FULL TEXT - American Water Works Association

Peer Reviewed
Evaluating the Efficacy of PAC and Water
Parameters to Remove Trace Organics
CH RISTINE O . VAL C A RC E , 1 E RIC A W. GO NZA G A, 1 AN D D AV I D W. MAZ Y C K 2
1Carbonxt,
Gainesville, Fla.
of Environmental Engineering Sciences, University of Florida, Gainesville
2Department
Multivariable regression modeling was used to
evaluate the efficacy of commonly used powdered
activated carbon (PAC) and water parameters for
selecting a commercial PAC to remove trace organic
contaminants from natural waters. Trace con­
centrations of radiolabeled methylisoborneol (MIB)
and atrazine were spiked into natural waters, including
surface and groundwater sources, and treated with
PACs at varying contact times and PAC doses. A
rigorous selection criterion was applied to the
regression modeling effort to optimize the R2-adjusted,
hypothesis test the significance and prevent multi­
collinearity of regression coefficients. The results
indicated that MIB and atrazine removal were affected
by the same PAC parameters: tannin value and phenol
value or trace capacity number in liquid phase
(TCNL). The TCNL and phenol value were strongly
correlated parameters. Similarly, ultraviolet absorbance
at 254 nm and specific ultraviolet absorbance at 254 nm
influenced performance and could be interchanged.
Surface area and iodine number did not appear to be
significant in the regression analysis.
Keywords: iodine number, phenol value, tannin value, trace capacity number
Powdered activated carbon (PAC) is routinely
applied to drinking water treatment systems and is the
most widely accepted control technology for the
removal of taste- and odor-causing compounds. This
technology is increasingly being used to remove trace
concentrations of other organic compounds, such as
endocrine disrupting compounds.
Although PAC is effective at removing trace organic
contaminants, it is a heterogeneous material whose characterization is complicated by diverse source materials
and activation techniques. Combinations of virgin material and activation techniques result in many products
with varying surface areas, pore size distributions, and
surface functionalities that ultimately affect adsorption
performance. Adsorption performance is further complicated by dynamic interactions between the PAC, water
chemistry, trace organic contaminants, and competing
background organic compounds (Tennant & Mazyck
2007, Newcombe et al. 2002, Newcombe & Drikas 1997,
Summers & Roberts 1988).
With so many commercial PACs, many treatment
operators select a carbon on the basis of conventional
manufacturers’ specifications, primarily iodine number
(Macleod & Simpson 1993), which is a standard
E50
procedure developed as a surrogate measure for surface
area. Because of the complex nature of adsorption phenomena, however, surface area and iodine number do not
correlate with adsorption performance (Zhang 2008,
Macleod & Simpson 1993). Therefore, selecting a PAC
based on iodine number may result in inadequate performance and increased doses and correspondingly increased
costs to meet target removal goals. There are many other
indexes (e.g., tannin, phenol, molasses) that have been
developed over the years to find a surrogate method that
correlates with performance—but how useful are they?
Despite scientific knowledge of adsorption mechanisms (Huang et al. 2011, Tennant & Mazyck 2007,
Moreno-Castilla 2004, Newcombe et al. 2002, Considine et
al. 2001, Gillogly et al. 1998, Newcombe & Drikas 1997,
Pendleton et al. 1997, Summers & Roberts 1988, Coughlin
& Ezra 1968), PAC selection may be subjected to long
and costly procedures that use jar test simulations to
mimic a full-scale plant use and attempt to model PAC
performance for a particular application. These modeling
efforts have been tested on atrazine, methylisoborneol
(MIB), and other trace organic compounds in which PAC
performance can be modeled quite accurately (Cook et
al. 2001, Gillogly et al. 1998, Huang et al. 1996, Greene
VAL CARCE E T AL . | M A R C H 2 0 1 7 • 1 0 9 :3 | J O U R N A L AW WA
2017 © American Water Works Association
et al. 1994). Setbacks to these methods are that new isotherm data have to be developed every time water characteristics change, for every PAC under consideration and
for each contaminant of concern. Water utilities are limited by the lack of a holistic approach that combines
scientific understanding into a simple method for PAC
selection and performance predictions.
PAC adsorption of trace organic compounds may
depend on several factors acting simultaneously, such as
the availability of high-energy-adsorption sites, functional
groups of the adsorbent and adsorbate, PAC pore size and
volume distributions, molecular size and contaminant
structure, background organic competition, and solution
chemistry interactions (Ding et al. 2008, Quinlivan et al.
2005, Li et al. 2002, Newcombe et al. 2002, Considine et
al. 2001, Franz et al. 2000, Pelekani & Snoeyink 1999,
Newcombe & Drikas 1997, Newcombe et al. 1997,
Pendleton et al. 1997, Summers & Roberts 1988).
Because these factors, among others, may be contributing simultaneously to adsorption, a more comprehensive
algorithm with multiple variables may correlate better
with performance rather than attempting to correlate
and predict performance with single, individual parameters such as iodine number.
This research aimed to use multivariable linear regression (MLR) analysis to assess whether a combination of
common PAC specifications and water quality parameters
could be used as a tool for PAC selection. Macleod and
Simpson (1993) did similar work in which different activated carbon indexes were correlated to MIB and geosmin removal. The shortcoming of that study was that a
single carbon parameter on its own was not able to correlate sufficiently with performance to provide a reliable
performance metric; tannin value—a surrogate measure
for mesopore volume—attained the highest correlation
for Macleod and Simpson (1993). Moreover, it did not
study the effect of water parameters. The research here
differed in its goal of being more comprehensive by analyzing the usefulness of multiple regression variables. The
resulting regression models can be used directly by water
utilities for selecting a PAC for their application on the
basis of measurable PAC characteristics (which may be
reported by PAC manufacturers). PAC characteristics
analyzed in the current study through multivariable
regression modeling included iodine number, phenol
value, tannin value, trace capacity number (TCN) in
liquid phase (TCNL), the pH at the point of zero charge
(pHpzc), moisture, surface area, average pore volume,
average pore size, and average mesopore volume. Water
characteristics analyzed included dissolved organic carbon (DOC), ultraviolet absorbance at 254 nm (UV254),
specific ultraviolet absorbance at 254 nm (SUVA254), pH,
total dissolved solids (TDS), and hardness.
Notwithstanding the likelihood of scrupulous regulations surrounding endocrine disrupting compounds in the
future and continual decreases in water quality making
taste and odor issues more frequent, such a tool has many
valuable advantages. An easy and economical method of
selecting an effective PAC for removing trace organic
compounds can improve PAC efficiency and thus conserve resources, minimize waste sludge, reduce operating
costs, and provide superior water quality.
MATERIALS AND METHODS
PAC and water characteristics. Commercial PACs were
sieved as received through a number 325 sieve (45 µm)
(ASTM 2009); what was retained in the 400 sieve (37 µm)
(ASTM 2009) was subsequently used for standard characterization procedures and adsorption experiments.
This maintained a constant average particle size for all
the PACs, minimizing the effects of particle size on
adsorption experiments (Crittenden et al. 1991, Lee &
Snoeyink 1980). PAC samples were stored in glass jars
and kept in vacuum-sealed desiccators. Table 1 lists the
PAC product name, manufacturer, raw carbon material,
and the corresponding identification letters assigned for
this study. PACs in Table 1 were thermally activated
unless otherwise specified.
A set of five blended PACs was made to increase the
sample size of PACs used for the regression of PAC
characteristics. Blending occurred after sieving the commercial PACs and was made by mixing 50% by weight
of two commercial PACs and shaking the mixture in a
glass jar manually for 1 min. The blended PACs are
specified in Table 1.
PAC characterizations following standard procedures
included the iodine number (ASTM 2011), tannin value
(AWWA 2010), and phenol value (AWWA 1978). The
TCNL used acetoxime as the adsorbate. The TCNL procedure can be found in AWWA Standard B604-12 (2012) as
the acetoxime number test; the two terminologies are synonymous. AWWA B604-12, Granular Activated Carbon
(AWWA 2012), uses the term “trace capacity” for the
carbon tetrafluoride number test. Following the terminology used by Zhang (2008), the carbon tetrafluoride number is synonymous with the TCN in the gas phase. The
two tests, TCNL and TCN in the gas phase, are correlated
to each other (Zhang 2008); therefore, either one can be
used to determine the TCN for the activated carbon as
long as the method (gas or liquid) is reported.
Brunauer–Emmett–Teller surface area, average pore size,
average pore volume, and average mesopore volume for
PAC samples were analyzed using nitrogen adsorption with
a high-speed surface area analyzer.1 The carbon pHpzc was
determined after mixing 2 g of carbon with 80 mL of
deionized water (DI) for 30 min.
Natural raw water samples were obtained from ground
and surface waters of the Midwest and southeastern
United States. Table 2 lists the water source and corresponding identification details assigned for this study.
Filtered water samples were analyzed for DOC, TDS,
SUVA254, UV254, pH, total hardness, and calcium hardness.
VA LC A R C E ET A L. | M A R C H 2017 • 109: 3 | JO U R NA L AWWA
2017 © American Water Works Association
E51
TABLE 1
Description of PAC products used in the current study
Carbon
Identification
Product Name
Manufacturer
Raw Carbon
Material
G
Aqua Nuchara
MeadWestvaco
Wood
H
Hydrodarco Exp 385
Norit
Lignite
I
Hydrodarco W
Norit
Lignite
J
Norit PAC 20 BF
Norit
Bituminous
K
Hydrocarbo B
Norit
Lignite
N
WPH-M
Calgon Carbon
Bituminous
P
CASPa
Norit
Wood
Q
WPH
Calgon Carbon
Bituminous
GJ
50/50 blend of Aqua Nuchar and Norit PAC 20 BF
GN
50/50 blend of Aqua Nuchar and WPH-M
GH
50/50 blend of Aqua Nuchar and Hydrodarco Exp 385
HN
50/50 blend of Hydrodarco Exp 385 and WPH-M
JN
50/50 blend of Norit PAC 20 BF and WPH-M
PAC—powdered activated carbon
aChemically
activated powdered activated carbon
DOC was analyzed using a solid sample combustion unit.2
A spectrophotometer3 was used to measure UV254 absorbance. SUVA254 was calculated as (UV254) × 100/DOC.
TDS and pH were measured using a pH/conductivity
meter.4 Total hardness and calcium hardness were determined using a hardness test kit5 with an ethylenediaminetetraacetic acid drop count titration method.
PAC adsorption experiments. Water samples spiked with
an initial 100 ng/L of C14-labeled MIB (or atrazine)6
were measured into a 40-mL vial with no headspace.
Concentrations of MIB or atrazine before spiking with
radiolabeled stock solutions were not measured and
assumed not to affect results because Gillogly et al.
(1998) found percent removal to be independent of
TABLE 2
Description of raw water sample
sources used in the current study
Water
Identification
E52
Source
A
Tampa Bay Regional Surface Water
Treatment Plant, Tampa, Fla.
B
Gainesville Regional Utilities groundwater,
Gainesville, Fla.
C
Greater Cincinnati Water Works
groundwater, Cincinnati, Ohio
D
Greater Cincinnati Water Works surface
water, Cincinnati, Ohio
E
Deionized water, University of Florida,
Gainesville, Fla.
initial concentration in the parts-per-trillion concentration range. The PAC was added to the vials in a DI slurry
form at various doses (5, 20, and 50 mg/L). Contact
times of 5, 30, and 240 min (4 h) were chosen to
approximate realistic point of contact before rapid mix,
during flocculation, and during sedimentation, respectively, but adsorption tests were carried out without
adding other treatment chemicals or making other modifications. The pH of the samples was not adjusted, and
the addition of PAC at these typical doses was not
observed to significantly alter the pH of the samples.
After the predetermined contact time, samples were
vacuum filtered using 0.45-μm filter paper in which 9 mL
of the filtrate was mixed with 11 mL of a scintillation
cocktail7 in a polyethylene scintillation vial. Effluent MIB
and atrazine concentrations were then analyzed using a
liquid scintillation analyzer8 that counted the counts per
minute. The counts per minute were converted to nanograms per liter of MIB (55 mCi/mmol) and atrazine
(160 mCi/mmol) using a standard curve. Blank samples
of all water were analyzed to verify the initial concentration and calculate percent removal of trace contaminant.
Raw data can be found in the University of Florida
I n s t i tutional Repository (http://ufdc.ufl.edu/
IR00004128/00001). All experiments were performed in
duplicate.
Multivariable regression analysis and validation. MLR
analysis was conducted using R-program v.2.13.1 (2011)
statistical software. The following four regression-model
selection criteria were used to determine the regression
parameters of PACs that correlated with the removal of
MIB or atrazine.
VAL CARCE E T AL . | M A R C H 2 0 1 7 • 1 0 9 :3 | J O U R N A L AW WA
2017 © American Water Works Association
•• Subset regression search using the leaps package
(Lumley 2009) for an exhaustive search with all possible variables to find the model with the highest R2
in each subset
•• Significance test on regression coefficients with P < 0.05
•• Subsets that optimized R2-adjusted and minimum
mean square error (MSE)
•• Absence of multicollinearity by calculating the variance inflation factor with the statistical software
package “car” (Fox & Weisburg 2011)
The mean prediction sum of squares (MSPRESS) is a
parameter used to validate the model with the given data
already incorporated in the model. Values close to the MSE
suggest the model should continue to make predictions of
future outcomes with reasonable variability (Kutner et al.
2004). The MSPRESS was calculated using the “boot”
package of the statistical software (Canty & Ripley 2011)
and used methods from Davison and Hinkley (1997), such
as the delete-one-at-a-time technique.
To further determine the robustness of the models
selected and test the viability of the algorithm as a method
for selecting PACs and predicting trace organic removal,
a set of data was reserved for validation. Similar to the
MSPRESS, the mean squared prediction errors (MSPR)
of new data should be similar to the MSE for the model
to prove robust (Kutner et al. 2004).
A sensitivity analysis using standardized coefficients was
performed using the statistical software’s “QuantPsyc”
package (Fletcher 2010). The magnitude of the regression
coefficients is not representative of the dependent variable’s
sensitivity to an individual regression variable because
regression models are empirical and parameters can have
different units and varying magnitudes. The standardized
coefficients can be used to determine the order of influence
that the regression parameters have on the dependent variable by standardizing the values of the data set through the
standard deviation of the variable and thus creating a
unitless regression model.
RESULTS AND DISCUSSION
Ten PAC characteristics and seven water characteristics
were chosen for this study to examine their correlation as
independent variables to adsorption performance using
MLR analysis. In addition, contact time and carbon dose
were independent variables in the MLR analysis. Contact
time and carbon dose were parameters relating to adsorption kinetics and PAC system designs. Including these
parameters met the objective of developing algorithms that
were practical and easy to use in water treatment facilities.
Regression analysis. The following four models (M1–
M4) combined the PAC and water characteristic variables
identified by the multilinear regression model selection
criteria to significantly influence the performance of PAC
for the removal of MIB and atrazine from natural water.
Carbon dose and contact time tend to have a logarithmic
relationship with percent removal; therefore, the data were
log transformed with respect to time and dose to linearize
the relationship with percent removal.
% removal = b0 + b1log(dose) + b2log(time)
– b3log(UV254) – b4log(tannin) + b5phenol
(M1)
% removal = b0 + b1log(dose) + b2log(time) (M2)
– b3log(UV254) – b4log(tannin) + b5TCNL
% removal = b0 + b1log(dose) + b2log(time) (M3)
– b3log(SUVA254) – b4log(tannin) + b5TCNL
% removal = b0 + b1log(dose) + b2log(time) (M4)
– b3log(SUVA254) – b4log(tannin) + b5phenol
where b is the regression coefficient, b1 is the regression
coefficient for the first variable, and so on.
Although the model selection process was performed
separately for each contaminant, the results are in
agreement with each other. In other words, the same
variables, without sacrificing too many degrees of freedom, explain 84–85% of the variability in the performance data for both trace organic contaminants. This
is significant because it suggests that, for physical
adsorption, performance may be explained independently of adsorption mechanisms.
Following the model selection criteria, these models
showed 95% confidence (P < 0.05) that their regression
coefficients were not equal to zero and that the variance
inflation factor of approximately 1 indicated no multicollinearity among the independent variables. The models used a standard deviation (square root of MSE) of
±11.6 – 13% removal. The values for the regression
coefficients and statistical analysis are summarized in
Table 3 for atrazine and Table 4 for MIB. The regression
analysis excluded data collected with water B1 so that
it could be used to validate the models by calculating
the MSPR of predicted values versus experimental values. Models 3 and 4 for atrazine using the SUVA254 had
higher MSPRs than the models with UV254 as a result
of models 1 and 2 overpredicting the performance of
PACs at a 50-mg/L dose and 4-h contact time. It is
important to remain cautious when interpreting prediction results of regression models on the basis of continuous numbers with no boundary because the empirical equation does not recognize the physical limitation
of 100% removal. As a result of atrazine’s high adsorption affinity, erroneous predictions higher than 100%
may occur for atrazine in comparison with similar models for MIB removal, which has a lower affinity for PAC
and therefore a slower rate of adsorption. Nonetheless,
the MSPR of most of the models was close to the MSE,
suggesting their validity in predicting performance.
Because regression models are empirical, they depend on
a set of specified units (refer to Tables 5 and 6 for the corresponding units specific to each parameter in models 1–4).
VA LC A R C E ET A L. | M A R C H 2017 • 109: 3 | JO U R NA L AWWA
2017 © American Water Works Association
E53
As such, the magnitudes of the regression coefficients cannot be used to determine which parameters have more
influence on percent removal. A sensitivity analysis using
standardized regression coefficients (unitless) reveals which
variables have a greater effect than others in determining
percent removal of the contaminant. Table 7 for atrazine
and Table 8 for MIB tabulate the standardized coefficients
for the regression models. The results indicate that PAC
dose had the greatest effect on percent removal for both
TABLE 3
Regression coefficients and statistical
summary for the removal of atrazine
Parameter
Model 1
(Phenol
and UV)
Model 2
(TCNL
and UV)
Model 3
(TCNL
and
SUVA)
Model 4
(Phenol
and
SUVA)
b0
10.3
23.6
54.7
41.2
b1
21.1
21.1
21.1
21.1
b2
48.1
47.9
48.3
48.4
b3
12.2
12.4
26.3
25.8
b4
28.0
30.5
31.8
29.3
b5
3.67
1.52
1.51
3.6
b2
0.8518
0.8517
0.8497
0.8489
MSE
135
135
137
138
MSPRESS
137
137
139
139
MSPR
140
142
213
196
b0—regression coefficient, MSE—mean square error, MSPR—mean
squared prediction error, MSPRESS—mean prediction sum of squares,
SUVA—specific ultraviolet absorbance, TCNL—trace capacity number in
liquid phase, UV—ultraviolet
TABLE 4
Regression coefficients and statistical
summary for the removal of MIB
Parameter
Model 1
(Phenol
and UV)
Model 2
(TCNL
and UV)
Model 3
(TCNL
and SUVA)
Model 4
(Phenol
and
SUVA)
b0
1.31
14.0
48.5
36.7
b1
49.9
49.7
50.0
50.1
b2
16.8
16.8
16.7
16.7
b3
15.6
15.4
29.4
29.4
b4
30.5
33.0
33.6
31.0
b5
3.02
1.29
1.25
2.84
b2
0.8542
0.8531
0.8416
0.8409
MSE
154
155
167
168
MSPRESS
155
156
168
169
MSPR
132
139
204
198
b0—regression coefficient, MIB—methylisoborneol, MSE—mean square
error, MSPR—mean squared prediction error, MSPRESS—mean prediction
sum of squares, SUVA—specific ultraviolet absorbance, TCNL—trace
capacity number in liquid phase, UV—ultraviolet
E54
MIB and atrazine. On the other hand, the PAC’s phenol
and TCNL values had the least effect of the model variables on percent removal for both trace contaminants. This
indicates that selecting a PAC with a low tannin value is
more important than selecting a PAC with a high phenol
or TCNL value. However, because of the leverage that
carbon N holds on the regression slope of the tannin value,
future research with other PACs would have to determine
the accuracy of this sensitivity analysis.
Because the rate of adsorption for MIB is less than that
of atrazine, providing the longest possible contact time,
especially in high UV or SUVA waters, is recommended.
From an operational perspective, the high influence that
PAC dose exerts on the regression outcome is advantageous. Treatment operators may have limited choices for
point of application and therefore treatment contact time,
but PAC dosing is flexible and can ensure that percent
removal goals are reached.
PAC characteristics. PAC characteristics and their values
for each sample are shown in Table 5. Studying the values
in Table 5, it becomes evident that the tannin value for
carbon N (4,800 mg/L) was extreme in comparison with
the rest of the samples: the tannin value ranged from 160
to 600 mg/L. The extreme tannin value of carbon N is the
only PAC characteristic in Table 5 that can explain the
similarly extreme variability between the adsorption
performance of carbon N and the rest of the PAC samples,
thus becoming a significant variable in the MLR model.
Furthermore, as shown in Figure 1, part A, carbon N also
has a great deal of leverage in determining the regression
coefficient (otherwise known as the slope of the regression variable) of the tannin value variable in the MLR
model. It is thus imperative to explore the significant
leverage that carbon N carries to determine whether there
is sufficient reason to remove it as an outlying data set.
Carbon N is a PAC product9 “specifically designed for
use in the treatment of potable water” (Calgon Carbon
Corporation 2011). In addition, its intended application
is to remove taste and odor compounds such as MIB,
pesticides such as atrazine, and other refractory organic
compounds (Calgon Carbon Corporation 2011). Unfortunately, the only quality control test reported by the
company for this product is iodine number (verified at
500 mg/g; Calgon Carbon Corporation 2011), so it was
not possible to verify the tannin value beyond triplicate
measures in the laboratory.
Although carbon N was an outlying subset of this
study, there was not sufficient reason to exclude it; it is a
representative sample of commercial PACs used for trace
organic compound removal in water treatment. Furthermore, supporting literature confirms the negative correlation between tannin value and MIB removal for a range
of other PAC products with tannin values ranging from
130 to 1417 mg/L (Macleod & Simpson 1993). Macleod
and Simpson (1993) investigated the effects of common
PAC indexes such as moisture, ash content, density, pH,
VAL CARCE E T AL . | M A R C H 2 0 1 7 • 1 0 9 :3 | J O U R N A L AW WA
2017 © American Water Works Association
PAC characteristics
TABLE 5
Moisture
%
Surface
Area
m2/g
Average
Pore Size
Å
Average
Pore Volume
cm3/g
Average
Mesopore
Volume
cm3/g
7.10
6.61
1,550
30
1.17
0.35
15
11.10
5.59
730
36
0.66
0.33
10
11.30
7.10
540
41
0.55
0.32
9.5
17
8.30
9.70
950
23
0.56
0.15
350
5.5
10
11.70
5.10
490
43
0.52
0.30
4,800
7.4
15
8.10
9.08
560
19
0.27
0.04
Carbon
Identification
Iodine
Number
mg/g
Tannin
Value
mg/L
Phenol
Value
%
TCNL
mg/g
pHpzc
su
G
1,100
260
3.0
1.2
H
700
260
8.0
I
530
370
5.6
J
950
310
K
540
N
500
P
880
160
1.9
0.02
3.20
12.80
1,520
33
1.24
0.55
Q
980
310
9.5
15
7.90
9.72
970
23
0.55
0.13
GJ
1,000
220
6.4
7.5
7.50
8.16
1,250
28
0.88
0.26
GN
800
250
5.2
6.9
7.30
7.85
1,200
28
0.84
0.23
GH
900
170
5.3
8.8
9.10
6.10
1,190
37
0.92
0.34
HN
600
400
7.4
14
10.30
7.34
630
28
0.44
0.17
JN
710
600
8.4
15
7.00
9.39
750
23
0.42
0.10
PAC—powdered activated carbon, pHpzc—pH at the point of zero charge, su—standard unit, TCNL—trace capacity number in liquid phase
Water characteristics
TABLE 6
Water
Identification
DOC
mg/L
UV254
Abs
SUVA254
L/mg/m
pH
su
TDS
mg/L
Total Hardness
mg/L CaCO3
Calcium Hardness
mg/L CaCO3
174
A
6.78
0.168
2.47
7.65
305
220
B1a
2.55
0.025
0.98
7.89
251
261
NA
B2a
1.05
0.023
2.19
7.65
227
407
280
B3a
1.01
0.012
1.19
7.74
228
180
120
B4a
0.78
0.011
1.41
7.51
227
380
313
B5a
0.77
0.011
1.43
7.47
115
185
120
C
0.96
0.017
1.73
7.46
277
NA
200
D
1.42
0.080
5.60
7.56
221
233
160
E
0.20
0.001
0.49
6.78
10
10
0
Abs—absorbance, CaCO3—calcium carbonate, DOC—dissolved organic carbon, NA—not applicable, su—standard unit, SUVA254—specific ultraviolet absorbance
at 254 nm, TDS—total dissolved solids, UV254—ultraviolet absorbance at 254 nm
aWater
B samples were collected on several occasions.
TABLE 7
Sensitivity analysis for atrazine
regression models
TABLE 8
Sensitivity analysis for MIB
regression models
Standardized
Coefficient
Model 1
Model 2
Model 3
Model 4
Standardized
Coefficient
Model 1
Model 2
Model 3
Model 4
b1
0.66
0.65
0.66
0.66
b1
0.63
0.63
0.63
0.64
0.36
0.36
0.35
0.35
b2
0.48
0.48
0.48
0.48
b2
b3
0.37
0.37
0.37
0.36
b3
0.38
0.38
0.36
0.36
0.44
b4
0.41
0.45
0.45
0.42
0.26
b5
0.20
0.21
0.20
0.19
b4
b5
0.42
0.26
0.46
0.27
0.48
0.27
MIB—methylisoborneol
VA LC A R C E ET A L. | M A R C H 2017 • 109: 3 | JO U R NA L AWWA
2017 © American Water Works Association
E55
FIGURE 1
MIB removal with PAC as a function of
tannin value
A Average percent removal of MIB in water E with
MIB Removal—%
5 mg/L PAC dose and 30 min contact time,
showing that carbon N has a great deal of
leverage to determine slope of regression line
100
90
80
70
60
50
40
30
20
10
0
Carbon N
100
1,000
Tannin Value—mg/L
10,000
B Macleod and Simpson (1993) DI experiments at
MIB Removal—%
5 mg/L PAC dose and approximately 30 min contact
time superimposed
Water E
Macleod and Simpson
100
90
80
70
60
50
40
30
20
10
0
100
1,000
Tannin Value—mg/L
10,000
DI—deionized water, MIB—methylisoborneol, PAC—powdered
activated carbon
FIGURE 2
Correlation between average mesopore
volume of PAC and corresponding
tannin value
Tannin Value—mg/L
10,000
1,000
R2 = 0.759
100
10
1
0.01
0.1
Average Mesopore Volume—cm3/g
1
PAC—powdered activated carbon
E56
molasses number, iodine number, surface area, phenol
number, and tannin value. The parameter in their study
that correlated best with MIB and geosmin removal in DI
and surface water from the Hillsborough River (water A;
Tampa, Fla.) was the tannin value. The example in Figure 1,
part B, superimposes data from Macleod and Simpson
(1993) with data from this study to suggest a logarithmic
relationship between MIB removal and tannin value.
Similar correlation existed at other contact times and PAC
dose with water E as well as water A and by Macleod and
Simpson (1993). The results of the MLR study confirmed
the conclusions of Macleod and Simpson (1993) that a
negative correlation existed with PAC adsorption performance and the PAC’s tannin value.
Tannin value is a PAC characteristic used as a surrogate
parameter for mesoporosity. Figure 2 shows tannin value
as a function of the average mesopore volume with both
variables on a log scale to linearize the power function
relationship. Because the tannin value seemed to have a
logarithmic relationship to both MIB removal and mesoporosity, the log transformation was used in the MLR
analysis. The smaller the tannin value, the more mesopores the activated carbon had because it could remove
more tannic acid with less carbon. In agreement with
Macleod and Simpson (1993), the results of MLR models
advocate using PACs with high mesoporosity. This is also
recommended by other researchers, who suggest that
large pores will prevent pore blockage by competing
background organic compounds and allow the MIB to
diffuse into smaller pores (Tennant & Mazyck 2007,
Quinlivan et al. 2005). Tannin value was also significant
for water E despite it being DI water with low background organic compounds. Therefore, even in the
absence of larger competing compounds, MIB and atrazine adsorption performance improved with PACs of low
tannin values and larger mesopore volumes. Macleod and
Simpson’s (1993) experiments in DI also confirm these
results. Larger mesopore volumes allow improved surface
diffusion kinetics and can be a contributing mechanism
for adsorption in organic free waters.
The addition of phenol value (models 1 and 4) and
TCNL (models 2 and 3) suggested that something other
than the physical nature of pore size distributions may
contribute to variability in PAC performance. As such, the
variability in performance between PACs of similar tannin
values in Figure 1, part B, may be attributed to differences
in enthalpy, energy of adsorption sites, and/or differences
in surface functionalities. For instance, Pendleton et al.
(1997) and Considine et al. (2001) explained the phenomenon in which oxygen functional groups created hydrophilic sites where water will preferentially adsorb. Their
enthalpy of displacement studies concluded that the small
enthalpy values are not strong enough to replace water
with MIB. Moreover, because of hydrogen bonding and
the polar nature of water, the water clusters formed at the
edge of carbon pores created a pore blockage effect that
VAL CARCE E T AL . | M A R C H 2 0 1 7 • 1 0 9 :3 | J O U R N A L AW WA
2017 © American Water Works Association
The models had similar results with the models that had
UV254 being slightly favored over the SUVA254 models.
UV254 is correlated to aromaticity and SUVA254 is the
fraction of DOC that is aromatic. It is possible for two
waters to have the same SUVA254 value but behave differently because the UV254 of the two waters may be
different. The results of the regression models imply that
UV254 is a good indicator of background competition
because of its relationship with molecular weight and
aromatic character of natural organic matter (NOM).
SUVA254, representative of the aromatic fraction of DOC,
is also a good indicator of background competition.
Figure 5, parts A and B, show MIB and atrazine removal,
respectively, as a function of UV254. Research studies have
shown that UV254 correlates with molecular weight and
aromaticity of NOM in solution (Uyguner & Bekbolet
2005, Chin et al. 1997, Peuravuori & Pihlaha 1997). The
low UV 254 absorbance values of the groundwater
samples in Figure 5 indicate waters with a lower concentration of aromatic NOM than the surface water
samples. The efficiency for trace organic removal of
MIB and atrazine diminished with increasing aromatic
Plot of phenol value as a function of
carbon pH
Phenol Value—%
FIGURE 3
10
9
8
7
6
5
4
3
2
1
0
0
2
4
6
8
10
12
14
Carbon pH—su
su—standard unit
FIGURE 4
TCNL—mg/g
prevented the physical migration and adsorption of MIB
into smaller pores (Tennant & Mazyck 2007). The water
cluster effect has also been linked to phenol adsorption
trends whereby less phenol was adsorbed in the more
acidic carbons (with fewer oxygen-functional groups)
(Franz et al. 2000, Coughlin & Ezra 1968). In relation to
the phenol value parameter, this would mean that higher
phenol values represented more basic carbons, which is
weakly suggested in Figure 3 with the PACs of the current
study. However, a larger set of PAC samples is necessary
to ascertain the significance of this relationship between
phenol value and carbon pHpzc.
On the other hand, there was a strong correlation
between the phenol value and the TCNL (Figure 4). The
TCNL may be a surrogate measure for high-energyadsorption sites. Its correlation to PAC performance of
MIB removal is supported by the work of Zhang (2008),
who investigated the viability of using TCN as an indicator of activated carbon performance to remove MIB
and geosmin. The TCN is an index developed to measure PAC’s ability to remove low-molecular-weight
organic compounds at low concentrations. Some carbon
manufacturers speculate that it is a better indicator of
performance than iodine number because, rather than a
general indication of surface area, the TCN indicates
high-energy-adsorption sites necessary for adsorption.
Zhang’s (2008) results support the use of the TCN
because there was a very strong correlation with removal
(R2 = 0.8236 for MIB).
For decades, countless surrogate methods and adsorbates have been studied to find a one-size-fits-all solution:
one index that measures both the available adsorption
sites and the energy or activity level of these PAC sites.
Unfortunately, before being fully vetted, many of these
methods became standard procedures that were accepted
and flaunted across the industry for removing a plethora
of contaminants in water and air. These indexes were
established under ideal water conditions (i.e., high concentrations of adsorbate and organic free water) and
therefore do not encompass the entire complexity of the
system to which they were applied.
This study attempted to thoroughly vet the applicability
of these indexes for realistic conditions (trace concentration of contaminants and a range of background matrixes)
to redirect the industry’s focus on those characteristics or
combination of characteristics that show promise.
The results show that iodine number, a well-established
and accepted index, cannot be a substitute for accurate
performance data. The lack of an iodine number in the
regression model analysis questions the efficacy of using
it as a common parameter for selecting PACs in trace
contaminant applications. Rather, the results support the
use of a combination of at least two parameters: tannin
value and either phenol or TCNL.
Water characteristics. The water characteristics and
their values for each water sample are shown in Table 6.
Correlation between TCNL
and phenol value
20
18
16
14
12
10
8
6
4
2
0
R2 = 0.9132
0
2
4
6
Phenol Value—%
8
10
TCNL—trace capacity number in liquid phase
VA LC A R C E ET A L. | M A R C H 2017 • 109: 3 | JO U R NA L AWWA
2017 © American Water Works Association
E57
concentration because there was a higher affinity for
adsorption of these aromatic compounds present in
larger concentrations than the trace organic compounds. Mechanisms for the reduced capacity of trace
organic compounds in natural waters as compared with
DI may be explained by direct site and pore blockage
competition of the NOM (Ding et al. 2008, Li et al.
2003, Newcombe et al. 2002, Pelekani & Snoeyink
1999, Newcombe et al. 1997).
Other water quality parameters, such as calcium hardness and DOC concentration, were not significant parameters in the regression models of natural waters. Although
waters B4 and B5 had similar UV254 levels, DOC concentrations, and SUVA254 values, the differences in TDS and
hardness concentrations had no significant effect on
adsorption performance. Because DOC concentrations
were 1 billion times greater in magnitude than the
concentration of the trace organic, the small variations
in DOC concentrations among the water samples had
no statistically significant effect. Rather, the aromatic
FIGURE 5
MIB removal with PAC as a function of
UV254 absorbance
DI
Groundwaters
Surface waters
MIB Removal—%
A Average percent removal of MIB treated with 5 mg/L of
carbon J for 4 h as a function of the initial UV254
100
90
80
70
60
50
40
30
20
10
0
0
0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18
UV254
B Average percent removal of atrazine treated with
20 mg/L of carbon J for 5 min as a function of the
initial UV254
Atrazine Removal—%
100
CONCLUSIONS
Tannin value was found to be a useful surrogate measure for mesopores; the conclusions of this study further
support that it may be useful for PAC selection and performance predictions. Lower tannin values indicate more
mesopores and better performance for the removal of
trace organic contaminants from natural waters. This
suggests it may improve bulk diffusion mechanisms
and/or minimize the detrimental effects of competing
background organics by reducing pore blockage effects.
However, because PAC adsorption is a complex phenomenon, it is unreasonable to assume that tannin value alone
can be applied universally. This research indicates that a
second PAC parameter is necessary: the phenol value or
TCNL value (found to correlate to each other). In addition, UV254 or SUVA254 may be used as a simple measure
of background effect to performance.
In summary, the trends elucidated from this regression
research can better guide treatment operators to select
appropriate PACs for their application. The regression
models (M1, M2, M3, and M4) may be used directly by
water utilities by simply plugging in known values for
the model variables to compare expected percent
removal performance for various PACs under consideration. Furthermore, despite structural and chemical
reactivity differences, the result of the multivariable
regression analysis identified the same PAC parameters
influencing the removal of MIB and atrazine, thus providing stronger evidence to move away from traditional
parameters such as iodine number and focus instead on
tannin and phenol/TCNL.
ACKNOWLEDGMENT
90
80
70
60
50
40
30
0
0.2
0.4
0.6
0.8
Initial UV254
1.0
1.2
DI—deionized water, MIB—methylisoborneol, PAC—powdered
activated carbon, UV254—ultraviolet absorbance at 254 nm
E58
concentration of the DOC will have a much greater
effect; therefore, the influence of NOM/DOC on adsorption is a product of the concentration of aromatic components. Likewise, although calcium is known to affect
the adsorption of NOM by way of charge neutralization
and calcium complexation (Lee & Snoeyink 1980), Bose
and Reckhow (1997) demonstrated that incremental
changes to the specific organic charge do level off above
calcium concentrations of 200 mg/L (as calcium carbonate), which was the average calcium hardness concentration for the natural waters in this study.
This material is based on work supported by the National
Science Foundation Graduate Research Fellowship
under Grant No. DGE-0802270. Special thanks go to
Gainesville Regional Utilities, Manatee County Water
Treatment Plant, Greater Cincinnati Water Works, Calgon
Carbon Corporation, Norit, and Mead WestVaco for
supporting the goals of this research by donating water
and PAC samples. Also, the authors thank Rick Loftis at
Engineering Performance Solutions for his generosity in
laboratory space and analytical equipment as well as his
support, encouragement, and advice.
VAL CARCE E T AL . | M A R C H 2 0 1 7 • 1 0 9 :3 | J O U R N A L AW WA
2017 © American Water Works Association
ABOUT THE AUTHORS
Christine O. Valcarce (to whom
correspondence may be addressed) is
R&D director at Carbonxt, 3951
NW 48th Terrace, Ste. 111,
Gainesville, FL 32066 USA;
[email protected]. This article
is part of her doctoral thesis. She
earned her bachelors, masters, and
doctoral degrees from the University of Florida,
Gainesville. Valcarce specializes in advanced activated
carbon research for air and water applications. Erica W.
Gonzaga is an applications engineer at Carbonxt.
David W. Mazyck is professor of environmental
engineering sciences, University of Florida, Gainesville.
https://dx.doi.org/10.5942/jawwa.2017.109.0012
ENDNOTES
1NOVA N2200e series surface area analyzer, Quantachrome Instruments,
Boynton Beach, Fla.
2TOC-5000A, solid sample combustion unit, Shimadzu, Tokyo, Japan
3DR/4000U Spectrophotometer, Hach, Loveland, Colo.
4Accumet Research AR20 pH/conductivity meter, Fisher Scientific,
Hampton, N.H.
5Model HA-4P hardness test kit, Hach, Loveland, Colo.
6American Radio Labeled Chemicals, St. Louis, Mo.
7Perkin Elmer, Waltham, Mass.
8Tri-Carb 2900TR Scintillation Analyzer, Packard Instruments, Mediden,
Conn.
9WPH-M, Calgon Carbon Corp., Moon Township, Penn.
PEER REVIEW
Weight and Aromaticity. Environmental Science & Technology,
33:6:1630. https://dx.doi.org/10.1021/es960404k.
Considine, R.; Denoyel, R.; Pendleton, P.; Schumann, R.; & Wong, S.,
2001. The Influence of Surface Chemistry on Activated Carbon
Adsorption of 2-methylisoborneol From Aqueous Solution. Colloids
and Surfaces: Physicochemical and Engineering Aspects. 179:2–
3:271. https://dx.doi.org/10.1016/S0927-7757(00)00647-6.
Cook, D.; Newcombe, G.; & Sztajnbok, P., 2001. The Application of
Powdered Activated Carbon for MIB and Geosmin Removal:
Predicting PAC Doses in Four Raw Waters. Water Research,
35:5:1325. https://dx.doi.org/10.1016/S0043-1354(00)00363-8.
Crittenden, J.C.; Reddy, P.S.; Arora, H.; Trynoski, J.; Hand, D.W.; Perram,
D.L.; & Summers, R.S., 1991. Predicting GAC Performance With Rapid
Small-Scale Column Tests. Journal AWWA, 83:1:77.
Coughlin, R.W. & Ezra, F.S., 1968. Role of Surface Acidity in the
Adsorption of Organic Pollutants on the Surface of Carbon.
Environmental Science & Technology, 2:4:291. https://dx.doi.
org/10.1021/es60016a002.
Davison, A.C. & Hinkley, D.V., 1997. Bootstrap Methods and Their
Applications. Cambridge University Press, Cambridge. https://dx.doi.
org/10.1017/CBO9780511802843.
Ding, L.; Snoeyink, V.L.; Mariñas, B.J.; Yue, Z.; & Economy, J., 2008.
Effects of Powdered Activated Carbon Pore Size Distribution on the
Competitive Adsorption of Aqueous Atrazine and Natural Organic
Matter. Environmental Science & Technology, 42:4:1227. https://dx.
doi.org/10.1021/es0710555.
Fletcher, T.D., 2010. QuantPsyc: Quantitative Psychology Tools. R
package version 1.4. https://cran.r-project.org/web/packages/
QuantPsyc/QuantPsyc.pdf (accessed Dec. 13, 2016).
Fox, J. & Weisburg, S., 2011 (2nd ed.). An {R} Companion to Applied
Regression. Sage: Thousand Oaks, Calif. http://socserv.socsci.
mcmaster.ca/jfox/Books/Companion (accessed Dec. 13, 2016).
Franz, M.; Arafat, H.A.; & Pinto, N.G., 2000. Effect of Chemical Surface
Heterogeneity on the Adsorption Mechanism of Dissolved Aromatics
on Activated Carbon. Carbon, 38:13:1807. http://dx.doi.org/10.1016/
S0008-6223(00)00012-9.
Date of submission: 04/25/2016
Date of acceptance: 10/11/2016
REFERENCES
ASTM, 2011. Standard Test Method for Determination of Iodine Number
of Activated Carbon. D4607-94. ASTM International, West
Conshohocken, Pa.
ASTM, 2009. Standard Specification for Wire Cloth and Sieves for
Testing Purposes. E11-09. ASTM International, West
Conshohocken, Pa.
AWWA, 2012. ANSI/AWWA Standard B604-12, Granular Activated
Carbon. AWWA, Denver.
AWWA, 2010. ANSI/AWWA Standard B600-10, Powdered Activated
Carbon. AWWA, Denver.
AWWA, 1978. ANSI/AWWA Standard B600-78, Powdered Activated
Carbon. AWWA, Denver.
Bose, P. & Reckhow, D.A., 1997. Modeling pH and Ionic Strength Effects
on Proton and Calcium Complexation of Fulvic Acid: A Tool for
Drinking Water-NOM Studies. Environmental Science & Technology,
31:3:765. https://dx.doi.org/10.1021/es9604469.
Calgon Carbon Corporation Product Bulletin, 2011. WPH-M Powdered
Activated Carbon. www.calgoncarbon.com/wp-content/uploads/
product-literature/WPH_M.pdf (accessed Dec. 13, 2016).
Canty, A. & Ripley, B., 2011. Boot: Bootstrap R (S-Plus) Functions. R
package version 1.3-2.
Chin, Y.; Aiken, G.R.; & Danielsen, K.M., 1997. Binding of Pyrene to
Aquatic and Commercial Humic Substances: The Role of Molecular
Gillogly, T.E.; Snoeyink, V.L.; Elarde, J.R.; Wilson, C.M.; & Royal, E.P., 1998.
14C-MIB Adsorption on PAC in Natural Water. Journal AWWA, 90:1:98.
Greene, B.E.; Snoeyink, V.L.; & Pogge, F.W., 1994. Adsorption of
Pesticides by Powdered Activated Carbon. AWWA Research
Foundation & AWWA, Denver.
Huang, W.; Lin, T.; & Chen, J., 2011. Kinetic Studies on Adsorption of
Odor-Causing Substances by Activated Carbon. Sustainable
Environment Research, 21:2:95.
Huang, C.; Van Benschoten, J.E.; & Jensen, J.N., 1996. Adsorption
Kinetics of MIB and Geosmin. Journal AWWA, 88:4:116.
Kutner, M.; Nachtsheim, C.; Neter, J.; & Li, W., 2004 (5th ed.). Applied
Linear Statistical Models. McGraw Hill, New York.
Lee, M.C. & Snoeyink, V.L., 1980. Removal of Organics by Adsorption in
the Purification of Potable Water and in the Treatment of Water for
Reuse. Research Report No. 153. University of Illinois Water
Resources Center, Urbana-Champaign, Ill. http://web.extension.
illinois.edu/iwrc/pdf/153.pdf (accessed Dec. 13, 2016).
Li, L.; Quinlivan, P.A.; & Knappe, D.R., 2002. Effects of Activated Carbon
Surface Chemistry and Pore Structure on the Adsorption of Organic
Contaminants From Aqueous Solution. Carbon, 40:12:2085.
https://dx.doi.org/10.1016/S0008-6223(02)00069-6.
Li, Q.; Snoeyink, V.L.; Mariaas, B.J.; & Campos, C., 2003. Elucidating
Competitive Adsorption Mechanisms of Atrazine and NOM Using
Model Compounds. Water Research, 37:4:773. https://dx.doi.
org/10.1016/S0043-1354(02)00390-1.
VA LC A R C E ET A L. | M A R C H 2017 • 109: 3 | JO U R NA L AWWA
2017 © American Water Works Association
E59
Lumley, T., 2009. Using Fortran Code by Alan Miller Leaps: Regression
Subset Selection. R package version 2.9. http://CRAN.R-project.org/
package=leaps (accessed Dec. 13, 2016).
Macleod, B. & Simpson, M., 1993. Relationships Between Powdered
Activated Carbon Performance for Geosmin and 2-Methylisoborneol
Removal and Common Physical Adsorption Indices. Proc. AWWA
1993 Water Quality Technology Conference, Miami, Fla.
Moreno-Castilla, C., 2004. Adsorption of Organic Molecules From
Aqueous Solutions on Carbon Materials. Carbon, 42:1:83.
https://dx.doi.org/10.1016/j.carbon.2003.09.022.
Newcombe, G.; Morrison, J.; Hepplewhite, C.; & Knappe, D., 2002.
Simultaneous Adsorption of MIB and NOM Onto Activated Carbon: II.
Competitive Effects. Carbon, 40:12:2147. https://dx.doi.org/10.1016/
S0008-6223(02)00098-2.
Peuravuori, J. & Pihlaja, K., 1997. Molecular Size Distribution and
Spectroscopic Properties of Aquatic Humic Substances. Analitica
Chimica Acta, 337:2:133. https://dx.doi.org/10.1016/S00032670(96)00412-6.
Quinlivan, P.A.; Li, L.; & Knappe, D.R., 2005. Effects of Activated
Carbon Characteristics on the Simultaneous Adsorption of
Aqueous Organic Micropollutants and Natural Organic Matter.
Water Research, 39:8:1663. https://dx.doi.org/10.1016/j.
watres.2005.01.029.
R Development Core Team, 2011. R: A Language and Environment
for Statistical Computing. R Foundation for Statistical Computing,
2.13.2, Vienna, Austria. www.R-project.org (accessed Dec. 13,
2016).
Newcombe, G. & Drikas, M., 1997. Adsorption of NOM Onto Activated
Carbon: Electrostatic and Non-Electrostatic Effects. Carbon,
35:9:1239. https://dx.doi.org/10.1016/S0008-6223(97)00078-X.
Summers, R.S. & Roberts, P.V., 1988. Activated Carbon Adsorption of
Humic Substances: II. Size Exclusion and Electrostatic Interactions.
Journal of Colloid Interface Science, 122:2:382. https://dx.doi.
org/10.1016/0021-9797(88)90373-6.
Newcombe, G.; Drikas, M.; & Hayes, R., 1997. Influence of Characterized
Natural Organic Material on Activated Carbon Adsorption: II. Effect on
Pore Volume Distribution and Adsorption of 2-Methylisoborneol. Water
Research, 31:5:1065. https://dx.doi.org/10.1016/S0043-1354(96)00325-9.
Tennant, M.F. & Mazyck, D.W., 2007. The Role of Surface Acidity and
Pore Size Distribution in the Adsorption of 2-Methylisoborneol via
Powdered Activated Carbon. Carbon, 45:4:858. https://dx.doi.
org/10.1016/j.carbon.2006.11.009.
Pelekani, C. & Snoeyink, V., 1999. Competitive Adsorption in Natural
Water: Role of Activated Carbon Pore Size. Water Research,
33:5:1209. https://dx.doi.org/10.1016/S0043-1354(98)00329-7.
Uyguner, C.S. & Bekbolet, M., 2005. Evaluation of Humic Acid
Photocatalytic Degradation by UV-Vis and Fluorescence
Spectroscopy. Catalysis Today, 101:3:267. https://dx.doi.org/10.1016/
j.cattod.2005.03.011.
Pendleton, P.; Wong, S.; Schumann, R.; Levay, G.; Denoyel, R.; &
Rouquero, J., 1997. Properties of Activated Carbon Controlling
2-Methylisoborneol Adsorption. Carbon, 35:8:1141. https://dx.doi.
org/10.1016/S0008-6223(97)00086-9.
E60
Zhang, X., 2008. Selecting Activated Carbon for Micropollutant Removal
in Drinking Water Treatment: Trace Capacity Number Test. Master’s
thesis, University of Toronto, Ont., Canada.
VAL CARCE E T AL . | M A R C H 2 0 1 7 • 1 0 9 :3 | J O U R N A L AW WA
2017 © American Water Works Association