Bioresource Technology 101 (2010) 7278–7285

Contents lists available at ScienceDirect

Bioresource Technology
journal homepage: www.elsevier.com/locate/biortech

Identification of the most sensitive parameters in the activated sludge model

implemented in BioWin software
Ewa Liwarska-Bizukojc a,*, Rafal Biernacki a,b
a
Institute of Environmental Engineering, Technical University of Lodz, Al. Politechniki 6, 90-924 Lodz, Poland
b
Water Supply System and Sewer – Zgierz Ltd., ul. A. Struga 45, 95-100 Zgierz, Poland

a r t i c l e i n f o a b s t r a c t

Article history: In order to simulate biological wastewater treatment processes, data concerning wastewater and sludge
Received 5 February 2010 composition, process kinetics and stoichiometry are required. Selection of the most sensitive parameters
Received in revised form 20 April 2010 is an important step of model calibration. The aim of this work is to verify the predictability of the acti-
Accepted 25 April 2010
vated sludge model, which is implemented in BioWin software, and select its most influential kinetic and
Available online 15 May 2010
stoichiometric parameters with the help of sensitivity analysis approach. Two different measures of sen-
sitivity are applied: the normalised sensitivity coefficient (Si,j) and the mean square sensitivity measure
Keywords:
(dmsqr
j ). It occurs that 17 kinetic and stoichiometric parameters of the BioWin activated sludge (AS) model
Activated sludge
Model parameters
can be regarded as influential on the basis of Si,j calculations. Half of the influential parameters are asso-
Sensitivity analysis ciated with growth and decay of phosphorus accumulating organisms (PAOs). The identification of the set
Wastewater of the most sensitive parameters should support the users of this model and initiate the elaboration of
determination procedures for the parameters, for which it has not been done yet.
Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction mainly on the targets of modelling. Several calibration methodolo-

Activated sludge models (ASM1, ASM2, ASM2d, and ASM3) pro-
posed by the International Water Association (IWA) task group on
mathematical modelling for design and operation of biological
wastewater treatment are the most commonly used mathematical
description for modelling biological wastewater treatment pro-
cesses. The activated sludge models (ASMs) or ASM-based models
are incorporated in most of the contemporarily used simulation
software, for example ASIM, BioWin, GPS-X, WEST, DESASS (Ger-
naey et al., 2004; Ferrer et al., 2008). In order to successfully apply
one of these models in simulations of wastewater treatment
processes, its calibration is necessary. Model calibration is defined
as the adaptation of a model so that it fitted a certain set of data
from the full-scale wastewater treatment plant (WWTP) under
study (Petersen et al., 2003). A methodology for the calibration of
the activated sludge plant models may be different and depends
Abbreviations: ASM, activated sludge model; AS/AD, activated sludge/anaerobic
digestion; AOB, Ammonia Oxidising Biomass; ANAMMOX, ANaerobic AMMonia
OXidisers; n, number of output variables; NOB, Nitrite Oxidising Biomass; OHOs,
Ordinary Heterotrophic Organisms; PE, population equivalent; PAOs, phosphorus
accumulating organisms; PHA, poly-hydroxy-alkanoates; SD, standard deviation;
xi, input variable; yi, output variable; WWTP, wastewater treatment plant.
* Corresponding author. Tel.: +48 42 631 35 97; fax: +48 42 631 35 17.
E-mail addresses: ewa.liwarska-bizukojc@p.lodz.pl (E. Liwarska-Bizukojc),
r.biernacki@wodkan.zgierz.pl (R. Biernacki).
gies have been elaborated so far (Melcer et al., 2003; Vanrolleghem
et al., 2003; Langergraber et al., 2004). Vanrolleghem et al. (2003)
proposed ‘‘Biomath-Calibration” protocol, which consolidated engi-
neering experience and a scientific approach in the area of ASMs
calibration. This protocol consists of four main stages: (1) definition
of the target(s), (2) the collection of the detailed information on the
activated sludge plant, (3) steady-state and dynamic calibration,
and (4) decision-making. A similar model calibration procedure
was described by Langergraber et al. (2004). This guideline is com-
posed of seven steps: (1) definition of objectives, (2) data collection
and model selection, (3) data quality control, (4) evaluation of mod-
el structure and experimental design, (5) data collection for simula-
tion study, (6) calibration/validation, and (7) study and evaluation
of success (Langergraber et al., 2004). Sin et al. (2005) summarised
and compared in detail two aforementioned and two other proce-
dures: the Dutch Foundation of Applied Water Research (STOWA)
calibration protocol and Water Environment Research Foundation
(WERF) protocol calibrations methodologies.
It was found that for a steady-state model calibration, the
parameters responsible for the long-term behaviour, i.e. yield coef-
ficient for heterotrophic biomass (YH), specific lysis (decay) rate
constant for heterotrophic biomass (bH), fraction of biomass lead-
ing to particulate products (fp) should be determined (Petersen
et al., 2003). At the same time for a dynamic calibration also
parameters connected with the short-term behaviour of the

0960-8524/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.biortech.2010.04.065
 ARTICLE IN PRESS

Water Research 39 (2005) 2459–2474

www.elsevier.com/locate/watres
Review
A critical comparison of systematic calibration protocols for
activated sludge models: A SWOT analysis
Gürkan Sin, Stijn W.H. Van Hulle, Dirk J.W. De Pauw, Ann van Griensven,
Peter A. Vanrolleghem
BIOMATH, Department of Applied Mathematics, Biometrics and Process Control, Ghent University, Coupure Links 653,
B-9000 Gent, Belgium
Received 14 September 2004; received in revised form 3 May 2005; accepted 4 May 2005

Abstract

Modelling activated sludge systems has gained an increasing momentum after the introduction of activated sludge
models (ASMs) in 1987. Application of dynamic models for full-scale systems requires essentially a calibration of the
chosen ASM to the case under study. Numerous full-scale model applications have been performed so far which were
mostly based on ad hoc approaches and expert knowledge. Further, each modelling study has followed a different
calibration approach: e.g. different influent wastewater characterization methods, different kinetic parameter
estimation methods, different selection of parameters to be calibrated, different priorities within the calibration steps,
etc. In short, there was no standard approach in performing the calibration study, which makes it difficult, if not
impossible, to (1) compare different calibrations of ASMs with each other and (2) perform internal quality checks for
each calibration study. To address these concerns, systematic calibration protocols have recently been proposed to
bring guidance to the modeling of activated sludge systems and in particular to the calibration of full-scale models. In
this contribution four existing calibration approaches (BIOMATH, HSG, STOWA and WERF) will be critically
discussed using a SWOT (Strengths, Weaknesses, Opportunities, Threats) analysis. It will also be assessed in what way
these approaches can be further developed in view of further improving the quality of ASM calibration. In this respect,
the potential of automating some steps of the calibration procedure by use of mathematical algorithms is highlighted.
r 2005 Elsevier Ltd. All rights reserved.

Keywords: Activated sludge models (ASMs); Calibration; Systematic protocols; SWOT analysis

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2460
2. Systematic protocols for activated sludge model calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2461
2.1. Summary of the four systematic calibration protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2461
2.1.1. The BIOMATH calibration protocol (Vanrolleghem et al., 2003) . . . . . . . . . . . . . . . . . . . 2461
2.1.2. The STOWA calibration protocol (Hulsbeek et al., 2002) . . . . . . . . . . . . . . . . . . . . . . . . 2463

Corresponding author. Tel.: +32 9 264 5937; fax: +32 9 264 6220.
E-mail address: gurkan.sin@ugent.be (G. Sin).

0043-1354/$ - see front matter r 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.watres.2005.05.006
 E. Liwarska-Bizukojc, R. Biernacki / Bioresource Technology 101 (2010) 7278–7285 7279

Nomenclature

baerob,A aerobic decay rate for AOB (d1) YlowPP fraction of P stored in releasable poly-P form
baerob,N aerobic decay rate for NOB (d1) (mg P mg P1)
bH aerobic decay rate for OHO (d1) YP/acetic amount of P released for 1 mg of acetate sequestered in
ka ammonification rate of soluble organic nitrogen com- the form of PHA (mg P mg COD1)
pound (d1) YP/PHA,aerobic amount of P stored per unit of PHA oxidised in aer-
K NH4 substrate (N—NHþ 1
4 ) half-saturation constant (mg N l ) obic conditions (mg P mg COD1)
KOA oxygen half-saturation constant for AOB (mg O2 l1) YP/PHA,seq amount of PHA stored when 1 mg of acetate or propio-
KOH oxygen half-saturation constant for OHOs (mg O2 l1) nate is sequestered (mg COD mg COD1)
KPP P uptake half-saturation constant (mg P l1) YPO4 amount of P released per 1 mg of substrate
KS substrate (COD) half-saturation constant (mg COD l1) (mg P mg COD1)
KS,PAO substrate (PHA) half-saturation constant (mg COD l1)
KS,P-limited substrate (PHA) half-saturation constant under phos- Greek symbols
phorus limiting conditions (mg COD l1) dmsqr
j the mean square sensitivity measure (–)
Ntot total nitrogen concentration (mg N l1) ganoxic,H anoxic growth factor
Ptot total phosphorus concentration (mg P l1) ganoxic,h rate reduction factor for hydrolysis under anoxic condi-
RAS recirculated activated sludge (m3 d1) tions (–)
Si,j the normalised sensitivity coefficient (–) lmaxA maximum specific growth rate of AOB (d1)
SRT sludge retention time (d) lmaxH maximum specific growth rate of OHO under aerobic
TKN Kjeldahl nitrogen concentration (mg N l1) conditions (d1)
TSS total suspended solids (mg l1) lmaxPAO maximum specific growth rate of PAO
YA yield coefficient for AOB (mg COD mg N1) lmaxP-limited maximum specific growth rate of PAO under phos-
YH yield coefficient for OHO under aerobic conditions phorus limiting conditions (d1)
(mg COD mg COD1)

system should be estimated. To this set of parameters belong: methods that: (1) can be quickly used to seek for the most important
maximum specific growth rate for heterotrophic biomass (lmaxH), inputs, (2) are based on differential analyses, (3) are based on
maximum specific growth rate for autotrophic biomass (lmaxA), sampling and (4) are based on variance methods (US EPA, 2009).
correction factor, which adjust for either the change in lmaxH asso- Choosing the appropriate sensitivity analysis method is usually a
ciated with anoxic conditions or for the fact that only a portion of compromise between the amount of information one wants from
biomass can denitrify (ganoxic,H), correction factor for hydrolysis of the analysis and the computational difficulties of the analysis (Sal-
slowly biodegradable organic matter under anoxic conditions telli et al., 2008). With regard to activated sludge models (ASMs)
(ganoxic,h), substrate half-saturation constant for heterotrophic sensitivity analysis is applied for selection of the parameters to be
organisms (KS), ammonium half-saturation coefficient for auto- adjusted within model calibration as well as to investigate, if the
trophic biomass (K NH4 ), oxygen half-saturation constant for hetero- parameters that were modified in this process indeed influence on
trophic organisms (KOH) and oxygen half-saturation constant for the model outputs significantly (Petersen et al., 2002; Gernaey
autotrophic organisms (KOA) (Petersen et al., 2003). et al., 2004). According to the authors of ‘‘Biomath-Calibration” pro-
Analysing the literature on ASMs application and calibration, it tocol, sensitivity analysis should be incorporated into the calibra-
is well seen that one of the major problems is to select a set of rel- tion protocol to minimise the efforts and optimise the overall
evant parameters, which are necessary to achieve good prediction calibration procedure (Vanrolleghem et al., 2003). Sin et al. (2009)
of the used model. These parameters should be calibrated or, if studied the issue of uncertainty analysis of WWTP models and the
possible, determined with the help of an appropriate methodology.
Identifiability analysis is performed to check the ability of obtain-
ing of a unique combination of parameters describing the behav-
iour of the system (Petersen et al., 2003). Theoretical and
practical identifiability should be distinguished. Theoretical identi-
fiability is a property of the model structure, whereas practical in-
cludes the quality of data (Petersen et al., 2003). It may happen
that theoretically identifiable parameters are practically unidenti-
fiable, if available data are invalid or too noisy. The problems in
identifying of ASM parameters are most probably caused by the
complicated nature of the ASMs and ASM-based models, and the
numerous parameters incorporated in them (Henze et al., 2000).
In Fig. 1, the changes of the number of parameters within ASMs
development are shown.
The authors of calibration guidelines for ASMs emphasized the
need of the selection of parameter subsets for model calibration
(Melcer et al., 2003; Vanrolleghem et al., 2003; Langergraber
et al., 2004). The calibration and/or determination of all model
parameters are expensive and time consuming processes. Thus, a
sensitivity analysis, which allows for selecting the most influential
parameters, is a very helpful tool in this task. The various techniques Fig. 1. Comparison of the number of kinetic and stoichiometric parameters for
of global and local sensitivity analysis can be categorised into the ASMs and the BioWin AS model.
7280 E. Liwarska-Bizukojc, R. Biernacki / Bioresource Technology 101 (2010) 7278–7285
issue of framing, and how the latter affects the interpretation of
uncertainty analysis results. According to this study, sensitivity
analysis would be a very valuable tool for the supplementation of
the uncertainty analysis of WWTP models (Sin et al., 2009).
Several attempts have been made to identify the most sensitive
parameters for individual models from ASM family (Weijers and
Vanrolleghem, 1997; van Veldhuizen et al., 1999; Brun et al.,
2002; Petersen et al., 2002; Makinia, 2006; Makinia et al., 2006).
Weijers and Vanrolleghem (1997) elaborated the procedure for
selecting of the best identifiable parameters in calibrating ASM1
to full-scale plant data. This procedure inter alia comprised the
selection of a reduced set of the most sensitive parameters by
means of sensitivity analysis. The reduced set found by Weijers
and Vanrolleghem (1997) included the following kinetic and stoi-
chiometric parameters: YH, yield coefficient for autotrophic bio-
mass (YA), lmaxH, bH, lmaxA, KS, KOA and ganoxic,H. Brun et al.
(2002) elaborated a systematic approach, which can be used to
identify the most important parameters in ASMs, upon the exam-
ple of ASM2d. They proposed the application of three different
diagnostic measures including the mean square sensitivity mea-
sure (dmsqr
j ). Makinia et al. (2006) identified the most influential
parameters for ASM2d and ASM3P models on the basis of the nor-
malised sensitivity coefficient (Si,j).
At the same time the sensitivity analysis for the complex ASM-
based models as the BioWin activated sludge (AS) model has
hardly ever been performed. In this study, the standard sensitivity
measures were applied in order to fill this gap. The aim of this work
is to verify the predictability of the BioWin AS model, which is
implemented in BioWin software, and select its most influential ki-
netic and stoichiometric parameters.
2. Methods
2.1. Wastewater treatment plant in Zgierz
The data used in this paper were obtained from the WWTP in
the city of Zgierz (Poland). This WWTP treats municipal wastewa-
ter from the city and several communes located in the neighbour-
hood and industrial wastewater originating mainly from small and
medium enterprises. The average pollutant load to the plant corre-
sponds to approximately 94,000 PE. The contribution of industrial
wastewater is usually in the range from 10% to 15%. The average
inflow of wastewater is equal to 11,500 m3 d1, whereas the max-
imal inflow achieves 20,000 m3 d1. The treated wastewater is di-
rectly discharged to Bzura River. The Zgierz WWTP went into
operation in 1995 and since then it has performed mechanical
and biological treatment of wastewater. The biological step con-
sists of a three-zone bioreactor and secondary clarifier run in the
Phoredox process configuration. The volumes of anaerobic, anoxic
and aerobic zone are equal to 857.5, 3536.4 and 19730.6 m3,
respectively. The surface area of the radial secondary clarifier is
1018 m2 (diameter equals to 36 m) and its total depth is 4.5 m. A
schematic flow diagram of the biological step of the Zgierz WWTP
is shown in Fig. 2. Influent and effluent flow rates, pH were mea-
sured on-line. Chemical Oxygen Demand (COD), Biochemical Oxy-
gen Demand (BOD5), total nitrogen (Ntot), Total Kjeldahl Nitrogen
(TKN), nitrate (N—NO þ
3 ), ammonium (N—NH4 ), total phosphorus
(Ptot), total suspended solids (TSS) were determined in the Zgierz
WWTP laboratory with the use of the standard methods. Each anal-
ysis was performed at least twice a week.
2.2. BioWin General Model
The simulations of biological wastewater treatment processes
were carried out with the help of BioWin v. 3.0 (EnviroSim
Associates Ltd., Canada). BioWin 3.0 uses the integrated activated
sludge/anaerobic digestion (AS/AD) model, which is referred to
as the BioWin General Model. This model is a combination of
the international models ASM1, ASM2d and ASM3 proposed by
the IWA and, in addition, anaerobic digestion model (ADM).
The BioWin integrated AS/AD model comprises 50 state variables
and 60 process expressions. These expressions are used to de-
scribe the biological processes occurring in activated sludge
and anaerobic digestion systems, several chemical precipitation
reactions, and gas–liquid mass transfer for six gases.
In this work, the part of the BioWin integrated AS/AD model
concerning activated sludge (AS) model is studied. It includes the
following functional categories: (1) growth and decay of Ordinary
Heterotrophic Organisms (OHOs), (2) growth and decay of methy-
lotrophs, (3) hydrolysis, adsorption, ammonification and assimila-
tive denitrification, (4) growth and decay of Ammonia Oxidising
Biomass (AOB), (5) growth and decay of Nitrite Oxidising Biomass
(NOB), (6) growth and decay of ANaerobic AMMonia OXidisers
(ANAMMOX) and (7) growth and decay of phosphorus accumulat-
ing organisms (PAOs). In order to describe these processes 78 ki-
netic parameters and 54 stoichiometric coefficients are included
in the BioWin AS model (compare Fig. 1).
2.3. Data for calibration and validation of the BioWin AS model
The input data for the simulations came from the Zgierz WWTP
and comprised the period of March 2008 for the calibration and
June 2009 for the validation. The values of operating parameters
and wastewater characteristics used for model calibration and val-
idation are presented in Tables 1a and 1b, respectively. These data
obtained from the Zgierz WWTP were subjected to standard statis-
tical elaboration. The confidence intervals were calculated with the
help of t-Student test. A significance level at 95% was assumed. Fi-
nally, the averaged data were introduced into BioWin. The simula-
tions were performed with the assumption of constant process
temperature 12 °C and controlled SRT (21 d) for model calibration,
and 17 °C, SRT equal to 22 d for model validation under steady-
state conditions. The results of calibration were then tested within
the validation process performed for the different period of time.
The applied procedure, which was described above, met the
requirements of a steady-state model calibration and validation
presented by Petersen et al. (2002).
2.4. Sensitivity analysis
Sensitivity analysis enables the evaluation of the extent to
which the parameters used in the model calibration can influence
the model outputs. A parameter with high sensitivity is the one, for
which a small variation in its value causes a large variation in the
response predicted by the model. Analogously, a parameter with
low sensitivity is the one that may be varied over a relatively wide
range inducing only a relatively small variation in the predicted re-
sponse. Pursuant to EPA guidelines (US EPA, 1987) the normalised
sensitivity coefficient (Si,j) is defined as a ratio of the percentage
change in the output variable (yi) to a 10% change in the input var-
iable (xi):
 
Dy =y 
Si;j ¼  i i  ð1Þ
Dxi =xi
In this work, a 10% increase of the input variable was applied for
the purpose of Si,j calculations. For each input variable, i.e. each ki-
netic or stoichiometric parameter calculations were made sepa-
rately. The influence of a parameter on the model output can be
interpreted as follows: (1) Si,j < 0.25 indicates that a parameter
has no significant influence on the model output, (2) 0.25 6 Si,j < 1
 E. Liwarska-Bizukojc, R. Biernacki / Bioresource Technology 101 (2010) 7278–7285 7281

Fig. 2. A schematic flow diagram of biological reactors at the Zgierz WWTP.

Table 1a Eight output variables (yi) characterising the quality of the efflu-
Operating parameters and wastewater characteristics of the Zgierz WWTP (data from ent were taken into account within the calculations of the sensitiv-
March 2008).
ity analysis measures Si,j and dmsqr j . These were the following

Parameter Unit Mean SD Minimum Maximum variables: COD, BOD5, Ntot, Ptot, N—NHþ 4 , N—NO3 , TKN and TSS.
Flow rate 3
m d 1
11,600 1700 8892 18,000 Kinetic parameters in the BioWin AS model are divided into the
Process temperature °C 11.8 0.64 10.3 12.9 following categories: Ammonia Oxidising Biomass (AOB), Nitrite
SRT d 21 0.6 20.3 22 Oxidising Biomass (NOB), ANAMMOX, Ordinary Heterotrophic
RAS flow rate m3 d1 10,425 1510 7991 16,177
Organisms (OHOs), phosphorus accumulating organisms (PAOs),
pH 7.5 0.17 7.1 7.8
TSS mg l1 298 34.4 268 346 pH and switching functions. At the same time the stoichiometric
COD mg O2 l1 890 233 435 1480 coefficients are divided into the same groups excluding pH and
BOD5 mg O2 l1 487 115 260 800 switching functions. Due to the fact that the investigated WWTP
Ntot mg N l1 59.27 5.07 56.3 65.13
system was operated according to the Phoredox scheme, the
TKN mg N l1 58.35 5.18 55 64.33
N—NHþ mg N l1 35.8 5.388 31.31 41.78
parameters related to ANAMMOX process had not to be subjected
4
Ptot mg P l1 8.3 1.72 5.1 13.7 to sensitivity analysis. It was confirmed by the calculations of the
normalised sensitivity coefficient Si,j. For all ANAMMOX parame-
ters Si,j was equal to zero. All in all, 71 kinetic and 46 stoichiometric
parameters of the BioWin AS model were analysed.
Table 1b
Operating parameters and wastewater characteristics of the Zgierz WWTP (data from
June 2009).
3. Results and discussion
Parameter Unit Mean SD Minimum Maximum
Flow rate m3 d1 11,770 3170 7550 20,047 3.1. Model calibration and validation
Process temperature °C 17 1.02 14 19.1
SRT d 22 0.9 19.8 23.2
RAS flow rate m3 d1 10,578 2815 6785 18,016 Model calibration is the estimation of model parameters to fit a
pH 7.5 0.13 7.2 7.8 certain experimental set of data obtained from the studied WWTP.
TSS mg l1 365 109 264 520 Due to the discrepancies between the measured and simulated re-
COD mg O2 l1 844 313 217 1498 sults for the effluent, when default values of parameters were used,
BOD5 mg O2 l1 360 157 55 840
Ntot mg N l1 48.1 19.2 19.6 61.3
calibration of the BioWin AS model occurred to be necessary. These
TKN mg N l1 47.1 19.1 18.8 59.9 discrepancies to a highest extent concerned the following output
N—NHþ 4
mg N l1 30.4 12 12.5 38 variables characterising the effluent: COD, BOD5, N—NHþ 4 and
Ptot mg P l1 12.6 5.37 1.5 28.7 N—NO 3.
First of all, the parameters responsible for the long-term behav-
iour, i.e. yield coefficient for heterotrophic biomass (YH), decay rate
means that a parameter is influential; (3) 1 6 Si,j < 2 means that a constants for heterotrophic biomass (bH), for NOB (baerob,N) and for
parameter is very influential; (4) Si,j P 2 means that a parameter AOB (baerob,A) were calibrated. However, it occurred to be insuffi-
is extremely influential (Petersen et al., 2003). cient. The simulated values of BOD5 and COD in the effluent were
The second sensitivity analysis measure used is the mean still lower than the measured ones. It was the most probably
square sensitivity measure (dmsqr ) introduced by Brun et al. caused by lower affinity of biomass to the substrate (wastewater)
j
(2002). This sensitivity measure is designed to assess the individ- than it was assumed by the authors of BioWin. The half-saturation
ual parameter importance in a least squares parameter estimation constant for heterotrophic biomass, which expresses the affinity of
context and is defined as: heterotrophic biomass to the carbonaceous substrate, was assumed
at the relatively low level of 5 mg COD l1 in BioWin (Table 2). The
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi influent to the Zgierz WWTP contained a significant amount of
u
u1 X n
industrial wastewater (mainly from textile industry) and the affin-
dmsqr ¼t  S2 ð2Þ
j
n i¼1 i;j ity of this kind of substrate is usually lower than for typical muni-
cipal wastewater (Liwarska-Bizukojc et al., 2008). As a result the
A high value of dmsqr
j indicates that a parameter has an impor- default value of KS had to be changed and finally was set to
tant influence on the simulation results, whereas the value of zero 15 mg COD l1. In Table 2, all calibrated parameters are listed and
means that the simulation results do not depend on a parameter. compared with the literature values. The calibrated parameter
7282 E. Liwarska-Bizukojc, R. Biernacki / Bioresource Technology 101 (2010) 7278–7285
Table 2
Comparison of default, calibrated and literature values of the selected parameters.
Parameter Unit Default Calibrated Literature
value value valuesa
YH mg COD mg COD1 0.666 0.74 0.5–0.7
bH d1 0.62 0.90 0.1–0.62
baerob,N d1 0.17 0.23 0.05–0.15
KS mg COD l1 5.0 15 5–30
a
Data for temperature range from 10 to 20 °C given by Henze et al. (2002).
values were higher than default ones, and in some cases exceeded
the literature values (Henze et al., 2002). The elevated values of two
calibrated parameters: aerobic decay for OHO (bH) and for NOB
(baerob,N) are associated with the relatively long SRT in the Zgierz
WWTP. On the one hand a sufficiently long SRT is necessary for
nitrification. On the other hand, if SRT increases, the build-up of
biomass debris is significant, but the active fraction of biomass
drops (Grady et al., 1999).
In order to evaluate the quality of calibration of the Biowin AS
model, it was proposed to check, if the simulated value is included
in the confidence interval estimated for the measured data of the
effluent. If it happens, the simulation is successful because there
is no significant statistical difference between simulated and mea-
sured values of the investigated variable (Petersen et al., 2003). The
simulated and measured values of the basic output variables char-
acterising the effluent quality are presented in Fig. 3a.
Model validation was the next step. The model is considered to
be validated when model predictions agree with measured values
from an independent dataset within the acceptable tolerances (US
EPA, 1993). The results of model validation, which are depicted in
Fig. 3b, indicate that the calibration was performed properly and
the calibrated model can be regarded valid.
3.2. Sensitivity analysis
In this study, kinetic and stoichiometric parameters associated
with growth and decay of AOB, NOB, OHOs and PAOs were sub-
jected to sensitivity analysis. Also the kinetic parameters responsi-
ble for pH inhibition and switching functions were taken into
account. So the sensitivity analysis comprised 71 kinetic and 46
stoichiometric parameters.
In Table 3, the values of the normalised sensitivity coefficient
(Si,j) are shown, if Si,j is higher than 0.25. The coefficient Si,j was cal-
culated with regard to eight output variables as it was written
above in Section 2. However, for one of them, i.e. TSS, the values
of Si,j were always below 0.25 and thus TSS was excluded from Ta-
ble 3. It occurred that out of 117 kinetic and stoichiometric param-
eters, which were studied, 17 can be regarded as sensitive. To be
more detailed, seven parameters can be classified as influential,
eight as very influential and two as extremely influential according
to the classification proposed by Petersen et al. (2003).
The model parameters, which occurred to be sensitive, repre-
sent kinetic and stoichiometric parameters associated with growth
and decay of AOB, OHOs, PAOs and one parameter belongs to
switching functions. Eight out of 17 sensitive parameters are asso-
ciated with growth and decay of PAOs. In this group two parame-
ters are extremely influential (Si,j P 2). These are YP/acetic and YlowPP
(Table 3). Several other studies also emphasized the influential role
of the model parameters associated with PAOs in ASMs and ASM-
based models (Brun et al., 2002; Petersen et al., 2003; Makinia
et al., 2006). Moreover, Makinia et al. (2006) found that the effluent
concentration of P—PO3 4 was affected by the greatest number of
parameters that could be classified as very influential (Si,j P 1).
There were 11 and 6 very influential parameters in ASM2d and
ASM3P, respectively (Makinia et al., 2006). It coincides with the
findings of this study. The effluent concentration of total phospho-
rus, which consists mainly of P—PO3 4 , was affected by 11 parame-
ters. Eight of them are associated with PAOs, whereas two belong
to OHOs group of parameters and one represents switching
functions.
Referring the results of the normalised sensitivity coefficient
(Si,j) calculations (Table 3) to the calibrated parameters (Table 2),
it occurred that three out of four calibrated parameters can be re-
garded as the influential parameters. It warrants the performance
of sensitivity analysis here. Only for one parameter baerob,N, the va-
lue of Si,j was below 0.25 and ranged between 0.02 and 0.1 depen-
dently on the output variable. However, the calibration of this
parameter was necessary in order to receive good prediction of ni-
trate concentration in the effluent.
Most of the parameters identified as influential by Weijers and
Vanrolleghem (1997) in ASM1 were found to be sensitive in this
study for the BioWin AS model. To be more precise, five out of eight
kinetic and stoichiometric parameters belonging to the reduced set
of parameters determined by Weijers and Vanrolleghem (1997)
were also found in this work as influential. These are the parame-
ters associated with growth and decay of OHOs, i.e. YH, lmaxH, bH,
KS and maximum specific growth rate of AOB (lmaxA) (compare
with Table 3). It indicates that irrespective of the used model some
basic parameters of Monod kinetics for OHOs and AOB should be
determined, preferably with the help of an appropriate experimen-
tal procedure. The methodology for this purpose, including inter
alia oxygen uptake rate tests and ammonium uptake rate tests,
has already been elaborated (Petersen et al., 2003).
In order to identify the parameters that cause the highest vari-
ability in the model outputs, the values of the mean square sensi-
tivity measure dmsqr were calculated for all kinetic and
j
stoichiometric parameters, which are valid for the tested Phoredox
system. The model parameters were set in the decreasing order so
that the parameter importance ranking could be obtained (Table
4). Although the discussion focuses on the model parameters after
calibration, the analogous ranking was elaborated additionally for
the BioWin AS model with the default model parameters (without
calibration). It was made in order to verify whether the model is
indeed sensitive to the changes in the parameter values that were
modified during the calibration procedure. For the purpose of the
better legibility only top 10 parameters for the calibrated (ranking
No. 1) and the default (ranking No. 2) parameters were compared
in Table 4. In both rankings half of the top 10 parameters represent
kinetic parameters and the other half – stoichiometric parameters.
However, the stoichiometric coefficients are on the first places of
these rankings and outclass kinetic parameters. Most of the stoi-
chiometric coefficients, which are regarded on the basis of dmsqr j
as the most important parameters, are associated with phosphorus
removal and belong to PAOs group of parameters. It confirmed the
earlier observations made in this work and data to be found in lit-
erature (Makinia et al., 2006; Sun et al., 2009). In both rankings on
the first two places the same stoichiometric parameters associated
with biological phosphorus removal processes, i.e. YP/acetic and
YlowPP, are located. The most important stoichiometric coefficient
YP/acetic is defined as the amount of phosphorus released for 1 mg
of acetate sequestrated in the form of PHA. Its default value is
equal to 0.49 mg P mg COD1. YP/acetic corresponds with the yield
of phosphorus release to substrate uptake Y PO4 , which is used in
ASM2, ASM2d and ASM3P models (Henze et al., 2000; Rieger
et al., 2001). The default value of Y PO4 in ASM2 and ASM2d models
is equal to 0.4 mg P mg COD1, whereas in model ASM3P it is
slightly lower and equal to 0.35 mg P mg COD1 (Henze et al.,
1999; Rieger et al., 2001). The authors of the BioWin AS model refer
the amount of phosphorus released to the substrate in the form of
acetate only. Thus, its default value differs from the default value of
Y PO4 in ASMs. The parameter located on the second place, YlowPP,
 E. Liwarska-Bizukojc, R. Biernacki / Bioresource Technology 101 (2010) 7278–7285 7283

Fig. 3. Measured vs. simulated values of the output variables characterising the effluent: (a) model calibration and (b) model validation.

expresses the fraction of phosphorus stored in releasable poly-
phosphate-P and its default value is equal to 0.94 mg P mg P1.
The rest of phosphorus is stored in the form of high molecular
weight non-releasable polyphosphate-P. This stoichiometric coeffi-
cient is typical for the BioWin AS model. Its value cannot exceed
one and seems to be relatively stable for PAOs. However, the meth-
ods for its estimation have not been elaborated so far. In the review
by Petersen et al. (2003), which summarised the achievements in
the estimation of kinetic and stoichiometric parameters of ASMs,
the parameters associated with PAOs were not included. These
authors wrote that it was difficult to differentiate between the
kinetics of growth and storage and, as a result, the reliable tests
for determination of kinetic parameters of PAOs were not elabo-
rated (Petersen et al., 2003). In the meantime several attempts
aiming at the elaboration of the procedure for the determination
of kinetic and stoichiometric parameters for PAOs have been made
(Pala and Bölükbas, 2005; Vaiopoulou and Aivasidis, 2007; Ferrai
et al., 2010). However, the estimated parameters are not always re-
ferred to substrate concentration expressed as COD (Pala and
Bölükbas, 2005; Vaiopoulou and Aivasidis, 2007).
On the next places in both rankings there are the stoichiometric
coefficients associated with PAOs and OHOs. The location of these
coefficients in the rankings is different. The third place in ranking
No. 1 occupies YP/PHA,seq, whereas in ranking No. 2 this parameter
is on the fourth place. YP/PHA,seq expresses the amount of PHA
stored when 1 mg of acetate or propionate is sequestrated and,
similarly to YlowPP is typical only for the BioWin AS model. Last
stoichiometric parameter, which refers to PAOs and is present
among the top 10 parameters, is YP/PHA,aerobic. This parameter ex-
presses the amount of phosphorus stored per unit of PHA oxidised
under aerobic conditions.
This study showed that the number of processes and, as a con-
sequence, the number of parameters related to PAOs is signifi-
cantly extended in the BioWin AS model in comparison to ASM2,
7284 E. Liwarska-Bizukojc, R. Biernacki / Bioresource Technology 101 (2010) 7278–7285

Table 3 well-known for environmental engineers. One difference between

The values of Si,j for the most sensitive (Si,j P 0.25) parameters of the calibrated two rankings just concerns kinetic parameters (Table 4). In ranking
BioWin AS model.
No. 2 maximum specific growth rate of OHOs (lmaxH) under aero-
Parameters Si,j bic conditions is replaced with maximum specific growth rate of
COD BOD Ntot Ptot N—NHþ
4
N—NO
3 TKN PAOs (lmaxPAO). Other kinetic parameters as lmaxA, baerob,A, bH
and K NH4 are present in both rankings, however on different places.
AOB
Kinetic It was recommended to use the respirometric methods for the pur-
lmaxA 1.481 pose of the determination of the above-mentioned parameters
K NH4 1.111 (Petersen et al., 2003). One of these kinetic parameters, i.e. decay
baerob,A 1.111 rate constant for heterotrophic biomass (bH) was tuned in the stea-
OHO
Kinetic
dy-state calibration (Table 2).
lmaxH 1.042 Comparing the results of the calibration and the sensitivity
KS 0.841 analysis, few words should be said about half-saturation constant
bH 1.024 0.270 0.300 for heterotrophs (KS). It is a very important parameter in Monod-
ka 0.330
type kinetics, which was changed within the calibration of the Bio-
Stoichiometric
YH 0.475 0.755 1.081 1.053 Win AS model (Table 2). This constant occurred to be influential on
the basis of Si,j calculations, and ranks on 11th place in both impor-
PAO
Kinetic tance rankings (Tables 3 and 4). Similarly to the parameters asso-
lmaxPAO 0.541 ciated with growth and decay of OHOs and AOB i.e. lmaxA,
lmaxP-limited 0.811 baerob,A, bH and K NH4 , half-saturation constant for OHOs can be also
KS,PAO 0.541 determined with the help of respirometric batch tests (Kappeler
KS,P-limited 0.541
Stoichiometric
and Gujer, 1992; Liwarska-Bizukojc et al., 2008; Damayanti et al.,
YP/PHA,aerobic 1.621 2010).
YP/PHA,seq 1.889
YP/acetic 2.703
YlowPPa 0.375 2.162 4. Conclusions
Switches
KPP 0.811 Sensitivity analysis is a very helpful tool to identify the most
a
Calculated for 10% decrease of YlowPP due to the upper limit set on this important and sensitive parameters of mathematical models, espe-
parameter. cially if a model contains many parameters and is, in some cases,
overparameterised with respect to the available data.
The application of the normalised sensitivity coefficient (Si,j) al-
Table 4
Kinetic and stoichiometric parameters importance rankings: (1) for the calibrated
lowed for the determination of the set of 17 most important
BioWin AS model and (2) without calibration. parameters in the BioWin AS model. Half of the influential param-
eters are associated with growth and decay of PAOs. It occurred
Ranking No. 1 Ranking No. 2
that the evaluation methods should be elaborated for some influ-
Parameter dmsqr Parameter dmsqr ential stoichiometric coefficients related to PAOs. At the same time
YP/acetic 1.022 YP/acetic 1.040 other widely known stoichiometric and kinetic parameters as YH,
YlowPP 0.833 YlowPP 0.966 lmaxH, KS, lmaxA, baerob,A, bH and K NH4 can be estimated with the
YP/PHA,seq 0.717 YP/PHA,aerobic 0.758
use of respirometric tests.
YH 0.663 YP/PHA,seq 0.756
YP/PHA,aerobic 0.613 lmaxA 0.569 This work should support both practitioners and scientists. The
lmaxA 0.572 YH 0.469 selection of the set of the most sensitive parameters of the BioWin
baerob,A 0.426 baerob,A 0.429 AS model is particularly important for the users of BioWin soft-
K NH4 0.425 K NH4 0.427
ware. At the same time the attention of scientists should be drawn
bH 0.420 bH 0.296
lmaxH 0.397 lmaxPAO 0.284
to the parameters, for which the determination procedures have
not been elaborated until now.

ASM2d and ASM3P models. Thus, some parameters introduced to Acknowledgements

the BioWin AS model are relatively new and their understanding
as well as determination needs further investigations. At the same
time this study confirmed that the well-known and commonly
used in biological systems yield coefficient YH is also very sensitive
in the Biowin AS model and demands calibration. It was also nec-
essary in this work (compare Tables 2 and 4).
Parameters associated with PAOs were not modified within
steady-state calibration performed here. It occurred that the mea-
sured and simulated phosphorus concentrations in the effluent
were so close to each other that there was no need to do it. How-
ever, the knowledge that these parameters are sensitive may prove
helpful in further dynamic calibration of the BioWin AS model. It
might also support other users in situation, when any discrepan-
cies between measured and simulated values of phosphorus con-
centrations appear.
Five kinetic parameters, which belong to the top 10 parameters,
are in general related to growth and decay of OHOs and AOB. All
these parameters have been already applied in ASMs and are
This work was financed by Ministry of Science and Higher Edu-
cation (Republic of Poland), Project No. N N209 102 637. We would
like to also thank the Head of the Zgierz WWTP Grzegorz Kacprzak
and laboratory staff in the Zgierz WWTP for their help in this
research.
References
Brun, R., Kuhni, M., Siegrist, H., Gujer, W., Reichert, P., 2002. Practical identifiability
of ASM2d parameters – systematic selection and tuning of parameters subsets.
Water Res. 36 (16), 4113–4127.
Damayanti, A., Ujang, Z., Salim, M.R., Olsson, G., Sulaiman, A.Z., 2010. Respirometric
analysis of activated sludge models from palm and oil effluent. Bioresour.
Technol. 101, 144–149.
Ferrai, M., Guglielmi, G., Andreotolla, G., 2010. Modelling respirometric tests for the
assessment of kinetic and stoichiometric parameters on MBBR biofilm for
municipal wastewater treatment. Environ. Modell. Softw. 25 (5), 626–632.
Ferrer, J., Seco, A., Serralta, J., Ribes, J., Manga, J., Asensi, E., Morenilla, J.J., Llavador, F.,
2008. DESSAS: a software tool for designing, simulating and optimising WWTPs.
Environ. Modell. Softw. 23 (1), 19–26.
 E. Liwarska-Bizukojc, R. Biernacki / Bioresource Technology 101 (2010) 7278–7285
Gernaey, K.V., van Loosdrecht, M.C.M., Henze, M., Lind, M., Jorgensen, S.B., 2004.
Activated sludge wastewater treatment plant modelling and simulation. State
of the art. Environ. Modell. Softw. 19 (9), 763–783.
Grady Jr., C.P.L., Daigger, G.T., Lim, H.C., 1999. Biological Wastewater Treatment.
Marcel Dekker, New York.
Henze, M., Gujer, W., Mino, T., Matsuo, T., Wenzel, M.C., Marais, C.V.R., 1999.
Activated sludge model no. 2d. Water Sci. Technol. 39 (1), 165–182.
Henze, M., Gujer, W., Mino, T., van Loosdrecht, M.C.M., 2000. Activated Sludge
Models ASM1, ASM2, ASM2d and ASM3, IWA, Sci. and Tech. Report No. 9. IWA
Publishing, London.
Henze, M., Harremoës, P., Jansen, J., Arvin, E., 2002. Wastewater Treatment.
Biological and Chemical Processes. Springer-Verlag, Berlin, Heidelberg.
Kappeler, J., Gujer, W., 1992. Estimation of kinetic parameters of heterotrophic
biomass under aerobic conditions and characterization of wastewater for
activated sludge modelling. Water Sci. Technol. 25 (6), 125–139.
Langergraber, G., Rieger, L., Winkler, S., Alex, J., Wiese, J., Owerdieck, C., Ahnert, M.,
Simon, J., Maurer, M., 2004. A guideline for simulation studies of wastewater
treatment plants. Water Sci. Technol. 50 (7), 131–138.
Liwarska-Bizukojc, E., Scheumann, R., Drews, A., Bracklow, U., Kraume, M., 2008.
Effect of anionic and nonionic surfactants on the kinetics of the aerobic
heterotrophic biodegradation of organic matter in industrial wastewater. Water
Res. 42 (4–5), 923–930.
Makinia, J., 2006. Performance Prediction of Full Scale Biological Nutrient Removal
Systems using Complex Activated Sludge Models. D.Sc Thesis, Univesität
Hannover, Hannover.
Makinia, J., Rosenwinkel, K.-H., Spering, V., 2006. Comparison of two model
concepts for simulation of nitrogen removal at a full scale biological nutrient
removal pilot plant. J. Environ. Eng. ASCE 132 (4), 476–487.
Melcer, H., Dold, P.L., Jones, R.M., Bye, C.M., Takacs, I., Stensel, H.D., Wilson, A.W.,
Sun, P., Bury, S., 2003. Methods for wastewater characterization in activated
sludge modelling. Water Environment Research Foundation (WERF),
Alexandria, VA, USA.
Pala, A., Bölükbas, Ö., 2005. Evaluation of kinetic parameters for biological CNP
removal from a municipal wastewater through batch tests. Process Biochem. 40
(2), 629–635.
Petersen, B., Gernaey, K., Henze, M., Vanrolleghem, P.A., 2002. Evaluation of an
ASM1 model calibration procedure on a municipal–industrial wastewater
treatment plant. J. Hydroinform. 4, 15–38.
7285
Petersen, B., Gernaey, K., Henze, M., Vanrolleghem, P.A., 2003. Calibration of
activated sludge models: a critical review of experimental designs. In: Agathos,
S.N., Reineke, W. (Eds.), Biotechnology for the Environment: Wastewater
Treatment and Modelling. Waste Gas Handling. Kluwer Academic Publishers,
Dordrecht.
Rieger, L., Koch, G., Kühni, M., Gujer, W., Siegrist, H., 2001. The EAWAG BIO-P
module for activated sludge model no. 3. Water Res. 35 (16), 3887–3903.
Saltelli, A., Ratto, M., Andres, T., Campolongo, R., Cariboni, J., Gatelli, D., Saisana, M.,
Tarantola, S., 2008. Global Sensitivity Analysis. The Primer. John Wiley and Sons
Ltd., Chichester.
Sin, G., van Hulle, S.W.H., De Pauw, D.J.W., van Griensven, A., Vanrolleghem, P.A.,
2005. A critical comparison of systematic calibration protocols for activated
sludge models: a SWOT analysis. Water Res. 39 (12), 2459–2474.
Sin, G., Gernaey, K.V., Neumann, M.B., van Loosdrecht, M.C.M., Gujer, W., 2009.
Uncertainty analysis in WWTP model applications: a critical discussion using an
example from design. Water Res. 43 (11), 2894–2906.
Sun, P., Wang, R., Fang, Z., 2009. Fully coupled activated (FCASM): model
development. Bioresour. Technol. 100, 4632–4641.
US EPA, 1987. QUAL2E – The Enhanced Stream Water Quality Model EPA/823/B-95/
003. Environmental Research Laboratory, Athens, GA, USA.
US EPA, 1993. Manual Nitrogen control EPA/625/R-93/010. US EPA, Washington, DC,
USA.
US EPA, 2009. Guidance on the Development, Evaluation and Application on
Environmental Models EPA 100/K-09-003. US EPA, Washington, DC, USA.
Vaiopoulou, A.E., Aivasidis, B.A., 2007. A semi-batch on-line method for biokinetics
determination in an enhanced biological phosphorus removal system. In:
Proceedings of European Congress of Chemical Engineering (ECCE-6), 16–20
September 2007, Copenhagen, Denmark.
van Veldhuizen, H.M., van Loosdrecht, M.C.M., Heijnen, J.J., 1999. Modelling
biological phosphorus and nitrogen removal in a full scale activated sludge
process. Water Res. 33, 3459–3468.
Vanrolleghem, P.A., Insel, G., Petersen, B., Sin, G., De Pauw, D., Nopens, I., Weijers, S.,
Gernaey, K., 2003. A comprehensive model calibration procedure for activated
sludge models. In: Proceedings of WEFTEC 2003, 76th Annual Technical
Exhibition and Conference, 11–15 October 2003, Los Angeles, CA, USA.
Weijers, S.R., Vanrolleghem, P.A., 1997. A procedure for selecting the most
important parameters in calibrating the activated sludge model no. 1 with
full scale plant data. Water Sci. Technol. 36 (5), 69–79.