Professional Documents
Culture Documents
Water Quality: and Disinfection Kinetics
Water Quality: and Disinfection Kinetics
Water quality
and disinfection kinetics
Water quality significantly affects
predictions of microbial inactivation.
A
Charles N. Haas, Josh Joffe,
Uma Anmangandla,
Joseph G. Jacangelo,
mendments to the Safe Drinking
and Mark Heath Water Act required that surface water suppliers in
the United States filter, disinfect, or both to protect
customer health. The Surface Water Treatment Rule
(SWTR) specified filtration and disinfection treat-
ment requirements for public water systems using
surface water sources or groundwater under the
direct influence of surface water. The US Environ-
mental Protection Agency (USEPA) has issued a guid-
ance manual1 to assist utilities in implementing SWTR
requirements.
Among the organisms capable of transmitting
waterborne diseases, Giardia lamblia has been classi-
fied as a significant pathogen, often causing severe
gastric difficulty. The USEPA Office of Drinking Water
has adopted the contact-time concept to quantify the
inactivation of G. lamblia
cysts by disinfection. The
The contact-time tables of the US Environmental Protection concept was applied to
Agency are based on survival data obtained in buffered demand- both filtered and nonfil-
free water. However, disinfection data from experiments tered systems to ensure
conducted at Drexel University showed that the choice of source adequate disinfection.
water influences the degree of inactivation. Sophisticated The SWTR required
disinfection kinetic models and statistical tests are used to that all surface water
analyze the sources of these discrepancies. This article treatment facilities provide
demonstrates that the quality of a particular source water is a adequate filtration and
significant factor in predicting disinfection effectiveness. disinfection in order to
achieve (1) a 99.9 percent
Stock monochloramine
solution. Preformed mono-
chloramine was prepared TABLE 2 Water quality characteristics of experimental waters
daily as needed by mixing
equal volumes of chlorine Water Quality Parameter Bull Run Willamette Palm Beach
and ammonium chloride
solutions at a 3:1 (Cl 2 :N) To tal o rganic c arbo n— mg/L 1 .0 –1 .7 0 .8 –7 .1 1 0 –1 2
True c o lo r— units <5 NA* 2 5 –4 3
weight ratio, yielding a 150 Ammo nia— mg/L as N <0 .0 2 NA 1 .8 –2 .4
± 10 mg/L as Cl 2 solution. pH 7 .0 –7 .2 5 .0 –8 .5 6 .9 –7 .1
To tal hardne s s — mg/L as CaCO3 7 –1 5 NA 2 5 4 –3 3 2
Each solution was prepared To tal alkalinity— mg/L as CaCO3 5 –1 1 1 4 –3 6 2 3 0 –2 6 6
in a pH 8 phosphate buffer. Turbidity— ntu 0 .2 6 –1 .4 8 0 .7 –5 .0 0 .2 8 –0 .8 5
Stock ozone solution.
* NA— no t available
Oxygen carrier gas contain-
ing approximately 5 percent
ozone was bubbled through
an ozone generator* for a TABLE 3 Three disinfection reactors run in parallel for batch experiments
minimum of 20 min at 20oC
Reactor Purpose Description
(40 min at 40oC) through 400
mL of buffered reagent-grade 1 Co ntro l Wate r + o rganis ms
water in a 500-mL gas 2 Re s idual me as ure me nt Wate r + o rganis ms + dis infe c tant
3 Survival me as ure me nt Wate r + o rganis ms + dis infe c tant
absorption flask. Ozone con-
centrations in the stock solu-
tion ranged from 20 to 30
mg/L. Effluent gas was neu- TABLE 4 Comparison of k* values*
tralized by passage through a
Number of Correlation
solution containing 132 g/L Disinfectant Water k*—1/min Experiments Coefficient
of sodium thiosulfate and 3
g/L of potassium iodide3 to Fre e c hlo rine BDF 0 .0 0 8 6 0 .9 9 6 9
Bull Run 0 .0 3 3 4 0 .9 6 5 8
eliminate excess ozone. The Willame tte Rive r 0 .0 4 8 3 0 .8 6 8 9
stock solution was refriger- Palm Be ac h 0 .1 7 6 2 0 .7 2 9 5
ated (4oC) in dark conditions Mo no c hlo ramine BDF 0 .0 0 0 3 6 4 0 .9 9 8 8
Bull Run 0 .0 0 1 5 0 .9 9 5 4
for 15 to 20 min until use. Willame tte Rive r 0 .0 0 1 5 4 0 .9 8 1 2
Ozo ne BDF 0 .0 7 3 3 0 .9 7 5 5
Bull Run 2 .2 0 5 3 0 .9 3 5 9
Microbial preparation Willame tte Rive r 9 .6 9 1 3 0 .9 8 4 9
E. coli. Sterile nutrient Palm Be ac h 3 6 .4 5 1 0 .9 9 9 9
broth preparation. Nutrient
* Re s idual analys is c o mpute d fro m C = C0 e xp (– k* t)
Si—observed
sterilized using a membrane 10
–3
–2
each filter was placed on m-T7 10
22 to 24 h. Yellow coliforms
–4
were counted. None of the 10
experimental waters exhibited –5
10
detectable coliform counts.
G. muris. Enumeration of 10
–6
Giardia cysts was done by the –7
10
procedures outlined by Hoff –7 –6 –5 –4 –3 –2 –1 –0
et al.7 The cysts were concen- 10 10 10 10 10 10 10 10
trated by centrifugation at Si *
500˘ g for 5 min and resus- Observed versus predicted survival ratio plotted for all four waters; 95 percent confidence
pension of the pellet in phos- limits for error bars are included
At least twice during the course of the experi- This was achieved using a software application
ment (at the beginning and the end), samples were program.* The results from calculations for all waters
withdrawn from the residual measurement reactor for and disinfectants are shown in Table 4. Correlation
the analysis of residuals. These samples were ana- coefficients close to 1 demonstrate the validity of the
lyzed immediately after collection to minimize the first-order hypothesis.
possibility of residual loss. The resulting data were A direct relationship relative to demand for each
used for analysis of the disinfectant residual. disinfectant was observed (i.e., k* increases in order of
Free chlorine and monochloramine. Chlorine chlorine + preammoniation < monochloramine < free
residuals were measured using the methods chlorine < ozone). The value of k* also varied by water.
described2 and at times corresponding to initial, mid- Disinfectant demand was observed from greatest to
point, and final experimental conditions as well as least for Palm Beach, Willamette River, Bull Run, and
intermediate times as required. BDF water, respectively. The low value for the BDF
Stock chlorine solution. Chlorine concentrations water was expected because dC/dt = 0 is a condition for
were determined by forward amperometric titration demand-free water. In summary, residual demand
for free chlorine.2 was a function of both disinfectant and water.
Stock monochloramine solution. After the com-
bined solutions were stirred for 15 min, the resul- Models for microbial inactivation
tant solution was checked for free chlorine and mono- The data analysis of this project 10 used more
chloramine using forward amperometric titration.2 sophisticated and statistically accurate kinetic mod-
Ozone. Ozone residuals were measured using the
methods described previously2 and at times corre- *Microsoft Excel 4.0 Goal Seek, Microsoft Corp., Redwood, Wash.
The first-order decay rate * Kill was le s s than kill in BDF wate r.
for the different water and † The re was no s ys te matic diffe re nc e .
‡ Kill was mo re than kill in BDF wate r.
disinfectant combinations § Only two time data po ints we re available .
had been estimated with a
high degree of correlation
(Table 4); therefore, it was
important to incorporate this term (k*) into the dis- SSE (Uz)/(N – z) z
z
infection kinetic models (Figure 1). These equations
were derived by substituting the first-order residual
}} # 1 + F 1 – a X }}
S SE 0/ ( N – z ) 1 N – z N 2–z
(4)
decay equation (Eq 1) into the original rate expres- in which Uz = the vector of kinetic parameters for
sions for the three kinetic models (Figure 1) and then the bound of the confidence region, N = the number
integrating them.11 The resulting log-survival equa- of observations, z = the number of parameters, a is
tions for both the Chick–Watson and Hom models taken at 5 percent for a 95 percent confidence inter-
with first-order decay are given in Figure 2. The val, and F = the cumulative Fisher F distribution with
approximate solution to the Hom model with decay z and N – z degrees of freedom at the upper 1-a per-
was employed in this analysis.14 centile. This equation was rearranged for the confi-
The Hom and Chick–Watson equations in Figure dence interval:
2 were regressed against the survival data, and the z z
first-order decay rate was fixed according to the
appropriate value given in Table 4. Previous work10
1
SSE (Uz) = SSE0 X 1 + F 1 – a X }}
N – z N2 –z
(5)
has demonstrated that the Hom model provided a A software program* was used to solve the pre-
statistically significant improvement in fit to the inac- vious equation for the error sum of squares SSE(Uz)
tivation process, relative to the Chick–Watson model by varying only one of the three kinetic parameters.
in particular. Therefore, the Hom approach rather An initial high guess for the parameter was used for
than the Chick–Watson approach embodied in the C upper-bound convergence, and an initial low guess
X T tables was deemed the model of choice. The bet- was used for lower-bound convergence. These ini-
ter fit of the Hom model is illustrated in Figure 3 for tial guesses were based on the optimal parameter
the case of Giardia with monochloramine. The Hom values determined using nonlinear regression.
model better accounts for the shoulder in the sur-
vival data. Water comparison analysis
The optimal kinetic model parameters (i.e., k, n, In the water comparison analysis, inactivations
and m) were calculated using nonlinear regression achieved in BDF water were compared with those of
according to the same procedure used for the resid- the Bull Run, Willamette River, and Palm Beach
uals. The SSE0 was minimized for the survival data waters. The authors wanted to test whether disinfec-
according to the equation tion experiments performed with BDF water could be
used to predict performance in other waters. The pre-
SSE0 = S(lnSi – lnSi*)2 (3) dicted parameters k, n, and m from the Hom model
fit to BDF water were substituted into the predicted
in which Si = the observed survival ratio and Si* = the survival equations for the other waters, using the Hom
predicted survival ratio as given in Figure 2. Soft- with k* model of Figure 2. Only the first-order resid-
ware* was used to iterate the values of the model ual decay rate k* was retained because it was unique
parameters to a small tolerance level until the objec- for each of the waters. This led to a new set of predicted
tive function given by Eq 3 was minimized. survival values (Si* in Figure 2) for each of the waters
To calculate a confidence region for the kinetic that deviated slightly from the predicted survival for
parameters and the resulting predicted survival, the BDF water (according to the differences in k*).
model prediction error was computed by propagating The predicted survival values (Si*) were plotted
the errors associated with the determination of the against the actual observed values (Siobserved) for each
kinetic parameters. Upper and lower bounds for each of the waters (Figure 4). A perfect agreement between
of the parameters k, n, and m were determined using
the F ratio test:15 *Microsoft Excel 4.0 Solver, Microsoft Corp., Redwood, Wash.