Chemical Engineering Journal 157 (2010) 551–557
Contents lists available at ScienceDirect
Chemical Engineering Journal
journal homepage: www.elsevier.com/locate/cej
Full-scale modelling of an ozone reactor for drinking water treatment
Wim T.M. Audenaerta,b,∗, Manly Callewaerta, Ingmar Nopensb, Jan Cromphoutc,
Robert Vanhouckec, Ann Dumoulina, Pascal Dejansa, Stijn W.H. Van Hullea,b
aEnBiChem Research Group, Department of Industrial Engineering and Technology, University College West Flanders, Graaf Karel de Goedelaan 5, 8500 Kortrijk, Belgium
bBIOMATH, Department of Applied Mathematics, Biometrics and Process Control, Ghent University, Coupure Links 653, 9000 Gent, Belgium
cVMW, Flemish Water Supply Company, Beliardstraat 73, 1040 Brussels, Belgium
a r t i c l ei n f o
Received 21 August 2009
Received in revised form
23 December 2009
Accepted 30 December 2009
Advanced oxidation processes
Full-scale drinking water production
a b s t r a c t
In 2003, the Flemish Water Supply Company (VMW) extended its drinking water production site in
Kluizen (near Ghent, Belgium) with a combined ozonation and biological granular activated carbon
(BGAC) filtration process. Due to this upgrade, biostability increased, less chlorination was needed and
drinking water quality improved significantly. The aim of this study was to describe the full-scale reactor
with a limited set of equations. In order to describe the ozonation process, a model including key pro-
cesses such as ozone decomposition, organic carbon removal, disinfection and bromate formation was
developed. Kinetics were implemented in WEST®and simulation results were compared to real data. The
predicting performance was verified with a goodness-of-fit test and key parameters were determined
through a local sensitivity analysis. Parameters involving optical density (both rate constants and sto-
ichiometric coefficients) strongly affect model output. Some parameters with respect to bromate and
bacteria showed to be only, but to a large extent, sensitive to their associated concentrations. A scenario
analysis was performed to study the system’s behavior at different operational conditions. It was demon-
strated that the model is able to describe the operation of the full-scale ozone reactor, however, further
data collection for model validation is necessary.
© 2010 Elsevier B.V. All rights reserved.
In order to produce high quality drinking water from surface
rine is often used as oxidising agent. However, it is well established
due to the formation of potential toxic organochlorine compounds
such as tetrachloroethene, trichloroethene and halo-acetic acids
. Besides this, emerging pollutants and endocrine disruptors
(EDCs) became important contaminants in water systems during
the last few decades. In order to minimize this by-product for-
mation and to remove harmful compounds, advanced oxidation
processes (AOPs) have already proven to be effective technologies
[2,3]. Besides the potential benefits in drinking water production,
AOPs have a large potential for the treatment of different types
of water and waste streams originating from waste water, such
as domestic and industrial effluents, sludge and membrane con-
∗Corresponding author at: BIOMATH, Department of Applied Mathematics, Bio-
metrics and Process Control, Ghent University, Coupure Links 653, 9000 Gent,
E-mail address: Wim.Audenaert@Howest.be (W.T.M. Audenaert).
centrates, swimming pool water and process water. Both low and
high concentrated flows can be treated with AOP techniques [4,5].
Most techniques are based on the formation of hydroxyl radi-
cals which are the strongest oxidators that can be used in water
treatment systems. Hydroxyl radicals can be generated in water
through different combinations of oxidants, like ozone and hydro-
gen peroxide, or by combining a single oxidant with UV radiation
. AOP techniques such as ozonation are either used for (1)
the complete or partial oxidation of the organic contamination,
(2) the oxidation of a specific contaminant or (3) the removal of
pathogens. However, it should be noted, that ozone also produces
disinfection by-products (DBP) such as bromate which is formed
out of bromide and considered to be a potential human carcino-
gen . Ozone is used for the removal of MIB (2-methylisoborneol)
in drinking water production installations as this component is
responsible for odour and taste problems . The stage 1 and
stage 2 rules, promulgated by US EPA (Environmental Protection
Agency), defined a MCL (maximum contaminant level) of 10?gl−1
bromate for systems using ozone . In Europe, a similar reg-
ulatory level is applied since 1998 (98/83/EC). If the incoming
water contains high bromide levels, bromate formation can be
restricted by, e.g. adding hydrogen peroxide (H2O2) to the ozona-
tion process, . In this case, hydroxyl radicals become important
1385-8947/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
W.T.M. Audenaert et al. / Chemical Engineering Journal 157 (2010) 551–557
Despite the many advantages and added value of the AOP tech-
niques, there still exist several bottlenecks and research questions
concerning these techniques. First, scale-up of lab-scale research
reactors to full-scale industrial reactors is often failing. Second,
further research on process control and optimization is necessary.
Third, the removal of organic components and micro-organisms
is not yet completely unravelled. Modelling of AOP processes
offers an elegant and cost-effective tool to tackle these research
Many different attempts have already been made to describe
pounds [10,11]. Two general accepted deterministic models for
ozone decomposition in “pure water” have been developed in the
early 1980s, both based on the first model of Weiss . The model
of Staehelin, Hoigné and Bühler, known as the SHB model, was
experimentally developed at acidic to neutral pHs [13,14], while
Tomiyasu, Fukutomi and Gordon (TFG) developed their model at
high pH values . A comparison of both, together with some
simulation results can be found in Ref. . On the other hand,
numerous empirical and semi-empirical studies describing ozone
decomposition [17,18], reactions with organic compounds [16,18],
by-product formation  and disinfection  were conducted in
the last decades. Lovato et al.  extended the SHB model with
an empirical approach by relating one of the 18 kinetic constants
to the solution pH. van der Helm described ozone decomposi-
tion and organic compound removal by using UV absorbance at
254nm (UVA254) as surrogate for the NOM concentration .
Sohn et al.  developed an empirical relation to predict bro-
mate formation related to several operational and water quality
In this contribution a simplified model is presented for the sim-
ulation of the full-scale ozone reactor of the Flemish Water Supply
Company (VMW) in Kluizen (Belgium). The aim of this study is to
describe this process with a limited set of equations, to determine
the key parameters and to perform different scenario analysis. The
results will be used as a starting point for further model develop-
ment, especially in terms of extending the mechanisms of organic
compound removal and biodegradability enhancement. As such,
the model might contribute to answering the research questions
The 60,000m3day−1water treatment plant (WTP) which is the
subject of this study is fed with raw water captured from low-
enhanced coagulation followed by sludge blanket clarification,
oxidation with chlorine, sand filtration and granular activated
carbon filtration (GAC). A final disinfection with chlorine was
applied [21,22]. The intensive chlorination resulted in high tri-
halomethanes (THM) levels and prevented biofilm growth on the
activated carbon granules. Due to the absence of biological activity
and limited contact time in the GAC (15min), NOM removal was
mainly by adsorption, which is limited to the first 10,000 bed vol-
umes (BV, a unit that expresses the volume of water that already
passed through the filter as a multiple of the volume of the filter
bed). In 2000 the earthy-musty taste and odour compound MIB
appeared in the feed water with concentrations above the odour
threshold value of 10ngl−1, caused by algal growth in the reser-
voirs. This, along with a lack of biostability of the water due to
high TOC levels, forced the drinking water company to search for
an effective oxidizing technique. An ozone production and mixing
unit was introduced, together with biological granular activated
carbon (BGAC) filters. Ozone was implemented for both disinfec-
tion and oxidation. Due to the excellent disinfection capacities, a
first chlorination step could be omitted. On the other hand, ozone
was also implemented to enhance biodegradability in favour of the
biofilm present in the BGAC to remove MIB and NOM. Only a final
chlorination step remained [21,22].
Fig. 1. Schematic overview of the ozone production and mixing unit.
2. Materials and methods
2.1. Ozone reactor
Ozone is produced from oxygen with two Wedeco EFFIZON®
ozone generation units, each with a production capacity of
4000gh−1(180g ozone per Nm3oxygen/ozone mixture). The
ozone generating elements consist of discrete borosilicate glass
tubes with a diameter less than 11mm. A part of the main water
stream is pumped up and pressured up with a high pressure
booster pump. The water is brought into a venture injector where
ozonated gas is introduced into the water. The side stream is then
re-introduced to the main water pipe prior to a static mixer. The
overview of the gas transfer process is given in Fig. 1. A sampling
point is located after the static mixer (before the activated car-
bon filters). After the mixing is completed, the water proceeds
to the BGAC filters. Gas flow as well as ozone concentrations
in the gas are continuously monitored in order to evaluate the
verts any residual or non-dissolved ozone to oxygen so that the
ozone concentration in the treated off-gas is lower than 0.1ppm
2.2. Biological granular activated carbon (BGAC) filters
The BGAC consists of ten pressure filters with a diameter of 6m
and operational pressure of 1.5–2bar. GAC is operated as a two-
stage filtration, the first filter stage operates between 25,000 and
50,000BV. After 50,000BV (two years), the carbon is reactivated
and the filter is switched to the second stage position, which oper-
ates between 0 and 25,000BV. A contact time of 6min is obtained
in the filters, above the carbon bed. Non-dissolved gas is collected
in an upward tee above each GAC filter and led away to the ozone
destruction system [21,22].
2.3. Modelling approach
The ozone reactor of the Flemish Water Supply Company was
implemented in the modelling and simulation platform WEST®
(MostforWater, Belgium) as two continuous stirred tank reactors
(CSTRs) in series. In the first tank with a volume of 10.68m3the
pipe before and with the static mixer. A second reactor represents
W.T.M. Audenaert et al. / Chemical Engineering Journal 157 (2010) 551–557
Fig. 2. Implementation of the ozone reactor in the simulation platform WEST®.
tion phase takes place. The simulation configuration as used in the
software program is given in Fig. 2.
Four ozone reactions were implemented in WEST®according to
•Reaction of optical density (OD) with ozone (organic carbon oxi-
•Bromide oxidation (bromate formation).
rect [10,23]. E.g., the disinfection reaction rate can be described by
the following equation, where molecular ozone as well as hydroxyl
radicals contribute to the oxidizing capacity of the system:
? = (kO3+ kOHRc)[O3][Xbact]withRc=[HO•]
with ?: reaction rate, in this example the inactivation rate of
bacteria (CFUm−3s−1); kO3: second order rate constant for the
direct reaction of molecular ozone with a specific compound, in
this example micro-organisms (m3g−1s−1); kOH: second order
rate constant for the indirect ozone reaction pathway of hydroxyl
of ozone in solution (gm−3); [Xbact]: the density of viable micro-
organisms, in this model expressed as colony-forming units per
When Rcis assumed to be constant and very little, both direct
and indirect reactions can be lumped into one:
? = kO3[O3][Xbact]
However, if the indirect mechanism plays an important role
in the oxidation of some calculated species, predictive capabili-
ties of the model will deteriorate because process efficiency in that
case highly depends on other process conditions such as scavenger
concentrations that are not included in the model. Probably rad-
ical reactions indeed occur in the waterworks because MIB was
removed after ozonation, while rate constants with regard to the
direct and indirect pathway are <10 and 3×109M−1s−1, respec-
Direct ozone decomposition is modelled assuming that ozone
follows a first order decay with a rate constant of 0.000485s−1
. UV absorbance (optical density) at 254nm was used as a sur-
rogate for the amount of organic material that reacts with ozone
sents a part of the organic pollution concentration as it specifically
gives a measure of the amount of aromatic and unsaturated com-
opportunities for modelling and control as it can be determined
on-line and consequently a huge amount of real-time and accurate
data are available. Accordingly, OD might be a useful parameter in
model-based control of WTPs.
2008 over a period of 300 days. For kOD, which is the rate constant
for the reaction of UVA254with ozone, an initial value in the range
of 0.1m3g−1s−1was used , while after calibration through
parameter estimation a value of 0.0135m3g−1s−1was found. The
stoichiometry or yield (Y) of the reaction is represented by YO3/OD,
with a numerical value of 0.22 . This implies that one unit of
OD (m−1) consumes 0.22g of ozone.
Chick–Watson model  (and references therein):
adapted fromthe classic
= −kX[O3][Xbact] (3)
For kX, the inactivation rate constant for a particular micro-
organism (m3g−1s−1), a value of 0.6m3g−1s−1was applied after
value of 1.29×10−14.
by-product of ozonation in bromide containing waters. Bromide
was assumed to be directly oxidized to bromate, although in real-
ity ozone first oxidizes bromide to form hypobromous acid and
forms bromate [23,26,27]. As such, one rate constant for bromate
formation was determined after calibration and the stoichiometric
The rate constant was found to be 0.00043m3g−1s−1.
The kinetics and stoichiometric coefficients used in the model
are presented in Table 1 as a Petersen matrix. This matrix presen-
tation offers a clear overview of the chemical reaction mechanisms
included in the model. Reaction rates are indicated in the right
column. Matrix elements are stoichiometry parameters.
The temperature and pH influence was not accounted for in
this study as these parameters remain almost constant during the
waterworks daily operation. Although, for example the inactiva-
tion constant for bacteria, kX, is temperature dependent . With
position rate when lower than 7, but at higher values, the rate
Petersen matrix representing the model used in this study.
ProcessO3(gm−3)OD (m−1m−3)Xbact(CFUm−3)Br−(gm−3) BrO3(gm−3) Reaction rate
Reaction of optical density with ozone
W.T.M. Audenaert et al. / Chemical Engineering Journal 157 (2010) 551–557
increases significantly . For instance, von Gunten and Hoigné
showed that the half-life of ozone is 10 times higher at pH 10 than
is formed when lowering the pH. Further, no gas transfer equations
were included in the model. A dissolved ozone concentration in
the influent of the first tank was defined at the beginning of each
simulation run. Finally, the activated carbon present in the second
However, this was not considered.
2.4. Data interpretation
The goodness-of-fit between experimental and simulated val-
, which is expressed as follows:
where yirepresents the simulated data points; ym,irepresenting
the measured data points.
A value of the TIC lower than 0.3 indicates a good agreement
with measured data .
A sensitivity analysis was performed in WEST®to determine
a major influence on the model output). The relative sensitivity
function (RSF) was adopted to evaluate the sensitivity of the model
output (concentration of ozone, OD, CFU and bromate) to a change
of model parameters (rate constants kO3, kOD, kXand kBrand the
stoichiomtric coefficients YO3/ODand YO3/X).
finite forward difference method with a perturbation factor of 0.1%
. This means that SFs were calculated by raising the nominal
parameter value with 0.1% as shown in following equation:
inal parameter value; ? is the perturbation factor.
RSF was calculated as follows:
RSF =SF × ?
y(t,?j+ ??j) − y(t,?j)
A RSF less than 0.25 indicates that the parameter is not influ-
ential. Parameters are moderately influential when 0.25<RSF<1.
When 1<RSF<2 and RSF>2, the parameter seems to be very and
extremely influential, respectively .
Ozone was measured spectrophotometrically with the indigo
reagent method at 600nm . Optical density was constantly
measured with a process integrated UV spectrometer at 254nm.
Bromide, bromate and total CFU were analysed according to stan-
dard methods . All analyses were performed by the Flemish
Water Supply Company.
3. Results and discussion
3.1. Modelling results
2008 over a period of 300 days. The influent flow rate and optical
density are represented in Fig. 3. The influent bacteria and bro-
mide concentration were on average 183CFUl−1and 138?gl−1,
Fig. 3. Influent flow rate and optical density.
respectively. The descending trend of the OD can be attributed to
improved settling and flotation performance in the pre-treatment
steps during that period.
As mentioned before, all samples were withdrawn before the
ulation results are representing the effluent of the first reactor in
Fig. 2. Fig. 4 shows the measured and calculated optical density
after the first reactor. A good agreement was obtained as the TIC
for this parameter is calculated to be 0.044 (<0.3). The deviation
of the applied rate constant for kODin comparison with literature
reported values can be explained by differences in organic carbon
content of the water.
On average, 25% of the OD was removed for both calculated and
modelled values. Concerning the number of CFU, an average log
sured removal (1.1log). Measured influent bacteria concentrations
vary between 80 and 200CFUl−1, while those of the effluent are in
are compared in Fig. 5. As can be seen, calculated values agree well
with experimental ones. This is confirmed with a calculated TIC of
0.084, although it has to be highlighted that more data points have
to be collected in future studies.
age calculated bromate concentration and the measured value are
model predictions are realistic. In future work, the validation pro-
cess will be repeated, especially for bromate formation. Measured
3.2. Sensitivity analysis
The initial parameter values used in the sensitivity analysis are
eters involving OD (kODand YO3/OD) strongly affect model output
Fig. 4. Comparison of measured and calculated optical density.
W.T.M. Audenaert et al. / Chemical Engineering Journal 157 (2010) 551–557
Fig. 5. Comparison of measured and calculated logarithmic bacteria removals.
Fig. 6. Predicted and measured bromate concentrations in the effluent.
compared to others (Table 2). kODhas a moderate effect on calcu-
influence with respect to OD itself. The same conclusions can be
made for YO3/OD, although moderate effects are slightly higher.
kXand kBronly influence their associated concentrations [X] and
rate estimation regarding these parameters will be necessary to
obtain realistic predictions of bacteria and bromate levels. kO3and
YO3/Xdo not exert an influence on simulation output. Bacteria (X)
form part of OCS (ozone consuming substances), but they consume
a negligible amount of ozone due to their extremely low concen-
trations . This explains that YO3/Xhas no effect on the bacteria
concentration as the ozone requirement for this reaction is met
under normal operational conditions.
3.3. Scenario analysis
The effect of applied ozone concentration and flow rate on cer-
tain key variables was evaluated. The normally operational ozone
Fig. 7. Scenario analysis, effect of flow rate on OD and bromate concentrations.
dose in the drinking water production centre is 2.5mgl−1. Scenar-
ios were calculated for concentrations varying from 0 to 5mgl−1.
Flow rate (and consequently hydraulic retention time) was varied
within a range of 0 (batch reactor) to 1m3s−1, while real influent
flow rates are in the range of 0.4–0.5m3s−1.
Effects of ozone dose on effluent OD and bromate formation are
shown in Fig. 7. This figure reveals that a compromise has to be
made regarding ozone dose and flow rate to comply with bromate
levels without losing OD (and bacteria) removal goals. Fig. 7 also
shows that the normally applied flow rates in this case are in a
mate level stays well below the standard of 10?gl−1. The influent
OD concentration was 11.3m−1.
As can be deduced from Fig. 8, the operational ozone concen-
tration of 2.5mgl−1is well chosen. Again, sufficient OD is removed
and the bromate guideline is met without problems.
3.4. Bromate formation
Sohn et al.  stated that most of the models for predicting
residual ozone and bromate formation take empirical functional
forms because the complexity of natural organic matter restricts
developing complete theoretically based chemical kinetic models.
They developed a multiple regression model that was compared to
experimental data from the VMW  (and references therein):
[BrO3] = 1.55 × 10−6× [TOC]−1.26× [pH]5.82
×[O3]1.57× [Br]0.73× t0.28× (1.035)T−20
Fig. 9 shows this comparison. The model developed in this
study was added to the graph. The regression model seems to
show a better prediction of the experimental data, as also indi-
cated when comparing the TIC values of both, the regression and
the kinetic models with values of 0.17 and 0.21, respectively.
Fig. 8. Scenario analysis, effect of ozone dose on OD and bromate concentrations.
W.T.M. Audenaert et al. / Chemical Engineering Journal 157 (2010) 551–557
Fig. 9. comparison of predicting performance of a regression model and the kinetic
model developed in this study.
However, both models agree well with reality (TIC<0.3). The over-
estimation of bromate by the kinetic model can be caused by
inaccurate parameter estimation with insufficient data as kBrhas
proven to be very influential to simulated bromate concentra-
tions (see Section 3.2). The assumptions that were made regarding
the bromate mechanism, together with not included temperature
effects can contribute to less prediction accuracy, together with
the fact that only 1 data point was available for calibration. Regres-
sion models are able to describe experimental data very well and
temperature or reactor correction factors can be easily added .
On the other hand, models based on well defined mechanisms
can give more substantiated insight in processes and are easier
In this study, a simplified kinetic model describing ozone
decomposition, organic carbon removal, disinfection and bromate
formation during ozonation applied in drinking water production
was developed. Calibration and simulation runs were based on his-
Supply Company waterworks in Kluizen, Belgium. It was demon-
strated that the developed model is able to predict excess ozone
concentration, OD removal, bacteria inactivation and bromate for-
mation, although further data collection and batch experiments
will be necessary to further validate the model. A sensitivity analy-
sis revealed that parameters involving optical density (both rate
constants and stoichiometric coefficients) strongly affect model
output. Some parameters with respect to bromate and bacteria
showed to be only, but to a large extent, sensitive to their asso-
ciated concentrations. OD seems to be a valuable parameter for the
application of model-based control and optimization strategies as
it can be determined on-line and consequently a huge amount of
real-time and accurate data are available.
With drinking water standards becoming more stringent, mod-
els will become an important tool to assess drinking water plant
ysis (particularly effect of ozone dosage on reactor performance)
and will play an important role in the drinking water modelling
studies in Flanders. The model will be used as starting point for a
more detailed model which includes radical reactions (i.e. model
of Staehelin, Hoigné and Bühler) [10,13,14] and a combination
of AOP techniques (ozone, UV, H2O2) to guarantee satisfactory
model predictions and to improve the applicability and optimiza-
pany (VMW) for the fruitful collaboration and making the data of
the water purification site available.
This research was partially funded by a University College West
Flanders PhD Research Grant and is in close collaboration with the
Veg-i-Trade FP7-KBBE-2009-3 project.
 J.J. Rook, Formation of haloforms during chlorination of natural water, J. Water
Treat. Exam. 23 (2) (1974) 234–243.
 M. Werderitsch, Full scale advanced oxidation treatment plant for ground-
water, in: M. Sievers, S. Geissen, S. Schäfer, B. Kragert, M. Niedermeiser
(Eds.), Proceedings of the 5th IWA International Conference/10th IOA-EA3G
Conference on Oxidation Technologies for Water and Wastewater Treat-
ment, Berlin, Germany, March 30–April 2, 2009, CUTEC publication series,
 C. Boucherie, C. Lecarpentier, N. Fauchon, M. Djaferand, V. Heim, “Ozone” and
“GAC filtration” synergy for emerging micropollutants removal on drinking
water treatment plant? in: M. Sievers, S. Geissen, S. Schäfer, B. Kragert, M. Nie-
dermeiser (Eds.), Proceedings of the 5th IWA International Conference/10th
IOA-EA3G Conference on Oxidation Technologies for Water and Wastewater
 J.C. Crittenden, R.R. Trussell, D.W. Hand, K.J. Howe, G. Tchobanoglous, Water
Treatment Principles and Design, John Wiley and Sons, Hoboken, USA, 2005.
 S.A. Parsons, Advanced Oxidation Process for Water and Wastewater Treat-
ment, IWA Publishing, London, UK, 2005.
involving ozone, hydrogen peroxide and ultraviolet radiation Ozone Sci. Eng. 9
 U. von, Gunten, Ozonation of drinking water: Part II. Disinfection and by-
product formation in presence of bromide, iodide or chlorine, Water Res. 37
 M. Sangehashi, K. Shiraishi, H. Fujita, T. Fujji, A. Sakoda, Ozone decomposition
of 2-methylisorboneol (MIB) in adsorption phase on high silica zeolites with
preventing bromate formation, Water Res. 39 (2005) 2926–2934.
 National Primary Drinking Water Regulations: Stage 2 Disinfectants and Dis-
infection Byproducts Rule, Fed. Regist. 71 (2) (2006).
 J. Beltràn, Ozone Reaction Kinetics for Water and Wastewater Systems, CRC
Press LLC, Florida, USA, 2004.
 A.W.C. van der Helm, Integrated modeling of ozonation for optimization of
drinking water treatment, Ph.D. Thesis, Faculty of Civil Engineering and Geo-
sciences, Delft University of Technology (150 pag.), 2007.
 J. Weiss, Investigations on the radical HO2in solution, Trans., Faraday Soc. 31
 R.E. Bühler, J. Staehelin, J. Hoigné, Ozone decomposition in water studied by
pulse radiolysis. 1. HO2/O2−and HO3/O3−as intermediates, J. Phys. Chem. 88
 J. Staehelin, R.E. Bühler, J. Hoigné, Ozone decomposition in water studied by
position in basic aqueous solution, Inorg. Chem. 24 (1985) 2962–2966.
 P. Westerhoff, R. Song, G. Amy, R. Minear, Applications of ozone decomposition
models, Ozone Sci. Eng. 19 (1997) 55–74.
 M.E. Lovato, C.A. Martín, A.E. Cassano, A reaction kinetic model for ozone
decomposition in aqueous media valid for neutral and acidic pH, Chem. Eng. J.
146 (2009) 486–497.
 A.W.C. van der Helm, P.W.M.H. Smeets, E.T. Baars, L.C. Rietveld, J.C. Van Dijk,
 J. Sohn, G. Amy, J. Cho, Y. Lee, Y. Yoon, Disinfectant decay and disinfectant
by-products formation model development: chlorination and ozonation by-
products, Water Res. 38 (10) (2004) 2461–2478.
 N.K. Hunt, B.J. Mari˜ nas, Inactivation of Escherichia coli with ozone: chemical
and inactivation kinetics, Water Res. 33 (1999) 2633–2641.
 J. Cromphout, R. Vanhoucke, Reduction of exploitation costs and improvement
of water quality by the implementation of ozonization at the waterworks
in Kluizen, in: Proceedings of the IOA-EA3G-VIVAQUA International Confer-
ence on Ozone and Related Oxidants in Advanced Treatment of Water for
sistant Pollutants and Control of By-products, Brussels, Belgium, May 15–16,
 J. Cromphout, E. Walraevens, R. Vanhoucke, Improvement of water quality in
the drinking water plant of Kluizen by the use of ozone in combination with
GAC, Tribune de L’Eau 58 (1) (2005) 15–18.
 U. von, Gunten, Ozonation of drinking water. Part I. Oxidation kinetics and
product formation, Water Res. 37 (2003) 1443–1467.
 W. Song, V. Ravindran, M. Pirbazari, Process optimization using a kinetic
model for the ultraviolet radiation–hydrogen peroxide decomposition of natu-
W.T.M. Audenaert et al. / Chemical Engineering Journal 157 (2010) 551–557 Download full-text
 P. Westerhoff, G. Aiken, G. Amy, J. Debroux, Relationships between the struc-
ture of natural organic matter and its reactivity towards molecular ozone and
hydroxyl radicals, Water Res. 33 (10) (1999) 2265–2276.
 W.R. Haag, J. Hoigné, Ozonation of bromide-containing waters: kinetics of for-
mation of hypobromous acid and bromate, Environ. Sci. Technol. 17 (1983)
 U. von Gunten, J. Hoigné, Bromate formation during ozonation of bromide-
Sci. Technol. 28 (1994) 1234–1242.
 H. Theil, Economic Forecasts and Policy, 2nd ed., North Holland Publishing Co.,
 X. Zhou, A new method with high confidence for validation of computer simu-
lation models for flight systems, Chin. J. Syst. Eng. Electron. 4 (1993) 43–52.
 D.J.W. De Pauw, P.A. Vanrolleghem, Practical aspects of sensitivity function
approximation for dynamic models, Math. Comp. Modell. Dyn. Sys. 12 (2006)
 T. Jiang, X. Liu, M.D. Kennedy, J.C. Schippers, P.A. Vanrolleghem, Calibrating a
side-stream membrane bioreactor using Activated Sludge Model No. 1, Water
Sci. Technol. 52 (2005) 359–367.
 H. Bader, J. Hoigné, Determination of ozone in water by the indigo method,
Water Res. 15 (1981) 449–456.
 Standard Methods for the Examination of Water and Wastewater, 18th ed.,
American Public Health Association Inc., New York, 1992.
 T. Huang, C.J. Brouckaert, M. Pryor, C.A. Buckley, Application of computational
fluid dynamics modelling to an ozone contactor, Water SA 30 (2004) 51–56.
 S.W.H. Van Hulle, J. Verstraete, J. Hogie, P. Dejans, A. Dumoulin, Modelling and
simulation of a nitrification biofilter for drinking water purification, Water SA
32 (2) (2006) 257–264.