310 D (APHA, 2005). For analysis of the dissolved
fraction, samples were first filtered using a 0.45 µm filter (Pall Corporation Supor®-450, Port
Washington, NY) (Edzwald and Tobiason, 1999). Equipment conditions are listed in Table 3.3.
Samples were collected in 40 mL amber glass vials with Teflon® lined screw caps (VWR
International, Mississauga, ON). For preservation they were acidified to a pH < 2 using 3 drops
of concentrated sulphuric acid. If storage was required, samples were refrigerated at 4oC until
analysis, which was typically performed within 1 – 3 days.
Page 35
20
Table 3.3: Total and dissolved organic carbon instrument conditions
Parameter Condition
Sample volume 15 mL
Oxidant volume and type 1000 µL of 100 g/L sodium persulphate
Acid volume and type 200 µL of 5% phosphoric acid
Reaction time 150 seconds
Detection time 120 seconds
Purge gas Nitrogen
Calibration and quality control of organic carbon measurements was completed using
standards prepared from dry potassium hydrogen phthalate (KHP) (Sigma-Aldrich Corp.,
Oakville, ON) in Milli-Q® water. A six-point calibration (0, 0.625, 1.25, 2.5, 5, and 10 mg/L)
was completed approximately once per month or as dictated by quality control data. Every 10
samples run, one blank (Milli-Q® water) and one check standard prepared to 3.0 mg/L TOC was
run. Figure 3.2 shows an example TOC calibration and Figure 3.3 shows quality control
standards run for DBP modelling work presented in section 4 of this document. The limits in the
quality control chart was prepared based on 8 quality control standards (3.0 mg/L TOC) as
discussed in section 3.2.1. Table 3.4 reports the standard deviation and method detection limit
as found from the 8 quality control standards.
Table 3.4: Accuracy and method detection limit of TOC analysis
Parameter Value
Manufacturer reported accuracy ± 5 % of the measured value
Quality control standard deviation 0.10 mg/L
MDL (mg/L) 0.35 mg/L
3.2.3 Ultraviolet absorbance
Ultraviolet absorbance (UVA) was measured at 254 nm using a CE 3055
spectrophotometer (Cecil Instruments, Cambridge, England). The method followed is described
in Standard Method 5910 B (APHA, 2005). A quartz cuvette with 1 cm path length was used.
Prior to measurements, the spectrophotometer was zeroed using Milli-Q® water.
Page 36
21
Figure 3.2: Example TOC calibration
Figure 3.3: TOC quality control chart (July to Sep., 2012)
Area count = 6217.8(TOC) + 2499 R² = 0.9989
0
10000
20000
30000
40000
50000
60000
70000
0 2 4 6 8 10 12
Are
a C
ou
nt
TOC (mg/L)
2.9
3
3.1
3.2
3.3
3.4
3.5
3.6
Tota
l org
anic
car
bo
n (
mg/
L)
Limit (3SD) Warning (2SD) Average
Otonabee River
Lake Ontario
Lake Simcoe
Ottawa River
Page 37
22
3.2.4 Liquid chromatography – organic carbon detection
Liquid chromatography – organic carbon detection (LC-OCD) analysis was conducted
using the method described by Huber et al. (2010). Samples were passed through a 0.45 µm
filter to remove any particulates prior to analyses. A weak cation exchange column (250 mm x
20mm, Toso, Japan) provided chromatographic separation of the NOM fractions. The mobile
phase used was a phosphate buffer exposed to UV at a flow rate of 1.1 mL/min (MLE, Dresden,
Germany). The samples first passed through the UVA detector (254 nm) and then through the
organic carbon detector (OCD). Prior to entering the OCD, the solution was acidified to convert
organic carbon to carbonic acid. During the acidification process, bound nitrogen (organic and
inorganic) was converted to nitrate which absorbs light in the UV region (Huber et al., 2011).
This allows for the simultaneous quantification of bound nitrogen in each separated fraction
using the UVA detector. Calibration was performed by the on-site technician using potassium
hydrogen phthalate. Data acquisition and processing was carried out using a customized
software program (ChromCALC, DOC-LABOR, Karlsruhe, Germany). All data from the OCD,
organic nitrogen detector (OND), and UVD were collected, however only OCD data was used in
the analysis. An example LC-OCD chromatogram is shown as Figure 3.4.
Figure 3.4: LC-OCD chromatogram of surface water (Huber et al., 2011).
Responses shown for organic carbon detection (OCD), UVA detection (UVD),
and organic nitrogen detection (OND). Peak A: biopolymers, Peak B: humic
substances, Peak C: building blocks, Peak D: low molecular weight acids, Peak
E: low molecular weight neutrals, Peak F: nitrate, Peak G: ammonium
Page 38
23
Table 3.5: Description of LC-OCD fractions (Huber et al., 2011)
Fraction Description
Biopolymers High molecular weight based on elution time
Saturated compounds based on lack of UV absorbance
Non-ionic based on not being retained by cation and anion exchange resins
Some compounds contain nitrogen indicating presence of polysaccharides
and proteins
Humic
substances
Humic acids elute prior to fulvic acids
Contains carboxylic groups
One study placed 28% of organic carbon in Fulvic acids to be related to
aromatic structures, 42% for humic acids.
Building
blocks
Shoulder to humic substance peak indicating similar properties to humic
substances but with lower molecular weight
Breakdown products from humic substances
LMW acids Elute separately due to an ion chromatographic effect
Anions at neutral pH
LMW neutrals Low molecular weight and low ion density
Hydrophilic to amphiphilic as it elutes close to the permeation volume of the
column
Hydrophobic
organic carbon
Material which dos not elute in a standard run time due to strong
hydrophobicity
3.2.5 Chlorine residuals
Free chlorine residuals were measured with a HACH Odyssey D5/250 spectrophotometer
(HACH, Loveland, CO) following the DPD colourimetric method described in Standard Method
4500-Cl G (APHA, 2005). Prior to sample measurement a Milli-Q® blank was used to zero the
instrument. For sample measurement, one DPD pillow was added to 25 mL of sample in a clear
cuvette, capped, and shaken to rapidly mix. Absorbance was measured and free chlorine residual
was determined through the instrument’s internal calibration.
Page 39
24
3.2.6 pH measurements
Sample pH was measured using a Model 8015 pH meter (VWR Scientific Inc.,
Mississauga, ON). A three point calibration (pH: 4, 7, and 10) was completed prior to measuring
a set of samples. During measurement samples were stirred using a magnetic stirrer and bar.
3.2.7 Fluorescence excitation-emission matrices
The FEEMs were collected using a Perkin Elmer Luminescence Spectrometer LS50B
with FLWinLab Version 3.0 (Waltham, Massachusetts, USA). The collection protocol stipulated
collection of intensity values at 10 nm increments within excitation-emission ranges of 250-380
nm 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 800 V. UV-grade polymethylmetacrylate cuvettes
with four optical windows were used. To account for background noise and scattering
interference, spectra for Milli-Q© water was obtained using the same instrument settings.
Intensity values from the Milli-Q© samples were subtracted from intensity values of sample
spectra to reduce noise effects.
3.2.8 Principal component analysis
Each fluorescence sample produced a total of 4214 intensity values at unique excitation-
emission wavelength pairs. Prior to PCA, intensity data of each sample had to be arranged by
row. All data manipulation and organization of the spectrum was carried out using in-house data
manipulation scripts written in Python 2.6 (Python Software Foundation), which have been
included as section 8.4. Using the PLS Toolbox (Eigenvector Research Inc., Manson, WA) for
MATLAB 7.12.0 (MathWorks, Natick, MA), principal component analysis (PCA) was applied
to the fluorescent excitation-emission matrices.
The model input data was pre-processed using the PLS toolbox. Autoscaling was applied
to the intensity values to reduce bias. Autoscaling centers the data on the mean and scales each
variable to a unit standard deviation. Without autoscaling, the PCA technique would favour
higher intensity values and skew the components found to those with higher fluorescence
intensities, neglecting smaller peaks. For model validation, a cross-validation method of random
subsets (subset size of approximately 5-10% of the sample size) was selected. From the PCA
model, loading values and score values were collected for analysis.
Page 40
25
3.2.9 Trihalomethanes
Trihalomethane (THM) analysis was conducted using the EPA Method 551.1. All four
THM species were analyzed (trichloromethane/chloroform, bromodichloromethane,
dibromochloromethan, and tribromomethane/bromoform). The liquid-liquid extraction method
outlined by the USEPA (USEPA, 1990) was carried out using MTBE (VWR International,
Mississauga, ON, ACS grade > 99.8%). Stock standard solutions (2000 µg/L of each specie)
from Supelco Inc. (Bellefonte, Pennsylvania, USA), were used for calibration and quality
control. All standards were kept in sealed glass vials at -4oC. MTBE extracts were analyzed with
a Hewlett Packard 5890 Series II Plus gas chromatograph (Mississauga, Ontario, CA) equipped
with an electron capture detector. A J&W Science DB-5.625 durabond column (length: 30m,
inner diameter: 0.25 mm, film: 0.25 µm) (Agilent Technologies Canada Inc., Mississauga,
Ontario, CA) was used. Injections were run in splitless mode, with helium as carrier gas and an
argon/methane (95%/5%) mix as makeup gas. Table 3.6 reports the instrument conditions and
temperature program used.
Table 3.6: THM analysis instrument conditions
Parameter Value
Injector temperature 200oC
Detector temperature 300oC
Temperature program
0 – 10 min: 35oC
10 – 16.25 min: 35oC – 60
oC
17.25 – 19.75 min: 60oC – 110
oC
21.25 – 25.58 min: 110oC – 240
oC
Carrier gas Helium
Makeup gas 95% argon, 5% methane
Flow rate 1.2 mL/min at 35oC
An eight-point calibration was completed at concentrations of 2, 4, 10, 20, 30, 40, 60, and
80 µg/L. A series of sixteen quality control standards were run to determine method detection
limits. Eight of the quality control standards were prepared to a concentration of 10 µg/L of each
specie with the remaining eight prepared to a target concentration of 60 µg/L. With every 10
samples analyzed, two check standards were run (10 and 60 µg/L) along with a blank (no
Page 41
26
addition of stock to Milli-Q®). All blank standards analyzed as part of this work did not display
background interference. An example calibration chart of each THM specie is shown as Figure
3.5. Method detection limits and standard deviations (SD) for the THM analysis is reported in
Table 3.7. Example quality control charts (Figure 3.6 and Figure 3.7) for trichloromethane at
both low and high concentrations. An overview of the method is presented as Table 3.8.
Table 3.7: Accuracy and method detection limit of THM analysis
Parameter TCM BDCM DBCM TBM
low high low high low high low high
Quality control standard
deviation (µg/L) 0.7 3.9 0.9 2.4 0.9 1.8 0.6 2.2
MDL (µg/L) 2.5 NAa
3.0 NAa 3.0 NA 2.1 NA
a
Low: 10 µg/L target, high: 60 µg/L target.
NAa Not Applicable: high quality standards not used to determine MDL
Figure 3.5: Typical calibration chart for THM species (July 2012)
y = 0.8919x + 0.3251 R² = 0.9920
y = 4.4197x - 2.1403 R² = 0.9864
y = 3.8725x + 2.9656 R² = 0.9764
y = 1.8124x + 0.7383 R² = 0.9905
0
50
100
150
200
250
300
350
400
0 10 20 30 40 50 60 70 80 90
Are
a C
ou
nt
Concentration (µg/L)
trichloromethane bromodichloromethane dibromochloromethane tribromomethane
Page 42
27
Figure 3.6: Trichloromethane (TCM) quality control - low concentration target: 10 ug/L
Figure 3.7: Trichloromethane (TCM) quality control - high concentration target: 60 ug/L
6
7
8
9
10
11
12
13
Tric
hlo
rom
eth
ane
Co
nce
ntr
atio
n (
ug/
L)
Otonabee Simcoe Ontario Ottawa
3 SD Limit 2 SD Warning Average
45
49
53
57
61
65
69
73
Tric
hlo
rom
eth
ane
Co
nce
ntr
atio
n (
ug/
L)
Otonabee Simcoe Ontario Ottawa
3 SD Limit 2 SD Warning Average
Page 43
28
Table 3.8: Trihalomethane analysis method outline
Collect quenched samples in 25 mL amber vials
Store samples in the dark at 4oC for up to 14 days
Bring samples to room temperature, prior to analysis
All samples and standards are prepared in 40 mL amber glass vials
Blanks: Add 25 mL of Milli-Q water into 40 mL amber vials to be prepared alongside samples
Standard Working Solution: (10 µg/mL)
1. Fill a 5 mL volumetric flask partially with methanol
2. Using a 50 µL syringe, add 25 µL of THM stock (2000 µg/mL – Supelco 48140-U) to the
volumetric flask below the surface of the methanol.
3. Add methanol to the volumetric flask to the 5 mL mark.
4. Cap with a glass stopper and invert 5 times to ensure mixing
Calibration: (2 µg/L - 80 µg/L)
1. Add appropriate amounts of the standard working solution to 25 mL of Milli-Q to
achieve an 8 point calibration between 2 and 80 µg/L.
Quality Standards: (10 µg/L and 60 µg/L)
1. Quality standards of 10 µg/L and 60 µg/L are made by respectively adding 25 µL and
150 µL of the standard working solution into 25 mL Milli-Q samples.
2. 1 quality standard of each concentration is prepared for every 10 samples run
3. A set of 8 quality standards at each concentration is prepared for every calibration
prepared
Extraction
1. To the 40 mL amber glass vial:
2. Add 1 teaspoon (5 mL) of sodium sulphate using a scoop
3. Add 4 mL of MTBE extraction solvent
4. Cap with Teflon-lined silicon septa and screw cap
Page 44
29
Table 3.8: Trihalomethane analysis method outline (continued)
5. Shake vial vigorously for approx. 10 seconds
6. Repeat for all samples and standards
7. Place all vials upright in the rack and shake for 3 minutes
8. Let the sample stand for 10 minutes prior to phase separation
9. Using a glass Pasteur pipette, extract the top organic layer (MTBE) from the vial and
transfer to a clean 1.8 mL GC vial. Allow for no headspace in the vial. Use a clean
pipette for each sample. Cap immediately
10. Place GC vials in freezer overnight to freeze out any remaining water
11. Once frozen, extract the top 1.5 mL from the GC vial and transfer to a new, clean 1.8 mL
GC vial
12. Analyze using GC ECD.
3.2.10 Haloacetic acids
Haloacetic acid analyses were performed using a liquid-liquid extraction method (EPA
Method 552.3) using MTBE (VWR International, Mississauga, ON, ACS grade > 99.8%). Nine
haloacetic acid species were included in the analysis: monochloroacetic acid (MCAA),
monobromoacetic acid (MBAA), dichloroacetic acid (DCAA), trichloroacetic acid (TCAA),
bromochloroacetic acid (BCAA), dibromoacetic acid (DBAA), bromodichloroacetic acid
(BDCAA), dibromochloroacetic acid (DBCAA), and tribromoacetic acid (TBAA). Standard
solutions from Supelco Inc. (Bellefonte, Pennsylvania, USA), were used for calibration and
quality control. Stock solution had variable concentration of each component (Table 3.9). All
standards were kept in sealed glass vials at -4oC. Samples were analyzed with a Hewlett Packard
5890 Series II Plus gas chromatograph (Mississauga, Ontario, CA) equipped with an electron
capture detector. A J&W Science DB-5.625 Durabond column (length: 30m, inner diameter:
0.25 mm, film: 0.25 µm) (Agilent Technologies Canada Inc., Mississauga, Ontario, CA) was
used. Injections were run in splitless mode, with helium as carrier gas and an argon/methane
(95%/5%) mix as makeup gas. Instrument conditions are summarized in Table 3.10.
A six-point calibration was completed at concentrations at varying concentration ranges,
dependent on the concentration of the specie within the stock solution. A series of eight quality
control standards prepared at a low dose were run to determine method detection limits. With
every 10 samples analyzed, a check standard was run along with a blank (no addition of stock to
Page 45
30
Milli-Q®) (Table 3.11). All blank standards analyzed as part of this work did not display
background interference. An example calibration chart is shown as Figure 3.8 for all HAA
species. Table 3.12 reports the method detection limits and standard deviations for the THM
analysis. An example quality control chart for dichloroacetic acid (DCAA) at a target
concentration of 5.6 µg/L is shown as Figure 3.9. Check standards for Lake Ontario water
exceeded the quality limits (maximum of 3.1 µg/L away from average). An overview of the
HAA analysis method is presented as Table 3.13.
Table 3.9: Stock HAA specie concentrations
HAA specie Stock concentration (µg/L)
Bromoacetic Acid 387.9
Bromochloroacetic Acid 406.5
Bromodichloroacetic Acid 435.8
Chloroacetic Acid 573.1
Chlorodibromoacetic Acid 1003
Dibromoacetic Acid 212.3
Dichloroacetic Acid 590.4
Tribromoacetic Acid 2200
Trichloroacetic Acid 204.0
Table 3.10: HAA analysis instrument conditions
Parameter Value
Injector temperature 200oC
Detector temperature 300oC
Temperature program
0 – 9.5 min: 50oC
9.2 - 22 min: 50oC – 65
oC
22 - 24 min: 65oC – 85
oC
24 - 30 min: 85oC – 205
oC
Carrier gas Helium
Makeup gas 95% argon, 5% methane
Flow rate 1.2 mL/min at 35oC
Page 46
31
Table 3.11: Calibration range and check standard target concentrations for HAAs
HAA Specie Calibration Range
(6 point calibration)
(µg/L)
Check Standard Target
Concentration (µg/L)
Bromoacetic Acid 1-31 7.8
Bromochloroacetic Acid 1-33 8.1
Bromodichloroacetic Acid 1-35 8.7
Chloroacetic Acid 1-46 5.6
Chlorodibromoacetic Acid 2-80 20.1
Dibromoacetic Acid 0.5-17 4.3
Dichloroacetic Acid 1-47 5.6
Tribromoacetic Acid 4-176 44
Trichloroacetic Acid 0.5-16 4
Figure 3.8: Typical HAA calibration curves (July 2012)
MCAA: y = 0.0286x - 0.0996, R2 = 0.9828 MBAA: y = 0.2961x - 0.4946, R2 = 0.9973
DCAA: y = 0.6234x + 0.1464, R² = 0.983
TCAA: y = 0.4563x - 0.6356, R2 = 0.9953
BCAA: y = 0.8294x - 0.9536, R2 = 0.9934 DBCAA: y = 0.5731x + 1.0879, R² = 0.9913
DBAA: y = 0.5383x + 0.0883, R² = 0.9951
CDBAA: y = 0.0917x - 0.042, R2 = 0.9926 TBAA: y = 0.7677x - 4.2011, R2 = 0.9948
0
10
20
30
40
50
60
70
80
0 10 20 30 40 50 60 70 80 90
Are
a C
ou
nt
Concentration (µg/L)
MCAA MBAA DCAA TCAA BCAA DBAA DBCAA CDBAA TBAA
Page 47
32
Figure 3.9: Example HAA quality control chart for DCAA
Table 3.12: Quality control standard deviation and HAA method detection limits
HAA Specie Quality Control Standard
Deviation (µg/L)
Method Detection Limit
(µg/L)
Bromoacetic Acid 1.5 5.2
Bromochloroacetic Acid 0.3 1.1
Bromodichloroacetic Acid 0.4 1.3
Chloroacetic Acid 0.2 0.7
Chlorodibromoacetic Acid 0.3 1.1
Dibromoacetic Acid 0.2 0.7
Dichloroacetic Acid 0.7 2.5
Tribromoacetic Acid 2.7 9.6
Trichloroacetic Acid 0.5 1.8
4
5
6
7
8
9
10
Dic
hlo
roac
eti
c ac
id C
on
cen
trat
ion
(u
g/L)
Otonabee Ontario Ottawa
3 SD Limit 2 SD Warning Average
Page 48
33
Table 3.13: HAA analysis method outline
Collect quenched samples in 25 mL amber vials
Store samples in the dark at 4oC for up to 14 days
Bring samples to room temperature, prior to analysis
All samples and standards are prepared in 40 mL amber glass vials
Sulfonamide Solution:
1. Add 15 mL of dietylene glycol, 15 mL of ether and 3 g of N-methyl-N-nitroso-p-toluene
sulphonamide (Diazald) to a 40 mL amber vial
2. Shake until completely dissolved. This solution can be used to make diazomethane and
can be stored at 4oC for up to 30 days
Diazomethane Generation:
1. Prior to extraction, diazomethane must be produced
2. Set up generation apparatus as shown in Figure 6521:3 in Standard Methods (APHA,
2005).
3. Fill the first tube with ether to a depth of 3 cm
4. Add potassium hydroxide solution (370 g/L KOsolution in Milli-Q water) to the second
tube so that it is just touching the base of the impinge.
5. Add sulphonamide solution above the KOusing a long Pasteur pipette; ensure no mixing
occurs
6. Add 4 mL of MTBE to the last tube and put it in a beaker of ice such that the impinge is
submerged in the MTBE.
7. Connect the nitrogen gas feed to the in port of the apparatus
8. Slowly turn on the gas flow and bubble the nitrogen gas through the apparatus slowly
until the MTBE solution becomes yellow
9. This solution may be stored at 4oC in a 25 mL amber vial for up to 24 hours
10. Allow for the diazomethane to warm up to room temperature prior to use in extraction.
Blanks:
1. Add 25 mL of Milli-Q water into 40 mL amber vials to be prepared alongside samples
Page 49
34
Table 3.13: HAA analysis method outline (continued)
Standard Working Solution: (10 µg/mL)
1. Fill a 5 mL volumetric flask partially with methanol
2. Using a 50 µL syringe, add 25 µL of HAA9 stock (2000 µg/mL Supelco-47787) to the
volumetric flask below the surface of the methanol.
3. Add methanol to the volumetric flask to the 5 mL mark.
4. Cap with a glass stopper and invert 5 times to ensure mixing
Calibration: (2 µg/L - 80 µg/L)
1. Add appropriate amounts of the standard working solution to 25 mL of Milli-Q to
achieve an 8 point calibration between 2 and 80 µg/L.
Quality Standards: (10 µg/L and 60 µg/L)
2. Quality standards of 10 µg/L and 60 µg/L are made by respectively adding 25 µL and
150 µL of the standard working solution into 25 mL Milli-Q samples.
3. 1 quality standard of each concentration is prepared for every 10 samples run
4. A set of 8 quality standards at each concentration is prepared for every calibration
prepared
Extraction
1. To the 40 mL amber glass vial:
2. Add 1 teaspoon (5 mL) of sodium sulphate using a scoop
3. Add 1 mL of sulphuric acid
4. Add 4 mL of MTBE extraction solvent
5. Cap with Teflon-lined silicon septa and screw cap
6. Shake vial vigorously for approx. 10 seconds
7. Repeat for all samples and standards
8. Place all vials upright in the rack and shake for 3 minutes
9. Let the sample stand for 10 minutes prior to phase separation
10. Using a glass Pasteur pipette, extract the top organic layer (MTBE) from the vial and
transfer to a clean 1.8 mL GC vial. Allow for 100 µL of headspace.
Page 50
35
Table 3.13: HAA analysis method outline (continued)
11. Using an auto pipette, transfer 100 µL of diazomethane to the top of the vial and cap
12. Place GC vials in freezer overnight to freeze out any remaining water
13. Once frozen, extract the top 1.5 mL from the GC vial and transfer to a new, clean 1.8 mL
GC vial
14. Analyze using GC ECD.
Page 51
36
4 Disinfection By-Product Modelling using Principal Component Analysis of
Fluorescence Excitation-Emission Matrices
4.1 Introduction
For decades the drinking water industry had understood the importance of monitoring and
studying the formation of by-products associated with disinfection processes. In North America
chlorine remains the most common disinfectant used by water utilities (AWWA, 2007) which,
when added to natural waters, forms organic halides through reactions with natural organic
material (NOM) (Richardson and Postigo, 2011). Control and regulations for utilities using
chlorine typically revolve around concentrations of two organic halide groups, trihalomethanes
(THMs) and haloacetic acids (HAAs) (Beggs et al., 2009). Since DBPs are principally formed
through the reaction of a disinfectant with NOM, any predictive model must incorporate NOM
concentration. Difficulties arise since NOM is comprised of a complex heterogeneous mixture
of humic acid, fulvic acid, proteins, carbohydrates, and other organic compound classes (Her et
al., 2003). Each of these classes has unique reactivity with disinfectants to form DBPs, which
implicates the importance of understanding the character, or concentrations of specific
compound classes, considered to be within the definition of NOM (Barrett et al., 2000).
Historically, predictive DBP models have been successfully linked to NOM estimation
parameters including total organic carbon (TOC), dissolved organic carbon (DOC), UV-
absorbance (UVA) at 254 nm, and specific UV-absorbance (SUVA) (Sadiq & Rodriguez, 2004).
In particular, SUVA (DOC divided by UVA at 254 nm), has been shown to correlate well with
DBP formation (Kitis et al., 2001; Barrett et al., 2000). SUVA may be used to provide an
estimate of humic content, which has been generally recognized as the NOM fraction most
reactive to form DBPs (Barrett et al., 2000). In contrast, both TOC and DOC provide an
estimate of the total amount of organic content, however cannot provide information regarding
NOM reactivity or structure. With knowledge that specific NOM fractions are more significant
in the formation of DBPs, it is postulated that methods which can characterize those fractions
will improve the accuracy of DBP formation models.
Analysis using fluorescence spectra has been gaining traction as a promising method for
determining information regarding organic matter structure and function in water (Zepp et al.,
2004; Bieroza et al., 2011). This type of analysis is completed through collection of
Fluorescence Excitation-Emission Matrices (FEEM). These matrices represent fluorescence
Page 52
37
intensity at various excitation-emission pairs. When compared to other NOM characterization
techniques, including liquid chromatograph organic carbon detection (LC-OCD), FEEM analysis
is fast and provides consistent results with high sensitivity (Peiris et al., 2008). Several
components of NOM fractions fluoresce and therefore can be identified in a fluorescence
spectrum (Her et al., 2003). Most importantly, different compounds and compound groups
exhibit unique fluorescence signatures that allow them to be distinguished from each other
(Coble et al., 1990; Zepp et al., 2004). Traditional analysis of FEEMs involves identifying peaks
of fluorescence intensity. The position (excitation-emission region) of the peak can be correlated
with a specific compound and the maximum intensity of the peak with concentration (Bieroza et
al., 2011). This method, referred to as ‘peak picking,’ is limited in that it neglects to capture the
heterogeneity of NOM fractions (Peiris et al., 2010). The fluorescence intensity at specific
excitation/emission pairs is influenced by several variables including concentration, quantum
yield, and absorptivity. Molecular structure has a noted effect on some of these factors which
can be common among different molecules. By ‘peak picking’ the nature of overlapping
structural properties between distinct compounds and many of the potentially relevant
chromophores are neglected (Persson and Wedborg, 2001).
To further extract information and, in particular, changes in NOM composition between
several samples, multivariate analysis techniques can be employed. Common advanced analysis
techniques involve using two-way principal component analysis (PCA) and multi-way parallel
factor (PARAFAC) analysis (Peiris et al., 2010). These multivariate techniques reduce the
number of variables needed to explain variance between samples in a given set (Persson and
Wedborg, 2001). Furthermore, this pattern recognition and data set simplification approach,
unlike ‘peak picking’, incorporates the entire fluorescence spectrum (Bieroza et al., 2011).
This work provides evidence regarding the application and robustness of principal
component analysis of FEEMs (FEEM-PCA) in the water treatment field. Specifically, FEEM-
PCA is compared to traditional NOM estimators including DOC, UVA, and SUVA for the
purpose of predicting the formation of THMs and HAAs from four different source waters to
water treatment facilities. In order to assess how FEEM-PCA may be related to other NOM
indicators, jar tests were conducted using a range of coagulant doses (5 – 70 mg/L), to provide a
large resulting range of NOM concentrations and associated DBP concentrations.
Page 53
38
4.2 Method overview
4.2.1 Source waters
Four distinct surface water sources in Ontario, Canada were used in this study. They
were selected to cover a range of NOM concentrations (2.4 – 5.9 mg/L DOC) and represent both
lakes and rivers. Water quality parameters are reported in Table 4.1.
Table 4.1: Source water characteristicsa
Otonabee River Lake Simcoe Lake Ontario Ottawa River
DOC (mg/L) 5.6 4.1 2.4 5.9
Alkalinity (mg/L
as CaCO3) 101 121 87 35
pH 8.3 8.2 8.3 7.4
a All waters were collected between June and September of 2012.
4.2.2 Jar tests
To simulate the processes of coagulation, flocculation, sedimentation, and filtration for
each water source a bench-scale jar test approach was applied. All waters were coagulated with
aluminum sulphate (alum) (General Chemical, Parsipanny, NJ). To ensure a range of DOC an
alum dose range of 5 – 70 mg/L (5, 10, 20, 30, 40, 50, 60, 70 mg/L alum or 0.45, 0.89, 1.78,
2.67, 3.56, 4.45, 5.34, 6.23 mg/L as Al) was applied to each of the four waters. Jar tests were
conducted using a PB-700 Standard Jar Tester paddle stirrer with six square, acrylic 2-L jars
(Phipps & Bird, Richmond, VA). The protocol followed was adapted from the USEPA’s
Enhanced Coagulation Guidance Manual (USEPA, 1999). Coagulation was simulated through
rapid mix (100 RPM) for 1.5 minutes and flocculation by reducing the mixing speed to 30 RPM
for 15 minutes. The water was then allowed to stand (no mixing) for 30 minutes to simulate
sedimentation. Vacuum filtration was performed on the settled water using 1.2 micron glass
microfibre filters (Whatman 934-AH, Florham Park, NJ) to represent anthracite-sand media
filters (Wassink, 2011). This finished water was analyzed for organic material (TOC, FEEM)
and pH. For specific analyses including DOC and LC-OCD the water was further filtered
through 0.45 micron membrane filters (Pall Corporation Supor®-450, Port Washington, NY) to
ensure that all particulates had been removed (USEPA, 1999).
Page 54
39
4.2.3 Disinfection by-product formation and analysis
Filtered water collected from the jar tests was chlorinated in order to form disinfection
by-products. To ensure equal conditions for DBP formation, the pH of the finished water was
adjusted to 7 ± 0.1 using sulphuric acid or sodium hydroxide prior to chlorination. Chlorine
dosages of 2.5 and 3.5 mg/L were applied to represent those used at water treatment plants in the
region. Chlorinated samples were sealed with no head space and incubated at 21 ± 1oC for 24
hours. Following 24 hours (± 0.5 hour) the free chlorine residual was measured (0.02 to 2.54
mg/L) and remaining chlorine quenched using ascorbic acid (100 mg/L ascorbic acid). Duplicate
25 mL samples of quenched solution were sealed with no head space and retained for THM and
HAA9 analysis. All waters were tested for THM formation and all except Simcoe was analyzed
for HAA9 formation.
Trihalomethane analysis was conducted using the EPA Method 551.1. The liquid-liquid
extraction method was carried out using MTBE (USEPA, 1990). This method allowed for
quantification of four THM species: trichloromethane (TCM), bromodichloromethane (BDCM),
chlorodibromomethane (CDBM), and tribromomethane (TBM). For HAA9, the EPA Method
552.3 was used. The nine HAA species analyzed using this method included: monochloroacetic
acid (MCAA), monobromoacetic acid (MBAA), dichloroacetic acid (DCAA), trichloroacetic
acid (TCAA), bromochloroacetic acid (BCAA), dibromoacetic acid (DBAA),
bromodichloroacetic acid (BDCAA), dibromochloroacetic acid (DBCAA), and tribromoacetic
acid (TBAA). For both analyses, stock standard solutions from Supelco Inc. (Bellefonte,
Pennsylvania, USA), were used for calibration and quality control. All standards were kept in
sealed glass vials at -4oC. Analyses were conducted using a Hewlett Packard 5890 Series II Plus
gas chromatograph (Mississauga, Ontario, CA) equipped with an electron capture detector and a
J&W Science DB-5.625 durabond column (length: 30m, inner diameter: 0.25 mm, film: 0.25
µm) (Agilent Technologies Canada Inc., Mississauga, Ontario, CA). Injections were run in
splitless mode, with helium as carrier gas and an argon/methane (95%/5%) mix as makeup gas.
4.2.4 DOC, TOC, and UVA measurements
Concentrations of DOC and TOC were obtained through heated persulfate oxidation
using an O.I. Analytical Aurora 1030 organic carbon analyzer (College Station, Texas, USA)
following Standard Method 5310 D (APHA, 2005). UVA at 254 nm was determined using a CE
3055 model spectrophotometer (Cecil Instruments, Cambridge, England) with a quartz cuvette
Page 55
40
following Standard Method 5910 B (APHA, 2005). The instrument was first zeroed at 254 nm
using Milli-Q® water prior to analysis of samples.
4.2.5 Fluorescence spectra collection
FEEMs were collected using a Perkin Elmer Luminescence Spectrometer LS50B
(Waltham, Massachusetts, USA) with FLWinLab Version 3.0. No pre-treatment of the samples
was applied, except that all had been adjusted to a pH of 7. Maintaining a common pH between
samples ensured that fluorescence characteristics of the acidic functional groups in humic
molecules remained constant (Mobed et al., 1996). Collection of intensity values at 10 nm
increments within excitation-emission ranges of 250-380 nm 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 (Peiris et al., 2010; 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. It has been shown that, while the
PMMA cuvettes used reduce the intensity of excitation wavelengths below 285 nm, this
approach is appropriate for purposes of distinguishing NOM elements using fluorescence (Peiris
et al., 2008). To account for background noise and scattering interference, spectra for Milli-Q®
water was obtained using the same instrument settings. Intensity values from the Milli-Q®
samples were subtracted from intensity values of sample spectra to reduce background noise
effects.
4.2.6 Fluorescence data analysis
Each sample produced a total of 4214 fluorescence intensity values at unique excitation-
emission wavelength pairs. In total, 35 unique samples were run in duplicate (70 FEEMs
collected in total). Prior to PCA, intensity data of each sample was arranged by row. All data
manipulation and organization of spectra was carried out using in-house data manipulation
scripts written in Python 2.6 (Python Software Foundation). This processing step generated
several unique data sets for PCA: one of all samples (XA: 70 x 4214 matrix), and one for each
water source (Otonabee XOE: 16 x 4214, Simcoe XSI: 18 x 4214, Ontario XON: 18 x 4214, Ottawa
XOT: 18 x 4214).
PCA is a multivariate data analysis technique that is used to approximate a large data
matrix through pattern detection. Common goals for PCA include the identification of
Page 56
41
similarities between samples, data simplification, and detection of outliers (Wold et al., 1987).
Approximation of the dominant data patterns is achieved through new, mutually independent,
variables that are mathematically represented by linear combinations of the original variables
(Persson and Wedborg, 2001). In generalized terms, the original data matrix X is decomposed
into the product of two small matrices, ti and pi which are referred to as the scores and loading
matrix, respectively (equation 4.1). The residuals generated from the approximation are
represented in another matrix E. For multiple samples, n, in the matrix X, the product and
residuals are summed (Wold et al., 1987; Peiris et al., 2010).
∑
The loading matrix, pi, is a projection matrix that relates the PCA space to the original
spectral space. The information in the loading matrix can be used to identify the spectral areas
which a specific principal component represents. The scores matrix, ti, represents projections of
the samples, or object, into the principal component (PC) space. They serve as an analog for
concentrations or relative importance of a PC for a given sample (Persson and Wedborg, 2001).
PCA was applied independently, creating 5 PCA models of the variance in each data set.
PCA was performed using the PLS Toolbox V6.7.1 (Eigenvector Research Inc., Manson, WA) in
MATLAB 7.12.0 (MathWorks, Natick, MA).
To track any potential instrumental changes, Milli-Q® samples that were collected over
the experimental period were compared using the uncorrected matrix correlation (UMC) method.
This method compares two entire matrices and indicates the similarity using a value from 0 to 1
(1 representing a perfect correlation). Using this technique the realtive mean square error
(RMSE) between the matrices can be estimated (Burdick and Tu, 1989). All matrix correlations
had RMSE values below 0.06 indicating a high degree of similarity between Milli-Q© spectra
and implying instrument and hardware stability during the experimental time period.
4.3 Results and discussion
4.3.1 Conventional parameter results from jar tests
Based on traditional NOM indicators (DOC, UVA, SUVA), it was observed that
increasing alum dose caused a decreasing concentration of organic material (Figure 4.1). Within
(4.1)
Page 57
42
the context of this work, these results are being used as reference data for determining the
strength of fluorescence results, presented later.
Figure 4.1: DOC, UVA, and SUVA results for all waters over the range of coagulant doses
Although the raw Ottawa River water was highest in DOC, it showed increased
sensitivity to coagulation in the range of 10 to 30 mg/L alum and DOC was reduced at a greater
rate when compared to the other waters (46% DOC reduction compared to 10 – 20% DOC
reduction between dosages of 10 to 30 mg/L alum). This effect was also observed in UVA
results. Furthermore, SUVA for Ottawa River water was not lower than that of the Otonabee
River at any alum dose, as was the case for DOC and UVA measurements. When considering
SUVA, the results suggest that while the Ottawa DOC was reduced the most, the reactive portion
of DOC was not as well removed at higher alum doses. This result was likely due to the low
alkalinity, 35 mg/L as CaCO3 for the Ottawa River, compared to 100 mg/L or greater for the
other water sources. The pH of Ottawa River water was reduced by the greatest amount over the
range of doses (7.3 to 5.2) due to its limited buffering capacity. This reduced pH improves the
efficiency of alum coagulation and ultimately results in greater NOM removal (Hu et al., 2006).
0.00
0.05
0.10
0.15
0.20
0.25
0 20 40 60 80
Alum dose (mg/L)
Lake Simcoe Otonabee River Lake Ontario Ottawa River
1.0
2.0
3.0
4.0
5.0
6.0
0 20 40 60 80
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0 20 40 60 80
DOC (mg/L) UVA (cm-1
) SUVA (cm-1
/ mg/L)
Error bars represent standard deviation of duplicate samples
Page 58
43
4.3.2 Fluorescence results from jar tests
Fluorescence intensity values for each coordinate pair were plotted to display the spectra
for a given sample. It has been previously established by others that fluorescence in distinct
spectral regions is associated with functional groups and molecular structure, while intensity is
proportional to concentration (Mobed et al., 1996; Her et al., 2003; Zepp et al., 2004). Based on
the location of the two main intensity peaks (Figure 4.2) and comparison to the literature, peak a
represents humic-acid type matter (Ex/Em: 270nm/450nm) while peak b represents fulvic-acid
type material common to fresh waters (Ex/Em: 340nm/440nm) (Stedmon and Markager, 2005;
Murphy et al., 2008). Peaks of high intensity at Ex/Em: 250-300 nm / 500 – 600 nm a and
Ex/Em: 300 – 380 nm / 300 – 380 nm are representative of second and first order Rayleigh
Scattering, respectively, which are both related to the concentration of particulates in the sample
(Peiris et al., 2010). Each of the four water sources differed in overall intensity of these two
peaks; however their presence and location was approximately consistent (Figure 4.2 to Figure
4.5). The excitation/emission location of peaks a and b (highest measured intensity) are reported
in Table 4.3. The similarity of peak locations between the four source waters speaks to the
similarity of their organic character, even when comparing lake and river sources.
Table 4.2: Excitation/emission peak location for each water source
Peak Excitation/Emission of peak (nm/nm)
Otonabee Simcoe Ontario Ottawa
a 280/430 280/437 280/428 280/437
b 340/434 340/431 340/429 340/443
Figure 4.2: Example fluorescence spectra (Otonabee River raw water)
a a
b
b
Note: 3D and 2D views of identical spectra
Page 59
44
Figure 4.3: Fluorescence spectra of Ottawa River raw water
Figure 4.4: Fluorescence spectra of Lake Simcoe raw water
Figure 4.5: Flourescence specctra of Lake Ontario raw water
Note: 3D and 2D views of identical spectra, intensities truncated at 200 a.u.
Note: 3D and 2D views of identical spectra, intensities truncated at 200 a.u.
a a
b
b
a
a b
b
a a
b b
Note: 3D and 2D views of identical spectra
Page 60
45
Five PCA models were created, one which included samples from all water sources and
four source-specific models. For each model the first 3 principal components explained the
majority of variance in the sample set (> 75% variance explained). A cut-off of 5% variance
explained by a PC was used to determine the extent of the PCA models. At rates below 5%, the
importance of the PC was deemed negligible and considered to be noise. A summary of the
variance explained by each principal component is shown in Table 4.3.
Table 4.3: Variance explained by each PC for each sample set
Principal
Component
Variance explained (%)
All Samples
(XA)
Otonabee
(XOE)
Simcoe (XSI) Ontario (XON) Ottawa (XOT)
1 74.5 87.4 85.5 56.0 76.5
2 13.8 6.2 8.3 9.6 14.6
3 6.3 NA NA 6.4 NA
Total Variance
Explained (%) 94.5 93.7 93.8 75.1 91.1
NA = Not applicable; variance explained by PC below 5%.
With the exception of Lake Ontario, the 3 PCs described > 91% of the variance for each
sample set. Since Lake Ontario has a low concentration of organic material (TOC ~ 2.5 mg/L)
and did not have a pronounced response to coagulation, the overall variance was lower. As such,
the signal (change due to coagulation) was small when compared to other water sources.
By plotting loading values and their coordinates (excitation/emission), we can discern
what each of the PCs represents in the original spectral space. Loading plots for a model
comprised of all samples is shown in Figure 4.6. Similar plots were produced for each of the
individual waters (not shown). Loading plots for PC 1 show that it represented humic material,
however there were no pronounced peaks which further reinforces the heterogeneity of the
organic material. Loading results for PC 2 show peaks in low excitation/emission regions
(Ex/Em: 250-290 nm / 300-350 nm), representative of protein-like substances (Stedmon and
Markager, 2005; Chen et al., 2003; Peiris et al., 2010). PC 3 was found to be similar to PC 2
with high loading values in the low excitation/emission range; however scattering regions were
more pronounced (Figure 4.6). Therefore, it is thought that PC 3 is mostly representative of
scattering and to a lesser degree, protein-like material.
Page 61
46
Figure 4.6: Loading plots for three principal components of the ‘all samples’ model from
sample set XA.
Scores are analogous to concentration of the spectral elements represented by each PC for
a given sample, and were plotted to show the changes relative to alum dose (Figure 4.7). PC 1
scores for humic-like substances mimic the DOC and UVA results well. The exception is the
more pronounced increase in the Ottawa River water score following a dose of 50 mg/L alum. .
PC 2 results show limited changes as a function of coagulant dose. This follows from the known
mechanisms of coagulation with alum; protein-like substances are not significantly impacted.
PC 1
PC 2
PC 3
Page 62
47
Ottawa River results differed from the other waters showing a pronounced increase in protein-
like concentration between alum doses of 30 and 50 mg/L.
Figure 4.7: Scores from all sample PCA model (XA)
4.3.3 Relationships between NOM indicators
One of the primary objectives of this work was to demonstrate the viability of using
FEEM-PCA to estimate NOM concentrations and character. As such, there is a need to compare
the results of FEEM-PCA with accepted and established measures of NOM.
Linear relationship strength between PC scores and traditional NOM indicators are
reported in Table 4.4 as R2 values. Overall, relationships can be considered strong to very strong
(R2 > 0.8). Patterns regarding which parameter is most correlated with FEEM-PCA are not
consistent between waters. With the exception of the Ottawa River, DOC is most correlated with
PC 1 score, followed by UVA and SUVA, respectively.
In an effort to improve the correlations, PC 2 score (protein-like) was added to that of PC
1 with the intention of producing a value more representative of ‘total’ NOM. The effect of
combining PC 1 and PC 2 scores on the correlations was not consistent between waters. In
comparison to correlations using only PC 1 scores, correlations with all samples decreased in
-100
-80
-60
-40
-20
0
20
40
60
80
0 20 40 60 80
Alum dose (mg/L)
Lake Simcoe Otonabee River Lake Ontario Ottawa River
Note: Error bars represent standard deviation of duplicate samples
-150
-100
-50
0
50
100
150
200
0 20 40 60 80
-150
-100
-50
0
50
100
150
200
250
0 20 40 60 80
PC 1 Score PC 2 Score PC 3 Score
Page 63
48
strength for DOC (0.90 to 0.87) and UVA (0.90 to 0.66), while increasing with SUVA (0.79 to
0.87). A similar effect was observed for Lake Ontario samples; decreasing for DOC (0.87 to
0.69) and UVA (0.72 to 0.54), increasing for SUVA (0.03 to 0.47). However, with Lake Ontario
the correlation with SUVA was very poor in both cases and the observed change should not be
considered significant. Correlations did not change significantly for Otonabee River and Lake
Simcoe. Correlations for the Ottawa River sample most notably increased through the addition
of PC 1 and PC 2 scores. Correlations with all three parameters are increased; DOC: 0.87 to 0.97,
UVA: 0.89 to 0.98, and SUVA 0.90 to 0.98. It is hypothesized that the increase observed for the
Ottawa River results is due to pronounced variability in the protein-like score when compared to
the other waters (Figure 4.7). While there appears to be advantages to using a combined PC
score (protein-like and humic-like) for NOM concentration estimation, it is greatly dependent on
the source water.
Table 4.4: Linear relationship strength (R2) between FEEM-PCA and traditional NOM
indicators
NOM
indicator
All samples Otonabee
River
Lake Simcoe Lake
Ontario
Ottawa
River
Linear relationship strength (R2) with PC 1 (humic-like)
DOC (mg/L) 0.90 0.99 0.95 0.87 0.87
UVA (cm-1
) 0.90 0.99 0.99 0.72 0.89
SUVA
(cm-1
/mg/L) 0.79 0.96 0.94 0.03 0.90
Linear relationship strength (R2) with PC 1 + PC 2 (humic-like + protein-like)
DOC (mg/L) 0.87 0.99 0.94 0.69 0.97
UVA (cm-1
) 0.66 0.98 0.97 0.54 0.98
SUVA
(cm-1
/mg/L) 0.87 0.95 0.93 0.05 0.98
4.3.4 Disinfection by-product formation prediction
Results from FEEM-PCA analyses were compared to traditional NOM indicators for the
purpose of predicting the formation of disinfection by-products (DBPs). Corresponding with the
observed decrease in NOM concentration (Figure 4.1), THM and HAA concentrations decreased
with increasing coagulant dose for all waters (Figure 4.8 and Figure 4.9). Total concentrations of
Page 64
49
THMs varied between 9.2 and 112 µg/L and total HAAs between 11.7 and 128 µg/L for all
waters and chlorine doses. As with organic content, the reduction in DBP concentrations with
increasing coagulant dose was more pronounced for the two river sources. Results from other
studies have shown similar results of decreasing DBP concentrations with increasing NOM
removal through coagulation (Teixeira et al., 2011; Sadiq and Rodriguez, 2004).
For THMs, only TCM and BDCM were observed above the method detection limits
(MDL) reported in section 3.2.9. For HAAs, only DCAA and TCAA were found above the
MDL (section 3.2.10). Speciation graphs for THMs are shown in the appendix as Figure 8.1 and
Figure 8.2. For both chlorine doses, in river source waters BDCM represented approximately
20-30% of total THMs, while in lake sources it represented 35-40% of the total. HAA speciation
graphs are shown in as Figure 8.3 and Figure 8.4. For river sources, TCAA represented
approximately 40-50% of the total HAAs formed, while it represented 30-40% at all chlorine
doses for lake sources. Others have presented similar speciation and total THM and HAA
concentration ranges for waters with low bromide ion concentrations (< 0.01 mg/L) and similar
organic content (2 – 7 mg/L TOC) (William et al., 1996; Ates et al., 2007). Results also
compared well to results reported in previous disinfection by-product studies performed on
Otonabee River water (Wassink, 2011).
Water quality parameters typically included in empirical models for the prediction of
disinfection by-product formation consist of reaction time, pH, water temperature, chlorine dose,
bromide ion concentration, and an indicator of NOM concentration (Chowdhury et al., 2007;
Sadiq and Rodriguez, 2004). In these experiments, reaction time, chlorination pH, temperature,
and dose were controlled to be equal for all waters and samples. Bromide ion concentrations
were not explicitly controlled, although the only brominated DBP observed above its MDL was
BDCM. Furthermore, under the assumption that bromide concentration would not be affected by
coagulation, source specific models would not require the inclusion of bromide. In future work,
it is suggested that bromide ion is measured for the purposes of creating a more accurate model
which can account for more than one source water.
Page 65
50
Figure 4.8: Total THMs concentration vs. alum dose.
Figure 4.9: Total HAA9 concentration vs. alum dose
0
20
40
60
80
100
120
0 20 40 60 80
Tota
l TH
M c
on
cen
trat
ion
(µ
g/L
)
Alum dose (mg/L)
Lake Simcoe Otonabee River Lake Ontario Ottawa River
0
20
40
60
80
100
120
140
0 20 40 60 80
Tota
l HA
A c
on
cen
trat
ion
(µ
g/L
)
Alum dose (mg/L)
Otonabee River Lake Ontario Ottawa River
Note: Error bars represent standard deviation of duplicates
Note: Error bars represent standard deviation of duplicates
0
20
40
60
80
100
120
0 20 40 60 80
0
20
40
60
80
100
120
140
0 20 40 60 80
Chlorine dose: 2.5 mg/L Chlorine dose: 3.5 mg/L
Chlorine dose: 2.5 mg/L Chlorine dose: 3.5 mg/L
Page 66
51
4.3.4.1 Non-linear models
Based on DBP models reported in literature, a non-linear power based model was
proposed to fit the data compiled in this work (Sadiq and Rodriguez, 2004). Since only the
organic concentration was not controlled at two chlorine doses, the model was simplified to only
these variables (equation 4.2).
where, c1 through c4 are constants determined via regression.
Using MATLAB, the coefficient of correlation (R2) for the non-linear model as well as
the mean absolute error (MAE) was determined. These two parameters were used to elucidate
differences in model strength and to compare various NOM indicators. MAE was used in place
of the more commonly calculated mean square error (MSE) since it is less affected by outliers in
the data. MSE’s tendency to skew results heavily due to outliers results from the term being
squared, when compared to MAE (Bermejo and Cabestany, 2001). The MAE was calculated by
summing the absolute value of residuals divided by the number of samples (equation 4.3).
∑ | |
⁄
where, fi is the predicted value corresponding with the measured value yi and n is the
number of samples.
Example plots of measured values versus predicted values for the regressions of all
samples (XA) are shown in Figure 4.10. Results of the non-linear models are reported in Table
4.5 and Table 4.6 for THMs and HAAs, respectively. Results from the non-linear models
generally showed very high correlations (R2 > 0.8), especially for river water sources. Lower
correlations for lake waters were likely associated with their lower organic content and smaller
range of DBP formation potential. The relatively low DBP concentrations formed in lake waters
likely increased the relative influence of noise and therefore reduced correlation strength.
Visually it can be seen that using DOC as the predictor (or SUVA which incorporates DOC), the
nature of Lake Ontario (low organic content) water was not captured by the model and results
fall well outside the 95% confidence intervals (Figure 4.10).
(4.2)
(4.3)
Page 67
52
Figure 4.10: Measured values vs. predicted values of THMs from non-linear regressions of
source-specific models
0
20
40
60
80
100
120
0 20 40 60 80 100 120
0
20
40
60
80
100
120
0 20 40 60 80 100 120
0
20
40
60
80
100
120
0 20 40 60 80 100 120
Lake Simcoe Otonabee River Lake Ontario Ottawa River
(a) DOC (b)
Note: Solid line represents line of best fit, dashed lines represent 95%
confidence intervals. n = 70.
0
20
40
60
80
100
120
0 20 40 60 80 100 120
0
20
40
60
80
100
120
0 20 40 60 80 100 120
Pre
dic
ted
To
tal T
HM
s (µ
g/L)
Pre
dic
ted
To
tal T
HM
s (µ
g/L)
Pre
dic
ted
To
tal T
HM
s (µ
g/L)
Pre
dic
ted
To
tal T
HM
s (µ
g/L)
Pre
dic
ted
To
tal T
HM
s (µ
g/L)
Measured Total THMs (µg/L)
Measured Total THMs (µg/L) Measured Total THMs (µg/L)
Measured Total THMs (µg/L)
Measured Total THMs (µg/L)
(b) UVA
(c) SUVA (d) PC1
(e) PC1 + PC2
Page 68
53
Table 4.5: Results of non-linear regression for various NOM indicators and total THM concentrations
NOM Indicator All samples Otonabee River Lake Simcoe Lake Ontario Ottawa River
R2 MAE R
2 MAE R
2 MAE R
2 MAE R
2 MAE
DOC (mg/L) 0.88 6.40 0.94 2.36 0.75 4.10 0.90 0.36 0.96 3.32
UVA (cm-1
) 0.94 3.85 0.94 2.62 0.75 3.99 0.81 0.56 0.96 3.42
SUVA (cm-1
/mg/L) 0.88 5.61 0.92 3.14 0.75 4.00 0.75 0.80 0.94 3.60
PC 1 0.82 7.36 0.95 2.42 0.75 4.20 0.77 0.66 0.94 4.00
PC 1 + PC 2 0.76 9.58 0.94 2.49 0.77 4.38 0.75 0.86 0.87 6.39
Table 4.6: Results of non-linear regression for various NOM indicators and total HAA concentrations
NOM Indicator All samples Otonabee River Lake Ontario Ottawa River
R2 MAE R
2 MAE R
2 MAE R
2 MAE
DOC (mg/L) 0.88 11.77 0.99 1.85 0.94 0.62 0.92 12.58
UVA (cm-1
) 0.94 7.05 0.98 1.95 0.81 1.24 0.92 12.64
SUVA (cm-1
/mg/L) 0.89 12.18 0.97 3.05 0.78 1.94 0.91 13.21
PC 1 0.78 17.54 0.99 2.27 0.77 1.60 0.94 11.07
PC 1 + PC 2 0.75 23.43 0.98 2.42 0.81 2.02 0.80 23.04
Page 69
54
The MAE for each water source and DBP type was used to rank and compare the strength
of each NOM indicator as a predictor for DBP formation. The order of the indicators from
strongest to weakest was DOC > UVA > PC 1 Score > SUVA > PC 1 + PC 2 score. It should be
noted that, especially for river sources, the differences of both MAE and R2 values are very small
when comparing organic indicators (less than 0.09 and 3.07, respectively). In these cases it may
be concluded that all indicators performed to an equivalent degree, except for the PC 1 + PC 2
score. Poor correlations and model strength when using the PC 1 + PC 2 (humic-like + protein-
like) combined scores suggests that the protein fraction had no influence on DBP formation.
This follows from trends reported in literature, which identifies humic-like substances as being
the principal component involved in DBP formation (Barrett et al., 2000).
For the model comprising all samples (XA), results showed that UVA better predicted the
formation of DBPs analyzed in this study than the other NOM indicators. For these same models
SUVA and DOC performed similarly followed by PC 1 score.
4.4 Conclusions
The purpose of this work was to investigate the application of principal component
analysis of fluorescence excitation-emission matrices as a predictor for disinfection by-products
resulting from chlorination. The FEEM-PCA technique was compared to traditional organic
concentration indicators: DOC, UVA, and SUVA. FEEM-PCA was a strong indicator of organic
content and had good correlation with both THM and HAA formation. Consistent with other
NOM indicators the FEEM-PCA technique was less successful at modelling DBP formation
from lake water sources when compared to river sources. Furthermore, the results of modelling
using pooled data from all water sources showed the FEEM-PCA technique performed worse
than other traditional indicators, implicating that this technique is best applied to singular water
sources. Furthermore, a reduction in correlation strength when protein-like results were included
supports the assertion that protein content does not strongly influence DBP formation.
Page 70
55
5 Contributions of spatial, temporal, and treatment impacts on natural organic matter
character using principal component analysis of fluorescence spectra
5.1 Introduction
The monitoring of Natural Organic Matter (NOM) has proven to be of great importance
in drinking water treatment. NOM has been shown to be directly related to disinfection by-
product formation (Sadiq and Rodriguez, 2004), negatively affects water treatment processes
including ultrafiltration (Peiris et al., 2009), and promotes biological growth in distribution
systems (Baghoth et al., 2011). In both surface and ground waters, NOM consists of a wide
array of compounds including humic acids, carbohydrates, and proteins. This diversity of
compounds within the definition of NOM is problematic when considering removal techniques
(Pifer and Fairey, 2012) since there is evidence that both functional groups and molecular weight
distributions of compounds are integral to determining treatability (Her et al., 2003).
Unfortunately, traditional NOM quantification techniques, such as total organic carbon (TOC),
provide only gross measurements and limited information regarding NOM character (Gone et al.,
2009). Two analysis techniques which have been shown to provide information on multiple
NOM fractions include liquid chromatography-organic carbon detection (LC-OCD) (Huber et
al., 2011) and multivariate analysis of fluorescence excitation-emission matrices (FEEM)
(Persson and Wedborg, 2001; Zepp et al., 2004; Peiris et al., 2010; Bieroza et al, 2011).
LC-OCD allows simultaneous quantification of several fractions of organic matter. It
consists of a DOC analyzer preceded by chromatographic separation on the basis of molecule
size and into hydrophilic and hydrophobic fractions (Huber and Frimmel, 1992). Hydrophilic
NOM is divided into five subcategories: biopolymers, humic substances, building blocks of
humic substances, low molecular weight (LMW) acids and LMW neutrals.
The use of fluorescence measurements has been gaining traction as a promising method
for determining information on organic matter structure and function in water (Bieroza et al.,
2011). This method recognizes that several components of NOM fractions have unique
fluorescence signatures and therefore can be identified in a fluorescence spectrum. The
fluorescence spectrum is a representation of 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 a specific compound and the maximum
intensity of the peak correlated with concentration (Bieroza et al., 2011). This method, referred
Page 71
56
to as ‘peak picking,’ is limited in that it neglects to capture the heterogeneity of NOM fractions
(Peiris et al., 2010). The fluorescence intensity at specific excitation/emission wavelength pairs
is influenced by several variables including concentration, quantum yield, and absorptivity.
Molecular structure has a noted effect on some of these factors, which can be common among
several different molecules. By ‘peak picking’ the nature of overlapping structural properties
between distinct compounds and many of the potentially relevant compounds are neglected
(Persson and Wedborg, 2001). When compared to other NOM characterization techniques, such
as LC-OCD, the multivariate analysis of FEEMs is much faster and lower cost, while still
providing consistent results with high sensitivity (Peiris et al., 2008).
To further extract information and, in particular, changes in NOM composition between
several samples, multivariate analysis techniques have been used. Common techniques applied
include two-way principal component analysis (PCA) and multi-way parallel factor analysis
(PARAFAC) (Peiris et al., 2010). These multivariate techniques reduce the number of variables
needed to explain variance between samples in a set (Persson and Wedborg, 2001). Furthermore,
this pattern recognition and data set simplification approach, unlike ‘peak picking’, incorporates
the entire fluorescence spectrum (Bieroza et al., 2011).
The purpose of this study was to investigate the spatial, temporal, and treatment impacts
on NOM character at four City of Toronto water treatment plants (F.J. Horgan, Island, R.C.
Harris, and R.L. Clark). Furthermore, for the secondary purpose of comparing and identifying
applications of NOM characterization techniques, both LC-OCD and PCA of FEEM were used.
All four of the treatment facilities draw from the same source, Lake Ontario, Canada, but differ
in treatment processes applied. In total the four plants serve a population of greater than 3.5
million people. Raw water from the four water intakes was analyzed to identify if NOM
character varied temporally and spatially. With the intention of informing treatment optimization
work, NOM character of the treated water was also monitored to quantify the performance of
different treatments (e.g., coagulants, filtration media).
5.2 Method overview
5.2.1 Water treatment plant characteristics
A summary of treatment processes used and water sources at each of the four plants is
shown in Table 5.1 and Table 5.2. Intake locations and depths are also illustrated in Figure 5.1.
Process flow diagrams of each plant are presented in section 8.3.
Page 72
57
Table 5.1: Intake descriptions (CTC Source Protection Committee, 2012)
Treatment Facility Intake location Intake depth Intake length
F.J. Horgan 1a 18 m (59 ft.) 2925 (9596 ft.)
R.C. Harris Northeast 15 (49 ft.) 2232 (7323 ft.)
Southwest 15 (49 ft.) 2125 (6972 ft.)
Island East 83 (272 ft.) 4848 (15905 ft.)
Middle 83 (272 ft.) 4662 (15295 ft.)
West 83 (272 ft.) 4696 (15407 ft.)
R.L. Clark 1a 11 (36 ft.) 1610 (5282 ft.)
a only one intake.
Table 5.2: Summary of treatment processes
Treatment facility Coagulation/Flocculation Coagulant Filter Media
F.J. Horgan Inline PACl in winter:
1.2 - 2 mg/L
Alum in summer
(above 10o
C): 4 -
5 mg/L in addition
to 0.4-0.5 mg/L
polyelectrolyte
1.5 - 2.2 m GAC
above 0.25 m sand
R.C. Harris Hydraulic flocculation,
sedimentation Alum: 3 - 8.25
mg/L
0.25 - 0.30 m GAC
or anthracite over
0.25 m sand Island Inline PACl: 1.8 mg/L 0.91 m anthracite
over 0.25 m sand R.L. Clark Mechanical flocculation,
sedimentation Alum: 5 - 8 mg/L 0.46 m anthracite
over 0.31 m sand
5.2.2 Total organic carbon
Organic carbon measurements were based on the wet oxidation method as described in
Standard Method 5310 D (APHA, 2005). Samples were acidified at the time of collection to pH
2 using sulphuric acid (VWR International, Mississauga, ON) for preservation. Concentrations
of total organic carbon were obtained through heated persulfate oxidation using an O.I.
Analytical Aurora 1030 organic carbon analyzer (College Station, Texas, USA).
Page 73
58
Figure 5.1: Toronto WTP intake locations.
.
N
Note: Intake locations interpreted from Source Water Protection Plan (CTC Source Protection Committee, 2012), intake
lines interpreted based on treatment facility location. Map prepared in Google Earth V 7.0.2.8415 (Image © 2013
TerraMetrics, © 2012 Google, Image NOAA, Image © 2013 DigitalGlobe).
R.L. Clark intake (11 m depth)
Island intakes (83 m depth)
R.C. Harris intakes (15 m depth)
F.J. Horgan intake (18 m depth)
9.06 km
Page 74
59
5.2.3 Liquid chromatography – organic carbon detection
LC-OCD analyses were conducted using the method described by Huber et al. (2010).
Samples were passed through a 0.45 µm filter to remove any particulates prior to analyses. A
weak cation exchange column (250 mm x 20mm, 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. The samples first passed
through a UVA detector (254 nm) followed by an organic carbon detector (OCD). Prior to
entering the OCD, the solution was acidified to convert carbonates to carbonic acid. During the
acidification process, bound nitrogen (organic and inorganic) was converted to nitrate which
absorbs light in the UV region (Huber et al., 2011). This allows for the simultaneous
quantification of bound nitrogen in each separated fraction using the UVA detector. 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).
5.2.4 Fluorescence spectra collection
Fluorescence excitation-emission matrices were collected using a Perkin Elmer
Luminescence Spectrometer LS50B (Waltham, Massachusetts, USA) with FLWinLab Version
3.0. Collection of intensity values at 10 nm increments within excitation-emission ranges of
250-380 nm 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. It has
been shown that, while the PMMA cuvettes used 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., 2008). To account scattering interference, blank subtraction was
performed using spectra collected of Milli-Q© water using the same instrument settings (Her et
al., 2003).
5.2.4.1 Fluorescence data analysis
PCA is a multivariate data analysis technique that approximates a large data matrix
through observed patterns. Common goals for PCA include identification of similarities between
Page 75
60
samples, data simplification, and detection of outliers (Abdi and Williams, 2010).
Approximation of dominant data patterns is achieved by obtaining new mutually independent
variables that are mathematically represented by linear combinations of the original variables
(Persson and Wedborg, 2001). In generalized terms, the original data matrix X is decomposed
into the product of two small matrices, ti and pi which are referred to as the scores and loading
matrix, respectively (equation 5.1).
∑
The residuals generated from the approximation are represented in another matrix, E. For
multiple samples, n, in the matrix X, the product and residuals are summed (Peiris et al., 2010).
The loading matrix, pi, is a projection matrix that relates the PCA space to the original spectral
space. Information in the loading matrix can be used to identify the spectral areas representing a
specific principal component. The scores matrix, ti, represents projections of the samples, or
object, into the principal component (PC) space. Scores are analogous to concentrations and
represent the relative importance of a PC for a given sample (Persson and Wedborg, 2001).
Each sample produced a total of 4214 fluorescence intensity values at unique excitation-
emission wavelength pairs. Prior to PCA, intensity data of each sample was arranged by row.
All data manipulation and organization of the spectrum was carried out using in-house data
manipulation scripts written in Python 2.6 (Python Software Foundation). Using the PLS
Toolbox (Eigenvector Research Inc.) for MATLAB 7.12.0, principal component analysis (PCA)
was applied to the fluorescent excitation-emission matrices. During the study period a total of
174 unique FEEMs were collected from all four WTPs.
The model input data was pre-processed using the PLS toolbox. Autoscaling was applied
to the intensity values to reduce bias. Autoscaling centers the data on the mean and scales each
variable to a unit standard deviation. Without autoscaling, the PCA technique would favour
higher intensity values and skew the components identified to those with higher fluorescence
intensities, neglecting smaller peaks. For model validation, a cross-validation method of random
subsets (subset size of 10 samples) was selected.
(5.1)
Page 76
61
5.2.5 Statistical analysis
Tukey’s method allows simultaneous pairwise comparisons of multiple sample means
(Dean and Voss, 1999). The method calculates the minimum significance difference (MSD) for
each possible pair of means within the group. The MSD can be thought of as a confidence
interval for the difference between two sample means which, when spans 0, indicates no
significant difference between the sample sets. In generalized terms, assume Tukey’s method is
applied to identify significance between sample means in a group of four sample sets. For any
two sample sets within the group, i and j, the MSD can be expressed as follows:
(
√ )√ (
)
∑ ∑ ( )
where, µi and µj are the means of sample sets i and j, qv,n-1,α is the value of the studentized range
distribution for: v sample sets = 4, n observations, r replicates, α confidence level (95%) =
0.05, and y is a measured value.
5.3 Results and discussion
All four treatment facilities utilize Lake Ontario as their source water. However, their
intake locations and depths differ considerably. It was of interest to understand raw and treated
water NOM fraction differences among the four treatment plants. NOM was quantified using
total organic carbon (TOC) measurements and was further characterized using both FEEM-PCA
and LC-OCD techniques.
5.3.1 Analysis of raw water
TOC results at all four water treatment facilities varied between 1.85 and 2.55 mg/L
during the study period; and periods of increasing or decreasing trends in TOC levels were
generally experienced simultaneously by all four sources. Using ANOVA coupled with Tukey’s
method for identifying significant differences, it was found that the raw water TOC from the
Island intake was significantly lower than the other three treatment plants (95% confidence).
Furthermore, R.C. Harris raw water TOC was significantly higher than both R.L. Clark and the
(5.2)
Page 77
62
Island treatment plants. The maximum absolute difference (between the Island and R.C.
Harris) was 0.13 mg/L TOC and considered to be low.
Differences in character were first examined using LC-OCD. Results for raw water
samples are shown as Figure 5.2. Using one-way ANOVA, concentrations of all fractions were
found to be significantly equal, including overall DOC, at a 95% confidence level. The largest
NOM fraction was found to be humic substances followed by building blocks for all locations.
The relative distribution of NOM fractions is considered typical for lake sources (Huber et al.,
2011).
Figure 5.2: Raw water LC-OCD results
The fluorescence spectra for each of the four water sources exhibited similar
characteristics. Two main peaks, a and b, were identified from the spectral plots of fluorescence
intensities as shown in Figure 5.3 (example spectrums of raw water). Through comparison with
published work, peak a represents humic-acid type matter (Ex/Em: 270nm/450nm) whereas peak
b is consistent with fulvic-acid type material common to fresh waters (Ex/Em: 340nm/440nm)
(Peiris et al., 2010; Murphy et al., 2008). Peaks of high intensity were observed in both first and
second order Rayleigh scattering regions (Ex/Em: 300 – 380 nm / 300 – 380 nm and Ex/Em:
250-300 nm / 500 – 600 nm) which have been shown by to be related to sub-incident wavelength
sized particulate concentrations (Aslan et al., 2005; Peiris et al., 2010).
0.0
0.5
1.0
1.5
2.0
2.5
DOC Bio-Polymers HumicSubstances
Building Blocks LMW Neutrals LMW Acids
Co
nce
ntr
atio
n (
mg/
L)
F.J. Horgan Island R.C. Harris R.L. Clark
Note: Error bars represent one standard deviation (n = 6)
Page 78
63
Figure 5.3: Example raw water spectra from all four water treatment plants
250300
350
300
400500
6000
10
20
Inte
nsity (
a.u
.)
Excitation (nm)Emission (nm)
Inte
nsity (
a.u
.)
0
5
10
15
20
Excitation (nm)
Em
issio
n (
nm
)
250 300 350300
350
400
450
500
550
600
Inte
nsity (
a.u
.)
0
5
10
15
20
250300
350
300
400500
6000
10
20
Inte
nsity (
a.u
.)
Excitation (nm)Emission (nm)
Inte
nsity (
a.u
.)
0
5
10
15
20
Excitation (nm)
Em
issio
n (
nm
)
250 300 350300
350
400
450
500
550
600
Inte
nsity (
a.u
.)
0
5
10
15
20
250300
350
300
400500
6000
10
20
Inte
nsity (
a.u
.)
Excitation (nm)Emission (nm)
Inte
nsity (
a.u
.)
0
5
10
15
20
Excitation (nm)
Em
issio
n (
nm
)
250 300 350300
350
400
450
500
550
600
Inte
nsity (
a.u
.)
0
5
10
15
20
(a)
(b)
(c)
(d)
(a) F.J. Horgan plant, (b) Island plant, (c) R.C. Harris plant, (d) R.L. Clark plant.
Note: Intensity values truncated at 20 a.u. (arbitrary units)
a b
a
b
a b
a
b
a b
a
b
a b
a
b
Page 79
64
The excitation/emission location of peaks a and b (highest measured intensity) are shown
for each intake in Table 5.3. Peak locations were used to compare the similarity of organic
character among the four intakes. The most pronounced difference in spectrums was the lower
intensity of Rayleigh scattering peaks for the Island plant.
Table 5.3: Excitation/emission location of peaks for each water source
Peak Excitation/Emission of peak (nm/nm)
F.J. Horgan Island R.C. Harris R.L. Clark
a 280/443 280/443 280/436 280/442
b 340/433 340/428 340/433 340/430
Principal component analysis was applied to the full set of 174 samples from all four
plants; a cut off of 5% variance explained by each PC was applied. In total, the PCA model
explained 88.3% of the variance in the sample set when considering the first four PCs (Table
5.4).
Table 5.4: Variance explained by PCA model
Principal
Component Variance explained by PC (%) Cumulative variance explained (%)
1 64.62 64.62
2 13.42 78.04
3 5.50 83.54
4 4.80 88.34
The physical meaning of each principal component was determined through analysis of
loading plots (Figure 5.4). The loading plot for PC 1 displayed high values in areas (Ex/Em 260
– 380 / 350 – 550 nm) representing general humic-like substances as indicated in Figure 5.4 (a).
PC 2 loading values indicated that it predominantly represented scattering regions with high
loading values in both the first and second order Rayleigh scattering regions. The loading plot
for PC 3 showed a maximum at Ex/Em of 250/500 nm and a pronounced minimum at Ex/Em of
310/340 nm, neither of which were found to relate to a previously reported organic component
(Bagoth et al., 2011; Peiris et al., 2010). PC 4 represents protein-like substances as indicated by
high loading values within the low Ex/Em region (250-300/300-350 nm) (Stedmon and
Markager, 2004).
Page 80
65
Figure 5.4: Loading plots for first 3 principal component
(a)
(b)
(c)
(d)
(a) PC 1, (b) PC 2, (c) PC 3, (d) PC 4
Page 81
66
Average and MSD values for raw water FEEM-PCA scores (Figure 5.5) show the intake
water quality for each plant to be very similar in NOM character. Of note is the significantly
lower scattering score (PC 2) for the Island plant, evident from the non-overlapping MSD.
Furthermore, the MSD of each PC provides insight into the amount of variance during the
sampling time period. Scores varied more for humic-like (PC 1) than for the other identified
fractions (MSD PC1: 10.9 a.u., PC2: 7.2 a.u., PC3: 2.5 a.u.). In addition, the low MSD for the
protein-like fraction (PC 3) indicated low variability in protein-like material for all locations.
Figure 5.5: Averages and MSD for raw water FEEM-PCA scores
5.3.2 Analysis of treatment efficiency
The effectiveness of treatment 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. One-way
ANOVA was carried out on the calculated differences from all four plants to observe significant
differences in treatment efficiency. Statistical analysis did not include data collected beyond
June 19th
, 2012 due to process and equipment changes at the F.J. Horgan plant. At this time,
ozonation commenced and the GAC filter media was replaced. Effects of these changes were
observed from the pronounced increase in TOC removal observed in July (Figure 5.6). The
one-way ANOVA coupled with Tukey’s method yielded results indicating significant effects on
PC 1 (humic-like) PC 2 (scattering) PC 4 (protein-like)
-30
-20
-10
0
10
20
30
40
50
Sco
re
F.J. Horgan WTP Island WTP R.C. Harris WTP R.L. Clark WTP
Page 82
67
TOC due to treatment. The F.J. Horgan, R.C. Harris, and R.L. Clark plants reduced TOC by
significantly equal amounts (0.16 – 0.21 mg/L), while the Island plant had a significant but very
minor effect (0.04 mg/L TOC reduction) (95% confidence). The observed TOC reduction
(approx. 8 - 11%) at all plants except for the Island (2% reduction) was comparable to treatment
performances at plants using lake waters with low organic content (2 – 4 mg/L TOC) (Volk et
al., 2000; Uyak and Toroz, 2007). The 2% reduction in TOC at the Island plant was likely due to
a non-optimized coagulant dose.
Figure 5.6: Comparison of treatment performance for all plants: TOC
The results of statistical analysis on changes of NOM fractions due to treatment, as
analyzed by LC-OCD are shown in Table 5.5. Significant changes (95% confidence) to overall
DOC and humics were observed. At the R.L. Clark plant, bio-polymers were reduced; at the F.J
Horgan and R.L Clark plants building blocks were reduced. The Island plant was an exception
where no significant differences between raw and treated water were found. The results are
supported through evidence indication coagulation most significantly affects humic substances
(Chow et al., 2008; Baghoth et al., 2011) and biopolymers (Volk et al., 2000). Diemert et al.
(2012) reported similar trends of pronounced reduction of humic substances with minor changes
to biopolymer fractions from coagulation of lake and river waters using alum and PACl.
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Feb-12 Mar-12 Mar-12 Apr-12 May-12 May-12 Jun-12 Jul-12 Jul-12 Aug-12
Tota
l org
anic
car
bo
n d
iffe
ren
ce (
raw
- t
reat
ed
) (m
g/L)
Date (Month-Day)
F.J. Horgan WTP Island WTP R.C. Harris WTP R.L. Clark WTP
GAC media replaced and
pre-ozonation applied at
F.J. Horgan WTP
Page 83
68
Figure 5.7: Comparison of treatment performance for all plants: LC-OCD
Table 5.5: Significant changes in LC-OCD fractions between raw and treated water
Plant DOC Bio-
Polymers Humics
Building
Blocks
LMW
Neutrals
LMW
Acids
F.J. Horgan Yes No Yes Yes No No
Island No No No No No No
R.C. Harris Yes No Yes No No No
R.L. Clark Yes Yes Yes Yes No No Yes = significantly different (95% confidence); No = significantly similar (95% confidence)
Similar to TOC and LC-OCD results, the difference between raw and treated water scores
associated with FEEM-PCA data was calculated to observe changes to NOM fractions as a
function treatment. It is important to note that score values reported form PCA are not calibrated
to external standards. Score values, and changes to them, should be taken only in context of
score values from the same PCA model. The average and MSD of each PC for all treatment
plants is shown as Figure 5.8. PC 1 scores from the F.J. Horgan WTP showed that humic-like
substances were reduced throughout the entire study period. As was noted for TOC and LC-
OCD results, a marked increase in humic-like removal (approx. 79 a.u. to 196 a.u.) at the end of
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
Dif
fere
nce
in c
on
cen
trat
ion
(ra
w -
tre
ate
d)
(mg/
L)
F.J. Horgan WTP Island WTP R.C. Harris WTP R.L. Clark WTP
Note: Error bars represent one standard deviation (n = 6)
DOC Bio-Polymers Humic Substances
Buliding Blocks
LMW Neutrals LMW Acids
Page 84
69
June to early July occured when ozonation was applied and GAC filter media was replaced at the
F.J. Horgan WTP. All other treatment plants also showed significant reduction of PC 1 scores
due to treatment (interpreted by the MSD not overlapping 0 a.u.). Reductions of 11.4 a.u., 30.1
a.u., and 15.1 a.u. were observed for the Island, R.C. Harris, and R.L. Clark WTPs. In terms of
scattering (PC2), treatment at the F.J. Horgan WTP resulted in an increase following
implementation of process changes (-5.7 a.u. to 63 a.u.). Prior to this period no significant effect
in scattering due to treatment was observed. It is possible that ozonation potentially caused
breakdown of compounds resulting in increased concentration of small particulates which are
associated with scattering response as suggested by Peiris et al. (2010). Treatment at the Island
WTP produced no observable change to scattering. For both R.C. Harris WTP and R.L. Clark
WTP there was a significant decrease in particulate concentrations due to treatment. Throughout
the the study period protein-like scores (PC 4) were significantly reduced at the F.J. Horgan
WTP, while increased at the other three WTPs.
In order to make a full assessment on the apparent effects (TOC, LC-OCD, and FEEM-
PCA) of process changes at the F.J. Horgan plant, additional sampling is required. Trends
observed in the month of July indicated that treated water quality has not reached a steady state
and therefore further monitoring may help to determine the significance of apparent effects.
Figure 5.8: Average difference in PC score due to treatment
PC 1 (humic-like) PC 2 (scattering) PC 4 (protein-like)
-20
0
20
40
60
80
100
Sco
re d
iffe
ren
ce (
raw
- t
reat
ed
)
F.J. Horgan WTP Island WTP R.C. Harris WTP R.L. Clark WTP
Note: Error bars represent standard error (Tukey’s method) n = 24
Page 85
70
5.4 Conclusions
Analysis of raw water organic content (TOC, LC-OCD, and FEEM-PCA) at four Lake
Ontario water treatment plants by indicated strong similarities between the source waters from
different intake locations. Through FEEM-PCA data analysis it was observed that the Island
plant raw water had consistently lower scattering scores throughout the study period. Scattering
information from FEEM-PCA is known to be related to particulate concentrations (Peiris et al.,
2010), although more work is needed to determine what is being represented by scattering scores
within the context of water treatment. This intake is both deeper and farther from shore when
compared to the other plant intakes.
This work demonstrated the ability of FEEM-PCA as a rapid NOM characterization
technique for low organic content source waters. In comparison to LC-OCD and TOC, the
fluorescence method was also able to identify significant changes in overall
scattering/particulates. Furthermore, fluorescence results were well supported by both TOC and
LC-OCD.
Assessments of changes to the character of NOM via LC-OCD and FEEM-PCA showed
that primarily the humic fraction was influenced by treatment processes at all plants. The most
significant changes were observed to have resulted from process changes at the F.J Horgan plant.
Future work should include increased sample locations within the plant to determine whether the
ozonation process or replacement of GAC media caused the notable increase in NOM removal
effectiveness.
Page 86
71
6 Pilot-scale coagulation optimization including NOM characterization with principal
component analysis of fluorescence spectra
6.1 Introduction
Implementing changes to water treatment processes has an inherent high degree of
uncertainty. Understanding the intricacies of the impacts of a process change at full-scale is not
necessarily conducive from bench scale tests where variables are often well controlled, water
volumes are vastly different, and there is difficulty in observing seasonal results caused by
changing source water quality. Pilot-scale studies have been shown as a means to address issues
associated with scale and reduce errors in process optimization (Knowles et al., 2012).
In order to take full advantage of pilot-scale studies, an appropriate experimental design
must be established that is able to identify when apparent process improvements are in fact, real.
Furthermore, the experimental design should show the reliability of scale-up from pilot to a full-
scale plant. At the core of pilot-based studies is the need to run two identical parallel trains
where one pilot train serves as a control, which may be constantly compared to the full-scale
performance, while the second acts as an experimental train. Fundamental to parallel pilot
studies is the validation of similarity between both pilot trains and the full-scale plant.
Validation involves establishing, with statistical significance, that with equal treatment applied,
both pilot trains and the full-scale train produce identical water quality at any given point in the
process. This allows for the conclusion that observed water quality differences in the
experimental train are due only to the controlled process changes. Furthermore, validation
permits for the conclusion that changes observed at pilot-scale will translate well if applied in a
full-scale scenario (Anderson et al., 1993).
Pilot-scale studies were conducted with the objective to identify impacts on water quality
resulting from use of aluminum sulphate (alum) or polyaluminum chloride (PACl) as coagulants.
The motivation for testing alternative coagulants was the need to limit sludge production
resulting from the coagulation process at the City of Peterborough, Ontario, Canada, water
treatment plant. Pilot-scale studies were aimed at determining the potential reduction of sludge
as well as to quantify changes to water quality and provide costs associated with implementing
PACl. As part of the water quality analysis, principal component analysis (PCA) of fluorescence
excitation-emission matrices (FEEM) was conducted to determine how NOM fractions were
affected by treatment processes and coagulant type.
Page 87
72
6.2 Methods
The source water for both full and pilot-scale trains is the Otonabee River, Ontario,
Canada. A summary of typical source quality (March 2011 to May 2012) is shown in Table 6.1
Table 6.1: Peterborough raw water quality yearly ranges
Parameter Range (Yearly)
Temperature (oC) 0.0 – 27.7
Total organic carbon (TOC) (mg/L) 5.81 – 7.33
Ultraviolet absorbance (UVA) (cm-1
) 0.140 – 0.260
pH 7.43 – 8.62
Turbidity (NTU) 0.27 – 1.47
6.2.1 Pilot plant
One of the pilot-scale treatment trains was dosed with alum and operated to mimic the
current operation of the full-scale plant to ensure that pilot-scale tests accurately reflect full-scale
conditions. The second pilot treatment train was dosed with PACl such that differences in
treated water quality could be observed. Figure 6.1 shows a simplified process flow diagram,
which identifies major unit processes common to both full-scale and pilot-scale facilities.
Both treatment trains were outfitted with a wide array of instrumentation to monitor water
quality in real time with online data collection and tabulation completed through a sophisticated
supervisory control and data acquisition (SCADA) system. Measurements included pH,
Flocculation Sedimentation Dual Media Filtration
(anthracite and sand)
Chlorine Contact
Raw Water
Coagulant,
Flash Mix
Chlorine
(disinfection)
To Distribution
Chlorine
(ensure residual)
Sodium Silicate
(pH control)
Fluoride
Figure 6.1: Simplified process flow diagram of pilot and full-scale treatment train
Page 88
73
temperature, total organic carbon (TOC), ultraviolet absorbance (UVA), turbidity, and particle
counts for raw water, settled water, and filtered water on all treatment trains. In addition, weekly
grab samples were collected at the pilot and full-scale plant including TOC, pH and FEEM for
raw water, settled water, and filtered water. Samples for THM and HAAs were collected post-
filtration for 4 days per week and chlorinated for 24 hours prior to being quenched.
6.2.2 Fluorescence excitation-emission matrices
FEEMs were collected using a Perkin Elmer Luminescence Spectrometer LS50B
(Waltham, Massachusetts, USA) with FLWinLab Version 3.0. No pre-treatment of the samples
was applied, however all had been adjusted to a pH of 7. Maintaining a common pH between
samples ensured that fluorescence characteristics of the acidic functional groups in humic
molecules remained constant (Mobed et al., 1996). Collection of intensity values at 10 nm
increments within excitation-emission ranges of 250-380 nm 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 (Peiris et al., 2010; 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. It has been shown that, while the
PMMA cuvettes used reduce the intensity of excitation wavelengths below 285 nm, this
approach is appropriate for purposes of distinguishing NOM elements using fluorescence (Peiris
et al., 2008). To account for background noise and scattering interference, spectra for Milli-Q®
water was obtained using the same instrument settings. Intensity values from the Milli-Q®
samples were subtracted from intensity values of sample spectra to reduce background noise
effects.
Scattering/particulate regions of the FEEM were removed in an effort to increase model
sensitivity to humic and protein-like substances (Bahram et al., 2006). Rayleigh scattering is
characterized by emitted photons at equal or double the wavelength of excitation photons,
referred to as second order Rayleigh scattering (FORS) and second order Rayleigh scattering
(SORS), respectively (Mobed et al., 1996). A region surrounding Rayleigh scattering was
removed to fully exclude any influence from the spectra (Zepp et al., 2004). In this study, points
satisfied by equations 6.1 and 6.2 were removed from the spectrum (intensities set to 0).
Approximate FORS and SORS regions omitted are visualized in an example spectrum (Figure
6.2).
Page 89
74
Figure 6.2: Example FEEM illustrating scattering regions (FORS and SORS)
6.2.3 Principal component analysis
Using the PLS Toolbox (Eigenvector Research Inc., Wenatchee, WA) for MATLAB,
principal component analysis (PCA) was applied to the fluorescent excitation-emission matrices.
Two models were constructed using different sample sets in order to increase sensitivity. Model
1 was created using all samples: raw water, settled water, and filtered water. Model 2 was
created using only treated water samples (settled and filtered water).
All data was pre-processed using the PLS toolbox prior to model creation. Autoscaling
was applied to the intensity values to reduce bias. Autoscaling centers the data on the mean and
scales each variable to a unit standard deviation. Without autoscaling, the PCA technique would
favour higher intensity values and skew the components to those with higher fluorescence
intensities, neglecting smaller peaks. For model validation, a cross-validation method of random
subsets was used.
6.2.4 Statistical assessment
Statistical assessments were made to determine similarity of full and pilot train treated
water as well as the treated water between the two pilots (alum vs. PACl). A difficulty
SORS
FORS
(6.1)
(6.2)
Page 90
75
encountered with the analysis was the cyclic seasonal nature of water quality data collected for
over a year. Analysis of yearly averages would have incorrectly ignored natural fluctuations in
raw and treated water qualities. Therefore, the influence of natural changes in water such as
temperature and pH on the results was removed through the calculation of differences rather than
absolute values. With this approach the null hypothesis was a difference between of 0 would
indicate similarity between the two trains. This approach is reliant on the assumption that
external environmental conditions affect all treatment trains equally. This extends to imply that
the effect of source water quality on finished water quality is independent of the coagulant type.
Two sets of statistical tests were performed, one utilizing the baseline data to observe any
differences in the three treatment trains (one full-scale, two pilot-scale) operated at equal
conditions, and the second to compare changes in treated water quality between the two pilots to
illustrate differences associated with a coagulant change. During baseline testing both pilot
trains received alum doses equal to the full-scale plant. It was expected that baseline data should
demonstrate significantly similar water quality between the two pilot and full-scale trains. It was
possible, however, that differences in scale and other physical differences between the trains
would contribute to small, consistent offsets in an absolute value. These cases were identified
when the baseline sample standard deviation was 10% or less of the mean. This implied that the
offset was consistent and the error was assumed to be carried through into the experimental data.
As such, this baseline ‘offset’ mean was taken as the new assumed mean difference during the
experimental phase of pilot testing.
6.2.5 Tukey’s method
Tukey’s method allows simultaneous pairwise comparisons of multiple sample means
(Dean and Voss, 1999). The method calculates the minimum significance difference (MSD) for
each possible pair of means within the group. The MSD can be thought of as a confidence
interval for the difference between two sample means which, when spans 0, indicates no
significant difference between the sample sets. In generalized terms, assume Tukey’s method is
applied to identify significance between sample means in a group of four sample sets. For any
two sample sets within the group, i and j, the MSD can be expressed as follows:
(6.3)
Page 91
76
(
√ )√ (
)
∑ ∑ ( )
where, µi and µj are the means of sample sets i and j, qv,n-1,α is the value of the studentized range
distribution for: v sample sets = 4, n observations, r replicates, α confidence level (95%) =
0.05, and y is a measured value.
Further details regarding statistical methods can be found in the appendix (section 8.2.3).
6.3 Results and discussion
Statistical results presented in this work are classified as either during the ‘baseline,’ or
‘experimental’ period. During baseline testing both pilot trains received alum doses equal to the
full-scale plant. The baseline period was used to establish the validity of pilot trials through
analyzing the similarity in water quality between the pilot trains and to the full-scale plant. In
the experimental period, PACl was used in one of the two pilot trains while the other remained
on alum. In order to directly compare performance of the two coagulants, dosages were selected
which resulted in equal filtered water TOC, determined through periodic jar testing (once per
month). Daily averages of TOC and UVA from April 2011 to May 2012 are shown in Figure
6.3.
Using paired t-tests, filtered water UVA and TOC for the alum and PACl pilot were
found to be significantly equal during baseline testing (95% confidence). While the difference in
UVA was found insignificant between pilots during the experimental phase, difference in TOC
was significant (0.03 mg/L) but, less than one standard deviation of quality control standards (0.1
mg/L) (Section 3.2.1). During both testing periods a significant, but slight, difference in UVA
between the alum pilot and full-scale was identified (≤ 0.01 a.u.). Comparison of alum pilot and
full-scale TOC demonstrated significant differences of 0.14 and 0.05 mg/L for baseline and
experimental periods, respectively.
Due to the large sample sizes (baseline n ~ 25; experimental n ~ 374), small differences
in absolute values between trains were found to be significant. Therefore, significance was also
assessed on the basis of the absolute difference in terms of ultimate impact on the process. For
Page 92
77
example, differences of ≤ 0.1 pH units (as seen during baseline testing) were not considered to
have a significant impact on water quality.
Figure 6.3: TOC and UVA of pilot plant settled water
The results of statistical tests performed on all monitored parameters are presented in the
appendix, section 8.1.3. Water treated with PACl exhibited a pH difference of 0.72 ± 0.12 in
settled water. The addition of alum depressed the pH of the water while the pre-hydrolyzed
PACl did not exhibit the same effect. Furthermore, it was shown that in comparison to PACl, the
alum treated settled water turbidity by 0.23 ± 0.23 NTU. Following filtration, no substantial
difference was observed between pilots (0.1 ± 0.1 NTU). A significant increase in total THM
concentrations in PACl treated water was observed (~20% increase), while there was no
significant change to HAA9 concentrations. Studies by others suggest that the elevated pH of the
PACl treated water caused the observed increase in THM formation (Singer, 1994; Sadiq and
Rodriguez, 2004). To test this hypothesis, a series of jar tests with both alum and PACl,
followed by chlorination at a controlled pH, was completed. The results of these tests showed
that, when alum or PACl treated water was chlorinated at an equal pH (7 and 8), total THMs
formed following 24 hours of contact time with free chlorine were equal. Additional details
showing the methodology and results from the pH controlled DBP formation tests are presented
in section 8.4.
0.04
0.06
0.08
0.10
0.12
0.14
0.16
2.30
2.50
2.70
2.90
3.10
3.30
3.50
3.70
3.90
4.10
4.30
UV
Ab
sorb
ance
25
4 n
m
TO
C (
mg/
L)
Month - YY
Alum Pilot Filtered Water TOC PACl Pilot Filtered Water TOC
Alum Pilot Filtered Water UVA PACl Pilot Filtered Water UVA
Page 93
78
Both the alum and PACl pilot filters were backwashed simultaneously and therefore had
equal filter run times. The monthly average total loss of head from a full filter run is presented
as Figure 6.4. Consistently, it was observed that alum pilot filters had greater loss of head when
compared to the PACl pilot. Average differences for individual months were statistically
confirmed using Tukey’s method at 95% confidence. It was thought this was a result from with
the increased turbidity observed in alum settled water. This implies that for equal TOC removal,
PACl provided more efficient particle removal.
Figure 6.4 Average monthly total loss of head for pilot plant filter
Overall, comparison of the “control” pilot and the full-scale plant showed that statistically
they are not considered equal, however, in most cases, the absolute differences were small. A
complete table indicating statistical results and the absolute difference for all measured
parameters is presented as Table 8.15 in the appendix (section 8.1.3).
6.4 Performance evaluation using fluorescence spectra
6.4.1 Spectral features of Peterborough water
Intensities from the acquired FEEMs showed two prominent peaks at approximately
Excitation/Emission (Ex/Em) = 340 nm/430 nm (peak α) and Ex/Em = 280 nm/430 nm (peak β).
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0.018
0.020
Loss
of
He
ad (
m·h
-1)
Month - YY
Alum Pilot Filtered Water PACl Pilot Filtered Water
Note: Error bars represent one standard deviation
Page 94
79
The literature indicate that fluorophores at these excitation/emission pairs correspond with
fulvic-like humic substances (peak α) (Coble et al., 1990) and humic acid type material (peak β)
(Sierra et al., 2005; Peiris et al., 2010). Several pronounced peaks were also observed between
Ex 320 and 380 nm and Em of 300 and 400 nm and are considered to be First Order Raleigh
Scattering (FORS) resulting from particulate and colloidal matter in the water (Peiris et al.,
2010) (Figure 6.5 and Figure 6.6).
Figure 6.5: Raw water FEEM – 2D view
Figure 6.6: Raw water FEEM - 3D view
Peak β,
humic
acid
Peak ,
fulvic acid
Peak β,
humic acid
Peak ,
fulvic acid
Inte
nsi
ty (
a.u
.)
Inte
nsi
ty (
a.u
.)
Page 95
80
6.4.2 Results of principal component analysis and model selection
Principal component analysis was completed using a set of 77 samples which included
raw, settled, and filtered water from the full-scale and pilot trains. In the context of this study
two types of comparisons were to be made; changes to PC scores between raw and treated water
as well as differences of PC scores between treated water samples. It was hypothesized that by
limiting the organic concentration range for model inputs, sensitivity of the output would be
increased for that same range.
Model 1 included all collected samples in order to assess changes between raw and
treated water.
Model 2 included only treated water samples (settled and filtered water) to examine
differences between alum or PACl treated water.
A summary of the variance explained by each PC for Models 1 and 2 is shown in Table
6.2. Cumulative variance explained by the first two components of Models 1 and Model 2 were
91.27% and 84.60%, respectively. Model 1 results show that more variance was explained by
PC 1 (83.13%), when compared to Model 2 (70.88%); PC 2 in Model 2 explained more variance
than in Model 1 (13.72% vs. 8.13%).
Table 6.2: Variance (%) explained by PCs for Models 1 and 2
Model # Samples Used for
Model Creation
Principal
Component #
Percent Variance
Explained by
This PC (%)
Cumulative
Percent Variance
Explained (%)
1 All Samples 1 83.13 83.13
2 8.13 91.27
2
Treated Water
Samples (Settled
and Filtered Water)
1 70.88 70.88
2 13.72 84.60
High loading values in specific excitation/emission regions indicate that the PC is
representative of the spectral data in the same excitation/emission region. For instance, it has
been previously established that fluorescence intensity at 340 nm/430 nm (excitation/emission)
indicates fulvic acid (Murphy et al., 2008). High loading values in this region would indicate
Page 96
81
that the PC is also representative of fulvic acid. For PCs 1 to 3, Figure 6.7 to Figure 6.10 show
loading values plotted with their corresponding excitation/emission coordinates. Loading plots
for each Model were included to demonstrate the similarity of loading values between Models 1
and 2.
There was a high degree of commonality of PC loading values for each model. PC 1 in
all models had pronounced loading peaks in regions associated with humic-like material. The
humic-like region of 270 nm excitation / 360 nm emission to 380 nm excitation / 600 nm
emission was determined through reference to literature (Peiris et al., 2010; Her et al., 2003;
Coble et al., 1990). For Model 2, there appeared to be some division of peaks within the large
humic zone which was not present in Model 1. Two localized peaks are visible at approximately
320 nm excitation / 500 nm emission and 360 nm excitation / 430 nm emission (Figure 6.9).
From both models, PC 2 results exhibited loading peaks in the 260-300 nm excitation / 300-400
nm emission region. This region is known to be associated with protein-like substances (Peiris et
al., 2010; Her et al., 2003). In addition to the protein-like region, both models exhibited
shoulder peaks which continued into emission values up to 475 – 500 nm. It is unknown at this
time what this region represents.
6.4.3 Performance of coagulation on NOM fraction removal
Changes to humic-like and protein-like material identifiable by FEEM-PCA were tracked
by calculating the difference between treated and raw water scores (raw – treated). Scores are
reported in arbritrary units (a.u.) and it should be noted that the score values are only significant
in relation to other scores calculated by the same model. Positive differences indicate a decrease
of scores and illustrate the effectiveness of treatment (Figure 6.11). Scores for PC 1 differences
showed a significant effect of coagulation on humic-like material (difference of 139 – 152 a.u.).
This degree of humic-like removal was also shown to be significantly equal between the full-
scale and two pilot trains based on their MDLs. For PC 2 scores, both the full-scale plant and the
alum pilot showed increased protein-like scores (-20.3 a.u. and -10.7 a.u., respectively) while the
PACl pilot showed a slight decrease (11.3 a.u.), due to treatment. These findings are consistent
with the literature that indicates humic-like material is effectively removed through coagulation
and that proteins are not significantly affected (Gone et al., 2009).
Page 97
82
Figure 6.7: Raw and treated water (Model 1), PC 1 loading plot
Figure 6.8: Raw and treated water (Model 1), PC 2 loading plot
Figure 6.9: Treated water (Model 2), PC 1 loading plot
Page 98
83
Figure 6.10: Treated water (Model 2), PC 2 loading plot
Figure 6.11: Impact of coagulation on PC scores. Averages and MDL (bars) for each treatment
train (n = 9 – 10).
6.4.4 Performance of filtration on NOM fraction removal
FEEM-PCA was also used to track changes to NOM fraction removal associated with
filtration (Figure 6.12). For both alum treatment trains (full and pilot-scale) there was a general
trend of a minor reduction in humic-like score (15.6 a.u. and 11.4 a.u.). The PACl pilot did not
show any significant decrease of humic-like score due to filtration. Trends for PC 2, the protein-
PC 1 PC 2
-40
-20
0
20
40
60
80
100
120
140
160
180
Dif
fere
nce
in P
C S
core
(ra
w -
tre
ate
d)
Full-scale Alum Pilot PACl Pilot
Page 99
84
like fraction were similar to PC 1. The full-scale plant and alum pilot filtration was found to
significantly decrease PC 2 score (25.1 a.u. and 11.3 a.u., respectively) while no significant
decrease was observed from the PACl pilot. These results were not unexpected as dual media
filtration (anthracite and sand) is not intended to remove the dissolved organic fractions.
Figure 6.12: Impact of filtration on PC scores. Averages and MDL (bars) for each treatment
train (n = 9 – 10).
6.4.5 Performance of FEEM-PCA scores related to common NOM indicators
FEEM-PCA was directly compared to total organic carbon (TOC), ultraviolet absorbance
(UVA), and specific ultraviolet absorbance (SUVA). Based on raw water and filtered water
samples (due to availability of online TOC and UVA data for these sampling points) PC 1 and
PC 2 scores were compared both independently as well as summed together as Total PC
(equation 6.4).
-10
0
10
20
30
40
Dif
fere
nce
in P
C S
core
(raw
- t
reat
ed
)
Full-scale Alum Pilot PACl Pilot
(6.4)
PC 1 PC 2
Page 100
85
The strength of the linear relationship between FEEM-PCA score and the traditional
NOM indicators was assessed using the coefficient of determination (R2). Two iterations of
correlations for Model 1 (containing both raw and treated water data) were performed, one that
included both raw and treated water samples and the other with only results for the treated water
subset. Linear correlations show that FEEM-PCA scores were most correlated with TOC (Table
6.3). The strongest linear correlations were observed between PC 1 and TOC using Model 1 and
including all samples (R2
= 0.94, 0.83). Including both raw and treated water samples increases
the linear correlation strength for each PC or TPC (e.g. PC 1 R2 = 0.52 to 0.94 when including
raw water samples). This is likely due to the wider TOC range (3 – 7 mg/L) when compared to
the range that included only treated water samples (3 – 4 mg/L TOC). A lack of correlation was
observed with any traditional NOM indicator and PC 2 (all R2 < 0.27). Furthermore, inclusion of
PC 2 scores (TPC) resulted in a loss of correlation.
Table 6.3: Strength of linear correlation between FEEM-PCA and common NOM indicators
Model Samples
Used PC #
Correlation
with TOC
(R2)
Correlation
with UVA
(R2)
Correlation
with SUVA
(R2)
Model 1
(all
samples)
Raw and
Treated
Water
1(humic-like
material) 0.94 0.82 0.14
2 (protein-like
material) 0.01 0.00 0.02
TPC 0.83 0.75 0.16
Model 1
(all
samples)
Treated
Water
Samples
1 (humic-like
material) 0.52 0.10 0.32
2 (protein-like
material) 0.04 0.01 0.04
TPC 0.13 0.02 0.06
Model 2
(treated
samples)
Treated
Water
Samples
1 (humic-like
material) 0.61 0.14 0.40
2 (protein-like
material) 0.24 0.12 0.27
TPC 0.29 0.04 0.15
TPC = Total PC Score
The results also show that Model 2 (containing only treated water data) was better
correlated with TOC, UVA, and SUVA, than Model 1 when using results for the same treated
water samples (R2 = 0.61 vs 0.52, 0.14 vs 0.1, 0.40 vs 0.32). The motivation behind creating two
models was the hypothesis that model sensitivity would be increased through limiting the organic
Page 101
86
concentration range of model inputs to be comparable to the samples of interest. The results
support the hypothesis and point to the necessity of selective model regression for increasing the
capabilities of FEEM-PCA.
In general, the identified high correlation strength of PC 1 with TOC is promising for
future work when using FEEM-PCA as an indicator in lieu of traditional measurement
techniques. To assess the utility of FEEM-PCA, an endpoint should be predicted with FEEM-
PCA as opposed to traditional NOM indicators. While correlation with TOC is useful, the
increased prediction strength of endpoints such as disinfection by-product formation could help
solidify the FEEM-PCA technique as a promising new NOM indicator.
6.5 Conclusions
Characterization of NOM changes due to coagulation showed that humic-like substances
were reduced. No effect on protein-like material was observed. The character of NOM in settled
water was equal between both alum and PACl treated water indicating no difference in removal
efficiencies for specific fractions. Filtration was shown only to have a minor and non-consistent
effect on humic-like substances, although it was not expected to reduce DOC. Results from
FEEM-PCA correlated well with other NOM measures including TOC, UVA, and SUVA. The
importance of sample selection for model regression was demonstrated through increased
sensitivity and correlations.
Page 102
87
7 Summary, Conclusions and Recommendations
7.1 Summary
The application of FEEM-PCA was investigated for DBP formation prediction from
chlorination, assessment of temporal and spatial differences in raw water, and assessment of
treatment process efficiencies. In all applications, the FEEM-PCA technique was compared to
commonly used NOM surrogates including DOC, UVA, and SUVA as well as using LC-OCD.
Water from the Ottawa and Otonabee Rivers, as well as Lake Ontario and Simcoe were
coagulated and chlorinated over a range of dosages. Results indicated that FEEM-PCA
correlated well with DOC and both measures provided excellent predictive strength in DBP
modelling. UVA and SUVA were also shown to be strong surrogates in this application.
Spatially and temporal changes to NOM character was investigated in Lake Ontario using
TOC, FEEM-PCA, and LC-OCD. In a five month study, the raw and treated water at four City
of Toronto water treatment plants was monitored. Results indicated that Lake Ontario NOM did
not vary significantly among the four treatment plants as well as over the study period. Using
FEEM-PCA, it was noted that the Island WTP had decreased scattering or particulate
composition in the raw water; an observation not possible when using other NOM
characterization techniques. Only small differences in treatment efficiency were observed,
especially at the Island plant which showed no impact on overall NOM quantity or character asa
function of treatment. The application of ozonation and replacement of GAC filter media at the
F.J. Horgan WTP was shown to significantly increasing the removal of humics, while at the
same time increasing overall protein and particulate content.
Pilot studies at the City of Peterborough WTP examined water quality differences
resulting from the use of two different coagulants; PACl and alum. FEEM-PCA was applied to
identify differences in NOM following individual treatment steps. Both coagulants were shown
to have equal impacts on NOM removal, with no differences in specific fraction removal
efficiencies.
7.2 Conclusions
The conclusions of the research can be summarized as follows:
1. FEEM-PCA is a viable NOM surrogate for DBP formation modelling. It was
strongly correlated with DOC and provided similar predictive strength during
modelling.
Page 103
88
2. Raw water was not found to vary spatially between the four City of Toronto water
treatment plants. Due to the low organic content of Lake Ontario (~2.0 mg/L TOC),
treatment provided little to no NOM removal.
3. FEEM-PCA was able to identify several notable differences between City of Toronto
raw and treated waters not observed by other NOM surrogates or characterization
methods.
a. Island WTP raw water was found to consistently have lower
scattering/particulates.
b. Application of ozonation and GAC was found to increase both scattering and
protein-like content in treated water at the F.J. Horgan WTP.
4. Alum and PACl showed equal efficiencies at removal of NOM fractions at pilot-scale
at the Peterborough WTP.
5. In comparison to alum, PACl caused no depression of pH prior to chlorination which
lead to ~20% increased THM formation.
6. In comparison to alum, application of PACl resulted in increased particulate removal
efficiency which translated into significantly lower loss of head build-up on media
filters.
7.3 Recommendations
Based on this work, the following recommendations are made for future study:
1. Directly compare NOM fractions identified through LC-OCD and FEEM-PCA.
While some LC-OCD and FEEM-PCA data was collected as part of this work, it was
limited to only Lake Ontario. Additional sampling from multiple water sources is
needed to obtain adequate concentration variations.
2. Further investigate multivariate model optimization through scattering region removal
or smoothing techniques.
3. Additional should be conducted following individual unit processes at the City of
Toronto treatment plants to better quantify NOM removal efficiencies.
4. Assess the ability of FEEM-PCA to identify low levels of wastewater influence in
drinking water sources.
Page 104
89
REFERENCES
Amy, G., Chadik, P.A., Chowdhury, Z.K., 1987. Developing models for predicting
trihalomethane formation potential and kinetics. Journal of the American Water Works
Association 79, 89-97.
Amy, G., Siddiqui, M., Ozekin, K., Zhu, H.W., Wang, C., 1998. Empirically based models for
predicting chlorination and ozonation by-products: haloacetic acids, chloral hydrate, and
bromate. EPA Report CX 819579.
Anderson, W.B., Douglas, I.P., Van Den Oever, J., Jasim, S.Y., Fraser, J.C., and Huck, P.M.
1993. Experimental Techniques for Pilot Plant Evaluation. Proceedings, AWWA Water
Quality Technology Conference, Miami, Florida, Part I, 343–364.
Aslan, K., Lakowicz, J.R., Geddes, C.D., 2005. Plasmon light scattering in biology and
medicine: new sensing approaches, visions and perspectives. Current Opinion in Chemical
Biology 9, 538-544.
Ates, N., Sule Kaplan, S., Sahinkaya, E., Kitis, M., Dilek, F.B., Yetis, U., 2007. Occurrence of
disinfection by-products in low DOC surface waters in Turkey. Journal of Hazardous
Materials 142, 526-534.
Boorman, G.A., Dellarco, V., Dunnick, J.K., Chapin, R.E., Hunter, S., Hauchman, F., Gardner
H., Cox, M., Sills, R.C., 1999. Drinking water Disinfection Byproducts: Review and
Approach to Toxicity Evaluation. Environmental Health Perspectives 107, 207-217
Baghoth, S.A., Sharma, S.K., Amy, G.L., 2011. Tracking natural organic matter (NOM) in a
drinking water treatment plant using fluorescence excitation-emission matrices and
PARAFAC. Water Research 45, 797-809.
Bahram, M., Bro, R., Stedmon, C., Afkhami, A., 2006. Handling of Rayleigh and Raman scatter
for PARAFAC modeling of fluorescence data using interpolation. Journal of Chemometrics
20, 99-105.
Page 105
90
Baker, A., Inverarity, R., Charlton, M., Richmond, S. 2003. Detecting river pollution using
fluorescence spectrophotometry: case studies from Ouseburn, NE England. Environmental
Pollution 124, 57-70.
Barrett, S.E., Krasner, S.W., Amy, G.L., 2000. ‘Natural Organic Matter and Disinfection By-
Products: Characterization and Control in Drinking Water – An Overview’. In Natural
Organic Matter and Disinfection By-Products (S.E. Barrett, S.W. Krasner, and G.L. Amy,
eds). Oxford University Press.
Beggs, K.M.H, Summer, R.S., McKnight, D.M., 2009. Characterizing chlorine oxidation of
dissolved organic matter and disinfection by-product formation with fluorescence
spectroscopy and parallel factor analysis. Journal of Geophysical Research 114, 10
Bermejo, S., and Cabestany, J., 2001. Oriented principal component analysis for large margin
classifiers. Neural Networks 14, 1447-1461.
Bieroza, M., Baker, A., Bridgeman, J., 2010. Classification and calibration of organic matter
fluorescence data with multiway analysis methods and artificial neural networks: an
operational tool for improved drinking water treatment. Environmentrics 22, 256-270
Chen, B., and Westerhoff, P., 2010. Predicting disinfection by-product formation potential in
water. Water Research 44, 3755-3762.
Chen, C., Zhang, X., Zhu, L., Liu, J., He, W., Han, H., 2008. Disinfection by-products and their
precursors in a water treatment plant in North China: seasonal changes and fraction analysis.
Science of the Total Environment 397, 140-147.
Chen, W.J., and Weisel, C.P. 1998. Halogenated DBP concentrations in a distribution system.
Journal of the American Water Works Association 90, 151-163.
Childress, A.E., Vrijenhoek, E.M., Elimelech, M., Tanaka, T.S., Beuhler, M.D., 1999.
Particulate and THM precursor removal with ferric chloride. Journal of Environmental
Engineering 125, 1054-1061.
Page 106
91
Chowdhurry, S., Champagne, P., McLellan, P.J., 2009. Models for predicting disinfection
byproduct (DBP) formation in drinking waters: A chronological review. Science of the Total
Environment 407, 4189-4206.
Coble, P.G., Green, S.A., Blough, N.V., Gagosian, R.B., 1990. Characterization of dissolved
organic matter in the Black Sea by fluorescence spectroscopy. Nature 348 (6300), 432-435
Coble. P.G., 1996. Characterization of marine and terrestrial DOM in sweater using excitation-
emission matrix spectroscopy. Marine Chemistry 51, 325-346.
Credit Valley, Toronto and Region, and Central Lake Ontario (CTC) Source Protection
Committee. 2012. CTC Region Source Water Protection Plan.
Dean, A. And Voss, D., 1999. Design and Analysis of Experiments. Springer-Verlag New York,
Inc., New York.
Diemert, S., (2012) The Impact of Coagulation on Endocrine Disrupting Compounds,
Pharmaceutically Active Copmpounds and Natural Organic Matter. (Master’s thesis).
University of Toronto, Ontario, CA.
Edzwald, J.K. and Tobiason, J.E., 1999. Enhanced coagulation: US requirements and a broader
view. Water Science and Technology 40, 63-70.
Gallard, H.H., and U. von Gunten, U., 2002. Chlorination of natural organic matter: kinetics of
chlorination and of THM formation. Water Research 36, 65-74.
Gone, D.L., Seidel, J., Batiot, C., Bamory, K., Ligban, R., Biemi, J., 2009. Using fluorescence
spectroscopy EEM to evaluate the efficiency of organic matter removal during coagulation-
flocculation of a tropical surface water (Agbo reservoir). Journal of Hazardous Materials 172,
693-699
Graves, C.G., Matanoski, G.M., Tardiff, R.G., 2001. Weight of Evidence for an Association
between Adverse Reproductive and Developmental Effects and Exposure to Disinfection By-
products: A Critical Review. Regulatory Toxicology and Pharmacology 34, 103-124.
Page 107
92
Her, N., Amy, G., McKnight, D., Sohn, J., Yoon, Y., 2003. Characterization of DOM as a
function of MW by fluorescence EEM and HPLC-SEC using UVA, DOC, and fluorescence
detection. Water Research 37, 4295-4303.
Hong, H.C., Liang, Y., Han, B.P., Mazumder, A., Wong, M.H., 2007. Modeling of
trihalomethane (THM) formation via chlorination of the water from Dongjiang River (source
water for Hong Kong’s drinking water). Science of the Total Environment 385, 48-45
Hrudey, S.E., 2009. Chlorination disinfection by-products, public health risk tradeoffs and me.
Water Research 43, 2057-2092
Hua, G., and Reckhow, D.A., 2007. Characterization of Disinfection Byproduct Precursors
Based on Hydrophobicity and Molecular Size. Environmental Science and Technology 41,
3309-3315.
Hua, G., and Reckhow, D.A., 2008. DBP formation during chlorination and chloramination:
Effect of reaction time. pH, dosage, and temperature. Journal of the American Water Works
Association 100, 82-95+12.
Huber, S.A., Balz, A., Abert, M., Pronk, W., 2011. Characterisation of aquatic humic and non-
humi matter with size-exclusion chromatography – organic carbon detection – organic
nitrogen detection (LC-OCD-OND). Water Research 45, 879-885.
Huber, S. and Frimmel, F.H., 1992. A liquid chromatographic system with multi-detection for
the direct analysis of hydrophilic organic compounds in natural waters. Fresenius Journal of
Analytical Chemistry 342, 198-200.
Hudson, N., Baker, A., Reynold, D., 2007. Fluorescence analysis of dissolved organic matter in
natural, waste and polluted waters – a review. River Research and Applications 23, 631-649.
Joliffe, I.T., 1986. Principal Component Analysis, Second Edition. Springer-Verlag New York,
New York.
Kitis, M., Karanfil, T., Kilduff, J.E., Wigton, A., 2001. The reactivity of natural organic matter
to disinfection by-products formation and its relation to specific ultraviolet absorbance.
Water Science and Technology 43, 9-16.
Page 108
93
Korshin, G.V., Li, CW., Benjamin, M.M., 1996. Monitoring the properties of natural organic
matter through UV spectroscopy: a consistent theory. Water Research 31, 1787-1795
Knowles, A.D., MacKay, J., Gagnon, G.A., 2012. Pairing a pilot plant to a direct filtration water
treatment plant. Canadian Journal of Civil Engineering 39, 689-700.
Krasner, S.W., Weinberg H.S., Richardson, S.D., Pastor, S.J., Chinn, R., Climenti, M.J., Onstad,
G.D., Thurston, A.D JR., 2006. Occurrence of a New Generation of Disinfection Byproducts.
Environmental Science and Technology 40, 7175-7185.
Liang, L. and Singer, P.C., 2003. Factors Influencing the Formation and Relative Distribution of
Haloacetic Acids and Trihalomethanes in Drinking Water. Environmental Science and
Technology 37, 2920-2928.
Matilainen, A., Lindqvist, N., Korhonen, S., Tuhkanen, T., 2002. Removal of NOM in the
different stages of the water treatment process. Environment International. 28, 457-465
Matilainen, A., Gjessing, E.T., Lahtinen, T., Hed, L., Bhatnagar, A., Sillanpää, M., 2011. An
overview of the methods used in the characterisation of natural organic matter (NOM) in
relation to drinking water treatment. Chemosphere 83, 1431-1442.
McBean, E., Zhu, Z., Zeng. W., 2008. Systems analysis models for disinfection by-prodcut
formation in chlorinated drinking water in Ontario. Civil Engineering and Environmental
Systems 25, 127-138.
McKnight, D.M., Boyer, E.W., Westerhoff, P.K., Doran, P.T., Kulbe, T., Andersen, D.T., 2001.
Spectrofluorometric characterization of dissolved organic matter for indication of precursor
organic material and aromaticity. Limnology and Oceanography 46, 38-48.
McQuarrie, J.P., and Carlson, K. 2003. Secondary benefits of aquifer storage and recovery:
Disinfection and by-product control. Journal of Environmental Engineering 129, 412-418
Mobed, J.J., Hemmingsen, S.L., Autry, J.I., McGown, L.B., 1996. Fluorescence
Characterization of IHSS Humic Substances: Total Luminescence Spectra with Absorbance
Correction. Environmental Science and Technology 30, 3061-3065.
Page 109
94
Montgomery, D.C., and Runger, G.C., 2003. Applied Statistics and Probability for Engineers 3rd
Edition. Wiley, New York.
Murphy, K.R., Stedmon, C.A., Waite, T.D., Ruiz, G.M., 2008. Distinguishing between terrestrial
and authochthonous organic matter sources in marine environments using fluorescence
spectroscopy. Marine Chemistry 108, 40-58.
Oxenford, J.L. 1997. ‘Disinfection By-Products: Current Practices and Future Directions’. In
Disinfection By-Products in Water Treatment (R.A. Minear and G.L. Amy, eds.), Boca Raton,
FL: CRC Press.
Peiris, B.R.H., Hallé, C., Haberkamp, J., Legge, R.L., Peldszus, S., Moresoli, C., Budman, H.,
Amy, G., Jekel, M., and Huck, P.M., 2008. Assessing nanofiltration fouling in drinking water
treatment using fluorescence fingerprinting and LC-OCD analyses. Water Science &
Technology: Water Supply 8 (4), 459-465.
Peiris, 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
Peiris, R.H., Hallé, C., Budman, H., Moresoli, C., Peldszus, S., Huck, P.M., Legge, R.L., 2010.
Identifying fouling events in a membrane-based drinking water treatment process using
principal component analysis of fluorescence excitation-emission matrices. Water Research
44, 185-194
Persson, T., Wedborg, M., 2001. Multivariate evaluation of the fluorescence of aquatic organic
matter. Analytica Chimica Acta 434, 179-192.
Pifer, A.D., and Fairey, J.L., 2012. Improving on SUVA254 using fluorescence-PARAFAC
analysis and asymmetric flow-field flow fractionation for assessing disinfection byproduct
formation and control. Water Research 46, 2927-2936.
Roch, T., 1997. Evaluation of total luminescence data with chemometrical models: a tool for
environmental monitoring. Analytica Chimica Acta 356, 61-74.
Page 110
95
Rodriguez, M.J. and Serodes, J., 1999. Assessing empirical linear and non-linear modelling of
residual chlorine in urban drinking water systems. Environmental Modelling and Software
14, 93-102.
Sadiq, R. And Rodriguez, M.J., 2004. Disinfection by-products (DBPs) in drinking water and
predictive models for their occurrence: a review. Science of the Total Environment 321, 21-
46.
Sérodes, JB., Rodriguez, M.J., Li, H., Bouchard, C., 2003. Occurrence of THMs and HAAs in
experimental chlorinated water of the Quebec City area (Canada). Chemosphere 51, 253-263
Sierra, M.M.D., Giovanela, M., Parlanti, E., Soriano-Sierra, E.J., 2005. Fluorescence fingerprint
of fulvic and humic acids from varied origins as viewed by single-scan and
excitation/emission matrix techniques. Chemosphere 58 (6), 715-733
Singer, P.C. 1994. Control of disinfection by-products in drinking water. Journal of
Environmental Engineering 120, 727-744.
Sobsy, M.D., 1989. Inactivation of Health-related Microorganisms in Water by Disinfection
Process. Water Science and Technology 21, 179-195.
Sohn, J., Amy, G., Jaeweon, C., Yonghun, L., Yeomin, Y., 2004. Disinfectant decay and
disinfection by-products formation model development: chlorination and ozonation by-
products. Water Research 38, 2461-2478.
Sohn, J., Amy, G., Yoon, Y., 2007. Process-train profiles of NOM through a drinking water
treatment plant. Journal of the American Water Works Association 99, 145-153.
Standard methods for the examination of water and wastewater (2005) APHA, AWWA&
WPCF, Washington, D.C.
Stedmon, C.S., Markager, S., Bro, R., 2003. Tracing dissolved organic matter in aquatic
environments using a new approach to fluorescence spectroscopy. Marine Chemistry 82, 239-
254
Page 111
96
Stramski, D. And Wozniak, S.B. 2005. On the role of colloidal particles in light scattering in the
ocean. Limnology and Oceanography 50, 1581-1591.
Teixeria, M.R., and Rosa-Vânia Sousa, S.M., 2011. Natural Organic Matter and Disinfection By-
products Formation Potential in Water Treatment. Water Resources Management 25, 3005 –
3015.
Thurman, E.M., 1985. Organic geochemistry of natural waters. Kluwer Academic, Hingham,
MA.
USEPA. (1990) Determination of Chlorination Disinfection Byproducts, Chlorinated Solvents,
and Halogentated Pesticide/Herbicides in Drinking Water by Liquid-Liquid Exctraction and
Gas Chromatography with Electron-Capture Detection. Method 551.1 revision 1.0.
USEPA, (1999) Microbial and disinfection by-product rules simultaneous compliance guidance
manual, EPA 815-R-99-015
USEPA. (2003) Determination of Haloacetic Acids and Dalapon in Drinking Water by Liquid-
Liquid Microextraction, Derivatization, and Gas-Chromatography with Electron Capture
Detection. Method 552.3 revision 1.0. EPA 815-B-03-002
Uyak, V., Toroz, I., Meric, S., 2005. Monitoring and modeling of trihalomethanes (THMs) for a
water treatment plant in Istanbul. Desalination 176, 127 – 141
Uyak, V. and Toroz, I., 2007. Disinfection by-product precursors reduction by various
coagulation techniques in Istanbul water supplies. Journal of Hazardous Materials 141, 320-
328.
Volk, C., Bell, K., Ibrahim, E., Verges, D., Amy, G., Lechevallier, M., 2000. Impact of
enhanced and optimized coagulation on removal of organic matter and its biodegradable
fraction in drinking water. Water Research 34, 3247-3257
Wang, G., Deng, Y, Lin, T., 2007. Cancer risk assessment from trihalomethanes in drinking
water. Science of the Total Environment 387, 86-95.
Page 112
97
Wassink, J. (2011). Coagulation optimization to minimize and predict the formation of
disinfection by-products. (Master’s thesis). University of Toronto, Ontario, CA.
Weishaar, J.L., Aiken, G.R., Bergamaschi, B.A., Fram, M.S., Fujii, R., Mopper, K., 2003.
Evaluation of specific ultraviolet absorbance as an indicator of the chemical composition and
reactivity of dissolved organic carbon. Environmental Science and Technology 37, 4702-
4708.
Williams, D.T., Lebel, G.L., Benoit, F.M., 1996. Disinfection by-products in Canadian drinking
water. Chemosphere 34, 299-316.
Wold, S., Esbensen, K., Geladi, P., 1987. Principal component analysis. Chemometrics and
Intelligent Laboratory Systems 2, 37-52.
Yange, X., Shang, C., Lee, W., Westerhoff, P., Fan, C. 2008. Correlations between organic
matter properties and DBP formation during chloramination. Water Research 42, 2329-2339.
Zepp, R.G., Sheldon, W.M., Moran, M.A., 2004. Dissolved organic fluorophores in southeastern
US coastal waters: correction method for eliminating Rayleigh and Raman scattering peaks in
excitation-emission matrices. Marine Chemistry 89, 15-36.
Page 113
98
8 Appendices
8.1 Raw Data
8.1.1 Disinfection by-product modelling (Section 4)
Page 114
99
Table 8.1: PC 1 Scores
Lake Simcoe Otonabee River Lake Ontario Ottawa River
Coagulant Dose (mg/L) Average STD Average STD Average STD Average STD
0 -1.96 0.98 134.31 9.16 -70.27 2.61 157.61 2.35
5 -1.30 0.39 NR NR -69.41 1.08 154.69 0.50
10 -7.09 0.15 130.59 9.47 -76.93 0.09 151.48 1.31
20 -22.66 0.22 101.81 1.99 -89.96 0.40 76.86 0.87
30 -34.49 0.26 79.63 2.03 -90.06 6.84 -20.51 0.73
40 -39.63 0.66 57.79 3.40 -88.78 0.87 -43.64 0.29
50 -46.45 0.30 32.32 4.50 -92.97 0.07 -46.42 1.83
60 -49.54 0.64 16.24 1.26 -92.25 1.13 -13.80 2.26
70 -53.26 0.32 8.92 3.38 -89.39 2.09 38.50 7.67
NR = Not Recorded
STD = Standard deviation
Table 8.2: PC 2 Scores
Lake Simcoe Otonabee River Lake Ontario Ottawa River
Coagulant Dose (mg/L) Average STD Average STD Average STD Average STD
0 31.49 0.17 53.36 7.23 -10.61 0.74 -40.55 1.92
5 8.58 0.51 NR NR -21.81 0.34 -56.11 0.08
10 9.70 0.65 51.82 12.01 -16.57 0.11 -75.12 1.23
20 16.35 1.29 47.43 3.72 -7.53 0.24 -59.77 0.37
30 15.66 0.44 44.33 4.01 -2.07 8.12 -6.53 0.97
40 14.57 0.30 44.03 3.89 -7.04 1.90 -9.78 1.29
50 12.85 0.58 39.02 5.13 -10.12 1.52 -13.57 0.32
60 9.67 0.51 33.68 2.54 -9.77 1.56 -43.25 0.28
70 7.80 1.13 30.92 3.96 -12.04 2.89 -69.01 2.60
NR = Not Recorded
STD = Standard deviation
Page 115
100
Table 8.3: TOC
Lake Simcoe Otonabee River Lake Ontario Ottawa River
Coagulant Dose (mg/L) Average STD Average STD Average STD Average STD
0 3.87 0.02 5.59 0.00 2.47 0.01 5.85 0.02
5 3.75 0.01 NR NR 2.31 0.00 5.76 0.11
10 3.61 0.01 5.33 0.05 2.19 0.00 5.32 0.01
20 3.53 0.12 4.90 0.03 1.97 0.00 4.11 0.02
30 3.09 0.04 4.55 0.02 1.66 0.23 3.09 0.05
40 2.93 0.04 4.12 0.00 1.68 0.02 2.74 0.04
50 2.69 0.00 3.74 0.10 1.60 0.02 2.18 0.01
60 2.54 0.00 3.37 0.02 1.53 0.01 2.17 0.00
70 2.31 0.00 3.15 0.00 1.54 0.02 2.34 0.01 NR = Not Recorded
STD = Standard deviation
Table 8.4: UVA
Lake Simcoe Otonabee River Lake Ontario Ottawa River
Coagulant Dose (mg/L) Average STD Average STD Average STD Average STD
0 0.065 NA 0.128 0.000 0.017 0.000 0.200 0.000
5 0.061 NA NR NR 0.015 0.000 0.198 0.006
10 0.058 NA 0.115 0.000 0.013 0.000 0.173 0.001
20 0.051 NA 0.103 0.001 0.012 0.001 0.111 0.001
30 0.045 NA 0.092 0.001 0.012 0.001 0.068 0.000
40 0.042 NA 0.078 0.000 0.010 0.001 0.055 0.000
50 0.037 NA 0.067 0.001 0.010 0.000 0.045 0.000
60 0.034 NA 0.059 0.000 0.012 0.000 0.043 0.001
70 0.032 NA 0.056 0.003 0.013 0.000 0.048 0.001 NR = Not recorded
NA = Not applicable (duplicates not collected)
STD = Standard deviation
Page 116
101
Table 8.5: SUVA
Lake Simcoe Otonabee River Lake Ontario Ottawa River
Coagulant Dose (mg/L) Average STD Average STD Average STD Average STD
0 1.68 0.01 2.29 0.00 0.69 0.00 3.42 0.01
5 1.63 0.01 NR NR 0.65 0.00 3.44 0.16
10 1.61 0.00 2.16 0.02 0.59 0.00 3.25 0.03
20 1.45 0.05 2.10 0.04 0.58 0.04 2.70 0.04
30 1.46 0.02 2.01 0.03 0.69 0.14 2.20 0.04
40 1.44 0.02 1.90 0.00 0.57 0.05 2.00 0.03
50 1.38 0.00 1.78 0.06 0.63 0.01 2.06 0.01
60 1.34 0.00 1.75 0.01 0.78 0.01 1.96 0.04
70 1.39 0.00 1.77 0.09 0.85 0.01 2.03 0.04 NR = Not Recorded
STD = Standard deviation
Page 117
102
Table 8.6: THMs
Coagulant
Dose
(mg/L)
Lake Simcoe Otonabee River Lake Ontario Ottawa River
TCM BDCM TCM BDCM TCM BDCM TCM BDCM
avg std avg std avg std avg std avg std avg std avg std avg std
Chlorine dose = 2.5 mg/L
0 23.0 0.8 12.8 0.8 56.1 5.0 14.2 0.0 7.7 1.1 4.1 0.8 84.8 2.5 19.1 0.8
5 31.4 0.7 16.3 0.7 NR NR NR NR 7.1 0.1 4.0 0.2 74.6 2.5 20.2 0.8
10 30.6 0.1 16.6 0.2 49.3 9.1 13.6 0.4 7.5 0.4 4.2 0.2 73.2 2.5 15.7 0.8
20 19.5 2.5 12.0 0.8 58.3 11.5 15.7 0.9 7.1 0.1 4.1 0.0 56.6 4.2 16.0 1.0
30 21.1 2.3 12.7 1.9 56.7 5.7 14.3 1.1 6.8 0.2 4.0 0.3 40.9 14.1 11.7 2.5
40 25.5 0.6 14.6 0.3 37.4 7.7 11.1 0.9 5.5 0.3 3.8 0.5 32.6 2.4 12.2 0.9
50 19.2 4.9 11.9 2.3 34.0 7.0 11.4 1.0 5.4 1.2 3.9 0.5 33.5 1.6 12.7 0.5
60 15.5 0.4 10.5 0.1 35.1 0.2 11.3 0.7 5.1 0.3 3.8 0.5 31.2 5.2 12.9 2.9
70 20.5 0.3 12.4 0.1 24.8 1.7 9.1 0.4 5.4 0.0 4.0 0.0 35.9 2.5 13.8 1.3
Chlorine dose = 3.5 mg/L
0 30.0 5.0 16 1.2 65.9 10.7 17.2 1.2 8.5 1.2 5.0 0.6 71.2 2.5 23.3 2.5
5 27.0 5.3 15.1 2.2 NR NR NR 2.2 7.3 0.5 4.9 0.8 86.4 2.5 24.6 2.5
10 32.7 0.4 17.5 0.1 60.6 7.8 17.4 0.1 7.9 0.1 5.0 0.3 63.3 7.5 20.1 3.5
20 25.3 5.7 15.3 2.7 61.1 10.4 17.4 2.7 7.7 0.2 5.1 0.1 41.9 2.5 17.2 2.5
30 19.6 0.5 11.9 0.6 51.1 6.7 15 0.6 7.0 0.3 4.7 0.2 44.8 3.6 17.1 1.6
40 19.8 0.1 12.3 0.3 45.9 0.3 12.9 0.3 7.0 0.2 4.8 0.1 32.6 8.3 13.1 2.7
50 23.6 1.3 13.9 0.9 39.9 7.3 12.8 0.9 6.7 0.9 4.5 0.4 38.0 1.7 14.8 0.9
60 22.8 0.2 13.5 0.2 31.3 1.4 11.1 0.2 6.9 0.1 4.7 0.0 28.0 5.3 12.3 1.0
70 18.5 3.2 12.0 1.1 30.4 5.2 10.5 1.1 7.0 0.2 4.8 0.3 32.2 10.9 13.2 3.4 NR = Not Recorded
std = Standard deviation
avg = average
Page 118
103
Table 8.7: HAAs
Coagulant
Dose
(mg/L)
Lake Simcoe Otonabee River Lake Ontario Ottawa River
DCAA TCAA DCAA TCAA DCAA TCAA DCAA TCAA
avg std avg std avg std avg std avg std avg std avg std avg std
Chlorine dose = 2.5 mg/L
0 4.0 0.2 4.6 0.1 52.6 5.8 52.7 3.7 21.5 1.2 12.0 0.8 71.7 3.8 64.5 3.5
5 0.7 0.7 6.8 0.7 NR NR NR NR 20.0 0.9 10.3 0.3 73.6 2.4 59.2 2.1
10 4.4 0.3 6.4 0.3 54.8 0.7 48.8 0.7 18.7 0.7 10.0 0.0 77.5 1.0 69.8 1.0
20 0.9 0.7 6.8 0.7 49.3 1.0 45.9 0.7 18.4 0.0 9.5 0.2 68.9 0.4 65.9 0.8
30 0.7 0.1 5.0 0.1 45.7 1.1 42.0 1.0 17.2 0.7 7.3 0.0 41.0 0.9 33.9 1.2
40 3.5 0.7 4.3 0.7 38.1 4.7 38.2 3.3 17.2 0.5 7.8 0.2 29.8 1.0 18.7 0.4
50 4.3 0.2 4.9 0.0 35.4 1.3 29.5 0.4 17.7 0.7 7.4 0.7 28.9 0.9 17.9 0.4
60 4.3 0.7 3.5 0.7 32.6 0.8 26.2 0.0 17.0 0.2 7.8 0.1 33.1 0.8 21.1 0.1
70 4.9 0.7 3.0 0.7 28.5 0.4 22.6 0.3 16.5 0.1 6.6 0.4 37.7 1.8 27.1 0.6
Chlorine dose = 3.5 mg/L
0 -0.3 0.5 4.6 0.0 59.3 3.0 66.3 4.0 21.9 1.1 13.6 1.2 106.9 3.9 111.6 4.8
5 -0.2 0.7 6.8 0.7 NR NR NR NR 21.9 0.4 12.7 0.2 111.0 1.4 111.4 0.3
10 1.9 2.2 5.9 0.4 58.3 0.1 62.3 0.8 20.9 0.6 11.4 0.3 103.0 2.8 112.9 2.9
20 5.3 0.7 5.7 0.7 56.1 0.4 59.2 1.6 20.9 0.1 10.5 0.1 89.2 0.7 99.7 0.7
30 2.5 2.5 4.7 0.1 49.4 2.7 48.8 2.7 20.7 0.5 9.8 0.2 45.4 2.1 36.5 1.2
40 0.0 0.7 4.7 0.7 42.6 3.5 41.6 2.0 19.0 0.4 8.7 0.3 32.2 0.8 21.8 0.6
50 4.3 0.1 4.6 0.0 38.1 1.8 33.9 1.2 19.3 0.4 9.7 0.1 31.4 0.7 21.8 0.1
60 4.5 0.7 3.3 0.7 34.5 1.4 30.5 1.2 19.4 1.0 9.4 0.6 36.3 0.5 25.3 0.8
70 4.4 0.7 3.3 0.7 32.2 0.8 28.4 0.5 19.3 0.3 8.9 0.2 41.1 1.8 31.8 0.5 NR = Not Recorded
std = Standard deviation
avg = average
Page 119
104
Figure 8.1: THM speciation graphs, chlorine = 2.5 mg/L
0.0
10.0
20.0
30.0
40.0
50.0
60.0
70.0
80.0
0 5 10 20 30 40 50 60 70
Co
nce
ntr
atio
n (
µg/
L)
Coagulant Dose (mg/L) BDCM TCM
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
0 5 10 20 30 40 50 60 70
0.0
10.0
20.0
30.0
40.0
50.0
60.0
0 5 10 20 30 40 50 60 70
0.0
10.0
20.0
30.0
40.0
50.0
60.0
70.0
80.0
90.0
0 5 10 20 30 40 50 60 70
Otonabee River Lake Simcoe
Lake Ontario Ottawa River
Page 120
105
Figure 8.2: THM speciation graphs, chlorine = 3.5 mg/L
0.0
10.0
20.0
30.0
40.0
50.0
60.0
70.0
80.0
90.0
0 5 10 20 30 40 50 60 70
Co
nce
ntr
atio
n (
µg/
L)
Coagulant Dose (mg/L) BDCM TCM
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
16.0
0 5 10 20 30 40 50 60 70
0.0
10.0
20.0
30.0
40.0
50.0
60.0
0 5 10 20 30 40 50 60 70
0.0
10.0
20.0
30.0
40.0
50.0
60.0
70.0
80.0
90.0
0 5 10 20 30 40 50 60 70
Otonabee River Lake Simcoe
Lake Ontario Ottawa River
Page 121
106
Figure 8.3: HAA speciation graphs, chlorine = 2.5 mg/L
0.0
20.0
40.0
60.0
80.0
100.0
120.0
0 5 10 20 30 40 50 60 70
Co
nce
ntr
atio
n (
µg/
L)
Coagulant Dose (mg/L) TCAA DCAA
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
0 5 10 20 30 40 50 60 70
0.0
20.0
40.0
60.0
80.0
100.0
120.0
140.0
160.0
0 5 10 20 30 40 50 60 70
Otonabee River Ottawa River
Lake Ontario
Page 122
107
Figure 8.4: HAA speciation graphs, chlorine = 3.5 mg/L
0.0
20.0
40.0
60.0
80.0
100.0
120.0
140.0
0 5 10 20 30 40 50 60 70
Co
nce
ntr
atio
n (
µg/
L)
Coagulant Dose (mg/L) TCAA DCAA
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
0 5 10 20 30 40 50 60 70
0.0
50.0
100.0
150.0
200.0
250.0
0 5 10 20 30 40 50 60 70
Otonabee River Ottawa River
Lake Ontario
Page 123
108
8.1.2 Toronto NOM project data tables
Table 8.8: Raw water LC-OCD results
Raw Water (mg/L)
FJ Horgan
Date DOC
Bio-
Polymers
Humic
Substances
Building
Blocks
LMW
Neutrals
LMW
Acids
10-Apr 2.00 0.23 0.79 0.48 0.23 0.11
25-Apr 2.03 0.18 0.82 0.47 0.32 0.09
10-May 2.07 0.22 0.81 0.44 0.22 0.09
05/23/12 2.11 0.22 0.80 0.43 0.28 0.09
06/06/12 2.06 0.31 0.83 0.44 0.18 0.09
Island
Date DOC
Bio-
Polymers
Humic
Substances
Building
Blocks
LMW
Neutrals
LMW
Acids
10-Apr 2.00 0.21 0.80 0.45 0.30 0.13
25-Apr 2.00 0.17 0.77 0.43 0.25 0.10
10-May 2.03 0.23 0.79 0.41 0.30 0.10
05/23/12 1.82 0.18 0.72 0.43 0.21 0.10
06/06/12 1.94 0.22 0.75 0.42 0.20 0.09
RC Harris
Date DOC
Bio-
Polymers
Humic
Substances
Building
Blocks
LMW
Neutrals
LMW
Acids
10-Apr 1.82 0.18 0.64 0.48 0.24 0.13
25-Apr 1.96 0.16 0.64 0.53 0.28 0.10
10-May 1.95 0.20 0.71 0.42 0.30 0.10
05/23/12 1.89 0.20 0.74 0.38 0.24 0.10
06/06/12 2.01 0.24 0.67 0.45 0.25 0.11
RL Clark
Date DOC
Bio-
Polymers
Humic
Substances
Building
Blocks
LMW
Neutrals
LMW
Acids
10-Apr 1.75 0.16 0.71 0.42 0.25 0.11
25-Apr 1.96 0.15 0.69 0.46 0.22 0.11
10-May 1.85 0.17 0.72 0.40 0.27 0.10
05/23/12 1.84 0.17 0.73 0.31 0.25 0.10
06/06/12 1.93 0.23 0.69 0.42 0.29 0.10
Page 124
109
Table 8.9: Treated water LC-OCD results
Treated Water (mg/L)
FJ Horgan
Date DOC
Bio-
Polymers
Humic
Substances
Building
Blocks
LMW
Neutrals
LMW
Acids
10-Apr 1.92 0.22 0.77 0.44 0.21 0.11
25-Apr 1.84 0.15 0.72 0.41 0.24 0.10
10-May 2.01 0.19 0.75 0.42 0.23 0.11
05/23/12 NR NR NR NR NR NR
06/06/12 1.94 0.27 0.77 0.35 0.21 0.09
Island
Date DOC
Bio-
Polymers
Humic
Substances
Building
Blocks
LMW
Neutrals
LMW
Acids
10-Apr 1.82 0.21 0.77 0.51 0.23 0.12
25-Apr 2.10 0.17 0.80 0.44 0.22 0.09
10-May 2.06 0.23 0.82 0.42 0.25 0.08
23-May 1.82 0.17 0.76 0.40 0.26 0.08
06-June 2.07 0.22 0.79 0.42 0.24 0.08
RC Harris
Date DOC
Bio-
Polymers
Humic
Substances
Building
Blocks
LMW
Neutrals
LMW
Acids
10-Apr 2.10 0.21 0.75 0.54 0.22 0.14
25-Apr 2.00 0.20 0.81 0.46 0.25 0.09
10-May 1.95 0.23 0.84 0.45 0.24 0.10
23-May 2.13 0.25 0.79 0.44 0.23 0.09
06-June 2.29 0.32 0.86 0.38 0.24 0.09
RL Clark
Date DOC
Bio-
Polymers
Humic
Substances
Building
Blocks
LMW
Neutrals
LMW
Acids
10-Apr 2.26 0.22 0.84 0.58 0.30 0.21
25-Apr 2.03 0.21 0.74 0.47 0.25 0.09
10-May 2.05 0.25 0.80 0.51 0.25 0.10
23-May 2.14 0.23 0.81 0.45 0.25 0.08
06-June 2.05 0.30 0.81 0.46 0.23 0.09
Page 125
110
Table 8.10: Treated water PC scores
Treated Water (mg/L)
FJ Horgan
Date PC1 PC2 PC3 PC4 PC5
04-Apr -1.39E+02 -3.82E+00 1.02E+01 2.37E+01 1.14E+01
10-Apr -8.71E+00 9.29E+01 -2.64E+01 1.97E+01 4.18E+00
12-Apr -2.81E+01 7.88E+01 -8.14E+00 -2.71E+01 -1.59E+01
17-Apr -1.28E+02 3.11E+01 1.26E+01 -2.06E+01 9.39E+00
19-Apr -1.28E+02 3.11E+01 1.26E+01 -2.06E+01 9.39E+00
25-Apr -1.28E+02 3.11E+01 1.26E+01 -2.06E+01 9.39E+00
02-May -8.38E+01 7.90E+00 -7.55E+00 2.05E+01 7.40E+00
04-May -7.74E+00 6.74E+01 -4.12E+01 2.75E+01 -8.29E+00
08-May -2.63E+01 1.85E+01 -1.33E+01 3.56E+01 -2.38E+00
10-May -5.59E+01 2.72E+01 -6.45E+00 1.82E+01 1.82E+00
15-May -5.81E+01 3.33E+01 -7.07E+00 1.51E+01 5.48E+00
Island
04-Apr 3.80E+01 -2.87E+01 -1.57E+01 4.26E+00 1.25E+01
10-Apr 1.46E+02 7.08E+00 6.81E+00 -6.34E+00 1.33E+01
12-Apr 2.46E+02 4.13E+01 3.44E+01 2.11E+01 5.93E+01
17-Apr 6.33E+00 -2.54E+01 7.81E+00 -3.99E+01 -2.76E+00
19-Apr 6.33E+00 -2.54E+01 7.81E+00 -3.99E+01 -2.76E+00
25-Apr 2.01E+01 -2.22E+00 1.93E+00 -1.53E+01 4.80E+00
02-May 7.76E+00 -4.17E+01 -1.39E+01 2.96E+00 -6.26E+00
04-May 1.60E+01 -3.84E+01 -1.67E+01 1.10E+01 -5.89E+00
08-May 9.44E+01 -6.95E+01 -2.06E+01 1.63E+01 -1.97E+01
10-May 2.80E+01 -3.77E+01 -5.15E+00 -9.20E+00 -5.42E+00
15-May 3.60E+01 -4.00E+01 -5.11E+00 -2.20E+01 5.41E-01
RC Harris
04-Apr -3.02E+01 -1.76E+01 5.46E+00 -9.46E+00 7.09E+00
10-Apr 9.56E+01 5.22E+01 9.06E+01 2.02E+01 -4.01E+01
12-Apr 1.37E+02 1.45E+01 -2.64E+01 -1.94E+01 -9.36E+00
17-Apr -3.20E+01 -1.02E+01 6.67E+01 2.50E+01 -1.75E+01
19-Apr -2.25E+01 -1.50E+01 4.38E+00 -3.90E+00 1.14E+01
25-Apr -1.91E+01 2.20E+00 -6.80E+00 -1.80E+00 3.87E+00
02-May -2.19E+01 -2.26E+01 -6.65E+00 1.39E+01 -2.50E+00
04-May 2.14E-01 -8.96E+00 -1.79E+01 1.94E+01 -5.10E+00
08-May 3.00E+01 -1.92E+01 -1.29E+01 3.09E+01 -1.21E+01
10-May -2.99E+01 4.05E+00 -1.77E+00 1.28E+01 2.18E+00
15-May -1.88E+01 -2.62E+00 -2.43E+00 7.93E-01 5.42E+00
Page 126
111
Table 8.10: Treated water PC scores (continued)
RL Clark
04-Apr -3.97E+01 -3.45E+01 1.87E+01 -2.16E+01 4.81E+00
10-Apr 8.32E+01 6.11E+01 -2.34E+01 -3.03E+01 -1.77E+01
12-Apr 9.05E+01 5.49E+01 -1.69E+01 -3.45E+01 -1.57E+01
17-Apr -3.56E+01 -2.33E+01 1.66E+01 -2.21E+01 -2.27E+00
19-Apr 2.23E+00 -2.92E+01 1.18E+00 -3.49E+00 1.20E+01
25-Apr 2.23E+00 -2.92E+01 1.18E+00 -3.49E+00 1.20E+01
02-May -4.79E+01 -1.26E+01 -3.35E+00 1.75E+01 -4.14E-01
04-May 3.80E+00 -2.35E+01 -1.15E+01 7.81E+00 -6.58E+00
08-May 3.69E+01 -4.72E+01 -7.05E+00 2.02E+01 -1.54E+01
10-May -2.25E+01 -2.89E+01 1.01E+01 -2.58E+00 6.91E-01
15-May -1.55E+01 -1.92E+01 2.68E+00 -1.01E+01 5.84E+00
Table 8.11: Raw water PC scores
Raw Water (mg/L)
FJ Horgan
Date PC1 PC2 PC3 PC4 PC5
04-Apr -4.50E+01 8.58E+00 1.39E+01 1.50E+00 -1.17E+01
10-Apr 2.87E+02 -6.59E+01 -7.47E-02 5.74E+01 8.05E+00
12-Apr -3.68E+00 -1.33E+01 9.15E-01 -1.28E+01 -1.34E+01
17-Apr -1.02E+02 6.00E+00 1.43E+01 1.60E+01 -1.31E+01
19-Apr -1.02E+02 6.00E+00 1.43E+01 1.60E+01 -1.31E+01
25-Apr -1.02E+02 6.00E+00 1.43E+01 1.60E+01 -1.31E+01
02-May 7.72E+01 8.00E+01 -3.04E+01 -1.49E+00 6.09E+01
04-May -8.11E-01 -1.49E+01 1.30E+01 -2.36E+01 3.11E+00
08-May -2.14E+01 -1.64E+00 4.29E+01 -1.83E+01 5.12E+00
10-May -2.08E+01 -1.62E+01 1.23E+01 -8.50E+00 3.38E+00
15-May -4.37E+01 -1.94E+01 8.57E+00 -1.65E+00 -3.49E+00
Island
04-Apr 1.32E+01 1.66E+01 -3.03E+01 7.52E+00 -1.17E+01
10-Apr 5.68E+01 8.17E+00 -4.32E+01 -1.01E+01 -1.68E+01
12-Apr 2.43E+02 1.30E+00 4.16E+01 -2.51E+01 -1.12E+01
17-Apr -1.50E+01 2.94E+01 -2.51E+01 1.45E+01 -8.32E+00
19-Apr -1.50E+01 2.94E+01 -2.51E+01 1.45E+01 -8.32E+00
25-Apr 3.31E+01 -5.31E+00 -4.23E+01 5.09E+00 -1.33E+01
02-May -3.16E+00 4.27E+01 -2.71E+00 -2.60E+00 -1.11E+01
04-May 2.46E+01 4.60E+01 -5.15E+00 -4.94E+00 -9.22E+00
08-May 4.45E+01 6.40E+01 6.16E+00 -7.45E+00 -8.54E+00
10-May 3.89E-02 3.16E+01 -1.38E+01 5.71E+00 -8.97E+00
15-May 6.57E+00 2.95E+01 -2.50E+01 1.07E+01 -1.45E+01
Table 8.11: Raw water PC scores (continued)
Page 127
112
RC Harris
04-Apr -8.48E+01 -2.36E+01 -8.73E+00 1.34E+01 -3.49E+00
10-Apr 5.24E+01 -3.68E+01 -6.78E+00 -2.67E+01 -7.64E+00
12-Apr 8.43E+01 -4.00E+01 -1.86E+01 -1.14E+01 -1.55E+01
17-Apr 3.24E+01 1.31E+01 3.89E+01 7.04E+01 5.82E+00
19-Apr -5.55E+01 -1.86E+01 -5.80E+00 4.06E+00 2.46E+00
25-Apr 7.75E+00 -1.08E+01 2.81E+01 1.47E+01 -7.24E-01
02-May -2.94E+01 1.39E+01 1.83E+01 -5.33E+00 -1.56E+00
04-May -1.76E+00 1.70E+01 1.79E+01 -7.64E+00 -1.05E+00
08-May 2.89E+00 4.35E+00 4.54E+01 -1.73E+01 6.22E+00
10-May -3.77E+01 -1.60E+01 2.27E+01 -3.80E+00 -1.24E-02
15-May -2.95E+01 -7.53E+00 -2.36E-01 7.28E+00 -2.81E+00
RL Clark
04-Apr -5.67E+01 -2.18E+01 -2.89E+01 6.84E+00 1.53E+01
10-Apr 2.29E+01 -5.72E+01 -2.39E+01 -2.78E+01 4.62E+00
12-Apr 5.07E+00 -5.32E+01 -1.93E+01 -2.21E+01 8.32E-01
17-Apr -9.05E+01 -2.67E+01 -4.17E+00 2.91E+00 2.63E+01
19-Apr -7.01E+01 -4.12E+01 -2.37E+00 2.60E+00 1.93E+01
25-Apr -7.01E+01 -4.12E+01 -2.37E+00 2.60E+00 1.93E+01
02-May -3.34E+01 9.69E+00 8.09E+00 -1.28E+01 7.00E+00
04-May 4.06E+01 3.37E+01 -6.60E+00 -1.48E+01 5.31E+00
08-May 6.41E+01 3.46E+01 2.18E+01 -2.44E+01 1.14E+01
10-May -2.20E+01 6.01E+00 -3.57E+00 -2.39E+00 1.28E+01
15-May -4.23E+01 -6.18E+00 -8.71E+00 3.10E+00 5.27E+00
Page 128
113
Table 8.12: Raw water TOC
Raw water (mg/L)
Date
F. J. Horgan
WTP
R. C. Harris
WTP Island WTP R. L. Clark WTP
28-Feb 2.09 2.10 2.07 2.14
1-Mar 2.07 2.09 2.06 2.06
6-Mar 2.11 2.13 2.11 2.02
8-Mar 2.10 2.12 NR 2.06
13-Mar 2.19 2.20 2.10 2.06
15-Mar 2.13 2.14 2.07 2.00
20-Mar 2.16 2.19 2.08 2.08
22-Mar 2.19 2.15 2.03 1.84
27-Mar 2.00 2.06 1.98 1.97
29-Mar 2.05 2.17 1.97 2.02
2-Apr NR 2.24 2.15 2.21
4-Apr 2.07 2.07 1.99 2.06
10-Apr 2.17 2.18 2.12 2.33
12-Apr 2.10 2.10 2.24 2.07
17-Apr NR 2.14 NR 2.08
19-Apr NR 2.10 2.01 2.04
24-Apr NR 2.27 2.14 NR
26-Apr 2.24 2.38 2.24 NR
1-May 2.54 2.35 2.28 2.37
3-May 2.24 2.33 2.29 2.26
8-May 2.32 2.41 2.28 2.46
15-May 2.01 2.06 2.00 2.10
17-May 2.21 2.25 2.20 2.29
22-May 2.14 2.20 2.01 2.27
24-May 2.06 2.25 2.06 2.27
29-May 2.40 2.52 2.24 2.41
31-May 2.14 2.21 2.04 2.08
5-Jun 2.10 2.15 1.98 2.08
7-Jun 2.19 2.21 2.05 2.17
12-Jun 2.19 2.17 2.10 2.03
14-Jun 2.01 2.07 1.93 1.93
19-Jun 2.21 2.31 2.05 2.19
21-Jun 2.34 2.36 2.14 2.34
26-Jun 2.27 2.37 2.23 2.33
28-Jun 2.14 2.22 2.06 2.12
3-Jul 2.00 2.00 1.85 1.84
5-Jul NR 1.97 1.88 1.83
10-Jul 2.09 2.29 1.87 2.16
12-Jul 2.47 2.26 2.29 2.01
17-Jul 2.18 2.21 1.90 2.12
31-Jul 2.26 2.28 1.95 2.27
NR = not recorded
Page 129
114
Table 8.13: Treated water TOC
Treated Water (mg/L)
Date
F. J. Horgan
WTP
R. C. Harris
WTP Island WTP R. L. Clark WTP
28-Feb 2.02 2.01 2.08 1.92
1-Mar 1.91 1.88 2.03 1.86
6-Mar 1.94 1.96 2.05 1.86
8-Mar 1.96 1.95 NR 1.96
13-Mar 1.87 1.93 2.09 2.07
15-Mar 1.94 1.93 1.95 1.95
20-Mar 1.90 1.96 2.04 1.95
22-Mar 1.90 1.99 2.05 2.02
27-Mar 1.73 1.84 1.92 1.78
29-Mar 1.75 1.86 1.96 1.86
2-Apr NR 2.06 2.09 2.05
4-Apr 1.83 1.85 1.96 1.86
10-Apr 1.99 1.98 2.09 1.98
12-Apr 1.88 1.87 1.98 1.90
17-Apr NR 2.09 NR 1.95
19-Apr NR 1.90 2.05 2.00
24-Apr NR 2.11 2.23 NR
26-Apr 2.10 2.09 2.20 NR
1-May 2.14 2.19 2.23 2.11
3-May 2.22 2.17 2.24 2.18
8-May 2.18 2.21 2.25 2.21
15-May 1.85 1.91 1.97 1.89
17-May 2.08 2.12 2.21 2.16
22-May 2.01 2.04 2.04 1.94
24-May 2.04 2.16 2.05 2.17
29-May 2.14 2.21 2.17 2.19
31-May 2.10 2.04 2.04 2.00
5-Jun 1.93 1.95 1.95 1.90
7-Jun 1.95 1.94 2.00 1.95
12-Jun 2.00 1.90 2.12 1.82
14-Jun 1.75 1.77 1.80 1.72
19-Jun 1.98 2.02 1.98 1.98
21-Jun 2.11 2.14 2.22 2.13
26-Jun 2.27 2.42 2.28 2.23
28-Jun 1.83 2.19 2.09 2.05
3-Jul 1.52 1.77 1.81 1.73
5-Jul NR 1.81 1.84 1.76
10-Jul 1.66 2.15 1.81 2.03
12-Jul 1.81 1.99 1.92 2.15
17-Jul 1.65 2.05 1.94 1.96
31-Jul 1.81 2.17 1.97 2.07
NR = not recorded
Page 130
115
8.1.3 Peterborough pilot coagulant optimization data tables
Table 8.14: FEEM-PCA scores
Model 1
Pilot plant raw
water
Full-scale plant
raw water
Full-scale plant
settled water
Full-scale plant
filtered water
Date PC1 PC2 PC1 PC2 PC1 PC2 PC1 PC2
1/13/2012 122.16 -4.69 123.16 -4.62 -32.79 -7.67 -38.81 -11.27
1/19/2012 148.43 1.00 137.09 -2.95 5.25 28.63 -17.77 -7.55
1/26/2012 144.68 -11.33 137.33 -19.17 0.55 20.56 -20.59 -16.38
2/2/2012 176.47 -2.17 148.21 4.78 1.77 26.04 -24.49 1.62
2/9/2012 122.88 -18.82 131.46 -9.81 -18.67 4.55 -38.51 -19.57
2/16/2012 115.86 -31.30 107.23 -37.32 -35.63 -18.70 NR NR
3/22/2012 53.60 -44.28 54.44 -19.48 -76.47 -33.95 -72.93 -29.11
3/29/2012 98.84 39.57 97.27 40.58 -39.22 67.78 -60.27 12.00
4/4/2012 85.37 -3.17 84.05 -1.96 -37.16 45.10 -55.38 4.67
4/16/2012 82.54 34.82 NR NR NR NR -49.20 17.04
PACl pilot
settled water
Alum pilot settled
water
PACl pilot
filtered water
Alum pilot
filtered water
1/13/2012 -33.48 1.19 -29.62 -1.83 -34.66 3.59 29.62 0.64
1/19/2012 16.15 46.79 -3.03 12.16 -11.37 13.37 -15.57 4.46
1/26/2012 -4.96 15.86 -6.85 8.51 -23.57 -4.85 -16.08 0.13
2/2/2012 -25.49 2.27 -22.11 -4.40 -30.49 -7.66 -23.39 -6.74
2/9/2012 -39.36 -6.53 -37.45 0.13 -41.80 -4.98 -37.34 1.18
2/16/2012 -50.70 -32.68 -47.24 -31.55 -53.99 -32.34 -43.73 -3.94
3/22/2012 -54.71 -7.29 -74.49 -26.82 -79.01 -29.03 -76.79 -40.57
3/29/2012 -47.25 29.36 -39.92 28.33 -60.65 13.58 -51.87 27.52
4/4/2012 -59.46 -6.25 -33.03 36.29 -65.88 -10.71 -49.02 10.27
4/16/2012 -65.42 -16.95 -53.10 4.13 -59.16 1.65 -48.37 20.25 NR = not recorded
Page 131
116
Table 8.15: Statistical comparison of pilot and full-scale treatment trains: water quality parameters
Parameter
Alum and PACl Pilot1
Alum Pilot and Full-scale Plant2
Significantly
Different?3
Average
Difference
Standard
Deviation
Significantly
Different?3
Average
Difference
Standard
Deviation
Settled Water
Turbidity (NTU)
Baseline No -0.03 0.06 Yes 0.12 0.13
Experimental Yes 0.23 0.23 Yes 0.12 0.13
Filtered Water
Turbidity (NTU)
Baseline Yes 0.00 0.00 Yes -0.03 0.14
Experimental Yes 0.01 0.01 Yes -0.01 0.02
Settled Water pH Baseline Yes 0.01 0.01 No 0.03 0.07
Experimental Yes -.72 0.12 Yes 0.02 0.08
Filtered Water pH Baseline No 0.00 0.03 Yes -0.29 0.03
Experimental Yes -0.73 0.13 Yes 0.07 0.55
Total Organic Carbon
(mg/L)
Baseline No 0.00 0.02 Yes -0.14 0.05
Experimental Yes -0.03 0.22 Yes -0.05 0.30
Ultraviolet
Absorbance
Baseline No 0.00 0.00 Yes -0.01 0.00
Experimental No 0.00 0.00 Yes 0.00 0.01
Particle Counts (2 – 3
µm) (counts/mL)
Baseline No -0.34 0.72 Yes -2.83 2.08
Experimental Yes 1.91 6.14 Yes -3.83 7.87
Particle Counts (3 – 5
µm) (counts/mL)
Baseline Yes 3.34 1.67 Yes -13.79 12.72
Experimental Yes 7.31 9.90 Yes -5.09 15.16
Particle Counts (5 – 7
µm) (counts/mL)
Baseline Yes 5.08 2.52 Yes -8.18 6.69
Experimental Yes 3.43 6.54 Yes -2.12 9.40
Particle Counts (7 – 10
µm) (counts/mL)
Baseline Yes 0.54 0.51 Yes -6.56 4.16
Experimental Yes 1.70 3.29 Yes -2.32 7.18
Particle Counts (10 –
15 µm) (counts/mL)
Baseline Yes 0.16 0.06 Yes -1.81 0.92
Experimental Yes 0.41 0.74 Yes -1.32 3.72
Particle Counts ( > 15
µm) (counts/mL)
Baseline Yes 0.04 0.03 Yes -0.66 0.30
Experimental Yes 0.07 0.14 Yes -0.42 0.80
Total Flow (L/min) Baseline No 0.00 0.00 NA
NA NA
Experimental No 0.00 0.01 NA
NA NA
Filter Flow (L/min) Baseline No 0.00 0.00 Yes -0.15 0.14
Experimental No 0.00 0.01 Yes -0.05 0.24 1 difference = control pilot – experimental pilot
2 difference = control pilot – full-scale plant
3 Assessments made with 95% confidence
NA: not applicable
Particle counts are from filtered water
Page 132
117
Table 8.16: Statistical comparison of pilot and full-scale treatment trains: chlorine residual and disinfection by-products
Parameter
Control and Experimental Pilot1
Control Pilot to Full-scale Plant2
Significantly
Different?3
Average
Difference
Standard
Deviation
Significantly
Different?3
Average
Difference
Standard
Deviation
Free Chlorine Residual
(SDS) (mg/L)
Baseline No 0.01 0.07 No -0.05 0.10
Experimental Yes 0.11 0.11 Yes -0.06 0.13
Total THMs (µg/L) Baseline No 2.3 8.9 No -0.3 6.6
Experimental Yes -7.6 10.2 Yes 2.7 8.0
Total HAAs (µg/L) Baseline No -3.0 5.6 No -3.9 7.2
Experimental No 0.8 21.0 No -3.2 10.5 1 difference = control pilot – experimental pilot
2 difference = control pilot – full-scale plant
3 Assessments made with 95% confidence
Page 133
118
8.2 Calculations
8.2.1 Analysis of variance
Analysis of variance (ANOVA) was used to identify statistically significant differences
between the four treatment facilities. One-way ANOVA was used to test the null hypothesis that
two or more sample sets were from the same population through application of the F-statistic
(Dean and Voss, 1999). The F statistic was calculated as follows (Montgomery and Runger,
2003):
∑
∑ ( )
⁄
where, K is the total number of groups, N is the total number of observations, ni is the number of
observation within the i’th group, is the sample mean of the i’th group, is the mean of all
observations, i denotes the group, j denotes the observation within the i’th group
If the calculated F statistic is greater than the critical F value (from a standardized
distribution) at a specified confidence level, the null hypothesis that all sample sets are derived
from the same population is rejected. The drawback to ANOVA is it does not specify the sample
sets within the group which are significantly different (Montgomery and Runger, 2003).
8.2.2 Tukey’s method
Tukey’s method allows simultaneous pairwise comparisons of multiple sample means
(Dean and Voss, 1999). The method calculates the minimum significance difference (MSD) for
each possible pair of means within the group. The MSD can be thought of as a confidence
interval for the difference between two sample means which, when spans 0, indicates no
significant difference between the sample sets. In generalized terms, assume Tukey’s method is
applied to identify significance between sample means in a group of four sample sets. For any
two sample sets within the group, i and j, the MSD can be expressed as follows:
(8.1)
(8.2)
Page 134
119
(
√ )√ (
)
where, µi and µj are the means of sample sets i and j, qv,n-1,α is the value of the studentized range
distribution for: v sample sets = 4, n observations, r replicates, α confidence level (95%) =
0.05.
∑ ∑ ( )
Where, y is a measured value
8.2.3 Statistics on water quality differences at Peterborough Pilot
The example calculation attempts to determine if there is a significant difference in TOC
between the control and experimental pilot trains.
Baseline:
To first remove the effects of source water quality, difference between the two parameters is
calculated.
where, Dn is the calculated difference on a given day (n), and n is the sample size (i.e. number of
days)
The average ( and standard deviation (S) of the sample set D is calculated. For
baseline TOC (n = 18) data the following values were found:
It is assumed that the ‘population’ mean or the expected difference between TOC for the
two pilots is 0. This leads to the postulation of the null and alternative hypotheses.
(8.3)
(8.4)
(8.5)
Page 135
120
where, Ho is the null hypotheis, H1 is the alternative hypothesis,
Depending on the circumstance, either Student’s t statistic or Tukey’s method was used
to test the null hypothesis.
Page 136
121
121
8.3 Process flow diagrams of City of Toronto water treatment plants
Figure 8.5: F.J Horgan process flow diagram
Figure 8.6: Island process flow diagram
Figure 8.7: R.C. Harris process flow diagram
Figure 8.8: R.L. Clark process flow diagram
Page 137
122
122
Figure 8.5: F.J Horgan process flow diagram
Page 138
123
123
Figure 8.6: Island process flow diagram
Page 139
124
124
Figure 8.7: R.C. Harris process flow diagram
Page 140
125
125
Figure 8.8: R.L. Clark process flow diagram
Page 141
126
126
8.4 Program code
8.4.1 Real RLS files (Python 2.6)
Compiles data from the fluorescence spectrophotometer into one data file. Output from
the software is emission wavelengths for every excitation wavelength.
import glob
import xlwt
outbook = xlwt.Workbook()
while True:
prefix = raw_input("sample prefix: ")
osheet = outbook.add_sheet("%s" % prefix)
filenames = []
rows = []
q =1
#determine number of files
for name in glob.glob('%s0*.RLS' % prefix):
q = q + 1
#print q
#put files in correct order
for i in range(1,q,1):
#puts files with just 1 digit first
for name in glob.glob('%s00%i.RLS' % (prefix, i)):
# print name
Page 142
127
127
filenames.append(name)
#then double digits
for name in glob.glob('%s0%i.RLS' % (prefix, i)):
# print name
filenames.append(name)
# print filenames
# go through an get values
for d in range(0,(q-1),1):
aFile = open(filenames[d], "rU")
f = 0
print aFile
for line in aFile.readlines():
rows = line.rstrip('\n').split('\t')
#print rows[1]
osheet.write(f, d, float(rows[1]))
f += 1
#columns = zip(*rows)
#print columns
aFile.close()
cont = raw_input("Continue? (0 to save file, 1 to continue): ")
if cont == '0':
break
outputname = raw_input("Output file name?: ")
outbook.save("%s.xls" % outputname)
print "saved as %s.xls" % outputname
Page 143
128
128
8.4.2 Manipulate for PCA input (Python 2.6)
Subtracts the Milli-Q® intensities (first worksheet for every day) and compiles all
samples in columns for PCA input. Iterates for all *.xls files in a folder.
import xlwt
import xlrd
import glob
import os
outbook = xlwt.Workbook()
namebook = xlwt.Workbook()
osheet = outbook.add_sheet("Output")
namesheet = namebook.add_sheet("Names")
osheetcol = -1
onsheetrow = 0
for name in glob.glob('*.XLS'):
title, ext = os.path.splitext(name)
book = xlrd.open_workbook("%s.xls" % title)
msheet = book.sheet_by_index(0)
numrows = msheet.nrows
numcols = msheet.ncols
numdata = int(numrows*numcols)
print title
Page 144
129
129
for i in range(1,book.nsheets,1):
asheet = book.sheet_by_index(i)
namesheet.write(onsheetrow,0,title)
namesheet.write(onsheetrow,1,asheet.name)
#osheet = outbook.add_sheet(asheet.name)
osheetrow = 0
osheetcol = osheetcol + 1
print(".")
#subtract milliQ values and remove emission columns
#counting by 2 removes emission column values (not the actual columns)
for q in range(0,asheet.ncols,1):
for p in range(asheet.nrows):
#USED FOR ELIMINATION OF SCATTERING REGIONS
#if q <= 10 and p >= ((24*q)+361):
# osheet.write(osheetrow, osheetcol, "0")
#elif q >= 5 and p <= (25*(q-4)):
# osheet.write(osheetrow, osheetcol, "0")
#else:
# osheet.write(osheetrow, osheetcol, (float(asheet.cell(p,q).value) -
float(msheet.cell(p,q).value)))
osheet.write(osheetrow, osheetcol, (float(asheet.cell(p,q).value) -
float(msheet.cell(p,q).value)))
osheetrow = osheetrow + 1
onsheetrow = onsheetrow + 1
Page 145
130
130
outbook.save("Output RS Elim ALL.xls")
namebook.save("Output Names.xls")
8.4.3 Deconvolute loadings (Python 2.6)
Takes loadings in column format and reorganizes into spectrum format.
import xlwt
import xlrd
bopen = raw_input('open file: ')
book = xlrd.open_workbook("%s.xls" % bopen)
outbook = xlwt.Workbook()
for i in range(book.sheet_by_index(0).ncols):
osheet = outbook.add_sheet('PC %i' % i)
rownum = 0
for p in range (0,13,1):
for q in range (0,601,1):
osheet.write(q, p, float(book.sheet_by_index(0).cell(rownum, i).value))
rownum = rownum + 1
outbook.save('%s results.xls' % bopen)
Page 146
131
131
8.5 Increased THM formation in PACl treated water
8.5.1 Background
Currently the Peterborough pilot plant is running two identical parallel treatment trains to
observe water quality differences caused by using either alum or poly-aluminum chloride (PACl)
as coagulants. The control scheme between the two pilot trains is to dose the respective
coagulants to match filtered water total organic carbon (TOC). While filtered water TOC levels
between alum and PACl treated water are matching, it has been observed that PACl treated water
produces higher concentrations of THMs (trihalomethanes) after contact with 2-3 mg/L of
sodium hypochlorite for 24 hours. Due to PACl being ‘pre-hydrolized’, it does not depress the
pH to a significant degree when added to water, unlike alum. This results in alum treated water
being approximately 0.6 pH units lower than PACl treated water. Literature indicates that more
basic waters are more favourable to THM production, therefore the observed difference in THM
formation could be due to the difference in treated water pH. Another possible explanation for
the observed THM concentrations is that alum and PACl have differing removal efficiencies for
specific NOM fractions which contribute to DBP formation.
8.5.2 Method: pH adjusted DBP Formation Tests
To address differences in pH, both alum and PACl treated water was chlorinated at a
common pH as well as without pH adjustment (baseline case). THM and HAA concentrations
samples (after 24 hours of chlorination) were measured and normalized by dividing by the initial
TOC concentration (measured prior to chlorination). The average normalized THM and HAA
concentrations of pH adjusted and non-adjusted samples were statistically compared using t-tests
to identify if there is a significant difference in concentrations. If pH is to explain the difference
in DBP concentrations the average normalized THM and HAA concentrations of alum and PACl
treated water was equal when chlorinated at an equal pH. Statistical comparison of normalized
DBP concentrations in the un-adjusted case was used to confirm the pilot observations that DBP
formation is not equal between alum and PACl when chlorinated at dissimilar pH values.
8.5.2.1 Jar Tests
To determine if differences in pH can account for the observed increased THM formation
in PACl treated water, bench-scale jar tests followed by 24 hour DBP formation potential tests
will be conducted. Jar tests will consist of coagulation, flocculation, sedimentation, and vacuum
Page 147
132
132
filtration. The jar tests will be conducted using a PB-700 Standard Jar Tester paddle stirrer with
six square, acrylic 2-L jars (Phipps & Bird, Richmond, VA). In addition to the raw water, both
alum and PACl will be collected from the pilot plant to ensure consistency of chemicals used.
Alum and PACl dosages for the jar tests will be equal to the dosages for the pilot plant at the
time of raw water collection. It is expected that these doses will be approximately 50 mg/L alum
and 25 mg/L HI705 PACl. Triplicates of each coagulant dose for the jar tests will be included to
give an indication of the precision of the results and to allow for paired t-tests to check for
statistical significance.
8.5.2.1.1 Peterborough Pilot Plant Samples
In conjunction with the jar tests, samples will be collected form the Peterborough pilot
plant. There will be no alterations to the operation of the pilot plant. Duplicate samples for
TOC, UV254, and FEEM will be collected from both pilot plant trains. TOC and UV254 samples
will be collected in a 40 mL vial with 3 drops of concentrated sulphuric acid for preservation.
FEEM samples will be collected in a separate 40 mL vial with no additives.
8.5.2.1.2 Analysis
Filtered samples from each jar and pilot plant samples will be subjected to TOC, UV254,
pH, and FEEM analysis. Analysis for these parameters will be completed at the University of
Toronto to ensure consistency of measurement. TOC and UV254 will indicate the amount of
organic material remaining in the treated water. To provide reference and allow for the
calculation of TOC reduction from each coagulant and dose, raw water TOC, UV254, and FEEM
will also be measured. These values will be ultimately used to normalize DBP concentrations and
provide a reference for FEEM results when comparing the two coagulants in terms of relative
NOM fraction removal.
8.5.2.2 DBP Formation Tests
Filtered samples from the jar tests will be chlorinated and allowed to stand at room
temperature (23oC) for 24 hours for DBP formation potential tests. Chlorine dose will be defined
by the dose used at the pilot plant for the day raw water is sampled. It is expected that this dose
will be approximately 3 mg/L. A solution of sodium hypochlorite at 600 mg/L is prepared to
dose the 250 mL sample bottles.
Page 148
133
133
Based on results from the pilot plant, it is expected that after filtration water treated with
alum will have a pH of approximately 7.1; while PACl treated water will match raw water pH
fairly closely with a pH of approximately 7.9. The hypothesis is that PACl treated water will
have a greater concentration of THMs than alum treated water, after 24 hours of contact time
with chlorine, due to the higher pH of PACl treated water. To test the validity of this claim, the
pH of a set of PACl and alum treated water samples will be reduced a target pH of 7 while
another set will be adjusted to a target pH of 8, prior to the DBP formation test. It is expected
that the concentration of DBPs will be equal between the two coagulants when they are
chlorinated at the same pH. Furthermore, the set of samples chlorinated at a pH of 8 will result
in higher THM concentrations and lower HAA concentrations compared to the set adjusted to a
pH of 7. Since the intention is to see the difference in DBP formation at different pH values,
non-pH adjusted samples of both alum and PACl treated water will be subjected to DBP
formation tests as a control measure and to provide baseline results.
For the jar tests, three 250 mL samples of filtered water will be collected from every jar.
Two of these 250 mL samples will be pH adjusted to 7 and 8 respectively and the final 250 mL
sample will be left un-adjusted. Chlorine residual of the samples will be measured after 24 hours
to determine the chlorine demand of each water sample. After quenching, samples from each
250 mL bottle will be collected for THM, HAA, TOC, UV254, and FEEM analysis. As with jar
test samples, TOC and UV254 will allow for tracking the changes to organic content due to
chlorination. These will primarily be used for comparison to FEEM results.
Similar to the jar tests, at the pilot plant three samples from each source will be collected
for pH adjustment and DBP formation tests (two for pH adjustment to 7 and 8, one for un-
adjusted). After pH adjustment chlorine will be spiked into the samples and capped. After 24
hours the chlorine residual will be measured and the remaining chlorine will be quenched. A set
of 25 mL vials will be prepared at UofT and shipped to Peterborough for the storage of
THM/HAA samples. Each 25 mL vial will have an appropriate amount of ammonium chloride
for quenching. These samples, once collected will be sent back to UofT for analysis.
Furthermore, from each of these adjusted samples, duplicates for TOC, UV254, and FEEM
analysis will be collected in 40 mL amber vials (1 for TOC/UV254 and 1 for FEEM).
Page 149
134
134
8.5.2.2.1 Analysis
The significance of the DBP results was determined through a series of paired t-tests.
The measured values, DBP concentrations, will be normalized with TOC by dividing DBP
concentration (µg/L) by TOC concentration (mg/L). Normalization of the total DBP
concentrations by a gross measure of organic content allows for comparison of the two
coagulant’s performance by the amount of available reactants. This will also allow for the
results from different doses for each coagulant to be included in a general analysis.
To test the hypothesis (that at a common pH DBP production will be equal between alum
and PACl treated water) a series of t-tests will be performed. Two pieces of evidence are needed
to prove the role of pH:
1. That normalized DBP concentrations between alum and PACl treated water are
statistically the same when chlorinated at a common pH (both pH 7 and 8).
2. That normalized DBP concentrations between alum and PACl treated water are
statistically different when chlorinated at dissimilar pH (unadjusted samples).
Another set of t-tests was set up to test the difference between jar test and pilot samples.
This will indicate how closely jar tests relate to the pilot plant. The set will attempt to show
statistical similarity between average normalized DBP concentrations of jar test and pilot plant
samples that underwent the same treatment (same coagulant, chlorinated at the same pH).
8.5.3 Results and Discussion
8.5.3.1 Jar test water quality
Water quality of the filtered water both from jar tests and pilot was characterized in order
to correlate initial conditions with DBP formation potential. As discussed in the methods
section, water quality was quantified through measurements of TOC, UVA254, pH, PCA of
FEEM, and LC-OCD. This section focuses on results for TOC, UVA, and pH while NOM
fractionation results are discussed in a later section.
Results were as expected for all parameters measured based on historical data from the
pilot plant. The coagulant doses were chosen to match the doses at the pilot plant on March 15th
,
2012 (50.7 mg/L Alum, 27.7 mg/L PACl). This was also when raw water was collected and sent
to UofT for the jar tests. The current operational scheme of the pilot was to maintain TOC levels
equal between the two trains. As expected, this was reflected in the results of the jar test. With
90% confidence, TOC levels between the alum and PACl treated water were found to be
Page 150
135
135
statistically equal. This was further reflected in t-tests showing that UVA and SUVA are also
statistically similar. The results also showed that the pH of alum and PACl treated water were not
similar. T-test results are shown in Table 8.17.
Table 8.17: Jar Test Water Quality T-Test Results
T-test parameters Critical Value
(90% confidence) T value
Significantly
Equal?
TOC of Alum vs. PACl
treated water 3.18 0.171 Yes
UVA of Alum vs.
PACl treated water 3.18 0.000 Yes
SUVA of Alum vs.
PACl treated water 3.18 0.494 Yes
pH of Alum vs. PACl
treated water 3.18 12.954 No
Water quality results between the triplicate jars showed good agreement and indicated
high precision in the jar test method. Standard deviations of TOC measurements between
triplicates were approximately 0.1 mg/L which was considered to be very low. UVA results
showed very low standard deviations which was anticipated for this measurement method. UVA
was less sensitive to NOM concentration and therefore will not reflect small changes in NOM
between samples.
To see how well the two measured parameters for NOM estimation related (TOC and
UVA), they were plotted as Figure 8.9 and Figure 8.10.
Figure 8.9: UVA vs. TOC for Jar Test Results: Full Range
R² = 0.958
0.000
1.000
2.000
3.000
4.000
5.000
6.000
7.000
0.000 0.020 0.040 0.060 0.080 0.100 0.120 0.140 0.160 0.180
TOC
(m
g/L)
UVA
Page 151
136
136
Figure 8.10: UVA vs. TOC for jar test results: filtered water samples
The entire range of results is shown in Figure 8.9, including raw water and the jar tests
blanks. In Figure 8.10 only filtered water samples from the jar test, without the Milli-Q blanks
and raw water are shown. The limited range is shown to demonstrate that there was a strong
relationship between UVA and TOC even when the value range was greatly limited. RSQ for
both ranges were very strong which implied a good linear correlation between TOC and UVA254
measurements.
8.5.4 Controls and Blanks
The TOC blanks (untreated Milli-Q) were found to have an average concentration of
0.363 mg/L TOC. This low amount of background interference was potentially from the
sulphuric acid used in the TOC analysis method. Alternatively, the background TOC levels
could have resulted from contaminated reagents used by the instrument. With this subtraction,
jar test blanks (AM50 1 and PM27 1) were found to have TOC concentrations of 0.590 and 0.577
mg/L, respectively. It is possible that the presence of low levels of TOC in these blanks came
from the filtration step of the jar tests. Some amount of the filter paper or substance on the filter
paper could have been introduced into the sample during vacuum filtration. The similarity of
TOC between the two jar test blanks indicates that all samples would contain a certain amount of
extra TOC introduced from the jar test method. Through the assumption that interference levels
were similar between all samples still allows for comparison, although blank results imply some
R² = 0.9317
3.450
3.500
3.550
3.600
3.650
3.700
0.073 0.074 0.075 0.076 0.077 0.078 0.079 0.080
TOC
(m
g/L)
UVA
Page 152
137
137
error in overall accuracy of the results. For jar test blanks, UVA254 showed similar trends in
background NOM.
A series of 6 TOC quality control samples made to 3.0 mg/L were run with the samples
to check the validity of the calibration. The average measured concentration of the 6 quality
control was 3.01 ± 0.06 mg/L which indicated a high level of accuracy and precision.
8.5.5 Water quality
Pilot plant water quality results were obtained through averaging online data collected
during March 15th
, 2012. TOC and UVA measurements were made a 2 minute intervals
throughout the day. Each water source did not have a dedicated instrument therefore
approximately every 2 hours the source being analyzed by the instrument was switched as to
allow for multiple sources to be monitored from one instrument. This worked out to
approximately 160 – 200 measurements for each water source every day. pH of each water
source was measured continuously throughout the day resulting in 720 measurements at 2 minute
intervals.
Statistical tests were performed on pilot plant data to demonstrate the similarity of NOM
concentrations between both pilot plant trains. Results are shown in Table 8.18.
Table 8.18: Pilot Plant Water Quality T-Test Results
T-test parameters Critical Value
(90% confidence) T value
Significantly
Equal?
TOC of Alum vs. PACl
treated water 1.96 11.77 No
UVA of Alum vs.
PACl treated water 1.96 9.50 No
SUVA of Alum vs.
PACl treated water 1.96 7.98 No
pH of Alum vs. PACl
treated water 1.96 1195.35 No
The t-test results for pilot plant data showed that no parameter was significantly equal
between trains (Alum vs. PACl). This was mainly due to the very large sample size of relatively
stable parameters. As discussed the sample numbers for these test were between 160 and 720.
This created an expected accuracy that exceeded the instrument’s capability. It is suggested that
in this case the values be accepted as equal since the differences between measured values is
Page 153
138
138
below the expected accuracy of the instrument. For example the TOC analyzer was accurate to
within 0.1 mg/L TOC. The average value for pilot plant 1 and 2 for March 15th
was 3.61 and
3.69, respectively, which was within the margin of error for the TOC analyzer. Using this
alternative way of looking at the data, similar conclusions to those found with jar test results can
be formed. This was further illustrated in the UVA254 measurements. There was a difference in
average absorbance of 0.001, which was far below the accuracy expected from the spectrometer.
Therefore, it was concluded TOC, UVA, and SUVA are equal between Alum and PACl treated
water at pilot-scale. Furthermore, the pH was not similar and was approximately 0.6 pH units
higher for the PACl treated water.
8.5.6 Comparison of pilot and jar test results
Results from the jar tests showed some discrepancies when compared to the pilot plant
data. An underlying effort of this study was to maintain variables as similar as possible between
the jar tests and the operation of the pilot plant.
The TOC of coagulated water from the jar tests compared to the pilot plant were similar
at approximately 3.6 mg/L TOC. As with the pilot plant samples, statistical significance using t-
tests had limited meaning since the number of samples for pilot plant data misrepresented the
accuracy of the instrument. However, all values for alum and PACl treated water for both jar
tests and pilot plant samples were within 0.1 mg/L TOC.
Similarity between pilot and jar tests did not extend past TOC of treated water samples.
Of particular note, was the difference in TOC for raw water. Based on 172 measurements, TOC
at the pilot plant was found to be 6.52 mg/L while duplicate samples at UofT were found to have
a TOC of 5.74 mg/L. Under an assumption that the TOC of treated water samples were the
same, as discussed above, a difference of 0.75 mg/L could not be attributed to instrument error.
If the discrepancy was due to instrument bias it would have been expected that all values would
be off by a similar amount. The implication was also that at the pilot plant, the same coagulant
dosages are responsible for the removal of an extra 0.75 mg/L TOC. One possible explanation
was the pilot plant filters were more effective than the 1.4 micron glass fibre filters used for jar
tests.
UVA measurements were found to be consistently higher at the pilot plant when
compared to jar test results. At pilot-scale, both trains were found to have a UVA254 of
approximately 0.11 while jar test results indicated an absorbance of 0.08 for both sample types.
Page 154
139
139
This result was unexpected based on the fact that TOC levels match between jar tests and pilot.
This was attributed to instrumental bias from one or both of the instruments.
Another significant between the jar tests and pilot results was the pH of all water
samples. It appears that pH measurements at UofT were higher by approximately 0.15 pH units.
The spread between Alum and PACl treated water was similar (0.68 for Jar Tests, 0.63 for pilot
samples). The similarity in spread would have indicated that the experiment is still comparable
between pilot and jar tests and that the error was due to instrumental bias. That being said there
is a larger implication in that the pH of raw water at UofT was measured to be above 8 while at
pilot it was measured to be below 8. During the pH adjustment prior to chlorination, PACl jar
tests water was adjusted down to a pH of 8 while PACl pilot plant samples were adjusted up to a
pH of 8. Since this instrumental bias was not accounted for in the experiment the water from
pilot and jar tests were very likely at dissimilar pH values when chlorinated.
8.5.7 DBP Formation Results
In sections 8.5.3.1, it was shown that NOM concentrations from both Alum and PACl
treated water were equal with doses of 50.7 mg/L and 27.7 mg/L, respectively. To perform the
DBP formation test, water samples were spiked with 2.6 mg/L of sodium hypochlorite and 24
hours of contact time was allowed. By maintaining the two concentrations of reactants for DBP
formation equal, the effect of pH and NOM fractions was observed.
8.5.7.1 Trihalomethane Formation
THM analysis showed that only chloroform (trichloromethane: TCM) and
bromodichloromethane (BDCM) were above detection limit in all samples. Total THMs were
calculated based on the sum of both TCM and BDCM concentrations. pH vs. Total THM
concentration is plotted to show the relationship between pH and THM formation. Figure 8.11
shows each replicate TTHM concentration and pH since pH did differ slightly between samples
even with adjustment. A second plot shows the average and standard deviation (as error bars) for
samples chlorinated at similar pH. Indications of sample groups are also shown on Figure 8.11.
A similar plot for pilot plant data is shown as Figure 8.12.
For both jar test and pilot-scale samples Total THM concentrations increased with
increasing pH. This conformed to expectations based on literature as well as pilot plant
historical data showing increased THM formation in PACl treated water. For jar tests and pilot
samples, average TTHM concentrations of un-adjusted PACl samples are 23% and 22% higher
Page 155
140
140
than un-adjusted Alum samples. This followed well with the historical data which shows 20-
30% higher TTHM concentrations in PACl treated water.
For jar test results, there was a relatively strong linear relationship between TTHM
concentration and pH. The RSQ for Alum and PACl treated water is 0.84 and 0.96, respectively.
A linear relationship was not as pronounced for pilot results with RSQ values of 0.70 and 0.81
for Alum and PACl treated water respectively. The reduced linearity of pilot results was likely
due to the large discrepancy between duplicates of samples adjusted to pH 7. It was not apparent
what this large deviation between duplicates at pH 7 was caused by, but it was consistent
between Alum and PACl treated water. It was concluded that the relationship between TTHM
concentration and pH appeared to be fairly linear between 7 and 8; however samples at more
varied pH values would be needed to confirm this assertion.
Figure 8.11: pH vs. Total THM Concentration for Jar Test Samples
35.00
40.00
45.00
50.00
55.00
60.00
6.8 7 7.2 7.4 7.6 7.8 8 8.2
Tota
l TH
M C
on
cen
trat
ion
(µ
g/L)
pH
Alum Jar Tests
PACl Jar Tests
Adjusted to pH 7
Un-adjusted
Alum Samples
Adjusted to pH
8 and un-
adjusted PACl
samples
Page 156
141
141
Figure 8.12: pH vs Total THM Concentration for Pilot Plant Samples
To account for small variances in NOM concentrations between samples, TTHM values
were normalized with respect to TOC. This was performed by dividing TTHM concentration
(µg/L) by pre-chlorination TOC (mg/L), which yields a normalized value of µg TTHM/mg TOC.
In a sense this can be regarded as a measure of formation potential, as in for every mg of TOC in
the water sample so many µg of THMs form.
T-tests were performed on the normalized TTHM values to investigate if formation
potential matches between Alum and PACl treated water at an equal pH. The results of these t-
tests are shown in Table 8.19 and Table 8.20.
.
Table 8.19: T-Tests for Normalized TTHM Jar Test Results
T-Test Critical Value
(90% confidence) T Value Significant Equal?
Alum vs. PACl at pH 7 2.78 1.894 Yes
Alum vs. PACl at pH 8 2.78 0.834 Yes
Alum UA vs. PACl UA 2.78 7.081 No
20.00
25.00
30.00
35.00
40.00
45.00
6.8 7 7.2 7.4 7.6 7.8 8
Tota
l TH
M C
on
cen
trat
ion
(µ
g/L)
pH
Pilot Plant 1 - Alum
Pilot Plant 2 - PACl
Adjusted to pH 7
Un-adjusted
Alum Samples
Adjusted to pH 8
Un-adjusted
PACl samples
Page 157
142
142
Table 8.20: T-Tests for Normalized TTHM Pilot Plant Results
T-Test Critical Value
(90% confidence) T Value Significant Equal?
Alum vs. PACl at pH 7 4.3 0.126 Yes
Alum vs. PACl at pH 8 4.3 2.014 Yes
Alum UA vs. PACl UA 4.3 12.3524 No
The statistical tests showed that when chlorinated at a common pH, water treated either
by Alum or PACl had the same potential for THM formation. Furthermore, the t-tests confirmed
that un-adjusted samples have differing THM formation potentials. These two pieces of
information together support the hypothesis that the observed higher THM concentrations in
PACl treated water were due to a difference in pH.
8.5.7.2 Haloacetic Acid Formation
All samples were analyzed for all nine species of HAAs although only bromoacetic acid
(MBAA), trichloroacetic acid (TCAA), dibromoacetic acid (DBAA), and chlorodibromoacetic
acid (CDBAA) were detected. As with THMs, plots of pH vs. Total HAA Concentration were
prepared.
Figure 8.13: pH vs. total HAA concentration jar tests
55.00
60.00
65.00
70.00
75.00
80.00
85.00
6.8 7 7.2 7.4 7.6 7.8 8 8.2
Tota
l HA
A C
on
cen
trat
ion
(µ
g/L)
pH
Alum Jar Test
PACl Jar Test
Adjusted to pH 7
Un-adjusted
Alum Samples
Adjusted to pH
8 and un-
adjusted PACl
samples
Page 158
143
143
Figure 8.14: pH vs. total HAA concentration pilot samples
Results from the jar tests showed an apparent decrease in Total HAA concentration when
chlorinated at pH 8 compared to at pH 7. This agreed with literature in that HAA formation is
favoured in more acidic conditions. For jar tests, there was little difference between un-adjusted
and pH 8 PACl samples since the filtered water was found to be approximately pH 8. Pilot plant
results do not show the same drop in HAA concentration at a higher pH, except for Alum treated
water. Historical data from the pilot plant showed only small differences (±10%) between HAA
concentrations from both trains. Furthermore, one train was not observed to be consistently
higher than the other.
Jar test results showed a stronger negative linear correlation between total HAA
concentration and pH. RSQ was found to be 0.836 and 0.775 for Alum and PACl samples,
respectively. This indicated some degree of linear relationship. For pilot results, RSQ was
found to be 0.796 and 0.443 for Alum and PACl samples, respectively. This further supported
the observation that pilot samples do not show the negative correlation between pH and total
HAA concentration. From these results, it was concluded that the effect of pH on HAA
concentration was not as pronounced as seen with THMs.
As with TTHMs, THAAs were normalized to account for small changes in TOC between
samples. HAA formation potential was then subjected to several t-tests in order to show if
35.00
40.00
45.00
50.00
55.00
60.00
6.8 7 7.2 7.4 7.6 7.8 8
Tota
l HA
A C
on
cen
trat
ion
(µ
g/L)
pH
Pilot Plant 1 - Alum
Pilot Plant 2 - PACl
Adjusted to pH 7 Un-adjusted
Alum Samples
Adjusted to pH 8
Adjusted to pH 8
Un-adjusted
PACl Samples
Page 159
144
144
samples treated at equal pH have equal potential for HAA formation. Results from the t-tests are
shown in Table 8.21 and Table 8.22.
Table 8.21: T-Tests for Normalized THAA Jar Test Results
T-Test
Critical Value
(90%
confidence)
T Value Significant Equal?
Alum vs. PACl at pH 7 2.78 4.3274 no
Alum vs. PACl at pH 8 2.78 3.9336 no
Alum UA vs. PACl UA 2.78 3.5685 no
Table 8.22: T-Tests for Normalized THAA Pilot Plant Results
T-Test
Critical Value
(90%
confidence)
T Value Significant Equal?
Alum vs. PACl at pH 7 4.3 1.9540 yes
Alum vs. PACl at pH 8 4.3 5.3141 no
Alum UA vs. PACl UA 4.3 0.9843 yes
8.5.8 Conclusion
In conclusion, this work demonstrated that the observed increased THM formation in
PACl treated water in comparison to alum was caused by the pH. Since PACl does not depress
the pH, THM formation is favoured. When chlorinated at equal pH, PACl and alum treated
water showed no statistical difference in THM formation. HAAs were not found to be
significantly affected by any pH changes and furthermore, were found to be significantly equal at
unadjusted pH values (alum: ~7.1, PACl: ~7.8).